"cbmpinr
f ,:,,:,.-,* U., :.
research&development
Table of Contents
Section Title / Contents
Table of Contents
BOSC Computational Toxicology Subcommittee Members
Meeting Agenda
Charge Questions
Computational Toxicology Implementation Plan for FY2009-2012
BOSC Poster Abstracts
CTRP Scientist Biosketches
Training, Outreach and Leadership by the NCCT
- Scientific Leadership Roles
- Mentoring
- Computational Toxicology Rotational Fellowship Program
- Communities of Practice Presentations
- Partnership Agreements (MTAs, MoUs, CRADAs, and lAGs)
CTRP Bibliographic Information
-NCCT
- New Starts
Prior BOSC Letter Reports and ORD Responses
- 2005 Report and Response
- 2006/2007 Report and Response
- 2008 Report and Response
UNITED STATES ENVIRONMENTAL PROTECTION AGENCY
901*01
" * n
"
111
• * (
iMi 11111 (
,^ • •
S!" ." •
" 1
COMPUTATIONAL
TOXICOLOGY"
-------
B'O-S-C
BOARD OF SCIENTIFIC COUNSELORS
September 2, 2009
COMPUTATIONAL TOXICOLOGY SUBCOMMITTEE
ROSTER
CHAIR
George Daston, Ph.D.
Research Fellow
Miami Valley Laboratories
The Proctor & Gamble Company
Members
James Clark, Ph.D.
Distinguished Scientific Associate
Exxon Mobil Research & Engineering Co.
Richard Di Giulio, Ph.D.
Professor
Nicholas School of the Environment and Earth Sciences
Duke University
AH Faqi, DVM, Ph.D. DABT
Director, Developmental and Reproductive Toxicology
MPI Research, Inc.
Lawrence Hunter, Ph.D.
Director, Center for Computational Pharmacology and Computational Bioscience Program
University of Colorado
M. Moiz Mumtaz, Ph.D.
Science Advisor
Division of Toxicology and Environmental Medicine
Agency for Toxic Substances and Disease Registry
Dennis Paustenbach, Ph.D., CIH, DABT
President and Founder
ChemRisk, Inc.
John Quackenbush, Ph.D.
Professor of Biostatistics and Computational Biology
Department of Biostatistics and Computational Biology
Dana-Farber Cancer Institute
A Federal Advisory Committee for the U. S. Environmental Protection Agency's Office of Research and Development
Previous I TOC
-------
Computational Toxicology Subcommittee Roster
Santiago Schnell, Ph.D.
Associate Professor of Molecular and Integrative Physiology
Brehm Investigator, Brehm Center for Type 1 Diabetes Research and Analysis
Research Associate Professor of Computational Medicine and Bioinformatics
University of Michigan Medical School
Cynthia Stokes, Ph.D.
Independent Consultant
Katrina Waters, Ph.D.
Computational Biology & Bioinformatics Group
Battelle
Previous
TOC
-------
BOARD OF SCIENTIFIC COUNSELORS
COMPUTATIONAL TOXICOLOGY SUBCOMMITTEE
DRAFT AGENDA
September 29-30, 2009
Hilton Raleigh-Durham Airport at Research Triangle Park
4810 Page Creek Lane
Durham, NC 27703
Telephone: (919) 941-6000
August 26, 2009
Tuesday, September 29, 2009
12:00 p.m. - 12:30 p.m. Registration
12:30 p.m. - 12:40 p.m.
12:40 p.m.-12:45 p.m.
12:45 p.m.-1:00 p.m.
1:00 p.m.-1:45 p.m.
1:45 p.m.-2:15 p.m.
2:15 p.m. -4:15 p.m.
4:15 p.m. -5:15 p.m.
5:15 p.m. - 6:15 p.m.
6:15 p.m.
Welcome and Introductions
- New Subcommittee Members
- Draft Charge
- Meeting Agenda
DFO Remarks
Dr. George Daston,
Subcommittee Chair
Ms. Lori Kowalski, Office of
Research and Development (ORD)
Computational Toxicology Research Mr. Lek Kadeli, ORD Acting
Program (CTRP) - Critical Component Assistant Administrator (AA)
of EPA Science in the 21st Century
CTRP Overview
Introduction to Poster Session I:
Informatics, Exposure Science,
ORD, and External Partners
Poster Session I
Poster Session I: Discussion
Comments on the CTRP
Dr. Robert Kavlock, Director,
National Center for Computational
Toxicology (NCCT)
Dr. Ann Richard, NCCT
Subcommittee/ORD
Subcommittee/ORD
Dr. Peter Preuss, Director, ORD/National Center
for Environmental Assessment; Mr. Jim Jones,
Deputy AA, EPA/Office of Prevention,
Pesticides, and Toxic Substances; Dr. John
Bucher, Associate Director, National
Toxicology Program, National Institutes of
Health; Dr. Cal Baier-Anderson, Senior Health
Scientist, Environmental Defense Fund
Adjourn
A Federal Advisory Committee for the U. S. Environmental Protection Agency's Office of Research and Development
Previous
TOG
Next
-------
BOSC COMPUTATIONAL TOXICOLOGY SUBCOMMITTEE SEPTEMBER 2009 MEETING AGENDA
Wednesday, September 30, 2009
8:00 a.m. - 8:30 a.m.
8:30 a.m. - 10:30 a.m.
10:30 a.m. - 11:30 a.m.
Introduction to Poster Session II: Dr. Thomas Knudsen, NCCT
High Throughput Screening, Toxicity
Predictions, Virtual Tissues, and
Uncertainty Analysis
Poster Session II
Poster Session II: Discussion
11:30 a.m. - 12:00 noon CTRP Future: Providing High
Throughput Decision Support
Tools for Screening and Assessing
Chemical Exposure, Hazard, and Risk
12:00 noon- 12:15 p.m. Public Comment
12:15 p.m. - 1:15 p.m. Working Lunch
1:15 p.m. - 3:30 p.m. Subcommittee Working Time
3:30 p.m. Adjourn
Subcommittee/ORD
Subcommittee/ORD
Dr. David Dix, Acting Deputy
Director, NCCT
Subcommittee
Subcommittee
Previous
TOC
-------
BOARD OF SCIENTIFIC COUNSELORS
August 24, 2009
DRAFT Charge
Computational Toxicology Subcommittee
September 29-30, 2009 Meeting
Background
The National Center for Computational Toxicology (NCCT) became operational on February 20,
2005. On April 25-26, 2005, the BOSC Computational Toxicology Subcommittee held its first
public meeting at the Office of Research and Development's (ORD) Research Triangle Park (RTF),
North Carolina facility, where the majority of NCCT staff is located. This meeting was intended as
the first of several consultative reviews of the Center's progress, and was prospective in nature, due
to the newness of the Center. The Subcommittee developed a letter report from the April meeting
which addressed six charge questions that concentrated on the NCCT's strategic goals; its
collaborations, and connectedness to the rest of the Agency and to outside scientists; its staffing
plan; and its thematic choices. The letter report was finalized by the BOSC Executive Committee
and transmitted to ORD in July 2005. A formal response of the NCCT to the review was provided
to the BOSC at their September 2005 Executive Committee meeting.
The Subcommittee met again in a June 2006 review to continue to provide the NCCT with advice
on the progress the Center had made, since Spring 2005, in fulfilling its mission and strategic goals.
The subcommittee addressed nine charge questions, which touched on the new extramural
bioinformatics centers, effectiveness of Center work in Agency research, use of computational tools,
feedback on the first generation computational toxicology implementation plan, effectiveness of
Center communication of research, outcomes, and responsiveness to stakeholder needs. A letter
report was finalized by the BOSC Executive Committee and transmitted to ORD in December 2006.
A formal response of the NCCT to the review was provided to the BOSC at their January 2007
Executive Committee meeting.
In December 2007, the Subcommittee met to discuss progress the NCCT had made with respect to
five NCCT activities: ToxCast; Information Management/Information Technology (IM/IT) -
Informatics; Virtual Liver; Developmental Systems Biology; and Arsenic Biologically-Based Dose
Response Model. A letter report was finalized by the BOSC Executive Committee and transmitted
to ORD in September 2008. A formal response of the NCCT to the review was provided to the
BOSC at their February 2009 Executive Committee meeting. That response included more details
on progress with informatics (ACToR, DSSTox, ToxRefDB), chemical prioritization (ToxCast and
ExpoCast), and systems modeling (v-Liver and v-Embryo); and an explanation of the disinvestment
in arsenic research.
September 2009 Review
The purpose of the September 2009 review is to provide the NCCT and broader ORD
Computational Toxicology Research Program (CTRP) with 1) advice/recommendations on the
progress the CTRP has made in the past four and a half years in fulfilling its mission and strategic
A Federal Advisory Committee for the U.S. Environmental Protection Agency's Office of Research and Development
Previous I TOG I Next
-------
BOSC COMPUTATIONAL TOXICOLOGY SUBCOMMITTEE SEPTEMBER 2009 DRAFT CHARGE
goals; and 2) advice/recommendations on whether the NCCT should continue as an established
organization beyond its original five-year charter. In particular, the subcommittee will address the
following questions:
Charge Question 1: What is your evaluation of the progress the CTRP has made in achieving its
original goals and objectives, and whether it has efficiently utilized available resources?
Charge Question 2: To what extent and how effectively has the CTRP utilized internal and external
partnerships to foster its goals?
Charge Question 3: What evaluation can you provide relative to the contributions of the CTRP to
the advancement of transforming the field of toxicity testing?
Charge Question 4: To what extent do the ORD intramural projects, the extramural STAR centers,
and the five stated CTRP management priorities described in the FY09-12 implementation plan
combine to efficiently support the goal of providing high throughput decision support tools for
screening and assessing chemical exposure, hazard and risk to human health?
Charge Question 5: The NCCT was established as an organization with a five-year charter ending
in February 2010, which would continue dependent on: 1) meeting established goals; and 2) having
continuing mission-critical goals and objectives. What recommendation(s) can you provide the
Agency regarding continuation of the NCCT as an established organization, and the criticality of its
goals and objectives to EPA?
Previous I TOC
-------
00011001t Qi 00
i 0 00i 10 1110111
COMPUTATIONAL
TOXICOLOGY
U.S. EPA OFFICE OF RESEARCH AND DEVELOPMENT
COMPUTATIONAL TOXICOLOGY RESEARCH PROGRAM
IMPLEMENTATION PLAN FOR FISCAL YEARS 2009-2012
Providing High Throughput Decision Support Tools for Screening
and Assessing Chemical Exposure, Hazard and Risk
BOSC Review Draft- 24 August, 2009
DISCLAIMER:
This document has been reviewed by the U.S. EPA Office of Research and Development (ORD)
and approved for public release, but does not necessarily constitute official Agency policy. This
Plan follows the first generation FY2005-2008 Computational Toxicology Research Program
(CTRP) Implementation Plan, and provides a strategic overview of research for FY2009-2012.
This Plan was reviewed by ORD senior management and members of the Science Council, as
well as the Computational Toxicology Subcommittee of the ORD Board of Scientific Counselors
(BOSC) on September 29-30, 2009, in RTF, NC.
Previous
TOC
-------
EPA CompTox Research Program FY2009-2012 BOSC Review Draft- 24 August, 2009
TABLE OF CONTENTS
LIST OF FIGURES iv
LIST OF TABLES iv
ACRONYMS v
EXECUTIVE SUMMARY vii
I. History of the CTRP and the NCCT 1
A. Defining the Mission of Computational Toxicology at EPA 1
B. Timeline of CTRP Development 2
C. Resources for the CTRP 4
1. Funding and NCCT Personnel 4
D. BOSC Reviews of the CTRP 6
E. NRC Report and EPA's Strategic Plan for 21 st Century Toxicology 7
1. NRC Report on Toxicity Testing in the 21st Century 7
2. EPA Strategic Plan for Evaluating the Toxicity of Chemicals 8
F. Significant Accomplishments of the CTRP: FY2006-2008 9
1. Accomplishments of the NCCT 9
2. Accomplishments by the CTRP ORD Intramural Partners 12
3. Accomplishments by STAR Grantees in the CTRP 16
G. Summary on Retrospective of the CTRP and NCCT 17
II. Revision of the CTRP for FY2009-2012 18
A. Maturation of the Program 18
B. CTRP Integration across Other ORD Laboratories and Centers 21
C. Regional and Program Interactions 26
D. Priority Areas for CTRP Management 29
1. Toxicity Predictions and Chemical Prioritizations Incorporating Exposure 29
2. Strengthening Cross-ORD Collaborations 29
3. Tox21: A Federal Partnership Transforming Toxicology 30
4. Communicating Computational Toxicology 32
a. EPA Program Office Training and Implementation of Computational Tools 32
b. Communities of Practice for Chemical Prioritization and Exposure Science 32
5. Developing Clients for Virtual Tissues 33
III. CTRP Project Summaries for FY2009-2012 35
A. Intramural Projects Coordinated by NCCT 35
1. ACToR 35
2. DSSTox 36
3. ToxRefDB 37
4. ChemModel 37
5. ToxCast™ 37
6. ExpoCast™ 37
7. v-Embryo™ 39
8. v-Liver™ 39
9. Uncertainy 39
B. Intramural Projects Coordinated by NERL and NHEERL 40
1. NERL 40
11
Previous I TOC
-------
EPA CompTox Research Program FY2009-2012 BOSC Review Draft- 24 August, 2009
2. NHEERL 41
C. Extramural STAR Grantee Projects 43
D. Summary Integration of the CTRP Projects for FY2009-2012 44
IV. Appendices A-l
A. Intramural CTRP Projects A-l
1. Project Plans A-l .a
a. ACToR - Aggregated Computational Toxicology Resource A-l .a. 1
b. DSSTox- Chemical Information Technologies in Support of Toxicology
Modeling A-1 .b. 1
c. ToxRefDB - Toxicity Reference Database A-l.c.l
d. ChemModel- The Application of Molecular Modeling to Assessing Chemical
Toxicity A-1 .d. 1
e. ToxCast™- Screening and Prioritization of Environmental Chemicals Based on
Bioactivity Profiling and Predictions of Toxicity A-l .e.l
f. ExpoCast™- Exposure science for screening, prioritization, and toxicity
testing A-l .f. 1
g. v-Embryo™ - Virtual Embryo A-l.g. 1
h. v-Liver M- Virtual Liver Project A-l.h.l
i. Uncertainty Analysis in Toxicological Modeling A-l.i. 1
2. Project Outcomes Table A-2.1
B. Extramural STAR Centers B-l
C. FY2004 "New Start" Award Bibliography C-l
D. EPA Strategic Plan for Evaluating the Toxicity of Chemicals D-l
in
Previous I TOC
-------
LIST OF FIGURES
Figure 1 -ORD Computational Toxicology Research Program Development 8
Figure 2-CTRP Budget History 10
Figure 3 -NCCT Organizational Chart 11
Figure 4 -Computational Toxicology in ORD 27
Figure 5 -The Future State: Using Hazard and Exposure Predictions to Prioritize Testing and
Monitoring 28
Figure 6 -Applying Computational Toxicology Along the Source to Outcome Continuum 38
LIST OF TABLES
Table 1-Project Outcomes Table A-1.2
IV
Previous
TOC
-------
EPA CompTox Research Program FY2009-2012 BOSC Review Draft- 24 August, 2009
ACRONYMS
ACToR Aggregated Computational Toxicology Resource
BOSC Board of Scientific Counselors
CPCP Chemical Prioritization Community of Practice
CoP Communities of Practice
CTISC Computational Toxicology Implementation and Steering Committee
CTRP Computational Toxicology Research Program
DNA Deoxyribonucleic acid
DSSTox Distributed Structure-Searchable Toxicity Database
EDC Endocrine Disrupting Compounds
EPA U.S. Environmental Protection Agency
ExpoCast™ Exposure Forecasting Project
ExpoCop Exposure Science Community of Practice
FTE Full Time Equivalents
FTTW Future of Toxicity Testing Workgroup
FY Fiscal Year
HTS High Throughput Screening
IAG Interagency Agreements
IRIS Integrated Risk Information System
KEGG Kyoto Encyclopedia of Genes and Genomes
LTG Long Term Goal
MICA Mechanistic Indicators of Childhood Asthma
MO A Modes or Mechanisms of Action
MOU Memorandum of Understanding
MTA Material Transfer Agreements
MYP Multi Year Plan
NCCT National Center for Computational Toxicology
NCGC NIH Chemical Genomics Center
NCEA National Center for Environmental Assessment
NCER National Center for Environmental Research
NCGC National Institutes of Health Chemical Genomics Center
NGO Non-Governmental Organization
NERL National Exposure Research Laboratory
NHEERL National Health and Environmental Effects Research Laboratory
NHGRI National Institutes of Health Chemical Genome Research Institute
NIEHS National Institute of Environmental Health Sciences
NIH National Institutes of Health
NPD National program Directors
NRC National Research Council
NRMRL National Risk Management Research Laboratory
NTP National Toxicology Program
OECD Organization for Economic Cooperation and Development
OPP Office of Pesticide Programs
OPPT Office of Pesticides, Prevention, and Toxics
OPPTS Office of Pesticides, Prevention, and Toxic Substances
Previous I TOC
-------
EPA CompTox Research Program FY2009-2012
BOSC Review Draft- 24 August, 2009
ORD Office of Research and Development
OSCP Office of Science Coordination and Policy
OW Office of Water
QSAR Quantitative Structure Activity Relationship
RFA Request for Applications
RNA Ribonucleic acid
SAB Science Advisory Board
SEE Senior Environmental Enrollee
STAR Science to Achieve Results
ToxCast™ Toxicity Forecasting Project
ToxRefDB Toxicity Reference Database
v-Embryo™ Virtual Embryo Project
v-Liver™ Virtual Liver Project
v-Liver-KB v-Liver Knowledgebase
VI
Previous
TOC
-------
EPA CompTox Research Program FY2009-2012
BOSC Review Draft- 24 August, 2009
EXECUTIVE SUMMARY
This document lays out the fiscal year 2009-2012 objectives of the U.S. Environmental
Protection Agency (EPA), Office of Research and Development (ORD) research program in
computational toxicology. Computational toxicology is the application of mathematical and
computer models to help assess chemical hazards and risks to human health and the
environment. Supported by advances in informatics, high-throughput screening (HTS)
technologies, and systems biology, EPA is developing robust and flexible computational tools
that can be applied to the thousands of chemicals in commerce, and contaminant mixtures found
in America's air, water, and hazardous-waste sites. The ORD Computational Toxicology
Research Program (CTRP) is composed of three main elements. The largest component is the
National Center for Computational Toxicology (NCCT), which was established in 2005 to
coordinate research on chemical screening and prioritization, informatics, and systems modeling.
The second element consists of related activities in the National Health and Environmental
Effects Research Laboratory (NHEERL) and the National Exposure Research Laboratory
(NERL). The third and final component consists of academic centers working on various aspects
of computational toxicology and funded by the EPA Science to Achieve Results (STAR)
program. Together these elements form the key components in the implementation of both the
initial strategy, A Framework for a Computational Toxicology Research Program (US EPA,
2003), and the newly released The U.S. Environmental Protection Agency's Strategic Plan for
Evaluating the Toxicity of Chemicals (US EPA, 2009). Key intramural projects of the CTRP
include digitizing legacy toxicity testing information toxicity reference database (ToxRefDB),
predicting toxicity (ToxCast™) and exposure (ExpoCast™), and creating virtual liver (V
Liver™) and virtual embryo (v-Embryo™) systems models. EPA funded STAR centers are also
providing bioinformatics, computational toxicology data and models, and developmental toxicity
data and models. All CTRP projects participate in the Agency's formal quality assurance (QA)
program and regularly undergo peer review. The models and underlying data are being made
publicly available through the Aggregated Computational Toxicology Resource (ACToR), the
Distributed Structure-Searchable Toxicity (DSSTox) Database Network, and other EPA
websites. Thus the CTRP is providing the foundation for advancing high-throughput toxicology
and risk assessments, and thereby closing critical data gaps for thousands of chemicals and
helping EPA better assess and manage chemical risk.
The CTRP is evolving beyond the initial focus on hazard identification and chemical
prioritization, as expressed in the new long term goal of providing high-throughput decision
support tools for assessing chemical exposure, hazard and risk. There is an increasing emphasis
on using high-throughput bioactivity profiling data in systems modeling to support quantitative
risk assessments, and greater involvement in developing complementary higher throughput
exposure models. Discussions are well underway between NCCT, NHEERL, NERL, the
National Risk Management Research Laboratory (NRMRL), the National Center for
Environmental Assessment (NCEA), and the National Center for Environmental Research
(NCER, managers of the STAR program) on how the CTRP will play a major role in future
integrated ORD programs centered on looking at chemical hazards and risks from a life cycle
viewpoint. This integrated approach will enable analysis of life stage susceptibility, and
understanding of the exposures, pathways and key events by which chemicals exert their toxicity
in developing systems (e.g., endocrine related pathways). The CTRP will be a critical component
Previous
vn
TOC
-------
EPA CompTox Research Program FY2009-2012 BOSC Review Draft- 24 August, 2009
in next generation risk assessments utilizing quantitative high-throughput data and providing a
much higher capacity for assessing chemical toxicity than is currently available.
This second generation CTRP implementation plan is highly consistent with the Agency's
priority for improving the management of chemical and contaminant risks. A January 2009 a
memo sent to all EPA employees from Administrator Jackson listed managing chemical risks as
one of five top priorities and stated:
"More than 30 years after Congress enacted the Toxic Substances Control Act, it is clear that
we are not doing an adequate job of assessing and managing risks of chemicals in consumer
products, the workplace and the environment. It is now time to revise and strengthen EPA's
chemicals management and risk assessment programs."
With contributions from across ORD, the CTRP will provide EPA program offices better
decision analysis tools for hazard and exposure screening and assessment, which can then be
used to better manage the risks of chemicals. The CTRP is acquiring an international reputation
for leadership in the introduction of innovative high-throughput technologies and computational
approaches for identifying toxicity pathways and characterizing response to environmental
exposures. It is through this effort that problems will be addressed, and solutions to EPA's
chemicals management and risk assessment programs will be developed.
Vlll
Previous I TOC
-------
Mission statement of Computational
Toxicology: To integrate modern
computing and information technologies
with molecular biology to provide the
Agency with decision support tools for
high-throughput risk assessment.
I. HISTORY OF THE CTRP AND THE NCCT
A. Defining the Mission of Computational Toxicology at EPA
Computational toxicology applies mathematical
and computer models and molecular biological
and chemical approaches to explore both
qualitative and quantitative relationships
between chemical exposure and adverse health
outcomes. Recent technological advances make
it possible to develop molecular profiles using
high-throughput and high content methods that
identify the impacts of environmental exposures on living organisms. With these tools, scientists
can produce a more-detailed understanding of the hazards and risks of a much larger number of
chemicals. The integration of modern computing with molecular biology and chemistry is
allowing scientists to better understand a chemical's progression through the environment to the
target tissue within an organism, and ultimately to the key steps that trigger an adverse health
effect. Currently, risk estimates are most often based on gross outcomes of disease such as
occurrence of cancer, a neurological disorder, or a visible birth defect. The National Research
Council, in its 2007 report Toxicity Testing in the 21st Century: A Vision and a Strategy called
for a concerted effort to move toxicology from a primarily descriptive science to a more
predictive one by utilizing largely human based in vitro studies to understand the biological
pathways by which chemically induced diseases occur. EPA's CTRP is working to aide this
transformation by evaluating the key molecular changes occurring in the function of critical
human toxicity pathways within cells, tissues, individuals and populations. The key will be
connecting these changes quantitatively and systemically to the types of adverse health effects
that have been the traditional basis of EPA risk assessments and to use this understanding to
reduce the current uncertainties in the extrapolation of effects across dose, species and
chemicals. The ORD CTRP is composed of three main elements. The largest component is the
NCCT, which was established in 2005 to coordinate research on chemical screening and
prioritization, informatics, and systems modeling. The second element consists of related
activities in the NHEERL and NERL laboratories of ORD. The third and final component
consists of academic centers working on various aspects of computational toxicology and funded
by the EPA STAR program.
The rapid and ongoing success of the CTRP is impacting hazard and exposure identification,
helping to close data gaps, identify toxicity pathways, suggest modes of action, and make for
more efficient utilization of precious resources on the highest priority chemicals. Besides these
initial outcomes from the higher throughput approach of the CTRP, informatics and modeling
efforts will provide more in-depth and quantitative molecular understanding of how biological
systems respond to environmental chemicals. These knowledgebases and in silico tools will
reduce or quantify uncertainties relating to biological susceptibility, species differences and dose
response as part of a faster and more intelligent targeted testing paradigm in support of
quantitative risk assessments.
Previous I TOC
-------
EPA CompTox Research Program FY2009-2012
BOSC Review Draft- 24 August, 2009
B. Timeline of CTRP Development
In FY2002 Congress ordered a redirection of $4 million from available EPA funds,
".. .for the research, development and validation of non-animal alternative chemical
screening and prioritization methods, such as rapid, non-animal screens and Quantitative
Structure Activity Relationships (QSAR), for potential inclusion in EPA's current and
future relevant chemical evaluation programs."
To fulfill this directive, the EPA embarked on development of a research program that: (1) was
consistent with the Congressional mandate; (2) complemented and leveraged related on-going
Agency sponsored efforts to consider alternative test methods; (3) further advanced the research
to support the Agency's mission; and (4) would not duplicate the mission and programs in this
area conducted by other agencies (see Figure 1 for a timeline of CTRP development). Thus the
Figure 1 ORD Computational Toxicology
Research Program Development
Congressional Redirect
EDC Pilot Projects
FY02
CTRP Framework
CTISC
7 CTRP Proof of Concepts
STAR Systems Biology RFA
FY04
1st CTRP
Implementation Plan
ToxCast Design
1st and 2nd STAR Centers
sBOSC II
FY06
NCCT Staffing Complete
MAS 21st Century Vision
EPA 21st Century Strategy
Tox21 MOU
ACToR Launch
DSSTox v2
3rd STAR Center
sBOSC III
FY08
FY03
CTRP Design Team
SAB and BOSC reviews
RTP CTRP Workshop
STAR HTS RFA
FY05
NCCT Launch
DSSTox v1
ToxCast Concept
sBOSC I
FY07
1st Title 42s
Chemical Prioritization
CoP
ToxCast Launch
ToxRefDB Launch
v-Tissues Launch
Intl Science Forum
FY09
2nd CTRP
Implementation Plan
ACToR v1
ExpoCast Launch
ToxRefDB v1
ToxCast I Complete
1st ToxCast Summit
ToxCast II Launch
V-Tissues 09
4th STAR Center
sBOSC IV
CTRP was initiated to target these goals and, in the process, significantly advance toxicology and
risk assessment as currently practiced by the Agency and the broader environmental sciences
community. In FY2002-2003 pilot projects were funded to demonstrate computational
toxicology could be adapted to the study of endocrine disruptors. Early successes of these efforts
included refinement of estrogen receptor ligand binding data for development of quantitative
Previous
TOC
-------
EPA CompTox Research Program FY2009-2012 BOSC Review Draft- 24 August, 2009
structure-activity models, evaluation of EPA-developed cell lines for detecting estrogen and
androgen activities from various species and the development of an alternative test method for
evaluating effects on steroidogenesis in H295R cells (see EDSP Assay Status).
With increasing attention to and expectations for the CTRP in FY2003, ORD developed^
Framework for a Computational Toxicology Research Program, which was published in
FY2004 and provided strategic direction for the program. This document was the product of a
cross-ORD design team of scientists and was endorsed by the Science Advisory Board (SAB).
ORD hosted a workshop in Research Triangle Park in late FY2003 to introduce the CTRP
framework (Kavlock et. al., 2003), from which the three objectives for EPA computational
toxicology were translated into the three initial Long Term Goals (LTGs) for the program:
• Risk assessors use improved methods and tools to better understand and describe the
linkages of the source-to-outcome paradigm,
• EPA Program Offices use advanced hazard characterization tools to prioritize and
screen chemicals for toxicological evaluation and
• EPA assessors and regulators use new and improved methods and models based on the
latest science for enhanced dose-response assessment and quantitative risk assessment.
With issuance of the CTRP framework, ORD began the process of implementing a more
formalized program. A cross-Agency working group, the Computational Toxicology
Implementation Steering Committee (CTISC) was formed in FY2004 to oversee the selection
and funding of projects across ORD. Seven cross-ORD projects were initiated as result of CTISC
action, and these "new start" projects became a critical component of the first generation CTRP
implementation plan. Greater detail on the accomplishments of these seven projects is provided
below in Section I.F.2.
In October 2004, then EPA Science Advisor and Assistant Administrator for ORD, Dr. Paul
Oilman announced the formation of the National Center for Computational Toxicology (NCCT),
which began official functions in February 2005. The announcement states:
"The Center will advance the science needed to more quickly and efficiently evaluate the
potential risk of chemicals to human health and the environment. The Center will
coordinate and implement EPA's research on computational toxicology to provide tools
to conduct more rapid risk assessments and improve the identification of chemicals for
testing that may be of greatest risk."
NCCT quickly became the hub for ORD CTRP research. NCCT formed key partnerships with
the other Laboratories and Centers within ORD, which formed the second critical element.
Partnerships with NHEERL, NERL, NRMRL, and NCEA helped in the execution of not only the
seven cross-ORD "new start" projects awarded by the CTISC in 2004, but also in several
original NCCT led projects including Distributed Structure-Searchable Toxicity Database
(DSSTox), ToxCast™, ToxRefDB, and the virtual tissues projects looking at liver and embryo.
Greater detail on the accomplishments and future plans for these and other NCCT projects is
provided below in Section I.F.I.
Previous I TOC
-------
EPA CompTox Research Program FY2009-2012
BOSC Review Draft- 24 August, 2009
The third critical component of the CTRP is extramural partners and research, much of which is
supported by NCER through the STAR program. In FY2003 and 2004 two separate STAR
Requests for Applications (RFA) that funded projects in HTS and systems biology were issued.
In FY2006 two STAR academic centers to support the advancement of bioinformatics in
environmental health were funded, with a third center for computational toxicology funded in
FY2008. An award for a fourth center, which will focus on pathways and models of
developmental toxicity, was made in late FY2009. Additional information on the STAR centers,
their accomplishments, and future plans is provided in Section IF.3.
In ORD 's Computational Toxicology Research Program Implementation Plan for FY2006-2008,
these various research efforts supporting the three LTGs identified in the CTRP framework were
grouped into five research tracks: (A) Development of Data for Advanced Biological Models;
(B) Information Technologies Development and Application; (C) Prioritization Method
Development and Application; (D) Providing Tools and System Models for Extrapolation across
Dose, Life Stage, and Species; and (E) Advanced Computational Toxicology Approaches to
Improve Cumulative Risk Predictions. Within these five tracks, all of the NCCT, ORD
intramural and STAR funded extramural project plans and interconnections were defined in this
first generation implementation plan. The next generation of the CTRP implementation plan
carries forward many of these same research components into FY2009-2012.
As noted in the EPA Strategic Plan for Evaluating the Toxicity of Chemicals (see Section I.E.2),
many environmental statutes require that EPA consider both human and ecological health risks,
and while the initial emphasis of the CTRP has been on improvements in human health
assessments in order to provide the critical mass and resources necessary to be successful, the
program also over time must address ecological concerns as well. As we look to the future, we
see opportunities to leverage existing efforts such as the Tox21 project (see Section II.D.3) and
research on alternative species (e.g., the zebrafish projects underway at NHEERL and at the
newest STAR Center) to help transition the CTRP to a more balanced human and ecological
assessment program.
C.
Resources for the CTRP
CTRP Budget History
1. Funding and NCCT Personnel
Funding of the CTRP has been relatively stable over the past several years, and this has allowed
the program to develop consistent with the
strategic plan. In FY2009, the program
was funded at ~$ 15M and 32 FTEs. Fl§ure 2
Approximately 50% of the resources are
allocated to NCCT, 25% to the STAR program,
and the remainder to NHEERL and NERL.
The majority of the FTEs (~22) are located in
the NCCT. Figure 2 displays the history of the
budget through to the President's FY2010
request, which would provide an increase in
tu-
35-
30
25-
on
15
10
5
n.
K
c
I
c:
|
_
I
~
=
1
1
DFTEs
• $ (in Ms)
Previous
TOC
-------
EPA CompTox Research Program FY2009-2012
BOSC Review Draft- 24 August, 2009
funding, initially to support Phase II of the ToxCast™, but then broader aspects of the program
in the out years.
The NCCT is organized into three primary functional groups: Chemical Prioritization, Systems
Modeling, and Informatics, with a small group of administrative support personnel (Figure 3). In
addition to the permanent federal staff of 23, there are a number of postdoctoral, predoctoral,
student contractors, and Senior Environmental Enrollees (SEE) (A Grantee Organization) who
support various aspects of the program. In 2006 and 2007 NCCT successfully recruited three
senior-level Title 42 scientists in bioinformatics and systems biology. These hires have proven
critical for NCCT to establish core research projects in predictive and systems toxicology and the
necessary informatics and computational infrastructure. More recently, in 2008 several junior
research positions were filled from scientists coming through the NCCT postdoctoral program.
Full CVs of the staff are provided at the NCCT website. The NCCT is currently recruiting a
public affairs specialist to improve both internal and external communications.
Figure 3 - NCCT Organizational Chart
ORD NCCT Organizational Chart
EPA Environmental Protection Agency
ORD Office of Research and Development
NCCT
National Center for
Computational
Toxicology
Director
RobertJCayjock
Deputy Director (on detail)
Jerry Hlancato
Administration
Management Analyst
KarenjDean
Program Analyst
Sandra Roberts
DoroUffi Goodson
SEE Grantee
Public Affairs Specialist
TBD
Chemical Prioritization
David Dt», ;/,;-Jrifl Dspu-j Director)
Dai\is Roffoff - UNC Graduate Student
Kefth Houi*
Stephen Little
James Rabinowitz
Systems Modeling
Elaine Cohen JHuisal
Toni_KnjJdsen
N
-------
EPA CompTox Research Program FY2009-2012 BOSC Review Draft- 24 August, 2009
D. BOSC Reviews of the CTRP
To help guide the CTRP, ORD established a standing subcommittee of the ORD BOSC to
provide review and advice to NCCT and the CTRP. To view all prior BOSC reviews and ORD
responses please visit the attached link (NCCT BOSC Reviews).
The BOSC first met in April 2005 to review the organization of NCCT, initial plans for
implementation, and progress of the early CTRP work. The panel commented very favorably on
the Center's early progress and the means outlined to achieve its goals. The composition of staff,
plans for future hiring, establishment of working partnerships, and the Center's strategic plan
were especially highlighted. Several recommendations were made for consideration. Two main
recommendations were to develop a formal implementation plan for the future, and secondly to
develop Communities of Practices (CoPs) within the EPA, which could serve as a networking
function for interested scientists. ORD's Computational Toxicology Research Program
Implementation Plan (FY 2006 - 2008) was completed in April of 2006, in time for the next
BOSC review. Two CoPs have been organized and are discussed later in this document. A few
other minor suggestions were also made, which were addressed in the formal ORD response to
the review.
On June 19-20, 2006, a second BOSC site visit was held to assess and evaluate NCCT progress
in executing the first implementation plan and incorporating prior BOSC recommendations. In
this review it was noted, in the 16 months of its existence, "NCCT has made substantial progress
in (1) establishing goals and priorities; (2) making connections within and outside EPA to
leverage staff's considerable modeling experience; (3) expanding its capabilities in informatics;
and (4) significant contributions to research and decision making throughout the Agency." The
BOSC also noted that "many of the recommendations made by the BOSC during its first review
have been acted on by NCCT." This review occurred just before NCCT hired the first two
scientists, under ORD's new Title 42 authority, which brought needed experience in informatics
and systems biology to address several of the key recommendations of the BOSC from this
second review.
The third review by the BOSC occurred on December 17-18, 2007. In this review, the BOSC
noted that NCCT continued to make substantial progress in setting priorities and goals, and
specifically acknowledged the "increased capabilities in bioinformatics through the funding of
two STAR centers and in informatics and systems biology through staff hires; expansion of its
technical approaches to even more programs within the Agency; and formation of an extensive
collaboration with the NIEHS and the NHGRI for its ToxCast™project." They went on to
recommend that: (1) client offices participate in future reviews to ensure all parties understand
how NCCT's efforts can address the most relevant needs of the Agency; (2) interactions with
risk assessors in the Agency be enhanced, particularly related to how ToxCast™ might be
utilized by them; (3) complementary efforts in exposure prioritization be undertaken; (4) the
directions and milestones be detailed and applications of the virtual tissues to risk assessment be
clarified; and (5) finally, the committee encouraged a more precise definition of the database for
compilation and rigorous quantitative analysis of the ToxCast™ data. Overall, the review
considered the goals of the Center to be well described, very ambitious and innovative, as well as
important for the future of research at EPA.
Previous I TOC
-------
EPA CompTox Research Program FY2009-2012 BOSC Review Draft- 24 August, 2009
The next BOSC review is scheduled for September 29-30, 2009, and will focus on the products
from the first CTRP implementation plan and the future directions outlined in this second
generation plan.
E. NRC Report and EPA's Strategic Plan for 21st Century Toxicology
Two significant activities have increased the visibility and importance of EPA's computational
toxicology research efforts over the past two years. The first is a report, commissioned by the
EPA, and from the NRC presented a vision of how toxicological evaluations should be
conducted in the future. The second activity, motivated by the NRC report, was development of
an EPA strategic plan to transform how the Agency addresses chemical toxicity.
1. NRC Report on Toxicity Testing in the 21st Century
The NRC released a report titled Toxicity Testing in the Twenty First Century: A Vision and
Strategy, in 2007, which outlines a long-term vision for developing novel approaches to
chemical toxicity characterization and prediction. This vision addresses several concerns about
the current "gold standard" methods for toxicity characterization which rely heavily on extensive
animal testing. These concerns are the desire to reduce the number of animals used in testing, to
reduce the overall cost and time required to characterize each chemical, and to increase the level
of mechanistic understanding of chemical toxicity.
The NRC report outlines an approach for toxicity determination in which each chemical would
be first characterized for a number of properties related to environmental distribution, exposure
risk, physico-chemical properties, and metabolism. These activities fall under the heading
"chemical characterization," this (with the exception of metabolism) excludes bioactivity in the
target organism. The second stage is "toxicity pathway characterization," in which a series of
cell-based and non-cell-based (in vitro) tests would be used to indicate which (if any) "toxicity
pathways" are activated by the test chemical. A major challenge posed by this approach is that
few such toxicity pathways are currently understood, and assays to probe these pathways are
therefore generally lacking. Next, for a subset of chemicals, "targeted testing" would be carried
out to refine our understanding of the effects of triggering specific toxicity pathways. Targeted
testing might involve additional in vitro assays or a limited amount of in vivo animal testing. A
final phase is "dose response and extrapolation modeling," that would use new and existing data
and models to perform low dose extrapolation, toxicokinetics, and exposure estimation. All
phases would have significant computer modeling components. The end result of these studies
would be a determination of the potential toxic effects (including mode and mechanism of
action) of a compound, as well as estimates of dose response behavior.
There is a relatively simple argument as to why an in v/Yro-based approach should be able to
predict whole animal or human toxicity. The effect of a chemical is ultimately due to direct or
indirect molecular interactions of the chemical with one or more cellular components. These
interactions can be receptor or enzyme binding, disruption of a lipid membrane, localized
production of free radicals, or non-specific dephosphorylation. Nonetheless, if two chemicals
have the same biological interactions and have the same distribution and kinetics within an
organism, then the two chemicals should present the same bioactivity profiles and potential toxic
Previous I TOC
-------
EPA CompTox Research Program FY2009-2012 BOSC Review Draft- 24 August, 2009
effects. This concept highlights a major benefit of the in vitro, mechanism-based approach; it
provides a way to extrapolate from one chemical to the next based on a set of relatively
inexpensive and quick biochemical or cellular assays. Achieving this vision however, will take
many years due to a number of circumstances that are described in the NRC report. It is
noteworthy that the NRC report did not downplay how difficult the task of developing this new
approach to toxicity testing would be, proposing a timeline of 20 years with annual investment
on the order of $100M to fully achieve this vision. However, for initial chemical screening and
prioritization a generalized set of assays for key biological targets and pathways can be
successfully implemented in a much shorter and less costly fashion.
2. EPA Strategic Plan for Evaluating the Toxicity of Chemicals
In response to the release of the NRC report, EPA established an intra-agency workgroup, the
Future of Toxicity Testing Workgroup (FTTW), under the auspices of the Science Policy
Council. The FTTW includes representatives from across the Agency, including the Regions and
all major Program Offices. It produced The U.S. Environmental Protection Agency's Strategic
Plan for Evaluating the Toxicity of Chemicals, which serves as a blueprint to ensure a leadership
role for EPA in pursuing the directions and recommendations presented in the 2007 NRC report.
The strategy presents the Agency's vision of how to incorporate a new scientific paradigm and
new tools into toxicity testing and risk assessment practices with ever-decreasing reliance on
traditional approaches. The overall goal of the strategy is to provide the tools and approaches
necessary to move from a near exclusive use of animal tests for predicting human health effects
to a process that relies more heavily on in vitro assays, especially those using human cell lines.
The program envisioned in this strategy builds upon the traditionally major components of the
risk assessment process (i.e., hazard identification, dose response, exposure assessment, and risk
characterization) by overlaying a toxicity pathways approach on the source—>• fate &
transport—>exposure—>-outcome continuum. Specific components include three integrative and
interactive focal areas:
• Toxicity Pathway-Based Chemical Screening and Prioritization. This research will
focus on (1) identifying toxicity pathways and deploying in vitro assays to characterize
the ability of chemicals to perturb those pathways, and (2) further development and
implementation of the ToxCast™ concept to establish the predictive relationship of the
new assays for identifying adverse outcomes in humans or ecological populations.
• Toxicity Pathway-Based Risk Assessment. This element will focus on reducing key
risk assessment uncertainties currently associated with the extrapolation of data from
animal studies to humans, from high doses to relevant human exposures, and to different
population susceptibilities (e.g., children). The program will achieve these ends by (1)
developing knowledgebases of toxicity pathways, toxicological responses, and
information on biological networks; (2) constructing dynamic computational models of
tissue biology that link diverse data together to understand the progression of events from
exposure to effect; and (3) demonstrating that this new vision of toxicity testing
adequately predicts human risk using case studies.
Previous I TOC
-------
EPA CompTox Research Program FY2009-2012 BOSC Review Draft- 24 August, 2009
• Institutional Transition. Implementing major changes in toxicity testing of
environmental chemicals and incorporating new types of toxicity data into risk
assessment will require significant institutional changes in terms of (1) how EPA
transitions to the use of new types of data and models; (2) how EPA deploys resources
necessary to implement the new toxicity testing paradigm, such as hiring of scientists
with particular scientific expertise and training of existing scientific staff; and (3) how
EPA educates stakeholders and the public.
While the workgroup identified a range of partners in this effort and some planning on the
relative role of these partners has been developed, the specific areas of work to be conducted and
funded by EPA versus these other partners needs to be further assessed. Decisions on these
relative roles will have a significant impact on EPA resources required to implement the vision.
Regardless, the CTRP will play a central role in all three goals of strategy - from identifying and
conducting high throughput screening on chemical libraries of interest, to developing systems
levels model of biology for application in risk assessment, to training program offices in the
understanding and use of the new technologies.
F. Significant Accomplishments of the CTRP: FY2006-2008
1. Accomplishments of the NCCT
With the CTRP being in existence for nearly six years, and the NCCT for more than four of
those six, a number of projects contained within the CTRP are now delivering important
accomplishments to regulatory offices within the Agency, toxicologists in the international
community, and the regulated chemical industry. Details of these accomplishments are excerpted
below, with greater details available in the descriptions of the ten NCCT-led projects provided in
Appendix IV.A., with additional information on the annual milestones and projected impacts of
the projects provided in a summary format in Appendix IV.C.
A number of projects on informatics, chemical prioritization and systems level models of biology
are beginning to provide the foundation for high throughput, decision making tools that EPA
program offices can apply to chemical hazard screening and risk assessment. These include:
• ToxRefDB: A relational, electronic toxicity reference database (ToxRefDB)
developed in partnership with the Office of Pesticide Programs that contains
results of over 30 years and $2B worth of rat and mouse chronic, rat
multigenerational, and rat and rabbit developmental studies for over 400
chemicals (Martin et al 2009a ; Martin etal 2009b: Knudsen etal 2009}. This
relational database is allowing the Agency, for the first time, to readily discern
patterns of toxicity in the assays and to assess the value of the design of the assays
in assessing toxicity. It will also be invaluable in the interpretation of the HTS
data derived in the ToxCast™ Program. This database will be expanded to
include Developmental Neurotoxicity Assays, and potentially results from the
Endocrine Disrupter Screening Program, thus affording a one stop shop for
animal bioassay data.
Previous I TOC
-------
EPA CompTox Research Program FY2009-2012 BOSC Review Draft- 24 August, 2009
• ACToR: The Aggregated Computational Toxicology Resource (ACToR)
provides an Internet based portal of information on chemical structure, bioassay
and toxicology data for environmental chemicals from 200 sources of public data
for over 500,000 chemicals, and provides a central integrated public resource for
all DSSTox, ToxCast™ and ToxRefDB data (Judson. etal2009\ ACToR was
released to the public in the second quarter of FY2009, and will undergo periodic
updates over the next several years as we add data and functionalities, and
respond to user feedback.
• DSSTox: The Distributed Structure-Searchable Toxicity Database Network
(DSSTox) was updated with several high-interest chemical inventories, including
ToxCast1 Phase I chemicals, EPA High Production Volume Chemicals, and the
2 major public genomics inventories (GEO and ArrayExpress), an on-line
structure-browser, and linkages and coordination with internal and outside
resources such as ACToR and PubChem (Williams, et al 2009; Williams, etal
2009}. DSSTox, in coordination with ACToR, is also responsible for all chemical
information registration and review for ToxCast™ and Tox21 (see the next page
for a more detailed explanation of Tox21) projects, and is the source of high
quality chemical structures for the OECD QSAR Toolbox.
• ToxCast™: ToxCast™ is a major effort to evaluate the comprehensive use of
HTS assays to provide biological fingerprints of activity that can be used to
predict adverse outcomes in rodents and in humans (Dixet al 2007). With the
initiation of 9 contracts in FY2007, the CTRP began to generate a myriad of
molecular in-vitro based data for the ToxCast™ program. Data collection for over
300 chemicals from over 500 assays in Phase I of this chemical prioritization
program was completed in quarter 2 of FY2009. An internal EPA workshop
regarding ToxCast™ was held in March 2008, and a public ToxCast Summit
workshop was held in May 2009 with over 30 national and international Material
Transfer Agreements (MTA) analysis partners participating. Analyses are
focusing along three dimensions: (1) comparing bioactivity profiles across
chemical classes; (2) correlating specific assays or pathways with toxic
phenotypes; and (3) correlating phenotypic syndromes with bioactivity profiles. It
is expected that analysis of the data will continue over several years, both by EPA
and external groups as the compendium of data represents a truly unique and
innovative resource. A significant number of additional partners were brought
into the program in FY2008 and FY2009 via MTAs. Additional contracts awards
are anticipated to augment the breadth of biological pathways contained with
Phase II, which will launch in late FY2009. A critical feature of Phase II is the
plan to include drugs that have failed in human clinical trials, thus providing
human toxicity data upon which to benchmark ToxCast™ bioactivity profiles.
One major pharmaceutical company, Pfizer, has already agreed to collaborate in
this regard, and Health and Environmental Sciences Institute has adopted the
concept as their Emerging Issues Proposal for 2009, promoting the opportunity
for a much wider group of pharmaceutical companies to contribute failed drugs
and clinical data. Thus Phase II of ToxCast™ will be generating HTS data on 700
10
Previous I TOC
-------
EPA CompTox Research Program FY2009-2012 BOSC Review Draft- 24 August, 2009
additional chemicals, some with animal toxicity information, or clinical and
additional data on human disease, susceptibility, and variability that will
contribute to the goal expanding, verifying and translating in vitro bioactivity into
predictions of potential toxicity.
• Tox21: To garner complementary expertise across the federal government to
transform the field of toxicology to a more predictive science, EPA signed a
Memorandum of Understanding (MOU) with the National Toxicology Program
(NTP)/National Institute of Environmental Health Sciences (NIEHS) and the
National Institutes of Health Chemical Genomics Center (NCGC)/ National
Institutes of Health Chemical Genome Research Institute (NHGRI) in February
2008 (Collins et al, 2008}. This "Tox21" consortium now has four active working
groups identifying chemicals, assays, informatic analyses and targeted testing as
plans proceed to have nearly 10,000 chemicals under study at the NCGC by the
fourth quarter of FY2009 (Kavlock et al, 2009}. Supported by Interagency
Agreements (lAGs) between the NTP and EPA, this effort will conduct 50 or
more HTS assays on this enlarged chemical library every year for the next several
years.
• Cumulative Risk: Tools of computational toxicology have been used in
integrating data for the cumulative risk assessments of cholinesterase inhibiting
pesticides (organophosphates and carbamates) and in developing a model for the
effects of iodomethane, a fumigant, on development. These methods have been
used to set safe exposure limits for these chemicals in the field and home.
• Communities of Practice (CoP): Though not projects or tools in the typical
sense, NCCT has formed several CoPs promoting the utilization of CTRP
research in specific areas of computational toxicology. Each CoP has a charter
and an open membership policy, and is co-chaired by a member of the NCCT.
The CoPs operate via activities such as teleconferences, face to face meetings,
team rooms, and workshops. By bringing together members from different parts
of EPA, ORD and the outside scientific, regulatory and regulated community,
CoPs help to promote the adoption of common practices and ontologies, guide
development of common databases and software usage, aid in construction of
training materials, provide recommendations on efficiencies of relevant
operations, and act as a public outreach mechanism for ORD activities. To date,
two CoPs have been established on Chemical Prioritization and Exposure
Science. Both of these are large, active groups that meet monthly and have
brought a wealth of ideas, interest and international collaboration to various
CTRP projects.
• International Science Forum on Computational Toxicology. In 2007, the
CTRP hosted EPA's annual Science Forum. This was the first EPA Science
Forum held outside of Washington, B.C. and featured an international overview
of the state of the science on computational toxicology (Kavlock et al 2008}.
More than 300 people attended this Forum on advances in computer sciences,
11
Previous I TOC
-------
EPA CompTox Research Program FY2009-2012 BOSC Review Draft- 24 August, 2009
molecular biology and chemistry, and systems models, that can be used to
increase the efficiency and the effectiveness by which the hazards and risks of
environmental chemicals are determined. The Forum surveyed the state of the art
in many areas of computational toxicology and identified key areas important to
move the field forward: proof-of-concept studies demonstrating the additional
predictive power gained; more researchers comfortable generating and working
with high throughput data and using it in computational modeling; and regulatory
authorities willing to embrace new approaches as they gain scientific acceptance.
Ideas from this Forum and efforts bridging from it were helpful in EPA's
development of The U.S. Environmental Protection Agency's Strategic Plan for
Evaluating the Toxicity of Chemicals.
2. Accomplishments by CTRP ORD Intramural Partners
The seven cross-ORD "new start" research projects initiated by the CTISC in FY2004 advanced
the field of computational toxicology and were important components in the first generation
CTRP implementation plan and establishment of the NCCT. Details on the organization, goals,
and cross-ORD participants for these projects were provided in the FY2006-2008 CTRP
Implementation Plan, and a comprehensive listing of publications from these projects is included
in Appendix C. Descriptions of major accomplishments are provided below:
Linkage of Exposure and Effects Using Genomics, Proteomics, and Metabolomics in
Small Fish Models: This project used a combination of whole organism endpoints,
genomic, proteomic, and metabonomic approaches, and computational modeling to (a)
identify new molecular biomarkers of exposure to endocrine disrupting compounds
(EDCs) representing several modes/mechanisms of action (MOA) and (b) link those
biomarkers to effects that are relevant for both diagnostic and predictive risk assessments
using small fish models. Data from the project have provided the basis for several
important predictive modeling efforts. For example, the development of a graphical
systems model focused on defining the HPG axis of small fish, which enables
consideration of the interactive nature of a perturbed system at multiple levels of
biological organization, ranging from changes in gene, protein and metabolite expression
profiles to effects in cells/tissues that directly influence reproductive success. A second
modeling effort involves development of a steady-state model for ovarian tissue to
predict synthesis and release of testosterone and estradiol. Results from the model were
successfully compared to data generated from the fathead minnow. Model-predicted
concentrations of the two steroids over time corresponded well with both baseline
(control) data, and information from experiments in which estradiol synthesis was
blocked by fadrozole. Modeling also has focused on the consequences of perturbations in
the HPG axis relative to effects in individuals and populations. Here, a population model
which employs a Leslie matrix in conjunction with the logistic equation (to account for
density dependence) was used to translate laboratory toxicity information into prediction
of population trajectories. In these analyses, changes in steroid or vitellogenin
concentrations in female fish first were related to fecundity, and then using this
relationship to population status in fish exposed to EDCs which inhibit production of
vitellogenin , most notably compounds that depress steroid synthesis (e.g., fadrozole,
12
Previous I TOC
-------
EPA CompTox Research Program FY2009-2012 BOSC Review Draft- 24 August, 2009
prochloraz, trenbolone). That analysis is unique in that it focuses on biochemical
endpoints, female steroids and vitellogenin, which reflect both toxic MOA of EDCs and
have a functional relationship to reproductive success (formation of eggs). As such,
within the overall systems framework for the project, this computational model can serve
as the basis via which genomic information can be quantitatively linked to responses in
populations.
A Systems Approach to Characterizing and Predicting Thyroid Toxicity Using an
Amphibian Model: The main objective of this work was to develop a hypothalamic-
pituitary-thyroid (HPT) model which is capable of integrating data from different levels
of biological organization into a coherent system. A simulation model has been
developed to describe the thyroid axis of X. laevis tadpoles. Information pertaining to
normal baseline HPT-axis development was collected to compare to the perturbed
system. Thyroid and pituitary gland culture systems were developed to investigate the
response of these components in isolation from the HPT-axis feedback mechanisms
within the animal. Genes responsive to TSH were also measured in the thyroid gland and
pituitary in vivo and in vitro in response to chemical exposure and TSH stimulation.
Genes involved in the T4 synthesis pathway that robustly responded to TSH, such as NIS
and thyroid peroxidase (TPO), were not changed when challenged with chemical alone.
Other thyroid-specific genes such as thyroid transcription factor and thyroglobulin were
not TSH-responsive. Efforts were made to obtain TPO activity from^ laevis, but the
small amount of tissue contributed to our inability to develop an assay. Therefore,
porcine thyroid glands were used to isolate and measure TPO activity. Twenty-four
chemicals have been tested in the in vitro assay for their capacity to inhibit TPO activity.
Although the majority of the tested chemicals were negative, several were identified as
TPO inhibitors. These positives were tested in the thyroid gland explant culture assays
and the TH released into the culture media was measured by RIA. One of the test
chemicals, a mercaptobenzothiazol, inhibited T4 release from the thyroid glands at
concentrations that were not overtly toxic to the gland. Chemicals that test positive in the
TPO inhibition assay and the ex vivo thyroid gland TH release assay will be tested in the
abbreviated amphibian metamorphosis assay to determine activity in the HPT-axis in
vivo. This will begin to provide information on the predictive ability of the in vitro and ex
vivo assays for identifying thyroid-disruptive chemicals.
Risk Assessment of the Inflammogenic and Mutagenic Effects of Diesel Exhaust
Particulates: A Systems Biology Approach: This project utilized a systems approach
to developing and applying predictive computational models that quantitatively describe
relationships between the composition of DEP and its genotoxic and inflammogenic
potencies. A significant accomplishment is derived from the development of a prototype
ESP sampler, which has resulted in a significant improvement in DEP collection yield
relative to conventional filter methods for particle capture that were in use previously.
These analyses have produced an unprecedented physicochemical characterization of the
DEP, including XRF analysis of metal content and GC/MS analysis of organic species, as
well as determinations of OC/EC, particle size and aerodynamic characteristics, etc. By
design, this Phase generated inflammogenicand mutagenic DEPs of varying composition
to address the central hypothesis regarding the inflammogenic and mutagenic potency of
13
Previous I TOC
-------
EPA CompTox Research Program FY2009-2012 BOSC Review Draft- 24 August, 2009
DEP. An extensive database has been generated and is currently being analyzed for
publication and use in model development.
The Mechanistic Indicators of Childhood Asthma (MICA) study: A Systems Biology
Approach to Improve the Predictive Value of Biomarkers for Assessing Exposure,
Effects, and Susceptibility in the Detroit Children's Health Study: This
computational toxicology effort integrated rodent and human research across the source-
to-outcome continuum to link gene expression with clinical outcomes and biomarkers of
exposure, early effect, and susceptibility for two broad classes of chemicals: polycyclic
aromatic hydrocarbons (PAHs) and metals. In collaboration with Michigan State
investigators, exposures of rodents to concentrated air particulates was completed using
state of the art mobile air exposure chambers. Genomic analysis of rodent blood and lung
tissues highlighted tissue-specific patterns of gene expression related to airborne
exposures. For the children's study (whose protocols were approved by three separate
Internal Review Boards), a 20-page respiratory health questionnaire was mailed to the
parents of 6,883 children aged 7 to 12 years recruited through the Henry Ford Health
System. An indoor/outdoor MICA-Air component was added using an innovative
participant-based air sampling approach that included measurements of nitrogen dioxide
and selected volatile organic compounds. A subset (205) of these children volunteered for
clinical examinations (including measurements of pulmonary function and exhaled nitric
oxide) collection of blood (for both genetic and gene expression analysis), and nail and
urine samples. Our analysis strategy represents a true "systems" approach in that each
type of data is examined in the context of the broader MICA data set. This systems
approach affords more robust conclusions, because the predictive value of biomarkers
from particular data slices can be assessed for biological and statistical validity against
the diverse set of supporting data.
An Approach to Using Toxicogenomic Data in Risk Assessment: Dibutyl Phthalate
(DBF) Case Study: To address how genomic data may be used most effectively in risk
assessment, this project was initiated with the goals of: 1) developing an approach to
using toxicogenomic data in risk assessment and 2) testing this approach in a case study.
Recognizing that genomic data type (e.g., species, organ, design, method) varies, the
approach included formulating questions that the toxicogenomic data may inform. Since
microarray data is often informative of the mechanism or mode of action (MOA) of a
chemical, the approach included an assessment of the toxicogenomic dataset in
conjunction with the toxicity dataset in order to relate the affected endpoints (identified in
the toxicity dataset evaluation) to the pathways (identified in the toxicogenomic dataset
evaluation) as a method for informing the mechanism of action. Dibutyl phthalate (DBP)
was selected for the case study, focusing on the male reproductive outcomes, since it has
a relatively large and consistent genomic dataset, phenotypic anchoring of certain gene
expression data for these outcomes, and an ongoing Integrated Risk Information System
(IRIS) assessment. The case study team concluded that the "genomic dataset" should
include all gene expression data (single gene, global gene expression, protein, RNA) in
the evaluation as these data taken together provide a stronger basis for reproducibility of
the global gene expression study findings. This evaluation found that the gene level
findings from the DBP genomic studies (i.e., microarray, RT-PCR, and protein
14
Previous I TOC
-------
EPA CompTox Research Program FY2009-2012 BOSC Review Draft- 24 August, 2009
expression) were highly consistent in both the identification of differentially expressed
genes (DEGs) and their direction of effect. This project also identified research needs for
toxicity and toxicogenomic studies for use in risk assessment. These include: 1) Parallel
study design characteristics with toxicogenomic studies (i.e., dose, timing of exposure,
organ/tissue evaluated) to obtain comparable toxicity and toxicogenomic studies to aid
our understanding of the linkage between gene expression changes and phenotypic
outcomes; 2) Exposure time-course microarray data to develop a regulatory network
model; 3) Generate TK data in relevant study (time, dose, tissue), and obtain relevant
internal dose measure to derive best internal dose metric; 4) Multiple doses in microarray
studies in parallel with phenotypic anchoring.
Development of Microbial Metagenomic Markers for Environmental Monitoring
and Risk Assessment: This project focused on the development of nonculture-based
genomic methods for environmental monitoring and risk assessment. The research
focused on the use of a microbial community genome (metagenome) approach to identify
novel nucleic acid sequence markers for fecal contamination and source identification.
The basic experimental design consisted of challenging genomic microbial community
DNA from different fecal samples in genome subtraction studies to enrich for host
specific microbial genes. Sequencing complete fecal metagenomes generated redundant
information, making difficult the selection of host-specific genes. To address this
limitation, a novel approach called genome fragment enrichment was developed to select
for DNA fragments present in a specific fecal microbial community and absent in other
fecal communities. A patent disclosure was filed on this specific method. In each case,
hundreds of enriched DNA fragments were sequenced and assigned a putative functional
role. While several assays were developed, these have to be further evaluated,
particularly to determine the geographic stability of these methods, not only in the U.S.,
but also in other parts of the world. To this end, we are working with researchers from
different regions in the U.S. and from countries such as Canada, Brazil, Austria,
Singapore, and Spain. Results of this research will overcome the current limitation of
assessing the microbial water quality by measuring bacterial densities, which are both
time consuming and do not provide information about the sources impacting a watershed.
Such information is necessary to implement adequate pollution control and remediation
practices.
Simulating Metabolism of Xenobiotic Chemicals as a Predictor of Toxicity: The
MetaPath research project is a collaboration with OPP scientists developing a capability
for forecasting the metabolism of xenobiotic chemicals of EPA interest, to predict the
most likely formed chemical metabolites, and to interface that information with toxic
effect models allowing prediction of parent chemical toxic potential and the identity of
chemical metabolites of equal or greater toxicity than the parent chemical. Key
milestones include developing and expanding the in vivo and in vitro liver metabolism
database especially for chemicals and transformation reactions underrepresented in the
current database; finalizing development of the searchable metabolism database and
continuing to populate existing databases with additional metabolism data; enhancing the
performance of the existing metabolic simulator by incorporating reliable metabolism
data and expansion of relevant transformation reactions; and conducting in vitro
15
Previous I TOC
-------
EPA CompTox Research Program FY2009-2012 BOSC Review Draft- 24 August, 2009
experimentation to verify maps and metabolites forecasted by the metabolic simulator
and evidence for enhanced estrogenicity.
3. Accomplishments by STAR Grantees in the CTRP
The STAR grants and centers have been a critical component of the CTRP from its inception. A
brief summary of the accomplishments of the three existing STAR centers for this program are
provided below. More information and fuller descriptions of these centers are available in
Appendix B.
NCER funded two STAR environmental bioinformatics centers as part of the CTRP in FY2006.
The Research Center for Environmental Bioinformatics and Computational Toxicology at the
University of Medicine & Dentistry of New Jersey (UMDNJ), Piscataway, NJ, and The Carolina
Environmental Bioinformatics Research Center at the University of North Carolina (UNC),
Chapel Hill, are operating as cooperative agreements and helping to facilitate the application of
bioinformatics tools and approaches to environmental health issues supported by the CTRP.
To date, the UMDNJ center has made progress expanding the framework of the FDA
Array Track to ebTrack, an integrated bioinformatics system for environmental research and
analysis enabling the integration, curation, management, first-level analysis and interpretation of
environmental and toxicological data from diverse sources. Other major accomplishments
include the enhancement of Shape Signatures QSAR technology for chemical hazard
identification; metabolic engineering tools for identifying important pathways within the overall
hepatocyte metabolism; and computational procedures for quantifying the structure of molecular
bionetworks via the S-space Network Identification Protocol (SNIP) and the Closed-Loop
Identification Protocol (CLIP).
Major accomplishments of The Carolina Center for Environmental Bioinformatics include the
development and refinement of a mouse model of variation in genetic susceptibility relevant to
human populations, pathway modeling in genomic analysis, and new methods in quantitative
structure activity (QSAR) modeling relevant to toxicity. This work complements other work in
the CTRP, utilizing the unique strengths of the STAR Center in genetics, toxicology, and
statistical modeling. An early outcome of this work included dissection of the genetic regulation
of liver gene expression. In addition, the Carolina Environmental Bioinformatics Center has
refined expression quantitative trait locus (eQTL) analysis procedures. These methods serve the
larger goal of elucidating the underlying mechanisms of toxicity. The Center has also developed
high-quality methods for testing biological pathway involvement in toxicogenomics studies, and
a novel hierarchical two-step approach to model chemical structure for in vitro/in vivo toxicity
data.
In FY2008 NCER funded a third center, through a cooperative agreement, The Carolina Center
for Computational Toxicology at the UNC at Chapel Hill. This Center is applying high-
performance computing techniques and resources to in silico multi-scale modeling applications
at the cellular, organ, and system-wide level. In its first year, the center has begun to implement
and design advanced mathematical approaches modeling biological systems, and biological-
chemical interactions represented in the ToxCast™, as well as other datasets.
16
Previous I TOC
-------
EPA CompTox Research Program FY2009-2012 BOSC Review Draft- 24 August, 2009
Another high-priority for EPA is to understand the molecular and cellular processes that, when
perturbed, result in developmental toxicity. With a project start date of November 2009 NCER
responded to this need by funding the Texas-Indiana Virtual STAR Center; Data-Generating in
vitro and in silico Models of Developmental Toxicity in Embryonic Stem Cells and Zebrafish at
the University of Houston, Texas A&M Institute for Genomic Medicine, and Indiana University.
This center that will bridge the interface of in vitro data generation and in silico model
development to answer critical biological questions related to toxicity pathways important to
human development. This research should result in an improved predictive capacity for
estimating outcomes or risk associated with developmental exposure to environmental toxicants.
G. Summary on Retrospective of the CTRP and NCCT
The first five years of the CTRP and NCCT have seen a great deal of progress and
accomplishment. These accomplishments have come from across ORD, from the STAR
grantees, and increasingly from the NCCT as the Center has become fully staffed and its projects
mature. As the plan from the first implementation comes to an end, the CTRP is poised to carry
out the vision of the NRC report on Toxicity Testing in the Twenty First Century: A Vision and
Strategy, and The U.S. Environmental Protection Agency's Strategic Plan for Evaluating the
Toxicity of Chemicals. This CTRP implementation plan for FY2009-2012 will go on to explain
revisions to the program and specific projects that will be management priorities into the future.
17
Previous I TOC
-------
EPA CompTox Research Program FY2009-2012 BOSC Review Draft- 24 August, 2009
II. REVISION OF THE CTRP FOR FY2009-2012
A. Maturation of the Program
The accomplishments of the CTRP noted above have enabled the program to evolve beyond the
early focus on hazard identification and chemical prioritization into broader areas of risk
assessment. In doing so, a number of research activities are becoming increasingly intertwined.
For example, results of the ToxCast™ program are now providing data for models being
developed in the virtual tissue programs, and in turn, the virtual tissue programs are beginning to
guide targeted testing needs by other parts of the program. As a result, there has become a
blending of activities across the three long term goals (LTG) contained in the first
implementation plan. While the LTGs served us well in the initial years of the program, we have
reduced them in the current plan to a single goal, namely—Providing High Throughput Decision
Support Tools for Screening and Assessing Chemical Exposure, Hazard and Risk. Following
The U.S. Environmental Protection Agency's Strategic Plan for Evaluating the Toxicity of
Chemicals, we will see increased emphasis on the use of new data types being generated in the
CTRP in quantitative risk assessment, and greater involvement in aspects of high throughput
exposure models and in analysis of results coming out of high throughput bioactivity profiling.
As noted in other places, discussions are already well underway between NCCT and NHEERL,
NERL and NCEA on the shape of such an expansion. This expansion will incorporate progress
from ToxCast™ and ExpoCast™, as well as the v-Liver™ and v-Embryo™ projects, into
informatics tools and databases that can be used for both research and regulatory applications.
For example, in FY2011 we expect ToxCast™ will be winding down Phase II of its three part
development program, providing in vitro and in vivo toxicity data on a total of 1,000 compounds
across hundreds of molecular targets and biological pathways. Over this same time period, the
Tox21 consortium will be building a chemical screening library of up to 10,000 chemicals, and
conducting approximately one assay per week on this library and providing additional in vitro
bioactivity data. Within the ACToR database and website, and in the analysis application
ToxMiner, all of this bioactivity data will be merged with chemical information and the exposure
data being curated from the ExpoCast™ project. At this stage in FY2011, steps will be taken to
provide EPA's program offices decision analysis tools that incorporate hazard and exposure data
for prioritization of further chemical testing. Prior to that time the CTRP will be hosting periodic
training courses on the new science and tools for program office and regional personnel. We
envision this decision analysis tool being part of the ToxMiner application. In addition to
statistically based predictive toxicology tools, ToxMiner will be able to reference the mapping of
ToxCast™ assays to biological pathways curated from sources such as Kyoto Encyclopedia of
Genes and Genomes (KEGG) into a computable format for use in ToxMiner.
Depending on the precise pace of progress in ToxCast™ and ExpoCast™, and the development
of screening and prioritization tools, additional resources could be freed up to support other
components of the CTRP in FY2011 and FY2012. However, chemical prioritization will remain
a priority area until it reaches the stated objectives. A natural extension of the chemical
prioritization projects is using this data for systems modeling in the v-Liver™ and v-Embryo ™
projects. It is these systems models that will help fully utilize the toxicity and exposure pathway
information coming out of ToxCast™ and ExpoCast™ into next generation, higher throughput
risk assessments. In conjunction with OPPTS, we have also begun an examination of the
18
Previous I TOC
-------
EPA CompTox Research Program FY2009-2012 BOSC Review Draft- 24 August, 2009
feasibility of using HTS tools to evaluate the potential hazards of nanomaterials. This effort will
start with carbon nanotubes and nanomaterials being used in the OECD Working Party on
Manufactured Nanomaterials and include materials submitted to the EPA via the
Premanufacturing Notification program of the Toxics Substances and Control Act. We expect
the application of bioactivity profiling of nanomaterials to become an increasingly important
component of the CTRP as we gain experience with their handling and screening results.
Another major consideration in the evolution of these programs, screening high priority
chemicals in toxicity pathways for testing prioritization, is to appropriately tailor them for
reducing to practice the research program for regulatory use. At current funding levels, the
burden of putting into practice the screening and prioritization of 5,000-10,000 chemicals using a
comprehensive battery of HTS assays could not be borne by the ORD CTRP alone.
Relative to the STAR program, the number of funded Centers will reach five (two on
environmental bioinformatics, one on computational toxicology, and one or two on virtual
tissues) by FY2010 and consideration will need to be given to the second generation Centers as
the first ones to be awarded begin to reach conclusion of their funding cycle.
For historical purposes, the following is provided as transition information relative to the original
three LTG structure, with notation of the degree of emphasis the research activities that will be
carried over into the new implementation plan. Much of the increased activity is predicated on
continual and increasing support of the CTRP program, such as is being witnessed in the FY2010
budget process. The following section describes research areas of increased and decreased effort
relative to the FY2006-FY2008 three LTGs.
Long-Term Goal 1 - EPA risk assessors use improved methods and tools to better
understand and describe linkages across the source to outcome paradigm. Work was
directed towards computational models and modeling systems that represented
comprehensive descriptions of the underlying biology of adverse impacts caused by
exposure to environmental agents. The whole-systems biology modeling approach was to
develop a range of models, from those describing pharmacodynamic connections
between exposure and effects to those describing complex endogenous pathways and the
perturbations in such pathways resulting from environmental exposures. Also, ways to
incorporate and use "omics" information in these models was explored. Finally,
attempts were made at formulating models of common, but complex, disease processes
which are then exacerbated by exposures to exogenous substances and stressors through
the development of virtual organ models, the first being the virtual liver and second the
virtual embryo.
• Increase - efforts to develop virtual models of tissues (liver and embryo) that link
across levels of biological organization from molecular to cellular to tissue level
responses
• Increase - Coordination of efforts across NCCT, NERL, NHEERL, and NCEA to
ensure models of aspects of the source to outcome paradigm can be integrated and
scaled to meet the increasing needs of chemical evaluations.
19
Previous I TOC
-------
EPA CompTox Research Program FY2009-2012 BOSC Review Draft- 24 August, 2009
• Decrease - Research efforts related to the validation and acceptance of PBPK
models
• Decrease after FY2009 - Linkage of exposure and effects using genomics,
proteomics, metabonomics in small fish models
Long-Term Goal 2 - EPA Program Offices use advanced hazard characterization tools
to prioritize and screen chemicals for toxicological evaluation. Molecular biological
tools were employed to develop fingerprints of biological activity of chemicals of
concern to the EPA. Computational models were applied to the fingerprints to derive
associations with classical measures of toxicity derived from animal studies so that
predictive models were developed leading to more efficient testing paradigms and
reduction in uncertainties in inter-species extrapolation. Proof-of-concept demonstration
of ToxCast™, the forecasting tool, is scheduled to provide a number of EPA program
offices with an extremely useful tool to improve the efficiency and effectiveness of
hazard identification and risk assessment methodologies. There were also new and
innovative ways developed to assimilate, evaluate, and use the myriad of data assorted
with molecular and chemical information. Increasingly integrated chemical-biological
effects databases are intended to spur new capabilities for data mining and chemical
categorization in conjunction with HTS data. Development of advanced computational
chemistry method also provided in silico means to predict complex interactions of
environmental chemicals with biochemical receptors which can then lead to adverse
effects.
• Increase - Support for Phase II of ToxCast™. The additional chemicals, which
could total up to 1,000, will include a greater number of pesticides, pesticidal
inerts, antimicrobials, industrial chemicals, water contaminants and failed drug
candidates. Phase II should be winding down sometime in FY2011, depending on
the exact pace of progress in FY 2009 and FY2010.
Increase — Develop methods to analyze data and relate the results from the
ToxCa;
levels.
ToxCast™ studies to potential for hazard and risk from realistic human exposure
• Increase - Interactions between NCCT and NHEERL on identification of
additional critical toxicity pathways to include in the chemical prioritization
program.
• Increase - In conjunction with NERL develop ExpoCast™, the exposure
component of a chemical prioritization process.
• Increase - Cross governmental collaborations to employ quantitative high
throughput screening assays to predict human toxicity by engagement with
relevant components of the NTP/NIEHS, NCGC/NHGRI and FDA in the Tox21
program.
20
Previous I TOC
-------
EPA CompTox Research Program FY2009-2012 BOSC Review Draft- 24 August, 2009
Long-Term Goal 3 - EPA risk assessors and regulators use new models based on the
latest science to reduce uncertainties in dose-response assessment, cross-species
extrapolation, and quantitative risk assessment. The intent of this goal was to develop
additional key modules for computational models of biological processes relevant to the
induction of toxicity for high priority environmental chemicals. These modules would
help assess the interaction of exposure to environmental chemicals with other processes
such as underlying disease and concomitant intake of pharmacological agents. As a
result, EPA will be less reliant on default assumptions for risk assessment and better able
to accurately characterize the true uncertainty associated with risk predictions for various
chemical classes (e.g., EDCs) under conditions more relevant to actual exposures and
lifestyles
• Decrease - Chemical specific efforts vis a vis tools that have greater generic
applicability.
• Increase - Analysis of the resulting HTS data to (1) provide mode of action
information to specific risk assessments being conducted by EPA; (2) provide
rationale for grouping of cumulative risk assessments based on toxicity pathways;
(3) design of a higher throughput risk assessment approach for chemicals based
upon exposure potential and perturbation of toxicity pathways; and (4) develop
methods for analyzing and quantifying the uncertainty in dose-response model
predictions.
These modifications to levels of effort and emphases are consonant with distillation of the three
LTGs from the FY2006-2008 implementation plan into a single goal--Providing High
Throughput Decision Support Tools for Screening and Assessing Chemical Exposure, Hazard
and Risk. This will more efficiently support the use of new data types being generated in the
CTRP in quantitative risk assessments, as well as providing high throughput hazard and exposure
data and models for screening. Since the majority of the HTS data is being generated for human
molecular targets and pathways, this will support a transition from the current dependence on
animal-based toxicology. In combination with clinical data from pharmaceutical partners,
expanding efforts bringing in human data on exposure, and multi-scale systems models, it should
be possible to improve both the pace and quality of risk assessments.
B. CTRP Integration across ORD Laboratories and Centers
Due to the relatively small size of NCCT and the ambitious nature of the CTRP mission, a key
part of the process of advancing the science involves developing partnerships that are both within
and external to ORD, so as to best leverage resources committed to the effort. Within ORD, the
majority of research in health, ecological, and risk assessment, and are contained within a variety
of Multi-Year Plans (MYPs) that incorporate efforts from multiple Laboratories and Centers and
are coordinated by National Program Directors (NPDs). In the case of the CTRP, the Director of
the NCCT also leads the ORD Computational Toxicology Program. At present, there is an
ongoing active dialogue between ORD Laboratories and Centers, and relevant NPDs, regarding
the future directions of the CTRP and other programs. To move beyond just dialogue, the NCCT
21
Previous I TOC
-------
EPA CompTox Research Program FY2009-2012
BOSC Review Draft- 24 August, 2009
has hosted or is hosting, rotational fellows from across ORD (NERL, NHEERL and NCEA to
date) that spend 4 months or more working with NCCT scientists on CTRP projects.
In response to the Administrator's priorities, a pilot effort is being considered by ORD that
addresses problems of broad national significance through highly integrated multi-disciplinary
research efforts. In the last several months, NCCT, NERL, NHEERL, NRMRL and NCEA have
been exploring opportunities for greater synergy in execution of the CTRP, and each Laboratory
and Center is actively developing implementation plans for such an effort. This integration will
include aspects of the Human Health Research Program MYP and the Safe Pesticides Safe
Products MYP. The challenge will be to ensure these plans are adequately integrated and the
Figure 4
Computational Toxicology in ORD
Virtual
Systems
Quantitative
Tools
Data
Repositories
Populations
MetaPath
QSAR
TTDB
ToxRefDB
Ambient
Monitoring
Personal
Monitoring
rTK
Biomonitoring
HTS
Assay Development
In vitro
Rodent models
Human studies
Environmental
Concentration
Environmen
Release
dividual
Exposure
Internal
Dose
Biological
outputs are suitably ambitious in nature. As EPA has just released its Strategic Plan for the
Evaluation of the Toxicity of Chemicals (U.S. EPA 2009), there is a window of opportunity to
continue this momentum within the CTRP, while also expanding efforts to include critical
aspects of exposure assessment (NERL), toxicity pathway coverage and targeted testing
(NHEERL), life cycle analysis of chemical use (NRMRL), and quantitative risk assessment
(NCEA). If successful, this integration of efforts across multiple laboratories, centers, and
research programs could meld the numerous ongoing ORD research efforts in computational
toxicology into a more functional integrated multidisciplinary research program.
22
Previous
TOC
-------
EPA CompTox Research Program FY2009-2012 BOSC Review Draft- 24 August, 2009
Figure 4 shows the stages in the source-to-outcome continuum on the horizontal axis, and
existing components of ORD research and development infrastructure needed to support risk
assessment on the vertical axis, and products/tools for decision support on the far right. At the
present, many of these efforts are not integrated, and in some cases not fully compatible and a
major emphasis in the near future will to bring more coordination and integration to these
currently diverse efforts. To overcome these challenges, integrated cross-disciplinary teams,
involving exposure scientists, effects researchers, risk assessors, risk managers and
computational modelers will need to be further developed.
Understanding of complex, interrelated environmental stressors and potential impacts on human
health has grown tremendously in recent years. Basic and clinical sciences, however, have
significantly outpaced risk assessment science. Insights into health and disease exist, but it is
unclear how to incorporate this information into risk assessment with current methodologies.
Risk assessment will have to address fundamental paradigm shift from a reliance on animal
toxicology data derived primarily from rodent bioassays. The need for this shift is made more
immediate by the challenges of applying new types of data stemming from advances in
computational toxicology and the huge volume of data that will be generated from the European
Union's Registration, Evaluation, and Authorization of Chemicals (REACH) Program. High
throughput data will need to be translated into knowledge to support science-based decisions in
risk assessment. The areas where this knowledge is expected to have an impact is in defining
toxicity pathways and informing risk assessors about interpretation of multiple modes of action
for toxicity, and providing insight into human variability in key pathways and human
susceptibility. The impacts of this information will be qualitative and quantitative. For this type
of information to be incorporated quantitatively numerous challenges exist not the least of which
is extrapolation from in vitro test systems to in vivo human health outcomes. However, the
challenges are not insurmountable if ORD, all of EPA and key extramural partners work together
in a coordinated, multidisciplinary fashion. The initial impacts of this new paradigm will
probably be seen in cases of chemicals lacking significant data sets but for which toxicity
predictions or rankings can be developed from HTS data. These results may be used to derive
estimates of relevant potency to chemicals that have much larger data bases and affect the same
toxicity pathways. In addition, challenges in extrapolating from effective concentration in vitro
will require additional considerations in the development of environmental exposure estimates.
Examples of promising CTRP efforts with antimicrobials and pesticidal inerts, EDSP
compounds, and industrial chemicals such as phthalates and perfluorinated chemicals; as well as
hepatocarcinogens and teratogens are all underway. Results from these studies have been
presented, and are being published in the peer reviewed literature, EPA databases, and decision
support software tools.
The CTRP will continue to provide critical research components, work on integrating these
efforts across ORD, and facilitate the institutional transition necessary to see these tools
integrated by EPA programs and regions. This includes quantitative and mechanistic
experimental data (e.g., ToxCast™) that is useful for chemical prioritization, but also supports
systems models that could be useful for quantitative risk assessments. Data generated from
experimental systems can be used to define toxicity pathways and link them to adverse events
via the mode-of-action framework. This definition of adverse, versus As toxicity pathways and
key events are defined, development of high throughput assays for measuring impact on these
23
Previous I TOC
-------
EPA CompTox Research Program FY2009-2012 BOSC Review Draft- 24 August, 2009
pathways, and bioindicators of downstream key events will provide the tools for HTS of
compounds and evaluation of outcome predictions in target populations. Future efforts will also
need to focus on the translation of the computational and high throughput methods into
information that can be used in risk assessment. This is most apparent with respect to actual use
in quantitative risk assessment (i.e. being able to use in vitro or computational methods to
develop a point of departure for an IRIS assessment or other type of risk assessment), but is also
relevant to prioritization (i.e. how does the relative potency information derived from the in vitro
assays or models compare to in vivo potency) if information from screening assays is going to be
used to drive further testing decisions. Key issues in developing the next generation of risk
assessments include:
• Defining Adversity- Before these methods can be used in risk assessment, it will be vital
to understand how 'perturbations of toxicity pathways' relate to adverse effects that are of
concern for human or ecological health.
• Development of decision methods and criteria for using high-throughput data in risk
assessment.
The large interdisciplinary projects required to meet the goals of this program are dependent on
well integrated data repositories such as ACToR. These databases not only provide inputs for
empirical models for prioritization/hazard identification, but also access to well-structured data
required for data mining needs to define and evaluate toxicity pathways, dose-response modeling
needs. As hazard identification and risk characterization tools are developed, a gradual shift from
reliance on empirical models supported by limited mechanistic information to incorporation of
detailed mode-of-action for predicting hazard can proceed. The first step in this process will be
the development of prioritization tools out of ToxCast™ which incorporate toxicity pathway
information to identify chemicals of most concern and to direct more detailed testing and
modeling toward chemicals and toxicity pathways of highest importance. As computational
models are developed that relate the perturbation of a toxicity pathway quantitatively to an
adverse outcome (aided by development of virtual tissues), it should become possible to not only
prioritize but also screen for chemicals where the evidence is convincing of either safety or
toxicity. The toxicity pathways and modes of action for those chemicals where evidence is
inconclusive would then become the subject of further experimental study. Eventually,
quantitative prediction of risk from HTS data could be derived based on evaluation of toxicity
pathway predictions from systems models of tissues and organs (e.g., v-Liver™ and
v-Embryo™), and perhaps even simulated target populations.
24
Previous I TOC
-------
EPA CompTox Research Program FY2009-2012 BOSC Review Draft- 24 August, 2009
Flgure 5 The Future State: Using Hazard and Exposure
Predictions to Prioritize Testing and Monitoring
High exposure potential^ im_
Low exposure potential
HE ~~ H
HE ToxCast targets
HE
ToxCast Low
Hazard
Prediction / \ Low Priority for
E E \ToxCast Hazard Prediction Bioactivity Profiling
/
HE
Lower Priority for
Testing and Monitoring
Intelligent, Targeted Testing
t
Human Biomonitoring
A major step in higher throughput modeling and prediction of exposure will come from the
CTRP ExpoCast1 project, a collaboration of NERL and the NCCT focused on providing
Biologically-Relevant Exposure Science for 21st Century Toxicity Testing. The ExpoCast™
project is described in more detail in the project plans attached in the appendices. It should be
noted that pre-existing and other NERL research, databases and models will be critical to the
success of ExpoCast™ and the broader goals of ORD's CTRP. As ExpoCast™, ToxCast™ and
other modeling efforts succeed, it would be possible to prioritize testing and monitoring of large
numbers of chemicals and other environmental contaminants. These prioritizations would be
feasible because they would be based on hazard (H) and exposure (E) predictions of calculable
and reasonable uncertainty that were derived solely from in silico and in vitro data (Figure 5).
Upon validation of these predictions and development of decision support tools and software for
translation into a form useful for chemical prioritization by EPA program offices.
As suggested by the NRC report "Toxicity Testing in the 21st Century: A Vision and a Strategy"
and The U.S. Environmental Protection Agency's Strategic Plan for Evaluating the Toxicity of
Chemicals, the majority of current CTRP efforts are centered on advancing toxicity testing for
assessing human health effects of environmental agents. However, under environmental
legislative mandates (e.g., the Toxic Substances Control Act; the Federal Insecticide, Fungicide,
and Rodenticide Act; and the Clean Water Act), most EPA programs regulate compounds to
ensure both environmental and human health risks are properly managed. Statutory language and
resulting policy typically require decisions for chemicals that encompass environmental and
human health risks such that the CTRP will also need eventually to develop higher throughput
and computational approaches for ecotoxicology and risk assessment. Notable progress has been
made in the previous CTRP implementation plan, and within other ORD research programs, on
the development and use of toxicity pathway models, toxicology knowledgebases (e.g.,
25
Previous I TOC
-------
EPA CompTox Research Program FY2009-2012 BOSC Review Draft- 24 August, 2009
ECOTOX), and systems biology models (e.g., small fish model) in the field of environmental
science. In follow up to the current CTRP implementation plan for FY2009-2012, opportunities
will exist to bring together relevant disciplines, data, and models across both human health and
environmental risk assessment applications. As noted previously, we will be using existing
resources and projects to leverage an increased attention to ecologically relevant areas, however,
given the challenges of developing a system for relevance to human health assessment, the near
term goals of the CTRP will have to remain largely focused on that area. It is expected that
future versions of the CTRP and related implementation plans will accommodate further
progress in ecotoxicology that can be incorporated and merged with the efforts relating to human
health in this current CTRP plan.
C. Regional and Program Interactions
The CTRP is working closely with Program Offices, including the Office of Pesticides,
Prevention, and Toxic Substances (OPPTS), and the Office of Water (OW) to create informatics
tools and databases of use to the goals of both research and regulation. Within the past year, full
day sessions have been spent briefing senior management in these and other Program and
Regional Offices on the activities of the CTRP. This has resulted in shared participation, input
and authorship between Program and CTRP staff in planning, research projects, scientific
publications, and promulgation of regulatory decisions. Many examples of collaborations with
Program Offices exist, including DSSTox, ACToR, ToxRefDB and ToxCast™. Regional input
has been more limited: briefings on the CTRP have been given to particular Regions (e.g., 6 and
7), and to the regional risk assessors group. However, additional dialogue is needed as products
begin to emerge from the program that could be of utility to the Regions, especially in relation to
the Toxics Release Inventory (TRI) and Superfund programs.
In moving forward over the next several years, the key points of engagement with the Programs
will be their use of the DSSTox, ToxRefDB and ACToR databases as sources of chemical,
toxicity and exposure information, and predictions and prioritizations based on results from
ToxCast™, ExpoCast™ and the Virtual Tissues projects. More detailed descriptions of what
these engagements will be are provided in the individual project descriptions in the appendices.
Some specific examples include the continued co-development of ToxRefDB to include
additional guideline test results, as well as to be positioned to record the output of the Tier 1
Endocrine Disrupter Screening batteries.
OPPTS-OPP: Driven by its needs to assess the effects of pesticidal inerts and anti-microbial
agents, both of which suffer from limited data availability, the Office of Pesticide Programs has
been a strong proponent of this new approach to toxicity testing. OPP has defined a Strategic
Direction for New Pesticide Testing and Assessment Approaches focused on developing and
evaluating new technologies in molecular, cellular, and computational sciences to supplement or
replace more traditional methods of toxicity testing and risk assessment. This integrated
approach to testing and assessment is moving toward a new paradigm where in vivo (animal)
testing is targeted to the most likely hazards of concern. As defined by OPP, the path forward
will include close collaborations with the CTRP, in order to predict chemical toxicity and
exposure through application of efficient and effective screening tools including new in vitro
assays that rapidly provide biological profiles of the toxicological potential of chemicals (i.e.,
ToxCast™). Exposure and biomonitoring data will also be critical to interpreting toxicity data
26
Previous I TOC
-------
EPA CompTox Research Program FY2009-2012 BOSC Review Draft- 24 August, 2009
and evaluating the effectiveness of the new testing and assessment paradigm. Over the next five
years, OPP plans to enhance its integrated approach to testing and assessment to better determine
what toxicity data are needed to further refine risk assessments for chemicals that do not have
extensive toxicity information (e.g., inert ingredients, certain antimicrobial and biochemical
pesticides, and metabolites and degradates of pesticide active ingredients). Over the next 10-15
years, as experience is gained and as understanding of toxicity pathways increases, an enhanced
integrated testing and assessment approach will be implemented for all pesticides including
conventional agricultural pesticides. This approach will fully integrate hazard and exposure data
along with advanced systems modeling based on new in vitro data and an understanding of
toxicity pathways to better predict risks and to determine what additional data are necessary (i.e.,
virtual tissues). The key goals of integrated approaches to testing and assessment are to:
• improve our ability to set priorities for what data to require,
• ensure that the data requirements are focused on the right issues, and
• efficiently reach the end result of effective risk assessment.
This approach would provide the ability to focus testing on pesticide chemicals and the effects
that could most likely result in harm. As a result, testing would
• use fewer animals,
• take less time,
• be less expensive in data generation and review, and
• explore a broader range of potential adverse effects.
These goals and approach are wholly consistent with the goals and approach of the CTRP, and
represent the close working relationship between OPP and ORD in developing these tools for
integrated approaches to testing and assessment.
CTRP researchers are working with OPP and other parts of OPPTS, and with the Organization for
Economic Cooperation and Development (OECD) to explore and evaluate regulatory application
of molecular screening assays in relation to chemical testing guidelines. The goals of this effort
are to provide tools for; 1) improving the understanding of mechanisms of toxicity; 2) identifying
biomarkers of toxicity and exposure; 3) reducing uncertainty in grouping of chemicals for
assessments, inter-species extrapolation, effects on susceptible populations, etc.; and 4) providing
alternative methods for chemical screening, hazard identification and characterization. Also, the
CTRP and OPP are working with the OECD, on reducing and refining current animal testing
guidelines using the ToxRefDB database and other tools (e.g., Extended One-Generation
Reproductive Toxicity Test Guideline).
OPPTS-OPPT: For the Office of Pollution Prevention and Toxics there will be several key
points of intersection with the CTRP over the next several years. These include using HTS
technologies and bioactivity profiling to lend biologically based support to strengthen and
potential revise of chemical categories currently in the new and existing chemicals programs; to
evaluate the relative hazard of chemicals being evaluated by the Design for the Environment
program; for providing toxicity pathway data on specific groups of chemicals of high concern
(e.g., the perflourinates); and for evaluation of the feasibility of HTS approaches to
characterization of the bioactivity profiles of manufactured nanoparticles. On the exposure side,
the CTRP expects to be engaged with exposure efforts within OPPT such as the IUR (Inventory
27
Previous I TOC
-------
EPA CompTox Research Program FY2009-2012 BOSC Review Draft- 24 August, 2009
Update Rule) as procedures are developed for broad scale predictions of exposure potential
across the life cycles of chemicals. CTRP researchers are working with OPPT and with the
OECD to access and incorporate CTRP tools (e.g., DSSTox, ToxRefDB, ER QSAR models) into
the OECD QSAR Application Toolbox. The Toolbox is an international effort to provide tools
for grouping chemicals, based on the understanding of mechanisms of toxicity in order to
extrapolate properties and effects from tested chemicals to untested chemicals.
OPPTS-OSCP: For more than a decade the Office of Science Coordination and Policy has led
EPA's work to fulfill mandates contained within the Food Quality Protection Act of 1996 to
identify chemicals that can interfere with the function of natural hormones (e.g., endocrine
disrupters). This is a mode of action that may result in significant adverse consequences in
developing organisms (e.g., embryo, fetus, neonates, and children) should they be exposed to
levels sufficient to cause perturbations of their endocrine systems. Animal models have shown
that the levels that impact developing organisms can be much lower than those impacting adults
and that exposures during development can lead to adverse effects not seen until adulthood. The
Agency has adopted a Tier 1 screening battery that is designed to detect whether a chemical
substance interacts with the estrogen, androgen or thyroid hormonal systems. Combinations of in
vitro and in vivo assays are used to provide complementary measurements that detect the
endocrine disrupting potential of a chemical. Given the rapid advances in computational and
molecular sciences, discussions are underway with OPPTS on the next generation of tools that
could be more efficiently applied to the large number of chemicals of potential concern for
ability to disrupt the function of the endocrine system. Included in these discussions are the
integrative potential of contributions from CTRP in HTS of chemicals, and providing informatics
and analysis solutions and tools; the Human Health Research Program in conducting targeted
follow up testing and exposure analysis; of the Endocrine Disrupters Research Program in
understanding mechanisms of action, developing methods for assessing cumulative risks, and
improving the ability to extrapolate results across species; of the Human Health Risk Assessment
Program in assessing the contributions of multiple exposures (e.g., chemicals with
common/different modes of endocrine action) across critical life stages; and the Risk
Management Research Program in providing information on chemical life cycles that pose the
greatest potential exposure and risk, and the development of tools for greener chemicals or
processes and other mitigation strategies.
The CTRP has identified a large collection of up to 10,000 chemicals that are of high priority for
EPA Program Offices. Using available funding, a subset of 700 will be used in Phase II of
ToxCast (cost per chemical of ~$20k). Going beyond the screening of these chemicals in the
full suite of ToxCast™ Assays (>500), ORD has the capacity under existing contracts to acquire
the entire collection of chemicals and to screen them in a subset of ToxCast M assays that cover a
broad spectrum of endocrine related activities. The cost of such HTS assays would be on the
order of $ 1 -4k per chemical depending on precisely which subset of ToxCast™ assays were
included. The use of HTS assays for receptor binding and transcriptional activation and (Q)SARs
for these endpoints to obtain empirical or predicted information on chemicals for which no data
were available could provide an ordered list of which chemicals have the highest potential to
interact with the estrogen, androgen and thyroid receptors. Along with exposure information, this
prioritized list could then be used to guide other effects research, exposure analyses, and
28
Previous I TOC
-------
EPA CompTox Research Program FY2009-2012 BOSC Review Draft- 24 August, 2009
assessment and management activities in the various parts of ORD. The results could be used to
select chemicals for entry into subsequent phases of EPA's EDSP.
OW- For the Office of Water, the CTRP expects to be engaged with the assessment of the
bioactivity of chemicals on the Candidate Contaminant List (CCL) and the derivation of
subsequent CCL lists. In the course of identifying and assessing CCL compounds, the OW has to
collect and analyze hazard and exposure data on thousands of the same compounds addressed in
ACToR, DSSTox and other CTRP databases. In future iterations of this process, the CTRP will
be able to assist the OW with data curation and analysis, and provide a wealth of new hazard and
exposure data that is computable, as well as prioritization tools designed to use this data.
D. Priority Areas for CTRP Management
1. Toxicity Predictions and Chemical Prioritizations Incorporating Exposure
With the publication of the first predictive bioactivity signatures from Phase I, and the initial
proof of concept of the ToxCast™ program, we will have hazard predictions that can be
incorporated into prioritizing chemicals for further screening and testing. Phase II, scheduled for
launch in later 2009, will explore greater diversity of chemical structures and classes in order to
evaluate the robustness of the signatures identified in Phase I. As indicated in Figure 5, having
viable exposure predictions or estimations will also be critical to the envisioned prioritization
scheme. ExpoCast™ will provide an overarching framework for the science required to
characterize biologically-relevant exposure, and can thus inform chemical prioritization by
linking information on potential toxicity of environmental chemicals to real-world health
outcomes. NCCT management will support continued collaborations with NERL and other
critical partners, within EPA and externally, to improve accessibility to EPA human exposure
data and create a consolidated EPA exposure database focused on measured concentrations in
biological media. As NCCT pushes ahead into Phase II of ToxCast™, expanding the compounds
in the screening program to include nanomaterials and chemicals with demonstrated human
toxicities (e.g., failed Pharmaceuticals), it will coordinate these efforts with ExpoCast™ in order
to maximize utility of datasets to develop predictive models and decision support software for
chemical prioritizations.
2. Strengthening Cross-OKD Collaborations
Given the broad nature of the challenges facing computational toxicology, the CTRP must
engage collaborative partners across ORD in order to be successful. Cross-ORD collaborations
have been a part of the CTRP from its inception, and will continue to be a dominant feature of
the program. Numerous collaborations from previous years will carry forward, including
linkages at the management level, such as the MOU with NHEERL and NERL to provide NCCT
with administrative support functions for funds control, extramural management, quality
assurance and information management. Key research partnerships have developed between
NCCT and the rest of ORD. In addition, NCCT continues to advise NCER on the formulation of
ideas for new computational toxicology RFAs, providing suggestions for scientific peer
reviewers, serving on relevancy reviews as appropriate, and collaborating with the cooperative
research partners of the STAR grants program. The ORD's multi-year planning process provides
29
Previous I TOC
-------
EPA CompTox Research Program FY2009-2012 BOSC Review Draft- 24 August, 2009
another opportunity for linkage between the CTRP and related research efforts. Each of the
MYPs is led by a National Program Director with support from staff of the Laboratories and
Centers. The NCCT participates actively in four of the MYP teams. Three contain similar
research activities for screening and prioritizing chemicals, i.e. Endocrine Disrupting Chemicals
(EDCs, LTG III), Safe Pesticides/Safe Products (SP2, LTG I), and Drinking Water Research
Program (DW, LTG II) whereas the fourth has a major focus on the incorporation of biologically
based mode-of-action information into quantitative risk assessment (the Human Health Research
Strategy and the Human Health MYP, LTG II). Consideration is also being given to how these
data and analyses can be incorporated into next generation risk assessments, in association with
ORD's Human Health Risk Assessment program and NCEA. NCCT management meets at least
quarterly with the National Program Directors for these MYPs, in part to continue dialogue on
sharing and coordination of resources between programs and to ORD Laboratories and Centers
beyond NCCT. Besides financial resources, NCCT is cultivating cross-ORD collaborations
through shared post-doctoral, other students and through NCCT rotational fellowships for ORD
scientists. To date, scientists from NHEERL, NERL and NCEA have participated in details of 4
months or longer in NCCT to work on collaborative research projects.
ORD is in the midst of a transformation to a system of integrated, multidisciplinary (IMD)
research projects. The CTRP is actively engaged in planning and discussions on one such IMD
proposal entitled "Decision Support Tools for Preventing, Reducing and Managing
Chemical Risks." This IMD project will address the tens of thousands of chemicals and millions
of products that current regulatory decision tools don't have the ability to assess, in terms of
impact on life-stage vulnerability, genetic susceptibility, disproportionate exposures and
cumulative risk. The project will incorporate some of the predictive, high-throughput tools for
exposure and hazard being developed as part of the CTRP, scale them up and with attention to
critical life stage impacts create prioritization algorithms and next generation risk assessments
that highlight viable management options for prevention, mitigation and risk reduction.
3. Tox21: A Federal Partnership Transforming Toxicology
The NRC report on Toxicity Testing in the 21st Century has significant implications for human
health risk assessment, and in order to accelerate progress in this area, two NIH institutes and
EPA have entered into a formal collaboration known as Tox21 to identify mechanisms of
chemically induced biological activity, prioritize chemicals for more extensive toxicological
evaluation, and develop more predictive models of in vivo biological response. Consistent with
the vision outlined by Krewski et al. in the NRC report, success in achieving these goals is
expected to result in methods for toxicity testing that are more scientific and cost effective as
well as models for risk assessment that are more mechanistically based. As a consequence, a
reduction or replacement of animals in regulatory testing is anticipated to occur in parallel with
an increased ability to evaluate the large numbers of chemicals that currently lack adequate
toxicological evaluation. Ultimately, Tox21 is expected to deliver biological activity profiles that
are predictive of in vivo toxicities for the thousands of understudied substances of concern to
regulatory authorities in the United States, as well as in many other countries.
The Tox21 collaboration is being coordinated through a five-year MOU, which leverages the
strengths of each organization. The MOU builds on the experimental toxicology expertise at the
30
Previous I TOC
-------
EPA CompTox Research Program FY2009-2012 BOSC Review Draft- 24 August, 2009
NTP, headquartered at the NIH/NIEHS; HTS technology of the NIH/NCGC, managed by the
NHGRI; and the computational toxicology capabilities of the EPA's NCCT. Each party brings
complementary expertise to bear on the application of novel methodologies to evaluate large
numbers of chemicals for their potential to interact with the myriad of biological processes
relevant to toxicity. A central aspect of Tox21 is the unique capabilities of the NCGCs high-
speed, automated screening robots to simultaneously test thousands of potentially toxic
compounds in biochemical and cell-based HTS assays, and an ability to target this resource
toward environmental health issues. As mentioned by Krewski et. al., EPA's ToxCast™ Program
is an integral and critical component for achieving the Tox21 goals laid out in the MOU.
To support the goals of Tox21, four focus groups—Chemical Selection, Biological
Pathways/Assays, Informatics, and Targeted Testing—have been established; these focus groups
represent the different components of the NRC vision described by Krewski et. al. The Chemical
Selection group is coordinating the selection of chemicals for the Tox21 compound library to test
at the NCGC. A chemical library of nearly 2,400 chemicals selected by NTP and the EPA is
already under study at the NCGC and results from several dozen HTS assays are already
available. In the near term, this library will be expanded to approximately 8,400 compounds,
with an additional 1,400 compounds selected by the NTP; 2,800 compounds selected by the EPA
and provided by the CTRP; and 2,800 clinically approved drugs selected by the NCGC.
Compound selection is currently based largely on the compound having a defined chemical
structure and known purity, on the extent of its solubility and stability in dimethyl sulfoxide
(DMSO), (the preferred solvent for HTS assays conducted at the NCGC), and on the compound
having low volatility. Implementing quality control procedures for ensuring identity, purity, and
stability of all compounds in the library is an important responsibility of this group. A subset of
the Tox21 chemical library will be included in Phase II of the ToxCast™ program, which will
examine a broader suite of assays in order to evaluate the predictive power of bioactivity
signatures derived in Phase I.
The Biological Pathways and Assays group is identifying critical cellular toxicity pathways for
interrogation using biochemical- and cell-based high-throughput screens and prioritizing HTS
assays for use at the NCGC. Assays already performed at the NCGC include those to assess (1)
cytotoxicity and activation of caspases in a number of human and rodent cell types, (2) up-
regulation of tumor suppressor p53, (3) agonist/antagonist activity for a number of nuclear
receptors, and (4) differential cytotoxicity in several cell lines associated with an inability to
repair various classes of DNA damage. Other assays under consideration include those for a
variety of physiologically important molecular pathways (e.g., cellular stress responses) as well
as methods for integrating human and rodent hepatic metabolic activation into reporter gene
assays. Based on the results obtained, this group will construct test batteries useful for
identifying hazard for humans and for prioritizing chemicals for further, more in-depth
evaluation. As Tox21 progresses, it will offer an excellent opportunity to incorporate assays
specifically relevant to the assessment of chemical hazard to wildlife. For example, as assays
become available for the reactivity of nuclear hormone receptors (e.g., estrogen and androgen)
from multiple species, we will be have ability to directly compare their responsiveness and test
whether a particular species might be more sensitive to perturbations than others.
31
Previous I TOC
-------
EPA CompTox Research Program FY2009-2012 BOSC Review Draft- 24 August, 2009
The Informatics Group is developing databases to store all Tox21-related data and evaluating the
results obtained from testing conducted at the NCGC and via ToxCast™ for predictive toxicity
patterns. To encourage independent evaluations and/or analyses of the Tox21 test results, all data
as well as the comparative animal and human data, where available, will be made publicly
accessible via various databases, including EPA's Aggregated Computational Toxicology
Resource (ACToR), NIEHS' Chemical Effects in Biological Systems (CEBS), and the National
Center for Biotechnology Information's PubChem.
As HTS data on compounds with inadequate testing for toxicity becomes available via Tox21,
there will be a need to test selected compounds in more comprehensive assays. The Targeted
Testing group is developing strategies and capabilities for this purpose using assays that involve
higher order testing systems (e.g., roundworms (Caenorhabditis elegans), zebrafish embryos,
rodents).
In addition to the testing activities, the MOU promotes coordination and sponsorship of
workshops, symposia, and seminars to educate the various stakeholder groups, including
regulatory scientists and the public, with regard to Tox21-related activities.
4. Communicating Computational Toxicology
As the CTRP has matured, communication of progress has developed beyond the publication of
peer-reviewed papers, to include implementation of software and databases, websites and other
applications. Given the importance of communicating and disseminating the products of the
CTRP, recruitment of a public affairs/communications specialist to the NCCT is currently
underway, to provide even greater efforts in this area.
a. EPA Program Office Training and Implementation of Computational Tools
The NCCT and CTRP partners from across ORD have given numerous seminars, held
multiple 1-2 day workshops, and provided specific training and installation of
computational tools for EPA program offices. These ad hoc approaches are now being
formalized into an online menu of lectures and tutorials on a broad range of
computational toxicology topics. Senior scientists from NCCT, NHEERL and NERL are
contributing to this resource, and a FY2009 NCCT recruitment of a communication
specialist will accelerate development of this effort. In FY2010, computational
toxicology training will initially focus on the tools ready for program office use,
including DSSTox, ACToR, ToxRefDB and the ToxMiner tool for analyzing ToxCast™
data.
b. Communities of Practice for Chemical Prioritization and Exposure Science
On the scientific level, the NCCT has initiated two Communities of Practice (CoP) in the
areas of Chemical Prioritization and Exposure Science that are intended to unite
practitioners in the designated fields. The concept of the CoPs was suggested by the
BOSC in April 2005, and has since been adopted as a primary means of communication
and integration of activities across ORD, the EPA, and outside entities and stakeholders.
32
Previous I TOC
-------
EPA CompTox Research Program FY2009-2012 BOSC Review Draft- 24 August, 2009
These efforts will serve to enhance communication and coordination, develop common
standards, promote consistency, evaluate and provide guidance on best practices,
recommend research priorities, and provide training to interested parties.
5. Developing Clients for Virtual Tissues
The virtual tissue projects are developing systems models of the liver and embryo that will make
the data from CTRP databases and chemical prioritization projects more useful in quantitative
risk assessments. Over the course of design and implementation of these systems models, interim
milestones and deliverables will need to be developed and communicated to Program Offices.
The process of communicating these products, and taking feedback from the Programs, will
serve to identify and establish longer-term clients for these ambitious projects that are utilizing
cutting-edge science that is very different from current regulatory practice.
The Virtual Liver (v-Liver™) project is actively engaging Program Office personnel to address
challenges in mode of action (MOA) elucidation and quantitative dose-response prediction for
chronic liver injury. In FY2009, chemicals that work through nuclear receptor pathways (e.g.,
CAR, PXR, PPARs) were chosen for a proof-of-concept. Information on these chemicals is
being used to populate the v-Liver™ Knowledgebase (v-Liver-KB) and develop a liver simulator
model. Program Office staff from OPPTS and NCEA were consulted on the selection of these
chemicals from key classes of pesticides and industrial chemicals. The first deliverable for risk
assessors will be v-Liver-KB, which formally organizes information on normal hepatic functions
and their perturbation by chemical stressors into pathophysiologic states. The v-Liver-KB will be
deployed as an interactive web-based and desktop tool to intuitively browse and query
physiologic knowledge on chemicals. This can then be used to hypothesize and test putative
MOA(s), and to link assay results from ToxCast™ and ToxRefDB with other evidence curated
from the literature. This system will provide computable information on key events that
transparently indicate the uncertainties and data gaps, and that make inferences on MOA from
experimental data. In addition, we will work closely with risk assessors to customize the system
for specific requirements. Beta versions of the liver simulator will be applied to Program Office
issues relating to key chemical classes- this will be an intensively collaborative process between
CTRP scientists and OPPTS and NCEA staff. CTRP scientists working on the v-Liver™ project
will also work with OPP staff on retrospective analyses of chronic and cancer in vivo test data,
further introducing OPP to the v-Liver-KB and the simulator as appropriate.
Motivation for the Virtual Embryo (v-Embyo™^ is the scientific need to understand mechanisms
of toxicity and predict developmental defects from complex datasets. The research goal is to
simulate embryonic tissues reacting to perturbation across chemical class, system, stage, genetic
makeup, dose and time. Data input is detailed knowledge of molecular embryology, as well as
high-throughput data from in vitro models on signaling pathways and cellular phenotypes. Initial
efforts have focused on retrospective analyses of regulatory data from reproductive and
developmental toxicity tests in ToxRefDB. This work has been a collaborative effort with OPP
that has fostered working relationships between the CTRP scientists and Program Office staff.
Next steps are to identify appropriate developmental and reproductive toxicities that will support
development of systems model, and are also of regulatory interest to Program Offices. This will
include the incorporation of high throughput data from morphogenetically-competent in vitro
33
Previous I TOC
-------
EPA CompTox Research Program FY2009-2012
BOSC Review Draft- 24 August, 2009
assays such as the Embryonic Stem Cell test and Zebrafish Embryo Test through collaborations
with NHEERL and outside partners. Virtual Embryo's first goals are to create a knowledgebase
and simulation engine that enable in silico reconstruction of key developmental landmarks,
which develop by regulating conserved signaling pathways and cellular processes and that are
sensitive to environmental chemicals of scientific and regulatory interest to EPA Program
Offices.
Previous
34
TOC
-------
EPA CompTox Research Program FY2009-2012
BOSC Review Draft- 24 August, 2009
III. CTRP PROJECT SUMMARIES FOR FY2009-2012
A. Intramural Projects Coordinated by NCCT
There are nine CTRP projects coordinated by investigators in NCCT which collectively comprise
the core of the CTRP moving forward into FY2009-2012. Individual project plans for each of
these append this document (Section IV). These projects span the source to outcome continuum
of toxicology research (Figure 6), providing critical components for next generation risk
assessments.
Figure 6
Applying Computational Toxicology Along
the Source to Outcome Continuum
Source/Stressor Formation
\
Environmental Cone.
\
ExpoCast
External Dose
1. ACToR - Aggregated Computational Toxicology Resource
Lead/Principal Investigator: Richard Judson
ACToR is a web-based informatics platform, organized at the top level by chemical and
chemical structure, which is indexing, collecting, and organizing many types of data on
environmental chemicals (Judson et al 2008}. Environmental chemicals are defined as those
likely to be in the environment, including all chemicals regulated or tracked by the EPA, as well
as related chemicals, such as pharmaceuticals, that find their way into water sources. ACToR is
indexing and linking to data from hundreds of sources, including the EPA, FDA, CDC, NIH,
academic groups, other governmental agencies (state and national) and international
organizations, such as the WHO. Information being indexed and gathered includes in vivo
toxicity, in vitro bioassay data, use levels, exposure information, chemical structure, regulatory
information and other descriptive data. Planning for the project began in mid-FY2007; beta
versions were available inside the EPA since early FY2008; and a public version became
available in December 2008. ACToR consists of a back-end database and a front-end web
Previous
35
TOC
-------
EPA CompTox Research Program FY2009-2012 BOSC Review Draft- 24 August, 2009
interface built on low-cost, publicly accessible applications and tools. Over the next 3 years,
ACToR will expand to include more publicly available resources and data, including more
information extracted from text reports and tabularized, and more information on chemical use
and exposure. The latter effort will be coordinated with the efforts of ExpoCast™ and NERL to
identify, index and extract data from exposure-related resources of highest interest and
importance to EPA programs. In planned upgrades to ACToR, the ability of users to perform
flexible searches across different layers of data will be enhanced, and customized data
downloads will be implemented. ACToR will serve as the primary vehicle to aggregate and
publicly disseminate all published data associated with the ToxCast™, ToxRef, and Tox21
research projects. Additionally, the ToxMiner and NCCT Chemical Repository systems are
being developed as part of ACToR. These are data repositories and data analysis engines for the
ToxCast™/Tox21 projects.
2. DSSTox- Chemical Information Technologies in Support of Toxicology Modeling
Lead/Principal Investigator: Ann Richard
The DSSTox project has implemented high quality data review procedures for standardized
chemical structure annotation, created linkages connecting diverse toxicity resources within and
outside of EPA, and published high quality EPA chemical inventories and toxicity data files
spanning over 10,000 substances. The broad utility of DSSTox data files for cheminformatics
and modeling applications provides significant opportunities to influence the course of predictive
modeling strategies and to encourage wider engagement of toxicologists in toxicity data
representation. The DSSTox project will continue efforts to expand chemical data file offerings
into less well-represented areas of toxicology (immunotoxicology, toxicogenomics, etc.), and
provide varied representations of summary toxicity endpoints. In addition, this research will
explore new representations of chemical structure in relation to the biology (e.g., analog
measures, chemical features, chemical classes), and new representations of biological endpoints
in relation to modeling (e.g., quantitative endpoints in terms of potency, summarized or grouped
effects, qualitative active and inactive classes). These efforts will be designed to complement and
augment projects in NCCT (ToxCast™, ACToR, and Tox21) that are working to improve
capabilities to access, mine, and integrate chemical-biological activity information from existing
and new data, both within and outside EPA, in support of toxicity prediction efforts. In close
coordination with ACToR, which is the primary informatics resource for ToxCast™ and Tox21
chemical and biological data, DSSTox will provide the initial chemical registration IDs, structure
annotation, and quality review of ToxCast™ and ToxRef inventories, as well as the expanded
Tox21 chemical testing library, helping to ensure quality and consistency of chemical
information across the various NCCT programs.
36
Previous I TOC
-------
EPA CompTox Research Program FY2009-2012 BOSC Review Draft- 24 August, 2009
3. ToxRefDB - Toxicity Reference Database
Lead/Principal Investigator: Matthew Martin
Thirty years of registration toxicity data and open literature studies have been historically stored
as hardcopy and scanned documents by the EPA and others. A significant portion of these data
have now been processed into standardized and structured toxicity data within the EPA's
Toxicity Reference Database (ToxRefDB), including chronic, cancer, developmental, and
reproductive studies from laboratory animals (Martin et al 2009a ; Martin et al 2009b\ Knudsen
et al 2009). ToxRefDB is a collaborative project between NCCT and OPP. These data are now
accessible and mineable within ToxRefDB and are serving as a primary source of validation for
U.S. EPA's ToxCast™ research program in predictive toxicology. In addition to providing
reference toxicity information to research efforts, ToxRefDB will be mined for information on
the role and impact of previous and current study guidelines on the regulation of environmental
chemicals. The initial collection of studies in March of 2006 focused primarily on the reviews of
registrant submitted toxicity studies on pesticide active ingredients. ToxRefDB design,
development and implementation were completed in mid-FY2006 with ongoing updates to the
standardized vocabulary and data entry tool interface. The entry of over 2,000 studies spanning
the majority of the ToxCast™ Phase I chemical set was completed late FY2008. The status of
these initial datasets are either published, in press, or submitted and are being made publicly
available through the ToxRefDB website. A web-based query tool for the entire contents of
ToxRefDB will become available to the public in 2009, in conjunction with a quarterly update of
the EPA ACToR program. ToxRefDB will continue to enter available data from chronic, cancer,
developmental, and reproductive studies with a focus on potential ToxCast™ Phase II chemicals.
The availability and entry of toxicity data into ToxRefDB will also guide the selection of
ToxCast™ Phase II chemicals. Over the next year, ToxRefDB will also be expanded to capture
developmental neurotoxicity (DNT) study data and possibly in vivo data submitted to the EDSP.
In addition, ToxRefDB is being used for retrospective analyses by various EPA, OECD and
other workgroups working on revisions to animal test guidelines and other projects.
4. ChemModel- Application of Molecular Modeling to Assessing Chemical Toxicity
Lead/Principal Investigator: James Rabinowitz
This project is using modern molecular modeling methods developed for the discovery of novel
pharmaceutical agents to computationally predict toxicant-biomolecular target interactions. A
library of computational models of relevant biomolecular targets is being developed. Molecular
modeling approaches may then be used to interrogate this library for the capacity of specific
environmental molecules to interact with each target. The endocrine system provides a test for
the utility of this approach because many of the pathways for toxicity and the macromolecular
targets in those pathways have been identified. Appropriate experimental crystal structures of
many of the receptor protein targets are available to create the computational library of targets.
The ultimate objective of this research is to develop a library of biomolecular targets for
chemical toxicity, and the methods appropriate for their application to predicting the capacity of
a chemical to interact with these targets. This library of targets may then be used in conjunction
with other approaches as part of a chemical prescreen.
37
Previous I TOC
-------
EPA CompTox Research Program FY2009-2012 BOSC Review Draft- 24 August, 2009
5. ToxCast™- Screening and Prioritization of Environmental Chemicals Based on
Bioactivity Profiling and Predictions of Toxicity
Lead/Principal Investigator: Keith Houck
The objective of the ToxCast™ research program is to develop a cost-effective and rapid
approach for screening and prioritizing a large number of chemicals for toxicological testing.
Using data from HTS bioassays developed in the drug discovery field, ToxCast™ is generating
data, constructing databases, and building computational models to forecast the potential human
toxicity of chemicals. HTS bioassays for ToxCast™ are also being provided by NHEERL
partners. These hazard predictions should provide EPA regulatory programs, including OPP,
with science-based information helpful in prioritizing chemicals for more detailed toxicological
evaluations, ultimately leading to reduced animal testing. Furthermore, the toxicity pathways
identified from this dataset and project will be critical to transforming the practice of risk
assessment for environmental chemicals and contaminants (in collaboration with NCEA and
EPA Program Offices). ToxCast™ is a multi-year effort that is divided into three distinct phases:
• Phase I: 300 chemicals assayed in over 600 different HTS bioassays, to create
predictive bioactivity signatures based on the known toxicity of the chemicals;
• Phase II: focused on confirmation and expansion of ToxCast™ predictive signatures,
generating HTS data on 700 additional chemicals;
• Phase III: ToxCast™ expanded to thousands of environmental chemicals for which
little toxicological information is available.
Once ToxCast™ has gone through successful initiation of Phase III, the data and toxicity
predictions will be ready for deployment throughout numerous EPA program offices. NCCT will
work to link these hazard predictions with exposure predictions, and create integrated database
analysis tools facilitating customized chemical prioritizations appropriate to specific programs.
Beyond the initial application of ToxCast™ data and tools to prioritizing chemicals for further
screening, testing and monitoring, secondary applications will include the Virtual Tissues
systems modeling projects, and next generation risk assessments with NCEA, NHEERL and
EPA program offices.
6. ExpoCast™- Exposure Science for Screening, Prioritization, and Toxicity Testing.
Lead/Principal Investigator: Elaine Cohen Hubal
ExpoCast™ will provide an overarching framework for the science required to characterize
biologically-relevant exposure as a critical part of the CTRP (Cohen Hubal, 2009). The
ExpoCast™ program will foster novel exposure science research to (1) inform chemical
prioritization, (2) understand system response to chemical perturbations and implications at the
individual and population levels, and (3) link information on potential toxicity of environmental
contaminants to real-world health outcomes. An important early component of ExpoCast™ will
be to consider how best to consolidate and link human exposure data for chemical prioritization
and toxicity testing. ExpoCast™ represents a strong collaboration between NCCT and NERL,
with both parties providing leadership and critical scientific contributions towards this
transformation of exposure science. Initial research will focus on identifying and evaluating
novel approaches for characterizing exposure to prioritize chemicals and developing modeling
approaches for considering exposure potential based on chemical properties, sources (e.g.,
consumer products), uses, lifecycle, and individual/population vulnerability. Beyond the initial
38
Previous I TOC
-------
EPA CompTox Research Program FY2009-2012 BOSC Review Draft- 24 August, 2009
application of ExpoCast™ data and tools to prioritizing chemicals for further screening, testing
and monitoring, secondary applications will include next generation risk assessments with
NCEA, NERL and EPA program offices.
7. v-Embryo™ - Virtual Embryo
Lead/Principal Investigator: Thomas Knudsen
Motivation of the Virtual Embryo (v-Embyro™) is scientific needs to understand mechanisms of
toxicity and predict developmental defects from complex datasets. EPA must evaluate
environmental chemicals for potential effects on development. Part of this challenge is to
understand mechanisms by which chemicals disrupt prenatal development. Unfortunately, the
mechanisms of prenatal developmental toxicity are not understood in sufficient depth or detail
for risk assessment purposes. Because embryonic tissues are regulated simultaneously by
pathways that control genetic patterning, molecular clocks, morphogenetic tissue
rearrangements, and cellular differentiation there is a need for computational (in silicd] models to
address this complexity. EPA's v-Embyro™ will comprise a framework to merge data and
knowledge about developmental processes, leading to cell-based computational models that can
be used to analyze mechanisms in developmental toxicity. This is a collaborative project between
NCCT, NHEERL, and NCEA. Data input is detailed knowledge of molecular embryology, high-
throughput data from in vitro models, signaling pathways, and cellular phenotypes. Output
models aim for the modular reconstruction of a developing embryo from cell-based models of
morphogenesis and differentiation.
8. v-Liver™ - The Virtual Liver Project
Lead/Principal Investigator: Imran Shah
The Virtual Liver (v-Liver™) computational paradigm represents tissues as cellular systems in
which discrete individual cell level responses give rise to complex physiologic outcomes. In this
model cell level responses are governed by a self-regulating network of normal molecular
processes, and adverse histopathologic effects arise due to chronic stimulation by environmental
chemicals. The v-Liver™ proof of concept (PoC) is being developed by: (i) focusing on
environmental chemicals responsible for hepatocarcinogenesis in rodent studies, (ii) organizing
mode-of-action (MOA) knowledge on the relevant molecular and cellular processes perturbed by
these chemicals; (iii) developing a tissue simulation platform to investigate the uncertainties in
MOA and neoplastic lesion formation; and (iv) evaluating in vitro assays to predict lesion
development across chemicals and doses. Virtual Liver is a collaborative project between NCCT
and NHEERL.
9. Uncertainty Analysis in Toxicological Modeling
Lead/Principal Investigator: R. Woodrow Setzer
The goals of this project are to develop standardized and more efficient computational
approaches for parameter estimation and model selection; standardize approaches for model
evaluation for PBPK and other dynamic models; and develop methods for constructing priors
(probabilistic summaries of current knowledge) for model parameters, based on existing
computational methods and data sets. The initial motivation for this work was the need to
39
Previous I TOC
-------
EPA CompTox Research Program FY2009-2012 BOSC Review Draft- 24 August, 2009
standardize and make more sophisticated parameter estimation and model evaluation for PBPK
models being used by OPP, and that emphasis will continue in the early phase of this project.
However, all models relevant to toxicological risk assessment have similar requirements, and this
project will coordinate closely with the Virtual Tissues and ToxCast™ projects. In particular, the
project will collaborate with ToxCast™ in developing approaches for quantifying uncertainty in
ToxCast™ predictions and prioritizations.
B. Intramural Projects Coordinated by NERL and NHEERL
Several key components and research projects of the CTRP are coordinated by NERL, NHEERL
or both in conjunction with NCCT and other collaborators. In addition,
1. National Exposure Research Laboratory
NERL is conducting human exposure research for screening, prioritization, and toxicity testing
in collaboration with and complementary to the ExpoCast™-project. Exposure science is crucial
for addressing many of our important and complex environmental health issues and is essential
in order for toxicity testing to be valuable in public health protection. There is a clear need for a
collaborative effort across the exposure and risk assessment community to ensure the required
exposure science, data and tools are ready to address immediate needs resulting from application
of high-throughput in vitro technologies for toxicity testing. A coherent program is required to
formulate significant exposure questions posed by these novel in vitro toxicity data, develop
creative approaches for applying existing exposure information and tools to address these
questions, and finally identify key exposure research needs to interpret the toxicity data for risk
assessment. The authors of the National Academies report (NRC, 2007) emphasize that
population-based data and human exposure information are required at each step of their vision
for toxicity testing and risk assessment. The collaborating NERL and NCCT scientists have
identified and will be conducting research in the following priority exposure research areas to
support chemical screening, prioritization, and toxicity testing (Sheldon and Cohen Hubal,
2009}: (1) accessible and linkable exposure databases; (2) exposure screening tools for
accelerated chemical prioritization; (3) computational tools for dose reconstruction and source-
to-outcome analyses; (4) tools for understanding the fundamental processes and factors
influencing human exposures; and (5) efficient monitoring methods to measure and interpret
biologically-relevant exposure metrics. Susceptibility, vulnerability, and life-stage aspects are
integral to each of these.
The NERL directed aspects of the ExpoCast™ program will include research required to
understand fundamental processes and factors influencing human exposures as well as
development of the tools required to facilitate efficient exposure assessment. This research will
be implemented in the following broad areas:
Prioritization and screening. Two related research activities will be implemented, the
first developing high quality, high quantity exposure databases aligned with the NCCT
databases, and the second developing and evaluating screening models for risk assessment.
NERL will inventory, compile and organize available environmental and exposure data into
readily accessible databases. These databases will be efficiently organized so that they are
40
Previous I TOC
-------
EPA CompTox Research Program FY2009-2012 BOSC Review Draft- 24 August, 2009
aligned and integrated with other NCCT databases, allowing the users to link source,
environmental, exposure, and effects data for chemicals and degradates/metabolites. Research
will be conducted to develop the next generation of predictive environmental fate and transport,
exposure and dose screening models that can be linked with the corresponding NCCT,
NHEERL, and other Agency toxicity screening-level models. Available screening models will be
inventoried and assessed. A workshop of international experts will be scheduled for evaluating
the available models currently being used by various international organizations. New/refined
models will be developed, based on the evaluation results, and linked with appropriate Agency
toxicity models for future risk assessments.
Linked exposure-dose models. Research will be conducted to efficiently link NERL's
environmental, exposure and dose models and databases for supporting Program Office specific
exposure and risk assessments. Emphasis will be on developing tools and approaches to facilitate
rapid assessment.
Biomarkers. Collaborative research with scientists from ORD, CDC, and academia will
be implemented to design studies for developing and evaluating tools to interpret the results of
exposure biomarker studies and link these results to indicators of first biological response. The
Metabolic Simulator and Metapath research tools will be upgraded and their utility for risk
assessments will be evaluated using industry provided data. Collaborative observational studies
will be conducted with CDC, NHEERL, and others to develop and refine models for relating
measured exposure biomarker results with environmental exposures (both forward and reverse
dosimetry). Research will be conducted with NCCT and NHEERL to understand how 'omics-
based exposure biomarker data, combined with chemical biomarker results, can be used to link
exposure with indicators of effects.
Observational studies. Collaborative research with NCS and/or STAR awardees will be
conducted to identify and characterize the key factors influencing children's exposures to
pesticides and other chemicals. Research activities will also be implemented to characterize real-
world exposures to multiple chemicals (mixtures) in targeted communities and/or vulnerable
populations (including children). Tools will be developed for predicting high exposures,
understanding the factors contributing to these exposures, providing input for the development
and evaluation of risk reduction strategies.
2. National Health and Environmental Effects Research Laboratory
Research within NHEERL is both parallel to, and integrated with the CTRP. While some of this
research has been ongoing for a number of years, a new effort to expand this program is
currently underway. The overall goal is the prediction of chemical toxicity to humans and
wildlife based on understanding fundamental biology and its perturbation by toxicants. The
approach is based on the elucidation of key events that link initiating events to adverse outcomes,
and leverages expertise in human studies, whole animal toxicology in a wide range of species,
and cellular and molecular biology to identify toxicity pathways associated with adverse health
and ecological outcomes. The four focus areas for NHEERL research and the integration of these
efforts into the CTRP is described below.
41
Previous I TOC
-------
EPA CompTox Research Program FY2009-2012 BOSC Review Draft- 24 August, 2009
Linkage of environmental exposure to perturbation of toxicity pathways. For the
appropriate application of high throughput assays based on toxicity pathways to chemical
screening (hazard identification) and for risk assessment, the environmental exposure levels and
their relation to the exposure at the cellular level must be known. Pharmacokinetic studies and
modes will be used to determine the relationship between external exposures and tissue dose in
vivo, while cellular and molecular biology studies of in vivo effects and subsequent development
of toxicity pathway assays will broaden the scope of screening assays.
Linkage of toxicity pathway perturbations to adverse outcomes. Research here is focused
on identifying toxicity pathways, key events, and modes of action (MOA) as they relate to
adverse outcomes and disease. Mode of Action (MOA) is defined as a sequence of key events
and processes, starting with interaction of an agent with a cell, proceeding through operational
and anatomical changes, and resulting in an adverse health effect. The use of global biology
measures ("omics") will be used to discover new toxicity pathways that will then be translated
into medium and high thoughput in vitro and non-mammalian screening assays. Results from
these new technologies will be compared with predictions from in vivo experiments. In cases
where molecular targets are established for chemical classes, quantitative structure-activity
relationship ([QJSAR) models and read-across methods will be developed to predict the
toxicological potential of untested chemicals. These efforts will be coordinated with activities in
OPPTS, OECD and the European Union to ensure the efforts meet Agency and International
needs for regulatory purposes.
Development of toxicity pathway assays. NHEERL is developing assays for
neurodevelopmental, immunotoxicity and cellular stress responses. As these assays are
established, these will be used to expand the breadth of assays in the ToxCast testing program.
To date, the ToxCast™ Phase I chemicals have been tested in several neurodevelopmental and
cellular stress assays. Through participation in the Tox21 MOU, NHEERL has contributed
additional toxicity pathway assays related to the assessment of stress responses in cellular
systems. In addition, NHEERL will be performing secondary screening and targeted testing to
explore insights and hypotheses generated from the ToxCast and Tox21 HTS efforts. These
follow up studies are designed to evaluate findings from the screening and provide quantitative
in vivo relationships.
Quantitative models for risk assessment. As quantitative relationships are established
between assay conditions and environmental exposures for humans and wildlife, a transition to
toxicity pathway-based risk assessment becomes technically feasible. Data to define these
quantitative relationships will be generated in collaboration with NCEA and other ORD partners
to ensure the suitability for use in quantitative modeling. Research integrated with NERL
modeling efforts will make PK models compatible with exposure models. Modeling of MOA is
being conducted in collaboration with the NCCT through modeling of Virtual
Tissues/Systems/Organisms, as well as less detailed BBDR modeling projects. Through the
integrated nature of this research, it is anticipated that in vivo models will be generated for
comparison with in vitro toxicity pathway screening results. In addition, models will be
developed which can take in vitro results directly as input and make quantitative in vivo
predictions for target organisms. The v-Liver™ and v-Embryo™ projects are two NCCT-
NHEERL collaborations currently underway in this area. An additional virtual cardiopulmonary
42
Previous I TOC
-------
EPA CompTox Research Program FY2009-2012 BOSC Review Draft- 24 August, 2009
project is currently being developed based on extensive NHEERL research in this area and the
CTRP "new start" project looking at the mechanistic indicators of childhood asthma (MICA).
Special considerations for ecology research. The actions presented thus far are
applicable to both health and ecological problems. Ecology has some specific issues, however,
which must be addressed in parallel with the efforts described above. First, while human health
risk assessment can be conducted on the basis of a susceptible subpopulation's risk of an adverse
outcome, most ecological concerns relate to the effect on the viability of the population. There
has been demonstrated success within NHEERL of linking MOA models to population models
which will be extended to toxicity pathway-based models to be used for ecological risk
assessment as relevant. Second, the NRC report highlights the many problems when
extrapolating among species and recommends that species extrapolation be avoided by focusing
on human cells for toxicity pathway assays. This is not possible for ecological risk assessment as
the number of relevant species is much too large for direct testing in each one. Therefore, the
identification of appropriate sentinel species and development of toxicity pathway assays in these
species will be coupled with the development of methods for species extrapolation.
C. Extramural STAR Grantee Projects
Implicit in many of the research projects contained with the CTRP, bioinformatics is one area of
research that is needed. This rapidly emerging technology is crucial to the computational
toxicology program and there remains a large gap in ORD relative to the ability to analyze the
high volumes of molecular data and to predict potential toxicity, modes of action, and ultimately
risk. To help bridge this gap, NCER has supported the establishment of two STAR
Environmental Bioinformatics Research Centers (EBRC). The Research Center for
Environmental Bioinformatics and Computational Toxicology at the University of Medicine &
Dentistry of New Jersey (UMDNJ), Piscataway, NJ, and The Carolina Environmental
Bioinformatics Research Center at the University of North Carolina, Chapel Hill, are operating
as cooperative agreements and helping to facilitate the application of bioinformatics tools and
approaches to environmental health issues supported by the CTRP.
In the next year, UMDNJ researchers will begin the design of new ebTrack interfaces to open
source databases and to various "external" and Center-developed modeling tools for facilitating
wider-deployment and applicability of the ebTrack/ArrayTrack system for integrative analyses of
various types of genomic, proteomic, and metabonomic data. Additionally, plans are underway
to refine the environmental bioinformatics Knowledge Base (ebKB) and to make a public beta
version of ebKB available; implement a modular "Virtual Liver" with alternative levels of detail
in describing physical structure of the liver with respect to toxicokinetic and toxicodynamic
processes with case studies focusing on environmentally relevant chemicals; and refine the
framework for DORIAN (Dose-Response Information Analysis) modules representing different
scales of biological complexity ranging from molecule-molecule interactions to biochemical
networks to virtual organs and systems.
For the Carolina Bioinformatics Center, goals for the next year include; (i) continuing progress
in dose-response pathway modeling and analysis of ToxCast™ Phase I data; (ii) the continuation
of QSAR modeling of multiple animal toxicity endpoints; and (iii) the development of QSAR
and other statistical models to use in vitro biological data to predict in vivo toxicity endpoints.
43
Previous I TOC
-------
EPA CompTox Research Program FY2009-2012 BOSC Review Draft- 24 August, 2009
Other plans include the development of specific data-mining algorithms for genomic databases,
and extending computational work on fast approaches for genome-wide expression QTL analysis
to human haplotypes.
The third STAR Center is the Carolina Center for Computational Toxicology at the University of
North Carolina, Chapel Hill. This Center is developing fine-scale predictive simulations of the
protein-protein/-chemical interactions in nuclear receptor networks; mapping chemical-perturbed
networks and devising modeling tools that can predict the pathobiology of compounds based on
a limited set of biological data; building tools that will enable toxicologists to understand the role
of genetic diversity between individuals in responses to toxicants; and creating unbiased
discovery-driven prediction of adverse chronic in vivo outcomes based on statistical modeling of
chemical structures and high-throughput screening.
Another high-priority for EPA is to understand the molecular and cellular processes that, when
perturbed, result in developmental toxicity. In response to this need, NCER has funded the
Texas-Indiana Virtual STAR Center; Data-Generating in vitro and in silico Models of
Developmental Toxicity in Embryonic Stem Cells and Zebrafish at the University of Houston,
Texas A&M Institute for Genomic Medicine, and Indiana University. As chemical production
increases worldwide, there is a concordant increased need for determining the hazard and risk to
human health at realistic exposure levels. The main objective of the proposed multidisciplinary
Texas Indiana Virtual STAR (TIVS) Center is to contribute to a more reliable chemical risk
assessment through the development of high throughput in vitro and in silico screening models
of developmental toxicity. Specifically, the TIVS Center aims to generate in vitro models of
murine embryonic stem cells and zebrafish for developmental toxicity. The data produced from
these models will be further exploited to produce predictive in silico models for developmental
toxicity on processes that are relevant also for human embryonic development.
D. Summary Integration of the CTRP Projects for FY2009-2012
The CTRP spans several ORD Laboratories and Centers, as well as the extramural STAR grants
program. Collectively these various components of the CTRP are developing new methods and
tools that will enhance our ability to predict adverse effects and understand the mechanisms
through which chemicals induce harm. Advances from the CTRP will give EPA the ability to
screen and assess a larger number of chemicals than traditional methods allow. In addition, EPA
is collaborating with other governmental and private organizations to leverage resources and
access complementary expertise in order to accelerate progress in high-priority research areas.
Throughout the various components of the CTRP, focus is maintained on addressing a number of
key science questions:
• What are the key linkages in the continuum between the source of a chemical in the
environment and its adverse outcomes?
• How can we develop predictive models for screening and testing?
• How can we improve quantitative risk assessment and reduce uncertainty by using
advanced computational techniques?
44
Previous I TOC
-------
EPA CompTox Research Program FY2009-2012
BOSC Review Draft- 24 August, 2009
• How can we enhance dose-response modeling, especially in low-dose ranges, to
include knowledge of molecular events?
In order to address these questions ORD's CTRP will continue to provide informatics, chemical
prioritization and systems modeling solutions for EPA. ACToR, DSSTox, ToxRefDB;
ToxCast™ and ExpoCast™; and v-Liver™ and v-Embryo™, are all excellent examples of
extensive, multidisciplinary and integrated projects that bring together the talents of ORD, along
with extramural scientists from EPA funded STAR centers to provide the high throughput
decision support tools for screening and assessing chemical exposure, hazard and risk.
Previous
45
TOC
-------
IV. APPENDICES
A. Intramural CTRP Projects
1. Project Plans
a. ACToR - Aggregated Computational Toxicology Resource
Lead/Principal Investigator: Richard Judson
Version: March 4, 2009
Research Issue/Relevance: The EPA faces a significant issue in that there are many chemicals
in wide-spread use, and which the agency regulates, for which there is little or no toxicology
information. The EPA ToxCast™ program is one effort that is addressing this problem by
screening many of these chemicals using high-throughput techniques and helping prioritize
which ones are candidates for more detailed testing. To be effective, ToxCast™ needs to know
what chemicals are in need of screening, and needs to know what is already known (or not) about
these chemicals. ACToR is providing this information. In addition, the ToxCast™, ToxRefDB
and Tox21 projects each have need for unified, sophisticated informatics support and
management of the large amount of chemical and biological information that are central aspects
of these projects. Additional capabilities tied to this informatics resource will be needed to
address the data analysis and toxicity prediction challenges of the ToxCast™/Tox21 projects.
A related issue is that other EPA programs as well as external stakeholders need easy access to
information on environmental chemicals both within and beyond the set of interest to ToxCast™.
Currently, toxicity and exposure data associated with environmental chemicals does not adhere
to standardized representations, and is widely dispersed across many databases and Internet
resources, many of which are difficult to access or search. The ACToR project is addressing this
challenge by creating a central, standardized, publicly accessible chemical-informatics platform
to enable searching and cross referencing of chemical-associated toxicity information to aid
prioritization and hazard identification of environmental chemicals.
Purpose/Objective/Impact: ACToR aims to provide a unified, centralized resource of data on
environmental chemicals including toxicology, in vitro assay data, and chemical structure
information. By gathering information on the type and location of toxicity or exposure data
associated with environmental chemicals into a single, searchable, publicly accessible web-site,
ACToR is providing the basis for chemical selection and screening within NCCT projects, such
as ToxCast™ and Tox21. ACToR is also coordinating with the DSSTox project to incorporate
quality chemical review and structure-annotation for the chemical data sets of highest interest to
the various NCCT projects. In addition to its use in supporting various NCCT and EPA projects,
ACToR is a publicly available EPA resource that enables other government agencies, industry,
and academic researchers to quickly search and collate toxicity-related information on chemicals
of interest. As such, it will promote and encourage other entities to adopt standards for chemical
representation and broadly survey chemical information pertaining to toxicology resources on
the Internet.
A-l.a.l
Previous I TOC
-------
EPA CompTox Research Program FY2009-2012 BOSC Review Draft- 24 August, 2009
Synopsis: ACToR is a web-based informatics platform, organized at the top level by chemical
and chemical structure that is indexing, collecting, and organizing many types of data on
environmental chemicals. Environmental chemicals are defined as those likely to be in the
environment, including all chemicals regulated or tracked by the EPA, as well as related
chemicals, such as Pharmaceuticals that find their way into water sources. ACToR is indexing
and linking to data from hundreds of sources, including the EPA, FDA, CDC, NIH, academic
groups, other governmental agencies (state and national) and international organizations, such as
the WHO. Information being indexed and gathered includes in vivo toxicity, in vitro bioassay
data, use levels, exposure information, chemical structure, regulatory information and other
descriptive data. Planning for the project began in mid-FY07; beta versions were available inside
the EPA since early FY08, and a public version became available in December 2008. ACToR
consists of a back-end database and a front-end web interface built on low-cost, publicly
accessible applications and tools. Over the next 3 years, ACToR will expand to include more
publicly available resources and data, including more information extracted from text reports and
tabularized, and more information on chemical use and exposure. The latter effort will be
coordinated with the efforts of ExpoCast™ and NERL to identify, index and extract data from
exposure-related resources of highest interest and importance to EPA programs. In planned
upgrades to ACToR, the ability of users to perform flexible searches across different layers of
data will be enhanced, and customized data downloads will be implemented. ACToR will serve
as the primary vehicle to aggregate and publicly disseminate all published data associated with
the ToxCast1 , ToxRef, and Tox21 research projects. Additionally, the ToxMiner and NCCT
Chemical Repository systems are being developed as part of ACToR. These are data repositories
and data analysis engines for the ToxCast/Tox21 projects.
Partnerships/Collaborations (Internal & External):
1. EPA ToxCast™ program - provide data for use in selecting chemicals and providing
toxicology data for validation; provide route for publication of data
2. Tox21 partnership - provide data for use in selecting chemicals and providing toxicology
data for validation; provide route for publication of data
3. DSSTox coordination - align methods for registering high-interest chemical inventories
(ToxCast™, ToxRef, Tox21, DSSTox published data files), utilizing DSSTox chemical
information quality review and structure-annotation within ACToR
4. EPA Centers and Offices (OPPT/OPP/NCEA/OW) - provide data on chemicals of
interest
Milestones/Products:
FY09
1. Initial public deployment.
2. Significant version 2, including refined chemical structure information.
3. Develop workflow for tabularization of data buried in text reports.
4. Integrate all ToxCast™ and ToxRefDB data.
5. Quarterly releases with new data.
A - 1 . a. 2
Previous I TOC
-------
EPA CompTox Research Program FY2009-2012 BOSC Review Draft- 24 August, 2009
FY10
1. Quarterly releases with new data.
2. Implementation of a process to gather tabular data on priority chemicals from text
reports.
3. Survey sources of chemical use and exposure data and import any remaining sources.
4. Develop flexible query interface and data download process.
5. Develop process to extract data from open literature.
FY11
1. Quarterly releases with new data
FY12
1. Quarterly releases with new data
Keywords:
Computational Toxicology; ACToR; ToxCast™; DSSTox; Database
QA Project Plan: Category II. ACToR development is guided by a series of Standard
Operating Procedures. These govern all aspects of the project, including data acquisition,
formatting, quality assurance, database filling and maintenance, and system administration. All
QA plans are archived in the EPA internal QA system.
A-l.a.3
Previous I TOC
-------
b. DSSTox- Chemical Information Technologies in Support of Toxicology Modeling
Lead/Principal Investigator: Ann M. Richard
Research Issue/Relevance: A central regulatory mandate of the EP is to assess the potential
health and environmental risks of large numbers of chemicals released into the environment,
often in the absence of relevant test data. Significant advances in toxicity prediction capabilities
are predicated on the ability to store, mine, and analyze information on many levels in relation to
chemicals and their effects on biological systems. Standardized, high quality chemical structure
annotation and searchability of toxicity-related information across the Internet and within EPA
programs is a crucial requirement for creating effective data linkages and gathering relevant data.
Equally important is the need to incorporate meaningful chemical structure and property
representations, based on principles of organic chemistry and biologically informed measures of
chemical similarity, into toxicity modeling efforts. Finally, successful modeling efforts will
depend upon suitable representations of biological activity, both HTS and in vivo, in relation to
chemical structure. NCCT's ToxCast™ and Tox21 projects are employing high-throughput
screening tests to probe biochemical target interactions, chemical pathways, and cellular
responses potentially relevant to toxicity for thousands of chemicals of high potential exposure
and environmental interest. The goal is to use these data, in conjunction with legacy data and
chemical structure considerations, to infer meaningful patterns and to develop models to predict
a range of in vivo bioassay responses. Biologically informed toxicity prediction models that
incorporate chemical structure-activity considerations are likely to provide the best means for
prioritizing large lists of chemicals for potential hazard - a pressing need for many EPA
programs - and charting a path forward for a more efficient and cost-effective screening and
testing paradigm.
Purpose/Objective/Impact: The current research will use the DSSTox project framework to
incorporate strict quality standards for chemical information across NCCT projects (ToxCast™,
ACToR, Tox21), and expand comparability and linkages of summary toxicity data in the context
of a standardized cheminformatics environments. This project will also use these data
foundations to promote and explore new ways to associate chemical structure with biological
activity, extending traditional structure-activity relationships (SAR) towards new paradigms for
biologically informed structure-based toxicity prediction. These efforts have the potential to
impact a wide variety of EPA program offices that heavily rely on chemical information
resources and have a need for structure-based data exploration, analog searching, and improved
toxicity prediction models when limited test data are available. These include programs within
OPPTS [e.g., Green Chemistry, PreManufacture-Notification Program (PMN), Office of
Pesticide Programs (OPP), HPV Testing Program], as well as EPA's IRIS Program, Office of
Water, and Office of Environmental Information. New information technologies that incorporate
strict quality standards and more flexible and diverse means for assessing biological and
chemical similarity will also improve the identification of lexicologically relevant analogs by
enhancing the ability to explore data and quantify associations across diverse chemical and
biological data domains.
Synopsis: The DSSTox project has implemented high quality data review procedures for
standardized chemical structure annotation, created linkages connecting diverse toxicity
A-l.b.l
Previous I TOC
-------
EPA CompTox Research Program FY2009-2012 BOSC Review Draft- 24 August, 2009
resources within and outside of EPA, and published high quality EPA chemical inventories and
toxicity data files spanning over 10,000 substances. The broad utility of DSSTox data files for
cheminformatics and modeling applications provides significant opportunities to influence the
course of predictive modeling strategies and to encourage wider engagement of toxicologists in
toxicity data representation. The DSSTox project will continue efforts to expand chemical data
file offerings into less well-represented areas of toxicology (immunotoxicology, toxicogenomics,
etc.), and provide varied representations of summary toxicity endpoints. In addition, this research
will explore new representations of chemical structure in relation to the biology (e.g., analog
measures, chemical features, chemical classes), and new representations of biological endpoints
in relation to modeling (e.g., quantitative endpoints in terms of potency, summarized or grouped
effects, qualitative active and inactive classes). These efforts will be designed to complement and
augment projects in NCCT (ToxCast™, ACToR, and Tox21) that are working to improve
capabilities to access, mine, and integrate chemical-biological activity information from existing
and new data, both within and outside EPA, in support of toxicity prediction efforts. In close
coordination with ACToR, which is the primary informatics resource for ToxCast™ and Tox21
chemical and biological data, DSSTox will provide the initial chemical registration IDs, structure
annotation, and quality review of ToxCast™ and ToxRef inventories, as well as the expanded
Tox21 chemical testing library, helping to ensure quality and consistency of chemical
information across the various NCCT programs.
Partnerships/Collaborations (Internal & External): The DSSTox project is being
coordinated and linked with a number of public efforts (ILSI, ToxML, LHASA UK, PubChem,
ChemSpider), and government research laboratories (NIEHS, NTP, FDA) that are promoting
controlled toxicity vocabularies, adopting data standards, and migrating diverse toxicity data into
the public domain. DSSTox is also aligned with major NCCT projects (ToxCast™, ToxRefDB,
ACToR, Tox21), providing key quality review procedures and cheminformatics support,
expanding DSSTox data file publications of toxicological data in support of predictive modeling,
and enhancing linkages to public resources such as PubChem for disseminating bioassay results
to the broader modeling community. In coordination with ACToR, partnerships and
collaborations with scientists across EPA (NHEERL, OPPT, OPP, NERL) are being forged to
improve cheminformatic capabilities across the Agency from a unified chemical structure
perspective, most recently extending into exposure data arenas (ExpoCast™). Research
collaborations are on-going with SAR modelers at UNC (A. Tropsha, H. Zhu) and in the data
mining and SAR community (C. Yang, R. Benigni, E. Benfenati) to improve methods to
incorporate biological considerations into SAR models. OECD is using DSSTox as a source of
high quality, quality controlled chemical structures and activities in their QSAR Toolbox.
Finally, we are pursuing closer collaborations with toxicogenomics resources such as the
National Center for Biotechnology Information (GEO), the European Bioinformatics Institute
(ChEBI and ArrayExpress), and the NIEHS CEBS toxicogenomics resource.
Milestones/Products:
FY09
1. Publish paper on ToxCast™ 320 chemical inventory from SAR modeling perspective.
2. Publish papers and coordinate efforts with NCBI (GEO) and EBI (ArrayExpress) to
structure-annotate and provide chemical linkages to microarray data for toxicogenomics.
3. Restart Chemoinformatics Communities of Practice using EPA's Science Portal.
A-l.b.2
Previous I TOC
-------
EPA CompTox Research Program FY2009-2012 BOSC Review Draft- 24 August, 2009
4. Publish DSSTox files for ToxRef and ToxCast™ inventories and selected summary
endpoints, and facilitate publication and linkage to the NLM PubChem project. Compile
and publish public genetic toxicity data and SAR predictions for ToxCast 320.
5. Continue expansion of DSSTox public toxicity database inventory for use in modeling.
6. Perform primary chemical review and structure annotation of the ToxCast™/Tox21
chemical testing libraries, coordinating with ACToR and within a central chemical
registry.
Publish DSSTox files for Tox21 inventory and selected summary endpoints, and facilitate
publication and linkage to the NLM PubChem project.
2. Publish DSSTox files for NTP study areas (Immunotox, Genetox, etc) to facilitate
incorporation into ACToR and encourage broader SAR examination;Explore new
approaches to SAR modeling based on feature categories within existing DSSTox files
and ToxCast™ data.
3. Explore new approaches to SAR modeling based on classifiers and feature categories.
4. Expand CEBS collaboration to incorporate DSSTox GEO and ArrayExpress files, create
chemical linkage to ILSI Developmental Toxicity database and facilitate structure-
searching.
5. Advise and assist efforts within ExpoCast™ to identify and chemically annotate
important exposure-related public data resources.
In collaboration with ACToR, establish procedures and protocols for automating
chemical annotation of new experimental data generated by NCCT Programs
(ToxCast™, Tox21) and in collaboration with CEBS or NHEERL.
2. Document and employ PubChem analysis tools in relation to published DSSTox and
ToxCast™ data inventory in PubChem.
3. Collaborate with SAR modeling efforts to predict ToxCast™ endpoints using in vitro
data.
4. Continue expansion of DSSTox public toxicity database inventory for use in modeling
with co-publication and linkage to ACToR and PubChem.
FY12
1. Redesign DSSTox website to provide hosting of donated chemical descriptors, properties
and predictions for high interest inventories.
2. Publish master tables of DSSTox IDs and high quality structures to serve as public data
registry for toxicology, particularly for EPA, FDA, and NTP datasets.
3. Promote use of chemical registry system from ToxCast™/Tox21, linked to DSSTox
content and integrated into ACToR, more broadly within EPA.
4. Collaborate with SAR modeling efforts to expand modeling to address Tox21 chemicals
and endpoints.
5. Continue expansion of DSSTox public toxicity database inventory into toxicity and
exposure areas not effectively linked to current databases.
Keywords: DSSTox; prediction; cheminformatics; structure-activity; SAR
A-l.b.3
Previous I TOC
-------
EPA CompTox Research Program FY2009-2012 BOSC Review Draft- 24 August, 2009
QA Project Plan: Category II. The DSSTox project, involving the compilation,
standardization, and review of chemical data largely extracted from secondary public data
sources has a large intrinsic QA component. All QA plans are archived in the EPA internal QA
system. Documentation and procedures contributing to overall QA objectives include:
> Maintenance of a DSSTox web page describing QA procedures for obtaining and
reviewing chemical information prior to inclusion in the DSSTox Master file:
http://www.epa.gov/ncct/dsstox/ChemicalInfQAProcedures.html
> File versioning, error tracking within data files, publication of Log files with version
history, and full documentation of published DSSTox data files
> Maintenance of the DSSTox Master File tables in ACCESS, cross referencing, and use of
automated scripts to perform field content error checks for new file generation;
> Coordinated publication in ACToR and PubChem checks for structure consistencies;
> Review of all newly published DSSTox data files by Source collaborators;
> Error reporting system associated with DSSTox published files and structure browser.
A-l.b.4
Previous I TOC
-------
c. ToxRefDB - Toxicity Reference Database
Lead/Principal Investigator: Matthew Martin
Version: April 02, 2009
Research Issue/Relevance: As the EPA moves toward a new chemical toxicity testing
paradigm, the vast library of laboratory animal toxicity study information that is publicly
available, and for which the agency has received and continues to receive, will provide context
for many of the technologies applied recently to toxicity testing and screening. However, the
animal toxicity study information has not been made electronically accessible, searchable, or
computable. ToxRefDB is the relational database designed and developed to electronically
capture all of the relevant information spanning thirty years of health effect data from the agency
and beyond. The EPA ToxCast™ program is one effort using high-throughput techniques to
prioritize chemicals for further testing. To be effective, ToxCast™ needs reference toxicity
information to provide the interpretive context for the large amounts of screening data.
ToxRefDB is providing the reference in vivo toxicity data for programs such as ToxCast™ in a
searchable and computable format.
Additionally, the utility of ToxRefDB is broader than use as a validation dataset for ToxCast™.
Regulatory scientists have begun to assess the role and impact of previous and current guideline
studies and components of those studies in the regulation and assessment of chemicals.
ToxRefDB will be the primary data source for numerous retrospective analyses and may have a
large impact on the future revisions to existing guideline studies.
Purpose/Objective/Impact: The ToxRefDB project has provided access to a wealth of in vivo
toxicity data in a structured and searchable format. These data are being released through a series
of manuscripts which are currently submitted for publication or in preparation. In addition,
ToxRefDB data will be publicly available through the ToxRefDB website. This will fill a major
gap in the environmental toxicology community as very limited resources in this area exist in the
public domain. Such information should have high utility in building and interpretation of
predictive toxicology models. In addition, researchers and regulatory scientists can access the
toxicity information to address numerous questions specific to hazard identification and
characterization of environmental chemicals along with retrospective analyses that will direct
and evaluate possible changes to guideline toxicity studies.
Synopsis: Thirty years of registration toxicity data and open literature studies have been
historically stored as hardcopy and scanned documents by the EPA and others. A significant
portion of these data, including chronic, cancer, developmental, and reproductive studies from
laboratory animals, have now been processed into standardized and structured toxicity data,
within the EPA's Toxicity Reference Database (ToxRefDB). These data are now accessible and
mineable within ToxRefDB and are serving as a primary source of validation for EPA's
ToxCast™ research program in predictive toxicology. In addition to providing reference toxicity
information to research efforts, ToxRefDB will be mined for information on the role and impact
of previous and current study guidelines on the regulation of environmental chemicals. The
initial collection of studies in March of 2006 focused primarily on the reviews of registrant
A-l.c.l
Previous I TOC
-------
EPA CompTox Research Program FY2009-2012 BOSC Review Draft- 24 August, 2009
submitted toxicity studies on pesticide active ingredients. ToxRefDB design, development and
implementation were completed in mid-FY2006 with ongoing updates to the standardized
vocabulary and data entry tool interface. The entry of over 2,000 studies spanning the majority of
the ToxCast™ Phase I chemical set was completed late FY2008. The statuses of these initial
datasets are either published, in press, or submitted and are being made publicly available
through the ToxRefDB website. Over the next year, a web-based query tool for the entire
contents of ToxRefDB will become available to the public and this will be performed in
conjunction with a quarterly update of the EPA ACToR program. ToxRefDB will continue to
enter available data from chronic, cancer, developmental, and reproductive studies with a focus
on potential ToxCast™ Phase II chemicals. The availability and entry of toxicity data into
ToxRefDB will also guide the selection of ToxCast™ Phase II chemicals. Over the next year,
ToxRefDB will also be expanded to capture developmental neurotoxicity (DNT) study data and
possibly in vivo data submitted through the agency as part of the endocrine disrupter screening
program (EDSP). The completion of retrospective analyses on reproductive toxicity studies will
be completed by late FY2009 and the use of ToxRefDB for additional analyses including rat and
mouse chronic/cancer study and rat and rabbit development study assessments.
Partnerships/Collaborations (Internal & External):
1. EPA ToxCast™ program - provide the reference toxicity information for interpreting the
screening data with respect to animal toxicity information
2. Tox21 partnership - provide the reference toxicity information for interpreting the
screening data with respect to animal toxicity information
3. EPA ACToR program - provide ACToR with ToxRefDB data for chemical indexing and
searchability
4. EPA Centers, Offices, and Labs (OPPT/QPP/OSCP/NHEERL) - offices provide legacy
toxicity data; provide searchable and computable toxicity data to offices; entry of
additional study types and from various sources
5. OECD (including BfR, RIVM, PMRA) - evaluation of current testing guidelines and
assessment of proposed new guidelines, e.g., extended one-generation reproduction
toxicity study
Milestones/Products:
FY09
1. Publication on ToxRefDB
2. Release of stand-along ToxRefDB data entry tool
3. ToxRefDB webpage online
4. Initial public release of selected chronic/cancer endpoints
5. Public release of selected reproductive toxicity endpoints
6. Public release of selected developmental toxicity endpoints
7. Collection of ToxCast™ Phase II chemical toxicity data
8. Public release of ToxRefDB web-based query tool
Tl\/f
9. Complete entry of targeted set of chemicals and study types for Phase II of ToxCast
10. Complete reproductive toxicity study retrospective analysis
A-l.c.2
Previous I TOC
-------
EPA CompTox Research Program FY2009-2012 BOSC Review Draft- 24 August, 2009
Quarterly releases with new data in conjunction with ACToR
Implementation of a process to gather and enter open literature studies
Expansion of ToxRefDB to capture DNT studies and EDSP data
Complete retrospective analyses on other major study types
Release of ToxRefDB live data entry tool
1. Quarterly releases with new data in conjunction with ACToR
FY12
1. Quarterly releases with new data in conjunction with ACToR
Keywords (three to five): Computational Toxicology; ToxRefDB; ToxCast™; ACToR;
Database
QA Project Plan: Category II. ToxRefDB development is guided by a series of Standard
Operating Procedures. These govern all aspects of the project, including data acquisition,
formatting, quality control and assurance, data entry and maintenance, and system
administration. All QA plans are archived in the EPA internal QA system.
A-l.c.3
Previous I TOC
-------
d. ChemModel- The Application of Molecular Modeling to Assessing Chemical Toxicity
Lead/Principal Investigator: James Rabinowitz
Research Issue/Relevance: Insufficient experimental information exists for the evaluation of
the potential of a large number of environmental chemicals to cause toxicity and other
environmental effects. Where data does exist often it is not ideal for this task. The Agency often
must make decisions about specific chemicals when lacking an ideal data set. Molecular
modeling approaches provide an approach for estimating relevant missing information. One
approach to this problem is to estimate the relevant missing information by extrapolation from
existing information on the chemical of interest and other similar chemicals making use of
molecular modeling approaches. Knowledge of the mechanisms of toxicity provides a rational
basis for application of these computational tools. The results of these models may be used in
conjunction with experimental information to inform decisions about the relevant chemical and
to prioritize the requirements for obtaining missing experimental data.
Purpose/Objective/Impact: The overall objective of this research is to develop an approach
(including the necessary tools) for the application of molecular modeling methods to Agency
problems, particularly problems resulting from the requirement to make preliminary decisions
about chemicals in a data poor environment. This includes the preliminary evaluation of
chemical toxicity and the prioritization of chemicals testing needs. Knowledge of the potential
mechanisms of toxicity provides a rational basis for extrapolation from existing information to
derive information about chemicals for which little data exists. The differential step in many
mechanisms of toxicity may be generalized as the interaction between a small molecule (a
potential toxicant) and one or more macromolecular targets. (The small molecule may be the
chemical itself or one of its descendants). Using modern molecular modeling methods developed
for the discovery of novel pharmaceutical agents, it is possible to computationally predict these
toxicant-biomolecular target interactions using a combination of direct computer modeling of
atomic interactions between the toxicant-target pair and correction factors derived from
experimentally-derived interactions with similar targets. To employ this approach, a library of
computational models of relevant biomolecular targets is being developed. Molecular modeling
approaches may then be used to interrogate this library for the capacity of specific environmental
molecules to interact with each target. These approaches were developed for the discovery of
new pharmaceuticals where the objective is to discover molecules that interact most potently
with the target. However, the Agency's need is to discover if chemicals of environmental interest
interact with the target, even if their interaction is much weaker than seen with potential
pharmaceutical agents. An objective of this research is to evaluate the relevant molecular
modeling methods in relationship this Agency requirement.
The endocrine system provides a test for the utility of this approach because many of the
pathways for toxicity and the macromolecular targets in those pathways have been identified.
Appropriate experimental crystal structures of many of the receptor protein targets are available
to create the computational library of targets. Additionally, experimental data of the capacity of a
library of chemicals to displace the natural ligand from the rat estrogen receptor is available from
a single source. While most of the chemicals in this library have been show not to interact with
the receptor in this laboratory assay, the few that displace estrogen are orders of magnitude less
A-l.d.l
Previous I TOC
-------
EPA CompTox Research Program FY2009-2012 BOSC Review Draft- 24 August, 2009
potent than the natural ligand. The data from this chemical library provides an opportunity to test
the toxicant-target approach and its capacity to separate less potent chemicals from a large
number of similar inactive chemicals. In addition this exploration addresses the specific Agency
need to evaluate the potential of chemicals to disrupt the endocrine system. A similar approach
may be used to investigate other health effects and fate and metabolism.
The ultimate objective of this research is to develop a library of biomolecular targets for
chemical toxicity, and the methods appropriate for their application to predicting the capacity of
a chemical to interact with these targets. This library of targets may then be used in conjunction
with other approaches as part of a chemical prescreen.
Synopsis: The differential step in many mechanisms of chemical toxicity may be generalized as
the interaction between a small molecule (a potential toxicant) and one or more macromolecular
targets. The small molecule may be the chemical itself or one of its descendents. Describing the
potential of a molecule to participate in interactions of this type is a source of insight chemical
toxicity. In this project a series of molecular models (148) for critical toxicity targets is being
developed and methods to evaluate the capacity of a small molecule to interact with these targets
assessed. These methods are adopted from those used in the design of novel pharmaceutical. A
study of a library of 280 environmental chemicals interacting with the estrogen receptor target is
in the final stages of completion. In this library 14 of the chemicals are weakly active (3 -5
orders of magnitude less active than estrogen) and the others are inactive. Modeling the potential
interaction of these chemicals with the rat estrogen receptor provides an ordered list of
molecules. The best results are achieved using a pharmacophore filter. With that approach all
14 active chemicals are identified in the first 22 chemicals. In addition to the importance of
these results relative to potential binding to the environmentally important estrogen receptor,
they indicate that this approach may be used to find chemicals that interact weakly with the
target. All 150 of the targets have been interrogated with the ToxCast™ chemicals. Based on
the results from the estrogen receptor study, pharmacophores for as many of the targets as
possible will be developed. The analysis of this data will proceed by comparison with specific
ToxCast™ endpoint data and in concert with short term data to evaluate more complex
biological endpoints. The logical extension is to consider the androgen receptor, where relevant
data for comparison and developing a pharmacophore are available. When sufficient data on
other biological macromolecules that are relevant to the Agency requirements become available
the current library of targets will be expanded. This library of targets will be used to study
chemicals and families of chemicals of importance to the Agency. The Toxicant-Target
approach described above models molecular identification processes. In collaboration with other
EPA scientist similar approaches are being applied to the steps that follow identification. A
study on the differential metabolism of pyrethroids and the effect of stereo-structure on
biological clearance is underway as is a study of perfluorinated chemicals.
Partnerships/Collaborations (Internal & External): Scientists from NHEERL/RTD have
provided the database of the interactions of molecules with the estrogen receptor. We continue to
interact with them relative to this data and the biological details of the computational modeling
effort. Scientists from NERL/HEASD are collaborators on the study of the metabolism and fate
pyrethroids. Collaboration with CDC scientists relative to using the target-toxicant approach to
investigate the interaction of environmental chemicals with nervous system enzymes and
A-l.d.2
Previous I TOC
-------
EPA CompTox Research Program FY2009-2012 BOSC Review Draft- 24 August, 2009
receptors is being developed. As described above these studies apply methods developed for
pharmaceutical discovery to model the capacity of an environmental chemical to interact with a
macromolecular target.
Milestones/Products:
FY09
1. Report on the capability of the target-toxicant paradigm to identify chemicals that bind
weakly to the estrogen receptor, including a description of the method.
2. Description of the library of 148 biological macromolecule targets.
3. Report on molecular modeling studies of the potential biological effects of the perfluoro
compounds.
FY10
1. Report on the metabolism of pyrethroids and the effects of three dimensional chemical
structures.
2. Description of additional targets added to the target library.
3. Report on the interaction of ToxCast™ chemicals with nuclear receptor targets and the
importance of pharmacophore filters.
FY11
1. Report on the integration of results from the target library and available experimental
parameters.
2. Report on the comparison of results with and without pharmacophore filter for
ToxCast™ chemicals.
FY12
1. Comparison of results using the target library with experimental determined activities
particularly when observed at the molecular level.
2. Report of the potential use of molecular modeling and the target toxicant paradigm for
regulatory purposes, including a discussion of the OECD principles either as they
currently exist or relative to molecular modeling specific principles.
Keywords:
Molecular modeling; Protein binding; toxicity prescreening; weak interactions
QA Project Plan: Category IV. The quality objectives for molecular modeling are to achieve
the best balance between reasonable computational speed and model performance. Development
is guided by a series of Standard Operating Procedures. These govern all aspects of the project,
including data acquisition, formatting, quality assurance, database filling and maintenance, and
system administration. All QA plans are archived in the EPA internal QA system.
A-l.d.3
Previous I TOC
-------
e. ToxCast™- Screening and Prioritization of Environmental Chemicals Based on
Bioactivity Profiling and Predictions of Toxicity
Lead/Principal Investigator: Keith Houck
Research Issue/Relevance: The objective of the ToxCast™ research program developed by
The NCCT of the EPA's ORD is to develop cost-effective innovative approaches to efficiently
screen and prioritize a large number of chemicals for toxicological testing. Using data from state-
of-the-art high-throughput screening (HTS) bioassays developed by the pharmaceutical industry,
ToxCast™ is building computational models to predict the potential human toxicity of
chemicals. These hazard predictions should provide the Agency's regulatory programs with
science-based information that will be helpful in setting priorities for more targeted toxicological
evaluations that will help the Agency focus on those chemicals and endpoints with the greatest
potential for causing adverse effects in humans. The ultimate goal of ToxCast™ is to deliver an
affordable, efficient, science-based system for categorizing chemicals according to their
predicted toxicities.
An essential component of the ToxCast™ research program is the development of a
standardized, reference database containing animal toxicity studies called ToxRefDB.
ToxRefDB is being populated with the results of guideline animal toxicity studies on pesticidal
active chemicals that are submitted to the Agency by manufacturers as a requirement of licensing
a pesticide product. ToxRefDB is, for the first time, providing a searchable, mineable historical
database for accessing a wealth of reference in vivo study data. Most importantly, ToxRefDB
will provide the essential interpretive context to anchor ToxCast™ in vitro data (i.e.,HTS and
genomic data) to animal toxicity endpoints with selected ToxRef in vivo outcomes serving as the
basis for developing predictive in vitro bioactivity profiles and signatures. Equally essential to
the overall success of this project will be the development of a suitable informatics and analysis
infrastructure for storing, relating and extracting patterns from all data associated with the
ToxCast™ project, including chemical, HTS, and in vivo data elements.
ToxCast™ databases and predictive models for the potential toxicity of environmental chemicals
will be useful to EPA program offices for chemical prioritization. For example, the Office of
Pesticide Programs (OPP) and the Office of Pollution Prevention and Toxics (OPPT) anticipates
taking advantage of ToxCast™ models and datasets to prioritize in vivo animal testing of
products that have limited toxicity data available such as:
• antimicrobial pesticides
• inert ingredients in pesticide products
• manufacturing process impurities
• metabolites and environmental degradates of concern
• new and existing industrial chemicals
After internal clearance and external peer review, the information in ToxRefDB and HTS data
generated on the chemicals screened in ToxCast™ will be publicly available at
www.epa.gov/ncct/toxrefdb and www.epa.gov/ncct/toxcast.
A-l.e.l
Previous I TOC
-------
EPA CompTox Research Program FY2009-2012 BOSC Review Draft- 24 August, 2009
Purpose/Objective/Impact: The objective of the ToxCast™ research program is to develop a
cost-effective and rapid approach for screening and prioritizing a large number of chemicals for
toxicological testing (Dix etal., 2007). Using data from HTS bioassays developed in the drug
discovery community, ToxCast™ is generating data, constructing databases, and building
computational models to forecast the potential human toxicity of chemicals. These hazard
predictions should provide EPA regulatory programs, including OPP, with science-based
information helpful in prioritizing chemicals for more detailed toxicological evaluations,
ultimately leading to reduced animal testing. ToxCast™ is currently in the proof-of concept
phase, wherein over 300 chemicals have been assayed in over 600 different HTS bioassays,
creating rich bioactivity profiles for these chemicals. The Phase I chemicals are primarily
conventional pesticide actives that have been extensively evaluated using traditional mammalian
toxicity testing, and hence have a number of well characterized toxicity outcomes (e.g.,
carcinogenicity; and developmental, reproductive and neural toxicity). These in vivo data, in
turn, have been extracted from the evaluations conducted by OPP scientists and were used to
construct and populate the ToxRefDB. Comparable toxicity data from other toxicity sources
(e.g., National Toxicology Program) are also being captured in ToxRefDB. A broader and more
diverse set of complementary data on thousands of chemicals is being identified and collated in
EPA's Aggregated Computational Toxicology Resource (ACToR). ACToR (and the analysis
component, ToxMiner) is providing the essential informatics infrastructure for housing,
integrating and analyzing all chemical and assay data associated the ToxCast™ project, also in
the context of a much larger world of web-accessible chemical-toxicological information.
DSSTox, in turn, is providing high quality, standardized chemical structure indexing for ACToR
and ToxCast™, including other high-interest Agency chemical-data inventories.
ToxRefDB is critical to developing predictive signatures, because it links ToxCast™ HTS in
vitro data to in vivo toxicity endpoints associated with the same chemicals. The toxicity data in
ToxRefDB and the HTS data generated in ToxCast™ will be made publicly available through
EPA websites and databases. The first manuscript on ToxRefDB was recently published (Martin
et al, 2008), presenting toxicity profiles from two-year rodent bioassays on 310 chemicals. A
similar analysis is nearing completion on multigeneration reproduction and prenatal
developmental test data for the ToxCast™ chemicals in ToxRefDB, profiling the toxicity
potential of this chemical set across generation, life-stage, and different classes of endpoints.
ToxCast™ is a multi-year effort that is divided into three distinct phases:
7. Phase I: 300 chemicals assayed in over 600 different HTS bioassays, to create predictive
bioactivity signatures based on the known toxicity of the chemicals;
8. Phase II: focused on confirmation and expansion of ToxCast™ predictive signatures,
generating HTS data at least 300 additional chemicals;
TA/f
9. Phase III: ToxCast expanded to thousands of environmental chemicals for which little
toxicological information is available.
Once ToxCast™ has gone through successful initiation of Phase III, the data and toxicity
predictions will be ready for deployment throughout numerous EPA program offices. NCCT will
work to link these hazard predictions with exposure predictions, and create integrated database
analysis tools facilitating customized chemical prioritizations appropriate to specific programs.
A-l.e.2
Previous I TOC
-------
EPA CompTox Research Program FY2009-2012 BOSC Review Draft- 24 August, 2009
Synopsis: The primary goals for ToxCast™ are the completion of Phase I data collection and,
concurrently, development of the informatics infrastructure to store and analyze these data,
derivation of predictive signatures from Phase I data, validation of these signatures with Phase II
data, and the application of these predictions to the prioritization of chemicals in various
chemical and nanomaterial testing programs. Success in meeting these four principal goals will
lead to secondary applications in developing human toxicity pathway analyses, and in high
throughput risk assessments.
Partnerships/Collaborations (Internal & External):
Tox21 with NTP and NCGC
OECD Molecular Screening Project
OECD Enhanced One Generation Reproductive Test Guideline
EPA/ORD/NHEERL DNT Team
EPA/ORD/NCEA Phthalate Team
MOUs and MTAs with over 20 external organizations collaborating on ToxCast™ assays,
chemicals and data analysis.
Milestones/Products:
FY09
1. Completion of ToxCast™ Phase I data collection.
2. Provide annotated Phase I data sets for public access.
3. Derivation of predictive signatures from ToxCast™ Phase I data.
Tl\/f
4. Multiple publications on ToxCast data sets.
5. Publication on signature generation (SCOUT item, April 2009).
6. Publication on analysis of NR pathways and toxicity.
7. Convene first ToxCast™ Data Summit for identifying promising prediction models from
intramural and extramural sources.
8. Finalize selection of chemicals for next 3-5 years of ToxCast™ and Tox21 projects.
9. Prioritize and order 4000-6000 chemicals in collaboration with other Tox21 partners.
FY10
1. Publications describing approaches to combining exposure, PK and in vitro assays to do
risk prioritization.
2. Evaluate compatibility of nanomaterials of diverse classes with ToxCast™ assays.
3. Prioritize and select assays to be run for ToxCast™ Phase II.
4. Completion of ToxCast1 Phase II data collection.
FY11
1. Confirmation of ToxCast™ predictive signatures with Phase II data.
2. Publications on signature confirmations and applications.
FY12
1. Application of ToxCast™ predictions to the prioritization of chemicals in various EPA
chemical and nanomaterial testing programs.
A-l.e.3
Previous I TOC
-------
EPA CompTox Research Program FY2009-2012
BOSC Review Draft- 24 August, 2009
Keywords (three to five): ToxCast™; high-throughput screening; hazard; predictive
toxicology; chemical prioritization
QA Project Plan: Category I. ToxCast™Quality Management Plan, includes: Information
Management QA Project Plan, NCGC QA Project Plan (for IA), and 9 separate Contractor QA
Plans and Records. All QA plans are archived in the EPA internal QA system.
Previous
A-l.e.4
TOC
-------
f. ExpoCast™- Exposure science for screening, prioritization, and toxicity testing.
Lead/Principal Investigator: Elaine Cohen Hubal
Research Issue/Relevance: High visibility efforts in toxicity testing and computational
toxicology raise important research questions and opportunities for exposure scientists. There is
a clear need for a collaborative effort across the exposure and risk assessment community to
ensure that the required exposure science, data and tools are ready to address immediate needs
resulting from application of high-throughput in vitro technologies for toxicity testing. A
coherent program is required to formulate significant exposure questions posed by these novel in
vitro toxicity data, develop creative approaches for applying existing exposure information and
tools to address these questions, and finally identify key exposure research needs to interpret the
toxicity data for risk assessment. The authors of the National Academies report (NRC, 2007)
emphasize that population-based data and human exposure information are required at each step
of their vision for toxicity testing, and that these exposure data will continue to play a critical
role in both guiding the development and use of the toxicity information. Exposure research
questions posed in this report include how to: (1) use information on host susceptibility and real-
world exposures to interpret and extrapolate in vitro test results; (2) use human exposure data to
select doses for toxicity testing so information on biological effects pertains to environmentally-
relevant exposures; and (3) relate human exposure data from biomonitoring surveys to
concentrations that perturb toxicity pathways to identify biologically-relevant exposures. The
NCCT has identified the need to include exposure information for chemical prioritization,
modeling system response to chemical exposures across multiple levels of biological
organization (through to the population level), and linking information on potential toxicity of
environmental contaminants to real-world health outcomes (Cohen Hubal et al., 2008).
Together, scientists from NCCT and NERL's Human Exposure Research Program have
identified and will be conducting research in the following priority exposure research areas to
support chemical screening, prioritization, and toxicity testing (Sheldon and Cohen Hubal,
2009): (1) accessible and linkable exposure databases; (2) exposure screening tools for
accelerated chemical prioritization; (3) computational tools for dose reconstruction and source-
to-outcome analyses; (4) tools for understanding the fundamental processes and factors
influencing human exposures; and (5) efficient monitoring methods to measure and interpret
biologically-relevant exposure metrics. Susceptibility, vulnerability, and life-stage aspects are
integral to each of these.
Purpose/Objective/Impact: The ExpoCast™ program is being initiated in FY09 to ensure the
required exposure science and computational tools are ready to address global needs for rapid
characterization of exposure potential arising from the manufacture and use of tens of thousands
of chemicals and to meet challenges posed by new toxicity testing approaches. The overall goal
of this project is to develop novel approaches and tools for screening, evaluating and classifying
chemicals, based on the potential for biologically-relevant human exposure, to inform
prioritization and toxicity testing. An emphasis will be placed on conducting research to mine
and translate scientific advances and tools in a broad range of fields to provide information that
can be used to support enhanced exposure assessments for decision making and improved
environmental health. Advanced exposure databases, computational tools, and analysis
approaches are required to prioritize chemicals, to design effective in vitro screening protocols,
A-l.f.l
Previous I TOC
-------
EPA CompTox Research Program FY2009-2012 BOSC Review Draft- 24 August, 2009
and to interpret the results of these screening tests for human health risk assessment. Approaches
for integrating information on genetic susceptibility, life-stage, and population-level
vulnerabilities with in vitro toxicity data are also required to improve public-health decision
making. This initiative will advance Agency tools for efficiently characterizing and classifying
chemicals based on potential for biologically-relevant exposures. The improved exposure science
and knowledge will subsequently inform the characterization of environmentally-relevant
toxicity.
Synopsis of NCCT Directed Research: ExpoCast™ will provide an overarching framework
for the science required to characterize biologically-relevant exposure in support of the Agency
computational toxicology program. Broadly and long-term, the ExpoCast™ program will foster
novel exposure science research to; (1) inform chemical prioritization, (2) understanding the
systems response to chemical perturbations resulting from environmentally relevant exposures
and how these translate to relevant biological changes at the individual and population levels, (3)
link information on potential toxicity of environmental contaminants to real-world health
outcomes. The NCCT directed aspects of the ExpoCast™ program will have a strong focus on
research required to interpret and translate in vitro hazard data in the context of real-world
exposures for risk assessment. Research will be conducted jointly with NERL to leverage
expertise and resources required to meet objectives of this multidisciplinary project.
An important early component of ExpoCast™ will be to consider how best to consolidate and
link human exposure and exposure factor data for chemical prioritization and toxicity testing.
Under the ExpoCast™ program NCCT and NERL scientists will collaboratively:
• Evaluate and recommend approaches for improving accessibility to EPA human exposure
and exposure factor data and for facilitating links between exposure and toxicity data
(e.g., through DSSTox and ACToR systems);
• Advocate for the creation of a consolidated EPA exposure database focused on measured
and predicted concentrations in exposure and biological media;
• Propose standards for human exposure data representation.
Early research activities will focus on identifying and evaluating novel approaches for
characterizing exposure to prioritize chemicals and developing modeling approaches for
considering exposure potential based on chemical properties, sources (e.g., consumer products),
uses, lifecycle, and individual/population vulnerability. Specific tasks pertinent to these goals
include:
• Analysis of extant exposure data to identify the critical metrics and develop simple
indices for representing biologically-relevant personal exposure over time, place,
lifestage, and lifestyle or behavior.
• Development of novel approaches for characterizing biologically-relevant exposure to
prioritize chemicals, some examples:
o Application of residential models for prioritizing SVOCs
o Development of dermal uptake model suite.
o Application of biomonitoring equivalent (BE) approach to interpret ToxCast™
data.
• Development of human exposure knowledgebase.
• Application of genomic tools and other biomarkers of exposure and susceptibility to
consider population-level vulnerabilities for toxicity testing and risk assessment.
A-l.f.2
Previous I TOC
-------
EPA CompTox Research Program FY2009-2012 BOSC Review Draft- 24 August, 2009
NCCT Partnerships/Collaborations (Internal & External): The ExpoCast™ project will be
conducted in close collaboration with the ToxCast™ program and through extensive
collaboration with NERL principle investigators in the Human Exposure Research Program.
Integration of research directed by the both NCCT and NERL will be enhanced by activities such
as the Computational Toxicology Rotational Fellow Program. To jumpstart joint research
activities, a NERL investigator will join the rotational fellows program in the summer of 2009 to
focus on improving access to human exposure data. Partnerships with the other labs and centers
will also be developed as the ExpoCast M program advances. For example, collaboration with
NHEERL on the Mechanistic Indicators of Childhood Asthma (MICA) study is providing the
opportunity to pilot advanced computational approaches for evaluating multi-factorial biomarker
data (including genomic data) across the exposure-outcome continuum to investigate the
interplay of environmental and genetic factors on complex disease.
External collaborations include research with Dr. John Little of Virginia Tech, to develop
improved tools for rapidly predicting exposure associated with SVOCs used or emitted in the
residential environment. Dr. Sean Hays and coworkers of Summit Toxicology will be exploring
application of the Biomonitoring Equivalent (BE) approach for interpretation of ToxCast™ data.
In a collaboration established through Bio-chem Redirect Program and implemented through the
ISTC (International Science and Technology Center), Dr. Petr Nikitin of the Natural Science
Center (NSC) of A.M. Prokhorov General Physics Institute, Russian Academy of Sciences is
leading research to develop a multi-channel immunosensor for detection of pyrethroids. The
primary goal of this project is development of a biochip technology that avoids sophisticated
labeling steps. This is an example of the type of research required to address needs for advanced
exposure monitoring tools. In collaboration with the ICCA-LRI, we are participating in planning
of a workshop focused on developing innovative tools to characterize biologically-relevant
environmental exposures and implication of these for health risks. Finally, in collaboration with
the Environmental Bioinformatics STAR center at UMDNJ, we will be presenting a symposium
at the ISES 2009 annual meeting.
Input to and feedback on ExpoCast™ will be solicited through the ExpoCoP (Exposure
Science Community of Practice) to facilitate integration and collaboration across the broader
scientific community. ExpoCoP includes representatives from the ORD labs and centers, Agency
program offices, other federal government agencies, academia, industry, and environmental
advocacy groups. International representatives also participate in the ExpoCoP.
Milestones/Products:
FY09
• EHP paper with NERL, "Exposure as part of a systems approach for assessing risk"
• Tox Sci Forum paper "Biologically-Relevant Exposure for Toxicity Testing"
• ExpoCoP monthly teleconference, ESC resource, face-to-face meeting at ISES 2009
• ICCA-LRI workshop, "Connecting Innovations in Biological, Exposure and Risk Sciences:
Better Information for Better Decisions"
• SVOC workshop, "Semi-Volatile Organic Compounds (SVOCs) in the Residential
Environment"
• ISES 2009 Annual Conference, "Transforming Exposure Science for the 21st Century"
• Symposium at ISES 2009, "Integrative Exposure Biology and Computational Toxicology for
Risk Assessment"
A-l.f.3
Previous I TOC
-------
EPA CompTox Research Program FY2009-2012 BOSC Review Draft- 24 August, 2009
• Survey and identify high priority exposure data resources for initial chemical indexing in
collaboration with ACToR and DSSTox
• ExpoCast™ conceptual framework and research plan
FY10
• Workshop to review/evaluate current exposure prioritization tools
• Position paper recommending standards for exposure data representation.
• White paper defining exposure space and plan for assessing exposure data landscape.
• White paper exploring development and application of human exposure knowledgebase.
• Begin implementation of standards across exposure databases of highest interest and utility for
NCCT projects
FY11
• Manuscript describing extant data analyses to identify critical determinants for exposure
classification and chemical prioritization based on potential for exposure.
• Guide incorporation and further development of simple exposure estimation tools within the
ACToR system for use in prioritization.
FY12
• Apply exposure index for prioritization to subset of ToxCast compounds to evaluate concept.
Keywords (three to five): Exposure Science, Chemical Prioritization and Toxicity Testing,
Vulnerable Populations, Susceptibility
QA Project Plan: Category III. Quality Assurance (QA) of modeling projects serves at least
two overlapping goals: 1) Verification - Reproducibility of results is essential for the scientific
method, and 2) Continuity - Proper documentation of results allows future researchers (or the
same researcher after a long period of time) to return to a project without excessive amounts of
time spent understanding what was done before. All QA plans are archived in the EPA internal
QA system.
A-l.f.4
Previous I TOC
-------
g. Virtual Embryo
Lead/Principal Investigator: Thomas B. Knudsen, PhD
Research Issue/Relevance: Research issue: The Virtual Embryo project (www.epa.gov/ncct/v-
Embryo/) is motivated by scientific and regulatory needs to understand mechanisms of
developmental toxicity. A key research issue is to model how the embryo reacts to
environmental chemicals as a 'complex system.' Navigating this complexity requires detailed
knowledge of molecular embryology, data on cellular systems using high-throughput (HTP)
screening approaches, and computational models of network dynamics and multi-cellular
function.
Morphogenesis entails a dynamic tissue flow that is driven by conserved cell signaling pathways
and cellular reaction networks that follow these pathways during stimulus, mutation or injury [1].
Our strategy is to modularize the embryo as a collection of tractable models that represent the
cell as a computational unit [2]. In this strategy, the cell is treated as an autonomous agent that
processes local signals and selects from a repertoire of core behaviors that include growth,
differentiation, mitosis, apoptosis, migration, adhesion, and cell-shape changes. Specific rules for
signal-response are programmed for molecular pathways, cellular dynamics, and unique biology
per morphogenetic system [3]. Sophisticated imaging techniques can reveal complex dynamics
of cell-cell relationships and a 'morphogenetic blueprint' of early development [4].
Although much is known about molecular signaling networks that drive morphogenesis,
considerably less is known about the nature of 'higher-order' processes that control collective
cellular behavior [3]. In complex systems, molecular networks can invoke higher-level processes
through signal-input and response-output relationships as determined by the timing and function
of signal strength and dynamic range [5]. We take this hypothesis in the context of
environmental stressors to embryonic development. Understanding network state relationships
will be required to predict non-linear dose-response relationships and when a breakdown of
higher-order control systems may occur [6]. Cell-based computational models have been used to
predict emergent properties that arise from cooperative transactions of cell groups behaving as a
self-regulating system [7]. Homeostasis, adaptation, and repair are a few examples of emergence
in a perturbed system [8].
Relevance: A new strategy for developmental toxicity testing involves screening multiple
chemicals through cell-based in vitro assays [1]. The goal is to build robust signatures of toxicity
that translate into in vivo predictions [9]. Because tissues are more than a cumulative sum of
individual cell behaviors, computational models can accelerate this effort [7]. A 'virtual tissue'
(VT) representation reconstructs a broad range of biological responses following cell-based rules
and systems-level controls. VT models can rapidly sweep parameter space following chemical or
genetic perturbation to predict aggregate cellular behaviors and higher-order responses [10].
A-l.g.l
Previous I TOC
-------
EPA CompTox Research Program FY2009-2012 BOSC Review Draft- 24 August, 2009
Purpose/Objective/Impact: Purpose: EPA's Virtual Embryo (v-Embryo™) will comprise a
knowledgebase (VT-KB) of relevant information and simulation engine (VT-SE) on the front
end of the modeling software. Proof-of-principle is underway to explore the range of potential
applications where comprehensive simulation is reliable and to find the limit in scale or stages
where studying the real embryo is not cost-effective. Computational models have been built by
others for in silico reconstruction of chondrogenesis [11], gastrulation [12], angiogenesis [13],
and somitogenesis [14]. These models were implemented as hybrid cellular automata using
CC3D open-source tissue simulation environment [www.CompuCell3D.org] which is being
evaluated for the Virtual Embryo. Initial models will focus on specific morphogenetic systems
that replay important concepts in experimental embryology and that are targets in developmental
toxicity. The end goal is a library of computer-driven simulations that can be manipulated in
silico and correlated with in vitro responses or in vivo phenotypes to predict developmental
toxicity. Applications for developmental toxicity align with EPA's strategy on the future of
toxicity testing [15] and can leverage unique pathway-based data for numerous chemicals tested
in mouse embryonic stem cells (mESC), free-living zebrafish (ZF) embryos, and ToxCast™
assays [16] to:
• Simulate key signaling pathways, interlocking genetic networks and cellular dynamics in
developing tissues;
• Model how embryonic cells react as agents to chemical exposure individually, and
collectively as a complex system;
• Analyze emergent behaviors and canalizing influences following stimulus / injury /
perturbation; and
• Understand how this complexity contributes to the differential susceptibility of
embryonic tissues across chemical, dose, stage, genetic makeup and time.
Objective: The main objective is to build data-rich models that can be used to analyze causal
relationships during environmental and genetic perturbation. One initial prototype will focus on
the powerful sine oculis network that controls early eye development - conserved morphogenetic
pathways [17], patterns of malformation and sensitivity to chemical perturbation [18]. There are
many advantages of focusing on eye malformations and specifically the molecular events related
to Pax6 leading to this endpoint: strong knowledgebase, relevant human phenotypes, and
underlying genetic susceptibility and environmental sensitivity. The underlying molecular
pathways and signaling networks scale to tissue-level developmental effects in many different
systems. A second prototype will focus on the patterning systems controlling early limb-bud
development. The specific objectives are as follows.
1. Build knowledgebase (VT-KB) and front end simulation engine (VT-SE) for developmental
processes and toxicities.
Rationale: VT-KB is required to initially store gene-gene and gene-phenotype associations
that will be used to provide the rules for cell-based modeling in the VT-SE. The framework
A-l.e.2
Previous I TOC
-------
EPA CompTox Research Program FY2009-2012 BOSC Review Draft- 24 August, 2009
will rely on information from the literature, data from national repositories (GO, EMAGE,
GXD, MPO, ZFIN, OMIM) and pathway analysis software. VT-KB design, development and
implementation are supported through ITS-ESE contract No.: 68-W-04-005, Task Order No.
058: Technical Support for Development of Developmental Systems Toxicity Network
(DevToxNet). Perl-scripts written for information extraction and assisted curation of relevant
facts from scientific literature returned an output matrix loaded to MySQL. Initial application
wll build gene-gene and gene-phenotype associations during eye and limb development for
rat, mouse, zebrafish, or human species. VT-SE comprises a front end to the modeling
software (DDLab, CC3D, C++, Python, Blender3D, GanttPV). An interactive tool is being
developed with support from the Environmental Modeling and Visualization Laboratory
under ITS-ESE contract No.: 68-W-04-005, Task Order No. 02: Virtual Embryo: Simulation
and Visualization Project Management Plan. Initial application of VT-SE is to construct a
cell-based, network-driven model for lens-retina induction that reconstructs the cellular
dynamics and morphogenetic blueprint of ocular dysmorphogenesis. This model can be
quantitative in terms of the degree of severity of chemical-induced defects (graded dose
response) and relative risk for incidence of responding embryos (quantal dose response).
2. Construct cell-based computational models for prototype morphogenetic processes and
embryonic modules.
Rationale: The second Specific Aim use the VT-SE to model modular embryonic systems,
specific morphogenetic events and their perturbation. The strategy will apply agent-based
models (ABMs). Initial prototype (optic cup, limb-bud) have well-characterized signaling
networks and differential susceptibility to chemical disruption; other systems will be added
over time. Both small prototypes are organized by complex self-regulating networks of signal
molecules commuted from cell signaling centers and described mathematically as Turing
gradients. Prevailing models entail reciprocal induction in which heterotypic interactions
between presumptive lens epithelium and prospective neural retina lead to formation of the
optic cup, and interactions between apical ectodermal ridge and underlying mesenchyme
drive polarized outgrowth of the paddle-shaped limb-bud. Both processes are organized by a
self-regulating network of genes and signaling gradients (FGFs, BMPs, SHHs). Whereas the
dual-reciprocating models set the stage for emergence of the optical neuraxis and
appendicular skeleton, respectively, they do not explain higher-order processes that control
geometry and size of these rudiments, nor do they account for differential susceptibility to
teratogens [18]. For this purpose, we propose the extended cellular large-Q Pott's model
(CPM) implemented in CC3D [19] and managed with Python software as part of the VT-SE.
3. Specify rules for component interactions of developmental pathways at the cellar and
molecular scales.
Rationale: Network structures for regulatory pathways and cellular systems will be portrayed
using a Boolean (on-off) formalism. A Boolean Network (BN) qualitatively captures system
behavior probabilistically (PBN) or deterministically (DBN): the former is more biologically
plausible whereas the latter is a modeling tool of the whole process, which enables us to
simulate, analyze, and manipulate different parts of the system. Both models incorporate
rule-based dependencies for gene-gene and cell-cell interactions that can be built with
A-l.e.3
Previous I TOC
-------
EPA CompTox Research Program FY2009-2012 BOSC Review Draft- 24 August, 2009
information from the VT-KB. States of the network as determined by wiring and rules will be
characterized using DDLab software, which identifies stable attractors in complex systems
[20]. The attractor concept implies that a finite number of stable cell states exist in a complex
system as pathways of differentiation or canalization. Order and timing has great importance
for predictive modeling genetic errors or cellular disruptions in the embryo caused by phase-
specific chemical effects. As such, the ability to introduce pre-defined or stochastic lesions
will enhance the functionality of VT-SE. Virtual Embryo is building a prototype interface
that manages the order and timing of gene expression and signaling events for Python-based
components in the VT-SE. This tool is being developed under ITS-ESE contract No.: 68-W-
04-005, Task Order No. 02: Virtual Embryo - Simulation and Visualization.
4. Analyze abnormal developmental trajectories predicted from cellular modes of action that
follow chemical perturbations.
Rationale: A high-fidelity computer program that links cellular processes with network-level
function can be evaluated for its capacity to evolve features not explicitly coded in the cell-
based model. As noted above, this emergence is important for manifesting the response to
genetic errors and cellular disruptions that are introduced to the model as targets of
developmental toxicity, based on simulated and experimental data. Such a computational
model can reveal an interaction of mechanisms at the cellular and molecular scales to
produce emergent phenomena that manifest as abnormal developmental phenotypes [10].
Rules will derive from simulated data and semi-arbitrary parameters in ocular or related
systems since it will take a major experimental effort to model parameters kinetically for all
relevant pathways, reactants and interactions. Eventually, such information would be helpful
to build a quantitative model. CC3D software advancements will be needed to implement
molecular motors for core cell behaviors and to parallelize this implementation.
Impact: This research aims to improve mechanistic understanding and predictive modeling of
developmental toxicity. Biological models that are simple enough so as to be computationally
feasible (tractable) and yet complex enough to compute integrated cellular behaviors (rational)
can reveal key events in multi-cellular organization, classify abnormal developmental trajectories
from genetic network inference, and predict chemical dysmorphogenesis from pathway-level
data. The initial focus on existing data, with use of the modeling effort to identify data gaps and
to help guide the design of experiments for generating additional data as needed, can produce
results that provide significant new information on likely dose-response and time-course
behaviors of developmental toxicants. Most developmental modeling has been qualitative,
showing the link between fundamental processes and morphogenesis. Virtual Embryo is moving
to a different, quantitative level through knowledge of molecular embryology and pathway-level
data from high-throughout screening efforts. That resource can have an impact on HTP
hypothesis testing (parameter sweeps) to inform experimental design, or to dry-run intractable
experiments complicated by time, scale, and cost (monetary, animal). Over the short-term, we
anticipate the work will draw greater attention to integrative thinking, the application of
computational models to understand mechanisms, and approaches for uncertainty analysis and
understanding how large uncertainties about parameter values will affect quantitative prediction.
Expanding the prototype models to broader representation of stages, tissues and species will be
an intermediate step towards the more visionary reconstruction of a 'Virtual Embryo'.
A-l.g.4
Previous I TOC
-------
EPA CompTox Research Program FY2009-2012 BOSC Review Draft- 24 August, 2009
Synopsis: Motivation of the Virtual Embryo is scientific needs to understand mechanisms of
toxicity and predict developmental defects from complex datasets. The research goal is to
simulate embryonic tissues reacting to perturbation across chemical class, system, stage, genetic
makeup, dose and time. Data input is detailed knowledge of molecular embryology, high-
throughput data from in vitro models, signaling pathways, and cellular phenotypes. Output
models aim for the modular reconstruction of a developing embryo from cell-based models of
morphogenesis and differentiation.
Milestones/Products:
FY09-10
1. project plan: Category III QAPP
2. recruit: postdoctoral fellow
3. manuscript: application of VT-KB to analyze ToxRefDB developmental toxicity studies
4. model: VT-KB based qualitative (structural) model of self-regulating ocular gene
network
5. model: VT-SE based cell-based computational model of lens-retina induction
6. manuscript: ocular morphogenesis, gene network inference, analysis and modeling
FY10-11
1. project plan: extend lens-retina model to other stages and species
2. model: incorporate pathway data from ToxCast™, mESC and ZF embryos
3. manuscript: sensitivity analysis for developmental trajectories and phenotypes
4. project plan: integrate with other morphogenetic models (ES cells, Zfish)
FY11-12
1. manuscript: test model against predictions for pathway-based dose-response relationship
2. manuscript: uncertainty analysis of models for complex systems
3. model: computer program of early eye development using rules-based architecture, cell-
based simulators and systems-wiring diagrams
Keywords (three to five): embryo development; systems biology; computational modeling
QA Project Plan: Category III. The proposed designation for Virtual Embryo is Quality
Assurance Category III. This designation recognizes its origin as a basic research project
(Category IV) that is moving into proof of concept phase (Category III). A Virtual Embryo
Quality Management Plan (QMP) will be constructed as the project moves into the proof of
concept phase.
Phase-I (development): first-generation ABMs based on the small prototype systems of lens
induction and polarized limb outgrowth (2009-10).
Phase-II (evaluation): sensitivity analysis using data for ToxCast™ chemicals that disrupt eye
and/or limb development or Tox21 assays of relevant signaling pathways (2010-11).
A-l.e.5
Previous I TOC
-------
EPA CompTox Research Program FY2009-2012 BOSC Review Draft- 24 August, 2009
Phase-Ill (expansion): uncertainty analysis of quantitative models that simulate reaction to
perturbation across chemical, system, stage, genetic makeup, dose and time (2011-12).
The anticipated Date of Elevation to Category-II is 2012. This is based on the premise that
research enabled by these models will reduce uncertainty in risk assessment for prenatal
developmental toxicity, through understanding of complex mechanisms of environmental
chemicals and their impact on complex developing systems. We also anticipate a successful
Virtual Embryo in the long-term can reduce the reliance on animal testing for prenatal
developmental toxicity. Many of the 10,000 chemicals EPA is concerned with do not have such
information available.
References
1. National Research Council, Committee on Toxicity and Assessment of Environmental Agents
(2007). Toxicity Testing in the Twenty-first Century: A Vision and a Strategy. Washington,
DC: National Academies Press Qittp://www.nap.edu/catalog/11970.html).
2. Thorne BC, Bailey AM, DeSimone DW and Peirce SM (2008) Agent-based modeling of
multicell morphogenetic processes during development. Birth Defects Res (part C) 81: 344-
353
3. Lewis J (2008) From signals to patterns: space, time, and mathematics in developmental
biology. Science 322: 399-403
4. Keller PJ, Schmidt AD, Wittbrodt J and Stelzer EHK (2008) Reconstruction of zebrafish early
embryonic development by scanned light sheet microscopy. Science 322: 1065-1069
5. Janes KA, Reinhardy HC and Yaffe MB (2008) Cytokine-induced signaling networks
prioritize dynamic range over signal strength. Cell 135: 343-354
6. Andersen ME, Yang RSH, French CT, Chubb LS and Dennison JE (2002) Molecular circuits,
biological switches, and nonlinear dose-response relationships. Env Hlth Persp 110: 971-978
7. Knudsen TB and Kavlock RJ (2008) Comparative bioinformatics and computational
toxicology. In: Developmental Toxicology Volume 3, Target Organ Toxicology Series. (B
Abbott and D Hansen, editors) New York: Taylor and Francis, Chapter 12, pp 311-360
8. Basanta D, Miodownik M and Baum B (2008) The evolution of robust development and
homeostasis in artificial organisms. PLOS Computat Biol 4(3): e!000030
9. Martin MT, Houck KA, McLaurin K, Richard A and Dix DJ (2007) Linking regulatory
toxicological information on environmental chemicals with high-throughput screening (HTS)
and genomic data. The Toxicologist CD - J. Soc Toxicol 96: 219-220
10. Andersen T, Newman R and Otter T (2009). Shape homeostasis in virtual embryos. Artificial
Lifel5(2):161-183.
A-l.g.6
Previous I TOC
-------
EPA CompTox Research Program FY2009-2012 BOSC Review Draft- 24 August, 2009
11. Chaturvedi R, Huang C, Kazmierczak B, Schneider T, Izaguirre JA, Glimm T, Hentschel HE,
Glazier JA, Newman SA and Alber MS (2005) On Multiscale approaches to three-dimensional
modeling of morphogenesis. JR Soc Interface 2: 237-253
12. Cui C, Yang X, Chuai M, Glazier JA and Weijer CJ (2005) Analysis of tissue flow patterns
during primitive streak formation in the chick embryo. Developmental Biol 284: 37-47
13. Mahoney AW, Smith BG, Flann NS and Podgorski GJ (2008). Discovering novel cancer
therapies: a computational modeling and search approach. In: IEEE Symposium on
Computational Intelligence in Bioinformatics and Bioengineering. Sun Valley, ID: CIBCB.
14. Glazier JA, Zhang Y, Swat M, Zaitlen B and Schnell S (2008) Coordinated action of N-CAM,
N-cadherin, EphA4, and ephrin B2 translates genetic prepatterns into structure during
somitogenesis in chick. Curr Topics Devel Biol 81:205-247
15. Future of Toxicity Testing Workgroup, US EPA (2009) The U.S. Environmental Protection
Agency's Strategic Plan for Evaluating the Toxicity of Chemicals
http://www.epa.gov/osa/spc/toxicitytesting/
16. Chapin R, Augustine-Rauch K, Beyer B, Daston G, Finnell R, Flynn T, Sunter S, Mirkes P,
O'Shea KS, Piersma A, Sandier D, Vanparys P and van Maele-Fabry G (2008) State of the art
in developmental toxicity screening methods and a way forward: a meeting report addressing
embryonic stem cells, whole embryo culture, and zebrafish. Birth Defects Res (Part B) 83:
446-456
17. Chow RL and Lang RA (2001) Early eye development in vertebrates. Annu Rev Cell Dev Biol
17: 255-296
18. Green ML, Singh AV, Zhang Y, Nemeth KA, Sulik KK and Knudsen TB (2007)
Reprogramming of genetic networks during initiation of the fetal alcohol syndrome. Devel
Dynam 236: 613-631
19. Cickovski TM, Huang C, Chaturvedi R, Glimm T, Hentschel HGE, Alber MS, Glazier JA,
Newman SA and Izaguirre JA (2005) A framework for three-dimensional simulation of
morphogenesis. IEEE/ACM Trans Comput Biol Bioinfor 2: 1-15
20. Wuensche A (1996) Discrete dynamics Lab (DDLab) Available from [http://www.ddlab.com]
A-l.e.7
Previous I TOC
-------
h. The Virtual Liver Project: v-Liver™
Lead/Principal Investigator: Imran Shah
Research Issue: The Virtual Liver project (http://www.epa.gov/ncct/virtual liver) is aimed at
providing decision support tools for evaluating chemical-induced adverse liver outcomes across
chemicals, doses and species using in vitro data. Considering nuclear receptor (NR) mediated
liver cancer as an archetypal chronic adverse outcome, we focus on the research issues: Which
molecular circuits and cellular states altered by chemicals lead to cell damage, death and
proliferation? How are these cellular perturbations propagated across tissues as lesions? How can
we organize this complexity computationally to develop Virtual Tissues?
Two important perturbed cellular phenotypes, or states, in carcinogensis are: (i) initiation, in
which chemical mutagens cause DNA damage rendering a cell resistant to apoptosis, inhibition
of cell proliferation; and (ii) promotion, in which mitogenic signals persistently stimulate the
initiated cell creating focal proliferation. Increasing evidence suggests that the nuclear receptor
(NR) superfamily mediates rodent hepatocarcinogenesis for a number of environmental
chemicals (Butler 1996). For example, di(2-ethylhexyl)-phthalate (DEHP) and perfluorooctanoic
acid (PFOA) are PPAR-a activators (Maloney and Waxman 1999); while pesticides like
conazoles and pyrethroidsactivate either PXR or both PXR and CAR (Kretschmer and Baldwin
2005). The role of NR mediated activity in molecular circuits is being actively explored through
genomic profiling, and the dose-dependence of specific molecular switches is being assayed
across hundreds of environmental chemicals in ToxCast™.
Propagating cellular alterations spatially requires information flow between cells, which
normally occurs on a backdrop of microanatomic spatial zones with heterogenous levels of
nutrients and distinct spatial distribution of intracellular states (Pette and Wimmer 1979;
Oinonen et. al. 1998). The microanatomic distribution of xenobiotics causes zonal alterations in
cell states (Kato et. al. 2001) that can progress to cell injury and even death. Hepatocyte (HC)
death stimulates neighbouring cells to replicate (regenerative proliferation). Necrotic death can
also lead to Kupffer cell (KC) activation, migration and release of inflammatory cytokines,
which can locally accelerate cell injury. There is evidence for such HC-KC interactions in
PPAR-a mediated hepatotoxicity and cancer (Rusyn 1998; Roberts 2007). Mitogens have also
been shown to disrupt gap junction communication between cells (Krutovskikh et. al. 1995),
which can reduce their homeostatic capacity. Advanced imaging, histomorphometry (HMP) and
molecular assays are making it feasible to extract local information on cells in a microanatomic
context.
To computationally model this level of biological complexity requires some simplifying
assumptions about the modular organization of physiologic events across scales. We hypothesize
a cell-oriented abstraction for developing "Virtual Tissues" with the following assumptions: (a)
tissues can be represented as a complex cellular system; (b) cells are the unit of function and can
be modeled as autonomous agents that use molecular circuits to make decisions; and (c) injury is
a collective response of the multi-agent system to persistent stress.
A-l.h.l
Previous I TOC
-------
EPA CompTox Research Program FY2009-2012 BOSC Review Draft- 24 August, 2009
Relevance: Current approaches for assessing the risk of adverse effects in humans are based on
animal testing, which is time consuming, resource intensive and fraught with uncertainty. Novel
strategies are necessary to efficiently and effectively evaluate the risk of thousands of
environmental chemicals. Integrative computational systems, in vitro models and assays offer an
avenue for more cost-effective and humane alternatives for the future of toxicity testing. Liver
toxicity is a frequent outcome in rodent testing and it is difficult to evaluate its relevance in
humans.
Purpose: The v-Liver™ will provide in silico decision support tools to: (a) analyze the mode-
of-action in light of available data and prior knowledge, and (b) quantitatively simulate an MOA
at environmentally relevant tissue doses. This will inform/evolve biologically-based dose-
response models to include more relevant physiologic details necessary for predicting the human
risk of injury at low doses.
Objective: The primary objective is to develop an integrated in silico/in vitro framework that:
(a) aids intelligent hypothesis generation about the plausible sequence of molecular, cellular and
tissue events perturbed by a test chemical, and (b) quantitatively simulates the risk of these
events in humans at environmentally relevant tissue doses using in vitro data. The v-Liver™
proof-of-concept (PoC) will focus on a subset of 20 environmental chemicals with known rodent
toxicity in ToxRefDB and in vitro data in ToxCast™. There are two specific goals of the PoC:-
/. The v-Liver™ Knowledgebase (KB).
Knowledgebased, or semantic, approaches (Karp 2001) are important for computationally
modeling incomplete and evolving insight on complex processes. They enable integration of
disparate biological information from literature, -omic data, or pathway databases at different
scales into coherent computable representation that is flexible, extensible and transparent. A
more important advantage of semantic approaches is their support for automated reasoning,
which is important for inferring plausible sequences of events perturbed by new chemicals.
Large-scale knowledgebased approaches have become feasible due to semantic web
technology and available ontologies for different levels of biological organization. The v-
Liver™ Knowledgebase (v-Liver-KB) will represent normal hepatic functions and their
perturbation by chemical stressors into pathophysiologic states using description logic
expressed in OWL and stored in the Sesame semantic repository. To facilitate the
construction of the v-Liver-KB, a Cytoscape plugin is being developed to synthesize
information from different biological databases into OWL using a custom ontology. This
system also supports SPARQL-based queries, interactive visualization and information
export in RDF. Information about the 20 PoC chemicals will be represented in the KB at
multiple biological levels describing events and causal relationships between these events
based on evidence from experiments or the literature. The main outputs of the KB will be:
1. A computable logical description of the molecular circuits and cellular states involved
in normal hepotocyte and Kupffer cell function based on literature
2. A computable logical description of perturbations in molecular switches and cell
states due to by PoC chemicals.
3. Interactive web-based tools to browse and interactively query/explore the v-Liver-KB
to analyze alternative MOA in light of HTS, omic or other cell based assays.
A-l.h.2
Previous I TOC
-------
EPA CompTox Research Program FY2009-2012 BOSC Review Draft- 24 August, 2009
4. Intelligent inference tools to explore alternative pathways perturbed by chemicals
based on existing information on partial orders in the KB.
//. The v-Liver™ Simulator (Sim).
The spatial model of the hepatic lobule will be developed using a multi-agent system (MAS)
(Axelrod, 1997; Epstein and Axtell, 1996; Athale et. al.. 2005) in which hepatocytes and
Kupffer cells will be modeled as autonomous agents. Information on molecular circuits in
the KB will be used to describe chemical-induced molecular perturbations of nuclear
receptors (NRs) namely, CAR, PXR and PPAR-a. We are developing a variation of
Probabilistic Boolean Networks (Kauffman 1993; Shmulevich et. al. 2002) to describe the
dynamics of individual agent decisions regarding state of stress, injury, cell cycle
progression, apoptosis, necrosis or migration (KC). These will be augmented and calibrated
using available literature and/or ToxCast™ data on the PoC chemicals. We believe this work
will advance the two-stage clonal growth models of cancer (Conolly and Andersen 1997) by
including relevant information on molecular pathways and cell-communication. The agent
population will be initially situated in 2-dimensional regular spatial grid to model in vitro
conditions and a simplified cross-section through the hepatic lobule. Portal to centrilobular
blood flow will be initially represented as a gradient of nutrients and xenobiotics (estimated
from organ dose), which can be extended to model more complex flows if necessary.
Simulating the MAS will generate a spatial distribution of cellular alterations that can be
interpreted as tissue lesions. Hence, the v-Liver™ Simulator (v-Liver-Sim) will dynamically
simulate the key molecular and cellular perturbations leading to adverse effects in hepatic
tissues. The predictions will be evaluated for the PoC chemicals using ToxCast™ data. The
main outputs of the v-Liver™ Simulator are:
• A large-scale tissue simulation engine to enable quantitative exploration of
alternative physiologic processes and their histopathologic outcomes.
• A computational interface to integrate the tissue simulator with a PBPK modeling
system to investigate individual exposure / population variability.
• Interactive tools to communicate with the simulation engine and to visualize
results of simulations across chemicals, MOAs, doses and species.
Impact: The v-Liver™ will impact the future of toxicity testing by providing computational
tools to explore the mode of action for new environmental chemicals using background
knowledge, chemical structure, and/or in vitro assays, and to provide an initial assessment of
hepatic lesion formation. Focusing on 20 NR activating environmental chemicals and their
hepatic lesions through a subset of molecular pathways will demonstrate the v-Liver ™proof-of-
concept (PoC). The project is also expected to contribute to on-going assessments of pesticides
and persistent toxics by providing useful information about the human relevance of any liver
effects and their putative dose-response. If successful, the Virtual Tissues will be able to leverage
available screening data from ToxCast™, fill any data gaps with targeted studies and reduce the
time, the cost and the requirement for as many animal studies.
Synopsis: The v-Liver™ computational paradigm represents tissues as cellular systems in
which discrete individual cell level responses give rise to complex physiologic outcomes. In this
model cell level responses are governed by a self-regulating network of normal molecular
A-l.h.3
Previous I TOC
-------
EPA CompTox Research Program FY2009-2012 BOSC Review Draft- 24 August, 2009
processes and adverse histopathologic effects arise due to chronic stimulation by environmental
chemicals. The v-Liver™ PoC is being developed by: (i) focusing on environmental chemicals
responsible for hepatocarcinogenesis in rodent studies; (ii) organizing mode-of-action (MOA)
knowledge on the relevant molecular and cellular processes perturbed by these chemicals; (iii)
developing a tissue simulation platform to investigate the uncertainties in MOA and neoplastic
lesion formation; and (iv) evaluating in vitro assays to predict lesion development across
chemicals and doses.
Partnerships/Collaborations (Internal & External): EPA Collaborations: NCCT ToxCast™
in vitro assays and their linkage with physiologic outcomes; NCCT v-Embryo™ and cell level
modeling of tissue responses; NCCT/NHEERL/PBPK modeling to infer internal dose; NHEERL
Genomics Core on MoA for PPAR-a activators; NHEERL PK Branch on hepatic xenobiotic and
T3/T4 metabolism. External Collaborators: UNC Carolina Center for Computational Toxicology;
UMDNJ PBPK Modeling.
Milestones/Products:
FY09
1. Prioritize proof of concept (PoC) environmental chemicals with clients
2. KB: Information about PoC chemicals using ToxCast assays
3. KB: Cytoscape KB visualization and analysis tool
4. Cell response: Initial molecular circuit describing hepatic cell functions
5. Tissue Simulator: Develop / use MAS framework
FY10
1. Tissue Simulator: Test liver lesions formation
2. Integrate molecular circuits for MOA chemicals in Tissues
3. Evaluate simulator using PoC chemicals and ToxCast™ data to predict outcomes
FY11
1. KB inference tool for analyzing MOA for new chemicals/mixtures
2. Extend lobule simulator to liver and integrate with PBPK model
FY12
1. Evaluate impact of genomic variation on cellular responses and lesion formation
2. Evaluate v-Liver™ for simulating human pathology outcomes using clinical data
Keywords (three to five): Virtual tissues; knowledgebases; mode-of-action modeling; dose-
response modeling; nuclear receptors mediated hepatocarcinogenesis.
QA Project Plan: Category III. QA of modeling projects serves at least two overlapping goals:
1) Verification - Reproducibility of results is essential for the scientific method, and 2)
Continuity - Proper documentation of results allows future researchers (or the same researcher
after a long period of time) to return to a project without excessive amounts of time spent
understanding what was done before. Since the v-Liver™ project requires multiple researchers
working over several years, ensuring both continuity of modeling efforts and reproducibility of
modeling results is vital. All QA plans are archived in the EPA internal QA system.
A-l.h.4
Previous I TOC
-------
i. Uncertainty Analysis in Toxicological Modeling
Lead/Principal Investigator: R. Woodrow Setzer
Research Issue/Relevance: The analysis of uncertainty in toxicological modeling is critical to
the EPA because the Agency is increasing its use of toxicological models in regulatory decisions,
and any use of model predictions in a rational decision process must consider the uncertainty of
those predictions. The recent National Academy report on risk assessment activities in the U.S.
EPA emphasizes the importance of incorporating a quantification of uncertainty in risk
assessments.
In the context of models, the easiest form of uncertainty to address is that about parameter
values, assuming we have the correct model. This is the sort of uncertainty that statistical
methodologies were designed to estimate, and is typically quantified through confidence
intervals or probability distributions. However, we are rarely completely confident about the
models we use, and often the uncertainty about the underlying processes taking place or the best
way to characterize those processes in a model can be quite substantial. This form of uncertainty
is prevalent throughout dose-response analysis, from simple empirical modeling used in
benchmark dose analysis, to pathway modeling used in virtual tissues, and is called model
uncertainty.
Both forms of uncertainty are quantified through comparisons of models with data, by
quantifying the degree to which different parameter values for a given model, and the best-fitting
parameter values among different models, yield model predictions that are consistent with the
data. "Consistency" is quantified through the mediation of an additional, statistical, model that
relates the biological model to the data by describing the variability in the data as it is affected by
the biological model.
In principle this process is straightforward, and there are standard statistical procedures for
carrying the process out. However, toxicological models comprise a wide array of modeling
techniques, for example, simple algebraic expressions, systems of ordinary or partial differential
or difference equations, and may involve agent-based or stochastic process models. Such models
can be quite complex, with many parameters whose values are known with varying degrees of
uncertainty. Statistical models usually need to accommodate multiple hierarchical levels of
variability: variation among studies and among individuals in addition to the usual measurement
error, which should be allowed to differ among studies and endpoints. Information for estimating
parameters and evaluating models themselves may come from well-characterized experimental
data as well as from tabulated (for example, physiological parameters like organ weights) or
computed (for example, computed partition coefficients for a physiologically-based
pharmacokinetic, or PBPK, model).
Thus, despite the existence of sound statistical theory, the application of good statistical practice
for these models can be difficult, requiring thoughtful application of both statistical and
computational expertise and quite a bit of 'art' to get good results. The key challenges in
uncertainty analysis for toxicological models in risk assessment are to develop computationally
A-l.i.l
Previous I TOC
-------
EPA CompTox Research Program FY2009-2012 BOSC Review Draft- 24 August, 2009
efficient, transparent, and statistically valid approaches that may be implemented without the
development of extensive case-specific programming.
Purpose/Objective/Impact: The objective of this project is to develop tools and best practices
to facilitate the quantification of uncertainty in toxicological models. Early efforts will focus on
PBPK models, specifically models being developed as part of a joint project between the
Agency's Office of Research and Development and Office of Pesticide Programs to develop a
cumulative risk assessment for the pyrethroid pesticides. However, similar methods are
applicable to other forms of toxicological models, and examples will be pursued in the virtual
tissue projects in NCCT, and in evaluating uncertainty in the ToxCast predictors.
In recent years, it has become clear that Bayesian statistical methods are ideal for estimating
parameters for PBPK models, because of the ease with which partial information about
parameters is included, through informative priors; the ease with which hierarchical models are
constructed; and the fact that, as long as informative priors are used, lack of identifiability of
model parameters, which is not generally possible to diagnose a priori, is not an impediment to
completing a valid analysis (failure of identifiability of parameters may be diagnosed from
analyses of the posterior parameter distribution). The same logic should apply to other
toxicological models. Thus, this project will center on tools to apply Bayesian methods to such
models.
More specific objectives of this project are:
Develop the computational tools to carry out Bayesian analyses of large dynamic models
as efficiently as possible. The standard approach to evaluating the posterior in a Bayesian
analysis of complex models is to generate random samples from the distribution by
Markov Chain Monte Carlo (MCMC) methods. For models that are expensive to
compute, such as those (like PBPK models) expressed as solutions to systems of
differential equations, an implementation that takes advantage of cluster computing may
take advantage of some parallel structures in the problem. This objective will develop a
standard computational approach to parallelizing such problems, and will explore
alternative approaches to implementing MCMC methods with an eye towards
computational efficiency. This objective also includes the development of modeling tools
to facilitate dynamic models in the statistical language R.
Describe a language specifically for expressing PBPK models. The language should be
extensible, be capable of incorporating systems models expressed in SBML, and should
use semantics specific to PBPK models to facilitate model checking. The language would
be ideal as a way to archive PBPK models and as the definitive way to communicate
PBPK models in the literature.
Adapt statistical model evaluation approaches to complex toxicological models, and
develop examples to demonstrate the behavior of different model evaluation
methodologies in the face of various model failures.
Develop a general approach to developing priors for chemical-specific PBPK model
parameters. There are already methods for computing chemical specific parameters either
from physical chemical properties or in vitro assays (depending upon the parameter). For
this objective, data sets of measured chemical-specific parameters will be compared to
values predicted from computational or in vitro methods. Statistical methods such as
A-l.i.2
Previous I TOC
-------
EPA CompTox Research Program FY2009-2012 BOSC Review Draft- 24 August, 2009
regression will be used to adjust the predictions, and the variance about the resulting
regression lines used to characterize the prior uncertainty about such predictions.
Develop approaches for quantifying uncertainty of ToxCast-like predictors involving
HTS data as inputs, explicitly evaluating the importance of variability and design of HTS
assays on prediction and prioritization uncertainty.
Examples for parameter estimation, model evaluation and overall uncertainty analysis
drawn from PBPK models for pyrethroid pesticides and the encompassing cumulative
risk analysis being conducted in collaboration with NHEERL, NERL, and OPP; selecting
among molecular and cellular models used in developing a virtual liver model, and others
developed in collaboration with the virtual tissues projects in NCCT.
Synopsis: This plan will target three areas: standardizing and making more efficient
computational approaches for parameter estimation and model selection; standardizing
approaches for model evaluation for PBPK and other dynamic models; and development of
methods for constructing priors (probabilistic summaries of current knowledge) for model
parameters, based on existing computational methods and data sets. The initial motivation for
this work was the need to standardize and make more sophisticated parameter estimation and
model evaluation for PBPK models, and that emphasis will continue in the early phase of this
project. However, all models relevant to toxicological risk assessment have similar
requirements, and this project will coordinate closely with the virtual tissue and dose-response
projects.
Partnerships/Collaborations (Internal & External): Internal: Jimena Davis (post doc);
Richard Judson, John Wambaugh, Imran Shah, Thomas Knudsen.
External: ORD/NHEERL: Mike Hughes, Kevin Crofton, Tim Shafer, Ginger Moser, Rory
Conolly; ORD/NERL: Rogelio Tornero, Valerie Zartarian, Xianping Xue; OPPTS/OPP: Anna
Lowit, David Miller, Ed Scollon; NIEHS/NTP: Mike DeVito.
Milestones/Products:
FY09
1. Contribution to 2009 SOT CED course on "Uncertainty and Variability in PBPK
Models".
2. Submission of ms(s) on improvement of computational efficiency in Bayesian analyses
of PBPK models using MCMC, and assessment of convergence (Wambaugh, Davis,
Garcia, Setzer).
3. Submission of ms on assessing fit of PBPK models (and, through example, other
complex mechanistic models), to data, whether the models are 'fit' to the data using
Bayesian or other methods, or are parameterized a priori from in vitro data (Wambaugh,
Davis, Garcia, Setzer).
4. Submission of ms reviewing approaches for constructing priors for PBPK model
parameters (that is, establishing a priori estimates for model parameters in advance of
using in vivo PK data, with a characterization of uncertainty) (Davis, Setzer, Tornero,
DeVito).
5. Draft ms on the estimation of model parameters and comparison of alternative model
forms for a PBPK model for permethrin (in preparation of SAP review in FY10) (Davis,
Setzer, Tornero, DeVito)
A-l.i.3
Previous I TOC
-------
EPA CompTox Research Program FY2009-2012 BOSC Review Draft- 24 August, 2009
6. Draft ms on global sensitivity analysis for exposure-dose model for permethrin, in
preparation for SAP review in FY10. (Davis, Setzer, Tornero, Zartarian, Chiu)
7. Estimation of PBPK parameters for deltamethrin and two other pyrethroids
complete.(Davis, Setzer, Tornero)
MS on permethrin PBPK parameter estimation and model comparison submitted
MS on global sensitivity analysis for permethrin exposure-dose model submitted
SAP review of permethrin PBPK exposure - dose model.
Submission of ms on parameter estimation for deltamethrin and two other pyrethroids
Completion of exposure-dose-effect model for 'mini-cumulative' risk assessment and
draft document for SAP review in early FY11, including global 'exposure to effect'
sensitivity analysis (Davis, Setzer, Tornero, DeVito, Crofton, Shafer, Lowit, Scollon,
Miller, ...)
6. Substantial completion of modeling and uncertainty analysis for vLiver (Setzer, Shah,
Wambaugh)
7. R package "RDynamic", for simplifying dynamic modeling in the statistical language R,
submitted to CRAN.
8. Description of ontologically-aware PBPK language, working name 'SemanticPK'
drafted, and translator to R completed (using code developed for RDynamic).
9. Discussion of uncertainty about ToxCast phase I predictions at 2010 SOT.
10. Problem identification and early stages of evaluating alternative pathway formulations in
vLiver model (Setzer, Shah, Wambaugh).
11. Problem identification, timelines, and early stages of evaluating alternative model
formulation for BBDR and virtual embryo models completed (Setzer, Conolly, Knudsen).
FY11
1. SAP on 'mini-cumulative' risk assessment for pyrethroids
2. Preparation for SAP for cumulative risk assessment for pyrethroids.
3. publication of virtual tissue and BBDR modeling results
4. Initial steps in constructing test datasets for analysis "competition" recommended by
UVPKM.
5. Submission of ms on "RDynamic" to the Journal of Statistical Software.
FY12
1. Submission of ms on 'SemanticPK'.
2. R package "RDynamic", for simplifying dynamic modeling in the statistical language R,
submitted to CRAN.
Keywords (three to five): uncertainty analysis; physiologically based pharmacokinetic models;
Statistics; Bayes methods; prior; Markov Chain Monte Carlo
QA Project Plan: Category III. This project is a Category III QA category due to the
significance of the pyrethroid cumulative risk assessment, to which this plan contributes.
Several quality objectives apply to this project: 1) Computer code developed for the project must
faithfully execute the intent of the code, whether that intent is described mathematically, or in
A-l.i.4
Previous I TOC
-------
EPA CompTox Research Program FY2009-2012
BOSC Review Draft- 24 August, 2009
terms of other software (for example, implementations of the same model in different
programming languages must give identical results for identical inputs); 2) Distribution versions
of software packages must be installable and useable by a reasonably sophisticated person; 3)
Analyses must be transparent: it must be possible to replicate all analyses from the archived files
and information; and 4) Data sources must be transparent; in particular, data from the literature
must be annotated to be adequate for purpose, and extent of literature searches must be
documented. All QA plans are archived in the EPA internal QA system.
Previous
A-l.i.5
TOC
-------
EPA CompTox Research Program FY2009-2012
IV. APPENDICES cont. -2. Project Outcomes Table
BOSC Review Draft- 24 August, 2009
1. Providing High Throughput Computational Tools for the Identification of Chemical Exposure, Hazard and Risk
Project Title
ACToR - Aggregated
Computational Toxicology
Resource
DSSTox- Distributed
Structure- Searchable
Toxicity Database Network
Outputs/Outcomes
FY09
• Initial public
deployment.
• Significant version 2,
including refined
chemical structure
information.
• Develop workflow for
tabularization of data
buried in text reports.
• Integrate all ToxCast
and ToxRefDB data.
• Quarterly releases with
new data.
• MS: SAR Perspective
of ToxCast 320 chemical
inventory.
• MSs: NCBI (GEO) and
FBI (Array Express)
structure-annotations
and linkages to
microarray data.
• Restart
Chemoinformatics CoP.
• Publish files for
ToxRefDB and ToxCast
inventories and selected
summary endpoints, and
facilitate publication and
linkage to PubChem.
Publish public genetic
toxicity data and SAR
predictions for ToxCast
320.
• Continue expansion of
DSSTox public toxicity
database inventory.
• Primary chemical
review and structure
annotation of
ToxCast/Tox21 libraries
within a central registry.
Outputs/Outcomes
FY10
• Quarterly releases with
new data.
• Implementation of a
process to gather tabular
data on priority chemicals
from text reports.
• Survey sources of
chemical use and
exposure data and import
any remaining sources.
• Develop flexible query
interface and data
download process.
• Develop process to
extract data from open
literature.
• Publish files for Tox21
inventory and selected
summary endpoints and
facilitate linkage to
PubChem.
• Publish files for NTP
study areas.
• Explore new approaches
to SAR based on feature
categories.
• Expand CEBS
collaboration to
incorporate DSSTox GEO
and ArrayExpress files
and create chemical
linkage to ILSI
Developmental Toxicity
database.
• Assist efforts within
ExpoCast regarding
chemically annotatation.
Outputs/Outcomes
FY11
• Quarterly releases with
new data.
• Establish procedures
and protocols for
automating chemical
annotation of new
experimental data
generated by NCCT and
in collaboration with
CEBSorNHEERL.
• Document and employ
PubChem analysis tools
in relation to published
DSSTox and ToxCast
data.
• Collaborate with SAR
modeling efforts to
predict ToxCast
endpoints.
• Continue expansion of
DSSTox public toxicity
database inventory for
use in modeling with co-
publication and linkage
to ACToR and
PubChem.
Outputs/Outcomes
FY12
• Quarterly releases with
new data.
• Redesign DSSTox
website to provide
hosting of donated
chemical descriptors,
properties and
predictions.
• Publish master tables
of DSSTox IDs and high
quality structures.
• Promote use of
chemical registry system
from ToxCast/Tox21
more broadly within EPA.
• Collaborate with SAR
modeling efforts to
expand modeling to
address Tox21
chemicals and
endpoints.
• Continue expansion of
DSSTox public toxicity
database inventory into
new toxicity and
exposure areas.
Expected Impacts
Enables access to cross-chemical data by EPA program offices,
NCCT, and other ORD organizations and external stakeholders.
Improves EPA data transparency.
Adoption of DSSTox chemical standards more broadly within
EPA and across other government Agencies to improve quality
and read-across capabilities.
Promote public data dissemination and encourage greater public
participation of industry and commercial sources in public toxicity
database and modeling efforts.
Facilitate improved toxicity prediction models and data mining
capabilities across wider span of endpoints and chemicals
impacting Hazard ID and risk assessment.
A-2.1
Previous
TOC
Next
-------
EPA CompTox Research Program FY2009-2012
IV. APPENDICES cont. -2. Project Outcomes Table
BOSC Review Draft- 24 August, 2009
Project Title
ToxRefDB
Chem Model -The
Application of Molecular
Modeling to Assessing
Chemical Toxicity
Outputs/O utcom es
FY09
• MSs: Chronic/cancer,
multigeneration and
developmental modules
• Release of stand-along
data entry tool.
• ToxRefDB webpage
online.
• Collection of ToxCast
Phase II chemical toxicity
data.
• Public release of
ToxRefDB web-based
query tool.
• Complete entry of
targeted set of chemicals
and study types for
Phase II of ToxCast.
• Complete reproductive
toxicity study
retrospective analysis.
• MS: Capability of the
target-toxicant paradigm
to identify chemicals that
bind weakly to the
estrogen receptor,
including a description of
the method.
• Description of the
library of 1 48 biological
macromolecule targets.
• MS: Molecular
modeling studies of the
potential biological
effects of the perfluoro
compounds.
Outputs/Outcomes
FY10
• Quarterly releases with
new data in conjunction
with ACToR.
• Implementation of a
process to gather and
enter open literature
studies.
• Expansion of ToxRefDB
to capture DNT studies
and EDSP data.
• Complete retrospective
analyses on other major
study types.
• Release of ToxRefDB
live data entry tool.
• MS: The metabolism of
pyrethroids and the effects
of three dimensional
chemical structures.
• Description of additional
targets added to the target
library.
• MS: The interaction of
Toxcast chemicals with
nuclear receptor targets
and the importance of
pharmacophore filters.
Outputs/Outcomes
FY11
• Quarterly releases with
new data in conjunction
with ACToR.
• MS:The integration of
results from the target
library and available
experimental
parameters.
• MS: The comparisons
of results with and
without pharmacophore
filter for Toxcast
chemicals.
Outputs/O utcom es
FY12
• Quarterly releases with
new data In conjunction
with ACToR.
• Comparison of results
using the target library
with experimental
determined activities.
• MS: Thel use of
molecular modeling and
the target toxicant
paradigm for regulatory
purposes, including a
discussion of the OECD
principles relative to
molecular modeling
specific principles.
Expected Impacts
Enables access of traditional toxicological data in a structured
and computable format extending the utility of the data beyond
chemical risk assessment and into broad research applications,
including guiding novel toxicity characterization methods and
transparent data-driven retrospective analyses leading to refined
animal use, a more predictive toxicology paradigm, and more
efficient chemical safety assessments.
An approach will be provided to prioritize chemicals for their
ability to influence the endocrine system by competing with
natural ligands for the binding sites of receptors. The application
of a method used to find strong acting drug like chemicals will be
evaluated for its capability of discovering weakly active
chemicals. The capability of these methods will be used for
finding weakly active chemicals. Parameters based on the
capacity of a chemical to interact with macromolecular targets for
toxicity will be available for applications of computational
methods to screen for or eventually predict chemical toxicity.
A-2.2
Previous
-------
EPA CompTox Research Program FY2009-2012
IV. APPENDICES cont. -2. Project Outcomes Table
BOSC Review Draft- 24 August, 2009
Project Title
T ox Cast™
ExpoCast™: Exposure
science for screening,
prioritization , andtoxicity
testing
Out puts/O utcom es
FYO9
• Completion of ToxCast
Phase 1.
• Provide Phase 1 data
sets to public.
• Derivation of predictive
signatures from ToxCast
Phase 1 data.
• MSs: ToxCast Phase 1
data sets.
• MS: Signature
generation
• MS: NR pathways and
toxicity.
- Convene first ToxCast
Data Summit for
identifying prediction
models.
• Finalize selection of
chemicals for Phase II of
ToxCast and Tox21 .
- ExpoCoP monthly
teleconference, ESC
resource, face-to-face
meeting at !SES 2009.
• MSs: EHP and ToxSci
on role of exposure in
the transforming
toxicology.
- SVOC workshop,
"Semi-Volatile Organic
Compounds (SVOCs) in
the Residential
Environment".
• Survey and identify
high priority exposure
data resources for initial
chemical indexing in
collaboration with
ACToR and DSSTox.
• ExpoCast conceptual
framework and research
plan.
Outputs/Outcomes
FY1O
• MSs: Describing
approaches to combining
exposure, PK, and in vitro
assays to do risk
prioritization .
• Evaluate compatibility of
nanomaterials of diverse
classes with ToxCast
assays.
• Prioritize and select
assays to be run for
ToxCast Phase II.
• Completion of ToxCast
Phase II data collection.
- Position paper
recommending standards
for exposure data
representation.
• White paper defining
exposure space and plan
for assessing exposure
data landscape.
• White paper exploring
development and
application of human
exposure knowledge base.
• Begin implementation of
standards across
exposure databases of
highest interest and utility
for NCCT projects.
• Award contracts.
Outputs/Outcomes
FY11
• Confirmation of
ToxCast predictive
signatures with Phase II
data.
• MSs: Signature
confirmations and
applications.
• MS: describing extant
data analyses to identify
critical determinants for
exposure classification
and chemical
prioritization based on
potential for exposure.
• Guide incorporation
and further development
of simple exposure
estimation tools within
the ACToR system for
use in prioritization.
Outputs/Outcomes
FY12
• MSs: Profiling of large
chemical sets for
potential for hazard.
• Apply exposure index
for prioritization to subset
of ToxCast compounds
to evaluate concept.
Expected Impacts
Program Offices will have tools to prioritize chemicals for targeted
toxicity testing and as well as insight on potential mechanisms of
toxicity.
Advance Agency tools for efficiently characterizing and
classifying chemicals based on potential for biologically-relevant
exposure; inform characterization of environmentally-relevant
toxicity.
A-2.3
Previous
-------
EPA CompTox Research Program FY2009-2012
IV. APPENDICES cont. -2. Project Outcomes Table
BOSC Review Draft- 24 August, 2009
Project Title
Virtual Embryo
(v-Embryo™ ) - The
Virtual Embryo Project
Virtual Liver -
(v-Liver1M)) - The Virtual
Liver Project
Uncertainty- Uncertainty
Analysis in Toxicological
Modeling
Outputs/O utcom es
FY09
• MS: application of VT-
KB to analyze ToxRefDB
developmental toxicity
studies. • VT-KB based
qualitative (structural)
model of self-regulating
ocular gene network.
• VT-SE based cell-
based computational
model of lens-retina
induction.
• MS: ocular
morphogenesis, gene
network inference,
analysis and modeling.
• Prioritize proof of
concept (PoC)
environmental chemicals
with clients. • Knowledge
Base (KB): Information
about PoC chemicals
using ToxCast assays.
• KB: Cytoscape KB
visualization and
analysis tool. • Cell
response: Initial
molecular circuit
describing hepatic cell
functions. • Tissue
Simulator: Develop /use
MAS framework.
• MS: Improvement of
computational efficiency
in Bayesian analyses of
PBPK models using
MCMC. • MS: Assessing
fit of PBPK models to
data using Bayesian or
other methods. • MS:
Reviewing approaches
for constructing priors for
PBPK model
parameters. • MS:
Estimation of model
parameters and
comparison of alternative
model forms for
permethin PBPK model.
• MS: Global sensitivity
analysis for exposure-
dose model for
permethrin. • Estimation
of PBPK parameters for
deltamethrin and two
other pyrethroids
complete.
Outputs/Outcomes
FY10
• Extend lens-retina model
to other stages and
species.
• Incorporate pathway data
from ToxCast, mESC and
ZF embryos.
• MS: Sensitivity analysis
for key biological
pathways.
• MS: Developmental
trajectories and
phenotypes in
computational models.
• Integrate with other
morphogenetic models
(ES cells, Zfish).
• Tissue Simulator: Test
liver lesions formation.
• Integrate molecular
circuits for MOA chemicals
in tissues.
• Evaluate simulator using
PoC chemicals and
ToxCast data to predict
outcomes.
• MS: permethrin PBPK •
MS on global sensitivity
analysis for permethrin
exposure-dose model
submitted. • MS:
Parameter estimation for
deltamethrin and two other
pyrethroids. • Completion
of exposure-dose-effect
model for 'mini-cumulative'
risk assessment.
• Substantial completion of
uncertainty analysis for
vLiver. • Description of
ontologically-aware PBPK
language and translation
to R. • Problem
identification for evaluating
alternative pathway
formulations in vLiver
model. • Problem
identification evaluating
alternative model
formulation for BBDR and
virtual embryo models.
Outputs/Outcomes
FY11
• MS: Test of model
against predictions for
pathway-based dose-
response relationship.
• MS: Uncertainty
analysis of models for
complex systems model.
• Model: Computer
program of early eye
development using
rules-based architecture,
cell-based simulators
and systems-wiring
diagrams.
• KB inference tod for
analyzing MOA for new
chemicals/mixtures.
• Extend lobule simulator
to liver and integrate
with PBPK model.
• SAP on 'mini-
cumulative' risk
assessment for
pyrethroids.
• Preparation for SAP for
cumulative risk
assessment for
pyrethroids.
• Publication of virtual
tissue and BBDR
modeling results
• Initial steps in
constructing test
datasets for analysis
"competition"
recommended by
UVPKM.
• Submission of ms on
"RDynamic"tothe
Journal of Statistical
Software.
Outputs/Outcomes
FY12
• MS: Integrated eye
morphogenesis.
• Integration of
computational models of
different systems.
• Evaluation of
integrative model with
data from ES cells, Zfish.
• Evaluate impact of
genomic variation on
cellular responses and
lesion formation.
• Evaluate v- Liver for
simulating human
pathology outcomes
using clinical data.
• Submission of ms on
'SemanticPK'.
• R package "R Dynamic",
for simplifying dynamic
modeling in the statistical
language R, submitted to
CRAN.
Expected Impacts
Framework for in silico reconstruction of the embryo to facilitate
navigation of complex relationships and predict systems-level
behavior (outcome) from data on biochemical, molecular and
cellular changes.
Proof-of-Concept decision support tools to enable tiered-testing:-
a) Reduce uncertainty in evaluating the effect of chemicals on
normal hepatic pathways.
b) Estimate dose-dependent adverse hepatic effects in
individuals and variability in populations.
Improved cumulative risk assessment for pyrethroid pesticides,
based on realistic quantitative assessment of uncertainties.
A-2.4
Previous
-------
EPA CompTox Research Program FY2009-2012 BOSC Review Draft- 24 August, 2009
IV. APPENDICES cont.
B. Extramural STAR Centers Projects
1. The Research Center for Environmental Bioinformatics and Computational Toxicology at the
University of Medicine & Dentistry of New Jersey (UMDNJ), Piscataway, brings together a
team of computational scientists with diverse backgrounds in bioinformatics, chemistry and
environmental science, from UMDNJ, Rutgers, and Princeton Universities, and the US Food and
Drug Administration's Center for Toxicoinformatics. The team is addressing multiple elements
of the source-to-outcome sequence for toxic pollutants as well as developing tools for toxicant
characterization. The computational tools developed through this effort will be extensively
evaluated and refined through collaboration between STAR Center scientists as well as with
colleagues from the three universities and the EPA. Particular emphasis is on methods that
enhance current risk assessment practices and reduce uncertainties. Researchers are also
developing a web accessible Environmental Bioinformatics Knowledge Base that will provide a
user-oriented interface to an extensive set of information and modeling resources.
2. The Carolina Environmental Bioinformatics Research Center at the University of North
Carolina, Chapel Hill, is developing new analytic and computational methods, creating efficient
user-friendly tools to disseminate the methods to the wider community, and applying the
computational methods to molecular toxicology and other studies. The Center brings together
multiple investigators and disciplines, combining expertise in biostatistics, computational
biology, chemistry, and computer science to advance the field of Computational Toxicology.
Researchers focus on providing biostatistician support to the Center by performing analyses and
developing new methods in Computational Biology. The Center is also creating a framework for
merging data from various technologies in a systems-biology approach.
3. Carolina Center for Computational Toxicology at the University of North Carolina, Chapel
Hill, University of North, will advance the field of computational toxicology through the
development of new methods and tools, as well as through collaborative efforts. The Center is
utilizing a bottom-up approach to predictive computational modeling of adverse effects of toxic
agents. The emphasis spans from the fine-scale predictive simulations of the protein-protein/-
chemical interactions in nuclear receptor networks, to mapping chemical-perturbed networks and
devising modeling tools that can predict the pathobiology of the test compounds based on a
limited set of biological data, to building tools that will enable toxicologists to understand the
role of genetic diversity between individuals in responses to toxicants, to unbiased discovery-
driven prediction of adverse chronic in vivo outcomes based on statistical modeling of chemical
structures, high-throughput screening and the genetic makeup of the organism. In each project,
new computer-based models will be developed and published that represent the state-of-the-art.
The tools produced within each project will be widely disseminated, and the emphasis will be
placed on their usability by the risk assessment community and the investigative toxicologists
alike. The synthesis of data from a variety of sources will move the field of computational
toxicology from a hypothesis-driven science toward a predictive science.
4. Texas-Indiana Virtual STAR Center; Data-Generating in vitro and in silico Models of
Developmental Toxicity in Embryonic Stem Cells and Zebrafish at the University of Houston,
Texas A&M Institute for Genomic Medicine, and Indiana University. Project Period:
Project start: November 1, 2009
Project end: October 31, 2012
B-l
Previous I TOC
-------
EPA CompTox Research Program FY2009-2012 BOSC Review Draft- 24 August, 2009
IV. APPENDICES cont.
Description
Objectives/Hypothesis:
As chemical production increases worldwide, there is increasing evidence as to their hazardous
effects on human health at today's exposure levels, which further implies that current chemical
regulation is insufficient. Thus, a restructuring of the risk assessment procedure will be required
to protect future generations. Given the very large number of man-made chemicals and the likely
complexity of their various and synergistic modes of action, emerging technologies will be
required for the restructuring. The main objective of the proposed multidisciplinary Texas
Indiana Virtual STAR (TIVS) Center is to contribute to a more reliable chemical risk assessment
through the development of high throughput in vitro and in silico screening models of
developmental toxicity. Specifically, the TIVS Center aims to generate in vitro models of murine
embryonic stem cells and zebrafish for developmental toxicity. The data produced from these
models will be further exploited to produce predictive in silico models for developmental
toxicity on processes that are relevant also for human embryonic development.
Approach:
The project is divided into three Investigational Areas; zebrafish models, murine embryonic stem
cells models and in silico simulations. The approaches are to:
1. Generate developmental models suitable for high throughput screening. Zebrafish
developmental models (transgenic GFP/EGFP/RFP models of crucial steps in
development) and embryonic stem cell (ESC) differentiation models (transgenic beta-geo
models of crucial steps in differentiation) will be generated. Important morphology
features and signaling pathways during development will be documented. The impact of
environmental pollutants on development and differentiation will be assessed in the
models. Finally, the models will be refined for high throughput screening and
automation.
2. Generate a computational model that faithfully recreates the major morphological
features of normal wild-type zebrafish development (ie- segmentation into somites,
proper patterning of vascular and neural systems) and the differentiation to three
primitive layers (endoderm, mesoderm and ectoderm) in mouse embryonic stem cells.
The data for simulations are produced from developed high information content zebrafish
and ESC models. Once a working model of normal development has been generated, we
will carry out a directed series of parameter sweeps to try to create developmental defects
in silico. We will compare the results of computationally created defects with
experimentally-generated defects in zebrafish and embryonic stem cells. Best matches
between the two datasets will suggest hypotheses about possible mechanisms by which
defects occur.
3. Perform proof-of-concept experiments of the in vitro and in silico test platforms with a
blind test of chemicals.
Techniques will be molecular biology techniques on zebrafish and ESC models, such as cloning,
imaging, in vitro differentiation and in vitro exposure studies, and in silico mathematical
simulations.
Expected Results (Outputs/Outcomes):
B-2
Previous I TOC
-------
EPA CompTox Research Program FY2009-2012
IV. APPENDICES cont.
BOSC Review Draft- 24 August, 2009
In collaboration with other initiatives taken in the field of chemical safety, our generated results
and models will contribute to large screening effort to prioritize chemicals for further risk
assessment. We will specifically contribute with:
• 9 transgenic fish lines validated for toxicity screening
• 16 embryonic stem cell models validated for toxicity screening
• High information content models on development and differentiation to produce data for
in silico simulations, within the project and elsewhere
• Computational models for developmental toxicology of normal development and of
mechanisms by which chemical perturbations cause experimentally-observed
developmental defects
• Information on developmental toxicity on 39 compounds.
Previous
B-3
TOC
-------
EPA CompTox Research Program FY2009-2012 BOSC Review Draft- 24 August, 2009
IV. APPENDICES cont.
C. FY2004 "New Start" Award Bibliography
Project Title: Linkage of Exposure and Effects Using Genomics, Proteomics, and Metabolomics in
Small Fish Models
Peer Reviewed Publications:
Ankley, G.T., K.M. Jensen, E.J. Durban, E.A. Makynen, B.C. Butterworth, M.D. Kahl, D.L. Villeneuve, A.
Linnum, L.E. Gray, M. Cardon, V.S. Wilson. 2005. Effects of two fungicides with multiple modes of action
on reproductive endocrine function in the fathead minnow (Pimephales promelas). Toxicol. Sci. 86, 300-
308.
Ankley, G.T., K.M. Jensen, M.D. Kahl, E.A. Makynen, L.S. Blake, K.J. Greene, R.D. Johnson and D.L.
Villeneuve. 2007. Ketoconazole in the fathead minnow (Pimephlaes promelas): reproductive toxicity and
biological compensation. Environ. Toxicol. Chem. 26, 1214-1223.
Ankley, G.T., D.H. Miller, K.M. Jensen, D.L. Villeneuve and D. Martinovic. 2008.
Relationship of plasma sex steroid concentrations in female fathead minnows to reproductive success and
population status. Aquat. Toxicol. 88, 69-74.
Ankley, G.T., D. Bencic, M. Breen, T.W. Collette, R. Connolly, N.D. Denslow,
S. Edwards, D.R. Ekman, K.M. Jensen, J. Lazorchak, D. Martinovic, D.H. Miller, E.J.
Perkins, E.F. Orlando, N. Garcia-Reyero, D.L. Villeneuve, R.-L.Wang , and K.
Watanabe. 2009. Endocrine disrupting chemicals in fish: Developing exposure
indicators and predictive models of effects based on mechanisms of action. Aquat.
Toxicol. 92, 168-178.
Breen, M.S., D.L. Villeneuve, M. Breen, G.T. Ankley and R.B. Conolly. 2007.
Mechanistic computational model of ovarian steroidogenesis to predict biochemical responses to endocrine
active compounds. Ann. Biomed. Engin. 35, 970-981.
Ekman, D.R., Q. Teng, K.M. Jensen, D. Martinovic, D.L. Villeneuve, G.T. Ankley and T. W. Collette.
2007. NMR analysis of fathead minnow urinary metabolites: a potential approach for studying impacts of
chemical exposures. Aquat. Toxicol. 85, 104-112.
Ekman, D.R., Q. Teng, D.L. Villeneuve, M.D. Kahl, K.M. Jensen, E.J. Durban, G.T. Ankley and T.W.
Collette. 2008. Investigating compensation and recovery of fathead minnow (Pimephales promelas)
exposed to 17a-ethynylestradiol with metabolite profiling. Environ. Sci. Tecnhol. 42, 4188-4195.
Ekman, D.R., Q. Teng, D.L. Villeneuve, M.D. Kahl, K.M. Jensen, E.J. Durban, G.T. Ankley and T.W.
Collette. 2009. Profiling lipid metabolites yields unique information on gender- and time-dependent
responses of fathead minnows (Pimephales promelas) exposed to 17a-ethynylestradiol. Metabolomics 5, 22-
32.
C-l
Previous I TOC
-------
EPA CompTox Research Program FY2009-2012 BOSC Review Draft- 24 August, 2009
IV. APPENDICES cont.
Garcia-Reyero, N., D.L. Villeneuve, K.J. Kroll, L. Liu, E.F. Orlando, K.H. Watanabe, M.S. Sepulveda, G.T.
Ankley and N.D. Denslow. 2009. Expression signatures for a model androgen and antiandrogen in the
fathead minnow ovary. Environ. Sci. Technol. 43, 2614-2619.
Garcia-Reyero, N., K.J. Kroll, L. Liu, E.F. Orlando, K.H. Watanabe, M.S. Sepulveda, D.L. Villeneuve, E.J.
Perkins, G.T. Ankley and N.D. Denslow. 2009. Gene expression responses in male fathead minnows
exposed to binary mixtures of an estrogen and antiestrogen. BMC Genomics, In Press.
Johns, S.M., M.D. Kane, N.D. Denslow, K.H. Watanabe, E.F. Orlando, D.L. Villeneuve,
G.T. Ankley and M.S. Sepulveda. 2009. Characterization of ontogenetic changes in gene expression in the
fathead minnow (Pimephalespromelets). Environ. Toxicol. Chem. 28, 873-880.
Martyniuk, C.J., S. Alvarez, S. McClung, D.L. Villeneuve, G.T. Ankley and N.D. Denslow.2009.
Quantitative proteomic profiles of androgen receptor signaling in the liver of fathead minnows (Pimephales
promelas) J. Proteome Res. In Press.
Martinovic, D., L.S. Blake, E.J. Durhan, K.J. Greene, M.D. Kahl, K.M. Jensen, E.A. Makynen, D.L.
Villeneuve and G.T. Ankley. 2008. Characterization of reproductive toxicity of vinclozolin in the fathead
minnow and co-treatment with an androgen to confirm an anti-androgenic mode of action. Environ. Toxicol.
Chem. 27, 478-488.
Miller, D.H., K.M. Jensen, D.L. Villeneuve, M.D. Kahl, E.A. Makynen, E.J. Durhan and G.T. Ankley.
2007. Linkage of biochemical responses to population-level effects: a case study with vitellogenin in the
fathead minnow (Pimephlaes promelas). Environ. Toxicol. Chem. 26, 521-527.
Perkins, E.J., N. Garcia-Reyero, D.L. Villeneuve, D. Martinovic, S.M. Brasfield, L.S.
Blake, J.D. Brodin, N.D. Denslow and G.T. Ankley. 2008. Perturbation of gene
expression and steroidogenesis with in vitro exposure of fathead minnow ovaries to
ketoconazole. Mar. Environ. Res. 66, 113-115.
Villeneuve, D.L., P. Larkin, I. Knoebl, A.L. Miracle, M.D. Kahl, K.M. Jensen, E.A. Makynen, E.J. Durhan,
B.J. Carter, N.D. Denslow and G.T. Ankley. 2007. A graphical systems model to facilitate hypothesis-
driven ecotoxicogenomics research on the brain-pituitary-gonadal axis. Environ. Sci. Technol. 40, 321-330.
Villeneuve, D., L. Blake, J. Brodin, K. Greene, I. Knoebl, A. Miracle, D. Martinovic and G.T. Ankley.
2007. Transcription of key genes regulating gonadal steroidogenesis in control and ketoconazole- or
vinclozolin-exposed fathead minnows. Toxicol. Sci. 98, 395-407.
Villeneuve, D.L., L.S. Blake, J.D. Brodin, J.E. Cavallin, E.J. Durhan, K.M. Jensen, M.D. Kahl, E.A.
Makynen, D. Martinovic, N.D. Mueller and G.T. Ankley. 2008. Effects of a 3p-hydroxysteroid
dehydrogenase inhibitor, trilostane, on the fathead minnow reproductive axis. Toxicol. Sci. 104, 113-123.
Villeneuve, D.L., N.D. Mueller, D. Martinovic, E.A. Makynen, M.D. Kahl, K.M. Jensen, E.J. Durhan, J.E.
Cavallin, D. Bencic and G.T. Ankley. 2009. Direct effects, compensation and recovery in female fathead
minnows exposed to a model aromatase inhibitor. Environ. Health Perspect. 117, 624-631.
Villeneuve, D.L., R.-L. Wang, D.C. Bencic, A.D. Biales, D. Martinovic, J.M.
C-2
Previous I TOC
-------
EPA CompTox Research Program FY2009-2012 BOSC Review Draft- 24 August, 2009
IV. APPENDICES cont.
Lazorchak, G. Toth and G.T. Ankley. 2009. Altered gene expression in the brain and ovaries of zebrafish
exposed to the aromatase inhibitor fadrozole: microarray analysis and hypothesis generation. Environ.
Toxicol. Chem. In Press.
Wang, R.-L., A. Biales, D. Bencic, D. Lattier, M. Kostich, D. Villeneuve, G.T. Ankley, J. Lazorchak and G.
Toth. 2008a. DNA microarray application in ecotoxicology: experimental design, microarray scanning, and
factors impacting transcriptional profiles in a small fish species. Environ. Toxicol. Chem. 27, 652-663.
Wang, R.-L., D. Bencic, A. Biales, D. Lattier, M. Kostich, D. Villeneuve, G.T. Ankley, J. Lazorchak and G.
Toth. 2008b. DNA microarray-based ecotoxicological discovery in a small fish specbs. Environ. Toxicol.
Chem. 27, 664-675.
Watanabe, K.H., K.M. Jensen, E.F. Orlando and G.T. Ankley. 2007. What is
normal? A characterization of the values and variability in reproductive endpoints of the fathead minnow,
Pimephales promelas. Comp. Biochem. Physiol. 146, 348-356.
Watanabe, K.H., Z. Li, K. Kroll, D.L. Villeneuve, N.J. Szabo, E.F. Orlando, M.S.
Sepulveda, T.W. Collette, D.R. Ekman, G.T. Ankley and N.D. Denslow. 2009. A physiologically-based
model of endocrine-mediated responses of male fathead minnows to!7a-ethinylestradiol. Toxicol. Sci. In
Press.
Project Title: Simulating Metabolism of Xenobiotic Chemicals as a Predictor of Toxicity
Peer Reviewed Publications:
Mazur, C. S.; Kenneke, J. F. 2008. Cross-species comparison of conazole fungicide metabolites using rat and
rainbow trout (Onchorhynchus mykiss) hepatic microsomes and purified human CYP 3A4. Environmental
Science and Technology, 42:947-954.
Mazur, C. S.; Kenneke, J.F.; Tebes-Stevens, C. Okino, M. S.; Lipscomb, J. C. 2007. In vitro metabolism of
the fungicide and environmental contaminant trans-bromuconazole and implications for risk assessment.
Journal of Toxicology and Environmental Health, Part A, 70:1241-1250.
Kenneke, J. F. 2006. Environmental fate and ecological risk assessment for the reregistration of antimycin A
(PC Code 006314), Appendix D: In vitro mammalian metabolism and Appendix G: Summary of antimycin
hydrolysis research, U.S. EPA, Office of Pesticide Programs, Reregistration Eligibility Decision (RED) on
Antimycin A
Project Title: Risk Assessment of the Inflammogenic and Mutagenic Effects of Diesel Exhaust
Particulates: A Systems Biology Approach
Peer Reviewed Publications:
Cao, D., Bromberg, PA and Samet, JM (2007). Diesel-induced Cox-2 expression involves chromatin
modification via degradation of HDAC1 and recruitment of p300. Am. J. Respir. Cell. Mol. Biol.37:232-239.
Cao, D., Tal, T., Graves, L., Gilmour, I., Linak, W., Reed, W., Bromberg, P., and Samet, J., Diesel Exhaust
Particulate (DEP)-Induced Activation of Stat3 Requires Activities of EGFR and SRC in Airway Epithelial
Cells, American Journal of Physiology: Lung, Cell, & Molecular Physiology, 292, L422-L429 (2007).
C-3
Previous I TOC
-------
EPA CompTox Research Program FY2009-2012 BOSC Review Draft- 24 August, 2009
IV. APPENDICES cont.
Cho, S.-H., Yoo, J.-L, Turley, A.T., Miller, C.A., Linak, W.P., Wendt, J.O.L., Muggins, F.E., and Gilmour,
M.I., Relationships between Composition and Pulmonary Toxicity of Prototype Particles from Coal
Combustion and Pyrolysis, Proceedings of the Combustion Institute, 32, in press (2008).
Ciencewicki, J., Gowdy, K., Krantz, Q.T., Linak, W.P., Brighton, L., Gilmour, M.I., and Jaspers, I., Diesel
Exhaust Enhanced Susceptibility to Influenza Infection is Associated with Decreased Surfactant Protein
Expression, Inhalation Toxicology, 19, 1121-1133 (2007).
DeMarini, D.M., Brooks, L.R., Warren, S.H., Kobayashi, T., Gilmour, M.I., and Singh P., Bioassay-Directed
Fractionation an dSalmonella Mutagenicity of Automobile and Forklift Diesel Exhaust Particles,
Environmental Health Perspectives, 112, 814-819 (2004).
Gottipolu, R.R., Wallenborn, J.G., Karoly, E.D., Schladweiler, M.C., Ledbetter, A.D., Krantz, Q.T., Linak,
W.P., Nyska, A., Johnson, J.A., Thomas, R., Richards, J.E., Jaskot, R.H., and Kodavanti, U.P. (2009)., One-
month Diesel Exhaust Inhalation Produces Hypertensive Gene Expression Pattern in Healthy Rats,
Environmental Health Perspectives. 117:38-46.
Gowdy, K., Krantz, Q.T., Daniels, M., Linak, W.P., Jaspers, I., and Gilmour, M.I., Modulation of Pulmonary
Inflammatory Responses and Anti-microbial Defenses in Mice Exposed to Diesel Exhaust, Toxicology &
Applied Pharmacology, 229, 310-319 (2008).
Linak, W.P., Yoo, J.I., Wasson, S.J., Zhu, W., Wendt, J.O.L., Huggins, F.E., Chen, Y., Shah, N., Huffman,
G.P., and Gilmour, M.I., Ultrafine Ash Aerosols from Coal Combustion: Characterization and Health
Effects, Proceedings of the Combustion Institute, 31, 1929-1937 (2007).
Reed, W., Gilmour, I., DeMarini, D., Linak, W. and Samet, J. (2008). Gene Expression Profiles of Human
Airway Epithelial Cells Exposed to Diesel Exhaust Particles of Varying Composition. In Preparation.
Saxena, RK, Williams, W & Gilmour, MI. (2007) Suppression of basal and cytokine induced expression of
MHC, ICAM 1 and B7 markers on mouse lung epithelial cells exposed to diesel exhaust particles. Am J
Biochem Biotech. 3(4). 187-192.
Saxena, RK., Gilmour, ML, & MD Hayes. Uptake of diesel exhaust particles by lung epithelial cells and
alveolar macrophages. Biotechnology, 2007, 3 (4). 187-192
Singh, P., DeMarini, D.M., Dick, C.A.J., Tabor, D., Ryan, J., Linak, W.P., Kobayashi, T., and Gilmour, M.I.,
Bioassay-Directed Fractionation, Physiochemical Characterization, and Pulmonary Toxicity of Automobile
and Forklift Diesel Exhaust Particles in Mice, Environmental Health Perspectives, 112(8) 820-825 (2004).
Stevens, T., Krantz, Q.T., Linak, W.P., Hester, S., and Gilmour, M.I., Increased Transcription of Immune
and Metabolic Pathways in Naive and Allergic Mice Exposed to Diesel Exhaust, Toxicological Sciences,
102(2), 359-370 (2008).
Stevens, T., Linak, WP., Gilmour, MI. Differential potentiation of allergic lung disease in mice exposed to
chemically distinct diesel samples. Tox Sci. 107(2), 522-534.
C-4
Previous I TOC
-------
EPA CompTox Research Program FY2009-2012 BOSC Review Draft- 24 August, 2009
IV. APPENDICES cont.
Stevens,!, Hester, S, & Gilmour MI. Differential transcriptional changes in mice exposed to chemically
distinct diesel samples. Submitted.
Tal., T., Bromberg, P.A., Kim, Y. and Samet, J.M. (2008). Tyrosine phosphatase inhibition induces
epidermal growth factor receptor activation in human airway epithelial cells exposed to diesel exhaust
Toxicol. Appl. Pharmacol. 233:382-388.
Project Title: Development of Microbial Metagenomic Markers for Environmental Monitoring and
Risk Assessment
Peer Reviewed Publications:
Lamendella R, Santo Domingo JW, Yannarell AC, Ghosh S, Di Giovanni G, Mackie RI, Oerther DB.
Evaluation of swine-specific PCR assays used for fecal source tracking and analysis of molecular diversity of
Bacteriodales-swine specific populations. Appl Environ Microbiol. 2009 Jul 24. [Epub ahead of print]
Lu J, Santo Domingo JW, Hill S, Edge TA. Microbial Diversity and Host-specificSequences of Canadian
Goose Feces. Appl Environ Microbiol. 2009 Jul 24.
Santo Domingo, J.W. and T.A. Edge. 2009. Identification of primary sources of faecal pollution. In Safe
Management of Shellfish and Harvest Waters. G. Rees., K. Pond, D. Kay and J. Santo Domingo. IWA
Publishing, London, UK.
Lee, Y.-J., M. Molina, and J.W. Santo Domingo, J.D. Willis, M. Cyterski, D.M. Endale, and O.C. Shanks.
2008. A temporal assessment of cattle fecal pollution in two watersheds using 16S rRNA gene-based and
metagenome-based assays. Appl. Environ. Microbiol. 74:6839-6847.
Lu, J. and J.W. Santo Domingo. 2008. Turkey fecal microbial community structure and functional gene
diversity revealed by 16S rRNA gene and metagenomic sequences. J. Microbiol. 46:469-477.
Lu, J., J.W. Santo Domingo, R. Lamendella, T.Edge, and S.Hill. 2008. Phylogenetic diversity and molecular
detection of gull feces. Appl. Environ. Microbiol. 74: 3969-3976.
Lamendella, R., Santo Domingo J.W., Kelty C, and Oerther DB. 2008. Occurrence of bifidobacteria in feces
and environmental waters. Appl. Environ. Microbiol. 74:575-584.
Santo Domingo, J.W., D.G. Bambic, T.A. Edge, and S. Wuertz. 2007. Quo vadis source tracking? Towards a
strategic framework for environmental monitoring of fecal pollution. Water Res. 41:3539-3552.
Lu, J., J.W. Santo Domingo, and O.C. Shanks. 2007. Identification of chicken-specific fecal microbial
sequences using a metagenomic approach. Water Res. 41:3561-3574.
Shanks, O., J.W. Santo Domingo, J. Lu, C.A. Kelty, and J. Graham. 2007. PCR Assays for the identification
of human fecal pollution in water. Appl. Environ. Microbiol. 73: 2416-2422.
Vogel, J.R., D.M. Stoeckel, R. Lamendella, R.B. Zelt, J.W. Santo Domingo, S.R. Walker, and D.B. Oerther.
2007. Identifying fecal sources in a selected catchment reach using multiple source-tracking Tools. J.
Environ. Qual. 36:718-729.
C-5
Previous I TOC
-------
EPA CompTox Research Program FY2009-2012 BOSC Review Draft- 24 August, 2009
IV. APPENDICES cont.
Lamendella, R., J. W. Santo Domingo, D. Oerther, J. Vogel, and, D. Stoeckel. 2007. Assessment of fecal
pollution sources in a small northern-plains watershed using PCR and phylogenetic analyses of Bacteroidetes
16S rDNA. FEMS Microbiol. Ecol. 59:651-660.
Santo Domingo, J.W., Lu, J., Shanks, O., Lamendella, R., Kelty, C. A., and Oerther, D.B. "Development of
host-specific markers for source tracking using a novel metagenomic approach," Water Environment
Federation, Proceedings of Disinfection 2007, Pittsburg, PA, February 4-7, 2007.
Shanks, O., J. W. Santo Domingo, R. Lamendella, C.A. Kelty, and J. Graham. 2006. Competitive
metagenomic DNA hybridization identifies host-specific genetic markers in cattle fecal samples. Appl.
Environ. Microbiol. 72:4054-4060.
Shanks, O., J. W. Santo Domingo, and J. Graham. 2006. Use of competitive DNA hybridization to identify
differences in the genomes of two closely related fecal indicator bacteria. J. Microbiol. Methods. 66:321-330.
Project Title: A Systems Approach to Characterizing and Predicting Thyroid Toxicity Using an
Amphibian Model
Peer Reviewed Publications:
Sternberg, Thoemke, Hornung, Tietge, and Degitz. Regulation of thyroid-stimulating hormone release from
pituitary by t4 during metamorphosis in Xenopus laevis (In review)
Serrano, Higgins Witthuhn, Korte, Hornung, Tietge, and Degitz In vivo assessment and potential diagnosis
of xenobiotics that perturb the thyroid pathway: Part I. Differential protein profiling of Xenopus Laevis brain
tissues by two-dimensional polyacrylamide gel electrophoresis and peptide-labeling with isobaric tags for
relative and absolute quantification (iTRAQ) following exposure to model T4 inhibitors. (In review)
Conners, K. , Jorte, J.J., Anderson G., Degitz SJ. Charaterization of thyroid hormone transporting protein
expression during tissue-specific metamorphosis in Xenopus tropicalis (In review)
Hornung, M.W. Degitz, S.J., Korte, L.M., Olson, J., Kosian, P.A., Linnum, A.L., Tietge, J.E. Inhibition of
thyroid hormone release from cultured amphibian thyroid glands by methimazole, 6-propylthiouracil, and
perchlorate. (Completed, NHEERL In-House review. To be submitted with the following Hornung et al.
paper).
Degitz, S.J., Hornung, M.W., Korte, J.J, Holcombe, G.W, Kosian, P.A., Thoemke, K.R., Helbing, C., Tietge,
J.E. In vivo and in vitro regulation of genes in the thyroid gland following exposure to the model T4
synthesis inhibitors methimazole, 6-propylthiouracil, and perchlorate. (In preparation. To be submitted with
above paper).
Hornung, M., Burgess, E. Tandem in vitro and ex vivo thyroid gland assays to screen xenobiotic chemicals
for thyroid hormone synthesis inhibition. (In preparation).
Nichols systems model paper (In preparation).
C-6
Previous I TOC
-------
EPA CompTox Research Program FY2009-2012 BOSC Review Draft- 24 August, 2009
IV. APPENDICES cont.
Tietge Butterworth Kosian Hammermeister Hornung Haselman Degitz Analysis of Thyroid Hormone and
Related lodo-Compounds in Complex Samples by Inductively Coupled Plasma Emission/Mass
Spectrometry. (In preparation)
Tietge, Butterworth, Haselman, Holcombe, Korte, Kosian, Wolfe, Degitz. Early Temporal Effects of Three
Thyroid Hormone Synthesis Inhibitors in Xenopus laevis. (In preparation)
Project Title: Mechanistic Indicators Of Childhood Asthma (Mica): A Systems Biology Approach For
The Integration Of Multifactorial Environmental Health Data
Peer Reviewed Publications:
Kim SJ, Dix DJ, Thompson KE, Murrell RN, Schmid JE, Gallagher JE, Rockett JC.
Effects of storage, RNA extraction, genechip type, and donor sex on gene expression profiling of human
whole blood. Clin Chem. Jun;53(6): 1038-45. (2007)
Vesper, S.,McKinstry C., Haugland., R., Neas, L., Hudgens, E., Heidenfelder, B., and Gallagher J.
Environmental Relative Moldiness Index (ERMIsm) as a Tool to Identify Mold Related Risk Factors for
Childhood Asthma Sci Total Environ. May 1;394(1): 192-6 (2008)
Johnson M, Hudgens E, Williams R, Andrews G, Neas L, Gallagher J, Ozkaynak H. "A Participant-Based
Approach to Indoor/Outdoor Air Monitoring in Community Health Studies" Journal of Exposure Science
and Environmental Epidemiology. (2008), 1-10 (2008).
Cohen Hubal E, Richards A., Shah I, Edwards S, Gallagher J, Kavlock R, Blancato, J Exposure Science and
the US EPA National Center for Computational Toxicology J Expo Sci Environ Epidemiol. November
(2008).
Heidenfelder B,. ReifD, Harkema, JR, Cohen Hubal E, Hudgens,E. Bramble L G. Wagner G, Harkema
JR, Morishita M, Keeler G, Edwards,SW and Gallagher J. Comparative Microrarray Analysis and
Pulmonary Changes in Brown Norway Rats Exposed to Ovalbumin and concentrated Air Particulates Tox
Sci. volume 108 2009 March 2 (2009)
Heidenfelder B, Johnson M, Hudgens E, Inmon J, Hamilton R, Neas L, and Gallagher J, Increased plasma
reactive oxidant levels and their relationship to blood cells, total IgE, and allergen-specific IgE in asthmatic
children Journal of Asthma accepted (2009) Williams AH, Gallagher JE, Hudgens E, Johnson MM,
Mukerjee S, Ozkaynak H, Neas LMN. EPA Observational studies of children's respiratory health in Detroit
and Dearborn, Michigan. Proceedings of AWMA 102nJune 16-19; Detroit, Michigan.(2009)
J. E Gallagher, E A Cohen Hubal, S.W. Edwards Invited book Chapter "Biomarkers of Environmental
Exposure" "Biomarkers of toxicity: A New Era in Medicine Editors Vishal S. Vaidya and Joseph V.
Bonventre Publisher:John Wiley and Sons, Inc. October 1, (2009)
Markey M. Johnson, Ron Williams, Zhihua Fan, Lin, Edward Hudgens, Jane Gallagher, Alan Vette, Lucas
Neas, Haluk Ozkaynak Indoor and outdoor concentrations of nitrogen dioxide, volatile organic compounds,
and polycyclic aromatic hydrocarbons among MICA-Air households in Detroit, Michigan submitted AWMA
(2009)
C-7
Previous I TOC
-------
EPA CompTox Research Program FY2009-2012
IV. APPENDICES cont.
BOSC Review Draft- 24 August, 2009
Gallagher, J Reif, D; Heidenfelder, B Neas, L; Hudgens, E Williams, A Inmon, J; Rhoney, S, Andrews G.,
Johnson, M Ozkaynak, H; Edwards, S, Cohen-Hubal, E Mechanistic Indicators of Childhood asthma (
MICA); A systems biology approach for the integration of multifactorial environmental health data
submitted: Journal of Exposure Science and Environmental Epidemiology (2009).
Previous
C-8
TOC
-------
EPA CompTox Research Program FY2009-2012
IV. APPENDICES cont.
D. EPA Strategic Plan for Evaluating the Toxicity of Chemicals
BOSC Review Draft- 24 August, 2009
EPA/100/K-09/001 I March 2009
www.epa.gov/osa
United States
Environmental Protection
Agency
The U.S. Environmental Protection
Agency's Strategic Plan for
Evaluating the Toxicity of Chemicals
Chemicals
Receptors / Enzymes / etc.
Direct Molecular Interaction
Pathway Regulation /
Genomics
Cellular Processes
Tissue / Organ / Organism Tox Endpoint
Office of the Science Advisor
Science Policy Council
Previous
D-l
TOC
-------
EPA CompTox Research Program FY2009-2012
IV. APPENDICES cont.
BOSC Review Draft- 24 August, 2009
PA 100/K-09/001
March 2009
The U.S. Environmental Protection Agency's
Strategic Plan for Evaluating
the Toxicity of Chemicals
Office of the Science Advisor
Science Policy Council
U.S. Environmental Protection Agency
Washington, DC 20460
Recyeted/Recyclatole
Pnnted with vegetab^-basKJ ink on paper thai
contains a m^imum of 50% pa$£-c
ftb@f and is pr
-------
EPA CompTox Research Program FY2009-2012
IV. APPENDICES cont.
BOSC Review Draft- 24 August, 2009
DISCLAIMER
Mention of trade names or commercial products does not constitute endorsement or
recommendation for use. Notwithstanding any use of mandatory language such as "must" and
"require" in this document with regard to or to reflect scientific practices, this document does not
and should not be construed to create any legal rights or requirements.
Previous
TOC
-------
EPA CompTox Research Program FY2009-2012 BOSC Review Draft- 24 August, 2009
IV. APPENDICES cont.
AUTHORS AND CONTRIBUTORS
Future of ToxicMy Testing Workgroup Co-Chairs
Michael Firestone, Office of Children's Health Protection and Environmental Education, U.S.
EPA
Robert Kavlock. Office of" Research and Development. U.S. EPA
Hal Zenick, Office of Research and Development, U.S. EPA
Science Policy Council Staff
Melissa Kramer, Office of the Science Advisor, U.S. EPA
Future of Toxieity Testing Workgroup Representatives
Marcia Bailey, Region 10, U.S. EPA
Arden Calvert, Office of the Chief Financial Officer. U.S. EPA
Laurel Celeste. Office of General Counsel, U.S. EPA
Vicki Dellarco, Office of Prevention. Pesticides, and Toxic Substances. U.S. EPA
Scott Jenkins, Office of Air and Radiation, U.S. EPA
Gregory Miller, Office of Policy, Economics, and Innovation, U.S. EPA
Nicole Paquette. Office of Environmental Information. U.S. EPA
Santhini Ramasamy, Office of Water. U.S, EPA
William Sette. Office of Solid Waste and Emergency Response. U.S. EPA
Other Contributors
Kathcrinc Anitole. Office of Prevention. Pesticides, and Toxic Substances. U.S. EPA
Hugh B;ii1on, Office of Research and Development, U.S. EPA
Norman Birchfield. Office of the Science Advisor, U.S. EPA
Michael Brody, Office of the Chief Financial Officer, I I.S. EPA
Rory Conolly. Office of Research and Development. U.S. EPA
David Dix. Office of Research and Development. U.S. EPA
Stephen Edwards, Office of Research and Development, U.S. EPA
Andrew Geller. Office of Research and Development, II.S. EPA
Karen Ilarncmik. Office of Prevention. Pesticides, and Toxic Substances, U.S. EPA
Jean Holmes, Office of Prevention, Pesticides, and Toxic Substances, I ",S. EPA
Richard Judson. Office of Research and Development. U.S. EPA
Thomas Knudsen, Office of Research and Development, U.S. EPA
Julian Preston, Office of Research and Development, U.S. EPA
Kathleen RalTaele, Office of the Science Advisor, U.S. EPA
Ram Ramabhadran, Office of Research and Development, U.S. EPA
James Samet, Office of Research and Development, U.S. EPA
Patricia Schmieder, Office of Research and Development, t LS. EPA
Banalata Sen, Office of Prevention, Pesticides, and Toxic Substances, U.S. EPA
Imran Shah. Office of Research and Development, U.S. EPA
Linda Sheldon. Office of Research and Development. U.S. EPA
Previous I TOC
-------
EPA CompTox Research Program FY2009-2012
IV. APPENDICES cont.
BOSC Review Draft- 24 August, 2009
John Vandenberg, Office of Research and Development. U.S. EPA
Maurice 7,eeman, Office of Prevention, Pesticides, and Toxic Substances, U.S. KPA
External Peer Reviewers
John R, Bucher, Ph.D., Associate Director, National Toxicology Program, National Institute of
Environmental Health Sciences
George Daston, Ph.D., Research Fellow, P&G
Daniel Krewski, Ph.D.. MHA, Professor and Director, Mclaughlin Centre for Population Health
Risk Assessment, I Iniversity of Ottawa
Martin Stephens, Ph.D., Vice President for Animal Research Issues, "Hie Humane Society of the
United States
Previous
TOC
-------
EPA CompTox Research Program FY2009-2012 BOSC Review Draft- 24 August, 2009
IV. APPENDICES cont.
TABLE OF CONTENTS
LIST OF FIGURES vi
LIST OF TABLES vi
ACRONYMS vii
1. Introduction ,. ,. ..,..,.,.. ,.,... 1
2. Regulatory Applications and Impacts 5
2.1 Chemical Screening and Prioritization 5
2.2 Toxicity Pathway-Based Risk Assessment 5
2.3 Institutional Transition 7
3. Toxicity Pathway Idt*ntitication and Chemical Screening and Prioritization 8
3.1 Strategic Goal 1: Toxicity Pathway Identification and Assay Development 10
3.2 Strategic Goal 2: Chemical Priorili/ation 11
4. Toxicity Pathway-Based Risk Assessment 12
4.1 Strategic Goal 3: Toxicity Pathway Know ledgebascs 13
4.2 Strategic Goal 4: Virtual Tissues, Organs, and Systems: Linking Exposure. Dosimctry, and
Response 14
4.3 Strategic Goal 5: Human Evaluation and Quantitative Risk Assessment 16
5. Institutional Transition 18
5.1 Strategic Goal 6: Operational Transition IX
5.2 Strategic Goal 7: Organizational Transition 20
5.3 Strategic Goal 8: Outreach 20
6. Future Steps 23
Appendix: Other Related Activities 24
References 27
Previous I TOC
-------
EPA CompTox Research Program FY2009-2012
IV. APPENDICES cont.
BOSC Review Draft- 24 August, 2009
LIST OF FIGURES
Figure 1. Toxicity Pathways 2
Figure 2. Toxicity Pathways Target Multiple Levels of Biological Organization...,,,.. 8
Figure 3. "I'oxCast™ 11
Figure 4. Toxicity Pathways to Dose-Response , 12
Figure 5. Knowledgehase Development 14
Figure 6. Relative (°o) Kmphasis of the Three Main Components of this Strategic Plan over its
Expected 20-year Duration 23
LIST OF TABLES
Table 1. Strategic Plan: Applications and Impacts
Previous
TOC
-------
EPA CompTox Research Program FY2009-2012
IV. APPENDICES cont.
BOSC Review Draft- 24 August, 2009
ACRONYMS
ACToR Aggregated Computational Toxicology Resource
FIFRA Federal Insecticide, Fungicide, and Rodenticide Act
F'lTW Future ofToxieily Testing Workgroup
HIS High ITiroughpul Screening
IRIS Integrated Risk Information System
NRC National Research Council of the National Academies
OPPTS Office of Prevention, Pesticides, and Toxic Substances
ORD Office of Research and Development
QSAR Quantitative Structure-Activity Relationship
SAR Structure-Activity Relationships
Previous
TOC
-------
EPA CompTox Research Program FY2009-2012 BOSC Review Draft- 24 August, 2009
IV. APPENDICES cont.
1. INTRODUCTION
KPA bases its regulatory' decisions on a wide range of tools and information that represent the best
available science. In some situations, where very limited or no animal toxieity data exist. EPA may
use tools such as structure-activity relationships (SAR) and quantitative structure-activity
relationship (QSAR) modeling, together with information on exposure to make decisions about
priority setting and the need for further evaluation (e.g., for new chemicals in the toxics program.
high production volume chemicals, and pesticide inerts). To establish regulatory standards, KPA
relies heavily on toxieity testing to evaluate clinical or pathological effects in experimental animal
models. As such, toxieity testing and related research is currently a multi-billion dollar activity that
engages thousands of research scientists, risk assessors, and risk managers throughout the world.
To that end, the historical path taken in toxieity testing of environmental agents has generally been
either to make incremental modifications to existing tests or to add additional tests to cover
endpoints not previously considered (e.g., developmental neuroloxieity). This approach has led
over time to a continual increase in the number of tests, cost of testing, use of laboratory animals,
and time to develop and review the resulting data. Moreover, the application of current toxieity
testing and risk assessment approaches to meet existing, and evolving, regulatory needs has
encountered challenges in obtaining data on the lens of thousands of chemicals to which people are
potentially exposed and in accommodating increasingly complex issues (e.g.. lifestage
susceptibility, mixtures, varying exposure scenarios, cumulative risk, understanding mechanisms
of toxieity and their implications in assessing dose-response, and eharacteri/alion of uncertainty) .
While the challenges of such information gaps are great, the explosion of new scientific tools in
computational, informational, and molecular sciences offers great promise to address these
challenges and greatly strengthen toxieity testing and risk assessment approaches. Proven benefits
have been demonstrated in allied fields such as medicine and phannaccuticals. Although untapped,
the potential application to toxieity testing and risk assessment has also been reeogni/ed by KPA as
witnessed by the issuance of a series of papers that provided guidance on the use of genotnic data/
To better anticipate the potential contribution ofnew technologies and scientific advances to issues
associated with toxieity testing and risk assessment, KPA commissioned the National Research
Council (NRC) in 2004 to review existing strategies (NRC, 2006) and develop a long range vision
for toxieity testing and risk assessment (NRC. 2007). In the subsequent release of Toxicily Testing
in the 21st Century: a 1'ision and a Strategy; a landmark transformation in toxieity testing and risk
assessment is envisioned that focuses on 'toxieity pathways."" This approach is based on the
rapidly evolving scientific understanding of how genes, proteins, and small molecules interact to
form molecular pathway's that maintain cell function. The goal is to determine how exposure to
environmental agents can perturb these pathways causing a cascade of subsequent key events
' These limitations have been described more fully in,-! Review of the Reference Dose and Reference Concentration
Processes: hup "www epa gov 'ncea;insi'RFD_FINAI,[ I J.pdf
1 Interim Policy on Genomics (20112): http/ www epa.gov/osa-spcigefiomics.htm, Uenomies White Paper (2004):
http:.'.w\vw.epa gov osa:pdfs.HFA-Genom]cs-\Vhite-Papcr,p(.1f; Interim Guidance for Microarray-Bascd Assays
COO 7): hUp:-.%ww.epa.govi'i>sa.i'spc.;pJf&'epa_inleriiTi_guidance_for_rnicroaiTay-ha.sed_assays-extemal-
review draft pdf
3 Toxieity pathways are cellular response pathways thai, when sufficiently perturbed, are expected to result in
adverse health effects.
Previous I TOC
-------
EPA CompTox Research Program FY2009-2012
IV. APPENDICES cont.
BOSC Review Draft- 24 August, 2009
leading to adverse health effects. This sequence of events is illustrated in Figure 1 wherein the
introduction of an environmental stressor may trigger such a cascade. Successful application of
these new scientific tools and approaches will inform and produce more credible decision making
with ail increased efficiency in design and costs and a reduction in animal usage.
Other agencies have also recognised
the need for this transformative shift,
including the National Toxicology
Program in their Roadmap for the
Future and the hood and Drug
Administration in their Critieal Path
Program. In anticipating the
emergence, and potential, of this new
scientific paradigm, EPA's Office of
Research and Development (ORD) and
some of the Agency's regulatory
programs have also begun to redirect
resources in intramural and extramural
research programs to "jump stall" the
process of transformation. For
example. ORD created the National
Center for Computational Toxicology
in 2006. Likewise, ORD National
Laboratories and Centers have also
Figure 1. Tttxicity Pathways. Toxicity pathways describe the
processes by which perturbations of normal biological processes
due to exposure to a slrcssor (e.g.. chemical) produce changes
sufficient to lead to cell injury and subsequent events (modified
from NRC, 2007).
begun to incorporate these new scientific tools to better support the research being conducted
tinder several of its mulliyear research plans. Several ongoing projects address the use of in vitro
assays in risk assessment and loxicity testing (e.g.. Ciuyton, el a/., 2008), and assessments under
the Integrated Risk Information System (IRIS) program arc describing and evaluating published
gcnomic data. KP.Ys Office of Prevention. Pesticides, and Toxic Substances (OPPTS) is also
actively involved in the development and transition of computational toxicology' tools into
regulatory practice. OPPTS has developed a multi-year strategic plan to advance computational
toxicology tools in its risk assessment and management paradigm. Current activities include
assisting ORD by providing the necessary databases to support the development of models for
efficiently and credibly predicting toxic potency and levels of exposure, beta testing the new
computer models, training staff, and initiating plans for successful international coordination and
stakeholder involvement. Furthermore, recognizing the need to partner to achieve the vision and
goals laid out by the NRC, EPA recently signed a Memorandum of Understanding for research
cooperation with the National Toxicology Program and the National Institutes of Health
Chemical Gcnomics Center as a substantive step forward in building collaborations across sister
federal agencies.6 EPA is also working actively at the international level with programs such as
the Organization for Economic Cooperation and Development (OECD) through the Molecular
4 Computational toxicology is the application of mathematical and computer models and molecular biological
approaches to improve the Agency's priorilizalion of data requirements and risk assessments (from .-I Framework
for a Computational Toxicology Research Program. HPA 600.14-03/065),
5 http://cfpub.cpa.gov nccairis/indc.x.cfhi
6 htlp://www.epa.gov/comptox/arlicles/com plox_mou.html
Previous
TOC
-------
EPA CompTox Research Program FY2009-2012 BOSC Review Draft- 24 August, 2009
IV. APPENDICES cont.
Screening Initiative, the Integrated Approaches for Testing and Assessment Workgroup. Test
Guideline Committees, and the QSAR Expert Group to ensure global harmoni/.ation of any new
approach that originates from the research program. A more complete listing of these
collaborations may be found in the appendix.
In response to the release of the NRC reports, KPA has established an intragency workgroup, the
Future of Toxicity Testing Workgroup (FTTW), under the auspices of the Science Policy
Council. The FTTW includes representatives from across the .Agency, including the Regions and
all major Program Offices, It has produced this current document, which will serve as a blueprint
lor ensuring a leadership role for KPA in pursuing the directions and recommendations presented
in the 2007 NRC report. This document presents a strategy that is consistent with the NEC's
directions and recommendations. It presents the Agency's vision of how to incorporate a new
scientific paradigm and new tools into toxicily testing and risk assessment practices with ever-
decreasing reliance on traditional apical approaches. The overall goal of this strategy is to
provide the tools and approaches to move from a near exclusive use of animal tests for predicting
human health effects to a process that relies more heavily on in vitro assays, especially those
using human cell lines. The topics to be covered include (1) the applications and impacts benefits
for various types of regulatory activities (Section 2), (2) the research to be conducted to facilitate
the screening and prioritization of environmental agents (Section 3), (3) the implementation of a
toxicity pathway-based approach to risk assessment (Section 4), and (4) the critical companion
component, namely, the institutional transition that must occur before the changes can be fully
implemented (Section 5).
As described in Section 6, the workgroup ro-cogniy.es thai the full implementation ofthe vision
set out in this strategy will require a significant investment of resources over a long period of
time. 'Hie workgroup has identified a range of partners in this effort, and some planning on the
relative role of these partners has begun, although the specific areas of work to be
conducted funded by EPA versus other partners needs further assessment. Decisions on the
relative roles will have a significant impact on KPA resources required to implement the vision.
Since the NRC charge and report centered on advancing toxicity testing for assessing human
health effects of environmental agents, this strategic plan is presented primarily within that
context. However, under environmental legislative mandates (e.g., the Toxic Substances Control
Act; the federal Insecticide, Fungicide, and Rodenticidc Act; and the Clean Water Act), most
KPA programs must regulate compounds to ensure both environmental and human health risks
are properly managed. Since statutory language and-'or resulting policy typically require single
regulatory decisions for a chemical(s) that encompass environmental and human health risks at
the same time, accelerated and cost effective approaches for both areas are critical to reali/e
programmatic benefits. As in the human health arena, development and application of
approaches described in this strategy apply to ecoloxicology and risk assessment as well. Notable
progress is being made within EPA Laboratories and Centers on the development and use of
toxicity pathway models and the creation of prioritization schemes, toxicology knowledgebases.
and systems biology models in the field of environmental science. The bringing together of
relevant disciplines to share data and integrate models is critical to fully achieve increased
efficiency in toxicity testing and a reduction in animal usage for both human health and
environmental risk assessment. Consequently, the Agency will be implementing this strategy in a
manner that addresses both human health and ecological risk assessment. Future versions ofthe
Previous I TOC
-------
EPA CompTox Research Program FY2009-2012
IV. APPENDICES cont.
BOSC Review Draft- 24 August, 2009
strategy will summarize progress made in advancing integrated testing and assessment capability
and revisit remaining challenges.
Previous
TOC
-------
EPA CompTox Research Program FY2009-2012 BOSC Review Draft- 24 August, 2009
IV. APPENDICES cont.
2. REGULATORY APPLICATIONS AND IMPACTS
The research arising from implementation of this strategy will change the nature of the methods,
models, and data that will inform the major components of the risk assessment process (i.e..
ha/ard identification, dose response, exposure assessment, and risk characterization). Without
attempting to he all-inclusive. Table 1 presents some of the major cross-office applications and
impacts of these new scientific approaches, with more in-depth discussion ofthe planned work
described in Sections 3-5. Hie three components of this strategic plan, namely, chemical
screening and prioritization. toxieity pathway-based risk assessment, and institutional transition,
are not independent elements but rather highly interactive and integrative efforts that will
maximize the value and application ofthe research generated.
2.1. Chemical Screening and Prioriti/ation
An ongoing need of several regulatory offices is to have tools to assist in chemical screening and
prioriti/ation, e.g., high production volume chemicals, air toxics, the drinking water Contaminant
Candidate Lists, and Superfund chemicals. These programs consider anticipated exposure and
hazard to select chemicals to evaluate in longer-term, whole-animal laboratory studies. An early
use for data developed under the new paradigm will be as an efficient and cost effective screen
for several types of chemical toxieity. Thus, risk assessors could use in silico (computer-based)
technologies and structure molecular bioactivity profiling from diagnostic high-throughput .in
vitro assays, along with predicted exposure/dose information, to predict chemicals most likely to
cause hazards of concern for humans. This approach will also enable risk assessors to determine
the specific effects, in vivo data, and exposures that would be most useful to assess, quantify, and
manage. As the technology develops, EPA will be able to screen previously untested chemicals
using libraries of chemical, molecular, biological, and toxicological data and models to identify
the types of adverse effects that they arc most likely to produce in standard animal bioassays.
More importantly. EPA will be able to gain better insight into whether such effects would likely
be manifest in humans under various exposure scenarios. As noted earlier, these needs arc
common to a number of federal agencies; discussions are underway to develop more common
paradigms among federal agencies to facilitate data sharing.
2.2. Toxicity Pathway-Based Risk Assessment
'lite current approach to risk assessment includes uncertainties associated with (1) the human
relevance of laboratory animal studies (species extrapolation). (2) the use of high doses in
animals to estimate risk associated with lower environmental/ambient exposures (dose
extrapolation), and (3) predicting the risk to susceptible populations. In recent years, the
consideration of such issues has been better informed by the incorporation of information on
potential modes of action through which toxicity may be expressed, '[lie approach outlined
earlier in Figure 1 focuses on perturbations in baseline biological processes that may lead down
toxicity pathways to adverse health outcome(s). Combining this information with distributional
data on population characteristics of exposure and dose (magnitude, frequency, and duration)
provides a scientifically based approach for reducing the uncertainties associated with current
risk assessments. By relying on a quantitative understanding of perturbations in toxicity
pathways that lead to adverse health effects, the new approach to toxicity testing and risk
assessment envisioned in this document will greatly increase EPA's capacity to assess individual
Previous I TOC
-------
EPA CompTox Research Program FY2009-2012
IV. APPENDICES cont.
BOSC Review Draft- 24 August, 2009
chemicals and their mixtures. "The new approach will also increase EPA's confidence that the
Agency's assessments adequately protect human health. Rcali/ation and acceptance of this new
approach will likely encounter numerous challenges, but the effort is expected to ultimately lead
to better protection of human health.
Table 1, Strategic Plan: Applications and Impacts
o
1
1
»w Approach
1
Toxicilj Pathway
Idtiitifiaifion and
Chemical Screening &
PHoritoiioii
Need to screen 10,000'*
of chemicals for wide
range of endpoints in a
manner thai considers
toxieity pathways and the
potential for human
exposure.
Need to limit cos! and
animal usage, improve
timeliness, and decrease
uncertainty in testing
decisions,
Identification oftoxicity
jxHhways for key
toxieologicrd endpoints.
Cornhsnc in siiico and
taoprofiles from UTS"
ulon» with QSAR
approaches linked to
animal study data
Offices would bo better
able to direct etibrts and
resources to chemicals
with greatest ixucntinl
nsk. Significant increase
in efficiency with marked
reduction in cosf for
toxieity testing.
Toxicily P;iHiwa>-B;*secI Risk Assessmenl
For many clieniicals, the current sapproaeh
relies on expensive animal EdsUifif tlutt Itike^
time to conduct and rc^ew. Untitattom in the
design of in vivo studies often jwcvent
complete evahiationoi'all endpomts ajid
hiizai'djisk s-ceiiunos of concern,
Limited imderAtanding of biotostcal
niechaj'ii&nis most often leads to lutcertain^' MS
assessing cumulative nsk or estrapolttlnig m
vitro to in vivo or across doses, lifesttsges,
species, or genetic diversi^
New scientific saidersEandingcaiKl fools in
molecular, computational, and information
science* consistent with applications in allied
aiea,s such as medicine aiidplianiiiiceuticals
represent a path forward
Reliance on increased understanding of how
perturbations of biologicLtl processes a!
environmentally relevant conccntratioas
tnt^er events (i.e., toxicilv [sjthway(s)) Htti3
may lead to adverse health ou teenies.
Develop linked cxposnre/dose models to-
inform dosing levels for toxieity testing and
inform risks.
More scientifically relevant data on which to
base EPA's regulatory decisions and/or
impact analyses dial rely ott Uwsc risk
assessments.
Institutional Transition
implementing the new approach will
reqtrira significant instil utiomtl investmeni
in operational and orgiani/ational transition
and in public outreach.
EPA lacks itppropriale ex|x:rl!se and
sufficient ftsndtng to fislly and most
elllcieatly utilise the new to>acit\' testing
ttfclmoiojzies when iiiakm.u regulakxr\'
decisions.
Fully adopting tlic new fmradigm slKndd
be supported by nice ham stkally based
proof-of- concept and venficafion studies
i-'urthcr, such atlo-pttoti will reijusre
adtijlioual training ol" existing staff and
hiring new staff tonversiinc in state-of-tht:-
scicnce knowletlge tn fields such as
toxicology, hiocbemtstry, hioinfonnatics,
etc.
A well informed public will have greater
confidence as EPA greatly expands the
nyniber of chemicals assessed for possible
nsks and unpro%res existing strategies for
ha/ard and risk assoAsnient!
7 High-Throughput Screening (UTS) refers to robotic technologies developed by the pharmaceutical industry for
drug development thai enable the ability to evaluate the effects of hundreds lo thousands of chemicals per day on
molecular, biochemical or cellular processes
Previous
TOC
-------
EPA CompTox Research Program FY2009-2012
IV. APPENDICES cont.
BOSC Review Draft- 24 August, 2009
23. Institutional Transition
Implementing major changes in toxicity testing of environmental contaminants and incorporating
new types of toxicity data into risk assessment will require significant institutional change
involving:
• Operational transition - how EPA will transition to the use of new types of data and
models for toxicity testing and risk assessment;
• Organizational transition how KPA will deploy resources necessary to implement the
new toxicily testing paradigm such as hiring of scientists with particular scientific
expertise and training of existing scientific staff and risk managers;
• Outreach - efforts by EPA to share information with the public and improve risk
communication.
The process of moving from research to regulatory acceptance for implementing new science
related to toxicity testing will be an iterative and long-term effort (likely encompassing more
than a decade). Essential to this iterative process will be the demonstration that the predictive
nature of these new approaches is superior to that of our current practices for toxicity testing and
risk assessment. It will be critical to begin activities geared toward regulatory acceptance early in
the process of implementing this strategic plan.
Previous
TOC
-------
EPA CompTox Research Program FY2009-2012
IV. APPENDICES cont.
BOSC Review Draft- 24 August, 2009
3. TOXICITY PATHWAY IDENTIFICATION AND CHEMICAL SCREENING
AND PRIORI TI/AT ION
The advancements in biotechnology brought about by the sequencing of the human genome and
the investment in high throughput screening tools to mine large chemical libraries for potential
drugs have for the first time allowed a broad scale, unbiased examination of the molecular and
cellular targets of chemicals. At this time, the examination of the relationships between the
molecular and cellular targets ol chemicals and the traditional endpoints of toxicity is at an early
stage of development. Even upon characterization of these types of relationships, significant
phenotypic data will be required to critically establish the role oftoxicity pathways in evaluating
hazards and risks. The great potential is that identification of a toxicity pathway and
development of an in vitro bioassay for studying its chemical interactions will enable evaluation
ol the effects ol thousands of chemicals in that pathway. Broadening this approach to the many
toxicity pathways present in living systems allows a new avenue for identifying those chemicals
that pose the greatest potential hazard. Knowledge of the toxicity pathways triggered by any one
chemical will also allow targeting of specific in vivo tests to more fully characterize the potential
hazard and risk. The identification oftoxicity pathways for key target tissues, organs, and
lifcstagcs. and their linkage across levels of biological organization and exposure pathways and
intensities are core elements of this strategy.
As indicated in Figure 2, chemicals may interact with a single pathway (the blue chemical) or
multiple pathways (the yellow chemical). Also, multiple pathways can lead to the same
expression of toxicity in the target organ as signaling pathways converge on common elements.
It is important to note that multiple
mechanisms of action for am
particular adverse response likely 1
exist, and thai many environmental
pollutants arc likely to have multiple
mechanisms of action. Two critical
components of the toxicity pathway
concept are (1) extending knowledge
of molecular perturbations and cell
signaling pathways to understand
linkages between levels of biological j^g 2. Toxteity pathway* Target Multiple Levels of
organization and (2) extending Biological Organization.
knowledge of in vitro and m vivo
markers relevant to adaptive changes and/or adverse outcomes (see Section 5). As the research
moves forward, it will be important to capture quantitative relationships between the molecular
events and the higher order changes. Demonstration of plausible connectivity along the
mechanism of action from initiating event to adverse outcome will serve as the rationale for
using data from subcellular or cell-based in vitro assays for not only chemical prioritixation but
also predictive risk assessment. As loxicity pathways are identified, relevant in vitro assays can
Chemicals —,
Receptors / Enzymos / etc.
Direcl Molecular Interaction
Pathway Regulation /
Genomics
Cellular Processes
Tissue f Organ / Organism Tox Endpoint
Mode of action is defined as a sequence of key events and processes, starting with interaction of an agent with a
cell, proceeding through operational and anatomical changes, and resulting in an adverse health effect. .Mechanism
of action implies a more detailed understanding and description of events, often at the molecular level, than is meant
by mode of action.
Previous
TOC
-------
EPA CompTox Research Program FY2009-2012 BOSC Review Draft- 24 August, 2009
IV. APPENDICES cont.
be utilized and their results compared to in vivo studies as appropriate given the need to predict
effects in humans or other species. While comparing responses to those in animal hioassays will
be an early milestone of this strategy, the ultimate goal is the prediction ofhunian risk,
'ITlcrefore. efforts will shift towards that goal as experience with the approach increases. An
added benefit to the loxicily pathway approach is that mixtures or their components could be
evaluated in this manner, and as knowledge grows, it will be possible to predict where
interaction with multiple toxicity pathways might be expected to lead to non-additive outcomes.
Tliis later activity will be an important outcome of the research highlighted in Section 4,2
(Strategic Goal 4) that is focused on the development of virtual tissue models. As noted below.
virtual tissue models will also provide a basis for predicting emergent properties of tissues by
integrating knowledge of molecular and cellular behaviors obtained from reductionist /» vitro
approaches,
In 2007, EPA launched ToxCast ' in order to develop a cost-effective approach for prioritizing
the loxicily testing of large numbers of chemicals in a short period of time. Using data from a
broad range of state-of-the-art UTS bioassays developed in the pharmaceutical industry.
ToxCast is building computational models to forecast the potential human toxicity of
chemicals. Results from the HIS bioassays are being analy/.ed for signatures of bioactivity that
correlate with known toxicities. These hazard predictions will provide EPA regulator*' programs
with science-based information helpful in prioriti/.ing chemicals for more detailed lexicological
evaluations, and lead to more efficient use of animal testing.
"Hie research described here focuses on two major strategic goals:
1) Identification of toxicity pathways and deployment of in vitro assays to characterize the
ability of chemicals to perturb those pathways in different biological contexts, and
2) Implementation of ToxCast . with an initial focus on providing input for chemical
prioriti/alion, shilling over time to providing input for dose-response modeling.
.-V key feature of ToxCast is the phased nature of implementation (see Strategic Goal 2, Section
3.2). from proof of concept, to forward validation, and finally to reduction to practice. The
number of chemicals will grow from the hundreds to the thousands, and the number of assays
will change as experience and biology dictate. As the number of chemicals and breadth of
toxicity pathways covered increase. ToxCast will improve as a unique resource to build chemo-
informalic-based predictions of chemicals" potential human toxicity. Such advancements should
help promote improved QSAR models and data upon which to build virtual tissue models.
Exposure science also plays a large role in this strategy. More simple and reliable screening
models are needed that predict exposures to chemicals so that information from the full source-
to-oulcome continuum is brought into consideration in the evaluation of chemicals a critically
important step for new chemicals that have not yet been released into the environment. Examples
of such simple methods and models for new chemicals can be found at EPA's Sustainable
Futures Initiative1". Additional such models should further evaluate exposure based on the life
cycle of intended product use and the physical-chemical properties of the chemicals. This
' htlp:--''\v\vw.epa.gov.'nccl/toxcast.1
1 ° http: ••' w w w epa.gov /oppt sf'
Previous I TOC
-------
EPA CompTox Research Program FY2009-2012 BOSC Review Draft- 24 August, 2009
IV. APPENDICES cont.
research should include the expansion of computational chemistry methods to further predict
exposures as well as methods to predict release into the environment during product life cycle.
Several additional screening-level models are currently under development in Canada and
Kurope. Research in this area should be coordinated with these groups to facilitate an
international approach for chemical screening. KPA should promote easy public access to all of
these additional models through the Internet,
3.1. Strategic Goal 1: Toxicity Pathway Identification and Assay Development
'ITie most systematic and extensive approach currently underway for screening and prioriti/ation
is EPA's ToxCast . Fully implementing the proposed strategy for more efficient toxicity testing
will utili/.e a combination of the more exploratory ToxCast chemical signature approach (see
Strategic Goal 2), and the more hypothesis-driven approaches to elucidating loxicity pathways.
Developing systems-based models will require comprehensive identification of the biological
processes that can result in toxicity when they are perturbed by chemical exposures. Therefore,
toxicity pathway identification and development of appropriate in vitro assays to characterize the
dose-response and time course of perturbations to those pathways will be needed. Measurement
ol chemical form and concentration from in vitro assays will also he important in hypothesis-
driven research that seeks to establish linkages between perturbations of toxicity pathways and
adverse effects, as well as for establishing structure-activity relationships. These research goals
will utilize a range of methods (e.g.. transcriptomic, proteomic, metabolomic, cellular, and
biochemical analyses) to identify toxicity pathways using in vivo and in vitro systems. The in
vitro assays and toxicity pathways already included in the ToxCast project will be a part of this
research, but additional assays providing greater coverage of relevant toxicity pathways will
need to be developed. For example, developmental neurotoxicity key responses are known to
include cell proliferation, apoptosis, differentiation (into different cell types and creating
different functionality architecture of a cell), neurite outgrowth, synaptogenesis, and myelination
(Cockc el al., 2007; Lien et al., 2007), but the underlying molecular pathways are not yet
completely identified. Through the informed use of newer "systems-based" approaches (Edwards
& Preston. 2008). the flow of molecular regulatory information underlying the control of these
cellular events can be characteri/.ed, classified, and modeled. To facilitate use in risk assessment,
these studies will be coupled with mechanism of action-based studies, including animal and
human components as described in Strategic Goal 4.
Current priorities for research include developing in vitro assays for the key targets of chemicals
in the environment for which limited knowledge is available (e.g., developmental neurotoxicity,
immunotoxicity. reproductive loxicity) as well as for relatively well-eharaeleri/.ed toxicity
pathways such as stress response signaling. Studies representative of the full range of human
variability will be necessary to characterize processes that may occur more readily in sensitive
populations (e.g., asthmatics) or at certain lifestages (e.g., prenatal development). Additional
emphasis needs to be placed on toxicities demonstrated to occur in humans. For example, clinical
trials or post-marketing surveillance for phamiaceuticals. as well as molecular and genetic
epidemiology studies, afford the opportunity to examine effects of chemicals already introduced
into the environment that may not currently he well assessed by in vivo animal toxicity studies.
Some of these pathways may be important for environmental chemicals with respect to human
variability or exposure to complex mixtures.
10
Previous I TOC
-------
EPA CompTox Research Program FY2009-2012
IV. APPENDICES cont.
BOSC Review Draft- 24 August, 2009
$ThouMnd«
Cancer
ReproTox
DevTox
NeuroTox
PulmonaryTox
ImmunoTox
HTS
-omics
Bianformaticsf
Machine Learning
3.2. Strategic Goal 2: Chemical Prioiitization
This strategy extends approaches that are currently under development for KPA s ToxCast
program to include greater coverage of toxicity pathways and chemicals. The goal of the
ToxCast program is to provide a comprehensive assessment ol toxicity pathways for a
relatively low cost per chemical (current estimates are in range of $20-23,000). ToxCast (see
Figure 3) was
designed to collect '" 1"ftD testin9
data from a wide
range of in vitro ¥
'
assays, mostly
; • . . .- •» ?.
mechanistic in nature, 0 ^
to prioriti/.e which
chemicals to test
further and which in
vivo studies were
likely most important.
This screening and
prioritization approach
provides a near-term benefit during an extended transition to the more comprehensive proposed
vision. As more comprehensive descriptions of processes involved in toxicological responses
become available, different assays may be identified to replace those in the initial ToxCast
effort, and the relationship to in vivo studies will shift from prioritization to providing input for
dose-response modeling.
ToxCast is being developed in a phased manner. During FYOK-O'A substantial progress will be
made on the first two phases ofthe ToxCast program (l)ix et al., 2007; Kavlock et al.. 2008).
Phase I is a proof of concept involving 320 chemicals that have robust in vivo animal toxicity
information. 'ITiese chemicals have been profiled using over 400 high and medium throughput in
vitro assays. From these in vitro bioactiv ity profiles, classifiers or signatures predictive of
chemicals" in vivo loxicily are being derived. Phase II will involve validation ofthe predictive
bioaetivily and expansion of the diversity of chemicals tested. Phase III is the most relevant to
this strategic plan, as it would begin to apply the knowledge gained in Phases I and II to the tens
of thousands of chemicals of concern to KPA regulatory offices. An adaptation ofthe approach
to evaluate the hazardous properties of nanomalerials is also anticipated.
Figure 3. ToxGast™ is using a variety of UTS assays to develop bux-ictivity
signatures that are predictive of effects in traditional toxicity testing approaches.
11
Previous
TOC
-------
EPA CompTox Research Program FY2009-2012
IV. APPENDICES cont.
BOSC Review Draft- 24 August, 2009
4. TOXICITY PATHWAY BASED RISK ASSESSMENT
ITie goals of the proposed new strategy for loxicily testing include collecting mechanistic data.
largely in vitro, for the purpose of predicting human risk from exposure to chemicals. Prediction
of in vivo effects in humans requires a combination of measurements and computer modeling to
link in vitro responses to tissue dosimetry to alterations in the structure and function oI tissues
and organs. A substantial challenge will be to address the range of human variability arising from
differences in age, life stage, genetics, disease susceptibility, epigenelics, diet, disease status, and
other factors that potentially influence or interact with toxicity pathways.
'ITie initial process for predicting human risk under this new approach could be summarized as
(1) characterizing or predicting potential human exposures; (2) estimating the resulting chemical
dosimelry (magnitude, frequency, and duration) For target pathways, tissues or organs; (3)
measuring toxicity pathway response at doses consistent with human exposures; (4) predicting
the in vivo human response resulting from pathway perturbations: (5) quantifying the range of
human variability and susceptibility; and (6) validating predictions ulili/.ing in vivo systems (e.g.,
laboratory animals, human data). In the current state of mechanistic toxicology (top row of
Figure 4). chemicals are administered to the test animals (usually at high doses), a variety of
Erviron mental
Chemicals
Molecubr
Sensing
Celular
Signafng
TBS ue
Responses
Knowledgsbase
Toxicity
Pathways
Molecular
Nstwxks
Celular
Networks
Virtual
Tissues
Dose-Response
Figure 4. Toxicitj Pathways to Dose-Response. The vertical arrows at each step in the process reflect the
iterative nature of experimentation and modeling needed to gain fill! understanding of both the toxicity pathway
determination and the relationship to normal biology.
biochemical approaches are used to detect alterations in molecular pathways, the data are mined
to describe the ensuing cellular alterations (e.g.. oxidative stress damage, mitochondrial
dysfunction), and tissue changes are confirmed at the level of morphology or function. The
12
Previous
TOC
-------
EPA CompTox Research Program FY2009-2012 BOSC Review Draft- 24 August, 2009
IV. APPENDICES cont.
bottom row of the figure depicts the vision for future ways of assessing risk, which includes
determining the key toxicity pathways, defining approaches for examining perturbations in
molecular networks, and translating the results to responses at the cell, and ultimately tissue and
organ level, using computational models of the relevant systems. 'Hie expectation is that
assessments in the future will utili/.e data from in vitro studies, and the need for in vivo animal
testing will be substantially reduced. However, until the state of science of this new approach has
reached a level oI confidence for use in regulatory decision making, the traditional approach to
toxicity testing will continue into the foreseeable future. With time, we expect that it will be
progressively augmented and ideally replaced by computational models that integrate the
information generated from non-animal sources into predictive models of response based upon
the underlying biology. The vertical arrows at each step in the process reflect the iterative nature
of experimentation and modeling needed to gain full understanding of both the toxicity pathway
determination and the relationship to unperturbed biology. One anticipated outcome of the
development of virtual tissues will be an increased understanding of the role of metabolism and
of intra- and inter-cellular signaling pathways. This understanding will lead to the development
of improved in vitro systems that, for example, might include combined cell-based systems to
provide metabolic competency or to better reflect the intercellular responses in heterogeneous
tissues.
As the transition progresses, it is important that increased emphasis will be placed on
examination of exposure concentrations that are expected to occur in the environment. The key
difference in future toxicity evaluations will be the transition to a focus on ways in which
molecular pathways (as detected by in vitro models) arc perturbed by chemical exposure
throughout the range of exposures from environmental to the higher dose levels commonly used
in contemporary toxicity studies. Dosimetry measurements coupled with computational
modeling will be critical for predicting in vivo exposure levels of concern and for determining
relevant in vitro concentrations. Some responses of targeted toxicity pathways can be evaluated
in simpler cell culture models, whereas, in other cases, multiple in vitro assays may be necessary
for the integration of multiple pathways that produce in vivo responses. Iliese situations would
require biologically based models for the responses as well as for chemical dosimetry in order to
predict the integrated in vivo response.
Implementing this new paradigm requires organization of existing scientific information;
computational methods for exposure, chemical dosimetry, and perturbations of biological
processes; and evaluation of the methods for risk assessment applications. The research program
to implement this element of the strategy is defined by three goals: development of toxicity
pathway and exposure knowledgebases; development of virtual tissues, organs, and systems; and
evaluation of human relevance.
4.1. Strategic Goal 3: Toxicity Pathway Knowledgebases
The underlying basis of the 2007 NRC report is that there are a finite number of toxicity
pathways (i.e., in the hundreds) that could be queried using in vitro assays to obtain insights into
the ability of chemicals to perturb those pathways. It refers to several stress pathways (e.g..
oxi dative stress response) and notes the general listing of signaling pathways in a previous NRC
report (2006), However, an inventory of toxicity pathways and their involvement in a variety of
toxicological responses needs to be created. Likewise, from exposure science there needs to be a
1.1
Previous I TOC
-------
EPA CompTox Research Program FY2009-2012
IV. APPENDICES cont.
BOSC Review Draft- 24 August, 2009
complementary effort focusing on those chemical properties and computational methods that
could be used to reliably predict behaviors in the environment and exposures. Ibis effort would
include information on stability in the environment, likely routes lor exposure, potential for
bioaccimiulalion, and extent of metabolism. Therefore, a strategic goal is the development of a
kno\vledgebase lor loxicity pathways and exposure. Knowledgebases differ from traditional
databases in the extent of integration of information and the inclusion of tools that can draw
inferences from amongst the diverse elements.
The knowledgebase would serve a variety of functions throughout the research and development
effort associated with implementing this new approach to loxieity testing and will become a
standard tool in the risk assessments of the future. ACToR (Figure 5), the Aggregated
Computational
Toxicology Resource
under development in
ORD. is an example of
the needed approach of
bringing together diverse
types of information into
a system where
interrelationships of
individual database
elements (e.g., traditional
toxicology, chemical
structure information.
high throughput
screening data, molecular
pathway analysis, chemical data repositories, peer reviewed published literature, and internal
Agency databases) can be explored and ulili/cd (.ludson el al., 2008). Key steps in development
of these knowledgebases include: (1) creating electronic repositories of existing loxieity
information; (2) developing semantics for describing toxicity pathways'. (3) automating pathway
inference tools to aid in discovering mechanistic links between genomie information and
molecular and cellular observations; and (4) creating a toolbox with a user-friendly interface to
organize, access, and analyze toxicity pathway assay results,
4.2. Strategic Goal 4: Virtual Tissues, Organs, and Systems: Linking Kxposure,
Dosimetry, and Response
Computational techniques relevant to this slralegy Tall into two general branches: knowledge-
discovery (data-collection, mining, and analysis) represented in Strategic Goal 3. and dynamic
computer simulalion (mathematical modeling at various levels of detail) described in ibis
section. The central premise of the latter approach is that critical effects of environmental agents
on molecular-, cellular-, tissue-, and organ-level pathways can be captured by computational
models that focus on the flow of molecular regulatory information (Kmidsen & Kavlock, 2008).
This information flow- is influenced by genetic and environmental signals, with the net outcome
being the emergent properties associated with baseline or abnormal collective cell behavior.
Thus, computational systems modeling will be used to predict organ injury due to chemical
exposure by simulating: (l)the dynamics and characteristics of exposure and dose, (2) the
Figure 5. Knowledgebase Development. ACToR brings together a diverse set of
currently unlinked resources available from internal and external sources into a
system with a user friendly interlace to readily mine and analyze loxieity data.
14
Previous
TOC
-------
EPA CompTox Research Program FY2009-2012 BOSC Review Draft- 24 August, 2009
IV. APPENDICES cont.
dynamics of perturbed molecular pathways. (3) their linkage with processes leading to alterations
ol cell state, and (4) the integration of the molecular and cellular responses into a physiological
tissue model. By placing a strong emphasis on understanding the biology of the system and the
key regulatory components, these virtual tissue models represent a significant opportunity to
better understand the linkage between chemically induced alterations in toxicity pathways and
effects at the organ level. This research represents an ambitious effort, conceivable for the first
time due to the current technological advances. Virtual tissue and organ system models will
initially include liver, cardiopulmonary function, selected immune system tissues, multi-organ
endocrine axes, and developing embryonic tissues. Development of these virtual tissue and organ
systems will require newly generated data to both fill data gaps identified within the iterative
process and test the predictive nature of these virtual systems. Comparative studies should
include pathways fundamentally reliant upon cell signaling (e.g.. cell proliferation, apoptosis.
cell adhesion), intermediary metabolism (e.g., glycolysis. oxygen utilization, fatty acid
biosynthesis), differentiation-specific functions (e.g., extracellular matrix remodeling), and other
categories as developed above (sec Strategic Goal 1) to ensure that predictions are broadly
applicable. The wealth of existing data from NTP assays, published reports, and previous EPA
intramural studies will be leveraged wherever possible with additional experiments designed to
fill data gaps. Such efforts will also help answer how well in vitro experimental systems
represent the full range of diverse cells present in the human body, how variability observed in
the human population can modify quantitative predictions of in vivo dose-response, how
exposure conditions influence outcomes, and how well the virtual tissue models represent the
underlying processes.
Not all toxicity pathways are likely to be expressed in every tissue, and likewise not all tissues
are likely to manifest adverse outcomes following chemical perturbation. Chemicals that affect
the same loxicity pathway can do so via a number of different (and overlapping) mechanisms,
and development of assays across loxicity pathways leading to the same outcome is a necessary
component of the proposed strategy. Some toxicities are manifest only when multiple cell types
and specific cell-cell interactions are present. Other to.xicilies may he dependent upon tissue
geometry and three-dimensional architecture. Examples include signaling between hepatocytes
and KupfTcr cells, or the many forms of signaling between epithelial and mesenchymal cells. As
such, developers of virtual cells, tissues, organs, and systems musl always bear in mind the need
to remain relevant to the processes critical to expressions of toxicity in vivo. Consistent with the
NRC vision (2007), this need will likely entail a continued although decreasing role for in vivo
systems for the foreseeable future.
A premise of the new to.xicily testing strategy is that computational methods combined with an
understanding of biological and exposure processes can be used to develop a more efficient mid
accurate approach for predicting risks from many chemicals, On the exposure side, models have
been developed and are available that predict fate and transport, environmental concentration,
exposures, and doses. These models work at multiple scales; for multiple sources, routes, and
pathways: and for multiple chemicals, although each model only addresses a single process or
compartment. Research is needed so that such models can take into account weathering of
contaminants, differences in bioavailability of contaminants, variations in exposures with age.
and variability in exposures within populations. Research is also needed to combine these models
across various scales to develop a linked source-to-oulcome modeling framework, to evaluate the
framework using multiple chemicals and exposure scenarios, and to improve the computational
15
Previous I TOC
-------
EPA CompTox Research Program FY2009-2012 BOSC Review Draft- 24 August, 2009
IV. APPENDICES cont.
efficiency for the approach. Ultimately, these exposure models will be linked to the virtual tissue
models for utili/.ing in vitro toxicity test results in quantitative risk assessments. Given the
complexity ot the challenges present in addressing each of these components, this effort
represents a long-term goal of the strategy. However, elTorts must begin now to put us on the
path to achieving the ultimate vision of Toxicity Testing in the 2la (.'entwy (NRC, 2007).
'ITic derived computational models must accurately describe the processes and mechanisms that
determine exposure and effect. They must have reliable input parameters in order to quantify
these processes. On the exposure side, our current understanding of processes and factors for
many classes ol chemicals and pathways (i.e., dermal and incidental ingestion) is limited. New
approaches will be evaluated that will allow us to address the most significant uncertainties.
Relational databases populated with data on exposures, exposure factors, activity patterns, and
biomarkers will be developed as described. Inlormatic approaches or applications of network
theory could potentially be used to provide a better understanding of important exposures, as
well as exposure/response relationships. In the 2007 NRC report, emphasis was placed on
biomarkers and their role in relating real world exposure to in vivo and in vitro biological
response. They were also proposed as primary1 indicators in surveillance programs for tracking
predicted exposures and health outcomes. Because of this emphasis, novel approaches for using
biomarkers and integrating them into new risk assessment approaches will be investigated for
chemicals already existing in the human environment. Perhaps such biomarker data can be used
to improve predictive exposure models that will be relied upon for new chemicals not yet
introduced into the environment.
4.3. Strategic Goal 5: Human Evaluation and Quantitative Risk Assessment
The critical challenge of this new vision for toxicity testing using mechanistic in vitro assays.
targeted in vitro or in vivo testing, and computational models is to demonstrate that it
successfully and adequately predicts human toxicological responses. Proof of concept efforts
need to address this challenge both retrospectivcly and prospcctivoly. Existing human data from
pharmaceutical and environmental studies will be used to the extent possible. Human data could
come from a range of sources including case reports, epidctniologieal studies (e.g.. from the
National Children's Study), and clinical trials. KPA has extensive experience obtaining human
clinical data following exposure to the criteria air pollutants (e.g.. o/one. particulale matter) and
other chemicals (e.g.. MTBK) . Engagement of the pharmaceutical industry and the Food and
Drug Administration to access toxicity findings from clinical trials ol"drugs that were
successfully registered or that failed to be registered would be a desirable component of this
ellbrt. Limited data may be available for some nutrients or dietary supplements as well.
Such efforts will help address the question of the extent to which key events (critical
perturbations) that are predictive of health endpoinls (e.g.. cancer, immunosuppression, kidney
disease) must be demonstrated or whether the perturbation of baseline biological processes
sufficient to induce substantial cellular level response (e.g., a stress response) should be
considered an adequate endpoint for risk assessment. Linking a specific pathway perturbation to
" All liPA conducted or supported research is subject to and must comply with liPA regulations on the protection of
human subjects Sec hUp Avww cpa.gov-Tcdrgstr'EPA-GENER.'-\L,12006jTcbruar\"Day-06/g 1045 htm,
htlp:'.'www epa.gov/oamrtpncv'fbrms/Kiftfi_l 7a.pdf
16
Previous I TOC
-------
EPA CompTox Research Program FY2009-2012 BOSC Review Draft- 24 August, 2009
IV. APPENDICES cont.
a particular target organ endpoint has the advantage of predicting outcomes that are already used
in risk assessment, while alternative approaches raise issues of which endpoints should and
should not be considered for risk assessment, Fhis approach is relatively straightforward for
some effects (e.g., hcmolysis of red blood cells by KGBK, where the effect and the mechanism of
action leading to it are qualitatively the same, even if quantitatively different). Linkage is more
complicated for effects observed in animals that may predict human effects that are related, but
not identical to, the outcomes in animals (e.g., developmental effects in an animal model may
predict developmental effects in humans, but the exact manifestation might be different). On the
other hand, as knowledge is gained about the interaction of chemicals with molecular targets, and
this knowledge is combined with information on how perturbations of those targets are translated
to responses in species-specific patterns (e.g., how activation oFeertain transcription factors lead
to species-specific tissues responses), it will be increasingly possible to predict human outcomes
from in vitro studies that identify mechanism of action. Clearly this aspect will need to be
addressed on a case-by-case basis as we gain experience.
To be most useful in evaluation of risk to humans, the pathway-based efforts should ideally be
tied to a known mechanism of action, such as via the use of quantitative biologically based, dose-
response models. Understanding of the relevant mechanism of action will enable the
identification of biomarkers for key event parameters (linked to toxicity pathways) that can be
monitored in human studies for those chemicals already released into the environment at
significant levels. 'ITtese biomarkers could be measured in observational human studies to
provide in vivo data to support the underlying pathway-based model. In addition, genetic
susceptibility in humans identified via whole genome association studies will provide support for
pathway-based models when genes critical for a key toxicity pathway are associated with
susceptibility. Finally, the use of quantitative models requires estimation of uncertainty and
variability in the predictions from in vitro assays and computational models. Formal methods for
model evaluation are essential for demonstrating the success of this new approach to toxicily
testing and risk assessment.
17
Previous I TOC
-------
EPA CompTox Research Program FY2009-2012 BOSC Review Draft- 24 August, 2009
IV. APPENDICES cont.
5. INSTITUTIONAL TRANSITION
Implementing major changes in toxicity testing ol environmental contaminants and incorporating
new types of toxicity data into risk assessment will require significant institutional changes. This
section will touch upon three major thrusts of implementing institutional transition: operational
transition, organizational transition, and outreach.
s.l. Strategic Goal 6: Operational Transition
Operational transition covers the technical aspects associated with EPA's implementation of a
new toxicity testing paradigm and associated changes in risk assessment. It will consider such
disparate topics as the importance of grounding the science, ensuring consistency of approaches
within HP A, and working with outside partners and issues associated with the use of new models
and tools.
'ITie KRC "envision|s| a future in which tests based on human cell systems can serve as better
models of human biologic responses than apical studies in different species." Achieving such a
future, however, will require substantial research to study and define various toxicity pathways.
In evaluating possible options for the future of toxicity testing, the NRC eventually chose an
option involving both in vitro and in vivo tests but based primarily upon human biology and the
attendant use of substantially fewer animal studies that would be focused on mechanism and
metabolism, llieir vision for the next 10 to 20 years relies on understanding perturbations of
critical cellular responses and the use of computational approaches for assessing hazard and risk.
A paradigm shift in toxicity testing based on pathway perturbation will likely require significant
methodological advances and future changes to EPA's risk assessment guidelines. Although it is
infeasible to denote a specific timeline for how long it will take to substantially complete the
strategic goals associated with toxicity pathway identification, chemical screening and
prioritization. and toxicity pathway-based risk assessment, this plan takes the view that advances
are likely to be gradual over the next decade or two. The good news is that toxicity testing
research efforts have already begun moving EPA and others towards the use of in silico
technologies and high throughput testing systems. The speed at which we are able to complete
this transition will depend on the availability of increased research funding. It is important to
note that our understanding of toxicity pathways for some apical endpoints (e.g., hepatotoxicity)
may be developed at a faster pace than others (e.g.. ncurotoxicity) thus, allowing more rapid
introduction of newer high-throughput in vilro testing methods.
Grounding the Science - From a broad regulatory perspective, data used by EPA to support
regulatory decisions will be shaped by the statutory language covering the action, regulatory
policies, and the resulting time and resources allocated to the assessment. Where appropriate, use
of data should be consistent with the EPA guidance articulated in a number of science policy and
guidance documents, including toxicity testing guidelines, risk assessment guidelines'".
information quality guidelines , and peer review guidance.1
12 hUp:^www,epa.gov/risk/guidance.htm
13 http '/www.epa.gov/qualily.inibrmationgURiclincs'
M http:"www epa.gov/peerreview pJf's/Peer8o20Revi
18
Previous I TOC
-------
EPA CompTox Research Program FY2009-2012 BOSC Review Draft- 24 August, 2009
IV. APPENDICES cont.
To implement this new paradigm, regulators, stakeholders, and the public will need to develop
confidence that the data generated can he used effectively and that public health will continue to
be protected. A step-wise implementation is envisioned: first, experience will be gained from
proof of concept studies using data from chemicals (e.g., pesticides) with a large set of toxieily
data developed using the current paradigm. Availability oI both new and traditional types of data
will allow extrapolation and comparison of results across methodologies.
Optimally, early success stories that meet programmatic needs in specific areas such as
mechanism of action analyses or cumulative risk assessments will demonstrate the broader
applicability ol computational toxicology within the Agency. Reliability of the testing paradigm
will need to be evaluated via a comprehensive development and review process, involving public
comment, harmonization with other agencies and international organizations, and peer review by
experts in the field. Bringing new methods into regulatory practice will require several phases
starting from the development of the science and technologies, to technology transfer and
building the regulatory infrastructure, to incorporation of the new tools into decision making.
Because this transformative paradigm will rely on new and complex science and will likely be
surrounded by some controversy, an important part of regulatory acceptance will be to conduct
research that will verify the approaches and models that will come to replace much of the way
toxicity testing and risk assessments are conducted in the Agency today. An important
component of the effort to develop new approaches to testing will be to translate the research
into regulatory applications.
Issues Associated With the Use of New Methods and Models - For this new paradigm to be
successful, new methods and models should be thoroughly evaluated prior to their application
and use in regulatory decision making. The computer-based models used by the Agency should
be publicly available. Testing methods should be accompanied by documentation that describes
(1) the method and its theoretical basis. (2) the techniques used to verity that the method is
accurate, and (3) the process used to evaluate whether the method and the results are sufficient to
provide an adequate basis for its use in regulatory decision making. Access to data to allow for
third party independent replication of results, to the extent practicable, is essential. Such review
is appropriate before the Agency relies on data from such a method.
Working With Outside Partners -The appendix provides details about the many outside parties
KPA will need to partner with in order to implement this strategic plan including:
• Other federal bodies such as the National Toxicology Program (NTP) and the Kill
Chemical Gcnomics Center (NCGC). with whom EPA has a memorandum of
understanding to collaborate:
• "Hie Interagency Coordinating Committee on the Validation of Alternative Methods
(ICC VAVI), which is made up of representatives from 15 federal agencies thai generate
or use toxicological data:
• Foreign governmental parties and programs such as REACH, which is the new European
Union Regulation on Registration. Evaluation, Authori/alion, and Restriction of
Chemicals that went into effect June 1, 2007;
'* See htlp;;''epa.gov;crem libraw ORFAfguidancedrafll -JB pJf
19
Previous I TOC
-------
EPA CompTox Research Program FY2009-2012 BOSC Review Draft- 24 August, 2009
IV. APPENDICES cont.
• The OECD (Organization tor Economic Co-Operation and Development), which
represents over 30 countries in the Americas, Europe and Asia;
• Academia;
• Chemical industry: and
• Non-governmental organizations.
Case Study Development - Significant challenges, such as interpretation and communication of
data obtained using new toxieity testing approaches, will emerge under a new paradigm for
toxicity testing, A key feature of a successful communication strategy will be to develop case
studies using new kinds of data that can serve as a basis to explore, evaluate, and most
importantly explain hazard, dose-response, and exposure information in a risk assessment
framework. Characterization of risk information, both qualitative and quantitative, in a manner
suitable lor communication to risk managers will be a significant challenge for the research and
risk assessment community, but it will be crucial if the new toxicity testing paradigm is to reach
its potential.
5.2. Strategic Goal 7: Organizational Transition
Organisational transition is meant to cover changes in direction over lime with regard to
deployment of human capital resources necessary to implement the new toxicity testing
paradigm such as hiring of scientists with particular scientific expertise and training of existing
scientific staff. For example, KPA has hired key new scientific stuff and initialed training
including three new training courses in genoniics designed and implemented by EPA's Risk
Assessment Forum. Additional resources and training programs will be needed in both KPA's
research program as well as its regulatory' and regional programs.
As noted in Section 2, several intra-agcncy. interageney, and international activities arc already
underway to begin the transformation that will change the nature of toxicity data generated and
how it is used to assess chemically induced risks to human health. Substantial funding will be
needed to provide the scientific basis for creating new testing tools; to verify the utility of new
testing tools including conducting peer review: to develop and standardize data-storage, data-
access, and data-management systems; to evaluate predictive power for humans; and to improve
the understanding of the implications of test results and how they can be applied in risk
assessments used in environmental decision-making.
EPA expects that the use of less expensive, high-throughput testing methods will allow for the
generation of toxicity data for thousands of currently untested or under-tested chemicals. The
availability of these new data will likely lead to the need for more staff to interpret the data for
many more chemicals and manage their risks. Additionally, toxieity databases such as EPA's
IRIS and models used to assess risks may need to undergo substantial changes in the long term
requiring future resources.
5.3. Strategic Goal 8: Outreach
Outreach consists of those efforts that will be used to help educate the public and stakeholders as
well as improve risk communication.
20
Previous I TOC
-------
EPA CompTox Research Program FY2009-2012 BOSC Review Draft- 24 August, 2009
IV. APPENDICES cont.
In reaching out to the public, it will be important to re-emphasize points made by EPA
Administrator Carol Browner in a 1995 memorandum to senior Agency staff about the Agency's
policy related to its new Risk Characterisation Program, lliis memorandum described tbe
importance of adhering to the "core values of transparency, clarity, consistency, and
reasonableness (which) need to guide each of us in our day-to-day work, from the lexicologist
reviewing the individual (scientific) study, to the exposure and risk assessors, to the risk
manager, and through It) the ultimate decision-maker. Further, "because transparency in
decision-making and clarity in communication will likely lead to more outside questioning of our
assumptions and science policies, we must be more vigilant about ensuring that our core
assumptions and science policies are consistent and comparable across programs, well grounded
Stakeholder Involvement Implementation of a paradigm shift in toxicity testing and related
changes to risk assessment methods and practices will require a sustained effort over many years
- remember that the NRC envisioned some 10 to 20 years to reach their goal. This transition to
new methods and approaches will need to be transparent, including efforts to share information
with both the public and risk managers. It will be critical to effectively communicate with
stakeholders (the public, scientists, federal and state agencies, industry, the mass media.
nongovernmental organizations) about tbe new tools and the overall program regarding its
strengths, limitations, and uncertainties. One way to enhance stakeholder involvement and ensure
cooperation is to hold periodic workshops where all parties can gather to share information and
progress; another tool is for EPA to establish a web portal to detail advancements in the science
and relate these to improvements in risk assessment methods and practice.
Collaboration among different elements in the research community involved in relevant research
on new testing approaches will be needed to take advantage of the new knowledge, technologies.
and analytical tools as they are developed, and collaboration between research and regulatory
scientists will be vital to ensure that the methods developed can be reliably used in risk
assessments of various types (initially qualitative, but ultimately both qualitative and
quantitative). Mechanisms for ensuring sustained communication and collaboration, such as data
sharing, will also be needed. Independent review and evaluation of the new toxieity testing
paradigm should be conducted to provide advice for mideourse corrections, weigh progress,
evaluate new and emerging methods, and make any necessary refinements in light of new
scientific challenges advances. This may be accomplished using existing KPA mechanisms for
peer review, e.g.. through reviews by the Board of Scientific Counselors, the Science Advisory
Board, and the FIFRA Scientific Advisory Panel. For testing that the Agency may wish to
require, performance standards should be considered so that individual methods from any
qualified source may be used. The NRC (2007) stressed thai "in vitro tests would be developed
not to predict the results of current [animal] apical toxicity tests but rather as [human] cell-based
assays that are informative about mechanistic responses of human tissues to toxic chemicals.
The [NRC] committee is aware of the implementation challenges that the new toxicity-testing
paradigm would face." Presumably, establishing regulatory7 confidence that the new approaches
are robust and protective of human health will be at the forefront of future challenges for EPA
and its partners.
16 http:"'www.epa.gov/oswer;nskassessnienl'pdr;1995_0521_risk_characteriitation_program.pdf
21
Previous I TOC
-------
EPA CompTox Research Program FY2009-2012
IV. APPENDICES cont.
BOSC Review Draft- 24 August, 2009
Rink Communication Communicating with policy makers and the public is an important part
of any risk management exercise. ITie complexity of the emerging toxieity testing paradigm and
how new types ofdala and information will be used to assess risk will make communication of
results challenging: consequently, the Agency must work to build public trust in the adopted
technologies. As the science moves away from well-established animal models, a significant
effort must be made to share information with risk assessors/managers and the public by clearly
describing test results and methodologies in a transparent manner. A fundamental aspect of
gaining public trust is transparency. Therefore, education and effective communication with
stakeholders (the public, scientists, regulatory authorities, industry, the mass media, and
nongovernmental organizations) on the strengths, limitations, and uncertainties of the new
tools paradigm will be critical.
Ciiven that these new methods will be less intuitive than looking for traditional effects in whole
animal studies, communication strategies will be very important. At this time, much of EPA's
effort in this area is presented on the Agency's National Center for Computational Toxicology
Web site. As the new toxicity testing paradigm continues to evolve, the Agency will need to be
vigilant in maintaining an interactive Web site to describe each individual assay or method in use
and where it fits into the exposure-response continuum.
When communicating about risk, it is important for the Agency to address the source, cause.
variability, uncertainty, and the potential adversity of tlie risks, including the degree of
confidence in the risk assessment methodology, the rationale for the risk management decision.
and the options for reducing risk (U.S. EPA. 1995; U.S. EPA, 1998). EPA will continue to
interact with stakeholders in order to develop and maintain effective informational tools.
7 hUp:''wvv\y.epa.gov/comptox.1
22
Previous
TOC
-------
EPA CompTox Research Program FY2009-2012
IV. APPENDICES cont.
BOSC Review Draft- 24 August, 2009
6, FUTURE STEPS
This strategic plan describes an ambitious and substantive change in the process by which
chemicals are evaluated for their toxicity. The NRC (2007) suggested that such a transformation
would require up to S100M per year in funding over a 10-20 year period to have a reasonable
chance of reaching the goals. Even including the resources of sister agencies, the overall federal
budget for the collaborative c(Torts does not approach the NRC proposed level of funding-
Decision on the relative role of KPA vis-a-vis other partners will have a major impact on the
resources that EPA needs to dedicate to this effort. These decisions will have to be made as the
strategy is implemented. Explanation of these decisions, their rationale, and implications will be
included in a subsequent implementation plan.
Regardless of whatever level of funding is ultimately applied to the vision of a more efficient and
effective chemical safety evaluation effort, translation of this strategy into research and activities
related to operational and organizational change will require development of an implementation
plan as well as periodic peer review of directions and progress. Representatives from those EPA
organizations most involved and impacted by the new vision will play key roles in the
implementation program. The Science Advisory Board and/or the Board of Scientific Counselors
will play key roles in the scientific peer review of the program. As noted in Section 4, there will
be a progression in the
implementation efforts from an
early focus on hazard
identification to a growing
emphasis on the use of toxicity
pathway characterization in risk
assessment. Support for
institutional transitions is also
expected to increase overtime as
the tools and technologies
emerge out of the research
programs and become available
for regulatory use. Figure 6
depicts one potential way that the
^ Screening^* rioritizat ion
Risk Assessnent
""Institutional Transit on
2010
2016 2020
Year
2025
level of effort of the three main
activities involved in this strategy
could change over time.
Figure 6. Relative (%) Emphitit of the Three Main Components of
this Strategic Plan over its Expected 20-year Duration.
Previous
TOC
-------
EPA CompTox Research Program FY2009-2012 BOSC Review Draft- 24 August, 2009
IV. APPENDICES cont.
APPENDIX: OTHER RELATED ACTIVITIES
Other US Government Activities
ITie National Toxicology Program (NTP) at the National Institute of Environmental Health
Sciences (NIKHS) coordinates loxicological testing programs within the Department of Health
and Human Services '. Similar to EPA, NTP is developing the use of computational models, in
vitro assays, and non-mammalian in vivo assays targeting key pathways, molecular events, or
processes linked to disease or injury for incorporation into a transformed chemical testing
paradigm.
Hie NIH Chemical (ienomics Center (NCCJC) of the National Human Genome Research
Institute conducts ultra high throughput screening assays as part of the Mill's Molecular
Libraries Initiative within the NIH Roadmap
A Memorandum of Understanding was recently signed by KPA, the NIP, and the NCC5C to
collaborate on generating a comprehensive map of the biological pathways alTected by
environmental chemical exposures and use this map to predict how potential chemical toxicants
will affect various types of cells, tissues, and individuals. 'I IK hope is to refine many of the
toxicity tests performed on animals and eventually supplant them with in vitro testing and
computational prediction (Collins et al., 2008).
In 2004 the Food and Drug Administration (FDA) produced a report20 addressing the need to
translate the rapid advances in basic biomedical sciences into new preventions, treatments and
cures. FDA holds large databases of human, animal, and in vitro data for screening drug
candidates for toxicity that may also be useful for screening environmental chemicals. The
FDA's National Center for Toxicological Research (NCTR) aims to develop methods for the
analysis and integration of gcnoinic, transeriptomic. proteomic. and metabolomic data to
elucidate mechanisms of toxicity31. NCTR has coordinated the Microarray Quality Control
(MAQC) project, with numerous partners including EPA (Shi et al.. 2006). In addition. NCTR
has provided its Array Track database to EPA for storage of genotnics data for research and
possible regulator.' use.
The Interagency Coordinating Committee on the Validation of Alternative Methods
(IC'C'VAM) was established by law in 2000 to promote development, validation, and regulatory
acceptance of alternative safely testing methods. ICC VAN! is made up of representatives from 15
federal agencies that generate or use toxicological data. Emphasis is on alternative methods that
will reduce, refine, and'or replace the use of animals in testing while maintaining and promoting
scientific quality and the protection of human health and the environment". The NTP
Interagency Center for the Evaluation of Alternative Toxicological Methods (NICE ATM)
administers and provides scientific support for 1CCVAM. ICCVAM/N1CEATM evaluates test
method submissions and nominations, prepares technical review documents, and organi/.es
18 http ntp mehs mh gi^ nip main p'Ot N211 uc mitidU\ewiiUuilpath\shitepdper.html
~' http ivutt Ida go\ mtr o> en icw mission htm
22 http ILC\ am niehs nih go\ .ihout m_(j \ him
24
Previous I TOC
-------
EPA CompTox Research Program FY2009-2012 BOSC Review Draft- 24 August, 2009
IV. APPENDICES cont.
scientific workshops and peer review meetings. For example. ICCVAM NICEATM recently
released a report" that describes two in vitro eytotoxieily tests that can be used for estimating
starting doses tor acute oral toxicity tests, thereby reducing the number of animals used.
Related Activities by Foreign Governments
A new European Union Regulation on Registration, Evaluation, Authorization, and
Restriction of Chemicals (REACH) went into effect June 1, 2007. 'Hie main goals of REACH
are (1) to improve the protection ol human health and the environment from risks associated \\ith
chemicals in commerce and (2) to promote alternative lest methods. REACH requires
manufacturers and importers to demonstrate they have appropriately identified and managed the
risks of substances produced or imported in quantities of one ton or more per year per company.
'Hie new Kuropean Chemicals Agency (KCHA)"' will manage the system databases, coordinate
evaluation of chemicals, and run a public database of hazard information" .
The Kuropean Centre for the Validation of Alternative Methods (KCVAM)" coordinates
the validation of alternative test methods in the European Union. ECVAM develops, maintains,
and manages a database on alternative procedures and promotes the development, validation, and
international recognition of alternative lest methods.
The Japanese Center for the Validation of Alternative Methods (JaCV'AM) is pail of the
Japanese National Institute of Health Sciences. JaCVAM has conducted validation studies for
alternative test methods and participates in international validation efforts",
The Korean Center for the Validation of Alternative Methods (KoC.'V AM) is a branch of
NITR, the National Institute of Toxicological Research. NITR is collaborating with the Korean
Societv for Alternatives to Animal Experiments (KSAAE) to refine methods in acute oral.
-)0
reproductive development, genetic, and endocrine toxicity testing"'.
The Organization for Economic Co-Operation and Development (OECD) represents 30
countries in the Americas (including the United States). Europe, and Asia. The OECD
"Guidelines for the Testing of Chemicals"' provides a collection of internationally harmonized
testing methods for a number of lexicological cndpoints using in vivo, in vitro, and even
alternative approaches." Test guidelines can be updated to reflect scientific advances and the
state of the science if member countries agree to do so. A few OECD workgroups and efforts
address issues relevant to this KPA strategy, e.g., the OECD QSAR Toolbox30 and the joint
OECD I PCS (International Programme for Chemical Safety) Toxicogenomics Working Group,
which has developed a proposal for a Molecular Screening Project, modeled after EPA's
ToxCast program.
23 http k«.\jm nidisnihgo\ melhoiK diuicloxmi nru Imcr.hlm
l> http ei,h,] europa eu re,u.h_en a,p
J' hup ci cuiopj ou em ironwent chonin.jK leach reach inlro htm
u http OA dm in. il
Jl hup www nihsgd jp english index html
m http w\\wMA mi ac ip ivue P \Kk pdl
"'http tiUinid souricoci Jorg\l ^vdHHltl 2? nw f ;'rpsv/pcntxIitaL;pl5 about.hlm?jnlissn 1607310.x
30 http www oeLd or
-------
EPA CompTox Research Program FY2009-2012 BOSC Review Draft- 24 August, 2009
IV. APPENDICES cont.
Academia
Numerous I .S. academic researchers and centers are funded by NIH or EPA s National Center
for Environmental Research to develop assays and analysis methods that might be helpful to the
goals of this KPA research strategy, This includes two Bioinlornialies Centers funded by KPA in
2006.
'Hie European Commission funds several large academic, government, and industry consortia
that are conducting research that could lead to effective in vitro toxicity tests. The CASC'ADK
Network of Excellence studies human health effects of chemical residues and contaminants in
food and drinking water, designing assays to elucidate estrogen, testosterone, and thyroid
hormone pathways for the development of mechanism- and disease-based test methods. The aim
of the carcinoCiKNOMK.'S"* project is to develop in vitro methods for assessing the
carcinogenic potential of compounds. ReProTect" is optimizing an integrated set of
reproductive developmental tests for a detailed understanding of gametogenesis, sleroidogenesis,
and embryogenesis that can support regulatory decisions.
Industry
The European Partnership for Alternative Approaches to Animal Testing (EPAA)" is a
joint initiative from the European Commission and a number of companies and trade federations.
Its purpose is to promote the development of alternative approaches to safety testing. The KPA A
focuses on mapping existing research; developing new alternative approaches and strategies; and
promoting communication, education, validation, and acceptance of alternative approaches,
Non-Governmental Organizations (NGOs)
The Comparative Toxicogenomics Database " (C I'D) elucidates molecular mechanisms by
which environmental chemicals affect human disease. CTD includes manually curated data
describing cross-species chemical-gene, protein interactions and chemical - and gene- disease
relationships to illuminate molecular mechanisms underlying variable susceptibility and
environmentally influenced diseases. These data will also provide insights into complex
chemical-gene and protein interaction networks.
The Johns Hopkins Outer for Alternatives to Animal Testing36 supports the creation.
development, validation, and use of alternatives to animals in research, product safety testing.
and education. Similarly. AItTox.org provides information on non-animal methods for toxicity
testing including a table'8 that summarizes the alternative testing methods by endpoint that have
been approved or endorsed internationally by at least one regulatory agency.
31 http••'••'www.cascadcncl org;
3J http:/'www,carcinogenotnics.eu
31 http ,'/www rcprotcct.cu,1'
M http:'•ec.curopa cu'cntcrprisc'cpaa- index en.htm
35 hup',''tld.mdibl.org/
36 hUp:'/3ltweb.|hsph,edttinde,\.htrn
3 http-.-';www.alttox.org/aboul.
K http;,-'.-www nlUox,org'lire'validalion-ra'ValidHled-ra-methods.btinl
26
Previous I TOC
-------
EPA CompTox Research Program FY2009-2012 BOSC Review Draft- 24 August, 2009
IV. APPENDICES cont.
REFERENCES
Coecke S, Goldberg AM, Allen S, Buzanska I,, Calamandrei G. Croftort K, Ilareng I., Ilartung T, Knaut
II, Ilonegger P. Jacobs VI, I.ein V, I.i A, Mundy \\, Owen D, Schneider S, Silbergeld E, Reum 1,
Trnovec I, Monnet-Tschudi 1% Bal-Price A. (2007) Workgroup report: incorporating in vitro alternative
methods for developmental neurotoxieity into international hazard and risk assessment strategies. Environ
Health I'erspect. 115(6):924-31.
Collins FS, Gray GM, BucherJR. (2008) Transforming Environmental Health Protection, Science
319:906-7.
Dix DJ, Houek KA. Martin MT, Richard AM, Set/er RW. Kavlock RJ. (2007) The ToxCasI program lor
prioritizing toxicity testing of environmental chemicals. Taxical Sci. 95( 1 ):5-l 2.
Kdwards S\V, Preston R.I. Systems Biology and Mode of Action Based Risk Assessment. Taxical. Sci. in
press, doi: 10.1093 toxsci 'kfnl 90.
Guylon KZ, Kyle AD. Aubrechl J. Cogliano VJ. Easlmond DA, Jackson M. Keshava N. Sandy MS,
Sonawane B, Zhang L, Waters MD, Smith MT. (2008) Improving prediction of chemical carcinogenicity
by considering multiple mechanisms and applying toxicogenomic approaches. Afutat. Res. In press,
doi:10,10]6j.mn-ev,20()8.U).001.
Judson R, Richard A. Dix D, llouck K, Elloumi !•', Martin M Cathey 1, Iransue TR, Spencer R, Wolf M.
(2008) ACToR—Aggregated Compulalional Toxicology Resource. Taxicol App]Pharmacol. 233(1):7-13.
Kavlock RJ, Ankle> G, Blancato J. Brccn M, Conolly R, Dix D, llouck K. Ilubal E. Judson R,
Rabinowilz J, liichard A, Sclzcr RW. Shah I. Villcncuvc D. Wcbcr E. (2007) Computational Toxicology
A State oi the Science Mini Review. Toxicol Sci. Dec 7.
Kavlock RJ. Dix DJ, llouck KA, Judson RS, Martin MT. Richard AM. (200K). ToxCasI™: Developing
predictive signatures for chemical toxicity. Proceedings of the 6th World Congress on Alternatives in the
Life Scie««'.v.TX-4-1. In Press.
Lein P, Locke P, Goldberg A. (2007) Meeling reporl: alternative!! lor developmental neuroloxicity testing.
Environ Health Perspeci. 11 5(5):764-8.
National Research Council of the Xalional Academies (NRC). (2006) Toxicity Testing lor Assessment of
Environmental Agents. The National Academies Press. Washington, DC.
National Research Council of the National Academies (NRC). (2007) Toxieity Testing in the 21"
Century: A Vision and A Strategy, 'llie National Academies Press. Washington, DC.
Shi L et al. (2006) The MicroArray Quality Control (MAQC) project shows inler- and intraplaiform
reproducibility of gene expression measurements. Hat Biatechnol. 24(9):1151 -61.
U.S. Environmental Protection Agency. (1995) Ecological risk: a primer for risk managers. Washington,
DC: U.S. Environmental Protection Agency. EPA. 734i'R-95<001. Available al htlp: www.epa.gov nscep .
U.S. Environmental Protection Agency. (1998) Guidelines for Ecological Risk Assessment. Washington,
DC: U.S. Environmental Protection Agency. EPA 630/R-95/002F.
http: cfpub.epa.gov. ncea.clhtrecordisplay.cfm?deid-12460
27
Previous I TOC
-------
EPA CompTox Research Program FY2009-2012
IV. APPENDICES cont.
BOSC Review Draft- 24 August, 2009
&EPA
United States
Environmental Protection
Agency
PRESORTED STANDARD
EPA
PERMIT i:
Office of the Science Advisor (G105R)
Washington. DC 2D46O
Official Business
Penalty tor Private Use
$300
Previous
TOC
-------
CTRP BOSC Poster Abstracts
September 29-30, 2009
Session I
Session II
Previous I TOC
-------
Session 1
Poster
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
Title
ACToR: Aggregated Computational Toxicology
Resource
DSSToxand Chemical Information Technologies in
Support of Predictive Toxicology
Characteristics and Applications of the ToxRefDB In Vivo
Datasets from Chronic, Reproductive and Developmental
Assays
Literature Mining and Knowledge Discovery Tools for
Virtual Tissues
ExpoCast: Exposure Science for Prioritization and
Toxicity Testing
Computational Approaches and Tools for Exposure
Prioritization and Biomonitoring Data Interpretation
Advanced Exposure Mertics for Chemical Risk Aanalysis
The U.S. "Tox21 Community" and the Future of
Toxicology
Supported STAR Research Centers Advance the Field of
Computational Toxicology
UNC The Carolina Environmental Bioinformatics Center
UMDNJ Environmental Bioinformatics & Computational
Toxicology Center (ebCTC): Research Collaborations in
Multi-Scale Modeling of Environmental Toxicants
UNC Systms Biology - Carolina Center for
Computational Toxicology
UTH - The Texas-Indiana Virtual STAR Center; Data-
Generating in vitro and in silico Models of Developmental
Toxicity in Embryonic Stem Cells and Zebrafish
Mechanistic Indicators of Childhood Asthma (MICA): A
Systems Biology Approach for the Integration of
Multifactorial Environmental Health Data
Linkage of Exposure and Effects Using Genomics,
Proteomics, and Metabolomics in Small Fish Models
Development of a Searchable Metabolite Database and
Simulator of Xenobiotic Metabolism
Risk Assessment of the Inflammogenic and Mutagenic
Effects of Diesel Exhaust Particulates: A Systems
Biology Approach
Development of Microbial Metagenomic Markers for
Environmental Monitoring
Develop a Systems Approach to Characterizing and
Predicting Thyroid Toxicity Using an Amphibian Model
Presenters Bio
R Judson
A Richard
M Martin
A Singh
E Cohen Hubal
C. Tan
T. Collette
R Tice (NTP/NIEHS)
D Segal (NCER)
F Wright (UNC)
B Welsh (UMDNJ)
I Rusyn (UNC)
Eva M. Bondesson (UTH)
J Gallagher (NHEERL)
T Collette for G. Ankley
(NERL)
J Jones (Athens)
J Samet (NHEERL)
Jorge W. Santo Domingo
(NERL)
S Degitz (NHEERL)
Previous
TOC
-------
ACToR: Aggregated Computational Toxicology Resource
Authors: Richard Judson1, Tommy Cathey2, Thomas Transue2, Ann Richard1, Doris
Smith3, James Vail3, Kaitlin Daniel1
1 National Center for Computational Toxicology, USEPA, RTP, NC 27711
2 Lockheed Martin, Contractor to the USEPA, RTP, NC 27711
3The National Caucus and Center on Black Aged, Inc., Senior Environmental
Employment Program, Grantee to the USEPA, RTP, NC 27711
The EPA Aggregated Computational Toxicology Resource (ACToR) is a set of
databases compiling information on chemicals in the environment from a large number
of public and in-house EPA sources. ACToR has 3 main goals: (1) The serve as a
repository of public toxicology information on chemicals of interest to the EPA, and in
particular to be a central source for the testing data on all chemicals regulated by all
EPA programs; (2) To be a source of in vivo training data sets for building in vitro to in
vivo computational models; (3) To serve as a central source of chemical structure and
identity information for the ToxCast™ and Tox21 programs. There are 4 main
databases, all linked through a common set of chemical information and a common
structure linking chemicals to assay data: the public ACToR system (available at
http://actor.epa.gov), the ToxMiner database holding ToxCast and Tox21 data, along
with results form statistical analyses on these data; the Tox21 chemical repository which
is managing the ordering and sample tracking process for the larger Tox21 project; and
the public version of ToxRefDB. The public ACToR system contains information on
~500K compounds with toxicology, exposure and chemical property information from
>400 public sources. The web site is visited by ~1,000 unique users per month and
generates ~1,000 page requests per day on average. The databases are built on open
source technology, which has allowed us to export them to a number of collaborating
organizations.
This work was reviewed by EPA and approved for publication but does not necessarily
reflect official Agency policy.
Previous I TOC
-------
DSSTox and Chemical Information Technologies in Support of Predictive
Toxicology
Ann M. Richard, National Center for Computational Toxicology, USEPA, RTP, NC
The EPA NCCT Distributed Structure-Searchable Toxicity (DSSTox) Database project
initially focused on the curation and publication of high-quality, standardized, chemical
structure-annotated toxicity databases for use in structure-activity relationship (SAR)
modeling. In recent years, the project has expanded to include: creation of DSSTox
files for high-interest EPA chemical inventories; strengthening structure-based linkages
among public resources; tailoring chemical and bioassay DSSTox content for
incorporation into NIH's PubChem; creating local structure-browsing capabilities of
DSSTox content and inventories; and expanding comparability of and linkages to gene
expression data. Within the NCCT, the DSSTox project framework is applying strict
quality standards for chemical information, pertaining to both generic (ACToR,
ToxRefDB) and actual test substances (ToxCast™, Tox21). Within these projects, we
are working to expand comparability and linkages of summarized toxicity data in the
context of a standardized cheminformatics environment. Future research will build on
these cheminformatics data foundations and enriched data resources to develop data
mining strategies to explore new and flexible ways to relate chemical structure to
biological endpoints (e.g., reactivity groupings, biofunctional or toxicity-informed
similarity, chemical feature space), and new representations of biological endpoints in
relation to structure-based modeling (e.g., HTS clusters, bioassay profiles, summarized
or grouped effects, qualitative active and inactive classes). Incorporating traditional
SAR concepts into this new HTS data-rich world poses conceptual and practical
challenges, but also holds great promise for improving toxicity prediction capabilities.
This work was reviewed by EPA and approved for publication but does not necessarily
reflect official Agency policy.
Previous I TOC
-------
Characteristics and Applications of the ToxRefDB In Vivo Datasets from
Chronic, Reproductive and Developmental Assays
Matthew Martin, NCCT/ORD, USEPA, Research Triangle Park, NC, USA.
ToxRefDB was developed to store data from in vivo animal toxicity studies. The
initial focus was populating ToxRefDB with pesticide registration toxicity data that
has been historically stored as hard-copy and scanned documents by the Office of
Pesticide Programs. A significant portion of these data have now been processed
into ToxRefDB in a standardized and structured format. ToxRefDB currently includes
chronic, cancer, sub-chronic, developmental, and reproductive studies on over 400
chemicals, many of which are pesticide active ingredients. These data are now
computable within ToxRefDB, and are serving as reference toxicity data for the
development of ToxCast™ predictive signatures as well as for retrospective
analyses assessing past performance of guideline toxicity studies and informing on
potential changes to current guidelines. The three primary datasets currently being
used for predictive modeling have been published and include chronic, reproductive
and developmental endpoints. The rat and mouse chronic data primarily focuses on
pathological endpoints related to progression and formation of tumors. The
reproductive data is from rat multi-generation studies and focuses on reproductive
performance measures, reproductive organ pathologies, and offspring survival
decrements. The developmental endpoints are from rat and rabbit prenatal studies
for which detailed anatomical information is collected on observed malformations
which were subsequently mapped to the developing system. Thus ToxRefDB
provides high quality, comparable data for over 200 of the of the 309 ToxCast™
chemicals from chronic, reproductive and developmental study types. This work was
reviewed by EPA and approved for publication but does not necessarily reflect
official Agency policy.
Previous I TOC
-------
Literature Mining and Knowledge Discovery Tools for Virtual Tissues
Singh AV1, Knudsen T B2 and Shah I2
1 Lockheed Martin, Contractor to the USEPA, RTP, NC
2 National Center for Computational Toxicology, USEPA, RTP, NC
Virtual Tissues (VTs) are in silico models that simulate the cellular fabric of tissues to
analyze complex relationships and predict multicellular behaviors in specific biological
systems such as the mature liver (v-Liver™) or developing embryo (v-Embryo™). VT
models require input of biological knowledge about the systems under investigation. We
are using VTs to model experimental data, such as ToxCast™ in vitro assays, with
information about molecular pathways, cellular networks and clinical phenotypes in target
organ systems. Knowledgebase development requires a flexible platform to extract and
organize relevant facts from the scientific literature and other sources of information. The
knowledge discovery workflow starts with information retrieval (IR) by user-defined input
on single or multiple keywords to retrieve relevant PubMed abstracts, followed by
information extraction (IE) and relationship mapping (RM). Currently, we use the publicly
available '@Note'1 tool for highly customizable named entity recognition (NER). A
vocabulary of terms was built to describe pathologically relevant concepts using publicly
available ontologies (www.OBOfoundrv.org/) including genes, pathways, anatomy, clinical
outcomes, and chemicals. The results from @Note are stored in a relational database for
statistical analyses to summarize relationships and map them to broader biological
concepts. The text-mining (TM) workflow is being implemented as a modular tool that
uses open-source libraries. We are using this workflow to extract relevant facts about
hepatocarcinogenesis and embryo dysmorphogenesis. This poster will provide specific
examples of predicted associations that were data-mined from ToxCast_320 chemicals
and 76 ToxRefDB endpoints. Biomedical literature mining aims to identify and extract
plausible patterns for explicit (IE) and implicit (TM) concepts that are previously known
and unknown, respectively and that can be used to better understand the inferred
associations in predictive modeling. [This work has been reviewed by EPA and cleared for
presentation, but does not reflect official Agency policy].
http://sysbio.di.uminho.pt/anote/wiki/index.php/Main Page.
Previous I TOC
-------
ExpoCast: Exposure Science for Prioritization and Toxicity Testing
Elaine A Cohen Hubal1 and Peter Egeghy2
1 National Center for Computational Toxicology, U.S. EPA, RTP, NC, USA
2 National Exposure Research Laboratory, U.S. EPA, RTP, NC, USA
The US EPA is completing the Phase I pilot for a chemical prioritization research
program, called ToxCast™. Here EPA is developing methods for using
computational chemistry, high-throughput screening, and toxicogenomic
technologies to predict potential toxicity and prioritize limited testing resources.
There is a clear need for a parallel and collaborative effort across the exposure
and risk assessment community to provide the exposure science required for
interpretation of high-throughput in vitro toxicity data. A coherent research
program is required to advance exposure characterization to translate advances
and findings in computational toxicology for enhanced risk assessment, informed
decision making and improved public health. US EPA is initiating the
ExpoCast™ program to ensure that the required exposure science and
computational tools are ready to address global needs for rapid characterization
of exposure potential arising from the manufacture and use of tens of thousands
of chemicals and to meet challenges posed by new toxicity testing approaches.
ExpoCast™ will provide an overarching framework for science required to
characterize biologically-relevant exposure in support of the Agency
computational toxicology program. The overall goal of this program is to develop
novel approaches and tools for evaluating and classifying chemicals, based on
potential for biologically-relevant human exposure, to inform prioritization and
toxicity testing. Broadly and long-term, the ExpoCast™ program will foster novel
exposure science research to (1) inform chemical prioritization, (2) understand
implications of system response to chemical perturbations at the individual and
population levels, (3) link information on potential toxicity of environmental
contaminants to real-world health outcomes. This presentation will introduce
EPA's ExpoCast™ program.
This work has been reviewed and approved by the US EPA for publication but
does not necessarily reflect Agency policies.
Previous I TOC
-------
Computational Approaches and Tools for Exposure Prioritization and
Biomonitoring Data Interpretation
Cecilia Tan, Eric Weber, John Kenneke, Marsha Morgan, Daniel Chang, Michael-Rock
Goldsmith, Rogelio Tornero-Velez, Curt Dary
National Exposure Research Laboratory, Office of Research and Development, U.S.
Environmental Protection Agency, Research Triangle Park, NC 27711 USA
The ability to describe the source-environment-exposure-dose-response continuum is
essential for identifying exposures of greater concern to prioritize chemicals for toxicity
testing or risk assessment, as well as for interpreting biomarker data for better
assessment of exposure or risk. To link each element in this continuum, scientists at
the National Exposure Research Laboratory (NERL) and the National Center for
Computational Toxicology (NCCT) are collaborating to develop, evaluate, and apply
various computational approaches and tools including predictive environmental fate
modeling (i.e., Environmental Fate Simulator (EPS)), exposure modeling, physiological
based pharmacokinetic (PBPK) modeling, and pharmacodynamic modeling, and
biologically based dose-response modeling. Specifically, NERL currently directs
research activities in the following areas: (1) EPS; (2) screening level PBPK modeling;
and (3) interpretation and use of biomonitoring data. The components of EPS include:
a computational tool for calculating physical and chemical properties based on chemical
structure; a reaction pathway simulator for predicting transformation pathways and
products; linked databases populated with measured/calculated molecular descriptors
necessary for predicting physical transport and chemical reactivity; an expert system for
environmental characterization data needed to estimate partitioning behavior and
reactivity; and the EPS software that provides seamless linkage of disparate models
and databases. Screening level PBPK models are used to link external exposures to
tissue dosimetry for improved dose response assessment. NERL scientists utilize a
combination of results from in vitro/in vivo studies and QSAR/computational chemistry
techniques to estimate chemical-specific parameters required for PBPK models. This
knowledge and expertise are especially important for chiral chemicals, which exist as
mixtures of stereoisomers having different physical and/or biological properties but are
frequently treated in toxicity testing and risk assessment as single chemicals. For
interpreting biomonitoring data, NERL is developing a framework to use the same
computational approaches (e.g., PBPK modeling) to assess the quantitative
relationships between biomarkers and human exposures. For example, an exposure
study is underway to estimate non-occupational exposure to pyrethroids based on
urinary biomarkers. This framework will identify the critical data gaps and uncertainties
in estimating human exposures and will help in designing future exposure and
epidemiological studies.
This work was reviewed by EPA and approved for publication but does not necessarily
reflect official Agency policy.
Previous I TOC
-------
Advanced Exposure Metrics for Chemical Risk Analysis
Timothy W. Collette, Quincy Teng, Drew Ekman, Jim Lazorchak, David Lattier,
Michael-Rock Goldsmith, Joachim Pleil, National Exposure Research Laboratory,
USEPA
Direct measurement of human exposure to environmental contaminants in real
time (when the exposure is actually occurring) is rare and difficult to obtain. This
frustrates both exposure assessments and investigations into the linkage
between chemical exposure and human disease. However, it is feasible to
obtain information on the levels of environmental contaminants (and their
metabolites and adducts) in the biofluids of individuals that may have been
exposed. Furthermore, it is feasible to obtain information on the occurrence of
specific diseases and other adverse conditions in various human demographics.
The Agency's exposure and risk assessments could be greatly improved if these
chemical biomarkers could be used to both reconstruct previous exposure
scenarios, and to predict the future likelihood of adverse effects. While progress
has, indeed, been made along these paths, biomarker methods based solely on
xenobiotics and their metabolites/adducts are inherently limited. This new
research program (still in the planning stages) is based on the belief that systems
biology approaches and 'omic-based biomarkers, when used in conjunction with
tradition biomarkers, offer great promise for both exposure reconstruction and for
elucidating the linkages between exposures and adverse outcomes.
A significant amount of research has already been devoted to the use of systems
biology approaches and 'omic techniques (transcriptomics, proteomics, and
metabolomics) to screen chemicals for hazardous effects. However, changes in
transcripts, proteins, and endogenous metabolites may, in some cases, be more-
certain indicators of chemical exposures than of apical chemical effects.
Nonetheless, these powerful new techniques have rarely been applied as
biomarkers of exposure. In comparison to (or in combination with) conventional
biomarkers, 'omic markers of exposure offer considerable promise for exposure
assessment. Note that 'omic markers are a unique pattern of a large number
and wide variety of transcript, protein, or endogenous metabolite changes.
These signatures may be more informative and more chemical-specific than a
conventional biomarker. Also, taking advantage of the earlier research in 'omic
markers for effects, these markers of exposure may, in some cases, be able to
identify exposure to a specific mode-of-action-active chemical. Indeed, 'omic
markers can sometimes serve as a linkage across the source-to-outcome
continuum, functioning concomitantly as markers of exposure, dose
characterization, and effects.
This abstract has been reviewed in accordance with the U. S. Environmental
Protection Agency's peer and administrative review policies and approved for
presentation and publication.
Previous I TOC
-------
The U.S. "Tox21 Community" and the Future of Toxicology
Raymond Tice, Ph.D. 1 Robert Kavlock, Ph.D.2, and Christopher Austin, M.D.3
1 Chief, Biomolecular Screening Branch, National Toxicology Program, National Institute
of Environmental Health Sciences, RTP, NC 27709
2 National Center for Computational Toxicology, USEPA, RTP, NC 27711
3Director, NIH Chemical Genomics Center, National Human Genome Research Institute
National Institutes of Health, Bethesda, MD 20892-3370
In early 2008, the National Institute of Environmental Health Sciences/National
Toxicology Program, the NIH Chemical Genomics Center, and the Environmental
Protection Agency's National Center for Computational Toxicology entered into a
Memorandum of Understanding to collaborate on the research, development, validation,
and translation of new and innovative test methods that characterize key steps in
toxicity pathways. A central component is the exploration of high throughput screening
assays and tests using phylogenetically lower animal species (e.g., fish, worms), as well
as high throughput whole genome analytical methods, to evaluate mechanisms of
toxicity. The goals of the "Tox21 Community" are to investigate the use of these new
tools to (1) prioritize substances for further in-depth toxicological evaluation, (2) identify
mechanisms of action for further investigation, and (3) develop predictive models for in
vivo biological response. Success is expected to result in test methods for toxicity
testing that are more mechanistically based and economically efficient; as a
consequence, a reduction or replacement of animals in regulatory testing is anticipated
to occur in parallel with an increased ability to evaluate the large numbers of chemicals
that currently lack adequate toxicological evaluation. The initial focus of this
collaboration has been on identifying toxicity-related pathways (and assays for those
pathways), establishing a Tox21 library of ~10000 compounds, and developing the
databases and bioinformatic tools needed to mine the resulting data. The coordinated
approaches being taken to achieve our goals, the lessons learned, and expectations for
the future will be presented. This work was reviewed by EPA and approved for
publication but does not necessarily reflect official Agency policy.
Previous I TOC
-------
NCER-Supported STAR Research Centers Advance the Field of Computational
Toxicology
Deborah Segal, National Center for Environmental Research (NCER), USEPA,
Washington, DC
Advances in genomics and computer methods have positioned computational
toxicology at the forefront in the development of predictive models of exposures to
pollutants and their subsequent health effects. These models will improve the scientific
foundation for conducting risk assessments. In an effort to advance the field, ORD's
National Center for Environmental Research (NCER), through the Science To Achieve
Results (STAR) program, has funded research centers that are integrating modern
computing and information technology with molecular biology and chemistry. The goals
of the STAR computational toxicology program are the following:
• Improve linkages across the source-to-outcome continuum
• Develop approaches for prioritizing chemicals for further screening and testing
• Produce better methods and predictive models for quantitative risk assessment
NCER has issued three Requests for Application (RFAs) for STAR computational
toxicology research centers. As a result of a 2005 RFA "Environmental Bioinformatics
Research Centers," EPA funded two centers that are developing statistical and
bioinformatics tools and approaches for predicting toxicity to chemical exposures. A
2007 RFA, "Computational Toxicology Centers: Development of Predictive
Environmental and Biomedical Computer-Based Simulations and Models," led to an
additional center. It is applying high-performance computing techniques and resources
to in silico multi-scale modeling applications at the cellular, organ, and system-wide
level to address the environmental problems and research needs facing the U.S. In July
of 2009, the most recent center grant was awarded following the issuance of the RFA,
"STAR Computational Toxicology Research Centers: In vitro and in silico Models of
Developmental Toxicity Pathways." This center will bridge the interface of in vitro data
generation and in silico model development to answer critical biological questions
related to toxicity pathways important to human development. The first three centers
are operating as cooperative agreements with ORD. This has enabled productive
collaborations between center Pis and EPA scientists and has resulted in numerous
jointly published journal articles in the peer reviewed literature.
This work was reviewed by EPA and approved for publication but does not necessarily
reflect official Agency policy.
Previous I TOC
-------
The Carolina Environmental Bioinformatics Center
Fred A. Wright, Alexander Tropsha, Leonard McMillan, Ivan Rusyn
The University of North Carolina at Chapel Hill, Chapel Hill, NC 27599
The Carolina Environmental Bioinformatics Center brings together multiple investigators
and disciplines, combining expertise in biostatistics, computational biology, chem-
informatics and computer science to advance the field of Computational Toxicology.
The Center is developing novel analytic and computational methods, creating efficient
user-friendly tools to disseminate to the wider community, and applying the
computational methods to data relevant to chemical toxicology. Effort is divided into
three Research Projects, with an emphasis on collaboration within the Center and with
the EPA. Project 1: Biostatistics in Computational Biology (PI Wright) provides
biostatistical support to the Center, performing analysis and developing new methods in
collaboration with EPA personnel and the computational toxicology community. This
Project's investigators have engaged in numerous collaborations with EPA in diverse
areas, ranging from pathway dose-response methods development to analysis of-
omics response to pesticide exposures. The Project has also contributed directly to the
EPA NCCT's involvement with the Microarray Quality Control Consortium II. In addition,
Project 1 is actively working on prediction methods for EPA ToxCast® data, bringing
together Center investigators and industry partners in machine learning methods for
toxicity prediction. Project 2: Chem-informatics (PI Tropsha) coordinates the compilation
and mining of data from relevant external databases and performs analysis and
methods development for building statistically significant and externally predictive
Quantitative Structure-Activity Relationship models of chemical toxicology data. In
addition, Project 2 is developing computational tools to perform these tasks, and has
longstanding collaborations with EPA investigators in DSSTox data analysis.
Moreover, Project 2 is providing important chemical descriptor data to improve toxicity
prediction for ToxCast® data analysis, and providing context for further prioritization in
chemical testing. Project 3: Computational Infrastructure for Systems Toxicology (Pis
Rusyn and McMillan) has created a framework for handling emerging -omics data on
genetic susceptibility in model organisms. In addition, Project 3 provides programming
expertise to create graphical tools that are used by partners within the Center and with
collaborators at the EPA. The synergy of the interactions within the Center, as well as
with the EPA and toxicological community, are strengthening and advancing the field of
computational toxicology through direct partnerships and the dissemination of tools
used by both bioinformatics and bench scientists.
Previous I TOC
-------
Environmental Bioinformatics & Computational Toxicology Center (ebCTC):
Research Collaborations in Multiscale Modeling of the Effects of Environmental
Toxicants
William J. Welsh and Panes G. Georgopoulos, University of Medicine and Dentistry -
R.W. Johnson Medical School
The USEPA-funded environmental bioinformatics and Computational Toxicology Center
(ebCTC) is a research consortium of the University of Medicine and Dentistry - R.W.
Johnson Medical School, Princeton University, Rutgers University, and the Center for
Toxicoinformatics of the US Food and Drug Administration. ebCTC augments USEPA's
research at its National Center for Computational Toxicology (NCCT) through the
development and application of novel computational methods supporting the
mechanistic assessment of health risks associated with environmental factors.
ebCTC brings together a multidisciplinary team of computational scientists and
engineers, with backgrounds in cheminformatics, enviroinformatics, bioinformatics, and
mechanistic process modeling, to address, collaboratively and simultaneously, multiple
elements of the "environmental health sequence" from "source" (e.g. the release or
formation of a "stressor", such as a chemical, radiological, or biological agent in an
environmental medium) to "outcome" (e.g. the development of an environmentally
caused disease). This effort is pursued through the study and elucidation of the cascade
of individual events and processes involved in the above source-to-outcome continuum
within a consistent and integrative multiscale analysis framework. The ebCTC approach
considers human health state as the result of coupled dynamic systems spanning
multiple scales of "biological space" (i.e. involving processes and interactions at the
scales of molecules, cells, tissues, organs, organisms, and populations). This
integrative analysis framework embraces a general "reverse engineering" approach,
that incorporates multiple diagnostic and prognostic modeling methodologies, to study
and reveal the hierarchical structures and functional dynamics of multiscale biological
systems in relation to their perturbations by behavioral and environmental influences.
The methods and computational tools that are being developed through the above effort
are extensively evaluated and refined through collaborative applications involving
ebCTC scientists and colleagues from the three consortium universities, USFDA, and
USEPA. Examples of ongoing (or completed) projects that involve extensive
collaborations of ebCTC with USEPA scientists include: the development and
application of methods for reconstructing population exposures to environmental
chemicals from biomarker data; the biologically-based modeling of multimedia,
multipathway, and multiroute population exposures to arsenic and arsenic compounds;
the development of biologically-based toxicokinetic and toxicodynamic models that
incorporate the effects of aging in physiological and biochemical process dynamics; the
incorporation of toxicogenomic data in human health risk analyses with dibutyl phthalate
(DBP) as a case study; the pathway analysis of microarray data collected following
exposures to conazoles; etc. ebCTC researchers are also interacting continuously and
extensively with USEPA scientists in numerous research activities, that include the
development of methods and tools related to the Virtual Liver project efforts, and the
analysis of data that are becoming available through the ToxCast project.
Previous I TOC
-------
Carolina Center for Computational Toxicology
Ivan Rusyn, Shawn Gomez, Timothy Elston, Fred Wright, Alexander Tropsha
The University of North Carolina at Chapel Hill, Chapel Hill, NC 27599
The Carolina Center for Computational Toxicology is engaged in a broad interdisciplinary
effort to devise novel tools, methods and knowledge to assist the regulatory agencies
and the greater environmental health sciences community in protecting the environment
and human health. Using publicly available data, the Center applies knowledge and
expertise of the individual investigators and teams to develop complex predictive
modeling solutions that range from mechanistic, interpretative data modeling to
discovery and decision support efforts. Furthermore, each of the Center's three
Research Projects is engaged in active collaboration with other projects within both the
Center and the EPA. Project 1: Biomedical modeling of chemical-perturbed networks
(Pis Gomez and Elston), is mapping chemical-perturbed networks and devising
modeling tools that can predict the pathobiology of the test compounds based on a
limited set of biological data. This group of investigators is actively collaborating with the
v-Liver® project and are examining the network biology relationships using ToxCast®
and ToxRefDB® data. Project 2: Toxico-genetic modeling (Pis Wright and Rusyn), is
building computational tools that will enable toxicologists to understand the role of
genetic diversity among individuals in responses to toxicants. This project is most
actively engaged in data analysis for the ToxCast®, ToxRefDB® and ACToR projects at
the EPA and is also conducting biological experiments in genetically defined human and
mouse cells in collaboration with the Tox21 partners (EPA, NIEHS/NTP and NCGC).
Project 3: Chem-informatics (PI Tropsha) is engaged in unbiased discovery-driven
prediction of adverse chronic in vivo outcomes based on statistical modeling of known
relationships between chemical structures, biological screening results, and the genetic
makeup of the organism. This project continues its historically tight interactions with
DSSTox, ACToR, ToxCast®, and ToxRefDB® teams within EPA through data analysis
and model development efforts. Collectively, the Center not only advances the field of
computational toxicology through the development of new methods, software, and applied
models but also advocates the application of the knowledge into practice through
collaborative efforts with EPA and other governmental stakeholders. The emphasis of our
work is on the usability of the outputs by the risk assessment community and the
investigative toxicologists thus facilitating the transition of the field of computational
toxicology from a hypothesis-driven toward a predictive science.
Previous I TOC
-------
NCER STAR GRANT ABSTRACT
EPA Grant Number: 83428901
Title: The Texas-Indiana Virtual STAR Center; Data-Generating in vitro and in silico
Models of Developmental Toxicity in Embryonic Stem Cells and Zebrafish
Investigator(s):
1. Prof. Jan-Ake Gustafsson (Contact PI) E-mail: jgustafsson@uh.edu
2. Prof. Richard H. Finnell E-mail: rfinnell@ibt.tamhsc.edu
3. Prof. James A. Glazier E-mail: glazier@indiana.edu
Institution(s):
1. University of Houston, Department of Biology and Biochemistry, Houston, TX 77204
2. The Texas A&M Institute for Genomic Medicine, Texas A&M University/Texas A&M
Health Science Center, Houston, TX 77030
3. Indiana University, Department of Physics, Bloomington, IN 47405-7003
EPA Project Officer: (leave blank)
Project Period:
Project start: November 1, 2009
Project end: October 31, 2012
Description
Objectives/Hypothesis:
As chemical production increases worldwide, there is increasing evidence as to their
hazardous effects on human health at today's exposure levels, which further implies
that current chemical regulation is insufficient. Thus, a restructuring of the risk
assessment procedure will be required to protect future generations. Given the very
large number of man-made chemicals and the likely complexity of their various and
synergistic modes of action, emerging technologies will be required for the restructuring.
The main objective of the proposed multidisciplinary Texas Indiana Virtual STAR (TIVS)
Center is to contribute to a more reliable chemical risk assessment through the
development of high throughput in vitro and in silico screening models of developmental
toxicity. Specifically, the TIVS Center aims to generate in vitro models of murine
embryonic stem cells and zebrafish for developmental toxicity. The data produced from
these models will be further exploited to produce predictive in silico models for
developmental toxicity on processes that are relevant also for human embryonic
development.
Approach:
The project is divided into three Investigational Areas; zebrafish models, murine
embryonic stem cells models and in silico simulations. The approaches are to:
1. Generate developmental models suitable for high throughput screening.
Zebrafish developmental models (transgenic GFP/EGFP/RFP models of crucial
steps in development) and embryonic stem cell (ESC) differentiation models
(transgenic beta-geo models of crucial steps in differentiation) will be generated.
Important morphology features and signaling pathways during development will
be documented. The impact of environmental pollutants on development and
Previous I TOC
-------
differentiation will be assessed in the models. Finally, the models will be refined
for high throughput screening and automation.
2. Generate a computational model that faithfully recreates the major morphological
features of normal wild-type zebrafish development (ie- segmentation into
somites, proper patterning of vascular and neural systems) and the differentiation
to three primitive layers (endoderm, mesoderm and ectoderm) in mouse
embryonic stem cells. The data for simulations are produced from developed
high information content zebrafish and ESC models. Once a working model of
normal development has been generated, we will carry out a directed series of
parameter sweeps to try to create developmental defects in silico. We will
compare the results of computationally created defects with experimentally-
generated defects in zebrafish and embryonic stem cells. Best matches between
the two datasets will suggest hypotheses about possible mechanisms by which
defects occur.
3. Perform proof-of-concept experiments of the in vitro and in silico test platforms
with a blind test of chemicals.
Techniques will be molecular biology techniques on zebrafish and ESC models, such as
cloning, imaging, in vitro differentiation and in vitro exposure studies, and in silico
mathematical simulations.
Expected Results (Outputs/Outcomes):
In collaboration with other initiatives taken in the field of chemical safety, our generated
results and models will contribute to large screening effort to prioritize chemicals for
further risk assessment. We will specifically contribute with:
• 9 transgenic fish lines validated for toxicity screening
• 16 embryonic stem cell models validated for toxicity screening
• High information content models on development and differentiation to produce
data for in silico simulations, within the project and elsewhere
• Computational models for developmental toxicology of normal development and
of mechanisms by which chemical perturbations cause experimentally-observed
developmental defects
• Information on developmental toxicity on 39 compounds
All the data produced in this project will be released to public databases. The developed
models will be automated for high throughput screening.
Supplemental Keywords:
Risk assessment, effects, dose-response, teratogen, organism, cellular, infants,
chemicals, toxics, aquatic ecosystem protection, pollution prevention, green chemistry,
public policy, environmental chemistry, biology, physics, genetics, mathematics,
modeling, measurement methods.
Previous I TOC
-------
Mechanistic Indicators of Childhood Asthma (MICA): A Systems Biology
Approach for the Integration of Multifactorial Environmental Health Data
Jane Gallagher1, David Reif2, Edward Hudgens 1, Ann Williams1, Mary Johnson1, Ron
Williams3, Haluk Ozkaynak3, Lucas Neas1, Brooke Heidenfelder1, Elaine Cohen Hubal2
and Stephen Edwards1
1 National Health Environmental Effects Research Laboratory, USEPA, RTP, NC
2 National Center for Computational Toxicology, USEPA, RTP, NC
3 National Exposure Research Laboratory, USEPA, RTP NC
Modern methods in molecular biology and advanced computational tools show promise
in elucidating complex interactions that occur between genes and environmental factors
in diseases such as asthma; however appropriately designed studies are critical for
these methods to reach their full potential. We used a case-control study to investigate
whether genomic data (blood gene expression), viewed together with a spectrum of
exposure, effects and susceptibility markers (blood, urine and nail), provide a
mechanistic explanation for the increased susceptibility of asthmatics to ambient air
pollutants. We studied 205 non-asthmatic and asthmatic children, (9-12 years of age)
who participated in a clinical study in Detroit, Michigan. The study combines a traditional
epidemiological design with an integrative approach to investigate the environmental
exposure of children to indoor-outdoor air. The study includes measurements of internal
dose (metals, total and allergen specific IgE, PAH and VOC metabolites) and clinical
measures of health outcome (immunological, cardiovascular and respiratory). A parallel
study using allergen sensitized Brown Norway rats included expression data from both
blood and lung to establish relationships between transcriptional changes in the blood
and corresponding changes in the target tissue. Expected immunological indications of
asthma have been obtained. In addition, initial results from our analyses point to the
complex nature of childhood health and risk factors linked to metabolic syndrome
(obesity, blood pressure and dyslipidemia). For example, 31 % and 34% of the
asthmatic MICA subjects were either overweight (BMI > 25) or hypertensive, (age and
gender adjusted blood pressure values > 90th percentile). This study represents a new
paradigm for epidemiological studies in which traditional health endpoints and
biomarkers are coupled with genetics and high content (Omics) data to expand the use
of mechanistic models for human risk assessment.
This work was reviewed by EPA and approved for publication but does not necessarily
reflect official Agency policy.
Previous I TOC
-------
Linkage of Exposure and Effects Using Genomics, Proteomics, and
Metabolomics in Small Fish Models
Timothy W. Collette , Gerald Ankley (NERL), Dan Villeneuve, et al.2L); Drew
Ekman, David Bencic, Rong-Lin Wang, et al. (NERL); Rory Conolly (NCCT);
Nancy Denslow (University of Florida); Natalia Garcia-Reyero (Jackson State
University); Dalma Martinovic (University of St. Thomas); Ed Perkins (US Army
Corps of Engineers); Karen Watanabe (Oregon Health Sciences University)
Knowledge of possible toxic mechanisms/modes of action (MOA) of chemicals
can provide valuable insights as to appropriate methods for assessing exposure
and effects, thereby reducing uncertainties related to extrapolation across
species, endpoints and chemical structure. However, MOA-based testing
seldom has been used for assessing the ecological risk of chemicals. This is in
part because past regulatory mandates have focused more on adverse effects of
chemicals (reductions in survival, growth or reproduction) than the MOA through
which these effects are caused. A recent departure from this involves endocrine-
disrupting chemicals (EDCs), where there is a regulatory need for USEPA to
understand both MOA and adverse outcomes. To achieve this understanding,
advances in predictive approaches are required whereby mechanistic changes
caused by chemicals at the molecular level can be translated into apical
responses meaningful to ecological risk assessment, such as effects on
development and reproduction, and ultimately population-level impacts.
This is a large, integrated project with collaborators from multiple ORD
laboratories/centers, other Federal agencies, and several universities (originally
through EPA's extramural grants program), that is employing two small fish
models, the fathead minnow (Pimephales promelas) and zebrafish (Danio rerio),
to develop better predictive tools for assessing the ecological risk of EDCs. For
this work, a systems-based approach is being used to delineate toxicity pathways
for 12 model EDCs (muscimol, fipronil, haloperidol, apomorphine, ketoconazole,
trilostane, prochloraz, fadrozole, flutamide, vinclozolin, 17(3-trenbolone and 17a-
ethinylestradiol) with different known or hypothesized toxic MOA. The studies
employ a combination of state-of-the-art genomic (transcriptomic, proteomic,
metabolomic), bioinformatic and modeling approaches, in conjunction with whole
animal testing protocols, to develop response linkages across biological levels of
organization, ranging from molecular alterations to population impacts.
This abstract has been reviewed in accordance with the U. S. Environmental
Protection Agency's peer and administrative review policies and approved for
presentation and publication.
Previous I TOC
-------
Development of a Searchable Metabolite Database and Simulator of Xenobiotic
Metabolism
W. JACK JONES1; Pat Schmieder and Rick Kolanczyk2; Ovanes Mekenyan3
1 National Exposure Research Laboratory, Ecosystems Research Division, USEPA, GA
2 National Health and Environmental Effects Research Laboratory, Mid-Continent
Ecology Division, USEPA, Duluth, MN
3 Laboratory of Mathematical Chemistry, Bourgas University, Bourgas, Bulgaria.
Methods and tools are needed by the EPA's Office of Prevention, Pesticides, and
Toxic Substances (OPPTS) to evaluate and prioritize chemicals for toxicity testing and
hazard assessment, and to enhance the interpretation of registrant data that is
submitted as part of the regulatory process to improve human health and ecological risk
assessments. An often overlooked process, the metabolic activation of chemicals
(production of potentially hazardous transformation products from parent chemicals of
concern), is considered to be an important factor for assessing risk to the environment
and human health. The primary goals of this project are to enhance the ability to
interpret metabolism data via development of a metabolism database (mammalian liver)
that is searchable by text and chemical structure and additionally to develop an in silico
capability for reliably forecasting the metabolism of xenobiotic chemicals of EPA
concern.
Metabolism data, collected from the peer-reviewed literature and from registrant-
submitted data (required for chemical registration/re-registration), has been coded for
risk assessor evaluation/use and for development, training and improvement of a
metabolic simulator. Metabolic pathway information is electronically stored in MetaPath,
a software system allowing sophisticated chemical structure/substructure search
queries to identify commonalities and differences in metabolites among chemicals,
species, dosing regimes, etc. The system depicts metabolic pathways and provides
rapid retrieval of metabolism study information and associated metadata including
metabolite quantities where available. The database will be used by OPPTS scientists
to increase efficiency of metabolism data access and analysis for performance of risk
assessments.
An initial version of a metabolic simulator is under development. The simulator
utilizes a library of more than 340 "functional-group" transformations targeting both in
vitro and in vivo mammalian liver metabolism. Literature-derived, experimentally
determined metabolic maps for diverse chemicals were used for initial simulator
training, with performance of the simulator enhanced by expanding the chemical domain
focus on collection of additional metabolism maps for transformations underrepresented
in the initial training set. Future research will include linking metabolism predictions with
exposure and toxic effects models to enhance prioritization tools for toxicity testing and
chemical assessments for large chemical lists of concern.
The potential impact of this work is significant as it provides much needed tools
to EPA Offices such as OPPTS and the scientific community for evaluating the potential
role of metabolism in enhancing or diminishing toxicity. Linkage of these tools with
exposure and toxic effects models will assist Agency scientists in prioritizing large
chemical lists for further toxicity evaluations, especially for data poor chemicals. These
tools will also allow risk assessors to more systematically and efficiently assess the
hazard of both parent chemicals and their potentially bioactive metabolites. This work
was reviewed by EPA and approved for publication but does not necessarily reflect
official Agency policy.
Previous I TOC
-------
1. Generate developmental models suitable for high throughput screening.
Zebrafish developmental models (transgenic GFP/EGFP/RFP models of crucial
steps in development) and embryonic stem cell (ESC) differentiation models
(transgenic beta-geo models of crucial steps in differentiation) will be generated.
Important morphology features and signaling pathways during development will
be documented. The impact of environmental pollutants on development and
differentiation will be assessed in the models. Finally, the models will be refined
for high throughput screening and automation.
2. Generate a computational model that faithfully recreates the major morphological
features of normal wild-type zebrafish development (ie- segmentation into
somites, proper patterning of vascular and neural systems) and the differentiation
to three primitive layers (endoderm, mesoderm and ectoderm) in mouse
embryonic stem cells. The data for simulations are produced from developed
high information content zebrafish and ESC models. Once a working model of
normal development has been generated, we will carry out a directed series of
parameter sweeps to try to create developmental defects in silico. We will
compare the results of computationally created defects with experimentally-
generated defects in zebrafish and embryonic stem cells. Best matches between
the two datasets will suggest hypotheses about possible mechanisms by which
defects occur.
3. Perform proof-of-concept experiments of the in vitro and in silico test platforms
with a blind test of chemicals.
Techniques will be molecular biology techniques on zebrafish and ESC models, such as
cloning, imaging, in vitro differentiation and in vitro exposure studies, and in silico
mathematical simulations.
Expected Results (Outputs/Outcomes):
In collaboration with other initiatives taken in the field of chemical safety, our generated
results and models will contribute to large screening effort to prioritize chemicals for
further risk assessment. We will specifically contribute with:
• 9 transgenic fish lines validated for toxicity screening
• 16 embryonic stem cell models validated for toxicity screening
• High information content models on development and differentiation to produce
data for in silico simulations, within the project and elsewhere
• Computational models for developmental toxicology of normal development and
of mechanisms by which chemical perturbations cause experimentally-observed
developmental defects
• Information on developmental toxicity on 39 compounds
All the data produced in this project will be released to public databases. The developed
models will be automated for high throughput screening.
Supplemental Keywords:
Risk assessment, effects, dose-response, teratogen, organism, cellular, infants,
chemicals, toxics, aquatic ecosystem protection, pollution prevention, green chemistry,
public policy, environmental chemistry, biology, physics, genetics, mathematics,
modeling, measurement methods.
Previous I TOC
-------
Risk Assessment of the Inflammogenic and Mutagenic Effects of Diesel Exhaust
Particulates: A Systems Biology Approach
James M. Samet, NHEERL, M. Ian Gilmour, NHEERL, William Reed, UNC-Chapel Hill,
David DeMarini, NHEERL, William Linak, NRMRL, Seung Cho, Arcadia Corp., Dongsun
Cao, UNC-Chapel Hill, Hugh Barton, NHEERL/NCCT
Diesel exhaust particulate matter (DEP) is a ubiquitous ambient air contaminant derived
from mobile and stationary diesel fuel combustion. Exposure to DEP is associated with
carcinogenic and immunotoxic effects in humans and experimental animals. At the
cellular level, these health effects are underlain by genotoxic and inflammatory
properties of chemical compounds present in DEP. DEP is composed of elemental,
inorganic and organic compounds that vary widely in composition with the source of the
fuel, engine operating conditions, sampling methods and other parameters. The
genotoxic and inflammatory potencies of DEP also vary with its physicochemical
properties, and these differences along with multiple health effects impede the
development of targeted regulatory strategies for mitigating the impact of DEP exposure
on human health. While traditional reductive toxicology approaches are not likely to
succeed in quantifying relationships between DEP composition and its numerous health
effects, generating a database for modeling the toxicological effects of DEP would
provide a framework for quantitative hazard identification. This project undertook a
systems approach towards the development of a predictive a computational model that
quantitatively describes relationships between the composition of DEP and its genotoxic
and inflammogenic potencies. In phase 1 (Specific Aim 1), 16 distinct DEP were
generated using a combination of fuels, engine types, engine loads and collection
temperatures. These DEP were characterized through extensive chemical and physical
analyses. In phase 2 (Specific Aims 2 and 3), the inflammogenic and genotoxic
potencies of each of the 16 DEP was determined quantitatively. Specific bioassays
measured the expression of the pivotal inflammatory mediator IL-8 in cultured human
lung cells in response to DEP exposure. Signaling mechanisms that regulate the
expression of IL-8/MIP-2 in response to DEP exposure were also examined in order to
provide mechanistic insight and support for the models. The genotoxicity of the 16 DEP
was assayed using bacterial mutagenicity assays. Phase 3 (Specific Aim 4) will utilize
the generated data to construct a series of statistical and mathematical models that
quantitatively relate DEP composition, its inflammogenic and mutagenic effects and the
relevant intracellular signaling mechanisms. Projects funded by this start-up award
produced new findings ranging from the physicochemical properties of DEP to the
molecular mechanisms of toxicity of DEP inhalation. Specifically, these projects
generated data on combustion factors that influence the chemical speciation of DEP,
identified the signal transduction mechanisms activated by DEP exposure of human
lung cells, and ranked and characterized the genotoxicity of DEP of varying
composition. The information provided by these projects has decreased uncertainty in
the risk assessment of DEP exposure and provided biological plausibility in support of
regulatory efforts aimed at mitigating the health effects of DEP inhalation. This work
was reviewed by EPA and approved for publication but does not necessarily reflect
official Agency policy.
Previous I TOC
-------
Development of Microbial Metagenomic Markers for Environmental Monitoring
Jorge W. Santo Domingo, NRMRL/ORD, AWBERC, Cincinnati, OH
Microbiological impairment of water is assessed by monitoring for the presence of
sanitary indicator bacteria. However, conventional methods used to detect bacterial
indicators do not provide any information on the sources impacting water. Recently
developed microbial source tracking (MST) methods have been used to determine the
sources of fecal pollution and pathogens affecting surface waters. While several studies
have reported the successful application of MST methods, none of the methods can
meet performance expectations in complex water systems. One of the basic problems
of MST is the need to rely on the use of culture based methods to assess the primary
sources of pollution. Our research group has developed and evaluated nonculture-
based genomic methods for environmental monitoring and risk assessment based on
fecal metagenome (microbial community genome) and 16S rRNA gene sequences. To
identify potential host-specific markers we developed a novel completive hybridization
approach called genome fragment enrichment (GFE). Using GFE we have identified
dozens of metagenome specific gene fragments which we then used in assay
development. Thus far, we have developed markers to track human, cattle, and poultry
sources of fecal pollution. In addition, using sequencing analyses of 16S rDNA clone
libraries we have identified several novel markers for waterfowl (i.e., gulls and geese).
We have applied the latter assays in samples collected from multiples sites in Lake
Ontario. The results from these studies have confirmed the importance of waterfowl as
primary sources of fecal pollution in the region. The methods developed in this study will
be useful in epidemiological studies and in the evaluation of risk management practices
designed to prevent, reduce, and eliminate pollution of recreational waters and waters
used as sources of drinking water.
This work was reviewed by EPA and approved for publication but does not necessarily
reflect official Agency policy.
Previous I TOC
-------
Develop a Systems Approach to Characterizing and Predicting Thyroid Toxicity
Using an Amphibian Model
Sigmund Degitz, Mike Hornung, Joseph Tietge, National Health and Environmetnal
Effects Research Laboratory, Mid-Continent Ecology Division, Duluth, MN
This research makes use of in vitro and in vivo approaches to understand and
discriminate the compensatory and toxicological responses of the highly regulated HPT
system. Development of an initial systems model will be based on the current
understanding of the HPT axis and the compensatory processes involved in thyroid
hormone homeostasis. Experiments have been conducted to better understand the
relationships of the critical sub-components of the system. Particular emphasis has
been placed on understanding the relative importance of gene expression in the
pituitary, thyroid, and peripheral tissues under normal conditions and following exposure
to chemicals known to interfere with thyroid hormone (TH) synthesis. These molecular
changes are being linked to functional measurements of key hormones and enzymes
that are part of the HPT pathway, all of which are being interpreted in the context of
organismal-level effects.
The primary goal of this work is to develop a sufficient understanding of the HPT so that
predictive models can be developed, testing protocols can be abbreviated, and efforts in
inter-species extrapolation can be improved. One of the most likely uses for a HPT
systems model is to aid in the understanding and discrimination of different modes of
action. As such, this work further enables the development of quantitative structure
activity relationships (QSARs) by providing a basis for sorting chemicals by mode of
action, a necessary step prior to quantifying features of chemical structure associated
with a particular type of toxicity. If these relationships can ultimately be established, then
predictive models can be developed to rank chemicals for future in vivo testing.
This work was reviewed by EPA and approved for publication but does not necessarily
reflect official Agency policy.
Previous I TOC
-------
Session II
Poster
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
Title
Bioactivity Profiling Results from ToxCast Phase I
Assays
ToxMiner™ Software Interface for Visualizing and
Analyzing ToxCast Data
Evaluating the Toxicity Pathways Using High-Throughput
Environmental Chemical Data
Modeling and Predicting Cancer from ToxCast Phase I
Data
Endocrine Profiling and Prioritization Using ToxCast Assay?
Computational Molecular Modeling Methods for
Screening for Chemical Toxicity: The Toxicant-Target
Approach
Development of high throughput methods for chemical
screening and prioritization
The Second Phase of ToxCast and Initial Applications to
Chemical Prioritization
Nuclear Receptor Activity and Liver Cancer Lesion
Progression
A Virtual Liver for Simulating Chemical-Induced Injury
Experimental Models for Quantitative Analysis of
Hepatocarcinogenesis
EPA'S Virtual Embryo: Modeling Developmental Toxicity
Predictive Modeling of Developmental and Reproductive
Toxicity Pathways
Adaptive Responses to Prochloraz Exposure that Alter
Dose-Response and Time-Course Behaviors
Applying Uncertainty Analysis to a Risk Assessment for
the Pesticide Permethrin
Methodology for Uncertainty Analysis of Dynamic
Computational Toxicology Models
Presenters Bio
K Houck
M. Martin
H Mortensen
R Judson
DReif
J Rabinowitz
WMundyetal(NHEERL)
D. Dix
I Shah
J Wambaugh
Chris Gorton (NHEERL)
T Knudsen
Hunter et al
RConolly
R Setzer
J Davis
Previous
TOC
-------
Bioactivity Profiling Results from ToxCast Phase I Assays
Keith Houck, National Center for Computational Toxicology, USEPA, RTP, NC
The ToxCast™ Phase I library of 309 chemicals (320 substances including replicates)
was profiled against 470 in vitro assays to generate data to build initial predictive
models of in vivo toxicity. The in vitro assays included nine different technologies
encompassing cell-free, high-throughput screening assays as well as cell-based assays
in a variety of cell lines, primary human cells as well as primary rat hepatocytes.
Concentration-response data were collected for all active chemical-assay combinations.
Data reproducibility was compared by calculating the concordance of results for the
chemical replicates. Cell-based assays were, in general, more sensitive to chemical
effects than were biochemical assays. Chemical responses were diverse with a wide
range of promiscuity. Many expected interactions were noted in the data, including
endocrine and xenobiotic metabolism enzyme activity. These data are being used to
build correlations to in vivo toxicity endpoints as captured in the relational database
ToxRefDB.
This work was reviewed by EPA and approved for publication but does not necessarily
reflect official Agency policy.
Previous I TOC
-------
ToxMiner™ Software Interface for Visualizing and Analyzing ToxCast Data
Matthew T. Martin2, Ling-Chieh Tsai1, David J. Dix2, Richard S. Judson2, David M. Reif2,
Russell S. Thomas1
1The Hammer Institutes for Health Sciences, 6 Davis Drive, Research Triangle Park, NC
27709
2National Center for Computational Toxicology, USEPA, RTP, NC 27709
The ToxCast™ dataset represents a collection of assays and endpoints that will require
both standard statistical approaches as well as customized data analysis workflows. To
analyze this unique dataset, we have developed an integrated database with Java-
based interface called ToxMiner. The software is organized around both chemical and
assay centric visualization and analysis tools. Currently, these visualization and
analysis tools include standard views of chemical and assay properties and results, the
ability to define subsets of chemicals and assay results based on specified criteria,
correlation matrices across assays, hierarchical clustering, and relative risk calculations.
Machine learning algorithms have been added using Weka. The ToxMiner software
and database will be made freely available. This work was reviewed by EPA and
approved for publication but does not necessarily reflect official Agency policy.
Previous I TOC
-------
Evaluating the Toxicity Pathways Using High-Throughput Environmental
Chemical Data
Holly M. Mortensen*, David Reif, David Dix, Keith Houck, Robert Kavlock,
Richard Judson, National Center for Computational Toxicology, USEPA, RTP,
NC
The application of HTS methods to the characterization of human phenotypic
response to environmental chemicals is a largely unexplored area of
pharmacogenomics. The U.S. Environmental Protection Agency (EPA), through
its ToxCast™ program, is developing predictive toxicity approaches that use in
vitro high-throughput screening (HTS) to profile and model the bioactivity of
environmental chemicals. Current efforts draw from the extensive use of HTS
technologies by pharma and biotech industries for the purposes of drug
discovery, with notable similarities and differences. Output from the first phase of
these experiments has been used to construct target gene lists that have been
linked with publically available information on gene and protein annotation,
molecular, biological, and cellular pathway/processes, as well as gene-disease
association information. These data are integrated, and can be accessed and
queried using the ToxMiner™ database. Currently there is no standard for
analysis of available gene-pathway interaction data, and most studies to date
have focused on a single data source; however, by looking across pathway data
sources we illustrate, using computational network methods, previously
undefined toxicity and toxicity-related pathway coverage in relation to global
pathway space. Finally, we illustrate what pathways are being affected by the
ToxCast™ chemicals screened in Phase I, and the relation of those pathways to
human disease.
This work was reviewed by EPA and approved for publication but does not
necessarily reflect official Agency policy.
Previous I TOC
-------
Modeling and Predicting Cancer from ToxCast Phase I Data
Authors: Richard Judson, David Dix, Robert Kavlock, Imran Shah, Keith Houck,
Thomas Knudsen, David Reif, Matt Martin, National Center for Computational
Toxicology, USEPA, RTP, NC
The ToxCast™ program is generating a diverse collection of in vitro cell free and
cell based HTS data to be used for predictive modeling of in vivo toxicity. We are
using this in vitro data, plus corresponding in vivo data from ToxRefDB, to
develop models for prediction and prioritization. This poster will focus on a set of
machine learning based models that produce toxicity signatures, which are
algorithms that yield a toxicity class prediction based on an association between
in vitro assay data and an in vivo endpoint, derived from training examples. We
demonstrate this approach with a signature for rat liver proliferative lesions using
data from the chemicals with rat chronic/cancer data in ToxRefDB, 61 of which
are positive for this endpoint. We also demonstrate the use of derived gene and
pathway perturbation scores which are more aggregated predictors that can be
used in machine learning approaches. Qualitative uses of these perturbation
scores are demonstrated with relation to other in vivo endpoints in rodents and
humans, including their use in predicting whether a chemical will be a probable
human carcinogen.
This work was reviewed by EPA and approved for publication but does not
necessarily reflect official Agency policy.
Previous I TOC
-------
Endocrine Profiling and Prioritization Using ToxCast Assays
David Reif1, Matthew Martin1, Keith Houck1, Richard Judson1, Thomas Knudsen1,
Shirlee Tan2, Vicki Dellarco3, David Dix1 and Robert Kavlock1
1 National Center for Computational Toxicology, USEPA, RTP, NC
20ffice of Science Coordination and Policy, Washington, DC
30ffice of Pesticide Programs; USEPA, Washington, DC
The U.S. EPA's Endocrine Disrupter Screening Program (EDSP) is charged with
screening pesticide chemicals and environmental contaminants for their potential to
affect the endocrine systems of humans and wildlife (http://www.epa.gov/endo/). The
prioritization of chemicals for testing is a goal shared by both the EDSP and the U.S.
EPA's ToxCast™ program (http://epa.gov/ncct/toxcast/), in which a battery of in vitro,
high-throughput screening assays (467) have assessed a library of 309 environmental
chemicals at a cost <1% of that required for full-scale animal testing. In order to aid the
EDSP, we describe putative endocrine profiles for the entire ToxCast™ library of 309
unique chemicals by focusing on assays involving the estrogen (n=5), androgen (n=4)
and thyroid (n=4) signaling pathways, as well as other nuclear receptors and xenobiotic
metabolizing enzymes (n=70) that have potential relevance to endocrine signaling.
Using these multi-assay profiles in combination with information on relevant chemical
properties, toxicity pathways, and in vivo study results, we present a flexible ranking
system by which chemicals can be prioritized for further screening. By incorporating
multiple sources of information (in vitro assays + chemical descriptors + pathways + in
vivo studies), this prioritization system offers a comprehensive look at a given
chemical's toxicity signature. Importantly, the signatures provide a transparent look at
the relative contribution of all information sources that determine an overall priority
ranking. The results demonstrate that combining multiple data sources into an overall
weight of evidence approach for prioritizing further chemical testing results in more
robust conclusions than any single line of support taken alone. This work was reviewed
by EPA and approved for publication but does not necessarily reflect official Agency
policy.
Previous I TOC
-------
Computational Molecular Modeling Methods for Screening for Chemical Toxicity:
The Toxicant-Target Approach
JR Rabinowitz and SB Little, National Center for Computational Toxicology, USEPA,
RTP, NC
The risk posed to human health and the environment by chemicals that result from
human activity often must be evaluated when relevant elements of the preferred data
set are unavailable. Therefore, strategies are needed that estimate this information and
prioritize the outstanding data requirements. Knowledge of the potential mechanisms
for activity provides a rational basis for the extrapolations inherent in the preliminary
evaluation of risk and the establishment of priorities for obtaining missing data for
environmental chemicals. The differential step in many mechanisms of toxicity may be
generalized as the interaction between a small molecule (a potential toxicant) and one
or more macromolecular targets. An approach based on computation of the interaction
between a potential molecular toxicant and a library of macromolecular targets of
toxicity has been proposed for chemical screening. A library of potential protein targets
for chemical toxicity has been developed from the Protein Data Bank
(www.rcsb.org/pdb). As a test of this approach the interaction between targets
constructed from the rat estrogen receptor and molecules in a data set of chemicals
tested for the capacity to compete with the natural ligand for that receptor have been
computed using molecular "docking" methods. These methods were developed to aid
in the discovery of new Pharmaceuticals (chemicals that bind strongly to the receptor).
In this application they are being tested for their capacity to identify molecules that bind
weakly to the receptor in a data base of primarily inactive chemicals. The data set
being studied (KIERBL in DSSTox) contains 280 chemicals plus 17 -estradiol. Of
these chemicals 14 compete with the natural ligand for the receptor and each binds 3-5
orders of magnitude more weakly than 17 -estradiol. Two different rapid computational
"docking" methods have been applied. Using these without consideration of the
geometry of binding between the toxicant and the target, all of the active molecules
were discovered in the first 16% of the chemicals. When a filter is applied based on the
geometry of a simplified pharmacophore for binding to the ER, the results are improved
and all of the active molecules were discovered in the first 8% of the chemicals. In
order to obtain no false negatives in the model that includes the pharmacophore filter
only 8 molecules of the 280 are false positives. These results indicate that molecular
"docking" algorithms that were designed to find the chemicals that act most strongly at a
receptor can efficiently separate weakly active chemicals from a library of primarily
inactive chemicals. The advantage of using a pharmacophore filter suggests that the
development of filters of this type for other receptors will prove valuable for other
potential targets. This approach may be used in conjunction with other molecular
parameters and bioassay data to address chemical prioritization. The evaluation of the
capability of these methods, or any multi-parameter method for chemical screening,
requires an understanding of the position of an untested chemical in the parameter
space of model. A method for determining this position in the space of relevant
parameters is being developed.
This work was reviewed by EPA and approved for publication but does not necessarily
reflect official Agency policy.
Previous I TOC
-------
Development of high throughput methods for chemical screening and
prioritization
T. Shafer1, W. Mundy1, S. Simmons1, R. Luebke1, K. Crofton1, K. Houck2, D. Dix2
1 National Health Environmental Effects Research Laboratory, USEPA, RTP, NC
2 National Center for Computational Toxicology, USEPA, RTP, NC
To implement the predictive toxicity testing envisioned in the NAS report on Toxicity
Testing in the 21st century, rapid and efficient in vitro screens that are clearly linked to
adverse outcomes are needed. While high throughput screens exist for many different
biological endpoints, there are adverse outcomes for which efficient screens currently
do not exist. These include developmental neurotoxicity, immunotoxicity, and cellular
stress response pathways. The goal of this research program is to develop cell-based
assays linked to adverse outcomes. Two stress response pathway-based assays that
measure the activation of oxidative stress and unfolded protein (heat shock) responses
in human liver cells have been developed to date. Screens for potential developmental
neurotoxicants are being developed by identifying and utilizing high throughput, in vitro
assays for critical developmental processes that are sensitive to perturbation by
toxicants. A new project will develop in vitro screening assays for immunotoxicity, based
on cellular signaling by antigen processing cells and effector cells. This work will focus
on toxicant-induced alteration in cytokine production which in turn modulates immune
function and development of allergic diseases. To date, screening assays for
proliferation, neurite outgrowth and cytotoxicity have been evaluated and utilized to
screen the ToxCast 320 at a single concentration. Compounds (~120) that altered an
endpoint >3 standard deviations from the control mean were further characterized with
complete concentration-response curves. Analysis of these data to determine relative
potency and efficacy of these compounds is ongoing. Additionally, the ToxCast 320
chemicals were screened using the two stress response pathway assays in human liver
cells. These data provided 15-point concentration-response curves for each compound,
which facilitated discrimination between "active" and "inactive" compounds and
generated relative potency (AC50) and efficacy information. These stress assays are
currently being used to screen compounds of interest to OPPT, and have also been
transferred to the NCGC for incorporation into their testing battery. Overall, these results
demonstrate the utility of cell-based HTS assays to screen and prioritize chemicals for
toxicity testing.
This work was reviewed by EPA and approved for publication but does not necessarily
reflect official Agency policy.
Previous I TOC
-------
The Second Phase of ToxCast and Initial Applications to Chemical Prioritization
David Dix1, Keith Houck1, Richard Judson1, Robert Kavlock1, Stephen Little1, Matthew Martin1, Holly
Mortensen1, David Reif1, Ann Richard1, Woodrow Setzer1, Andrew Beam2, Daniel Rotroff3, Maritja
Wolf4
1 National Center for Computational Toxicology, USEPA, RTP, NC
2Dept. of Statistics, North Carolina State University, Raleigh, NC
3 Department of Environmental Sciences and Engineering, University of North Carolina at Chapel Hill,
NC
4Lockheed Martin (Contractor to U.S. EPA), Research Triangle Park, NC
Tens of thousands of chemicals and other contaminants exist in our environment, but only a fraction of
these have been characterized for their potential hazard to humans. ToxCast™ is focused on closing
this data gap and improving the management of chemical risk through a high throughput screening
(HTS) system providing bioactivity profiles in a broad range of pathways relevant to carcinogenic,
mutagenic, reproductive and chronic toxicity. In Phase II of ToxCast™ 700 additional chemicals will
be screened in many of same 467 assays of Phase I, as well as additional HTS assays being added to
the program. These include cell-free assays, cell-based assays in a variety of human and rodent
primary cells and cell lines, and assays in zebrafish and other non-mammalian species. The 700
chemicals being prepared for Phase II of ToxCast™ are a subset of the 10,000 chemical library being
assembled for Tox21 HTS testing by NTP-NCGC-EPA. Approximately 400 Phase II chemicals are
pesticidal actives, inerts or antimicrobials, or industrial chemicals rich in existing animal toxicity data
and thus useful for verifying and expanding predictive toxicity signatures and pathways from Phase I
screening. Approximately 150 Phase II chemicals are pesticidal inerts, antimicrobials, or industrial
chemicals with limited toxicity data and in need of chemical categorization and prioritization. Another
150 Phase II chemicals are failed pharmaceutical compounds with animal and human toxicity data for
direct confirmation of human toxicity pathways and predictors. ToxCast™ HTS assays will provide an
in vitro threshold concentration for specific biological targets, genes, pathways or predictors, which in
combination with high throughput pharmacokinetic modeling can provide an estimated dose at which a
similar in vivo activity is expected. This estimated in vivo equivalent dose can then be used in
combination with chemical structures from DSSTox, and existing in vitro and in vivo toxicity data from
ToxRefDB for predictive modeling and chemical prioritization. Upon successful completion of Phase
II, the ToxCast™ program will be prepared to conduct rapid, quantitative and high-quality hazard
characterizations and subsequent prioritizations on thousands of chemicals. In combination with
ExpoCast™ profiling of exposure potential, these chemical prioritizations can eventually be based
upon both hazard and exposure characterizations, providing ranking of chemicals for entry into
targeted testing specific to carcinogenic, mutagenic, developmental and reproductive, or chronic
toxicity. Decision support software for chemical prioritizations is being provided by ToxMiner™, an
integrated database and interface for ToxCast™ data analysis, visualization and uncertainty
assessment. ToxMiner is user-friendly and will be freely available, facilitating widespread
implementation. ToxCast data will also be available through ACToR, other EPA websites, and
PubChem. ToxCast provides a means to generating meaningful data on the thousands of untested
environmental chemicals, and with associated tools a way to use this data to guide more intelligent,
targeted testing of environmental chemicals in the future. This abstract was reviewed by EPA and
approved for publication, but may not necessarily reflect official Agency policy.
Previous I TOC
-------
Nuclear Receptor Activity and Liver Cancer Lesion Progression
Imran Shah, Keith Houck, Richard S. Judson, Robert J. Kavlock, Matthew T. Martin, John
Wambaugh, David J. Dix , National Center for Computational Toxicology, US EPA, RTP, NC
Nuclear receptors (NRs) are ligand-activated transcription factors that control diverse cellular
processes. Chronic stimulation of some NRs is a non-genotoxic mechanism of rodent liver
cancer with unclear relevance to humans. We explored this question using human CAR, PXR,
PPARa, LXR, ER, AR and Ahr activity assays for 309 environmental chemicals from
ToxCast™, and liver histopathology data from long-term rodent testing studies from
ToxRefDB. The chemicals activated multiple human NRs in combinations that were
informative about rodent liver injury. In addition, some surprising relationships were observed
between the degree of human NR activity and the severity of hepatic lesions progressing to
cancer. The results have implications for nuclear receptor chemical biology and the
extrapolation of in vitro data for predicting liver cancer in humans. In this poster we report on
this analysis, highlighting putative relationships between NRs and cancer lesion progression.
Furthermore, we describe the selection of chemicals and cellular endpoints for modeling NR-
mediated mitogenic, mutagenic and cytotoxic processes involved in hepatocarconogenesis.
This work was reviewed by EPA and approved for publication but does not necessarily reflect
official agency policy.
Previous I TOC
-------
A Virtual Liver for Simulating Chemical-Induced Injury
J Wambaugh, J Jack, and I Shah, National Center for Computational Toxicology,
USEPA, RTP, NC
The US EPA Virtual Liver - vLiver™ -is a tissue simulator that is designed to predict
histopathologic lesions - the gold-standard for toxicity. We have developed an
approach for a biologically motivated model of a canonical liver lobule. The simulated
lobule is composed of discrete representations of hepatic cells that can each determine
their state and fate in response to their local environment, which is determined by a
dynamic graph of the interactions between cells and vascular segments.
We are simultaneously developing two interacting models - one model for the cellular
dynamics that drives how an individual hepatocyte responds to its local environment,
including local concentrations of endogenous and xenobiotic compounds, and a second,
tissue model that determines the local environment of each simulated hepatocyte,
including the impact of whole-organism environmental exposure. The two interacting
models provide a working framework in which research focusing on refining specific
aspects of a single scale, i.e. cellular or tissue, can determine consequences on both
scales.
We are investigating the molecular mechanisms underlying chemically-induced
physiological changes in hepatocytes. Specifically, we are focusing on the roles of a
subset of the nuclear receptor superfamily - the so-called adopted orphan nuclear
receptors. Through in silico models, we hope to elucidate the biochemical processes
governing the important hepatocellular processes associated with the diseases and
disorders of liver toxicity. The chemical induction of nuclear receptors has been linked
to a variety of important cellular processes in the liver, including proliferation, steatosis,
apoptosis, necrosis, and hyperplasia. Nuclear receptor-mediated effects can have
drastic consequences in rodent hepatocytes. Building in silico models will help to reveal
the differences between the human and rodent cell behavior with respect to nuclear
receptor activation/inhibition.
The Virtual Liver cellular dynamics model requires two modules of cellular signaling
networks: one for the effects of chemicals on nuclear receptor activation, crosstalk, and
regulation of gene expression, and a second describing the effects of nuclear receptor-
mediated gene expression on the cell signaling pathways, ultimately predicting changes
in cellular phenotypes. Whereas the first module is an investigation into gene
expression, the second module is the realization of that gene expression within the
context of normal cellular function. The second module provides a causal link between
nuclear receptor-mediated gene expression and cellular changes - including,
proliferation, survival, death, and disease (cancer).
This poster will present preliminary aspects of the of the first cellular dynamics module.
Literature curation, the ToxCast™ data set, and the v-Liver™ Knowledgebase are being
used to establish a nuclear receptor crosstalk and gene expression simulation model.
We are interested in modeling these activities with a threshold networks approach to
Previous I TOC
-------
Boolean networks that, while relatively simple, is capable of capturing the dynamics of
cellular processes with very low computational overhead.
The Virtual Liver tissue model for microdosimetry makes use of a modified
physiologically-based pharmacokinetic (PBPK) approach to determining local
concentrations throughout the simulated lobule. Based upon ordinary differential
equations, this microdosimetry approach bypasses computationally-intensive fluid
dynamics to rapidly determine the impact of environmental exposure to individual
hepatocytes within the simulated lobule. By providing a spatially-extended environment,
the tissue model is intended to allow physiologically based models for inter-cellular
communication and lesion progression as allowed by the cellular dynamics model. We
will also be presenting results of our microdosimetry model as they pertain to lobule
layout, hetergeniety of the hepatocellular environment, and the consequences of oral
vs. inhalation exposure.
The Virtual Liver ultimately provides a framework for making predictions of in vivo
consequences based upon in vitro data. The cellular dynamics model is intended to be
calibrated with in vitro measures of chemical activity, while the tissue model can be
calibrated with histopathology slides and pharmacokinetic data. As a chemically-
perdurable simulation of a homeostatic tissue function, virtual tissues will be powerful
tools for 21st Century Toxicology.
This work was reviewed by EPA and approved for publication but does not necessarily
reflect official Agency policy.
Previous I TOC
-------
Experimental Models for Quantitative Analysis of Hepatocarcinogenesis
Chris Gorton2, John Jack1, John Wambaugh1 and Imran Shah1
1 National Center for Computational Toxicology (NCCT), US EPA, RTP, NC.
National Health and Environmental Effects Lab (NHEERL), US EPA, RTP, NC
Predictive models of chemical-induced liver cancer face the challenge of bridging
causative molecular mechanisms to adverse clinical outcomes. The latent sequence of
intervening events from chemical insult to toxicity are poorly understood because they
span multiple levels of biological organization and timescales. The availability of high-
throughput molecular assays provide a global view of epigenetic, transcriptional and
pathway level changes that can shed much needed light on the regulatory networks
perturbed by xenobiotic stressors. A key challenge in this process is to resolve the role
of these networks in the normal homeostatic response of cells as opposed to
irreversible alterations due to persistent stress. To link molecular mechanisms to
neoplastic lesions will require quantitative assays on molecular changes and altered
cellular phenotypes, as that is the level of biological organization at which tissue
damage becomes manifest. This poster outlines the collaboration between NHEERL
and NCCT on an integrative experimental strategy aimed at developing a model of NR-
mediated hepatocarconogenesis (The US EPA Virtual Liver v-Liver™). As a proof of
concept we are using 20 nuclear receptor (NR) activating chemicals from the EPA
ToxCast™ Program to design short-term in vitro and in vivo studies to generate data on
a range of molecular and cellular endpoints.
This work was reviewed by EPA and approved for publication but does not necessarily
reflect official agency policy.
Previous I TOC
-------
EPA's Virtual Embryo: Modeling Developmental Toxicity
Knudsen TB1, Singh AV2, Rountree MR3, DeWoskin RS4 and Spencer RM2
1 National Center for Computational Toxicology, USEPA, RTP, NC
2 Lockheed Martin, Contractor to the USEPA, RTP, NC
3 National Center for Computational Toxicology, USEPA, RTP, NC
4 National Center for Environmental Assessment, USEPA, RTP, NC
Embryogenesis is regulated by concurrent activities of signaling pathways organized into
networks that control spatial patterning, molecular clocks, morphogenetic rearrangements
and cell differentiation. Quantitative mathematical and computational models are needed
to better understand how genetic errors and biochemical disruptions may perturb these
complex processes, leading to developmental defects. EPA's Virtual Embryo (v-
Embryo™) is an effort to build cell-based computational models using detailed knowledge
of molecular embryology and data from the ToxCast™ high-throughput in vitro screening
effort. The end goal is a library of simulations that can be manipulated in silico and
correlated with in vitro responses or in vivo phenotypes in predictive modeling of
developmental processes and toxicities. The Specific Aims of the project are to: build a
virtual tissue knowledgebase (VT-KB) relevant to development; construct a virtual tissue
simulation engine (VT-SE) for embryonic systems; specify rules for component
interactions of developmental signaling pathways; and analyze abnormal developmental
trajectories that follow perturbations. Software for these purposes includes open-access
programming environments such as CompuCellSD, Python, BioTapestry and GanttPV.
Initial models for proof of principle are focusing on two systems with extensive
experimental embryology and targets for disruption by environmental chemicals: limb-bud
development and optic cup development. The modeling effort can enhance EPA efforts
applying the latest scientific knowledge in quantitative models of dose-response
relationships and uncertainty analysis of developmental and reproductive toxicity. [This
work has been reviewed by EPA and cleared for presentation, but does not reflect official
Agency policy].
Previous I TOC
-------
Predictive Modeling of Developmental and Reproductive Toxicity Pathways
Hunter ES1, Padilla S1, and Knudsen TB2
1 National Health and Environmental Effects Research Laboratory, USEPA, RTP, NC
2 National Center for Computational Toxicology, USEPA, RTP, NC
EPA must evaluate environmental chemicals for potential effects on development and
reproduction. Mechanistic information is essential to understanding how chemicals perturb
development; unfortunately, the mechanisms of prenatal developmental toxicity are not
understood in sufficient depth or detail for risk assessment purposes. This presentation
will explore new approaches for pathway-based prediction of developmental toxicity using
data from high-throughout screening (HTS) assays and from alternative models. These in
vitro platforms include: ToxCast™ HTS biochemical assays (NovaScreen), murine
embryonic stem (ES) cell lines, and zebrafish (ZF) embryos. Anchoring in vivo data was
obtained from Toxicity Reference Database (ToxRefDB) and includes chemical-endpoint
relationships for developmental endpoints in pregnant rat and rabbit studies. Results have
been obtained for an initial pass of the ToxCast_320 chemical library in the ES cell and ZF
embryo assays, using one concentration level (25 |j,M and 80 |j,M, respectively).
Chemicals were ranked by activity in the ES cells (cytotoxicity, myosin heavy chain
immunoreactivity) and ZF larva (malformations, embryo lethality). Data inclusive of all the
diverse platforms and species studied here was obtained for ~215 chemicals. A
preliminary analysis of the single concentration data is being undertaken to correlate
relative developmental activities of chemicals against ES and ZF systems with the
NovaScreen and with ToxRefDB, to identify potential target pathways leading to adverse
developmental and reproductive endpoints. Although many of the important molecular
components of embryogenesis are highly conserved in these species, the developmental
processes and strategies can differ markedly; therefore, initial efforts are focused on data
standardization and calibration. [This work has been reviewed by EPA and cleared for
presentation, but does not reflect official Agency policy].
Previous I TOC
-------
Adaptive Responses to Prochloraz Exposure That Alter Dose-Response and Time-
Course Behaviors
Rory Conolly1, Miyuki Breen2, Dan Villneneuve3, Gary Ankley3
1 National Center for Computational Toxicology, USEPA, RTP, NC
2 North Carolina State University, Graduate Program with the USEPA, RTP, NC
3 National Health Effects and Environmental Research Laboratory, USEPA, RTP, NC
Dose response and time-course (DRTC) are, along with exposure, the major determinants of
health risk. Adaptive changes within exposed organisms in response to environmental stress
are common, and alter DRTC behaviors to minimize the effects caused by stressors. In this
project, we are analyzing how several feedback regulatory loops in fathead minnows
compensate for endocrine stress due to the fungicide Prochloraz. Affected endpoints include
estradiol (E2) levels, ovarian aromatase mRNA, and vitellogenin levels. The data show, for
example, a significant decrease in E2 levels followed by a return to baseline during prolonged
exposure to Prochloraz. Characterization of the mechanisms that underlie these kinds of
adaptive changes will build toward a refined description of DRTC behavior for Prochloraz,
thereby helping us to better understand when exposures pose health risks and when they do
not. In addition, this project will help us to evaluate the possibility that activation of stress
response pathways is itself a useful regulatory endpoint, i.e., the possibility that it is
appropriate to regulate exposures such that stress response pathways are not overwhelmed
and without explicit consideration of downstream, more apical endpoints.
This work was reviewed by EPA and approved for publication but does not necessarily reflect
official agency policy.
Previous I TOC
-------
Applying Uncertainty Analysis to a Risk Assessment for the Pesticide
Permethrin
R. Woodrow Setzer1, Jimena Davis1, Rogelio Tornero-Velez2, Jianping Xue2,
Valerie Zartarian2
1 National Center for Computational Toxicology, 2National Exposure Research
Laboratory, ORD, US EPA, RTP, NC
We discuss the application of methods of uncertainty analysis from our previous
poster to the problem of a risk assessment for exposure to the food-use pesticide
permethrin resulting from residential pesticide crack and crevice application.
Exposures are simulated by the SHEDS (Stochastic Human Exposure and Dose
Simulation) model, which is loosely coupled to a PBPK model for human internal
dose estimation. This presentation discusses approaches for quantifying the
uncertainties at several points in the coupled model: parameter estimation in the
PBPK model; extrapolation to a human model; exposure parameters in SHEDS;
and evaluation of overall uncertainty of the predictions of the coupled model and
application of sensitivity analysis to identify the most important contributors to
that uncertainty. Uncertainties in each component model are characterized as
probability distributions on the parameters of that model. In the case of the
PBPK model, the uncertainty distribution is derived from prior information about
parameter values as well as in vitro data specific to permethrin pharmacokinetics,
and is computed using Bayesian statistical methods. Extrapolating the PBPK
model from rodents to humans involves changing physiological parameters and
extrapolating from rodent to human chemical-specific parameter values.
Uncertainties here are estimated both from limited human data and from
experience in using similar extrapolation methods in other chemicals.
Uncertainties in the SHEDS model are derived from an understanding of the
uncertainty of the component distributions describing pesticide use and
parameters governing pesticide fate and human behavior relating to exposure.
The final output of the coupled model is a probability distribution of exposures
that characterizes the distribution of internal dose for a defined population. We
use Monte-Carlo methods to propagate the uncertainty in each of the
components to make confidence bands around this probability distribution.
Finally, global sensitivity analysis allows us to identify individual components of
uncertainty which contribute most to the overall uncertainty in the coupled
model's predictions. This work was reviewed by EPA and approved for
publication but does not necessarily reflect official Agency policy.
Previous I TOC
-------
Methodology for Uncertainty Analysis of Dynamic Computational Toxicology
Models
Jimena Davis, John Wambaugh, Ramon I. Garcia, R. Woodrow Setzer, National Center
for Computational Toxicology, USEPA, RTP, NC
The task of quantifying the uncertainty in both parameter estimates and model
predictions has become more important with the increased use of dynamic
computational toxicology models by the EPA. Dynamic toxicological models include
physiologically-based pharmacokinetic (PBPK) models, closely-related (and often
coupled) pharmacodynamic models or biologically-based dose-response (BBDR)
models, and models such as those for virtual tissues. Given a set of values for
biological parameters describing the subject and chemical parameters describing the
compound, biological models can make predictions that both allow assessment of the
understanding of how data came to be (interpolation) as well as what might occur under
different conditions (extrapolation). Careful consideration must be given to determining
the value of model parameters and uncertainty about those values as well as the
selection of one model over another given the overall uncertainty about different
models. Quantitative uncertainty analyses are necessary for fully vetting models for
applications such as risk assessments. Along with uncertainty, in these types of
systems variability must also be accounted for on various levels, such as variation in
model parameters across individuals in a population as well as variation in experimental
data. Thus, the analysis of computational toxicology models requires valid statistical
methodologies that are capable of handling both uncertainty and variability accurately.
Several methodological issues are generic to these dynamic toxicological models.
Often, values must be determined for parameters in the absence of hard chemical-
specific data, contributing parameter uncertainty. We discuss approaches for
developing prior distributions quantifying the uncertainty in chemical-specific parameters
based on comparisons of measured values with predicted values from computational or
in vitro methods. Such informative priors are beneficial both in the context of Bayesian
estimation, and for assessing uncertainty of predictions from dynamic models for which
in vivo data are entirely lacking. One application wherein informative priors are
particularly useful that is of interest to the Agency focuses on the development of better
quantitative approaches for cumulative risk assessments of linked exposure-dose-
effects models. A major part of this effort involves formulating PBPK models, which
include well known physiological parameters as well as unknown physicochemical
parameters that describe the uptake and disposition of chemicals or toxins through the
body. Using PBPK models as a motivating example, we discuss some of the
advantages and drawbacks associated with the use of hierarchical Bayesian analysis in
model calibration, uncertainty analysis, and model evaluation. Conventional
computational methods for estimating parameters and evaluating their uncertainty,
which were developed for substantially simpler non-linear models, require lengthy
computations (e.g., weeks or even months) when applied to dynamic models. We
discuss some attempts to standardize this analysis, address the issue of efficient
computational time for deterministic (e.g., PBPK) models, and deal with uncertainty in
stochastic models (e.g., agent-based virtual tissue models). Finally, we discuss
evaluating how well models describe data and approaches to evaluating model
uncertainty. This work was reviewed by EPA and approved for publication but does not
necessarily reflect official Agency policy.
Previous I TOC
-------
CTRP Scientist Biosketches
Maria Bondesson-Bolin
Elaine Cohen Hubal
Timothy W. Collette
Rory Conolly
Jon "Christopher Gorton
Jimena L. Davis
Sigmund J, Degitz Jr.
David J. Dix
Peter Paul Egeghy
Jane E. Gallagher
Keith A. Houck
Edward "Sid" Hunter, III
John Jack
William Jack Jones
Richard Judson
Robert J. Kavlock
Thomas B. Knudsen
Stephen Blair Little
Matthew T. Martin
Holly M. Mortensen
William R. Mundy
James Rabinowitz
David M. Reif
Ann M. Richard
Ivan Rusyn
James M. Samet
Jorge W. Santo Domingo
Deborah Segal
R. Woodrow Setzer
Imran Shah
AmarV. Singh
Cecilia Tan
Raymond R. Tice
John F. Wambaugh
William J. Welsh
Fred A. Wright
Previous I TOC I Next
-------
BIOGRAPHICAL SKETCH
NAME
Maria Bondesson-Bolin
POSITION TITLE
Research Assistant Professor
EDUCATION/TRAINING
INSTITUTION AND LOCATION
Uppsala University, Sweden
Karolinska Institutet, Sweden
Ludwig Institute, Karolinska Institutet, Sweden
Karolinska Institutet, Sweden
DEGREE
(if applicable)
1988
1995
1995-96
1996
YEAR(s)
BS
PhD
Postdoc
Postdoc
FIELD OF STUDY
Microbiology
Cell & Molecular Biology
Oncology & Pathology
Cell & Molecular Biology
A. Positions and Honors.
Positions
1998-02
2002
2003
2003-2009
2009 - Present Research Assistant Professor, University of Houston, Houston, TX
Research Assistant Professor, Department of Cell and Molecular Biology, Karolinska Institutet
Senior Researcher, Department of Cell and Molecular Biology, Karolinska Institutet
Research Secretary, Swedish Research Council/Medicine
Project Manager, CASCADE, Dept. of Biosciences and Nutrition, Karolinska Institutet
Honors
1998
2002
Swedish Natural Science Research Council/Swedish Research Council, Research Assistant
Professorship
Dissertation committee for Maria Lindebro; Mechanisms of regulation of dioxin receptor function
B. Peer-reviewed publications
1. Svensson C., Bondesson M., Nyberg E., Linder S., Jones N. and Akusjarvi G. Independent Transformation
Activity by Adenovirus-5 E1A-Conserved Regions 1 or 2 Mutants. Virology 182:553-561, 1991
2. Linder S., Popowics P., Svensson S., Marshall H., Bondesson M. and Akusjarvi G. Enhanced Invasive Properties
of Rat Embryo Fibroblasts Transformed by Adenovirus E1A Mutants with Deletions in the Carboxy-terminal Exon.
Oncogene 7:439-443, 1992
3. Bondesson M., Svensson C., Linder S. and Akusjarvi G. The Carboxy-terminal Exon of the Adenovirus E1A
protein is Required for E4F-dependent Transcription Activation. The EMBO Journal 11:3347-3354, 1992
4. Bondesson M., Mannervik M., Akusjarvi G. and Svensson C. An Adenovirus E1A Transcriptional Represser
Domain Functions as an Activator when Tethered to a Promoter. Nucleic Acids Research 22:3053-3060, 1994
5. Bondesson M. Transcriptional regulation by the adenovirus E1A proteins. Thesis 1995
6. Bondesson M., Ohman K., Mannervik M., Fan S. and Akusjarvi G. Adenovirus E4 Open Reading Frame 4 Protein
Autoregulates E4 Transcription by Inhibiting E1A Transactivation of the E4 Promoter. Journal of Virology 70:3844-
3851,1996
7. Wahlstrb'm G., Vennstrb'm B. and Bondesson M. The adenovirus E1A oncoprotein is a potent coactivator for
thyroid hormone receptors. Molecular Endocrinology 1999 13:1119-1129
8. Castro D., Arvidsson M., Bondesson M. and Perlmann T. Activity of the Nurrl carboxyl-terminal domain depends
on cell type and integrity of the activation function 2. The Journal of Biological Chemistry 1999 274:37483-37490
9. Ichimura K., Bondesson M., Goike H., Schmidt E., Moshref A. and Collins VP. Deregulation of the p14
ARF/MDM2/p53 pathway is a prerequisite for human astrocytic gliomas with G1/S transition control gene
abnormalities. Cancer Research 2000 60, 417-424
-------
10. Nygard M., Wahlstrom G.M., Tokumoto Y.M., Gustafsson M.V. and Bondesson M. (2003) Hormone-dependent
repression of the E2F-1 gene by thyroid hormone receptors. Molecular Endocrinology 17, 79-92
11. Gustafsson MV, Zheng X, Pereira T, Gradin K, Jin S, Lundkvist L, Ruas JL, Poellinger L, Lendahl U and
Bondesson M (2005). Hypoxia requires notch signaling to maintain the undifferentiated cell state. Developmental
Cell 9, 617-628
12. Nygard M, Becker N, Demeneix B, Pettersson K and Bondesson M (2006). Thyroid hormone-mediated negative
transcriptional regulation of necdin expression. Journal of Molecular Endocrinology 36(3), 517-30
13. Turowska O, Nauman A, Pietrzak M, Poptawski P, Master A, Nygard M, Bondesson M, Tanski Z and
Puzianowska-Kuznicka M (2007) Overexpression of E2F1 in clear cell Renal Cell Carcinoma: a potential impact
of erroneous regulation by thyroid hormone nuclear receptors, Thyroid, 11:1039-48
14. Bondesson M, Jonsson J, Pongratz I, Olea N, Craved! J-P, Zalko D, Hakansson H, Halldin K, Di Lorenzo D, Behl
C, Manthey D, Balaguer P, Demeneix B, Fini JB, Laudet V, Gustafsson J-A, (2009) A CASCADE of Effects of
Bisphenol A, In press Reproductive Toxicology, doi:10.1016/j.reprotox.2009.06.014
Other scientific publications
15. Demeneix B, Gustafsson JA, Bondesson M et al. (2005) Vote REACH for the safer management of chemicals in
EU. Financial Times Nov7
16. Bondesson M and Gustafsson JA (2006) Chemical Contaminants in food: The CASCADE Network of Excellence.
Food Science and Technology 20; 34-36
C. Research Support.
Ongoing Research Support
Agency: Swedish FORMAS Project Period: 11/01/2008 - 12/31/2009
Title: In vitro methods for endocrine disruption
Role on Project: PI Total grant: $143,000
Agency: Swedish Research Council Project Period: 01/01/2007 - 12/31/2009
Title: Effects of environmental pollutants on nerve cell differentiation
Role on Project: PI Total grant: $ 108,000
Agency: EU / I.D.# LSHM-CT-2005-018652 Project Period: 03/01/2006 - 02/31/2011
Title: CRESCENDO "Nuclear receptors during development and aging"
Coordinator: Barbara Demeneix, Ph.D, Professor (France)
Co-P.I.: Jan-Ake Gustafsson, M.D., Ph.D., Professor
Role on Project: Work Package Manager Total grant: $14.000.000 (for 20 research groups)
Agency: EU / I.D.# FOOD-CT-2004-506319 Project Period: 02/01/2004 - 02/28/2010
Title: CASCADE "Chemicals as contaminants in the food chain: An NOE for research, risk assessment and education"
Coordinator and P.I.: Jan-Ake Gustafsson, M.D., Ph.D., Professor
Role on Project: Project Manager (to 03/31/2009) Total grant: $ 18.000.000 (for 25 research groups)
Agency: US-EPA Project Period: 11/01/2009-10/31/2012
Title: Title: The Texas-Indiana Virtual STAR Center; Data-Generating in vitro and in silico Models of Developmental
Toxicity in Embryonic Stem Cells and Zebrafish
P.I.: Jan-Ake Gustafsson M.D., Ph.D., Professor
Role on Project: Project Manager Total grant: $ 3.190 993 (for 3 research groups)
-------
Principal Investigator/Program Director (Last, First, Middle): Cohen Hubal, Elaine
BIOGRAPHICAL SKETCH
Provide the following information for the key personnel and other significant contributors in the order listed on Form Page 2.
Follow this format for each person. DO NOT EXCEED FOUR PAGES.
NAME
Elaine Cohen Hubal
eRA COMMONS USER NAME
POSITION TITLE
Chemical Engineer
EDUCATION/TRAINING (Begin with baccalaureate or other initial professional education, such as nursing, and include postdoctoral training.)
INSTITUTION AND LOCATION
Massachusetts Institute of Technology,
Cambridge, MA
North Carolina State University, Raleigh, NC
North Carolina State University, Raleigh, NC
CUT
DEGREE
(if applicable)
S.B.
M.S.
Ph.D.
Predoc
Fellow
YEAR(s)
1984-1988
1990-1992
1992-1996
1992-1996
FIELD OF STUDY
Chemical Engineer
Chemical Engineer
Chemical Engineer
A. POSITIONS and HONORS
Research and Professional Experience:
March 2004 - present Research Scientist. NCCT, US EPA
May 2004 - March 2004 Research Scientist. One-year detail to ETD. NHEERL, US EPA
May 2003 - Apr 2004 Acting Associate Director for Human Exposure Modeling. HEASD, NERL
Oct 2002 - Apr 2003 Acting Associate Director for Human Exposure Measurements. HEASD, NERL
1997-2004 Chemical Engineer. HEASD, NERL
1996-1997 Research Chemical Engineer. RTI, RTP, NC
1988-1990 Chemical Engineer. Camp Dresser & McKee, Boston, MA
Selected Awards and Honors:
• USEPA 2008 Scientific and Technological Achievement Awards (STAA), for research that improves our
understanding of children's exposure to pesticides in the residential environment;
• USEPA 2007 Children's Environmental Health Excellence Award
• USEPA Bronze Medal for Commendable Service, for developing a risk assessment resource, a Framework
for Assessing Health Risks of Environmental Exposures to Children 2006
• USEPA Gold Medal for Exceptional Service, in recognition of advancing the scientific basis for assessing
and monitoring children's environmental exposures through the development of agency-wide risk
assessment guidance 2006
Invited Lectures/Symposia (selected):
Cohen Hubal, EA. Does exposure science imitate art? Plenary. International Council of Chemical
Associations Long Range Research Initiative (ICCA-LRI) workshop: Connecting Innovations in Biological,
Exposure and Risk Sciences: Better Information for Better Decisions. Charleston, South Carolina. June
2009
Cohen Hubal, EA. Biologically relevant exposure science for toxicity testing. Presented to:
The Strategic Science Team of the ACC Long-Range Research Initiative
Washington, D.C. May 13, 2009
Cohen Hubal, EA, T Pastoor. Improving Exposure Science and Dose Metrics for Toxicity Testing, Screening,
Prioritizing, and Risk Assessment. ILSI Health and Environmental Sciences Institute (HESI) Annual
Emerging Issues Forum, Tucson, AZ, Jan 20, 2009.
Previous
TOC
-------
Principal Investigator/Program Director (Last, First, Middle): Cohen Hubal, Elaine
Cohen Hubal, E, D Reif, S Edwards, L Neas, E Hudgens, J Gallagher. Mechanistic Indicators of Childhood
Asthma (MICA) Study. 11th SAC Seminar: New Trends in Chemical Toxicology. Moscow, Russian
Federation. September 22-25, 2008.
Cohen Hubal, EA. Considering Susceptibility: Translation for Risk Management. ICCA-LRI workshop:
Twenty-first century approaches to toxicity testing, biomonitoring, and risk assessment. Amsterdam, The
Netherlands. June 16-17, 2008.
Cohen Hubal, EA. Computational Toxicology. Workshop on Toxicogenomics in Risk Assessment. Toxicology
And Risk Assessment Conference. Cincinnati, OH April 14-15, 2008.
Cohen Hubal, EA. Invited panelist. Bridging Human and Ecological Exposure Sciences: A Window of
Opportunity. ISEA Annual Meeting, Durham, NC. October 14-18, 2007.
Cohen Hubal, EA. Exposure Assessment. Fundamentals of Human Health Risk Assessment with a Case
Study Approach. Continuing Education Course AMOS. Society of Toxicology Continuing Education
Course, Charlotte, North Carolina, March 25, 2007.
Cohen Hubal, EA. Application of Nanotechnology-Enabled Sensor Technologies for Monitoring Human
Exposure to Environmental Contaminants. EPA/ORCAS Nanotechnology Applications in Environmental
Health Workshop. RTP, NC, April 20, 2006.
Cohen Hubal, EA. Application of Micro- and Nanoscale Sensor Technologies for Monitoring Human Exposure
to Environmental Contaminants. ILSI HESI Emerging Issues Forum. Puerto Rico, January 17, 2006.
Cohen Hubal, EA. Framework to Use Biomonitoring Data to Inform Exposure Assessment in Children.
WHO/I PCS Workshop on Advances in the Use of Biomarkers in Children, Buenos Aires, Argentina,
November 17-18, 2005.
Leadership Provided to Scientific Community (selected):
• Chair Exposure Science for Screening Prioritizing and Toxicity Testing Community of Practice (ExpoCoP).
June 2008-present.
• Editorial Board Journal of Exposure Science and Environmental Epidemiology. January 2007 - present.
• Co-chair International Society of Exposure Science (formerly ISEA) 2009 Annual Meeting: Transforming
Exposure Science for the 21st Century, Nov 1-6, Minneapolis, MN.
• Co-chair International Council of Chemical Associations Long Range Research Initiative (ICCA-LRI)
workshop: Connecting Innovations in Biological, Exposure and Risk Sciences: Better Information for Better
Decisions. Charleston, South Carolina. June 2009
• Member National Children's Study Data Access Committee. 2008- present
• World Health Organization Temporary Adviser to plan the I PCS international workshop on "Identifying
Important Life Stages for Monitoring and Assessing Risks from Exposures to Environmental
Contaminants." 2009-present
• Program planning committee for the International Society of Exposure Analysis 2007 Annual Meeting.
Chair symposium: Computational Toxicology. Durham, NC. October 14-18, 2007.
• Member ILSI Health and Environmental Sciences Institute, Sensitive Subpopulations Working Group,
2006-present.
• Member ILSI Health and Environmental Sciences Institute, Biomonitoring Working Group 2004-present
• Co-organized /co-chaired with Richard Judsen, Session title "Genetic Variation, Gene-Environment
Interactions and Environmental Risk Assessment" for International Science Forum on Computational
Toxicology, May 21-23, 2007, RTP, NC.
• Cohen Hubal, EA. Exposure Science for Computational Toxicology. US EPA NCCT Course on
Computational Toxicology. Research Triangle Park, NC. March 4, 2008.
• NCCT Cosponsor with NCER. Program planning committee for US EPA Workshop on Research Needs for
Community-Based Risk Assessment. Session organizer/chair: Data needs and measurement methods for
CBRA. Research Triangle Park, NC. October 18-19, 2007.
• Member US EPA Risk Assessment Forum. June 2004 - 2009.
Previous I TOC
-------
Principal Investigator/Program Director (Last, First, Middle): Cohen Hubal, Elaine
B. SELECTED PUBLICATIONS
Cohen Hubal, EA. Biologically-Relevant Exposure Science for 21st Century Toxicity Testing
Toxicol. Sci., Advance Access published on July 14, 2009; doi: doi:10.1093/toxsci/kfp159
Sheldon, LS, and EA Cohen Hubal. Exposure as Part of a Systems Approach for Assessing Risk. Environ
Health Perspect doi:10.1289/ehp.0800407 [Online 8 April 2009]
Cohen Hubal EA, Richard AM, Imran S, Gallagher J, KavlockR, Blancato J, Edwards S. (2009) Exposure
Science and the US EPA National Center for Computational Toxicology. Journal Expo Sci Environ
Epidemiol Available online: Nov 5 2008 [Epub ahead of print]
Xu, Y., EA Cohen Hubal, PA Clausen, JC Little. Predicting Residential Exposure to Phthalate Plasticizer
Emitted from Vinyl Flooring-A Mechanistic Analysis. Environ. Sci. Technol., DOI: 10.1021/es801354f
available online February 19, 2009.
Heidenfelder, BL, DM Reif, EA Cohen Hubal, EE Hudgens, LA Bramble, JG Wagner, JR Harkema, M
Morishita, GJ Keeler, SW Edwards, JE Gallagher. (2009) Comparative microarray analysis and pulmonary
morphometric changes in brown Norway rats exposed to ovalbumin and/or concentrated air particulates.
Toxicol Sci. 108(1), 207-221.
Sanchez, Y, K Deener, E Cohen Hubal, C Knowlton, D Reif, D Segal. Research needs for community based
risk assessment. J Expo Sci Environ Epidem. 2009 Feb 25 [Epub ahead of print]
Cohen Hubal, EA, MG Nishioka, WA Ivancic, M Morara, P Egeghy. (2008) Comparing surface residue transfer
efficiencies to hands using polar and non-polar fluorescent tracers. Environmental Science & Technology
42 (3), 934-939.
Kavlock, RJ, G Ankley, J Blancato, M Breen, R Conolly, D Dix, K Houck, E Cohen Hubal, R Judson, J
Rabinowitz, A Richard, RWSetzer, I Shah, D Villeneuve, and E Weber. (2008) Computational toxicology:
A state of the science mini review. Toxicological Sciences 103(1), 14-27.
Cohen Hubal, EA, J Moya, SG Selevan. (2008) A lifestage approach to assessing children's exposure.
Developmental and Reproductive Toxicology. Birth Defects Res (Part B) 83:522-529.
Gallagher J., Hudgens E., Heidenfelder B.N., Reif D.M., Neas L, Wlliams A., Harkema J. , Hester S., Edwards
S.E., Cohen Hubal EA. Mechanistic indicators of children's asthma study (MICA): A systems biology
apporoach for the integration of multifactorial environmental health data. Submitted.
Xu, Y., EA Cohen Hubal, PA Clausen, JC Little. Predicting Residential Exposure to Phthalate Plasticizer
Emitted from Vinyl Flooring - Sensitivity, Uncertainty, and Implications for Biomonitoring. Submitted.
Gallagher, JE, EA Cohen Hubal, SW Edwards. Invited Chapter "Biomarkers of Environmental Exposure"
"Biomarkers of toxicity: A New Era in Medicine Editors Vishal S. Vaidya and Joseph V. Bonventre
PublisherJohn Wley and Sons, Inc. Submitted.
RN Hines, D Sargent, H Autrup, LS Birnbaum, RL Brent, NG Doerrer, EA Cohen Hubal, DR Juberg, C Laurent,
R Luebke, K Olejniczak, CJ Portier, WSIikker. Approaches for Assessing Risks to Sensitive Populations:
Lessons Learned from Evaluating Risks in the Pediatric Population. Submitted.
Reif, DM, JE. Gallagher, BL Heidenfelder, EE Hudgens, W Jones, CL Wlliams-DeVane, LM Neas, EA Cohen
Hubal, SW Edwards. Elucidating Asthma Phenotypes via Integrated Analysis of Blood Gene Expression
Data with Demographic and Clinical Information. In Preparation.
Reif, DM, CL Wlliams-DeVane, EA Cohen Hubal, W Jones, EE Hudgens, BL Heidenfelder, LM Neas, JE.
Gallagher, SW Edwards._Systems Modeling of Gene Expression, Demographic and Clinical Data to
Determine Disease Endotypes In preparation.
Firestone M, J Moya, E Cohen Hubal, V Zartarian. (2007) Identifying childhood age groups for exposure assessments
monitoring. Risk Analysis 27(3): 701-714.
Ryan, PB, TA Burke, EA Cohen Hubal, JJ Cura, TE McKone. (2007) Using Biomarkers to inform cumulative
risk assessment. Environ Health Perspect 115:833-84
deFur, PL, GW Evans, EA Cohen Hubal, AD Kyle, RA Morello-Frosch, D Williams. (2007) Vulnerability as a
function of individual and group resources in cumulative risk assessment. Environ Health Perspect
115:817-824.
Cohen Hubal, EA, P Egeghy, K Leovic, G Akland. (2006) Measuring potential dermal transfer of a pesticide to
children in a daycare center. Environ Health Perspect 114(2)264-269.
Previous I TOC
-------
Principal Investigator/Program Director (Last, First, Middle): Cohen Hubal, Elaine
Cohen Hubal, EA. (2006) Uso de los Datos de Biomonitoreo para Informar sobre la Evaluation de la
Exposition Infantil [Using Biomonitoring Data to Inform Exposure Assessment in Children] Acta
Toxicologica Argentina [Journal of the Argentinan Society of Toxicology]. 14(suplemento)17-19.
Barone S Jr, RC Brown, S Euling, E Cohen Hubal, CA Kimmel, S Makris, J Moya, SG Selevan, B Sonawane, T
Thomas, C Thompson. (2006) Vision General de al Evaluation del Riesgo en Salud Infantil Empleando un
Enfouque por Etapas de Desarrollo [Overview of a Life Stage Approach to Children's Health Risk
Assessment] Acta Toxicologica Argentina. 14(suplemento)7-10.
Birnbaum, LS, EA Cohen Hubal. (2006) Polybrominated diphenyl ethers: a case study for application of
biomonitoring data to characterize exposure. Environ Health Perspect 114:1770-1775.
Previous
TOC
-------
Principal Investigator/Program Director (Last, First, Middle): Collette, Timothy W.
BIOGRAPHICAL SKETCH
Provide the following information for the key personnel and other significant contributors in the order listed on Form Page 2.
Follow this format for each person. DO NOT EXCEED FOUR PAGES.
NAME
Timothy W. Collette
eRA COMMONS USER NAME
POSITION TITLE
Research Chemist
EDUCATION/TRAINING (Begin with baccalaureate or other initial professional education, such as nursing, and include postdoctoral training.)
INSTITUTION AND LOCATION
Berry College, Rome, GA
University of Georgia, Athens, GA
DEGREE
(if applicable)
B.S.
Ph.D.
YEAR(s)
1981
1985
FIELD OF STUDY
Chemistry
Physical Chemisrty
A. POSITIONS and HONORS
Research and Professional Experience:
1985 - Present Research Chemist, Processes and Modeling Branch, Ecosystems Research Division,
NERL, U.S. EPA, Athens, GA
1981 - 1985 Teaching Assistant and Research Assistant, Department of Chemistry, University of
Georgia, Athens, GA.
Selected Awards and Honors : (from 1998 - 2009)
1999 STAA - Methods and Monitoring Category
2000 STAA - Methods and Monitoring Category
2000 Bronze Medal - Application of Raman Spectroscopy
2000 Athens Federal Executives Association Public Service Recognition Award
2001 Sigma Xi Outstanding Research Paper Award - University of Georgia Chapter
2002 STAA - Methods and Monitoring Category
2002 STAA - Review Article Category
2003 STAA - Methods and Monitoring Category
2004 Office of Pollution Prevention and Toxics Mission Award - PFOA Workgroup
2004 Bronze Medal - Computational Toxicology Design Team
2004 Bronze Medal - Promoting Strong Science in Agency Decisions
2004 Commendation from the Office of Acquisition Management
2004 STAA - Methods and Monitoring Category
2006 Gold Medal - Perchlorate Risk Characterization Team
2007 STAA - Fate and Transport Category
2008 STAA - Methods and Monitoring Category
Invited Lectures/Symposia (selected): (selected from about 35 during 1998 - 2009)
"The Value of GC-IR for Environmental Contaminant Identification, T.W. Collette, 49th Pittsburgh Conference,
New Orleans, LA., March (1998).
"Optimization of Modern Dispersive Raman Spectrometers for Molecular Speciation of Organics in Water,"
T.W. Collette and T.L. Wlliams, 26th Annual Meeting of the Federation of Analytical Chemistry and
Spectroscopy Societies, Vancouver CANADA, October (1999).
"Perchlorate in Fertilizers?: Analysis by Raman Spectroscopy" T.W. Collette and T.L. Wlliams, 220th National
Meeting of The American Chemical Society, Washington, DC., August (2000).
"Speciation of Complex Organic Contaminants in Water with Raman Spectroscopy" T.W. Collette and T.L.
Wlliams, 30th International Symposium on Environmental Analytical Chemistry, Espoo, FINLAND, June
(2000).
Previous
TOC
-------
Principal Investigator/Program Director (Last, First, Middle): Collette, Timothy W.
"Determination of Perchlorate in Some Fertilizers And Plant Tissue by Raman Spectroscopy" T.W. Collette and
T.L. Williams, 222th National Meeting of The American Chemical Society, Chicago IL, August (2001).
"The Role of Raman Spectroscopy in the Analytical Chemistry of Potable Water" T.W. Collette and T.L.
Williams, 222th National Meeting of The American Chemical Society, Chicago IL, August (2001).
"Raman Analysis of Fertilizer and Plant Tissue Extracts for Perchlorate Contamination" T.W. Collette and T.L.
Williams, Eastern Analytical Symposium, Atlantic City NJ, October (2001).
"Principals of Infrared Spectroscopy and Spectral Interpretation" T.W. Collette, Joint Oil Analysis Program
Technology Showcase, U.S. Air Force, Pensacola, FL, April (2004).
"The Athens Lab's Role in EPA's Computational Toxicology Program", T.W. Collette. University of Georgia,
Interdisciplinary Toxicology Program Retreat, Athens, GA, February (2006).
"Metabolomics in Small Fish Toxicology and Ecological Risk Assessments," T. Collette, D. Ekman, Q. Teng, D.
Villeneuve, and G. Ankley, 3rd International Conference of the Metabolomics Society, University of
Manchester, UNITED KINGDOM, June (2007).
"Metabolomics in Small Fish Toxicology and Other Environmental Applications", T. Collette, Fort Johnson
Marine Science Seminar Series, Charleston, South Carolina, October (2007).
"Assessing Exposures to Regulated Chemicals using Metabolomics with Multiple Analytical Techniques", T.
Collette, D. Ekman, W. Garrison, M. Henderson, and Q. Teng, 2007 Eastern Analytical Symposium,
Somerset, New Jersey, November (2007).
"Fish Toxicogenomics: Moving into Monitoring and Regulation", NERC International Opportunity Workshop,
Pacific Environmental Science Centre, North Vancouver, CANADA, April (2008).
Assistance/Leadership Provided to the Scientific Community:
Society of Applied Spectroscopy, Chair, National Tellers Committee: 2006 - present
Advisory Board: Comprehensive Analytical Chemistry: 1998 - 2007
Society of the Sigma Xi, University of Georgia Chapter, Admission Committee: 2002 - 2005
Editorial Advisory Board: Vibrational Spectroscopy: 1999-2005
Coordinating Committee: International Symposium on Environ. Anal. Chem.: 1996 - 2001
Special Issue Editor: Vibrational Spectroscopy: September 2000
Program Committee: SPIE Symposium on Environmental and Industrial Sensing: 1999
Exhibit Chairman: 11th International Conference on Fourier Transform Spectroscopy: 1997
Program Committee: International Symposium on Environ. Anal. Chem.: 1997, 1999, 2000
Associate Editor: Vibrational Spectroscopy: 1995 -1999
Assistance/Leadership Provided to the Agency:
EPA Cross-ORD Post-doc Recruitment Workgroup: 2005
EPA/ORD Computational Toxicology Implementation and Steering Committee: 2004 - present
EPA/OPPT Telomer Degradation Workgroup 2004 - present
EPA/OPPT PFOA Monitoring Workgroup: 2003 - present
EPA/ORD Safe Pesticides, Safe Products Long Term Goal 3 Workgroup: 2003 - present
EPA/ORD Computational Toxicology Research Initiative Design Team: 2002-2003
B. SELECTED PUBLICATIONS (selected from 1998 - 2008, total more than 60)
"Identification of New Ozone Disinfection Byproducts in Drinking Water," S.D. Richardson, A.D. Thruston,
Jr., T.V. Caughran, P.M. Chen, T.W. Collette, T.L. Floyd, K.M. Schenck, B.W. Lykins, Jr., G. Sun, and G.
Majetich, Environ. Sci. Technol. 33, 3368-3377 (1999).
"Perchlorate Identification in Fertilizers," S. Susarla, T.W. Collette, A.W. Garrison, N.L. Wolfe, and S.C.
McCutcheon, Environ. Sci. Technol. 33, 3469-3472 (1999).
"Identification of New Drinking Water Disinfection Byproducts Formed in the Presence of Bromide," S.D.
Richardson, A.D. Thruston, Jr., T.V. Caughran, P.M. Chen, T.W. Collette, T.L. Floyd, K.M. Schenck,
B.W. Lykins, Jr., G. Sun, and G. Majetich, Environ. Sci. Technol. 33, 3378-3383 (1999).
"Optimization of Raman Spectroscopy for Speciation of Organics in Water," T.W. Collette, T.L. Williams, and
J.C. D'Angelo, Appl. Spectres. 55, 750-766 (2001).
Previous I TOC
-------
Principal Investigator/Program Director (Last, First, Middle): Collette, Timothy W.
"Raman Spectroscopic Analysis of Fertilizers and Plant Tissue for Perchlorate" T.L Williams, R.B. Martin, and
T.W. Collette, Appl. Spectros. 55, 967-988 (2001).
"Analysis of Hydroponic Fertilizer Matrixes for Perchlorate: Comparison of Analytical Techniques " T.W.
Collette, T.L Williams, E.T. Urbansky, M.L Magnuson, G.N. Hebert, and S.H. Strauss, Analyst, 128, 88-
97 (2003).
"Degradation of Chlorpyrifos in Aqueous Chlorine Solutions: Pathways, Kinetics, and Modeling", S.E. Duirk and
T.W. Collette, Environ. Sci. Techno!., 40, 546 - 551 (2006).
"Monitoring the Speciation of Aqueous Free Chlorine from pH 1-12 with Raman Spectroscopy to Determine the
Identity of the Potent Low-pH Oxidant", D.P. Cherney, S.E. Duirk, J.C. Tarr, and T.W. Collette, Appl.
Spectros., 60, 764 - 772 (2006).
"Raman Spectroscopy-Based Metabolomics For Differentiating Exposures to Triazole Fungicides Using Rat
Urine," D.P. Cherney, D.R. Ekman, D.J. Dix, and T.W. Collette, Anal. Chem. 79, 7324-7332 (2007).
"NMR Analysis of Male Fathead Minnow Urinary Metabolites: A Potential Approach for Studying Impacts of
Chemical Exposures", D.R. Ekman, Q. Teng, K.M. Jensen, D. Martinovic, D.L. Villeneuve, G.T. Ankley,
and T.W. Collette, Aquatic Tox. 85, 104 - 112 (2007).
"Chlorpyrifos Transformation by Aqueous Chlorine in the Presence of Bromide and Natural Organic Matter",
S.E. Duirk, J.C. Tarr, and T.W. Collette, J. Agric. Food. Chem. 56, 1328 - 1335 (2008).
"Investigating Compensation and Recovery of Fathead Minnow (Pimephales Promelas) Exposed to 17 -
Ethynylestradiol with Metabolite Profiling", D.R. Ekman, Q., Teng, D.L. Villeneuve, M. D. Kahl, K.M.
Jensen, E.J. Durhan, G.T. Ankley, and T. W. Collette, Environ. Sci. Technol. 42, 4188-4195 (2008).
"International NMR-based Environmental Metabolomics Intercomparison Exercise", M.R. Viant, D.W. Bearden,
J.G. Bundy, I.W. Burton, T.W. Collette, D.R. Ekman, V. Ezernieks, T.K. Karakach, C.Y. Lin, S. Rochfort,
J.S. de Ropp, Q. Teng, R.S. Tjeerdema, J.A. Walter, and H. Wu, Environ. Sci. Technol. 43, 219-225
(2009).
"Spectral Relative Standard Deviation: A Practical Benchmark in Metabolomics", H.M. Parsons, D.R. Ekman,
T.W. Collette, and M.R. Viant, Analyst. 134, 478-485 (2009).
"A Direct Cell Quenching Method for Cell-Culture Based Metabolomics", Q. Teng, W. Huang, T.W. Collette,
D.R. Ekman, and C. Tan, Metabolomics. 5, 199-208 (2009).
"Profiling Lipid Metabolites Yields Unique Information on Sex- and Time-dependent Responses of Fathead
Minnows (Pimephales promelas) Exposed to 17a-Ethynylestradiol", D.R. Ekman, Q., Teng, D.L.
Villeneuve, M. D. Kahl, K.M. Jensen, E.J. Durhan, G.T. Ankley, and T. W. Collette, Metabolomics. 5, 22
- 32 (2009).
"A Computational Model of the Hypothalamic-Pituitary-Gonadal Axis in Male Fathead Minnows Exposed to
17 -Ethinylestradiol and 17(3-Estradiol", K.H. Watanabe, Z. Li, K. Kroll, D.L. Villeneuve, N. Garcia-
Reyero, E.F. Orlando, M.S., Sepulveda, T.W. Collette, D.R. Ekman, G.T. Ankley and N.D. Denslow,
Toxicol. Sci., 109, 180-192 (2009).
"Endocrine-Disrupting Chemicals in Fish: Developing Exposure Indicators and Predictive Models of Effects
based on Mechanism of Action", G.T. Ankley, D.C. Bencic, M.S. Breen, T.W. Collette, et al. Aquatic Tox.,
92, 168-178(2009).
"Integrating Omic Technologies Into Aquatic Ecological Risk Assessment And Environmental Monitoring:
Hurdles, Achievements And Future Outlook", G. Van Aggelen, G. Ankley, W. Baldwin, D. Bearden, W.
Benson, J. Chipman, T. Collette, et al., Accepted - Environ. Health Perspect. (2009).
Previous I TOC
-------
Principal Investigator/Program Director (Last, First, Middle): Conolly, Rory
BIOGRAPHICAL SKETCH
NAME
Rory Conolly
eRA COMMONS USER NAME
Bongoeight
POSITION TITLE
Senior Research Biologist
EDUCATION/TRAINING (Begin with baccalaureate or other initial professional education, such as nursing, and include postdoctoral training.)
INSTITUTION AND LOCATION
Harvard College, Cambridge, MA
Harvard School of Public Health
Imperial Chemical Industries, Cheshire, England
DEGREE
(if applicable)
A.B.
Sc.D.
YEAR(s)
1972
1978
1978- 1979
FIELD OF STUDY
Biology
Physiology/Toxicology
Biochemical Toxicology
A. POSITIONS and HONORS
Research and Professional Experience:
1979-1986
1986-1988
1988-1989
1989-1995
1995-2004
2001-2004
2004-2005
2004-2005
2005-present
Assistant Professor of Toxicology, The University of Michigan
Research Manager, NSI Tech. Services Corp., Dayton, OH
Deputy Director, NSI Technology Services Corp., Dayton, OH
Scientist, Chemical Industry Institute of Toxicology, RTP, NC
Senior Scientist, CUT Centers for Health Research, RTP, NC
Director, Center for Computational Biology & Extrapolation Modeling, CUT Centers for
Health Research, RTP, NC
Director, Center for Computational Systems Biology & Human Health Assessment, CUT
Centers for Health Research, RTP, NC
Senior Investigator, Computational Systems Biology & Human Health Assessment, CUT
Centers for Health Research, RTP, NC
Senior Research Biologist, National Center for Computational Toxicology, ORD, U.S. EPA,
RTP, NC
Professional Societies and Affiliations:
1981 -present
1985-present
1985-present
1997-1998
1998-2005
1998-present
2001 -2001
2002 - present
2004 - 2005
2005 - present
2009 - present
Member, Society of Toxicology
Member, Society for Risk Analysis
Member, American Association for the Advancement of Science
President, Risk Assessment Specialty Section, Society of Toxicology
Member, U.S. EPA FIFRA Science Review Board
Adjunct Professor of Biomathematics, North Biomathematics, North Carolina State
University, Raleigh, NC
President, Biological Modeling Specialty Section, Society of Toxicology
Faculty Affiliate, Department of Environmental and Radiological Health Sciences, Colorado
State University, Fort Collins, CO
Member, NAS Board on Environmental Studies and Toxicology
Adjunct Professor of Environmental Science, Wright State University, Dayton, OH
Councilor, Risk Assessment Specialty Section, Society of Toxicology
Honors and Awards:
1991 Outstanding Presentation in Risk Assessment, Annual Meeting of the Society of Toxicology
1999 Outstanding Presentation in Risk Assessment, Annual Meeting of the Society of Toxicology
2003 Outstanding Presentation in Risk Assessment, Annual Meeting of the Society of Toxicology (2 awards)
2004 Best Published Paper in Risk Assessment, Risk Assessment Specialty Section, Society of Toxicology
2005 Arnold J. Lehman Award for career achievement in risk assessment, Society of Toxicology
2009 EPA Bronze medal, Perchlorate team, For exceptional assistance to the Office of Water on an
important, highly visible, and scientifically complex health assessment of perchlorate.
PHS 398/2590 (Rev. 09/04, Reissued 4/2006)
Page 1
Previous
TOC
-------
Principal Investigator/Program Director (Last, First, Middle): Conolly, Rory
Selected Invitations at National & International Symposia (last 3 years):
"Computational Modeling to Evaluate Candidate Modes of Action for Arsenic", presented at the Workshop
session "How Can Biologically-Based Modeling of Arsenic Kinetics and Dynamics Inform the Risk
Assessment Process?" 46th Annual Meeting, Society of Toxicology, Charlotte, NC, USA, March 27, 2007.
"Data needs and parameter estimation for the 2-stage clonal growth (CG) model", presented at the
symposium, Wither Clonal Growth Modeling, Society for Risk Analysis Annual Meeting, Henry B. Gonzalez
Convention Center, San Antonio, TX, December 12, 2007.
"Integration of Pharmacokinetic (PK) and Pharmacodynamic (PD) Modeling of Arsenic to Inform the Risk
Assessment Process", teleconference presentation to the Socitey of Toxicology Risk Assessment Specilaty
Section, March 12, 2008 (with Elaina Kenyon and Hisham el-Masri).
"Use of Mode of Action (MoA) Information in Biologically Based Modeling for Arsenic:
Dose-Response (DR) and Time-Course (TC) Considerations", presented at the symposium, Incorporation
of Mode of Action into Mechanistically-Based Quantitative Models, 47th Annual Meeting, Society of
Toxicology, Seattle WA, March 20, 2008.
"Computational Modeling in Concert with Laboratory Studies: Application to B cell", presented at the
symposium, Dioxin Toxicity: Mechanisms, Models & Potential Health Risks, Michagan State Univeristy,
Superfund Program Conference, MSU Kellogg Center Lincoln Room, East Lansing, Ml, October 20-21,
2008.
"Biologically based dose-response modeling. The potential for accurate description of the linkages in the
applied dose-tissue dose-health effect continuum", presented at the Workshop, From Genes to Organs:
Advancements in Modeling Biological Systems, Society of Toxicology 48th Annual Meeting, Baltimore, MD,
March 15-19,2009.
"Studying the Basic Biology of B Cell Differentiation to Understand the Effects of 2, 3, 7, 8-tetrachlorodibenzo-
p-dioxin (TCDD) on Immune Function", Superfund Basic research Program Webinar, presented as part of
the Spring/Summer 2009 edition of Risk eLearning "Computational Toxicology: New Approaches for the
21st Century." Co-presented with Norbert E. Kaminski. June 24, 2009.
Selected Expert Committees/Advisory Panels/Organizing Activities
"Computational Systems Biology: The Integration of Data Across Multiple Levels of Biological Organization to
Understand How Perturbations of Normal Biology Become Adverse Health Effects", NRC Committee on
Models in the Regulatory Process, Workshop on Emerging Issues for Regulatory Environmental Modeling,
National Academy of Sciences, Washington, DC, December 2, 2005.
Organized the minisymposium "Computational systems biology and health risks of environmental chemicals"
as part of the joint SIAM/SMB Conference on the Life Sciences, Brownstone Hotel, Raleigh, NC, July 31 -
August 4, 2006.
Organized the workshop "Systems Biology and the Health Risks of Environmental Chemicals" as part of the
Seventh International Conference on Systems Biology, Yokohama, Japan, October 9-13, 2006.
Co-organized (with Richard Phillips, Exxon-Mobil) the symposium "Computational Toxicology- Industry-Wde
Initiative Plus EPA/NEHS Initiative" as part of the 27th Annual Meeting of the American College of
Toxicology, Renaissance Esmeralda Resort & Spa, Indian wells, California, November 5-8, 2006.
Invited discussion leader, "Additivity to background as a potential source of linearity. Applicability of the general
argument of Crump, Hoel, Langley and Peto, 1976. JNCI 58:1537-41", NRC Workshop on the Implications
of Receptor-Mediated Events on Dose-Response, NRC, Washington, DC, May 3-4, 2007.
"(A biologist's perspective on) Estimating low-dose risk from high-dose data and its associated uncertainty",
NRC Workshop on Quantitative Approaches to Characterizing Uncertainty in Human Cancer Risk
Assessment Based on Bioassay Results, NRC, Washington, DC, June 5, 2007.
Invited participant, IOM Brainstorming Session "Approximating Dose-Response Relationships Using Limited
Data", Institute of Medicine, Washington, DC, June 18, 2007.
Co-chair, ILSI Health And Environmental Sciences Institute, Emerging Issue Subcommittee on Methodology
for Intermittent/ Short-term Exposure to Carcinogens (MISTEC), September 2008 - present.
Invited participant, ILSI-HESI Risk Assessment Brainstorming Session, Washington, DC, August 24-25, 2009.
Co-organizer, SOT 2010 Annual Meeting Workshop "Does Background Disease Lead to Low-Dose Linearity?",
with Harvey Clewell and Lorenz Rhomberg.
PHS 398/2590 (Rev. 09/04, Reissued 4/2006) Page 2
Previous I TOC
-------
Principal Investigator/Program Director (Last, First, Middle): Conolly, Rory
Selected Assistance/Advisory Support to the Agency
"Computational Toxicology and New Directions in Risk Assessment", SOT Contemporary Concepts in
Toxicology Workshop: Probabilistic Risk Assessment (PRA): Bridging Components Along the Exposure-
Dose-Response Continuum, Washington, DC, July 25 - 27, 2005.
Analysis of a PBPK/PD model for perchlorate for the Office of Water, 2008-2009.
B. SELECTED PUBLICATIONS (selected from MOO).
Clewell, H.J., III, Quinn, D.W., Andersen, M.E., and Conolly, R.B. (1995). An improved approximation to
the exact solution of the two-stage clonal growth model of cancer. Risk. Anal. 15, 467-473.
Goldsworthy, T.L., Conolly, R.B., and Fransson-Steen, R. (1996). Apoptosis and cancer risk
assessment. Mutat. Res. 365, 71-90.
Kramer, D.A. and Conolly, R.B. (1997). Computer simulation of clonal growth cancer models. I.
Parameter estimation using an iterative absolute bisection algorithm. Risk Anal. 17,115-126.
R.B. Conolly and M.E. Andersen. (1997). Hepatic foci in rats after diethylnitrosamine initiation and
2,3,7,8-tetrachlorodibenzo-p-dioxin promotion: Evaluation of a quantitative two-cell model and of
CYP1A1/1A2 as a dosimeter. Toxicol. Appl. Pharmacol. 146, 281-293.
Conolly, R.B., Beck, B.D., and Goodman, J.I. (1999). Stimulating research to improve the scientific
basis of risk assessment. Toxicol. Sci., 49, 1-4.
You, L, Archibeque-Engle, S., Casanova, M., Conolly, R.B., and Heck, H.d'A. (1999). Transplacental
and lactational transfer of p,p'-DDE in Sprague-Dawley rats. Toxicol. Appl. Pharmacol. 157,134-
144.
Keys, D.A., Wallace, D.G., Kepler, T.B., and Conolly, R.B. (2000). Quantitative evaluation of alternative
mechanisms of blood disposition of di(n-butyl) phthalate and mono(n-butyl) phthalate in rats.
Toxicol. Sci. 53,173-184.
Haag-Gronlund, M., Conolly, R.B., Scheu, G., Warngard, L., and Fransson-Steen, R. (2000). Analysis
of rat liver foci growth with a quantitative two-cell model after treatment with 2,4,5,3',4'-
pentachlorobiphenyl. Toxicol. Sci. 57, 32-42.
Conolly, R.B., Lilly, P.O., and Kimbell, J.S. (2000). Simulation modeling of the tissue disposition of
formaldehyde to predict nasal DNA-protein cross-links in F344 rats, rhesus monkeys, and humans.
Environ. Hlth. Perspect. 108(suppl 5), 919-924.
Ou, Y.C., Conolly, R.B., Thomas, R.S., Xu, Y., Andersen, M.E., Chubb, L.S., Pitot, H.C., and Yang,
R.S.H. (2001). A clonal growth model: Time-course simulations of liver foci growth following penta-
or hexachlorobenzene tratment in a medium-term bioassay. Cancer Res. 61, 1879-1889.
Conolly, R.B., Kimbell, J.S., Janszen, D., Schlosser, P.M., Kalisak, D., Preston, J., and Miller, F.J.
(2003). Biologically motivated computational modeling of formaldehyde carcinogenicity in the F344
rat. Toxicol. Sci. 75, 432-447.
Ou, Y.C., Conolly, R.B., Thomas, R.S., Gustafson, D.L, Long, M.E., Dobrev, I.D., Chubb, L.S., Xu, Y.,
Lapidot, S.A., Andersen, M.E., and Yang, R.S.H. (2003). Stochastic simulation of hepatic
preneoplastic foci development for four chlorobenzene congeners in a medium-term bioassay.
Toxicol. Sci. 73, 301-314.
Conolly, R.B. and Lutz, W.K. (2004). Non-monotonic dose-response relationships: Mechanistic basis,
kinetic modeling, and implications for risk assessment. Toxicol. Sci. 77:151-157.
Gaylor, D.W., Lutz, W.K., and Conolly, R.B. (2004) Statistical analysis of non-monotonic dose response
relationships: Research design and analysis of nasal cell proliferation in rats exposed to
formaldehyde. Toxicol. Sci. 77:158-164.
Tan, Y.-M., Butterworth, B.E., Gargas, M.L. and Conolly, R.B. (2003). Biologically motivated
computational modeling of chloroform cytolethality and regenerative cellular proliferation. Toxicol.
Sci. 75:192-200.
Paul S. Price, P.S., Conolly, R.B., Chaisson, C.F., Gross, E.A., and Young, J.S. (2003). Modeling inter-
individual variation in physiological factors used in PBPK models of humans. CRC Crit. Rev.
Toxicol. 33:469-503.
Conolly, R.B., Kimbell, J.S., Janszen, D.J., Schlosser, P.M., Kalisak, D., Preston, J., and Miller, F.J.
(2004). Human respiratory tract cancer risks of inhaled formaldehyde: Dose-response predictions
PHS 398/2590 (Rev. 09/04, Reissued 4/2006) Page 3
Previous I TOC
-------
Principal Investigator/Program Director (Last, First, Middle): Conolly, Rory
derived from biologically-motivated computational modeling of a combined rodent and human
dataset. Toxicol. Sci. 82:279-296
Andersen, M.E., Dennison, J.E., Thomas, R.E., and Conolly, R.B. (2005). New directions in incidence-
dose modeling. TRENDS Biotechnol. 23, 122-127.
Conolly, R.B., Gaylor, D.W., and Lutz, W.K. (2005). Population variability in biological adaptive
responses to DMA damage and the shapes of carcinogen dose-response curves. Toxicol. Appl.
Pharmacol. 207, S570-S575.
Lutz, W.K., Gaylor, D.W., Conolly, R.B., and Lutz, R.W. (2005). Nonlinearity and thresholds in dose-
response relationships for carcinogenicity due to sampling variation, logarithmic dose scaling, or
small differences in individual susceptibility. Toxicol. Appl. Pharmacol. 207, S565-S569.
Zhang, Q., Andersen, M.E., and Conolly, R.B. (2006). Binary gene induction and protein expression in
individual cells. Theor. Biol. Med. Modelling 3:18, DOI:10.1186/1742-4682-3-18.
Tan, Yu-Mei, Liao, K.H., Conolly, R.B., Blount, B.C., Mason, A.M., and Clewell, H.J. (2006). Use of a
physiologically based pharmacokinetic model to identify exposures consistent with human
biomonitoring data for chloroform. J. Toxicol. Environ. Health, Part A, 69, 1727-1756,
DOI: 10.1080/15287390600631367.
Breen, M.S., Villeneuve, D.L., Breen, M., Ankley, G.T., and Conolly, R.B. (2007). Mechanistic
computational model of ovarian steroidogenesis to predict biochemical responses to endocrine
active compounds. Annals Biomed. Engineering, 35, 970-981, DOI: 10.1007/sl0439-007-9309-7.
Conolly, R.B., and Thomas, R.S. (2007). Biologically motivated approaches to extrapolation from high
to low doses and the advent of systems biology: The road to toxicological safety assessment.
Human and Ecological Risk Assessment 13, 52-56.
Rhomberg, L.R., Baetcke, K., Blancato, J., Bus, J., Cohen, S., Conolly, R., Dixit, R., Doe, J., Ekelman, K.,
Fenner-Crisp, P., Harvey, P., Hattis, D., Jacobs, A., Jacobson-Kram, D., Lewandowski, T., Liteplo, R.,
Pelkonen, O., Rice, J., Somers, D., Turturro, A., West, W., and Olin, S. (2007). Issues in the design and
interpretation of chronic toxicity and carcinogenicity studies in rodents: Approaches to dose selection. Crit.
Rev. Toxicol. 37: 729-837, DOI: 10.1080/10408440701524949.
Liao, K.H., Tan, Y.-M., Conolly, R.B., Borghoff, S.J., Gargas, M.L., Andersen, M.E., and Clewell, H.J., III.
(2007). Bayesian estimation of pharmacokinetic and pharmacodynamic parameters in a
mode-of-action-based cancer risk assessment for chloroform. Risk Anal. 27, 1535-1551.
Nong, A., Tan, Y.-M., Krolski, M.E., Wang, J., Lunchick, C., Conolly, R.B., and Clewell, H.J., III. (2008).
Bayesian calibration of a physiologically based pharmacokinetic/pharmacodynamic model of carbaryl
cholinesterase inhibition. J. Toxicol. Environ,. Health 71, 1363-1381.
Ankley, G.T., Bencic, D.C., Breen, M.S., Collette, T.W., Conolly, R.B., Denslow, N.D., Edwards, S.W., Ekman,
D.R., Garcia-Reyero, N., Jensen, K.M., Lazorchak, J.M., Martinovic, D., Miller, D.H., Perkins, E.J., Orlando,
E.F., Villeneuve, D.L., Wang, R.-L, and Watanabe, K.H. (2009). Endocrine disrupting chemicals in fish:
Developing exposure indicators and predictive models of effects based on mechanism of action. Aquatic
Toxicology doi:10.1016/j.aquatox.2009.01.013.
Breen, M.S., Breen, M., Terasaki, N., Yamazaki, M., and Conolly, R.B. (2009). Computational model of
steroidogenesis in human H295R cells to predict biochemical response to endocrine active chemicals:
Model development for metyrapone. Submitted to Environmental Health Perspectives.
Bhattacharya, S., Andersen, M.E., Conolly, R.B., Thomas, R.S., Kaminski, N.E., and Zhang, Q. A
transcriptional regulatory switch underlying B cell terminal differentiation and Its disruption by dioxin.
Submitted to PLOS Biology.
Zhang, Q., Bhattacharya, S., Crawford, R.B., Kline, D.E., Conolly, R.B., Thomas, R.S., Kaminski, N.E., and
Andersen,M.E., Stochastic modeling of B lymphocyte terminal differentiation and Its suppression by dioxin.
Submitted to BMC Systems Biology.
Kitchin, K.T., and Conolly, R.B. Arsenic induced carcinogenesis - oxidative stress as a possible mode of action
and future research needs for more biologically based risk assessment. Submitted to Chemical research in
Toxicology.
Luke, N.S., Sams, R.S., III, DeVito, M.J., Conolly, R.B., and EI-Masri, H.A. Development of a quantitative
model incorporating key events in a hepatotoxic mode of action to predict tumor incidence. Submitted to
Toxicological Sciences.
PHS 398/2590 (Rev. 09/04, Reissued 4/2006) Page 4
Previous I TOC
-------
Principal Investigator/Program Director (Last, First, Middle): Gorton, Jon Christopher
BIOGRAPHICAL SKETCH
NAME
Jon "Chris"topher Gorton
eRA COMMONS USER NAME
POSITION TITLE
Senior Research Biologist
EDUCATION/TRAINING (Begin with baccalaureate or other initial professional education, such as nursing, and include postdoctoral
training.}
INSTITUTION AND LOCATION
Grinnell College, Grinnell, IA
University of Kansas Medical Center, Kansas
City, KS
DEGREE
(if applicable)
B.S.
Ph.D.
YEAR(s)
1979
1984
FIELD OF STUDY
Biology/Chemistry
Biochemistry
A. POSITIONS and HONORS
Research and Professional Experience:
2009-Present Senior Research Biologist, NHEERL, Integrated Systems Toxicology Division
2006- 2009 Leader, National Health Environmental Effects Research Laboratory (NHEERL)
Toxicogenomics Core Facility
2005-Present Senior Research Biologist, NHEERL, Environmental Carcinogenesis Division
2002-2005 Consultant, ToxicoGenomics, Chapel Hill, NC
1989-2002 Research Biologist, CUT, Centers for Health Research, RTF, NC
1987-1989 Research Fellow, Duke University, Durham, NC
Invited Lectures/Symposia (last 2 years):
Gorton, J. C. (2006) Invited presentation, "Modulation of xenobiotic metabolizing enzyme
expression by caloric restriction through PGC-1 alpha and PPARalpha" Experimental
Biology Annual Meeting, San Francisco, CA. April, 2006.
Gorton, J. C. (2007) Invited presentation, "Nuclear receptor activation - risk assessment and
regulatory implications" Perfluoroalkyl Acids and Related Chemistries: Toxicokinetics and
Mode-of-Action Workshop, Arlington, VA, Feb. 14-17, 2007
Gorton, J. C. (2007) Invited presentation, "Nuclear receptor transcriptional networks"
University of Tennessee, Memphis. Sept., 2007.
Gorton, J. C. (2007) Invited presentation, "Nuclear receptors and aging" Gordon Conference on
Aging, Les Diableretes, Switzerland. Sept., 2007
Gorton, J. C. (2007) Invited presentation, "Nuclear receptor transcriptional networks" University
of Nice, Nice, France. October, 2007.
Gorton, J. C. (2007) Invited presentation, "Toxicogenomics of nuclear receptors" Bayer Crop
Science, Sofia-Annipolis, France, October, 2007.
Gorton, J. C. (2007) Invited presentation, "Nuclear receptor transcriptional networks" University
of Burgandy, Dijon, France October, 2007.
Gorton, J. C. (2007) Invited presentation, "Toxicogenomic dissection of the PFOA
transcriptional profile" EuroTox 2007, Amsterdam, Netherlands, October, 2007.
Pagel
Previous
TOC
-------
Principal Investigator/Program Director (Last, First, Middle): Gorton, Jon Christopher
Gorton, J. C. (2008) Invited presentation, "Toxicogenomic dissection of the PFOA
transcriptional profile" PFAA-II meeting, EPA, June, 2008.
Gorton, J. C. (2009) Invited presentation, "Toxicogenomic dissection of nuclear receptor action"
NC State University. March, 2009.
Assistance/Leadership Provided to the Scientific Community:
• Advisory appointments (since 1998): NIH/NIEHS: Chaired 11 review panel committees;
reviewer on 30 panel committees. Member, Scientific Committee, Critical Assessment of
Techniques for Microarray Data Analysis, 2000-2005; Member, Peroxisome Proliferator
Case Study Work Group, ILSI, Washington, B.C., 2001-2003; Reviewer, TCE Risk
Assessment, NY Department of Health, 2005.
• Society of Toxicology activities: Member, Continuing Education Committee, 2006-2009;
Chair, Continuing Education Committee, 2008-2009; Co-chair, 8 symposia, roundtables,
continuing education courses since 1999; Secretary/Treasurer, Carcinogenesis Specialty
Section, 2006-2008.
• Editorial board appointments: Archives of Toxicology, 2001-2003; Cell Biology and
Toxicology, 2001-present; International Journal of Toxicology, 1998-2008; Toxicological
Sciences, 2001-present; Toxicology, 1999-present; Toxicology Letters, 2001-present; PPAR
Research, 2005-present; Chemico-Biological Interactions, 2005-present; Journal of
Pharmacology and Experimental Therapeutics, 2009- present.
• Academic affiliations: Adjunct Assistant Professor, Curriculum in Toxicology, University of
North Carolina, Chapel Hill, NC; Adjunct Assistant Professor, Integrated Toxicology
Program, Duke University, Durham, NC; Associate Member, Graduate Faculty, University
of Louisiana, Monroe, LA
• Miscellaneous: Organizing Committee, Dioxin '91; founder and leader, Triangle Array
Users Group, April 1999-2006; Chair, platform session, Gordon Research Conference on
Mycotoxins and Phycotoxins, Waterville, ME, 2005; Chair, symposium, "Obesity as a
modulator of chemical toxicity", AAAS annual meeting, San Francisco, CA. 2007; Member,
Genomics Committee, ILSI/HESI 2006-present. Member, microarray quality control
(MAQC) committee, 2006-present.
Assistance/Leadership Provided to the Agency:
• Co-chair, PPARalpha workgroup, Risk Assessment Forum, 2005-present.
• Member, Data Analysis Workgroup, 2007-present.
• Member, Virtual Liver Project, 2006-present.
• Leader, NHEERL Toxicogenomics Core, 2006-present.
B. SELECTED PUBLICATIONS (selected more than 65 total).
Valles E.G., Laughter A.R., Dunn C.S., Cannelle S., Swanson C.L., Cattley R.C. and Corton
J.C. (2003) Role of the peroxisome proliferator-activated receptor alpha in responses to the
hepatocarcinogenic phthalate, diisononyl phthalate (DINP). Toxicology. 191, 211-25.
Laughter, A.R., Dunn, C.S., Howroyd, P., Cattley, R.C., Swanson, C., Corton, J.C. (2004) Role
of the peroxisome proliferator-activated receptor alpha in responses to trichloroethylene and
metabolites, trichloroacetate and dichloroacetate. Toxicology. 203, 83-98.
Page 2
Previous I TOC
-------
Principal Investigator/Program Director (Last, First, Middle): Gorton, Jon Christopher
Howroyd, P., Swanson, C., Dunn, C., Cattley, R.C. and Corton, J.C. (2004) Decreased
Longevity and Acceleration of Age-Dependent Lesions in Mice Lacking the Nuclear
Receptor Peroxisome Proliferator-Activated Receptor a (PPARa). lexicological Pathology.
32,591-599.
Corton, J.C., Apte, U., Anderson, S.P., Limaye, P., Yoon, L., Latendresse, J., Dunn, C.S.,
Everitt, J.I., Voss, K.A., Swanson, C., Wong, J.S., Gill, S.S., Chandraratna, R.A.S., Kwak,
M.-K., Kensler, T.W., Stulnig, T.M., Steffensen, K.R., Gustaffson, J.-A. and Mehendale, H.
(2004) Caloric restriction mimetics include nuclear receptor agonists. Journal of Biological
Chemistry. 279, 46204-12.
Anderson, S.P., Dunn, C.S., Laughter, A.R., Yoon, L., Swanson, C., Stulnig, T.M., Steffensen,
K.R., Chandraratna, R.A.S., Gustafsson, J-A and Corton, J.C. (2004) Overlapping
Transcriptional Programs Regulated by Peroxisome Proliferator-Activated Receptor a,
Retinoid X Receptor and Liver X Receptor in Mouse Liver. Molecular Pharmacology. 66,
1440-1452.
Anderson, S.P., Howroyd, P., Liu, J., Swanson, C., Bahnemann, R., and Corton, J.C. (2004) The
transcriptional response to a peroxisome proliferator-activated receptor a (PPARa) agonist
includes increased expression of proteome maintenance genes. Journal of Biological
Chemistry. 279, 52390-52398.
Corton, J.C. and Lapinskas, P. (2005) Peroxisome proliferator-activated receptors: role in
phthalate-induced male reproductive tract toxicity? Toxicological Sciences. 83, 4-17.
Lapinskas, P.J., Brown, S., Swanson, C., Cattley, R.C., and Corton, J.C. (2005) Role of
peroxisome proliferator-activated receptor alpha in mediating phthalate-induced liver
toxicity. Toxicology.207, 149-163.
Stauber, A.J., Brown-Borg, H., Liu, J., Laughter, A., Staben, R.A., Coley, J.C., Swanson, C.,
Voss, K.A., Kopchick, J.J. and Corton, J.C. (2005) Constitutive expression of peroxisome
proliferator-activated receptor alpha and regulated genes in dwarf mice. Molecular
Pharmacology. 67, 681-694.
Xiao S., Anderson S.P., Swanson C., Bahnemann R., Voss K.A., Stauber A.J. and Corton J.C.
Activation of the Nuclear Receptor Peroxisome Proliferator-Activated Receptor alpha
(PPARa) Enhances Hepatocyte Apoptosis. . Toxicological Sciences. 92, 368-77.
Martin M.T., Brennan R., Hu W., Ayanoglu E., Lau C., Ren H., Wood C.R., Corton J.C.,
Kavlock R.J., Dix D.J. (2007). Toxicogenomic Study of Triazole Fungicides and
Perfluoroalkyl Acids in Rat Livers Predicts Toxicity and Categorizes Chemicals Based on
Mechanisms of Toxicity. Toxicol Sci. 97:595-613.
Fostel JM, Burgoon L, Zwickl C, Lord P, Corton JC, Bushel PR, Cunningham M, Fan L,
Edwards SW, Hester S, Stevens J, Tong W, Waters M, Yang C, Tennant R.. (2007). Towards
a checklist for exchange and interpretation of data from a toxicology study. Toxicol Sci.
99:26-34.
MAQC Consortium. (2007). The MicroArray Quality Control (MAQC) project shows
interplatform reproducibility of gene expression measurements. Nature Biotechnology. 24,
1151-61
Boedigheimer MJ, Wolfinger RD, Bass MB, Bushel PR, Chou JW, Cooper M, Corton JC, Fostel
J, Hester S, Lee JS, Liu F, Liu J, Qian HR, Quackenbush J, Pettit S, Thompson KL. (2008).
Sources of variation in baseline gene expression levels from toxicogenomics study control
animals across multiple laboratories. BMC Genomics. 12, 285.
Previous I TOC
-------
Principal Investigator/Program Director (Last, First, Middle): Gorton, Jon Christopher
Rosen MB, Lee JS, Ren H, Vallanat B, Liu J, Waalkes MP, Abbott BD, Lau C, Corton JC.
(2008). Toxicogenomic dissection of the perfluorooctanoic acid transcript profile in mouse
liver: evidence for the involvement of nuclear receptors PPAR alpha and CAR. Toxicol Sci.
103,46-56.
Rosen MB, Abbott BD, Wolf DC, Corton JC, Wood CR, Schmid JE, Das KP, Zehr RD, Blair
ET, Lau C. (2008). Gene Profiling in the Livers of Wild-Type and PPARalpha-Null Mice
Exposed to Perfluorooctanoic Acid (PFOA). Toxicol Pathol. 36, 592-607.
Lee JS, Ward WO, Wolf DC, Allen JW, Mills C, DeVito MJ, Corton JC. (2008). Coordinated
changes in xenobiotic metabolizing enzyme gene expression in aging male rats. Toxicol Sci.
106,263-83.
Corton, JC (2008) Role of PPARalpha in mediating the effects of trichloroethylene and
metabolites. Critical Reviews in Toxicology. 27,1.
Ren H, Vallanat B, Nelson DM, Yeung LW, Guruge KS, Lam PK, Lehman-McKeeman LD,
Corton JC. (2009) Evidence for the involvement of xenobiotic-responsive nuclear receptors
in transcriptional effects upon perfluoroalkyl acid exposure in diverse species. Reprod
Toxicol. 27, 266-77
Elam MB, Cowan GS Jr, Rooney RJ, Hiler ML, Yellaturu CR, Deng X, Howell GE, Park EA,
Gerling 1C, Patel D, Corton JC, Cagen LM, Wilcox HG, Gandhi M, Bahr MH, Allan MC,
Wodi LA, Cook GA, Hughes TA, Raghow R. (2009). Hepatic Gene Expression in Morbidly
Obese Women: Implications for Disease Susceptibility. Obesity (Silver Spring). In press.
Carmen Gonzalez M, Corton JC, Cattley RC, Herrera E, Bocos C. (2009). Peroxisome
proliferator-activated receptor alpha (PPARalpha) agonists down-regulate alpha2-
macroglobulin expression by a PPARalpha-dependent mechanism. Biochimie. 91, 1029-35.
Michael J. Boedigheimer, Jeff W. Chou, Matthew Cooper, J. Christopher Corton, Jennifer Fostel,
Raegan O'Lone, P. Scott Pine, John Quackenbush, Karol L. Thompson, and Russell D.
Wolfinger (2009). Sources of Variance in Rat Liver and Kidney Baseline Gene Expression in
a Large Multi-Site Dataset. Submitted. In Batch effects and experimental noise in microarray
studies: sources and solutions, ed. Scherer, A. John Wiley & Sons Ltd, Publisher
Rosen MB, Lau C, Corton JC. (2009). Does Exposure to Perfluoroalkyl Acids Present a Risk to Human
Health? Toxicol Sci. In press.
Submitted
Beena Vallanat, Steven P. Anderson, Holly M. Brown-Borg, Hongzu Ren, Sander Kersten,
Sudhakar Jonnalagadda, Rajagopalan Srinivasan and J. Christopher Corton (2009). Analysis
of the Heat Shock Response in Mouse Liver Reveals Transcriptional Dependence on the
Nuclear Receptor Peroxisome Proliferator-Activated Receptor alpha (PPARa). BMC
Genomics. Submitted.
Gail M. Nelson, Gene J. Ahlborn, James W. Allen, Hongzu Ren, J. Christopher Corton, Michael
P. Waalkes, Kirk T. Kitchin, and Don A. Delker (2009). Impact of Life Stage and Duration
of Exposure on Liver Gene Expression in Arsenic-Treated Male C3H Mice. Toxicology.
Submitted.
Ren H, Aleksunes, LM, Wood, C, Vallanat B, George, M, Klaassen, CD, Corton JC. (2009)
Characterization of Peroxisome Proliferator-Activated Receptor a (PPARa) - Independent
Effects of PPARa Activators in the Rodent Liver: Di-(2-ethylhexyl) phthalate Activates the
Constitutive Activated Receptor. Tox Sci Submitted.
Page 4
Previous I TOC
-------
Principal Investigator/Program Director (Last, First, Middle): Davis, Jimena L.
BIOGRAPHICAL SKETCH
NAME
Jimena L. Davis
eRA COMMONS USER NAME
POSITION TITLE
Mathematical Statistician
EDUCATION/TRAINING (Begin with baccalaureate or other initial professional education, such asnursing, and include postdoctoral training.)
INSTITUTION AND LOCATION
Clemson University, Clemson, SC
North Carolina State University, Raleigh, NC
North Carolina State University, Raleigh, NC
NCCT, US EPA, Research Triangle Park, NC
DEGREE
(if applicable)
B.S.
M.S.
Ph.D.
YEAR(s)
2003
2005
2008
2008-
FIELD OF STUDY
Mathematical Sciences
Applied Mathematics
Computational
Mathematics
Uncertainty Analysis and
Risk Assessment
A. POSITIONS and HONORS
Research and Professional Experience:
2008-Present Cross ORD Postdoctoral Fellow, National Center for Computational Toxicology, US EPA,
RTP, NC (Mentors: Woodrow Setzer and Rogelio Tornero-Velez)
2003-2008 Research Assistant, Department of Mathematics, North Carolina State University, Raleigh, NC
(Advisor: H.T. Banks)
2006 Student Intern, Computational Biology Department, Sandia National Laboratories,
Albuquerque, NM (Advisor: Elebeoba E. May)
2002 Participant, Research Experience for Undergraduates in Computational Number Theory and
Combinatorics, Clemson University, Clemson, SC
Professional Societies and Affiliations:
2003-present Member, Society of Industrial and Applied Mathematics
2003-present Member, American Mathematical Society
Honors and Awards:
2009 Superior Accomplishment Recognition Award from the Office of Pesticide Programs
2004 - 2008 Department of Energy Computational Science Graduate Student Fellowship,
North Carolina State University, Raleigh, NC
2006 & 2007 Association for the Concerns of African American Graduate Students Academic Achievement
Award - College of Physical and Mathematical Sciences,
North Carolina State University, Raleigh, NC
2003 - 2004 Statistical and Applied Mathematical Sciences Institute Fellowship,
North Carolina State University, Raleigh, NC
2003 - 2004 Mathematics Department Fellowship,
North Carolina State University, Raleigh, NC
2003 Faculty Scholarship Award,
Clemson University, Clemson, SC
2003 summa cum laude Graduate,
Clemson University, Clemson, SC
2002 & 2003 Beta Kappa Chapter of Phi Sigma Pi Scholarship Award,
Clemson University, Clemson, SC
2001 - 2003 Multicultural Achievement Award,
PHS 398/2590 (Rev. 09/04, Reissued 4/2006)
Page J
Previous
TOC
-------
Principal Investigator/Program Director (Last, First, Middle): Davis, Jimena L.
Clemson University, Clemson, SC
2001 - 2003 Susan & Harry Frampton Scholarship,
Clemson University, Clemson, SC
2000 Mathematical Sciences Freshman Award,
Clemson University, Clemson, SC
1999 - 2003 Palmetto Fellows Scholarship,
Clemson University, Clemson, SC
1999 - 2003 Coca-Cola Clemson Scholarship,
Clemson University, Clemson, SC
Selected Expert Committees/Advisory Panels/Organizing Committees:
2009 -present President, EPA-RTP Networking and Leadership Training Organization (NLTO)
B. SELECTED PUBLICATIONS
Banks H.T., Davis J.L., and Hu S., "A Computational Comparison of Alternatives to Including Uncertainty in
Structured Population Models," CRSC Tech. Rpt. CRSC-TR09-14, North Carolina State University, June
2009; Three Decades of Progress in Systems and Control, to appear
Banks H.T., Davis J.L., Ernstberger S.L., Hu S., Artimovich E., and DharA.K., "Experimental Design and
Estimation of Growth Rate Distributions in Size-Structured Shrimp Populations," CRSC Tech. Rpt. CRSC-
TR08-20, North Carolina State University, November 2008; Inverse Problems, to appear
Davis J.L., "Uncertainty Quantification in the Estimation of Probability Distributions on Parameters in Size-
Structured Populations," Ph.D. Dissertation (2008)
Banks H.T., Davis J.L., Ernstberger S.L., Hu S., Artimovich E., DharA.K., and BrowdyC.L, "A Comparison of
Probabilistic and Stochastic Formulations in Modeling Growth Uncertainty and Variability," CRSC Tech.
Rpt. CRSC-TR08-03, North Carolina State University, February 2008; Journal of Biological Dynamics,
Volume 3, pages 130 - 148 (2009)
Banks H.T. and Davis J.L., "Quantifying Uncertainty in the Estimation of Probability Distributions," CRSC Tech.
Rpt. CRSC-TR07-21, North Carolina State University, December 2007; Mathematical Biosciences and
Engineering, Volume 5, pages 647 - 667 (2008)
Banks H.T. and Davis J.L., "A Comparison of Approximation Methods for the Estimation of Probability
Distributions on Parameters," CRSC Tech. Rpt. CRSC-TR05-38, North Carolina State University, October
2005; Applied Numerical Mathematics, Volume 57, pages 753 - 777 (2007)
Calkin N., Davis J., James K., Perez E., and Swannack C., "Computing the Integer Partition Function,"
Mathematics of Computation, Volume 76, pages 1619 - 1638 (2007)
Davis J., Fricks J., Macabea J., Stroud L., White G., and Wong A., "Evaluating a Physiologically Based
Pharmacokinetic Model Proposed for Use in Risk Assessment," 2003 Industrial Mathematics Modeling
Workshop for Graduate Students, CRSC Tech. Rpt. CRSC-TR04-07, North Carolina State University,
March 2004
C. SELECTED PRESENTATIONS
"Quantifying Uncertainty in the Estimation of Probability Distributions on Parameters," SIAM 2008 Annual
Meeting Graduate Student Workshop on Diversity, San Diego, CA, July 2008 (oral presentation)
"Quantifying Uncertainty in the Estimation of Probability Distributions," Department of Energy Computational
Science Graduate Fellows' Annual Conference, Washington, DC, June 2008 (oral presentation)
"Estimation of Probability Distributions on Parameters in Size-Structured Populations," North Carolina State
Mathematics Department Graduate Recruitment Weekend, Raleigh, NC, February 2008 (oral presentation)
"Uncertainty Quantification in the Estimation of Probability Distributions on Parameters," Applied Mathematics
Graduate Student Seminar, Raleigh, NC, January 2008 (oral presentation)
"A Comparison of Approximation Methods for the Estimation of Probability Distributions on Parameters,"
Infinite Possibilities Conference, Raleigh, NC, November 2007 (oral presentation)
"Using Confidence Bands to Quantify Uncertainty in the Estimation of Probability Distributions," Atlantic Coast
Conference on Mathematics in the Life and Biological Sciences, Blacksburg, VA, May 2007 (oral
PHS 398/2590 (Rev. 09/04, Reissued 4/2006) Page 2
Previous I TOC
-------
Principal Investigator/Program Director (Last, First, Middle): Davis, Jimena L.
presentation)
"A Study of Computational Approaches for Parameter Estimation in the Escherichia coli K-12 Central Metabolic
System," 2006 Student Symposium, Sandia National Laboratories, Albuquerque, NM, August 2006 (oral
presentation)
"A Computational and Statistical Comparison of Approximation Methods for the Estimation of Probability
Distributions on Parameters," Department of Energy Computational Science Graduate Fellows' Annual
Conference, Washington, DC, June 2006 (poster presentation)
"Comparison of Two Approximation Methods in the Estimation of Growth Rate Distributions in Size-Structured
Mosquitofish Populations," SIAM-SEAS Conference, Charleston, SC, March 2005 (oral presentation)
"Distributions of Growth Rates in Size-Structured Mosquitofish Population Models," Journees Jeunes, Paris,
France, March 2004 (oral presentation)
D. WORKSHOPS
Genomes to Global Health: Modeling of Infectious Diseases, Statistical and Applied Mathematical Sciences
Institute, Research Triangle Park, NC, September 2004
SAMSI/CRSC Undergraduate Workshop, North Carolina State University, Raleigh, NC, June 2004
Mathematics Meets Biology: Epidemics, Data Fitting, and Chaos, MAA Prep Workshop, University of
Louisiana at Lafayette, Lafayette, LA, May 2004
Data Mining and Machine Learning, Statistical and Applied Mathematical Sciences Institute, Research Triangle
Park, NC, September 2003
2003 Industrial Mathematics Modeling Workshop for Graduate Students, Center for Research in Scientific
Computation, North Carolina State University, Raleigh, NC, July 2003
E. CONTINUING EDUCATION
"Characterizing Variability and Uncertainty with Physiologically-Based Pharmacokinetic Models," Society of
Toxicology, Baltimore, Maryland, March 2009
PHS 398/2590 (Rev. 09/04, Reissued 4/2006) Page 3
Previous I TOC
-------
Principal Investigator/Program Director (Last, First, Middle): Degitz Jr., Sigmund J.
BIOGRAPHICAL SKETCH
Provide the following information for the key personnel and other significant contributors in the order listed on Form Page 2.
Follow this format for each person. DO NOT EXCEED FOUR PAGES.
NAME
Sigmund J, Degitz Jr.
eRA COMMONS USER NAME
POSITION TITLE
Toxicologist
EDUCATION/TRAINING (Begin with baccalaureate or other initial professional education, such as nursing, and include postdoctoral training.)
INSTITUTION AND LOCATION
Northland College, Ashland, Wl
University of Illinois, Urbana-Champaign
DEGREE
(if applicable)
B.S
Ph.D.
YEAR(s)
1991
1996
FIELD OF STUDY
Biology
Toxicology
A. POSITIONS and HONORS
Research and Professional Experience:
1998-Present Toxicologist, U.S. EPA, Duluth, MN
1996-1998 Postdoctoral Research Fellow, University of North Carolina, Chapel Hill
Selected Awards and Honors:
University of Minnesota-Duluth, Adjunct Associate, Integrated Biosciences Graduate Program
Bronze Medal OECD Support Team. For leadership in development of internationally-harmonized EDC Test
methods through the OECD. 2003
Bronze Medal Promoting Strong Science in Agency Decisions. For significant achievements in working with
program offices to promote the use of strong science in Agency decisions. 2003
STAA award level 3
STAA award level 2
B. SELECTED PUBLICATIONS
C.
Olmstead AW, Kosian PA, Korte JJ, Holcombe GW, Woodis KK, Degitz SJ. Sex reversal of the amphibian,
Xenopus tropicalis, following larval exposure to an aromatase inhibitor. AquatToxicol. 2009 Jan 31;91(2):143-
50.
Olmstead AW, Korte JJ, Woodis KK, Bennett BA, Ostazeski S, Degitz SJ. Reproductive maturation of the
tropical clawed frog: Xenopus tropicalis. Gen Comp Endocrinol. 2009 Jan 15;160(2):117-23
Helbing, CC, Ji L, Bailey CM, Veldhoen N, Zhang F, Holcombe GW, Kosian PA, Tietge JE, and SJ Degitz.
(2007) Identification of gene expression indicators for thyroid axis disruption in a Xenopus laevis
metamorphosis screening assay. Part 2: Effects on the tail and hindlimb
Helbing CC, Bailey CM, Ji L, Gunderson MP, Zhang F, Veldhoen N, Skirrow RC, Mu R, Lesperance M,
Holcombe GW, Kosian PA,Tietge JE, and SJ Degitz. (2007) Identification of gene expression indicators
for thyroid axis disruption in a Xenopus laevis metamorphosis screening assay. Part 1: Effects on the
brain AquatToxicol. 2007 May 31;82(4):227-41.
Douglas JF, Denver R, Degitz SJ, Tietge JE, and LWTouart (2007)The hypothalmic-pituitary-thyroid (HPT)
axis in frogs and its role in frog development and reproduction. (In press)
Villeneuve DL, Miracle AL, Jensen KM, Degitz SJ, Kahl MD, Korte JJ, Greene KJ, Blake LS, Linnum AL,
Ankley GT (2006) Development of quantitative real-time PCR assays for fathead minnow (Pimephales
promelas) gonadotropin beta subunit mRNAs to support endocrine disrupter research. Comp Biochem
Physiol C Toxicol Pharmacol.
Ankley GT, Daston GP, Degitz SJ, Denslow ND, Hoke RA, Kennedy SW, Miracle AL, Perkins EJ, Snape J,
Tillitt DE, Tyler CR, and D Versteeg (2006) Toxicogenomics in Regulatory Ecotoxicology
Environmenal Science and Technology 40(13):4055-4065.
Zhang F, Degitz SJ, Holcombe GW, Kosian PA, Tietge J, Veldhoen N, Helbing CC. (2006) Evaluation of gene
i
Previous
TOC
-------
Principal Investigator/Program Director (Last, First, Middle): Degitz Jr., Sigmund J.
expression endpoints in the context of a Xenopus laevis metamorphosis-based bioassay to detect
thyroid hormone disrupters. Aquatic Toxicology. 76(1):24-36.
Degitz SJ, Holcombe GW, Flynn KM, Kosian PA, Korte JJ, Tietge JE.(2005) Progress towards development of
an amphibian-based thyroid screening assay using Xenopus laevis. Organismal and thyroidal
responses to the model compounds 6-propylthiouracil, methimazole, and thyroxine. Toxicological
Sciences. 87(2):353-64.
Tietge, JE, GW, Holcombe, KM, Flynn, PA, Kosian, JJ, Korte, LE, Anderson, DC, Wolf, and SJ, Degitz. (2005).
Metamorphic inhibition of Xenopus laevis by sodium perchlorate: effects on development and thyroid
histology. Environmental Toxicology and Chemistry, 24:926-933.
Ankley GT, SJ Degitz, SA Diamond, and JE Tietge (2004). Assessment of environmental steressors potentially
responsible for malformations in North American anuran amphibians. Ecotoxicology and Environmental
Safty58, 7-16.
Degitz SJ, Rogers JM, Zucker RM, Hunter ES 3rd. (2004) Developmental toxicity of methanol: Pathogenesis in
CD-1 and C57BL/6J mice exposed in whole embryo culture.
Birth Defects Res Part A Clin Mol Teratol. 70,179-84
Degitz SJ, Zucker RM, Kawanishi CY, Massenburg GS, Rogers JM. (2004) Pathogenesis of methanol-induced
craniofacial defects in C57BL/6J mice. Birth Defects Res Part A Clin Mol Teratol. 70, 172-8.
Rogers JM, Brannen KC, Barbee BD, Zucker RM, Degitz SJ. (2004) Methanol exposure during gastrulation
causes holoprosencephaly, facial dysgenesis, and cervical vertebral malformations in C57BL/6J mice.
Birth Defects Res Part B Dev Reprod Toxicol. 71, 80-8.
Kosian PA, EA Makynen, GT Ankley, and SJ Degitz.(2003) Bioconcentration and Metabolism of All-Trans
Retinoic Acid by Three Native North American Ranids. Toxicological Sciences 74, 147-156.
Degitz SJ, PA Kosian, GW Holcombe, JE Tietge, EJ Durhan, and GT Ankley.(2003) Comparing the effects of
Retinoic Acid on amphibian limb development and lethality: chronic exposure results in lethality not limb
malformations. Toxicological Sciences 74, 139-146
Degitz, SJ, EJ Durhan, PA Kosian, GT Ankley, and JE Tietge. (2003) Development toxicity of methoprene and
its degradation products in Xenopus laevis. Aquatic Toxicology 64, 97-105.
Gray, LE, Ostby, V. Wilson, C. Lambright, K. Bobseine, P. Hartig, A. Hotchkiss, C. Wold, J. Furr, M. Price, L
Parks, R. Cooper, T. Stoker, S. Laws, S. Degitz, K.M. Jensen, M.D. Kahl, JJ. Korte, E.A. Makynen,
J.E. Tietge, and G.T. Ankley (2002) Xenoendocrine disrupters - tiered screening and testing: Filling key
data gaps. Toxicology. 181-182, 371-82.
Simon R, JE Tietge, B Michalke, SJ Degitz, and KWSchramm. (2002) Iodine species and the endocrine
system: thyroid hormone levels in adult Danio rerio and developing Xenopus laevis. Anal Bioanal
Chem 372, 481-5
Tietge, JE, SA Diamond, GT Ankley, DL DeFoe, GW Holcombe, KM Jensen, SJ Degitz, GE Elonen, and E
Hammer (2000). Ambient solar UV-B causes mortality in larvae of three species of Rana.
Photochemistry and Photobiology 74, 261-268.
Degitz SJ, PA Kosian, EA Makynen, KM Jensen and GT Ankley (2000) Stage- and Species-specific
Developmental Toxicity of All- Trans Retinoic Acid in Four Native North American Ranids and Xenopus
laevis. Toxicological Sciences 57, 264-274.
Ankley GT, JE Tietge, GW Holcombe, DL DeFoe, SA Diamond, KM Jensen, and SJ Degitz.(2000) Effects of
laboratory ultraviolet light and natural sunlight on survival and development of Rana pipiens. Can. J.
Zool. 78, 1092-1100.
Previous I TOC
-------
Principal Investigator/Program Director (Last, First, Middle): DJX, David J.
BIOGRAPHICAL SKETCH
Provide the following information for the key personnel and other significant contributors in the order listed on Form Page 2.
Follow this format for each person. DO NOT EXCEED FOUR PAGES.
NAME
David J. Dix
eRA COMMONS USER NAME
POSITION TITLE
Research Biologist
EDUCATION/TRAINING (Begin with baccalaureate or other initial professional education, such as nursing, and include postdoctoral training.)
INSTITUTION AND LOCATION
University of Illinois at Chicago
Rush University, Chicago IL
North Carolina State University, Raleigh NC
National Institute of Environmental Health
Sciences, RTP, NC
DEGREE
(if applicable)
B.S.
Ph.D
Post-doc
Post-doc
YEAR(s)
1985
1990
1990-1992
1992-1995
FIELD OF STUDY
Biological Sciences
Physiology
Biochemistry
Reproductive and
Developmental
Toxicology
A.
POSITIONS and HONORS
Research and Professional Experience:
2009-present Acting Deputy Director, National Center for Computational Toxicology, U.S. EPA/ORD, RTP NC
2005-present Research Biologist, National Center for Computational Toxicology, U.S. EPA/ORD, RTP NC
2008-present Adjunct Associate Professor, Department of Environmental Sciences and Engineering,
University of North Carolina, Chapel Hill, NC.
2004-2005 Lead Research Biologist, Genomic Effects Team, U.S. EPA, ORD/NHEERL/RTD, RTP, NC.
2001-2007 Adjunct Assistant Professor, Department of Molecular and Environmental Toxicology, North
Carolina State University, Raleigh, NC.
1997-1998 Adjunct Assistant Professor, Department of Biology, North Carolina Central University, Raleigh,
NC.
1995-2004 Research Biologist, Gamete and Early Embryo Branch, Reproductive Toxicology Division,
National Health and Environmental Effects Research Laboratory, Office of Research and
Development, U.S. Environmental Protection Agency, RTP, NC.
1992-1995 Intramural Research Fellow, Laboratory of Reproductive and Developmental Toxicology,
National Institute of Environmental Health Sciences, Research Triangle Park, NC (Mentor: Dr.
E.M. Eddy).
1990-1992 Postdoctoral Research Associate, Department of Biochemistry, North Carolina State University,
Raleigh, NC (Mentor: Dr. Elizabeth Theil).
Professional Societies and Affiliations:
1999-2006 Society for the Study of Reproduction
2001-present Society of Toxicology
Honors and Awards:
2006 EPA Superior Accomplishment Recognition Award (SARA) for creating and chairing the EPA Chemical
Prioritization Community of Practice.
2007 SARA for fostering coordination between ToxCast and OECD Molecular Screening Project.
2007 EPA Quality Step Increase for outstanding performance.
2007 ORD Bronze Medal Award for characterizing toxicity pathways of conazoles.
2008 Promotion through Technical Qualifications Board to GS-15.
2008 ORD Scientific and Technological Achivevement Award Level II for metabolomics program.
PHS 398/2590 (Rev. 09/04, Reissued 4/2006)
Page J
Previous
TOC
-------
Principal Investigator/Program Director (Last, First, Middle): DJX, David J.
Selected invitations at National & International Symposia:
International Society Regulatory Toxicology Pharmacology annual meeting, Nov 2005, Baltimore, MD. "ORD's
Computational Toxicology Research Program".
NAS/NRC workshop on Applications of Genomic Signatures, Dec 2005, Welches, OR. "Antifungal
Toxicogenomics".
CARCINOGENOMICS 1st Annual EU Consortium Meeting, Nov 2007, Valencia, SPAIN. "EPA's ToxCast
Program: Prioritizing the Toxicity Testing of Environmental Chemicals".
4th International Conference on Toxicogenomics (ICT), Korean Society of Toxicogenomics and
Toxicoproteomics. Incheon, Korea, Nov 2008. "EPA's ToxCast Program..."
40th Annual Symposium of the Society of Toxicology of Canada, Montreal, Quebec, Dec 2008. "The U.S.
EPA's ToxCast Program..."
Selected Expert Committees/Advisory Panels/Organizing Committees:
1995-present Mentored 9 undergraduate, 7 graduate, and 6 postdoctoral students; one EPA intern.
2005-present Organizing Committee of Microarray Quality Control Project.
2005-present Editorial Board, Toxicological Sciences.
2007-present Editorial Board, Systems Biology in Reproductive Medicine.
2007-present Scientific Advisory Board, EU project CARCINOGENOMICS.
2009-present Scientific Advisory Board, EU project ChemScreen.
Selected Assistance/Advisory Support to the Agency:
2006-present Contracting Officer's Representative (Project Officer) on eight ToxCast contracts.
2006-present Developing ToxRefDB for OPP, OECD and extramural use.
2007-2008 ORD Future of Toxicology Working Group.
2009-present Chemical prioritization effort for OPPTS.
B. SELECTED PUBLICATIONS (selected from a total of 75 peer-reviewed).
Tully DB, JC Luft, JC Rockett, H Ren, JE Schmid, CR Wood, DJ Dix (2005). Reproductive and genomic effects
in testes from mice exposed to the water disinfectant byproduct bromochloroacetic acid. Reproductive
Toxicology 19(3):353-366.
Bao W, JE Schmid, AK Goetz, H Ren, DJ Dix (2005). A database for tracking toxicogenomic samples and
procedures. Reproductive Toxicology 19(3):411-419.
Ostermeier GC, RJ Goodrich, MP Diamond, DJ Dix, SA Krawetz (2005). Towards using stable spermatozoal
RNAs for prognostic assessment of male factor fertility. Fertility and Sterility, 83:1687-94.
Barton HA, Tang J, Sey YM, Stanko JP, Murrell RN, Rockett JC, Dix DJ (2006). Metabolism of myclobutanil
and triadimefon by human and rat cytochrome P450 enzymes and liver microsomes. Xenobiotica 36:793-
806.
Denslow ND, JKColbourne, DJ Dix, JH Freedman, CC Helbing, S Kennedy, PL Wlliams (2006). Selection of
surrogate animal species for comparative toxicogenomics. In: Emerging Molecular and Computational
Approaches for Cross-Species Extrapolations. Eds. W Benson and R Di Giulio. SETAC Press, Florida.
Dix DJ, Gallagher K, Benson WH, Groskinsky BL, McClintock JT, Dearfield KL, Farland WH (2006). A
framework for the use of genomics data at the EPA. Nat Biotechnol 24:1108-11.
Goetz AK, Bao W, Ren H, Schmid JE, Tully DB, Wood C, Rockett JC, Narotsky MG, Sun G, Lambert GR, Thai
SF, Wolf DC, Nesnow S, Dix DJ (2006). Gene expression profiling in the liver of CD-1 mice to characterize
the hepatotoxicity of triazole fungicides. Toxicol Appl Pharmacol 215:274-84.
Kim SJ, Dix DJ, Thompson KE, Murrell RN, Schmid JE, Gallagher JE, Rockett JC (2006). Gene expression in
head hair follicles plucked from men and women. Ann Clin Lab Sci 36:115-26.
Kim YK, Suarez J, Hu Y, McDonough PM, Boer C, Dix DJ, Dillmann WH (2006). Deletion of the inducible 70-
kDa heat shock protein genes in mice impairs cardiac contractile function and calcium handling associated
with hypertrophy. Circulation 113:2589-97.
Rockett JC, Narotsky MG, Thompson KE, Thillainadarajah I, Blystone CR, Goetz AK, Ren H, Best DS, Murrell
RN, Nichols HP, Schmid JE, Wolf DC, Dix DJ (2006). Effect of conazole fungicides on reproductive
development in the female rat. Reprod Toxicol 22:647-58.
PHS 398/2590 (Rev. 09/04, Reissued 4/2006) Page 2
Previous I TOC
-------
Principal Investigator/Program Director (Last, First, Middle): DJX, David J.
Shi L et al. (2006). The MicroArray Quality Control (MAQC) project shows inter- and intraplatform
reproducibility of gene expression measurements. Nat Biotechnol 24:1151-61.
Tully DB, Bao W, Goetz AK, Blystone CR, Ren H, Schmid JE, Strader LF, Wood CR, Best DS, Narotsky MG,
Wolf DC, Rockett JC, Dix DJ (2006). Gene expression profiling in liver and testis of rats to characterize the
toxicity of triazole fungicides. Toxicol Appl Pharmacol 215:260-73.
Cherney DP, Ekman DR, Dix DJ, Collette TW(2007). Raman spectroscopy-based metabolomics for
differentiating exposures to triazole fungicides using rat urine. Anal Chem 79:7324-32.
Dix DJ, Houck KA, Martin MT, Richard AM, Setzer RW, Kavlock RJ (2007). The ToxCast program for
prioritizing toxicity testing of environmental chemicals. Toxicol Sci 95:5-12.
Goetz AK, Ren H, Schmid JE, Blystone CR, Thillainadarajah I, Best DS, Nichols HP, Strader LF, Wolf DC,
Narotsky MG, Rockett JC, Dix DJ (2007). Disruption of testosterone homeostasis as a mode of action for
the reproductive toxicity of triazole fungicides in the male rat. Toxicol Sci 95:227-39.
Kim SJ, Dix DJ, Thompson KE, Murrell RN, Schmid JE, Gallagher JE, Rockett JC 2007). Effects of storage,
RNA extraction, genechip type, and donor sex on gene expression profiling of human whole blood. Clin
Chem 53:1038-45.
Martin MT, Brennan RJ, Hu W, Ayanoglu E, Lau C, Ren H, Wood CR, Gorton JC, Kavlock RJ, Dix DJ (2007).
Toxicogenomic study of triazole fungicides and perfluoroalkyl acids in rat livers predicts toxicity and
categorizes chemicals based on mechanisms of toxicity. Toxicol Sci 97:595-613.
Platts AE, Dix DJ, Chemes HE, Thompson KE, Goodrich R, Rockett JC, Rawe VY, Quintana S, Diamond MP,
Strader LF, Krawetz SA (2007). Success and failure in human spermatogenesis as revealed by
teratozoospermic RNAs. Hum Mol Genet 16:763-73.
Judson R, Richard A, Dix D, Houck K, Elloumi F, Martin M, Cathey T, Transue TR, Spencer R, Wolf M.
(2008). ACToR-Aggregated Computational Toxicology Resource. Toxicol Appl Pharmacol. 15;233(1):7-13.
Kavlock RJ, Ankley G, Blancato J, Breen M, Conolly R, Dix D, Houck K, Hubal E, Judson R, Rabinowitz J,
Richard A, Setzer RW, Shah I, Villeneuve D, Weber E. (2008). Computational toxicology-a state of the
science mini review. Toxicol Sci. 2008 May;103(1):14-27.
Barrier M, Dix DJ, Mirkes PE (2009). Inducible 70 kDa heat shock proteins protect embryos from teratogen-
induced exencephaly: Analysis using Hspa1a/a1b knockout mice. Birth Defects Res A Clin Mol Teratol
28;85(8):732-740.
Goetz AK, Dix DJ (2009). Mode of action for reproductive and hepatic toxicity inferred from a genomic study of
triazole antifungals. Toxicol Sci. 110(2):449-62.
Goetz AK, Dix DJ (2009). Toxicogenomic effects common to triazole antifungals and conserved between rats
and humans. Toxicol Appl Pharmacol. 238(1 ):80-9.
Judson R, Richard A, Dix DJ, Houck K, Martin M, Kavlock R, Dellarco V, Henry T, Holderman T, Sayre P, Tan
S, Carpenter T, Smith E (2009). The toxicity data landscape for environmental chemicals. Environ Health
Perspect. 117(5):685-95.
Knudsen TB, Martin MT, Kavlock RJ, Judson RS, Dix DJ, Singh AV (2009). Profiling the activity of
environmental chemicals in prenatal developmental toxicity studies using the U.S. EPA's ToxRefDB.
Reprod Toxicol. 28(2):209-19.
Martin MT, Judson RS, Reif DM, Kavlock RJ, Dix DJ (2009). Profiling chemicals based on chronic toxicity
results from the U.S. EPA ToxRef Database. Environ Health Perspect. 117(3):392-9.
Martin MT, Mendez E, Corum DG, Judson RS, Kavlock RJ, Rotroff DM, Dix DJ (2009). Profiling the
reproductive toxicity of chemicals from multigeneration studies in the toxicity reference database.
Toxicol Sci. 110(1):181-90.
Goetz AK, JC Rockett, H Ren, I Thillainadarajah, DJ Dix (in press). Inhibition of Rat and Human
Steroidogenesis by Triazole Antifungals. Systems Biol Repro Med. In press.
Houck KA, DJ Dix, RS Judson, RJ Kavlock, J Yang, EL Berg (2009). Profiling Bioactivity of the ToxCast
Chemical Library Using BioMAP Primary Human Cell Systems. J Biomolec Screen. In press.
PHS 398/2590 (Rev. 09/04, Reissued 4/2006) Page 3
Previous I TOC
-------
Principal Investigator/Program Director (Last, First, Middle): Egeghy, Peter Paul
BIOGRAPHICAL SKETCH
NAME
Peter Paul Egeghy
eRA COMMONS USER NAME
POSITION TITLE
Research Environmental Health Scientist
EDUCATION/TRAINING (Begin with baccalaureate or other initial professional education, such asnursing, and include postdoctoral training.)
INSTITUTION AND LOCATION
University of California, Berkeley
University of California, Berkeley
California State University, Northridge
University of North Carolina, Chapel Hill
DEGREE
(if applicable)
B.A.
M.P.H.
M.S.
Ph.D.
YEAR(s)
1990
1993
1994
2001
FIELD OF STUDY
Physical Environmental
Sciences
Environmental Health
Industrial Hygiene
Environmental Sciences
& Engineering
A. POSITIONS and HONORS
Research and Professional Experience:
2009-Present Research Fellow (detail), National Center for Computational Toxicology, US EPA, Research
Triangle Park, NC
2004-Present Research Environmental Health Scientist, National Exposure Research Laboratory, US EPA,
Research Triangle Park, NC
2001-2004 Environmental Health Scientist (post-doctoral fellow), National Exposure Research
Laboratory, US EPA, Las Vegas, NV
1994-1996 Assistant Environmental Health and Safety Manager, Olive View-UCLA Medical Center, Los
Angeles, CA
Professional Societies and Affiliations:
2004-present Member, International Society of Exposure Science (ISES)
2006-present Member, International Association for Breath Research (IABR)
2006-2009 Member, Society for Risk Analysis (SRA)
Honors and Awards:
2009 National Exposure Research Laboratory Special Achievement Award: Technical Support to
the Office of Pesticide Programs and the Health Effects Division
2008 USEPA Level I Scientific and Technological Achievement Award (STAA): Identifying Important
Sources, Pathways, and Routes of Children's Exposures to Pesticides in Their Environments
2008 USEPA Level II Scientific and Technological Achievement Award (STAA): Improved
Understanding of Children's Exposure to Pesticides in the Residential Environment
2007 Children's Environmental Health Excellence Award: Science Achievement.
2007 National Exposure Research Laboratory Special Achievement Award: Children's Exposure
Team Support of the Agency's Mission
2007 USEPA Superior Accomplishment Recognition Awards (Team): for contributions to the
Moncure Air Quality Screening Study
2006 USEPA Superior Accomplishment Recognition Awards (Team): for contributions to the
Analysis of Children's Exposure Factors Data.
1999-2001 NIEHS Biostatistics Traineeship Award
Selected Expert Committees/Advisory Panels/Organizing Committees:
2009 Organizing Committee Member, 2009 International Society of Exposure Science Conference,
Minneapolis, MN
PHS 398/2590 (Rev. 09/04, Reissued 4/2006)
Page J
Previous
TOC
-------
Principal Investigator/Program Director (Last, First, Middle): Egeghy, Peter Paul
2008 Expert Panel Member, Novel Approaches for Assessing Exposure for School-Aged Children in
Longitudinal Studies, US EPA STAR Grant Announcement, Washington, DC
2007-2008 Invited Author, USEPA's Scientific and Ethical Approaches for Observational Exposure
Studies, Research Triangle Park, NC
2007 Expert Panel, United States Army Research Institute of Environmental Medicine Workshop on
Research Issues Related to JP8 Exposure Assessment, Natick, MA
2006-2007 Advisory Board Member, North Carolina Central University Environmental Risk and Impact in
Communities of Color (ERICC) Initiative, Durham, NC
2005 Co-Chair, USEPA/NERL Workshop for the Analysis of Children's Exposure Measurements
Data, Research Triangle Park, NC
2004 Expert Panel, State of EPA Mercury Science Teleconference Workshop Series
2003 Committee Member, USEPA HEASD Authorship Guidance Council, RTP, NC
2002-2004 Expert Panel, USEPA/CDC NHANES Biomarker Interpretation Committee
2001 Section Lead, Biological Monitoring in Exposure Assessment, ISEA Exposure Assessment
and Epidemiology Workshop, Charleston, SC
B. SELECTED PUBLICATIONS
Stout DM 2nd, Bradham KD, Egeghy PP, Jones PA, Croghan CW, Ashley PA, Pinzer E, Friedman W,
Brinkman MC, Nishioka MG, Cox DC. (2009) American Healthy Homes Survey: a national study of
residential pesticides measured from floor wipes. Environ Sci Technol. 43(12):4294-300.
Cohen Hubal E, Nishioka M, Ivancic W, Morara M, Egeghy P. (2008) Comparing surface residue transfer
efficiencies to hands using polar and non-polar fluorescent tracers. Environ Sci Technol. 42(3):934-9.
Lin YS, Egeghy PP, Rappaport SM. (2008) Relationships between levels of volatile organic compounds in air
and blood from the general population. J Expo Sci Environ Epidemiol. 18(4):421-9.
Morgan MK, Sheldon LS, Thomas KW, Egeghy PP, Croghan CW, Jones PA, Chuang JC, Wlson NK. (2008)
Adult and children's exposure to 2,4-D from multiple sources and pathways. J Expo Sci Environ Epidemiol.
18(5):486-94.
Tulve NS, Egeghy PP, Fortmann RC, Whitaker DA, Nishioka MG, Naeher LP, Hilliard A. (2008) Multimedia
measurements and activity patterns in an observational pilot study of nine young children. J Expo Sci
Environ Epidemiol. 18(1):31-44.
Egeghy PP, Sheldon LS, Fortmann RC, Stout II DM, Tulve NS, Cohen Hubal EA, Melnyk LJ, Morgan MK,
Jones PA, Whitaker DA, Croghan CW, Coan A. (2007) Important Exposure Factors for Children: An
Analysis of Laboratory and Observational Field Data Characterizing Cumulative Exposure to Pesticides.
U.S. Environmental Protection Agency, Washington, DC, EPA/600/R-07/013.
Nakayama S, Strynar MJ, Helfant L, Egeghy P, Ye X, Lindstrom AB. (2007) Perfluorinated compounds in the
Cape Fear Drainage Basin in North Carolina. Environ Sci Technol. 41(15):5271-6.
Kim D, Andersen ME, Chao YC, Egeghy PP, Rappaport SM, Nylander-French LA. (2007) PBTK modeling
demonstrates contribution of dermal and inhalation exposure components to end-exhaled breath
concentrations of naphthalene. Environ Health Perspect. 115(6):894-901.
Kirrane E, Loomis D, Egeghy P, Nylander-French L. (2007) Personal exposure to benzene from fuel emissions
among commercial fishers: comparison of two-stroke, four-stroke and diesel engines. J Expo Sci Environ
Epidemiol. 17(2):151-8.
Tulve NS, Driver J, Egeghy PP, Evans J, Fortmann RC, Kissel JC, McMillan N, Melnyk LJ, Morgan MK, Starr
JM, Stout DM, Strynar MJ. (2006) The U.S. EPA National Exposure Research Laboratory's (NERL's)
Workshop on the Analysis of Children's Measurement Data. US EPA, Washington, DC, EPA/600/R-06/026.
Chao YC, Kupper LL, Serdar B, Egeghy PP, Rappaport SM, Nylander-French LA. (2006) Dermal exposure to
jet fuel JP-8 significantly contributes to the production of urinary naphthols in fuel-cell maintenance
workers. Environ Health Perspect. 114(2):182-5.
Cohen Hubal EA, Egeghy PP, Leovic KW, Akland GG. (2006) Measuring potential dermal transfer of a
pesticide to children in a child care center. Environ Health Perspect. 114(2):264-9.
Egeghy PP, Quackenboss JJ, Catlin S, Ryan PB. (2005) Determinants of temporal variability in NHEXAS-
Maryland environmental concentrations, exposures, and biomarkers. J Expo Anal Environ Epidemiol.
15(5):388-97.
PHS 398/2590 (Rev. 09/04, Reissued 4/2006) Page 2
Previous I TOC
-------
Principal Investigator/Program Director (Last, First, Middle): Egeghy, Peter Paul
Serdar B, Egeghy PP, Gibson R, Rappaport SM. (2004) Dose-dependent production of urinary naphthols
among workers exposed to jet fuel (JP-8). Am J Ind Med. 46(3):234-44.
Egeghy PP, Hauf-Cabalo L, Gibson R, Rappaport SM. (2003) Benzene and naphthalene in air and breath as
indicators of exposure to jet fuel. Occup Environ Med. 60(12):969-76.
Serdar B, Egeghy PP, Waidyanatha S, Gibson R, Rappaport SM. (2003) Urinary biomarkers of exposure to jet
fuel (JP-8). Environ Health Perspect. 111(14): 1760-4.
Rhodes AG, LeMasters GK, Lockey JE, Smith JW, Yiin JH, Egeghy P, Gibson R. (2003) The effects of jet fuel
on immune cells of fuel system maintenance workers. J Occup Environ Med. 45(1):79-86.
Egeghy PP, Nylander-French L, Gwin KK, Hertz-Picciotto I, Rappaport SM. (2002) Self-collected breath
sampling for monitoring low-level benzene exposures among automobile mechanics. Ann Occup Hyg.
46(5):489-500.
Egeghy PP, Tornero-Velez R, Rappaport SM. (2000) Environmental and biological monitoring of benzene
during self-service automobile refueling. Environ Health Perspect. 108(12): 1195-202.
C. SELECTED PRESENTATIONS
Egeghy PP. Exposure and EPA's National Exposure Research Laboratory. International Council of Chemical
Associations (ICCA) Long-Range Research Initiative (LRI) Workshop. Charleston, SC. June 2009 (oral
presentation)
Egeghy PP, Tulve NS, Adetona O; Naeher LP. Using Pesticide Screening Questions to Identify the More
Highly Exposed Participants in a Larger Cohort. International Society for Environmental Epidemiology
(ISEE) and International Society of Exposure Analysis (ISEA) Joint Annual Conference, Pasadena,
California, October, 2008 (oral presentation)
Egeghy PP, Quackenboss JJ. Within- and Between-Person Variation in Environmental Concentrations of
Metals, PAHS, and Pesticides Measured in NHEXAS-Maryland. Conference Internationale d'Epidemiologie
et d'Exposition Environnementales, Paris, September 2006 (poster presentation)
Egeghy PP, Stout II DM, Tornero-Velez R, Furtaw Jr EJ. Prediction of Airborne Pesticide Distributional
Parameters by Physiochemical Properties. Conference Internationale d'Epidemiologie et d'Exposition
Environnementales, Paris, September 2006 (poster presentation)
Egeghy PP, Thomas KW. Estimates of Age-Specific Urinary Excretion Rates for Creatinine among Children.
Conference Internationale d'Epidemiologie et d'Exposition Environnementales, Paris, September 2006
(oral presentation)
Egeghy PP, Tulve N, Stout II DM, Morgan M, Melnyk L, Cohen Hubal E, Fortmann R, Sheldon L. Identifying
Important Factors Influencing Children's Exposures to Pesticides. EPA Science Forum, May 2006 (poster
presentation)
Egeghy PP, Morgan MK, Croghan CW, Sheldon LS. Reliability of Biomarkers of Pesticide Exposure among
Children and Adults in CTEPP-Ohio. International Society of Exposure Analysis (ISEA) 15th Annual
Conference, Tucson, AZ, October 2005 (oral presentation)
Egeghy PP, Catlin S, Ryan PB, Quackenboss JJ. Determinants of Residential Lead Exposure. International
Society of Exposure Analysis, Stresa, Italy, September, 2003 (oral presentation)
Egeghy PP, Quackenboss JJ, Ozkaynak AH, Ryan PB. Alternative Exposure Measurements to Improve
Epidemiological Study Designs: Determinants of Temporal Variability in Environmental Concentrations and
Biomarkers. National Children's Study Meetings, Baltimore, MD, December, 2002 (poster presentation)
Egeghy PP. Mixed Models Analysis of Urbanization Level on Chlorpyrifos Exposure. Presented at International
Society of Exposure Analysis 2002 Conference, Vancouver, Canada, August, 2002 (poster presentation)
D. WORKSHOPS
Connecting Innovations in Biological Exposure and Risk Sciences: Better Information for Better Decisions.
International Council of Chemical Associations (ICCA) Long-Range Research Initiative (LRI), Charleston,
SC. June 2009
Research Approaches to Assessing Public Health Impacts of Risk Management Decisions Workshop, US
EPA, Research Triangle Park, NC, January 2008
Biomonitoring Workshop, International Life Sciences Institute (ILSI) Health and Environmental Sciences
PHS 398/2590 (Rev. 09/04, Reissued 4/2006) Page 3
Previous I TOC
-------
Principal Investigator/Program Director (Last, First, Middle): Egeghy, Peter Paul
Institute (HESI), Research Triangle Park, NC, September 2004
Exposure Research Needs for Addressing Cumulative Risk, US EPA, Research Triangle Park, NC, February
2004
Human Exposure and Dose Modeling Workshop, US EPA, Research Triangle Park, NC, October 2002
Exposure Assessment and Epidemiology Workshop, International Society of Exposure Analysis, Charleston,
SC, November 2001
E. CONTINUING EDUCATION
"Facilitative Leadership Seminar," Western Management Development Center, Denver, CO, March 2007
"Exposure Assessment for Environmental Chemicals using Biomonitoring," Conference Internationale
d'Epidemiologie et d'Exposition Environnementales, Paris, September 2006
"Team Building Team Leadership Seminar," Western Management Development Center, Denver, CO, August
2003
"Igniting Leadership at All Levels," US EPA Office of Research and Development Leadership Summit,
Baltimore, MD, January 2003
PHS 398/2590 (Rev. 09/04, Reissued 4/2006) Page 4
Previous I TOC
-------
Principal Investigator/Program Director (Last, First, Middle): Gallagher, Jane E.
BIOGRAPHICAL SKETCH
Provide the following information for the key personnel and other significant contributors in the order listed on Form Page 2.
Follow this format for each person. DO NOT EXCEED FOUR PAGES.
NAME
Jane E. Gallagher
eRA COMMONS USER NAME
POSITION TITLE
Research Scientist
EDUCATION/TRAINING (Begin with baccalaureate or other initial professional education, such as nursing, and include postdoctoral training.)
INSTITUTION AND LOCATION
State University of New York at Potsdam
Purdue University West Lafayette IN
University of Utah at Salt Lake City
University of North Carolina at Chapel Hill NC
DEGREE
(if applicable)
B.S.
M.S.
M.S.
Ph.D.
YEAR(s)
1976
1981
1982
1986
FIELD OF STUDY
Chemistry
Civil Engineer
Chemical Engineer
Environmental
Toxicology
A. POSITIONS and HONORS
Research and Professional Experience:
1987-Present Health Research Scientist, Epidemiology Biomarker, Branch Human Studies Division, NHEERL,
USEPA RTP, NC
1999-Present Adjunct Asst. Professor- UNC School of Public Health- Environmental Sciences and
Engineering
1986-1987 National Research Council (NRC) - postdoctoral fellow. Genetic Toxicology Division US
EPA, Research Triangle Park, NC
1982-1986 Biologist, Laboratory of Pulmonary Pathobiology, National Institute Environmental Health
Sciences (NIEHS RTP, NC).
1981-1982 NIOSH traineeship Rocky Mountain Center for Occupational Medicine, Salt Lake City, Utah
Selected Awards and Honors (recent):
Civil Engineering Alumni Achievement Award- Purdue University -2006
Science and Technology Achievement Award (3) - Most recently "for contributions leading to a better
understanding of PM health effects using animal and complementary in vitro human cell test systems" 2005.
Invited Lectures/Symposia (recent):
Mechanistic Indicators of Childhood Asthma (MICA) Study, 17th Annual Conference of the International Society
of Exposure Analysis, Durham/Research Triangle Park, NC. October 2007.
Mechanistic Indicators of Childhood Asthma (MICA) -Integrating Environmental, Clinical and Susceptibility
Markers to Improve the Impact of Human Air Pollution Studies. Public Health Implications of
Biomonitoring September 24-25 USA EPA 2007.
Mechanistic Indicators of Childhood Asthma-A Computational Toxicology Study" US/Canadian Research
Studies Workshop. Detroit Michigan October 21, 2005.
Integrating Biomonitoring Data into Risk Assessment 31st Annual Meeting of The Toxicology Forum at the
Given Institute Aspen, Colorado July10-14, 2005.
Office of Children's Health Protection, Washington, DC "Mechanistic Indicators of Childhood Asthma",2005.
Society of Toxicology Annual Meeting- National Children Study symposium "Validation of non-invasive
Biological Sources for Application in Environmental Epidemiology Studies" Baltimore, MD. 2004.
Environmental Carcinogenesis Division- "Application of biomarkers of arsenic exposure/effect in Fallen,
Nevada. Research Triangle Park, NC - 2005.
Office of Research and Development Annual National Children Studies progress meeting "Non-invasive
Samples for Gene/Environmental Studies" Research Triangle Park, NC 2002, 2003.
Previous
TOC
-------
Principal Investigator/Program Director (Last, First, Middle): Gallagher, Jane E.
Assistance/leadership provided to the scientific community:
• Editorial Board: Mutation Research 2000 - present.
• Journal Reviewer: Tox. Appl. Pharm., Tox. Sci., Inhal. Tox., Toxicol., Environ. Health, EHP.
• Adjunct Asst. Professor Environmental Sciences and Engineering, UNC Chapel Hill 2002-present.
• Research Advisor: Scott Rhoney, Master student School of Public Health UNC Chapel Hill NC
• Co-Advisor mentor for a community based air toxics emissions project: NC State
• Reviewer Biomarkers RFAs for National Center for Environmental Research
• Reviewer for World Health Organization- NC^-PAH Health Effects document 2003.
Assistance/leadership provided to the agency:
• Development of shared Access Data base for the integration of multifactorial environmental health data for
shared NERL NHEEL and NCCT users. 2008.
• Panel member: "Integrating Biomonitoring Data into Risk Assessment 31 st Annual Meeting of The
Toxicology Forum at the Given Institute Aspen, Colorado JulylO-14, 2005.
• Topic Lead: Office of Research and Development (ORD) Human Health Risk Review- Prepared Board of
Scientific review materials and coordinated cross-divisional session on "Oxidative Stress and Human
Health"- Research Triangle Park, NC Feb 2005.
• Briefing for EPA's Office of Children's Health and Protection "Mechanistic Indicators of Childhood
Asthma - Washington, DC 2005.
• Principle Investigator- Successful application for an Office of Research and Development NCCT
competitive grant proposal (0.9 Million) involving investigators from National Exposure Research Lab
Cincinnati and EPA's National Human Exposure and Effects Lab 2005.
• Assistance to Office of Science Policy, US EPA- Assisted Office of Science Policy in the identification of
EPA ongoing asthma research for EPA's Computer Retrieval of Information on Scientific Projects
Initiative- 2004.
B. PUBLICATIONS: (2007-2009)
Kim SJ, Dix DJ, Thompson KE, Murrell RN, Schmid JE, Gallagher JE, Rockett JC.
Effects of storage, RNA extraction, genechip type, and donor sex on gene expression profiling of human whole
blood . Clin Chem. Jun;53(6): 1038-45. (2007)
Vesper, S.,McKinstry C., Haugland., R., Neas, L., Hudgens, E., Heidenfelder, B., and Gallagher J.
Environmental Relative Moldiness Index (ERMIsm) as a Tool to Identify Mold Related Risk Factors for
Childhood Asthma Sci Total Environ. May 1;394(1): 192-6 (2008)
Johnson M, Hudgens E, Williams R, Andrews G, Neas L, Gallagher J, Ozkaynak H. "A Participant-Based
Approach to Indoor/Outdoor Air Monitoring in Community Health Studies" Journal of Exposure Science and
Environmental Epidemiology. (2008), 1-10 (2008).
Cohen HubalE, Richards A., Shah I, Edwards S, Gallagher J, Kavlock R, Blancato, J Exposure Science and
the US EPA National Center for Computational Toxicology J Expo Sci Environ Epidemiol. November (2008).
Heidenfelder B,. ReifD, Harkema, JR, Cohen Hubal E, Hudgens,E. Bramble L G. Wagner G, Harkema JR,
Morishita M, Keeler G , Edwards,SW and Gallagher J. Comparative Microrarray Analysis and Pulmonary
Changes in Brown Norway Rats Exposed to Ovalbumin and concentrated Air Particulates Tox Sci. volume
108 2009 March 2 (2009)
Previous I TOC
-------
Principal Investigator/Program Director (Last, First, Middle): Gallagher, Jane E.
Heidenfelder B, Johnson M, Hudgens E, Inmon J, Hamilton R, Neas L, and Gallagher J, Increased plasma
reactive oxidant levels and their relationship to blood cells, total IgE, and allergen-specific IgE in asthmatic
children Journal of Asthma accepted (2009)
Williams AH, Gallagher JE, Hudgens E, Johnson MM, Mukerjee S, Ozkaynak H, Neas LN. EPA
Observational studies of children's respiratory health in Detroit and Dearborn, Michigan. Proceedings of
AWMA 102nJune 16-19; Detroit, Michigan.(2009)
J. E Gallagher, E A Cohen Hubal, S.W.Edwards Invited book Chapter "Biomarkers of Environmental
Exposure" "Biomarkers of toxicity: A New Era in Medicine Editors Vishal S. Vaidya and Joseph V. Bonventre
Publisher: John Wiley and Sons, Inc. October 1, (2009)
Markey M. Johnson, Ron Williams, Zhihua Fan, Lin, Edward Hudgens, Jane Gallagher, Alan Vette, Lucas
Neas, Haluk Ozkaynak Indoor and outdoor concentrations of nitrogen dioxide, volatile organic compounds, and
polycyclic aromatic hydrocarbons among MICA-Air households in Detroit, Michigan submitted AWMA
(2009)
Gallagher, J Reif, D; Heidenfelder, B Neas, L; Hudgens, E Williams, A Inmon, J; Rhoney, S, Andrews G.,
Johnson, M Ozkaynak, H; Edwards, S, Cohen-Hubal, E Mechanistic Indicators of Childhood asthma ( MICA);
A systems biology approach for the integration of multifactorial environmental health data submitted: Journal
of Exposure Science and Environmental Epidemiology (2009)
In preparation
David M. Reif, Jane E. Gallagher, Brooke L. Heidenfelder, Ed E. Hudgens, Wendell Jones, ClarLynda
Williams-DeVane, Lucas M. Neas, Elaine A. Cohen Hubal, Stephen W. Edwards Elucidating Asthma
Phenotypes via Integrated Analysis of Blood Gene Expression Data with Demographic and Clinical Information
(Nature Genetics) 2009
David M. Reif*, ClarLynda Williams-DeVane*, Elaine A. Cohen Hubal, Wendell Jones, Ed E. Hudgens,
Brooke L. Heidenfelder, Lucas M. Neas, Jane E. Gallagher, Stephen W. Edwards
* Authors contributed equally. Systems Modeling of Gene Expression, Demographic and Clinical Data to
Determine Disease Endotypes PLOS Comp Bio 2009
Previous I TOC
-------
Principal Investigator/Program Director (Last, First, Middle): Houck, Keith, A.
BIOGRAPHICAL SKETCH
Provide the following information for the key personnel and other significant contributors in the order listed on Form Page 2.
Follow this format for each person. DO NOT EXCEED FOUR PAGES.
NAME
Keith A. Houck
eRA COMMONS USER NAME
khouck
POSITION TITLE
Toxicologist
EDUCATION/TRAINING (Begin with baccalaureate or other initial professional education, such as nursing, and include postdoctoral training.)
INSTITUTION AND LOCATION
DEGREE
(if applicable)
YEAR(s)
FIELD OF STUDY
Guilford College, Greensboro, NC
University of North Carolina, Chapel Hill
Duke University, Durham, NC
Genentech, Inc.
B.S.
M.S.
Ph.D.
Post-doc
1980
1982
1989
1989-1992
Biology
Chemistry
Pathology/Toxicology
Molecular and Cellular Biology
A.
POSITIONS and HONORS
Research and Professional Experience:
2006-present Toxicologist, National Center for Computational Toxicology, USEPA, NC
2006-present Joint Research Committee Member/Consultant, Cystic Fibrosis Foundation Therapeutics
2005-Present Lead Generation Team/Consultant, NINDS Spinal Muscular Atrophy Project
2005 Independent Consultant, Dept. Cell and Tissues Engineering, Becton-Dickinson, RTP, NC
1994-2004 Research Advisor, Lilly Research Laboratories, Eli Lilly & Co., RTP, NC.
1992-1994 Senior Biologist, Sphinx Pharmaceuticals, RTP, NC.
1989-1992 Postdoctoral Fellow, Molecular Biology Department, Genentech, Inc.
1982-1985 Senior Research Analyst, Pathology Department, Duke University
Professional Societies and Affiliations:
1992-present American Association for the Advancement of Science
2001-present Society of Biomolecular Sciences (Member: Conference Committee)
2007-present Society of Toxicology (Specialty Section: Nanotoxicology)
Honors and Awards:
2000 Changing the World Award, Eli Lilly & Co.
2003 Best Paper Award, Society of Biomolecular Sciences
Selected invitations at National & International Symposia:
U.S. EPA ToxCast Data Analysis Summit. RTP, NC, May 14-15, 2009, "Characteristics of the ToxCast In Vitro
Datasets from Biochemical and Cellular Assays".
15th Annual Conference of the Society of Biomolecular Sciences, Lille, France, April 26-30, 2009, "Use of
Primary Human Cell Systems for Creating Predictive Toxicology Profiles".
GlobalChem Conference 2009, Baltimore, MD, April 8, 2009, "Evaluation of the ToxCast Suite of Cellular and
Molecular Assays for Prediction of In Vivo Toxicity".
The Committee on Toxicity of Chemicals in Food, Consumer Products and the Environment (COT) 'toxicology
for the21st century', Manor Hotel, Meriden, UK February 11, 2009, "The US EPA's ToxCast Program for
the Prioritization and Prediction of Environmental Chemical Toxicity".
Predictive Human Toxicity and ADME/TOX Studies, Brussels, Belgium, January 20-21, 2009, "Prediction of
Toxicity Potential using In Vitro Screening Assays from the EPA's ToxCast Program".
CASCADE Workshop on Synergies from Global Chemical Screening Programs (US-EPA/OECD) Oct. 1, 2007,
"EPA's ToxCast™ Program for Predicting Hazard and Prioritizing Toxicity Testing of Environmental
Chemicals".
PHS 398/2590 (Rev. 09/04, Reissued 4/2006)
Page J
Previous
TOC
-------
Principal Investigator/Program Director (Last, First, Middle): Houck, Keith, A.
13th Annual Conference of the Society of Biomolecular Screening, Montreal, Canada, April 15-19 2007,
"Toxicity Profiling Using High-Throughput and High-Content Technologies".
Fall Symposium of the National Capital Area Chapter Society of Toxicology, Bethesda, MD, Dec. 11, 2006,
"Environmental chemical hazard prediction by high-throughput screening and genomics approaches in the
ToxCast program of the US Environmental Protection Agency".
Selected Expert Committees/Advisory Panels/Organizing Committees:
2010 Co-chair, Society of Biomolecular Sciences Regional Meeting, RTP, NC.
2009 ECETOC Workshop on Guidance on Identifying Endocrine Disrupting Effects, Barcelona, Spain.
2009 NIH Nanomaterial Grand Opportunity Grant Review Panel
2009 NIH ARRA Challenge Grant Review
2008-present Clinical Chemistry and Clinical Toxicology Devices Panel of the Medical Devices Advisory
Committee, Center for Devices and Radiological Health, Food and Drug Administration
2008 NIH Roadmap RFA on Assay Development for High Throughput Molecular Screening Grant
Review Panel
2007 Co-chair, symposium, "Toxicity Profiling Using High-Throughput and High-Content
Technologies" 13th Annual Conference of the Society of Biomolecular Screening, Montreal,
Canada, April 15-19 2007.
2007-present Lecturer, North Carolina Central University, The Brite Center, Dept. of Pharmaceutical Sciences
2006 NIH Roadmap RFA on Assay Development for High Throughput Molecular Screening Grant
Review Panel
Selected Assistance/Advisory Support to the Agency:
2006-present TOCOR, contracts for ToxCast.
2002-2003 Genomics Coordinator, NHEERL Genomics and Proteomics Committee.
2002-2003 Principal Investigator, CRADA with Duke University to produce DNA microarrays for NHEERL.
2003-2006 Project Officer, GeneChip processing contract for ORD/NHEERL.
2004-2005 Principal Investigator, CRADA with Affymetrix to develop in vitro toxicogenomics.
2004-2005 Project Officer, contract for toxicogenomics with Iconix Pharmaceuticals for ORD.
2004-present Co-Chair, Genomics Task Force Data Analysis Workgroup of EPA Science Policy Council.
2006-present Contracting Officer's Representative (Project Officer) on eight ToxCast contracts.
2003-present Consulting with Office of Pesticide Programs on conazole reproductive toxicity.
2006-present Developing ToxRefDB for OPP use
2007 ORD Future of Toxicology Working Group
B. SELECTED PUBLICATIONS (selected from a total of 64 peer-reviewed).
Ostermeier GC, DJ Dix, D Miller, P Khatri, SA Krawetz. Spermatozoal RNA profiles of normal fertile men.
Lancet, 360, 772-777, 2002.
Rockett JC, RJ Kavlock, C Lambright, L Parks, JE Schmid, VS Wlson, C Wood, DJ Dix. DNA arrays to monitor
gene expression in rat blood and uterus following 17-beta-estradiol exposure - biomonitoring
environmental effects using surrogate tissues. Toxicological Sciences, 69, 49-59, 2002.
Richburg, JH, Johnson, K, Schoenfeld, HA, Meistrich, ML and Dix, DJ (2002). Defining the cellular and
molecular mechanisms of toxicant action in the testis. Toxicology Letters 135: 167-183.
Hampton CR, A Shimamoto, CL Rothnie, J Griscavage-Ennis, A Chong, DJ Dix, ED Verrier, TH Pohlman
(2003). HSP70.1 and -70.3 are required for late-phase protection induced by ischemic preconditioning of
mouse hearts.Am J Physiol Heart Circ Physiol 285: H866-H874.
Schmitt E, A Parcellier, S Gurbuxani, C Cande, A Hammann, M Celia Morales, CR Hunt, DJ Dix, RT Kroemer,
F Giordanetto, M Jaattela, JM Penninger, A Pance, G Kroemer, C Garrido (2003). Chemosensitization by
a non-apoptogenic heat shock protein 70-binding apoptosis inducing factor mutant. Cancer Research,
63(23): 8233-
Hunt CR, DJ Dix, GG Sharma, RK Pandita, A Gupta, M Funk, TK Pandita (2004). Genomic instability and
enhanced radiosensitivity in Hsp70.1/3-deficient mice. Molecular and Cellular Biology, 24(2):899-911.
PHS 398/2590 (Rev. 09/04, Reissued 4/2006) Page 2
Previous I TOC
-------
Principal Investigator/Program Director (Last, First, Middle): Houck, Keith, A.
Rockett JC, ME Burczynski, AJ Fornace Jr, PC Herrmann, SA Krawetz, DJ Dix (2004). Surrogate tissue
analysis: monitoring toxicant exposure and health status of inaccessible tissues through the analysis of
accessible tissues and cells. Toxicology and Applied Pharmacology 194:189-199.
Rockett JC, P Patrizio, JE Schmid, NB Hecht, DJ Dix (2004). Gene expression patterns associated with
infertility in humans and rodent models. Mutation Research, 549:225-240.
Tully DB, JC Luft, JC Rockett, H Ren, JE Schmid, CR Wood, DJ Dix (2005). Reproductive and genomic effects
in testes from mice exposed to the water disinfectant byproduct bromochloroacetic acid. Reproductive
Toxicology 19(3):353-366.
Bao W, JE Schmid, AK Goetz, H Ren, DJ Dix (2005). A database for tracking toxicogenomic samples and
procedures. Reproductive Toxicology 19(3):411-419.
Ostermeier GC, RJ Goodrich, MP Diamond, DJ Dix, SA Krawetz (2005). Towards using stable spermatozoal
RNAs for prognostic assessment of male factor fertility. Fertility and Sterility, 83:1687-94.
Barton HA, Tang J, Sey YM, Stanko JP, Murrell RN, Rockett JC, Dix DJ (2006). Metabolism of myclobutanil
and triadimefon by human and rat cytochrome P450 enzymes and liver microsomes. Xenobiotica 36:793-
806.
Denslow ND, JKColbourne, DJ Dix, JH Freedman, CC Helbing, S Kennedy, PL Williams (2006). Selection of
surrogate animal species for comparative toxicogenomics. In: Emerging Molecular and Computational
Approaches for Cross-Species Extrapolations. Eds. W Benson and R Di Giulio. SETAC Press, Florida.
Dix DJ, Gallagher K, Benson WH, Groskinsky BL, McClintock JT, Dearfield KL, Farland WH (2006). A
framework for the use of genomics data at the EPA. Nat Biotechnol 24:1108-11.
Goetz AK, Bao W, Ren H, Schmid JE, Tully DB, Wood C, Rockett JC, Narotsky MG, Sun G, Lambert GR, Thai
SF, Wolf DC, Nesnow S, Dix DJ (2006). Gene expression profiling in the liver of CD-1 mice to characterize
the hepatotoxicity of triazole fungicides. Toxicol Appl Pharmacol 215:274-84.
Kim SJ, Dix DJ, Thompson KE, Murrell RN, Schmid JE, Gallagher JE, Rockett JC (2006). Gene expression in
head hair follicles plucked from men and women. Ann Clin Lab Sci 36:115-26.
Kim YK, Suarez J, Hu Y, McDonough PM, Boer C, Dix DJ, Dillmann WH (2006). Deletion of the inducible 70-
kDa heat shock protein genes in mice impairs cardiac contractile function and calcium handling associated
with hypertrophy. Circulation 113:2589-97.
Rockett JC, Narotsky MG, Thompson KE, Thillainadarajah I, Blystone CR, Goetz AK, Ren H, Best DS, Murrell
RN, Nichols HP, Schmid JE, Wolf DC, Dix DJ (2006). Effect of conazole fungicides on reproductive
development in the female rat. Reprod Toxicol 22:647-58.
Shi L et al. (2006). The MicroArray Quality Control (MAQC) project shows inter- and intraplatform
reproducibility of gene expression measurements. Nat Biotechnol 24:1151-61.
Tully DB, Bao W, Goetz AK, Blystone CR, Ren H, Schmid JE, Strader LF, Wood CR, Best DS, Narotsky MG,
Wolf DC, Rockett JC, Dix DJ (2006). Gene expression profiling in liver and testis of rats to characterize the
toxicity of triazole fungicides. Toxicol Appl Pharmacol 215:260-73.
Cherney DP, Ekman DR, Dix DJ, Collette TW(2007). Raman spectroscopy-based metabolomics for
differentiating exposures to triazole fungicides using rat urine. Anal Chem 79:7324-32.
Dix DJ, Houck KA, Martin MT, Richard AM, Setzer RW, Kavlock RJ (2007). The ToxCast program for
prioritizing toxicity testing of environmental chemicals. Toxicol Sci 95:5-12.
Goetz AK, Ren H, Schmid JE, Blystone CR, Thillainadarajah I, Best DS, Nichols HP, Strader LF, Wolf DC,
Narotsky MG, Rockett JC, Dix DJ (2007). Disruption of testosterone homeostasis as a mode of action for
the reproductive toxicity of triazole fungicides in the male rat. Toxicol Sci 95:227-39.
Kim SJ, Dix DJ, Thompson KE, Murrell RN, Schmid JE, Gallagher JE, Rockett JC 2007). Effects of storage,
RNA extraction, genechip type, and donor sex on gene expression profiling of human whole blood. Clin
Chem 53:1038-45.
Martin MT, Brennan RJ, Hu W, Ayanoglu E, Lau C, Ren H, Wood CR, Gorton JC, Kavlock RJ, Dix DJ (2007).
Toxicogenomic study of triazole fungicides and perfluoroalkyl acids in rat livers predicts toxicity and
categorizes chemicals based on mechanisms of toxicity. Toxicol Sci 97:595-613.
Platts AE, Dix DJ, Chemes HE, Thompson KE, Goodrich R, Rockett JC, Rawe VY, Quintana S, Diamond MP,
Strader LF, Krawetz SA (2007). Success and failure in human spermatogenesis as revealed by
teratozoospermic RNAs. Hum Mol Genet 16:763-73.
PHS 398/2590 (Rev. 09/04, Reissued 4/2006) Page 3
Previous I TOC
-------
Principal Investigator/Program Director (Last, First, Middle): Hunter, Edward S.
BIOGRAPHICAL SKETCH
NAME
Edward S, Hunter
eRA COMMONS USER
Hunter
,111 (Sid)
NAME
POSITION TITLE
Research Toxicologist
EDUCATION/TRAINING (Begin with baccalaureate or other initial professional education, such as nursing, and include postdoctoral training.)
INSTITUTION AND LOCATION
Hampden-Sydney College, Virginia
Old Dominion University, Virginia
University of North Carolina at Chapel Hill
DEGREE
(if applicable)
B.S.
M.S.
Ph.D.
YEAR(s)
1975-80
1980-83
1982-86
FIELD OF STUDY
Chemistry
Toxicology
Anatomy/Embryology
A. POSITIONS and HONORS
Research and Professional Experience:
May 2009 - Acting Chief Systems Biology Branch, ISTD, NHEERL, ORD, US EPA, RTF, NC
May 2009 - Research Toxicologist, Systems Biology Branch, ISTD, NHEERL, ORD, US EPA, RTF, NC
2008 - 2009 Acting Director Reproductive Toxicology Division, NHEERL, ORD, US EPA, RTF, NC
2005 - 2007: Acting Chief Gamete and Early Embryo Biology Branch, RTD, NHEERL, US EPA, RTF, NC
1993 - 2009: Research Toxicologist, Developmental Biology Branch, RTD, NHEERL, US EPA, RTF, NC
1990 - 1993: Toxicologist, Developmental and Reproductive Toxicology Group, NTP, NIEHS, RTF, NC
1987 - 1990: Research Assistant Professor, Department of Cell Biology and Anatomy, UNC-Chapel Hill, NC
1986 - 1987: Research Associate, Department of Anatomy, UNC-Chapel Hill, NC
Adjunct Associate Professor, Curriculum in Toxicology, UNC-Chapel Hill, NC
Professional Societies and Affiliations:
Memberships: Teratology Society
Editorial Boards: Section editor -Teratology - 2000-2002
Honors and Awards:
1999 - Clark Fraser Young Investigator Award, Teratology Society; 2003 - NHEERL Award: Development and
implementation of genomics, proteomics and bioinformatics capabilities within NHEERL. US EPA, ORD,
NHEERL
Selected invitations at National & International Symposia:
Developmental and Reproductive effects of drinking water contaminants. Gordan Research Conference.
Disinfection Byproducts. Mount Holyoak College, August, 2009; Effects of a drinking water concentrate on
Reproductive Development in a rat Bioassay. American Water Works Association Meeting, San Diego, CA,
June, 2009; Reproductive Development in a Multi-Generational Rat Bioassay of Drinking Water Concentrates
in the Four Lab Study, Toxicology and Risk Assessment Conference, Dayton, OH, Apr 2008; Embryonic
development: Gastrulation. Continuing Education Course. Society of Toxicology, Seattle, WA, Mar 2008;
Understanding Pathways of Toxicity: Making sense of changing signals. Symposium Lecture. Teratology
Society, Vancouver, BC, June 2004; Long-Range Research Initiative Annual Science Meeting. American
Chemistry Council, Miami Florida, May, 2004; Toxic Damage to Developmental Signals, Spring Symposium,
Integrated Toxicology Program, Duke University, Feb 2003; Drinking Water and Reproduction Symposium,
Society of Toxicology, Nashville, April, 2002; Departmental Seminar, Department of Toxicology, North
Carolina State University, Raleigh, NC, November, 2001; International Conference on Genes and Gene
PHS 398/2590 (Rev. 09/04, Reissued 4/2006)
Page 1
Continuation Format Page
Previous
TOC
-------
Principal Investigator/Program Director (Last, First, Middle): Hunter, Edward S.
Delivery for Diseases of Alcoholism, Alcohol Research Center, UNC at Chapel Hill, North Carolina, May,
2001; The role of apoptosis in developmental neurotoxicity and neurodegeneration in adults, Society of
Toxicology Symposium, San Francisco, CA, March, 2001; Early Craniofacial Development: Life among the
Signals, Triangle Consortium for Reproductive Biology, Research Triangle Park, January, 2001.
Selected Expert Committees/Advisory Panels/Organizing Committees:
Implementation planning for ToxTesting in the 21st Century - NHEERL, US EPA (2009-); Chair - NHEER
Toxicogenomics Core Advisory Group (2008-); Revisioning the Health Divisions of NHEERL, US EPA
(2006-7); Coordinator, NHEERL Health Division Proteomics Users Group (2004-5); In Vitro Assays for
Developmental Toxicity Assessments Planning Committee. ILSI HESI Developmental and Reproductive
Toxicology Technical Committee. June 2005; Long-Range Research Initiatives. American Chemistry Council.
May, 2004; Science Policy Council Technical Framework on Genomics for EPA, Performance-Based Quality
Assurance Workgroup (2004-5); Statistically-based structure-activity relationship (SAR) Systems for
Developmental Toxicity: Limitations and Challenges. ILSI Risk Science Institute, Sept. 2003; NHEERL
Genomics and Proteomics Committee (2001-6, Chair 2006).
Symposium Organization: Using Embryonic Stem Cells for Developmental Toxicity. In Vitro Assays for
Developmental Toxicity Assessments. ILSI HESI Developmental and Reproductive Toxicology Technical
Committee. February 2007. Co-Organier with Philippe VanParys; Genomics and Proteomics in Reproductive
and Developmental Toxicity. Society of Toxicology, Salt Lake City, UT, March 2003. Co-Organizer with Kim
Treinen; Drinking Water and Reproduction, Symposium Teratology Society, Florida, June, 2000.
B. SELECTED PUBLICATIONS.
Johnson CS, Zucker RM, Hunter ES 3rd. Sulik KK. (2008). Perturbation of retinoic acid (RA)-mediated limb
development suggests a role for diminished RA signaling in the teratogenesis of ethanol. Birth Defects Res
A Clin Mol Teratol. 79(9):631-41.
E.S. Hunter, III and Phillip Hartig (2008). Chapter 7: Targeted Gene changes affecting Developmental toxicity.
In: Developmental Toxicology. B. Abbott and D. Hanson Eds.
Rice G, Teuschler LK, Speth TF, Richardson SD, Miltner RJ, Schenck KM, Gennings C, Hunter ES 3rd.
Narotsky MG, Simmons JE. (2008) Integrated disinfection by-products research: assessing reproductive
and developmental risks posed by complex disinfection by-product mixtures. J Toxicol Environ Health A.
2008;71(17): 1222-34.
Chapin R, Augustine-Rauch K, Beyer B, Daston G, Finnell R, Flynn T, Hunter S. Mirkes P, O'Shea KS,
Piersma A, Sandier D, Vanparys P, Van Maele-Fabry G. (2008). State of the art in developmental toxicity
screening methods and a way forward: a meeting report addressing embryonic stem cells, whole embryo
culture, and zebrafish. Birth Defects Res B Dev Reprod Toxicol. Aug;83(4):446-56.
E. S. Hunter. III. Ellen Rogers, Ann Richard, Neil Chernoff (2006) Bromochloro-haloacetic acids: Effects on
mouse embryos in vitro and QSAR considerations. Reproductive Toxicology. 21(3):260-6.
E. S. Hunter. III. Maria Blanton, Ellen Rogers, Leonard Mole, Neil Chernoff (2006). Short-term exposures to
haloacetic acids produces dysmorphogenesis in mouse conceptuses in vitro. Reproductive
Karoly, E. , J.E. Schmid, E.S. Hunter. Ill (2005). Ontogeny of transcript profiles during mouse early
craniofacial development. Repro. Tox. 19:339-352.
S. Degitz, J Rogers and E.S. Hunter. Ill (2004). Developmental toxicity of methanol: Pathogenesis in CD-I and
C57BL/6J mice exposed in whole embryo culture. Birth Defects Research Part A.70:179-184
Corey S. Johnson, Maria R. Blanton, E. S. Hunter. 111(2004). Effects of ethanol and hydrogen peroxide on
mouse limb bud mesenchyme differentiation and cell death. In vitro Cell Dev Biol Anim 40: 108-112.
J.E. Simmons, L.K. Teuschler, C. Gennings, T.E. Speth,, S.D. Richardson, R.J. Miltner, M.J. Narotsky, K.M.
Schenck, E.S. Hunter. III. R.C. Hertzberg and G. Rice (2004). Component-based and whole-mixture
PHS 398/2590 (Rev. 09/04, Reissued 4/2006) Page 2 Continuation Format Page
Previous I TOC
-------
Principal Investigator/Program Director (Last, First, Middle): Hunter, Edward S.
techniques for addressing the toxicity of drinking-water disinfection by-products mixtures. Journal of
Toxicology and Environmental Health Part A, 2004, 67: 741-754.
L. White, E.S. Hunter. III. M.W. Miller, M.F. Ehrich, S. Barone, Jr. (2003). The role of apoptosis in
neurotoxicology. In: In vitro Neurotoxicology: Principles and Challenges. Humana Press.
E.H. Rogers, E.S. Hunter. III. M.B. Rosen, J.M. Rogers, C. Lau, P.C. Hartig, B.M. Francis, N. Chernoff (2003).
Lack of evidence for intergenerational reproductive effects due to prenatal and postnatal undernutrition in
the female CD-I mouse. Reproductive Toxicology. 17:519-525.
M.L. Fascineli, E.S. Hunter. III. Wilma De Grava Kempinas (2002). Fetotoxicity caused by the interaction
between zinc and arsenic in mice. TCM, 22:315-327.
N. Chernoff, E.S. Hunter III. L. L. Hall, M. B. Rosen, C. F. Brownie, D. Malarkey, M. Marr, J. Herkovits
(2002) . Lack of Teratogenicity of Microcystin-LR in the Mouse and Toad. Applied Toxicology. 22: 13-17.
J.M. Rogers and E.S Hunter. III. (2001). Redox redux: a closer look at conceptal low molecular weight thiols.
ToxicolSci. Jul;62(l):l-3.
E.S. Hunter. Ill and D.J. Dix (2001) Heat shock proteins Hsp70-l and Hsp70-3 are necessary and sufficient to
prevent arsenite-induced dysmorphology in mouse embryos. Molecular Reproduction and Development,
59: 285-293.
G.R. Klinefelter, E.S. Hunter. Ill and M. G. Narotsky (2001) Reproductive and Developmental Toxicity
Associated With Disinfection By-Products of Drinking Water. International Life Sciences Institute.
E.S. Hunter. Ill and Phillip Hartig (2000). Transient Modulation Of Gene Expression In The Neurulation Staged
Mouse Embryo. New York Academy of Sciences. 919, 278-283.
K.W. Ward, E.H. Rogers and E.S. Hunter. Ill (2000). Comparative pathogenesis of haloacetic acid and protein
kinase inhibitor embryotoxicity in mouse whole embryo culture. Toxicol Sci 53, 118-26.
R.M. Zucker, E.S. Hunter. Ill and J.M Rogers (1999). Apoptosis and morphology of mouse embryos by
confocal laser scanning microscopy. Methods: A companion to Methods in Enzymology. 18, 473-480.
K.W. Ward, E.H. Rogers and E.S. Hunter. Ill (1998). Dysmorphogenic effects of a specific protein Kinase C
inhibitor during neurulation. Reprod. Toxicol., 12(5), 525-534.
Hartig, P.C. and E.S. Hunter. Ill (1998). Gene Delivery to the neurulating embryo during culture. Teratology.
58:103-112.
R.M. Zucker, E.S. Hunter. Ill and J. Rogers (1998). Confocal laser scanning microscopy of embryo apoptosis.
Cytometry. 33: 348-354.
G.A. Boorman, V. Dellarco, J.K. Dunnick, R.E. Chapin, E.S. Hunter. III. F. Hauchman, H. Gardner, M. Cox,
R.C. Sills (1999). Drinking water disinfection byproducts: Review and approach to toxicity evaluation.
Environmental Health Perspectives, 107(Suppl 1): 207-217.
Tabacova, S., E.S. Hunter. Ill and L. Balabaeva (1997). Potential role for oxidative damage in developmental
toxicity of arsenic. IN: Arsenic: Exposure and Health Effects. Abernathy, R.L. Calderon and W.R.
Chappell, Eds, Chapman and Hall, NY, pp. 135-144.
E.S. Hunter. III. E.H. Rogers, J.E. Schmid and A. Richard (1996). Comparative effects of haloacetic acids in
whole embryo culture. Teratology. 53:352-360.
A.M. Richard and E.S. Hunter. Ill (1996). Quantitative structure activity relationship for the developmental
toxicity of haloacetic acids in mammalian whole embryo culture. Teratology. 54:57-64.
E.S. Hunter. Ill, and T.W. Sadler (1996). Direct effects of cocaine and cocaine metabolites on embryonic
development in whole embryo culture. Toxicology In vitro. 10:407-414.
Tabacova, S., E.S. Hunter. Ill and B.C. Gladen (1996). Developmental toxicity of inorganic arsenic in whole
embryo culture: Oxidation state-, dose-, time-, and gestational age dependence. Tox. Applied Pharm. 138:
298-307.
E.S. Hunter. III. (1996). Alterations of Intermediary Metabolism as a mechanisms of abnormal development
during early organogenesis. In: Handbook of Experimental Pharmacology. Vol 124 I: Drug Toxicity in
Embryonic Development I. R.J. Kavlock and G.P. Daston, Eds., pp. 371-406.
PHS 398/2590 (Rev. 09/04, Reissued 4/2006) Page 3 Continuation Format Page
Previous I TOC
-------
Principal Investigator/Program Director (Last, First, Middle): Hunter, Edward S.
E.S. Hunter, III and J. A. Tugman (1995). Inhibitors of glycolytic metabolism effect neurulation staged mouse
conceptuses in vitro. Teratology, 52: 317-323.
Hunter, E.S., III. L.E. Kotch, R.E. Cefalo, and T.W. Sadler (1995). Effects of cocaine administration during
early organogenesis on prenatal and postnatal development in mice. Fundamental and Applied
Toxicology. 28: 177-186.
Hunter. E.S.. III. J.A. Tugman, K.K. Sulik and T.W. Sadler. (1994). Effects of Short Term Exposure to Ethanol
on Mouse Embryos in vitro. Toxicology In vitro. 8: 413-421.
Sadler, T.W. and E.S. Hunter, III (1994). Principles of abnormal development: Past, present and future. In:
Developmental Toxicology, Second Edition. (C.A. Kimmel and J. Buelke-Sam, Eds). Raven Press, New
York pp. 53-63.
Sadler, T.W., K.M. Deno and E.S. Hunter, III (1993). Effects of altered maternal metabolism during
gastrulation and neurulation stages of embryogenesis. In: Maternal Nutrition and Pregnancy Outcome.
(C.L. Keen, A. Bendix, and C.C. Willhite, Eds). Annals NY Acad Sci.
Hunter, E.S., III and T.W. Sadler (1992). The role of the visceral yolk sac in hyperglycemia-induced
embryopathies in mouse embryos in vitro. Teratology. 45:195-203.
Hunter, E.S., III and T.W. Sadler (1989). Fuel mediated teratogenesis: Altered embryonic glucose metabolism
as a result of hypoglycemia in vitro. Am J Physiol. 257:E269-E276.
Hunter, E.S.,111, and T.W. Sadler (1988). Embryonic metabolism of fetal fuels in whole embryo culture.
Toxicology in Vitro. 2:163-167.
Sadler, T.W., E.S. Hunter. III. W. Balkan, L. Shum, W.E. Horton, Jr., and R.E. Wynn (1988). Effects of
maternal diabetes on embryogenesis. Am. J. Perinatol. 5:319-326.
Hunter, E.S., III. W. Balkan, and T.W. Sadler (1988). Improved development of pre-somite mouse embryos in
whole embryo culture. J. Exp. Zool. 245:264-269.
Hunter, E.S., III and T.W. Sadler (1987). D-(-)-beta-Hydroxybutyrate induced effects on mouse embryos in
vitro. Teratology. 36: 259-264.
Hahn, V.K.M., E.S. Hunter, III. R.M. Pratt, J. Zendegui, B.C. Lee (1987). Expression of rat transforming
growth factor alpha mRNA during development occurs predominantly in the maternal decidua. Molec. &
Cell. Biol. 7: 2335-2343.
Hunter, E.S., III, and T.W. Sadler (1987). A potential mechanism of DL-beta-hydroxybutyrate induced
malformations in mouse embryos. Am. J. Physiol., 253: E72-E80.
Sadler, T.W. and E.S. Hunter, III (1987). Hypoglycemia: How little is too much for the embryo?. Am. J. Obstet.
Gynecol, 157: 190-193.
Hunter, E.S., III, and T.W. Sadler (1987). Metabolism of D- and DL-beta-hydroxybutyrate by mouse embryos
in vitro. Metabolism, 36: 558-561.
Sadler, T.W., E.S. Hunter, III. W. Balkan, and R.E. Wynn. (1987). The role of maternal serum factors in
diabetes-induced embryopathies as studied in whole embryo culture. In: Approaches to Elucidate
Mechanisms in Teratogenesis. (F. Welsch, ed). Hemisphere Publishing Corp., Washington, pp. 109-
Horton, W.E., Jr., T.W. Sadler, and E.S. Hunter, III (1985). Effects of hyperketonemia on mouse embryonic
and fetal glucose metabolism in vitro. Teratology, 31: 227- 233.
Sadler, T.W., W.E. Horton, Jr., and E.S. Hunter, III (1985). Mammalian embryos in culture: A new approach in
investigating normal and abnormal developmental mechanisms. In: Developmental Mechanisms: Normal
and Abnormal. J.W. Lash and L. Saxen, eds. A.R. Liss, Inc., NY,NY, pp. 227-240.
Sadler, T.W., W.E. Horton, Jr., E.S. Hunter, III (1984). Mechanisms of diabetes-induced congenital
malformations as studied in mammalian embryo culture. In: Diabetes and Pregnancy: Teratology,
Toxicology and Treatment. C.C. Peterson, K. Furhmann and L. Jovanovic, eds. Praeger Press,
Philadelphia, PA. pp. 51-71.
PHS 398/2590 (Rev. 09/04, Reissued 4/2006) Page 4 Continuation Format Page
Previous I TOC
-------
Post-Doc (Last, First, Middle): Jack, John, R.
BIOGRAPHICAL SKETCH
Provide the following information for the key personnel and other significant contributors in the order listed on Form Page 2.
Follow this format for each person. DO NOT EXCEED FOUR PAGES.
NAME
John Jack
eRA COMMONS USER NAME
POSITION TITLE
Post-doctoral Research Associate
(Mathematician)
EDUCATION/TRAINING (Begin with baccalaureate or other initial professional education, such as nursing, and include postdoctoral training.)
INSTITUTION AND LOCATION
State University of New York at Potsdam
State University of New York at Potsdam
Louisiana Tech University
Environmental Protection Agency at Research
Triangle Park, NC
DEGREE
(if applicable)
B.A.
M.A.
Ph.D.
Post-doc
YEAR(s)
2004
2004
2009
2009-
Present
FIELD OF STUDY
Mathematics
Mathematics
Computational Analysis
and Modeling
Computational
Toxicology
A.
POSITIONS and HONORS
Research and Professional Experience:
2009-present Postdoctoral Research Associate, National Center for Computational Toxicology, ORD,
Environmental Protection Agency, Research Triangle Park, NC
2006-2009 Research Assistant, Institute for Micromanufacturing, Louisiana Tech University, Ruston, LA
(Mentor: Dr. Andrei Paun)
2008-2009 Teaching Assistant, Department of Computer Science, Louisiana Tech University, Ruston, LA
2005-2007 Teaching Assistant, Department of Mathematics, Louisiana Tech University, Ruston, LA
2003-2004 Peer Tutor, Department of Mathematics, SUNY Potsdam, Potsdam, NY
2003 Research Experience for Undergraduates (REU), Department of Mathematics, SUNY Potsdam,
Potsdam, NY (Mentor: Dr. Harold Ellingson)
2001 Research Experience for Undergraduates (REU), Department of Physics, SUNY Potsdam,
Potsdam, NY (Mentor: Dr. Biman Das)
Professional Societies and Affiliations:
2001-2004 Member of Pi Mu Epsilon
Honors and Awards:
2008-2009 LI Graduate Research Fellow, Louisiana Optical Network Interface (LONI) Institute
2008 Ph.D. Student of the Year, Institute for Micromanufacturing, Louisiana Tech University, Ruston,
LA
2007-2008 NSF Fellowship, Louisiana Tech University, Ruston, LA
2006-2008 University of Louisiana System Fellowship, Louisiana Tech University, Ruston, LA
2005-2007 Teaching Assistantship, Department of Mathematics, Louisiana Tech University, Ruston, LA
2003-2004 Vice President, Pi Mu Epsilon, SUNY Potsdam, Potsdam, NY
Selected invitations at National & International Symposia:
Workshop on Language Theory, The International Conference on Unconventional Computation, 2007,
Kingston, ON, CA. "Simulating Apoptosis Using Discrete Methods: a Membrane System and a Stochastic
Approach. International Conference on Unconventional Computation".
The 13th International Meeting on DNA Computing, 2007, Memphis, TN. "Modeling the Effects of HIV-1 Virions
and Proteins on Fas-Induced Apoptosis of Infected Cells".
PHS 398/2590 (Rev. 09/04, Reissued 4/2006)
Page J
Previous
TOC
-------
Post-Doc (Last, First, Middle): Jack, John, R.
The Joint AMS/MAA Conference, 2002, San Diego, CA. "Trivariate Statistics for Heavily Saturated Optical
Systems".
Selected Expert Committees/Advisory Panels/Organizing Committees:
2006 Room Chair, LA-MS section MAA meeting
Selected Assistance/Advisory Support to the Agency:
None
B. SELECTED PUBLICATIONS (selected from a total of 64 peer-reviewed).
J. Jack, A. Paun. Discrete Modeling of Biochemical Signaling with Memory Enhancement. LNBI Transactions
on Computational Systems Biology.
J. Jack, A. Paun, A. Rodriguez-Paton. Effects of HIV-1 Proteins on the Fas-mediated Apoptotic Signaling
Cascade: a Computational Study of Latent CD4+ T Cell Activation. WMC9. 227-246, 2008.
J. Jack, A. Rodriguez-Paton, O.H. Ibarra, A. Paun. Discrete Nondeterministic Modeling of the Fas Pathway.
International Journal of Foundations of Computer Science. 15:5, 1147-1167, 2008.
M. DeCoster, R. Masvekar, S. Maddi, J. Jack, and J. McNamara. Cellular Morphological and Biochemical
Changes During Apoptosis In Vitro: Links to Modeling. LBRN Work-inprogress Seminar. 2008.
J. Jack, F. Romero-Campero, M. Perez-Jimenez, O.H. Ibarra, A. Paun. Simulating Apoptosis Using Discrete
Methods: a Membrane System and a Stochastic Approach. International Conference on Unconventional
Computation. 2007.
J. Jack, A. Paun. Modeling the Effects of HIV-1 Virions and Proteins on Fas-Induced Apoptosis of Infected
Cells. DNA13. 20077
A. Allan, M. Dunne, J. Jack, J. Lynd, H. Ellingson. Classification of the Group of Units in the Gaussian
Intergers modulo n. Pi Mu Epsilon Journal [accepted] (8pp).
B. Das, E. Drake, J. Jack, M. Cianciosa. Higher Order Multivariate Statistics in the Intensity Fluctuations For a
Heavily Saturate Amplified Spontaneous Emission. The 35th Meeting of the Division of Atomic, Molecular
and Optical Physics. 2004.
B. Das, E. Drake, J. Jack. Trivariate Characteristics of Intensity Fluctuations for Heavily Saturated Optical
Systems. Applied Optics. 43, 834-840, 2004.
PHS 398/2590 (Rev. 09/04, Reissued 4/2006) Page 2
Previous I TOC
-------
Principal Investigator/Program Director (Last, First, Middle): Jones, Jack W.
BIOGRAPHICAL SKETCH
Provide the following information for the key personnel and other significant contributors in the order listed on Form Page 2.
Follow this format for each person. DO NOT EXCEED FOUR PAGES.
NAME
William Jack Jones
eRA COMMONS USER NAME
POSITION TITLE
Research Microbiologist
EDUCATION/TRAINING (Begin with baccalaureate or other initial professional education, such as nursing, and include
postdoctoral training.)
INSTITUTION AND LOCATION
Clemson University
Clemson University
DEGREE
(if applicable)
B.S.
Ph.D.
YEAR(s)
1973
1981
FIELD OF STUDY
Microbiology
Microbiology
Professional Experience
1991 -Present
2001 - 2006
1989-1991
1984-1989
1981 -1984
1976-1981
Research Microbiologist. USEPA, ORD, NERL.
Metabolism and Bioremediation Research Team Leader. U.S. EPA,
NERL.
Associate Professor. School of Applied Biology, Georgia Institute of
Technology.
Assistant Professor. School of Applied Biology, Georgia Institute of
Technology.
Postdoctoral Research Associate. Dept. of Microbiology, University of Illinois.
Research Associate. Dept. of Microbiology, Clemson University.
Selected Awards and Honors
2007, NERL Special Achievement Award
2006, 2005, 2004, 2004, 2003, 2000; EPA Scientific and Technological Achievement Awards
(STAA)
2005, US EPA Computational Toxicology Research Program New Start Competitive Research
Award
2003, Facilitation Impact Award (Southeast Association of Facilitators)
1998, USEPA Bronze Medal for Commendable Service: EPA Bioremediation Team
1998, Technical Review Panel: Dow Chemical Environmental Research
1996, USEPA/ORD Internal Grant Competition Award
1995, Research Award: USEPA/ARCS (Assessment and Remediation of Contaminated
Sediments)
1994-1995, NSF/EPA Environmental Research Grants Review Panel
1994-1998, US DOD (ESTCP) Environmental Research Review Panel
1992-present, National Research Council Research Adviser
Professional Societies
American Society for Microbiology (1975-present)
Sigma Xi (1990-present)
World Health Organization (WHO) International Program on Chemical Safety (2000-2004)
Previous
TOC
-------
Principal Investigator/Program Director (Last, First, Middle): Jones, Jack W.
Invited Lectures/Symposia
Jones, W.J., R.C. Kolanczyk, O. Mekenyan, A. Protzel, G. Dannan, S. Abel, and P.K.
Schmieder. 2007. Designing Pesticide Metabolic Pathway/Degradate Databases for
Registrant Submitted Health Effects/Ecological Effects Data. Presentation at BOSC Safe
Pesticides/Safe Products (SP2) Review, Research Triangle Park, NC.
Chapkanov, A., G. Chankov, S. Temelkov, J. Jones, and O. Mekenyan. 2007. Design and
Performance of a Xenobiotic Metabolism Database Manager for Metabolic Simulator
Enhancement and Chemical Risk Analysis. Presentation at SETAC Europe 17th Annual
Meeting, May 20-24, 2007, Porto, Portugal.
Jones, W.J., O. Mejenyan, R. Kolanczyk, and P. Schmieder. 2007. Simulating Metabolism to
Enhance Effects Modeling. Presentation at BOSC Safe Pesticides/Safe Products (SP2)
Review, Research Triangle Park, NC.
Jones, W.J., P.K. Schmieder, R.C. Kolanczyk, and O. Mekenyan. 2006. Simulating Metabolism
of Xenobiotic Chemicals as a Predictor of Toxicity. Presented at BOSC Review of the
Computational Toxicology Program, Research Triangle Park, NC.
Jones, W.J., O'Niell, W.L, Mazur, C.S., Kenneke, J.F., and Garrison, A.W. 2003.
Enantioselective Transformation of Chiral PCBs and Fipronil in Anoxic Sediments.
Presented at: 23rd International Symposium on Halogenated and Persistent Organic
Pollutants, Boston, MA, August 24-29, 2003.
Weber, E.J., T.W. Collette, W. J. Jones, J.F. Kenneke, C.S. Mazur, L.A. Suarez, C.T. Stevens,
J.W. Washington, K. Wolfe, G.W. Bailey, and R.S. Parmar. 2003. Modeling Chemical Fate
and Metabolism for Computational Toxicology. Presented at EPA Science Forum 2003,
Washington, DC.
Selected Publications
Jones, W.J., C.S. Mazur, J.F. Kenneke and A.W. Garrison. 2007. Enantioselective Microbial
Transformation of the Phenylpyrazole Insecticide Fipronil in Anoxic Sediments. Environ. Sci.
Technol. 41:8301-8307.
Jarman, J.L., W.J. Jones, L.A. Howell, and A.W. Garrison. 2005. Application of Capillary
Electrophoresis to Study the Enantioselective Transformation of Five Chiral Pesticides in
Aerobic Soil Slurries. Journal of Agricultural Food Chemistry 53:6175-6182.
Tebes-Stevens, C.L. and W.J. Jones. 2004. Estimation of Microbial Reductive Transformation
Rates for Chlorinated Benzenes and Phenols Using a QSAR Approach. Environ. Toxicol.
Chem. 23:1600-1609.
Mazur, C.S., W.J. Jones and C. Tebes-Stevens. 2003. H2 Consumption during the Microbial
Reductive Dehalogenation of Chlorinated Phenols and Tetrachloroethene. Biodegradation
14:285-295.
Pakdeesusuk, U., W.J. Jones, C.M. Lee, A.W. Garrison, W.L. O'Niell, D.L Freedman, J.T.
Coates, and C.S. Wong. 2003. Changes in Enantiomeric Fractions during Microbial
Reductive Dechlorination of PCB132, PCB149, and Aroclor 1254 in Lake Hartwell Sediment
Microcosms. Environ. Sci. Technol. 37:1100-1107.
Mazur, C.S. and W.J. Jones. 2001. Hydrogen Concentrations in Sulfate-Reducing Estuarine
sediments during PCE dehalogenation. Environ. Sci. Technol. 35:4783-4788.
Jones, W.J. and N.D. Ananyeva. 2001. Correlations between pesticide transformation rate and
microbial respiration activity in soils of different ecosystems. Biol. Fertil. Soils 33 (6):477-
483.
Garrison, A.W., V.A. Nzengung, J.K. Avants, J.J. Ellington, W.J. Jones, D. Rennels, and N.L.
Wolfe. 2000. Phytodegradation of p,p-DDT and the Enantiomers of o,p-DDT. Environ. Sci.
Technol. 34:1663-1670.
Pardue, J.H., S. Kongara, and W.J. Jones. 1996. Effect of Cadmium on Reductive
Dechlorination of Trichloroaniline. Environ. Toxicol. Chem. 15:1083-1088.
Previous I TOC
-------
Principal Investigator/Program Director (Last, First, Middle): Jones, Jack W.
Liu, S.M. and W.J. Jones. 1995. Biotransformation of Dichloroaromatic Compounds in
Nonadapted and Adapted Freshwater Sediment Slurries. Appl. Microbiol. Biotechnol.
43:725-732.
Presentations
Protzel, A., G. Dannan, R. Kolanczyk, O. Mekenyan, S. Abel, P. Schmieder, and J. Jones. 2006.
Development of a Structure-Searchable Database for Pesticide Metabolites and
Environmental Degradates. Presented at the Society for Toxicology Annual Meeting, San
Diego, CA, March 5-9, 2006.
Mekenyan, O.G., S.D. Dimitrov, T.S. Pavlov, W.J. Jones, and P.K. Schmieder. 2005.
Metabolism and Metabolic Activation of Chemicals: In Silico Simulation. Presented at the 3rd
International Symposium on Computational Methods in Toxicology and Pharmacology
Integrating Internet Resources, Shanghai, China, October 29-November 1, 2005.
Serafimova, R., H. Aladjov, R. Kolanczyk, P. Schmieder, Y. Akahori, J. Jones, and O.
Mekenyan. 2005. QSAR evaluation of ER Binding Affinity of Chemicals and Metabolites.
Presented at Society of Environmental Toxicology and Chemistry 26th Annual Meeting,
Baltimore, MD, Nov. 13-17, 2005.
Kolanczyk, R., M. Tapper, B. Nelson, V. Wehinger, J. Denny, D. Kuehl, B. Sheedy, C. Mazur, J.
Kenneke, J. Jones, and P. Schmieder. 2005. Increased Endocrine Activity of Xenobiotic
Chemicals as Mediated by Metabolic Activation. Presented at Society of Environmental
Toxicology and Chemistry 26th Annual Meeting, Baltimore, MD.
Mekenyan, O., J. Jones, P. Schmieder, S. Kotov, T. Pavlov, and S. Dimitrov. 2005.
Performance, reliability, and improvement of a tissue-specific metabolic simulator.
Presented at Society of Environmental Toxicology and Chemistry 26th Annual Meeting,
Baltimore, MD, Nov. 13-17, 2005.
Garrison, A.W., W.J. Jones, T.E. Wese, B. J. Konwick, M.A. Tapper, and M.K. Morgan. 2003.
The enantiomers of chiral pollutants pose different risks. Presented at: 24th Annual Society
of Environmental Toxicology and Chemistry Meeting, Austin, TX, November 9-13, 2003.
Jones, W.J., O'Niell, W.L, Mazur, C.S., Kenneke, J.F., and Garrison, A.W. 2003.
Enantioselective transformation of chiral PCBs and fipronil in anoxic sediments. Presented
at: 23rd International Symposium on Halogenated and Persistent Organic Pollutants,
Boston, MA, August 24-29, 2003.
Garrison, A.W., Jones, W.J., Wese, T.E., Washington, J.W., Jarman, J.L, and Avants, J. 2003.
Observations of enantioselectivity in the fate, persistence and effects of modern pesticides.
Presented at: 23rd International Symposium on Halogenated Organic Pollutants and
Persistent Organic Pollutants, Boston, MA, August 24-29, 2003.
Pakdeesusuk, U., C. M. Lee, D.L. Freedman, J.T. Coates, C.S. Wong, W.J. Jones, and A.W.
Garrison. 2002. Enantioselectivity in the biodegradation of PCB atropisomers. Presented at:
23rd Annual Society of Environmental Toxicology and Chemistry Meeting, Salt Lake City,
UT, November 16-20, 2002.
Tebes-Stevens, C.L., and W.J. Jones. 2000. QSAR analysis of sorption-corrected rate
constants for reductive biotransformation of halogenated aromatics. Presented at: 220th
American Chemical Society National Meeting, Washington, DC, August 20-24, 2000.
Previous I TOC
-------
Principal Investigator/Program Director (Last, First, Middle): JlldSOfl, Richard
BIOGRAPHICAL SKETCH
Provide the following information for the key personnel and other significant contributors in the order listed on Form Page 2.
Follow this format for each person. DO NOT EXCEED FOUR PAGES.
NAME
Richard Judson
eRA COMMONS USER NAME
POSITION TITLE
Bioinformatician
EDUCATION/TRAINING (Begin with baccalaureate or other initial professional education, such as nursing, and include postdoctoral training.)
INSTITUTION AND LOCATION
Rice University, Houston TX
Princeton University, Princeton NJ
Princeton University, Princeton NJ
University of Houston
DEGREE
(if applicable)
BA
MA
Ph.D.
Post-Doc
YEAR(s)
1981
1984
1988
1988-1989
FIELD OF STUDY
Chemistry / Physics
Chemistry
Chemistry
Chemistry / Physics
A. POSITIONS and HONORS
Research and Professional Experience:
2006-Present Bioinformatician, National Center for Computational Toxicology, USEPA, RTF NC
2008-Present Adjunct Professor, Univ. of North Carolina, Chapel Hill, Dept of Environmental Sciences and Engineering
2005-2006 President, GAMA BioConsulting, Guilford CT
2001-2005 Senior Vice President, Chief Scientific Officer, Genaissance Pharmaceuticals, New Haven CT
1999-2001 Senior Vice President, Informatics, Genaissance Pharmaceuticals, New Haven CT
1997-1999 Group Leader, Bioinformatics, CuraGen Corp., New Haven, CT
1990-1996 Senior Member Technical Staff, Sandia National Laboratories, Livermore, CA
1981-1983 CRT Process Engineer, Tektronix, Inc., Beaverton, OR
Honors and Awards
• EPA Bronze Medal for Commendable Service (2008)
• EPA / OEI CIO's John Cooper Partnership Award (2008)
• EPA Web Workgroup Award (2009) for ACToR
Selected invitations at National & International Symposia:
• Indiana University-Perdue University Indianapolis, Wiley lecture in Computational Chemistry, 10/02, "Informatics From Chem to
Bio: The Intersection of Science, Engineering, Medicine, and Computing"
• Princeton University, 10/02 "Open Problems in Genomics, Bioinformatics and Medicine"
• American Heart Association Scientific Sessions 6/03 "Apolipoprotein E Haplotypes are Associated with Baseline Levels and
Statin-Induced Decreases in C-Reactive Protein"
• BIOPHEX 2004 (Toronto), 9/04 "Regulatory Issues in Pharmacogenetics Testing"
• Michigan State University Department of Radiology, 6th Annual Molecular Imaging Workshop, 7/05 "Pharmacogenomics of
Statin Response"
• IBC Diagnostic Biomarkers, 12/05 (Cambridge MA) "Technical, Business and Regulatory Challenges to The Development of
Drug / Genetic Test Combinations"
• Pfizer Conference on Long QT Issues in Drug Development (Philadelphia), 9/05 "Using Genetic Testing to Manage Risks in
Thorough QT Trials"
• Molecular Diagnostics and Personalized Medicine (Boston), 9/05 "Technical, Business and Regulatory Challenges to the
Development of Drug / Genetic Test Combinations"
• EPA Environmental Information Symposium (Savannah), 12/06 "Developing Computer Systems and Databases for High
Throughput Toxicity Testing Prioritization"
• TRAC 2007 (Cincinnati), 4/07 "ACToR: Aggregated Computational Toxicology Resource"
• EPA Science Forum (RTP, NC), 5/07 "Computational Toxicology - Where is the Data?"
• University of Louisville, Center for Bioinformatics (Louisville KY), 6/07 "Computational Toxicology and Chemical
Prioritization"
i
Previous
TOC
-------
Principal Investigator/Program Director (Last, First, Middle): JlldSOfl, Richard
• International Society for Exposure Analysis Annual meeting (Durham NC), 10/07, "ACToR: Aggregated Computational
Toxicology Resource"
• North Carolina State University, School of Veterinary Medicine (Raleigh NC), 1/08 "Applications of Genetic Variation to Human
and Animal Health".
• North Carolina State University, Bioinformatics Program (Raleigh NC), 3/08 "Introduction to the EPA ToxCast Program"
• ebCTC Second Annual Toxicology Symposium (New Brunswick), 3/08 "Overview of the EPA ToxCast Program"
• North Carolina State University, Department of Environmental Science and Toxicology (Raleigh NC), 10/08 "The EPA ToxCast
Program"
• Society of Risk Analysis Annual Meeting (Boston MA) 12/08 "Toxicity Signatures from the ToxCast Program"
• Society of Toxicology Annual Meeting (Baltimore MD) 3/09 "Pathway-Based Concentration Response Profiles from
Toxicogenomics Data"
• ICCA / LRI Annual Meeting (Charleston, SC) 6/09 Overview of the EPA ToxCast Program
• International Implications of the U.S. National Research report on Toxicity Testing in the 21st Century 6/09 (Ottawa, ON),
Overview of the EPA ToxCast Program
Selected Assistance/Advisory Support to the Agency:
2007 - Present: PI for ACToR (Aggregated Computational Toxicology Resource) Project
2007 - Present: Co-Pi for ToxCast Project
2007-present: Mentored 1 undergraduate, 3 graduate, and 2 postdoctoral students
Selected Expert Committees/Advisory Panels/Organizing Committees:
2007-2008: Member of EPA/ORD Genomics Task Force, responsible for data management strategy
2007: Lecturer for Genomics Training Course developed for OPPTS
2007: Session Co-Chair for 2007 EPA Science Forum
2007-Present: Member of ORD IT Governance Board
2009-Present: Consultant to the FDA NCTR Scientific Advisory Board
2009: Co-chair of First EPA ToxCast Data Analysis Summit
B. SELECTED PUBLICATIONS (selected from 69 total).
• R. Judson, J.C. Stephens, A. Windemuth, "Tracking the Causative Genetic Variant: A Gene-based Haplotype Approach",
Conference Proceedings "Genome Targets to Drug Candidates", London November 29 (1999).
• P. Uetz, L. Giot, G. Cagney, T. A. Mansfield, R. S. Judson, J. R. Knight, D. Lockshon, V. Narayan, M. Srinivasan, P. Pochart, A.
Qureshi-Emili, Y. Li, B. Godwin, D. Conover, T. Kalbfleisch, G. Vijayadamodar, M. Yang, M. Johnston, S. Fields, J. Rothberg,
"A Comprehensive Analysis of Protein-protein Interactions in Saccharomyces Cerevisiae", Nature, Vol. 403 623 (2000).
• R. Judson, J. C. Stephens, A. Windemuth, "The Predictive Power of Haplotypes in Clinical Response", Pharmacogenomics, Vol.
1 15(2000).
• C. M. Drysdale, D. McGraw, C. B. Stack, J. C. Stephens, R. S. Judson, K. Nandabalan, K. Arnold, G. Ruano, S. Liggett,
"Complex Promoter and Coding Region, Beta(2)-Adrenergic Receptor Haplotypes Alter Receptor Expression and Predict In Vivo
Responsiveness", Proc. Nat. Acad. Science, USA, Vol. 97 10483-10488 (2000).
• J. C. Stephens, J. A. Schneider, D. A. Tanguay, J. Choi, T. Acharya, S. E. Stanley, R. Jiang, C. J. Messer, A. Chew, J.-H. Han, J.
Duan, J. L. Carr, M. S. Lee, B. Koshy,. M. Kumar, G. Zhang, W. R. Newell, A. Windemuth, C. Xu, T. S. Kalbfleisch, S. Shaner,
K. Arnold, V. Schulz, C. M. Drysdale, K. Nandabalan, R. S. Judson, G. Ruano, G. F. Vovis "Haplotype Variation and Linkage
Disequilibrium in 313 Human Genes", Science, Vol. 293, 489-493 (2001).
• R. Judson, B. Salisbury, J. Schneider, A. Windemuth and J. C. Stephens, "How many SNPs does a genome-wide haplotype map
require?", Pharmacogenomics, Vol. 31-13 (2002).
• R. Judson, "Using multiple drug exposure levels to optimize power in pharmacogenetic trials", Journal of Clinical Pharmacology,
Vol. 44 816-824 (2003).
• R. Jiang, J. Duan, A. Windemuth and J. C. Stephens, R. Judson, C. Xu, "Genome-wide evaluation of public SNP databases",
Pharmacogenomics, Vol. 4 779-789 (2003).
• B. Winkelmann, M. Hoffman, M. Nauck, A. Kumar, K. Nandabalan, R. Judson, B. Boehm, A. Tall, G. Ruano, W. Marz,
"Haplotypes of the cholesterol ester transfer protein gene predict lipid-modifying response to statin therapy", The
Pharmacogenomics Journal, Vol. 3 284-296 (2003).
• A. Windemuth, M. Kumar, K. Nandabalan, B. Koshy, C. Xu, M. Pungliya, R. Judson "Genome-wide Association of Haplotype
Markers to Gene Expression Levels", in "The Genome of Homo sapiens"
(Cold Spring Harbor Symposia on Quantitative Biology LVIII) (2004).
• R. Judson, C.D. Brain, B. Dain, A. Windemuth, G. Ruano, C. Reed, "New and confirmatory evidence of an association between
APOE genotype and baseline C-reactive protein in dyslipidemic individuals", Atherosclerosis, Vol. 177, 345-351 (2004).
Previous I TOC
-------
Principal Investigator/Program Director (Last, First, Middle): JlldSOfl, Richard
• R. Judson, A. Moss, "Pharmacogenomics in drug development: when and how to apply", in Cardiac Safety in clinical research
and drug development, practical guidelines, J. Morgenroth & I. Gussak, Eds. (Humana Press 2005).
• MJ. Ackerman. I. Splawski, J.C. Makielski, DJ. Jester, M.L. Will, K.W. Timothy, M.T. Keating, G. Jones, M. Chadha, C.R.
Burrow, J.C. Stephens, C.X. Xu, R. Judson, M.E. Curran "Spectrum and prevalence of cardiac sodium channel variants among
Black, White, Asian and Hispanic individuals: Implications for arrhythmogenic susceptibility and Brugada/Long QT syndrome
genetic testing", Heart Rhythm, Vol. 1, 600-607 (2005).
• C. Reed, M. Kalnik, K. Rakin, M. Athanasiou, R. Judson, "Restoring Value to Stalled Phase II Compounds: The Case for
Developing a Novel Compound for Depression Using Pharmacogenetics", Pharmacogenomics, Vol. 6, 95-100 (2005).
• R. Judson, B. Salisbury, "Technologies for Nutrigenomic Association Studies", in Nutrigenomics: The Future of Integrative
Metabolism, Eds: B. German, S. Watkins, M. Roberts, (Elsevier Press, 2006).
• R. Judson, B. Salisbury, C. Reed, M. Ackerman, "Pharmacogenetics in Thorough QT Trials", Molecular Diagnosis & Therapy,
Vol.10, 153-162(2006).
• R. Judson, "Pharmacogenetics in Drug Development and Research" in Electrical Diseases of the Heart: Genetics, Mechanisms,
Treatment, Prevention edited by Gussak, Antzelevitch, Wilde, Friedman, Ackerman and Shen (Springer, 2008).
• R.J. Kavlock, G. Ankley, J. Blancato, M. Breen, R. Conolly, D. Dix, K. Houck, E. Hubal, R. Judson, J. Rabinowitz, A. Richard,
R.W. Setzer, I. Shah, D. Villeneuve, E. Weber, "Computational Toxicology- A State of the Science Mini Review",
Toxicological Sciences, Vol. 103, 14-27 (2008)
• R.J Kavlock, D. Dix, K. Houck, R. Judson, M. Martin, A. Richard, "ToxCast: Developing Predictive Signatures for Chemical
Toxicity", AATEX Journal WC6 Proceedings Vol. 14, 623 (2008).
• A. Richard, C. Yang, R. Judson, "Toxicity Data Informatics: Supporting a New Paradigm for Toxicity Prediction", Toxicology
Mechanisms and Methods, Toxicology Mechanisms and Methods Vol. 18, 103-118 (2008)
• C.J. Verzilli, T. Shah, J.P. Casas, J. Chapman, M. Sandhu, S. Debenham, M.S. Boekholdt, K.T. Khaw, N. Wareham, R. Judson,
E.J. Benjamin, S. Kathiresan; M.G. Larson, J.Rong, R. Sofat, S.E. Humphries, L. Smeeth, G. Cavalleri, J.C. Whittaker, A.D.
Hingorani, "Bayesian meta analysis of genetic association studies with different sets of markers", Am. J.Hum.Gen. Vol. 82, 859-
872 (2008).
• R. Judson, F. Elloumi, R.W. Setzer, Z. Li, I. Shah, "A Comparison of Machine Learning Algorithms for Chemical Toxicity
Classification Using a Simulated Multi-Scale Data Mode" BMC Bioinformatics Vol. 9, 241 (2008)
• R. Judson, A. Richard, D. Dix, K. Houck, F. Elloumi, M. Martin, T. Cathey, T.R. Transue, R. Spencer, M. Wolf, "ACToR-
Aggregated Computational Toxicology Resource", Toxicology and Applied Pharmacology, Vol. 233, 7-13 (2008).
• R. Judson, A. Richard, DJ. Dix, K. Houck, M. Martin, R. Kavlock, V. Dellarco, T. Henry, T. Holderman, P. Sayre, S. Tan, T.
Carpenter, E. Smith, "The Toxicity Data Landscape for Environmental Chemicals", Environmental Health Perspectives (2008, in
press)
• M.T. Martin, R. Judson, D. Reif, DJ. Dix, R. Kavlock, "Profiling Chemicals Based on Chronic Toxicity Profiles from the U.S.
EPA ToxRef Database", Environmental Health Perspectives, Vol. 117, 392-399 (2009)
• M.T. Martin, E. Mendez, D.G. Corum, R.S. Judson, R. J. Kavlock, D.Rotroff, D. J. Dix, "Profiling the Reproductive Toxicity of
Chemicals from Multigeneration Studies in the Toxicity Reference Database (ToxRefDB)", Toxicological Sciences (2009, in
press).
• T.B. Knudsen, M.T. Martin, RJ. Kavlock, R.S. Judson, DJ. Dix, and A.V. Singh, "Profiling the Activity of Environmental
Chemicals in Prenatal Developmental Toxicity Studies using the U.S. EPA's ToxRefDB", Neurotoxicology and Teratology, Vol.
31,241(2009)
• A.W. Knight, S. Little, K. Houck, D. Dix, R. Judson, A. Richard, N. McCarroll., G. Akerman, C. Yang, L. Birrell, R. M.
Walmsley "Evaluation of High-throughput Genotoxicity Assays Used in Profiling the US EPA ToxCastTM Chemicals",
Regulatory Toxicology and Pharmacology (in press 2009)
ISSUED PATENTS
• U.S. Patent 6,586,183 - Association of Beta 2-Adrenergic Receptor Haplotypes with Drug Response
• U.S. Patent 6,931,326 - Methods for Obtaining and Using Haplotype Data
• U.S. Patent 6,944,767 - Methods and Apparatus for Ensuring the Privacy of Personal Medical Information
• U.S. Patent 7,058,517 - Methods for Obtaining and Using Haplotype Data (2)
• U.S. Patent 7,250,258 - CDK Genetic Markers Associated with Galantamine Response
Previous I TOC
-------
Principal Investigator/Program Director (Last, First, Middle): KavlOCk, Robert J.
BIOGRAPHICAL SKETCH
NAME
Robert J. Kavlock
eRA COMMONS USER NAME
Kavlock
POSITION TITLE
Research Biologist
EDUCATION/TRAINING (Begin with baccalaureate or other initial professional education, such as nursing, and include postdoctoral training.)
INSTITUTION AND LOCATION
University of Miami
University of Miami
Federal Executive Institute (Class 321)
DEGREE
(if applicable)
B.S.
Ph.D.
YEAR(s)
1969-73
1973-77
2006
FIELD OF STUDY
Biology
Embryology
Leadership
A. POSITIONS and HONORS
Research and Professional Experience:
2005- Director, National Center for Computational Toxicology, ORD, USEPA
2004-2005 Special Assistant (Computational Toxicology) to NHEERL Director
1999-2000: Acting Associate Director for Health, NHEERL (June-January)
1989-2004: Director, Reproductive Toxicology Division, NHEERL, USEPA, RTP, NC
1981-1989: Chief, Perinatal Toxicology Branch, DTD, HERL, USEPA, RTP, NC
1979-1981: Res. Biologist, Perinatal Toxicology Branch, DTD, HERL, USEPA, RTP, NC
1977-1979: Research Associate, Dept. of Biology, Univ. of Miami, Coral Gables, FL
Adjunct Associate Professor, Department of Pharmacology, Duke University
Adjunct Assistant Professor, Department of Zoology, NCSU
Professional Societies and Affiliations:
Memberships: Society of Toxicology, including Developmental and Reproductive Toxicology Specialty Section
and the North Carolina Society of Toxicology; Teratology Society
Editorial Boards: Toxicological Sciences (1994-2000); Teratogenesis, Carcinogenesis and Mutagenesis (to
2003); Journal of Toxicology and Environmental Health, Part B (current); Birth Defects Research, Part B
(2003-present); Neurotoxicology and Teratology (2006-present); Environmental Health Perspectives
(Associate Editor, 2006-present)
Honors and Awards:
US Human Society North American Alternative Award, 2008; US EPA/ORD Statesman of the Year, 2007; US
EPA Bronze Medals, 2004; Computational Toxicology Design Team, 1998, for efforts on Harmonized
Reproductive Testing Guidelines; US EPA Science Achievement Award, 1995 for efforts on validation of
benchmark dose methodology; US EPA Scientific and Technological Achievement Awards: Level I, 1994;
Level II, 1983, 1984, 1984, 1986, 1993; Level III, 1983, 1984, 1985, 1985, 1987, 1989, 1992, 1993 for various
peer reviewed scientific publications; US EPA Silver Medal, 1985 for development of an in vivo screening
procedure for developmental toxicity; Best Paper of the Year Award, Fundamental and Applied Toxicology,
1995; President, Teratology Society, 2001; President, Reproductive and Developmental Toxicology Specialty
Section, 1997; President, North Carolina Society of Toxicology, 1999
Selected invitations at National & International Symposia:
ZEBET 20th Year Anniversary, Berlin, October 2009; /World Congress on Alternatives to Animals, Rome,
September 2009; National Research Council Symposium on Toxicity Based Risk Assessment, Washington
DC, May 2009; ILSI Developmental Toxicity - New Directions Workshop, Washington DC, April 2009;
California Institute of Regenerative Medicine Stem Cells in Predictive Toxicology, Berkeley, 2008; Gene
Environmental Interactions in Reproduction, Malmo, Sweden, Feb 2008; European Chemicals Agency,
PHS 398/2590 (Rev. 09/04, Reissued 4/2006)
Page 1
Continuation Format Page
Previous
TOC
-------
Principal Investigator/Program Director (Last, First, Middle): KavlOCk, Robert J.
October 2007; Duke University SBRP Symposium on HTS Assays, October 2007; 6th World Congress on
Alternatives to Animals in Research, Tokyo, August 2007; American Chemistry Council, Washington, August
2007; 2nd Low Dose Workshop on Low Dose Effects of Environmental Toxicants, Berlin, April 2007; US EPA
Office of Pesticide Programs, Washington, Feb.2007; Duke University School of the Environment, Durham, NC
Jan. 2007; US EPA Science Policy Council, Dec. 2006; National Academy of Science Committee of Risk
Assessment, Washington, Dec 2006; 4th International Academic Conference on Environmental and
Occupational Medicine, Kunming, China, Oct. 2006; US EPA Office of Drinking Water, Sept. 2006; American
Association of Pharmaceutical Sciences, San Antonio, Oct. 2006; US EPA Regional Science Liaisons, April
2006; ORNL Bioinformatics Summit, April 2006; Society of Toxicology, San Diego, March 2006; US EPA
Region 6, Dallas, Jan 2006; European Commission Science Delegation, RTP, Jan 2006; Arizona State
University Workshop on Genetics and Environmental Regulation; Jan. 2005; National Academy of Science
Committee on the Future of Toxicology, Jan. 2005; National Academy of Science Workshop on Sustainability
in the Chemical Industry, Feb. 2005; US EPA National Risk Management Research Laboratory, Cincinnati,
Feb. 2005; US EPA Office of Science Coordination and Policy, Mar 2005; Tox Forum, Aspen, July, 2005.
2007; July 2005; Oak Ridge National Laboratory Ecogenomics Meeting, Knoxville, July, 2005; 5th World
Congress on Alternatives to Animals in Research, Berlin, August 2005; Board of Scientific Councilors, Jan,
2004; FDA Science Forum, Washington DC, Jan 2004; US/EU Bilateral Meeting on Chemical Safety,
Charlottesville, VA, Apr 2004; EPA Office of International Affairs United Kingdom Science Exchange, Aug
2004; National Academy of Sciences Future of Toxicology Committee, Sept 2004; National Toxicology
Program Workshop on Thyroid Toxicity, Washington DC, Apr 2003; US EPA Science Advisory Board,
Washington DC, Sept 2003.
Selected Expert Committees/Advisory Panels/Organizing Committees:
Expert Panel Member, Integrated Testing of Pesticides, Canadian Council of Academies (2009-); Expert,
WHO Working Group of the Health of Effects of DDT, Geneva, June 2009; Reviewer, European 7th
Framework Proposals for the Innovative Medicines Initiative, Brussels, February 2009; Co-Chair, Tox21
Working Group (2007-); NIEHS SBRP Peer Review Panel, Sept 2007; Chair, EPA International Science
Forum on Computational Toxicology, 2007; OCED Molecular Screening Initiative Working Group (2005-
present); WHO/I PCS Working Group on Principles for Evaluating Health Risks to Children, 2003-2006; Chair,
EPA Workshop on a Framework for Computational Toxicology, 2003; Chair, WHO/I PCS and Japan MOE
Workshop on Research Needs for Endocrine Disrupters, 2003; I LSI Workgroup on Human Framework for
Using MOA Information to Evaluate Human Relevance of Animal Toxicity Data, 2002-2004; EPA/NIEHS/ACC
Scientific Frontiers in Developmental Toxicity Risk Assessment, 2002; American Chemistry Council Focal Area
Leader, Long Range Research Initiative, 2002-2005; NTP/NIEHS Endocrine Disrupters Low-Dose Peer
Review, 2000; I PCS/WHO Steering Group for International State-of-Science Assessment of Endocrine
Disrupters, 1997-2002; ; Reviewer, European Commission Framework Calls, 2001, 2002, 2004, 2007; NIH
ALTX-4 Study Section, Standing Member, 1997-2001; CUT Science Advisory Committee, 1996-2001; Chair,
NTP Center for Evaluation of Risk to Human Reproduction Expert Panel on Phthalates, 1999-2000 and 2005);
IARC Monograph Working Groups, Volumes 36, 41, 47, 54, 58, 73, and 79; IARC Handbooks of Cancer
Prevention, Volumes 2 and 4.
Selected Assistance/Advisory Support to the Agency:
Co-Chair, EPA Science Policy Council Future of Toxicity Testing Working Group (2007-2009); Chair, US
EPA/ORD Technical Qualifications Review Board for Science and Technology Positions (2007- 2009); Chair,
EPA/ORD Computational Toxicology Design Team (2003) and Implementation Steering Group, 2004-present;
NHEERL Genomics Program Steering Committee, 2001-2002; Endocrine Disrupter Methods Validation
Subcommittee (EPA FACA), 2001-2003; Co-Organizer, Japanese NIES/US EPA Workshop on EDCs, Tokyo,
February 2000; NHEERL Human Health Research Strategy Implementation Team, 2001-2003; Chair,
NHEERL Branch Chief Career Ladder Committee, 1997-1998; Chair, ORD Endocrine Disrupter Research
Strategy Committee, 1995-1998; Chair, EPA Workshop to Develop Research Needs for Endocrine Disrupters,
1995; Chair, HERL Communications Issues Committee, 1992 ;EPA Working Group on Harmonized Testing
Guidelines for Reproductive and Developmental Toxicity, 1991-1999; Co-Chair, HERL(NHEERL) Technical
Qualifications Board, 1989-1997; Co-Chair, ORD RIHRA Topic IV Subcommittee (Biologically Based Dose
PHS 398/2590 (Rev. 09/04, Reissued 4/2006) Page 2 Continuation Format Page
Previous I TOC
-------
Principal Investigator/Program Director (Last, First, Middle): KavlOCk, Robert J.
Response Models), 1988-1993; Chair, OHR/HERL Mission Statement Committee, 1987; EPA Working Group
on Developmental Toxicity Testing Guidelines, 1984-1985.
B. SELECTED PUBLICATIONS (selected more than 180 total).
Kramer, MG, Firestone, M, Kavlock, R and Zenick, H (2009). The Future of Toxicity Testing for Environmental
Contamainants. Environ. Health Perspectives 117:A283-A284.
Knudsen, TB, Martin, MT, Kavlock, RJ, Judson, RS, Dix, DJ and Singh, AV (2009). Profiling the Activity of
Environmental Chemicals in Prenatal Developmental Toxicity Studies using the US EPAs ToxRefDB.
Reproductive Toxicology 28:209-219.
Martin, MT, Mendez, E, Corum, DG, Judson, RS, Rotroff, DM and Dix, DJ (2009). Profiling the Reproductive
Toxicology of Chemicals from Multigeneration Studies in ToxRefBD. Toxicol. Sci. 11:181-190.
Martin, MT., Judson, RS., Reif, DM, Kavlock, RJ and Dix, DJ (2009). Profiling Chemicals Based on Chronic
Toxicity Results from the US EPA ToxRef Database. Environ. Health Perspectives 117:392-399.
Kavlock, R. Austin, CP and Tice, RR (2009). Toxicity Testing in the 21st Century: Implications for Human
Health Risk Assessment. Risk Analysis 29: 485-487.
Judson, R, Richard, A., Dix, DJ, Martin, M, Kavlock, R, Dellarco, Henry, T, Holderman, T., Sayre, P, Tan, S,
Carpenter, T, and Smith, E (2009). The Toxicity Data Landscape for environmental chemicals. Environ.
Health Perspectives 177:685-695.
Cohen Hubal, EA, Richard, AM, Shah, I, Gallagher, J, Kavlock, R, Blancato, J and Edwards, SW(2008).
Exposure science and the US EPA National Center for Computational Toxicology. Journal of Exposure
Science and Environmental Epidemiology, 1-6.
Houck, KA and Kavlock, RJ (2008). Understanding mechanisms of toxicity: Insights from drug discovery.
Toxicol and Appl. Pharm. 277:163-178.
Rogers, JM and RJ Kavlock (2008). Developmental toxicity. In: Casarett & Doull's Toxicology: The
Basic Science of Poisons, 7th ed. CD Klaassen, editor. McGraw-Hill, Inc., New York, NY, 301-331.
Dix, DJ, Houck, KA, Martin, MT, Richard, AM, Setzer, RWand Kavlock, RJ (2007). The ToxCast Program for
Prioritizing Toxicity Testing of Environmental Chemicals. Toxicol. Sci., 95(1); 5-12.
Martin, MT, Brennan, R, Hu, W, Ayanoglu, E, Lau, C, Ren, H, Wood, CR, Gorton, JC, Kavlock, RJ and Dix, D.
(2007). Toxicogenomic Study of Triazole Fungicides and Perfluoroalkly Acids in Rat Livers Accurately
Categorizes Chemicals and Identifies Mechanisms of Toxicity. Toxicol. Sci. 97(2): 595-613.
Kavlock, R, Barr, D, Boelkeheide, K, Breslin, W, Breysse, P, Chapin, R, Gaido, K, Hodgson, E, Marcus, M,
Shea, K and Wlliams, P. (2006). NTP-CERHR Expert Panel update on the reproductive and developmental
toxicity of di(2-ethylhexyl phthalate. Repro. Toxicol. 22:291-399.
Kavlock, RJ, Ankley, GT, Collette, T, Francis, E, Hammerstrom, K, Fowle, J, Tilson, H, Schmieder, P, Veith,
GD, Weber, W, Wolf, DC, and Young, D. (2005). Computational Toxicology: framework, partnerships and
program development. Repro. Tox. 19:281-290.
Kavlock. RJ and Cummings, A (2005). Mode of Action: Reduction of Testosterone Availability-Molinate-
induced Inhibition of Spermatogenesis. Crit. Rev. Tox. 35:685-690.
Cummings, A and Kavlock, RJ (2005). A systems biology approach to developmental toxicology. Repro.
Toxicol. 19:281-290.
Cummings, A and Kavlock, RJ (2004). Gene-environment interactions: A review of effects on reproduction and
development. Critical Reviews in Toxicology 34:461-485.
Rockett, JC, Kavlock, RJ, Lambright, C, Parks, LG, Schmid, JE, Wlson, VS, Wood, C and Dix, DJ
(2002). DNA arrays to monitor gene expression in rat blood and uterus following 17(3-estradiol
exposure: biomonitoring environmental effects using surrogate tissues. Tox. Sci. 69:49-59
Damstra T, Barlow, S, Bergman A, Kavlock R and Van Der Kraak, G, editors (2002). International
Programme On Chemical Safety Global Assessment: The State-Of-The-Science Of Endocrine
Disrupters. World Health Organization, Geneva.
Setzer RW, Lau C, Mole ML, Copeland MF, Rogers JM, and Kavlock RJ. (2001). Toward a biologically based
dose-response model for developmental toxicity of 5-fluorouracil in the rat: a mathematical construct.
Toxicol Sci.; 59(1):49-58.
Barlow, S, RJ Kavlock, JA Moore, SL Schantz DM Sheehan, DL Shuey, and JM Lary (1999). Teratology
Society Position Paper: The developmental toxicity of endocrine disrupters to humans. Teratology
60(6):365-375.
PHS 398/2590 (Rev. 09/04, Reissued 4/2006) Page 3 Continuation Format Page
Previous I TOC
-------
Principal Investigator/Program Director (Last, First, Middle): KavlOCk, Robert J.
Reiter, LW, C DeRosa, RJ Kavlock, G Lucier, MJ Mac, J Melillo, RL Melnick, and T Sinks (1998). The U.S.
Federal framework for research on endocrine disrupters and an analysis of research programs supported
during Fiscal Year 1996. Environ Health Persp 106(3):105-113.
Kavlock, RJ and GP Daston (1997). Handbook of Experimental Pharmacology, Vol. 124 Drug Toxicity in
Embryonic Development. Vol I and II: Advances in Understanding Mechanisms of Birth Defects:
Morphogenesis and Processes at Risks. (610 pgs). Springer-Verlang, Heidelberg, Germany, ISBN 3-540-
61259-9; ISBN 3-540-61261-0.
Cooper, RL and RJ Kavlock (1997). Endocrine disrupters and reproductive development: A weight-of-
evidence overview. Journal of Endocrinology 152:159-166.
Kavlock, RJ and GT Ankley (1996). A perspective on the risk assessment process for endocrine-disruptive
effects on wildlife and human health. Risk Analysis 16(6):731-739.
Kavlock, RJ, GP Daston, C DeRosa, P Fenner-Crisp, LE Gray, S Kaattari, G Lucier, M Luster, MJ Mac, C
Maczka, R Miller, J Moore, R Rolland, G Scott, DM Sheehan, T Sinks and HA Tilson (1996). Research
needs for the risk assessment of health and environmental effects of endocrine disrupters: A Report of the
U.S. EPA sponsored workshop. Environmental Health Perspectives Vol. 104, Supplement 4, pp 715-740.
Kavlock, RJ and RWSetzer (1996). The road to embryologically based dose-response models.
Environmental Health Perspectives 104 (Suppl 1):107-121.
Kavlock, RJ, BC Allen, EM Faustman, CA Kimmel (1995). Dose response assessments for developmental
toxicity:IV. Benchmark doses for fetal weight changes. J Fund Appl Toxicol 26:211-222.
Faustman, EM, BC Allen, RJ Kavlock, and CA Kimmel (1994). Dose-Response Assessment for
Developmental Toxicity: I. Characterization of Data Base and Determination of NOAELs. Fundamental
and Applied Toxicology 23:478-486.
Shuey, DL, C Lau, RJ Kavlock, JM Rogers, TR Logsdon, RM Zucker, KH Elstein, MG Narotsky, and RW
Setzer (1994). Biologically-Based Dose-Response Modeling in Developmental Toxicology: Biochemical
and Cellular Sequelae of 5-Fluorouracil Exposure in the Rat Fetus. Toxicol. Appl. Pharm. 126: 129-144.
Rogers, JM, ML Mole, N Chernoff, BD Barbee, Cl Turner, TR Logsdon and RJ Kavlock (1993). The
Developmental Toxicity of Inhaled Methanol in the CD-1 Mouse, with Application of Quantitative Dose-
Response Modeling for Estimation of Benchmark Doses. Teratology 47:175-188.
Oglesby, LA, MT Ebron-McCoy, TR Logsdon, F Copeland, PE Beyer and RJ Kavlock (1992). In Vitro
Embryotoxicity of a Series of Para-Substituted Phenols: Structure, Activity and Correlation with In Vivo
Data. Terato/ogy45(1):11-33.
Kavlock, RJ, GA Green, GL Kimmel, R Morrissey, E Owens, JM Rogers, TW Sadler, HF Stack, MD Waters
and F Welsch (1991). Activity Profiles of Developmental Toxicity: Design Considerations and Pilot
Implementation. Terato/ogy43:159-185.
Kavlock, RJ (1990). Structure-activity relationships in the developmental toxicity of substituted phenols: In
vivo effects. Teratology 41(1):43-59.
Kavlock, RJ, R Short and N Chernoff (1987). Further evaluation of an in vivo teratology screening procedure.
Teratogenesis, Carcinogenesis, and Mutagenesis7:7^Q.
Kavlock, RJ, N Chernoff and EH Rogers (I985). The effect of acute maternal toxicity on fetal development in
the mouse. Teratogenesis, Carcinogenesis, and Mutagenesis 5(I):3-I3.
Kavlock, RJ and JA Gray (I982). Evaluation of renal function in neonatal rats. Biol. of the Neonate 41:279-288.
Chernoff, N and RJ Kavlock (1982). An in vivo screen utilizing pregnant mice. J. of Toxicology and
Environmental Health 10:541-550.
Gray, LE Jr., RJ Kavlock, N Chernoff, J Ferrell, J McLamb and J Ostby (1982). Prenatal exposure to the
herbicide TOK destroys the rodent Harderian gland. Science 215:293-294.
PHS 398/2590 (Rev. 09/04, Reissued 4/2006) Page 4 Continuation Format Page
Previous I TOC
-------
Principal Investigator/Program Director (Last, First, Middle): Knildsen, Thomas B.
BIOGRAPHICAL SKETCH
Provide the following information for the key personnel and other significant contributors in the order listed on Form Page 2.
Follow this format for each person. DO NOT EXCEED FOUR PAGES.
NAME
KNUDSEN, Thomas B.
eRA COMMONS USER NAME
TOKNUD01
POSITION TITLE
Developmental Systems Biologist (title 42)
National Center for Computational Toxicology
EDUCATION/TRAINING (Begin with baccalaureate or other initial professional education, such as nursing, and include postdoctoral training.)
INSTITUTION AND LOCATION
Albright College, Reading PA
Thomas Jefferson University, Philadelphia PA
Children's Hospital, Cincinnati OH
Emory University, Atlanta GA
DEGREE
(if applicable)
B.S.
Ph.D.
Postdoc
Postdoc
YEAR(s)
1976
1981
1981-82
1982-86
FIELD OF STUDY
Biology
Anatomy
Cell Biology
Developmental Biology
A. POSITIONS and HONORS
Research and Professional Experience:
1986-90 Assistant Professor, Dept. of Anatomy, E. Tenn. State University, Johnson City TN
1990-03 Asst./Assc./Full Prof, (tenured) Path. Anat. Cell Biol., Thomas Jefferson University, Philadelphia PA
2003-pres. Editor-in-Chief, Reproductive Toxicology (Elsevier)
2004-07 Professor (tenured), Mol. Cell & Craniofacial Biol., U of Louisville, School of Dentistry, Louisville KY
2004-07 Director, Systems Analysis Laboratory, U of Louisville, Birth Defects Center
2004-pres. Center for Genetics and Molecular Medicine, U Louisville, Louisville KY
2005-07 Professor of Biochemistry and Molecular Biology (joint appointment), U Louisville
2007 Center for Environ. Genomics & Integrative Biology, U Louisville, Louisville KY
2007 Clinical and Translational Science Institute, Louisville, Biomedical Informatics Group
2007-pres. Adjunct Professor, University of Louisville
2007-pres. Developmental Systems Biologist (title 42), NCCT, US Environmental Protection Agency, RTP NC
Professional Societies and Affiliations:
Memberships: American Association for the Advancement of Science (AAAS), Sigma Xi, Teratology
Society, Society of Toxicology, European Teratology Society
Editorial Boards: Reproductive Toxicology (Elsevier), Editor-in Chief (2003 - present); Birth Defects Research
(Part C), 2002 - present; Developmental Dynamics, 2002 - present; Co-Editor,
Developmental Toxicology (Comprehensive Toxicology Series - Elsevier)
Honors and Awards:
Fellowships: Predoctoral trainee, T32 HD07075 (1977-80); NIH Postdoctoral trainee, F32 HD06212 (1982-85)
Federal grants (PI): NIH/NICHD grant R29 HD24143 (1989-94); NIH/NICHD grant RO1 HD30302 (1993-98);
US EPA grant CR 824445 (1995-99); NIH/NIEHS grant RO1 ES09120 (1998-01); NIH/NIEHS grant 1 R13
ES012410 (2003) ; US EPA-NCERQA grant R 827445-01-0 (1999-03): NIH/NIAAA grant RO1AA13205 (2001-
08); NIH/NIEHS grant 1 R13 ES013116 (2004) ; NIH/NIEHS grant 2 RO1 ES09120 (2001-07); NIH/NIEHS
Training grant T32 ES07282 (1998-08) ; NIH/NIEHS grant 1 R21 ES013821-01 (2005-08)
Special Recognition: Wilson Publication Award, Teratology Society (2002); University Scholar, U Louisville
(2003-08); ResearchlLouisville: 3rd place, Innovation in Biotechnology (2004); Distinguished Alumni Award
(2008) Thomas Jefferson University; Keynote Speaker, McGill University Pharmacology Research Day (2009).
Leadership roles (selected advisory panels, organizing committees, workshops, editorial boards):
Editor-in-Chief, Reproductive Toxicology (2003 - present); President of the Teratology Society (2007-08);
Chairman, Program Committee, 47th Annual Meeting of the Teratology Society; Council of the Teratology
Society (1999-02 and 2005-09); Scientific Liason Task Force, Society of Toxicology (2008-12); European
PHS 398/2590 (Rev. 09/04, Reissued 4/2006)
Page J
Biographical Sketch Format Pag
Previous
TOC
-------
Principal Investigator/Program Director (Last, First, Middle): Knildsen, Thomas B.
Commission, Expert Panel (FP7); Steering Committee, First International Workshop on Virtual Tissues (EPA,
April 21-23, 2009); Steering Committee, ILSI-HESI DART Workshop on "Developmental Toxicology New
Directions", Leader - Working Group on New Technologies (2009); Co-Organizer, Symposium on "Gene
Regulatory Networks in Developmental Biology and Computational Toxicology", Teratology Society 2009;
Director, Systems Analysis Laboratory, ULSD Birth Defects Center (2004-07); Workgroup "Consensus Panel
on Renaming the Peripheral Benzodiazepine Receptor (PER)" (2004-05); Chairman, Committee on
Bioinformatics in Teratology (2002-05); Organizer, symposium on "Systems Biology: a new venue for exploring
mechanisms of developmental toxicity", Society of Toxicology (2004); Organizer, workshop on "Microarray
Data Analysis and Bioinformatics"; Teratology Society (2002); Organizer, symposium on "Pluripotent Stem
Cells in and of the Embryo", Teratology Society (2001); Organizer, symposium on "Genomics in Birth Defects
Research", Teratology Society (1998); Director, Education Course on "Risk Assessment in Developmental
Toxicology", Teratology Society (1996); FASEB delegate from the Teratology Society (1998-01); NIH Human
Embryology & Development II Study Section (1994-98).
Speaking invitations (2007-09) at National & International Symposia (selected from 80 total):
Genes for low-dose regulation of the embryonic transcriptome (Hormesis - 6th Int Conf, Amherst 2007)
Computational Toxicology: New Approaches to Improve Environ. Hlth Protection (RIVM, Bilthoven NL2007)
Genomics as a tool in developmental toxicology (Eurotox, Amsterdam 2007)
Computational Toxicology: New Approaches to Improve Env. Hlth Protection (AIHA-SOT, Louisville 2007)
Computational Framework to Predict Toxicity & Prioritize Testing of Env. Chems. (NCAC-SOT 2007)
Virtual tissues and artificial life simulators: applications in birth defects research (Philadelphia 2008)
Virtual tissues and developmental systems biology (Gordon Research Conference, Andover 2008)
Computational embryology (CASCADE training in Developmental Risk Assessment, Berlin 2008)
Virtual tissue models in developmental toxicity research (SOT, Baltimore 2009)
Virtual tissue models and computational embryology (McGill University, Montreal 2009)
Modeling the Embryo (Gordon Research Conference, New London 2009)
Gene regulatory networks and the underlying biology of developmental toxicology (ETS, Aries FR 2009)
B. SELECTED PUBLICATIONS (70 total).
Chinsky JM, Ramamurthy V, Fanslow WC, Ingolia DE, Blackburn MR, Shaffer KT, Higley HR, Trentin JJ,
Rudolph FB, Knudsen TB, and Kellems RE (1990) Developmental expression of adenosine deaminase in
the upper alimentary tract of mice Differentiation 42: 172-183.
Airhart MJ, Roberts MA, Knudsen TB and Skalko, RG (1990) Axonal guidance of adenosine deaminase
immunoreactive primary afferent fibers in developing mouse spinal cord Brain Res Bull 25: 299-309.
Hong L, Mulholland J, Chinsky JM, Knudsen TB, Kellems RE and Glasser SR (1991) developmental
expression of adenosine deaminase during decidualization in the rat uterus Biol Reprod 43: 83-93.
Knudsen TB, Blackburn MR, Chinsky JM, Airhart MJ and Kellems RE (1991) Ontogeny of adenosine
deaminase in the mouse decidua and placenta: immunolocalization and embryo transfer studies. Biol.
Reprod. 43: 171-184.
Knudsen TB, Winters RS, Otey SK, Blackburn MR, Airhart MJ, Church JK and Skalko RG (1992) Effects of
(R)-deoxycoformycin (pentostatin) on intrauterine nucleoside catabolism and embryo viability in the
pregnant mouse. Teratology 45: 91-103.
Blackburn MR, Gao X, Airhart MJ, Skalko RG, Thompson LF and Knudsen TB (1992) Adenosine levels in the
early postimplantation mouse uterus. Quantitative analysis by HPLC-fluorometric detection and spatio-
temporal regulation by 5'-nucleotidase and adenosine deaminase. Dev. Dynam. 194: 155-168.
Gao X, Blackburn MR and Knudsen TB (1994) Activation of apoptosis in early mouse embryos by 2'-
deoxyadenosine exposure. Teratology 48: 1-12.
Gao X, Knudsen TB, Ibrahim MM and Haldar S (1995) Bcl-2 relieves deoxyadenylate stress and suppresses
apoptosis in Pre-B leukemia cells. Cell Death Different. 2: 69-78.
Puffinbarger NK, Hansen KR, Resta R, Laurent AB, Knudsen TB, Madara JL and Thompson LF (1995)
Production and characterization of multiple antigenic peptide antibodies to the adenosine A2b receptor. Mol
Pharmacol 47: 1126-1132.
PHS 398/2590 (Rev. 09/04, Reissued 4/2006) Page Biographical Sketch Format Page
Previous I TOC
-------
Principal Investigator/Program Director (Last, First, Middle): Knildsen, Thomas B.
Ibrahim MM, Weber IT and Knudsen TB (1995) Mutagenesis of human adenosine deaminase to active forms
that partially resist inhibition by pentostatin. Biochem. Biophvs. Res. Commn. 209: 407-416.
Wubah JA, Ibrahim MM, Gao X, Nguyen D, Pisano MM and Knudsen TB (1996) Teratogen-induced eye
defects mediated by p53-dependent apoptosis. Current Biology 6:60-69.
Resta R, Hooker SW, Laurent AB, Rahman SMJ, Franklin M, Knudsen TB, Nadon ML and Thompson LF
(1997) Insights into thymic purine metabolism and adenosine deaminase deficiency revealed by transgenic
mice overexpressing ecto-5'-nucleotidase (CD73). J Clin Invest 99: 676-683.
Blackburn MR, Knudsen TB and Kellems RE (1997) Genetically engineered mice demonstrate that adenosine
deaminase is essential for early postimplantation development. Development 124: 3089-97
Knudsen TB (1997) Genetic and Cellular Pathways in Teratogen-induced Cell Death. In Comprehensive
Toxicology (Vol. 10), Sipes IG, McQueen CA and Gandolfi AJ (eds.), New York: Pergamon, pp. 529-534.
Knudsen TB and Wubah JA (1998) Transgenic animal models. Functional analysis of developmental toxicity
as illustrated with the p53 suppressor model. In Handbook of Developmental Neurotoxicology, Slikker, W.
Jr. and Chang, L.W. (eds.), San Diego: Academic Press, pp 209-221.
Ibrahim MM, Razmara M, Nguyen D, Donahue RJ, Wubah JA and Knudsen TB (1998) Altered expression of
mitochondrial 16S ribosomal RNA in p53-deficient mouse embryos revealed by differential display.
Biochem Biophvs Acta 1403: 254-264.
Blackburn MR, Wubah JA, Thompson LF and Knudsen TB (1999) Transitory expression of the A2b adenosine
receptor during implantation chamber development. Devel Dynam 216: 127-136.
Knudsen TB (1999) HPLC-based mRNA differential display. In: Developmental Biology Protocols (vol. II),
Tuan, R.S. and Lo, C.W. (eds.). Totowa: Humana Press, Inc., pp 337-341.
Knudsen TB (2000) Mitochondrial transduction of teratogenesis. Teratology 62: 238-239.
Donahue RJ, Razmara M, Hoek JB and Knudsen TB (2001) Direct influence of the p53 tumor suppressor on
mitochondrial biogenesis and function. FASEB J 15: 635-644.
Wubah JA, Setzer RW, Lau C, Charlap JH and Knudsen TB (2001) Exposure-disease continuum for2-chloro-
2'-deoxyadenosine, a prototype ocular teratogen. I. Dose-response analysis. Teratology 64: 154-169.
O'Hara MF, Charlap JH, Craig RC and Knudsen TB (2002) Mitochondrial transduction of ocular teratogenesis
during methylmercury exposure. Teratology 65:131-144.
Lau C, Narotsky MG, Lui D, Best D, Setzer RW, Mann PC, Wubah JA and Knudsen TB (2002) Exposure-
disease continuum for 2-chloro-2'-deoxyadenosine, a prototype teratogen. II. Induction of lumbar hernia in
the rat and species comparison for the teratogenic responses. Teratology 66: 6-18.
Knudsen TB, Charlap JH and Nemeth KA (2003) Microarray applications in developmental toxicology. In:
Perspectives in Gene Expression. K. Appasani, ed. Eaton Publishing/BioTechniques Press, Westboro MA.
pp 173-194.
Charlap JC, Donahue RJ and Knudsen TB (2003) Exposure-disease continuum for 2-chloro-2'-
deoxyadenosine, a prototype ocular teratogen. 3. Intervention with PK11195. Birth Defects Res (A) 67:
108-115.
O'Hara MF, Nibbio BJ, Craig RC, Nemeth KR, Charlap JH and Knudsen TB (2003) Mitochondrial
benzodiazepine receptors regulate oxygen homeostasis in the early mouse embryo. Repro Tox 17: 365-375.
Lee JW, Park J, Jang B and Knudsen TB (2004) Altered Expression of Genes Related to Zinc Homeostasis in
Early mouse Embryos Exposed to Di-2-ethylhexyl phthalate. Toxicol Lett 152: 1-10.
Knudsen TB and Green ML (2004) Response characteristics of the mitochondrial DNA genome in
developmental health and disease. Birth Defects Res (C) 72: 313-329.
Singh AV, Knudsen KB and Knudsen TB (2005) Computational systems analysis of developmental toxicity:
design, development and implementation of a birth defects systems manager (BDSM). Reprod Tox 19:
421-439.
Knudsen TB (2005) How can we use bioinformatics to predict which agents will cause birth defects? In:
Primer in Teratology (B Hales and A Scialli, eds) Teratology Society Chapter 20, pp- 58-59
Szabo G, Hoek JB, Darley-Usmar V, Hajnoczky G, Knudsen TB, Mochly-Rosen D, and Zakhari, S (2005) RSA
2004: Combined basic research satellite symposium - session three: Alcohol and Mitochondrial
Metabolism: AT the crossroads of life and death. Alcoh. Clin. Exp. Res. 29: 1749-1752.
Nemeth KA, Singh AV and Knudsen TB (2005) Searching for biomarkers of developmental toxicity with
microarrays: normal eye morphogenesis in rodent embryos. Toxicol Appl Pharmacol. 206: 219-228.
PHS 398/2590 (Rev. 09/04, Reissued 4/2006) Page Biographical Sketch Format Page
Previous I TOC
-------
Principal Investigator/Program Director (Last, First, Middle): Knildsen, Thomas B.
Knudsen KB, Singh AV and Knudsen TB (2005) Data input module for Birth Defects Systems Manager.
Reprod Tox 20: 369-375.
Slikker W Jr, Young JF, Corley RA, Dorman DC, Conolly RB, Knudsen TB, Erstad BL, Luecke RH, Faustman
EM, Timchalk C and Mattison DR (2005) Improving predictive modeling in pediatric drug development:
pharmacokinetics, pharmacodynamics and mechanistic modeling. Ann NY Acad Sci 1053: 505-518.
Kinane DF, Shiba H, Stathopoulou PG, Zhao H, Lappin DF, Singh AV, Eskan MA, Beckers S, Weigel S, Alpert
B and Knudsen TB (2006) Gingival epithelial cells heterozygous for Toll-like receptor 4 polymorphisms
Asp299Gly and Thr399lle are hypo-responsive to Porphyromonas gingivalis. Genes & Immun 7: 190-200.
Papadopoulos V, Baraldi M, TR Guilarte, Knudsen TB, Lacapere JJ, Lindemann P, Norenberg MD, Nutt D,
Poupon MF, Weizman A, Zhang MR and Gavish M. (2006) TspO: New Nomenclature for the peripheral-
type Benzodiazepine receptor/ recognition Site (PER) based on its structure and molecular function.
Trends in Pharmacol Sci. 8: 402-409.
Green ML, Singh AV, Zhang Y, Nemeth KA, Sulik KK and Knudsen TB (2007) Reprogramming of genetic
networks during initiation of the fetal alcohol syndrome. Devel Dynam 236: 613-631.
Singh AV, Knudsen KB and Knudsen TB (2007) Integrative Analysis of the mouse embryonic transcriptome.
Bioinformation 1: 24-30.
Singh AV, Rouhka EC, Rempala GA, Bastian CD and Knudsen TB (2007) Integrative database management
for mouse development: systems and concepts. Birth Defects Res (C): 81: 1-19.
Calabrese EJ, Bailer J, Bachmann KA, Bolger PM, Borak J, Cai L, Cedergreen N, Chiueh CC, Cherian MG,
Clarkson TW, Cook RR, Diamond DM, Doolittle DJ, Dorato MA, Duke SO, Feinendegen L, Gardner DE,
Hart RW, Hastings KL, Hayes AW, Hoffman GR, Jaworowski Z, Johnson TE, Keller JG, Klaunig JE,
Knudsen TB, Kozumbo WJ, Lettieri T, Liu S-Z, Maisseu A, Maynard K, Masoro EJ, Mothersil C, Newlin
DB, Oehme FW, Phalen RF, Philbert MA, Rattan SIS, Riviere JE, Rodricks J, Sapolsky RM, Scott BR,
Seymour C, Smith-Sonneborn J, Snow ET, Spear L, Stevenson DE, Thomas Y, Williams GM and Mattson
MP (2007) Biological Stress Response Terminology: Integrating the concepts of adaptive response and
preconditioning stress within a hormetic dose-response framework Toxicol Appl Pharmacol 222: 122-128
Deaciuc IV, Song Z, Peng X, Barve SS, Song M, He Q, Knudsen TB, Singh AV, and McClain CJ (2008)
Genome-wide transcriptome expression in the liver of a mouse model of high carbohydrate diet-induced
liver steatosis and its significance for the disease. Hepatol International 2: 39-49
Barthold JS, McCahan, Singh AV, Knudsen TB, Si X, Campion L and Akins RE (2008) Altered expression of
muscle and cytoskeleton-related genes in a rat strain with inherited cryptorchidism. J Androl. 29:352-366.
Datta S, Turner D, Singh R, Ruest LB, Pierce WM Jr and Knudsen TB (2008) Fetal Alcohol Syndrome (FAS)
in C57BL/6 mice detected through proteomics screening of the amniotic fluid Birth Defects Res (Part A) 82:
177-186
Knudsen TB and Kavlock RJ (2008) Comparative bioinformatics and computational toxicology. In:
Developmental Toxicology 3rd edition. (B Abbott and D Hansen, editors) New York: Taylor and Francis,
Chapter 12, pp 311-360
Knudsen TB, Martin NT, Kavlock RJ, Judson RS, Dix DJ and Singh AV (2009) Profiling the Activity of
Environmental Chemicals in Prenatal Developmental Toxicity Studies using the U.S. EPA's ToxRefDB.
Reproductive Toxicol 28: 209-219
Benakanakere MR, Li Q, Eskan MA, Singh AV, Galicia JC, Stathopoulou P, Knudsen TB and Kinane DF
(2009) MicroRNA-105 modulates TLR-2 responses in human oral keratinocytes. J Biol Chem (In Press)
Ema M, Ise R, Katoc H, Oneda S, Hirose A, Hirata-Koizumi M, Singh AV, Knudsen TB, and lharad T (2009)
Fetal malformations and early embryonic gene expression response in cynomolgus monkeys maternally
exposed to thalidomide, (submitted June 2009)
Knudsen TB, Houck K, Judson RS, Singh AV, Weissman A, Mortensen H, Reif D, Dix DJ, and Kavlock RJ
(2009) Biochemical activities of 320 ToxCast™ chemicals evaluated Across 239 functional targets,
(submitted July 2009)
Judson RS, Houck KA, Kavlock RJ, Knudsen TB, Martin MT, Mortsensen HM, Reif DM, Richard AM, Rotroff
DM, Shah I and Dix DJ (2009) Predictive in vitro screening of environmental chemicals - the ToxCast
project, (submitted August 2009)
PHS 398/2590 (Rev. 09/04, Reissued 4/2006) Page Biographical Sketch Format Page
Previous I TOC
-------
Staff Scientist (Last, First, Middle): Little, Stephen B.
BIOGRAPHICAL SKETCH
Provide the following information for the key personnel and other significant contributors in the order listed on Form Page 2.
Follow this format for each person. DO NOT EXCEED FOUR PAGES.
NAME
Stephen Blair Little
eRA COMMONS USER NAME
POSITION TITLE
Chemist
EDUCATION/TRAINING (Begin with baccalaureate or other initial professional education, such as nursing, and include postdoctoral training.)
INSTITUTION AND LOCATION
Gardner-Webb University, Boiling Springs, NC
North Carolina State University, Raleigh NC
North Carolina State University, Raleigh NC
Indiana University, Bloomington, Indiana
DEGREE
(if applicable)
B.S.
B.S.
M.A.
Graduate
Certificate
YEAR(s)
1977
1981
2001
2009
FIELD OF STUDY
Mathematics
Biochemistry
Toxicology
Chemical Informatics
A.
POSITIONS and HONORS
Research and Professional Experience:
2005-present Computational Chemist, National Center for Computational Toxicology, ORD
1995-2005 Chemist with the Environmental Carcinogenesis Division, NHEERL, US Environmental
Protection Agency, RTP, NC
1993-1995 Research Scientist, contract support scientist with Integrated Laboratory Systems, Inc. working
at U.S. EPA, Environmental Carcinogenesis Division, RTP, NC
1984-1993 Research Assistant, contract support scientist with Environmental Health Research and Testing,
Inc. working at U.S. EPA, Environmental Carcinogenesis Division, RTP, NC
1982 - 1984 Research Technician, BSRC , UNC-CH, Chapel Hill, NC
1981-1982 Research Technician, Cancer Research Center, UNC-CH, Chapel Hill, NC
1980-1981 Research Chemist (GS-5), ACB, U.S. EPA, RTP, NC
1977-1979 Technical Assistant for Acme United Corporation, Fremont, NC
Professional Societies and Affiliations:
1982-Present American Chemistry
2002 - Present Society of Toxicology
1988 - Present Genetics and Environmental Mutagenesis Society
Honors and Awards:
1995 Time off Award - for special technical assistance
1997 Outstanding Performance Award
1998, 2000 On-The-Spot (OTS) awards
2001 Group cash award.
2001 "S" Special Accomplishment Recognition Award - forsignificant contribution to the EPA cosponsored
EMS International Breast Cancer meeting
2003 Scientific and Technological Achievement Award - honorable mention
2004 "S" Award - for serving as Conazole QA TSR technical expert on review team
2006 Time off Award 2006 - group award for establishing NCCT
Selected Expert Committees/Advisory Panels/Organizing Committees:
2003-2005 Genetics and Environmental Mutagenesis Society Scientific Board of Councilors
2007-2009 Genetics and Environmental Mutagenesis Society Scientific Board of Councilors
PHS 398/2590 (Rev. 09/04, Reissued 4/2006) Page J
Previous
TOC
-------
Staff Scientist (Last, First, Middle): Little, Stephen B.
B. PUBLICATIONS (since 1995)
Lewis-Bevan, L., Little, S.B. and Rabinowitz, J.R. Quantum Mechanical Studies of the Structure and
Reactivities of the Diol Epoxides of Benzo[c]phenanthrene, Chemical Research in Toxicology, 8: 499, 1995.
Rabinowitz, J.R., Little, S.B. and Lewis-Bevan, L. The Effect of Crowding in the Bay/Fjord Region on the
Structure and Reactivities of Polycyclic Aromatic Hydrocarbons and their metabolites: Quantum Mechanical
Studies, Polycyclic Aromatic Compounds, 11:237, 1996.
Rabinowitz, J.R., Gifford, E.M. and Little, S.B. The Interactions between Chlorinated Dioxins and a Positively
Charged Molecular Probe: A New Molecular Interaction Potential, Journal of Computational Chemistry, 19:
673-684, 1998.
Little, S. B., Rabinowitz, J. R., Wei, P. and Yang W. A Comparison of Calculated and Experimental
Geometries for Crowded Polycyclic Aromatic Hydrocarbons and Their Metabolites, Polycyclic Aromatic
Compounds, 14:53-61, 1999.
Rabinowitz, J.R., Little, S.B and Brown, K.W. "Why Does 5-Methyl Chrysene Interact with DMA Like Both a
Planar and a Non-Planar Polycyclic Aromatic Hydrocarbon? Quantum Mechanical Studies", International
Journal of Quantum Chemistry, 88: 99-106. 2002.
Rabinowitz, J.R., Little, S. B., Brown, K.W., Benzo[a]pyrene and benz[c]phenanthrene: the effect of structure
on the binding of water molecules to the diol epoxices. Chemical Research in Toxicology. 15: 1069-1079.
2002.
Rabinowitz, J. R., Goldsmith, M-R., Little, S. B., and Pasquinelli, M. A. (2008) Computational Molecular
Modeling for Evaluating the Toxicity of Environmental Chemicals: Prioritizing Bioassay Requirements.
Environmental Health Perspectives, 116, 573-577.
Knight, AW, Little, S, Houck, K, Dix, D, Judson, R, Richard, A, McCarroll, N, Akerman, G, Yang, C, Birrell, L,
Walmsley, RM, Evaluation of High-throughput Genotoxicity Assays Used in Profiling the US EPA ToxCast™
Chemicals, Regulatory Pharmacology and Toxicology (2009) in press.
Rabinowitz, J; Little, S; Laws, S; Goldsmith, M-R, Molecular Modeling for Screening Environmental Chemicals
for Estrogenicity: Use of the Toxicant-Target Approach, Chemical Research in Toxicology, (2009) accepted for
publication..
PHS 398/2590 (Rev. 09/04, Reissued 4/2006) Page 2
Previous I TOC
-------
Principal Investigator/Program Director (Last, First, Middle): Martin, Matthew T.
Biographical Sketch
Provide the following information for the key personnel and other significant contributors in the order listed on Form Page 2.
Follow this format for each person. DO NOT EXCEED FOUR PAGES.
NAME
Matthew T. Martin
eRA COMMONS USER NAME
POSITION TITLE
Biologist
EDUCATION/TRAINING (Begin with baccalaureate or other initial professional education, such as nursing, and include postdoctoral training.)
INSTITUTION AND LOCATION
James Madison University, Harrisonburg, VA
University of North Carolina at Chapel Hill
University of North Carolina at Chapel Hill
University of North Carolina at Chapel Hill
DEGREE
(if applicable)
BS
MS
PhD
YEAR(s)
2003
2008
Currently
Enrolled
Currently
enrolled
FIELD OF STUDY
Integrated Science and
Technology (ISAT)
Environmental Sciences
and Engineering
Environmental Sciences
and Engineering
Bioinformatics and
Computational Biology
Training Program
A. POSITIONS and HONORS
Research and Professional Experience:
2005-Present Biologist, National Center for Computational Toxicology, USEPA, NC
2005 Database Analyst. CH2M Hill Inc. Herndon, VA
2003-2005 Environmental Scientist. Versar Inc. Springfield, VA
Professional Societies and Affiliations:
2008-present Society of Toxicology
Honors and Awards:
2008 Superior Accomplishment Recognition Award (SARA) for development and
application of the Toxicity Reference Database
2008 SARA for enabling the comprehensive retrospective analysis of legacy toxicity data
in collaboration with scientists in the Office of Pesticide Programs (OPP)
2007 SARA Office of Pesticide Programs Strategic Vision on developing an Integrated
and Hypothesis Based Assessment Paradigm
2003 ISAT Outstanding Senior Thesis Award
Selected Expert Committees/Advisory Panels/Organizing Committees:
2008-present OECD Extended One Generation Reproductive Toxicity Study (EOGRTS) Working Group
Selected Assistance/Advisory Support to the Agency:
2006-present Developing ToxRefDB for OPP use
B. PUBLICATIONS (8 Published & 7 Submitted).
PHS 398/2590 (Rev. 09/04, Reissued 4/2006)
Page J
Biographical Sketch Format Page
Previous
TOC
-------
Principal Investigator/Program Director (Last, First, Middle): Martin, Matthew T.
Martin MT, Rotroff DM, Reif DM, Dix DJ, Judson RS, Kavlock RJ, Houck KA. Identifying Toxicity-
Dependent Nuclear Receptor Activation by Monitoring Transcription Factor Activity (2009),
manuscript submitted.
Rotroff DM, Beam A, Martin MT, K. Freeman, E. LeCluyse, K. Houck, R. Judson, R. Kavlock, D. J.
Dix ,S. Ferguson. Modulation of Xenobiotic Metabolizing Enzyme and Transporter Gene
Expression in Primary Cultures of Human Hepatocytes by ToxCast Chemicals. (2009) Toxicology
and Applied Pharmacology, Submitted.
Knudsen et al. Biochemical Activities of 309 ToxCast™ Chemicals Evaluated Across 239 Functional
Targets (2009), manuscript submitted.
Judson RS, Houck KA, Kavlock RJ, Martin MT, Mortensen HM, Reif DR, Richard AM, Rotroff DM,
Shah I, Dix DJ. Predictive In Vitro Screening of Environmental Chemicals - The ToxCast Project
(2009) manuscript to submitted to EHP.
Shah et al. Human Nuclear Receptor Activity Stratifies Rodent Hepatocarcinogens (2009), manuscript
submitted.
Martin MT, Rotroff DM, Reif DM, Dix DJ, Judson RS, Kavlock RJ, Houck KA. Identifying Toxicity-
Dependent Nuclear Receptor Activation by Monitoring Transcription Factor Activity (2009),
manuscript submitted.
Judson RS, Houck KA, Kavlock RJ, Knudsen TB, Martin MT, Mortensen HM, Reif DM, Richard AM,
Rotroff DM, Shah I & Dix DJ. High throughput screening of toxicity pathways perturbed by
environmental chemicals (2009), manuscript submitted.
Martin MT, Judson RS, Reif DM, Kavlock RJ, Dix DJ. Profiling Chemicals Based on Chronic Toxicity
Results from the U.S. EPA ToxRef Database. Environmental Health Perspectives 117:1-8 (2009).
Martin MT, Mendez E, Corum DG, Judson RS, Kavlock RJ, Rotroff DM and Dix DJ. Profiling the
Reproductive Toxicity of Chemicals from Multigeneration Studies in the Toxicity Reference
Database (ToxRefDB). Toxicol Sci, doi: 10.1093/toxsci/kfp080 [online 10 April 2009] (2009).
Knudsen TB, Martin MT, Kavlock RJ, Judson RS, Dix DJ & Singh AV. Profiling the Activity of
Environmental Chemicals in Prenatal Developmental Toxicity Studies using the U.S. EPA's
ToxRefDB Reproductive Toxicol 28, in press (2009).
Kavlock RJ, DJ Dix, KA Houck, RS Judson, MT Martin, AM Richard. (2007). ToxCast: Developing
predictive signatures for chemical toxicity. Alt. Animal Test Experiment. 14, Special Issue, 623-
627.
Judson R, Richard A, Dix DJ, Houck K, Martin M, Kavlock RJ, Dellarco V, Henry Holderman T, Sayre
P, Tan S, Carpenter T and Smith E. The Toxicity Data Landscape for Environmental Chemicals.
Environmental Health Perspectives 2009; In Press: doi.1289/ehp.0800168 [Online 22 December
2008].
Page_2
Previous I TOC
-------
Principal Investigator/Program Director (Last, First, Middle): Martin, Matthew T.
Judson RS, Richard AM, Dix DJ, Houck K, Elloumi F, Martin MT, Cathey T, Transue TR, Spender R,
Wolf M. 2008b. ACToR - Aggregated Computational Toxicology Resource. Toxicology and
Applied Pharmacology. doi:10.1016/j.taap.2007.12.037 [Online 11 June 2008].
Martin MT, Brennan RJ, Hu W, Ayanoglu E, Lau C, Ren H, Wood CR, Gorton JC, Kavlock RJ, Dix DJ.
Toxicogenomic study of triazole fungicides and perfluoroalkyl acids in rat livers predicts toxicity
and categorizes chemicals based on mechanisms of toxicity. Toxicol Sci.,97(2):595- 613. (2007)
Dix, DJ, Houck, KA, Martin, MT, Richard, AM, Setzer, RWand Kavlock, RJ. TheToxCast Program for
Prioritizing Toxicity Testing of Environmental Chemicals. Toxicol. Sci., 95(1); 5-12. (2007).
Previous
Page_3
TOC
-------
Principal Investigator/Program Director (Last, First, Middle): Mortensen, Holly M.
BIOGRAPHICAL SKETCH
NAME
Holly M. Mortensen
eRA COMMONS USER NAME
POSITION TITLE
Research Biologist
EDUCATION/TRAINING (Begin with baccalaureate or other initial professional education, such as nursing, and include postdoctoral training.)
INSTITUTION AND LOCATION
University of Maryland (College Park, MD)
University of Maryland (College Park, MD)
Stanford University (Stanford, CA)
University of California (Davis, CA)
DEGREE
(if applicable)
Ph.D.
M.S.
M.S.
B.S.
YEAR(s)
2008
2005
2001
1999
FIELD OF STUDY
Human Genetics
Biology
Anthropological
Genetics
Biological Anthropology
A. POSITIONS and HONORS
Research and Professional Experience:
2008-Present Research Biologist, National Center for Computational Toxicology
Office of Research and Development U.S. Environmental Protection Agency,
Research Triangle Park, NC
2008-2001 Graduate Research Assistant, Department of Biology,
University of Maryland, College Park MD
(Advisors: Sarah A. Tishkoff)
2006 HHMI Teaching and Learning Fellow: Course Development in Bioinformatics
1999-2001 Graduate Research Assistant, Departments of Genetics and Anthroplogical Sciences,
Stanford University, Stanford, CA
(Advisors: Joanna Mountain, Luca Cavalli-Sforza)
1997-1999 Research Assistant, Department of Biological Anthropology, University of California, Davis, CA
(Advisor: David Glenn Smith)
Professional Societies and Affiliations:
2008-present Society of Toxicology (SOT)
2005-present American Association for the Advancement of Science (AAAS)
2004-present American Association of Anthropological Genetics (AAAG)
2001-present American Society of Human Genetics (ASHG)
1999-present American Association of Physical Anthropologists (AAPA)
Honors and Awards:
2006 Invited Participant, Wellcome Trust Advanced Course: Working with the HapMap
Wellcome Trust Genome Campus, Hixton, Cambridge, UK
2006 Howard Hughes Medical Institute Teaching and Learning Fellowship
University of Maryland, College Park, MD
2004 AAAG Student Prize for Paper Presentation
AAPA yearly meeting, Tampa, FL
2004 Invited Participant: Course on Molecular Evolution
Marine Biological Laboratory, Woods Hole, MA
2001-2008 NSF-IGERT Fellowship in Hominid Paleobiology
2001-2003 Eugenie Clark Fellowship
Page 1
Previous
TOC
-------
Principal Investigator/Program Director (Last, First, Middle): Mortensen, Holly M.
University of Maryland, College Park, MD
2002 Jacob K. Goldhaber Travel Grant
University of Maryland, College Park, MD
2001 Ford Foundation Graduate Research Grant for Ecological Research
Stanford University, Stanford, CA
2000 Mary M. Wohlford Fellowship, Morrison Institure for Population and Resource Studies
Stanford University, Stanford, CA
1999-2001 Foreign Language Area Studies (FLAS) Fellowship, Committee for African Studies
Stanford University, Stanford, CA
1994-1996 Cal Grant A, California Student Aid Commission, CA
Teaching Experience:
2006 Teaching Assistant, HHMI Course Development in Bioinformatics
University of Maryland, College Park, MD
2002-2003 Teaching Assistant, Introductory Genetics
University of Maryland, College Park, MD
2004 Teaching Assistant, Principles of Biology
University of Maryland, College Park, MD
2000 Teaching Assistant, Laboratory Research Methods in Anthropological Genetics
Stanford University, Stanford, CA
1997 Undergraduate Teaching Assistant, Human Evolutionary Biology
University of California, Davis, CA
B. SELECTED PUBLICATIONS
Tishkoff, S.A. Reed,F.A., Friedlaender, F.R., Ehret, C., Ranciaro, A., Froment, A., Hirbo, J.B. Awomoyi, A.A.,
Bodo.J., Doumbo, O., Ibrahim, M., Juma, AT., Kotze, M.J., Lema, G., Moore, J.H., Mortensen H.M.,
Nyambo, T.B., Omar, S.A., Powell, K., Pretorius, G.S., Smith, M.W., Thera, M.A., Wambebe, C., Weber,
J.L., Williams, S.M. May 2009, The Genetic Structure and History of Africans and African Americans,
Science 324 (5930): 1035-1044.
Mortensen, H.M., July 2008, Genetic Variation at the N-Acetyltransferase (NAT) Genes in Global Human
Populations, Doctoral Dissertation, Department of Biology, University of Maryland, College Park.
Tishkoff, S. A., Gonder, M.K., Henn, B.M., Mortensen, H.M., Fernandopulle, N., Gignoux.C., Lema, G.,
Nyambo, T. B, Underhill, P.A., Ramakrishnan.U., Reed, F. A., Mountain, J. L, July 2007, History of
click-speaking populations of Africa inferred from mtDNA and Y chromosome genetic variation,
Molecular Biology and Evolution, 24 (10): 2180-2195.
Gonder, M. K., Mortensen, H.M., Reed, F.A., Tishkoff, S.A., 2007, Whole mtDNA Genome Sequence Analysis
of Ancient African Lineages, Molecular Biology and Evolution 24 (3):757-768.
Tishkoff, S. A., Reed, F. A., Ranciaro, A., Voight, B. F., Babbitt, C. C., Silverman, J. S., Powell, K.,
Mortensen, H. M., Hirbo, J. B., Osman, M., Ibrahim, M., Omar, S. A., Lema, G., Nyambo, T. B., Ghori,
J., Bumpstead, S., Pritchard, J. K., Wray, G. A., and Deloukas, P., 2007, Convergent adaptation of
human lactase persistence in Africa and Europe, Nature Genetics 39 (1):31-40.
Knight, A., Underhill, P.A., Mortensen, H.M., Zhivotovsky, L.A., Henn, B.M., Ruhlen, M., Mountain
J.L., 2003, African Y Chromosome and mtDNA Divergence Provides Insight into the History of Click
Languages. Current Biology, 13: 464-473.
Malhi, R.S., Mortensen, H.M., Eshleman, J.A., Kemp, B.M., Lorenz, J.G., Kaestle, F. A., Johnson, J.R.,
Gorodezky, C., Smith. D.G., 2003, Native American mtDNA Prehisory in the American Southwest.
American Journal of Physical Anthropology. 120: 108-124.
Mortensen, H.M., 2001, Evidence of African Prehistory: Y Chromosome and mtDNA Analysis of
Four Linguistically Divergent African Groups (the Hadzabe, Datoga, Irawq, and Sukuma of North-
eastern Tanzania). MS Thesis. Department of Anthropological Sciences, Stanford University.
Page 2
Previous I TOC
-------
Principal Investigator/Program Director (Last, First, Middle): Mortensen, Holly M.
C. SELECTED PRESENTATIONS
Mortensen, H.M. (May 2009) Mapping Human Toxicity and Disease Pathways in ToxCast™. ToxCast Data
Analysis Summit. US EPA Research Triangle Park, NC. USA. (presentation).
Mortensen, H.M., Dix, D., Houck, K., Kavlock, R., Judson, R. (May 2009) Using the ToxMiner™ Database
for Identifying Disease-Gene Associations in the ToxCast™ Dataset. National Academy of Science.
Symposium on Toxicity Pathway-Based Risk Assessment: Preparing for Paradigm Change. Washington,
DC. USA (poster).
Houck, K., Mortensen, H.M., Witt, K.L., Menghang, X. (April 2009)Call for Nominations of Quantitative High-
Throughput Screening Assays from Relevant Human Toxicity Pathways. The Society for Biomolecular
Screening, Lille, France.
Mortensen, H.M., Dix, D., Houck, K., Kavlock, R., Shah, I., Judson, R. (March 2009) The ToxCast™ Pathway
Database: Identifying Toxicity Signatures and Potential Modes of Action from Chemical Screening Data.
Society of Toxicology, Baltimore, MD, USA (poster).
Mortensen, H.M., Awadalla, P., Tishkoff, S.A. (October, 2007) The Role of Natural Selection in Shaping
Genetic Variation at the N-acetyltransferase (NAT) Genes in African and Global Populations. American
Society of Human Genetics, San Diego, CA, USA (poster)
Mortensen, H.M., Gonder, M. K., Tishkoff, S.A. (April, 2004) Ancient Migrations and Population
Expansions in East Africa: Genetic Evidence for Tanzanian Prehistory. American Association of Physical
Anthropology. Tampa, FL, USA (presentation)
Mortensen, H.M., Gonder, K., Tarazona-Santos, E., Hirbo, J., Powell, K., Knight, A., Mountain, J., Tishkoff,
S.A, 2003, Genetic History of Hunting and Gathering Populations of Tanzania. American Association of
Physical Anthropology, Tempe, AZ, USA (poster)
Powell, K.B, Mortensen, H.M, Tishkoff, S.A. , 2003, The evolution of Lactase Persistence in African
Populations. American Association of Physical Anthropology, Tempe, AZ, USA (poster)
Mortensen, H.M., Gonder, M. K., Tarazona-Santos, E., Hirbo, J. B., Mountain, J., Tishkoff. S. A., 2002,
Genetic history of linguistically diverse Tanzanian populations inferred from mtDNA. American Society of
Human Genetics yearly meeting, Baltimore, MD, USA (poster)
Page 3
Previous I TOC
-------
Principal Investigator/Program Director (Last, First, Middle): Mlindy, William R.
BIOGRAPHICAL SKETCH
NAME
William R. Mundy
eRA COMMONS USER NAME
Mundy
POSITION TITLE
Research Toxicologist
EDUCATION/TRAINING (Begin with baccalaureate or other initial professional education, such as nursing, and include postdoctoral
training.}
INSTITUTION AND LOCATION
University of Massachusetts
University of Kentucky
University of Kentucky
DEGREE
(if applicable)
B.S.
M.S.
Ph.D.
YEAR(s)
1979
1983
1987
FIELD OF STUDY
Environmental
Science
Toxicology
Toxicology
A. POSITIONS and HONORS
Research and Professional Experience:
2009 - Present Research Toxicologist, Integrated Systems Toxicology Division, NHEERL,
USEPA
1990-2009
1987-1990
1981 -1987
1980-1981
Research Toxicologist, Neurotoxicology Division, NHEERL, USEPA
NIH Staff Fellow, National Institute of Environmental Health Sciences
Research Assistant, Graduate Program in Toxicology, Univ. of Kentucky
Toxicology Technician, Litton Bionetics Inc., Rockville, MD
Professional Societies and Affiliations:
Memberships: American Society for Neurochemistry, International Neurotoxicology Association,
North Carolina
Society for Neuroscience, Society for Neuroscience, Society of Toxicology
Honors and Awards:
USEPA Scientific and Technological Achievement Award, 1999; USEPA Scientific and
Technological Achievement Award, 2000; Neurotoxicology Division SAINT Award, 2002;
NHEERL Strategic Goal Award: Future Issues, 2003; NHEERL Teamwork Award, 2003;
USEPA ORD Honor Award: Bronze Medal, 2003; USEPA Scientific and Technological
Achievement Award (Honorable Mention), 2006
Selected invitations at National & International Symposia:
Annual Meeting of the Society of Toxicology, Salt Lake City, UT, March 9, 2003; Annual
Summer Meeting of the Toxicology Forum, Aspen, CO, June 14, 2003; Mid-year Advisory Board
meeting for the Johns Hopkins Center for Alternatives to Animal Testing (CAAT), June 10, 2004;
ECVAM-CEFIC Workshop on validation of alternative approaches for developmental
neurotoxicity: models and endpoints, European Commission Joint Research Centre, Ispra, Italy,
April 19-21, 2005; Duke Center for Drug Discovery, High Content Cellular Imaging Symposium,
Durham, NC, October 8, 2008; CAAT TestSmart DNT2 meeting, Reston, VA, November 12,
2008
Pagel
Previous
TOC
-------
Principal Investigator/Program Director (Last, First, Middle): Mlindy, William R.
Selected Expert Committees/Advisory Panels/Organizing Committees:
Advisory Board Member, Johns Hopkins University Center for Alternatives to Animal Testing
(CAAT)
B. SELECTED PUBLICATIONS (selected more than 65 total).
Inglefield, J.R., Mundy, W.R. and Shafer, T.J.: Inositol 1,4,5-triphosphate receptor-sensitive
Ca2+ release, store-operated Ca2+ entry, and cAMP responsive element binding protein
phosphorylation in developing cortical cells following exposure to polychlorinated biphenyls.
J. Pharmacol. Exp. Ther. 297:762-773. 2001.
Altmann, L, Mundy, W.R., Ward, T.R., Fastabend, A. and Lilienthal, H.: Developmental
exposure of rats to a reconstituted PCB mixture or Aroclor 1254: Effects on long-term
potentiation and [3H]MK-801 binding in occipital cortex and hippocampus. Toxicol Sci. 61:321-
330,2001.
Inglefield, J.R., Mundy, W.R., Meacham, C.A., and Shafer, T.J.: Identification of calcium-
dependent and - independent signaling pathways involved in polychlorinated biphenyl-induced
CREB phosphorylation in developing cortical neurons. Neuroscience 115:559-573, 2002.
Parran, O.K., Barone Jr., S., and Mundy, W.R.: Methylmercury decreases NGF-induced TrkA
autophosphorylation and neurite outgrowth in PC12 cells. Dev. Brain Res. 141:71-81, 2003.
Parran, O.K., Barone Jr., S., and Mundy, W.R.: Methylmercury inhibits TrkA signaling through
the ERK1/2 cascade after NGF stimulation of PC12 cells. Dev. Brain Res. 149:53-61, 2004.
Das, K.P., Freudenrich, T.M., and Mundy, W.R.: Assessment of PC12 cell differentiation and
neurite growth: a comparison of morphological and neurochemical measures. Neurotoxicol.
Teratol. 26:397-406, 2004.
Mundy, W.R., Freudenrich, T.M., Crofton, K.M., and DeVito, M.J.: Accumulation of PBDE-47 in
primary cultures of rat neocortical cells. Toxicol. Sci. 82:164-169, 2004.
Meacham, C.A., Freudenrich, T.M., Anderson, W.L., Sui, L., Lyons-Darden, T., Barone Jr., S.,
Gilbert, M.E., Mundy, W.R., and Shafer, T.J.: Accumulation of methylmercury or
polychlorinated biphenyls in in vitro models of rat neuronal tissue. Toxicol. Appl. Pharmacol.
205:177-187.2005.
Mundy, W.R., and Freudenrich, T.M.: Apoptosis of cerebellar granule cells induced by organotin
compounds found in drinking water: involvement of MAP kinases. Neurotoxicology 27:71-81,
2006.
Coecke, S., Goldberg, A.M., Allen, S., Buzanska, L., Calamandrei, G., Crofton, K., Hareng, L.,
Hartung, T., Knaut, H., Honegger, P., Jacobs, M., Lein, P., Li, A., Mundy, W., Owen, D.,
Schneider, S., Silbergeld, E., Reum, T., Trnovec, T., Tschudi-Monet, F. and Bal-Price, A.:
Workgroup report: Incorporating in vitro alternative methods for developmental neurotoxicity
into international hazard and risk assessment strategies. Environ. Health Perspect. 115:924-
931,2007.
Viberg, H., Mundy, W. and Eriksson, P.: Neonatal exposure to decabrominated diphenyl ether
(PBDE 209) results in changes in BDNF, CaMKII and GAP-43, biochemical substrates of
neuronal survival growth and synaptogenesis. Neurotoxicology 29:152-159, 2008.
Radio, N.M. and Mundy, W.R.: Developmental neurotoxicity testing in vitro: Models for
assessing chemical effects on neurite outgrowth. Neurotoxicology 29:361-376, 2008.
Mundy, W.R., Robinette, B., Radio, N.M. and Freudenrich, T.M.: Protein biomarkers associated
with growth and synaptogenesis in a cell culture model of neuronal development. Toxicology
249:220-229, 2008.
Page 2
Previous I TOC
-------
Principal Investigator/Program Director (Last, First, Middle): Mlindy, William R.
Radio, N.M., Breier, J.M., Shafer, T.J. and Mundy, W.R.: Assessment of chemical effects on
neurite outgrowth in PC12 cells using high content screening. Toxicol. Sci. 105:106-118,
2008.
Breier, J.M., Radio, N.M., Mundy, W.R. and Shafer, T.J.: Development of a high-throughput
screening assay for chemical effects on proliferation and viability of immortalized human neural
progenitor cells. Toxicol. Sci. 105:119-133, 2008.
Harrill, J.A., Li, Z., Wright, F.A., Radio, N.M., Mundy, W.R., Tornero-Velez, R., and Crofton,
K.M.: Transcriptional response of rat frontal cortex following acute In Vivo exposure to the
pyrethroid insecticides permethrin and deltamethrin. BMC Genomics 9:546, 2008.
Radio, N.M., Freudenrich, T.M., Robinette, B.L., Crofton, K.M., Mundy, W.R.: Comparison of
PC12 and cerebellar granule cell cultures for evaluating neurite outgrowth using high content
analysis. Neurotoxicol. Teratol. Xxx:xx-xxx, 2009.
Previous
TOC
-------
Principal Investigator/Program Director (Last, First, Middle): RabiflOWitZ, James R.
BIOGRAPHICAL SKETCH
NAME
James Rabinowitz
eRA COMMONS USER NAME
POSITION TITLE
Research Physicist
EDUCATION/TRAINING (Begin with baccalaureate or other initial professional education, such as nursing, and include postdoctoral training.)
INSTITUTION AND LOCATION
Alfred University, Alfred, NY
Uppsala University, Uppsala Sweden
State University of New York at Buffalo
Institute for Environmental Medicine, NYU
Medical Center, Tuxedo, NY
Harvard School of Public Health, Continuing
Education, Boston, MA
DEGREE
(if applicable)
B.A.
Certificate
Ph. D.
Post Doc
YEAR(s)
1962
1969
1972
1973
1994
FIELD OF STUDY
Physics
Solid State Physics,
Computational
Chemistry and
Theoretical Biology
Physics
Environmental Medicine
Analyzing Risk:
Science, Assessment
and Management
A. POSITIONS and HONORS
Research and Professional Experience:
1968 -1972 Research Associate, Center for Theoretical Biology, State University of New York at Buffalo,
Buffalo, NY.
1972 -1973 Postdoctoral Fellow, Institute of Environmental Medicine, New York University Medical Center,
Tuxedo, NY.
1973 -1974 Guest Scientist, Northeast Radiological Health Laboratory, BRH/HEW/USPS, Winchester, MA.
1973 -1977 Associate Research Scientist, Institute of Environmental Medicine, New York University Medical
Center, Tuxedo, NY.
1977 -1980 Research Scientist, Science and Technology Research Center, New York Institute of
Technology, Dania, FL.
1980 -1983 Research Physicist, CBB, EBD, HERL, ORD, EPA, RTP, NC.
1983 -1995 Research Physicist, CMB, GTD, HERL, ORD, EPA, RTP, NC.
1991 -2000 Lecturer, Molecular Modeling Course, Department of Pharmaceutical Chemistry and Carolina
Seminars Series, UNC, Chapel Hill, NC.
1995-2001 Research Physicist, BPB, ECD, NHEERL, ORD, EPA, RTP, NC.
2001-2005Research Physicist, MTB, ECD, NHEERL, ORD, EPA, RTP, NC.
2005-present Research Physicist, NCCT, ORD, EPA, RTP, NC.
Professional Societies and Affiliations:
American Association for the Advancement of Science
International Society for Quantum Biology and Pharmacology
American Chemical Society; Section on Chemical Toxicology; Section on Computers in Chemistry
Honors and Awards:
Bronze Medal for Commendable Service, U.S. EPA, for Obtaining High Performance Computer Platforms for
Environmental Research and Risk Assessment, 1992
Scientific and Technology Achievement Award, US EPA , 1985, 1997
Special Acts Award, NHEERL Supercomputing and High Performance Computing , 1997.
ClO's John Cooper Partnership Award, Office of Environmental Information , 2008
PHS 398/2590 (Rev. 09/04, Reissued 4/2006) Page 1
Previous
TOC
-------
Principal Investigator/Program Director (Last, First, Middle): RabiflOWitZ, James R.
Selected Invitations at National & International Symposia:
Invited presenter and participant at GE program on alternative methods. Endocrine Disruption, Metabolism and
Skin with the intent to look at new approaches to in vitro toxicity testing and how these approaches can be
used to predict human health consequences. The Center for Alternatives to Animal Testing, Johns Hopkins
Bloomburg School of Public Health, Baltimore Maryland 2007.
Invited speaker and participant at the ECVAM Workshop on Molecular Modelling Approaches for Human
Hazard Assessment of Chemicals Feb.20-22, 2006 Ispra Italy.
Invited keynote speaker for the American Chemical Symposium -Molecular Modeling in Environmental
Chemistry- sponsored by the Geological Chemistry of the ACS, with additional co-sponsors, Philadelphia, PA,
2004
Speaker at the American Chemical Society Symposium -Computational Toxicology- sponsored by the
Chemical Toxicology Section. Cosponsored by the Computers in Chemistry Section, NY, NY -2003
Invited lecturer at the EURESCO Conference Computational Biophysics: Integrating Theoretical Physics and
Biology, Biophysics from First Principles EURO Conference: From Electronic to the Mesoscale, European
Science Foundation, San Feliu de Guixols, Spain -2002
Speaker Molecular Modeling Applications for Environmental Problems, Computers in Chemistry, American
Chemical Society, New Orleans, LA -1996
Speaker Theoretical Calculations in Cancer Research: Progress and Perspectives, International Society of
Quantum Chemistry and Pharmacology, St. Andrews, Scotland -1995.
Selected Expert Committees/Advisory Panels/Organizing Committees:
Chairman session Bioinformatics and Computational Toxicology 48th Annual Meeting of the Society of
Toxicology. 2009.
Organizing Committee International Science Forum on Computational Toxicology 2007
Organizer of American Chemical Society Symposium -Computational Toxicology- sponsored by the
Chemical Toxicology Section. Cosponsored by the Computers in Chemistry Section, NY, NY -2003
Organizer and Chairman Symposium on Molecular Modeling Applications for Environmental Problems,
Computers in Chemistry, American Chemical Society, New Orleans, LA -1996
Organizing Committee and Co-chairman Theoretical Calculations in Cancer Research: Progress and
Perspectives, International Society of Quantum Chemistry and Pharmacology, St. Andrews, Scotland -1995.
Executive Committee of the International Society for Quantum Biology and Pharmacology. 1999 - 2002
Reviewed research articles for various journals.
Reviewed Proposals for the Petroleum Research Fund, National Science Foundation and NIOSH since 1997
Consultant on Research Project at the University of Rhode Island 2001-2004
Selected Assistance/Advisory Support to the Agency:
Initiated the Agency's participation in the U.S.-Poland Maria Sklodowska-Curie Fund for Research through
consultation with the Agency's Office of International Affairs
Reviewer, Organobromine Waste Review for RCRA, Office of Solid Waste 1997
Member of the Working Group, Hazardous Waste Identification Rule, ORD for OSW. 1997-1998
Member of the Supercomputer Working Group (The name changed to the High Performance Computing
Committee) 1990-present.
Member of the Endocrine Disrupter Research Implementation Plan Team 2000 -2003.
Member of the Scientific Office of the Future advisory committee and users group 2004 - 2006.
Member of NERL SP2 Steering Committee 2007
B. SELECTED PUBLICATIONS
L Lewis-Bevan, SB Little and Jr Rabinowitz (1995) Quantum Mechanical Studies of the Three Dimensional
Structure of the Diol-epoxides of Benzo(c)Phenanthrene. Chemical Research in Toxicology 8, 499-505.
JR Rabinowitz, SB Little and EM Gifford (1998) The Interactions between Chlorinated Dioxins and a
Positively Charged Molecular Probe: A New Molecular Interaction Potential, Journal of Computational
Chemistry 19, 673-684.
PHS 398/2590 (Rev. 09/04, Reissued 4/2006) Page 2
Previous I TOC
-------
Principal Investigator/Program Director (Last, First, Middle): RabiflOWitZ, James R.
SB Little, JR Rabinowitz, P Wei. and W Yang (1999) A comparison of calculated and experimental
geometries for crowded polycyclic aromatic hydrocarbons and their metabolites, Polycyclic Aromatic
Compounds 14, 53-61.
DM Marini, ML Shelton MJ Kohan, EE Hudgens, TE Kleindienst, LM Ball, DB Walsh, JG de Boer, L Lewis-
Bevan, JR Rabinowitz, LD Claxton, J Lewtas (2000) Mutagenicity in lung of big blue mice and induction
of tandem-base substitutions in salmonella by the air pollutant peroxyacetyl nitrate (PAN): predicted
formation of intrastrand cross-links, Mutation Research 457, 41 - 55.
JR Rabinowitz, SB Little and KW Brown, (2001) Why does 5-methyl chrysene interact with DMA as both a
planar and nonplanar polycyclic aromatic hydrocarbon, International Journal of Quantum Chemistry 88,
99-106.
KW Brown, SB Little, JR Rabinowitz (2002) Benzo[a]pyrene and Benzo[c]phenanathrene: The effect of
structure on the binding of water molecules to the diol epoxides, Chemical Research in Toxicology 15,
1069-1079.
JR Rabinowitz, SB Little, EM Gifford (2004) Molecular Interaction Potentials for the development of
structure activity relationships, In Quantitative Structure-Activity Relationships for Pollution Prevention,
Toxicological Screening, Risk Assessment and Web Applications, Society of Environmental Toxicology
and Chemistry, 93 - 104.
JR Rabinowitz, M.-R Goldsmith, SB Little, and MA Pasquinelli (2008) Computational Molecular Modeling
for Evaluating the Toxicity of Environmental Chemicals: Prioritizing Bioassay Requirements.
Environmental Health Perspectives, 116, 573-577.
JR Rabinowitz, SB Little, SC Laws and M-R Goldsmith (2009) Molecular Modeling for Screening
Environmental Chemicals for Estrogenicity: Use of the Toxicant-Target Approach Chemical Research in
Toxicology accepted for publication
PHS 398/2590 (Rev. 09/04, Reissued 4/2006) Page 3
Previous I TOC
-------
Principal Investigator/Program Director (Last, First, Middle): Reif, David M.
BIOGRAPHICAL SKETCH
NAME
David M. Reif
POSITION TITLE
Statistician
EDUCATION/TRAINING
INSTITUTION AND LOCATION
U.S. Environmental Protection Agency (RTP, NC)
Vanderbilt University (Nashville, TN)
Vanderbilt University (Nashville, TN)
College of William and Mary (Williamsburg, VA)
DEGREE
Post-doc
Ph.D.
M.S.
B S
(Monroe Scholar)
YEAR(s)
2006-
2008
2002-
2006
2003-
2005
1998-
2002
FIELD OF STUDY
Computational
Toxicology
Human Genetics
Applied Statistics
Biology
A. POSITIONS and HONORS
Research and Professional Experience:
2008-Present Statistician, National Center for Computational Toxicology,
U.S. Environmental Protection Agency, Research Triangle Park, NC
2008-Present Visiting Scholar, Department of Statistics,
North Carolina State University, Raleigh, NC
2006-2008 Biologist (federal post-doc), National Center for Computational Toxicology,
U.S. Environmental Protection Agency, Research Triangle Park, NC
(advisor: Elaine Cohen Hubal)
2002-2006 Graduate Research Assistant, Center for Human Genetics Research,
Vanderbilt University, Nashville, TN
(advisors: Jason Moore and Jonathan Haines)
1999-2001 Research Assistant, Department of Biology,
College of William and Mary, Williamsburg, VA
(advisor: Patty Zwollo)
Honors and Awards (Selected):
2009 OTS Award, National Center for Computational Toxicology, U.S. EPA
2007 OTS Award, Human Studies Division, U.S. EPA
2005 International Travel Grant, Vanderbilt University, Nashville, TN
2003-2005 NIH Training Grant in Human Genetics, Vanderbilt University, Nashville, TN
2002 Phi Sigma Biology Honors Fraternity, College of William and Mary, Williamsburg, VA
2001 Omicron Delta Kappa Leadership Fraternity, College of William and Mary, Williamsburg, VA
2001 Monroe Scholarship Supplemental Award for International Research, College of William and
Mary, Williamsburg, VA
2000-2001 Howard Hughes Medical Institute Undergraduate Research Grant, College of William and Mary,
Williamsburg, VA
Selected Expert Committees/Advisory Panels/Organizing Committees:
2009 Chair, NCCT Seminar Series, Research Triangle Park, NC, USA.
2008-2009 Program Committee, "Bioinformatics and Computational Biology", Genetic and Evolutionary
Computation Conference.
2007-present Grant Reviewer, National Science Foundation.
2007
Session Chair, "Statistical and Mathematical Modeling for Community-Based Risk Assessment",
Community Based Risk Assessment Workshop, National Center for Environmental Research,
Research Triangle Park, NC, USA.
PHS 398/2590 (Rev. 09/04, Reissued 4/2006)
Page 1 (of 4)
(Updated August 2009)
Previous
TOC
-------
Principal Investigator/Program Director (Last, First, Middle): Reif, David M.
2006-present Reviewer for: Bioinformatics; PLoS Genetics; Pharmacogenomics; BMC Bioinformatics; Human
Genetics; Journal of Exposure Science and Environmental Epidemiology; Journal of Infectious
Disease; Biotechniques; Medical Science Monitor;IEEE/ACM Transactions on Computational
Biology and Bioinformatics; PLoS One; Journal of Statistical Software
Selected Service/Assistance/Advisory Support to the Agency and the Scientific Community:
2009 Task Order Contract Officer Representative (TOCOR), ToxCast in vitro screening contracts.
2009 "Gene-Environment Interactions", Science Cafe presented by the North Carolina Museum of
Science, Raleigh, NC, USA [public seminar].
2009 "Lessons on Modern Toxicology: How Darwin Saw It Coming", EPA Greenversations weblog.
2009 "How our shared genetic history shapes responses to a changing environment", SMART lecture
series, Panther Creek High School, Gary, NC [seminar].
2008 Career Panel, Durham Public Schools "Stay in School" program, Durham, NC, USA.
2008 "Human Genetics", EPA-Shaw University Research Associateship Program, Chapel Hill, NC,
USA [seminar].
2007 "Genotyping technology and analysis in cancer research", Association for Biomedical Research,
Chapel Hill, NC, USA [seminar].
2006-present EPA Speaker's Bureau
Teaching Experience:
2008 Course Director and Lecturer, Introduction to R,
North Carolina State University, Raleigh, NC, USA [semester course].
2006 Teaching Assistant & Guest Lecturer, Statistics for Biomedical Researchers,
Vanderbilt University, Nashville, TN, USA [semester course].
2006 Guest Lecturer, General Biology I & II,
Nashville State Technical College, Nashville, TN, USA [semester course].
B. SELECTED PUBLICATIONS
Peer-Reviewed:
2009 Sanchez Y., Deener K., Cohen-Hubal E., Reif P.M., Segal D.A. Research Needs for
Community Based Risk Assessment. Journal of Exposure Science and Environmental
Epidemiology, 1(10).
2009 Reif P.M., Motsinger A.A., McKinney B.A., Edwards K.M., Chanock S.J., Rock M.T., Crowe Jr.
J.E., Moore J.H. Integrated Analysis of Genetic and Proteomic Data Identifies Biomarkers
Associated with Systemic Adverse Events Following Smallpox Vaccination. Genes and
Immunity, 10(2).
2009 Heidenfelder B.N., Reif P.M., Harkema J., Cohen-Hubal E., Gallagher J, Edwards S.E.
Comparative microarray analysis and pulmonary morphometric changes in Brown
Norway rats exposed to ovalbumin and/or concentrated airborne particulates.
Toxicological Sciences, 108(1). [Cover Article].
2008 Martin M.T., Judson R.S., Reif P.M., Pix P.J. Profiling Chemicals Based on Chronic Toxicity
Results from the U.S. EPA ToxRef Database. Environmental Health Perspectives, 116(11).
2008 Reif P.M., McKinney B.A., Motsinger A.A., Chanock S.J., RockM.T., Moore J.H., Crowe Jr. J.E.
Genetic basis for systemic adverse events following smallpox vaccination. Journal of
Infectious Piseases, 198(1).
2008 Motsinger A.A., Reif P.M., Fanelli T.J., Ritchie M.P. A Comparison of analytical methods for
genetic association studies. Genetic Epidemiology, 32(6).
2008 Hardison N.E., Fanelli T.J., Pudek S.M., Reif P.M., Richie M.P., Motsinger A.A. A Balanced
Accuracy Fitness Function Leads to Robust Analysis using Grammatical Evolution
Neural Networks in the Case of Class Imbalance. Genetic and Evolutionary Computation
Conference.
2007 Motsinger A.A., Reif P.M. Embracing Complexity: Gene-gene and Gene-Environment
Interactions. In: Genes, Genomes, and Genomics, vol. 3.
PHS 398/2590 (Rev. 09/04, Reissued 4/2006) Page 2 (of 4) (Updated August 2009)
Previous I TOC
-------
Principal Investigator/Program Director (Last, First, Middle): Reif, David M.
2007 Motsinger A.A., Ritchie M.D., Reif P.M. Novel Methods for Detecting Epistasis in
Pharmacogenomics Studies. Pharmacogenomics, 8(9).
2007 McKinney B.A., Reif P.M., White B.C., Crowe J.C., Moore J.H. Evaporative cooling feature
selection for genotypic data involving interactions. Bioinformatics, 23(16).
2007 Reif P.M., Israel M.A., Moore J.H. Exploratory visual analysis of statistical results of
microarray experiments comparing high and low grade glioma. Cancer Informatics, 2(1).
2007 Motsinger A.A., Reif P.M., Fanelli T.J., Pavis A.C., Ritchie M.P. Linkage disequilibrium in
genetic association studies improves the power of Grammatical Evolution Neural
Networks. IEEE Symposium on Computational Intelligence in Bioinformatics and
Computational Biology.
2006 McKinney B.A., Reif P.M., Moore J.H., Crowe Jr. J.E. Cytokine Expression Patterns
Associated with Systemic Adverse Events Following Smallpox Immunization. Journal of
Infectious Diseases, 194(4).
2006 Reif P.M., Motsinger A.A., McKinney B.A., Crowe Jr. J.E., Moore J.H. Feature Selection using
Random Forests for the Integrated Analysis of Multiple Data Types. IEEE Symposium on
Computational Intelligence in Bioinformatics and Computational Biology.
2006 McKinney B.A., Reif P.M., Ritchie M.P., Moore J.H. Machine learning for detecting gene-
gene interactions. Applied Bioinformatics, 5(2).
2006 Motsinger A.A., Reif P.M., Pudek S.M., Ritchie M.P. Understanding the Evolutionary
Process of Grammatical Evolution Neural Networks for Feature Selection in Genetic
Epidemiology. IEEE Symposium on Computational Intelligence in Bioinformatics and
Computational Biology.
2006 Reif P.M., Moore J.H. Visual analysis of statistical results from microarray studies of
human cancer. Oncology Reports, 15(5). [Cover Article].
2005 Wilke R., Reif P.M., Moore J.H. Combinatorial pharmacogenetics. Nature Reviews Drug
Discovery, 4(11).
2005 White B.C., Gilbert J., Reif P.M., Moore J.H. A statistical comparison of grammatical
evolution strategies in the domain of human genetics. IEEE Congress on Evolutionary
Computation, 6(2).
2005 Reif P.M., Pudek S.M., Shaffer C.M., Wang J., Moore J.H. Exploratory Visual Analysis of
Pharmacogenomic Results. Biocomputing, 9th ed.
2004 Reif P.M., White B.C., Moore J.H. Integrated Analysis of Genetic, Genomic, and Proteomic
Data. Expert Reviews in Proteomics, 1 (1).
2003 Reif P.M., White B.C., Olsen N.J., Aune T.A., Moore J.H. Complex function sets improve
symbolic discriminant analysis of microarray data. In: Cantu-Paz, E. et al. (eds.) Lecture
Notes in Computer Science, 2724.
Submitted:
2009 Knudsen T., Houck K., Judson R.S., Singh A.V., Mortensen H.A., Reif P.M., Pix P.J., Kavlock
R.J. Biochemical Activities of 320 ToxCast Chemicals Evaluated Across 239 Functional
Targets.
2009 Gallagher J., Reif P.M., Hudgens E., Heidenfelder B.N., Neas L, Williams A., Harkema J.,
Hester S., Edwards S.E., Cohen-Hubal E. Integration of Exposure, Effects, and
Susceptibility Data to Improve the Predictive Value of Biomarkers for Asthmatic Children.
2009 Martin M.T., Pix P.J., Judson R.S., Kavlock R.J., Reif P.M., Richard A.M., Rotroff P.M.,
Makarov S., Romanov S., Medvedev A., Houck K. Assessment of the Impact of
Environmental Chemicals on Key Transcription Regulators and Correlation to Toxicity
Endpoints.
2009 Judson R.S., Houck K., Kavlock R.J., Knudsen T., Martin M.T., Mortensen H.M., Reif P.M.,
Richard A.M., Rotroff P.M., Shah I., Pix P.J. Predictive In Vitro Screening of Environmental
Chemicals - The ToxCast Project.
2009 Judson R.S., Houck K., Kavlock R.J., Knudsen T., Martin M.T., Reif P.M., Rotroff P.M., Shah I.,
Pix P.J. Predicting Rodent and Human Chemical Carcinogenicity Using In Vitro Assays.
PHS 398/2590 (Rev. 09/04, Reissued 4/2006) Page 3 (of 4) (Updated August 2009)
Previous I TOC
-------
Principal Investigator/Program Director (Last, First, Middle): Reif, David M.
2009 Rotroff D.M., Ferguson S.S., Beam A.L, DixD.J., Farmer A., Freeman K.M., Houck K., Judson
R.S., Reif P.M., LeCluyse E.L. Xenobiotic Metabolizing Enzyme and Transporter Gene
Expression in Primary Cultures of Human Hepatocytes Modulated by ToxCast Chemicals.
In Preparation:
2009 Reif P.M., Heidenfelder B.N., Gallagher J., Hudgens E., Neas L, Cohen-Hubal E., Edwards
S.E. Integrating demographic, clinical, and environmental exposure information to
identify genomic biomarkers associated with subtypes of childhood asthma.
2009 Reif P.M., Cohen-Hubal E., Hudgens E., Heidenfelder B.N., Edwards S.E., Gallagher J.
Genetic associations with subtypes of childhood asthma.
2009 Williams-Pevane C., Gallagher J., Reif P.M., Cohen-Hubal E, Heidenfelder B.N., Harkema J.,
Edwards S.E. Comparing gene expression patterns in blood and lung tissue of
immunologically-challenged rats exposed to concentrated airborne particulates.
2009 Elloumi F., Judson R.S., Pix P.J., Shah I., Knudsen T., Reif P.M., Singh A.V., Li Z., Wright F.A.
Deriving Toxicogenomics Pathway-based Concentration Response Profiles.
2009 Reif P.M., Pix P.J., Houck K., Martin M.T., Judson R.S., Kavlock R. J. Biological profiling of
endocrine related effects of chemicals using ToxCast.
C. SELECTED PRESENTATIONS
2009 "Summary of Approaches and Predictions", ToxCast Data Analysis Summit, U.S. EPA,
Research Triangle Park, NC, USA [seminar].
2008 "Pata integration adds etiological context to a childhood asthma gene expression study",
Genetics Department Seminar Series, North Carolina State University, Raleigh, NC, USA
[invited seminar].
2008 "Integrating multiple data sources to discriminate subtypes of childhood asthma", Visiting
Pulmonary Scholar Seminar Series, University of North Carolina, Chapel Hill, NC, USA
[invited seminar].
2008 "Integrating demographic, clinical, and environmental exposure information to identify genomic
biomarkers associated with subtypes of childhood asthma", International Society of Exposure
Analysis-International Society of Environmental Epidemiology (ISEA-ISEE) Joint Annual
Meeting, Pasadena, CA, USA [oral presentation].
2008 "Pata integration for the Mechanistic Indicators of Childhood Asthma Study", Human Studies
Division Seminar Series, U.S. EPA, Chapel Hill, NC, USA [seminar].
2007 'Detection and characterization of gene-gene and gene-environment interactions in common
human diseases and complex clinical endpoints", Therapeutic Applications of Computational
Biology and Chemistry (TACBAC), Wellcome Trust Conference Centre, Cambridge, UK
[invited seminar].
2006 "Genetic factors associated with adverse events following smallpox vaccination", Vanderbilt
University Genetics Retreat, The Hermitage, TN, USA [oral presentation].
2006 "Integrated analysis of genetic and proteomic data", Vanderbilt University Genetics Interest
Group, Nashville, TN, USA [seminar].
2005 "Exploratory Visual Analysis", Pacific Symposium on Biocomputing (PSB), Lihue, HI, USA
[oral presentation].
2003 "Symbolic Piscriminant Analysis of microarray data using complex function sets", Genetic and
Evolutionary Computation Conference (GECCO), Chicago, IL, USA [oral presentation].
D. WEBSITES
http://epa.gov/ncct/index.html
www.epistasis-list.org
www4.stat.ncsu.edu/~dmreif/Site/ST610A.html
PHS 398/2590 (Rev. 09/04, Reissued 4/2006) Page 4 (of 4) (Updated August 2009)
Previous I TOC
-------
Principal Investigator/Program Director (Last, First, Middle): Richard, Ann M.
BIOGRAPHICAL SKETCH
NAME
Ann M. Richard
eRA COMMONS USER NAME
NA
POSITION TITLE
Research Chemist
EDUCATION/TRAINING (Begin with baccalaureate or other initial professional education, such as nursing, dnd include pOStdOCtOTdl
INSTITUTION AND LOCATION
State University of New York at Oswego, NY
University of North Carolina at Chapel Hill, NC
DEGREE
(if applicable)
B.A./B.S.
Ph.D.
YEAR(s)
1978
1983
FIELD OF STUDY
Math/Chemistry
Physical Chemistry
A. POSITIONS and HONORS
Research and Professional Experience:
2005-present Principal Investigator, NCCT, US EPA, RTP, NC
1999-2005 Research Chemist, Molecular Toxicology Branch, ECD/NHEERL, US EPA, RTP, NC
2001-2002 Acting Chief, Molecular Toxicology Branch, ECD, NHEERL, US EPA, RTP, NC
1997-1999 Res. Chemist, Biochemistry Pathobiology Branch, ECD, NHEERL, US EPA, RTP, NC
1987-1997 Res. Chemist, Carcinogenesis Metabolism Branch, HERL, US EPA, RTP, NC
Professional Societies and Affiliations:
QSAR & Modeling Society, Genetics & Environmental Mutagen Society
Honors and Awards:
NHEERL Strategic Goal 4: Leadership in the Scientific Community, Plan Team Award, 2000
NHEERL Special Award for Developing Process for Advancing NHEERL Genomics & Proteomics Program
Superior Achievement Cash Awards, 2001, 2002, 2003, 2005, 2006, 2007, 2008, 2009
NHEERL Strategic Goal 5 Award: Future Issues, Genomics, Proteomics & Bioinf. Program, 2003
CompTox Program 1-yr Augmented Funding Award for DSSTox Database Project Expansion, 2004-2005
CompTox Program New-Start Research Award for CEBS/DSSTox Collaboration Project, 2005-2008
ORD Science & Technology Achievement Award, Level III, 2006
ORD Communication Award, 2006
ORD Science & Technology Achievement Awards, Level II, Level III, and Honorable Mention, 2007
EPA Office of Environmental Information Award: Improvements in Computational Toxicology, 2008
Selected invitations at National & International Symposia:
ADMET-2 Conference, San Diego.CA, 2/2005
Soc. of Toxicology, Public Database Resources, New Orleans, LA, 3/2005
ILSI - HESI Developmental Toxicity Database Workgroup Meeting, Washington, DC, 3/2005
Chemoinformatics Symposium, NC Central University, Durham, NC, 4/2005
NSF International Health Advisor Board Meeting, Ann Arbor, Ml, 4/2005
National Cancer Institute Workshop on Government Databases, Fredrick, MD, 7/2005
American Chemical Soc. Nat. Mtg., International Science Policy Symposium, Washington, DC, 8/2005
Int. Congress of Environ. Mutagens, San Francisco, CA, 9/2005
Data in Life Science Workshop, European Science Agency, Milan, Italy, 11/2005
ISRTP Workshop, Progress & Barriers to Incorporating Alt Tox Methods in the US, Baltimore, MD, 11/2005
Leadscope In Silico Toxicology Consortium, Bethesda, MD, 4/2006
EU Project on the Collection & Evaluation of (Q)SAR Models for Mutag. and Carcinog.; Rome, Italy, 6/2006
LHASA International Collaborative Group Meeting, Washington DC, 9/2006
EPA STAR Graduate Fellowship Conference, Washington DC, 9/2006
Vanderbilt University, Institute for Chemical Biology, Dept. Seminar, Nashville, TN, 2/2007
1
Previous
TOC
-------
Principal Investigator/Program Director (Last, First, Middle): Richard, Ann M.
Soc. for Biomolecular Sci., Toxicity Profiling Using HTS and HCS Technologies, Montreal, Canada, 4/2007
Teratology Society Annual Meeting, Pittsburg, PA, 6/2007
Comp. Methods in Toxicology & Pharmacology: Integrating Internet Resources, Moscow, Russia, 9/2007
Int. Society of Exposure Analysis Annual Meeting, Durham, NC 10/2007
eCheminfo Workshop on Predictive ADMET & Toxicology, Philadelphia, PA, 10/2007
Genetics & Environmental Mutagenesis Society Meeting, Chapel Hill, NC, 10/2007
Health Canada's New Substances Assessment & Control Bureau, Ottawa, Canada, 11/2007
SCARLET Workshop on In Silico Methods for Carcinogenicity and Mutagenicity, Milan, Italy, 4/2008.
QSAR & Omics Technologies and Systems Biology, Uppsala, Sweden, 9/2008.
eCheminfo Workshop on Predictive ADMET & Toxicology, Philadelphia, PA, 10/2008.
LHASA Limited Symposium: New Horizons in Toxicity Prediction, Cambridge, UK, 12/2008.
Ohio State University, Guest Lecturer, Molecular Informatics, Columbus, OH, 3/2009.
Proctor & Gamble, Computational Toxicology Seminar Series, Cincinnati, OH, 3/2009.
Joint Research Centre, Inst. Health & Consumer Protection, Ispra, Italy, 5/2009.
US FDA Center for Food Safety & Applied Nutrition, Comp. Tox. Workshop, Silver Spring, MD, 7/2009.
American Chemical Soc., Div. Chem. Inf., Chemical Text Mining & Databases, Wash. DC, 8/2009.
Int. Congress on Environ. Mutag., New Data Initiatives & Predic. Tox., Florence, Italy, 8/2009.
Selected Expert Committees/Advisory Panels/Organizing Committees:
Editorial Board, Mutation Research, 1994-present.
Editorial Board, Chemical Research in Toxicology, 1999-2002, 2003-2005, 2008-present.
Editorial Board, SAR & QSAR in Environmental Research, 2008-present.
Manuscript reviewer: Bioinformatics, Carcinogenesis, Chem.I Res. Tox., Chemico-Biol.l Interact.,
Chemosphere, Environ. Molec. Mutag., Environ. Health Perspec.s, Environ.I Tox. Chem., J. Chem. Info.
Comp. Sci., J. Amer. Chem. Soc., J. Comp. Chem., Mutat. Res, SAR QSAR Environ. Tox., Sci. Total
Environ., Toxicology, Tox. Sci., Regul. Tox. Pharmacol.
Advisory committee for Predictive Toxicity Challenge I, Freidburg, Germany, 2001
ILSI-Health Effects Sci. Inst. Working Group on SAR Toxicity Database, 2002
Organizing Committee, Prediction of AcuteToxicity Workshop, Mclean, VA, May 5-7, 2003
OECD Workgroup on Use & Regulatory Acceptance of QSARs, 2003-present
Organizing Committee, ADMET I Conference, San Diego, Feb. 11-13, 2004
Organizing Committee, International Congress on Environ. Mutagenesis, San Francisco, Sept., 2004
LeadScope LIST Workgroup for Implementation of ToxML standard ontologies, Mar 2004-2006
Review Committee, TERA - Health Canada QSAR Advisor Review Board, Apr 2005
I LSI Working Group on Prediction of Developmental Toxicity, 2002-present
LHASA VITIC SAR Database Advisory Committee, 2005
National Toxicology Program High Throughput Screening Assays Workshop, Arlington, VA, 12/05
EU Project on Evaluation of (Q)SAR Models for Mutagenicity & Carcinogenicity; Rome, Italy, June 22-23, 2006
Developmental Neurotoxicity Database Workshop, RTP, NC, Nov 14-15, 2006
Organizing Committee, Comp. Methods in Toxicology & Pharmacology, Moscow, Russia, 9/2007
Selected Assistance/Advisory Support to the Agency:
ORD Strategy GoalS Team - Leader in Environmental Research, 1999
NHEERL Strategy GoalS Team - Leader in Environmental Research, 1999
NHEERL Multi-Year Implementation Planning Committee for Goal 4 - Safe Communities, 2001
NHEERL Genomics and Proteomics Steering Committee, 2001-2002
NHEERL Genomics & Proteomics Committee Member, Bioinformatics Coordinator, 2002-2004
EPA Goal 4 Safe Pesticides/Safe Products Multi-Year Plan Steering Committee, 2003-present
EPA Research & Science Architecture (RST) Target Workgroup, 2003-present
Chemoinformatics Communities of Practice Coordinator, 2005-present
EPA Hiring Committee Lead for Senior Title 42 NCCT Bioinformatics Hire, 2006
EPA Science Connector Workgroup, NCCT Representative, 2007
Tox21 - EPA Chemical Working Group Lead, 2009.
Previous I TOC
-------
Principal Investigator/Program Director (Last, First, Middle): Richard, Ann M.
B. SELECTED PUBLICATIONS (2004 to present, 26 out of 60 total).
Richard, A.M. DSSTox Website launch: Improving public access to databases for building structure-toxicity
prediction models. Preclinica, 2:103-108, 2004.
Balu, N., Padgett, W.T., Lambert, G., Swank, A.E., Richard, A.M., Nesnow, S. Identification and
characterization of novel stable deoxyguanosine and deoxyadenosine adducts of benzo[a]pyrene-7,8-
quinone from reactions at physiological pH. Chem. Res. Toxicol. 17:827-838, 2004.
Kundu, B., Richardson, S.D., Swartz, P.O., Matthews, P.P., Richard, A.M., DeMarini, D.M. Mutagenicity in
Salmonella of Halonitromethanes: A Recently Recognized Class of Disinfection By-products in Drinking
Water. Mutat. Res. 562:39-65, 2004.
Kundu, B., Richardson, S.D., Granville, C.A., Shaughnessy, D.T., Hanley, N.M., Swartz, P.O., Richard, A.M.,
DeMarini, D.M. Comparative Mutagenicity of Halomethanes and Halonitromethanes in Salmonella TA100:
Structure-Activity Analysis and Mutation Spectra. Mutat. Res. 554:335-350, 2004.
Julian, E., Wllhite, C.C., Richard, A.M., DeSesso, J.M. Challenges in Constructing Statistically-Based SAR
Models for Developmental Toxicity. Birth Defects Research Part A, 70:902-911, 2004.
Granville, C.A., Ross, M.K., Tornero-Valez, R., Hanley, N.M., Grindstaff, R.D., Gold, A., Richard, A.M.,
Funasaka, K., Tennant, A.H., Kligerman, A.D., Evans, M.V., DeMarini, D. Genotoxicity and metabolism of
the source-water contaminant 1,1-dichloropropene: Activation by GSTT1-1. Mutat. Res. 572:98-112, 2005.
Fostel, J., Choi, D., Zwickl, C., Morrison, N., Rashid, A., Hasan, A., Bao, W., Richard, A., long, W., Bushel, P.,
Brown, R., Bruno, M., Cunningham, M., Dix, D., Eastin, W., Frade, C., Garcia, A., Heinloth, A., Irwin, R.,
Madenspacher, J., Merrick, A., Papoian, T., Paules, R., Rocca-Serra, P., Sansone, A., Stevens, J., Tomer,
K., Yang, C., Waters, M. Chemical Effects in Biological Systems - Data Dictionary (CEBS-DD): A
Compendium of Terms for the Capture and Integration of Biological Study Design Description,
Conventional Phenotypes and 'Omics Data. Tox. Sci., 88:585-601, 2005.
Hunter, S.E., Rogers, E., Blanton, M., Richard, A.M., Chernoff, N. Bromochloro-haloacetic acids: Effects on
mouse embryos in vitro and QSAR considerations, Birth Defects Research Part A, 21:260-266, 2005.
Yang, C., Richard, A.M., Cross, K.P. The art of data mining the minefields of toxicity databases to link
chemistry to biology. Curr Comput-Aided Drug Design, 2(2):135-150, 2006.
Richard A.M., Gold, L.S., Nicklaus, M.C. Chemical structure indexing of toxicity data on the internet: Moving
towards a flat world. Current Opinion in Drug Discovery & Develop., 9(3):314-325, 2006.
Richard A.M. Future of Predictive Toxicology: An Expanded View of "Chemical Toxicity" - Future of Toxicology
Perspective. Chem. Res. Toxicol., 19:1257-1262, 2006.
Dix, D.J., Houck, K.A., Martin, M.T., Richard, A.M., Setzer, W., Kavlock, R.J. The ToxCast program for
prioritizing toxicity testing of environmental chemicals. Tox. Sci., 95:5-12, 2007
Benigni, R., Netzeva, T.I., Benfenati, E., Bossa, C., Franke, R., Helma, C., Hulzebos, E., Marchant, C.,
Richard, A., Woo, Y.T., Yang, C. The expanding role of predictive toxicology: an update on the (Q)SAR
models for mutagens and carcinogens. J. Environ. Sci. Health C, 25:53-97, 2007.
Benigni, R., Bossa, C. Richard, A.M., Yang, C. A novel approach: Chemical relational databases, and the role
of the ISSCAN database on assessing chemical carcinogenicitiy. Annali dell' Inst. Super, di Sanita, 2007.
Richard, A., Yang, C., Judson, R. Toxicity Data Informatics: Supporting a New Paradigm for Toxicity
Prediction. Tox. Mech. Meth., 18:103-118, 2008.
Yang, C., Arnby, C.H., Arvidson, K., Aveston, S., Benigni, R., Benz, R.D., Boyer, S., Contrera, J., Dierkes, P.,
Han, X., Jaworska, J., Kemper, R.A., Kruhlak, N.L., Matthews, E.J., Rathman, J.F., Richard, A.M.
Understanding Genetic Toxicity through Data Mining: The Process of Building Knowledge by Integrating
Multiple Genetic Toxicity Databases, Tox. Mech. Meth., 18:277-295, 2008.
Kavlock, R.J., Ankley, G., Blancato, J., Breen, M., Conolly, R., Dix, D., Houck, K., Hubal, E., Judson, R.,
Rabinowitz, J., Richard, A., Setzer, R.W., Shah, I., Villeneuve, D., Weber, E. Computational Toxicology-A
State of the Science Mini Review, Tox. Sci, 103:14-27, 2008.
Zhu, H., Rusyn, I., Richard, A., Tropsha, A. The Use of Cell Viability Assay Data Improves the Prediction
Accuracy of Conventional Quantitative Structure Activity Relationship Models of Animal Carcinogenicity,
Environ. Health Perspec., 116:506-513, 2008.
Judson, R., Richard, A., Dix, D., Houck, K., Elloumil, F., Martin, M., Cathey, T., Transue, T., Spencer, R., Wolf,
M. ACToR - Aggregated Computational Toxicology Resource, Toxicol. Appl. Pharmacol., 233:7-13, 2008.
Hubal, E., Richard, A., Shah, I., Gallagher, J., Kavlock, R., Blancato, J., Edwards, S. Exposure Science and
the US EPA National Center for Computational Toxicology, J. Expos. Sci. Environ. Epidem. 1-6, 2008.
Previous I TOC
-------
Principal Investigator/Program Director (Last, First, Middle): Richard, Ann M.
BenfenatM, E., Benigni, R., DeMarini, D.M., Helma, C., Kirkland, D., Martin, T.M., Mazzatorta, P., Meunier, J.-
R., Ouedraogo-Arras, G., Richard, A.M., Schilter, B., Schoonen, W.G.E.J., Snyder, W.G.E.J., Yang, C.,
Young, D.M. Predictive models for carcinogenicity and mutagenicity: frameworks, state-of-the-art and
perspectives, J. Environ. Sci. Health C, 27:57-90, 2009.
Judson, R., Richard, A., Dix, D.J., Houck, K., Martin, M., Kavlock, R., Dellarco, V., Henry, T., Holderman, T.,
Sayre, P., Tan, S., Carpenter, T., Smith, E. The toxicity data landscape for environmental chemicals,
Environ. Health Perspec., 117:685-695, 2009.
Zhu, H., Ye, L, Richard, A., Golbraikh, A., Wright, F.A., Rusyn, I., Tropsha, A. A novel two-step hierarchical
quantitative structure-activity relationship modeling work flow for predicting acute toxicity of chemicals in
rodents, Environ. Health Perspec., 117:1257-1264, 2009.
Williams-Devane, C., Wolf, M.A., Richard, A.M. DSSTox chemical-index files for exposure-related experiments
in ArrayExpress and Gene Expression Omnibus: Enabling toxico-chemogenomics data linkages,
Bioinformatics, 25:692-694, 2009.
Wlliams-Devane, C., Wolf, M.A., Richard, A.M. Towards a public toxicogenomics capability for supporting
predictive toxicology: Survey of current resources and chemical indexing of experiments in GEO and
ArrayExpress, Tox. Sci., 109:358-371, 2009.
Knight, A.W., Little, S., Houck, K., Dix, D., Judson, R., Richard, A, McCarroll, N., Ackerman, G., Yang, C.,
BirrelH, L., Walmsley, R.M. Evaluation of high-throughput genotoxicity assays used in profiling the US EPA
ToxCast™ Chemicals, Reg. Toxicol. Pharmacol. doi:10.1016/j.yrtph.2009.07.004, 2009.
DSSTOX WEBSITE & PUBLICATIONS
Richard, A.M., EPA DSSTox Website, launched March 2004; most recent update March 2009:
https://www.epa.gov/ncct/dsstox/
Richard, A.M., Transue, T., EPA DSSTox Structure-Browser, launched August 2007; v2.0, 2008:
http://www.epa.gov/dsstox structurebrowser/
Published DSSTox Data Files: http://www.epa.gov/ncct/dsstox/DataFiles.html (4 of 14 published DSSTox Data
Files listed below):
i. Gold, L.S., Slone, T.H., Williams, C.R., Burch, J.M., Stewart, T.W., Swank, A.E., Beidler, J., Richard,
A.M. (2003) DSSTox Carcinogenic Potency Database Summary Tables for Rats and Mice, Hamsters,
Dogs, and Non-Human Primates, SDF Files and Documentation, Last updated 2008:
CPDBAS_v5d_1547_20Nov2008; http://www.epa.gov/ncct/dsstox/sdf cpdbas.html
ii. Backus, G.S, Wolf, M.A., Burch, J., Richard, A.M. (2006) DSSTox EPA Integrated Risk Information
System (IRIS) Toxicity Review Data: SDF File and Documentation, Last updated 2008:
IRISTR_v1b_544_15Feb2008, www.epa.gov/ncct/dsstox/sdf iristr.html
iii. Houck, K., Dix, D., Judson, R., Martin, M., Wolf, M., Kavlock, R.,Richard, A.M. (2007) DSSTox EPA
ToxCast High Throughput Screening Testing Chemicals Structure-Index File: SDF File and
Documentation, Last updated 2009: TOXCST_v3a_320_12Feb2009,
http://www.epa.gov/ncct/dsstox/sdf toxcst.html
iv. Williams-Devane, C.R., Wolf, M.A. Richard, A.M. (2008) DSSTox European Bioinformatics Institute
(EBI) ArrayExpress Repository for Gene Expression Data (ARYEXP and ARYEXP_Aux): SDF Files
and Documentation, Last Updated: ARYEXP_v2a_958_06Mar2009,
ARYEXP_Aux_v2a_2556_06Mar2009, www.epa.gov/ncct/dsstox/sdf arvexp.html
v. Williams-Devane, C.R., Wolf, M.A., Richard, A.M. (2008) DSSTox National Center for Biotechnology
Information (NCBI) Gene Expression Omnibus (GEO) Series Experiments (GEOGSE and
GEOGSE_Aux): SDF Files and Documentation, Last Updated: GEOGSE_v2a_1179_09Mar2009,
GEOGSE_Aux_v2a_2700_09Mar2009, www.epa.gov/ncct/dsstox/sdf geogse.html
vi. S. Laws, J. Kariya, M. Wolf, and A.M. Richard (2009) DSSTox EPA Estrogen Receptor Ki Binding
Study (Laws et al.) Database - (KIERBL): SDF File and Documentation, Launch version:
KIERBL_v1a_278_17Feb2009, www.epa.gov/ncct/dsstox/sdf kierbl.html
Previous I TOC
-------
Principal Investigator/Program Director: Rusyn, Ivan
BIOGRAPHICAL SKETCH
Provide the following information for the key personnel and other significant contributors in the order listed on Form Page 2.
Follow this format for each person. DO NOT EXCEED FOUR PAGES.
NAME
Ivan Rusyn
eRA COMMONS USER NAME
I RUSYN
POSITION TITLE
Associate Professor
EDUCATION/TRAINING (Begin with baccalaureate or other initial professional education, such as nursing, and include postdoctoral training.)
INSTITUTION AND LOCATION
Ukrainian State Med. University, Kiev, Ukraine
Inst. Physiol. Chem. 1, University of Dusseldorf
University of North Carolina at Chapel Hill
University of North Carolina at Chapel Hill
Massachusetts Institute of Technology
DEGREE
(if applicable)
M.D. (w. Hons.)
Postdoctoral
Ph.D.
Postdoctoral
Postdoctoral
YEAR(s)
1994
1995-1996
2000
2000-01
2001-02
FIELD OF STUDY
Medicine
Free radical biology
Toxicology
DNA damage &
repair
Toxicogenomics
Professional Positions:
2002 - Associate (since 2007) and Assistant (2002-2007) Professor, Department of Environmental Sciences &
Engineering, University of North Carolina at Chapel Hill
2002 - Associate (since 2007) and Assistant (2002-2007) Professor, Associate Director (since 2007), Curriculum
in Toxicology, UNC-Chapel Hill
2005 - Scientific Co-Director, Carolina Environmental Bioinformatics Research Center
2003 - Member, Lineberger Comprehensive Cancer Center, UNC-Chapel Hill
2003 - Member, Bowles Center for Alcohol Studies, UNC-Chapel Hill
2003 - Member, Center for Environmental Health & Susceptibility, UNC-Chapel Hill
2002 - Member, Carolina Center for Genome Sciences, UNC-Chapel Hill
2001 - 2002 Res. Assoc., Biological Engineering Division, Massachusetts Inst. of Technology
2000 - 2001 Res. Fellow, Dept. of Environmental Sci. & Engineering, UNC-Chapel Hill
1996 - 2000 Res. Assist., Curriculum in Toxicology, UNC-Chapel Hill
1995 - 1996 Guest Researcher & Fellow of German Academic Exchange Service (DAAD),
Institute for Physiological Chemistry I, University of Dusseldorf, Germany
1994- 1995 Intern, Department of Otolaryngology, Kiev Regional Clinical Hospital, Ukraine
Significant Professional Activities:
2009-present Member, Standing Committee on Use of Emerging Science for Environmental Health Decisions, Board on
Life Sciences, Board on Environmental Studies & Toxicology, National Academies, Washington, DC
2008-present Member, Committee on Tetrachloroethylene, Board on Environmental Studies and Toxicology, National
Research Council of the National Academies, Washington, DC
2008-present Member, Board on Publications, Society of Toxicology, Reston, VA
2006-2007 Member, Working Group on IARC Monograph Volume 96 on "Alcoholic beverage consumption, acetaldehyde
and urethane" International Agency for Research on Cancer (IARC), Lyon, France
2006-2008 Scientific Program Committee, Society of Toxicology, Reston, VA
2005-2008 Expert Consultant, 12th Report on Carcinogens, NTP/NIEHS, Research Triangle Park, NC
Honors and Awards:
2008
2002
2000
2000
2000
2000
2000
Achievement Award, Society of Toxicology
Transition to Independent Position Award, NIEHS
Individual Postdoctoral National Research Service Award, NIEHS
Leon & Bertha Golberg Memorial Postdoctoral Fellowship, UNC-Chapel Hill
AACR - Bristol Myers Squibb Oncology Young Investigator Scholar Award
Carl C. Smith Mechanisms Specialty Section Award, Society of Toxicology
Young Investigator Award, Society for Free Radical Research International
1
Previous
TOC
-------
Principal Investigator/Program Director: Rusyn, Ivan
1998/99/2001 Young Investigator Award, The Oxygen Society
1995-96 Research Fellowship, German Academic Exchange Service (DAAD)
1994 First Class Honors Diploma, Ukrainian State Medical University, Kiev, Ukraine
Recent Publications (from 82 total):
Gatti, D.M., Harrill, A.M., Wright, F.A., Threadgill, D.W., and Rusyn, I. Replication and narrowing of gene expression
quantitative trait loci using inbred mice. Ma mm Genome In Press.
Harrill, A.M., Watkins, P.B., Su, S., Ross, P.K., Harbourt, D.E., Stylianou, I.M., Boorman, G.A., Russo, M.W., Sackler,
R.S., Harris, S.C., Smith, P.C., Tennant, R., Bogue, M., Paigen, K., Harris, C., Contractor, T., Wiltshire, T., Rusyn, I.,
and Threadgill, D.W. Mouse population-guided resequencing reveals that variants in CD44 contribute to
acetaminophen-induced liver injury in humans. Genome Res In Press.
Kim, S., Collins, L.B., Boysen, G., Swenberg, J.A., Gold, A., Ball, L.M., Bradford, B.U., and Rusyn, I. Liquid
chromatography electrospray ionization tandem mass spectrometry analysis method for simultaneous detection of
trichloroacetic acid, dichloroacetic acid, S-(1,2-dichlorovinyl)glutathione and S-(1,2-dichlorovinyl)-L-cysteine.
Toxicology 262:230-238, 2009.
Zhu, H., Ye, L, Richard, A., Golbraikh, A., Rusyn, I., and Tropsha, A. A novel two-step hierarchical quantitative structure
activity relationship modeling workflow for predicting acute toxicity of chemicals in rodents. Envr Health Persp
117:1257-1264,2009.
Harrill, A.H., Ross, P.K., Threadgill, D.W., and Rusyn, I. Population-based discovery of toxicogenomics biomarkers for
hepatotoxicity using a laboratory strain diversity panel. Toxicol Sci 110:235-243, 2009.
Pogribny, I.P., Tryndyak, V.P., Bagnyukova, T.V., Melnyk, S., Montgomery, B., Ross, S.A., Latendresse, J.R., Rusyn, I.,
and Beland, F.A. Hepatic epigenetic phenotype predetermines individual susceptibility to hepatic steatosis in mice fed
a lipogenic methyl-deficient diet. JHepato/51: 176-186, 2009.
Kim, S., Kim, D., Pollack, G., Collins, L., Rusyn, I. Pharmacokinetic analysis of trichloroethylene metabolism in male
B6C3F1 mice: Formation and disposition of trichloroacetic acid, dichloroacetic acid, S-(1,2-dichlorovinyl)glutathione
and S-(1,2-dichlorovinyl)-L-cysteine. Toxicol Appl Pharmacol 238: 90-99, 2009.
Ross, P.K., Woods, C.G., Bradford, B.U., Kosyk, O., Gatti, D.M., Cunningham, M.L., and Rusyn, I. Time-course
comparison of xenobiotic activators of CAR and PPARalpha in mouse liver. Toxicol Appl Pharmacol 235:199-207,
2009.
Gatti, D.M., Sypa, M., Rusyn, I., Wright, F.A., and Barry, W.T. SAFEGUI: Resampling-based tests of categorical
significance in gene expression data made easy. Bioinformatics 25:541-542, 2009.
Gatti, D.M., Shabalin, A.A., Lam, T.C., Wright, F.A., Rusyn, I., and Nobel, A.B. FastMap: Fast eQTL mapping in
homozygous populations. Bioinformatics 25: 482-489, 2009.
Harrill, A.H., and Rusyn, I. Systems biology and functional genomics approaches for the identification of cellular
responses to drug toxicity. Expert Opin Drug Metab Toxicol 4:1379-1389, 2008.
Stotts, D., Lee, K., and Rusyn, I. Supporting computational systems science: Genomic analysis tool federations using
aspects and AOP. In: Mandoiu, I., Sunderraman, R., and Zelikovsky A. (Eds.): Bioinformatics Research and
Applications, Proceedings of the 4th International Symposium ISBRA 2008, Springer Berlin/Heidelberg, LNBI 4983,
pp. 457-468, 2008.
Tsuchiya, M., Kono, H., Matsuda, M., Fujii, H., and Rusyn, I. Protective effect of Juzen-taiho-to on hepatocarcinogenesis
is mediated through the inhibition of Kupffer cell-induced oxidative stress. Int J Cancer 123:2503-2511, 2008.
Bradford, B.U., O'Connell, T.M., Han, J., Kosyk, O., Shymonyak, S., Ross, P.K., Winnike, J., Kono, H., and Rusyn, I.
Metabolomic profiling of a modified alcohol liquid diet model for liver injury in the mouse uncovers new markers of
disease. Toxicol Appl Pharmacol 232: 236-243, 2008.
Pogribny, I.P., Tryndyak, V.P., Boureiko, A., Melnyk, S., Bagnyukova, T.V., Montgomery, B., and Rusyn, I. Mechanisms
of peroxisome proliferator-induced DMA hypomethylation in rat liver. Mutat Res 644:17-23, 2008.
Han, J., Danell, R.M., Patel, J.R., Gumerov, D.R., Scarlett, C.O., Speir, J.P., Parker, C.E., Rusyn, I., Zeisel, S., and
Borchers, C.H. Towards high throughput metabolomics using ultrahigh field Fourier transform ion cyclotron resonance
mass spectrometry. Mefabo/oin/cs4:128-140, 2008.
Zhu, H., Rusyn, I., Richard, A., and Tropsha A. The use of cell viability assay data improves the prediction accuracy of
conventional quantitative structure activity relationship models of animal carcinogenicity. Envr Health Persp 116:506-
513,2008.
Pogribny, I.P., Rusyn, I. and Beland, F.A. Epigenetic aspects of genotoxic and non-genotoxic hepatocarcinogenesis:
Studies in rodents. Environ Mol Mutagen 49:9-15, 2008.
Rusyn, I., Fry, R.C., Begley, T.J., Klapacz, J., Svensson, J.P., Ambrose, M., and Samson, L.D. Transcriptional networks
in S. cerevisiae linked to an accumulation of base excision repair intermediates. PLoS OW£2:e1252, 2007.
Woods, C.G., Kosyk, O., Bradford, B.U., Ross, P.K., Burns, A.M., Cunningham, M.L., Qu, P., Ibrahim, J.G. and Rusyn I.
Gene expression in mouse liver reveals a temporal shift in molecular pathways that mediate effects of peroxisome
proliferators. Toxicol Appl Pharmacol 225.2Q7-277, 2007.
Previous I TOC
-------
Principal Investigator/Program Director: Rusyn, Ivan
Pogribny, IP., Tryndyak, V.P., Woods, C.G., Witt, S.E., and Rusyn, I. Epigenetic effects of the continuous exposure to
peroxisome proliferator WY-14,643 in mouse liver are dependent upon peroxisome proliferator activated receptor a.
Mutat Res 625:62-71, 2007.
Peffer, R., Moggs, J.G., Pastoor, T., Currie, R.A., Wright, J., Millburn, G., Waechter, F., and Rusyn, I. Mouse liver effects
of Cyproconazole, a triazole fungicide: Role of the constitutive androstane receptor. Tox/co/Sc/99:315-325, 2007.
Woods, C.G., Burns, A.M., Bradford, B.U., Ross, P.K., Kosyk, O., Swenberg, J.A., Cunningham, M.L., and Rusyn I. WY-
14,643-induced cell proliferation and oxidative stress in mouse liver are independent of NADPH oxidase. Toxicol Sci
98:366-374, 2007.
Beyer, R.P., et al. Multi-center study of acetaminophen hepatotoxicity reveals the critical importance of biological
endpoints in genomic analyses. Toxicol Sci 99:326-337, 2007.
Gatti, D., Maki, A., Chesler, E.J., Kirova, R., Lu, L, Wang, J., Williams, R.W., Perkins, A., Langston, M.A., Threadgill,
D.W., and Rusyn, I. Genome-level analysis of genetic regulation of liver gene expression networks. Hepatology
46:548-557, 2007.
Roberts, A., McMillan, L., Wang, W., Parker, J., Rusyn, I., Threadgill, D. Inferring missing genotypes in large SNP panels
using fast nearest-neighbor searches over sliding windows. B/o/nforinatfcs23:i401-07, 2007.
Woods, C.G., Vanden Heuvel, J.P., and Rusyn, I. Genomic profiling in nuclear receptor-mediated toxicity. Toxicol Pathol
35:474-494, 2007.
Maki, A., Kono, H., Gupta, M., Asakawa, M., Suzuki, T., Matsuda, M., Fujii, H., and Rusyn, I. Predictive power of
biomarkers of oxidative stress and inflammation in patients with hepatitis C virus-associated hepatocellular carcinoma.
X\nnSurgOnco/14:1182-90, 2007.
Hammond, L., Albright, C., He, L., Rusyn, I., Watkins, S.M., Doughman, S.D., Lemasters, J.J., and Coleman, R.A.
Increased oxidative stress is associated with balanced increases in hepatocyte apoptosis and proliferation in glycerol-
3-phosphate acyltransferase-1 deficient mice. Exp Mol Pathol 82:210-219, 2007.
Pogribny, I.P., Tryndyak, V.P., Muskhelishvili, L., Rusyn, I., and Ross, S.A. Methyl deficiency, alterations in global histone
modifications and carcinogenesis. J Nutr 137:2168-2228, 2007.
Yamashina, 8., Ikejima, K., Rusyn, I., and Sato, N. Glycine as a potent anti-angiogenic nutrient for tumor growth J
Gastroenterol Hepatol22:S62-S64, 2007.
Roberts, R.A., Ganey, P.E., Ju, C., Kamendulis, L.M., Rusyn, I., and Klaunig, J.E. Role of the Kupffer cell in mediating
hepatic toxicity and carcinogenesis. Toxicol Sci 96:2-15, 2007.
Woods, C.G., Burns, A.M., Maki, A., Bradford, B.U., Cunningham, M.L., Connor, H.D., Kadiiska, M., Mason, R.P., Peters,
J.M., and Rusyn, I. Sustained formation of a-(4-pyridyl-1-oxide)-/V-fe/?-butylnitrone radical adducts in mouse liver by
peroxisome proliferators is dependent upon peroxisome proliferator-activated receptor-a, but not NADPH oxidase.
Free Radio S/o/Med 42:335-342, 2007.
Rusyn, I., Peters, J.M., and Cunningham, M.L. Modes of action and species-specific effects of di-(2-ethylhexyl)phthalate
in the liver. Crit Rev Toxicol 36:459-479, 2006.
Powell, C.L., Kosyk, O., Ross, P.K., Schoonhoven, R., Boysen, G., Swenberg, J.A., Heinloth, A.M., Boorman, G.A.,
Cunningham, M.L., Paules, R.S., and Rusyn, I. Phenotypic anchoring of acetaminophen-induced oxidative stress with
gene expression profiles in rat liver. Toxicol Sci 93:213-222, 2006.
Kono, H., Woods, C.G., Maki, A., Connor, H.D., Mason, R.P., Rusyn, I., and Fujii, H. Electron spin resonance and spin
trapping technique provide direct evidence that edaravone prevents acute ischemia-reperfusion injury of the liver by
scavenging free radicals. Free Radio Res 40:579-588, 2006.
Powell, C.P., Kosyk, O., Bradford, B.U., Parker, J.S., Lobenhofer, EX., Denda, A., Uematsu, F., Nakae, D., and Rusyn, I.
Temporal correlation of pathology and DMA damage with gene expression in a choline deficient model of rat liver
injury. Hepato/ogy42:1137-1147, 2005.
Rusyn, I., Asakura, S., Li, Y., Kosyk, O., Koc, H., Nakamura, J., Upton, P.B., and Swenberg, J.A. Effects of ethylene
oxide and ethylene inhalation on DMA adducts, apurinic/apyrimidinic sites and expression of base excision DMA repair
genes in rat brain, spleen, and liver. DNA Repair 4:1099-1110, 2005.
Bammler, T., et al. Standardizing global gene expression analysis between laboratories and across platforms. Nat
Methods 2:351-356, 2005.
Bradford, B.U., Kono, H., Isayama, F., Kosyk, O., Wheeler, M.D., Akiyama, T.E., Bleye, L., Krausz, K.W., Gonzalez, F.J.,
Koop, D.R., and Rusyn, I. Cytochrome P450 CYP2E1, but not NADPH oxidase is required for ethanol-induced
oxidative DNA damage in rodent liver. Hepatology 41:336-344, 2005.
Ongoing Research Support:
R01 AA016258 8/06 - 7/10 0.50 academic month
NIH $431,307 1.00 summer month
Metabolomic and toxicogenetic study of ethanol toxicity
Previous I TOC
-------
Principal Investigator/Program Director: Rusyn, Ivan
The aim of this proposal is to define a "liver toxicity susceptibility state" in mouse liver in response to ethanol by combining
knowledge of toxicology, metabolomics, gene expression profiling and mouse genetics.
R01 ES15241 12/07-11/12 2.40 academic month
NIH $190,907
Bioengineering partnership to improve chemical hazard testing paradigms
This proposal will apply an integrative systems approach to: develop a 3D microscale mouse liver tissue bioreactor that
can be applied to high-throughput screening of chemicals; build, test and validate a quantitative structure-toxicity
relationship model that takes into account genetic diversity among individuals; and validate a fiscally sensible in vivo and
in vitro toxicity screening paradigm for a class of allylbenzene derivatives by producing knowledge anchored on the
genetic variability present within the population.
RD 833825 04/08-03/12 2.00 academic month
EPA STAR $817,943 0.88 summer month
Carolina Center for Computational Toxicology
The Center will develop complex predictive modeling solutions that span from mechanistic- to discovery-based efforts.
RD 832720 (Wright)9/06 - 8/10 0.60 academic month
US EPA Star $314,659
Carolina Environmental Bioinformatics Research Center
Project 3: Computational Infrastructure for Systems Toxicology (I. Rusyn, PI)
The objective of this proposal is to develop novel analytic and computational methods, create user-friendly tools to
disseminate the methods to the wide toxicology community, and enhance and advance the field of Computational
Toxicology.
P42 ES005948 (Swenberg) 04/06 - 03/10 1.50 academic month
NIEHS $219,825
Environmental exposure and effect of hazardous chemicals: Project #2 (I. Rusyn, PI)
The primary objective of the UNC Superfund Basic Research Program is to advance multidisciplinary research on
addressing scientific issues that underpin assessment human risk and the development of improved methods for
remediation of hazardous waste sites.
R01 ES012689 (Swenberg) 2/05 - 1/10 0.36 academic month
NIH/NIEHS $231,919
Adducts as Quantitative Markers of Butadiene Mutagenesis
The aim of this project is to study mutagenesis of butadiene and its metabolites in rodents and humans. New biomarkers
of butadiene exposure will be developed, and applied for research on the mechanisms of action.
R21 GM076059 (Tropsha) 6/06 - 5/10 0.36 academic month
NIH $207,353
Robust Computational Framework for Predictive ADME-Tox Modeling
This proposal seeks to establish a universally applicable and robust predictive ADME-Tox modeling framework based on
rigorous Quantitative Structure Activity/Property Relationships (QSAR/QSPR).
Previous I TOC
-------
Principal Investigator/Program Director (Last, First, Middle): Samet, James M.
BIOGRAPHICAL SKETCH
Provide the following information for the key personnel and other significant contributors in the order listed on Form Page 2.
Follow this format for each person. DO NOT EXCEED FOUR PAGES.
NAME
James M. Samet
eRA COMMONS USER NAME
POSITION TITLE
Senior Research Scientist
EDUCATION/TRAINING (Begin with baccalaureate or other initial professional education, such as nursing, and include postdoctoral
training.)
INSTITUTION AND LOCATION
University of Florida
University of North Carolina at Chapel Hill,
NC
University of North Carolina at Chapel Hill,
NC
Wake Forest University School of Medicine
University of North Carolina at Chapel Hill,
NC
DEGREE
(if applicable)
B.S.
M.S.
Ph D
r i i . L-/ .
Post Doc
Post Doc
YEAR(s)
1985
1990
1992
1 \J\J£-
1992-1994
1994-1996
FIELD OF STUDY
Microbiology and Cell
Science
Toxicology
Environmental
Sciences
Firo^anoiH
^IwWOdl IWIU
Biochemistry
Environmental Health
A. POSITIONS and HONORS
Research and Professional Experience:
2007-Present Senior Principal Investigator, Clinical Research Branch, Human Studies
Division, NHEERL
1997-Present Adjunct Associate Professor, Curriculum in Toxicology, University of North Carolina at
Chapel Hill
1997-2007 Principal Investigator, Clinical Research Branch, Human Studies Division, NHEERL
2001-2002 Acting Chief, Clinical Research Branch, Human Studies Division 1996-Present
Diplomate, Certified in General Toxicology, The American Board of Toxicology
1995-1997 Research Associate, Center for Environmental Medicine and Lung Biology,
University of North Carolina at Chapel Hill,
Selected Awards and Honors:
2003 EPA Gold Medal for Exceptional Service
2002 EPA Science and Technology Achievement Award (Level 1)
1993 Science Policy Fellow American Association for the Advancement of
Science/Environmental Protection Agency, Office of Health and Environmental
Assessment, U.S. EPA, Washington, DC 20460
B. SELECTED PUBLICATIONS
1. Cao, D., Bromberg, P.A. and Samet, J.M. (2009). Diesel Particle-Induced Transcriptional
Expression Of P21 Involves Activation Of Egfr, Src And Stat3. Am. J. Respir. Cell. Mol. Biol. In
press.
2. Lenz, A-G, Kim, YM, Hinze-Heyn, H., Karg, E, Lentner, B, Samet, JM, Schuiz, H and Maier, KL
-------
Principal Investigator/Program Director (Last, First, Middle): Samet, James M.
(2009). Effect of zinc oxide particles on cytokine mRNA expression in alveolar epithelial cells
exposed at the air-liquid interface. Submitted
3. Samet, J.M., Rappold, A., Graff, D., Cascio, W.E., Berntsen, J.H., Huang, Y-C, T., Herbst, M.,
Bassett, M., Montilla, T., Hazucha, M.J., Bromberg. P.A. and Devlin, R.B. (2009). Concentrated
Ambient Ultrafme Particle Exposure Induces Cardiac Changes In Young Healthy Volunteers. Am. J.
Respir. Crit. Care Med. In Press.
4. Silbajoris, R., Huang, J.M., Cheng, W.-C, Dailey, L, Tal, T.L. Jaspers, I., Ohio, A.J., Bromberg, P.A.
and Samet, J.M. Nanodiamond particles induce il-8 expression through a transcript stabilization
mechanism in human airway epithelial cells (2009). Nanotoxicology. In Press.
5. Wu, W., Silbajoris, R.A., Cao, D, Bromberg, P.A., Zhang, Q., Peden, D. B. and Samet, J.M.. (2008).
Regulation of cyclooxygenase-2 expression by cAMP response element and mRNA stability in a
human airway epithelial cell line exposed to zinc. Toxicol. Appl. Pharmacol. In press.
6. Tal, T.L., Silbajoris, R.A., Bromberg, P.A., Kim, Y. and Samet, J.M. (2008). Epidermal growth factor
activation by diesel particles is mediated by tyrosine phosphatase inhibition. Toxicol. Appl. Pharmacol.
233:382-8.
7. Wu W, Silbajoris RA, Cao D, Bromberg PA, Zhang Q, Peden DB, Samet JM. Regulation of
cyclooxygenase-2 expression by camp response element and mm'a stability in a human airway
epithelial cell line exposed to zinc. Toxicol Appl Pharmacol 2008;231 (2):260-266.
8. Wu, W., Madden, M, Kim, Y, Silbajoris, R., Jaspers, L, Graves, L.M., Bromberg, P.A. and Samet, J.M.
(2006). Transcriptional and Posttranscriptional Regulation of COX-2 Expression in Human Airway
Epithelial Cells Exposed to Zinc Ions. Bioch. Bioph. Res. Commun. In press.
9. Wang, X., Samet, J. M., and Ohio, A. J. (2006). Asbestos-induced activation of cell signaling pathways
in human bronchial epithelial cells. Exp Lung Res 32,229-243.
10. Dewar BJ, Gardner OS, Chen CS, Earp HS, Samet JM, and Graves LM. (2006) Capacitative
calcium entry contributes to the differential transactivation of the epidermal growth factor receptor
in response to thiazolidinediones. MolPharmacol 72:1146-1156,2007.
11. Cao, D., Bromberg, PA and Samet, JM (2007). Diesel-induced COX-2 expression involves chromatin
modification via degradation of HDAC1 and recruitment of p300. Am. J. Respir. Cell. Mol.
Biol.37:232-239.
12. Samet, JM, Graff, D, Benstsen, J, Ghio AJ, Huang T and Devlin, RB. (2007) A comparison of studies
on the effects of controlled exposure to fine, coarse and ultrafine ambient particulate matter from a
single location. Inhal. Toxicol. 19(Suppl. 1): 29-32.
13. Cao, D, Tal, TL, Graves, LM, Gilmour, I, Linak, W, Reed, W, Bromberg, PA, and Samet, JM. (2006).
Diesel exhaust particulates (DEP)-induced activation of Stat3 requires activities of EGFR and Src in
airway epithelial cells. Am. J. Physiol: Lung Cell. Mol. Biol. 292: L422-L429.
14. Kim, Y.M., Cao, D., Reed, W., Wu, W., Jaspers, L, Tal, T., Bromberg, P.A. and Samet, J.M. (2006).
Zn +-induced NF-kB-dependent transcriptional activity involves site-specific P65/REL-A
phosphorylation. Cellular Signaling. 19:538-546.
15. Kim, Y.M., Reed, W., Wu, W., Bromberg, P.A.,Graves, L.M. and Samet, J. M. (2006). Zn2+-
induced IL-8 expression involves AP-1, JNK, and ERK activities in human airway epithelial cells.
Am. J. Physiol: LungMol. Cell. Biol. 290: L1028-1035.
16. Wang, X.C., Wu, Y.M., Samet, J.M. and Ghio, A.J. (2006). [Expression of phosphorylated
ERK1/2 induced by crocidolite fibers in BEAS-2B cells]. Zhonghua Lao Dong Wei Sheng Zhi Ye
Bing Za Zhi. 24:597-600.
-------
Principal Investigator/Program Director (Last, First, Middle): SantO Domingo, Jorge W.
BIOGRAPHICAL SKETCH
Provide the following information for the key personnel and other significant contributors in the order listed on Form Page 2.
Follow this format for each person. DO NOT EXCEED FOUR PAGES.
NAME
Jorge W. Santo Domingo
eRA COMMONS USER NAME
POSITION TITLE
Microbiologist
EDUCATION/TRAINING (Begin with baccalaureate or other initial professional education, such as nursing, and include postdoctoral training.)
INSTITUTION AND LOCATION
University of Puerto Rico, Rio Piedras, PR
University of Puerto Rico, Rio Piedras, PR
Michigan State University, East Lansing, Ml
Michigan State University, East Lansing, Ml
DOE - WSRC, Aiken, SC
US EPA, Cincinnati, OH
DEGREE
(if applicable)
B.S.
M.S.
Ph.D.
Post Doc
Post Doc
Post Doc
YEAR(s)
1984
1987
1994
1995
1998
2002
FIELD OF STUDY
Biology
Biology
Microbiology
Molecular Biology
Molecular Biology
Water Microbiology
A. POSITIONS and HONORS
Research and Professional Experience:
8/84 - 4/87 Research Assistant, Microbial Ecology Laboratory, University of Puerto Rico
4/87 - 9/87 Research Associate, Microbial Ecology Laboratory, University of Puerto Rico
7/87 - 7/93 Graduate Research Assistant, Microbiology Department, Michigan State University
10/93 - 10/94 Research Associate, Microbiology Department, Michigan State University
2/95-6/98 DOE Post Doctoral Fellow, Oak Ridge Institute for Science and Education
7/98 - 10/00 EPA Federal Post Doctoral Fellow, ORD-NERL, Cincinnati, OH
11/00-5/09 Microbiologist, US-EPA, Cincinnati, OH
06/00-present Research Microbiologist, US-EPA, Cincinnati, OH
Selected Awards and Honors (since 2000):
US EPA On spot award. 2000. Training of MCEARD technicians in molecular biology methods
US EPA Superior Accomplishment Recognition Award. 2001. Development of rapid methods to detect
fecal enterococci in recreational waters
US EPA Team Award. 2002. Establishment of a new research program in molecular biology within the
Microbial Contaminant Control Branch
US EPA Superior Accomplishment Recognition Award. 2002. Mentoring of WSWRD technician during a
detail in MCCB
US EPA Scientific and Technological Achievement Award, Level II. 2003. Technical Information
Impacting Attainment of Clean Water Act Microbial Water Quality Goals (TMDLs)
National Risk Management Research Laboratory Honor Award, Support to Agency's Mission, Microbial
Source Tracking, 2004
Excellence in Review Award, Environmental Science and Technology, 2005.
US EPA Superior Accomplishment Recognition Award. 2006. For patent application entitled "Development
of Cow-Specific Primers and Identification of Cow-Specific DNA Sequences Using Genome Fragment
Enrichment"
Who's Who in Science and Engineering. Marquis Who's Who. 2006-2007
US EPA Superior Accomplishment Recognition Award. 2007. For work in projects resulting in numerous
publications
US EPA Superior Accomplishment Recognition Award. 2007. For research paper entitled "Identification of
Bacterial DNA Markers for the Detection of Human Fecal Pollution in Water"
i
Previous
TOC
-------
Principal Investigator/Program Director (Last, First, Middle): SantO Domingo, Jorge W.
ASM monthly magazine Microbe highlights research paper entitled "Identification of Bacterial DNA
Markers for the Detection of Human Fecal Pollution in Water", June 2007
US EPA Scientific and Technological Achievement Award, Level II. 2007. Scientific and Technological
Achievement in the Field of Genomics.
US EPA Scientific and Technical Achievement Award, Level II. 2008. Scientific and Technological
Achievement in the Field of Water Quality Monitoring and Indicators of Fecal Pollution
Other Experience and Professional Memberships:
Chair of Center of Excellence for Environmental Genomics Committee: 2008 -present
Member of the Molecular Biology Subcommittee of Standard Methods 9001: 2004-2008
Member of the Computational Toxicology Steering Committee (EPA): 2003 -2006
Member of the EPA Genomics Task Force Workgroup (EPA): 2006 - 2007
Member of WERF Project Steering Committee: 2002 - 2006
Member of the American Society for Microbiology: 1985 -present
Associate Editor of the Journal of Environmental Quality: 2002 - present
Ad hoc reviewer for: Applied and Environmental Microbiology, Environmental Science and Technology,
Microbiology Ecology, Microbiology, FEMS Microbiology Ecology, Bioremediation Journal, Water
Research, Water Environment Research, Journal of Applied Microbiology, Letters in Applied
Microbiology, Canadian Journal of Microbiology, Transactions of the American Society of Agricultural
Engineers, Electronic Journal of Biotechnology, NSF, NOAA, USDA, CICEET, SBRI, Gulf of Mexico
Program 2006 RFP
B. SELECTED PUBLICATIONS
Lu, J., J. W. Santo Domingo, S. Hill, and T. A. Edge. 2009.Microbial Diversity and Host-specific Sequences of
Canadian Goose Feces. Appl. Envir. Microbiol. In Press
Lamendella, R., J. W. Santo Domingo, A. C. Yannarell, S. Ghosh, G. Di Giovanni, R. I. Mackie, and D. B.
Oerther. 2009. Evaluation of swine-specific PCR assays used for fecal source tracking and analysis of
molecular diversity of Bacteriodales-swine specific populations. Appl. Envir. Microbiol. In Press
Lee, Y.-J., M. Molina, and J.W. Santo Domingo, J.D. Willis, M. Cyterski, D.M. Endale, and O.C. Shanks.
2008. A temporal assessment of cattle fecal pollution in two watersheds using 16S rRNA gene-based and
metagenome-based assays. Appl. Environ. Microbiol. 74:6839-6847.
Lu, J. and J. W. Santo Domingo. 2008. Turkey fecal microbial community structure and functional gene
diversity revealed by 16S rRNA gene and metagenomic sequences. J. Microbiol. 46:469-477.
Lu, J., J. W. Santo Domingo, R. Lamendella, T.Edge, and S.Hill. 2008. Phylogenetic diversity and molecular
detection of gull feces. Appl. Environ. Microbiol. 74: 3969-3976.
Lamendella, R., Santo Domingo J. W., Kelty C, and Oerther DB. 2008. Occurrence of bifidobacteria in feces
and environmental waters. Appl. Environ. Microbiol. 74:575-584.
Shanks OC, Atikovic E, Blackwood AD, Lu J, Noble RT, Santo Domingo J.W., Seifring S, Sivaganesan M,
Haugland RA. 2008. Quantitative PCR for genetic markers of bovine fecal pollution. Appl. Environ.
Microbiol. 74:745-752.
Santo Domingo, J. W., D.G. Bambic, T.A. Edge, and S. Wuertz. 2007. Quo vadis source tracking? Towards a
strategic framework for environmental monitoring of fecal pollution. Water Res. 41:3539-3552.
Lu, J., J. W. Santo Domingo, and O.C. Shanks. 2007. Identification of chicken-specific fecal microbial
sequences using a metagenomic approach. Water Res. 41:3561-3574.
Shanks, O., J. W. Santo Domingo, J. Lu, C.A. Kelty, and J. Graham. 2007. PCR Assays for the identification of
human fecal pollution in water. Appl. Environ. Microbiol. 73: 2416-2422.
Previous I TOC
-------
Principal Investigator/Program Director (Last, First, Middle): SantO Domingo, Jorge W.
Vogel, J.R., D.M. Stoeckel, R. Lamendella, R.B. Zelt, J.W. Santo Domingo, S.R. Walker, and D.B. Oerther.
2007. Identifying fecal sources in a selected catchment reach using multiple source-tracking Tools. J.
Environ. Qual. 36:718-729.
Lamendella, R., J. W. Santo Domingo, D. Oerther, J. Vogel, and, D. Stoeckel. 2007. Assessment of fecal
pollution sources in a small northern-plains watershed using PCR and phylogenetic analyses of
Bacteroidetes 16S rDNA. FEMS Microbiol. Ecol. 59:651-660.
Revetta, R.P., Santo Domingo, J.W., Kelty, C.A., Humrighouse, B.H., Oerther, D.B., Lamendella, R.,
Keinanen-Toivola, M., and Williams, M. "Molecular diversity of drinking water microbial communities: a
phylogenetic approach," Water Environment Federation, Proceedings of Disinfection 2007, Pittsburg, PA,
February 4-7, 2007.
Santo Domingo, J. W., Lu, J., Shanks, O., Lamendella, R., Kelty, C. A., and Oerther, D.B. "Development of
host-specific markers for source tracking using a novel metagenomic approach," Water Environment
Federation, Proceedings of Disinfection 2007, Pittsburg, PA, February 4-7, 2007.
Shanks, O., J. W. Santo Domingo, R. Lamendella, C.A. Kelty, and J. Graham. 2006. Competitive metagenomic
DNA hybridization identifies host-specific genetic markers in cattle fecal samples. Appl. Environ.
Microbiol. 72:4054-4060.
Shanks, O., J. W. Santo Domingo, and J. Graham. 2006. Use of competitive DNA hybridization to identify
differences in the genomes of two closely related fecal indicator bacteria. J. Microbiol. Methods. 66:321-
330.
Keinanen-Toivola, M.M., R.P. Revetta, and/. W. Santo Domingo. 2006. Identification of active bacterial
communities in drinking water biofilms using 16S rRNA-targeted clone libraries. FEMS Microbiol. Letters.
257:182-188.
Devereux, R., P. Rublee, J. H. Paul, K. G. Field and J. W. Santo Domingo. 2006. Development and
applications of microbial ecogenomic indicators for monitoring water quality: report of a workshop
assessing the state of the science, research needs and future directions. Environ. Monit. Assess. 116:459-
479.
Humrighouse, B. H., Santo Domingo, J.W., Revetta, R.P., Lamendella, R., Kelty, C., and Oerther, D.B.
"Microbial characterization of drinking water systems receiving groundwater and surface water as the
primary sources of water," Water Distribution System Analysis, Proceedings of the Annual Meeting,
Cincinnati, OH, August 27-30, 2006.
Batz, M.B., M.P. Doyle, J.G. Morris, Jr., J. Painter, R. Singh, R.V. Tauxe, M.R. Taylor, D.M.A. Lo Fo Wong,
F. Angulo, R. Buchanan, H.G. Claycamp, C. Smith DeWaal, J.W. Santo Domingo, K. Field, D. Goldman,
S. O'Brien, M. Moore, E. Ribot, S. Sundlof, and C. Woteki. 2005. Attributing illness to food. Emerging
Infectious Diseases. Available from http://www.cdc.gov/ncidod/EID/volllno07/04-0634.htm
Williams, M.W., J. W. Santo Domingo, and M.C. Meckes. 2005. Population diversity in model potable water
biofilms receiving chlorine or chloramines residual. Biofouling J. 21: 279-288.
Dick, L.K., A.E. Bernhard, T.J. Brodeur, J.W. Santo Domingo, J.M. Simpson, S.P. Walters, and K.G. Field.
2005. Host distributions of uncultivated fecal Bacteroidales reveal genetic markers for fecal source
identification. Appl. Environ. Microbiol. 71:3184-3191.
Simpson, JM, JW Santo Domingo, DJ Reasoner. 2004. Assessment of equine fecal contamination: the search for
alternative bacterial source-tracking targets. FEMS Microbiology Ecology 47:65-75.
Williams, M, J Santo Domingo, M Meckes, C Kelty, H Rochon. 2004. Phylogenetic Diversity of Drinking Water
Bacteria in a Distribution System Simulator. Journal of Applied Microbiology. In press
Invited oral presentations (last five years)
Swine and avian MST research. Gulf of Mexico Program MST Research Review. St Pete Beach, FL. February
2009.
What have we learned after 10 years of MST research: its importance in marker development and future
directions. Gulf of Mexico Alliance MST Workshop. Plenary session. St Pete Beach, FL. February 2009.
Previous I TOC
-------
Principal Investigator/Program Director (Last, First, Middle): SantO Domingo, Jorge W.
Monitoring fecal pollution using molecular tools. Department of Biological Sciences, Northern Kentucky
University. Highland Heights, KY. November 2008
Molecular tools for assessing microbial water quality. Department of Earth and Environmental Engineering,
Columbia University, New York, NY. April 2008.
Microbial source tracking: it takes two to tango. Department of Microbiology. University of Tennessee.
Knoxville, TN. April 2008.
Microbial source tracking: a tool box for environmental monitoring. Ohio State University, Ohio Agricultural
Research and Development Center, Wooster, OH. October 2007.
Introduction to microbial source tracking: a library independent perspective. EPI-Net workshop. Chicago, IL
September 2007.
Microbial Forensics and Environmental Monitoring of Fecal Pollution: From Phylogenetics to Metagenomics.
In Environmental Forensics: Microbial Clues to Contamination. ASM Annual Meeting, Toronto, Canada.
May 2007.
Microbial Source Tracking. Ontario Ministry of the Environment. Toronto, Canada. May 2007.
Fecal Source Tracking: an essential approach to microbial water quality monitoring. Department of Crop and
Soil Sciences. University of Kentucky, Lexington, KY. February 2007.
Determining the Sources of Fecal Pollution Using Molecular Methods. Allegheny Branch American Society for
Microbiology (ABASM) meeting. Plenary session. LaTrobe, PA, October 2006.
Microbial Source Tracking: A Necessary Tool for Environmental Monitoring and Risk Assessment. National
Beaches Conference, Niagara Falls NY, December 2006
Environmental Monitoring of Polluted Waters Using Source Tracking Molecular Tools: Lessons Learned and
Future Needs. Water Management Association of Ohio (WMAO) 35th Annual Fall Conference. Columbus,
OH, November 2006
Application of a Library Independent Method used in the Identification of Fecal Pollution Sources in
Environmental Waters. Sustainable Beaches Conference. St. Petersburg, FL, October 2005.
Assumptions and Limitations of MST. Microbial Source Tracking Workshop. Sponsored by Water
Environment Research Foundation and the Metropolitan Water District of Southern California, San
Antonio, TX, February 2005
Previous I TOC
-------
Principal Investigator/Program Director (Last, First, Middle): Segal, Deborah
BIOGRAPHICAL SKETCH
NAME
Deborah Segal
eRA COMMONS USER NAME
Segal
POSITION TITLE
Environmental Health Scientist
EDUCATION/TRAINING (Begin with baccalaureate or other initial professional education, such as nursing, and include postdoctoral training.)
INSTITUTION AND LOCATION
George Washington University
Johns Hopkins University
DEGREE
(if applicable)
B.A.
M.H.S.
YEAR(s)
1987
2000
FIELD OF STUDY
Political
Communications
Toxicology
A. POSITIONS and HONORS
Research and Professional Experience:
2007- Project Officer, National Center for Environmental Research (NCER), ORD, USEPA
2000-2007 Science Review Administrator, NCER, ORD, USEPA
1997-2000: Predoctoral Fellow, Johns Hopkins Bloomberg School of Public Health, Division of Toxicology
1992-1997: Science Communications Officer, American Psychological Association
1990-1992: Technical Production Editor, APA Books, American Psychological Association
1989-1990: Assistant Editor, Broadcasting Yearbook, Broadcasting Publications, Inc.
1988-1989: Editorial Assistant, Broadcasting Yearbook, Broadcasting Publications, Inc.
Professional Societies and Affiliations:
International Society for Exposure Analysis, Member
Honors and Awards:
EPA Bronze Medal for Commendable Service, for Creating and Maintaining the Preeminent Peer Review
Process for Evaluating Environmental Research in the United States, 2006; NIEHS Predoctoral Fellowship
Award, 1997-2000.
Selected Invitations at National & International Symposia:
"EPA's STAR Research Program in Computational Toxicology," presented at the 2nd & 3rd Annual Systems
Toxicology Symposiums: Piscataway, NJ, 2008, 2009. "Overview of EPA's Workshop on Research Needs for
Community-Based Risk Assessment," presented at the Symposium on Exposure Science for Community-
Based Cumulative Risk Assessment, International Society for Risk Assessment Annual Conference, Durham,
NC, 2007.
Selected Expert Committees/Advisory Panels/Organizing Committees:
Chaired/Organized session at the "Research Approaches to Assessing Public Health Outcomes of Risk
Management Decisions Workshop" ((January 2008) featuring the research of new grantees to the
"Environmental Health Outcome Indicators" STAR program. Directed all aspects of "Research Needs for
Community-Based Risk Assessment Workshop" (October 2007) designed to identify data gaps and prioritize
research needs for community risk assessments that consider interactions between chemical and non-
chemical stressors. Chaired and Organized session at U.S. EPA STAR Graduate Fellowships Conferences:
"Toxicogenomics" (October 2004) and "Use of Models in the Risk Assessment Process" (October 2006).
Selected Assistance/Advisory Support to the Agency:
PHS 398/2590 (Rev. 09/04, Reissued 4/2006)
Page 1
Continuation Format Page
Previous
TOC
-------
Principal Investigator/Program Director (Last, First, Middle): Segal, Deborah
Chair of Workgroup on Lab/Center/Office (L/C/O) Involvement in NCER Activties; Cochair of Peer Review
Improvement Committee (Chair, Subcommittee: Peer Review Summaries and FACA); Serves on NCER Policy
Implementation Workgroup; Served on Mode of Action Workgroup.
B. Publication
Sanchez, Deener, Cohen Hubal, Knowlton, Reif, & Segal (2009). "Research needs for community-based risk
assessment: findings from a multi-disciplinary workshop." Journal of Exposure Science and Environmental
Epidemiology, Advance Online Publication, 25, 1-10.
PHS 398/2590 (Rev. 09/04, Reissued 4/2006)
Previous
Page 2
TOC
Continuation Format Page
-------
Principal Investigator/Program Director (Last, First, Middle): Setzer, WoodfOW R.
BIOGRAPHICAL SKETCH
NAME
R. Woodrow Setzer
eRA COMMONS USER NAME
POSITION TITLE
Mathematical Statistician
EDUCATION/TRAINING
INSTITUTION AND LOCATION
University of Chicago, Chicago, Illinois
SUNY at Stony Brook, Stony Brook, New York
University of North Carolina, Chapel Hill
National Research Council Fellow, USEPA, RTP,
NC
DEGREE
(if applicable)
B.A
Ph.D.
Post-doc
Post-doc
YEAR(s)
1974
1983
1987
1989
FIELD OF STUDY
Mathematics
Ecology and Evolution
Biostatistics
Biostatistics and Risk
Assessment
A. POSITIONS and HONORS
Research and Professional Experience:
2009 - Present
2009 - Present
2005 - Present
2002 - 2005
2000-2009
1993-2002
1989-1993
1987-1989
1984-1987
1984
Adjunct Professor, Department of Biostatistics, North Carolina State University
Mathematical Statistician, NCCT, ORD, EPA
Mathematical Statistician, PKB, ETD, NHEERL, ORD EPA
Adjunct Associate Professor, Department of Biostatistics, School of Public Health,
University of North Carolina at Chapel Hill
Mathematical Statistician, BRSS, NHEERL, ORD
Health Scientist, HERL, ORD EPA
Postdoctoral Fellow, DTD, HERL, ORD EPA
Postdoctoral Fellow, Department of Biostatistics,
School of Public Health, University of North Carolina, Chapel Hill, NC
Lecturer, Department of Ecology and Evolution, State University of New York, Stony Brook,
NY
Professional Societies and Affiliations:
Society for Risk Analysis
Biometrics Society
American Statistical Association
Honors and Awards:
1. Level I USEPA Science and Technology Achievement Award for BBDR Modeling of the Developmental
Toxicityof5-FU, 1994
2. Level III USEPA Science and Technology Achievement Award for Dose-Response Relationship in Multi-
stage Carcinogenesis, 1994
3. Honorable Mention, USEPA Science and Technology Achievement Award for A New Mechanism for the
Exogenous Mitigation of 5-Fluorouracil-Induced Toxicity, 1997
4. USEPA Silver Medal for the Organophosphate Cumulative Risk Assessment, 2003
5. USEPA Bronze Medal for Commendable Service for Development of Benchmark Dose Software, 2004
6. USEPA Silver Medal for Scientific Workgroups for EPA's Guidelines for Carcinogen Risk Assessment and
Supplemental Guidance for Assessing Susceptibility from Early-Life Exposure to Carcinogens, 2006
7. USEPA Bronze Medal for Commendable Service for work with the Malathion RED team, 2006.
Pagel
Previous
TOC
-------
Principal Investigator/Program Director (Last, First, Middle): Setzer, WoodfOW R.
8. Honorable Mention, USEPA Science and Technology Achievement Award for Defining Different
Populations of Supernumerary Ribs and Assessing their Biological and Regulatory Significance, 2006
9. USEPA Bronze Medal for Commendable Service for Information Technology Improvement Project Team,
2007
10. Level III USEPA Science and Technology Achievement Award for Facilitating the Evaluation and
Utilization of Physiologically Based Pharmacokinetic (PBPK) Models in Risk Assessment, 2007
11. Honorable Mention, USEPA Science and Technology Achievement Award for ToxCast: A Biologically and
Chemically Based System for EPA Program Offices to Prioritize Toxicity Testing of Chemicals
12. USEPA Bronze Medal for "Successful Completion of a Decade's Work Involving Cutting Edge Science and
Innovation on the N-Methyl Carbamate Cumulative Risk Assessment", 2008
13. USEPA Bronze Medal for Efforts as Part of a Team Instrumental in Developing and Implementing the
BMD Methodology for Use in IRIS Assessments, 2008.
14. Award for Exceptional or Outstanding ORD Technical Assistance to the Regions or Program Offices, 2008.
15. Level III USEPA Science and Technology Achievement Award for the Analysis of the Risk Assessment
Implications of Early-Life Exposure to Carcinogens Considering Mode of Action, 2008.
16. Superior Accomplishment Award from the Office of Pesticide Programs, 2009.
Selected Invitations at National & International Symposia:
Risk Assessment Using EPA Benchmark Dose Software Version 1.2. A full day workshop presented (with
J. Gift) at the annual meeting of the Society for Risk Analysis, December 5, 1999
Calculating and Using Benchmark Doses (BMD). Federal/State Toxicology and Risk Analysis Committee,
May 21-23, 2001.
Populations and PK Models. NERL/NHEERL Exposure to Dose Modeling Workshop, Research Triangle
Park, NC, July 10-11,2001.
Basic Statistical Analysis of Developmental Toxicity Studies, in Experimental Design and Biostatistics, a
mini-education course at the annual meeting of the Teratology Society, Scottsdale, AZ, June 25, 2002.
Use of NOAEL, benchmark dose, and other models for human risk assessment of hormonally active
substances. SCOPE/IUPAC International Symposium on Endocrine Active Substances, Yokohama, Japan,
November 17-21,2002.
Cumulative Risk Analysis for Organophosphorus Pesticides. Society of Toxicology, Salt Lake City, UT,
March 9-13, 2003.
Mediating the Meeting between Model and Data: Statistical Issues for PBPK Modeling. International
Workshop on Uncertainty and Variability in Physiologically Based Pharmacokinetic (PBPK) Models, Research
Triangle Park, NC. October 31 - November 2, 2006.
International Workshop on Uncertainty and Variability in Physiologically Based Pharmacokinetic Models.
An International Workshop on the Development of Good Modelling Practice for PBPK Models. Chania, Crete,
Greece. April 26 - 28, 2007.
Lessons Learned: Modeling Cancer Data. ILSI Europe Workshop on the Application of Margin of Exposure
(MoE) Approach to Compounds in Food which are both Genotoxic and Carcinogenic, Rhodes, Greece, October
1-3,2008.
Scientific Workshop to Inform EPA's Response to National Academy of Science Comments on the Health
Effects of Dioxin in EPA's 2003 Dioxin Reassessment. Cincinnati, Ohio, 18-20 February, 2009.
WHO International Workshop on Principles of Characterizing and Applying PBPK Models in Risk
Assessment. Berlin, Germany, 6-9 July, 2009.
Selected Expert Commiittees/Advisory Panels/Organizing Committees:
Member, Editorial Board, Toxicology Methods, 1994 - 1998
President-Elect, Research Triangle Chapter, Society for Risk Analysis, 2001 - 2002
Chair, Research Triangle Chapter, Society for Risk Analysis, 2002 - 2003
Page 2
Previous I TOC
-------
Principal Investigator/Program Director (Last, First, Middle): Setzer, WoodfOW R.
Affiliate Member of the Biostatistics and Epidemiological Methods Facility Core, University of North Carolina
at Chapel Hill Center for Environmental Health and Susceptibility
ILSI HESI Dose Dependent Transitions in Mechanisms of Toxicity Committee 2002 - 2003.
Invited Participant, WHO/IPCS Author's Workshop on Dose-Response Modeling, Geneva, Switzerland, 2004
Invited Participant, EFSA/WHO International Conference, "Risk Assessment of Compounds that are both
Genotoxic and Carcinogenic" Brussels, Belgium, 16-18 November, 2005
ILSI-Europe Expert Group on the Application of the Margin of Exposure Approach to Genotoxic Carcinogens in
Food. 2006-2008.
Associate Editor for Journal of Statistical Software, 2007 - present.
Publications Officer, Risk Analysis Specialty Section, American Statistical Association, starting January, 2010.
Selected Assistance/Advisory Support to the Agency:
Planning Committee and epidemiology session Chair Mn/MMT Workshop held in Research Triangle Park, NC,
March 12-15, 1991
Co-Chair, Organizing Committee for the First HERL Symposium: Biological Mechanisms and Quantitative Risk
Assessment, 1992 - 1993.
IRIS RfD/C Committee, 1994-1995
Chair, Risk Assessment Forum Technical Panel, Benchmark Dose Technical Guidance Document, 1998 -
2005
Statistical Consultant/Collaborator with the National Center for Environmental Assessment for Development of
EPA's Benchmark Dose Software. 1993 - present.
Co-Chair, International Workshop on Uncertainty and Variability in Physiologically Based Pharmacokinetic
Models (2006 - present; workshop Oct 31 - Nov 2, 2006).
Member, NCEA Statistical Working Group (2005 - present).
Member, ORD Information Technology Improvement Project Working Group (2006).
B. SELECTED PUBLICATIONS (selected from 56 peer-reviewed).
1. Scheerer JB, Xi L, Knapp GW, Setzer RW, Bigbee WL, and Fuscoe JC (1999) Quantification of Illegitimate
V(D)J Recombinase-Mediated Mutations in Lymphocytes of Newborns and Adults. Mutation. Research.
431:291-303.
2. Hurst CH, DeVito MJ, Setzer RW, and Birnbaum LS (2000) Acute Administration of 2,3,7,8-
Tetrachlorodibenzo-p-dioxin (TCDD) in Pregnant Long Evans Rats: Association of Measured Tissue
Concentrations with Developmental Effects. Toxicological Sciences 53: 411-420.
3. Lau C, Andersen ME, Crawford-Brown D, Kavlock RJ, Kimmel CA, Knudsen TB, Muneoka K, Rogers JM,
Setzer RW, Smith G, and Tyl R (2000). Evaluation of Biologically Based Dose-Response Modeling for
Developmental Toxicity: A Workshop Report. Regulatory Toxicology and Pharmacology, 31: 190-199.
4. DeWoskin RS, Barone S Jr., Clewell HJ, Setzer RW (2001) Improving the development and use of
biologically based dose response models (BBDR) in risk assessment. Human and Ecological Risk
Assessment, 6: 1091 - 1120.
5. Lau C, Mole ML Copeland MF, Rogers JM, Kavlock RJ, Shuey DL, Cameron AM, Ellis DH, Logsdon TR,
Merriman J, and Setzer RW (2001) Toward a biologically based dose-response model for developmental
toxicity of 5-fluorouracil in the rat: Acquisition of experimental data. Toxicological Sciences, 59: 37-48.
6. Setzer RW, Lau C, Mole ML, Copeland FM, Rogers JM, and Kavlock RJ (2001). Toward a biologically-
based dose-response model for developmental toxicity of 5-fluorouracil in the rat: a mathematical
construct. Toxicological Sciences, 59: 49-58.
7. Shaughnessy DT, Setzer RW, and DeMarini DM (2001). Effect of the antimutagens vanillin and
cinnamaldehyde on the spontaneous mutation spectra of Salmonella TA104. Mutation Research, 480-
481: 55-69.
8. Wubah JA, Setzer RW, and Knudsen TB (2001). Exposure-disease continuum for 2-chloro-2'-
deoxyadenosine (2CdA), a prototype ocular teratogen. 1. Dose-response analysis. Teratology, 64: 154-
169.
9. Lau C, Narotsky MG, Lui D, Best D, Setzer RW, Mann PG, Wubah JA, and Knudsen, TB (2002).
Page 3
Previous I TOC
-------
Principal Investigator/Program Director (Last, First, Middle): Setzer, WoodfOW R.
Exposure-disease continuum for 2-chloro-2'-deoxyadenosine (2-CdA), a prototype! teratogen: Induction of
lumbar hernia in the rat and species comparisons for the teratogenic responses. Teratology 66: 6-18.
10. Knapp GW, Setzer RW, Fuscoe JC (2003). Quantitation of aberrant interlocus T-cell receptor
rearrangements in mouse thymocytes and the effect of the herbicide 2,4-dichlorophenoxyacetic acid.
Environmental Molecular Mutagenesis, 42: 37-43.
11. Rogers JM, Setzer RW, Branch S, Chernoff N (2004). Chemically induced supernumerary lumbar ribs in
CD-1 mice: size distribution and dose response. Birth Defects Research, 71: 17—25.
12. Smialowicz RJ, Burgin DE, Williams WC, Diliberto JJ, Setzer RW, Birnbaum LS (2004). Xyp1A2 is not
required for 2,3,7,8-tetrachlorodibenzo-p-dioxin-induced immunosuppression. Toxicology, 197, 15—22.
13. Slikker W, Andersen ME, Bogdanffy MS, Bus JS, Cohen SD, Conolly RB, David RM, Doerrer NG, Dorman
DC, Gaylor DW, Hattis D, Rogers JM, Setzer RW, Swenberg JA, Wallace K (2004). Dose-dependent
transitions in mechanisms of toxicity. Toxicology and Applied Pharmacology, 201: 203 ~ 225.
14. Slikker W, Andersen ME, Bogdanffy MS, Bus JS, Cohen SD, Conolly RB, David RM, Doerrer NG, Dorman
DC, Gaylor DW, Hattis D, Rogers JM, Setzer RW, Swenberg JA, Wallace K (2004). Dose-dependent
transitions in mechanisms of toxicity: case studies. Toxicology and Applied Pharmacology, 201: 226 -
294.
15. Clark LH, Setzer RW, Barton HA (2004) Framework for evaluation of physiologically-based
pharmacokinetic models for use in safety or risk assessment. Risk Analysis 24: 1697 - 1717.
16. Barton HA, Cogliano VJ, Flowers L, Valcovic L, Setzer RW, Woodruff TJ (2005). Assessing Susceptibility
from Early-Life Exposure to Carcinogens. Environmental Health Perspectives, 113: 1125 - 1133.
17. Dix DJ, Houck KA, Martin MT, Richard AM, Setzer RW, Kavlock RJ. (2007). The ToxCast Program for
Prioritizing Toxicity Testing of Environmental Chemicals. Toxicological Sciences 95: 5 - 12.
18. Barton HA, Chiu WA, Setzer RW, Andersen ME, Bailer AJ, Bois FY, DeWoskin RS, Hays S, Johanson G,
Jones N, Loizou G, MacPhail RC, Portier CJ, Spendiff M, Tan Y-M (2007). Characterizing Uncertainty and
Variability in Physiologically-based Pharcokinetic (PBPK) Models: State of the Science and Needs for
Research and Implementation. Toxicological Sciences, 99: 395 - 402.
19. Kavlock RJ, Ankley G, Blancato J, Breen M, Conolly R, Dix D, Houck K, Hubal E, Judson R, Rabinowitz J,
Richard A, Setzer RW, Shah I, Villeneuve D, Weber E. (2008). Computational toxicology - a state of the
science mini review. Toxicological Sciences 103: 14-27.
20. Judson R, Elloumi F, Setzer RW, Li Z, Shah I. (2008) A comparison of machine learning algorithms for
chemical toxicity classification using a simulated multi-scale data model. BMC Bioinformatics. 2008 May
19;9:241.
21. Wambaugh JF, Barton HA, Setzer RW. 2009. Comparing models for PFOA pharmacokinetics using
Bayesian analysis. Journal of Pharmacokinetics and Pharmacodynamics 35: 683 - 712.
22. Lou I, Wambaugh JF, Lau C, Hanson RG, Lindstrom AB, Strynar MJ, Zehr RD, Setzer RW, Barton HA.
2009. Modeling Single and Repeated Dose Pharmacokinetics of PFOA in Mice. Toxicological Sciences
107: 331 -341.
23. Rodriguez, CE, Setzer RW, Barton HA. 2009. Pharmacokinetic Modeling of Perfluorooctanoic Acid
During Gestation and Lactation in the Mouse. Reproductive Toxicology, doi:
10.1016/j.reprotox.2009.02.009
Page 4
Previous I TOC
-------
Principal Investigator/Program Director (Last, First, Middle):
Shah, I
BIOGRAPHICAL SKETCH
Provide the following information for the key personnel in the order listed on Form Page 2.
Follow this format for each person. DO NOT EXCEED FOUR PAGES.
NAME
Imran Shah
POSITION TITLE
Computational Systems Biologist, National Center
for Computational Toxicology, ORD, EPA
EDUCATION/TRAINING (Begin with baccalaureate or other initial professional education, such as nursing, and include postdoctoral training.)
INSTITUTION AND LOCATION
Imperial College of Science, Technology Medicine,
London, UK,
George Mason University, Fairfax, Virginia, USA
George Mason University, Fairfax, Virginia, USA
DEGREE
(if applicable)
B.Sc.
M.S.
Ph.D.
YEAR(s)
1989
1993
1999
FIELD OF STUDY
Physics
Applied And Engineering
Physics
Computational Biology
A. Positions and Honors.
Research and Professional Experience
2006-present Computational Systems Biologist, National Center for Computational Toxicology, Office of
Research and Development, US Environmental Protection Agency, Research Triangle Park,
North Carolina.
2004-2006 Head of Computational Systems Biology, Icoria, Research Triangle Park, North Carolina.
2000-2004 Assistant Professor of Bioinformatics, Department of Preventive Medicine and Biometrics, and
Department of Pharmacology, School of Medicine, University of Colorado Health Sciences
Center, Denver, Colorado.
1997-2004 Director of the Doctoral Program in Bioinformatics, School of Medicine, University of Colorado
Health Sciences Center, Denver, Colorado.
2001-2004 Director of Integrated Informatics, Department of Pharmacology, School of Medicine, University
of Colorado Health Sciences Center, Denver, Colorado.
2001-2004 Adjunct Assistant Professor of Computer Science and Engineering, University of Colorado,
Denver, Colorado.
1999-2001 Adjunct Assistant Professor of Computational Sciences & Informatics, School of Computational
Sciences, George Mason University, Fairfax, Virginia.
1998-2000 Bioinformatics Research Scientist, American Type Culture Collection (ATCC),
Manassas, Virginia.
1996-1997 Graduate Fellow, School of Computational Sciences
George Mason University, Fairfax, Virginia.
1995-1996 Bioinformatics Software Developer, The Institute for Genomic Research (TIGR), Rockville,
Maryland.
1993-1994 Software developer, Vision Lab, Department of Computer Science,
George Mason University, Fairfax, Virginia.
1991-1994 Graduate Research Assistant, School of Computational Sciences
George Mason University, Fairfax, Virginia.
Professional Societies and Affiliations:
1997-present International Society for Computational Biology (ISCB).
1999-2006 American Association for Artificial Intelligence (AAAI).
2008-present Society of Toxicology (SOT).
PHS 398/2590 (Rev. 05/01)
Page
Biographical Sketch Format Page
-------
Principal Investigator/Program Director (Last, First, Middle):
Honors
1996-1997 Predoctoral Fellowship, School of Computational Sciences, George Mason University, Fairfax,
Virginia.
1992 NASA Summer Fellowship for High Performance Computing, NASA Goddard Space Flight
Center, Greenbelt, Maryland.
2008 Environmental Protection Agency, Office of Research and Development, Bronze Medal
Award for Genomics Training.
Selected invitations at National & International Symposia:
"Inference in High-throughput Biology." 29th Annual Meeting of the Statistical Society of Canada. Vancouver,
Canada, June 2001.
"Pathway Visualization in Bioinformatics." Computer Graphics and Visualization Techniques for Bioinformatics
and Medical Applications. Center for Computational Biology, University of Colorado, Denver, October 2002.
"Computational Biology with Common Lisp." Bioinformatics Session, International Lisp Conference. San
Francisco, California, October 2002.
"Computational Inference of Metabolic Pathways." 10th International Meeting on Microbial Genomes. Lake
Arrowhead, California, September 2002.
"Microbial Metabolic Pathway Prediction." Department of Microbiology, School of Medicine, University of
Colorado, Denver, Colorado, September 2002.
"Metabolic Pathway Inference by Heuristic Search." The BioPathways Conference. Edmonton, Canada, August
2002.
"Alcoholism and Gene Expression Arrays." Workshop on Neuroinformatics. Center for Computational Biology,
University of Colorado, Denver, April 2002.
"Elucidating cis-Regulatory Control using Gene Arrays." The First Rocky Mountain Regional Bioinformatics
Meeting, Aspen, Colorado, December 2003.
"Computational Elucidation of Transcription Control Modules from Gene Expression Arrays." Human Medical
Genetics Program, University of Colorado Health Sciences Center, May 2003.
"Pathway Elucidation, in silico." Department of Chemical Engineering, The University of Queensland,
Queensland, Australia, July 2003.
"Integrating Biomarkers to Understand Fatty Liver Disease." Metabolic Profiling. Durham, North Carolina,
November 2005.
"Predicting Dose-Response at The System Level." McKim Conference, Duluth, Minnesota, September 2007.
"Representing Chemical-Induced Liver Injury for Multiscale Tissue Modeling." Conference on Semantics in
Healthcare & Life-Sciences (C-SHALS 2008), Boston, Massachusetts, February 2008.
"The Virtual Liver Project: Simulating Tissue Injury through Molecular and Cellular Processes." Systems
Biology of the Liver, International Conference on Systems Biology, Goteborg, Sweden, August 2008.
"Simulating Hepatic Tissue Lesions as Virtual Cellular Systems." Society of Toxicology, March 2009.
"The EPA Virtual Liver Project." International Workshop on Virtual Tissues, Research Triangle Park, North
Carolina, April 2009.
"The Virtual Liver: Simulating Hepatic Tissue Lesions as Cellular Systems." Toxicology and Risk Assessment
Conference, Cincinatti, Ohio, April 2009.
"Cell Behaviours and Virtual Tissues." Cell Behaviour Ontology Workshop, National Institute for General
Medical Sciences, Bethesda, Maryland, May 2009.
Selected Expert Committees/Advisory Panels/Organizing Committees:
2000-2005 Member of the Center for Computational Pharmacology, Program in Biomolecular Structure
University of Colorado Health Sciences Center, Denver, Colorado.
2001-2006 National Human Genome Research Institute, National Institute for Mental Health, National Heart
Lung and Blood Institute, National Science Foundation and Department of Energy grant
application review.
2001-2005 Member of the Center for Computational Biology, University of Colorado, Denver, Colorado.
2002 Co-organizer of the first workshop on Bioinformatics, Center for Computational Biology,
University of Colorado, Denver.
PHS 398/2590 (Rev. 05/01) Page Continuation Format Page
-------
Principal Investigator/Program Director (Last, First, Middle):
2002-2004 Member of BioPAX: An ontology for representing and exchanging biochemical pathways.
2003-2004 Program Committee for the "Innovative Applications of Artificial Intelligence (IAAI-03)"
Conference, Acapulco, Mexico.
2007 Session Chair, "Modeling Signaling as a Determinant of System Behavior", International Forum
on Computational Toxicology, EPA, Research Triangle Park, North Carolina.
2008-2009 Co-chair of the "First International Workshop on Virtual Tissues (v-Tissues 2009)", Research
Triangle Park, North Carolina.
2007 Environmental Protection Agency, Office of Research and Development, Future of Toxicology
Working Group.
B. Selected peer-reviewed publications
Cohen Hubal, E.A., Richard, A. M., Shah I.. Gallagher, J., Kavlock, R., Blancato, J., and Edwards, S.
Exposure science and the U.S. EPA National Center for Computational Toxicology. J Expos Sci Environ
Epidemiol (November 5, 2008).
Judson, R., Elloumi, F., Setzer, R. W., Li, Z., &Shah. I. A comparison of machine learning algorithms for
chemical toxicity classification using a simulated multi-scale data model. BMC Bioinformatics 9, 241(2008).
Kavlock, R. J., Ankley, G., Blancato, J., Breen, M., Conolly, R., Dix, D.,Houck, K., Hubal, E., Judson, R.,
Rabinowitz, J., Richard, A., Setzer, R.W., Shah. I.. Villeneuve, D., Weber, E. (2007). Computational toxicology
a state of the science mini review, Toxicol Sci, Advance Access, Dec 7, 2007.
Lapadat, R., DeBiasi, R.L., Tyler, K.L., Johnson, G.L, and Shah. I. Genes induced by reovirus have a distinct
modular c/s-regulatory architecture. Current Genomics, 6(7):501-513, 2005.
McShan, D., Upadhyaya, M. and Shah. I. Symbolic inference of xenobiotic metabolism. Pac. Symp. Biocomp.
9:545-56,2004.
McShan, D., Upadhyaya, M. and Shah. I. Heuristic Search for Metabolic Engineering: cfe novo synthesis of
vanillin. Comp. and Chem. Engg., Bioinformatics special issue (in press). 2004.
Hink, R.L., Hokanson, J.E., Shah. I.. Long, J.C., Goldman, D. Sikela, J.M. Investigation of DUSP8 and CALCA
in alcohol dependence. Addiction Biology. 8(3):305-312, 2003.
McShan, D., Rao, S. and Shah. I. PathMiner: Predicting metabolic pathways by heuristic search.
Bioinformatics. 19(13): 1692-1698, 2003.
McShan, D., and Shah. I. Distributed Intelligent Agents in Lisp for Bioinformatics (DIAL-B). Agents in
Bioinformatics, Autonomous and Multiagent Systems, 56-59, 2002.
Shah. I. and Hunter, L. Visual management of large scale data mining projects. Pac. Symp. Biocomp. 5:275-
287, 2000.
Shah. I. and Hunter, L. Visualization based on the Enzyme Commission nomenclature. Pac. Symp. on
Biocomp. 3:142-152, 1998.
Shah. I. and Hunter, L. Identification of Divergent Functions in Homologous Proteins by Induction over
Conserved Modules. Intell. Syst. Mol. Biol. 6:157-164, 1998.
Shah. I. and Hunter, L. Predicting Enzyme Function from Sequence: A Systematic Appraisal. Intell. Syst. Mol.
Biol. 5:276-283, 1997.
C. Research Support.
Principal investigator on "Modeling Metabolic Pathways: A Bioinformatics Approach." Funded by the National
Science Foundation, Department of Energy and Office of Naval Research, from September 1, 2000 to March
22, 2004.
Co-Principal investigator on "Integrated Neuroinformatics Resource for Alcoholism Research." Funded by the
National Institute for Alcohol Abuse and Alcoholism, from August 1, 2001 to March 22, 2004.
Investigator on "Gene Array Technology Center for Alcohol Research." Funded by the National Institute for
Alcohol Abuse and Alcoholism, from April 1, 2001 to March 22, 2004.
Co-Principal investigator on "Application of expression analysis to study disease pathogenesis." Funded by the
National Heart, Lung and Blood Institute, from October 2002 to March 22, 2004.
Principal investigator on "Target Assessment Technology Suite." Funded by the National Institute of Standards,
Advanced Technology Program, from 2002-2006.
PHS 398/2590 (Rev. 05/01) Page Continuation Format Page
-------
Principal Investigator/Program Director (Last, First, Middle):
BIOGRAPHICAL SKETCH
Provide the following information for the key personnel and other significant contributors in the order listed on Form
Page 2.
Follow this format for each person. DO NOT EXCEED FOUR PAGES.
NAME
Singh, Amar V.
eRA COMMONS USER NAME
POSITION TITLE
Scientific Systems Analyst
EDUCATION/TRAINING (Begin with baccalaureate or other initial professional education, such as nursing, and
INSTITUTION AND LOCATION
DEGREE
(if applicable)
YEAR(s)
FIELD OF STUDY
Botany/Biochem/Zool.
Biotechnology
Project Management
Botany (Bioinformatics)
Yuvaraja's College, Mysore India B.S. 1993
University of Mysore, Mysore India M.S. 1995
University of Louisville, Louisville, KY US Certificate 2006
CCS University, Meerut India PhD 2009
A. POSITIONS AND HONORS (In chronological order)
Research and Professional Experience:
1995-96 Assistant Manager, Jayson's Agritech Pvt. Ltd., Mysore, India
1996-97 Project Assistant, Central Food Technological Research Institute, Mysore, India
1997-99 Research Fellow, Department of Biotechnology, University of Mysore, Mysore, India
1999-00 Clinical Data Reviewer, Clinical Data Management Center, GSK, Bangalore, India
2000-01 Research Scientist in Bioinformatics, Avestha Gengraine Technologies Pvt. Ltd., India
2001-02 Group Leader, Bioinformatics, Avestha Gengraine Technologies Pvt. Ltd., India
2002-03 Manager, Bioinformatics, Mascon Global Ltd., Princeton NJ
2003-03 Research Asst. Thomas Jefferson University, Philadelphia PA
2003-05 Research Associate, Systems Analysis Laboratory, U of Louisville Birth Defects Center, Louisville KY
2005-07 Research Scientist, Birth Defects Center, U of Louisville Birth Defects Center, Louisville KY USA
2005-07 Bioinformatics Manager, Systems Analysis Laboratory, U of Louisville, Louisville KY
2006-07 Operation Manager, Center for Environmental Genomics and Integrative Biology, UoL Louisville KY
2007-pres Scientific Systems Analyst, Lockheed Martin Contractor at National Center for Computational Toxicology
(NCCT) US EPA, RTP Durham NC
Professional Societies and Affiliations:
International Society of Computational Biologists (ISCB), Teratology Society, Asia Pacific Bioinformatics Network
(ApBioNET), African Society for Bioinformatics and Computational Biology (ASBCB), BioClues (Indian Bioinformatics
Society)
Honors and Awards:
1997 Lady Tata Memorial Fellowship (Lady Tata Memorial Trust, Mumbai India).
1998 Senior Research Fellowship (Central Scientific and Industrial Research, Govt of India New Delhi, India)
2006 Young Investigator Travel Awards, 46th Annual Meeting of The Teratology Society Meeting at Tucson,
Arizona.
Special Recognition: Elected to Inbios Management Group, Bioinformatics Society of India (INBIOS) (2002-pres) ;
ResearchILouisville: 3rd place, Innovation in Biotechnology (2004); Committee Co-Chair, Issues and Protocols in
Bioinformatics Education, Bioinformatics Society of lndia(2005-pres). Member, Web Site Committee Teratology
Society (2005-2009; Chair of the Committee 2007-2008); Member Constitution and By-Laws Committee Teratology
Society (2009-pres); Core Council Member and Secretary BioClues; Organizing Committee Member for First Virtual
Bioinformatics Conference lnBix'10 in India.
PHS 398/2590 (Rev. 09/04)
Page J
Biographical Sketch Format Page
Previous
TOC
-------
Principal Investigator/Program Director (Last, First, Middle):
SELECTED INVITATIONS AT NATIONAL & INTERNATIONAL SYMPOSIA:
Invited speaker, "Systems Biology in Understanding Developmental Defects—Efforts and Challenges", First virtual
conference on "Bioinformatics to Systems Biology" November 16, 2007, jointly organized by Bioinformatics.Org and
the ISCB Regional Student Group, Denmark
B. SELECTED PEER-REVIEWED PUBLICATIONS (in chronological order).
1. S. Ahuja, S. K. Bagga*, R. Keith, G. G. Nair, A. V. Singh and R. V. S. V. Vadlamudi. (2002) Intellectual
Property Rights, Indian Journal of Pharmaceutical Sciences AI-PEAR-GP Discussion of the month, Nov-Dec
2002 Issue.
2. Dr. S.Bagga, A.V.Singh and S. Goswami. (2002) Gene Prediction: A New Frontier in Pharmaceutical
Research, II Anniversary Chronicle Pharmabiz Specials Dec 26 2002.
3. S. K. Bagga*, Laura McCarthy, S. Z. Rahman, K. Jhawar, S. Ahuja, N.Udupa, A. V. Singh and R. V. S. V.
Vadlamudi (2003). Power Plants: Green Pharmacy, Indian Journal of Pharmaceutical Sciences, May-June
2003.
4. S. K. Bagga*, A. V. Singh, Vibhav Garg, Sulip Goswami and R. V. S. V. Vadlamudi. (2003) Computer Aided
versus Wet Lab Drug Discovery, Indian Journal of Pharmaceutical Sciences, Jan-Feb 2003.
5. Singh AV, Knudsen KB and Knudsen TB (2005) Computational systems analysis of developmental toxicity:
design, development and implementation of a birth defects systems manager (BDSM). Reprod. Toxicol. 19: 421-
439.
6. Nemeth KA, Singh AV and Knudsen TB (2005) Searching for biomarkers of developmental toxicity with
microarrays: normal eye morphogenesis in rodent embryos. Toxicol Appl Pharmacol 206(2):219-28.
7. Knudsen KB, Singh AV and Knudsen TB (2005) Data input module for Birth Defects Systems Manager Reprod.
Toxicol. 20(3):369-75.
8. Kinane DF, Shiba H, Stathopoulou PG, Zhao H, Lappin DF, Singh AV, Eskan MA, Beckers S, Weigel S, Alpert
B and Knudsen TB (2006) Gingival epithelial cells heterozygous for Toll-like receptor 4 polymorphisms
Asp299Gly and Thr399lle are hypo-responsive to Porphyromonas gingivalis. Genes & Immunity Apr;7(3):190-
200.
9. Maia L. Green, AmarV. Singh, Yihzi Zhang, Kimberly A. Nemeth, Kathleen K. Sulik, and Thomas B. Knudsen.
(2007) Reprogramming of genetic networks During Initiation of the Fetal Alcohol Syndrome. Dev Dyn.
Feb;236(2):613-31.
10. AmarV Singh, Kenneth B Knudsen and Thomas B Knudsen. (2007) Integrative Analysis of the mouse
embryonic Transcriptome. Bioinformation, 1(10), 406-413.
11. AmarV Singh, Eric Rouchka, Greg Rempala, Caleb Bastian and Thomas B Knudsen. (2007) Integrative
Database Management for Mouse Development: Systems and Concepts Review. Birth Defects Research (Part
C) 81:1-19.
12. Deaciuc IV, Song Z, Peng X, Barve SS, Song M, He Q, Knudsen TB, Singh AV, and McClain CJ (2008)
Genome-wide transcriptome expression in the liver of a mouse model of high carbohydrate diet-induced liver
steatosis and its significance for the disease. Hepatol International, Volume 2, Number 1 (March, 2008) 2 :39-49
13. Barthold JS, McCahan, Singh AV, Knudsen TB, Si X, Campion L and Akins RE (2008) Altered expression of
muscle and cytoskeleton-related genes in a rat strain with inherited cryptorchidism. J. Androl. cryptorchidism. J
Androl. 29: 352-366
14. Rouchka EC, Phatak AW, and Singh AV. (2008), Effect of single nucleotide polymorphisms on Affymetrix®
match-mismatch probe pairs Bioinformation 2(9):405-11.
15. Benakanakere MR, Li Q, Eskan MA, Singh AV, Zhao J, Galicia JC, Stathopoulou P, Knudsen TB, Kinane DF
(2009) Modulation of TLR2 Protein Expression by miR-105 in Human Oral Keratinocytes. J Biol Chem.
284(34):23107-15
PHS 398/2590 (Rev. 09/04) Page _2 Biographical Sketch Format Page
Previous I TOC
-------
Principal Investigator/Program Director (Last, First, Middle):
16. Knudsen TB, Martin MT, Kavlock RJ, Judson RS, Dix DJ, and Singh AV (2009). Profiling the activity of
environmental chemicals in prenatal developmental toxicity studies using the U.S. EPA's ToxRefDB. Reprod
Toxicol. 28(2):209-19
17. Thomas B. Knudsen, Keith A. Houck, Richard S. Judson, Amar V. Singh, Arthur Weissman, Holly M.
Mortensen, David M. Reif, R. Woodrow Setzer, David J. Dix, and Robert J. Kavlock (2009) Biochemical
Activities of 309 ToxCast™ Chemicals Evaluated Across 239 Functional Targets. (Submitted to Nature Chem.
Biol)
18. Ema M, Iseb R, Katoc H, Onedad S, Hirosea A, Hirata-Koizumia M, Nishidac Y, Singh AV, Knudsen
TB and Ihara T (2009) Fetal malformations and early embryonic gene expression response in
cynomolgus monkeys maternally exposed to thalidomide Repro. Tox (Under Revision)
19. Singh AV, Yang C, Kavlock RJ, and Richard AM (2010) Developmental Toxicology Research Strategies:
Computational Toxicology. Comprehensive Toxicology, 2nd Edition (editors: GP Daston and TB Knudsen),
Elsevier: New York (Submitted)
PHS 398/2590 (Rev. 09/04)
Previous
Page _3
TOC
Biographical Sketch Format Page
-------
Principal Investigator/Program Director (Last, First, Middle):
Tan, Cecilia
BIOGRAPHICAL SKETCH
NAME
Tan, Cecilia
POSITION TITLE
Research Physical Scientist
EDUCATION
INSTITUTION AND LOCATION
National Cheng Kung University, Tainan, Taiwan
Harvard School of Public Health, Boston, MA
University of North Carolina at Chapel Hill,
Chapel Hill, NC
North Carolina State University
DEGREE
(if applicable)
B.S.
M.S.
Ph.D.
M.B.A.
YEAR(s)
1995
1997
2001
2009
FIELD OF STUDY
Environmental
Environmental Health
Environmental Sciences
and Engineering
Business Administration
A. POSITIONS and HONORS
Research and Professional Experience:
2009- Research Physical Scientist, National Exposure Research Laboratory, ORD,
USEPA
2003-2009 Associate Director, Center for Human Health Assessment, CUT Centers for
Health Research, RTP, NC
2001-2003 Post-doctoral Fellow, CUT Centers for Health Research, RTP, NC
1997-2001 NIOSH Trainee, The University of North Carolina at Chapel Hill, Chapel Hill, NC
1996-1997 Industrial Hygienist, Massachusetts Institute of Technology, Cambridge, MA
1993-1995 Research Assistant, National Cheng Kung University, Tainan, Taiwan
Selected Awards and Honors:
Risk Assessment Specialty Section Best Abstract Award, Society of Toxicology, New Orleans,
LA, 2005
Risk Assessment Specialty Section Best Abstract Award, Society of Toxicology, Salt Lake City,
UT, 2003
National Institute for Occupational Safety and Health pilot and research training grant, 2000
National Institute for Occupational Safety and Health traineeship award, 1998-2001
Department of Education Graduate Assistants in Areas of National Need (GAANN) Fellowship,
1997-1998
Hazardous Substances Academic Training Program Fellowship, 1996-1997
1994 Chinese Institute of Engineers Student Paper Contest Best Paper Award, Taiwan, 1994
The 5th Academic Essay Contest Best Essay Award, National Cheng Kung University, Taiwan,
1994
Invited Lectures/Symposia (selected):
Probabilistic reverse dosimetry: using pharmacokinetic modeling to estimate population-scale
distributions of exposure from biomonitoring data. Society of Risk Analysis Annual Meeting,
Boston, MA, December 2008.
Application of pharmacokinetic modeling to relate PFOA exposures and blood concentrations in
human populations. ISEA/ISEE Annual Meeting, Pasadena, CA, October 2008.
Conducting reverse dosimetry with physiologically based pharmacokinetics models to estimate
population exposure from data collected in populations-scaled biomonitoring studies.
Society of Risk Analysis Annual Meeting, Baltimore, MD, December 2006.
Application of pharmacokinetic modeling to estimate PFOA exposures associated with
measured blood concentrations in human populations. Society of Risk Analysis Annual
Page 1
Previous
TOC
-------
Principal Investigator/Program Director (Last, First, Middle): Tan, Cecilia
Meeting, Baltimore, MC, December 2006.
Reconstructing human chloroform exposure from biomonitoring data with a physiologically
based pharmacokinetic model. Society of Risk Analysis Annual Meeting, Orlando, FL,
December 2005.
Computational modeling of chloroform cytolethality and regenerative proliferation for cancer risk
assessment. Society of Risk Analysis Annual Meeting, Orlando, FL, December 2005.
The use of computational modeling in systems biology. NAASO Annual Scientific Meeting,
Vancouver, BC, October 2005.
Use of biologically based computational modeling in mode of action-based risk assessment - an
example of chloroform. U.S. EPA Workshop on Optimizing the design and interpretation of
epidemiologic studies to consider alternative disinfectants of drinking water, Raleigh, NC,
June 2005.
Scientific challenges: biomonitoring & its relevance to toxicology. Human biomonitoring: an
ICCA workshop for the global chemical industry. Charles de Gaulle, France, June 2005.
Modeling aggregate exposure and cumulative risk assessment of mixtures of common modes of
action. SOT Contemporary concepts in toxicology, Atlanta, GA, February 2005.
The use of physiologically based pharmacokinetic/pharmacodynamic modeling in quantitative
safety assessment. Predictive ADME/Toxicology Meeting. San Diego, CA, January 2005.
A physiologically based pharmacokinetic/pharmacodynamic model for N-methyl carbamate
pesticide carbaryl. The 25th American College of Toxicology Annual Meeting, Palm Springs,
CA, November 2004.
PBPK modeling to analyze human blood and exhaled breath for chloroform. Forum on
disinfection by-products - exploring the current science on disinfection by-products,
Research Triangle Park, NC, September 2004.
PBPK Modeling Courses:
A Course on Physiologically Based Pharmacokinetic (PBPK) Modeling in Drug Development
and Evaluation. Westin Alexandria, Alexandria, VA, 6-10 April, 2009.
2009 Society of Toxicology Annual Meeting Continuing Education Course - Characterizing
variability and uncertainty with physiologically based pharmacokinetic models. Title:
Variability in exposure and internal dosimetry assessed with PBPK models. Baltimore, MD,
15 March, 2009.
A Course on Physiologically Based Pharmacokinetic (PBPK) Modeling and Risk Assessment.
The Hamner Institutes for Health Research, Research Triangle Park, NC, 11-15 February,
2008.
A Short Course on Interpretation of Biomonitoring Data Using Physiologically Based
Pharmacokinetic (PBPK) Modeling. 2007Joint ISEA/ISEE Annual Meeting, Durham, NC, 14
October, 2007.
A Course on Interpretation of Biomonitoring Data Using Physiologically Based Pharmacokinetic
(PBPK) Modeling. CUT Centers for Health Research, Research Triangle Park, NC, 25-29
September, 2006.
A Course on Physiologically Based Pharmacokinetic (PBPK) Modeling and Risk Assessment.
CUT Centers for Health Research, Research Triangle Park, NC, 6-10 February, 2006.
A Course on Physiologically Based Pharmacokinetic (PBPK) Modeling and Risk Assessment.
CUT Centers for Health Research, Research Triangle Park, NC, 26-30 September, 2005.
Conference session chair:
Interpreting human biomonitoring data in the context of risk assessment: issues and challenges
2. Society of Risk Analysis Annual Meeting, Baltimore, MD, December 2006.
Forum on disinfection by-products - exploring the current science on disinfection by-products.
CUT Centers for Health Research, Research Triangle Park, NC, September 2004.
Page 2
Previous I TOC
-------
Principal Investigator/Program Director (Last, First, Middle): Tan, Cecilia
B. SELECTED PUBLICATIONS
Tan, Y., Clewell, H., Andersen, M. (2008) Time dependencies in perfluorooctylacids disposition
in rat and monkeys: a kinetic analysis. Toxicol. Lett. 177, 38-47.
Clewell, H., Tan, Y., Campbell, J., Andersen, M. (2008) Quantitative interpretation of human
biomonitoring data. Toxicol. Appl. Pharmacol. 231, 122-133.
Nong, A., Tan, Y., Krolski, M., Wang, J., Lunchick, C., Conolly, R., Clewell, H. (2008) Bayesian
calibration of a physiologically based pharmacokinetic/pharmacodynamic model of carbaryl
cholinesterase inhibition. J. Toxicol. Environ. Health A 71, 1363-1381.
Teeguarden, J., Bogdanffy, M., Covington, T., Tan, Y., Jarabek, A. (2008) A PBPK model for
evaluating the impact of aldehyde dehydrogenase polymorphisms on comparative rat and
human nasal tissue acetaldehyde dosimetry. Inhal. Toxicol. 20(4),375-90.
Dorman, D., Struve, M., Wong, B., Gross, E., Parkinson, C., Willson, G., Tan, Y., Campbell, J.,
Teeguarden, J., Clewell, H., Andersen, M. (2008) Derivation of an inhalation reference
concentration based upon olfactory neuronal loss in male rats following subchronic
acetaldehyde inhalation. Inhal Toxicol. 20(3),245-56.
Schroeter, J., Kimbell, J., Gross, E., Wilson, G., Dorman, D., Tan, Y., Clewell, H. (2008)
Application of physiological computational fluid dynamics models to predict interspecies
nasal dosimetry of inhaled acrolein. Inhal Toxicol 20(3),227-43.
Das, K., Grey, B., Zehr, R., Wood, C., Butenhoff, J., Chang, S., Ehresman, D., Tan, Y., Lau, C.
(2008) Effects of perfluorobutyrate exposure during pregnancy in the mouse. Toxicol Sci
105(1),173-81.
Chang, S., Das, K., Ehresman, D., Ellefson, M., Gorman, G., Hart, J., Noker, P., Tan, Y., Lieder,
P., Lau, C., Olsen, G., Butenhoff, J. (2008). Comparative pharmacokinetics of
perfluorobutyrate in rats, mice, monkeys, and humans and relevance to human exposure via
drinking water. Toxicol. Sci. 104(1)40-53.
Hays, S., Aylward, L, LaKind, J., Bartels, M., Barton, H., Boogaard, P., Brunk, C., DiZio, S.,
Dourson, M., Goldstein, D., Lipscomb, J., Kilpatrick, M., Krewski, D., Krishnan, K., Nordberg,
M., Okino, M., Tan, Y., Viau, C., Yager, J. (2008) Guidelines for the derivation of
Biomonitoring Equivalents: report from the biomonitoring equivalents expert workshop.
Regul. Toxicol. Pharmacol. 51 (S3), S4-S15.
LaKind, J., Aylward, L., Brunk, C., DiZio, S., Dourson, M., Goldstein, D., Kilpatrick, M., Krewski,
D., Bartels, M., Barton, H., Boogaard, P., Lipscomb, J., Krishnan, K., Nordberg, M., Okino,
M., Tan, Y., Viau, C., Yager, J., Hays, S. (2008). Guidelines for the communication of
Biomonitoring Equivalents: report from the Biomonitoring Equivalent Expert Workshop.
Regul. Toxicol. Pharmacol. 51 (S3), S16-S26
Tan, Y., Liao, K., Clewell, H. (2007) Reverse dosimetry: interpreting trihalomethanes
biomonitoring data using physiologically based pharmacokinetic modeling. J. Expo Sci
Environ Epidemiol 17,591-603.
Liao, K., Tan, Y., Clewell, H. (2007) Development of a screening approach to interpret human
biomonitoring data on volatile organic compounds: reverse dosimetry on biomonitoring data
for trichloroethylene. Risk Anal. 27(5), 1223-1236.
Liao, K., Tan, Y., Conolly, R., Borghoff, S., Gargas, M., Andersen, M., Clewell, H. (2007)
Bayesian estimation of pharmacokinetic and pharmacodynamic parameters in a mode-of-
action based cancer risk assessment for chloroform. Risk Anal. 27(6), 1535-1551.
Barton, H., Chiu, W., Setzer, W., Andersen, M., Bailer, A., Bois, F., Dewoskin, R., Hays, S.,
Johanson, G., Jones, N., Loizou, G., Macphail, R., Portier, C., Spendiff, M., Tan, Y. (2007)
Characterizing uncertainty and variability in physiologically based pharmacokinetic models:
state of the science and needs for research and implementation. Tox. Sci. 99, 395-402.
Tan, Y., Liao, K., Conolly, R., Blount, B., Mason, A., Clewell, H. (2006) Use of a physiologically
based pharmacokinetic model to identify exposures consistent with human biomonitoring
data for chloroform. J. Toxicol. Environ. Health 69, 1727-56.
Page 3
Previous I TOC
-------
Principal Investigator/Program Director (Last, First, Middle): Tan, Cecilia
Andersen, M., Clewell, H., Tan, Y., Butenhoff, J., Olsen, G. (2006) Pharmacokinetic modeling of
saturable, renal resorption of perfluoroalkylacids in monkeys - probing the determinants of
long plasma half-lives. Toxicol. 227, 156-164.
Tan, Y., Butterworth, B., Gargas, M., Conolly, R. (2003) Biologically motivated computational
modeling of chloroform cytolethality and regenerative cellular proliferation. Tox. Sci. 75, 192-
200.
Tan, Y., Flynn, M., Buller, T. (2002) Field evaluation of models for predicting worker exposure
during spray painting. Ann. Occup. Hyg. 46(1), 103-112.
Tan, Y. and Flynn, M. (2002) Methods for estimating the transfer efficiency of a compressed air
spray gun. Appl. Occup. Environ. Hyg. 17(1), 39-46.
Tan, Y. and Flynn, M. (2000) Experimental evaluation of a mathematical model for predicting
transfer efficiency of a high volume - low pressure air spray gun. Appl. Occup. Environ. Hyg.
15(10), 785-793.
Tan, Y., DiBerardinis, L, Smith, T. (1999) Exposure assessment of laboratory students. Appl.
Occup. Environ. Hyg. 14(8), 530-8.
Book Chapters
Qiang, Z., Tan, Y., Bhattacharya, S., Andersen, M. (2008) Computational systems biology
modeling of dosimetry and cellular response pathways. Drug Efficacy, Safety, and Biologies
Discovery - Emerging Technologies and Tools. Ed. Ekins and Xu. John Wiley & Sons, Inc.,
Hoboken, NJ.
Tan, Y. and Clewell, H. (2008) Probabilistic reverse dosimetry modeling for interpreting
biomonitoring data (submitted to Ed. Andersen and Krishnan).
Tan, Y., Yang, Y., Andersen, M., Clewell, H. (2008). Exposure science: Pharmacokinetic
modeling (submitted to Ed. Meliker, J.).
Page 4
Previous I TOC
-------
Principal Investigator/Program Director (Last, First, Middle): TiC6, Raymond R.
BIOGRAPHICAL SKETCH
NAME
Raymond R. Tice
eRA COMMONS USER NAME
POSITION TITLE
Chief, Biomolecular Screening Branch
National Toxicology Program
National Institute of Environmental Health
Sciences
EDUCATION/TRAINING (Begin with baccalaureate or other initial professional education, such as nursing, and include postdoctoral
training.)
INSTITUTION AND LOCATION
John Hopkins University
DEGREE
(if applicable)
Ph.D.
YEAR(s)
1976
FIELD OF STUDY
Biology
Raymond R. Tice, Ph.D. is Chief of the NTP Biomolecular Screening Branch (BSB). The BSB is
responsible for coordinating the NTP High Throughput Screening (HTS) Initiative and plays a
key role in the efforts of the Tox21 Community, which is an outgrowth of a 2008 Memorandum
of Understanding between the NTP, the NIH Chemical Genomics Center, and EPA's National
Center for Computational Toxicology to collaborate on the research, development, validation,
and translation of new and innovative test methods that characterize key steps in toxicity
pathways.
Tice has served as President of the Environmental Mutagen Society and as Vice-President of
the International Association of Environmental Mutagen Societies. He is the recipient of NIH
Director's Group Awards for activities associated with the NIH Molecular Libraries Initiative and
with the development of the ICCVAM Five-Year Plan (2008-2012). In late 2008, along with
Christopher Austin, Ph.D., of the NIH Chemical Genomics Center and Robert Kavlock, Ph.D., of
EPA's National Center for Computational Toxicology, Tice received the North American
Alternative Award from the Humane Society of the United States for "outstanding scientific
contributions to the advancement of viable alternatives to animal testing."
During his career, he has served on over 50 international expert panels and committees that are
primarily related genetic toxicology and more recently to validation of alternative test methods.
He has published 130 scientific papers and book chapters, edited four symposia proceedings
and contributed to 23 electronic review publications, in support of the NTP chemical nomination
process, and to 35 NICEATM-ICCVAM publications. Tice is a member of the editorial boards of
Mutation Research and Environmental and Molecular Mutagenesis.
Tice received his Ph.D. in biology in 1976 from Johns Hopkins University in Baltimore,
Maryland. He was employed by the Medical Department at Brookhaven National Laboratory,
Upton, New York from 1976 to1988, and by Integrated Laboratory Sciences, Inc., Durham,
North Carolina from 1988 to 2005, where his last position was as Senior Vice-President for
Research and Development. He joined NIEHS in 2005 as the Deputy Director of the NTP
Interagency Center for the Evaluation of Alternative Toxicological Methods (NICEATM) and was
promoted to Chief of the Biomolecular Screening Branch in 2008.
Selected Publications
1. Parham F, Austin C, Southall N, Xia M, Tice R, Portier C. Dose-response modeling of high-
throughput screening data. Submitted to Journal of Biomolecular Screening.
PHS 398/2590 (Rev. 09/04, Reissued 4/2006)
Page J
Previous
TOC
-------
Principal Investigator/Program Director (Last, First, Middle): TiC6, Raymond R.
2. Xia M, Huang R, Sun Y, Semenza GL, Aldred SF, Witt KL, Inglese J, Tice RR, Austin CP.
Identification of chemical compounds that induce 1 HIF-1a activity. Toxicological Sciences 2009;
doi:10.1093/toxsci/kfp123.
3. Kavlock RJ, Austin CP, Tice RR. Toxicity Testing in the 21st Century: Implications for Human
Health Risk Assessment, Risk Analysis 2009; 29(4):485-487.
4. Witt KL, Livanos E, Kissling GE, Torous DK, Caspary W, Tice RR, Recio L. Comparison of
flow cytometry- and microscopy-based methods for measuring micronucleated reticulocyte
frequencies in rodents treated with nongenotoxic and genotoxic chemicals. Mutation Res 2008;
649:101-113.
5. Xia M, Huang R, Witt KL, Southall N, Fostel J, Cho M-H, Jadhav A, Smith CS, Inglese J,
Portier CJ, Tice RR, Austin CP. Compound cytotoxicity profiling using quantitative high-
throughput screening. Environ Health Perspect 2007; 116:284-291.
6. Balls M, Amcoff P, Bremer S, Casati S, Coecke S, Clothier R, Combes R, Corvi R, Curren R,
Eskes C, Fentem J, Gribaldo L, Haider M, Hartung T, Hoffmann S, Schectman L, Scott L,
Spielmann H, Stokes W, Tice R, Wagner D, Zuang V. The principles of weight of evidence
validation of test methods and testing strategies. The report and recommendations of ECVAM
workshop 58. Altern Lab Anim. 2006; 34(6):603-20.
7. Burlinson B, Tice RR, Speit G, Agurell E, Brendler-Schwaab SY, Collins AR, Escobar P,
Honma M, Kumaravel TS, Nakajima M, Sasaki YF, Thybaud V, Uno Y, Vasquez M, Hartmann
A. Fourth International Workgroup on Genotoxicity Testing: Results of the in vivo Comet Assay
Workgroup. Mutation Res 2006; 627:31-35.
8. Manson J, Brabec MJ, Buelke-Sam J, Carlson GP, Chapin RE, Favor JB, Fischer LJ, Hattis
D, Lees PS, Perreault-Darney S, Rutledge J, Smith TJ, Tice RR, and Working P. NTP-CERHR
expert panel report on the reproductive and developmental toxicity of acrylamide. Birth Defects
Res B Dev Reprod Toxicol 2005; 74:17-113.
9. Witt KL, Tice RR, Wolfe G, Bishop JB. Genetic damage detected in CD-1 mouse pups
exposed perinatally to 3'-azido-3'-deoxythymidine or dideoxyinosine via maternal dosing,
nursing, and direct gavage. II: Effects of the individual agents compared to combination
treatment. Environ Mol Mutagen. 2004; 44:321-328.
10. Hartmann A, Agurell E, Beevers C, Brendler-Schwaab S, Burlinson B, Clay P, Collins A,
Smith A, Speit G, Thybaud V, Tice RR. Recommendations for conducting the in vivo alkaline
Comet assay. Mutagenesis 2003; 18:45-51.
11. Tice RR, Agurell E, Anderson D, Burlinson B, Hartmann A, Kobayashi H, Miyamae Y, Rojas
E, Ryu JC, Sasaki Y. The single cell gel/comet assay: guidelines for in vitro and in vivo genetic
toxicology testing. Environ Mol Mutagen 2000; 35:206-221.
PHS 398/2590 (Rev. 09/04, Reissued 4/2006) Page 2
Previous I TOC
-------
Principal Investigator/Program Director (Last, First, Middle): Wambaugh, John F.
BIOGRAPHICAL SKETCH
Provide the following information for the key personnel and other significant contributors in the order listed on Form Page 2.
Follow this format for each person. DO NOT EXCEED FOUR PAGES.
NAME
John F. Wambaugh
POSITION TITLE
Physical Scientist
EDUCATION/TRAINING (Begin with baccalaureate or other initial professional education, such as nursing, and include postdoctoral training.)
INSTITUTION AND LOCATION
University of Michigan, Ann Arbor, Ml
Georgia Institute of Technology, Atlanta, GA
Duke University, Durham, NC
Duke University, Durham, NC
NCCT, US EPA, Research Triangle Park, NC
DEGREE
(if applicable)
B.S.
M.S.
M.S.
Ph.D.
YEAR(s)
1995-1999
1999-2001
2001-2005
2001-2006
2006-2008
FIELD OF STUDY
Physics
Physics
Computer Science
Physics
Statistical Analysis of
Biological Models
A. POSITIONS and HONORS
Research and Professional Experience:
2008-present
2006-2008
2002-2006
2001-2003
2001
1999-2001
Physical Scientist, National Center for Computational Toxicology, US EPA, RTP, NC
Postdoctoral Physicist, National Center for Computational Toxicology, US EPA, RTP, NC
(mentors: Hugh Barton and Woodrow Setzer)
Research Assistant, Department of Physics, Duke University, Durham NC
(advisor: Robert Behringer)
Teaching Assistant, Department of Physics, Duke University, Durham, NC
Visiting Student, Center for Nonlinear Science, Los Alamos National Laboratory, Los Alamos,
NM (advisors: Charles Reichhardt, Cynthia Olson-Reichhardt)
Teaching Assistant, School of Physics, Georgia Institute of Technology, Atlanta, GA
Professional Societies and Affiliations:
2007-present
2006-present
2001-present
1995-present
Member, Sigma Xi, Duke University Chapter
Member, Society of Toxicology: Biological Modeling and Risk Assessment Specialty Sections,
North Carolina Society of Toxicology
Member, American Physical Society: Division of Fluid Dynamics, Statistical and Nonlinear
Physics Topical Group, Forum on Graduate Student Affairs
Member, American Association of Physics Teachers
Honors and Awards:
2006 Dynamics Days Student Travel Award,
Dynamics Days 2006 Conference in Baltimore, MD
2005 Duke University Graduate School Student Travel Award,
American Physical Society, Division of Fluid Dynamics Meeting in Chicago, IL
2004 Duke University Graduate School Student Travel Award,
American Physical Society, Division of Fluid Dynamics Meeting in Seattle, WA
2003 Duke University Graduate School Student Travel Award,
American Physical Society, Division of Fluid Dynamics Meeting in East Rutherford, NJ
Updated 11/13/2007
Page 1
Previous
TOC
-------
Principal Investigator/Program Director (Last, First, Middle): Wambaugh, John F.
B. SELECTED PUBLICATIONS (selected from 14 total).
1. "Modeling Single and Repeated Dose Pharmacokinetics of PFOA in Mice," I. Lou, J.F. Wambaugh, C.
Lau, R.H. Hanson, A.B. Lindstrom, M.J. Strynar, R.D. Zehr, and H.A. Barton, Toxicological Sciences
107, 331-341 (2009)
2. "Comparing Models for PFOA Pharmacokinetics Using Bayesian Analysis," J.F. Wambaugh, H.A.
Barton, and R.W. Setzer, Journal of Pharmacokientics and Pharmacodynamics 35, 683-712 (2008)
3. Wambaugh, J.F., Majmudar, T.S., Tighe, B.P., Socolar, J.E.S. and Behringer, R.P., "Experimental
observation of spatial scaling in dense granular matter," in preparation
4. Wambaugh, J.F., Hartley, R.R., and Behringer, R.P., "Force networks and elasticity in granular silos,"
arXiv: 0801.3387
5. Wambaugh, J.F., Matthews, J.V., Gremaud, P.A. and Behringer, R.P., "Response to perturbations in
granular flow," Physical Review E 76, 051303 (2007)
6. Wambaugh, J.F., "Graph Percolation as an Analog to Granular Force Networks," cond-mat/0603314,
(2006)
7. Mort, P., McKenzie, K., Wambaugh, J.F., and Behringer, R.P., "Granular flow through an Orifice -
Effect of granule size and shape distributions," Proceedings of Fifth World Congress on Particle
Technology (2006)
8. Wambaugh, J.F. and Behringer, R.P., "Asymmetry-induced circulation in granular hopper flows,"
Powders & Grains 2005, pages 915-918 (2005).
9. Wambaugh, J.F., Marchesoni, F. and Nori, F., "Shear and Loading in Channels: Oscillatory Shearing
and Edge Currents of Superconducting Vortices," Physical Review B 67, 144515 (2003)
10. Wambaugh, J.F., Reichhardt C., and Olson, C.J., "Ratchet-Induced Segregation and Transport of Non-
Spherical Grains," Physical Review E 65, 031308 (2002)
11. Wambaugh, J.F., Reichhardt, C., Olson, C.J., Marchesoni, F., and Nori, F., "Superconducting Fluxon
Pumps and Lenses," Physical Review Letters 83, 5106 (1999)
C. PRESENTATIONS WITH ABSTRACTS
1. "Examining Models for the Pharmacokinetics of Perfluorooctanoic acids", Annual Meeting of the Society
of Toxicology in Baltimore, Maryland, March 2009 (talk)
2. "Enhancing the Modeling of PFOA Pharmacokinetics with Bayesian Analysis," Annual Meeting of the
Society of Toxicology in Seattle, Washington, March 2008 (poster)
3. "Assessing Uncertainty in the Toxicology of PFOA," International Science Forum on Computational
Toxicology, Research Triangle Park, North Carolina, May 2007 (poster)
4. "Bayesian Analysis of Parameters for Pharmacokinetic Models," Annual Meeting of the Society of
Toxicology in Charlotte, North Carolina, March 2007 (poster)
5. "Joint Analysis of PFOA Plasma Concentration and Excretion Data," Perfluoroalkyl Acids and Related
Chemistries, Society of Toxicology Workshop in Arlington, Virginia, February 2007 (poster)
6. "Spatial Distribution of Forces within Granular Materials," APS Division of Fluid Dynamics Annual
Meeting in Tampa, Florida, November 2006 (talk)
7. "Impact of Particle Elasticity on Granular Force Networks," APS March Meeting in Baltimore, Maryland,
March 2006 (talk)
8. "Square Amplitude Granular Waves," Dynamics Days 2006 Meeting in Bethesda, Maryland, January
2006 (talk)
9. "Sensitivity of Granular Hopper Flows to Boundary Conditions," APS Division of Fluid Dynamics Annual
Meeting in Chicago, Illinois, November 2005 (talk)
10. "Asymmetry-induced circulation in granular hopper flows," Powders and Grains 2005, Stuttgart,
Germany, June 2005 (poster)
11. "Circulation in Asymmetric Granular Hoppers," APS Division of Fluid Dynamics Annual Meeting in
Seattle, Washington, November 2004 (talk)
12. "Observed Deviations from Janssen Model in Granular Silos," Dynamics Days 2004 Meeting in Chapel
Hill, North Carolina, January 2004 (poster)
13. "Elastic Effects in Granular Pressure Profiles," APS Division of Fluid Dynamics Annual Meeting in East
Rutherford, New Jersey, November 2003 (talk)
14. "Asymmetry-Induced Circulation in Conical Granular Flows," APS Division of Fluid Dynamics Annual
Updated 11/25/2008 Page 2
Previous I TOC
-------
Principal Investigator/Program Director (Last, First, Middle): Wambaugh, John F.
Meeting in East Rutherford, New Jersey, November 2003 (talk)
15. "Ratchet-Induced Segregation and Transport of Non-Spherical Grains," APS March Meeting in
Indianapolis, Indiana, March 2002 (talk)
16. "A New System for Accessing Transfer Function Coefficients for an Architectural Computer-Aided
Thermal Optimization Tool," Fifth International Building Performance Simulation Association Meeting,
Prague, Czech Republic, September 1997 (poster)
D. INVITED PRESENTATIONS
1. "Biological Modeling for Toxicology at the US Environmental Protection Agency," Department of
Physics and Chemistry, Coastal Carolina University, Conway, South Carolina, April 2009
E. WORKSHOPS
1. "PFAA Days II", Environmental Protections Agency, Research Triangle Park, North Carolina, June
2008
2. "Perfluoroalkyl Acids Research Planning Workshop," Environmental Protection Agency, Research
Triangle Park, North Carolina, August 2007
3. "Perfluoroalkyl Acids and Related Chemistries: Toxicokinetics and Mode-of-Action," Society of
Toxicology, Arlington, Virginia, February 2007
4. "Uncertainty and Variability in Pharmacokinetic Models", Environmental Protection Agency, Research
Triangle Park, North Carolina, October 2006
5. "Multiscale Model Development and Control Design Workshop on Fluctuations and Continuum
Equations for Granular Flow", Statistical and Applied Mathematical Sciences Institute, Research
Triangle Park, North Carolina, April 2004
F. CONTINUING EDUCATION
1. "Characterizing Variability and Uncertainty with Physiologically-Based Pharmacokinetic Models,"
Society of Toxicology, Baltimore, Maryland, March 2009
2. "Characterizing Modes-of-Action and Their Relevance in Assessing Human Health Risks," Society of
Toxicology, Baltimore, Maryland, March 2009
3. "Beginning and Intermediate Modeling and Simulation Techniques using acsIX, with Applications to
Computational Biology," Aegis Software, Research Triangle Park, North Carolina, May 2008
4. "Dose-Response Modeling for Occupational and Environmental Risk Assessment," Society of
Toxicology, Seattle, Washington, March 2008
5. "Use of Data for Development of Uncertainty Factors in Non-Cancer Risk Assessment," Society of
Toxicology, Seattle, Washington, March 2008
6. "Physiologically Based Pharmacokinetic Modeling for Risk Assessment Applications," Society of
Toxicology, Charlotte, North Carolina, March 2007
7. "Fundamentals of Human Health Risk Assessment with a Case Study Approach," Society of
Toxicology, Charlotte, North Carolina, March 2007
8. "Interpretation of Biomonitoring Data Using Physiologically Based Pharmacokinetic Modeling," CUT,
Center for Human Health Assessment, Research Triangle Park, North Carolina, September 2006
Updated 11/25/2008 Page 3
Previous I TOC
-------
Principal Investigator/Program Director (Last, First, Middle): Welsh, William James
BIOGRAPHICAL SKETCH
Provide the following information for the key personnel and other significant contributors in the order listed on Form Page 2.
Follow this format for each person. DO NOT EXCEED FOUR PAGES.
NAME
William J. Welsh
eRA COMMONS USER NAME
Welsh04
POSITION TITLE
Norman H. Edelman Professor in Bioinformatics &
Molecular Design, Department of Pharmacology
EDUCATION/TRAINING (Begin with baccalaureate or other initial professional education, such as nursing, and include postdoctoral training.)
INSTITUTION AND LOCATION
St. Joseph's University, Phila, PA
University of Pennsylvania, Phila, PA
University of Cincinnati, Cincinnati, OH
DEGREE
(if applicable)
B.S.
Ph.D.
Postdoc
YEAR(s)
1969
1975
1979-82
FIELD OF STUDY
Chemistry
Physical Chemistry
Comput Biophys Chem
A. Positions and Honors
Positions and Employment
1975-79 Research Chemist, Procter & Gamble Co. (Cinti., OH)
1979-86 Professor of Chemistry, College of Mount St. Joseph
Research Associate Professor of Chemistry, University of Cincinnati
1986-90 Assistant Professor of Chemistry, Univ. of Missouri-St. Louis (UM-St. Louis)
1990-95 Associate Professor, UM-St. Louis
1995-2001 Professor, UM-St. Louis
2001- Professor, Dept. of Pharmacology, UMDNJ-RWJMS
2001- Director, UMDNJ Informatics Institute
2003- Norman H. Edelman Chaired Professorship, UMDNJ-RWJMS
2005- Director, UMDNJ Environmental Bioinformatics & Computational Toxicology Center
Other Experience and Professional Memberships
1985 Extramural Research Associate, National Institutes of Health, Bethesda, MD
1991-1999 Member, Editorial Board, Journal of Computational and Theoretical Polymer Science
1997-2001 Associate Director, UM-St. Louis Ctr. for Molecular Electronics
1999-2001 Associate Director, University of Missouri Bioinformatics Center
1999-2001 Director, UM-St. Louis Center for Cheminformatics
2000-2003 Member, Editorial Advisory Bd, Journal of Computer Information and Chemical Sciences
2001- Member, Cancer Institute of New Jersey
2002- Member of Graduate Faculty, UMDNJ-Rutgers U. Environ. & Occup. Health Sci. Inst. (EOHSI)
2002- Member, New Jersey Center for Biomaterials
2003- Member of Graduate Faculty, Medicinal Chemistry, School of Pharmacy, Rutgers Univ.
2003- Editorial Board, Journal of Molecular Graphics & Modelling
2003- Editorial Board, Chemical Research in Toxicology Journal
2005- Editorial Board, Cancer Informatics Journal
Honors
1985 Teacher of the Year Award, College of Mount St. Joseph, Ohio
1998 St. Louis Award, St. Louis Section of the American Chemical Society
2001 Entrepreneur of the Year Award, University of Missouri
2001 Chancellor's Award for Research and Creativity, UM-St. Louis
2003 Norman H. Edelman Endowed Professorship in Bioinformatics, UMDNJ-RWJMS
2004 John C. Krantz, Jr. Lectureship Award, University of the Sciences in Philadelphia (USP).
2009 Honorary Society Plenary Lectureship, Georgia State University, Atlanta GE
B. Selected peer-reviewed publications in chronological order (from over 400 publications)
Tumor-targeted bioconjugate based delivery of camptothecin: design, synthesis and in vitro evaluation,
Paranjpe PV, Chen Y, Kholodovych V, Welsh WJ, Stein S, Sinko PJ. J Control Release 100, 275-92 (2004).
1
Previous
TOC
-------
Principal Investigator/Program Director (Last, First, Middle): Welsh, William James
Structural model of the Plasmodium CDK, Pfmrk, a novel target for malaria therapeutics. Peng Y, Keenan SM,
Welsh WJ. J Mol Graph Model. 24(1):72-80 (2005).
Rational inhibitor design and iterative screening for identification of plasmodial cyclin dependent kinase
inhibitors, Keenan SM, Geyer JA, Welsh WJ, Prigge ST, Waters NC. Comb Chem High Throughput Screen
8(1):27-38 (2005).
Discovery of novel triazole-based opioid receptor antagonists. Zhang Q, Keenan SM, Peng Y, Nair AC, Yu SJ,
Howells RD, Welsh WJ. J Med Chem. 49(14):4044-7(2006).
Discovery of broad spectrum protein kinase inhibitors to probe the malarial cyclin dependent protein kinase
Pfmrk. Woodard CL, Keenan SM, Gerena L, Welsh WJ, Geyer JA, Waters NC. Bioorg Med Chem Lett.
17(17):4961-6(2007).
Highly Potent Triazole-Based Tubulin Polymerization Inhibitors. Zhang Q, Peng Y, Wang XI, Keenan SM,
Arora S, Welsh WJ. J. Med. Chem. 50, 749-754 (2007).
Shape signatures: new descriptors for predicting cardiotoxicity in silico. Chekmarev DS, Kholodovych V,
Balakin KV, Ivanenkov Y, Ekins S, Welsh WJ. Chem Res Toxicol. 21(6):1304-14 (2008).
New predictive models for blood-brain barrier permeability of drug-like molecules. Kortagere S, Chekmarev D,
Welsh WJ, Ekins S. Pharm Res. 25(8):1836-45 (2008).
Novel microtubule polymerization inhibitor with potent antiproliferative and antitumor activity. Arora S, Wang XI,
Keenan SM, Andaya C, Zhang Q, Peng Y, Welsh WJ. Cancer Research. 69(5): 1910-5 (2009).
Specific interactions between the viral coreceptor CXCR4 and the biguanide-based compound NB325 mediate
inhibition of human immunodeficiency virus type 1 infection. Thakkar N, Pirrone V, Passic S, Zhu W,
Kholodovych V, Welsh WJ, Rando RF, Labib ME, Wgdahl B, Krebs FC. Antimicrob Agents Chemother.
53(2):631-8 (2009).
Evaluations of the trans-sulfuration pathway in multiple liver toxicity studies. Schnackenberg LK, Chen M, Sun
J, Holland RD, Dragan Y, long W, Welsh W, Beger RD. Toxicol Appl Pharmacol. 235(1):25-32 (2009).
Hybrid scoring and classification approaches to predict human pregnane X receptor activators. Kortagere S,
Chekmarev D, Welsh WJ, Ekins S. Pharm Res. 26(4):1001-11 (2009).
Novel microtubule polymerization inhibitor with potent antiproliferative and antitumor activity. Arora S, Wang XI,
Keenan SM, Andaya C, Zhang Q, Peng Y, Welsh WJ. Cancer Res. 69(5): 1910-5 (2009).
The major human pregnane X receptor (PXR) splice variant, PXR.2, exhibits significantly diminished ligand-
activated transcriptional regulation. Lin YS, Yasuda K, Assem M, Cline C, Barber J, Li CW, Kholodovych V, Ai
N, Chen JD, Welsh WJ, Ekins S, Schuetz EG. Drug Metab Dispos. 37(6): 1295-304 (2009).
ebTrack: an environmental bioinformatics system built upon ArrayTrack. Chen M, Martin J, Fang H, Isukapalli
S, Georgopoulos PG, Welsh WJ, Tong W. BMC Proc. 3 Suppl 2:S5 (2009).
Structure-activity relations of nanolipoblockers with the atherogenic domain of human macrophage scavenger
receptor A. Plourde NM, Kortagere S, Welsh W, Moghe PV. Biomacromolecules. 10(6):1381-91 (2009).
Understanding nuclear receptors using computational methods. Ai N, Krasowski MD, Welsh WJ, Ekins S. Drug
Discov Today 14(9-10):486-94 (2009).
NetCSSP: web application for predicting chameleon sequences and amyloid fibril formation. Kim C, Choi J,
Lee SJ, Welsh WJ, Yoon S. Nucleic Acids Res. 37, W469-73 (2009).
Predicting inhibitors of acetylcholinesterase by regression and classification machine learning approaches with
combinations of molecular descriptors. Chekmarev D, Kholodovych V, Kortagere S, Welsh WJ, Ekins S. Pharm
Res. 26(9):2216-24 (2009).
Application of Screening Methods, Shape Signatures and Engineered Biosensors in Early Drug Discovery
Process. Hartman I, Gillies AR, Arora S, Andaya C, Royapet N, Welsh WJ, Wood DW, Zauhar RJ. Pharm Res.
Jul 22 (2009). [Epub ahead of print]
Novel delta opioid receptor agonists exhibit differential stimulation of signaling pathways. Peng Y, Zhang Q,
Arora S, Keenan SM, Kortagere S, Wannemacher KM, Howells RD, Welsh WJ. Bioorg Med Chem. Jul 9
(2009). [Epub ahead of print]
Previous I TOC
-------
BIOGRAPHICAL SKETCH
NAME
Fred A. Wright
eRA COMMONS USER NAME
Fred_Wright
POSITION TITLE
Professor
EDUCATION/TRAINING (Begin with baccalaureate or other initial professional education, such as nursing, and include postdoctoral training.)
INSTITUTION AND LOCATION
SUNY at Buffalo, New York
The University of Chicago
DEGREE
(if applicable)
B.A.
PhD
YEAR(s)
1989
1994
FIELD OF STUDY
Psychology and
Statistics
Statistics
Positions and Employment
1994-1997
1997-Jan2002
Feb 2002-2008
Jul 2008-present
Assistant Adjunct Professor, Family & Preventive Medicine, University of California, San
Diego.
Assistant Professor, Division of Human Cancer Genetics, The Ohio State University,
Columbus, OH.
Associate Professor, Department of Biostatistics, University of North Carolina, Chapel Hill.
Professor, Department of Biostatistics, University of North Carolina, Chapel Hill.
Honors and Awards
2004 Elected to Delta Omega Public Health Honor Society
Phi Beta Kappa
Other Experience and Professional Memberships
Professional memberships in American Association for the Advancement of Science, American Society of
Human Genetics, American Statistical Association.
Selected peer-reviewed publications (in chronological order)
Becker, LB, Han B, Meyer P, Wright FA, Rhodes K, Smith D, Barrett J: Racial differences in the incidence of
cardiac arrest and subsequent survival. New England Journal of Medicine, 329: 600-606, 1993.
Kong A and Wright F: Asymptotic theory for gene mapping. Proceedings of the National Academy of
Sciences, USA, 91: 9705-9709, 1994.
Takiyyuddin MA, Parmer RJ, Kailasam MT, Cervenka JH, Kennedy B, Ziegler M, Lin MC, Li J, Grim CE,
Wright FA, O'Connor DT: Chromogranin A in human hypertension: influence of heredity. Hypertension, 26:
213-220, 1995.
Winqvist R, Hampton G, Mannerma A, Blanco G, Alavaikko M, Kiviniemi H, Taskinen PJ, Evans G, Wright FA,
Newsham I, Cavenee W: Loss of heterozygpsity for chromosome 11 in primary human breast tumors is
associated with poor survival after metastasis. Cancer Research, 55: 2660-2664, 1995.
Paulson TG, Wright FA, Parker BA, Russak V, Wahl GM: Microsatellite instability correlates with reduced
survival and poor disease prognosis in breast cancer. Cancer Research, 56:4021-4026, 1996.
Wright FA: The phenotypic difference discards sibpair QTL linkage information. American Journal of Human
Genetics, 60: 740-742, 1997.
Wright FA and Kong A: Linkage mapping in experimental crosses: the robustness of single-gene models.
Genetics, 146: 417-425, 1997.
Rock CL, Flatt S, Wright FA, Faerber S, Newman V, Kealey S, Pierce JP: Responsiveness of carptenoids to a
high-vegetable diet intervention to prevent breast cancer recurrence. Cancer Epidemiology, Biomarkers,
and Prevention, 6: 617-623, 1997.
Pierce JP, Faerber S, Wright FA, Rock CL, Newman V, Flatt S, Kealey S, Hryniuk W: Feasibility of a
randomized trial of a high-vegetable diet to prevent breast cancer. Nutrition and Cancer, 28:282-288, 1997.
Rock CL, Newman V, Flatt SW, Faerber SF, Wright FA, Pierce, JP: Nutrient intakes from foods and
dietary supplements in women at risk for breast cancer recurrence. Nutrition and Cancer, 29:122-139,
1997.
Dao TT, Kailasam MT, Parmer RJ, Le HV, LeVerge RL, Kennedy BP, Ziegler MG, Insel PA, Wright FA,
O'Connor DT: Expression of altered alpha-2 adrenergic phenotypic traits in normotensive humans at
genetic risk of hereditary (essential) hypertension. Journal of Hypertension, 16: 779-792, 1998.
Newman V, Rock CL, Faerber S, Flatt SW, Wright FA, Pierce JP: Dietary supplement use by women at risk
for breast cancer recurrence. The Women's Healthy Eating and Living Study Group. Journal of the
American Dietetic Association, 98: 285-292, 1998.
PHS 398/2590 (Rev. 09/04, Reissued 4/2006)
Page
Biographical Sketch Format Page
-------
Rayburn K, Martinez R, Escobedo M, Wright FA, Farias M: Glycemic effects of various species of nopal
(opuntia sp.) in type 2 diabetes mellitus. Texas Journal of Rural Health, 16: 68-76, 1998.
Hryniuk W, Frei E, Wright FA: A single scale for comparing dose-intensity of all chemotherapy regimens in
breast cancer: Summation dose-intensity. Journal of Clinical Oncology, 16: 3137-3147, 1998.
De La Chapelle A, Wright FA: Linkage disequilibrium mapping in isolated populations: The example of Finland
revisited. Proceedings of the National Academy of Sciences, USA, 95: 12416-12423, 1998.
Sadler GR, Thomas AG, Dhanjal SK, Gebrekristos B, Wright FA: Breast cancer screening adherence in
African-American women - Black cosmetologists prompting health. Cancer, 83:1836-1839, 1998.
Fierer J, Walls L, Wright F, Kirkland TN: Genes influencing resistance to Coccidioides immitis and the interleukin-10
response map to chromosomes 4 and 6 in mice. Infect Immun, 67:2916-2919,1999.
Wright, FA, O'Connor, DT, Yoneda, LU, Kutey, G, Roberts, E, Berry, C, Weber, JL, Timberlake, D, Schlager, G:
Genome scan for blood pressure loci in mice. Hypertension, 34:625-630,1999.
Lin S, Irwin ME, Wright FA: A multiple locus analysis of the COGA data set. Genetic Epidemiology 17 (Suppl
7j: S229-234, 1999.
O'Connor DT, Takiyyuddin MA, Printz MP, Dinh TO, Barbosa JA, Rozansky DJ, Mahata SK, Wu H, Kennedy
BP, Ziegler MG, Wright FA, Schlager G, Parmer RJ: Catecholamine storage vesicle protein expression in
genetic hypertension. Blood Pressure 8: 285-295, 1999.
Costello JF, Fruhwald MC, Smiraglia DJ, Rush LJ, Robertson GP, Gao X. Wright FA, Feramisco JD,
Peltomaki P, Lang JC, Schuller DE, Yu L, Bloomfield CD, Caligiuri MA, Yates A, Nishikawa R, Huang H-J
S, Petreilli NJ, Zhang X, O'Dorisio MS, Held WA, Cavenee WK, Plass C: Aberrant CpG island methylation
has non-random and tumor type-specific patterns. Nature Genetics 24:132-138, 2000.
Hoffman HM, Wright FA, Broide DH, Wanderer AA, Kolodner RD: Identification of a locus on chromosome
1q44for Familial Cold Urticaria. American Journal of Human Genetics 66:1693-1698, 2000.
Borrego S, Ruiz A, Saez ME, Gimm O, Gao X, Lopez-Alonso M, Wright FA, Antinolp G, Eng C: RET
genotypes comprising specific haplptypes of polymorphic variants predispose to isolated Hirschsprung
disease. Journal of Medical Genetics 37: 572-578, 2000.
Desai DC, Lockman JC, Chadwick RB, Gao X, Percesepe A, Evans GR, Miyaki M, Yuen ST, Radice P, Maher
ER, Wright FA, de la Chapelle A: Recurrent germline mutation in MSH2 arises frequently de novo.
Journal of Medical Genetics 37: 646-652, 2000.
Lin S, Cheng R, Wright FA: Genetic crossover interference in the human genome. Annals of Human Genetics,
65:79-93,2001.
Virtaneva Kl, Wright FA, Tanner SM, Yuan B, Lemon WJ, Caligiuri MA, Bloomfield CD, de la Chapelle A,
Krahe R: Gene expression profiling reveals fundamental biological differences in AML with isolated trisomy
8 and normal cytogenetics. Proceedings of the National Academy of Sciences USA, 98:1124-1129, 2001.
Huang J, Kuismanen SA, Liu T, Chadwick RB, Johnson CK, Stevens MW, Richards SK, Meek JE, Gao X,
Wright FA, Mecklin JP, Jarvinen HJ, Gronberg H, Bisgaard ML, Lindblom A, Peltomaki P: MSH6 and
MSH3 are rarely involved in genetic predisposition to non-polypotic colon cancer. Cancer Research, 61:
1619-1623,2001.
D, Zhao WD, Wright FA, Yuan H-Y, Wang J-P, Sears R, BaerT, Kwon D-H, ordon D, Gibbs S, Dai D, Yang Q,
Spitzner J, Krahe R, Stredney D, Stutz A, Yuan B: Assembly, annotation and integration of UniGene
clusters into the human genome draft. Genome Research, 11: 904-918, 2001.
Wang D, Cheng R, Gao X, Lin S, Wright FA: Transformation of sibpair values for the Haseman-Elston
method. American Journal of Human Genetics 68:1238-1249, 2001.
Rush LJ, Dai Z, Smiraglia DJ, Gao X, Wright FA, Fruhwald M, Costello JF, Held WA, Yu L, Krahe R, Kolitz JE,
Bloomfield CD, Caligiuri MA, Plass C: Novel methylation targets in de novo acute myeloid leukemia with
prevalence of chromosome 11 loci. Blood, 97:3226-33, 2001.
Wright FA, Lemon WJ, Zhao WD, Sears R, Zhuo D, Wang J-P, Yang H-Y, Baer T, Stredney D, Spitzner J,
Stutz A, Krahe R, Yuan B: A draft annotation and overview of the human genome. Genome Biology, 2:
research0025.1-0025.18, 2001.
Fruhwald MC, O'Dorisio SM, Smith L, Dai Z, Wright FA, Paulus W, Jurgens H, Plass C: Hypermethylation as a
potential prognostic factor and a clue to a better understanding of the molecular pathogenesis of
medulloblastoma - results of a genomewide methylation scan. Klinische Padiatrie, 213: 1-7, 2001.
Smiraglia DJ, Rush LJ, Fruhwald MC, Dai Z, Held, WA, Costello JF, Lang JC, Eng C, Li B, Wright FA, Caligiuri
MA, Plass C: Excessive CpG island hypermethylation in cancer cell lines versus primary human
malignancies. Human Molecular Genetics, 10: 1413-1419, 2001.
Fruhwald MC, O'Dorisio SM, Dai Z, Tanner SM, Balster DA, Gao X, Wright FA, Plass C: Aberrant promoter
methylation of novel rather than known methylation targets is a common abnormality in medulloblastomas
- Implications for tumor biology and potenial clinical utility. Oncogene 20: 5033-5042, 2001.
Dai Z, Lakshmanan RR, Zhu W-G, Smiraglia DJ, Rush LJ, Fruhwald MC, BrenaRM, Li B, Wright FA, Ross P,
OttersonGA, Plass C. Global Methylation Profiling of Lung Cancer Identifies Novel Methylated Genes.
Neoplasia, 3: 314-323, 2001.
-------
Wu L, Saavedra HI, Timmers C, Sang L, Nuckolls F, Nevins JR, Wright FA, Robinson ML, Leone G: the E2F1,
E2F2, and E2F3 transcription activators are essential for cellular proliferation. Nature, 414: 457-62, 2001.
Huang Y, Prasad M, Lemon WJ, Hampel H, Wright FA, Kornacker K, LiVoIsi V, Frankel W, Kloos RT, Eng C,
Pellegata N, de la Chapelle A: Gene expression in papillary thyroid carcinoma reveals highly consistent
profiles. Proc Natl Acad SciUSA:15044-9, 2001.
Lemon WJ, Palatini JJ, Krahe R, Wright FA. Theoretical and experimental comparisons of gene expression
indexes for oligonucleotide arrays. Bioinformatics. 11:1470-1476, 2002.
Pierce JP, Faerber S, Wright FA, Rock CL, Newman V, Flatt SW, Kealey S, Jones V, Caan BJ, Gold EB, Haan
M, Hollenbach KA, Jones L, Marshall JR, Ritenbaugh C, Stefanick ML, Thomson C, Wasserman L,
Natarajan L, Gilpin E: A randomized trial of the effect of a plant-based dietary pattern on additional breast
cancer events and survival: The Women's Healthy Eating and Living (WHEL) Study. Control Clin Trials.
23:728-756, 2002
Yoon H, Liyanarachchi S, Wright FA, Davuluri R, Lockman JC, de la Chapelle A, Pellegata NS: Gene
expression profiling of isogenic cells with different TP53 gene dosage reveals numerous genes that are
affected by TP53 dosage and identifies CSPG2 as a direct target of p53. Proc Natl Acad Sci USA.
99:15632-15637,2002
Borrego S, Wright FA, Fernandez RM, Williams N, Lopez-Alonso M, Davuluri R, Antinolo G, Eng C: A
founding locus within the RET proto-oncogene may account for a large proportion of apparently sporadic
Hirschsprung disease and a subset of cases of sporadic medullary thyroid carcinoma. Am J Hum Genet.
72:88-100,2003.
Tanner SM, Aminoff M, Wright FA, Liyanarachchi S, Kuronen M, Saarinen A, Massika O, Mandel H, Broch H,
de la Chapelle A: Amnionless, essential for mouse gastrulation, is mutated in recessive hereditary
megaloblastic anemia. Nat Genet 33:426-429, 2003.
Cheng R, Ma JZ, Wright FA, Lin S, Gao X, Wang D, Elston RC, Li MD: Nonparametric disequilibrium mapping
of functional sites using haplotypes of multiple tightly linked single-nucleotide polymorphism (SNP)
markers. Genetics, 164:1175-1187, 2003.
Wright FA: Information perspectives of Haseman-Elston regression. Hum Hered, 55:132-142. 2003.
Miller BJ, Wang D, Krahe R, Wright FA: Pooled analysis of loss of heterozygosity in breast cancer: a genome
scan provides comparative evidence for multiple tumor suppressors and identifies novel candidate regions.
Am J Hum Genet, 73:748-767, 2003.
Bachinski LL, Udd B, Meola G, Sansone V, Bassez G, Eymard B, Thornton CA, Moxley RT, Harper PS,
Rogers MT, Jurkat-Rott K, Lehmann-Horn F, Wieser T, Gamez J, Navarro C, Bottani A, Kohler A, Shriver
MD, Sallinen R, Wessman M, Zhang S, Wright FA, Krahe R: Confirmation of the type 2 myotonic
dystrophy (CCTG)n expansion mutation in patients with proximal myotonic myopathy/proximal myotonic
dystrophy of different European origins: a single shared haplotype indicates an ancestral founder effect.
Am J Hum Genet, 73:835-48, 2003.
Wang D, Lauria M., Yuan B, Wright FA: Mega Weaver: A Simple Iterative Approach for BAG Consensus
Assembly. In Proc. Second Asia-Pacific Bioinformatics Conference (APBC2004), Dunedin, New Zealand.
CRPIT, 29. Chen, Y.-P. P., Ed. ACS, 2004.
Hu J, Yin G, Morris JS, Zhang L, Wright FA (2004). Entropy and survival-based weights to combine Affymetrix
array types in the analysis of differential _expression and survival. Methods of Microarray Data Analysis
IV, Critical Assessment of Microarray Data Analysis (CAMDA), eds. J.S. Shoemaker and S.M. Lin, 95-108,
2004.
Graham MR, Virtaneva K, Porcella SF, Barry WT, Gowen BB, Johnson CR, Wright FA, Musser JM: Group A
Streptococcus transcriptome dynamics during growth in human blood reveals bacterial adaptive and
survival strategies. Amer J Path, 166: 455-465, 2005.
Barry WT, Nobel AB, Wright FA: Significance analysis of functional categories in gene expression studies: a
structured permutation approach. Bioinformatics, 21:1943-1949, 2005.
Hu J, Zou F, Wright FA: Practical FDR-based sample size calculations in microarray experiments.
Bioinformatics, 21:3264-3272, 2005.
Drumm ML, Konstan MW, Schluchter MD, Handler A, Pace R, Zou F, Zariwala M, Fargo D, Xu A, Dunn JM,
Darrah RJ, Dorfman R, Sandford AJ, Corey M, Zielenski J, Dime P, Goddard K, Yankaskas JR, Wright
FA, Knowles MR; Gene Modifier Study Group. Genetic modifiers of lung disease in cystic fibrosis. N EnglJ
Med. 353:1443-1453, 2005.
Hu J, Wright FA, and Zou F: Estimation of Expression Indexes for Oligonucleotide Arrays Using the Singular
Value Decomposition. Journal of the American Statistical Association, 101:41 -50, 2006.
Graham MR, Virtaneva K, Porcella SF, Gardner DJ, Long RD, Welty DM, Barry WT, Johnson CA, Parkins LD,
Wright FA, Musser JM. Analysis of the transcriptome of group A Streptococcus in mouse soft tissue
infection. Am J Pathol. 169:927-42, 2006.
-------
Nadler JJ, Zou F, Huang H, Moy SS, Lauder JM, Crawley JN, Threadgill DW, Wright FA, Magnuson TR. Large
scale gene expression differences across brain regions and inbred strains correlates with a behavioral
phenotype. Genetics, 174:1229-1236, 2006.
Sterrett A, Wright FA: Inferring the Location of Tumor Suppressor Genes by Modeling Frequency of Allelic
Loss. Biometrics, 63:33:40, 2007.
Hu J, Wright FA: Assessing differential gene expression with small sample sizes in oligonucleotide arrays
using a mean-variance model. Biometrics, 63:41-9, 2007.
Wright FA, Huang H, Guan X, Gamiel K, Jeffries C, Barry WT, Pardo- Manuel F, Sullivan PF, Wilhelmsen KC,
Zou F: Simulating association studies: a data-based resampling method for candidate regions or whole
genome scans Bioinformatics, 23: 2581-2588, 2007.
Huang H, Zou F, Wright FA: Bayesian analysis of frequency of allelic loss data. Journal of the American
Statistical Association, 102(480): p. 1245-1253, 2007.
Barry WT, Nobel AT, Wright FA: A statistical framework for testing functional categories in microarray data.
Annals of Applied Statistics, 2(1): 286-315, 2008.
Lee S, Sullivan PF, Zou F, Wright FA: Comment on a Simple and Improved Correction for Population
Stratification. American Journal of Human Genetics, 82(2): 524-526, 2008
Ghosh A, Zou F, Wright FA: Estimating Odds Ratios in Genome Scans: An Approximate Conditional
Likelihood Approach. American Journal of Human Genetics, 82(5): p. 1064-74, 2008
Harrill JA, Li Z, Wright FA, Radio NM, Mundy WR, Tornero-Velez R, Crofton KM. Transcriptional response of
rat frontal cortex following acute In Vivo exposure to the pyrethroid insecticides permethrin and
deltamethrin. BMC Genomics, 9(1):546, 2008
Gatti DM, Shabalin AA, Lam TC, Wright FA, Rusyn I, Nobel AB. FastMap: Fast eQTL mapping in homozygous
populations. Bioinformatics, 25(4): 482-489, 2008.
Sullivan PF, Lin D, Tzeng JY, van den Oord E, Perkins D, Stroup TS, Wagner M, Lee S, Wright FA, Zou F, Liu
W, Downing AM, Lieberman J, Close SL. Genomewide association for schizophrenia in the CATIE study:
results of stage 1. Molecular Psychiatry, 13(6):570-584, 2008.
Zou F, Nie L, Wright FA, Sen PK: A robust QTL mapping procedure. Journal of Statistical Planning and
Inference, 139(3): 978-989, 2009
Gatti DM, Sypa M, Rusyn I, Wright FA, Barry WT. SAFEGUI: resampling-based tests of categorical
significance in gene expression data made easy. Bioinformatics, 25(4): 541-542, 2009
Li Z, Wright FA, Royland J. Age-dependent variability in gene expression in male Fischer 344 rat retina.
ToxicolSci. 107(1):281-92, 2009.
Zhu H, Ye L, Richard A, Golbraikh A, Wright FA, Rusyn I, Tropsha A. A novel two-step hierarchical
quantitative structure-activity relationship modeling work flow for predicting acute toxicity of chemicals in
rodents. Environ Health Perspect, 117(8): 1257-64,
Gatti DM, Harrill AH, Wright FA, Threadgill DW, Rusyn I. Replication and narrowing of gene expression
quantitative trait loci using inbred mice. Mamm Genome. 2009 Jul 17. [Epub ahead of print]
Blackman SM, Hsu S, Ritter SE, Naughton KM, Wright FA, Drumm ML, Knowles MR, Cutting GR. A
susceptibility gene for type 2 diabetes confers substantial risk for diabetes complicating cystic fibrosis.
Diabetologia, 52(9):1858-65, 2009 September.
Sun W, Wright FA, Tang Z, Nordgard SH, Loo PV, Yu T, Kristensen VN, Perou CM. Integrated study of copy
number states and genotype calls using high-density SNP arrays. Nucleic Acids Res., 2009 Jul 6. [Epub
ahead of print]
Byrnes A, Jacks A, Dahlman-Wright K, Evengard B, Wright FA, Pedersen NL, Sullivan PF. Gene expression
in peripheral blood leukocytes in monozygotic twins discordant for chronic fatigue: no evidence of a
biomarker. PLoS One, 5;4(6):e5805, 2009 June.
Taylor-Cousar JL, Zariwala MA, Burch LH, Pace RG, Drumm ML, Galloway H, Fan H, Weston BW, Wright FA,
Knowles MR; Gene Modifier Study Group. Histo-blood group gene polymorphisms as potential genetic
modifiers of infection and cystic fibrosis lung disease severity. PLoS One, 4(1):e4270, 2009.
-------
Scientific Leadership Roles
Name
Elaine Cohen Hubal
Activity Type
Co-Chair
Editorial Board
Member
World Health
Organization
Temporary Adviser
Program Planning
Committee
Co-Chair
Program Planning
Committee (and
Chair)
Member
Member
Chair
Organization
International Council of Chemical
Associations Long Range Research
Initiative (ICCA-LRI) workshop:
Connecting Innovations in
Biological, Exposure and Risk
Sciences: Better Information for
Better Decisions. Charleston, SC
Journal of Exposure Science and
Environmental Epidemiology
National Children's Study Data
Access Committee
Plan the IPCS international
workshop on "Identifying Important
Life Stages for Monitoring and
Assessing Risks from Exposures to
Environmental Contaminants."
US EPA/ICCA meeting on Public
Health Applications of Human
Biomonitoring. Chair plenary
session: International Perspectives.
Research Triangle Park, NC
International Society of Exposure
Science (formerly ISEA) 2009
Annual Meeting, Minneapolis, MN
The International Society of
Exposure Analysis Annual Meeting.
Chair symposium: Computational
Toxicology. Durham, NC
ILSI Health and Environmental
Sciences Institute, Sensitive
Subpopulations Working Group
ILSI Health and Environmental
Sciences Institute, Biomonitoring
Working Group
Exposure Science for Screening
Prioritizing and Toxicity Testing
Community of Practice (ExpoCoP)
Dates of Service
June 2009
January 2007-Present
2008-Present
2009-Present
September 24-25, 2007
2009
October 14-1 8, 2007.
2006-2009
2004-Presesnt
June 2008-Present
Previous
TOC
-------
Scientific Leadership Roles
Name
Elaine Cohen Hubal
(cont.)
Jimena Davis
David Dix
Activity Type
Member program
planning committee
Member
President
Member
Member
Organizing Committee
Editorial Board
Chair
Adjunct Assistant
Professor
Member
Organization
US EPA Workshop on Research
Needs for Community-Based Risk
Assessment. Session
organizer/chair: Data needs and
measurement methods for CBRA.
RTP, NC
US EPA Risk Assessment Forum
EPA-RTP Networking and
Leadership Training Organization
(NLTO)
Society of Industrial and Applied
Mathematics
American Mathematical Society
FDA Microarray Quality Control
Project
Toxicological Sciences
Multiple symposium sessions at
successive SOT annual meetings
Dept. Environmental and Molecular
Toxicology, North Carolina State
University
Society of Toxicology
Dates of Service
October 18-1 9, 2007
June 2004 -2009
2009-Present
2003-Present
2003-Present
2005-2008
2005-Present
2003-Present
2001-2008
2001 -Present
Previous I TOC
-------
Scientific Leadership Roles
Name
David Dix (cont.)
Keith Houck
Activity Type
Co-Chair
Organizing Committee
Member
Member
Adjunct Assoc.
Professor
Member
Editorial Board
Session Co-Chair
Member
Lecturer
Organization
First EPA ToxCast Data Analysis
Summit
International Council of Chemical
Associations Long Range Research
Initiative (ICCA-LRI) workshop:
Twenty-First Century Approaches to
Toxicity Testing, Biomonitoring, and
Risk Assessment
Amsterdam, The Netherlands
OECD Extended One Generation
Reproductive Toxicity Study
(EOGRTS) Working Group
OECD Molecular Screening Project
Working Group
Dept. Environmental Sciences and
Engineering, School of Public
Health, Univ. of North Carolina at
Chapel Hill
EU CarcinoGenomics Scientific
Advisory Board
Systems Biology in Reproductive
Medicine
2007 EPA Science Forum
NIH Roadmap RFA on Assay
Development for High Throughput
Molecular Screening Grant Review
Panel
North Carolina Central University,
The Brite Center, Dept. of
Pharmaceutical Sciences
Dates of Service
2009
2008
2008-Present
2006-Present
2008-Present
2007-Present
2007-Present
2007
2008
2007-Present
Previous
TOC
-------
Scientific Leadership Roles
Name
Keith Houck (cont.)
Richard Judson
Activity Type
Member
Member: Conference
Committee
Specialty Section:
Nanotoxicology
Co-chair
Review Panel
Review Panel
Member
Member
Lecturer
Organization
American Association for the
Advancement of Science
Society of Biomolecular Sciences
Society of Toxicology
Society of Biomolecular Sciences
Regional Meeting, RTP, NC
NIH Nanomaterial Grand
Opportunity Grant Review Panel
NIH ARRA Challenge Grant Review
Clinical Chemistry and Clinical
Toxicology Devices Panel of the
Medical Devices Advisory Cmte, Ctr
for Devices and Radiological
Health, FDA
EPA/ORD Genomics Task Force,
responsible for data management
strategy
Genomics Training Course
developed for OPPTS
Dates of Service
1992-Present
2001-Present
2007-Present
2010
2009
2009
2008-Present
2007-2008
2007
Previous
TOC
-------
Scientific Leadership Roles
Name
Richard Judson
(cont.)
Robert Kavlock
Activity Type
Session Co-Chair
Member
Member
Adjunct Assistant
Professor
Consultant
Co-chair
Chair
Reviewer
Organization
2007 EPA Science Forum
ORD IT Governance Board
Tox21 Workgroup
UNC Dept. of Environmental
Sciences and Engineering
FDA NCTR Scientific Advisory
Board
First EPA ToxCast Data Analysis
Summit
EPA International Science Forum
on Computational Toxicology
NIEHS SBRP Peer Review Panel
Dates of Service
2007
2007-Present
2008-Present
2008-Present
2009-Present
2009
2007
September 2007
Previous I TOC
-------
Scientific Leadership Roles
Name
Activity Type
Organization
Dates of Service
Member (and Chair)
OCED Molecular Screening
Initiative Working Group
2005-Present
Chair
S/T Technical Qualifications Review
Board
2008-Present
Chair
Managing Chemical Risks
Integrated Multidisciplinary
Research Working Group
2009-
Reviewer
European 7th Framework Proposals
for the Innovative Medicines
Initiative, Brussels
February 2009
Member
Robert Kavlock (cont.)
Society of Toxicology, including
Developmental and Reproductive
Toxicology Specialty Section and
the North Carolina Society of
Toxicology;
Current
Member
Teratology Society
Current
Expert Panel Member
Integrated Testing of Pesticides,
Canadian Council of Academies
2009-Present
Expert
WHO Working Group of the Health
of Effects of DDT, Geneva
June 2009
Co-Chair
Tox21 Working Group
2007-Present
Editorial Board
Journal of Toxicology and
Environmental Health, Part B
Current
Previous
TOC
-------
Scientific Leadership Roles
Name
Activity Type
Organization
Dates of Service
Editorial Board
Neurotoxicology and Teratology
2006-present
Associate Editor
Environmental Health Perspectives
2006-present
Robert Kavlock (cont.)
Editorial Board
Birth Defects Research, Part B
2003-present
Organizing Committee
World Congress on Alternatives to
Animals- Nos 6 (2007) and 7 (2009)
2007 and 2009
Organizing Committee
ILSI New Directions in
Developmental Toxicity
2009
Thomas Knudsen
Editorial Board -
Editor-in Chief
Reproductive Toxicology (Elsevier)
2003-Present
Editorial Board
Birth Defects Research (Part C)
2002 - Present
Editorial Board
Developmental Dynamics,
2002 - Present
Editorial Board •
Co-Editor
Co-Editor, Developmental
Toxicology (Comprehensive
Toxicology Series - Elsevier)
2002 - Present
President
Teratology Society
2007-08
Previous I TOC
-------
Scientific Leadership Roles
Name
Thomas Knudsen
(cont.)
Stephen Little
Activity Type
Chairman, Program
Committee
Scientific Liaison Task
Force
European
Commission
Steering Committee
Steering Committee
Co-Organizer
Member
Member
Member
Board of Directors
Councilor
Organization
47th Annual Meeting of the
Teratology Society; Council of the
Teratology Society
Society of Toxicology
Expert Panel (FP7)
First International Workshop on
Virtual Tissues (EPA)
ILSI-HESI DART Workshop on
"Developmental Toxicology New
Directions", Leader- Working
Group on New Technologies
Symposium on "Gene Regulatory
Networks in Developmental Biology
and Computational Toxicology",
Teratology Society
American Chemistry
Society of Toxicology
Genetics and Environmental
Mutagenesis Society
Genetics and Environmental
Mutagenesis Society
Dates of Service
1999-02 and 2005-09
2008-12
2009
April 21-23, 2009
2009
2009
1982-Present
2002-Present
1988-Present
2007-Present
Previous
TOC
-------
Scientific Leadership Roles
Name
Matthew Martin
James Rabinowitz
David Reif
Activity Type
Member
Member
Member
Chairman (SOT)
Member
Chair
Member
Program Committee
Grant Reviewer
Course Director and
Lecturer
Organization
OECD Extended One Generation
Reproductive Toxicity Study
(EOGRTS) Working Group
American Association for the
Advancement of Science
International Society for Quantum
Biology and Pharmacology
Bioinformatics and Computational
Toxicology 48th Annual Meeting of
the Society of Toxicology
American Chemical Society;
Section on Chemical Toxicology;
Section on Computers in Chemistry
NCCT Seminar Series, Research
Triangle Park, NC, USA
NHEERL Data Analysis Working
Group
"Bioinformatics and Computational
Biology", Genetic and Evolutionary
Computation Conference
National Science Foundation
Introduction to R,
North Carolina State University,
Raleigh, NC, USA [semester
course.
Dates of Service
2008-Present
Current
Current
2009
Current
2009
2007-Present
2008-2009
2007-Present
2008
Previous
TOC
-------
Scientific Leadership Roles
Name
Ann Richard
Activity Type
Consultant
Consultant
Editorial Board
OpenTox Consortium
Editorial Board
Editorial Board
Organizing Committee
NCCT Representative
EPA Lead
Organization
LeadScope LIST Workgroup, for
Implementation of ToxML standard
ontologies
ILSI Working group on Prediction of
Developmental Toxicity
SAR and QSAR in Environmental
Research
Advisory role
Mutation Research
Chemical Research in Toxicology
and SAR & QSAR in Environmental
Research
Computational Methods in
Toxicology and Pharmacology:
Integrating Internet Resources,
Moscow, Russia.
EPA Science Connector Workgroup
Tox21 EPA Chemical Working
Group
Dates of Service
March 2004-Present
2002-Present
2008-Present
2008-Present
1994-Present
2008-2010
September 2007
2007
2009
10
Previous I TOC
-------
Scientific Leadership Roles
Name
R. Woodrow Setzer
Activity Type
Adjunct Assoc.
Professor, Dept. of
Biostatistics
Adjunct Professor
Statistical Consultant/
Collaborator
Member: • (Past
Chair) Workgroup
drafting Technical
Guidance for
Benchmark Dose
Analysis
Member
Member
Member
Associate Editor
Publication Officer
Organization
UNC Dept. Biostatistics, School of
Public Health, Chapel Hill, NC
Department of Biostatistics, North
Carolina State University
National Center for Environmental
Assessment for Development of
EPA's Benchmark Dose Software
Risk Assessment Forum
ILSI-Europe Expert Group on the
Application of the Margin of
Exposure Approach to Genotoxic
Carcinogens in Food
Risk Assessment Forum's Point of
Departure Workgroup
NCEA Statistical Working Group
Journal of Statistical Software
Risk Assessment Specialty Section,
American Statistical Association
Dates of Service
2000-2009
2009-Present
1993- Present
2005-Present
2006-2008
2004-Present
2005-Present
2005-Present
2010-2012
11
Previous
TOC
-------
Scientific Leadership Roles
Name
Imran Shah
John Wambaugh
Activity Type
Member
Member
Co-Chair
Session Chair
Member
Member
Member
Member
Member
Organization
International Society for
Computational Biology (ISCB).
Society of Toxicology
First International Workshop on
Virtual Tissues (v-Tissues 2009),
RTP, NC
"Modeling Signaling as a
Determinant of System Behavior",
International Forum on
Computational Toxicology, EPA,
Research Triangle Park, North
Carolina.
EPA/ORD, Future of Toxicology
Working Group
Sigma Xi, Duke University Chapter
Society of Toxicology: Biological
Modeling and Risk Assessment
Specialty Sections, North Carolina
Society of Toxicology
American Physical Society: Division
of Fluid Dynamics, Statistical and
Nonlinear Physics Topical Group,
Forum on Graduate Student Affairs
American Association of Physics
Teachers
Dates of Service
1997-Present
2008-Present
2008-2009
2007
2007
2007-Present
2006-Present
2001 -Present
1995-Present
12
Previous
TOC
-------
NCCT Mentoring
Name
Adebowale Adenji
Andrew Beam
Michael Breen
Miyuki Breen
Kelly Chandler
Jimena Davis
Robert DeWoskin
Peter Egeghy
Fathi Elloumi
Ramon Garcia
Michael-Rock
Goldsmith
Mentor
E.Cohen Hubal
R Judson
R. Conolly
R. Conolly
T. Knudsen (NCCT) &
S. Hunter (NHEERL)
R. W. Setzer (NCCT)
&
Rogelio Tornero-Velez
(NERL)
T. Knudsen
E Cohen Hubal
R. Judson
R.W. Setzer
J. Rabinowitz
Program
Predoctoral
B.S. Computer
Science,
Computer
Engineering, &
Electrical
Engineering at
NCSU
Masters in
Biomathematics,
North Carolina
State University
Ph.D.,
Biostatisics,
University of
North Carolina,
Chapel Hill
Postdoctoral
Ph.D., Biomedical
Engineering, Case
Western Reserve
University, OH
Ph.D., Molecular
Physiology &
Biophysics, Vanderbilt
University, TN
Ph.D., Computational
Mathematics, North
Carolina State
University, NC
Ph.D., Software
Engineering,
University of Tunisia
Ph.D., Chemistry,
Duke University, NC
Other
ORD Regional
Scientist Program
NCCT Fellow
NCCT Fellow
Tenure
06/06 - 09/06
08/09 - current
12/05 -current
09/09 - current
08/08 - current
07/09 - current
06/09 - current
04/07-01/09
02/05 - 08/09
01/06-01/07
Current Position
Region 7
Research Physical
Scientist in EPA/NERL
Toxicologist in
EPA/NCEA
Research Environmental
Health Scientist in
EPA/NERL
UNC
Research Physical
Scientist in EPA/NERL
Previous
TOC
-------
NCCT Mentoring
Name
Amber Goetz
John Jack
Nicole Kleinstreuer
Holly Mortensen
Melissa Pasquinelli
Jason Pirone
David Reif
Chester Rodriguez
Daniel Rotroff
Nisha Schuler Snipes
Mentor
D. Dix
1. Shah
R. Conolly
RJudson(NCCT)&
SueEuling(NCEA)&
Mitchell Kostich
(NHEERL)
J. Rabinowitz
1. Shah
E.Cohen Hubal
H. Barton (NCCT) &
R.W. Setzer (NCCT)
D. Dix
T Knudsen
Program
Predoctoral
Ph.D.,
Environmental
and Molecular
Toxicology,
NCSU
B.S. Biological
Sciences, North
Carolina State
University, NC
Postdoctoral
Ph.D., Computational
Analysis and
Modeling, Louisiana
Tech University, LA
Ph.D., Centre for
Bioengineering,
University of
Canterbury, NZ
Ph.D., Human
Genetics, University of
Maryland, MD
Ph.D., Theoretical
Chemistry, Carnegie
Mellon University, PA
Ph.D.,
Biomathematics and
Toxicology, North
Carolina State
University, NC
Ph.D., Human
Genetics, Vanderbilt
University, TN
Ph.D., Pharmacology,
University of
California, LA, CA
Ph.D. Philosophy, Cell
and Cancer Biology,
University of
Cincinnati, OH
Other
Tenure
08/05 - 06/07
07/09 - current
08/09 - current
09/08 - current
09/04 - 07/06
01/08-05/09
09/06-11/08
01/06-09/09
08/09 - current
09/09 - current
Current Position
Syngenta Crop
Protection, Greensboro,
NC
NCSU/College of Textiles
Faculty
UNC-CH/Applied
Mathematics Program
Statistician in EPA/NCCT
Toxicologist in
EPA/OPP/HED (DC)
Previous
-------
NCCT Mentoring
Name
Rogelio Tornero-Velez
Beena Vallanat
John Wambaugh
Clarlynda Williams-
Devane
Michael Zager
Yuchao (Maggie) Zhao
Mentor
J. Blancato
1. Shah
H. Barton (NCCT)&
R.W. Setzer (NCCT)
A. Richard
H. Barton
R. Conolly
Program
Predoctoral
Ph.D.,
Bioinformatics,
North Carolina
State University
Postdoctoral
Ph.D., Environmental
Sciences, University
of North Carolina, NC
Ph.D., Physics, Duke
University, NC
Ph.D., Applied
Mathematics, North
Carolina State
University, NC
Ph.D., Environmental
Engineering, North
Carolina State
University, NC
Other
NCCT Fellow
Tenure
1/05-3/05
06/08-10/08
07/06-11/08
10/03-12/08
03/05-12/05
03/06 - 02/08
Current Position
Physical Scientist in
EPA/NERL
NHEERL
Physical Scientist in
EPA/NCCT
Post-Doc in
EPA/NHEERL
Pfizer, Inc. in San Diego,
CA
California EPA
Previous
-------
UNITED STATES ENVIRONMENTAL PROTECTION AGENCY
10 o . o o i i i • i . o . q
i 00110111.0 ;
>" 0 1 1 i • 0 00101
CQtfPUTATlbNAL
ij! ooo TOXICOLOGY11
COMPUTATIONAL TOXICOLOGY
ROTATIONAL FELLOWSHIP PROGRAM
Office of Research and Development
National Center for Computational Toxicology
Research Triangle Park, NC
April 28, 2008
Previous
TOC
-------
Computational Toxicology Rotational Fellowship Program
Participants to Date
Ms. Beena Vallanat from the National Health and Environmental
Effects Research Laboratory (NHEERL) was selected as our first
Computational Toxicology Rotational Fellow. This Fellowship Program
is intended to help translate the technologies and approaches being
developed within the NCCT to other parts of the Agency. Beena
began her four-month fellowship on September 15, 2008 and worked
primarily with NCCT's Dr. Imran Shah. Her goal was to gain a greater
understanding of computational approaches for integrating disparate
data streams for elucidating toxicologic process in risk assessment.
Specifically, she worked on analyzing published DEHP time course
gene expression data in mice to computationally infer transcriptional networks linked to
cell proliferation, a key event in non-genotoxic hepatocarcinogenesis. The concordance
of the regulatory network models will be evaluated using archived data sets, and
through experimental validation of predicted transcription factors and microRNA. This
project provided useful information about key genetic-regulatory events following DEHP
exposure that precede cell proliferation and provide a novel strategy to analyze
expression profiles for risk assessment. This was a great opportunity for interaction,
collaboration and facilitation of the work being conducted by NHEERL's Toxicogenomics
Core and the NCCT Virtual Liver project.
Drs. Egeghy and DeWoskin are currently on a scientist rotation for the next four to six
months in The Computational Toxicology Rotational Fellowship Program. Dr. Peter
Egeghy comes to NCCT from the National Exposure Research Laboratory and Robert
Dewoskin comes from the National Center for Environmental Assessment.
Dr. Egeghy began his rotation on June 01, 2009 and will be working
with Dr. Cohen Hubal on the ExpoCast™ Program. ExpoCast™ is
being initiated to ensure the required exposure science and
computational tools are ready to address global needs for rapid
characterization of exposure potential arising from the manufacture
and use of tens of thousands of chemicals. An important early
component of ExpoCast™ will be to consider how best to consolidate
and link human exposure data for chemical prioritization and toxicity
testing. Dr. Egeghy will help identify high priority human exposure data
resources for initial chemical indexing in collaboration with ACToR and
DSSTox, lead development of standards for exposure data representation, and direct
initial implementation of these standards for the most critical data. As a fellow with the
NCCT, Dr. Egeghy will foster the cross-ORD collaborations required to facilitate
progress on these activities.
Previous
TOC
-------
Dr. DeWoskin began his rotation on July 6, 2009 and will be working
on an integrated multidisciplinary project with Dr. Knudsen to develop
an agent-based model that simulates cellular changes leading to
pattern disruption in limb formation, as part of EPA's Virtual Embryo
project (http://www.epa.gov/ncct/v-Embryo/). This research fellowship
will provide research experience in newer technologies for predictive
modeling of developmental toxicity, as well as training in other
complimentary areas of NCCT research including the application of
systems-based in vitro assays and machine-learning algorithms to
derive cell signaling networks and build functional models for pathway-based risk
assessment. As a fellow with the NCCT, Dr. DeWoskin will foster the cross-ORD
collaborations and advance NCEA's understanding of the latest data and methods for
quantitative characterization of the Mode of Action.
Previous
TOC
-------
Introduction
The Office of Research and Development's (ORD) National Center for
Computational Toxicology (NCCT) coordinates and implements EPA's research in the
field of computational toxicology. Within the source-to-outcome framework, the NCCT
conducts and sponsors research to provide models for fate and transport of chemicals,
environmental exposures to humans and wildlife, delivery of the chemical to the target
site of toxicity, molecular and cellular pathways of toxicity, and ultimately systems level
understanding of biological processes and their perturbation. For priority setting
activities, the NCCT helps establish and distribute databases of high quality toxicological
information, utilizes high throughput screening tools for understanding the potential to
interfere with toxicity pathways across chemicals and chemical classes, develops
systems level models of underlying biology to predict toxicity at the organ level, and
formulates structure activity models on important toxicity pathways. To improve
quantitative risk assessment, it applies newly developed methods and tools to
understanding determinants of susceptibility, interspecies differences, dose
extrapolation, and risks of exposure to mixtures. NCCT employees also serve as
scientific reviewers and advisors in providing technical assistance in the broad area of
computational toxicology, to other Laboratories and Centers in ORD, EPA Program
Offices, Regions and the States. NCCT communicates the results of its efforts through
peer reviewed publications, consultations, presentations, databases, publicly available
computational models, training sessions, and web sites. Another facet of the NCCT is to
serve as a source of training in computational toxicology by offering seminars, mini-
courses, symposia, and staff details. To expand upon its training mission and further
facilitate a more in-depth understanding of computational toxicology, NCCT has
developed this Computational Toxicology Rotational Fellows Program (CTRFP).
This rotational fellowship program allows the temporary assignment of scientists
from other EPA organizations to NCCT. This will enhance the work of EPA and build
relationships and collaborations to better equip EPA in addressing the difficult challenges
of toxicology in the 21st Century. In addition, this program will enhance the personal
satisfaction and professional development of those EPA employees who are involved in
the program. Candidates participating in this program will be detailed to unclassified
Previous I TOC
-------
developmental assignments for up to 1 year in duration, and are expected to return to
their home organization with enhanced skills in computational toxicology.
Purpose
The CTRFP will provide EPA scientists the opportunity to expand their knowledge
and experience and enhance their professional growth while promoting cross-
organization experiences that broaden employee understanding of ORD's Computational
Toxicology Program. In addition, the program will assist in developing a motivated,
flexible, and agile workforce equipped to meet the complex environmental challenges
facing the Agency now and in the future.
The fellowships will be structured to meet the goals identified by program
participants. Every effort will be made to ensure that assignment activities enhance or
build each participant's portfolio of skills and competencies related to the area of
computational toxicology.
There will be 1 or 2 fellowship positions available at any given time, the final
number is dependent on the negotiated costs for each fellowship. The CTRFP provides
rotational opportunities for permanent EPA employees in grades GS-12 through GS-15
and ST. The program formally recruits and competitively selects candidates for
participation. To gain full benefit from the program, participants must fulfill the
fellowship at the NCCT location-RTP, NC.
Program Features
1. Participant Eligibility
A. All permanent EPA employees in grades GS-12 through GS-15 and ST may
apply to participate in CTRFP, provided they have been in their current positions for at
Previous I TOC
-------
least one year and have received favorable (i.e. fully successful, exceeds expectations or
outstanding) performance ratings.
B. Each selected fellow should have an Individual Development Plans (IDP) in
place that identifies CTRFP as a developmental activity, and the competencies or skills
he/she wishes to develop through program participation.
C. Employees must receive approval of their first- and second-level supervisors
to participate in the CTRFP.
2. Selection Process and Procedures
A. Participation in the CTRFP is open to all EPA organizations and will be
administered and managed by the NCCT Program Manager.
B. NCCT will use the attached template (Appendix A) for announcing and
selecting candidates for CTRFP rotational opportunities, and will ensure that all eligible
employees receive fair consideration for program selection.
C. Interested and eligible EPA employees will submit complete application
materials (see specifics in Appendix A) for program consideration.
D. The Deputy Director of NCCT will convene an evaluation panel comprised of
senior ORD employees/managers, including NCCT employees. The evaluation panel is
responsible for making recommendation(s) to the NCCT Director.
3. No Grade/Series Changes in Fellowship
A. CTRFP is a developmental program designed to build and or enhance
employee skills and competencies, not to provide promotional opportunities for
employees.
B. CTRFP candidates will be detailed to unclassified fellowships/projects vs.
specific positions; therefore, employee job series and grade levels will not change.
C. Employees will return to their positions of record in home organizations
following CTRFP rotations.
4. Duration of Fellowships
Fellowships under this program are intended to be 4-months to a maximum of 1
year in total duration, depending on the fellowship project plan. The project will be
implemented in 120-day increments. Details will be extended or terminated to affect
the agreed upon total duration.
Previous I TOC
-------
5. Documentation
A. CTRFP assignments will be affected and documented on SF-52s, "Requests for
Personnel Action." Upon notification of selection for the CTRFP, the employee's
home organization will complete the SF-52s, obtain the signatures of the
employee's first-and second-level supervisors and route to the employee's
servicing personnel office. A copy of all SF-52s will be sent to the CTRFP
Manager in NCCT.
B. CTRFP assignments will be officially documented as 120-day "details," and may
be extended for up to maximum of 1 year.
C. At the completion of CTRFP rotations, home offices will process "termination of
detail" actions via an SF-52.
D. An assignee's official position of record, including title, occupational series, and
grade level, will not change as a result of participation in CTRFP.
6. Performance Management
According to EPA's Performance Appraisal and Recognition System (PARS)
training manual, supervisors are to develop summary ratings for EPA employees on
detail assignments for 120 days or more, requiring the establishment of specific
performance criteria based on the essential duties and responsibilities of assignments.
The NCCT supervisor will:
A. Establish performance plans for CTRFP assignees with critical elements (CEs)
based on the essential duties and responsibilities associated with CTRFP
fellowships or projects.
B. Communicate performance expectations to CTRFP assignees within 30 days of
the effective date of assignments.
C. Complete performance evaluations with assignees at the conclusion of CTRFP
rotations.
D. Provide written evaluations with summary ratings to home supervisors (also
referred to supervisors of record) and assignees at the end of rotation period.
E. Home supervisors will consider CTRFP summary ratings in determining overall
ratings at the conclusion of rating period.
7. Program Funding
Previous I TOC
-------
A. FTE will continue to be covered by the fellow's home office; however a portion
(up to a maximum of 50%) of the PC&B costs may be paid by NCCT. This is a
negotiable item.
B. A portion of the travel and training expenses to and from fellowships will be
paid by each assignee's home office. NCCT is willing to pay a portion of the travel and
training expenses and will negotiate the amount with the selected fellow's home office.
C. All travel and training required by NCCT during the fellowship will be paid by
NCCT. Any travel and training required by the fellow's home organization will be paid by
the home organization.
Roles and Responsibilities
1. Home Office Supervisors
Home office supervisors provide important coaching, guidance, feedback, and
support to assignees. In addition to responsibilities set forth in this guidance, home
office supervisors should:
A. Write a letter of recommendation for the candidate to include in the application
package.
B. Assist candidates in building specific, measurable individual development plans
(IDPs) that set forth the expectations of both the participating office and the
candidate, as well as training/education to support skill advancement;
C. Discuss learning experiences upon assignment completion and identify lessons
learned; and
D. Include summary ratings for CTRFP assignees in determining overall performance
ratings.
2. NCCT Supervisor
The NCCT supervisor, like home supervisors, provides important instruction,
guidance, and feedback to CTRFP assignees. The success of a rotational experience for
both the candidate and the NCCT is to a great extent, a function of the understanding
each other's expectations. The NCCT supervisor will:
A. Assist in preparing rotational agreements that set forth expectations of both the
participating office and the candidate;
Previous I TOC
-------
B. Provide an in-depth orientation on the organization, its structure, and office
protocol;
C. Provide regular positive and constructive feedback on performance and task
completion;
D. Provide office space as well as the supplies, computers and other tools and
equipment needed to be successful during the rotation. Upon termination of the
rotation, all space, supplies, equipment, etc. provided by NCCT will be retained
by NCCT.
E. Ensure the establishment of PARS plans, monitor performance, and provide
written evaluations with summary ratings at the conclusion of the rotation.
3. CTRFP Assignees
A. Define personal development objectives within IDPs.
B. Meet with home office supervisors to discuss how fellowships will support IDP
objectives.
C. Prepare application materials including resumes/curriculum vitae and statements
of interest—which address: the knowledge, skills and abilities they will contribute
during the fellowship; the desired goals and accomplishments they seek from the
fellowship and expect to bring back to their home organization; and the possible
opportunities for future collaborations utilizing the field of computational
toxicology.
D. Present final seminar on experience prior to completion of detail
E. Write narrative summaries of rotational experiences at the conclusion of
rotational experience and provide to the information to both home and rotation
supervisors.
Rotation Agreements
1. Rotation agreements will be developed in conjunction with the NCCT
Supervisor, after the participants are selected and will include the following
information:
A. The time frame for the rotation;
B. Funding arrangements and anticipated trips home;
Previous I TOC
-------
C. Project needs for supplies, computers and other expenses;
D. A plan and timeline for what specific tasks are to be performed or skills
developed during the rotation;
E. Estimated time and topic for candidate's seminar upon completion of the rotation.
2. Agreements should be signed by each assignee, his/her home office
supervisor, and the NCCT supervisor, and a copy of the agreement should be
provided to the CTRFP Program Manager in NCCT.
Travel Information
For general Agency travel-related information, please see: "On the Way with
EPA: A Reference Guide for Travel," published by the EPA Office of the Comptroller, April
1999. This program meets the criteria of a rotational assignment; therefore, after
approval by the employee's Training Officer, training/expense funds (rather than travel)
may be used to pay for the per diem costs associated with this program.
1. Timeframe for Rotation Planning
Candidates must submit all application materials to the CTRFP Program Manager
by the announced deadline. Failure to make these arrangements in a timely manner
may delay the start of the employee's rotation.
2. Travel Authorizations
Prior to travel, each assignee is to prepare a Travel Authorization (TA) to cover a
4-month (120-day) rotation, plus additional TAs for each extension for up to a maximum
of 1-year, according to the final agreement.
3. Estimating Rotation Expenses
In order to calculate the costs of a fellowship outside of the assignee's
geographical location, the round trip airfare/train fare/POV expenses for the appropriate
number of round trips, per diem (at 55% of the location's daily allowance), lodging (at a
maximum of 55%), must be included. The round trip airfare/train fare/POV must be
paid from the travel ceiling and is not included in the example below. The 55% lodging
Previous I TOC
-------
and per diem is based on what the Agency allows for 120-day rotational details. An
example is provided below for the breakout of how the 55% lodging and per diem
(meals and incidental expenses) works for a candidate doing a rotation to RTP is (based
on 2008 per diem rates): Per diem allowance = $49/DAY; Maximum full lodging (based
on hotel rates) = $97/day.
ESTIMATED TRAINING/EXPENSE COSTS
(TRAVEL FUNDS NOT INCLUDED)
EXPENSE FUNDS
Candidate MI&E allowance
$49 X 0.55 X # days = $
+$36 X .055X2 days =
$39.60
Candidate lodging
maximum
$97X0.55X# days =$
TOTAL
4 MONTHS
$3,273.60
$6,402
$9,675.60
6 MONTHS:
4,837.42
$9,603
$14,440.42
1 YEAR
$9,674.84
$19,206
$28,880.84
(We hope to lower this figure substantially by using nice, furnished apartments with all
utilities and local phone included.)
4. Rotation Housing
All housing arrangements must be coordinated with and approved by NCCT. The
NCCT administrative staff will assist the candidate in locating the best available housing
within a reasonable distance of the NCCT-RTP facility.
A. Federal travel policy does not permit EPA to pay any lodging costs for
candidates who choose to stay with relatives or other federal/EPA employees during
their rotations. (We can still pay for incidentals and per diem, however.)
B. To be reimbursed for lodging, candidates must submit a rental certificate or
official receipt from a housing complex or landlord.
5. Travel Vouchers
Previous
TOC
-------
A. Travel vouchers are the mechanism used to claim reimbursement for travel
expenses. Each assignee will submit travel vouchers on a monthly basis, for the
duration of the rotation.
B. Requests for advance payments for rotational costs can be submitted on day
one of a rotational assignment. Advance payments can include air fare, lodging, per
diem, subsistence, and transportation for the first month of a rotation; however,
vouchers must be submitted monthly so that reimbursement of authorized travel
expenses can be paid.
C. Candidates should complete Direct Deposit Forms to ensure that travel
voucher reimbursements go directly into checking/savings accounts rather being sent to
home addresses.
D. The maximum travel reimbursement for assignees who choose to drive their
private vehicles to RTP, NC is limited to the amount of the least expensive roundtrip
government airfare from his/her home office to RTP, NC
Previous
10
TOC
-------
APPENDIX B
UNITED STATES ENVIRONMENTAL PROTECTION AGENCY
PREPARE ON YOUR HOME ORGANIZATION'S LETTERHEAD
MEMORANDUM OF UNDERSTANDING
DATE:
SUBJECT: Application for Computational Toxicology Rotational Fellowship Program
(CTRFP)
FROM:
THRU:
THRU:
TO:
(Applicant's Name, Organization)
(Applicant's First-line Supervisor's Name, Organization)
(Applicant's Second-line or Division Director's Name)
Karen Dean
CTRF Program Manager
NCCT (MD-B-205-01)
This memorandum and its attachments provide the information required for my
application to the CTRFP. Below are my signature and those of my first and second-level
supervisor which certify the accuracy of the information provided below and signify our
understanding of the terms and commitments being made as part of this application.
Applicant Information:
Organization and Mail
Code:
Position Title:
Grade/Series:
Time in Current Position:
Office Telephone Number:
Travel Preparer's Name &
Phone#
11
Previous
TOC
-------
FUNDING AND FTE
1. If the applicant named above is accepted to the CTRFP, we are aware of and commit to
cover a portion of their travel/training expenses to and from the rotational assignment, as
well as for those costs during the rotation. Our portion of the travel and expense costs
will be negotiated with NCCT, but the maximum amount to be covered by NCCT is 50%
of the total costs.
2. We understand that the FTE will continue to be charged to the candidate's home
organization. However, PC&B costs are negotiable and NCCT may cover up to 50% of
the total costs while on the rotation.
3. All travel and training required by NCCT during the rotational assignment will be paid by
NCCT.
4. In the rare event that the home organization requires the candidate to travel or be trained
during the rotational assignment, the home organization will pay for those expenses.
PROGRAM FEATURES
1. CTRFP is a developmental program designed to build and/or enhance employee skills
and competencies in computational toxicology, not to provide promotional
opportunities for employees.
2. CTRFP candidates will be detailed to unclassified rotational assignments/projects vs.
specific positions; therefore, employee job series and grade levels will not change.
3. Employees will return to their positions of record in home organizations following
CTRFP rotations.
4. Rotational assignments under this program will be a minimum of 120 days, up to a
maximum of 1 year, depending on the projects undertaken during the rotation.
5. For candidates requiring temporary travel, home visits may be funded, as agreed to in
advance by the assignee, NCCT supervisor and home supervisor.
6. For candidates requiring temporary housing, NCCT staff will assist in finding suitable
lodging within per diem while on the rotation.
PERFORMANCE MANAGEMENT
We certify that the applicant's latest performance rating was favorable (e.g.,fully
successful, exceeds expectations or outstanding).
According to EPA's Performance Appraisal and Recognition System (PARS) training
manual, the NCCT supervisor will develop summary ratings for EPA employees on detail
assignments for 120 days or more, requiring the establishment of specific performance
criteria based on the essential duties and responsibilities of assignments.
12
Previous I TOC
-------
APPLICANT & MANAGEMENT CERTIFICATION
We certify and agree to the terms and conditions stated in this MOU.
Applicant
Signature/Date
First Line Supervisor
Signature/Date
Second-Line Supervisor/Division Director Signature/Date
13
Previous
TOC
-------
APPENDIX C
Candidate's Statement of Interest
Computational Toxicology Rotational Fellowship Program
Describe how the CTRFP aligns with what you hope to achieve at EPA over the next few
years. Please limit to no more than 1 page. Please use Times Roman 12 point font.
Address the following elements in your description:
1. The knowledge, skills and abilities you will contribute during the fellowship.
2. The desired goals and accomplishments you seek from the fellowship and what you
expect to bring back to your home organization.
3. The possible opportunities to utilize the field of computational toxicology in your
work once you return to your home organization, including future collaborations.
14
Previous I TOC
-------
APPENDIX D
COMPUTATIONAL TOXICOLOGY ROTATIONAL FELLOWSHIP PROGRAM
DOCUMENTATION REQUIREMENTS
CHECKLIST
I. APPLICATION MATERIALS TO BE SUBMITTED BY CANDIDATE:
Required Items
Application MOU signed by candidate & managers (see Appendix B)
Candidate's Statement of Interest (see Appendix C)
Candidate's Biosketch, CV or Resume
Candidate's Supervisor's Recommendation
Completed
II. SELECTED CANDIDATE'S HOME ORGANIZATION PREPARES:
Required Items
SF-52s for duration of assignment, including termination SF-52
(provide copy to NCCT)
IDP which includes CTRFP
Travel Authorizations for CTRFP
(for joint funding, done in conjunction with NCCT)
Completed
III. SELECTED CANDIDATE AND NCCT SUPERVISOR PREPARE:
Required Items
Rotation Agreement
PARS plans
PARS summary ratings
at end of assignment
Completed
IV. UPON COMPLETION OF ROTATION, SELECTED CANDIDATE PREPARES
Required Items
Rotational Assessment
Exit Seminar
Completed
15
Previous
TOC
-------
UNITED STATES ENVIRONMENTAL PROTECTION AGENCY
ROTATIONAL FELLOWSHIP PROGRAM
What Is this program and how is it impiemented?
0 temporary assignment of scientists from other EPA organizations to ORD/NCCT
0 1 or 2 fellowships ongoing at the same time
0 duration is 4-months to maximum of 1 year, depending on the project
0 employee is detailed to unclassified PD and stays at current permanent grade
0 must be geographically located to RTP, NC during the rotation
0 return to home organization upon completion of rotation
What is the purpose of this program?
0 promote cross-organization experiences
0 broaden employee understanding of ORD's Computational Toxicology Program
0 build relationships and collaborations to better equip EPA in addressing the
difficult challenges of toxicology in the 21st Century
0 enhance the work of EPA
Who is Eligible?
0 permanent EPA employees in grades GS-12 thru GS-15 and ST (not open to EPA
post-docs and other term or temporary employees)
0 in current position for at least one year
0 have received favorable (i.e. fully successful, exceeds expectations or outstanding)
performance ratings
0 approval by first & second-level supervisors required
Who funds the candidate while on the rotation?
0 FTE will continue to be covered by the fellow's home organization
0 a portion (up to a maximum of 50%) of the PC&B costs may be negotiated with
and paid by NCCT
0 a portion of the travel and training expenses to and from fellowships may be
negotiated with NCCT and the costs may be shared by assignee's home office and
NCCT
How do interested employees apply?
0 receive approval of their first- and second-level supervisors
0 complete application materials including—statement of interest; supervisor's
recommendation, resume/biosketch/CV and application MOU
What is the timeline of events?
0 application due date is June 13, 2008
0 interviews completed by July 15, 2008
0 selection(s) made by August 1, 2008
0 rotation(s) begin by September 14, 2008
Previous
TOC
Next
-------
EPA Communities of Practice Presentation Documents
Chemical Prioritization
A Tiered Approach for the Use of Non-Testing Methods in the Regulatory Assessment of
Chemicals. Dr. Andrew Worth, Systems Toxicology Unit, Institute for Health & Consumer
Protection, Joint Research Centre, European Commission. 06/25/2009.
http://www.epa.gov/ncct/practice community/Worth CPCP%20presentation.pdf
The U.S. EPA's 2006 Inventory Update Reporting (IUR) Data on Chemical Substances. Dr.
Susan Sharkey, USEPA - OPPT. 05/28/2009.
http://www.epa.gov/ncct/practice communitv/lUR Overview CPCP 28mav2009.pdf
Is ChAMP a Winning Strategy? Dr. Cal Baier-Anderson, Environmental Defense Fund.
04/22/2009.
http://www.epa.gov/ncct/practice communitv/CPCP%20april2009%20CBA%20ChAMP%20RBP
%20Eval%20for%20ToxCast%2009.pdf
The ToxCast 320 Chemical Library in Cultures of Primary Human Hepatocytes: qNPAs as
Wndows into Chemical-Induced Hepatocyte Biology. Dr. Stephen S. Ferguson, CellzDirect.
11/18/2008.
http://www.epa.gov/ncct/practice community/CZD-
EPA%20ToxCast%20320%20CPCP%20presentation-11-20-08.pdf
The ToxCast Chemical Universe and ACToR (Aggregated Computational Toxicology
Resource). Dr. Richard Judson, USEPA-NCCT. 10/17/2008.
http://www.epa.gov/ncct/practice community/Judson%20CPCP%20Landscape%20Oct%20200
Srev.ppt
Screening for Chemical Effects on Neuronal Proliferation and Neurite Outgrowth Using High-
Content/High-Throughput Microscopy. Dr. Joseph M. Breier, Curriculum in Toxicology,
University of North Carolina at Chapel Hill. 09/25/2008.
http://www.epa.gov/ncct/practice communitv/CPCP-9-25-08%20Final.pdf
Chemical Modulation of Gap Junctional Intercellular Communication in Toxicology. Dr. James E.
Trosko, Center for Integrative Toxicology, Food Safety Toxicology Center, Dept.
Pediatrics/Human Development, College of Human Medicine, Michigan State University.
08/25/2008.
http://www.epa.gov/ncct/practice communitv/Trosko lecture 2008.pdf
Solidus Bioscience's MetaChip Technology for High-Throughput In Vitro Assessment of
Chemical and Drug Candidate Toxicity. Solidus. 07/27/2008.
http://www.epa.gov/ncct/practice communitv/Solidus Biosciences Toxcast July 08.pdf
U.S. EPA Use of QSAR and Category Approaches in Profiling Hazards of Industrial Chemicals.
Dr. Tala Henry, USEPA- OPPT. 06/11/2008.
http://www.epa.gov/ncct/practice_community/USEPA_Use%20of_QSAR_and_Category_Appro
aches_Jun08.pdf
Previous I TOC
-------
Overview of the Contaminant Candidate List 3. Dr. Thomas Carpenter, USEPA-OGWDW.
05/22/2008.
http://www.epa.gov/ncct/practice communitv/CCL3 Community of%20Practice 052208.pdf
Screening Chemicals in Commerce to Identify Possible Persistent and Bioaccumulative
Chemicals: New Results and Future Work. Presentation by Ted Smith from EPA's Great Lakes
National Program Office. Dr. Edwin Smith, USEPA - Region 5. 04/24/2008.
http://www.epa.gov/ncct/practice community/Smith CPCP presentation 24apr2008.pdf
Chemical Prioritization and Risk Assessment in the 21st Century -A Highly Personal
Perspective. Dr. Melvin Andersen, The Hamner Institutes for Health Sciences. 03/27/2008.
http://www.epa.gov/ncct/practice community/Andersen EPA CPCP 27mar2008.pdf
Opportunities for Collaboration on In Vitro Testing Proposal. HESI Representatives; Dr. Jiri
Aubrecht (Pfizer), Dr. Albert Fornace (Georgetown University), Dr. Robert Schiestl, (UCLA),
Syril Pettit, M.E.M. 02/28/2008.
http://www.epa.gov/ncct/practice community/vitro testing.pdf
BioSeek - ToxCast Phase I Project Update. Dr. Ellen Berg, BioSeek, Inc. 01/24/2008
http://www.epa.gov/ncct/practice communitv/BioSeek ToxCast Summary 24Jan08.pdf
Exposure Science
Holistic Mass Balance Modeling Approach for Chemical Screening and Priority Setting. Jon
Arnot, University of Toronto Scarborough. 07/14/2009.
http://www.epa.gov/ncct/practice community/exposure science/MassBalanceMethodsforChemi
calScreening-ExpoCoP.pdf
Connecting Environment, Biology, and Behavior for Human Exposure and Risk Assessment:
Integrative Modeling Approaches. Dr. Panos G. Georgopoulos, Rutgers University. 05/05/2009.
http://www.epa.gov/ncct/practice community/exposure science/050509 Panos.pdf
GExFRAME A Web-Based Framework for Accessing Global Consumer Exposure Data,
Scenarios, and Models. Dr. Muhilan Pandian, infoscientific. 04/14/2009.
http://www.epa.gov/ncct/practice community/exposure science/041409 Pandian.pdf
Gene Expression Profiles: Biomarkers of Inter-Individual Susceptibility to Environmental Agents
And Indicators of Exposure. Dr. Rebecca Fry, University of North Carolina at Chapel Hill.
03/10/2009.
http://www.epa.gov/ncct/practice community/exposure science/031009 Fry.pdf
Biomonitoring Equivalents as Screening Tools for Interpretation of Human Biomonitoring Data.
Sean M. Hays, M.S., M.S. and Lesa L. Aylward, M.S. 02/10/2009.
http://www.epa.gov/ncct/practice community/exposure science/021009 Hays.pdf
Multimedia MultipathwayModeling of Emissions to Impacts: screening with USEtoxand
advanced spatial modeling with IMPACT. Dr. Olivier Jolliet, iMod-lmpact and Risk Modeling
School of Public Health, EHS, University of Michigan. 01/13/2009.
http://www.epa.gov/ncct/practice community/exposure science/011309 Jolliet.pdf
Previous I TOC
-------
Prioritization of HPV Chemicals under the Chemical Assessment and Management Program
(ChAMP). Drs. Nhan Nguyen and Cathy Fehrenbacher, USEPA-OPPT. 12/08/2008.
http://www.epa.gov/ncct/practice community/exposure science/120808 Fehrenbacher.pdf
Assessing the Exposure-Dose-Toxicity Relationship within the EPA's ToxCast Program. Dr.
Russell Thomas, The Hamner Institutes for Health Sciences. 11/04/2008.
http://www.epa.gov/ncct/practice community/exposure science/110408 Thomas.pdf
Chemical Exposure Priority Setting Tool (CEPST). Dr. Mike Jayjock, The Lifeline Group, Inc.
10/07/2008.
http://www.epa.gov/ncct/practice community/exposure science/100708 Jayjock.pdf
Characterizing Exposure to Indoor VOCs and SVOCs using Simple Mass-Transfer Models. Dr.
John Little, Virginia Polytechnic Institute and State University. 10/07/2008.
http://www.epa.gov/ncct/practice community/exposure science/100708 Little.pdf
European Centre for Ecotoxicology and Toxicology of Chemicals Targeted Risk Assessment
(ECETOC TRA) Tool. Rosemary Zaleski and Chris Money on behalf of the ECETOC TRA Task
Force. 09/09/2008.
http://www.epa.gov/ncct/practice community/exposure science/090908 Zaleski.pdf
TOXICO-CHEMINFORMATICS: DSSTox and Chemical Structure Annotation for improved data
access. Dr. Ann Richard, USEPA-NCCT. 08/12/2008.
http://www.epa.gov/ncct/practice community/exposure science/081208 Richard.pdf
Considering Exposure in Priority Setting Categorization of the Domestic Substances List under
the Canadian Environmental Protection Act (CEPA). Dr. Bette Meek, McLaughlin Centre
University of Ottawa. 07/08/2008.
http://www.epa.gov/ncct/practice community/exposure science/070808 Meek.pdf
Short Term Exposure Prioritization Needs for ToxCast™. Drs. Elaine Cohen Hubal and Richard
Judson, USEPA-NCCT. 06/17/2008.
http://www.epa.gov/ncct/practice community/exposure science/ToxCast Exposure Info Need
s.pdf
Chemical Selection for ToxCast: EPA's Program for Predicting Toxicity and Prioritizing
Chemical Testing. Dr. Richard Judson, USEPA-NCCT. 05/27/2008.
http://www.epa.gov/ncct/practice community/exposure science/052708Judson.pdf
Previous I TOC
-------
NCCT PARTNERSHIP AGREEEMENTS
Material Transfer Agreements (MTAs):
For ToxCast Data Generation:
BASF SE and its affiliate Metanomics GmbH, Berlin, DE - 05/08
Biolog, Inc., Hayward, CA - 02/09
CellzDirect Inc, Durham, NC -02/08
Centronix Corp., Manchester, UK-01/08
Iconix & Affymetrix Inc., Santa Clara, CA - 06/06
Imperial College of Science, Technology and Medicine, London, UK-11/08
Invitrogen Corp., Madison, Wl - 02/08
Solidus Biosciences Inc., Troy, NY-01/08
WatchFrog S.A., Evry, FR - 10/08
Zygogen, LLC, Atlanta, GA - 07/08
For Clinical Data Sharing:
Pfizer Inc., New York, NY - 04/09
For ToxCast/ToxRef Data Sharing:
Cogenics Inc., Morrisville, NC - 09/07
Gene Logic Inc, Gaithersburg, MD - 10/07
Genedata Inc, Lexington, MA - 04/09
GeneGo Inc, St. Joseph, Ml - 06/08
Germany Federal Institute (BfR) for Risk Assessment, Berlin, DE - 02/08
National Institute for Public Health and the Environment (RIVM), Bilthoven, NL-
12/08
SimBioSys Inc., Toronto, ON - 12/07
U.S. EPA, National Center of Environmental Assessment (NCEA), Washington,
DC - 03/09
U.S. EPA, National Exposure Research Laboratory (NERL), RTP, NC - 02/09
U.S. EPA, National Exposure Research Laboratory, Athens, GA-04/09
U.S. EPA, National Health & Environmental Effects Research Laboratory
(NHEERL), RTP, NC - 04/08
U.S. EPA, Office of Pollution Prevention & Toxics, Risk Assessment Division
(OPPT), Washington, DC - 03/09
For ToxCast Data Analysis:
Advanced Chemistry Development, Toronto, ON - 02/09
Albert-Ludwigs-Universitat Freiburg, Freiburg, GE - 03/09
BioSeek Inc., South San Francisco, CA - 04/09
Bull & Associates, Inc., Springfield, VA - 03/09
Cambridge Cell Networks Ltd, Cambridge, UK-05/08
Cellumen Inc., Pittsburgh, PA-03/09
Department of Chemical & Biomolecular Engineering, The Ohio State University,
Columbus, OH - 02/09
Previous I TOC
-------
Department of Pharmacology, University of Medicine and Dentistry of New
Jersey-Robert Wood Johnson Medical School, Piscataway, NJ - 03/08
Douglas Connect GmbH, Project Coordinator of OpenTox, Zeiningen, CH - 02/09
Drexel University College of Medicine, Philadelphia, PA - 04/09
Exponent, Inc. Health Practice, Philadelphia, PA-04/09
FMC Corporation, Princeton, NJ - 04/09
Food Standards Agency, London, UK- 03/09
Fraunhofer Institute of Toxicology and Experimental Medicine (ITEM), Hannover,
GE - 03/09
Helmholtz Zentrum Munchen (GmbH), Neuherbert, GE - 02/09
Ideaconsult Ltd, Sofia, BG - 03/09
In Silico Toxicology, Basel, CH - 03/09
Institute of Biomedical Chemistry of Russian Academy of Medical Sciences,
Moscow, RU - 03/09
Istituto di Ricerche Farmacologiche "Mario Negri," Milano, IT - 06/09
Istituto Superiore DI Sanita, Rome, IT - 03/09
Jawaharlal NEHRU University, New Delhi, IN -03/09
Leadscope Inc., Columbus, OH - 04/09
Lhasa Limited, Leeds, UK-05/08
Louisiana Tech University, Ruston, LA - 03/09
Max Planck Institute for Molecular Genetics, Berlin, DE - 02/08
Michigan State University, Lansing, Ml - 06/08
National Center for Toxicological Research, Jefferson, AR - 02/09
National Institute of Advanced Industrial Science & Technology (AIST), Ibaraki,
JP-02/09
National Technical University of Athens, Athens, GR - 04/09
North Carolina State University, Raleigh, NC - 04/08
NovaScreen Biosciences Corp., Hanover, MD - 02/09
OpenTox Consortium, David Gallagher, Beaverton, OR - 03/09
Princeton University, Princeton, NJ -05/08
RegeneMed Inc., San Diego, CA-04/09
SABiosciences Corp., Frederick, MD - 02/09
Saint-Petersburg State Polytechnical University, Saint-Petersburg, RU - 02/09
SAS Institute Inc., Gary, NC - 03/09
Seascape Learning Co. Pvt. Ltd, Ne Delhi, IN - 04/09
Simulations Plus Inc., Lancaster, CA-03/09
Summit Toxicology, LLP, Lyons, CO - 02/09
Syngenta Crop Protection Inc., Greensboro, NC - 03/09
Technische Universitat Munchen Dept of Informatics, Garching, GE - 03/09
The Dow Chemical Co., Midland, Ml - 02/09
The Institute of Biomedical Sciences, East China Normal University, Shanghai,
CN - 02/09
Toxicogenomic Informatics & Solutions, LLC, Lansing, Ml - 03/08
U.S. Food & Drug Administration, Office of Food Additive Safety, Center for Food
Safety and Applied Nutrition, College Park, MD - 02/09
University of Insubria, Varese, IT -03/09
University of Kansas, Lawrence, KS - 03/09
University of North Carolina at Chapel Hill, Chapel Hill, NC - 8/08
University of North Carolina School of Global Public Health, -Chapel Hill, NC -
03/09
Previous I TOC
-------
Memoranda of Understanding (MOUs)
BRITE Institute Center of Excellence, North Carolina Central University, Durham, NC -
03/08
National Institute of Environmental Health Sciences/National Toxicology Program, RTP,
NC & National Institutes of Health Chemical Genomics Center, Bethesda, MD -
01/08
The Hamner Institutes for Health Sciences, RTP, NC - 10/07
U.S. Army Center for Environmental Health Research, Ft Detrick, MD - 04/08
University of Cincinnati, Reading, OH - 04/08
Cooperative Reserch and Development Agreements (CRADAs)
Illumina Inc, San Diego, CA - 12/07
L'OREAL, Paris, France - 09/08
Interagency Agreements (lAGs)
Department of Health & Human Services-NIEHS Div of Intramural Research, RTP, NC -
08/05 [Funds In]
Department of Health & Human Services-NIH Chemical Genomics Center, Bethesda,
MD-12/06 [Funds Out]
Previous I TOC
-------
NCCT Bibliography 2005 - 2009
2009
1. Ankley GT, Bencic DC, Breen MS, Collette TW, Conolly RB, Denslow ND, Edwards SW,
Ekman DR, Garcia-Reyero N, Jensen KM, Lazorchak JM, Martinovic D, Miller DH, Perkins
EJ, Orlando EF, Villeneuve DL, Wang RL, Watanabe KH. Endocrine Disrupting Chemicals
In Fish: Developing Exposure Indicators and Predictive Models of Effects Based On
Mechanism of Action. Aquatic Toxicology 92(3): 168-78 (2009).
2. Barrier M, Dix DJ, Mirkes PE. Inducible 70 kDa Heat Shock Proteins Protect Embryos from
Teratogen-induced Exencephaly: Analysis Using Hspa1a/a1b Knockout Mice. Birth Defects
Res A Clin Mol Teratol 28; 85(8):732-740. (2009).
3. Benakanakere MR, Li Q, Eskan MA, Singh AV, Zhao J, Galicia JC, Stathopoulou P,
Knudsen TB, Kinane DF. Modulation of TLR2 Protein Expression by Mir-105 in Human Oral
Keratinocytes. J Biol Chem. 284(34):23107-15 (2009).
4. Benfenati E, Benigni R, Demarini DM, Helma C, Kirkland D, Martin TM, Mazzatorta P,
Ouedraogo-Arras G, Richard AM, Schilter B, Schoonen WG, Snyder RD, and Yang C.
Predictive Models For Carcinogenicity And Mutagenicity: Frameworks,State-of-the-Art, and
Perspectives. Journal of Environmental Science and Health. Part C, Environmental
Carcinogenesis Reviews. 27(2):57-90, (2009).
5. Goetz AK and Dix DJ. Mode of Action for Reproductive and Hepatic Toxicity Inferred From A
Genomic Study of Triazole Antifungals. Toxicological Sciences. Society of Toxicology,
110(2):449-62, (2009).
6. Goetz AK and Dix DJ. Toxicogenomic Effects Common to Triazole Antifungals and
Conserved Between Rats and Humans. Toxicology and Applied Pharmacology, 238(1):80-9,
(2009).
7. Heidenfelder BL, Reif DM, Harkema JR, Cohen Hubal EA, Hudgens EE, Bramble LA,
Wagner JG, Morishita M, KeelerGJ, Edwards SW, Gallagher JE. Comparative Microarray
Analysis and Pulmonary Changes in Brown Norway Rats Exposed To Ovalbumin and
Concentrated Air Particulates. Toxicol Sci. 108(1), 207-221. (2009).
8. Judson R, Richard A, Dix DJ, Houck K, Martin M, Kavlock R, Dellarco V, Henry T,
Holderman T, Sayre P, Tan S, Carpenter T, Smith E. The Toxicity Data Landscape for
Environmental Chemicals (Journal). Environmental Health Perspectives, 117(5):685-95,
(2009).
9. Kavlock RJ, Austin CP, and Tice RR. Toxicity Testing in the 21st Century: Implications for
Human Health Risk Assessment. Risk Analysis, 29(4):485-7; discussion 492-7 (2009).
10. Knudsen TB, Martin MT, Kavlock RJ, Judson RS, Dix DJ, Singh AV. Profiling The Activity Of
Environmental Chemicals In Prenatal Developmental Toxicity Studies Using The U.S. EPA's
ToxRefDB. Reprod Toxicol. 28(2):209-19 (2009).
11. Kramer MG, Firestone M, Kavlock R, and Zenick H. The Future of Toxicity Testing For
Environmental Contaminants. Environ. Health Perspect, 117(7):A283-A284. (2009).
Previous I TOC
-------
12. Lou I, Wambaugh JF, Lau C, Hanson RG, Lindstrom AB, Strynar MJ, Zehr RD, Setzer RW,
and Barton HA. Modeling Single and Repeated Dose Pharmacokinetics of PFOA in Mice (J).
Toxicological Sciences, 107(2):331-41, (2009).
13. Martin MT, Judson RS, Reif DM, Kavlock RJ, Dix DJ. Profiling Chemicals Based On Chronic
Toxicity Results From The U.S. EPA ToxRef Database. Environ Health Perspect.
117(3):392-9. (2009).
14. Martin MT, Mendez E, Corum DG, Judson RS, Kavlock RJ, Rotroff DM, Dix DJ. Profiling
The Reproductive Toxicity of Chemicals From Multigeneration Studies In The Toxicity
Reference Database. Toxicol Sci. 110(1): 181-90. (2009).
15. Reif DM, Motsinger AA, Mckinney BA, Edwards KM, Chanock SJ, Rock MT, Crowe JE Jr,
Moore JH. Integrated Analysis Of Genetic And Proteomic Data Identifies Biomarkers
Associated With Systemic Adverse Events Following Smallpox Vaccination. Genes and
Immunity, 10(2). (2009).
16. Rodriguez, CE., Setzer RW, and Barton HA. Pharmacokinetic Modeling of Perfluorooctanoic
Acid During Gestation And Lactation In The Mouse. Reproductive Toxicology, (3-4):373-86,
(2009).
17. Sheldon, LS, and Cohen Hubal EA. Exposure as Part of a Systems Approach for Assessing
Risk. Environ Health Perspect 117(8): 119-1194 (2009).
18. Thompson, CM., Johns DO., Sonawane B., Barton HA., Hattis D., Tardif R., and Krishnan K.
Database for Physiologically Based Pharmacokinetic (PBPK) Modeling: Physiological
Parameters for Health and Health-Impaired Elderly. Journal of Toxicology and
Environmental Health - Part B: Critical Reviews, 12(1):1-12, (2009).
19. Williams-Devane CR. Wolf MA, and Richard AM. DSSTox Chemical-Index Files for
Exposure-Related Experiments in Arrayexpress and Gene Expression Omnibus: Enabling
Toxico-Chemogenomics Data Linkages. Bioinformatics, 25(5):692-694, (2009).
20. Williams-Devane CR. Wolf MA, and Richard AM. Toward A Public Toxicogenomics
Capability For Supporting Predictive Toxicology: Survey Of Current Resources And
Chemical Indexing Of Experiments In GEO And ArrayExpress. Toxicological Sciences,
109(2):358-371, (2009).
21. Xu Y, Cohen-Hubal EA, Clausen PA, and Little JC. Predicting Residential Exposure To
Phthalate Plasticizer Emitted From Vinyl Flooring - A Mechanistic Analysis. Environmental
Science & Technology, 43(7):2374-80, (2009).
22. Zhu H, Ye L, Richard AM, Golbraikh A, Wright FA, Rusyn I, and Tropsha A. A Novel Two-
Step Hierarchial Quantitative Structure Activity Relationship Modeling Workflow for
Predicting Acute Toxicity of Chemicals in Rodents. Environmental Health Perspectives,
117:1257-1264, (2009).
Previous I TOC
-------
In Press
Cohen Hubal EA, Richard AM, Shah I, Gallagher J, Kavlock R, Blancato J, Edwards SW.
Exposure Science and the U.S. EPA National Center for Computational Toxicology. J Expo Sci
Environ Epidemiol. 2008 Nov 5. [Epub ahead of print].
Cohen Hubal EA. Biologically-Relevant Exposure Science for 21st Century Toxicity Testing
Toxicol. Sci., 2009 July 14. [Epub ahead of print] In Press Doi: Doi:10.1093/Toxsci/Kfp159.
Ema M, Iseb R, Katoc H, Onedad S, Hirosea A, Hirata-Koizumia M, Nishidac Y, Singh Av,
Knudsen Tb And lhara T (2009) Fetal Malformations And Early Embryonic Gene Expression
Response In Cynomolgus Monkeys Maternally Exposed To Thalidomide Repro. Tox In press.
Goetz, AK, Rockett JC, Ren H, Thillainadarajah I, and Dix DJ. (2009) Inhibition of Rat and
Human Steroidogenesis By Triazole Antifungals. Systems Biology in Reproductive Medicine, In
press.
Houck KA, Dix DJ, Judson RS, Kavlock RJ, Yang J, Berg EL. Profiling Bioactivity of The
Toxcast Chemical Library Using Biomap Primary Human Cell Systems. J Biomolec Screen.
(2009) In press.
Knight AW, Little S, Houck k, Dix D, Judson R, Richard A, McCarroll N, Akerman G, Yang C,
Birrell L, Walmsley RM. Evaluation of High-Throughput Genotoxicity Assays Used in Profiling
The US EPA Toxcast Chemicals, Regulatory Pharmacology and Toxicology (2009) In press.
Rabinowitz JR; Little SB; Laws SC, Goldsmith MR. Molecular Modeling for Screening
Environmental Chemicals For Estrogenicity: Use of The Toxicant-Target Approach, Chemical
Research In Toxicology, 2009 Aug 31. [Epub ahead of print] In press.
Sanchez YA, Deener K, Hubal EC, Knowlton C, Reif D, Segal D. Research needs for
community-based risk assessment: findings from a multi-disciplinary workshop. J Expo Sci
Environ Epidem. 2009 Feb 25 [Epub Ahead Of Print] In press.
2008
1. Aylward, LL,Barton HA, and Hays SM. Biomonitoring Equivalents (Be) Dossier for Toluene
(Cas No. 108-88-3). Regulatory Toxicology and Pharmacology, 51(3 Suppl):S27-36, (2008).
2. Barthold JS, Mccahan SM, Singh AV, Knudsen TB, Si X, Campion L, and Akins RE. Altered
Expression of Muscle And Cytoskeleton-Related Genes In A Rat Strain With Inherited
Cryptorchidism. J Androl. 29(3):352-366. (2008).
3. Benigni R, Bossa C, Richard AM, and Yang C. A Novel Approach: Chemical Relational
Databases, and the Role of the Isscan Database on Assessing Chemical Carcinogenity.
Annals of The Institute Of Superiore Sanita 44(1):48-56, (2008).
4. Cohen-Hubal EA, Nishioka MG, Ivancic WA, Morara M, and Egeghy PP. Comparing Surface
Residue Transfer Efficiencies to Hands Using Polar and Non-Polar Florescent Tracers.
Environmental Science & Technology. American Chemical Society, Washington, DC,
42(3):934-9, (2008).
Previous I TOC
-------
5. Cohen Hubal EA, Moya J, Selevan SG. A Lifestage Approach to Assessing Children's
Exposure. Developmental and Reproductive Toxicology. Birth Defects Res (Part B)
83(6):522-529. (2008).
6. Datta S, Turner D, Singh R, Ruest LB, Pierce WM Jr And Knudsen TB. Fetal Alcohol
Syndrome (FAS) in C57BL/6 Mice Detected Through Proteomics Screening of the Amniotic
Fluid Birth Defects Res (Part A) 82(4): 177-186. (2008).
7. Deaciuc IV, Song Z, Peng X, Barve SS, Song M, He Q, Knudsen TB, Singh AV, and Mcclain
CJ. Genome-Wide Transcriptome Expression In The Liver of A Mouse Model of High
Carbohydrate Diet-Induced Liver Steatosis And Its Significance For The Disease. Hepatol
International, 2(1): 39-49 (2008).
8. Hardison NE, Fanelli TJ, Dudek SM, Reif DM, Richie MD, Motsinger AA. A Balanced
Accuracy Fitness Function Leads To Robust Analysis Using Grammatical Evolution Neural
Networks In The Case Of Class Imbalance. Genetic and Evolutionary Computation
Conference. (2008).
9. Harris, LA, and Barton HA. Comparing Single and Repeated Dosimetry Data for
Perfluorooctane Suflonate in Rats. Toxicology Letters, 181(3):148-156, (2008).
10. Houck KA and Kavlock RJ. Understanding Mechanisms of Toxicity: Insights from Drug
Discovery. Toxicol and Appl. Pharm. 227(2): 163-178. (2008).
11. Judson R, Richard A, Dix D, Houck K, Elloumi F, Martin M, Cathey T, Transue Tr, Spencer
R, Wolf M. Actor-Aggregated Computational Toxicology Resource. Toxicol Appl Pharmacol.
15;233(1):7-13. (2008).
12. Judson R. Pharmacogenetics in Drug Development and Research in Electrical Diseases of
the Heart: Genetics, Mechanisms, Treatment, Prevention Edited By Gussak, Antzelevitch,
Wilde, Friedman, Ackerman and Shen (Springer, 2008).
13. Judson R, Elloumi F, Setzer RW, Li Z, and Shah I. A Comparison of Machine Learning
Algorithms For Chemical Toxicity Classification Using A Simulated Multi-Scale Data Model.
BMC Bioinformatics. 19(9):241, (2008).
14. Kavlock RJ, Ankley G, Blancato J, Breen M, Conolly R, Dix D, Houck K, Hubal E, Judson R,
Rabinowitz J, Richard A, Setzer RW, Shah I, Villeneuve D, and Weber E. Computational
Toxicology: A State of the Science Mini Review. Toxicological Sciences 103(1), 14-27.
(2008).
15. Knaak, JB., Dary CC, Okino MS, Power FW, Zhang X, Thompson CB, Tornero-Velez R.,
and Blancato.JN. Parameters for Carbamate Pesticide QSAR and PBPK/PD Models for
Human Risk Assessment. Environmental Contamination and Toxicology. Springer-Verlag,
New York, NY, 193:53-210, (2008).
16. Knudsen TB and Kavlock RJ. Comparative Bioinformatics and Computational Toxicology. In:
Developmental Toxicology 3rd Edition. (B Abbott And D Hansen, Editors) New York: Taylor
And Francis.Chapter 12, PP 311-360 (2008).
Previous I TOC
-------
17. Loizou, G., Spendiff M, Barton HA, Bessems J, Bois FY, d'Yvoire MB, Buist H, Clewell HJ
3rd, Meek B, Gundert-Remy U, Goerlitz G, and Schmitt W. Development Of Good Modelling
Practice For Phsiologically Based Pharmacokinetic Models For Use In Risk Assessment:
The First Steps. Regulatory Toxicology and Pharmacology, 50(3):400-411, (2008).
18. Motsinger AA, Reif DM, Fanelli TJ, Ritchie MD. A Comparison of Analytical Methods for
Genetic Association Studies. Genetic Epidemiology, 32(6). (2008).
19. Nong A, Tan YM, Krolski ME, Wang J, Lunchick C, Conolly RB, and Clewell HJ 3rd.
Bayesian Calibration of A Physiologically Based Pharmacokinetic/Pharmacodynamic Model
of Carbaryl Cholinesterase Inhibition. J. Toxicol. Environ, Health 71, 1363-1381. (2008).
20. Rabinowitz JR, Goldsmith MR, Little SB, and Pasquinelli MA. Computational Molecular
Modeling For Evaluating The Toxicity Of Environmental Chemicals: Prioritizing Bioassay
Requirements. Environmental Health Perspectives, 116(5), 573-577. (2008).
21. Reif DM, Mckinney BA, Motsinger AA, Chanock SJ, Rock MT, Moore JH, Crowe JE Jr.
Genetic Basis for Systemic Adverse Events Following Smallpox Vaccination. Journal of
Infectious Diseases, 198(1). (2008).
22. Richard AM, Yang C, and Judson R. Toxicity Data Informatics: Supporting a New Paradigm
for Toxicity Prediction. Toxicology Mechanisms and Methods. 18(2 & 3):103-118, (2008).
23. Rodriguez CE, Sobol Z, Schiestl RH. 9,10-Phenanthraquinone Induces DMA Deletions and
Forward Mutations Via Oxidative Mechanisms In The Yeast Saccharomyces Cerevisiae,
Toxicology In Vitro 22(2):296-300 (2008).
24. Rogers JM and Kavlock RJ. Developmental Toxicity. In: Casarett & Doull's
Toxicology: The Basic Science of Poisons, 7th Ed. Cd Klaassen, Editor. Mcgraw-Hill,
Inc., New York, NY, 301-331. (2008).
25. Rouchka EC, Phatak AW, and Singh AV. Effect of Single Nucleotide Polymorphisms On
Affymetrix® Match-Mismatch Probe Pairs Bioinformation 2(9):405-11. (2008)
26. Thompson, CM., Sonawane B, Barton HA, Dewoskin RS, Lipscomb JC, Schlosser P, Chiu
W., and Krishnan K. Approaches for Applications of Physiologically Based Pharmacokinetic
Models in Risk Assessment. Journal of Toxicology and Environmental Health - Part B:
Critical Reviews, 11(7):519-47, (2008).
27. Verzilli C, Shah T, Casas JP, Chapman J, Sandhu M, Debenham SL, Boekholdt MS, Khaw
KT, Wareham NJ, Judson R, Benjamin EJ, Kathiresan S, Larson MG, Rong J, Sofat R,
Humphries SE, Smeeth L, Cavalleri G, Whittaker JC, Hingorani AD. Hingorani. "Bayesian
Meta Analysis of Genetic Association Studies with Different Sets of Markers", Am.
J.Hum.Gen. 82(4):859-872 (2008).
28. Wambaugh, JF, Barton HA, and Setzer RW. Comparing Models for Perfluorooctanoic Acid
Pharmacokinetics Using Bayesian Analysis. Journal of Pharmacokinetics and
Pharmacodynamics, 35(6):683-712, (2008).
Previous I TOC
-------
29. Yang C, Hasselgren CH, Boyer S, Arvidson K, Aveston S, Diekes P, Benigni R, Benz RD,
Contrera J, Kruhlak NL, Matthews EJ, Han X, Jaworska J, Kemper RA, Rathman JF, and
Richard AM. Understanding Genetic Toxicity through Data Mining: The Process of Building
Knowledge by Integrating Multiple Genetic Toxicity Databases Toxicology Mechanisms and
Methods. 18(2 & 3):277-295, (2008).
30. Yoon, M. and Barton HA. Predicting Maternal Rat And Pup Exposures: How Different Are
They? Toxicological Sciences. Society of Toxicology, 102(1):15-32, (2008).
31. Zhu H, Rusyn I, Richard AM,and Tropsha A. Use Of Cell Viability Assay Data Improves The
Prediction Accuracy of Conventional Quantitative Structure-Activity Relationship Models of
Animal Carcinogenicity. Environmental Health Perspectives, 116(4):506-513, (2008).
2007
1. Barton HA., Chiu WA, Setzer RW, Andersen ME, Bailer AJ, Bois FY, Dewoskin RS, Hays S,
Johanson G, Jones N, Loizou G, Macphail RC, Portier C, Spendiff M, and Tan YM.
Characterizing Uncertainty And Variability In PBPK Models: State of The Science And
Needs For Research And Implementation. Toxicological Sciences. Society of Toxicology,
99(2):395-402, (2007).
2. Benigni R, Netzeva Tl, Benfenati E, Bossa C, Franke R, Helma C, Hulzebos E, Marchant C,
Richard A, Woo YT, Yang C. The Expanding Role of Predictive Toxicology: An Update on
the (Q)SAR Models For Mutagens and Carcinogens. J. Environ. Sci. Health C, 25:53-97,
(2007).
3. Blancato, JN., Evans MV, Power FW, and Caldwell JC. Development and Use of PBPK
Modeling and the Impact of Metabolism on Variability in Dose Metrics for the Risk
Assessment of Methyl Tertiary Butyl Ether (MTBE). Journal of Environmental Science and
Health, 1:29-51, (2007).
4. Breen MS, Villeneuve DL, Breen M, Ankley GT, and Conolly RB. Mechanistic Computational
Model of Ovarian Steroidogenesis to Predict Biochemical Responses to Endocrine Active
Compounds. Annals Biomed. Engineering, 35(6), 970-981. (2007).
5. Cherney DP, Ekman DR, Dix DJ, Collette TW. Raman Spectroscopy-Based Metabolomics
For Differentiating Exposures To Triazole Fungicides Using Rat Urine. Anal Chem
79(19):7324-32. (2007).
6. Chiu, W., Barton HA., Dewoskin RS, Schlosser P, Thompson CM, Sonawane B, Lipscomb
JC, and Krishnan K. Evaluation of Physiologically Based Pharmacokinetic Models for Use in
Risk Assessment. Journal of Applied Toxicology, 27(3):218-237, (2007).
7. Conolly R, Blancato, JN. Development and Use of PBPK Modeling and the Impact of
Metabolism on Variability In Dose Metrics for The Risk Assessment Of Methyl Tertiary Butyl
Ether (MTBE), Journal of Environmental Protection Science, 1: 29-51, (2007).
8. Conolly RB, and Thomas RS. Biologically Motivated Approaches to Extrapolation From
High To Low Doses And The Advent Of Systems Biology: The Road To Toxicological Safety
Assessment. Human and Ecological Risk Assessment 13(1), 52-56.(2007).
Previous I TOC
-------
9. Cummings AM, Stoker TE, and Kavlock RJ. Gender-Based Differences in Endocrine and
Reproductive Toxicity. Environmental Research, 104(1):96-107, (2007).
10. Defur PI, Evans GW, Cohen Hubal EA, Kyle AD, Morello-Frosch RA, Williams D.
Vulnerability as a Function of Individual and Group Resources In Cumulative Risk
Assessment. Environ Health Perspect 115(5):817-824. (2007).
11.Dix DJ, Houck KA, Martin MT, Richard AM, Setzer RWand Kavlock RJ. The Toxcast
Program For Prioritizing Toxicity Testing of Environmental Chemicals. Toxicol. Sci., 95(1); 5-
12. (2007).
12. Firestone M, Moya J, Cohen Hubal E, Zartarian V, Xue J. Identifying Childhood Age Groups
for Exposure Assessments Monitoring. Risk Analysis 27(3): 701-714. (2007).
13. Goetz AK, Ren H, Schmid JE, Blystone CR, Thillainadarajah I, Best DS, Nichols HP, Strader
LF, Wolf DC, Narotsky MG, Rockett JC, Dix DJ. Disruption of Testosterone Homeostasis as
a Mode of Action for the Reproductive Toxicity of Triazole Fungicides in the Male Rat.
Toxicol Sci 95(1):227-39. (2007).
14. Green ML, Singh AV, Zhang Y, Nemeth KA, Sulik KK, and Knudsen TB. Reprogramming Of
Genetic Networks During Initiation of the Fetal Alcohol Syndrome. Dev Dyn. 236(2):613-31.
(2007).
15. Hilborn ED, Carmichael WW, Scares CM, Yuan M, Servaites JC, Barton HA, and Azevedo
SM. Serologic Evaluation of Human Microcystin Exposure. Environmental Toxicology,
(22)5:459-463, (2007).
16. Kavlock RJ, Dix DJ, Houck KA, Judson RS, Martin MT, Richard AM. Toxcast: Developing
Predictive Signatures for Chemical Toxicity. Alt. Animal Test Experiment. 14, Special Issue,
623-627. (2007).
17. Kim SJ, Dix DJ, Thompson KE, Murrell RN, Schmid JE, Gallagher JE, Rockett JC. Effects
of Storage, RNA Extraction, Genechip Type, and Donor Sex On Gene Expression Profiling
of Human Whole Blood. Clin Chem 53(6): 1038-45. (2007).
18. Liao KH, Tan YM, Conolly RB, Borghoff SJ, Gargas ML, Andersen ME, and Clewell HJ 3rd.
Bayesian Estimation of Pharmacokinetic and Pharmacodynamic Parameters in a Mode-of-
Action-Based Cancer Risk Assessment for Chloroform. Risk Anal. 27(6), 1535-1551. (2007).
19. Martin MT, Brennan RJ, Hu W, Ayanoglu E, Lau C, Ren H, Wood CR, Gorton JC, Kavlock
RJ, Dix DJ. Toxicogenomic Study of Triazole Fungicides And Perfluoroalkyl Acids In Rat
Livers Predicts Toxicity and Categorizes Chemicals Based On Mechanisms of Toxicity.
Toxicol Sci 97(2):595-613. (2007).
20. Mckinney BA, Reif DM White BC, Crowe JC, Moore JH. Evaporative Cooling Feature
Selection ForGenotypic Data Involving Interactions. Bioinformatics, 23(16). (2007).
21. Motsinger AA, Reif DM. Embracing Complexity: Gene-Gene and Gene-Environment
Interactions. In: Genes, Genomes, and Genomics, Vol. 3. (2007).
Previous I TOC
-------
22. Motsinger AA, Ritchie MD, Reif DM. Novel Methods for Detecting Epistasis in
Pharmacogenomics Studies. Pharmacogenomics, 8(9). (2007).
23. Motsinger AA, Reif DM, Fanelli TJ, Davis AC, Ritchie MD. Linkage Disequilibrium In Genetic
Association Studies Improves The Power Of Grammatical Evolution Neural Networks, leee
Symposium on Computational Intelligence In Bioinformatics and Computational Biology.
Linkage Disequilibrium in Genetic Association Studies Improves the Power of Grammatical
Evolution Neural Networks (2007).
24. Platts AE, Dix DJ, Chemes HE, Thompson KE, Goodrich R, Rockett JC, Rawe VY, Quintana
S, Diamond MP, Strader LF, Krawetz SA. Success And Failure In Human Spermatogenesis
As Revealed By Teratozoospermic RNAs. Hum Mol Genet 16(7):763-73. (2007).
25. Power, F., Blancato, JN. Malathion Exposure During Lice Treatment: Use of Exposure
Related Dose Estimating Model (ERDEM) and Factors Relating To The Evaluation Of Risk,
U.S. Environmental Protection Agency, Washington, DC, EPA/600/R-07/023 (NTIS PB2007-
106971), (2007).
26. Reif DM, Israel MA, Moore JH. Exploratory Visual Analysis of Statistical Results of
Microarray Experiments Comparing High and Low Grade Glioma. Cancer Informatics, 2(1).
(2007).
27. Rhomberg LR, Baetcke K, Blancato J, Bus J, Cohen S, Conolly R, Dixit, R, Doe J, Ekelman
K, Fenner-Crisp P, Harvey P, Hattis D, Jacobs A, Jacobson-Kram D, Lewandowski T,
Liteplo R, Pelkonen O, Rice J, Somers D, Turturro A, West W, and Olin S. Issues in the
Design and Interpretation of Chronic Toxicity and Carcinogenicity Studies in Rodents:
Approaches To Dose Selection. Crit. Rev. Toxicol. 37(9): 729-837. (2007).
28. Rodriguez CE, Mahle DA, Gearhart JM, Mattie DR, Lipscomb JC, Cook RS, and Barton HA.
Predicting Age-Appropriate Pharmacokinetics Of Six Volatile Organic Compounds In The
Rat Utilizing Physiologically Based Pharmacokinetic Modeling. Toxicological Sciences.
Society of Toxicology, 98(1):43-56, (2007).
29. Ryan PB, Burke TA, Cohen Hubal EA, Cura JJ, Mckone TE. Using Biomarkers to Inform
Cumulative Risk Assessment. Environ Health Perspect 115:833-84 (2007)
30. Singh AV, Knudsen KB and Knudsen TB. Integrative Analysis of the Mouse Embryonic
Transcriptome. Bioinformation, 1(10), 406-413. (2007).
31. Singh AV, Rouchka E, Rempala G, Bastian C and Knudsen TB. Integrative Database
Management for Mouse Development: Systems and Concepts Review. Birth Defects
Research (Part C) 81:1-19. (2007).
32. Vahter M, Gochfeld M, Casati B, Thiruchelvam M, Falk-Filippson A, Kavlock R, Marafante E,
and Cory-Slechta D. Implications of Gender Differences for Human Health Risk Assessment
and Toxicology. Environmental Research, 104(1):70-84, (2007).
33. Wambaugh JF, Matthews JV, Gremaud PA, and Behringer RP. "Response to Perturbations
in Granular Flow. Physical Review E 76, 051303 (2007).
Previous I TOC
-------
34. Yoon M, Madden MC, and Barton HA. Extrahepatic Metabolism in Cyp2e1 in PBPK
Modeling of Lipophilic Volatile Organic Chemicals: Impacts on Metabolic Parameter
Estimation and Prediction of Dose Metrics. Journal of Toxicology and Environmental Health,
70(18):1527-1541, (2007).
2006
1. Barone S Jr, Re Brown, S Euling, E Cohen Hubal, Ca Kimmel, S Makris, J Moya, Sg
Selevan, B Sonawane, T Thomas, C Thompson. Vision General De Al Evaluacion Del
Riesgo En Salud Infantil Empleando Un Enfouque Por Etapas De Desarrollo [Overview Of A
Life Stage Approach To Children's Health Risk Assessment] Acta Toxicologica Argentina,
14(Suplemento) 7-10. (2006).
2. Barton HA, Tang J, Sey Ym, Stanko JP, Murrell RN, Rockett JC, Dix DJ Metabolism Of
Myclobutanil And Triadimefon By Human And Rat Cytochrome P450 Enzymes And Liver
Microsomes. Xenobiotica 36(9):793-806. (2006).
3. Barton HA, Tang J, Sey YM, Stanko JP, Murrell RN, Rockett JC, Dix DJ. Metabolism of
Myclobutanil and Triadimefon by Human and Rat Cytochrome P450 Enzymes and Liver
Microsomes. Xenobiotica, 36(09):793-806, (2006).
4. Barton HA, PastoorTP, Baetcke K, Chambers JE, Diliberto J, Doerrer NG, Driver JH,
Hastings CE, lyengar S, Krieger R, Stahl B, Timchalk C. The Acquisition and Application of
Absorption, Distribution, Metabolism, and Excretion (ADME) Data in Agricultural Chemical
Safety Assessments. Critical Reviews in Toxicology, 36(1):9-35, (2006).
5. Birnbaum LS and Cohen-Hubal EA. Polybrominated Diphenyl Ethers: A Case Study for
Using Biomonitoring Data to Address Risk Assessment Questions. Environmental Health
Perspectives, 114(11):1770-1775, (2006).
6. Blancato, JN. Exposure Related Dose Estimating Model (ERDEM) A Physiologically-Based
Pharmacokinetic And Pharmacodynamic (PBPK/PD) Model for Assessing Human Exposure
and Risk, EPA Report- U.S. Environmental Protection Agency, Washington, DC,
EPA/600/R-06/061 (NTIS PB2006-114712), (2006).
7. Blancato, JN. Computational Environmental Toxicology, Mcgraw-Hill Yearbook of Science
and Technology, Chapter 1, PP 72-75 (2006).
8. Carmichael NG, Barton HA, Boobis AR, Cooper RL, Dellarco VL, Doerrer NG, Fenner-Crisp
PA, Doe JE, Lamb JC 4th, and PastoorTP. Agricultural Chemical Safety Assessment: A
Multisector Approach to the Modernization of Human Safety Requirements. Critical Reviews
in Toxicology, 36(1):1-7, (2006).
9. Cohen Hubal EA, Egeghy PP, Leovic KW, Akland GG. Measuring Potential Dermal Transfer
of a Pesticide to Children in a Child Care Center. Environ Health Perspect 114(2)264-269.
(2006).
10. Cohen Hubal EA. Uso De Los Datos De Biomonitoreo Para Informar Sobre La Evaluacion
De La Exposition Infantil [Using Biomonitoring Data To Inform Exposure Assessment In
Children] Acta Toxicologica Argentina [Journal Of The Argentinan Society Of Toxicology].
14(Suplemento) 17-19. (2006).
Previous I TOC
-------
11. Denslow ND, Colbourne JK, Dix DJ, Freedman JH, Helbing CC, Kennedy S, Williams PI
Selection of Surrogate Animal Species For Comparative Toxicogenomics. In: Emerging
Molecular and Computational Approaches for Cross-Species Extrapolations. Eds. W
Benson and R Di Giulio. CRC Press, Florida. (2006).
12. Dix DJ, Gallagher K, Benson WH, Groskinsky BL, Mcclintock JT, Dearfield KL, Farland WH.
A Framework for the Use of Genomics Data at the EPA. Nat Biotechnol 24(9): 1108-11.
(2006).
13. Goetz AK, Bao W, Ren H, Schmid JE, Tully DB, Wood C, Rockett JC, Narotsky MG, Sun G,
Lambert GR, Thai SF, Wolf DC, Nesnow S, Dix DJ Gene Expression Profiling In The Liver of
Cd-1 Mice To Characterize The Hepatotoxicity of Triazole Fungicides. Toxicol Appl
Pharmacol 215(3):274-84. (2006).
14. Hunter SE, Rogers E, Blanton M, Richard AM, Chernoff N. Bromochloro-Haloacetic Acids:
Effects on Mouse Embryos In Vitro And QSAR Considerations, Birth Defects Research Part
A, 21(3):260-266, (2006).
15.Kavlock RJ, Barr D, Boekelheide K, Breslin W, Breysse P, Chapin R, Gaido K, Hodgson E,
Marcus M, Shea K, and Williams P. NTP Center For The Evaluation of Risks To Human
Reproduction: Expert Panel Update on the Reproductive and Developmental Toxicity of
Di(2-Ethylhexyl) Phthalate . Reproductive Toxicology, 22(3):291-399, (2006).
16. Kim SJ, Dix DJ, Thompson KE, Murrell RN, Schmid JE, Gallagher JE, Rockett JC. Gene
Expression In Head Hair Follicles Plucked From Men And Women. Ann Clin Lab Sci
36(2): 115-26. (2006).
17. Kim YK, Suarez J, Hu Y, McDonough PM, Boer C, Dix DJ, Dillmann WH. Deletion of the
Inducible 70-Kda Heat Shock Protein Genes In Mice Impairs Cardiac Contractile Function
and Calcium Handling Associated With Hypertrophy. Circulation 113(22):2589-97. (2006).
18. Luderer U, Collins TF, Daston GP, Fischer LJ, Gray RH, Mirer FE, Olshan AF, Setzer RW,
Treinen KA, Vermeulen R. NTP-CERHR Expert Panel Report on the Reproductive and
Developmental Toxicity of Styrene, Birth Defects Research (Part B) 77(2):110-193 (2006).
19. Potter LK, Zager MG, and Barton HA. A Mathematical Model for the Androgenic Regulation
of the Prostate in Intact and Castrate Adult Male Rats. American Journal of Physiology.
American Physiological Society, 291(5):E952-E964, (2006).
20. Richard AM, Gold LS, Nicklaus MC. Chemical Structure Indexing Of Toxicity Data on the
Internet: Moving Towards a Flat World. Current Opinion in Drug Discovery & Develop,
9(3):314-325, (2006).
21. Richard AM The Future Of Toxicology-Predictive Toxicology: An Expanded View Of
Chemical Toxicity. 10 Chemical Research in Toxicology. American Chemical Society,
Washington, DC, 9(September): 1257-1262, (2006).
22. Rockett JC, Narotsky MG, Thompson KE, Thillainadarajah I, Blystone CR, Goetz AK, Ren
H, Best DS, Murrell RN, Nichols HP, Schmid JE, Wolf DC, and Dix DJ. Effect of Conazole
Fungicides on Reproductive Development in the Female Rat. Reproductive Toxicology,
22(4):647-658, (2006).
10
Previous I TOC
-------
23. Shi L et al, MAQC Consortium. The Microarray Quality Control (MAQC) Project Shows Inter-
and Intraplatform Reproducibility of Gene Expression Measurements. Nat Biotechnol
24(9): 1151-61. (2006).
24. Sun G, Thai SF, Lambert GR, Wolf DC, Tully DB, Goetz AK, George MH, Grindstaff RD, Dix
DJ, Nesnow S. Fluconazole-lnduced Hepatic Cytochrome P450 Gene Expression And
Enzymatic Activities In Rats And Mice. Toxicology Letters, 164(1):44-53, (2006).
25. Tan YM, Liao KH, Conolly RB, Blount BC, Mason AM, and Clewell HJ. Use of A
Physiologically Based Pharmacokinetic Model to Identify Exposures Consistent With Human
26. Biomonitoring Data for Chloroform. J. Toxicol. Environ. Health, Part A, 69(18), 1727-1756.
(2006).
27. Tolson JK, Dix DJ, Voellmy RW, Roberts SM. Increased Hepatotoxicity of Acetaminophen in
Hsp70i Knockout Mice, Toxicol Appl Pharmacol. 2006 N 1; 210(1-2):157-62 (2006).
28. Tully DB, Bao W, Goetz AK, Blystone CR, Ren H, Schmid JE, Strader LF, Wood CR, Best
DS, Narotsky MG, Wolf DC, Rockett JC, Dix DJ. Gene Expression Profiling In Liver and
Testis of Rats to Characterize the Toxicity of Triazole Fungicides. Sciencedirect Elsevier
(Ed.), Toxicology and Applied Pharmacology, 215(3):260-273, (2006).
29. Wambaugh, JF, Graph Percolation As An Analog To Granular Force Networks. Cond-
Mat/0603314, (2006).
30. Yang C, Richard AM, Cross KP. The Art of Data Mining the Minefields of Toxicity Databases
to Link Chemistry to Biology. Curr Comput-Aided Drug Design, 2(2):135-150, (2006).
31. Yoon, M., Madden MC, and Barton HA. Developmental Expression of Aldehyde
Dehydrogenase in Rat: A Comparison of Liver and Lung Development. Toxicological
Sciences. Society of Toxicology, 89(2):386-398, (2006).
32. Zhang Q, Andersen ME, and Conolly RB. Binary Gene Induction and Protein Expression in
Individual Cells. Theor. Biol. Med. Modelling 5; 3:18, (2006)
2005
1. Andersen ME, Dennison JE, Thomas RE, and Conolly RB. New Directions In Incidence-
Dose Modeling. Trends Biotechnol. 23(3):122-127. (2005).
2. Bao W, Schmid JE, Goetz AK, Ren H, Dix DJ A Database For Tracking Toxicogenomic
Samples And Procedures. Reproductive Toxicology 19(3):411-419. (2005).
3. Barton HA, Cogliano VJ, Flowers L, Valcovic L, Setzer RW, Woodruff TJ. Assessing
Susceptibility from Early-Life Exposure to Carcinogens. Environmental Health Perspectives,
113(9): 1125-1133, (2005).
4. Barton HA, Computational Pharmacokinetics During Developmental Windows of
Susceptibility. Journal of Toxicology and Environmental Health - Part A: Current Issues,
68(11-12):889-900, (2005).
11
Previous I TOC
-------
5. Cohen Hubal EA, Suggs JC, Nishioka MG, Ivancic WA, Characterizing Residue Transfer
Efficiencies Using a Fluorescent Imaging Technique, Journal of Exposure Analysis and
Environmental Epidemiology. 15(3):261-270. (2005).
6. Conolly RB, Gaylor DW, and Lutz WK. Population Variability in Biological Adaptive
Responses to DMA Damage and the Shapes of Carcinogen Dose-Response Curves.
Toxicol. Appl. Pharmacol. 207(2 suppl):570-75. (2005).
7. Cummings A, and Kavlock R A Systems Biology Approach to Developmental Toxicology.
Repro. Toxicol, 19(3):281-290. (2005).
8. Firestone M, Cohen Hubal EA, E.A.Guidance On Selecting Age Groups For Monitoring And
Assessing Childhood Exposures To Environmental Contaminants, EPA Report - Risk
Assessment Forum, U.S. Environmental Protection Agency, Washington, DC. EPA/630/P-
03/003F. (2005).
9. Fostel J, Choi D, Zwickl C, Morrison N, Rashid A, Hasan A, Bao W, Richard A, Tong W,
Bushel PR, Brown R, Bruno M, Cunningham ML, Dix D, Eastin W, Frade C, Garcia A,
Heinloth A, Irwin R, Madenspacher J, Merrick BA, Papoian T, Paules R, Rocca-Serra P,
Sansone AS, Stevens J, Tomer K, Yang C, Waters M. Chemical Effects In Biological
Systems-Data Dictionary (CEBS-DD): A Compendium of Terms For The Capture and
Integration of Biological Study Design Description, Conventional Phenotypes, and 'Omics
Data. Tox. Sci., 88(2):585-601. (2005).
10. Granville CA, Ross MK, Tornero-Velez R, Hanley NM, Grindstaff RD, Gold A, Richard AM,
Funasaka K, Tennant AH, Kligerman AD, Evans MV, DeMarini D. Genotoxicity And
Metabolism of the Source-Water Contaminant 1,1-Dichloropropene: Activation By Gstt1-1.
Mutat. Res. 572(1-2):98-112, (2005).
11.Kavlock R, Ankley GT, Collette T, Francis E, Hammerstrom K, Fowle J, Tilson H, Toth G,
Schmieder K, Veith GD, Weber E, Wolf E, Wolf DC, and Young D. Computational
Toxicology: Framework, Partnerships, And Program Development. Reproductive Toxicology.
19(3):265-280, (2005).
12. Kavlock R, and Cummings A. Mode of Action: Inhibition of Androgen Receptor Function-
Vinclozolin-lnduced Malformations in Reproductive Development. Critical Reviews in
Toxicology, 35(8-9):721-726, (2005).
13. Kavlock, R. J. And A. M. Cummings. Mode of Action: Reduction of Testosterone Availability-
Molinate-lnduced Inhibition Of Spermatogenesis. Critical Reviews in Toxicology. Crc Press
Lie, Boca Raton, Fl, 35(8-9):685-690, (2005).
14. Lutz WK, Gaylor DW, Conolly RB, and Lutz RW. Nonlinearity and Thresholds in Dose-
Response Relationships for Carcinogenicity Due To Sampling Variation, Logarithmic Dose
Scaling, or Small Differences in Individual Susceptibility. Toxicol. Appl. Pharmacol. 207(2
suppl):565-69. (2005).
15. Ostermeier GC, Goodrich RJ, Diamond MP, Dix DJ, Krawetz SA. Toward Using Stable
Spermatozoal Rnas For Prognostic Assessment Of Male Factor Fertility. Fertility and
Sterility, 83(6): 1687-94. (2005).
12
Previous I TOC
-------
16.Seed, J, Carney EW, Corley RA, Crofton KM, Desesso JM, Foster PM, Kavlock RJ, Kimmel
G, Klaunig J, Meek ME, Preston RJ, Slikker W Jr, Tabacova S, Williams GM, Wiltse J,
Zoeller RT, Fenner-Crisp P, and Patton D. Overview: Using Mode of Action and Life Stage
Information To Evaluate the Human Relevance of Animal Toxicity Data. Critical Reviews in
Toxicology, 35(8-9):663-672, (2005).
17. Teeguarden JG, Waechter JM Jr., Clewell HJ 3rd, Covington TR, and Barton HA. Evaluation
of Oral and Intravenous Route Pharmacokinetics, Plasma Protein Binding and Uterine
Tissue Dose Metrics of BPA: A Physiologically Based Pharmacokinetic Approach.
Toxicological Sciences, 85(2):823-838, (2005).
18. Teeguarden JG, .Deisinger PJ, PoetTS, English JC, FaberWD, Barton HA, Corley RA, and
Clewell HJ 3rd Derivation Of A Human Equivalent Concentration For N-Butanol Using A
Physiologically Based Pharmacokinetic Model For N-Butyl Acetate and Metabolites N-
Butanol And N-Butyric Acid. Toxicological Sciences. 85(1):429-446, (2005).
19. Tully DB, Luft JC, Rockett JC, Ren H, Schmid JE, Wood CR, Dix DJ Reproductive And
Genomic Effects In Testes From Mice Exposed To The Water Disinfectant Byproduct
Bromochloroacetic Acid. Reproductive Toxicology 19(3):353-366. (2005).
13
Previous I TOC
-------
BMC Bioinformatics ^
D , .. , r:-,,,
Research article jv^yjj v,y*
A comparison of machine learning algorithms for chemical toxicity
classification using a simulated multi-scale data model
Richard Judson*1, Fathi Elloumi1, R Woodrow Setzer1, Zhen Li2 and
Imran Shah1
Address: National Center for Computational Toxicology, Office of Research and Development, U.S. Environmental Protection Agency, Research
Triangle Park, North Carolina 27711, USA and 2Deptof Biostatistics University of North Carolina, Chapel Hill, 3126 McGavran-Greenberg Hall,
CB #7420, Chapel Hill, NC 27599-7420, USA
Email: Richard Judson* - judson.richard@epa.gov; Fathi Elloumi - elloumi.fathi@epa.gov; R Woodrow Setzer - setzer.woodrow@epa.gov;
Zhen Li - zli@bios.unc.edu; Imran Shah - shah.imran@epa.gov
* Corresponding author
Published: 19 May 2008 Received: 22 January 2008
BMC Bioinformatics 2008, 9:241 doi: 10.1 186/1471 -2105-9-241 Accepted: 19 May 2008
This article is available from: http://www.biomedcentral.eom/l47l-2IOS/9/24l
© 2008 Judson et al; licensee BioMed Central Ltd.
This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.Org/licenses/by/2.0).
which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
Abstract
Background: Bioactivity profiling using high-throughput in vitro assays can reduce the cost and
time required fortoxicological screening of environmental chemicals and can also reduce the need
for animal testing. Several public efforts are aimed at discovering patterns or classifiers in high-
dimensional bioactivity space that predict tissue, organ or whole animal toxicological endpoints.
Supervised machine learning is a powerful approach to discover combinatorial relationships in
complex in vitro/in vivo datasets. We present a novel model to simulate complex chemical-
toxicology data sets and use this model to evaluate the relative performance of different machine
learning (ML) methods.
Results: The classification performance of Artificial Neural Networks (ANN), K-Nearest
Neighbors (KNN), Linear Discriminant Analysis (LDA), Naive Bayes (NB), Recursive Partitioning
and Regression Trees (RPART), and Support Vector Machines (SVM) in the presence and absence
of filter-based feature selection was analyzed using K-way cross-validation testing and independent
validation on simulated in vitro assay data sets with varying levels of model complexity, number of
irrelevant features and measurement noise. While the prediction accuracy of all ML methods
decreased as non-causal (irrelevant) features were added, some ML methods performed better
than others. In the limit of using a large number of features, ANN and SVM were always in the top
performing set of methods while RPART and KNN (k = 5) were always in the poorest performing
set. The addition of measurement noise and irrelevant features decreased the classification
accuracy of all ML methods, with LDA suffering the greatest performance degradation. LDA
performance is especially sensitive to the use of feature selection. Filter-based feature selection
generally improved performance, most strikingly for LDA.
Conclusion: We have developed a novel simulation model to evaluate machine learning methods
for the analysis of data sets in which in vitro bioassay data is being used to predict in vivo chemical
toxicology. From our analysis, we can recommend that several ML methods, most notably SVM and
ANN, are good candidates for use in real world applications in this area.
Previous I TOC
Page 1 of 16
(page number not for citation purposes)
-------
BMC Bioinformatics 2008, 9:241
http://vvww.biomedcentral.eom/1471-2105/9/241
Background
A daunting challenge faced by environmental regulators
in the U.S. and other countries is the requirement that
they evaluate the potential toxicity of a large number of
unique chemicals that are currently in common use (in
the range of 10,000-30,000) but for which little toxicol-
ogy information is available. The time and cost required
for traditional toxicity testing approaches, coupled with
the desire to reduce animal use is driving the search for
new toxicity prediction methods [1-3]. Several efforts are
starting to address this information gap by using relatively
inexpensive, high throughput screening approaches in
order to link chemical and biological space [1,4-21]. The
U.S. EPA is carrying out one such large screening and pri-
oritization experiment, called ToxCast, whose goal is to
develop predictive signatures or classifiers that can accu-
rately predict whether a given chemical will or will not
cause particular toxicities [4]. This program is investigat-
ing a variety of chemically-induced toxicity endpoints
including developmental and reproductive toxicity, neu-
rotoxicity and cancer. The initial training set being used
comes from a collection of~300 pesticide active ingredi-
ents for which complete rodent toxicology profiles have
been compiled. This set of chemicals will be tested in sev-
eral hundred in vitro assays.
The goal of screening and prioritization projects is to dis-
cover patterns or signatures in the set of high throughput
in vitro assays (high throughput screening or HTS, high
content screening or HCS, and genomics) that are strongly
correlated with tissue, organ or whole animal toxicologi-
cal endpoints. One begins with chemicals for which toxi-
cology data is available (training chemicals) and develops
and validates predictive classification tools. Supervised
machine learning (ML) approaches can be used to
develop empirical models that accurately classify the tox-
icological endpoints from large-scale in vitro assay data
sets. This approach is similar to QSAR (quantitative struc-
ture activity relationship), which uses inexpensive calcu-
lated chemical descriptors to classify a variety of chemical
phenotypes, including toxicity. By analogy, one could use
the term QBAR (for quantitative bio-assay/activity rela-
tionship) to describe the use of in vitro biological assays to
predict chemical activity. The QBAR strategy we describe
here is also related to biomarker discovery from large-
scale -omic data that is used to predict on- or off-target
pharmacology in drug development, or to discover accu-
rate surrogates for disease state or disease progression.
The QBAR in vitro toxicology prioritization approach faces
a number of inter-related biological and computational
challenges. First, there may be multiple molecular targets
and mechanisms by which a chemical can trigger a biolog-
ical response. Assuming that these alternative biological
mechanisms of action are represented in the data, multi-
ple techniques (including ML methods) may be required
to discover the underlying relationships between bio-
assays and endpoint activity. Second, our present under-
standing of biological mechanisms of toxicity (often
referred to as toxicity pathways) is relatively limited, so
that one cannot a priori determine which of a set of assays
will be relevant to a given toxicity phenotype. As a conse-
quence, the relevant features may be missing from the
data set and (potentially many) irrelevant features may be
included. Here, by relevant features we mean data from
assays that measure processes causally linked to the end-
point of interest. By extension, irrelevant features include
data from assays not causally linked to the endpoint. The
presence of multiple irrelevant assays or features must be
effectively managed by ML methods. Third, due to the
high cost of performing the required in vivo studies, there
are limited numbers of chemicals for which high quality
toxicology data is available, and typically only a small
fraction of these will clearly demonstrate the toxic effect
being studied. The small numbers of examples and unbal-
anced distribution of positive and negative instance for a
toxicological endpoint can limit the ability of ML meth-
ods to accurately generalize. In order to develop effective
QBAR models of toxicity, these issues must be considered
in the ML strategy.
Four critical issues for evaluating the performance of ML
methods on complex datasets are: (1) the data set or
model; (2) the set of algorithms evaluated; (3) the
method that is used to assess the accuracy of the classifica-
tion algorithm; and (4) the method that is used for feature
selection. In order to address the first issue, it was neces-
sary to develop a model of chemical toxicity that captured
the key points of the information flow in a biological sys-
tem. The mathematical model we use is based on the fol-
lowing ideas.
1. There are multiple biological steps connecting the ini-
tial interaction of a molecule with its principle target(s)
and the emergence of a toxic phenotype. The molecular
interaction can trigger molecular pathways, which when
activated may lead to the differential activation of more
complex cellular processes. Once enough cells are
affected, a tissue or organ level phenotype can emerge.
2. There will often be multiple mechanisms that give rise
to the same phenotype, and this multiplicity of causal
mechanisms likely exists at all levels of biological organi-
zation. Multiple molecular interactions can lead to a sin-
gle pathway being differentially regulated. Up-regulation
of multiple pathways can lead to the expression of the
same cellular phenotype. This process continues through
the levels of tissue, organ and whole animal. One can
think of the chain of causation between molecular triggers
Previous
TOC
Page 2 of 16
(page number not for citation purposes)
-------
BMC Bioinformatics 2008, 9:241
http://vvww.biomedcentral.eom/1471-2105/9/241
and endpoints as a many-branched tree, potentially with
feedback from higher to lower levels of organization.
3. The number of assays one needs to measure is large,
given our relative lack of knowledge of the underlying
mechanism linking direct chemical interactions with toxic
endpoints.
4. The number of example chemicals for which detailed
toxicology information is available is relatively limited
due to the high cost of generating the data. In most cases,
if a chemical is known to be significantly toxic, further
development and testing is halted, so it is unusual to have
complete, multi-endpoint toxicity data on molecules that
are toxic for any given mode. A corollary is that the
number of positive examples for any given toxicity end-
point will be very limited, rarely making up more than
10% of all cases. This will limit the power to find true
associations between assays and endpoints. A related issue
is that most publicly available data sets that one can use
for toxicology modeling are heavily biased toward posi-
tive or toxic chemicals, because much less public effort is
put into performing extensive studies on chemicals that
are negative examples. The ToxCast data set is addressing
this selection bias by gathering complete data from a set
of chemicals without regard to their ultimate toxicity.
5. The available toxicity endpoint data tends to be categor-
ical rather than quantitative. This is due to the nature of
the in vivo experiments used to evaluate chemical toxicity.
Typically, too few animals are tested under any given con-
dition to pinpoint the lowest effective dose or the rate of
phenotypic toxicity at a particular dose. Instead, if a toxic
effect is seen at a rate statistically above that seen with a
negative control, the chemical will be classified as causing
that toxicity.
We have developed a simple simulation model which
takes into account these ideas. Here we motivate the struc-
ture of the model, while the Methods and Results sections
provide details. We will illustrate the ideas behind our
model with the multiple known pathways that can lead to
rodent liver tumors. Several nuclear receptors, including
CAR (constitutive androstane receptor), PXR (pregnane-X
receptor) and AHR (aryl hydrocarbon receptor), when
activated by a xenobiotic, can upregulate a common set of
Phase I, Phase II and Phase III metabolizing enzyme path-
ways [22-24]. Each of these pathways can, when continu-
ally activated, lead to cellular phenotypes that include cell
proliferation, hypertrophy and cell death. A second, paral-
lel route is activated by compounds that bind to PPARa
(peroxisome proliferator-activated receptor a) and lead to
cellular hypertrophy and cellular proliferation [24,25]. In
a third mechanism, chemicals can directly interact with
DNA, causing the activation of DNA damage repair path-
ways, which can in turn lead to cell death and cellular pro-
liferation. All three of these cellular phenotypes are
potential precursors to liver tumors [26]. This collection
of interconnected direct molecular targets, target-induced
pathways, cellular or tissue phenotypes, and their connec-
tions to the endpoint of liver tumors are illustrated in Fig-
ure 1.
Our model also assumes that a given chemical can interact
with multiple molecular targets. It is well known that
many drug compounds interact with multiple targets, as
reflected in the phenomenon of off-target toxicity. Rele-
vant to the pathways shown in Figure 1, Moore et al
showed that there are compounds that simultaneously
activate both CAR and PXR pathways [27]. Preliminary
data from the ToxCast program allows us to quantify the
magnitude of this multi-target effect. From a set of 183
biochemical targets (primarily receptors, enzymes and ion
channels), the 320 ToxCast chemicals[28] (mostly pesti-
cides) were active against an average of 4.2 targets with a
maximum of 35, a minimum of 0 and a standard devia-
tion of 5.8.
The connections shown in Figure 1 are not deterministic
but instead depend on multiple factors including the
strength and duration of the initial chemical-target inter-
action. Some pathways are more likely than others to lead
to the manifestation of particular cellular processes, and
some cellular processes are more likely than others to lead
to liver tumors. Based on this, one could assign a proba-
bility or strength to each arrow in Figure 1. The probabil-
ity that a given chemical will cause liver tumors is then a
complex integral over the individual step-to-step proba-
bilities, modulated by the target interaction strengths for
the particular chemical.
There is a vast literature on the evaluation of the perform-
ance of different ML methods, but for the present applica-
tion the literature concerning the analysis of microarray
genomics data sets and for QSAR applications are most
relevant. Here we describe a pair of representative studies.
Ancona et al. [29] used three algorithms (Weighted Voting
Algorithm (WVM), Regularized Least Squares (RLS), Sup-
port Vector Machine (SVM)) to classify microarray sam-
ples as either tumor or normal. They examined the
number of training examples that would be required to
find a robust classifier. In their example, SVM and RLS
outperformed WVM. Statnikov et al. studied all of the
major classification issues in the context of multi-category
classification using microarray data in cancer diagnosis
[30]. They compared multi-category SVM (MC-SVM), k-
nearest neighbors (KNN) and several artificial neural net-
work (ANN) implementations and showed that MC-SVM
was far superior to the other algorithms they tested in
their application.
Previous
TOC
Page 3 of 16
(page number not for citation purposes)
-------
BMC Bioinformatics 2008, 9:241
http://vvww.biomedcentral.eom/1471-2105/9/241
Nuclear
DNA
Direct Molecular Targets
Molecular Pathways / Processes
Cellular /Tissue Processes
Tissue / Organ Endpoint
Figure I
Connections between molecular targets, pathways, cellular processes and endpoints. This is illustrated for 5 molecular targets
(nuclear DNA, and the nuclear receptors CAR, PXR, AHR and PPARa), three molecular pathways, and three cellular pheno-
types, with liver tumors being the final endpoint. The connections have differing strengths or probabilities and are modulated
by the collection of interactions of a given chemical with the molecular targets.
The literature on machine learning methods in QSAR is
equally vast and extends back for 15 years or more. Much
of this work (like much of QSAR in general) is focused on
the (relatively easy) task of predicting activity against
molecular targets. A representative approach to target
interaction prediction is the paper by Burbridge et al. com-
paring SVM to several other algorithms for the prediction
of binding to dihydrofolate reductase [31]. Lepp et al per-
formed a similar study that showed SVM performed well
in finding predictive QSAR models for a series of 21
molecular targets [32]. The recent state of the science for
predicting whole animal toxicity using ML and QSAR
methods were reviewed by Helma and Kramer [33],
Benigni and Giuliani [34] and by Toivonen et al. [35].
They describe the outcome of an experiment (the Predic-
tive Toxicology Challenge) in which 17 groups submitted
111 models using a training set of 509 NTP compounds
for which mouse carcinogenicity data was available. The
goal was to predict the carcinogenicity of a set of 185 test
compounds. Only 5 of the 111 models performed better
than random guessing and the highest positive predictive
value for these was 55%, and this model had a false posi-
tive rate of 37%. These 5 models[36] include rule-based
methods using chemical fragments plus calculated physi-
cochemical properties, a decision tree model, and one
using a voting scheme across several standard ML meth-
ods. It is difficult to draw many conclusions about the per-
formance of ML methods from this exercise, which failed
to produce significantly predictive methods. The authors
of these reviews speculate that the cause is a combination
of toxicity data being too noisy, the training and test
chemical spaces being too large, and structure based
approaches being inadequate to predict phenotypes as
complex as whole animal toxicity.
One of the key issues in systematically comparing the per-
formance of ML methods is that of estimating accuracy in
an unbiased way. For example, Ntzani and loannidis [37]
report that many of the early studies using microarray data
to classify tumor samples did not perform appropriate
cross validation, which has led to inflated predictions of
classification accuracy. This observation prompted our
use of independent validation sets. Molinaro et al.
showed that 10-fold cross validation performed well for
assessing accuracy of genomics classifiers [38]. Leave one
out cross-validation (LOOCV) typically performed some-
what better, but had a significantly higher computational
cost. This was assessed by Molinaro et al. in the context of
using linear discriminant analysis (LDA), ANN, diagonal
discriminant classifiers (DDA), classification and regres-
sion trees (CART) and ensemble classifiers. The Molinaro
study data set (300 samples and 750 independent varia-
Previous
TOC
Page 4 of 16
(page number not for citation purposes)
-------
BMC Bioinformatics 2008, 9:241
http://vvww.biomedcentral.eom/1471-2105/9/241
bles), which used simulated genomics data, was similar in
size to the present work. Baldi, et al. [39] have systemati-
cally addressed the issue of ML performance metrics. They
describe a number of accuracy metrics including the bal-
anced accuracy or Q-score we use in this paper. The Q-
score is the average of the sensitivity and specificity. This
is most useful in the case where the classification variable
is dichotomous and where the number of positive and
negative cases in a training set is not well balanced. They
also emphasize that the actual prediction accuracy is
related to the similarity of the training and test set.
Finally, Sima and Dougherty examined the issue of find-
ing an optimal subset of features with which to train a
classification algorithm [40]. They compare sequential
floating forward search (SFFS) [41] and T-test feature selec-
tion. This latter can fail when variables are only predictive
when they act together. These authors' basic conclusion is
that there are optimal subsets of features, but that poor
classification performance can be due to either a failure to
find an optimal subset or to the inability of any subset to
allow accurate classification. This study examined SVM,
KNN (n = 3) and LDA as classification algorithms. These
authors suggest that automated feature selection methods
have inherent limitations and that one should use biolog-
ically-based selection when possible. Baker and Kramer
used the nearest centroids rule to select small subsets of
genes that could be used as robust classifiers from genom-
ics data sets [42]. Kohavi assessed the behavior of cross
validation methods to assess classifier accuracy for the
C4.5 and Naive Bayes algorithms [43]. This author con-
cludes that k-fold cross validation with k = 10 provides a
good estimate of classification accuracy balanced against
modest computational requirements.
In summary, the goal of the analyses we present is to eval-
uate a machine learning approach to develop classifiers of
in vivo toxicity using in vitro assay data. In order to develop
an appropriate ML strategy, we generate simulated QBAR
data using a mathematical model whose structure and
parameters are motivated by an idealized biological
response to chemical exposure based on the following
concepts: (a) chemicals interact with multiple molecular
targets; (b) exposure to chemicals can stimulate multiple
pathways that lead to the same toxicological endpoint;
and (c) there are multiple levels of biological organization
between the direct molecular interaction and the "apical"
endpoint. Additional parameters for generating simulated
data include model complexity, the level of noise in the
features, the number of chemicals to be screened and the
number of irrelevant features. We focus on the special case
where there is a large imbalance between the fraction of
positive and negative examples, which is found to be the
case from our toxicological data [44]. The performance of
ML methods is analyzed as a function of these parameters.
Results
We evaluated the performance of different ML methods
on simulated data sets generated by a biologically moti-
vated analytic model. Data sets were simulated based on
two levels of complexity; the number of irrelevant assays
or input features in the data (data not causally connected
with the endpoint being predicted); the number of chem-
icals or instances; and the presence or absence of measure-
ment noise in the data. In all cases, all of the relevant
features (causal for the endpoint being predicted) were
included in the data set.
The network depiction of the simulation models SI (less
complex) and S2 (more complex) are illustrated in Figures
2 and 3. These networks closely resemble the one shown in
Figure 1, which models the connections leading from direct
molecular interactions with DNA and a variety of nuclear
receptors and to liver tumors. Structurally, the simulation
models are feed-forward networks that causally link direct
molecular interactions (M-nodes) with a final organism-
level toxicity endpoint, by way two levels of intervening
biological processes. Direct molecular interactions trigger
pathway processes (P-nodes) which in turn trigger cellular
processes (C nodes). Only if the cellular processes are acti-
vated to a sufficient level is the final endpoint manifested.
Of equal importance is the fact that many assays will be
measured that are not causally linked to the endpoint.
These irrelevant nodes are termed R-nodes for random. Our
simulations typically include many more R than M nodes
or features. Rules for linking molecular interaction
strengths to the endpoint are described in the Methods sec-
tion. The essential points for the present discussion are that
a given chemical can interact with one or more input nodes
and that the spectrum of input interactions uniquely deter-
mines the value of the endpoint.
The performance of LDA (Linear Discriminant Analysis),
KNN (k-Nearest Neighbors), SVM (Support Vector
Machines), ANN (Artificial Neural Networks), NB (Naive
Bayes) and RPART (Recursive Partitioning and Regression
Trees) was evaluated both with and without filter-based
feature selection, using 10-way cross-validation testing, as
well as validation with independent data sets which
included 300 instances. For each set of conditions (ML
method, model, number of features, number of chemi-
cals, inclusion of measurement noise, and the presence or
absence of filter-based feature selection), training was car-
ried out on 10 independent samples drawn from a simu-
lated data set of 10,000 chemicals. For all evaluations,
10% of the chemicals were positive and 90% were nega-
tive for the endpoint being predicted. As mentioned pre-
viously, this imbalance between positive and negative
examples reflects the situation with the data sets we are
modeling in which the adverse phenotypes being studied
are rare. Predicted performance was evaluated using K-
Previous
TOC
Page 5 of 16
(page number not for citation purposes)
-------
BMC Bioinformatics 2008, 9:241
http://vvww.biomedcentral.eom/1471-2105/9/241
0.1
Figure 2
Model SI. The "M" nodes represent assays that measure direct molecular interactions with a chemical. These interactions can
activate pathways ("P" nodes) which can in turn activate cellular processes ("C" nodes). Finally, the activation of cellular proc-
esses can lead to the presence of an organ or organism-level endpoint. For Model SI, an additional 300 random or "R" nodes
were included in the input set of features, so that a total of 308 features are examined. Numerical values shown along edges
are values of wik used in Equation I.
fold cross-validation with K = 10. For each of the 10 sam-
ples, we recorded the number of true positives (TP), false
positives (FP), true negatives (TN) and false negatives
(FN), sensitivity and specificity and the balanced accuracy
or Q-score, which is the average of the sensitivity and spe-
cificity. To independently test the performance of the ML
method, an independent validation set was drawn from
the simulated data set and evaluated with the classifica-
tion models for each of the 10 training sets. The results
(TP, FP, TN, FN, sensitivity, specificity, Q-score) from
these 10 data sets were also saved. The approach is out-
lined in Figure 4.
Figure 3
Model S2. All symbols are as described in Figure I. There are a total of 24 "M" nodes plus 300 "R" nodes for a total of 324 fea-
tures to be examined.
Previous
TOC
Page 6 of 16
(page number not for citation purposes)
-------
BMC Bioinformatics 2008, 9:241
http://vvww.biomedcentral.eom/1471-2105/9/241
Biological
Model
Simulated
Data
10 Iterations per
ML method and
condition set
Cross
Validation
Sample
k=10 spite
Independent
Validation
Sample
Mean Q for
k= 10 splits
and SD
Q(X-val) and Q(l-val)
Figure 4
Schematic view of the learning method employed. A large simulated data set is created from the model. From this large data
pool, multiple independent samples are drawn and either used for cross validation training and validation (X-val) (left hand
branch) or independent model validation (l-val) (right hand branch). For cross validation training, we use standard K-fold cross
validation with K = 10. The cross validation performance is the average of the 10 partitions. The classification model ("fit") used
in the right hand, independent validation branch is constructed using the entire data set for the left hand branch. For each clas-
sifier and each set of conditions, a total of 10 samples are drawn for the cross validation and 10 for the independent validation
processes. From this collection of results, we derive means and standard deviations for the balanced accuracy or Q-score.
Previous
TOC
Next
Page 7 of 16
(page number not for citation purposes)
-------
BMC Bioinformatics 2008, 9:241
http://vvww.biomedcentral.eom/1471-2105/9/241
The overall performance results of the different ML meth-
ods for the independent validation tests are shown in Fig-
ure 5. All results in this figure are calculated using Model
SI (Figure 2) for the case where the training and valida-
tion sets contained 300 chemicals or instances. Each panel
shows the Q-score trend for the ML methods as a function
of the number of features included. Horizontal lines are
drawn at Q = 0.9, which is a point that guarantees at least
80% sensitivity and specificity, and at Q = 0.5, which
occurs when sensitivity = 0 (all cases are predicted to be
negative for the endpoint). The far left point is the case
where only the causal features are used. Error bars (+ 1
SD) are given for the LDA results to provide an estimate of
the level of variation. The other methods showed similar
levels of variation. The figure shows the Q-Score curves as
a function of increasing number of irrelevant input fea-
tures in four blocks. In each block, each curve shows the
Q-score for one ML method beginning with just the causal
features (Nfeature = 8) and then increasing the number of
irrelevant features until Nfeature = 308. In the first block, the
curves generally show a decrease in performance going
from Nfeature = 8 to Nfeature =308, which means that the
accuracy of all learning methods generally decreased as
irrelevant features were added.
causal features, while with the maximum number of irrel-
evant features ANN, NB and SVM performed the best and
LDA the worst, at least in the absence of feature selection.
With the exception of LDA, the performance of different
ML methods stabilized after around 100 irrelevant fea-
tures. With the maximum number of irrelevant features
the classification accuracy of KNN and RPART were inter-
mediate between that of the highest group (ANN, SVM,
NB) and the lowest (LDA).
The second block from the left shows the classification
accuracy of the ML methods without feature selection but
with the addition of measurement noise. With no irrele-
vant features the classification accuracy of all ML methods
was significantly lower than in the absence of noise, as
expected. LDA showed the same maximum negative per-
formance trend with the addition of irrelevant features.
The main difference from the previous case (no noise) was
that the performance of KNN (k = 3) was close to that of
ANN, NB and SVM as the number of irrelevant features
increased. As before, RPART and KNN (k = 5) did not per-
form well. In general, the classification performance of
LDA degraded the most with addition of noise while other
methods remained more stable.
The response of different ML methods to the addition of
noise varied: LDA and ANN performed the best with only
The third block from the left shows the classification accu-
racy of the ML methods with filter-based feature selection
o
CT>
O
cq
o
Independent Validation
-------
BMC Bioinformatics 2008, 9:241
http://vvww.biomedcentral.eom/1471-2105/9/241
(T-test) in the absence of noise. Comparing the perform-
ance of the ML methods with the first block (no noise, no
feature selection), most ML methods performed better
with feature selection but their overall ranking was the
same. The exception was LDA, which showed the greatest
improvement in performance, tied with SVM and ANN
with the greatest Q-score. Feature selection also decreased
the overall variability in classification performance
between the different ML methods.
The fourth and final block represents the performance
results for the ML methods with noise and the use of T-test
feature selection. Compared with block 2, where feature
selection was not used, the performance of most ML
methods increases slightly. LDA showed a significant
increase in performance. Compared with block 3, the per-
formance of all techniques was significantly lower when
irrelevant features were added. Overall, LDA, NB, SVM,
ANN and KNN (N = 3) were quite stable i.e. their perform-
ance did not vary tremendously with the addition of noise
and irrelevant features.
An alternate way to examine the data is to fix the number
of features and look at trends as a function of number of
chemicals sampled. These curves (not shown) display the
expected trends that as the number of chemicals increases,
there is a corresponding improvement in performance.
The effects of the variant conditions are basically the same
as has already been shown.
Table 1 summarizes the results for both models SI and S2
for the limiting case where all 300 irrelevant features are
included. For all results, 300 chemicals were used. The
table is organized into 4 blocks, the same as in Figure 5,
but the rows within each block are sorted by decreasing
values of Q-score. Values of sensitivity, specificity or Q-
score > 0.8 are bolded. Rows shaded in gray have Q-score
values less than the best Q-score in that block minus one
standard deviation for the best performing method. From
this table, one can see that specificity is typically high and
that sensitivity is typically low. With a small number of
positive cases, a safe classification scheme is to assume
that most cases will be negative. The ML methods chiefly
differ by their ability to correctly predict the positive cases,
which is reflected in the sensitivity. In all cases, KNN (k =
5) and RPART perform poorly relative to the best ML
method. In the absence of feature selection, LDA also per-
forms poorly. SVM and ANN are always among the best
performers. NB and KNN (k = 3) are intermediate in per-
formance robustness (i.e. relative lack of sensitivity to
added noise and number of irrelevant features). The
trends for model S2 are not significantly different from
those for the simpler model SI. The addition of measure-
ment noise significantly degraded the performance of all
ML methods, and this degradation is mainly reflected in
poorer sensitivity, i.e. the ability to correctly predict posi-
tive cases.
Discussion
Developing predictive classifiers for complex biological
data sets is a challenging problem because there are gener-
ally more features than instances (curse of dimensional-
ity); the classification variable and input features are
noisy; and there are many irrelevant features (i.e. ones
that are measured but which have no causal connection to
the value of the classification variable). We have devel-
oped a test bed for representing biologically motivated
models and have used it to provide insight into the rela-
tive classification performance of different ML methods.
Though true in vitro biological systems are more complex
and dynamic than our model, our approach provides
empirical insight into the relative performance of different
learning methods as a function of the absence and pres-
ence of experimental noise and the number of features. In
particular, we have focused on the situation which is com-
mon in toxicology data sets, namely where there is an
imbalance between the number of positive and negative
examples.
We find several main trends from our simulated data by
systematically analyzing different ML methods on the
same testing, training and validation data. First, most ML
methods perform well in the presence of a small number
of causal features, but most show significant degradation
in performance as irrelevant features are added, which is
well-known [45]. Second, all ML methods perform better
with filter-based feature selection as irrelevant features are
added. Third, the performance depends upon noise in the
input features. While most ML methods perform well in
the absence of noise, some are more stable than others.
Fourth, in the presence of noisy and irrelevant features,
and with feature selection, most ML methods perform
similarly, with the exceptions of RPART and KNN (k = 5)
which performed significantly worse. The models (Figures
2 and 3) resemble generalized artificial neural networks,
leading one to suspect that ANN methods should perform
well. In general this is true, although (see Figure 5) other
methods always performed at least as well.
We found that the accuracy predicted using k-fold cross
validation was statistically indistinguishable from that
seen with an independent validation set except in the case
of KNN (k = 3 or 5) with no feature selection. In this case,
the k-fold cross validation predicted a higher accuracy
than was seen with independent validation. This is the
only situation where we detected over-fitting using the
training data. This phenomenon disappeared when we
tested KNN against a more balanced data set in which
there were equal numbers of positive and negative exam-
ples. All other parameters were unchanged. Issues arising
Previous
TOC
Page 9 of 16
(page number not for citation purposes)
-------
BMC Bioinformatics 2008, 9:241
http://vvww.biomedcentral.eom/1471-2105/9/241
Table I: Performance (mean and SD) of the ML methods.
Model
SI
SI
S2
S2
SI
SI
S2
SI
S2
S2
SI
SI
S2
S2
SI
SI
SI
S2
S2
S2
S2
SI
SI
S2
S2
SI
SI
S2
SI
SI
SI
S2
S2
S2
SI
SI
S2
S2
SI
S2
S2
SI
SI
SI
S2
SI
S2
S2
SI
S2
SI
S2
SI
S2
SI
S2
Learner
ANN
SVM
SVM
NB
NB
KNN(k=3)
ANN
CART
KNN(k=3)
LDA
KNN(k = S)
LDA
KNN(k = S)
CART
SVM
LDA
ANN
LDA
ANN
NB
SVM
NB
KNN(k=3)
KNN(k=3)
CART
CART
KNN(k = 5)
KNN(k = S)
KNN(k=3)
ANN
NB
KNN(k=3)
ANN
NB
SVM
CART
KNN(k = 5)
CART
KNN(k = 5)
SVM
LDA
LDA
LDA
SVM
NB
NB
LDA
SVM
ANN
KNN(k=3)
KNN(k=3)
ANN
KNN(k = S)
KNN(k = 5)
CART
CART
Noise
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
Feature Selection
None
None
None
None
None
None
None
None
None
None
None
None
None
None
T-test
T-test
T-test
T-test
T-test
T-test
T-test
T-test
T-test
T-test
T-test
T-test
T-test
T-test
None
None
None
None
None
None
None
None
None
None
None
None
None
None
T-test
T-test
T-test
T-test
T-test
T-test
T-test
T-test
T-test
T-test
T-test
T-test
T-test
T-test
0.7 1
0.68
0.66
0.63
0.62
0.58
0.56
0.56
0.47
0.7
0.45
0.66
0.34
NA
0.75
0.74
0.73
0.67
0.7
0.65
0.64
0.6 1
0.55
0.52
0.58
0.54
0.45
0.37
0.54
0.53
0.48
0.49
0.44
0.43
0.4 1
0.4
0.33
0.34
0.3
0.3
0.59
0.6
0.55
0.5
0.53
0.52
0.52
0.48
0.5 I
0.48
0.48
0.48
0.36
0.32
0.35
0.25
SD(Sens)
O.I2
0.076
0.086
O.I3
0.096
O.I3
O.I9
O.I3
O.I6
O.I I
O.I3
O.I3
O.I
NA
0.089
O.I I
O.I3
O.I I
O.I I
O.I3
O.I I
0.078
O.I6
O.I7
O.I2
O.I7
O.I2
O.I3
O.I I
O.I3
O.I3
O.I3
O.I I
0.062
0.09
O.I2
0.087
O.I7
O.I I
0.079
0.069
O.I2
O.I I
O.I I
0.084
0.083
0.087
O.I
O.I
O.I4
O.I
O.I
0.095
0.047
O.I4
0.093
0.96
0.99
0.99
0.98
0.99
0.98
0.98
0.96
0.98
0.76
0.98
0.69
0.99
NA
0.98
0.99
0.97
0.98
0.96
0.98
0.98
0.97
0.99
0.99
0.94
0.96
I
0.99
0.98
0.97
0.99
0.98
0.98
0.99
I
0.94
0.99
0.96
0.99
I
0.72
0.64
0.98
0.99
0.98
0.98
0.97
0.99
0.97
0.99
0.98
0.96
I
0.99
0.95
0.96
SD(Spec)
0.038
0.0063
0.0095
0.0072
0.0049
0.0 1 3
0.008
0.0 1 8
0.0 1 4
0.05 I
0.0 1 4
0.048
0.0 1 6
NA
0.0 1 I
0.0057
0.023
0.009
0.0 1 3
0.0 1 4
0.0074
0.0 1 2
0.008
0.0044
0.025
0.038
0.0035
0.0056
0.0 1 I
0.0 1 9
0.0054
0.009 1
0.0 1 6
0.0056
0.003 I
0.04 1
0.0066
0.03
0.0085
0.0039
0.05
0.037
0.0078
0.004
0.0 1
0.0088
0.0 1 I
0.0056
0.0 1 5
0.0083
0.0 1
0.0 1
0.0049
0.0036
0.023
0.027
0.84
0.83
0.82
0.8 1
0.8
0.78
0.77
0.76
0.73
0.73
0.72
0.67
0.66
NA
0.87
0.86
0.85
0.83
0.83
0.8 1
0.8 1
0.79
0.77
0.76
0.76
0.75
0.73
0.68
0.76
0.75
0.74
0.74
0.7 1
0.7 1
0.7
0.67
0.66
0.65
0.65
0.65
0.65
0.62
0.77
0.75
0.75
0.75
0.75
0.74
0.74
0.73
0.73
0.72
0.68
0.66
0.65
0.6
SD(Q)
0.068
0.039
0.04 1
0.063
0.047
0.067
0.09 1
0.065
0.077
0.05 I
0.064
0.079
0.049
NA
0.043
0.052
0.055
0.054
0.05 1
0.064
0.057
0.039
0.077
0.082
0.06 1
0.07
0.062
0.066
0.057
0.066
0.067
0.065
0.049
0.029
0.045
0.053
0.043
0.08
0.054
0.039
0.038
0.068
0.055
0.053
0.042
0.042
0.042
0.052
0.045
0.07
0.05 1
0.049
0.047
0.023
0.07 1
0.044
This data is compiled for the special case where 300 chemicals were used, as a function of model, feature selection and level of measurement noise. The
results are organized into 4 blocks, corresponding to the 4 blocks in Figure 5. Within a block, rows are ordered by decreasing values of Q-Score. The
results give the average sensitivity, specificity and Q-score along with their corresponding standard deviations. All ML methods were trained using 300
chemicals. The values come from 10 independent validation runs with unique samples of 300 chemicals. Values of sensitivity, specificity and Q-score >
0.8 are bolded. Rows where the Q-score is less than that of the best Q-score in the block minus one standard deviation for the best row are shaded.
Previous
TOC
Page 10 of 16
(page number not for citation purposes)
-------
BMC Bioinformatics 2008, 9:241
http://vvww.biomedcentral.eom/1471-2105/9/241
S1 - Deterministic Assays
S2 - Deterministic Assays
0 .
CN
C
CD
Q.
§ o
o
-10
-5
0
\^
5
Component 1
Figure 6
Distribution of chemicals in feature/assay space for model
and S2. The data is projected into 2 dimensions using multi-
dimensional scaling. Chemicals that are negative for the end-
point are indicated by black circles, and chemicals positive for
the endpoint are represented by red crosses. For ease of vis-
ualization only a randomly selected set of 500 chemicals are
shown.
from unbalanced data sets have been previously analyzed.
Japkowicz et al. found that classifier performance of
imbalanced datasets depends on the degree of class imbal-
ance, the complexity of the data, the overall size of the
training set and the classifier involved [46]. Sun et al. also
observed that given a fixed degree of imbalance, the sam-
ple size plays a crucial role in determining the "goodness"
of a classification model [47]. The KNN method is sensi-
tive to imbalanced training data [48,49], and the class dis-
tribution of our simulation data is highly skewed with the
positive to negative rate of 1:9, thus the sample size very
likely explains the different performance between the
training and validation sets.
One of the important limitations of this work is that the
performance of classifiers is biased by our model of chem-
ical-induced bioactivity and toxicity. We assume a static
deterministic model of a biological system without feed-
back. An important aspect of the chemical simulation
model is the use of multiple chemical classes, each of
which contains a collection of chemicals that behave sim-
ilarly (as measured by their molecular interaction spec-
trum). As described in the methods section, a chemical
m
o
CN
•£ LO
-------
BMC Bioinformatics 2008, 9:241
http://vvww.biomedcentral.eom/1471-2105/9/241
ing, all-positive clusters is very obvious when all features
are included.
We have focused on the performance of single classifiers,
but voting methods which combine the predictions of
multiple individual methods have been used. Statnikov et
al. studied ensemble classifiers or voting schemes, which
attempt to combine multiple sub-optimal classifiers to
improve overall accuracy. That paper evaluated the utility
of selecting very small subsets of genes (as few as 25 out
of> 15,000) for classification. This has the effect of greatly
reducing the danger of over-fitting from small numbers of
samples. Additionally, these authors demonstrated how
to evaluate the comparative performance of different algo-
rithms using permutation testing. Two conclusions from
the Statnikov et al. work on cancer diagnosis using micro-
array data are relevant to the present study. First, they
observe that SVM methods outperformed KNN and ANN.
Our findings show that the relative rankings of these 3
methods is a complex function of the number of irrele-
vant features, the level of noise and the use (or not) of fea-
ture selection. Second, the authors observed that the
ensemble classification methods tended to do worse than
single methods. Although we did not evaluate the per-
formance of ensemble based classification, our results
(Table 1 or Figure 5) do not suggest that voting would lead
to a decrease in performance, as long as the voting rule
was that the chemical was labeled positive if any method
predicted it to be positive.
The present work limited the number of chemicals to 300
and features to 300, which corresponds to the number of
chemicals and assays we are using in the first phase of the
ToxCast program. Despite the relatively small size of the
data set, we were able to evaluate key issues in supervised
learning from noisy and irrelevant data. We plan to
expand the number of features and instance in future
work as we gain additional insights from experimental
data. Additionally, we intend to more fully explore the use
of dimensionality reduction (e.g. through correlation
analysis of closely related features), feature selection and
classifier ensembles in future work.
Conclusion
The prediction of chemical toxicity is a significant chal-
lenge in both the environmental and drug development
arenas. Gold standard in vivo toxicology experiments in
rodents and other species are very expensive and often do
not directly provide mechanism of action information.
The alternative, which has been widely pursued in the
pharmaceutical industry, is to screen compounds using
use in vitro or cell based assays and to use the results of
these assays to prioritize compounds for further efficacy
and safety testing. These in vitro screening techniques are
now being introduced in a significant way into the world
of environmental chemical safety assessment. Here, there
are unique challenges due to the modest amount of in vivo
toxicology data that can be used to develop screening
models, and due to the broad chemical space covered by
environmental chemicals whose toxicology is poorly
characterized. The EPA is carrying out a significant screen-
ing and prioritization program called ToxCast, whose
eventual aim is to screen a large fraction of the commonly
used environmental chemicals and to prioritize a subset
of these for more detailed testing. The present analysis
provides a novel simulation model of the linkage between
direct chemical-target interactions and toxicity endpoints,
and uses this model to develop guidelines for using ML
algorithms to discover significant associations between in
vitro screening data and in vivo toxicology.
We find several main trends from our simulated data set
by systematically analyzing different ML methods on the
same testing, training and validation data. First, most ML
methods perform well in the presence of a small number
of causal features, but most show significant degradation
in performance as irrelevant features are added, which is
well-known [45]. Second, all ML methods perform better
with filter-based feature selection as irrelevant features are
added. Third, while most ML methods perform well in the
absence of measurement noise, some are more stable than
others. Fourth, in the presence of noisy and irrelevant fea-
tures, and with feature selection, most ML methods per-
form similarly well, with the main exceptions being
RPART and KNN which underperformed the other meth-
ods.
Methods
Simulation Models
We use two models of the networks connecting direct
molecular interactions with a test chemical and the pres-
ence or absence of a toxic endpoint. Direct molecular
interactions determine values of the M assays in the mod-
els. These interactions can trigger pathway processes (P-
nodes), which can in turn trigger cellular events (C-
nodes), which can finally lead to the expression of a toxic
endpoint. In addition to the M nodes, there are a large and
variable number of random or R nodes with which a
chemical can interact. Throughout the paper, we refer to
the M and R nodes as causal and irrelevant node or fea-
tures, respectively. A simulated chemical is uniquely char-
acterized by its spectrum of activity for the direct
molecular interaction assays (M + R nodes). The value of
the i-th M (or R) assay for chemical c is given by M;(c) and
is randomly generated from a gamma distribution (shape
= 3/2, rate = 0.5, ~95% of values are between 0 and 8).
This is the type of distribution one could see for -log(fe)
where k is a binding or inhibition constant for a molecule
interacting with protein target Figure 8 shows the distribu-
tion of values for the M and R assays or features.
Previous
TOC
Page 12 of 16
(page number not for citation purposes)
-------
BMC Bioinformatics 2008, 9:241
http://vvww.biomedcentral.eom/1471-2105/9/241
10 15 20
Assay Value
30
Figure 8
Distribution from which the M and R assay values are drawn.
This is a gamma distribution with shape = 3/2 and rate = 0.5
The model guarantees that if two molecules have the same
spectrum of direct physical interactions, they will exhibit
the same downstream biology, including whether or not
they cause the endpoint to be activated. By altering the
interaction strength connecting nodes in the model, one
can simulate differing degrees of coupling between multi-
ple molecular targets and the downstream processes they
control.
These networks simulate the ability for an endpoint to be
triggered by multiple independent mechanisms. In model
SI, there are 2 major mechanisms, driven by the inde-
pendent cellular processes Cl and C2 (see Figure 2). A col-
lection of chemicals may contain some substances that
trigger the endpoint through one mechanism and some
through the other. Some chemicals may trigger both. This
interplay of multiple mechanisms is characteristic of
many toxicological and disease processes and will allow
us to evaluate the ability of classification algorithms to
identify multiple paths from input to output in a biologi-
cal system.
For all P and C nodes, values are calculated using weights
for the edges leading into a node plus the values of the
parents:
where Xi (c) is the value for node Nt (c) in level L e [M,
R, P, C, Endpoint] for chemical c, and wik and wijk are
weights for the linear and quadratic interaction terms. The
quadratic term in Equation 1 simulates the presence of
cooperativity between upstream processes that is neces-
sary to trigger downstream processes. In order to test
binary classification algorithms, we assign chemicals to
the positive (1) class if the value of X;(c) for the endpoint
node is in the top 2% of the distribution, and to the neg-
ative (0) class otherwise. The weights values wik and wijk are
either 1.0 or 0.1 and are assigned sequentially through the
network using the repeating series (1.0, 1.0, 0.1, 0.1, 0.1).
For the simulations, 2 different model networks were
used, called SI and S2. The networks are shown in Figures
2 and 3. Model SI has 2 parents for each node. Model S2
has 4 C-level parents of the endpoint, 3 P-level parents for
each C node and 2 M-level parents for each P node. Note
that for S2, certain M-level molecular interactions can trig-
ger more than one of the major mechanisms. Figure 2 dis-
plays the values of the weights used for the linear portion
of the model. Both networks contained a total of 400
input layer nodes or molecular assays (SI: 8 M+392 R; S2:
24 M+374 R), although the simulations only made use of
up to 3 00 R nodes.
Simulation Data Sets
For each model (SI and S2), a set of 100,000 chemicals
was created with 2% being assigned to the positive end-
point class. The chemicals are not generated completely
randomly, but were instead created from 500 chemical
classes, each with 200 examples. To create a class, a first
example was randomly generated (M and R assays drawn
from the gamma distribution) and then the other exam-
ples are created from the exemplar by randomly adding
normally distributed variation (SD = 1) to each M and R
assay. The chemical class value (1... 5 00) was retained with
each chemical. From this large set of chemicals, a sam-
pling population was created by drawing 10,000 chemi-
cals from the larger set, but enriching the fraction of
positive cases to 10%. This represents a very broad uni-
verse of chemicals.
From the set of 10,000 chemicals, multiple samples were
drawn and used in the classification training and testing
process. The only data given to the classification algo-
rithms are the values for the M and R assays or features
and the endpoint classification. A sample was character-
ized by the following variables:
1. Model (SI, S2)
(1)
Previous
TOC
Page 13 of 16
(page number not for citation purposes)
-------
BMC Bioinformatics 2008, 9:241
http://vvww.biomedcentral.eom/1471-2105/9/241
Table 2: Classification or ML methods used, along with reference to the R library used.
ML Method Description
Library
KNN K-nearest neighbors (N = 3,5) MLInterfaces [50]
NB Naive Bayes el071 [51]
LDA Linear Discriminant Analysis MLInterfaces [50]
SVM Support Vector Machine (kernel = radial, cost = 100) e!07l[5l]
AN N Artificial Neural Networks (size = 10, range = 0.5, decay = 0.0001, maxit = 200, MaxN Wts = 10000) e 1071 [51 ]
RPART Recursive Partitioning and Regression Trees (method = class, cp = 0, usesurrogate = 2 el071 [51]
2. The number of chemicals (50,100,200,300)
3. The number of random or irrelevant features (R nodes)
(50,100,200,300)
4. Whether or not measurement noise was added to the
original M and R assay values. If so, normally distributed
noise (SD = 2) was added to each assay's value.
Classification Methodology
Each classification algorithm or ML method was evaluated
using the balanced accuracy or Q-score [39], which is the
average of the sensitivity and specificity for prediction.
This is a useful metric in the present situation because the
fraction of positive cases is small and the Q-score gives
equal weight to the accuracy of predicting positive and
negative cases. In each sample, the fraction of chemicals
that is positive for the endpoint is small (10%), so a good
first approximation would be to predict that all chemicals
will be negative for the endpoint. The Q-score for this
default prediction is 0.5, whereas a perfect prediction will
score 1.0.
Each ML method was evaluated against a set of 10 samples
or training sets, each using k-wise cross validation, with k
= 10 [43]. The model that was produced from each of the
training samples was evaluated against a separate valida-
tion sample. The training and validation samples were
drawn from the same distribution. We calculated distribu-
tions of Q-score for both the training samples (the results
of the k-fold cross validation) and the validation samples.
We call these the "predicted" and "true" Q-scores.
For each sample set described above, we evaluated per-
formance for a series of ML methods with no feature selec-
tion and with T-test filter feature selection. In the latter
case, the best 20% of features were selected, with a mini-
mum number of 8. (Note that the features (M-nodes) are
not strictly normally distributed, but are instead drawn
from a gamma distribution overlaid with normally dis-
tributed variation.) To manage the large number of indi-
vidual runs, a simple MySQL database was created with 2
tables called queue and result. The queue table contains all
run parameters and the result table holds all of the rele-
vant results. The relevant parameters in the queue table
are [model (SI, S2), measurement noise (0/2), number of
features, number of chemicals, ML method, feature selec-
tion mode (none or T-test)]. In all cases, the fraction of
positive cases in the sample was 10%. Figure 4 illustrates
the overall approach.
Classification AlgorithmslML methods
Table 2 lists the ML methods that were evaluated, along
with any non-default parameters. Parameters for each of
the machine learning methods were tuned so that the per-
formance (Q score) was acceptable (> 0.9) when tested
against model SI when the ML method was presented
with all of the true features, no irrelevant features, and
when no noise was added to the features. Default param-
eters were used for KNN, NB, LDA and RPART. For SVM,
the cost function was varied over the range from 1 to 1000
and a value of 100 was selected. ANN was the only
method requiring significant tuning. Approximately 20
combinations of the parameters listed in Table 1 were
tested prior to arriving at an acceptable set. All code was
written in R (version 2.5.1) using the MLInterfaces imple-
mentation of all ML methods. The code was parallelized
using snow and Rmpi and run on a Linux workstation
cluster and an SGI Altix 4700.
List of abbreviations
The following abbreviations are used in the manuscript:
ML: Machine Learning; KNN: k-Nearest Neighbors; NB:
Naive Bayes; LDA: Linear Discriminant Analysis; SVM:
Support Vector Machine; ANN: Artificial Neural Network;
CART: Classification and Regression Trees; RPART: Recur-
sive Partitioning and Regression Trees; HTS: High
Throughput Screening; HCS: High Content Screening.
Authors' contributions
RJ, IS, WS, ZL, FE participated in the design of the experi-
ment, in the design of the analysis strategy, in the formu-
lation of the conclusions and in implementation of the
analysis software. In addition, RJ developed the simula-
tion model and its software implementation and per-
formed the analysis runs. IS developed the bulk of the
final analysis software. RJ and IS drafted the manuscript.
All authors read and approved the final manuscript
Previous
TOC
Page 14 of 16
(page number not for citation purposes)
-------
BMC Bioinformatics 2008, 9:241
http://vvww.biomedcentral.eom/1471-2105/9/241
Acknowledgements
The authors wish to thank Edward Anderson and Govind Gawdi for help
with software configuration and code parallelization. Disclaimer: This man-
uscript has been reviewed by the U.S. EPA's National Center for Compu-
tational Toxicology and approved for publication. Approval does not signify
that the contents necessarily reflect the views and policies of the agency,
nor does mention of trade names or commercial products constitute
endorsement or recommendation for use.
References
I. Bhogal N, Grindon C, Combes R, Balls M: Toxicity testing: creat-
ing a revolution based on new technologies. Trends Biotechnol
2005, 23:299-307.
2. Directive 2003/I5/EC of the European Parliament and of the
Council of 27 February 2003 amending Council Directive 767
768/EEC on the approximation of the laws of Member States
relating to cosmetic products [http://ec.europa.eu/enterprise/
cosmetics/html/consolidated dir.htm]
3. REACH [http://ec.europa.eu/environment/chemicals/reach/
reach intro.htm]
4. Dix DJ, Houck KA, Martin MT, Richard AM, Setzer RW, Kavlock RJ:
The ToxCast program for prioritizing toxicity testing of
environmental chemicals. Toxicol Set 2007, 95:5-12.
5. Inglese J, Auld DS, Jadhav A, Johnson RL, Simeonov A, Yasgar A,
Zheng W, Austin CP: Quantitative high-throughput screening:
a titration-based approach that efficiently identifies biologi-
cal activities in large chemical libraries. Proc Not/ Acad Set USA
2006, 103:1 1473-1 1478.
6. Lamb J, Crawford ED, Peck D, Modell JW, Blat 1C, Wrobel MJ, Lerner
J, BrunetJP, Subramanian A, Ross KM, Reich M, Hieronymus H, Wei
G, Armstrong SA, Haggarty SJ, demons PA, Wei R, Carr SA, Lander
ES, Golub TR: The Connectivity Map: using gene-expression
signatures to connect small molecules, genes, and disease.
Science 2006, 313:1929-1935.
7. Strausberg RL, Schreiber SL: From knowing to controlling: a
path from genomics to drugs using small molecule probes.
Science 2003, 300:294-295.
8. Fliri AF, Loging WT, Thadeio PF, Volkmann RA: Biological spectra
analysis: Linking biological activity profiles to molecular
structure. Proc Natl Acad Sd USA 2005, 102:261-266.
9. Austin CP, Brady LS, Insel TR, Collins FS: NIH Molecular Libraries
Initiative. Science 2004, 306:1138-1139.
10. Bredel M, Jacoby E: Chemogenomics: an emerging strategy for
rapid target and drug discovery. Nat Rev Genet 2004, 5:262-275.
I I. Klekota J, Brauner E, Roth FP, Schreiber SL: Using high-through-
put screening data to discriminate compounds with single-
target effects from those with side effects. J Chem /nf Model
2006,46:1549-1562.
12. Kikkawa R, Fujikawa M, Yamamoto T, Hamada Y, Yamada H, Horii I:
In vivo hepatotoxicity study of rats in comparison with in
vitro hepatotoxicity screening system. J Toxicol Sd 2006,
31:23-34.
13. Fliri AF, Loging WT, Thadeio PF, Volkmann RA: Analysis of drug-
induced effect patterns to link structure and side effects of
medicines. Nat Chem Biol 2005, 1:389-397.
14. MelnickJS, Janes J, Kim S, Chang JY, Sipes DG, Gunderson D, Jarnes
L, Matzen JT, Garcia ME, Hood TL, Beigi R, Xia G, Harig RA, Asatryan
H, Yan SF, Zhou Y, Gu XJ, Saadat A, Zhou V, King FJ, Shaw CM, Su
Al, Downs R, Gray NS, Schultz PG, Warmuth M, Caldwell JS: An effi-
cient rapid system for profiling the cellular activities of
molecular libraries. Proc Natl Acad Sd USA 2006, 103:3153-3158.
15. O'Brien PJ, Irwin W, Diaz D, Howard-Cofield E, Krejsa CM, Slaughter
MR, Gao B, Kaludercic N, Angeline A, Bernardi P, Brain P, Hougham
C: High concordance of drug-induced human hepatotoxicity
with in vitro cytotoxicity measured in a novel cell-based
model using high content screening. Arch Toxicol 2006,
80:580-604.
16. Scherf U, Ross DT, Waltham M, Smith LH, Lee JK, Tanabe L, Kohn
KW, Reinhold WC, Myers TG, Andrews DT, Scudiero DA, Eisen MB,
Sausville EA, Pommier Y, Botstein D, Brown PO, Weinstein JN: A
gene expression database for the molecular pharmacology
of cancer. Not Genet 2000, 24:236-244.
17. Smith SC, DelaneyJS, Robinson MP, Rice MJ: Targeting chemical
inputs and optimising HTS for agrochemical discovery. Comb
Chem High Throughput Screen 2005, 8:577-587.
18. Tietjen K, Drewes M, Stenzel K: High throughput screening in
agrochemical research. Comb Chem High Throughput Screen 2005,
8:589-594.
19. Walum E, HedanderJ, Garberg P: Research perspectives for pre-
screening alternatives to animal experimentation On the
relevance of cytotoxicity measurements, barrier passage
determinations and high throughput screening in vitro to
select potentially hazardous compounds in large sets of
chemicals. Toxicol Appl Pharmacol 2005, 207:393-397.
20. Paolini GV.Shapland RH.van Hoorn WP, Mason JS, Hopkins AL: Glo-
bal mapping of pharmacological space. Nat Biotechnol 2006,
24:805-815.
21. Krewski D, DAcostaJ.Anderson M.Anderson H,JB III, Boekelheide
K, Brent R, Charnley G, Cheung V, Green S, Kelsey K, Kervliet N, Li
A, McCray L, Meyer O, Patterson DR, Pennie W, Scala R, Solomon
G, Stephens M, J Yager J, Zeize L: Toxicity Testing in the Twenty-first Cen-
tury: A Vision and a Strategy Washington D.C.: National Academies
Press; 2007.
22. Wang H, LeCluyse EL: Role of orphan nuclear receptors in the
regulation of drug-metabolising enzymes. Clin Pharmacokinet
2003,42:1331-1357.
23. Okey AB: An aryl hydrocarbon receptor odyssey to the
shores of toxicology: the Deichmann Lecture, International
Congress of Toxicology-XI. Toxicol Sci 2007, 98:5-38.
24. Heuvel JP Vanden, Thompson JT, Frame SR, Gillies PJ: Differential
activation of nuclear receptors by perfluorinated fatty acid
analogs and natural fatty acids: a comparison of human,
mouse, and rat peroxisome proliferator-activated receptor-
alpha, -beta, and -gamma, liver X receptor-beta, and retin-
oid X receptor-alpha. Toxicol Sci 2006, 92:476-489.
25. McMillian M, Nie AY, Parker JB, Leone A, Kemmerer M, Bryant S,
HerlichJ, Yieh L, BittnerA, Liu X, Wan J, Johnson MD: Inverse gene
expression patterns for macrophage activating hepatotoxi-
cants and peroxisome proliferators in rat liver. Biochem Phar-
macol 2004, 67:2141-2165.
26. Williams GM, latropoulos MJ: Alteration of liver cell function
and proliferation: differentiation between adaptation and
toxicity. Toxicol Pathol 2002, 30:41-53.
27. Moore LB, Parks DJ, Jones SA, Bledsoe RK, Consler TG, Stimmel JB,
Goodwin B, Liddle C, Blanchard SG, Willson TM, Collins JL, Kliewer
SA: Orphan nuclear receptors constitutive androstane
receptor and pregnane X receptor share xenobiotic and
steroid ligands. \ Biol Chem 2000, 275:15 122-15 127.
28. ToxCast [http://www.epa.gov/ncct/toxcast]
29. Ancona N, Maglietta R, Piepoli A, D'Addabbo A, Cotugno R, Savino
M, Liuni S, Carella M, Pesole G, Perri F: On the statistical assess-
ment of classifiers using DNA microarray data. BMC Bioinfor-
matics 2006, 7:387-401.
30. Statnikov A, Aliferis CF, Tsamardinos I, Hardin D, Levy S: A compre-
hensive evaluation of multicategory classification methods
for microarray gene expression cancer diagnosis. Bioinformat-
ics 2005, 21:63 1-643.
31. Burbridge R, Trotter M, Buxton B, Holden S: Drug design by
machine learning: support vector machines for pharmaceu-
tical data analysis. Computers & Chemistry 2001, 26:5-14.
32. Lepp Z, Kinoshita T, Chuman H: Screening for new antidepres-
sant leads of multiple activities by support vector machines.
J Chem Inf Model 2006, 46:158-167.
33. Helma C, Kramer S: A survey of the Predictive Toxicology
Challenge 2000-2001. Bioinformatics 2003, 19:1179-1182.
34. Benigni R, Giuliani A: Putting the Predictive Toxicology Chal-
lenge into perspective: reflections on the results. Bioinformat-
ics 2003, 19:1 194-1200.
35. Toivonen H, Srinivasan A, King RD, Kramer S, Helma C: Statistical
evaluation of the Predictive Toxicology Challenge 2000-
2001. Bioinformatics 2003, 19:1 183-1 193.
36. The Predictive Toxicology Challenge (PTC) for 2000-2001
[http://www.predictive-toxicology.org/ptc/ffROC]
37. Ntzani EE, loannidis JP: Predictive ability of DNA microarrays
for cancer outcomes and correlates: an empirical assess-
ment. Lancet 2003, 362:1439-1444.
Previous
TOC
Page 15 of 16
(page number not for citation purposes)
-------
BMC Bioinformatics 2008, 9:241
http://vvww.biomedcentral.eom/1471-2105/9/241
38. Molinaro AM, Simon R, Pfeiffer RM: Prediction error estimation:
a comparison of resampling methods. Bioinformatics 2005,
21:3301-3307.
39. Baldi P, Brunak S, Chauvin Y, Andersen CA, Nielsen H: Assessing
the accuracy of prediction algorithms for classification: an
overview. Bioinformatics 2000, 16:412-424.
40. Sima C, Dougherty ER: What should be expected from feature
selection in small-sample settings. Bioinformatics 2006,
22:2430-2436.
41. Pudil P: Floating Search Methods in Feature Selection. Pattern
Recognition Letters 1994, 15:1 I 19-1 125.
42. Baker SG, Kramer BS: Identifying genes that contribute most to
good classification in microarrays. BMC Bioinformatics 2006,
7:407-414.
43. Kohavi R: A Study of Cross Validation and Bootstrap for
Accuracy Estimation and Model Selection. International Joint
Conference on Artificial Intelligence; Montreal. IJCAI 1995. Unpaged
44. Martin MT, Houck KA, McLaurin K, Richard AM, Dix DJ: Linking
Regulatory Toxicological Information on Environmental
Chemicals with High-Throughput Screening (HTS) and
Genomic Data. The lexicologist CD - An official Journal of the Society
of Toxicology 2007, 96:219-220.
45. Almuallim H, Dietterich TG: Learning With Many Irrelevant
Features. Proceedings of the Ninth National Conference on Artificial
Intelligence 1991:547-552.
46. Japkowicz N, Stephen S: The class imbalance problem: A sys-
tematic study. Intelligent Data Analysis 2002, 6:429-450.
47. Sun Y, Kamel MS, Wong AKC, Wang Y: Cost-sensitive boosting
for classification of imbalanced data. Pattern Recognition 2007,
40:3358-3378.
48. Zhang J, Mani I: kNN Approach to Unbalanced Data Distribu-
tions: A Case Study involving Information Extraction. ICML
2003.
49. Li LH, T TM, Huang D: Extracting Location Names from Chi-
nese Texts Based on SVM and KNN. 2005 IEEE International
Conference on Natural Langrage Processing And Knowledge Engineering
2005, 10:371-375.
50. MLInterfaces: towards uniform behavior of machine learn-
ing tools in R [http://bioconductor.org/packages/1.8/bioc/vignettes/
MLInterfaces/inst/doc/M Llnterfaces.pdf]
51. The e 1071 package [http://cran.r-project.org/web/packages/
el07l/el07l.pdf]
Publish with BioMcd Central and every
scientist can read your work free of charge
"BioMed Central will be the most significant development for
disseminating the results of biomedical research in our lifetime."
Sir Paul Nurse, Cancer Research UK
Your research papers will be:
• available free of charge to the entire biomedical community
• peer reviewed and published immediately upon acceptance
• cited in PubMed and archived on PubMed Central
• yours — you keep the copyright
Submit your manuscript here:
http://www.biomedcentral.eom/i nfo/publishing_adv.asp
£ J BioMedcentral
Previous
TOC
Page 16 of 16
(page number not for citation purposes)
-------
48
ANN IST SUPER SANHA 2008 | VOL. 44, No. 1: 48-56
A novel approach: chemical relational databases,
and the role of the ISSCAN database
on assessing chemical carcinogenicity
Romualdo Benigni(a), Cecilia Bossa(a), Ann M. Richard(b) and Chihae Yang(c)
(a>Dipartimento di Ambiente e Connessa Prevenzione Primaria, Istituto Superiors di Sanitd, Rome, Italy
National Center for Computational Toxicology, US Environmental Protection Agency, Research
Triangle Park, North Carolina, USA
(c)LeadScope Inc., Columbus, Ohio, USA
Summary. Mutagenicity and carcinogenicity databases are crucial resources for toxicologists and
regulators involved in chemicals risk assessment. Until recently, existing public toxicity databases
have been constructed primarily as "look-up-tables" of existing data, and most often did not contain
chemical structures. Concepts and technologies originated from the structure-activity relationships
science have provided powerful tools to create new types of databases, where the effective linkage of
chemical toxicity with chemical structure can facilitate and greatly enhance data gathering and hy-
pothesis generation, by permitting: a) exploration across both chemical and biological domains; and
b) structure-searchability through the data. This paper reviews the main public databases, together
with the progress in the field of chemical relational databases, and presents the ISSCAN database
on experimental chemical carcinogens.
Key words: database, mutagenicity, carcinogenicity, chemical structure.
Riassunto ( Un approccio innovative: i database chimico relazionali e il ruolo del database ISSCAN per
la valutazione della cancerogenesi chimica). Basi di dati di cancerogenesi e mutagenesi sono essenziali
per la stima del rischio chimico. Finora queste si presentavano essenzialmente come tavole statiche,
ma i progress! nel campo delle relazioni struttura-attivita hanno permesso di creare nuove tipologie
dove 1'unione del dato tossicologico con la struttura chimica permette di legare ricerche in ambiti
chimico e biologico, e di esplorare i dati dal punto di vista strutturale. Questo articolo presenta le
principal! basi di dati pubbliche assieme agli sviluppi delle nuove banche dati chimico relazionali, e
illustra la banca dati ISSCAN sui cancerogeni chimici.
Parole chiave: basi di dati, mutagenesi, cancerogenesi, struttura chimica.
INTRODUCTION
Currently, the public has access to a variety of
databases containing mutagenicity and carcino-
genicity data. These resources are crucial for the
toxicologists and regulators involved in the risk as-
sessment of chemicals, which necessitate access to
all the relevant literature, and capability to search
across toxicity databases using both biological and
chemical criteria. In this field, rapid progress has
taken place both in terms of initiatives and tech-
nological innovation. In particular, public Internet
resources to support biological and toxicological
activity evaluation of chemicals have expanded
greatly and are ushering in a new era of public in-
formation access and data mining in support of
toxicity assessment.
In the context of the recent dramatic changes in
regulations and regulatory needs worldwide, the
progress in toxicological databases, and in database
technology is particularly timely and provides an
absolutely sine qua non tool for the regulatory imple-
mentations. As a matter of fact, increasing demands
and expectations are being placed on predictive tox-
icology in support of the new European REACH
legislation and other pieces of legislation worldwide
[1], and the need emerges for more structured or-
ganization and harnessing of legacy toxicity data,
and maximal utilization of these data [2]. Until now,
the assessment of chemical risk in the European
Union (EU) has been largely based on traditional
toxicology. However legislative, societal and practi-
cal realities (too many chemicals, too few resources)
have created new inducements and opportunities to
encourage use and acceptance of "alternative" ap-
proaches, which can reduce substantially the need
for experimental toxicological testing.
Address for correspondence: Romualdo Benigni, Dipartimento di Ambiente e Connessa Prevenzione Primaria, Istituto
Superiore di Sanita, Viale Regina Elena 299, Rome, Italy. E-mail: romualdo.benigni@iss.it.
Previous
-------
ISSCAN DATABASE
49
In 2003, the European Commission (EC) adopted a
legislative proposal for a new chemical management
system called REACH (Registration, Evaluation
and Authorisation of Chemicals). Article 13(1) of
the legal text of the draft REACH regulation states
that [3]: "Information on intrinsic properties of
substances may be generated by means other than
tests, in particular through the use of qualitative or
quantitative structure-activity relationship models
or from information from structurally related sub-
stances (grouping or read-across), provided that the
conditions set out in Annex XI are met".
REACH is expected to introduce a dramatic change in
the present EU regulatory schemes [4]. It will provide a
basis for the use of structure-activity relationships mod-
els, together with other "non-testing" approaches, for
predicting the environmental and lexicological proper-
ties of chemicals, in the interests of time-effectiveness,
cost-effectiveness and animal welfare. According to
an assessment carried out by the European Chemicals
Bureau (ECB), the in vivo mutagenicity studies, shortly
followed by carcinogenicity, are posing high demand
for test-related recourses [5,6].
In particular, the science of the relationships between
chemical structure and the biological activity of mol-
ecules is expected to play a new role and support three
distinct activities: category formation, "read-across",
and (Quantitative) Structure-Activity Relationships
((Q)SAR). A chemical category is a group of chemi-
cals whose physicochemical and human health and/or
environmental toxicological properties are likely to be
similar or follow a regular pattern as a result of struc-
tural similarity. If this similarity is recognized with suf-
ficient evidence, all the chemicals in the category can be
considered (and regulated) in the same way. Another
approach to fill data gaps is read-across. In the read-
across approach, endpoint information (e.g., carcino-
genicity) for one chemical is used to predict the same
endpoint for another chemical, which is considered to
be "similar" in some way (usually on the basis of struc-
tural similarity). Regarding the third approach, the sci-
entific foundation of (Q)SAR models lies in physical
organic chemistry, where features of a chemical and
its properties are used to estimate chemical behaviour
and activity solely from the knowledge of chemical
structure. (Q)SAR modeling has been widely used in
pharmacology, toxicology and physical chemistry [7],
and its capabilities and limitations are relatively well
understood [8-10]. Regarding the use of (Q)SAR, a
recent project supported by the European Chemicals
Bureau (ECB) surveyed the models for mutagenicity
and carcinogenicity in the public domain: the results
are summarized in [4] and [11].
The extensive use of estimation techniques such as
(Q)SARs, read-across and grouping of chemicals,
where appropriate and in a suitably constrained
context, has the potential to effect huge reductions
in use of animals for modeled toxicity endpoints. At
the same time, all these approaches need to be fed
by adequate amounts of good quality data and da-
tabases.
DATABASES OF CHEMICAL
MUTAGENS AND CARCINOGENS
IN THE PUBLIC DOMAIN
Among the sources of freely available data pertain-
ing to toxicity on chemical substances, one of the
principal resources is the TOXNET database of the
National Library of Medicine (NLM) (http://toxnet.
nlm.nih.gov/). TOXNET is a cluster of different da-
tabases, collecting information on toxicology, haz-
ardous chemicals, environmental health, and toxic
releases. From the website, it is possible to search
across and within the databases by several identifi-
ers, such as chemical name, CAS (Chemical Abstract
Service) number, molecular formula, classification
code, locator code, and structure or substructure
(with the CHEMID PLUS protocol). Among the
TOXNET databases, the Chemical Carcinogenesis
Research Information System (CCRIS) and the
GENE-TOX databases deal specifically with muta-
genicity and carcinogenicity data.
CCRIS contains over 8000 chemical records with
animal carcinogenicity, mutagenicity, tumor promo-
tion, and tumor inhibition test results provided by
the National Cancer Institute (NCI). Test results
have been reviewed by experts and all the records
are written in a standardized textual format.
GENE-TOX was developed by the US Environ-
mental Protection Agency (USEPA) and contains
genetic toxicology (mutagenicity) test data, result-
ing from expert peer review of the open scientific
literature, on over 3000 chemicals. The GENE-TOX
program was established within EPA to select assay
systems for evaluation, review data in the scientific
literature, and recommend proper testing protocols
and evaluation procedures for these systems.
Another repository of experimental carcinogenicity
data available on the web is the Carcinogenic Potency
Database (CPDB) (http://potency.berkeley.edu/cpdb.
html). This database collects the results from over
6000 chronic, long-term animal cancer bioassays on
over 1500 chemicals published in the general literature
through 1997 and by the National Cancer Institute/
National Toxicology Program through 1998. CPDB
is organized alphabetically by chemical name. All
experiments of a chemical are listed under the name
of the test agent; for each experiment, information
is included on test animals, features of experimental
protocol, and carcinogenicity results in detail, includ-
ing literature citation. CPDB is downloadable in pdf,
xls or txt formats, and searchable by chemical name,
CAS number, or author. Most recently, chemical-spe-
cific summary data pages have been provided on the
CPDB website to make these data more accessible
through chemical or structure searching (see, e.g., the
result of a search on acetaldehyde: http://potency.ber-
keley.edu/chempages/ACETALDEHYDE.html).
The US National Toxicology Program (NTP) makes
available on the web (http://ntp.niehs.nih.gov/) data
from more than 500 long-term toxicology and carcino-
genesis bioassays collected by the NTP and its pred-
ecessor, the National Cancer Institute's Carcinogenesis
Previous
-------
50
Romualdo Benigni, Cecilia Bossa, Ann M. Richard, et al.
Testing Program, and organized in a database at the
National Institute of Environmental Health Sciences
(NIEHS). These data can be accessed as technical
reports; the user can browse them directly or make
text searches (by chemical name or CAS number, for
example), or download the reports in pdf format. In
addition, detailed experimental study data, to the level
of individual animal observations, are housed in an
Oracle NTP on-line database, with limited searchable
access to detailed data on thousands of experiments
provided to the public on the NTP website.
To enhance their structure-searchability and use in
modeling applications, both the CPDB and the on-
line NTP database have been "chemically-indexed"
by the USEPA's National Center for Computational
Toxicology DSSTox (Distributed Structure-Searchable
Toxicity) database project (www.epa.gov/ncct/dsstox/),
which emphasizes quality procedures for accurate and
consistent chemical structure annotation of toxico-
logical experiments. Chemical structures and summary
mutagenicity and carcinogenicity data have been pub-
lished for the entire CPDB inventory (www.epa.gov/
ncct/dsstox/sdf_cpdbas.html; recently updated), along
with the URL address locating the specific chemical
data webpage on the CPDB website provided for each
indexed chemical substance. Chemical structures and
indicators of data availability (1 = yes, 0 = no) have also
been provided for the entire chemical inventory of the
online NTP database, for each of the 4 main NTP study
areas (Developmental, Immunological, Genetox, and
Chronic Cancer Bioassays) (see below for more infor-
mation on the DSSTox project).
From the International Agency for Research on Can-
cer (IARC) website it is possible to access the IARC
Monographs on the Evaluation of Carcinogenic Risks
to Humans (www-cie.iarc.fr/). In these documents,
independent assessments by international experts of
the carcinogenic risks to humans posed by a variety
of agents, mixtures and exposures, are published. The
Monographs are searchable by key word, CAS number,
synonym or chemical name.
Recently, a very useful tool that is expanding access
to a wide range of toxicological databases, as well as
other public biological activity databases available
on the web has been created by the National Center
for Biotechnology Information (NCBI) through
the PubChem project (http://pubchem.ncbi.nlm.
nih.gov). PubChem is a public information system
(tightly integrated into the cluster of biological
and literature databases hosted at NCBI, such as
PubMed http://www.ncbi.nih.gov/entrez/query.fcgi)
that links chemical identifiers (such as chemical
name, CAS number and chemical structures) to bio-
logical activity knowledge of substances. It should
be remarked that PubChem is not an independently
curated database, but rather a user-depositor system
that aggregates standardized data from many sourc-
es, providing a tool to interrogate databases in the
public domain in the US (including both toxicologi-
cal and biomedical ones). The PubChem interfaces
provide extensive query capabilities on textual and
numeric information, as well as a comprehensive set
of structure-based query methodologies. PubChem
was originally created to house all the bioassay data
of the NIH Molecular Libraries Initiative Screening
program, whose goal is to process hundreds of thou-
sands of chemicals through up to several thousands
of high-throughput bioassay screens, using chem-
istry to probe biology at the fundamental cellular
and protein receptor level (http://nihroadmap.nih.
gov/molecularlibraries/). PubChem has expanded,
however, as a user-depositor public data repository,
housing large amounts of public bioassay data, in-
cluding the NLM TOXNET and USEPA DSSTox
inventories. PubChem has also significantly expand-
ed its tools and capabilities for analyzing chemicals
across bioactivity space, through summary activity
assignments (active or inactive, or a binned range
of activities).
Recent reviews [12-14] surveyed the current status
of public toxicity databases in terms of their diverse
content and structure, and provide a useful comple-
ment to the information summarized above.
NEW NEEDS AND NEW TOOLS:
CHEMICAL RELATIONAL DATABASES
Until recently, many existing public toxicity data-
bases have been constructed primarily as "look-up-
tables" of existing data, and most often did not con-
tain chemical structures. These databases typically
utilize chemical names (usually common or com-
mercial names) and CAS numbers which are non-
unique and commercially registered and, therefore,
unsuitable for a unique, public identifier. In addi-
tion, often the organization of the data follows that
of the literature on paper, and does not lend easily
itself to informatics implementation.
Recently, concepts and computer techniques that
originated from the structure-activity relationships
science have provided powerful tools to create new
types of databases, where the ability to retrieve data
is strongly improved both in qualitative and quan-
titative terms. In fact, whereas the indexing (iden-
tifier) elements in traditional databases, such as
names and CAS numbers, are non-unique, prone to
errors and devoid of intrinsic information, chemical
structure as a chemical identifier has universally un-
derstood meaning and scientific relevance. Chemical
structure and chemical concepts (e.g., reactive func-
tional groups, acidity, hydrophobicity, electrophilic
reactivity, free radical formation) provide a common
language and framework for exploring the similar-
ity among chemicals and the underlying chemical
reactivity bases for diverse toxicological outcomes.
Hence, chemical structure should be considered an
essential identifier and scientifically useful metric
for chemical toxicity databases. Effective linkage of
chemical toxicity data with chemical structure infor-
mation can facilitate and greatly enhance data gath-
ering and hypothesis generation in conjunction with
(Q)SAR modeling efforts [15].
Previous
-------
ISSCAN DATABASE
51
Thus, a crucial point is that of collecting and stand-
ardizing portions of the existent knowledge in a way
that allows: a) exploration across both chemical and
biological domains; and b) structure-searchability
through the data. These characteristics may be gained
when chemical structures and toxicity data are incor-
porated into what is termed a Chemical Relational
Database (CRD). CRD is a special type of relational
database whose main informational unit is a chemi-
cal structure and whose fields are attributes or data
associated with that chemical structure.
In order to be accessed with a CRD application,
the information has to be stored in specialized file
formats. Among them, Structure Data File (SDF)
format has become as the most widely used public
standard for exchange of structure/data informa-
tion on chemicals. SDF files are simple text files
that adhere to a strict format for representing mul-
tiple chemical structure records and associated
data fields. Each record in the file is composed of
a "structure" section where the 2D or 3D structure
of the molecule is represented as MOLfile format,
and a second section composed of numerical or text
data fields (Figure 1). Hence, SDF files are very ver-
satile: they can accommodate many types of data,
are easily edited and manipulated by programming
scripts, and could be easily ported to other types of
standard formats, such as the mark-up languages,
XML and CML (for further information on issues
related to chemical annotation [4, 15]).
SEARCHING CAPABILITY
OF CRD DATABASES
Even though simple, useful searches can be performed
with widely available informatics tools, such as XLS
(Excel-readable) files of chemicals with annotated tox-
icity and/or properties, where it is possible, eg., to re-
trieve substances within a predefined range of toxicity
values. However, coding the chemical structure in the
SDF file allows one to perform remarkably more com-
plex searches by using specialized CRD software pro-
grams: most commercially available CRD applications
provide substructure and functional group search fea-
tures, different algorithms for searching compounds
chemically similar to query ones (similarity search),
and text and data field search functions (for informa-
tion on commercial and public software applications,
see the DSSTox website: www.epa.gov/ncct/dsstox/
SDFViewerBrowserCRDs.html).
When the SDF file is imported into a CRD ap-
plication, it is possible to do structure/text/data rela-
tional searching across records in the database. All
these operations are collectively termed "data min-
ing" [14]. In Figure 2, as an example the substructure
searching results using aniline as query structure are
depicted. The result of the search consists of all
chemicals in a database (i.e., an SDF file) contain-
ing aniline as basic substructure. In this way, it is
possible to identify subsets of chemicals according
to any structural query (i.e., functional group, or
molecular substructure).
10 9 0
0.3600
-0.8880
0.71 6,0'''
-o.teSo
/'1800
/-1 .7920
/ -1 .0560
/ 1 .7960
0.3040
-1 .2040
4 10 1 0
4910
\
3. 8 1 0
3\ 420
2 N7 1 0
\
2 6\1 0
1 5 Ks 0
1 3 1 t>
1220
M END,---""
> ^Substance
^V
o ,.
0^980'
,-''{.3840
-0.4000
-1 .4600
1 .6400
0.7560
2.4760
-0.6200
-2.4800
-1.3160
000
000
000
000
000
000
000
»%0 0 0
"(N-Q.. 0
ID> (41) *
0.0000
0.0000
0.0000
0.0000
0.0000
0.0000
0.0000
0.0000
0.0000
0.0000
-~
c
c
c
c
H
H
H
H
H
H
T"V20
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
o "o"
0 0
0 0
0 0
0 0
0 0
0 0
0 0
0 0
0 0
""9.
0
0
0
0
0
0
0
0
0
. 0
0
>-sO
0
0
0
0
0
0
0
0
Ov
0
0
0
0
0
0
0
0 0
0 0
*0 0
b\o
0 0>
0 0
0 0
0 0
0 0
0 0
0 0
0 0
0 0
0 0
, 0 0
\0 0
\ 0
0\ 0
0 0
0 0
,
1
1
/
2D Structure
••„
\
,--
'*'
,''
/
s
!> (41)
(41)
(4.1.).'''
*'
• Text and data fields
Fig. 1 | A sample SDF file, contain-
ing both structural (top) and data
(bottom) information.
Previous
-------
Romualdo Benigni, Cecilia Bossa, Ann M. Richard, et al.
13 LeadScope Enterprise - Enterprise: gonzo:9 - secure
File Edit View Tools Help
New Project ^ New Query ^ Refine Query HP Structures .X Kernove Display Studies Display Known Drug Information
jSfsfgJIggjfljMBSaSjSiBliliig^^^^^^^^^^B
Structures j Spreadsheet | Graphs |
Fig. 2 Example of substructure searching in a database of diverse chemicals. All the chemicals including aniline as a substructure are
highlighted. The search was performed with the program LeadScope (LeadScope Inc., Ohio),
Another very useful feature with the addition of
visual analytic tools is the possibility of character-
izing a database by its component functional groups
or chemical classes. An example of this capability
is presented in Figure 3, by applying a CRD ap-
plication to the SDF file. The figure shows that the
chemicals in the database are divided into chemical
classes, and the frequency in each class is given. In
addition, it is possible to add colors to each class
bar, pointing visually to the abundance in each class
of the chemicals active and inactive for some select-
ed property (e.g., carcinogenicity). Visualization of
the retrieved data makes easier and more immediate
the understanding of the results of the query.
The above data mining capabilities can be expand-
ed to perform more complex searches, by formulat-
ing queries where specific combinations of struc-
tures, data and text (i.e., "chemical profiles") are
searched for in the database at the same time.
Another crucial operation that can be performed
on structural databases is that of calculating chemi-
cal similarity between pairs of chemicals [16]. Based
on the structural motifs in common to two chemi-
cals, the degree of similarity can be quantified on,
e.g., a 0 to 1 scale, and the resulting similarity value
can be used as supporting evidence in the process of
identifying categories of similar chemicals.
A more sophisticated use of data mining approaches
allowed by modern CRD applications is the identi-
fication of one or more common structural patterns
among groups of chemicals with similar character-
istics or profiles (e.g., toxicity). Such patterns, when
identified, can be used as predictive models to estimate
the toxicity of other chemicals, with similar structural
patterns [14].
THE DSSTOX DATABASE PROJECT
In view of the powerful opportunities provided
by the CRD technology, a major problem is that of
transforming the available databases according to
the new standards. A considerable progress is repre-
sented by PubChem that allows the user to browse
through the US public databases individually and
collectively according to structural criteria. However,
even though this design permits a user to explore
and download all or portions of the available infor-
mation, there is no quality review of the structural
inventory of PubChem in relation to bioassay data,
which come from a large number of user-depositors
Previous
-------
ISSCAN DATABASE
53
or sources with various levels of quality review ap-
plied to their data; hence, it is largely a "user-beware"
public resource. New initiatives are now being devel-
oped to address this concern in the world of toxicity
data. An example of project designed to provide the
user with self-contained data files that can be read-
ily incorporated into CRD and used freely is the
Distributed Structure-Searchable Toxicity (DSSTox)
Database Network, which is a project of the USEPA
(www.epa.gov/comptox/).
A primary objective of the DSSTox website (www.
epa.gov/ncct/dsstox) is to serve as a central com-
munity forum for publishing standard-format,
structure-annotated chemical toxicity data files for
open-access, public use, and for use in CRD appli-
cations. DSSTox efforts include the careful quality
annotation of chemical structures, standardization
and documentation of toxicity data in collaboration
with toxicity data experts, and open public access to
toxicity databases.
In the initial phase, data files were not structure-
searchable on the DSSTox web site itself, but the
data files could be downloaded in their entirety and
freely used. Since September 2007, a DSSTox struc-
ture-browser offered on the DSSTox website allows
structure/substructure/similarity-searching through
all DSSTox data file content, and can be additional-
ly accessed from off-site collaborators (e.g., CPDB,
EPA IRIS, NTP) for website searching through ei-
ther local content (e.g., just the content of the origi-
nator's website) or broader searching through the
DSSTox inventory and, soon to be added, providing
external links to PubChem.
At present, the DSSTox data file cluster includes
six separate databases: CPDBAS - Carcinogenic Po-
tency Project Summary Tables (Source, LS Gold,
CarcinogenicPotencyProject,UCBerkeley);DBPCAN
- EPA Disinfection By-products Carcinogenicity Es-
timates Database (Source, YT Woo, USEPA, Office
of Pollution Prevention & Toxics); EPAFHM - EPA
Fathead Minnow Acute Toxicity Database (Source,
C. Russom, USEPA, Mid-Continental Ecology Di-
vision-Duluth); NCTRER - FDA NCTR Estrogen
Receptor Binding Database (Source, Weida Tong
and Hong Fang, National Center for Toxicological
Research, Jefferson, Arkansas); FDAMDD - FDA
Maximum Recommended Daily Dose (Source, Edwin
Matthews and R. Daniel Benz, US FDA, Rockville,
MD), and the newest data file, IRISTR (Source,
USEPA's Integrated Risk Information System Toxicity
Reviews), which includes 34 toxicity-related content
fields. Additionally, the DSSTox file inventory includes
2 structure-locator files, HPVCSI (USEPA's High
Production Volume Challenge Program) and NTPBSI
63 Projtet Browstr
F*J E* ¥i»¥ To* Wir»dw» Heip
Feature Comttnattoe Oustere Sesrttotes R-6ron» ft*J MaM
HMogriwi | scatterpM i Fealures
y*jfe
jpfoieet Hetarchv v
- FundjoraM groups •">
* (K"d «Kiy»«Jf ' |
serf halrdf j
* tieohd
* SM-iyd*
ffi *ene
66 **yn* Features
amMme
tfi mufms
iS ands :
Sis bwonyoupt 1
» carbanwte ,
'« cartjonyl
* catboxarwte i
Js carboxyWe
* cwboxyte aatl i
* ether ]
'*> jusnrfne 1
* hafcte
,.
A, hv^azrie
< >
IF*** [ \ Q
tes 0
Fitwed
72?
•^SX 2
1
150
13
129
3
1
1 ! " 238
2
1
14
258
.. ,, . 60
54
35
109
3
1S7
) 10 100 IK
X- logartNwc v
$5
Ida)
727;
21
1
150!
13
»29[
3
li
238
I
1
14
258
eo!
54
35
109]
3
187 1
SS
Flare: is CoweeT"!
Frequency
/"
^
Cetartor. £a«e v
^m ,«. ,:BH
tt .3,0 -25 ,2-0 29 2S 3,0 >»
Ready
Sets57
To»al774
Mean 2 55
Dev_079 Ftere*774
Selected 0
Fig. 3 Example of classification of the chemicals in a database by chemical classes. The analysis was performed with the program
LeadScope (LeadScope Inc., Ohio).
Previous
-------
Romualdo Benigni, Cecilia Bossa, Ann M. Richard, et al.
(National Toxicology Program Bioassay) contain-
ing URL addresses to chemical-specific data pages,
and 2 structure-index files containing only a chemi-
cal structure listing, NTPHTS (National Toxicology
Program High-Throughput Screening) and TOXCST
(EPA's National Center for Computational Toxicology
ToxCast testing program).
Each DSSTox database is published as a separate
and distinct module that adheres to standard con-
ventions in SDF data file format, file names, chemi-
cal structure fields, and minimum documentation re-
quirements. Together with the SDF file, the DSSTox
provides an MS Excel-readable file (.xls) (reporting
the non-structural data), and an Acrobat-readable
file (.pdf) which displays the traditional graphical
representation of the chemicals. In addition, the
DSSTox website provides a detailed guide on the use
of files, and a rich documentation on the entire sub-
ject of databases and related concepts [12, 17]. The
collected DSSTox published inventory contains over
six thousand unique chemical substances relevant to
toxicology and can be merged for structure-search-
ing, or ported into CRD applications.
THE ISSCAN DATABASE
ON CHEMICAL CARCINOGENS
As pointed out above, currently the public has
access to a variety of toxicity databases; however,
these publicly available data may not be immedi-
ately suitable for use. One general issue is that of
data quality, both from a chemical and biological
perspective. Beyond its most obvious meaning (data
"must" be of good quality, otherwise any inference
based on them is simply devoid of any value), there
are more subtle problems linked to this issue. For ex-
ample, for each chemical the CCRIS (as well as the
CPDB) reports all the available experimental results.
There are cases where more than one experiment,
with contradictory results, exist for a given chemi-
cal. There are also cases where the experimental pro-
tocols differ to a large extent. In all these cases, the
database user has to employ her/his expert judge-
ment to make an activity assignment. Together with
the data issue and linked to it, is that of the data
standardization, which can become extremely criti-
cal for some more formalized applications, such as
QSAR analyses [9]. These approaches need highly
summarized representations of the activity of the
chemicals (i. e., a unique number for the potency of
the active compounds; a dichotomous classification
into actives/inactives). But the large public databas-
es often do not meet these modeling requirements.
One example is the NTP on-line database that in-
cludes high-level detail on animal bioassays and ge-
netic toxicity experiments for several thousands of
chemicals, respectively, but which does not provide
ready access to data for the entire chemical study
inventory, relational access to particular slices of the
data, or aggregate summarizations of the data ac-
cording to the requirements of QSAR modeling.
To alleviate the above problems, at the Istituto Superi-
ore di Sanita (ISS) a new database on chemical carcino-
gens called ISSCAN: "Chemical carcinogens: structures
and experimental data" has been built. The data can be
freely downloaded from the ISS website: www.iss.it/
ampp/dati/cont.php?id=233&lang=l&tipo=7 or from
the DSSTox site: wwwepa.gov/ncct/dsstox/sdf_isscan_
external.html.
The ISSCAN database contains information on
chemical compounds tested with the long-term car-
cinogenicity bioassay on rodents (rat, mouse). The
specific characteristics of the ISSCAN database in re-
spect to other databases should be emphasized. First,
the ISSCAN initiative is aimed at providing the scien-
tific and regulatory community with carcinogenicity
calls that have been re-checked, in order to ensure the
quality of the data. The data were cross-checked on
different sources of information available; contradic-
tions were solved going back to the original papers,
and results based on insufficient protocols were not
included. Second, the biological data (carcinogenicity
and Salmonella mutagenicity) were coded in numeri-
cal terms that can be used directly for QSAR analy-
ses. This aspect of being QSAR-ready eliminates the
intermediate passage of data transformation that of-
ten is problematic for the QSAR practitioner without
specific toxicological expertise.
The general structure of the database is inspired
by that of the DSSTox. The ISSCAN database is
composed of standard chemical data fields, such
as 2D structure, chemical name and synonyms,
CAS registry number, molecular weight, chemical
formula and SMILES notation, together with bio-
logical data fields: carcinogenic potency in rat and
mouse, mutagenicity in Salmonella typhimurium
(Ames test), carcinogenicity results in the four ex-
perimental groups most commonly used for the can-
cer bioassay, carcinogenicity results from the NTP
experimentation (when available), overall carcino-
genicity, together with the source of carcinogenicity
data. Figure 4 displays the information reported by
ISSCAN for a representative chemical.
From the website it is possible to download four
different files:
1) an SDF file containing chemical structures to-
gether with chemical and biological data;
2) a PDF file with a detailed explanation and guid-
ance of use;
3) a PDF file with 2D chemical structures of the
substances;
4) an XLS file of the data.
At present, the second updated version of ISSCAN
is available, including 890 chemicals tested for ro-
dent carcinogenicity (the main primary sources of
data are the NTP, CPDB, CCRIS, and IARC re-
positories). It is our plan to accomplish the evalu-
ation of the remaining chemicals by the year 2008.
Since the SDF file cannot be read by users without
specialized software applications, it is also our plan
to make available on our website a tool suitable for
simple analyses.
Previous
-------
ISSCAN DATABASE
55
It should be emphasized that this type of project
(ISSCAN) is not in opposition to other databases
(e.g., CCRIS, CPDB) that follow the philosophy of
reporting vast amounts of data at different hierar-
chical levels, also including contradictory evidence
when existing. In contrast and complementary to
these efforts, the ISSCAN initiative is aimed at pro-
viding the end-user with information that is revised
and re-organized for a specific aim, whereas the
above databases have the important role of keeping
track of all the available information. Even when
the knowledge contribution of portions of such
databases looks very minor (e.g., data from experi-
ments with few animals and old protocols), this - in
a different context - may turn out to be very useful
for, e.g., planning further studies.
CONCLUSIONS
The key to a rapid progress in the field of chemical
toxicity databases exploitation is that of combining
information technology with the chemical structure
as identifier of the molecules. This permits an enor-
mous range of operations (e.g., retrieving chemicals
or chemical classes, describing the content of data-
bases, finding similar chemicals, crossing biological
and chemical interrogations, etc.) that other more
classical databases cannot allow. In the foreseeable
future, this trend will become even more pervasive:
a clear demonstration of this trend is the creation by
NCBI of the chemically-interrogable PubChem da-
tabase fully integrated with the traditional, textual
PubMed (http://www.ncbi.nlm.nih.gov/sites/entrez)
repository of biomedical information. At the same
time, there is a proliferation of new tools aimed at
Formula
FW
Substance ID
Mouse Female Cane
SAL
Rat Male Cane
TD50_Rat
TD50_Mouse
Rat_Female_Canc
Cane
MolWeight
Mouse_Male_Canc
Mouse_Male_NTP
ChemName
Rat_Male_NTP
Reference
SMILE
Rat Female NTP
CAS
Mouse_Female_NTP
Synonyms
C15H13NO
223.2699
2
ND
3 HN'
ND \
NP /^\ >
ND 4. ff—
-------
56
Romualdo Benigni, Cecilia Bossa, Ann M. Richard, et al.
animal toxicity on a chosen set of compounds: the
standardization of data and CRD-accessibility will
be a necessary requirement in order to fully exploit
the value of these data (for more information, see:
www.epa.gov/ncct/toxcast/).
A cknowledgements
This work was partially granted by the EU FP6 Contract n.
037017 OSIRIS "Optimized strategies for risk assessment of in-
dustrial chemicals through Integration of non-test and test in-
formation"
Disclaimer
This manuscript does not necessarily reflect the views and policies
of the USEPA, nor does mention of trade names or commercial
products constitute endorsement or recommendation for use.
Submitted on invitation.
Accepted on 16 December 2007.
References
1. Organisation for Economic Co-operation and Development.
Report on the Regulatory Uses and Applications in OECD
Member Countries of (Quantitative) Structure-Activity Re-
lationship ((Q)SAR) Models in the Assessment of New and
Existing Chemicals. 58. 2006. OECD Series on Testing and
Assessment. Paris: OECD; 2006. (ENV Monograph No. 58).
2. Richard AM. Future of predictive toxicology. An expanded
view of "chemical toxicity", future of toxicology perspec-
tive. Chem Res Toxicol2006;19-.1257-62.
3. Commission of the European Communities. Proposal concern-
ing the registration, evaluation, authorisation and restriction of
chemicals (REACH). (COM(2003)644Final). Bruxelles: EU;
2003.
4. Benigni R, Netzeva TI, Benfenati E, Bossa C, Franke R,
Helma C, Hulzebos E, Marchant C, Richard A, Woo Y-T,
Yang C. The expanding role of predictive toxicology: an up-
date on the (Q)S AR models for mutagens and carcinogens. /
Environ Sci Health C 2007;25:53-97.
5. Pedersen F, de Brujin J, Munn SJ, and Van Leeuwen, K.
Assessment of additional testing needs under REACH. Effects
of (Q)SARs, risk based testing and voluntary industry initia-
tives. Ispra: Joint Research Centre; 2003. (JRC report EUR
20863 EN).
6. Van der Jagt K, Munn SJ, Torslov J, de Brujin J. Alternative
approaches can reduce the use of test animals under REACH.
Addendum to the Report "Assessment of addtional testing
needs under REACH. Effects of (Q)SARs, risk based test-
ing and voluntary industry initiatives". Ispra: Joint Research
Centre; 2004. (JRC Report EUR 21405 EN).
7. Hansch C, Leo A. Exploring QSAR 1. Fundamentals and ap-
plications in chemistry andbiology. Washington DC: American
Chemical Society; 1995.
8. Hansch C, Hoekman D, Leo A, Weininger D, Selassie CD.
Chem-bioinformatics: comparative QSAR at the interface be-
tween chemistry and biology. Chem Rev 2002;102:783-812.
9. Franke R, Gruska A. General introduction to QSAR. In:
10.
11
Benigni R (Ed.). Quantitative structure-activity relationhsip
(QSAR) models of mutagens and carcinogens. Boca Raton:
CRC Press; 2003. p. 1-40.
Benigni R. Structure-activity relationship studies of chemi-
cal mutagens and carcinogens: mechanistic investigations
and prediction approaches. Chem Rev 2005;105:1767-800.
Benigni R, Bossa C, Netzeva TI, Worth AP. Collection and
evaluation of ( Q)SAR models for mutagenicity and carcino-
genicity. Office for the Official Publications of the European
Communities. EUR - Scientific and Technical Research Series.
Luxenbourg; 2007. (EUR 22772 EN). Available from: http://
ecb.jrc.it/documents/QSAR/EUR_22772_EN.pdf; last vis-
ited 21/11/2007.
12
Richard AM, Williams CR. Public sources of mutagenicity
and carcinogenicity data: use in structure-activity relation-
ship models. In: Benigni R (Ed.). Quantitative Structure-
Activity Relationship (QSAR) models of mutagens and car-
cinogens. Boca Raton: CRC Press; 2003. p. 145-74.
Yang C, Benz RD, Cheeseman MA. Landscape of current
toxicity databases and database standards. Curr Opinion Drug
Discov Develop 2006;9:124-33.
Yang C, Richard AM, Cross KP. The art of data mining the
minefields of toxicity databases to link chemistry to biology.
Curr Comput Aid Drug Des2006;2:l35-50.
Richard AM, Gold LS, Nicklaus MC. Chemical structure in-
dexing of toxicity data on the Internet: moving toward a flat
world. Curr Opinion Drug Discov Develop 2006;9:314-25.
Gallegos Saliner A. Mini-review on chemical similarity and pre-
diction of toxicity. Curr Comput Aid Drug Des2006;2:l05-22.
17. Richard AM. DSSTox web site launch: improving public ac-
cess to databases for building structure-toxicity prediction
models. Preclinica 2004;2:103-8.
18. Dix DJ, Houck KA, Martin MT, Richard AM, Setzer MW,
Kavlock RJ. The ToxCast program for prioritizing toxicity
testing of environmental chemicals. Toxicol Sci2007;95:5-l2.
13
14
15
16
Previous
-------
Research
A Novel Two-Step Hierarchical Quantitative Structure-Activity Relationship
Modeling Work Flow for Predicting Acute Toxicity of Chemicals in Rodents
Hao Zhu,1 Lin Ye,1 Ann Richard,2 Alexander Golbraikh,1 Fred A. Wright,3 Ivan Rusyn,4'" and Alexander Tropsha1'"
laboratory for Molecular Modeling, Division of Medicinal Chemistry and Natural Products, School of Pharmacy, University of North
Carolina at Chapel Hill, Chapel Hill, North Carolina, USA; 2National Center for Computational Toxicology, Office of Research and
Development, U.S. Environmental Protection Agency, Research Triangle Park, North Carolina, USA; 3Department of Biostatistics,
and 4Department of Environmental Sciences and Engineering, School of Public Health, University of North Carolina at Chapel Hill,
Chapel Hill, North Carolina, USA
BACKGROUND: Accurate prediction of in vivo toxicity from in vitro testing is a challenging problem.
Large public—private consortia have been formed with the goal of improving chemical safety assess-
ment by the means of high-throughput screening.
OBJECTIVE: A wealth of available biological data requires new computational approaches to link
chemical structure, in vitro data, and potential adverse health effects.
METHODS AND RESULTS: A database containing experimental cytotoxicity values for in vitro half-
maximal inhibitory concentration (IC5o) and in vivo rodent median lethal dose (LD5o) for more
than 300 chemicals was compiled by Zentralstelle zur Erfassung und Bewertung von Ersatz- und
Ergaenzungsmethoden zum Tierversuch (ZEBET; National Center for Documentation and
Evaluation of Alternative Methods to Animal Experiments). The application of conventional
quantitative structure—activity relationship (QSAR) modeling approaches to predict mouse or rat
acute LD5o values from chemical descriptors of ZEBET compounds yielded no statistically signifi-
cant models. The analysis of these data showed no significant correlation between IC^Q and LD^Q.
However, a linear IC5Q versus LD50 correlation could be established for a fraction of compounds.
To capitalize on this observation, we developed a novel two-step modeling approach as follows.
First, all chemicals are partitioned into two groups based on the relationship between IC5o and
LD5o values: One group comprises compounds with linear IC5o versus LD5o relationships, and
another group comprises the remaining compounds. Second, we built conventional binary clas-
sification QSAR models to predict the group affiliation based on chemical descriptors only. Third,
we developed ^-nearest neighbor continuous QSAR models for each subclass to predict LD50 values
from chemical descriptors. All models were extensively validated using special protocols.
CONCLUSIONS: The novelty of this modeling approach is that it uses the relationships between in vivo
and in vitro data only to inform the initial construction of the hierarchical two-step QSAR models.
Models resulting from this approach employ chemical descriptors only for external prediction of
acute rodent toxicity.
KEY WORDS: acute toxicity, computational toxicology, IC50, LD50, LOAEL, NOAEL, QSAR.
Environ Health Perspect 117:1257-1264 (2009). doi:10.1289/ehp.0800471 available via tittp://
dx.etoi.org/ [Online 3 April 2009]
Development of accurate and predictive
in vitro toxicity testing methods that could
be used as alternatives for lengthy and costly
in vivo experiments has long been an elusive
goal for both industry and regulatory agencies
(National Research Council 2007). New, bold
research programs were recently established at
the National Toxicology Program (Xia et al.
2008) and the U.S. Environmental Protection
Agency (U.S. EPA) (Dix et al. 2007) and
coordinated at the interagency level by the
U.S. government (Collins et al. 2008) to
address this important challenge in a system-
atic way. The overall goal of these initiatives
is to explore a diverse array of in vitro toxicity
assays, such as cell-based and cell-free high-
throughput screening (HTS) techniques, as
well as toxicogenomic technologies, to evaluate
the toxic potential of chemicals and prioritize
candidates for animal testing. However, the
utility of in vitro data as indicators of in vivo
effects will be fully realized only if rigorous
correlation between the toxicity of chemi-
cals in vitro and in vivo can be established
(National Research Council 2007; Rabinowitz
et al. 2008).
Many previous studies have indicated that
the correlation between the in vitro toxicity
results and animal toxicity test data (e.g., acute,
subacute, subchronic, and chronic rodent tox-
icity test results) is generally poor. Most nota-
bly, in 2001, the Interagency Coordinating
Committee on the Validation of Alternative
Methods (ICCVAM) hosted a workshop to
assess the relationship between cytotoxicity
and rodent acute toxicity for > 300 diverse
compounds; the data were compiled by the
Zentralstelle zur Erfassung und Bewertung
von Ersatz-und Ergaenzungsmethoden zum
Tierversuch (ZEBET; the National Center for
Documentation and Evaluation of Alternative
Methods to Animal Experiments) [ICCVAM
and National Toxicology Program Interagency
Center for the Evaluation of Alternative
Toxicological Methods (NICEATM) 2001].
It was concluded that there is no clear cor-
relation between cytotoxicity [half-maximal
inhibitory concentration (ICjo)] and acute
toxicity [median lethal dose (LD50)] data in
rodents. Similarly, poor correlation was found
between in vitro cytotoxicity and in vivo
rodent carcinogenicity, even when a diverse
set of in vitro end points from HTS was used
(Xia et al. 2008; Zhu et al. 2008).
Cheminformatics approaches such as quan-
titative structure-activity relationship (QSAR)
modeling have been widely used in toxicology
(Dearden 2003; Johnson et al. 2004). Several
software packages, such as Toxicity Prediction
by Komputer Assisted Technology (TOPKAT)
(Venkatapathy et al. 2004) and Multiple
Computer-Automated Structure Evaluation
(MultiCASE) (Matthews et al. 2006), have
been developed and actively used by both
industry and regulatory agencies. However,
existing modeling tools generally do not achieve
good external accuracy of prediction for com-
pounds not used in model development, and
few QSAR models have been successful in pre-
dicting in vivo toxicity end points for diverse
sets of environmental compounds (Benigni
et al. 2007; Stouch et al. 2003).
Address correspondence to A. Tropsha, 327 Beard
Hall, University of North Carolina, Chapel Hill,
NC 27599-7360 USA. Telephone: (919) 966-2955.
Fax: (919) 966-0204. E-mail: alex_tropsha@unc.edu
*These authors contributed equally to this work.
Supplemental Material is available online
(doi: 10.1289/ehp.0800471.SI viahttp://dx.doi.org/)
We thank T. Martin [U.S. Environmental
Protection Agency (EPA)], and J. Strickland and
M. Jackson (ILS, Inc., Durham, NC) for providing
some of the data used in this study. We also thank
W. Setzer (U.S. EPA) for his interest in this study
and valuable comments on the moving M-regression
method.
This work was supported, in part, by grants from
the National Institutes of Health (GM076059 and
ES005948) and the U.S. EPA (RD83272001 and
RD83382501).
The research described in this article has not been
subjected to each funding agency's peer review
and policy review and therefore does not neces-
sarily reflect their views, and no official endorse-
ment should be inferred. The manuscript has been
reviewed by the U.S. EPA National Center for
Computational Toxicology and approved for pub-
lication. Approval does not signify that the con-
tents necessarily reflect the views and policies of the
agency, nor does mention of trade names or com-
mercial products constitute endorsement or recom-
mendation for use.
The authors declare they have no competing
financial interests.
Received 26 November 2008; accepted 3 April 2009.
Environmental Health Perspectives • VOLUME 117 I NUMBER 81 August 2009
Previous
1257
-------
Zhu etal.
There are several possible reasons that previ-
ous attempts to establish relationships between
in vitro and in vivo toxicity data were largely
ineffective. These include, among other fac-
tors, inadequate attention paid to the chemical
diversity of the compounds used for screening
and modeling and, consequently, unjustified
confidence in the ability of models to extrapo-
late significantly outside the chemistry space
of the training set. Furthermore, the conven-
tional QSAR modeling efforts have been dis-
connected from the growing efforts to employ
in vitro screening (i.e., HTS data) to predict
in vivo outcomes. Recently, we have proposed
the use of hybrid chemical—biological descrip-
tors, that is, a combination of conventional
chemical descriptors with HTS profile data
regarded as biological descriptors. We have
demonstrated that these hybrid descriptors
afford QSAR models with significantly higher
accuracy of prediction of rodent carcino-
genicity versus models using chemical descrip-
tors alone, and much higher accuracy versus
models that used biological in vitro data alone
(Zhuetal. 2008).
These recent studies suggest that the
explicit consideration of chemical structure (in
the form of chemical descriptors) along with
in vitro assay data could potentially account
for discrepancies between in vitro and in vivo
results and produce more accurate predictive
models of in vivo toxicity. To validate this
hypothesis further, in this study we used the
ZEBET data set (ICCVAM and NICEATM
2001) for which previous attempts to establish
the direct in vitrolin vivo correlation proved
largely unsuccessful (Freidig et al. 2007). We
have observed that chemicals can be partitioned
into two classes based on comparison between
cytotoxicity and acute toxicity data: a) those for
which the linear in vitro (lC$o)/in vivo (LDjo)
correlation could be demonstrated and K) those
that correlate poorly. Furthermore, and of cen-
tral importance for applying our models to the
external set of chemicals for which no in vitro
data exist, we have built binary QSAR models
that could discriminate between compounds in
these two classes with reasonable accuracy based
on their chemical features alone. Finally, we
have established rigorous and externally predic-
tive class-specific QSAR models of rodent acute
toxicity measured by LD50 values. We show
that a two-step hierarchical QSAR modeling
work flow where compounds are first assigned
to a class using binary QSAR models and
then their LDj0 values is predicted using class-
specific continuous QSAR models affords accu-
rate prediction of LD50 values for compounds
not included in the training set. In addition, we
show that this two-step model's statistical pre-
diction accuracy compares favorably with cur-
rently available commercial toxicity predictors.
Our studies suggest that the two-step QSAR
modeling work flow can improve performance
of predictive acute toxicity models for diverse
organic compounds and aid in prioritizing
compounds for rodent toxicity testing.
Materials and Methods
Data sets. The ZEBET database consists of
data for 361 chemicals compiled from litera-
ture studies and published in a consolidated
ICCVAM report (ICCVAM and NICEATM
2001). Every compound in this data set has
at least one cytotoxicity result (IC50) and at
least one type of rodent acute toxicity value
(rat or mouse LD50). We defined ZEBET
criteria to select cytotoxicity data for this data
set as follows: a) at least two different ICjQ
values were available, either from different
cell types or from different cytotoxicity end
points; V) cytotoxicity data were obtained
with mammalian cells; c) cytotoxicity data
obtained with hepatocytes were not accept-
able; and d) chemical exposure time in
the cytotoxicity tests was at least 16 hr.
Furthermore, only the results obtained from
the following cytotoxicity tests were accepted:
a) cell proliferation measured by cell num-
ber, protein, DNA content, DNA synthe-
sis, or colony formation; b) cell viability and
metabolic indicators, including metabolic
inhibition test (MIT-24), 3-(4,5-dimethyl-
thiazol-2-yl)-2,5-diphenyltetrazolium
bromide (MTT) assay, 3-(4,5-dimethyl-
thiazol-2-yl)-5-(3-carboxymethoxyphenyl)-2-
(4-sulfophenyl)-2H-tetrazolium (MTS) assay,
and sodium 3,3-(l-[(phenylamino)carbonyl]-
3,4-tetrazolium)-bis(4-emthoxy-6-nitro)ben-
zene sulfonic acid hydrate (XTTC); r) cell
viability and membrane indicators, including
neutral red uptake, trypan blue exclusion, cell
attachment, and cell detachment; and d) dif-
ferentiation indicators.
For the purpose of this work, we
curated the data set to select the subset of
organic compounds and excluded inorganic
and organometallic compounds, as well as
compound mixtures, because conventional
chemical descriptors used in QSAR studies
could not be computed in these cases. There
were 254 and 235 compounds that had rat
or mouse LDjQ (millimole/kilogram-body
weight/day) values, respectively. Only LD50
values published in the Registry of Toxic
Effects of Chemical Substances (RTECS)
(Norager et al. 1978; Ruden and Hansson
2003) were used. The distributions of log(l/
LDjo) values of ZEBET compounds, with
the exception of a single outlier, were from
-2.61 to 2.30 for the rat and from -2.50 to
2.19 for the mouse. We considered one com-
pound, 2,3,7,8-tetrachlorodibenzo-^>-dioxin
(CAS 1746-01-6), an activity outlier because
its log(l/LD50) value was -4.21 for the rat,
which deviated significantly from the activ-
ity range of the data set. After excluding this
single outlier, the data sets used for modeling
consisted of 253 compounds for the rat and
235 compounds for the mouse. An additional
set of 115 compounds with complete data
(both LD50 and 1050) for the rat was recently
released by ICCVAM, which we used for vali-
dation (referred to as the ICCVAM data set).
[For raw data, see Supplemental Material,
Table 1 (doi:10.1289/ehp.0800471.Sl).]
The data on rat chronic lowest observed
adverse effect levels (LOAELs) and rat chronic
no observed adverse effect levels (NOAELs)
were compiled from an internal low-dose
toxicity data set established in our labora-
tory [see Supplemental Material, Table 2
(doi:10.1289/ehp.0800471.SI)]. These data
include a combination of multiple toxicity
phenotypes, such as liver toxicity and kidney
toxicity. Compared with the ZEBET data set,
42 unique compounds have both rat LOAEL
and in vitro IC50 values, and 41 compounds
have both NOAEL and in vitro ICjQ values.
Because of limited availability of LOAEL and
NOAEL data, we used these two data sets
only to illustrate the data partitioning algo-
rithm and did not build any QSAR models
for them.
QSAR modeling approaches. We used
the /^-nearest neighbor (£NN) QSAR model-
ing approach that has been developed in our
group (Zheng and Tropsha 2000). In brief,
the method is based on the £NN principle and
the variable selection procedure. It employs
the leave-one-out cross-validation procedure
(LOO-CV) and a simulated-annealing algo-
rithm for the variable selection. The proce-
dure starts with the random selection of a
predefined number of descriptors from all
descriptors. If £NN > 1, the estimated activi-
ties yt of compounds excluded by the LOO
procedure are calculated using the following
formula:
[1]
k- 1
where y: is the activity of the ^'th compound.
We define weights w~ as
U>jj = 1 — '
[2]
where d~ is the Euclidean distance between
compound i and its ^'th nearest neighbor.
Further details of the algorithms and work
flow are provided elsewhere (Medina-Franco
et al. 2005; Ng et al. 2004; Shen et al. 2002;
Zheng and Tropsha 2000).
We developed rat and mouse LDj0
QSAR models for ZEBET compounds using
DRAGON chemical descriptors (DRAGON
1258
VOLUME 1171 NUMBER 8 I August 2009 • Environmental Health Perspectives
Previous
-------
Two-step QSAR for acute rodent toxicity prediction
for Windows, version 5.4; Teleste s.r.l.,
Milan, Italy). Before model construction,
23 compounds with rat LD50 results and 24
compounds with mouse LD50 results were
selected at random to serve as external vali-
dation sets. The remaining 230 rat and 211
mouse compounds were used as modeling
sets, and each was divided multiple times
into training/test sets using the sphere exclu-
sion approach (Golbraikh et al. 2003). We
characterized the statistical significance of the
models with the standard LOO-CV R2 (q1)
for the training sets and the conventional R2
for the test sets when modeling real values
(i.e., continuous QSAR). For classification
modeling, we used correct classification rates
expressed as a fractional value between 0 and
1. The model acceptability cutoff values of
the LOO-CV accuracy of the training sets
and the prediction accuracy for test sets were
both set to 0.65 for classification models. For
continuous models, the acceptability thresh-
olds for LOO-CV regression q for the train-
ing sets and R2 values for the test set were
both set at 0.5. Models that did not meet
both training and test set cutoff criteria were
discarded.
Moving M-regression for data partitioning,
We used a novel approach related to a class
of M-regression methods (Andersen 2007),
which we termed "moving M-regression," to
select compounds for which there is a strong
correlation between IC50 and LD50 values
(class 1). The approach is a variant of the least
squares regression that takes into account
only those data points contained within a
band around the regression line jyregr = ax
+ b. For each y, only the points within the
interval \y — dt, y + d,] are candidates for class
1, whereas points outside of this band are
excluded from class 1. If the line y = ax + b
is moved, some new points will enter the
band, whereas some other points will leave
it, which may result in a higher regression R2;
this also explains why we use the term "mov-
ing M-regression." For each point, we define
{xj, yt}, «'=!,...,», the moving M-regression
inclusion function, as
1, if y± e [ax; + b-
0, otherwise.
+ b +
[3]
Thus, the moving M-regression line can be
found by minimizing the following expression:
n
F(a,b) = X T\(xi>yi)(yi - oxi - b)2• M
i= 1
Function _Fis not differentiable at all points
(XP yj) such that yt = axj + b — d\ and jy, = /ay +
b + d2. For practical purposes, we approximate
T|(A;, jy,) by sums of two sigmoid functions:
where PI and PI are large (- 100) positive
parameters. Indeed, as PI and P2 approach
infinity, the expression on the right side
of Equation 5 approaches the right side of
Equation 3. Small approximation errors in
the vicinity of points {axj + b — d-[, y,} and {/ay
+ b + d2, yi\ approach zero as both PI and PI
approach infinity. It is as if the data points are
gradually included within, or excluded from,
the band when the regression line is mov-
ing. Finally, replacing T|(:xy, yj) by Equation 3,
we obtain
F(a,b) =
[6]
To optimize F(il, b), the following system
of equations is to be solved:
dF
[7]
Equations 7 are nonlinear, so depending
on the data set and parameters PI, P2, d\, and
d2, they can have multiple solutions (a, b).
In these studies, Ay and jy, were the in vitro
log(l/IC5o) and in vivo log(l/LD5o) values,
respectively, for a data set of compounds
under study. Instead of using Equation 6,
Whole
in vitro/in vivo
dataset
r
we determined the compounds that belong
to class 1 by maximizing the number of data
points within the band. With this correction,
our target function takes the form
F*(a,b) =
l+exp[/>2(.y,.
[8]
To obtain the baseline toxicity regression
(see Results), we opted to minimize the num-
ber of outliers below the regression line. For
this purpose, we added additional terms for
the lower border of the band weighted by an
arbitrary parameter a. Thus, we minimized
the following target function:
F**(a,b) =
" < 1 _i_ ™
£ | 1 + a
[9]
The initial point (a, b) for minimization
of F** was selected manually. PI and PI were
equal to 100, d\ and d2 were equal to 0.4,
and a was equal to 1. To optimize Equation
9, the system of Equations 7 in which F is
replaced by F** should be solved. Figure 1
summarizes the data analytical work flow that
we employed in this study for rodent acute
toxicity modeling.
Model validation. We validated training
set models by evaluating their external predic-
tive power on the test sets as described above.
Compounds
above the
line
Generate regression
line between IC5(I and
LD5(I using moving
regression method
— h.
Compounds
on the line
Compounds
Figure I.The work flow of the two-step /rNN QSAR LD60 modeling.
Environmental Health Perspectives • VOLUME 117 I NUMBER 81 August 2009
Previous
1259
-------
Zhu etal.
Furthermore, a 5-fold external CV analysis
was performed for the original ZEBET data
set: the data set was randomly split into five
equal-size subsets of compounds and five
independent sets of calculations were con-
ducted each time using 80% of the whole data
set as a modeling set and the remaining 20%
compounds as a test set. In addition, robust-
ness of QSAR models was verified using a
Y-randomization (randomization of response)
approach as follows. We randomly divided
the modeling set compounds into class 1 and
class 2 subsets and developed £NN QSAR
LDjo models for each subset using the same
protocol and the same cutoff criteria (if and
R > 0.5) as for compounds in classes 1 and
2 that we generated by means of the moving
regression. The purpose of this was to see if
statistically significant QSAR models could
be obtained for any random division of the
original data into two classes. Independently,
we applied the test to compounds in unique
classes 1 and class 2 by randomizing their
LDjo values and redeveloping training set
models. Both Y-randomization tests were
repeated 10 times.
1 o
Results
Failure of conventional QSAR modeling of
rodent acute toxicity. The modeling set includ-
ing 230 compounds with known rat LD50
data was partitioned into 32 training and test
sets and the conventional £NN QSAR model-
ing approach was applied to all training sets
as detailed in "Materials and Methods." We
characterized each training set model by its
q2 value; for the five best training set models,
these values ranged between 0.5 and 0.57.
These five models were used for predicting
LDjo values for the respective external valida-
tion set (23 compounds). However, for each
of these models the R value for this exter-
nal set was < 0.5. When we used other types
of in-house or commercial QSAR methods
(e.g., support vector machine or partial least
square) and other types of descriptors (e.g.,
MolConnZ descriptors or molecular operating
environment descriptors), we obtained no sta-
tistically significant predictive QSAR models
(data not shown). Likewise, modeling of the
mouse data set (211 modeling compounds
and 24 external validation compounds) was
unsuccessful (data not shown). This negative
Log (1/IC5.
Log (1/IC5.
Figure 2. The identification of the baseline correlation between cytotoxicity (IC60) and various types of in vivo
toxicity testing results. (A) Rat LD50. (B) Mouse LD50. (C) Rat LOAEL. (D) Rat NOAEL. C1, class 1; C2, class 2.
Table 1. The results of data partitioning for the compounds with rat LD60, mouse LD60, rat chronic LOAEL,
and rat chronic NOAEL data in ZEBET data set using cytotoxicity IC50 values.
Model
Rat LD50 (original set)
Mouse LD50
Rat LOAEL
Rat NOAEL
Rat LD50 (full data set)
No. of C1 compounds
137
119
21
19
258
C1 ratio (%)
60
56
49
46
61
No. of C2 compounds
93
92
21
22
167
C2 ratio (%)
40
44
51
54
39
result corroborates the well-known inability of
conventional QSAR modeling approaches to
arrive at statistically significant and externally
predictive models of in vivo toxicity.
Data partitioning using the moving
M-regression approach. It is well known that
in vitro cytotoxicity correlates poorly with in vivo
toxicity end points when any relatively large set
of compounds is considered. The ZEBET data
set is no exception; cytotoxicity (IC50) correlates
with acute toxicity (LD50) for only a fraction
of the compounds in either the rat (Figure 2A)
or mouse (Figure 2B) data sets. Most of the
compounds are more toxic in vivo than in vitro.
Similar patterns could be found between cyto-
toxicity and other in vivo toxicity end points, for
example, rat chronic LOAEL and rat chronic
NOAEL (Figures 2C,D).
To devise a mathematical means for iden-
tifying compounds with strong in vitrolin vivo
correlation, we extended concepts that have
been previously employed in calculating the
"baseline regression" that correlated the aquatic
toxicity of (some) chemicals with the loga-
rithm of the «-octanol/water partition coef-
ficient (log P) (Klopman et al. 1999, 2000;
Mayer and Reichenberg 2006). Here, we have
developed a novel approach, termed "moving
M-regression," to identify a subset of com-
pounds with strong ICjQ versus LDj0 correla-
tion. Using this method, we have partitioned
compounds in the modeling set into two
classes: class 1, compounds with acute toxicity
that linearly correlate with cytotoxicity; and
class 2, compounds with acute toxicity that do
not correlate well with cytotoxicity, with these
points positioned above the regression line.
This analysis for the rat ZEBET data set
resulted in 122 compounds assigned to class
1, that is, within the linear regression cor-
relation band between LD50 and IC50 val-
ues. The points corresponding to 93 out of
108 remaining compounds are located above
the regression line band and are classified as
class 2, whereas 15 compounds fall below the
regression line (Figure 2A). Although these
compounds are likely to be activity outliers, in
the absence of an objective rationale for their
outlier status, we merged them into class 1 to
obtain the highest coverage of the resulting
models and to provide a more realistic mea-
sure of external predictivity. Figure 2A and
Equation 10 show the correlation between
the LDjQ and ICjQ values of the resulting 137
class 1 compounds:
Log(l/LD50)=-l.l +0.4
xlog(l/IC5()),
[10]
Abbreviations: C1, Class 1; C2, Class 2.
with R2 = 0.74, SE = 0.36, and n = 137. We
also applied this approach to analyze the rela-
tionship between the in vitro ICjQ and other
in vivo toxicity data, including mouse LDjQ,
rat chronic LOAEL, and rat chronic NOAEL
1260
VOLUME 1171 NUMBER 8 I August 2009 • Environmental Health Perspectives
Previous
-------
Two-step QSAR for acute rodent toxicity prediction
(Table 1). The same trend was found for all
data sets, that is, in all cases the data were
partitioned into two classes: a) points on
the baseline and V) points off the baseline
(Table 1, Figure 2). We found the ratio of
class 1 to class 2 compounds to be similar for
each of the four in vitrolin vivo toxicity data
sets. This result further supports the generality
of the "moving M-regression" approach.
Hierarchical QSAR modeling of the parti-
tioned rodent toxicity data. Using class assign-
ments from the data partitioning described
above, we employed a two-step QSAR
approach (Figure 1) for d) classification
modeling (i.e., establishing that compounds
assigned to classes 1 and 2 based on their
biological activity data could be subdivided
into the same classes based on their chemi-
cal structure), and b) predictive continuous
modeling for all compounds in each class
(i.e., estimation of the LDj0 based on chemi-
cal structure, not ICjQ data). For ZEBET rat
data, we generated three modeling sets: set 1,
230 compounds (137 class 1 vs. 93 class 2)
for classification modeling; set 2, 137 class 1
compounds; and set 3, 93 class 2 compounds
for developing two continuous rat LD50
models. The analysis of these three data sets
resulted in 252 classification models, as well
as 1,207 continuous LDj0 models for class 1
compounds and 40 continuous LDj0 models
for class 2 compounds that satisfied the sta-
tistical significance threshold criteria. Table 2
lists the statistical figures of merit for the best
models obtained for these three model-
ing sets.
To demonstrate that these QSAR mod-
els have significant external prediction accu-
racy, we have employed several concurrent
approaches for model validation. First, fol-
lowing our general model validation work
flow (Tropsha and Golbraikh 2007), we
used 23 compounds excluded randomly
from the entire data set as an external valida-
tion set. The following two-step prediction
protocol for external compounds was used:
a) £NN classification models were used to
assign compounds to class 1 or class 2; and
b) depending on the outcome, the respective
class-specific continuous QSAR models was
employed to predict the LD50 values for each
compound. The results demonstrate that the
overall accuracy of prediction for this exter-
nal set is reasonably good. In the first step,
the classification model had 65% prediction
accuracy (the fraction of correctly identified
class 1 and class 2 compounds). In the second
step, we obtained R2 = 0.70, mean absolute
error(MAE) = 0.39, and prediction cover-
age (i.e., the fraction of the external set com-
pounds within the applicability domains of
the models) of 74% for the external test set
when combining the predictions for class 1
and class 2 compounds.
Second, we performed a 5-fold external
CV analysis to test the robustness of the mod-
eling outcome using 253 rat ZEBET com-
pounds. The dataset was randomly split into
five equal-size subsets of compounds and the
modeling procedure was repeated five times,
using each subset as a test set and the remain-
ing four subsets as training set, as detailed
in "Materials and Methods." The statistical
results of this exercise were as follows: sloperegr
= 0.45 ± 0.01, #2regr = 0.71 ± 0.04, #2ext =
0.55 ± 0.05, MAE = 0.44 ± 0.04, coverage =
73 ± 3%.
Third, we performed Y-randomization tests
to establish whether our models are statisti-
cally robust. Random partitioning of the com-
pounds into two classes (10 times) produced
only three (for class 1) and 28 (for class 2)
models that satisfied the criteria ofq2/R2 > 0.5,
compared with 1,207 and 40, respectively,
models for "moving M-regression"—assisted
partitioning. Randomizing LDj0 data gener-
ated no model with q IR > 0.5 for class 1 and
class 2 compounds.
Fourth, we performed additional
Y-randomization analyses by randomly mov-
ing or rotating the correlation line (including
negative correlation) and redefining com-
pounds into classes 1 and 2. The randomly
assigned class 1 and class 2 sets were used
to develop QSAR LDj0 models individually
and the procedure was repeated 10 times. We
found that at most, a very small number (< 7)
of acceptable (Q2 > 0.5, R2 > 0.5) models
could be developed.
Similar modeling results were obtained
using the ZEBET mouse LD50 data. After par-
titioning 211 modeling set compounds into
119 class 1 compounds and 92 class 2 com-
pounds, we developed 843 classification mod-
els for class 1 versus class 2, 236 continuous
LD50 models for class 1 compounds, and 356
models for class 2 compounds. A two-step
prediction protocol for evaluation of the 24
external compounds resulted in similarly good
external prediction accuracy: R2 = 0.69, MAE
= 0.42, and prediction coverage of 54%.
As a true external validation challenge, we
have used our model to make predictions for
the 115 compounds with rat LD50 data in the
new ICCVAM data set. We compiled this
data set after we finished the development of
the above-described QSAR LD50 models, so
it could be viewed as a true "blind" validation
test. The statistical parameters of the predic-
tion results for these compounds were R2 =
0.57, MAE = 0.48, and prediction coverage of
70%. Although somewhat less accurate than
the results of the previous external prediction,
this validation reinforces the statistical signifi-
cance and utility of the model.
Y-randomization tests were also performed
for the mouse LDj0 data set. Similar to the
rat data, after 10 random assignments of com-
pounds into the two classes, we developed,
at most, 4 (for class 1) and 38 (for class 2)
models (q2IR2 > 0.5), compared with 843
and 236 models, respectively, when we used a
classification model. Randomization of LD50
values produced no significant models.
Stability of the in vitro andin vivo moving
M-regression parameters. Because the regres-
sion correlation between in vitro (ICjo) and
in vivo (LD5o) data is required to classify the
modeling set compounds and, subsequently,
to create the £NN classification models, this
linear correlation is an essential factor to
determine the robustness of our final mod-
els. Hence, the slope of the correlation and
associated correlation coefficient (R2) should
remain stable when new compounds are
added into the modeling set. To validate this
Table 2. Statistical information for the five most statistically significant kNN QSAR models based on three
modeling sets.
Model
N-training
Pred-training
N-test
Pred-test
The best kNN classification model for 137 class 1 versus 93 class 2 compounds
1
2
3
4
5
173 0.84
147 0.86
193 0.83
165 0.86
173 0.81
55
74
37
59
55
0.73
0.70
0.73
0.70
0.75
1
1
1
1
1
The best kNN continuous model for 137 class 1 compounds
1
2
3
4
5
103 0.66
103 0.73
111 0.71
115 0.65
77 0.73
34
34
26
22
60
0.81
0.71
0.74
0.79
0.71
3
2
3
5
2
The best kNN continuous model for 93 class 2 compounds
1
2
3
4
5
80 0.61
77 0.67
80 0.69
80 0.65
79 0.63
13
16
13
13
14
0.84
0.77
0.74
0.76
0.78
2
1
1
2
2
Abbreviations: NNN; number of the nearest neighbors used for prediction; N-test, number of compounds in the test set;
N-training, number of compounds in the training set; Pred-test, the overall predictivity of the test set (correct classifica-
tion rate for classification models, IP for continuous models); Pred-training, the overall predictivity of the training set (cor-
rect classification rate for classification models, cf for continuous models).
Environmental Health Perspectives • VOLUME 117 I NUMBER 81 August 2009
1261
Previous
-------
Zhu etal.
supposition, we compiled all available ZEBET
and ICCVAM compounds with rat LD50
data to create a new modeling set, including
the original modeling set (230 compounds),
the external validation set (23 compounds),
and additional data (115 compounds). We
also included the compounds previously not
used for modeling (inorganic, organometallic,
and mixtures) because we used no chemi-
cal descriptors in this validation. Using the
moving M-regression approach for all 425
compounds with IC50 and LD50 values, the
resulting in vitrolin vivo correlation param-
eters are similar to those obtained from our
original modeling set in EquationlO:
Log(l/LD50)=-l.l +0.36
xlog(l/IC5()),
[11]
with R2 = 0.71, SE = 0.37, and n = 258. The
proportions of class 1 and class 2 compounds
and outliers among these 425 compounds
were also comparable to those of the original
modeling set of 230 compounds (Table 1).
We conclude that adding new compounds
into the modeling set, which should be impor-
tant to improve the final model by enriching
its chemical and biological diversity, does not
affect the in vitrolin vivo regression statistics.
Comparison between the two-step hierar-
chical LD50 QSAR model and TOPKAT. We
compared the performance of our modeling
approach with that of TOPKAT software, ver-
sion 6.1 (Accelrys 2009; Enslein 1988). Two
types of comparison were considered. First,
we have analyzed 27 of the 115 ICCVAM
compounds that have been used neither for
building our model nor in the TOPKAT
2.0
. o.o
-o
£ -0.5
o
=5 -1.0
o>
£ -1.5
-2.0
-2
~S
o
-2-101 2
Experimental Log (1/LD50)
Figure 3. The correlation between experimental
and predicted LD50 values for 27 external com-
pounds within the applicability domain (A) using
TOPKAT and (6) using the two-step model devel-
oped in this study.
LD50 training set. Figure 3 shows the correla-
tion between the experimental and predicted
LD50 values obtained from our model versus
TOPKAT. The R2 and MAE of TOPKAT
were 0.16 and 0.78, respectively, for all
27 compounds, which is considerably less than
the same statistical parameters for prediction
of the same data set using our model, R and
MAE of 0.64 and 0.38, respectively. For seven
compounds that were outside of the applicabil-
ity domain for our model, the R2 and MAE
using TOPKAT were 0.60 and 0.50, respec-
tively, whereas our model produced values of
0.86 and 0.29, respectively (Table 3).
Second, we have used our models to pre-
dict acute toxicity compounds in the RTECS
(Norager et al. 1978) data set (data were kindly
provided by Todd Martin from the U.S.
EPA), which contains approximately 7,000
compounds with rat LD50 data. We removed
compounds that we found within the ZEBET
data set, as well as inorganic compounds and
mixtures. This procedure produced a library of
4,003 compounds spanning a diverse chemical
space of organic molecules for which experi-
mental rat LD50 data are available.
Because the size of the RTECS library is
much larger than that of our original model-
ing set, we drew from our experience in using
QSAR models for virtual screening (Oloff
et al. 2005) and narrowed the model applica-
bility domain. Consequently, predictions were
made only for compounds that had greater
than 70% confidence level in assigning them
to either class 1 or class 2 in step 1 of our
work flow (i.e., we required that > 70% of
all QSAR models meeting our acceptability
domain criteria would predict a compound
in the same class). We determined that there
were 1,562 compounds (out of 4,003) that
were not included in the training set of
TOPKAT rat LD50 model and for which pre-
dictions could be made based on the afore-
mentioned criteria. The TOPKAT model
predicted LD50 values for these compounds
with an R2 = 0.16 and MAE = 0.78 (Figure
4, Table 3). The same parameters for the two-
step QSAR model were 0.26 and 0.65, respec-
tively. After implementing the applicability
domain filter, we made predictions for 965
RTECS chemicals. TOPKAT model had
parameters of R2 = 0.22 and MAE = 0.65; the
same parameters for the two-step model were
0.33 and 0.54, respectively (Table 3), which is
better than or comparable to prediction accu-
racy of various commercial QSAR modeling
packages (Moore et al. 2003), albeit there is
room for improvement.
It should be noted that the predic-
tion accuracy of the two-step model can be
improved by applying stricter criteria in the
classification step. For instance, a 90% cutoff
for correct class prediction results in prediction
model statistics of R2 = 0.62 and MAE = 0.42,
but the coverage of the model diminishes con-
siderably to include 101 compounds (Table 3).
The performance of TOPKAT for the same
101 compounds is poor: R2 = 0.26 and MAE
= 0.66. Considering that the TOPKAT
LDjo training set contains many more com-
pounds (- 6,000) than the training set used
to develop the two-step model (- 200), it is
noteworthy that higher prediction accuracy
can be achieved using our modeling approach
for a much larger data set. Furthermore, our
approach outperforms TOPKAT consistently
over a range of error thresholds either for
965 RTECS compounds or for 101 RTECS
compounds (Figure 4). In addition, we used
the Wilcoxon test to calculate the ^-values for
the differences in MAEs obtained using two
methods. Both for the whole set (965 com-
pounds) and for the reduced set (101 com-
pounds), the improvement achieved by our
method, compared with TOPKAT, is statisti-
cally significant (p < 0.005).
One obvious reason that the prediction
accuracy of our models for RTECS com-
pounds is lower than that obtained from the
external validation set of ICCVAM data is the
difference in "activity" ranges of compounds
in these two data sets. For example, the activ-
ity (log 1/activity, in millimolar units) of
ZEBET compounds ranges from -2.61 to
2.30, whereas the activity range of RTECS
compounds is considerably larger, from —3.34
to 4.21. It should be stressed that the £NN
method used in our study cannot extrapolate
TableS. Comparison between TOPKAT and the two-step model prediction of the external compounds.
Two-step model
TOPKAT
Measure No applicability domain With applicability domain No applicability domain With applicability domain
Prediction of 27 new ZEBET compounds
ff 0.64 0.86 0.16 0.60
MAE 0.38 0.29 0.78 0.50
Coverage (%) 100 67 100 67
Prediction of 1,562 RTECS compounds with 70% confidence level
ff 0.26 0.33 0.19 0.22
MAE 0.65 0.54 0.76 0.65
Coverage (%) 100 62 100 62
Prediction of 1,562 RTECS compounds with 90% confidence level
ff 0.42 0.62 0.19 0.26
MAE 0.60 0.42 0.84 0.66
Coverage (%) 12 6 12 6
1262
Previous
VOLUME 1171 NUMBER 8 I August 2009 • Environmental Health Perspectives
TOC
-------
Two-step QSAR for acute rodent toxicity prediction
in the activity space because external com-
pound activity is predicted by averaging the
activities of nearest-neighbor compounds in
the training set as described in "Materials and
Methods." The MAE for the prediction of
RTECS compounds that have experimen-
tal activity above 2 or below —2 is 1.14 log
units. On the other hand, the MAE for the
prediction for RTECS compounds that have
experimental activity between -2 and 2 is
considerably lower, 0.52 log units. The likely
explanation for the better performance of our
models in the latter range is that more than
90% of our modeling set compounds have
rat LDjo activity in the same range, between
—2 and 2. Increasing the diversity and activ-
ity range of compounds in the modeling set
should significantly improve the prediction
accuracy of our models.
Discussion
The conventional wisdom in mechanistic and
regulatory toxicology is that predictions of the
in vivo toxicity end points from in vitro meas-
ures, even within the same species, are difficult.
However, an approximate linear correlation
between in vitro IC50 and rodent LD50, two of
the widely acceptable benchmark parameters
used for regulatory purposes, can be established
for a significant fraction of the compounds.
Indeed, we confirmed this notion by quantita-
tive analysis of the ICjQ/LDjQ relationships and
devised an objective, computational means to
partition compounds into two groups: those
having good linear fit within a defined band, or
those falling outside the band and exhibiting, for
the most part, higher in vivo than in vitro toxic-
ity. Our hypothesis to explain this observation is
that, whereas cytotoxicity assays can reflect some
of the toxicity mechanisms resulting in adverse
health effects at the whole-animal level, the
in vitro tests cannot fully reproduce the complex
mechanisms of the in vivo toxicity. For example,
it is well known that many compounds are not
toxicants themselves but have metabolites that
are toxic. We argue that the two-step predic-
tion model based on chemical descriptors only
„ 1.0
ffl 0.6
o
Two-step model
(965 RTECS)
Two-step model
(101 RTECS)
TOPKAT
(965 RTECS)
TOPKAT
(101 RTECS)
- --Randomsampling
1 2
Prediction errors
Figure 4. Fraction of compounds versus prediction
errors obtained by the two-step rat LD60 model,
TOPKAT, and random sampling for 965 and 101
RTECS compounds.
that we developed in our studies also assists
in identification of the compound subset that
may act directly (i.e., without being biotrans-
formed) and through mechanisms likely to
be predictive of the potential in vivo effects. A
similar argument was presented previously in
ecotoxicity research where log P was found to
be a mechanistically relevant predictor (Verhaar
et al. 2000).
To further substantiate this argument,
we considered the top 10 chemical fragment
descriptors that were used most frequently in
statistically significant QSAR models, that is,
descriptors with the highest discriminatory
power [see Supplemental Material, Table 3
(doi:10.1289/ehp.0800471.SI)]. It is note-
worthy that the aromatic primary amine,
"hydrazine," and "sulfonamide" moieties,
found within compounds that are known to
be toxic both in vitro and in vivo (Alaejos
et al. 2008; Carr et al. 1993; Toth 1988),
were found predominantly in compounds
of class 1. On the other hand, "pyrrolidine"
and "aromatic tertiary amine" moieties,
which require biotransformation (Domagala
1994), were predictors for class 2. We have
also demonstrated that this objective divi-
sion of the data set into two major groups
affords robust hierarchical QSAR models, an
assertion further supported by successive chal-
lenges to the models with external data sets,
CV, and randomization of data.
The approach advocated in this study for
biologically informed partitioning of structure-
activity relationship data differs from conven-
tional cheminformatics clustering approaches.
Traditional methods partition compounds into
multiple subgroups based on their chemical
structure properties only (i.e., chemical descrip-
tors). The underlying reasoning for chemically
based clustering is that similar structures are
expected to have similar biological properties
and mechanisms of activity. However, it is a
well-known limitation of structure-activity
relationships that the absence or presence of a
functional group or other minor change of the
chemical structure may result in a large change
of biological activity (Maggiora 2006). In our
studies, the conventional chemical structure-
based clustering method did not yield any sta-
tistically meaningful models, either global or
local. The distribution of pairwise chemical
similarities for all compounds within the mod-
eling sets (class 1 vs. class 2) of rat LDj0 values
using DRAGON descriptors is very similar
(data not shown). This observation reconfirms
that chemical clustering would not have par-
titioned compounds in a way similar to the
biological data-based partitioning.
Conclusions
Although the cytotoxicity data generally show
weak correlation with rodent acute toxicity,
we have demonstrated that these data can be
used to inform and improve QSAR model-
ing of in vivo acute toxicity. We have devel-
oped a novel two-step £NN QSAR modeling
approach that affords a successful prediction
of acute toxicity (LDj0) values from chemical
structure for both rats and mice. Furthermore,
we predicted LDj0 values for external com-
pounds with accuracy, exceeding that of previ-
ously published QSAR models developed with
the commercial (TOPKAT) software. It should
be stressed that although in vitro cytotoxic-
ity data have been used to establish the rules
for partitioning most compounds into two
classes, the ultimate models, both classification
and continuous, employ chemical descriptors
only. This vital feature of our approach makes
it possible to achieve accurate predictions of
rodent acute toxicity directly from chemical
structure alone, even bypassing the need for
in vitro studies of new compounds. We believe
that this biological-data—based partitioning
approach using in vitro toxicity data for the
modeling set only, coupled with subsequent
chemical-structure—based classification and
continuous QSAR modeling techniques, holds
promise for modeling other complex in vivo
toxicity end points. This approach charts a
future course for combining in vitro screening
methods and QSAR modeling to prioritize
chemicals for in vivo animal toxicity testing.
REFERENCES
Accelrys. 2009. Predictive Toxicology - DS TOPKAT. Available:
http://accelrys.com/products/discovery-studio/toxicology/
[accessed 30 June 2009].
Alaejos MS, Pino V, Afonso AM. 2008. Metabolism and toxicol-
ogy of heterocyclic aromatic amines when consumed
in diet: influence of the genetic susceptibility to develop
human cancer. A review. Food Res Int 41:327-340.
Andersen M. 2007. Modern Methods for Robust Regression.
Thousand Oaks, CA:Corwin Press.
Benigni R, Netzeva Tl, Benfenati E, Bossa C, Franke R, Helma C,
et al. 2007. The expanding role of predictive toxicology: an
update on the (Q)SAR models for mutagens and carcino-
gens. J Environ Sci Health C Environ Carcinog Ecotoxicol
Rev 25:53-97.
Carr A, Tindall B, Penny R, Cooper DA. 1993. In vitro cytotoxicity
as a marker of hypersensitivity to sulphamethoxazole in
patients with HIV. Clin Exp Immunol 94:21-25.
Collins FS, Gray GM, Bucher JR. 2008. Toxicology. Transforming
environmental health protection. Science 319:906-907.
Dearden JC. 2003. In silico prediction of drug toxicity. J Comput
Aided MolDes 17:119-127.
Dix DJ, Houck KA, Martin MT, Richard AM, Setzer RW,
Kavlock RJ. 2007. The ToxCast program for prioritizing toxic-
ity testing of environmental chemicals. Toxicol Sci 95:5-12.
Domagala JM. 1994. Structure-activity and structure-side-effect
relationships for the quinolone antibacterials. J Antimicrob
Chemother 33:685-706.
Enslein K. 1988. An overview of structure-activity relationships
as an alternative to testing in animals for carcinogenicity,
mutagenicity, dermal and eye irritation, and acute oral
toxicity. Toxicol Ind Health 4:479-498.
Freidig AP, Dekkers S, Verwei M, Zvinavashe E, Bessems JG,
van de Sandt JJ. 2007. Development of a QSAR for worst
case estimates of acute toxicity of chemically reactive
compounds. Toxicol Lett 170:214-222.
Golbraikh A, Shen M, Xiao Z, Xiao YD, Lee KH, Tropsha A. 2003.
Rational selection of training and test sets for the develop-
ment of validated QSAR models. J Comput Aided Mol Des
17:241-253.
ICCVAM and NICEATM. 2001. Report of the International
Workshop on In Vitro Methods for Assessing Acute
Environmental Health Perspectives • VOLUME 117 I NUMBER 81 August 2009
Previous
1263
-------
Zhu etal.
Systemic Toxicity. Interagency Coordinating Committee
on the Validation of Alternative Methods and National
Toxicology Program Interagency Center for the Evaluation
of Alternative Toxicological Methods Report 01-4499
Bethesda, MD:National Institutes of Health.
Johnson DE, Smith DA, Park BK. 2004. Linking toxicity and
chemistry: think globally, but act locally? Curr Opin Drug
DiscovDevel 7:33-35.
Klopman G, Saiakhov R, Rosenkranz HS. 2000. Multiple computer-
automated structure evaluation study of aquatic toxicity II.
Fathead minnow. Environ Toxicol Chem 19:441-447.
Klopman G, Saiakhov R, Rosenkranz HS, Hermens JLM. 1999.
Multiple Computer-Automated Structure Evaluation pro-
gram study of aquatic toxicity 1: guppy. Environ Toxicol
Chem 18:2497-2505.
Maggiora GM. 2006. On outliers and activity cliffs—why QSAR
often disappoints. J Chem Inf Model 46:1535; doi: 10.1021/
ci700332k [Online 28 December 2007].
Matthews EJ, Kruhlak NL, Cimino MC, Benz RD, Contrera JF.
2006. An analysis of genetic toxicity, reproductive
and developmental toxicity, and carcinogenicity data:
II. Identification of genotoxicants, reprotoxicants, and car-
cinogens using in silico methods. Regul Toxicol Pharmacol
44:97-110.
Mayer P, Reichenberg F. 2006. Can highly hydrophobic organic
substances cause aquatic baseline toxicity and can they
contribute to mixture toxicity? Environ Toxicol Chem
25(101:2639-2644.
Medina-Franco JL, Golbraikh A, Oloff S, Castillo R, Tropsha A.
2005. Quantitative structure-activity relationship analysis
of pyridinone HIV-1 reverse transcriptase inhibitors using
the k nearest neighbor method and QSAR-based database
mining. J Comput Aided Mol Des 19:229-242.
Moore DR. Breton RL, MacDonald DB. 2003. A comparison of
model performance for six quantitative structure-activity
relationship packages that predict acute toxicity to fish.
Environ Toxicol Chem 22:1799-1809.
National Research Council. 2007. Toxicity Testing in the 21st
Century: A Vision and a Strategy. Washington, DC:National
Academies Press.
Ng C, Xiao Y, Putnam W, Lum B, Tropsha A. 2004. Quantitative
structure-pharmacokinetic parameters relationships
(QSPKR) analysis of antimicrobial agents in humans using
simulated annealing k-nearest-neighbor and partial least-
square analysis methods. J Pharm Sci 93:2535-2544.
Norager 0, Town WG, Petrie JH. 1978. Analysis of the regis-
try of toxic effects of chemical substances (RTECS] files
and conversion of the data in these files for input to the
environmental chemicals data and information network
(ECDIN]. J Chem Inf Comput Sci 18:134-140.
Oloff S, Mailman RB, Tropsha A. 2005. Application of validated
QSAR models of D1 dopaminergic antagonists for data-
base mining. J Med Chem 48:7322-7332.
Rabinowitz JR. Goldsmith MR, Little SB, Pasquinelli MA. 2008.
Computational molecular modeling for evaluating the tox-
icity of environmental chemicals: prioritizing bioassay
requirements. Environ Health Perspect 116:573-577.
Ruden C, Hansson SO. 2003. How accurate are the European
Union's classifications of chemical substances. Toxicol
Lett 144:159-172.
Shen M, LeTiran A, Xiao Y, Golbraikh A, Kohn H, Tropsha A.
2002. Quantitative structure-activity relationship analysis
of functionalized amino acid anticonvulsant agents using
k nearest neighbor and simulated annealing PLS methods.
J Med Chem 45:2811-2823.
Stouch TR, Kenyon JR. Johnson SR, Chen XQ, Doweyko A, Li Y.
2003. In silico ADME/Tox: why models fail. J Comput Aided
Mol Des 17:83-92.
Toth B. 1988. Toxicities of hydrazines: a review. In Vivo 2:209-242.
Tropsha A, Golbraikh A. 2007. Predictive QSAR modeling work-
flow, model applicability domains, and virtual screening.
Curr Pharm Des 13:3494-3504.
Venkatapathy R, Moudgal CJ, Bruce RM. 2004. Assessment of
the oral rat chronic lowest observed adverse effect level
model in TOPKAT, a QSAR software package for toxicity
prediction. J Chem Inf Comput Sci 44:1623-1629.
Verhaar HJ, Solbe J, Speksnijder J, van Leeuwen CJ,
Hermens JL. 2000. Classifying environmental pollutants:
part 3. External validation of the classification system.
Chemosphere 40:875-883.
Xia M, Huang R, Witt KL, Southall N, Fostel J, Cho MH, et al.
2008. Compound cytotoxicity profiling using quantita-
tive high-throughput screening. Environ Health Perspect
116:284-291.
Zheng W, Tropsha A. 2000. Novel variable selection quantita-
tive structure-property relationship approach based on
the k-nearest-neighbor principle. J Chem Inf Comput Sci
40:185-194.
Zhu H, Rusyn I, Richard A, Tropsha A. 2008. Use of cell viability
assay data improvesthe prediction accuracy of conventional
quantitative structure-activity relationship models of animal
carcinogenicity. Environ Health Perspect 116:506-513.
1264
VOLUME 1171 NUMBER 8 I August 2009 • Environmental Health Perspectives
Previous
-------
Toxicology and Applied Pharmacology 233 (2008) 7-13
Contents lists available at ScienceDirect
Toxicology and Applied Pharmacology
journal homepage: www.elsevier.com/locate/ytaap
ACToR — Aggregated Computational Toxicology Resource
Richard Judson a'*, Ann Richard a, David Dix a, Keith Houck a, Fathi Elloumia, Matthew Martin a,
Tommy Catheyb, Thomas R. Transueb, Richard Spencer b, Maritja Wolfb
a National Center for Computational Toxicology, US. Environmental Protection Agency, 109 T.W. Alexander Drive, Research Triangle Park, NC 27711, USA
b Lockheed Martin, A Contractor to the US. Environmental Protection Agency, 109 T.W. Alexander Drive, Research Triangle Park, NC 27711, USA
ARTICLE INFO
Article history:
Available online 11 July 2008
Keywords:
ToxCast
ACToR
Database
HTS
Screening
Priori tization
ABSTRACT
ACToR (Aggregated Computational Toxicology Resource) is a database and set of software applications that
bring into one central location many types and sources of data on environmental chemicals. Currently, the
ACToR chemical database contains information on chemical structure, in vitro bioassays and in vivo
toxicology assays derived from more than 150 sources including the U.S. Environmental Protection Agency
(EPA), Centers for Disease Control (CDC), U.S. Food and Drug Administration (FDA), National Institutes of
Health (NIH), state agencies, corresponding government agencies in Canada, Europe and Japan, universities,
the World Health Organization (WHO) and non-governmental organizations (NGOs). At the EPA National
Center for Computational Toxicology, ACToR helps manage large data sets being used in a high-throughput
environmental chemical screening and prioritization program called ToxCast™.
Published by Elsevier Inc.
Introduction
Computational Toxicology is an emerging field that aims to use
modern computational and molecular biology techniques to under-
stand and predict chemical toxicity. A particular area where this
approach is being applied is in chemical screening and prioritization.
In the U.S., there are an estimated 30,000 unique chemicals in wide
commercial use (>1 t/year) (Muir and Howard, 2006), and only a
relatively small subset of these has been sufficiently well character-
ized for their potential to cause human or ecological toxicity to
support regulatory action. This "data gap" is well documented (EPA,
1998; Allanou et al., 1999; Birnbaum et al., 2003; Guth et al., 2005;
Applegate and Baer, 2006; Krewski et al., 2007). The standard
approach to determine a chemical's toxicity profile involves perform-
ing in vivo studies on rodents and other species, and can take
2-3 years and cost millions of dollars per chemical. Clearly, this
strategy is neither practical nor viable for evaluating tens of thousands
of chemicals; hence, the large inventories of existing chemicals for
which little or no test data are available. An alternative approach is to
attempt to cover much larger regions of chemical space by employing
more efficient in vitro methods. One strategy applies relatively
inexpensive and rapid high-throughput screening (HTS) assays to a
large set of chemicals, followed by the use of these results to prioritize
a much smaller subset of chemicals for more detailed analysis. The
* This work was reviewed by EPA and approved for publication but does not
necessarily reflect official agency policy.
* Corresponding author. Fax: +1 919 541 3085.
E-mail address: judson.richard@epa.gov (R. Judson).
0041-008X/S - see front matter. Published by Elsevier Inc.
doi:10.1016/j.taap.2007.12.037
"prioritization score" for a chemical would be based on derived
signatures, or patterns extracted from the HTS data, which are
predictive of particular effects or modes of chemical toxicity.
Chemicals of known toxicity comprise the reference or training set
that is used to develop and validate predictive signatures. HTS assays
that yield data for the predictive signatures would then be run on
chemicals of unknown toxicity (the test chemicals), and a prioritiza-
tion score for those chemicals would be produced. The U.S. EPA has
made a significant investment in this approach through the recent
launch of the ToxCast™ research program (Dix et al., 2007). ToxCast is
screening hundreds, and eventually thousands of environmental
chemicals using hundreds of HTS assays towards the two goals of
developing predictive toxicity signatures, and using these signatures
to prioritize chemicals for further testing. In this EPA context, the term
"environmental chemicals" refers primarily to industrial chemicals
and pesticides used or produced in large enough quantities to pose
significant potential for human or ecological exposure.
There are multiple computational aspects to this approach. First,
some of the screening assays themselves may be computational (in
silica). Second, a robust database and data analysis infrastructure are
required to manage the large data volumes produced by a large-scale
HTS program. Third, one needs high quality in vivo toxicology data on
as large and diverse group of chemicals as possible in order to develop
and validate the predictive signatures. Currently, such toxicity data are
available from a number of sources, but these data are widely
dispersed and often not sufficiently annotated or fully accessible for
computational use.
To support the EPA's ToxCast screening and prioritization effort, as
well as other EPA programs, we are developing a system called ACToR,
Previous
-------
R.Judson et al. / Toxicology and Applied Pharmacology 233 (2008) 7-13
for Aggregated Computational Toxicology Resource. ACToR is a set of
linked databases and software applications that bring together many
types and sources of data on environmental chemicals into one central
location. Currently, the ACToR chemical and assay databases contain
information on chemical structure, in vitro bioassays and in vivo
toxicology assays derived from more than 150 sources including the
EPA, CDC, FDA, N1H, state agencies, corresponding government
agencies in Canada, Europe and Japan, universities, the World Health
Organization and NGOs. An important set of data collections comes
from the DSSTox project (Distributed Structure-Searchable Toxicity)
(Richard and Williams, 2002) at the EPA which produces curated
collections of chemical structures with corresponding assay data. The
design of ACToR has followed that of the NIH PubChem Project in
many respects, but has been generalized to allow for the broader types
of data that are of interest to toxicologists and environmental
regulators. The current ACToR web interface is also designed to
meet the needs of scientists focused on the study of chemical toxicity.
This paper briefly outlines the design of the ACToR database and
the types of data it contains, and will illustrate its utility in the context
of developing training and validation data sets for chemical screening
and prioritization projects.
Materials and methods
Organization of the database. The current version of ACToR is focused
mainly on capturing information on chemicals and assays of chemical-
biological effects. Plans are underway to extend this to capture relevant
genomic and biological pathway information. The organizing
principles for the design of the chemical/assay system are largely
derived from the PubChem project, which is capturing chemical
structure and HTS information on millions of chemicals in its role as
the main data repository for the NIH Molecular Libraries Roadmap
(Austin et al., 2004). The main organizing principle of PubChem centers
on the three main types of data that are catalogued: substances
indexed by substance identifier (SID), compounds (i.e., chemical
structures, indexed by compound identifier (C1D), and bioassays
indexed by assay identifier (AID). A PubChem substance is a single
chemical entity submitted by one data source and often corresponds to
Generic Chemical
(from ACTOR)
ACTOFLGCID : INTEGER
NAME : TEXT
CASRN:TEXT
0..1
1..*
0..1
Substance
(from ACTOR)
ACTOR_SID : INTEGER
NAME : TEXT
CASRN:TEXT
0..*
Assay Result
(from ACTOR)
ACTOR_RSID : INTEGER
VALUE : REAL
1
0..*
Assay Component
(from ACTOR)
ACTOR_ACID : INTEGER
NAME:TEXT
0..1
Compound
(from ACTOR)
ACTOR_CID : INTEGER
NAME:TEXT
SMILES : TEXT
MOLFILE : TEXT
0..1
0..*
Data Collection
(from ACTOR)
ACTOR_DCID : INTEGER
NAME : TEXT
0..*
Assay
(from ACTOR)
ACTOR_A!D : INTEGER
NAME:TEXT
0..*
Phenotype
(from ACTOR)
ACTOR_PID : INTEGER
NAME:TEXT
Fig. 1. Schematic, entity-relationship (ER) diagram of the ACToR database schema showing key relationships between substances, compounds, generic chemicals, data collections and
assays. The annotations on the connecting lines are the range of number of entities for each relationship. For instance, a substance can have 0 or 1 generic chemicals (indicated by 0..1),
while a generic chemical can have 1 to many substances (indicated by 1.. ). The zero/one-to-many relationships are implemented through separate tables (not shown). The actual
schema contains 44 tables.
Previous
-------
R. Judson et al. / Toxicology and Applied Pharmacology 233 (2008) 7-13
the physical substance on which some experiment was performed. A
compound is a generic chemical entity that corresponds to a unique
chemical structure. Since a substance is defined as being both data
source and experiment-specific, many substances (SIDs) may map to a
single compound (CID). A bioassay, indexed by AID, represents a
specific type of test data associated with one or more substances.
In ACToR, these ideas are generalized somewhat, although the
model is close enough such that all data from PubChem can be easily
loaded into ACToR. In ACToR, a substance is similarly defined as a
unique chemical from a single "data collection" (see below) and is
minimally characterized by a data collection-specific SID and a
chemical name. Most often, the substance will also have synonyms,
a CAS (Chemical Abstracts Service) registry number (CASRN) and
multiple other parameters. A compound always has an associated
chemical structure and a data collection-specific CID, in addition to
optional parameters derived directly from chemical structure, such as
SMILES and InChI representations and a molecular weight. Note that
since ACToR is in essence a "super-aggregator", pulling in large
external data collections such as PubChem, it also stores the source-
labeled CIDs from each independent collection (e.g., PubChem CID,
DSSTox CID). The data collection-specific SIDs and CIDs are called
SOURCE_NAME_S1D and SOURCE_NAME_CID and are alphanumeric
strings of the form PUBCHEM_1234 or S1DS.2345. Additionally, ACToR
internally uses sequentially generated unique numeric SIDs and CIDs.
Data on chemicals across data collections are aggregated using the
concept of a generic chemical, which for this purpose takes the place
of the compound in PubChem. The vast majority of chemicals in
PubChem have defined structures, while many environmental
chemicals are complex, and often undefined, mixtures. However,
most environmental chemicals, along with their related toxicity data
are indexed by a more discriminating CAS registry number (CASRN)
rather than by chemical structure. Because of this, ACToR aggregates
information based on CASRN. A generic chemical is defined by a
CASRN, a preferred name, an ACToR CID and a unique generic chemical
identifier or GCID. All data on all substances sharing a particular
CASRN are attached to the corresponding generic chemical. An
advantage of using CASRN is that different numbers will be assigned
to a pure substance versus a mixture of isomeric forms or a mixture of
unrelated compounds. All of these cases, however, may share a
common compound PubChem CID and representative structure.
Disadvantages of using CASRN include the fact that they are not
always available or unique for a given substance (e.g. CASRN can be
retired and replaced), they do not typically distinguish to the level of
compound purity grade (e.g., analytical vs. technical grade), and they
are tied to a non-public registry system (Chemical Abstracts Service
(CAS) SciFinder). Nonetheless, CASRN are sufficiently general to serve
as the basis for aggregation. Because only a small fraction of PubChem
substance records contain a CASRN, we perform a second level of
aggregation and pull in all PubChem substances that share the
structure or PubChem CID associated with a particular GCID.
Currently, the two main sources of chemical structure data in ACToR
are EPA DSSTox and PubChem. Because DSSTox structures are quality
reviewed, hand curated, and reconciled with chemical name and
CASRN, they are always preferred over structures automatically
generated and provided by disparate sources in PubChem. Fig. 1
illustrates the basic relationships between substance, compounds,
generic chemicals and assays, which are described next.
In ACToR, an assay is a collection of data values associated with a
set of substances and can be represented in a rectangular matrix. An
assay is associated with an AID, a name, a category, and one or more
phenotypes. Examples of assay categories are listed in Table 1 and
reflect our focus on chemical toxicity and its origin in detailed
molecular biological interactions. As one can see, the concept of an
assay as implemented in ACToR is purposely broad so as to capture any
information potentially relevant to understanding toxicity and
evaluating risk for environmental chemicals. An assay also can have
one or more components, which correspond to the columns of the
rectangular data matrix. Each component is defined by an assay
component identifier (ACID), the corresponding AID, a name, a
description, units (when applicable) and a data type (FLOAT, INTEGER,
CATEGORICAL, TEXT, BOOLEAN, URL). The actual data values are called
assay results and are linked to the assay, the assay component and the
original data-collection-specific substance.
Because ACToR is intended to support hazard identification and
risk assessment, assays can be labeled by a series of "phenotypes" for
which they contain information. The set of phenotypes implemented
in ACToR spans both traditional toxicology study areas: general
chemical hazard, acute toxicity, subchronic toxicity, chronic toxicity,
carcinogenicity, developmental toxicity, reproductive toxicity, neuro-
toxicity, developmental neurotoxicity, immunotoxicity, dermal toxi-
city, respiratory toxicity, genotoxicity and ecotoxicity.
Data sources. ACToR is importing data from a large number of public
sources (currently >150), which are referred to as data collections. A
data collection will usually include a set of substances and may have
corresponding compounds (chemical structures) and one or more
assays. The largest source of data currently in ACToR in terms of
substances and assay data points is PubChem, which is itself a
Table 1
Categories of assays in ACToR
Assay category
Description
Examples
Physicochemical
Biochemical
Genomics
Cellular
Tissue
In vivo toxicology (tabular primary)
In vivo toxicology (study listing primary)
In vivo toxicology (tabular secondary)
In vivo toxicology (summary calls)
In vivo toxicology (summary report via URL)
Regulatory
Chemical Category
Physical and chemical properties (in vitro and/or in silica)
Biochemical (non-cell-based) (in vitro and/or in si/ico)
Gene expression values or signatures
Cell-based assay
Tissue slice assays
Tabulated results from primary animal-based studies
of chemical effect
Primary studies are available but have not been tabulated
Tabulated data from secondary sources of in vivo
toxicology studies
Derived summary determinations of risk
Links to text reports on the web for which specific data
values are not directly accessible in tabular form
Listings of chemicals that fall under specific environmental
laws or government mandates
Listing of structural or use categories, often intended
for prioritization efforts
Molecular weight, logR boiling point
Enzyme inhibition or receptor binding constants
Result of in vitro or in vivo microarray analysis
Cell culture cytotoxicity
Tissue slice cytotoxicity
Clinical chemistry, histopathology, developmental
and reproductive assays
Clinical chemistry, histopathology, developmental
and reproductive assays
Clinical chemistry, histopathology, developmental
and reproductive assays
Chemicals determined to pose a defined risk
of human cancer
Reports from EPA Integrated Risk Information System (IRIS),
National Toxicology Program (NTP)
U.S. Toxic Substances Control Act (TSCA)
Phthalate
Previous
TOC
-------
10
R. Judson eta/./ Toxicology and Applied Pharmacology 233 (2008) 7-13
Table 2
Summary statistics for the ACToR database
Data collections
232
Source-specific substances
Compounds (chemical structures)
Generic chemicals
Generic chemicals with structure
Assays
Assay components
Assay results
964,083
404,196
504,871
390,379
1592
10,733
6,118,231
Assay results are individual data points for a single substance and a single assay
component. The numbers only include substances having CAS registry numbers. A
much large number of substances, compounds and assay results are included from
PubChem, but are not currently indexed as generic chemicals.
compilation of multiple data sources (57 of which have data included
in ACToR). Most assay data in PubChem comes from HTS assays run by
the Molecular Libraries Screening Centers Network (MLSCN) (Austin
et al., 2004) on compounds from the Molecular Libraries Small
Molecule Repository (MLSMR). However, the vast majority of
chemicals in PubChem have no assay data and come from collections
of molecular structures from chemical manufacturer catalogs (e.g.,
SIGMA) or virtual screening libraries (e.g., ZINC).
The balance of the data collections within ACToR pertains more
specifically to environmental chemicals. These collections are from
the US EPA, CDC, FDA, NIH, equivalent agencies in Europe, Japan and
Canada, the World Health Organization, universities and several states
and NGOs. Some of these specific sources are described below.
To be included in ACToR, a data collection must meet several
criteria. First, it has to be publicly available with no restrictions on
redistribution. An important goal of the ACToR project is to create a
widely usable, freely distributable, open-source system. Any conclu-
sions drawn from these data should be subject to independent
confirmation, which is made possible by this open-source data model.
Second, the collection should contain information on environmen-
tally-relevant chemicals. We have not to this point included a number
of data sets focused exclusively on pharmaceutical compounds,
although toxicological information on these compounds is potentially
informative. Third, if a source consists of a web-accessible database,
we require an index of the chemicals in the database in order to link
that web resource back into ACToR. Several web databases that allow
local searching by name or CASRN do not provide full access or a list of
the chemicals needed for indexing; hence, these are not currently
included in the database. However, there are a number of important
data collections without publicly available indexes, so the ACToR user
interface provides URL links to allow the user to search these
databases on a chemical-by-chemical basis based on CASRN and/or
name. TOXNET and its component databases are the main data
collections in this category. In addition to compiling data from other
databases, selected tabular information from the primary toxicology
literature is also being captured in ACToR.
Software aspects. The ACToR database is implemented using
MySQL. Software to preprocess and load data is written in Perl and
the web interfaces are written in Java. The use of 100% open-source
software will allow the entire system to be easily distributed to other
interested groups.
Search and browsing. The current version of the database allows
one to browse by data collection or assay, and to search chemicals by
name and structure using a chemical drawing applet and standard
chemical similarity algorithms.
Results
Table 2 gives summary statistics on the current composition of the
database. As already mentioned, the vast majority of substances come
from PubChem, although the overlap of that set with chemicals of
environmental interest is relatively small. ACToR contains all sub-
stances, compounds and assay results from PubChem, but the table
only gives counts for chemicals that can be indexed by CAS registry
number, which yields just over 500,000 unique or generic chemicals.
To illustrate the utility of ACToR, we show how the aggregated data
can be used to evaluate sets of chemicals for use in developing and
validating toxicology signatures for a screening and prioritization
approach. This approach is more fully explored elsewhere (Judson
et al., in preparation). Our focus is on environmental chemicals having
sufficiently high production and use volumes such that there is
potential for human and ecological exposure. The sets of chemicals
used are summarized in Table 3. Because of overlaps between these
lists, the current total number of generic chemicals considered is
11,139. This exact number will fluctuate over time because the
chemicals included in the lists periodically change due to altered
use-patterns, introduction of new chemicals, and discontinuation of
use of others. On average, each of these chemical substances has
information derived from 2-3 sources, although some of chemicals are
found in a dozen or more data collections. Of these chemicals, 7512
have an associated chemical structure and C1D assigned. Of the
chemicals without a structure, many are mixtures or complex
substances (e.g., mica, milk, mink oil and molasses, all of which are
pesticide inert ingredients).
The primary in vivo toxicology assays (either tabulated or not) are
those derived from National Toxicology Program (NTP) studies and from
ToxRefDB. The majority of data currently in the ToxRefDB database, a
component of the larger ACToR system, contains summary results of
primary toxicology studies submitted to the EPA on pesticide active
ingredients (Martin et al., 2007). Typically these data have been
extracted from EPA Office of Pesticide Programs (OPP) evaluations of
studies based on EPA Office of Prevention, Pesticides and Toxic
Substances (OPPTS) harmonized test guidelines (http://www.epa.gov/
opptsfrs/home/guidelin.htm). ToxRefDB captures details of study design
and dose series data from the areas of histopathology, clinical chemistry,
hematology, gross anatomy, pathology (neoplastic and non-neoplastic),
urinalysis and mortality. Data is aggregated at the level of animal
treatment group (dose and time). Summary data from ToxRefDB is
entered into the ACToR assay tables. NTP primary tabular data is also
being entered into ToxRefDB for chemicals not covered by OPPTS
sources. The DSSTox program has indexed all of the studies in the NTP
database by chemical and study type, and this index is in ACToR.
The secondary in vivo toxicology study data is derived from Risk-
Based Concentrations (RBC); WHO Classifications of Pesticide
Table 3
Sources of lists of environmental chemicals
Data Collection
Chemicals
HPV — High Production Volume chemicals produced or 2810
imported in quantities > 1 M Ib/year
http://www.epa.gov/hpv/pubs/update/hpvchmlt.htm
IUR — Inventory Update Rule, chemicals produced or 5375
imported in quantities > 10,000 Ib/year (2002 list)
(also referred to as MPVs or medium Production volume chemicals)
http://www.epa.gov/oppt/iur/tools/data/2002-comp-chem-records.htm
Pesticide active ingredients including anti-microbials and food-use 3476
pesticides http: //www.epa.gov/pesticides/factsheets/registration.htm
Pesticide inert ingredients (or "other ingredients") 3850
http://www.epa.gov/opprd001/inerts/lists.html
TRI — Toxic Release Inventory provides reports of toxic chemical 577
releases and other waste management activities
http: //www.epa.gov/tri/
Drinking water chemical contaminants, disinfection by-products, 120
and chemical contaminant candidates
http://www.epa.gov/safewater/ccl/index.html
EDC — Draft list of chemicals considered for endocrine disruption 73
screening http://www.epa.gov/endo/pubs/edspoverview/index.htm
Previous
TOC
-------
R. Judson et al. / Toxicology and Applied Pharmacology 233 (2008) 7-13
Table 4
URLs for sources of data described in this article
URL
Data Source
CDC Agency for Toxic Substances and Disease Registry (ATSDR)
California EPA Proposition 65
Cancer Potency Database (DSSTox)
Chemical Abstracts Service (CAS) SciFinder
Center for the Evaluation of Risks to Human Reproduction (CERHR)
CERCLA Priority List of Hazardous Substances
eChemPortal
DrugBank
EPA Disinfection By-products Database (DSSTox)
EPA Fathead Minnow Database (DSSTox)
EPA HPV Challenge Program
EPA HPV Information System
EPA Integrated Risk Assessment System (IRIS)
EPA Pesticide Fact Sheets (Conventional Chemicals)
EPA Office of Pesticides (OPP) Inert (other) Pesticide
Ingredients
EPA Risk-Based Concentrations (RBC)
EPA ToxCast Program
European substances Information System (ESIS)
EXTOXNET Pesticide Information Profiles
FDA Everything Added to Food in the United States
FDA Maximum Daily Dose Database
Health Canada Priority Substance Lists
INCHEM Concise International Chemical Assessment Documents
INCHEM Environmental Health Criteria Monographs
INCHEM International Agency for Research on Cancer (IARC)
ITER TERA Risk Assessments
Ministry of Health Labor and Welfare (Japan) Risk Assessments
Molecular Libraries Small Molecule Repository (MLSMR)
National Toxicology Program (NTP)
NIH Molecular Libraries Roadmap
NTP llth Report on Carcinogens (RoC)
OECD Screening Information Data Sets (SIDS) for High Volume Chemicals
PubChem
TOXNET
WHO Classifications of Pesticide Hazard
http://www.atsdr.cdc.gov/toxfaq.html
http://www.oehha.ca.gov/prop65/prop65Jist/Newlist.html
http://potency.berkeley.edu, http://www.epa.gov/ncct/dsstox/sdLcpdbas.html
http: //www.cas.org/
http://cerhr.niehs.nih.gov/chemicals/index.html
http://www.atsdr.cdc.gov/cercla/051ist.html
http://webnet3.oecd.org/echemportal/ParticipatingDb.aspx
http://redpoll.pharmacy.ualberta.ca/drugbank
http://www.epa.gov/ncct/dsstox/sdLdbpcan.html
http://www.epa.gov/ncct/dsstox/sdLepafhm.html
http: //www.epa.gov/hpv/
http://www.epa.gov/ncct/dsstox/sdLhpvcsi.html
http://www.epa.gov/iris, http://www.epa.gov/ncct/dsstox/sdLiristr.html
http://www.epa.gov/opprd001/factsheets
http://www.epa.gov/opprd001/inerts/lists.html
http://www.epa.gov/reg3hwmd/risk/human/index.htm
http: //www.epa.gov/comptox/toxcast/
http://ecb.jrc.it/esis
http: //extoxnetors tedu
http://vm.cfsan.fda.gov/-dms/eafus.html
http://www.epa.gov/ncct/dsstox/sdLfdamdd.html
http://www.hc-sc.gc.ca/ewh-semt/contaminants/exis tsub/categor/_result_substance/index_e.html
http://www.inchem.org/pages/cicads.html
http://www.inchem.org/pages/ehc.html
http://www.inchem.org/pages/iarc.html
http: //www. tera.org
http://wwwdb.mhlw.go.jp/ginc/html/dbl.html
http://mlsmr.glpg.com/MLSMR_HomePage/project.html
http://ntp.niehs.nih.gov/, http://www.epa.gov/ncct/dsstox/sdLntpbsi.html
(http://nihroadmap.nih.gov/molecularlibraries/)
http://ntp.niehs.nih.gov/ntpweb/index.cfm7objectid-035E5806-F735-FE81-FF769DFE5509AFOA
http://www.chem.unep.ch/irptc/sids/OECDSIDS/indexcasnumb.htm
http: //pubchem.ncbi.nlm.nih.gov
http://toxnet.nlm.nih.gov
http://www.inchem.org/documents/pds/pdsother/class.pdf
Hazards; the Cancer Potency Database (CPDB) (Richard et al., 2006);
the EPA Fat Head Minnow Database (Russom et al., 2007); the FDA
Maximum Daily Dose Database (Matthews et al., 2004); and IRIS
(Integrated Risk Information System) (Richard et al., 2007). With the
exceptions of RBC and the WHO pesticide data, the data for these sets
are taken from the DSSTox database. Web site URLs for all of the data
sources used for this analysis are given in Table 4.
The category of in vivo toxicology (summary calls) describes
sources where experts have reviewed the toxicology literature and
have made a definitive statement about a particular chemical and
endpoint, for instances labeling a chemical as a proven human
carcinogen. Data sources for this category are California EPA
Determination of Cancer and Developmental Risks (Proposition 65);
CERCLA Priority List of Hazardous Substances; the FDA Everything
Added to Food in the United States List (EAFUS); Health Canada
Priority Substance Lists; EPA OPP Inert (other) Pesticide Ingredients
categories; NTP llth Report on Carcinogens (RoC); and the Disinfec-
tion By-products Database (Woo et al., 2007).
Data sources included under Toxicity Summary Reports on the
Web are Cancer Potency Database; National Toxicology Program (NTP)
reports (Burch et al., 2007); IRIS; EPA HPV Information System; CDC
Agency for Toxic Substances and Disease Registry (ATSDR); Center for
the Evaluation of Risks to Human Reproduction (CERHR); DrugBank;
EPA Pesticide Fact Sheets; EXTOXNET Pesticide Information Profiles;
INCHEM Concise International Chemical Assessment Documents
(CICAD); INCHEM Environmental Health Criteria Monographs (EHC);
INCHEM International Agency for Research on Cancer (IARC); ITER
TERA Risk Assessments; Ministry of Health Labor and Welfare (Japan)
Risk Assessments; NTP llth Report on Carcinogens (RoC); and OECD
Screening Information Data Sets (SIDS) for High Volume Chemicals;
ESIS (European chemical Substances Information System) and its
subsets ESIS HPV, ESIS LPV (low production volume); ESIS PBT
(Persistent Bioaccumulating Toxins), and ESIS ORATS (Online Eur-
opean Risk Assessment Tracking System). Note that some sources of
data are included in multiple assay categories.
Table 5 summarizes the amount of toxicology information that we
have currently captured in ACToR, selected by assay categories for the
set of 11,139 environmental chemicals being analyzed. About half of
these chemicals have some publicly available toxicology data within
the sets of information we have currently compiled. Primary in vivo
toxicology data is available for 1447 chemicals (13%), and secondary in
vivo toxicology data is available for a total of 1405 chemicals (13%). A
total of 5205 chemicals (47%) have one or more summary in vivo
toxicity calls or determinations, which are derived by experts who
have curated data from the primary scientific literature. Finally 5244
chemicals (47%) have one or more summary text reports on chemical
toxicity available on the web. However, many of these (especially from
the ESIS LPV list) simply state that no hazard or toxicology information
is available for that chemical. We emphasize once again that these are
conservative numbers as there are still large collections of data yet to
Table 5
(Number of chemicals)x(number of assays) in ACToR for the 11,139 environmental
chemicals being analyzed
Assays 0 1 2 3 4 5 >5
In vivo toxicology (study listing primary) 9692 686 341 85 42 51 242
In vivo toxicology (tabular secondary) 9734 861 240 163 91 43 7
In vivo toxicology (summary calls) 5934 2776 950 419 258 184 618
In vivo toxicology (summary report via URL) 5895 2706 1121 573 344 203 297
Previous
TOC
-------
12
R.Judson et al. / Toxicology and Applied Pharmacology 233 (2008) 7-13
Table 6
Number of the ToxCast Phase 1308 chemicals for which data is captured in ACToR from
primary guideline studies or from IRIS assessments for key areas of toxicology
Phenotype
Acute toxicology
Subchronic toxicology
Chronic toxicology
Carcinogenicity
Developmental toxicology
Reproductive toxicology
Immuno-toxicology
Genotoxicity
Neuro-toxicology
ToxRefDB
0
235
274
263
274
251
0
0
0
IRIS
126
0
126
126
126
126
126
126
126
NTP
0
0
38
38
4
0
1
60
0
Total
126
235
291
285
290
Til
126
147
126
% Coverage
41
77
95
93
95
91
41
78
41
be compiled and loaded into the database. The bottom line, though, is
that there is little detailed in vivo toxicology information for the
majority of these environmental chemicals.
The EPA ToxCast program is a major driver of the development of
the ACToR system. The goal of ToxCast is to develop and test methods
for chemical screening and prioritization by linking the results of in
vitro assays to in vivo toxicity data (Dix et al., 2007). The most rigorous
chemical toxicity testing data are derived from whole animal human
health guideline studies, many of which are being captured in
ToxRefDB and ACToR, initially for a set of 308 unique chemicals that
are being used in Phase 1 of ToxCast. OPP-required guideline studies
and NTP studies are the primary source of the in vivo data that will be
used in ToxCast. A secondary source is the data from IRIS assessments.
Table 6 shows the number of chemicals in the ToxCast Phase 1 set of
308 that have data from OPP guideline studies, from NTP or from IRIS
for a set of key areas including acute toxicity, subchronic toxicity,
chronic toxicity, carcinogenicity, developmental toxicity, reproductive
toxicity, immunotoxicity, neurotoxicity and genotoxicity. As one can
see, there is currently good coverage for many of these areas, but
several will require searches through other ACToR-catalogued data
collections in order to build the complete analysis data set.
Discussion
This paper briefly describes ACToR (Aggregated Computational
Toxicology Resource), which is a set of linked databases and analysis
tools that aggregate a large number of data sets of relevance to
environmental chemicals and toxicology. The utility of the system was
illustrated with an example showing the amount of data available from
multiple sources that can be used for developing training and validation
sets for high-throughput chemical screening and prioritization efforts.
ACToR is not alone in its goal of aggregating large sets of chemical
structure and assay data. PubChem is the largest effort currently
available, with information on more than 10 M unique chemical
compounds. PubChem currently focuses on aggregating data from in
vitro HTS assays as the primary data repository for the MLSCN.
PubChem allows more generalized types of assay data to be submitted
and displayed, but their query engine is not tailored to the types of
custom toxicology-based queries needed for our purposes. However,
their underlying data model maps easily into the ACToR application
and serves as a useful model for our internal data organization. This
has allowed us to import all of the PubChem data and easily integrate
it with other data sources. Another important comparison is with
TOXNET which is a collection of multiple data sources covering many
aspects of chemical toxicity. TOXNET has a common search engine that
allows the user to easily find data from multiple sources. However, it is
a closed system which does not allow a user to pull together datasets
that are useful for computational purposes. One unique aspect of the
ACToR system is that it is pulling together the data from PubChem
(focused on chemical structure and HTS in vitro assay data) and
TOXNET (focused on in vivo toxicology data) and combining it in a way
that it can be used for computational analysis. We are in the process of
extracting selected tabular data from TOXNET to include directly into
the ACToR database.
eChemPortal is an Organization for Economic Co-operation and
Development (OECD) effort very similar to ACToR. It is aggregating
information on HPVs and pesticides among others. eChemPortal
currently contains links to 7 large database systems, some of which
contain what in ACToR are multiple individual databases (e.g.,
1NCHEM contains 11 individual databases). Unlike eChemPortal,
which provides links to web pages for the component databases,
ACToR extracts tabular data from the individual sources and makes it
searchable in an aggregated fashion. A system called Vitic is being
developed as a collaboration between 1UCLID and a number of
pharmaceutical companies with the goal of being an international
toxicology information center (Judson et al., 2005). Finally, the
European substances Information System or ESIS provides links to a
number of databases including EPA HPV, 1UCL1D and E1NECS
(European INventory of Existing Commercial chemical Substances).
The CEBS (Chemical Effects in Biological Systems) project at the N1EHS
is constructing a multi-domain information repository to hold the
detailed results and summaries of in vivo and in vitro toxicology
experiments (Waters et al., 2003).
We have made use of several reviews of the toxicology data
landscape to select data collections to be included in ACToR. Yang et al.
have recently published two such reviews (Yang et al., 2006a; Yang
et al., 2006b). In 2001 and 2002, a pair of review collections was
published surveying the landscape of toxicity data available on the
internet (Brinkhuis, 2001; Poore et al., 2001; Felsot, 2002; Junghans
et al., 2002; Patterson et al., 2002; Polifka and Faustman, 2002;
Russom, 2002; Winter, 2002; Wolfgang and Johnson, 2002; Young,
2002; Richard and Williams, 2003).
We foresee several key uses for ACToR. One is the derivation of
training and validation data sets for ToxCast and other chemical
screening and prioritization efforts. The second is to serve as a unique
resource for researchers developing fully computational models linking
chemical structure with in vitro and in vivo assays. Third, this large
structure-searchable database can be a valuable resource for reviewers
within the EPA and other regulatory agencies who are examining new
chemicals submitted for marketing approval. Reviewers can use the
system to search for structural analogs of the novel compounds and, if
available, easily locate potentially informative in vitro bioassay and in
vivo toxicology data on related compounds. This can, in turn, inform
their decisions on the novel chemicals under review.
ACToR is a rapidly evolving system. Future developments will
involve bringing in additional sources of information; extraction of
tabular data from on-line text documents linked to chemicals;
addition of more curated chemical structures; and the construction
of a more flexible query and data export interface. Additionally,
this system will allow the construction of workflow processes for
prioritization of data capture, quality control and chemical
prioritization scoring. The system is currently being used at the
EPA to support the ToxCast chemical screening and prioritization
program. We are working towards public release of the system in
2008.
Conflict of interest disclosure statement
The authors declare that they have no conflicts of interest.
References
Allanou, R., Hansen, B., van det Bill, Y., 1999. Public availability of data on EU high
production volume chemicals. http://ecb.jrc.it/documents/Existing-Chemicals/
PUBLIC_AVAILABILITY_OF_DATA/.
Applegate, J., Baer, K., 2006. Strategies for closing the data gap. http://ecb.jrc.it/
documents/Existing-Chemicals/PUBLIC_AVAILABILITY_OF_DATA/.
Austin, C.P., Brady, L.S., Insel, T.R., Collins, F.S., 2004. NIH molecular libraries initiative.
Science 306,1138-1139.
Birnbaum, L.S., Staskal, D.F., Diliberto, J.J., 2003. Health effects of polybrominated
dibenzo-p-dioxins (PBDDs) and dibenzofurans (PBDFs). Environ. Int 29, 855-860.
Previous
TOC
-------
R. Judsori et al. / Toxicology and Applied Pharmacology 233 (2008) 7-13
13
Brinkhuis, R.P., 2001. Toxicology information from US government agencies. Toxicology
157, 25-49.
Burch, J., Eastin, W.C., Bowden, B., Wolf, MA, Richard, A. M., 2007. DSSTox national
toxicology program bioassay on-line database structure-index locator file: SDF file
and documentation, www.epa.gov/ncct/dsstox/.
Dix, D.J., Houck, K.A., Martin, M.T., Richard, A.M., Setzer, R.W., Kavlock, R.J., 2007. The
ToxCast program for prioritizing toxicity testing of environmental chemicals.
Toxicol. Sci. 95, 5-12.
EPA, 1998. Chemical hazard data availability study. Office of pollution prevention and
toxics, http://www.epa.gov/hpv/pubs/general/hazchem.pdf.
Felsot, A.S., 2002. WEB resources for pesticide toxicology, environmental chemistry, and
policy: a utilitarian perspective. Toxicology 173,153-166.
Guth, J., Denison, R., Saas, J., 2005. Background paper for reform no. 5 of the
Louisville charter for safer chemicals: require comprehensive safety data for all
chemicals. http://www.louisvillecharter.org/downloads/CharterBkgrdPaper5.pdf.
Judson, P.N., Cooke, P.A., Doerrer, N.G., Greene, N., Hanzlik, R.P., Hardy, C, Hartmann, A.,
Hinchliffe, D., Holder, J., Muller, L, Steger-Hartmann, T, Rothfuss, A., Smith, M.,
Thomas, K., Vessey, J.D., Zeiger, E., 2005. Towards the creation of an international
toxicology information centre. Toxicology 213,117-128.
Judson, R., Dix, D.J., Houck, K., Richard, A.M., Martin, M.T., Kavlock, R.J., Dellarce, V.,
Holderman, T, Tan, S., Carpenter, T, Smith, E., In preparation. The toxicity data
landscape for environmental chemicals.
Junghans, T.B., Sevin, I.E., lonin, B., Seifried, H., 2002. Cancer information resources:
digital and online sources. Toxicology 173,13-34.
Krewski, D.D., Acosta, J., Anderson, M., Anderson, H., Ill, J.B., Boekelheide, K., Brent, R.,
Charnley, G., Cheung, V., Green, S., Kelsey, K., Kervliet, N., Li, A., McCray, L, Meyer, 0.,
Patterson, D.R., Pennie, W, Scala, R., Solomon, G., Stephens, M., J Yager, J., Zeize, L,
2007. Toxicity testing in the twenty-first century: a vision and a strategy. National
Academies Press, Washington D.C.
Martin, M.T., Houck, K.A., McLaurin, K., Richard, A.M., Dix, D.J., 2007. Linking regulatory
toxicological information on environmental chemicals with high-throughput
screening (HTS) and genomic data. The Toxicologist CD—An official Journal of the
Society of Toxicology 96, 219-220.
Matthews, E.J., Kruhlak, N.L, Weaver, J.L, Benz, R.D., Contrera, J.F., 2004. Assessment of
the health effects of chemicals in humans: II. Construction of an adverse effects
database for QSAR modeling. Curr. Drug. Discov. Technol. 1, 243-254.
Muir, D.C., Howard, PH., 2006. Are there other persistent organic pollutants? A
challenge for environmental chemists. Environ. Sci. Technol. 40, 7157-7166.
Patterson, J., Hakkinen, P.J., Wullenweber, A.E., 2002. Human health risk assessment:
selected Internet and world wide web resources. Toxicology 173,123-143.
Polifka, J.E., Faustman, E.M., 2002. Developmental toxicity: web resources for evaluating
risk in humans. Toxicology 173, 35-65.
Poore, L.M., King, G., Stefanik, K., 2001. Toxicology information resources at the
Environmental Protection Agency. Toxicology 157,11-23.
Richard, A.M., Williams, C.R., 2002. Distributed structure-searchable toxicity (DSSTox)
public database network: a proposal. Mutat. Res. 499, 27-52.
Richard, A., Williams, C., 2003. Public Sources of mutagenicity and carcinogenicity data:
use in structure-activity relationship models. In: Benigni, R. (Ed.), QSARS of
mutagens and carcinogens. CRC Press, New York, pp. 145-173.
Richard, A.M., Gold, L.S., Nicklaus, M.C., 2006. Chemical structure indexing of toxicity
data on the internet: moving toward a flat world. Curr. Opin. Drug. Discov. Devel. 9,
314-325.
Richard, A.M., M.A., W, J., B., 2007. DSSTox EPA Integrated Risk Information System
(IRIS) toxicity review data: SDF file and documentation.
Russom, C.L., 2002. Mining environmental toxicology information: web resources.
Toxicology 173, 75-88.
Russom, C.L, Williams, C.R., Stewart, T.W, Swank, A.E., Richard, A.M., 2007. DSSTox EPA
Fathead Minnow Acute Toxicity Database (EPAFHM): SDF files and documentation.
www.epa.gov/ncct/dsstox/.
Waters, M., Boorman, G., Bushel, P., Cunningham, M., Irwin, R., Merrick, A., Olden, K.,
Paules, R., Selkirk, J., Stasiewicz, S., Weis, B., Van Houten, B., Walker, N., Tennant, R.,
2003. Systems toxicology and the Chemical Effects in Biological Systems (CEBS)
knowledge base. EHP. Toxicogenomics. Ill, 15-28.
Winter, C.K., 2002. Electronic information resources for food toxicology. Toxicology 173,
89-96.
Wolfgang, G.H., Johnson, D.E., 2002. Web resources for drug toxicity. Toxicology 173,
67-74.
Woo, Y.T., Williams, C.R., Fields, N., Richard, A.M., 2007. DSSTox EPA Water Disinfection
By-Products with Carcinogenicity Estimates Database (DBPCAN): SDF files and
documentation.
Yang, C., Benz, R.D., Cheeseman, M.A., 2006a. Landscape of current toxicity databases
and database standards. Curr. Opin. Drug. Discov. Devel. 9,124-133.
Yang, C., Richard, A.M., Cross, K.P., 2006b. The art of data mining the minefields of
toxicity databases to link chemistry to biology. Curr. CompuL-Aided. Drug. Dis. 2,
135-150.
Young, R.R., 2002. Genetic toxicology: web resources. Toxicology 173,103-121.
Previous
TOC
-------
Journal of Andrology, Vol. 29, No. 3, May/June 2008
Copyright © American Society of Andrology
of and in
a
JULIA S. BARTHOLD,* SUZANNE M. MCCAHAN,* AMAR V. SINGH,t THOMAS B. KNUDSERf
XIAOLI SI,* LIAM CAMPION,* AND ROBERT E. AKINS*
From the * Nemours Biomedical Research and Division of Urology, A.I. duPont Hospital for Children, Wilmington,
Delaware; and the ^National Center for Computational Toxicology, US Environmental Protection Agency, Research
Triangle Park, .North Carolina,
ABSTRACT: Development of the fetal gubernaculum is a prereq-
uisite for testicular descent and dependent on insulin-like 3 and
androgen, but knowledge of downstream effectors is limited. We
analyzed transcript profiles in gubernaculum and testis to address
changes occurring during normal and abnormal testicular descent in
Long Evans wild-type (wt) and cryptorchid (orl) fetuses. Total RNA
from male wt and orl gubernacula (gestational days [GD] 18-20), wt
female gubernacula (GD18), and testis (GD17 and 19) was
hybridized to Affymetrix GeneChips, Statistical analysis of temporal,
gender, and strain-specific differences in gene expression was
performed with the use of linear models analysis with empirical
Bayes statistics and analysis of variance {gubernaculum) and linear
analysis (testis). Overrepresented common gene ontology functional
categories and pathways were identified in groups of differentially
expressed genes with the Database for Annotation, Visualization,
and Integrated Discovery. Transcript profiles were dynamic in wt
males between GD18-19 and GD20, comparatively static in orl
GD18-2G gubernaculum, and similar in wt and orl testis. Functional
analysis of differentially expressed genes in wt and orl gubernaculum
identified categories related to metabolism, cellular biogenesis, small
GTPase-mediated signal transduction, cytoskeleton, muscle devel-
opment, and insulin signaling. Genes involved in androgen receptor
signaling, regulated by androgens, or both were overrepresented in
differentially expressed gubernaculum and testis gene groups.
Quantitative reverse transcription polymerase chain reaction (RT-
PCR) confirmed differential expression of genes related to muscle
development, including Myog, Tnnt2, Fst, Igf1, IgfbpS, Id2, and
Msx1. These data suggest that the orl mutation results in a primary
gubernacular defect that affects muscle development and cytoskel-
etal function and might alter androgen-regulated pathways.
Key words: Gubernaculum, undescended testis, gene expres-
sion profiling, fetus.
J Androl 2008;29:352-366
Cryptorchidism, or undescended testis, is one of the
most common congenital anomalies in humans,
occurring in 2%-3% of all boys (Barthold and
Gonzalez, 2003). The cause of nonsyndromic crypt or-
chidism, in most cases, is unknown; however, the
prevalence of sporadic and familial nonsyndromic
cryptorchidism supports multifactorial susceptibility
on the basis of contributions from specific genetic loci
interacting with environmental factors, which could
include endocrine-disrupting chemicals having antian-
drogenic, estrogenic, or both effects (Mahood et al,
2006). Genetic contributions to cryptorchidism are not
well understood, but the anomaly might be inherited in
as many as 25% of cases, with autosomal dominant
inheritance being the most common pattern and the
Supported by NTH grant P20 RR-020173-01. The authors have
nothing to disclose.
Correspondence to: Dr Julia Barthold, Division of Urology, A.I.
duPont Hospital for Children. 1600 Rockland Rd, Wilmington, DE
19803 (e-mail: jbarthol@nernours.org).
Received for publication August 30. 2007; accepted for publication
December 17, 2007.
DOI: 10.2164/jandrol. 107.003970
mean hcritability in first-degree male relatives calculated
to be .67 (Czeizel et al, 1981; Elert et al, 2003).
Completion of testicular descent in mammals is
dependent on the gubernaculum, which is an append-
age of the anterior abdominal wall comprising a core
of mesenchymal cells with associated extracellular
matrix and localized striated muscle (Radhakrishnan
et al, 1979; Costa et al, 2002). In the rat fetus, the
gubernaculum is visible at gestational day 14 (GDI4)
in both sexes (Radhakrishnan et al, 1979). The female
gubernaculum contains both mesenchymal and poorly
organized muscle cells, and further growth fails to
occur after GDI6. In males, the gubernaculum enlarges
after GD16, increases dramatically in size between
GDIS and 20, then becomes exteriorized by everting
into an extra-abdominal location around the time of
birth (GD22). The mesenchymal portion of the
gubernaculum disappears, leaving an outer layer of
muscle, which persists as a sac of cremaster muscle and
surrounds the scrota! testis. Eversion of the gubernac-
ulum-cremaster complex occurs rapidly, but the mech-
anisms controlling its development and motility are
poorly understood.
352
-------
Barthold et al • Gene Expression in Fetal Gubernaculum and Testis of Cryptorchid Rats
353
In vitro studies of gubernacular development and
phenotypic analysis of cryptorchid genetic mouse
models suggest that the testis is required for proper
development of the ipsilateral gubernaculum and
implicate secretion of the Leydig cell hormones insu-
lin-like factor 3 (InsO) and, to a lesser degree,
testosterone (Emmen et al, 2000) in gubernacular
development. Targeted deletion of either InslB or Rxfp2
is associated with high intra-abdominal testes in
homozygous male mice and delayed testicular descent in
heterozygotes (Zimmermann et al, 1999; Overbeek et al,
2001). Development of the fetal gubernaculum is femi-
nized in homozygous InsBIRxfp2 mutants (Tomiyama et
al, 2003). By contrast, mice and rats with spontaneous
androgen receptor defects or that have been exposed to
the antiandrogen flutamide (Spencer et al, 1991) show a
milder phenotype. Although a model of testicular descent
separates INSL3- and androgen-dependent phases into
distinct events (Hutson and Hasthorpe, 2005), both
hormones stimulate proliferation of fetal gubernacular
cells. Moreover, generalized expression of both the
INSL3 receptor RXFP2 (relaxin/insulin-like family re-
ceptor peptide 2, also known as LGR8 or GREAT) and
the androgen receptor is present in the fetal gubernaculum
(Emmen et al, 2000; Scott et al, 2005). Canonical InsBI
Rxfp2 signaling involves the cAMP/protein kinase A
(PKA) pathway via activation of the cAMP response
element (CRE; Halls et al, 2005), but information
regarding downstream effectors is limited.
The Long Evans orl rat strain is an inbred colony at
high risk for spontaneous cryptorchidism (Mouhadjer et
al, 1989). Approximately two-thirds of offspring are
affected, and up to 75% of cases occur unilaterally, with
the left side more frequently affected (unpublished
observations); overall, approximately 35%^tt% of
testes fail to descend (Barthold ct al, 2006). The orl
gubernaculum is reduced in size between GDIS and 20,
but the testis descends normally during this time
(Barthold et al, 2006). By the first day of life, however,
normal eversion fails to occur in about half of orl
gubernacula, and subsequent aberrant lateral migration
occurs with final localization of the ipsilateral testis in
the superficial inguinal pouch, anterior to the rectus
muscle. This is a unique animal model of cryptorchidism
in that the phenotype is similar to that seen most
commonly in the human population.
Because of the complexity of genetic pathways and
their interactions that are known in the various models of
cryptorchidism, a comprehensive screen of transcript
profiles is useful to address the changes associated with
the gubernaculum and testis in orl rats. In this study, we
use microarray analysis to study gene expression in the
developing fetal gubernaculum and testis. Our data
indicate that expression of genes involved in energy
Table 1 . Samples
Gubernaculum
wt
W18a-tf
W18a-d
W19a-d
WL20a-d
WR20a-c
orl
O18a-d
O19a-d
OL20a-d
OR20a-c
Testis
wt
W17a-e
W19a-e
orl
O17a-e
O19a-e
used for microarray
Gestational Age, d
18
18
19
20
20
18
19
20
20
17
19
17
19
analysis
Side
Right + left
Left
Left
Left
Right
Left
Left
Left
Right
Litter No,
1
2
3
4
4
5
6
7
7
8
3
9
6
Abbreviations; wt, Long Evans wild-type strain; orl, Long Evans
cryptorchid strain.
a Female (all others male),
pathways and in the functionally related categories of
muscle development, cytoskeleton organization and
biogenesis, and small GTPase-mediated signal transduc-
tion is altered in normal and orl fetuses during prenatal
growth of the gubernaculum. By contrast, we found fewer
strain-specific differences in fetal testicular gene expres-
sion, suggesting that genetic variants with gubernaculum-
specific effects predispose orl rats to cryptorchidism.
ancf
Animals
Breeding colonies of orl and wt rats were maintained in a reverse
light cycle room following protocols approved by the institu-
tional Animal Care and Use Committee. Female estrus cycles
were assessed with vaginal smears, and animals were mated in
the afternoon to generate timed pregnancies, which were
identified by visualization of sperm in vaginal smears the
following day, defined as GDI. Pregnant females were
euthanatized via CC>2 inhalation during the late morning of
GDI7-20. The caudal half of each fetus was immediately
collected in RNAlater (Applied Biosystems, Foster City,
California) and stored at 4°C for at least 24 hours to facilitate
microdjssection. Fetal testes and gubernacula were separated by
microdissection, and samples were collected as noted in Table 1.
For GDIS 19, left-sided samples were used because of a higher
incidence of left cryptorchidism observed in the orl strain
(unpublished observations) and the possibility of intrinsic left-
right asymmetry in males. At GD20, both left and right
gubernacula were removed from 3 fetuses. In females, left and
right gubernacula from the same fetus were pooled.
-------
354
Journal of Andrology • May/June 2008
RNA Extraction
Total RNA was purified from single gubernacula or testes with
the RNeasy Mini Kit (Qiagen, Valencia, California) and the
RNase-free DNase Set (Qiagen). RNA was quantified on the
basis of A26o with the use of an ND-1000 Ultraviolet-visible
spectrophotometer (NanoDrop Technologies, Wilmington,
Delaware). Overall integrity of the total RNA was verified with
a 2100 Bioanalyzer (Agilent Technologies, Santa Clara,
California) before processing for microarrays to assure consis-
tency across samples.
Microarray Sample Processing
RNA samples (Table 1) were assessed with Affymetrix Rat
Expression Array 230A (Affymetrix, Santa Clara, California).
This microarray contains 15 866 probe sets representing
approximately 10 500 genes and 2700 ESTs. Some genes are
represented by more than 1 probe set. Whereas NCBI Entrez
Gene lists approximately 38 000 genes for Rattus norvegicus, the
230A GcneChips interogates nearly one-third of known rat
genes. For testes, 1 p,g of total RNA from single organs was
labeled with the One-Cycle cDNA Synthesis Kit (Affymetrix).
This involved cDNA synthesis followed by in vitro transcription
with T7-RNA polymcrase and biotinylatcd nucleotidc. Because
of the smaller yield of RNA from gubernacula, 30 ng of total
RNA from single organs was amplified and labeled with the
GeneChip Two-Cycle cDNA Synthesis Kit (Affymetrix). After
cDNA synthesis and in vitro transcription with T7-RNA
polymerase, the resulting cRNA was used as a template for a
second round of cDNA synthesis, which was followed by in
vitro transcription in the presence of biotinylated nucleotide.
Biotinylated cRNA was hybridized to 230A GeneChips. Arrays
were washed, stained with strepavidin phycoerythrin conjugate,
and scanned at DuPont Haskell Laboratory for Health and
Environmental Sciences in a Hybridization Oven 640 (Affyme-
trix), GeneChip Fluidics Station (Affymetrix), and GeneArray
Scanner (Affymetrix) with Affymetrix protocols and reagents.
Standard Affymetrix quality control measures were consistent
across all hybridizations of the same tissue type.
Analysis of Microarray Data
As a measure of the quality of hybridization, the raw and
normalized probe intensity distributions for each GeneChip
were determined with histogram plots within the AffylmGUI
interface for the limma package of Bioconductor (Wettenhall
et al, 2006). Representations of 5' and 3' regions of transcripts
in the labeled cRNA were examined with the Affy package of
Bioconductor to verify consistency within tissue types.
Expression values were calculated with the MAS 5 algorithm
from raw probe intensities using GCOS (Affymetrix). Expres-
sion values were also calculated with the GC robust multiarray
average (GC-RMA) algorithm (Wu et al, 2004) from raw
probe intensities within AffylmGUI. All further analyses were
performed with the GC-RMA expression values unless
otherwise noted. Global gene expression patterns and overall
variability between samples were examined by principal
component analysis (PCA), which was performed in MeV
version 3.1 (Saeed et al, 2003). Two methods were used to
identify differentially expressed genes: 1) the LIMMA linear
models approach with the empirical Bayes statistic (B & 3) and
the multiple testing adjustment method of Holm, used within
AffylmGUI (referred to as linear analysis), and 2) calculation of
MASS expression ratios using GDIS wt as a reference
denominator followed by the scaling of Iog2-transformed values
to a median of 0.0 and standard deviation of 0.50 with statistical
analysis in GeneSpring version 7.2 using analysis of variance
with a Benjamini and Hochberg false discovery rate of .001
(referred to as reference denominator method). Differentially
expressed genes were filtered for a GC-RMA average expression
value of greater than or equal to 50 in at least 1 sample group in
the analysis and then separated into groups according to the
Iog2 ratio of average expression values for groups of samples.
Some groups of genes were further separated with K-means
clustering in MeV. Plots of expression profiles were created with
the statistical package R (http://www.r-projcct.org/). Gene
groups with mean expression levels as well as the entire data
set are available via Accession number GSE7755 at the
Gene Expression Omnibus (http://www.ncbi.nlm.nih.gov/geo).
Groups of differentially expressed probe sets were examined for
statistical overrepresentation of Gene Ontology (GO) Biological
Function categories and biological pathways as defined in the
Kyoto Encyclopedia of Genes and Genomes (KEGG; http://
www.genome.jp/kegg/) with the Database for Annotation,
Visualization, and Integrated Discovery (DAVID) 2007 (Dennis
et al, 2003; http://david.abcc.ncifcrf.gov/).
Real-Time Reverse Transcription Polymerase
Chain Reaction
Real-time reverse transcription polymerase chain reaction (RT-
PCR) was used to validate trends in selected array-derived data
from gubernaculum. cDNA was synthesized from 150 ng of total
RNA (n > 6 samples per group) with the High-Capacity cDNA
Archive Kit (Applied Biosystcms). Amplifications were per-
formed in triplicate using TaqMan Gene Expression Assays (see
Table 2 for details) and TaqMan Universal PCR Master Mix in
an ABI Prism 7900HT. Levels of target mRNA expression were
determined by the 2~~AACT method (Livak and Schmittgen, 2001)
with tripeptidyl peptidase 2 (Tpp2) as control and total rat
embryonic RNA (Agilent Technologies) as calibrator. Nonpara-
mctric statistical analyses of differences between strains were
performed in SPSS (version 14.0; SPSS Tnc) as indicated.
Global Gene Expression
PCA analysis using all probe sets on the RAE 230A array
was performed to examine global trends in gene
expression for gubernaculum and testis samples (Fig-
ure 1). Gubernaculum samples (Table 1) clustered into 4
main groups: 1) GDIS females, 2) GDIS 19 wt males, 3)
GD20 wt males, and 4) 13 of the 15 orl samples. The 2
remaining orl samples, OlSa and O19d, clustered with
GD20 and GDI8-19 wt males, respectively. These
samples were considered outliers and were excluded from
-------
Barthold et al • Gene Expression in Fetal Gubernaculum and Testis of Cryptorchid Rats
355
Table 2. Real-time
ABI Assay
Rn00432087_m1
Rn00673944_m1
Rn00518185_m1
Rn01 494289 ml
Rn01495280_m1
Rn00710306_m1
Rn00563116_m1
Rn00667535 ml
Rn00567418_m1
Rn01399583_m1
Rn01438455_m1
Rn01483694_m1
Rn01 43741 0_m1
Rn00584577_m1
reverse transcription polymerase
Gene Symbol
Bmp4
Des
Dusp6
Fst
Id2
Igf1
IgfbpS
Msx1
Myog
Nfkbl
Olfml
Tnnt2
Tpp2
Wnt4
chain reaction assays
Gene Name
Bone morphogenetic protein 4
Desmin
Dual-specificity phosphatase 6
Follistatin
Inhibitor of DMA binding 2
Insulin-like growth factor 1
Insulin-like growth factor-binding protein 5
Homeobox, msh-like 1
Myogenin
Nuclear factor of kappa light chain gene enhancer in B-cells 1 , p105
Olfactomedin 1
Troponin T2, cardiac
Tripeptidyl peptidase II
Wingless-related MMTV integration site 4
Abbreviation: MMTV, mouse mammary tumor virus.
further analysis. Linear analysis of left vs right GD20
samples failed to reveal significant differences; however,
further analyses were limited to left-sided samples. Testis
samples clustered into GDI? and 19 groups with
intermixing of the 2 strains at each time point. After
initial global data analysis, our overall strategy was to 1)
identify expression profiles of developmentally regulated
genes in a defined window of gestation in wt males, 2)
identify genes differentially expressed between strains,
and 3) analyze these groups of genes functionally.
Expression Profiles in Normal and Cryptorchid Strains
Gubernaculum—We studied changes in gene expression
across normal development of the gubernaculum by
comparison of the 3 male wt groups (12 samples). The
W18&19.-, . W20
' '"
gene expression profile in GDI8-19 samples was
remarkably similar, with only 11 differentially expressed
genes identified. By contrast, comparison of GDIS and
20 wt samples returned 1023 probe sets that were
differentially regulated, suggesting a major switch in
gene expression between GD18-19 and GD20. With the
use of K-means clustering, we further classified these
genes into 2 lists with declining expression (n = 371) or
increasing expression (n = 652) after GDIS (Figure 2).
The expression profiles show marked sexual dimor-
phism of these genes at GDIS, suggesting that they
participate in male-specific gubernacular development.
Surprisingly, by GD20, the expression profile in normal
males approximates that of GDIS females, as we
observed in PCA analysis (Figure 1).
B
GD17
GD19
Figure 1. Principal component analysis (PCA) of all gubernacular (A) and testicular (B) samples for all probe sets on the Affymetrix Rat 230A
GeneChip. Black and gray dots represent wild-type (wt) and Cryptorchid (orl) samples, respectively. Labels refer to samples within adjacent
dotted circles except as noted. (A) For gubernaculum, all gestational day (GD)18 and 19 wt (W) samples and all but 2 orl (O) samples were
noted to cluster together. The 2 outlier orl samples cluster between GD20 wt and GD18 female (F) samples (gray dots), which are closely
aligned. (B) For testis, GD17 and 19 samples cluster together with no clear separation between strains.
Previous
TOC
-------
356
Journal of Andrology • May/June 2008
A.
in
O
w *>..
(/) 0
. ?"i
0
CM
Figure 2. Expression profiles of genes differentially expressed
between GD18 and 20 in wt gubernaculum (n = 1023). Genes are
classified on the basis of K-means clustering: (A) Increasing
expression between GD18 and 20 (n = 652) and (B) decreasing
expression between GD18 and 20 (n = 371). Error bars show mean
Z-scores ± SD for probe sets in each subgroup. Vertical lines
separate the samples by gender (F indicates wt female; W, wt male)
and gestational age in days (18-20).
Because of minimal differences between GDIS and
20, we analyzed strain-specific differences at GDIS and
20 only. Linear analysis identified 2401 probe sets that
were differentially expressed between wt and orl males at
these time points. The reference denominator method,
which takes into account all time-, gender-, and strain-
specific differences in expression relative to GDIS,
returned 3707 probe sets. After filtering for low
expression, these lists were combined to generate a final
list of 3589 probe sets associated with wt versus orl
differences.
Log2 ratio of expression values was used to divide the
final list into groups of genes with either higher (orl-
high, n = 1681) or lower (orl-low, n = 1908) expression
in orl relative to wt samples. With the use of K-means
clustering, we identified 3 major expression profiles in
each of these groups (Figure 3). The observed pattern
suggests little change in expression of differentially
regulated genes in orl fetuses between GDIS and 20.
Interestingly, a tendency toward reciprocal patterns of
female, wt, and orl expression is seen in the correspond-
ing clusters from each group. Two clusters (cluster 1 of
each group) show the greatest differences between
normal males and females at GDIS. It is noteworthy
that the corresponding orl expression profiles in these
clusters are feminized.
Testis—Comparable linear analysis of GDI7 and 19
wt samples identified fewer (n = 818) differentially
regulated genes at these 2 time points in testis compared
with gubernaculum. More significantly, few genes were
differentially expressed in the testicular samples of wt
compared with orl fetuses. Linear analysis of strain
differences at GDI7 or 19 yielded only 349 differentially
expressed probe sets when combining the 2 lists. Of
these, expression was higher in wt males in 248 sets and
lower in 101.
Functional Annotation of Differentially Expressed Genes
We performed functional analysis of groups of genes
using DAVID and analyzed common GO Biological
Process annotations. Two groups of probe sets were
analyzed separately: genes differentially expressed be-
B.
Q.
fi^
Cluster 1
Cluster 2
c ^
o
'«
w o-
X
UJ
?tt.
?iTS
+>*
Ti^:
r*?'%
••••^•«
ft*f?t
00 00 O) O CO O)
T- i- i- CJ T-i-
LJ- o o o 55
O
OJ
CO 00 O) O CO O) O
i— i— i— CNJ i— i— OJ
u- o o o 55 5
Cluster 3
if.
t"*+
-----
ii*
t?;
.;;
:"iT
ft
tffftif
' ii !
/t-
S^
jl_
""*"
•jj
p
^
CO CO O) O 00 O) O
i— i— i— OJ i— i— C«l
u- oo o 55 5
Figure 3. Expression profiles of differentially expressed gene groups in orl and wt gubernaculum are shown: (A) Genes with higher expression
in orl (orl-high) and (B) genes with lower expression in orl (orl-low). Profiles were generated by K-means clustering of all samples in the 2 gene
groups, and the 3 major resultant clusters for each gene group are shown. Error bars show mean Z-scores ± SD for probe sets in orl-high
clusters 1 (n = 577), 2 (n = 502), and 3 (n = 448) and for orl-low clusters 1 (n = 604), 2 (n = 851), and 3 (n = 451). Vertical lines separate the
samples by gender and strain (F indicates wt female; O, orl male; W, wt male) and gestational age in days (18-20).
Previous
Next
-------
Barthold et al • Gene Expression in Fetal Gubernaculum and Testis of Cryptorchid Rats
357
Table 3. Selected functional gene ontology (GO) biological process annotations represented by differentially expressed genes3-
Gubernaculum
Testis
GO Biological Process
Metabolism
Cell organization and biogenesis
Biosynthesis
Cytoskeleton organization and biogenesis
Cell cycle
Localization
Transport
Muscle development
Generation of precursor metabolites and energy
Actin filament-based process
Small GTPase-mediated signal transduction
Cell division
Muscle contraction
Apoptosis
Phosphorylation
Transcription from RNA polymerase II promoter
Growth
Cellular morphogenesis
Development
No, of Genes
1205
392
298
104
141
502
433
44
126
48
63
34
41
121
122
127
53
91
379
P Value No, of Genes P Value
3.8 x 10~54
1.2 X 10~27
4,7 X 10~22
3.4 X 1G~10
2,3 X 10~9
4.4 X 10~8
1.8 x 10~7
2.0 X 1Q~6 10 4.9 X 1Q~4
5.7 x 10 6
1.6 x 10~5
2,1 X 10'~5
2.6 X 1Q'~5
6.7 X 10 5
1.3 x 10~4
1.6 x 10~4
2,8 x 10~4
4.0 X 10~4 H 2.7 X 10~3
5.9 X 10 4 14 2.1 X 10 2
6.3 x 10~3 60 3.3 x 1Q~5
' Modified Fisher exact P values are shown for selected overrepresented GO biological process terms identified by analysis of differentially
expressed genes in fetal gubernaculum (n = 3428 DAVID IDs) and testis (n = 347 DAVID IDs) with the use of Database for Annotation,
Visualization, and Integrated Discovery (DAVID; http://niaid.abcc.ncifcrf.gov/home.jsp).
tween wt and orl in gubernaculum (n = 3589; converted
to 3428 unique DAVID IDs) and the group differen-
tially expressed between wt and orl in testis (n = 349,
converted to 347 DAVID IDs). The analyses from each
group were compared for recurring functional themes.
Selected nonredundant categories are shown in
Table 3, and expression data for selected genes in these
groups are shown in Table 4. Multiple categories related
to general physiologic processes such as metabolism and
biosynthesis were identified. When analyzing all genes
differentially expressed between wt and orl, small GTPase
signal transduction was the most significantly represented
signaling pathway in GO. We also identified categories
related to small GTPase signaling, including muscle
development and cytoskeletal organization and biogen-
esis. These data are consistent with the known morpho-
logical changes occurring in the gubernaculum during
this time frame, including significant growth and
maturation of muscle (Radhakrishnan et al, 1979; Cain
et al, 1995). Comparatively few GO annotations were
Table 4. Selected genes differentially expressed in wt and orl fetal gubernaculunf
Probe Set ID
Muscle development
1368725_at
1368302_at
1387232_at
1374904__at
1367652__at
1375518__at
1388185_at
1386993_at
1387181_at
1388335_at
1398248_s_at
1 36831 0_at
1367600_at
1369928_at
Gene Title"
Jagged 1
Homeobox, rnsh-Iike 1
Bone morphogenetic protein 4
Sine oculis homeobox homolog 1 (drosophila)
Insulin-like growth factor-binding protein 3
Titin
Retinoblastoma 1
Myosin, heavy polypeptide 7, cardiac muscle, beta
Myogenic factor 6
Transgelin 2
Myosin, heavy polypeptide 6, cardiac muscle, alpha
Myogenin
Desmin
Actin, alpha 1, skeletal muscle0
Gene Symbol
Jagl
Msx1
Bmp4
Six1
Igfpb3
Ttn
Rb1
Myh7
Myf6
Tagln2
Myh6
Myog
Des
Actal
Log2(orl/wt)
3.96
2.44
1.9
1.88
1.94
1.36
-0.81
1.01
1.08
-1.14
-1.2
1.4
1,63
-1.7
-------
358
Journal of Andrology • May/June 2008
Table 4, Continued
Probe Set ID
1372569_at
1369375_a_at
1 367S70__at
1388298__at
1387348_at
1367628_at
Muscle contraction
1 36761 7_at
1367592_at
1370857_at
1367572_at
1370198_at
1368838__at
1368724__a_at
1371239_s_at
1387787_at
Cytoskeleton organization and
1367654__at
1387080_at
1372692_at
1370875_at
1387227_at
1383822__at
1368893__at
1371885_at
1 375881 _at
1367605_at
1399105_at
1374523_at
138846Q__at
1370184__at
13989QO__at
Small GTPase-mediated signal
1 37321 5_at
1389292__at
1367475_at
1374239_at
1388892_at
138873Q_at
1 36821 7_at
1 38971 0_at
1371255_at
1368096_at
1 372521 _at
1 37251 3_at
1398838_at
1 37706 1_at
1 37388 1_at
1370168_at
1 36804 1_at
1370130_at
1388729__at
Development
1375532_at
1 368641 _at
1377064_at
Gene Titleb
Four and a half LIM domains 3 (predicted)
Calpain 3
Transgelin
Myosin, light polypeptide 9, regulatory (predicted)
Insulin-like growth factor-binding protein 5
Lectin, galactose-binding, soluble 1
Aldolase A
Troponin T2, cardiac
Smooth muscle alpha-actin
Myosin, light polypeptide 3
Triadin
Tropomyosin 4
Tropomyosin 1 , alpha
Tropomyosin 3, gamma
Myosin, light polypeptide 2
biogenesis
Fat tumor suppressor homolog (Drosophi/a)
Chondroitin sulfate proteoglycan 6
Tyrosine kinase, non-receptor, 2
Villin 2
Wiskott-Aldrich syndrome protein-interacting protein
Bicaudal D homolog 2 (Drosophi/a)
CAP, adenylate cyclase-associated protein, 2 (yeast)
Cytoskeleton-associated protein 1 (predicted)
Destrin
Profilin 1
Bridging integrator 3
6-phosphogluconolactonase (predicted)
Capping protein (actin filament), gelsolin-like
Cofilin 1 , non-muscle
Dynactin 3 (predicted)
transduction
Active BCR-related gene (predicted)
RAB18, member RAS oncogene family
Cell division cycle 42
FERM, RhoGEF and pleckstrin domain protein 2 (predicted)
RAB2B, member RAS oncogene family
CDC42 effector protein (Rho GTPase-binding) 4 (predicted)
RalA-binding protein 1
Son of sevenless homolog 1 (Dmsophilaf
Harvey rat sarcoma viral (v-Ha-ras) oncogene homolog
RAB7, member RAS oncogene family-like 1
Rho family GTPase 2
Ras-related C3 botulinum toxin substrate 1
RAB7, member RAS oncogene family RhoGAP involved in
beta-catenin-A/-cadherin and NMDA receptor signaling
(predicted)
Rho, GDP dissociation inhibitor (GDI) beta Tyrosine 3-
monooxygenase/tryptophan S-monooxygenase activation
protein, theta polypeptide
Synaptojanin 2 binding protein
Ras homolog gene family, member A
Harvey rat sarcoma oncogene, subgroup R (predicted)
Inhibitor of DNA binding 2°
Wingless-related MMTV integration site 4
Dual-specificity phosphatase 6
Gene Symbol
Fhl3_predicted
Capn3
Tag/n
My!9_predicted
IgfbpS
Lgals 1
Aldoa
Tnnt2
Acta2
My/3
Trdn
Tpm4
Tpm1
Tpm3
My/2
Path
Cspg6
Tnk2
Vil2
Waspip
Bicd2
Cap2
Ckap 1__predicted
Dstn
Pfn1
Bin3
Pgls__predicted
Capg
cm
Dctn3_predicted
Abr predicted
Rab18
Cdc42
Farp2__predicted
Rab2b
Cdc42ep4_predicted
Ralbpl
Sos1
Hras
Rab7l1
Rnd2
Pad
Rab7
RICS_predicted
Arhgdib
Ywhaq
Synj2bp
RhoA
Rras_j)redicted
Id2
Wnt4
Dusp6
Log2(orl/wt)
-1.99
1.29
2,4
-2,52
-3,21
4,62
1,15
1,79
-2,32
-2.33
2,35
2,54
-2,57
-2,85
3,25
4,31
2,2
2,09
1.98
1.51
1,32
-1,62
-1,98
2,04
2.1
-2.22
-2,26
2,26
2,31
-2.75
2,78
2,38
2.12
1.71
1.5
1.28
1.27
0.96
-1.04
-1.58
-1,84
1,88
-2
2.07
2,15
-2,16
2,23
3,05
3,08
5.15
3.2
2.75
-------
Barthold et al • Gene Expression in Fetal Gubernaculum and Testis of Cryptorchid Rats
Table 4. Continued
359
Probe Set ID
1369008__a__at
1388856_at
1387843_at
1 370221 _at
1390119_at
1376755__at
1370747_at
1372447_at
137Q968_at
1373829__at
1368395__at
1370224_at
1375043_at
1367712__at
1388154__at
1389403__at
1386940_at
Insulin signaling pathway
1376779__at
1386950__at
1398799_at
1367573_at
1368116_a_at
1386888__at
Focal adhesion
1369955_at
1370267_at
1372905_at
1370333__a__at
1367760__at
137Q427__at
1370155_at
1389723_at
1383Q75__at
1388138__at
1371664_at
1387777_at
1386863__at
1368385__a__at
1398836_s_at
1387346_at
Other
1390638__at
1372964__at
139Q355__at
1368509_at
1389670_at
Gene Title"
Olfactotnedin 1
Kit ligand
Follistatin
WNT1 inducible signaling pathway protein 1
Secreted frizzled-related protein 2
Retinoic acid receptor, beta
Fibroblast growth factor 9
Fibroblast growth factor receptor 1d
Nuclear factor of kappa light chain gene enhancer
in b-cells 1, p105
Fibroblast growth factor receptor 2d
Glypican 3d
Signal transducer and activator of transcription 3
FBJ murine osteosarcoma viral oncogene homolog
Tissue inhibitor of metalloproteinase 1
E2F transcription factor 5
Bone morphogenetic protein 7
Tissue inhibitor of metalloproteinase 2
Forkhead box O1 A
Protein phosphatase 1, catalytic subunit, beta isoform
Eukaryotic translation initiation factor 4E
Ribosomal protein S6
Ribosomal protein S6 kinase, polypeptide 1
Eukaryotic translation initiation factor 4E binding protein 1
Procollagen, type V, alpha 1
Glycogen synthase kinase 3 beta
Vinculin (predicted)
Insulin-like growth factor 1
Mitogen-activated protein kinase 1
Platelet-derived growth factor, alpha
Procollagen, type I, alpha 2
Phosphoinositide-3-kinase, regulatory subunit 4, p150
(predicted)
Cyclin D1
Thrombospondin 4
Paxillin
Integrin-linked kinase
Protein phosphatase 1, catalytic subunit, alpha isoform
Growth factor receptor-bound protein 2
Actin, beta
Integrin beta 1 (fibronectin receptor beta)
Similar to Eph receptor A4 (predicted)"
AT-rich interactive domain 5B (MRF1-like) (predicted)06
Ryanodine receptord
Bardet-Biedl syndrome 2 homolog (human)d
Similar to homeobox protein A10 (predicted)8
Gene Symbol
Olfml
Kit!
Fst
Wispl
Sfrp2
Rarb
Fgf9
Fgfrl
Nfkbl
Fgfr2
Gpc3
Stat3
Fos
Timp 1
E2K
Bmp7
Timp2
Foxo 1a
Ppplcb
Eif4e
Rps6
Rps6kb1
Eif4ebp1
ColSal
Gsk3b
Vcl_predicted
Igf1
Map2k1
Pdgfa
Coi1a2
Pik3r4_predicted
Ccndl
Thbs4
Pxn
Ilk
Ppplca
Grb2
Actb
Itgbl
Rgd 1560587_predicted
AridSb pred
Ryr1
Bbs2
Rgd1566402_jiredicted
Log2(orl/wt)
2.57
2.31
2.09
2.04
1.96
1.93
1.57
1.55
1.53
1.26
1.11
1.06
0.9
1.07
-1.29
-1.81
2.18
1
0.72
-0.77
-1.52
2.33
2.61
2.34
1.88
1.68
1.38
1.15
1.12
1.03
1.01
0.88
0.73
0.62
0.93
1.5
-1.68
-2.76
2.97
2.46
2.26
1.85
1.18
2.1
Abbreviations: GAP, GTPase-activating protein; GDP, guanosine diphosphate; MMTV, mouse mammary tumor virus; NMDA, /V-methyl-o-
as pa rate.
a Greatest difference in Affymetrix 230A probe set mean expression levels between orl and wt samples at gestational day 18 or 20 expressed
as log,(orl/wt).
b Bold gene titles indicate that real-time reverse transcription polymerase chain reaction also was performed.
c Probe set with E annotation.
d Gene associated with human cryptorchidism according to Online Mendelian Inheritance in Man (OMIM; http://www.ncbi.nlm.nih.gov/omim/).
e Gene associated with cryptorchidism in mice.
-------
360
Journal of Andrology • May/June 2008
Tables. Selected Kyoto Encyclopedia of Genes and Genomes (KEGG) pathways represented by differentially expressed genes*
Gubernaculum
Testis
KEGG Pathway
Ribosome
Oxidative phosphorylation
Proteasome
ATP synthesis
Valine, leucine, and isoleucine degradation
Glycolysis/gluconeogenesis
Fatty acid metabolism
Insulin signaling pathway
Focal adhesion
No, of Genes
41
48
19
19
17
19
16
35
49
P Value No, of Genes P Value
1
4,
1
7,
4,
,8
,1
,9
.5
.2
4.0
4,
1
5,
,2
,7
.0
X
X
X
X
X
X
X
X
X
10
10
10
10
10
10
10
10
10
-17
-9
-9
-6
~4 5 9.6 x 10~3
-3
-3
-2
2 12 6.2 X 10 3
a Modified Fisher exact P values are shown for selected overrepresented KEGG pathway terms identified by analysis of differentially expressed
genes in fetal gubernaculum (n = 3428 DAVID IDs) and testis {n = 347 DAVID IDs) with the use of DAVID {http://niaid.abcc.ncifcrf.gov/
home.jsp).
identified in the differentially expressed testis gene group.
However, these include muscle development, which might
indicate selective effects of the mutation in myoid cells.
WTith the use of DAVID, we identified overrepresented
KEGG pathways for the 2 groups of genes differentially
expressed in wt and or! gubernaculum and testis. Selected
pathways are shown in Table 5, with the most genes
associated with focal adhesion in both gubernaculum and
testis. We separately analyzed overrepresentation of 755
known androgen-regulated and androgen-signaling path-
way genes (rittp://www.netpath.org/pathways?path_id=
NctPath_2; Bolton ct al, 2007) not included in DAVID
with a Fisher's exact test patterned after EASE (Expres-
sion Analysis Systematic Explorer) methodology (Hos-
ack et al, 2003). Of 622 present on the microarray, 199 (P
= .039) and 30 (P = .004) androgen-associated genes are
differentially expressed in gubernaculum and testis,
respectively. Together, the functional category and
pathway analyses suggest altered regulation of related
processes and pathways linked to energy and metabolism,
muscle and cytoskeleton organization, and altered
expression of androgen-regulated genes.
Expression of Genes Linked to Cryptorchidism or
Testicular Descent
We analyzed the expression patterns of candidate genes
annotated on the 230A chip. HoxalO, Epha4, and
AridSB, genes associated with cryptorchidism in mice
with spontaneous or targeted mutations (http://www.
informatics.jax.org/) are present in the list of differen-
tially expressed genes (Table 4). Of the previously
reported tyrosine kinases expressed in the fetal
GD16.5 mouse gubernaculum (Verma-Kurvari and
Parada, 2004), several failed to show significant
gubernacular expression during the interval studied
(Rous sarcoma virus [c-Src], spleen tyrosine kinase
v-Erb erythroblastic leukemia viral oncogene
homolog 4 [Erbb4], and Eph receptor B4 [Ephb4]) using
the microarray methodology. Others were expressed
during this time frame at comparable levels in females
and both male strains (platelet-derived growth factor
receptor alpha [Pdgfra], insulin-like growth factor 1
receptor [Igflr], c-src tyrosine kinase [Csk], kinase insert
domain protein receptor [Kdr, also known as Vegfr2 or
Flkl], and v-abl Abclson murine leukemia viral onco-
gene homolog 1 [Abll; data not shown]). Expression of
protein tyrosine kinase 2 (Ptk2, encoding FAK or focal
adhesion kinase) is sexually dimorphic at GDIS and
increases significantly in wt males between GDIS and 20
but is not differentially expressed between strains. None
of the functional annotation analyses of testicular gene
expression identified patterns suggestive of altered
hormone synthesis in orl fetal testis and representative
Leydig cell-specific genes, including probe sets for
isoforms of Cypllal, C'ypl7al, HsdBb, and Hsdl7b,
which were highly, and not differentially, expressed in
both strains (data not shown). Ins I3 expression was
higher in or! fetal GD17 and 19 testis, but the differences
were not significant on the basis of our linear analysis.
Comparison of Differentially Expressed Gubernaculum
and Testis Genes
Of the 349 testis genes differentially expressed between
the 2 strains, 117 were also differentially expressed in wt
compared with orl gubernaculum. Few of these genes
were down- or up-regulated in both testis and guber-
naculum of orl fetuses. These include insulin-like growth
factor-binding protein 5, interferon-induced transmem-
brane proteins 1 and 3, osteoglycin, and calcium/
calmodulin-dependent protein kinase II; delta (lower
expression in orl); and Epha4 (higher expression in orl).
Many genes encoding ECM proteins, including several
-------
Barthold et al • Gene Expression in Fetal Gubernaculum and Testis of Cryptorchid Rats
361
A.
Dression
).5 0.5 1.5
X '
LUin
$1.
00
0
^
O>
O
^P
0
CM
0
?%
00
5
t£;
O>
5
tti;?H
0
CM
B.
.
t/> 0
g-9
*
1 1
h
T
C
>
1
)
1
"-
-_
'1
o
i-
C
-
1
>
>
..,
i
i
-r— -,-
M^
5
-,- -r
1*1*
0)
5
Figure 4. Expression profiles of ribosomal genes in gubernaculum
and testis. Mean Z-scores ± SD for 62 probe sets in (A) male
gubernaculum and (B) testis are shown. Vertical lines separate the
samples by strain (O indicates orl; W, wt) and gestational age in
days (18-20).
procollagens, laminins, basigin, matrix Gla protein,
chondroitin sulfate proteoglycan 2, and spondin 1, show
reduced expression in orl testis.
We directly compared mean normalized expression of
62 ribosomal and mitochondria! ribosomal genes in
gubernaculum and testis (Figure 4). In contrast to the
highly differential expression in gubernaculum, expres-
sion levels of these genes are comparable in testis
samples from the 2 strains. Absolute expression levels
between testis and gubernaculum were not directly
comparable because of the differences in the protocols
used (ie, single compared with double amplification of
RNA samples). However, these data together with
global expression data suggest that altered expression
profiles are much more prominent in orl gubernaculum
than in testis. These data suggest that altered signaling
in the orl strain has a more profound effect on gene
expression in fetal gubernaculum than testis.
Validation of Array-Derived Expression Profiles
To determine whether the expression profiles obtained
from the microarrays were consistent with the relative
amounts of mRNA present in parallel samples, real-time
RT-PCR validation was carried out. Expression levels of
selected genes in specific pathways and functional
groups (Table 4, bold) were analyzed. We identified 2
candidate control genes, tripeptidyl peptidase 2 (Tpp2)
and lumican, with mean GC-RMA expression levels
showing minimal variation across all groups; real-time
RT-PCR results showed differences in raw Ct values of
less than 0.5 for both genes with more consistency seen
in Tpp2 expression across samples (data not shown). We
analyzed target genes related to Tgfp/Wnt/Hedgehog
(Bmp4, U2, Msxl, Wnt4, Fst), MAPK (Dusp6, Nfkbl),
and insulin-like growth factor signaling (Igfl, IgfbpS);
neurogenesis (Olfinl) and myogenesis (Myog, Tnnt2,
Des) in gubernacula, testes (6-12 samples/group from 2-
3 litters), or both relative to Tpp2. Compared with the
microarray data, we identified similar expression pat-
terns for most wt and orl samples at the 2 time points,
and differences in mRNA levels by RT-PCR were
statistically different for many of these genes at GD20
but not GDIS (Figure 5). The most significant differ-
ences between strains were noted for gubernacular genes
associated with neuromuscular development (Olfml,
Msxl, Myog, Tnnt2, and Id2),
Discussion
We characterized transcript profiles of fetal gubernac-
ulum and testis in an animal model of inherited
cryptorchidism and a wild-type strain to identify genetic
pathways that are activated during rapid growth of fetal
gubernaculum. Global analysis of samples suggests
delayed, feminized, or both patterns in the mutant orl
gubernaculum compared with the wt strain. We observe
2 major trajectories in the normal rat gubernaculum
between GDIS and 20 and sexually dimorphic expres-
sion of these genes at GDIS, with little change in gene
expression between GDIS and 19 in males. Functional
analysis of the normal pattern of gene expression is
consistent with growth, cellular proliferation, and
muscle development that are known to occur in the rat
gubernaculum during this time frame. The dynamic
changes resolve at GD20 to a level of expression similar
to that of the GDIS female, suggesting completion of
male-specific development. Similarly, by GD20, expres-
sion of Rxfp2 (Insl3 receptor) mRNA is markedly
diminished (Barthold et al, 2006) and antiandrogen
exposure does not prevent testicular descent (Spencer et
al, 1991), suggesting that the critical phase of InslS and
androgen stimulation of the gubernaculum occurs prior
to this time.
By contrast, GDI8-20 gene expression in orl fetuses is
substantially less varied, with significant overlap in the
function of genes that are more highly expressed in orl
males and in females, suggesting that loss of male-
specific signaling is already present at GDIS. The
observation that strain-specific expression profiling of
fetal testis shows relatively few differences also supports
a model of cryptorchidism in the orl strain in which
Previous
Next
-------
362
Journal of Andrology • May/June 2008
Wnt4
Fst Bmp4
Arrav RT-PCR
IS
S
5
.1 §
U =.
=
J
20 IS 20
I
*
_,
5
*
0
T I
-L
***
1
J
E.
X
Id
\v o \v o \v o w o
Olfml
Anav RT-PCR Arrav RT-PCR
18 20 18 20 18
e
0
!
0
i
f
i
o
•f;
0
d
cs
_
1
1
-L
*
0
11
&s
X
wl
0
'
_L <
20 18 20
1
X
Ifj
«M
0
S
2
T
-•
1 )
--
**
I
W O W O W O W O W 0 W 0 W 0 W 0
Msx1 Igf1
Anav RT-PCR
18
5
c.
«1
S
T
20 IS 20
1
0
o
*f
0
**
,1
-L
***
I
1
ression
&
Cd
\V 0 \V O W 0 W O
Nfkbl
Arrav RT-PCR Anav RT-PCR
18 20 18 20 18
1
;
i
o
i
I
q
o
-t
q
1 j_
L
***
••
i
s
oC
§1
I3
c._
x S
a <=
^
o
i
20 18 20
|
i
S
O
M
tf.
"— '
--
--
*
z
\V O W 0 W O W 0 \\ 0 W 0 \V 0 W 0
Dusp6 IgfbpS
Arrav RT-PCR
IS
e
0
|l
o
_
I
20 IS 20
T
1
T
if
w
s
*f.
«
-'•
t
--
11
1
E.
w o w o w o w o
Des
Myog
Array RT-PCR Array
20 20 20
o
*r.
*~~
ession
SOI)
c.
£g
O
I
30
"™"
^
T;
__
--
_^
•
e
s
ession
Cc C
pc J
3
:
J -|-
,1
S
S
1
T
* — 1—
Arrav RT-PCR Arrav RT-PCR
18 20 18 20 18
D
O
S5
0
ME
j
1
-
h-
0
o
,
, <•
--
*#*
I
0
e m
0
iA O
-------
Barthold et al • Gene Expression in Fetal Gubernaculum and Testis of Cryptorchid Rats
363
delayed or incomplete development of the gubernacu-
lum is a major contributing factor. Our functional
analysis suggests that many general pathways related to
metabolism, energy and growth are altered in the orl
gubernaculum, consistent with the decreased size of the
fetal orl gubernaculum (Barthold et al, 2006). We also
identified several specific, related pathways and biolog-
ical processes represented by differentially expressed
gubernacular genes, including small GTPase signal
transduction, focal adhesion, actin-filament based pro-
cess, cytoskeleton organization and biogenesis, and
muscle development. Although no database of andro-
gen-regulaled genes in the fetal gubernaculum exists, we
observed that genes regulated by androgens, involved in
androgen receptor signaling in other cell types, or both
were overrepresented in our lists of differentially
expressed genes from both testis and gubernaculum.
These data suggest that androgen receptor signaling
may be altered in the orl fetus, although additional
studies are needed.
The respective roles of Tnsl3 and androgen in cell-
specific development of the gubernaculum remains
undefined. In vitro, Rxfp2 activation increases cAMP
production via the stimulatory G-protein Gas, activates
the CRE reporter, and, in cooperation with testosterone,
stimulates proliferation of fetal gubernacular cells (Em-
men et al, 2000; Halls et al, 2007). Rxfp2 is one of many
G-protein coupled receptors that activates cAMP/PKA,
a response that regulates multiple developmental pro-
cesses, including neurogenesis and myogenesis (Lonze
and Ginty, 2002; Chen et al, 2005). However, little is
known of the downstream effects of cAMP/PKA
signaling in fetal gubernaculum beyond cellular prolif-
eration. In vivo, hyperplasia and extracellular matrix
production in the fetal gubernaculum is followed by
maturation of muscle precursors that become peripher-
ally oriented to form the striated cremaster muscle
(Radhakrishnan et al, 1979; Wensing, 1986). Cultured
GDI7 rat gubernacula enlarge in response to synthetic
androgen R1881 without Insl3 but contain poorly
organized myosin-positive cells within the mesenchymal
core. However, when cultured with testis, they contain a
defined outer layer of muscle (Emmen et al, 2000).
Marked atrophy of the fetal gubernaculum with loss of
the inner mesenchymal core is characteristic of both
fnslS and Rxfp2 null mice (Kubota et al, 2001). These
data suggest that In si 3 may play a role in the regulation
of both matrix remodeling and muscle development
within the gubernaculum.
Other rodent data support a role for Insl3IRxfp2 and
additional candidate genes in regulation of myogenesis
during development of the gubernaculum. Rxfp2 mRNA
is present throughout the gubernaculum at GD16 in rat
(Scott et al, 2005), but by GDI9, binding sites for Insl3
arc localized to the outer muscle layer (McKinnell et al,
2005), whereas the androgen receptor continues to be
expressed in both mesenchyme and muscle (Staub et al,
2005). A cell-specific developmental role for androgens in
the gubernaculum is not clear; however, after prenatal
exposure to the antiandrogen flutamide prior to GD17,
both mesenchymal and muscular compartments of the
GD20 rat gubernaculum are reduced in size (Cain et al,
1995) and embryonic muscle isoforms persist in adult
cremaster muscle (Tobe et al, 2002). HoxctlO transcripts
are also expressed throughout the GDI5.5 gubernacu-
lum, and histological studies of the postnatal cremaster
muscle in HoxalO (—/—) males show disordered myo-
genesis (Satokata et al, 1995).
Expression patterns of specific genes that are involved
in muscle development, contraction, or both (Table 4;
Figure 5) support our global functional analysis results
and suggest that terminal differentiation of muscle is
delayed or disrupted in orl gubernaculum. Expression
levels of Myog and Myf6, myogenic regulatory factors
that control later stages of muscle differentiation
(Sartorelli and Caretti, 2005), are reduced, whereas
several genes that arc down-regulated during or inhibit
terminal differentiation of muscle (Melnikova et al,
1999; Ohkawa et al, 2006), including Igfl, M2, Msxl,
and representatives of the fibroblast growth factor
family, show increased expression in the orl fetal
gubernaculum. We also identified altered expression of
several genes that promote skeletal muscle development,
including IgfbpS, Ilk, Bmp7, and Fst (Huang et al, 2000;
Amthor et al, 2002). Expression of Rps6kbl, Eif4e, and
Eif4ebpl are reduced in orl gubernaculum. These genes
are effectors of insulin and a mammalian target of
rapamycin (mTOR) signaling that regulate protein
synthesis and cell size (Ruvinsky and Meyuhas, 2006);
mTOR signaling is also critical for myoblast fusion
(Park and Chen, 2005). Reduced expression of these
genes is consistent with the global reduction in protein
synthesis, as well as the reduced expression of muscle-
specific genes that we observed in orl gubernaculum,
with previous microarray data showing increased
expression of energy and metabolism genes and de-
creased expression of genes involved in DNA replication
and transcription during skeletal myotube maturation
(Park and Chen, 2005).
In addition to muscle-specific genes, we identified
altered expression of genes related to small GTPase
signal transduction, cytoskeleton organization and
biogenesis, and focal adhesion. The Rho GTPases
encode proteins that are responsive to G-protein-
coupled receptor and receptor tyrosine kinase signaling
and are critical for cytoskeletal reorganization, cell
motility, axon guidance, and myogenesis (Kj oiler and
Hall, 1999; Bishop and Hall, 2000; Bryan et al, 2005).
-------
364
Journal of Andrology • May/June 2008
Several, including RhoA, Racl, Cdc42, and RhoC, arc
differentially expressed between strains (Table 4). Focal
adhesions are sites of cell attachment to the extracellular
matrix comprising integrins, cytoskeletal proteins, and
signaling molecules (Sastry and Burridge, 2000). Possi-
ble roles for focal adhesion signaling in the developing
gubernaculum include regulation of myoblast matura-
tion (Clemente et al, 2005), migration (Mitra et al,
2005), or both; formation of costameres (Z-bands
anchoring myoflbrils to the sarcolemma) (Quach and
Rando, 2006); and axon pathfinding (Robles and
Gomez, 2006). Expression of the mRNA for several
genes that participate in focal adhesion signaling,
including Ptk2, Kdr (Flkl), Src (y-src), and Csk (Sastry
and Burridge, 2000; Mitra et al, 2005) is present in the
GD16.5 mouse gubernaculum (Verma-Kurvari and
Parada, 2004). Csk encodes a tyrosine kinase linked to
focal adhesion turnover and regulation of the actin
cytoskeleton (McGarrigle et al, 2006), and the corre-
sponding protein is localized to both mesenchymal and
muscle layers in GDI6.5 mouse gubernaculum. Ilk, a
key component of integrin-mediated signaling that plays
a role in the switch from myogcnic proliferation to
differentiation (Huang et al, 2000), is expressed at lower
levels in orl gubernaculum.
Although the genetic basis for human cryptorchidism
remains largely unknown, review of gene defects
associated with cryptorchidism as compiled in Online
Mendelian Inheritance in Man (OMIM) supports our
present data. Several syndromes that include cryptor-
chidism are linked to genes that participate in small
GTPase signaling (SOS1, KRAS, FGD1), actin cyto-
skeleton regulation (FLNA, FLNE), muscle develop-
ment (ACTA I), or muscle contraction (RYRJ), Expres-
sion of some of these genes is altered in orl
gubernaculum (Table 4). In humans, the gubernaculum
is comprised primarily of mesenchyme, but striated
muscle is present within its distal portion in addition to
the surrounding cremaster muscle (Tayakkanonta, 1963;
Barteczko and Jacob, 2000; Costa et al, 2002), whose
role, if any, in testicular descent is unclear. It is notable,
however, that cryptorchidism is present in multiple
forms of congenital myopathy (OMIM) and is also
present in males with Prune-Belly syndrome (Jennings,
2000) and at a higher frequency in males with abdominal
wall defects (Kaplan et al, 1986). Moreover, altered
structure and function of the cremaster muscle has been
reported in cryptorchid boys (Tanyel et al, 2000). These
observations taken in combination with our present
study suggest that muscle patterning might play a more
important role in development and function of the
gubernaculum than previously recognized.
Limitations of this study include our analysis of
tissue-specific compared with cell-specific gene expres-
sion and the requirement for amplification of gubernac-
ulum but not testis. Although the amplification,
analysis, or both could be biased toward a particular
cell type, we have been unable to identify any clear
differences in cellular composition of wt compared with
orl gubernaculum using cell-specific immunostaining
(data not shown). Because of differences in RNA
processing, we avoided direct comparisons of gene
expression in testis and gubernaculum. Also, the
possibility exists that early transcriptosomal changes
associated with male specific gubernacular development
were missed because gene expression is already sexually
dimorphic in wt at GDIS. Therefore, although we can
identify global expression profiles that reflect develop-
ment of wt and orl gubernacula, we cannot determine
whether differences in gene expression are the cause or
result of abnormal development. Moreover, because
testicular descent does not occur until the postnatal
period in the rat, we are unable to determine at the fetal
stage which gubernacula (less than half) will be
associated with cryptorchid testes. Despite this, the
expression profiles of individual orl samples are highly
similar and markedly different from wt in gubernaculum
but not testis and therefore likely phenotype-specific as
opposed to strain-specific. The basis for reduced
penetrance of the phenotype remains unknown at this
time but might be related to a dosage effect determined
by environmental factors, modifying loci, or both. To
date, we have no evidence that maternal- or paternal-
specific factors determine phenotype because the fre-
quency of cryptorchidism in offspring does not appear
to be related to paternal phenotype or identity of the
dam (unpublished observations).
Analysis of gene expression in fetal tissues of wild-
type and cryptorchid orl mutant rats suggests that a
primary gubernacular defect that directly or indirectly
affects muscle function might predispose to cryptorchi-
dism in the affected strain. Further studies will be
necessary to elucidate the mechanism of gubernacular
dysfunction in the orl rat.
fifeferences
Amthor H. Christ B, Rashid-Doubell F, Kemp CH, Lang E. Patel K.
Follistatin regulates bone morphogenetic protein-7 (BMP-7)
activity to stimulate embryonic muscle growth. Dev Biol. 2002;243:
115-127.
Barteczko KJ, Jacob MI. The testicular descent in human. Origin,
development and fate of the gubernaculum Hunteri, processus
vaginalis peritonei, and gonadal ligaments. Adv Anal Embryo! Cell
Biol. 2000;156:III-X, 1-98.
Barthold JS, Gonzalez R. The epidemiology of congenital cryptorchi-
dism, testicular ascent and orchiopexy. / Urol. 2003;170:2396-
2401.
-------
Barthold et al • Gene Expression in Fetal Gubernaculum and Testis of Cryptorchid Rats
365
Barthold JS, Si X, Stabley D, Sol-Church K. Campion L, McCahan SM.
Failure of shortening and inversion of the perinatal gubernaculum in
the cryptorchid Long-Evans orl rat. / Vrol. 2006:176:1612 1617.
Bishop AL, Hall A. Rho GTPases and their effector proteins.
Biochem J. 2000;348(2):241-255.
Bolton EC, So AY, Chaivorapol C, Haqq CM, Li H, Yamamoto KR.
Cell- and gene-specific regulation of primary target genes by the
androgcn receptor. Genes Dev. 2007;21:2005-2017.
Bryan BA, Li D, Wu X, Liu M. The Rho family of small GTPases:
crucial regulators of skeletal myogenesis. Cell Mol Life Sci.
2005;62:1547-1555.
Cain MP, Kramer SA, Tindall DJ, Husmann DA. Flutamide-induced
cryptorchidism in the rat is associated with altered gubernacular
morphology. Urology. 1995:46:553 558.
Chen AE, Ginty DD, Fan CM. Protein kinasc A signalling via CREB
controls myogenesis induced by Wnt proteins. Nature. 2005:433:
317-322.
Clemente CF, Corat MA, Saad ST, Franchini KG. Differentiation of
C2C12 myoblasts is critically regulated by FAK signaling.
Am J Physiol. 2005;289:862R-870R.
Costa WS, Sampaio FJ, Favorite LA, Cardoso LE. Testicular
migration: remodeling of connective tissue and muscle cells in
human gubernaculum testis. J Vrol. 2002;167:2171-2176.
Czeizel A, Erodi E, Toth J. Genetics of undescended testis. J UroL
1981;126:528-529.
Dennis G Jr, Sherman BT, Hosack DA, Yang J. Gao W, Lane HC,
Lctnpicki RA. DAVID: Database for Annotation, Visualization,
and Integrated Discovery. Genome Biol. 2003;4:P3.
Elert A, Jahn K, Heidenreich A, Hofmann R. Der familiare
Leistenhoden [The familial undescended testis]. Klin Padiatr.
2003;215:40-45.
Emm en JM, McLuskey A, Adham IM, Engel W. Grootegoed JA,
Brinkmann AO. Hormonal control of gubernaculum development
during testis descent: gubernaculum outgrowth in vitro requires
both insulin-like factor and androgen. Endocrinology. 2000:141:
4720-4727.
Halls M. Bathgate R, Roche P. Summers R. Signaling pathways of the
LGR7 and LGR8 receptors determined by reporter genes. Ann N Y
Acad Sci. 2005;1041:292-295.
Halls ML, Bathgate RA. Summers RJ. Comparison of signaling
pathways activated by the relaxin family peptide receptors, RXFP1
and RXFP2, using reporter genes. / Pharmacol Exp Thcr.
2007;320:281-290.
Hosack DA, Dennis G Jr, Sherman BT, Lane HC, Lempicki RA.
Identifying biological themes within lists of genes with EASE.
Genome Biol. 2003;4:70R.
Huang Y, Li J, Zhang Y, Wu C. The roles of integrin-linked kina.se in
the regulation of myogenic differentiation. J Cell Biol. 2000:150:
861 872.
Hutson JM, Hasthorpe S. Abnormalities of testicular descent. Cell
Tissue Res. 2005:322:155-158.
Jennings RW. Prune belly syndrome. Semin Pedialr Surg. 2000;9:
115-120.
Kaplan LM, Koyle MA, Kaplan GW, Farrer JH, Rajfer J. Association
between abdominal wall defects and cryptorchidism. ./ UroL
1986:136:645-647.
Kjoller L, Hall A. Signaling to Rho GTPases. Exp Cell Res.
1999;253:166-179.
Kubota Y, Nef S, Farmer PJ, Temelcos C, Parada LF, Hutson JM.
Leydig insulin-like hormone, gubernacular development and
testicular descent. ./ Vrol. 2001:165:1673-1675.
Livak KJ, Schmittgcn TD. Analysis of relative gene expression data
using real-time quantitative PCR and the 2(-Delta Delta C(T))
Method. Methods (San Diego). 2001:25:402-408.
Lonze BE, Ginty DD. Function and regulation of CREB family
transcription factors in the nervous system. Neuron. 2002:35:605
623.
Mahood IK, McKinnell C, Walker M, Hallmark N, Scott H, Fisher
JS, Rivas A, Hartung S, Ivell R. Mason JI, Sharpe RM. Cellular
origins of testicular dysgenesis in rats exposed in utero to di(n-
butyl) phthalate. Inl J Androl. 2006;29:148-154, (discussion)
181-185.
McGarrigle D, Shan D, Yang S, Huang XY. Role of tyrosine kina.se
Csk in G protein-coupled receptor- and receptor tyrosine kinase—
induced fibroblast cell migration. J Biol Chem. 2006;281:10583-
10588.
McKinnell C. Sharpe RM, Mahood K. Hallmark N. Scott H, Ivell R,
Staub C, Jegou B. Haag F, Koch-Nolte F, Hartung S. Expression
of insulin-like factor 3 protein in the rat testis during fetal and
postnatal development and in relation to cryptorchidism induced
by in utero exposure to di (n-butyl) phthalate. Endocrinology.
2005:146:4536-4544.
Melnikova IN, Bounpheng M, Schatteman GC, Gilliam D, Christy
BA. Differential biological activities of mammalian Id proteins in
muscle cells. Exp Cell Res. 1999:247:94-104.
Mitra SK, Hanson DA. Schlaepfer DD. Focal adhesion kinase: in
command and control of cell motility. Nat Rev Mol Cell Biol.
2005;6:56-68.
Mouhadjer N, Pointis G, Malassine A. Bedin M. Testicular steroid
sulfatase in a cryptorchid rat strain. J Steroid Biochem. 1989:34:
555-558.
Ohkawa Y, Marfella CG, Imbalzano AN. Skeletal muscle specification
by myogenin and Mef2D via the SWI/SNF ATPase Brgl. EMBO
(Eur Mol Biol Organ) J. 2006;25:490-501.
Overbeek PA, Gorlov IP, Sutherland RW, Houston JB, Harrison WR,
Bocttgcr-Tong HL, Bishop CE, Agoulnik Al. A transgcnic
insertion causing cryptorchidism in mice. Genesis. 2001;30:26—35.
Park IH, Chen J. Mammalian target of rapamycin (rnTOR) signaling
is required for a late-stage fusion process during skeletal myotube
maturation. ./ Biol Chem. 2005:280:32009-32017.
Quach NL, Rando TA. Focal adhesion kinase is essential for
costamerogenesis in cultured skeletal muscle cells. Dev Biol.
2006:293:38-52.
Radhakrishnan J. Morikawa Y, Donahoe PK, Hendren WH.
Observations on the gubernaculum during descent of the testis.
Investig Vrol. 1979:16:365 368.
Robles E, Gomez TM. Focal adhesion kinase signaling at sites of
integrin-mediated adhesion controls axon pathfinding. Nal Neu-
rosci. 2006:9:1274-1283.
Ruvinsky I, Meyuhas O. Ribosomal protein S6 phosphorylation: from
protein synthesis to cell size. Trends Biochem Sci. 2006;31:342—348.
Saeed AI. Sharov V. White J, Li J, Liang W, Bhagabati N, Braisted J,
Klapa M, Currier T, Thiagarajan M, Sturn A, Snuffin M,
Rezantsev A, Popov D, Ryltsov A, Kostukovich E, Borisovsky I.
Liu Z, Vinsavich A, Trush V, Quackcnbush J. TM4: a free, open-
source system for microarray data management and analysis.
BioTechniques. 2003:34:374-378.
Sartorelli V, Caretti G. Mechanisms underlying the transcriptional
regulation of skeletal myogenesis. Citrr Opin Genet Dev. 2005;!5:
528-535.
Sastry SK, Burridge K. Focal adhesions: a nexus for intracellular
signaling and cytoskeletal dynamics. Exp Cell Res. 2000;261:25—36.
Satokata I, Benson G, Maas R. Sexually dimorphic sterility
phenotypes in HoxalO-deficient mice. Nature. 1995:374:460-463.
Scott DJ, Fu P. Shen PJ, Gundlach A. Layfield S, Riesewijk A.
Tomiyama H, Hutson JM, Trcgcar GW, Bathgate RA. Charac-
terization of the rat INSL3 receptor. Ann N Y Acad Sci. 2005;1041:
13-16.
-------
366
Journal of Andrology • May/June 2008
Spencer JR. Torrado T, Sanchez RS, Vaughan ED Jr, Imperato-
McGinley J. Effects of flutamide and finasteride on rat testicular
descent. Endocrinology. 1991:129:741 748.
Staub C, Rauch M, Ferriere F, Trepos M, Dorval-Coiffec I. Saunders
PT, Cobellis G, Flouriot G, Saligaut C, Jcgou B. Expression of
estrogen receptor ESR1 and its 46-kDa variant in the gubernac-
ulum testis. Siol Reprod. 2005;73:703-712.
Tanyel FC, Erdem S, Altunay H, Ergun L. Ozcan Z. Alabay B.
Buyukpamukcu N, Tan E. Distribution and morphometry of fiber
types in cremaster muscles of boys with inguinal hernia or
undescended testis. Pathol Res Pract. 2000;196:613 617.
Tayakkanonta K. The gubernaculum testis and its nerve supply.
Ami N Z J Surg. 1963;33:61-67.
Tobe T. Toyota N, Matsuno Y, Komiyama M, Adachi T. Ito H. Mori
C. Embryonic myosin heavy chain and troponin T isoforms remain
in the cremaster muscle of adult rat cryptorchidism induced with
flutamide. Arch Histol Cvtol 2002:65:279-290.
Tomiyama H, Hutson JM, Truong A, Agoulnik AI. Transabdominal
testicular descent is disrupted in mice with deletion of insulinlike
factor 3 receptor. J Pcdiatr Surg. 2003:38:1793 1798.
Verma-Kurvari S. Parada LF. Identification of tyrosine kinases
expressed in the male mouse gubernaculum during development.
Dev Dyn. 2004;230:660-665.
Wensing CJ. Testicular descent in the rat and a comparison of this
process in the rat with that in the pig. Anat Rec. 1986;214:154 160.
Wettenhall JM, Simpson KM, Satterley K, Smyth GK. affylmGUT: a
graphical user interface for linear modeling of single channel
microarray data. Bioinformatics (Oxf). 2006;22:897-899.
Wu Z. Irizarry RA, Gentleman R, Martinez-Miirillo F. Spencer F. A
model-based background adjustment for oligonucleotide expres-
sion arrays. J Am Stat Assoc. 2004;99:909-917.
Zimmermann S. Steding G, Emmen JM, Brinkmann AO, Nayernia K,
Holstein AF, Engel W, Adham IM. Targeted disruption of the InsB
gene causes bilateral cryptorchidism. Mol Endocrinol. \999; 13:681-691.
-------
This article was downloaded by: [US EPA Environmental Protection Agency]
On: 2 September2009
Access details: /Access Details: [subscription number 789514190]
Publisher Taylor & Francis
Informa Ltd Registered in England and Wales Registered Number: 1072954 Registered office: Mortimer House,
37-41 Mortimer Street, London W1T 3JH, UK
JOURNAL OF
TOXICOLOGY AND
ENVIRONMENTAL
HEALTH
Journal of Toxicology and Environmental Health, Part B
Publication details, including instructions for authors and subscription information:
http://www. info rmaworld. co m/smpp/title~content=t713667286
Approaches for Applications of Physiologically Based Pharmacokinetic Models
in Risk Assessment
Chad M. Thompson a; Babasaheb Sonawane a; Hugh A. Barton b; Robert S. DeWoskin c; John C. Lipscomb d;
Paul Schlossera; Weihsueh A. Chiu a; Kannan Krishnan e
" National Center for Environmental Assessment, Office of Research and Development, U.S. Environmental
Protection Agency, Washington, DC, USA b National Center for Computational Toxicology, Office of Research
and Development, U.S. Environmental Protection Agency, Research Triangle Park, North Carolina, USA c
National Center for Environmental Assessment, Office of Research and Development, U.S. Environmental
Protection Agency, Research Triangle Park, North Carolina, USA d National Center for Environmental
Assessment, Office of Research and Development, U.S. Environmental Protection Agency, Cincinnati, Ohio,
USA e Groupe de recherche interdisciplinaire en sante et Departement de sante environnementale et sante
au travail, Universite de Montreal, Montreal, Canada
Online Publication Date: 01 August 2008
To cite this Article Thompson, Chad M., Sonawane, Babasaheb, Barton, Hugh A., DeWoskin, Roberts., Lipscomb, John C.,
Schlosser, Paul, Chiu, Weihsueh A. and Krishnan, Kannan(2008)'Approaches for Applications of Physiologically Based
Pharmacokinetic Models in Risk Assessment', Journal of Toxicology and Environmental Health, Part B,11:7,519 — 547
To link to this Article: DOI: 10.1080/10937400701724337
URL: http://dx.doi.org/10.1080/10937400701724337
PLEASE SCROLL DOWN FOR ARTICLE
Full terms and conditions of use: http://www.informaworld.com/terms-and-conditions-of-access.pdf
This article may be used for research, teaching and private study purposes. Any substantial or
systematic reproduction, re-distribution, re-selling, loan or sub-licensing, systematic supply or
distribution in any form to anyone is expressly forbidden.
The publisher does not give any warranty express or implied or make any representation that the contents
will be complete or accurate or up to date. The accuracy of any instructions, formulae and drug doses
should be independently verified with primary sources. The publisher shall not be liable for any loss,
actions, claims, proceedings, demand or costs or damages whatsoever or howsoever caused arising directly
or indirectly in connection with or arising out of the use of this material.
Previous
-------
Journal of Toxicology and Environmental Health, Part B, 11:519-547, 2008
Copyright © Taylor & Francis Group, LLC
ISSN: 1093-7404 print /1521-6950 online
DOI: 10.1080/10937400701724337
j Taylor £t Francis
APPROACHES FOR APPLICATIONS OF PHYSIOLOGICALLY BASED
PHARMACOKINETIC MODELS IN RISK ASSESSMENT
Chad M. Thompson1, Babasaheb Sonawane1, Hugh A. Barton2, Robert S. DeWoskin3,
John C. Lipscomb4, Paul Schlosser1, Weihsueh A. Chiu1, and Kannan Krishnan5
1 National Center for Environmental Assessment, Office of Research and Development,
U.S. Environmental Protection Agency, Washington, DC, USA
2National Center for Computational Toxicology, Office of Research and Development,
U.S. Environmental Protection Agency, Research Triangle Park, North Carolina, USA
3National Center for Environmental Assessment, Office of Research and Development, U.S.
Environmental Protection Agency, Research Triangle Park, North Carolina, USA
4National Center for Environmental Assessment, Office of Research and Development,
U.S. Environmental Protection Agency, Cincinnati, Ohio, USA
5Groupe de recherche interdisciplinaire en sante et Departement de sante environnementale
et sante au travail, Universite de Montreal, Montreal, Canada
Physiologically based pharmacokinetic (PBPK) models are particularly useful for simulating exposures to environmental
toxicants for which, unlike pharmaceuticals, there is often little or no human data available to estimate the internal dose
of a putative toxic moiety in a target tissue or an appropriate surrogate. This article reviews the current state of knowl-
edge and approaches for application of PBPK models in the process of deriving reference dose, reference concentration,
and cancer risk estimates. Examples drawn from previous U.S. Environmental Protection Agency (EPA) risk assessments
and human health risk assessments in peer-reviewed literature illustrate the ways and means of using PBPK models to
quantify the pharmacokinetic component of the interspecies and intraspecies uncertainty factors as well as to conduct
route to route, high dose to low dose and duration extrapolations. The choice of the appropriate dose metric is key to the
use of the PBPK models for the various applications in risk assessment. Issues related to whether uncertainty factors are
most appropriately applied before or after derivation of human equivalent dose (or concentration) continue to be
explored. Scientific progress in the understanding of life stage and genetic differences in dosimetry and their impacts on
variability in susceptibility, as well as ongoing development of analytical methods to characterize uncertainty in PBPK
models, will make their use in risk assessment increasingly likely. As such, it is anticipated that when PBPK models are
used to express adverse tissue responses in terms of the internal target tissue dose of the toxic moiety rather than the
external concentration, the scientific basis of, and confidence in, risk assessments will be enhanced.
Improving the scientific basis for human health risk and safety assessments is an ongoing concern
for regulatory agencies. Often, the requisite data in humans for directly assessing health risks are not
available or are limited; therefore, developing risk and safety estimates requires extrapolation(s) across
species, exposure routes, durations, and exposure levels. By utilizing physiological, biochemical, and
physicochemical data, physiologically based pharmacokinetic (PBPK) models can perform these extrap-
olations using scientifically grounded principles, as well as characterize the variability and uncertainty
therein. Although data-intensive, PBPK models have the added benefit of affording predictions of inter-
nal dosimetry in humans that otherwise cannot be measured directly without potentially harming
patients or study participants. This article reviews the current state of knowledge and approaches for
application of PBPK models in cancer and noncancer risk assessment. Use of these models for perform-
ing the extrapolations often needed in risk assessment is discussed generally, followed by descriptions
This article is an edited excerpt from the final report "Approaches for the Application of Physiologically Based Pharmacokinetic Models
and Supporting Data in Risk Assessment" (U.S. EPA, 2006a), and was originally funded by U.S. Environmental Protection Agency under con-
tract RFQ-DC-03-00328. For more information about this report please see http://cfpub.epa.gov/ncea/cfm/recordisplay.dm?deid = 157668.
The National Center for Environmental Assessment has reviewed and approved this article for publication. Such approval does not
signify that the contents reflect the views or policy of the U.S. Environmental Protection Agency, nor does mention of trade names con-
stitute endorsement or recommendation for use.
Address correspondence to Kannan Krishnan, DSEST, 2375 Chemin de la Cote Ste Catherine Room 4105, Universite de Montreal,
Montreal, PQ, Canada H3T1A8. E-mail: kannan.krishnan@umontreal.ca
519
Previous
TOC
-------
520 C. M. THOMPSON ET AL
of the specific applications of PBPK models in reference concentration (RfC) derivation, reference dose
(RfD) derivation, and cancer risk assessment, along with illustrative examples from chemical-specific
human health risk assessments from regulatory agencies and peer-reviewed literature. Other applica-
tions such as (1) use of biomonitoring data to infer exposure, (2) evaluation of chemical mixtures, and
(3) linking of PBPK and pharmacodynamic modeling are also discussed. This article primarily focuses on
practices at the U.S. Environmental Protection Agency (EPA); however, PBPK models are used in regu-
latory assessments in Europe and North America (International Workshop on the Development of
Good Modeling Practice for PBPK Models, Chania, Greece, April 2007), and thus the approaches
herein may be applicable to the broader risk assessment community.
RATIONALE FOR USING DOSIMETRY MODELS IN RISK ASSESSMENT
Frequently, questions have been raised concerning the interpretations of toxicological studies,
including health risk assessments. These questions may revolve around observations of toxicities at
high doses that are not apparent at lower doses, or findings in one species that are not observed in
another species. In other cases, toxicity is observed in several species but at different exposure con-
centrations or exposure doses. Historically, questions about how to best apply information from
experimental animal toxicity studies or high occupational exposures to protecting public health has
led to a growing recognition that pharmacokinetic analyses, particularly pharmacokinetic models
that estimate dosimetry, are valuable tools to help provide some answers.
Advantages of Dosimetry Models
Pharmacokinetics involves the study of the time course of the concentrations or amounts of a
parent chemical or metabolite(s) in biological fluids, tissues and excreta, as well as the construction
of mathematical models to interpret such data (Wagner, 1981; Benet et al., 1996; Renwick, 2001).
The time course of the concentration of a chemical or its metabolite in biota is determined by the
rate and extent of absorption, distribution, metabolism, and excretion (ADME). The pharmacoki-
netics, or ADME, of a substance determines the delivered dose or the amount of chemical available
for interaction in the tissue(s) of interest. Relating adverse response(s) observed in biota to an appro-
priate measure of delivered dose (e.g., concentration of the toxic chemical in the target tissue)
rather than administered dose or exposure concentration is likely to improve the characterization of
many dose-response relationships (Clewell et al., 2002a; Slikker et al., 2004).
Various modeling approaches are used to characterize exposures and the resulting delivered
doses. These approaches reflect differences in chemical and physical characteristics (e.g., stable or
reactive gases, particulate matter, lipophilic organics, water-soluble compounds), differences in
pharmacokinetic properties, and the ability of compounds to produce contact site or systemic toxic
effects (U.S. EPA, 1994; Andersen & Jarabek, 2001; Overton, 2001; U.S. EPA, 2004).
Exposure to many drugs and toxicants occurs via the oral route resulting in systemic effects,
which were analyzed using relatively simple (e.g., one- and two-compartment) pharmacokinetic
models (O'Flaherty, 1981; Renwick, 2001). Generally, these models consist of a central compart-
ment that represents the whole body when distribution occurs nearly instantaneously (one-
compartment model) or plasma when an additional compartment (two-compartment model) is
necessary to describe a slower distribution phase such as sequestration into fat. Compartment
models help characterize the kinetic behavior of a chemical and are useful for deriving values
describing distribution in the body and clearance from the plasma (e.g., half-life).
The values derived from a compartment model analysis may only apply to the conditions of the
study from which the experimental data were obtained. However, these models have expanded to
include PBPK models that contain multiple compartments and explicit mathematical descriptions of
physiological processes and tissues most likely to affect chemical disposition (e.g., absorption from
the gut or lung, cardiac output, metabolism in the liver, renal clearance). Such models more closely
represent the biological determinants of a chemical's disposition in the body and predict the
internal dose that would result from different exposure regimens—including exposure conditions
for which no data are available or ethically obtainable.
Previous I TOC
-------
APPLICATIONS OF PBPK MODELS IN RISK ASSESSMENT
521
Clearly, the need to predict the behavior of chemicals in exposed organisms is a driving force
behind PBPK model development. Early PBPK models were developed to predict the behavior of
volatile anesthetics, including compounds now used exclusively as industrial chemicals (Krishnan &
Andersen, 2001). The general principles developed in these early PBPK modeling efforts for system-
ically distributed compounds are also applicable to other chemical classes. For example, the respi-
ratory tract is a frequent site of both exposure and toxicity, and it has been a particular focus for a
range of modeling approaches, including those developed to simulate the kinetics of gases of vari-
ous reactivities and solubilities, as well as particulate matter (PM) (U.S. EPA, 1994). More recently,
the kinetics of reactive gases and PM within the respiratory tract have been simulated using
advanced approaches such as two-dimensional and three-dimensional computational fluid dynam-
ics (CFD) modeling (Kimbell etal., 1993; Overton, 2001; Martonen etal., 2001; U.S. EPA, 2004).
The role of metabolism is another significant factor in the development of PBPK models.
Saturable metabolism often results in nonlinear relationships between the level of administered dose
and the levels of the internal dose for a parent compound or metabolite. In combination with other
physiological and chemical events, the resulting administered dose-response relationship can quickly
become difficult to resolve with simple analytical tools. PBPK models provide a reliable means to
account for interactions and nonlinearities among multiple processes and to provide insight into
whether the parent chemical or the metabolite is the toxic moiety leading to adverse effects. As
depicted in Figure 1, dose-response relationships that appear unclear or confusing at the adminis-
tered dose level may become more understandable when expressed on the basis of internal dose of
the chemical. The major advantage of constructing dose-response relationships on the basis of inter-
nal or delivered dose is that it provides a stronger biological basis for conducting extrapolations and
for comparing responses across studies, species, routes, and dose levels (Clewell & Andersen, 1985;
Andersen et al., 1987; Aylward et al., 1996; Benignus et al., 1998; Melnick & Kohn, 2000).
30
linear
Weibull
0 1000 2000 3000 400C
Exposure concentration (ppm)
B
30
25
8-15
CD
^1°1
5
0
10 20 30 40 50
Rate metabolized (mg/hr)
FIGURE 1. Relationship between the exposure concentration and adverse response for a hypothetical chemical. (A) Hypothetical exam-
ple of a chemical for which the correlation between dose and response is weak or complex, along with equally plausible curve fits (linear,
Hill, and Weibull). This dose-response relationship is improved when it is based on an appropriate measure of internal dose (B).
Previous
TOC
-------
522
C. M. THOMPSON ET AL
Advantages of using PBPK Models in Risk Assessment
Regulatory agencies such as the U.S. EPA derive dose-response values based on the current
understanding of a dose-response relationship. Reference values for noncancer effects corre-
spond to an estimate of a daily exposure to the human population (including sensitive subgroups)
that is likely to be without an appreciable risk of deleterious effects during a lifetime. The refer-
ence values developed at the U.S. EPA include RfC for chronic inhalation exposures and RfD for
chronic oral exposures. For chronic oral and inhalation cancer risk assessments with an unknown
or a linear mode of action (MOA) (e.g., mutagenic carcinogens), U.S. EPA typically develops unit
risk estimates, a probability of developing cancer over a lifetime per unit of exposure, including
the cancer slope factor (CSF) for oral exposures and the inhalation unit risk (IUR). The underlying
assumption in these derivations is that the typical human exposure concentrations or applied
doses of a parent chemical result in internal exposures to the putative toxic form of the chemical
in a target organ that will be less than or equal to a level that is not associated with significant
adverse responses during a lifetime (noncarcinogens) or that yields a likely risk at or below the
estimated lifetime risk (carcinogens).
Although a key factor in the induction of adverse effects is the presence of the toxic form of a
chemical in the target organ, it is rare that data are available on the time course of the toxic moiety
in the target tissue(s) in humans. Even in animal studies, it is more practical to obtain measures of
blood, plasma, and urinary concentrations of toxic chemicals and their metabolites than the actual
toxic moiety level in the relevant tissue. Pharmacokinetic models can therefore be used to estimate
the tissue concentration of toxic substances.
Among the compartmental pharmacokinetic models, PBPK models are generally the most
useful for conducting extrapolations needed to derive reference values because they model
the underlying physiological and chemical processes that determine chemical disposition, and
may be used to predict target organ concentrations for relevant exposure scenarios
(Himmelstein & Lutz, 1979; Rowland, 1985; Leung, 1991; Andersen, 1995; Krishnan &
Andersen, 2001, 2007; Reddy et al. 2005). Figure 2 depicts the process of PBPK model
development. By simulating kinetics and estimating dose metrics of chemicals, PBPK models
make explicit and perhaps reduce the uncertainty related to interspecies, intraspecies,
route-to-route, duration, and high-dose to low-dose extrapolations needed to derive RfC,
RfD, and cancer unit risk estimates. The following sections discuss how the PBPK models are
used in health risk assessment.
| Reflne Model
Simulation |
^
>
not OK
<
^
Comparison
to Data
/
/
C\K
Application in
Risk Assessment
FIGURE 2. Basic flow chart of PBPK model development. Note that physiological and biochemical constants used in deterministic mod-
eling could be replaced with statistical values when using probabilistic modeling approaches.
Previous
TOC
-------
APPLICATIONS OF PBPK MODELS IN RISK ASSESSMENT 523
APPLICATION OF PBPK MODELS IN RISK ASSESSMENT
This article focuses on applications of PBPK models in risk assessment and provides literature
references to guide the reader interested in further information. An extensive listing of more than
1000 references relevant to PBPK modeling for environmental chemicals is provided as an online
Appendix to "Approaches for the Application of Physiologically Based Pharmacokinetic Models and
Supporting Data in Risk Assessment" (U.S. EPA, 2006a).
Pharmacokinetic Model and Data Needs
The quantitative dose-response assessment portion of the risk assessment process can be used
to determine a point of departure (POD) for one or more of the most sensitive critical effects. The
POD is the dose-response point that marks the beginning of a low-dose extrapolation, and it can be
the no-observed-adverse-effect level (NOAEL) or the lowest-observed-adverse-effect level (LOAEL)
for an observed incidence or the lower bound on dose for an estimated incidence or change in
response level from a benchmark dose (BMD) analysis. Frequently, the POD is an external exposure
concentration (or administered dose) that relates to the observed responses in laboratory or field
studies. Infrequently, the POD is an internal dose metric (e.g. blood concentration) in a toxicity
study which is also designed to collect appropriate pharmacokinetic measurements. Generally,
however, if an assessment intends to use an internal dose metric such as blood concentration, a
PBPK model will be needed to estimate the internal dose(s) from the external exposure or doses
(including the POD) that were used in a toxicity study.
PBPK models are often intended to estimate target tissue dose in species and under exposure
conditions for which little or no data exist. Thus, if a complete pharmacokinetic data set were avail-
able, then there may be no need to develop a PBPK model. Such an optimal pharmacokinetic data
set for risk assessment would consist of the time-course data on the most appropriate dose metric
associated with exposure scenarios and doses used in the critical studies chosen for the assessment
(e.g., animal bioassays or human clinical and epidemiological studies) and relevant human expo-
sure conditions. An example of such a dose metric is the concentration of a toxic metabolite in tar-
get tissue over a 24-h period in the test species and humans. This information would be obtained
for the window of exposure and route and scenario of exposure associated with the critical study, as
well as for the window of susceptibility, appropriate route, and exposure scenarios in humans.
In almost all cases, however, the optimal data set is not available, and often the available animal
pharmacokinetic data may be limited. In the absence of experimental kinetic data on the biologically
active form of a chemical in target tissues, data on blood concentration of the parent chemical, urinary
metabolite levels, or fraction absorbed may be used as a surrogate for the tissue levels. These and
other subsets of pharmacokinetic data are used to develop a PBPK model to estimate the level of the
toxic moiety of interest, and the uncertainty in those estimates can be formally characterized.
Choosing PBPK Models Appropriate for Risk Assessment
PBPK models are most often used in risk assessments to estimate tissue and blood concentra-
tions of a toxic moiety (parent chemical or metabolite) resulting from the dosing regimens in the
animal toxicity or human studies that are the basis for deriving dose-response values (e.g., RfC, RfD,
CSFs). Figure 3 depicts four basic criteria for selecting appropriate PBPK models. The first criterion,
though appearing self-evident, is quite fundamental because not all models available in the scien-
tific literature are parameterized for the specific species and lifestage that were used in the critical
toxicological study under consideration in dose-response analysis. For example, a PBPK model for a
particular compound may be available for rats, yet the critical study in a dose-response analysis is in
mice. Similarly, the second criterion relates to the routes of exposure that can be simulated by an
existing PBPK model and those of interest for a health risk assessment. The third criterion is that a
model chosen for risk assessment applications would be able to provide simulations of the tissue
dose of the toxic moiety or an appropriate dose metric related to the MOA, exposure scenarios,
and routes associated with the critical study and likely human exposures. When a PBPK model does
not adequately meet these three criteria, additional data collection and model development may
Previous I TOC
-------
524
C. M. THOMPSON ET AL
Is the PBPK
model available
for the test
species and
humans?
Are the
parameters for
simulating
relevant routes
available?
Experimental
Data
Collection
Does the mode
simulate dose metrics
of relevance to risk
assessment?
Has the model been
evaluated and peer-
reviewed?*
FIGURE 3. Flow chart for selecting PBPK models appropriate for risk assessment. Four basic criteria for model use in risk assessment are
shown in diamond shape boxes. A detailed discussion of model evaluation is described in U.S. EPA (2006a) and Chiu et al. (2007).
be necessary. Finally, the fourth criterion is that PBPK models intended for use in risk assessment
need to be peer-reviewed; otherwise, efforts may be directed toward this end. Approaches for eval-
uating PBPK models are discussed in greater detail elsewhere (Clark et al., 2004; U.S. EPA, 2006a;
Chiu etal., 2007).
Evaluation of Dose Metrics for PBPK Model-Based Assessments
Relating blood and tissue concentrations with response in exposed organisms has long been
recognized in pharmacology (Wagner, 1981). As mentioned previously, in pharmacokinetics, the
target tissue dose that most closely relates to an adverse response is often referred to as the internal
"dose metric" (Andersen & Dennison, 2001). Dose metrics used in risk assessment applications,
Previous
TOC
-------
APPLICATIONS OF PBPK MODELS IN RISK ASSESSMENT 525
ideally, reflect the biologically active form of the chemical (parent chemical, metabolites, or
adducts), its level (concentration or amount), duration of internal exposure (instantaneous, daily,
lifetime, or a specific developmental period), and intensity (peak, average, or integral), as well as
the appropriate biological matrix (e.g., blood, target tissue, surrogate tissues). When using PBPK
models in risk assessment, the basic data needed are: (1) a POD for a critical effect from a toxicity
study, (2) a peer-reviewed PBPK model for the relevant animal test species and humans, and (3) a
dose metric that is reliably predicted by the PBPK model, is appropriate for the risk assessment, and
is supported by the MOA (if known). The methods and challenges associated with the first data
need (identification of critical effects and PODs for an assessment) remain the same regardless of
whether one uses PBPK models or not; approaches for identifying PODs are found elsewhere (U.S.
EPA, 1994, 2005a). The criteria and issues associated with the evaluation of PBPK models for use in
risk assessment are discussed in detail elsewhere (U.S. EPA, 2006a; Chiu et al., 2007). It is worth
noting, however, that although an animal model is frequently needed, if sufficient pharmacokinetic
data are available for the toxic moiety of interest in an animal species, then only a human PBPK
model may be necessary to predict internal dose in humans. The third data need, identification of
the appropriate dose metric, is the subject or the remainder of this subsection.
The dose metric, or the appropriate form of potential toxic moiety most closely associated with
the response, varies from chemical to chemical, depending on the MOA and critical effect. It has
two basic properties: the moiety and the measure thereof (Table 1). The dose metric for PBPK-
based risk assessment is chosen following the identification of the potential toxic moiety and eval-
uation of the relationship with the endpoint of concern. Useful frameworks for evaluating
hypothesized MOAs are included in Guidelines for Carcinogen Risk Assessment (U.S. EPA, 2005a)
and A Framework for Assessing Health Risks of Environmental Exposures to Children (U.S. EPA,
2006b). Although the framework specifically deals with carcinogens, the concepts are broadly
applicable to noncancer MOAs. The framework provides useful discussion related to evaluating
multiple MOAs (particularly over dose ranges) and for assessing relevance to humans. Furthermore,
available data on closely related chemicals may be used to infer the likely toxic moiety. Similarly,
the toxicity data for various exposure routes and modes of administration may be compared to infer
the potential toxic moiety (IPCS, 2005).
After the identification of the potential toxic moiety, the appropriate measure of tissue exposure
to the toxic moiety is selected (Figure 4). For example, peak concentration was related to some
neurotoxic effects of solvents (Bushnell, 1997; Benignus et al., 1998; Pierce et al., 1998;
MacDonald et al., 2002), such as the correlation of concentration of trichloroethylene at the time of
TABLE 1. Examples of dose metrics for exploring
dose-response relationships
Potential moieties
Parent compound
Peak concentration
Average concentration
Amount or quantity
AUC (integral)
Metabol ite
Peak concentration
Average concentration
Amount or quantity
Rate of production
Cumulative rate of amount formed/time/L tissue
AUC (integral)
Miscellaneous
Receptor occupancy (extent or duration)
Macromolecular adduct formation
Depletion of cofactors
Previous I TOC
-------
526
C. M. THOMPSON ET AL
Maximum
1
Integrated
i
Average
1
Integrated
1
^
r
Cmax
(mg/L tissue)
AUC
(mg/L tissue x time)
mg formed
per unit time
(e.g. mg/hr/L tissue)
mg formed
overtime specified
(e.g. mg/hr/L tissue)
DOSE METRICS
AUC
rng/L tissue x time
(parent or metabolite)
'evaluate correlations
to toxic responses
FIGURE 4. Examples of dose metrics for risk assessment application.
toxicity testing with observed effects on behavioral and visual functions (Boyes et al., 2000). For tet-
rachlorodibenzodioxin, tissue concentrations of the chemical measured during a critical period of
gestation were reported to predict the intensity of developmental responses (Hurst et al., 2000).
The gender-specific genotoxic effects of benzene in mice are related to differences in the rate of
oxidative metabolism (Kenyon etal., 1996).
For chronic effects of chemicals, the integrated concentration of the toxic form of chemical in
target tissue over time (i.e., the area under the concentration curve or AUC), typically determined
as the daily average, is often considered a reasonable dose metric (Andersen et al., 1987; Collins,
1987; Voisin et al., 1990; Clewell et al., 2002a). For carcinogens producing reactive intermediates,
the amount of metabolite produced per unit time and the amount of metabolite in target tissue over
a period of time (e.g., mg metabolite/L tissue during 24 h) were used as dose metrics (Andersen et al.,
1987; Andersen & Dennison, 2001). For developmental effects, the dose surrogate is defined in the
context of the window of exposure for a particular gestational event (Welsch et al., 1995). Although
the AUC and rate of metabolite formation figure among the most commonly investigated dose met-
rics, other surrogates of tissue exposure may also be appropriate for risk assessment purposes,
depending on the chemical and its MOA (e.g., maximal concentration [Cmax] of the toxic moiety,
duration and extent of receptor occupancy, macromolecular adduct formation, or depletion of glu-
tathione) (Clewell et al., 2002a). Table 2 lists dose metrics that have been used in a number of
PBPK models published in the peer-review literature. Although these models and dose metrics were
used for cancer and noncancer risk assessments performed in the scientific literature, they have not
necessarily been used in risk assessments carried out by regulatory agencies.
When the appropriate dose metric cannot be readily identified, evaluation of the relationship
with the endpoint of concern may be undertaken with each of the dose metrics to identify the one
that exhibits the best association (Andersen et al., 1987; Kirman et al., 2000). This becomes partic-
ularly important when there are limited or confusing data on the plausible MOA of the chemical. At
a minimum, the appropriate dose metric can be identified as the one that demonstrates a consis-
tent relationship with positive and negative responses observed at various dose levels, routes, and
scenarios in a given species. In other words, the level of the dose metric would be lower for expo-
sure conditions that elicit no effect and higher for conditions that elicit toxic responses, regardless of
-------
APPLICATIONS OF PBPK MODELS IN RISK ASSESSMENT
527
TABLE 2. Dose metrics used in published PBPK models for assessing risk
Chemical
Acrylonitrile
Bromotrifluoromethane
Butoxyethanol (2-)
Chloroform
Chloropentafluorobenzene
1,4-Dioxane
Ethyl acrylate
Ethylene glycol ethers
Formaldehyde
Heptafluoropropane
Isopropanol
Methoxyacetic acid
Methyl chloroform
Methylmercury
Methyl methacrylate
Methylene chloride
Styrene
Tetrachlorodibenzodioxin
Toluene
Trichloroethylene
Vinyl acetate
Vinyl chloride
Endpoint
Brain tumors
Cardiac sensitization
Forestomach lesions
and tumors
Liver cancer
Hepatic effects and
kidney tumor
Liver toxicity
Liver tumors
Forestomach tumors
Developmental
toxicity
Cancer
Cardiac sensitization
Neurobehavioral
effects
Developmental/
reproductive effects
Developmental effects
Hepatic effects
Neurological effects
Nasal lesions
Cancer
Lung tumors (mouse)
Biochemical responses
Cancer risk
Behavioral effects
Renal toxicity
Neurotoxicity
Cancer (liver lung and
kidney)
Olfactory
degeneration and
tumor development
Angiosarcoma
Dose metric
Peak metabolite concentration in target tissue
Concentration of parent chemical at the end
of exposure
Concentration of butoxyethanol/ butoxy
acetic acid in forestomach
Amount of metabolites covalently bound
to biological macromolecules L liver
per day; % cell kill/day
Maximal rate of metabolism per unit kidney
cortex volume
AUC of parent chemical in liver
Time-weighted average concentration
in liver over lifetime
Liver AUC
Tissue-specific glutathione depletion
Peak concentration and average daily AUC of
the alkoxyacetic acid (metabolite) in blood
DNA-protein cross-links
Concentration of parent chemical at the end
of exposure
Peak blood concentration
AUC of isopropanol and its metabolite
(acetone)
AUC of parent chemical (gestational day 1 1 )
Maximal concentration of parent chemical
(gestational day 8)
Area under the liver concentration vs. time
curve
Fetal brain concentrations
Amount metabolized/time/volume
nasal tissue
Rate of glutathione transferase metabolites
produced/L liver/time
Steady-state concentration of ring oxidation
metabolite mediated by CYP2F
Body burden
Time-weighted receptor occupancy
Up/down regulation of receptor occupancy
Fraction of cells induced
Brain concentrations at the time of testing
Metabolite production/L kidney/day
Blood concentration of metabolite
(trichloroethanol)
Amount metabolized/kg/day; AUC for
trichloroacetic acid or dichloroacetic
acid/L plasma; production of
thioacetylating intermediate from
dichlorovinylcysteine in kidney
Intracellular pH change associated with
the production of acetic acid; proton
concentration in olfactory tissue
mg metabolized/L liver; mg metabolite
produced/L liver/day
Reference
Kirman etal. (2000)
Vinegar and Jepson (1 996)
Poet etal. (2003)
Reitzetal. (1 990a)
Meek et al. (2002)
Clewell and Jarnot (1 994)
Leung and Paustenbach
(1 990)
Reitzetal. (1 990b)
Frederick etal. (1992)
Sweeney etal. (2001)
Schlosseretal. (2003);
Casanova etal. (1996)
Vinegar and Jepson (1 996)
Gentry etal. (2002)
Gentry etal. (2002)
Clarke etal. (1993, 1992)
Welschetal. (1995)
Reitzetal. (1 988a)
Gearhartetal. (1995)
Andersen et al. (2002,
1999)
Andersen etal. (1987)
Cruzan et al. (2002)
Kim et al. (2002)
Andersen etal. (1993)
Portieretal. (1993)
Conolly and Andersen
(1997)
Van Asperen etal. (2003)
Barton and Clewell (2000)
Barton and Clewell (2000)
Clewell et al. (2000);
Fisher and Allen (1993)
Bogdanffy et al. (2001,
1999)
Clewell etal. (2001); Reitz
etal. (1996b)
Previous
TOC
-------
528 C. M. THOMPSON ET AL
TABLE 3. Relationship between tumor prevalence and dichloromethane metabolites3
Exposure (ppm) Microsomal pathway6 GSTC pathway6 Tumor number
0 006
2000 3575 851 33
4000 3701 1811 83
aAdapted from Andersen et al. (1 987); methylene chloride dose response in female mice.
6mg metabolized/L tissue/d via each pathway.
cGlutathione S-transferase.
the route and exposure scenario (Clewell etal., 2002a). For example, Andersen etal. (1987) used a
PBPK model for dichloromethane (DCM) to examine which putative metabolites are responsible for
inducing liver and lung tumors in mice exposed to 2000 or 4000 ppm, 6 h/d, 5 d/wk for lifetime. In
brief, DCM undergoes oxidation by microsomal cytochrome P-450 enzymes or conjugation to
glutathione by glutathione S-transferase (GST). These authors designed a mouse PBPK model to cal-
culate the tissue dose of metabolites arising from exposure scenarios comparable to those used in
the relevant cancer bioassay study. The predicted amounts of metabolites formed by each pathway
were compared to the observed tumor incidence. Because the parent chemical was nonreactive,
Andersen et al. (1987) considered it an unlikely candidate responsible for the tumorigenicity.
Hence, only the relationship between the tissue exposure to its metabolites and tumor incidence
was examined (Table 3). Although the dose metric based on cytochrome P-450 oxidative pathway
varied little between 2000 and 4000 ppm, the flux through the GST pathway rose with increasing
dose of DCM and corresponded well with the degree of DCM-induced liver tumors at these expo-
sure concentrations. Similar results were obtained for lung tumors (data not shown).
Where there is an inadequate basis for prioritizing one dose metric over another, some suggest
using the most conservative one (i.e., the dose metric estimating the highest risk or lowest accept-
able exposure level) to be health protective (Clewell et al., 2002a). For more on dose metric selec-
tion, see (Clewell etal., 2002a; U.S. EPA, 2006a).
Overview of Extrapolations Possible With PBPK Models
The scientific literature (or database) for a specific chemical is often incomplete, thus extrapola-
tions are frequently required in risk assessment. Generally, there are five types of extrapolations
possible with PBPK models. Interspecies extrapolations are most common, as toxicity data is most
often derived in experimental animals. Because animal exposures are often performed in the con-
text of work days and work weeks, the exposure patterns frequently utilized in animal studies need
to be adjusted for duration, particularly when deriving dose-response values for continuous expo-
sure. Route extrapolations are often advantageous in risk assessment, as high-quality studies for all
routes of interest to human exposure are rarely available in the scientific literature. Similarly, high-
dose to low-dose extrapolations are necessary to account for the fact that experimental assays often
employ exposure doses high enough to elicit a response over more than one dose or concentration.
Finally, intraspecies extrapolations may be carried out to extrapolate findings in one group (usually
healthy adult animals or humans) to other susceptible populations.
Interspecies Extrapolation The application of PBPK models for interspecies extrapolation of
tissue dosimetry (Rowland, 1985) is performed in several steps. First, a model for the species in a
critical toxicity study is developed. The next step involves using species-specific or allometrically
scaled physiological parameters in the model and replacement of the chemical-specific parameters
(e.g., metabolic rates, protein binding constants) with appropriate estimates for the species of inter-
est (e.g., humans) for the risk assessment. In this approach, the qualitative determinants of pharma-
cokinetics are considered to be invariant among the various mammalian species unless qualitative
differences between species have been identified. In this case, differences would be factored into
the existing structure of PBPK models (e.g., if different metabolic pathways existed among species)
but, obviously, data describing these species differences are required.
Previous I TOC
-------
APPLICATIONS OF PBPK MODELS IN RISK ASSESSMENT 529
For conducting interspecies extrapolation of pharmacokinetic behavior of a chemical, quanti-
tative estimates of model parameter values (e.g., physiological parameters, partition coefficients,
and metabolic rate constants) in the second species are required. Physiological parameters are typ-
ically available from primary or secondary literature (Brown et al., 1997). The tissue:air partition
coefficients of chemicals appear to be relatively constant across species, whereas blood:air parti-
tion coefficients show some species-dependent variability. Therefore, the tissue:blood partition
coefficients for human PBPK models can be calculated by dividing the rodent tissue:air partition
coefficients by the human blood:air partition values (Krishnan & Andersen, 2001). The tissue:air
and blood:air partition coefficients for volatile organic chemicals may also be predicted using
appropriate data on the content of lipids and water in human tissues and blood (Poulin & Krishnan,
1996a, 1996b).
Kinetic constants for metabolizing enzymes do not necessarily follow any readily predictable
pattern, making the interspecies extrapolation of xenobiotic metabolism difficult. Therefore, the
metabolic rate constants for xenobiotics are best obtained in the species of interest. In vivo
approaches for determining metabolic rate constants are not always feasible for application in
humans. The alternative is to obtain such data under in vitro conditions (Lipscomb et al., 1998,
2003). A parallelogram approach may be used to predict values for the human PBPK model on the
basis of metabolic rate constants obtained in vivo in rodents as well as in vitro using rodent and
human tissue fractions (Reitz et al., 1988b; Lipscomb et al., 1998). Alternatively, for chemicals
metabolized by enzymes conserved across species, Vm3X has been scaled to the 0.75 power of body
weight and Km assumed invariant. This approach may be useful as a crude approximation, in the
absence of direct measurements of metabolic parameters. Once the human PBPK model is con-
structed and parameterized for a particular chemical, its dosimetric predictions should be com-
pared with human data (if available) to assess whether the model adequately simulates the
pharmacokinetics of the chemical in humans.
An example of rat-to-human extrapolation of the kinetics of toluene using a PBPK model is pre-
sented in Figure 5A. Here the structure of the PBPK model developed in rats was kept unchanged,
but the species-specific parameters were determined either by scaling or experimentally, as
described earlier. This reparameterized model was then able to predict accurately the kinetics of
toluene in humans (seeTardif et al., 1997). Whenever the human data for a particular chemical are
not available for evaluation purposes, a corollary approach using human data on similar chemicals
may be attempted (Jarabek et al., 1994).
There are some instances where PBPK models may be used for interspecies extrapolation of
toxicity studies without the need of an animal PBPK model. For example, an RfC for methanol was
proposed (Starr & Festa, 2003) using a mouse developmental toxicity study (Rogers et al., 1993) in
which blood methanol levels were also reported. By using the blood methanol level at the POD
from the mouse study, a previously published human methanol PBPK model (Bouchard et al.,
2001) was used to predict the inhalation concentration associated with the same internal blood
methanol level in humans. This example highlights the advantage afforded by toxicity studies that
also include pharmacokinetic measurements. It should be noted, however, that the utility of this
specific example is dependent on whether methanol, as opposed to a methanol metabolite, is a
suitable dose metric for methanol-induced toxicity.
Route-to-Route Extrapolation The extrapolation of the kinetic behavior of a chemical from
one exposure route to another is performed by including appropriate equations to represent each
exposure pathway. For simulating the intravenous administration of a chemical, a single input rep-
resenting the dose administered to the animal is included in the equation for mixed venous concen-
tration. Oral gavage of a chemical dissolved in a carrier solvent may be modeled using approaches
ranging from simple zero- or first-order absorption to multicompartment, multiabsorption process
models (Roth et al., 1993), and dermal absorption has been modeled by including a diffusion-lim-
ited compartment to represent skin as a portal of entry (Krishnan & Andersen, 2001). After the
equations describing the route-specific entry of chemicals into systemic circulation are included in
the PBPK model, it is possible to conduct extrapolations of pharmacokinetics and dose metrics. This
approach is illustrated in Figure 5B for oral-to-inhalation extrapolation of the kinetics of chloroform
Previous I TOC
-------
530
C. M. THOMPSON ET AL
A: Interspecies
c 100 -, Rat
o
'ra ID-
B: Route
'S
0.1
3 0.01 -
w
o 0.001 -
> 0.0001
8 12 16
Time (hours)
20
24
.1
1
"c
§
•a
o
o
_a
w
o
C
0)
100~i
10-
^p 1 <
D)
S 0.1 -
0.01-
0 001 ™
Oral
rx
"^•••^
"^^-v
"™^-^
""~^-v.
'*"""»„
01234567
Time (hours)
10 -,
1 -
0.1 -
0.01 -
0.001 -
0.0001
Human
C: Duration
I 10°-i
•E 10-
8
I? 1-
|| 0.1-
| 0-01 -
g 0.001 -
0.0001
10-1
1 -
0.1 -
0.01 -
0.001 -
0.0001
16 24 32
Time (hours)
40
48
8 12 16
Time (hours)
20
24
16 24 32
Time (hours)
40
48
100..
> 1-
0.1
0.01
Inhalation
345
Time (hours)
D: Low-Dose
300
0) 'T~-
£ £ 200
100
500 1000
Exposure Concentration (ppm)
1500
15 n
10 ^
II 5^
500 1000 1500
Exposure Concentration (ppm)
FIGURE 5. Examples of extrapolations afforded by PBPK models. (A) Interspecies extrapolation: Exposure of humans to 1 7.7 ppm tolu-
ene for 24 h results in the same AUC (3.8 mg/L/h) as exposure of rats to 50 ppm toluene for 6 h. It should be noted that if the pharmaco-
kinetics in humans were identical to rats, then an equivalent 24-h exposure would have yielded 12.5 ppm (i.e., 24 hx12.5 ppm is
identical to 6 hxSO ppm). Based on Tardif etal. (1997). (B) Route: Oral dose (1 mg/kg) and inhalation dose (83 ppm, 4 h) of chloroform
result in same AUC. Based on Corley et al. (1990). (C) Duration: Inhalation exposure of rats to 9.7 ppm toluene for 24 h results in an
AUC (2.4 mg/L/hr) equivalent to 50 ppm toluene for 4 h. Based on Tardif et al. (1997). (D) Low dose: Top panel, 1 2-h AUC of toluene,
indicates apparently linear kinetics. Bottom panel, 12-h AUC of a metabolite of toluene under the same exposure conditions, indicates
that the kinetics of toluene are nonlinear. Based on Tardif etal. (1997).
Previous
TOC
-------
APPLICATIONS OF PBPK MODELS IN RISK ASSESSMENT 531
in rats. Accordingly, 4-h inhalation exposure to 83.4 ppm chloroform is equal to an oral dose of 1
mg/kg, as determined with PBPK models on the basis of equivalent dose metric (i.e., parent chemi-
cal AUC in blood). Note that the peak concentrations differ approximately by 10-fold; thus, if peak
concentration was thought to be the appropriate dose metric, higher inhalation exposures would be
required to produce the same peak as a 1 mg/kg oral dose.
Duration Adjustment On the basis of equivalent dose metric, duration-adjusted exposure
values are obtained with PBPK models (Andersen et al., 1987; Bruckner et al., 2004; Simmons et
al., 2005). For example, if the appropriate dose metric of a chemical were the AUC, it would ini-
tially be determined for the exposure duration of the critical study using the PBPK model and then
the atmospheric concentration for a continuous exposure (e.g. a day, a window of exposure, or a
lifetime) yielding the same AUC is determined by iterative simulation. Figure 5C depicts an example
of 4-h to 24-h extrapolation of the pharmacokinetics of toluene in rats, based on equivalent 24-h
AUC (2.4 mg/L/h). According to the modeling results, rats exposed to 50 ppm toluene for 4 h or 9.7
ppm for 24 h receive equivalent doses (i.e., AUC). It should be noted that extrapolations across long
durations may not be warranted, as lifestage changes and pharmacodynamic adaptations (e.g., sen-
sitization and desensitization) may be operational (Clewell et al., 2002a).
High-Dose to Low-Dose Extrapolation PBPK models facilitate high-dose to low-dose extrap-
olation of tissue dosimetry by accounting for the dose dependency of relevant processes (e.g., satu-
rable metabolism, enzyme induction, enzyme inactivation, protein binding, and depletion of
glutathione reserves) (Clewell & Andersen, 1987). The description of metabolism in PBPK models
has frequently included a capacity-limited metabolic process that becomes saturated at high doses.
Nonlinearities arising from mechanisms other than saturable metabolism, such as enzyme induc-
tion, enzyme inactivation, depletion of glutathione reserves, and binding to macromolecules, can
also be described with PBPK models (Clewell & Andersen, 1987; Krishnan et al., 1992). A PBPK
model intended for use in high-dose to low-dose extrapolation needs equations and parameters
describing dose-dependent phenomena if they occur in the range of interest for the assessment.
Because the determinants of nonlinear behavior may not be identical across species and age groups
(e.g., maximal velocity for metabolism, glutathione concentrations), these parameters are required
for each species/age group. During the conduct of high-dose to low-dose extrapolation, no change
in the parameters of the PBPK model should be required except for the administered dose or expo-
sure concentration.
An example of high-dose to low-dose extrapolation is presented in Figure 5D. In this figure,
both the blood AUC and the amount metabolized over a period of time (12 h) are plotted as a
function of exposure concentrations of toluene. For conducting high-dose to low-dose simulation in
this particular example, only the numerical value of the exposure concentration (which is an input
parameter for the PBPK model) was changed during every model run. All other model parameters
remained the same. The model simulations shown in Figure 5D indicate the nonlinear nature of
blood AUC as well as the amount of toluene metabolized per unit time in the exposure concentra-
tion range simulated. In such cases, the high-dose to low-dose extrapolation of tissue dosimetry
should not be conducted assuming linearity but, rather, should be performed using tools such as the
PBPK models.
Estimating Intraspecies Variability
The magnitude of interindividual variability in internal dose may be assessed using PBPK mod-
els for RfC and RfD derivations. For this purpose, population distributions of parameters, particu-
larly those relating to physiology and metabolizing enzymes (i.e., genetic polymorphisms), are
specified in a Monte Carlo approach, such that the PBPK model output corresponds to distributions
of dose metric in a population. Using the Monte Carlo approach, repeated computations based on
inputs selected at random from statistical distributions for each input parameter (e.g., physiological
parameters, enzyme content/activity with or without the consideration of polymorphism) are con-
ducted to provide a statistical distribution of the output, i.e., tissue dose. Using the information on
the dose metric corresponding to a high percentile (e.g., 95th) and the 50th percentile, the magni-
tude of interindividual variability (pharmacokinetic component) is computed for risk assessment
Previous I TOC
-------
532
C. M. THOMPSON ET AL
re
2
Q.
50th percentile
....ill
95th percentile
• interindividual variability .
Concentration
FIGURE 6. Estimation of interindividual variability. In this example, interindividual variability describes the variation between the 50th
(median) and 95th percentile values of a dose metric simulated with a probabilistic PBPK model.
purposes (Figure 6). An important challenge for implementing this approach is to adequately
describe codependencies of parameters (e.g., tissue volume and blood flow); assuming indepen-
dence of parameters will overestimate the population variability. Another challenge is to adequately
characterize population distributions reflecting the range of factors important to pharmacokinetics,
including genetics, age, lifestyle, health, and nutritional status.
Although past efforts largely have characterized the impact of the distributions of parameters in
the adult population, variability analyses also need to address different lifestages (e.g., pregnancy,
children, geriatric). Age-specific changes in physiology, tissue composition, and metabolic activity
were incorporated into the same model structure used for adults (O'Flaherty, 1994; Corley et al.,
2003; Nong et al., 2006). Published examples of modeling different ages describe predictions for a
range of chemicals with different properties (Clewell et al., 2002b, 2004; Sarangapani et al., 2003;
Ginsberg et al., 2004). However, some lifestages, notably pregnancy and lactation, require different
model structures (i.e., describing the mother and the offspring) (Gentry et al., 2003, 2004; Corley et
al., 2003). Characterization of population variability across ages and lifestages as well as adult vari-
ability is an ongoing area of development (U.S. EPA, 2006c).
Applications of PBPK Models in RfC and RfD Derivation
The applications of PBPK models in RfD and RfC derivations are very similar (U.S. EPA, 2006a)
and therefore the present article describes only the approaches for applying PBPK models in RfC
derivation. An RfC value corresponds to an estimate (with uncertainty spanning perhaps an order of
magnitude) of continuous inhalation exposure (mg/m3) for a human population, including sensitive
subgroups, that is likely to be without an appreciable risk of deleterious noncancer effects during a
lifetime (U.S. EPA, 1994). Notationally, RfC is defined as:
= POD[HEc]/UF
where POD[HEC] is the POD (NOAEL, LOAEL, or BMC) dosimetrically adjusted to a human
equivalent concentration (HEC), and UF represents uncertainty factors to account for the extrapola-
tions associated with the POD (i.e., interspecies differences in sensitivity, human intraspecies vari-
ability, subchronic-to-chronic extrapolation, LOAEL-to-NOAEL extrapolation, and incompleteness
of database)
The starting point for an RfC derivation is the identification of the POD for the critical effect in a
key study. Subsequent steps involve (a) adjustment for the difference in duration between experi-
mental exposure (e.g., 6 h) and expected human exposure (24 h), (b) calculation of the HEC, and
(c) application of uncertainty factors (UFs). Briefly described in this section are the default methods
often employed by the U.S. EPA in performing the aforementioned steps in RfC derivation, and the
potential benefits afforded by PBPK models in performing the same steps.
Previous
TOC
-------
APPLICATIONS OF PBPK MODELS IN RISK ASSESSMENT 533
RfC Derivation: Determining a Point of Departure Points of departure are derived from
human epidemiological studies or animal toxicological studies. A POD for RfC derivation cannot be
identified or established with only pharmacokinetic data or PBPK models in the absence of dose-
response data. Integrated pharmacokinetic-pharmacodynamic models (Gearhart et al., 1990,
1994; Timchalk et al., 2002) may be capable of predicting response and thus estimating a POD in
the future, and are an area of ongoing research.
Typically, the POD used in the RfC process would be inhalation route specific, and would cor-
respond to exposure concentrations in an experimental or field study (NOAEL, LOAEL) or would be
derived from statistical analysis of dose-response data to obtain a BMC or to its lower confidence
limit (95th percentile), BMCL, associated with a specified response level (generally in the range of 1
to 10% above background; e.g., BMCL10%) (U.S. EPA, 1994, 2000a).
RfC Derivation: Route-to-Route Extrapolation When information on the POD is available
only for a noninhalation route of exposure (e.g., oral route), route-to-route extrapolation may be
conducted (Pauluhn, 2003). Historically, the NOAEL (mg/kg/d) associated with an oral exposure
route was converted to milligrams per day and then to the equivalent inhaled concentration on the
basis of human breathing rate and body weight. Data on the route-specific fraction absorbed, when
available, are used to determine the equivalent inhalation concentration on the basis of equivalent
absorbed doses (U.S. EPA, 1999a). Such simplistic approaches, however, assume that the rates of
ADME and tissue dosimetry of chemicals are the same for a given total dose, regardless of the expo-
sure route and intake rate. These approaches neglect the route-specific differences in pharmacoki-
netics, such as first-pass clearance. First-pass clearance arises when chemicals undergo extensive
metabolism in tissues at portals of entry; this may include the intestines and liver for orally absorbed
compounds or the lungs for inhaled compounds (Benet et al., 1996). Therefore, route-to-route
extrapolation using a more complete pharmacokinetic modeling approach, as previously described
in the Route to Route Extrapolation section, is preferable.
RfC Derivation: Duration Adjustment An RfC addresses continuous exposure of human
populations, so the POD used in its derivation should correspond to 24-h/d exposures (U.S. EPA,
1994). PODs are frequently obtained from animal exposures or occupational exposures that occur
for 6 to 8 h/d, 5 d/wk, so an adjustment to a continuous 24-h exposure, resulting in a lower concen-
tration for continuous exposures, is conducted on the basis of hours per day and days per week
(i.e., 6/24x5/7) (U.S. EPA, 2002). This simple adjustment assumes that "Haber's Rule" applies, i.e.,
that for a given chemical Cxt=k, where C and t are the concentration (mass per unit volume) and
time needed (at that concentration) to produce some adverse effect, and k is a constant associated
with that adverse effect. This approach hypothesizes that doubling the concentration will halve the
time needed to produce a comparable effect level. In pharmacokinetics, the integration of Cxt over
the exposure-response time frame of interest is also referred to as the AUC. If the AUC is not the
dose metric most associated with the adverse effect or various Cxt = k regimens do not result in a
comparable effect level, then Haber's Rule is not applicable (U.S. EPA, 2002). For example, when
data indicate that a given toxicity is more dependent on concentration than on duration (time), this
adjustment would not be used. If the appropriate measure of internal dose is uncertain, the U.S.
EPA uses adjustment to a continuous inhalation exposure based on the Cxt relationship as a matter
of health-protective policy (U.S. EPA, 2002). For additional insights into Haber's Rule (as one in a
family of power functions) and its use in risk assessment, the reader is referred elsewhere (Miller
etal., 2000).
PBPK models may be used to estimate the value of a proposed internal dose metric that
would result from various administered doses (Jarabek etal., 1994; U.S. EPA, 2002). PBPK mod-
els often do not address pharmacodynamic events and assume that these events do not alter the
kinetics for within-day exposures (<24 h). Consistent with U.S. EPA policy (U.S. EPA, 2002), the
dose metric of a chemical for the exposure scenario of the critical study is initially determined
using the PBPK model (e.g., 6 h/d, 5 d/wk); then the atmospheric concentration for a continuous
exposure (24 h/d) during a lifetime or a particular window of exposure that yields the same dose
metric is determined by iterative simulation (see example presented earlier in the Duration
Adjustment section).
Previous I TOC
-------
534 C. M. THOMPSON ET AL
RfC Derivation: Dosimetric Adjustment Factor (Interspecies Extrapolation) In the RfC pro-
cess, a dosimetric adjustment factor (DAF) is applied to the duration-adjusted POD to account for
pharmacokinetic differences between test species and humans to derive HEC (U.S. EPA, 1994). The
DAF depends on the nature of the inhaled toxicant and the MOA as well as the endpoint (local
effects vs. systemic effects). Dosimetry data, if available, in the test animals and humans (including
deposition data, region-specific dosimetry, blood concentration of systemic toxicants) are used to
estimate the DAF. In the absence of such data, knowledge of critical parameters or mathematical
models in the test species and humans is useful in estimating the DAF.
For highly reactive or water-soluble gases that do not significantly accumulate in blood (e.g.,
hydrogen fluoride, chlorine, formaldehyde, volatile organic esters), the DAF is derived for estimates
of the delivery of chemical to different regions of the respiratory tract, based on regional mass trans-
fer coefficients and differences in surface area and ventilation rates (U.S. EPA, 1994). For poorly
water-soluble gases that produce remote effects (e.g., xylene, toluene, styrene), PBPK models are
identified as the preferred approach. Absent a PBPK model, the DAF is determined on the basis of
the ratio of blood:air partition coefficients in animals and humans (U.S. EPA, 1994). For gases that
are water soluble with some blood accumulation (e.g., acetone, ethyl acetate, ozone, sulfur diox-
ide, propanol, isoamyl alcohol) and that have the potential for both respiratory and remote effects,
some combination of the above approaches may be used.
An alternative to the use of DAFs, discussed in the RfC guidance (U.S. EPA, 1994), is to employ
more elaborate or chemical-specific models to make interspecies extrapolations. Various computa-
tional tools are available to determine the uptake and deposition of gases and PM in nasal pathways
and the respiratory tract (Kimbell et al., 1993; Jarabek et al., 1994; Asgharian et al., 1995; Bush et al.,
1998; Iran etal., 1999; Hannaetal., 2001; Bogdanffy and Sarangapani, 2003; U.S. EPA, 2004). PBPK
models are frequently used for systemically distributed gases and vapors, but in conjunction with other
models (e.g., CFD), they may be used for locally acting gases with contact site effects. A limitation of
DAFs is that they do not account for metabolism of gases to more reactive moieties, so PBPK modeling
approaches would clearly be preferable for these compounds if adequate data are available.
RfC Derivation: Application of Uncertainty Factors The uncertainty and variability factors
(UFs) used in RfC and RfD derivation account for extrapolations: (1) from test animals to humans
(interspecies, UFA), (2) for variability within the human population to protect the most sensitive
population (intraspecies variability, UFH), (3) across duration of exposure (subchronic to chronic,
UFS), (4) from LOAEL to NOAEL (UFL), and (5) for poor-quality or missing data in the scientific liter-
ature (database deficiency, UFD) (Jarabek et al., 1994; U.S. EPA, 1994). The product of all UFs
generally should not exceed 3000 (U.S. EPA, 2002), otherwise there may not be sufficient data for
quantitative analysis. If the NOAEL for a chemical with an adequate database was identified in a
chronic study, only the UFA and UFH are used in the assessment. The conventional default value for
UFA of 10 is used in RfC and RfD derivation as an approximation of cross-species scaling resulting in
equivalent effects. Similarly, the default value for UFH of 10 is presumed adequate to account for
variability in the human kinetic and dynamic processes following exposure and to protect poten-
tially sensitive human subpopulations. While the incorporation of data in UF will reduce the uncer-
tainty in the extrapolation, the data may not support a reduction of the uncertainty factor value, as
the data may demonstrate differences greater than or less than the default value.
The values for UFA and UFH arising from historical use and science policy are supported by
empirical information for pharmacokinetics and pharmacodynamics (e.g., isoenzyme levels,
enzyme activity levels, tissue volumes, breathing rates, cell proliferation rates) (Dome et al., 2001 a,
2001 b, 2002). Extrapolations across species or estimates of interindividual variability (e.g., differ-
ences arising from genetic polymorphisms), however, are best done on the basis of chemical-
specific determinants of disposition and effects. Initially, evaluation of various specific determinants
of interspecies differences or human variability is useful, but simple pooling of these determinants
without accounting for covariance or nonlinear interactions may lead to unrealistic estimates for
either UFA or UFH (Lipscomb, 2004). The net impact of various determinants on the UFA and UFH is
more properly evaluated within the integrated and physiologically based context of a PBPK or
biologically based dose-response (BBDR) model.
Previous I TOC
-------
APPLICATIONS OF PBPK MODELS IN RISK ASSESSMENT 535
When data are available to go beyond default uncertainty values, these UFs are subdivided into
their toxicokinetic and toxicodynamic components (IPCS, 2005; U.S. EPA, 2005b). The World
Health Organization's International Programme on Chemical Safety (IPCS) produced guidance on
the development of chemical-specific adjustment factors (CSAFs) (IPCS, 2005). Although the princi-
ples of using chemical-specific data in developing values for UFs have long been endorsed by the
U.S. EPA (U.S. EPA, 1994), and many of the guiding principles in the IPCS document are also com-
ponents of the U.S. EPA risk assessment approach, the U.S. EPA does not use CSAFs per se, due in
part to differences in calculation methods. For instance, the U.S. EPA often separates the pharma-
cokinetic and pharmacodynamic components of the UFA equally (i.e., 10 or 3.16, generally
rounded to 3 each), whereas the IPCS advocates 10°'6 (4.0) and 10°'4 (2.5), respectively (IPCS,
2005; U.S. EPA, 2005c).
When sufficient chemical-specific data are available for PBPK modeling, such models are useful
for characterizing the magnitude of the pharmacokinetic component of the UFA as well as the UFH
used in the RfC and RfD processes. When using PBPK models to adjust for pharmacokinetic differ-
ences between species, a factor of 3 (one-half order of magnitude) is generally retained to account
for remaining uncertainties (U.S. EPA, 1994, 2003; Jarabek, 1995a; Clewell et al., 2002a). How-
ever, chemical-specific information on the pharmacodynamic aspect of inter- and intraspecies dif-
ferences may inform a further reduction or increase of these UFs from default values. It should be
recognized that PBPK and BBDR models are not currently suitable for predicting the magnitude of
LOAEL-NOAEL, subchronic-chronic, or database UFs, although research in these areas is ongoing.
In addition, issues related to whether UFs are most appropriately applied before or after derivation
of a HEC or human equivalent dose continue to be explored (Clewell et al., 2002a).
Applications of PBPK Models in Cancer Risk Assessment
The dose-response assessment portion of cancer risk assessment may vary, depending on MOA
considerations. A CSF is based on a linear extrapolation from the POD (i.e., high-dose to low-dose
extrapolation), or a nonlinear analysis may be applied (U.S. EPA, 2005a). Either approach may also
require interspecies or route-to-route extrapolations for the POD. An integrated PBPK-BBDR
model could likely improve the characterization of a chemical carcinogen's dose-response relation-
ships (e.g., a PBPK model coupled to a clonal expansion and progression model); however, most
such coupled models are still in the developmental stage. PBPK models improve estimation of the
internal dose metric for a chemical carcinogen and play an important role in making explicit or
reducing the uncertainties associated with the high-dose to low-dose, interspecies, route to route,
and intraspecies extrapolations used in the cancer risk assessment process.
Cancer Risk Assessment: High-Dose to Low-Dose Extrapolation The oral CSF or the IUR is
determined by modeling the relationship between the cancer response and the administered dose
or exposure concentration (U.S. EPA, 2005a). According to the cancer guidelines outlined by U.S.
EPA, either a nonlinear or linear extrapolation from the POD is conducted, as appropriate for the
MOA of the carcinogen (U.S. EPA, 2005a). The use of internal dose or delivered dose in such anal-
ysis has been encouraged.
Because high doses of chemicals are often administered in rodent cancer bioassays, the number
of tumors observed in such studies is not always directly proportional to the exposure dose. Thus,
the dose-response relationships may appear complex, in part due to nonlinearity in the pharmaco-
kinetic processes occurring at high exposure doses. In other words, the target tissue dose of the
toxic moiety may be disproportional to the administered doses used in animal bioassays. Therefore,
dose-response analysis based on an appropriate dose metric may result in linearization of the rela-
tionship (Andersen et al., 1987; Clewell et al., 2002a). The slope factor derived using the dose met-
ric-response curve has units of (dose metric)"1. For nonlinear analyses, a POD based on an applied
or external exposure can be converted to an equivalent internal dose metric and subsequently run
to predict the corresponding applied or external exposure dose in humans.
Cancer Risk Assessment: Interspecies Extrapolation For gases and PM, the default proce-
dure for interspecies extrapolation involves the derivation of an HEC (U.S. EPA, 1994; Jarabek,
1995a, 1995b). For oral exposures, when a PBPK model is not available, the U.S. EPA performed
Previous I TOC
-------
536 C. M. THOMPSON ET AL
interspecies scaling of doses according to body mass raised to the three-fourths power (BWOJS) (U.S.
EPA, 2002, 2005a). This procedure presumes that equal doses in these units (i.e., in mg/kg°-7S/d),
when administered daily over a lifetime, will result in equal lifetime cancer risks across mammalian
species. The three-fourths power scaling relationship (sometimes called Kleiber's law, from his orig-
inal proposition in a 1932 article) is generally attributed to differences in rates of basal metabolism.
There remains considerable dissent as to the generality of the BW°'7S scaling factor, the underlying
biological rationale, and the value of the exponent (e.g., some proponents advocate a BWa67
scaling based on surface area differences), particularly for toxicological effects of xenobiotic
chemicals in contrast to endogenous anabolic and catabolic processes (Agutter & Wheatley, 2004).
Nonetheless, BW°-7S scaling remains the current U.S. EPA default approach for oral cancer assess-
ments (U.S. EPA, 2005a).
The nature and slope of the dose-response relationship for carcinogens may not be identical in
test species and humans due to pharmacokinetic and pharmacodynamic differences (Monro,
1994). If appropriate data are available in both the test species and humans (e.g., tissue or blood
concentrations), then interspecies extrapolations of an equivalent carcinogenic or safe dose may be
conducted. In the absence of a complete data set, PBPK models provide a means to characterize
the relationship between the applied dose and the internal dose of a carcinogen in the species of
interest for subsequent extrapolation to humans (Andersen etal., 1987).
Cancer Risk Assessment: Intraspecies Extrapolation Cancer risk assessments have not gen-
erally considered intraspecies variability in pharmacokinetics or pharmacodynamics; however, use
of upper bounds on a maximum likelihood estimate (MLE) is generally thought to be conservative
and thus protective of susceptible populations. One advantage of using PBPK models is that this
assumption can be tested when sufficient data are available to link risk to an identifiable pharmaco-
kinetic outcome or dose metric. These models also describe the impact of variations in pharmacok-
inetic determinants such as polymorphisms in xenobiotic metabolizing enzymes. For instance, the
impact of interindividual variability, including polymorphisms in glutathione transferase theta, on
the disposition of methylene chloride was shown to impact cancer risk (Portier & Kaplan, 1989,
Casanova etal., 1996, el-Masri etal., 1999; Jonsson & Johanson, 2001).
PBPK models are useful in evaluating differences in chemical disposition among adults and chil-
dren (Clewell et al., 2002b, 2004; Gentry et al., 2003; Price et al., 2003; Ginsberg et al., 2004). In
this regard, susceptibility to children might be assessed by examining whether the dose metric is
expected to be higher in the young. In addition, recent U.S. EPA guidance suggests that additional
adjustment factors (age-dependent adjustment factors, ADAFs) to the cancer slope or unit risk value
be considered to account for enhanced susceptibility in early life (i.e., to neonates and young chil-
dren) from exposure to carcinogens exhibiting a mutagenic MOA (U.S. EPA, 2005b). It should be
noted that ADAFs generally account for increased susceptibility related to adult/child differences in
pharmacodynamic processes such as cell proliferation rates and numbers of cells with proliferative
potential (U.S. EPA, 2005b); thus even if pharmacokinetic differences are accounted for when
extrapolating from a POD, these adjustment factors would still likely be made for chemicals with a
mutagenic MOA unless, perhaps, the carcinogenicity data were derived from young animals or
humans. Furthermore, when assessing the less-than-lifetime exposures occurring in childhood, the
guidelines stipulate consideration of adult-children differences in key exposure factors (e.g., skin
surface area, drinking water ingestion rates) (U.S. EPA, 2005b).
Cancer Risk Assessment: Route Extrapolation As with RfC and RfD derivation, PBPK mod-
els can facilitate the conduct of route-to-route extrapolation by accounting for the route-specific
rate and magnitude of absorption, first-pass effect, and metabolism (Clewell & Andersen, 2004).
The slope factor or the POD associated with one exposure route may be translated into applied
dose for another exposure route by simulating the tissue dose of toxic moiety associated with the
exposures by each route (Gerrity et al., 1990; U.S. EPA, 2000b).
Brief Summary of Select Examples of PBPK Model Use in Risk Assessment
Regulatory agencies in the United States, Canada, and Europe are increasingly applying PBPK mod-
els in risk assessments, though there are far more models published in the peer-reviewed literature.
Previous I TOC
-------
APPLICATIONS OF PBPK MODELS IN RISK ASSESSMENT 537
Upon evaluation, some published models are found adequate for risk assessment applications,
while in other cases the published models reflect research efforts that did not necessarily address
the species, exposure routes, target tissues, or other factors that would be required for risk assess-
ment applications (DeWoskin et al., 2007). Examples from Canada and Europe in which models
were evaluated and deemed adequate or inadequate for risk assessment purposes are described in
a publication from the International Workshop on the Development of Good Modeling Practice for
PBPK Models (Loizou et al., 2008). Described next are two examples where PBPK models have
been used by the U.S. EPA; more detailed examples are described elsewhere (U.S. EPA, 2006a;
DeWoskin etal., 2007).
Example of Reference Value Derivation and Application of Uncertainty Factors The RfD der-
ivation for ethylene glycol monobutyl ether (EGBE) compared four dose-response modeling
approaches: (1) a default approach, (2) a PBPK modeling approach, (3) a BMD analysis, and (4) a
combined PBPK/BMD approach. Each approach necessitated the application of different UFs in
deriving the RfD values (Table 4). More about this assessment can be found elsewhere (U.S. EPA,
1999b; DeWoskin et al., 2007), but it is briefly described here to demonstrate the application of
PBPK modeling in reference value derivation, and more broadly to illustrate how various dose-
response modeling approaches can influence UFs applied in such derivations.
Two critical effects observed from EGBE exposure are hemolysis and hepatocellular toxicity. In
rats, the primary toxic metabolite responsible for these effects are 2-butoxyacetic acid (BAA); the
peak concentration of this metabolite in blood (i.e., Cmax BAA) was identified as an appropriate
dose metric. In each modeling approach, the UFD and UFS were set to 1. A value of 10 was chosen
for the UFH due to lack of data on human variability in EGBE sensitivity. For interspecies extrapola-
tion, in vivo and in vitro studies indicate that, pharmacodynamically, humans are less sensitive than
rats to the hematologic effects of EGBE; therefore the nonpharmacokinetic portion of the UFA was
also set to unity.
All other UFs were impacted by the choice of modeling approach. In the standard default and
BMD approaches, the UFA was set to 3 because no pharmacokinetic adjustments were considered
in developing a POD[HEC]. In contrast, the PBPK and combined PBPK/BMD approaches estimated a
POD[HEC] based on the dose metric BAA; thus with interspecies differences in pharmacokinetics
accounted for, the UFA was set to 1. In regards to extrapolating from a LOAEL to a NOAEL, UFL was
set to 3 for the default and PBPK approaches because data indicated that the LOAEL was very near
the threshold level for the critical effects of concern. In contrast, the UFL was set to 1 for the BMD
analyses because of the minimal and precursive nature (cell swelling) of the critical effect, and the
fact that a BMDOS for a minimally adverse effect is typically deemed to be equivalent to a NOAEL
for continuous data sets.
In the default approach, the rat LOAEL for hepatocellular cytoplasmic changes was chosen
because it provided the more conservative RfD among the data amenable to default approaches.
The derivation of this RfD is shown in Table 4. In the PBPK modeling approach, the RfD is based on
hemolysis, and a LOAEL of 59 mg/kg-d was chosen as the POD from which to estimate Cmax BAA in
blood using the PBPK model developed by Corley et al. (1994, 1997). The model predicted Cmax
BAA in rat blood of 103 |iM. Subsequently, the human PBPK model simulations indicated that 7.6
mg/kg-day EGBE in drinking water would yield the same Cmax BAA of 103 |iM in humans. This
human equivalent dose (HED) of 7.6 mg/kg-d was divided by the UFs in Table 4.
BMD analysis yielded a BMDOS of 49 mg/kg-d as the POD. In a combined approach (BMD/
PBPK), the rat PBPK model was exercised to simulate blood concentrations of BAA resulting from
TABLE 4. Comparison of RfD derivations for EGBE
Default PBPK BMD BMD/PBPK
POD
UFH,A,L
RfD
55 mg/kg-d
10,3,3
0.6 mg/kg-d
7.6 mg/kg-d
10,1,3
0.3 mg/kg-d
49 mg/kg-d
10,3, 1
2 mg/kg-d
5.1 mg/kg-d
10, 1,1
0.5 mg/kg-d
Previous I TOC
-------
538
C. M. THOMPSON ET AL
each of the exposure concentrations in the toxicity study. This BMD analysis yielded a BMDLOS of
64 uM, which corresponds to an HED of 5.1 mg/kg-d as the POD. Derivations of the RfD values
from both of these BMD approaches are shown in Table 4.
It is important to note in this example that the POD for the various approaches differed by
about an order of magnitude depending on whether a PBPK model was used. However, after
applying the various UFs, the RfD values were in relatively closer agreement. This result should not
necessarily be interpreted as coalescence around a "correct answer" because each POD was
reduced by different UF values—each somewhat subjective in nature. Thus, which RfD value to
choose should not arise from the modeling approach per se, but rather the judicious use of data to
support a particular RfD value (PBPK/BMD approach in this case). Perhaps most importantly, an
analysis of which modeling approach is truly best will not be possible until further methodologies
are derived for quantifying the uncertainty in the various approaches. To this end, a recent work-
shop was held to describe the state of the science as well as research and implementation needs for
statistical analyses to characterize uncertainty and variability in PBPK models (Barton et al., 2007).
Example in Cancer Risk Assessment Returning to the earlier example of DCM, in addition to
dose metric evaluation, the PBPK model for DCM was used for low-dose extrapolation (Andersen et
al., 1987). The model prediction of the target tissue dose of the glutathione conjugate resulting from
6-h inhalation exposures to 1-4000 ppm DCM is presented in Figure 7. The estimation of target tis-
sue dose of DCM-glutathione conjugate by linear back-extrapolation gives rise to a 21-fold higher
estimate than that obtained by the PBPK modeling approach. This discrepancy arises from the non-
linear behavior of DCM metabolism at high-exposure concentrations. At exposure concentrations
exceeding 300 ppm, the cytochrome P-450-mediated oxidation pathway is saturated, giving rise to
a corresponding disproportionate increase in the flux through glutathione conjugation pathway. By
accounting for the species-specific differences in metabolism rates and physiology in the PBPK
re
•O
•D
£
Q.
re
•ffi
•D
IS
s
O
10000
1000
100
0.01-^
10
100
PPM in air
1000
10000
FIGURE 7. PBPK model predictions of glutathione (GST)-pathway metabolites in mouse liver. The three curves are for a linear extrapo-
lation from the bioassay exposures of 2000 and 4000 ppm (solid thin line), the expected tissue dose based on model parameters for the
mouse (solid thick line), and the expected dose based on human model parameters (dotted line). The curvature occurs because oxida-
tion reactions become saturated as inhaled concentration increases above several hundred ppm. Reprinted Andersen et al. (1987), with
permission from Elsevier.
Previous
TOC
-------
APPLICATIONS OF PBPK MODELS IN RISK ASSESSMENT 539
model, the target tissue dose for humans was estimated to be some 2.7 times lower than that for the
mouse. The target tissue dose-based slope factor was subsequently used for characterizing the cancer
risk associated with human exposures (Andersen et al., 1987; Reitz et al., 1989; Haddad et al.,
2001 b). The case of DCM exemplifies how PBPK models can be used to improve the dose-response
relationship on the basis of appropriate dose metrics, thus leading to scientifically sound conduct of
interspecies and high-dose to low-dose extrapolations essential for cancer risk assessments.
OTHER APPLICATIONS OF PBPK MODELS IN RISK ASSESSMENT
Use of Pharmacokinetic Data and Models in Variability and Uncertainty Analysis
There is increasing focus on the application of variability and uncertainty analyses with PBPK
models. Although characterizing the variability and uncertainty in a PBPK model may be considered
an aspect of model development, such analyses also have the potential to improve risk assessments
by also characterizing the uncertainty in the overall characterization of risk. For instance, evaluation
of alternative dose metrics (e.g., Cmax vs. AUC) is useful in characterizing the uncertainty in the use
of a PBPK model; examples include DCM (Andersen et al., 1987) and retinoic acid (Clewell et al.,
1997). Likewise, uncertainty and variability may also be addressed through specialized analyses
such as probabilistic simulations using distributions for physiological, biochemical, and physico-
chemical parameters as well as measurement error contributing to uncertainty. Approaches for
incorporating human variability in risk assessment are reviewed in the section Estimating Intraspe-
cies Variability; both variability and uncertainty in PBPK models are discussed in greater detail else-
where (U.S. EPA, 2006a, Chiu et al., 2007; Barton et al., 2007).
Use of Pharmacokinetic Data and Models in Exposure Assessment and Interpretation
of Biomonitoring Data
The conventional approach to exposure assessment involves the calculation of applied dose for
each route of exposure based on information about the concentration of the chemical in the
medium, frequency and duration of exposure, rate of contact with the medium, and body weight
of the individual (Paustenbach, 2000). As described throughout this article, when sufficient data are
available, PBPK models allow for the prediction of absorbed or internal dose when the exposure
concentration is known (as in experimental settings). Once an internal dose is calculated for a given
species, route or duration, the model is adjusted for alternative species and exposure scenarios and
essentially exercised in reverse in order to determine exposure concentrations associated with the
internal dose metric. From this approach, it is evident that PBPK models have the potential to utilize
pharmacokinetic measurements or other biomarker data to reconstruct previous exposure(s) to
environmental toxicants or interpret biomonitoring data (Krishnan et al., 1992; Fennell et al., 1992;
Csanady et al., 1996; Timchalk et al., 2001, 2004; Tan et al., 2006). Indeed, such models were
used to establish biological exposure indices (e.g., breath, blood, or breath concentrations) to pro-
tect workers from harmful exposures to solvents (Perbellini et al., 1990; Leung, 1992; Kumagai &
Matsunaga, 1995; Thomas et al., 1996; Droz et al., 1999) or in epidemiology studies to reconstruct
human exposures over time (Roy & Georgopoulos, 1998; Canuel et al., 2000). In this regard, com-
prehensive PBPK models are being developed that provide estimates of an internal tissue dose from
multiroute (oral, inhalation, dermal) or multichemical exposures (Roy et al., 1996; Rao & Ginsberg,
1997; Corley et al., 2000; Liao et al., 2002; Levesque et al., 2002).
Use of Pharmacokinetic Data and Models in Risk Assessment of Chemical Mixtures
PBPK models facilitate risk assessment of chemical mixtures by estimating the change in dose
metrics due to multichemical interactions (Haddad et al., 2001 a). Tissue dosimetry-based assess-
ments for mixtures require adequately evaluated PBPK models for the mixture (in the test species
and in humans), as well as dose-response values for the individual chemicals (e.g., CSF, RfD, RfC).
An approach for using PBPK models in risk assessment of mixtures of systemic toxicants or carcino-
gens exhibiting threshold mechanism of action, would consist of (Haddad et al., 2001 a):
Previous I TOC
-------
540 C. M. THOMPSON ET AL
1. Characterizing the dose metrics associated with dose-response values for the mixture components.
2. Obtaining predictions of dose metrics of each mixture component in humans, based on informa-
tion on exposure levels provided as input to the mixture PBPK model.
3. Determining the sum total of the ratios of the results of steps (1) and (2) for each component
during mixed exposures for each target organ or critical effect.
Similarly, for carcinogens with slope factor (Haddad et al., 2001 a):
1. The dose metric-based slope factor is established for each component using the animal PBPK
model.
2. The dose metric associated with human exposure concentrations is established using mixture
PBPK models.
3. The results of steps (1) and (2) are combined to determine the potentially altered cancer
response during mixed exposures.
Risk assessments based on the use of PBPK models for single chemicals and mixtures, as
detailed in previous sections, account for only the pharmacokinetic aspect or, more specifically,
target tissue exposure to toxic moiety. If these tissue exposure simulations are combined with phar-
macodynamic models, then better characterization of dose-response relationships and prediction of
PODs (NOAEL, BMD, and BMC) may become possible.
Linkage of PBPK Models With Pharmacodynamic Models
The identification of PODs by simulation may become possible with the use of BBDR models.
These models would require the linkage of quantitative descriptions of pharmacokinetics and phar-
macodynamics via mechanism of action. Accordingly, the output of PBPK models is linked to the
pharmacodynamic (PD) model using an equation that reflects the researcher's hypothesis of how
the toxic chemical participates in the initiation of cellular changes leading to measurable toxic
responses. For example, certain DMA adducts induce mutations, some metabolites kill individual
cells, and expression of growth factors can stimulate cell proliferation. In each of these cases, the
temporal change in the dose metric simulated by the PBPK model is linked with mathematical
descriptions of the process of adduct formation, cytotoxicity, or proliferation in the BBDR models to
simulate the quantitative influence of these processes on tumor outcome.
For example, using a PD model, the rate of chloroform metabolism (|imol/g liver/h) was related
to fraction of liver cells killed (Page et al., 1997). In this model, a series of differential equations was
used to simulate time-dependent changes in the number of hepatocytes in the liver as a function of
basal rates of cell division and death, chloroform-induced cytolethality, and regenerative replica-
tions (Conolly & Butterworth, 1995; Page et al., 1997).
Table 5 presents a list of PD models for cancer and noncancer endpoints. A characteristic
of several of these PD models is that they are able to simulate the normal physiological processes
(e.g., cell proliferation rates, hormonal cycle) and additionally account for the ways in which chem-
icals perturbate such phenomena, leading to the onset and progression of injury. However, PD
models that are linked with PBPK models are not available for a number of adverse effects and
modes of action. This situation is a result, in part, of the complex nature of these models and the
extensive data requirements for development and evaluation of these models for various exposure
and physiological conditions.
With the availability of integrated pharmacokinetic-pharmacodynamic (PK-PD) models, the sci-
entific basis of the process of estimating/extrapolating PODs and characterizing the dose-response
curve will be significantly enhanced. Additionally, such a modeling framework will facilitate a quan-
titative analysis of the impact of pharmacodynamic determinants on the toxicity outcome, such that
the magnitude of the pharmacodynamic component of the interspecies and intraspecies factors
may be characterized more confidently. Even though some PBPK models have been used in RfD,
RfC, and unit risk estimate derivation, the need for applying such models (where possible) should
be continuously explored.
Previous I TOC
-------
APPLICATIONS OF PBPK MODELS IN RISK ASSESSMENT
541
TABLE 5. Examples of biologically based models of relevant endpoints and toxicological processes
Toxicity endpoint
or process Features Chemical studied References
Cancer
Cholinesterase
inhibition
Developmental
toxicity
Estrus cycle
Gene expression
Granulopoiesis
Nephrotoxicity
Teratogenic effect
Simulation of relative roles of initiation,
promotion, cytolethality, and
proliferation
Simulation of dose-dependent
inhibition of plasma cholinesterase,
red blood cell acetyl cholinesterase
and brain acetyl cholinesterase,
and nontarget B-esterase
Simulation of altered cell kinetics as
the biological basis of developmental
toxicity
Simulation of interactions of estradiol
and luteinizing hormone
Simulation of induction of CYP1A1/2
protein expression in multiple tissues
Simulation of loss of proliferating cells
and loss of functional cells
Simulation of induction of renal
2|i-globulin in male rat kidney as a
function of proteolytic degradation
and hepatic production
Sensitivity distribution of embryo
as a function of age and stage of
development
2-Acetylaminofluorine
Chloroform
Dimethylnitrosamine
Formaldehyde
Polychlorinated biphenyls
Pentachlorobenzene
Saccharin
Organophophates
Methylmercury
Endocrine-modulating
substances
Tetrachlorodibenzodioxin
Cyclophospham ide
2,2,4-Trimethyl-2-phenol
Hydroxyurea
Conolly et al. (2003); Tan et al. (2003);
Thomas et al. (2000); Conolly and
Andersen (1997); Conolly and Kimbell
(1994); Chen (1993); Luebeck et al.
(1991); Cohen and Ellwein (1990);
Moolgavkar and Luebeck (1990);
Moolgavkar and Knudson (1981 );
Moolgavkar and Venzon (1979);
Armitageand Doll (1957)
Timchalk et al. (2002); Gearhart et al.
(1994, 1990)
Faustman et al. (1 999); Leroux et al.
(1996)
Andersen etal. (1997)
Santostefanoetal. (1998)
Steinbach et al. (1 980)
Kohn and Melnick (1999)
Lueckeetal. (1997)
SUMMARY
Practical computer implementation of PBPK modeling has been feasible for over 20 yr. Despite
an early impact of PBPK modeling on the risk assessment of dichloromethane, PBPK model usage in
risk assessment has been relatively sparse. Today, however, there is growing interest, willingness
and expertise to apply these models in health risk assessments. Scientific progress in the under-
standing of lifestage and genetic differences in dosimetry and their impacts on variability in suscep-
tibility, as well as ongoing development of analytical methods to characterize the uncertainty in
PBPK models makes future PBPK model use in risk assessment increasingly likely. As such, it is
anticipated that when PBPK models are used to express adverse tissue responses in terms of the
internal target tissue dose of the toxic moiety rather than the external concentration, the scientific
basis of, and confidence in, risk assessments will be enhanced.
REFERENCES
Agutter, P. S., and Wheatley, D. N. 2004. Metabolic scaling: Consensus or controversy? Jheor. Biol. Med. Model. 1:1 3.
Andersen, M. E., Mills, J. J., Gargas, M. L., Kedderis, L., Birnbaum, L. S., Neubert, D., and Greenlee, W. F. 1993. Modeling receptor-
mediated processes with dioxin: implications for pharmacokinetics and risk assessment. Risk Ana!. 13:25-36.
Andersen, M. E. 1995. Development of physiologically based pharmacokinetic and physiologically based pharmacodynamic models for
applications in toxicology and risk assessment. Toxicol. Lett. 79:35-44.
Andersen, M. E., Clewell, H. J. Ill, Gargas, M. L., Smith, F. A., and Reitz, R. H. 1987. Physiologically based pharmacokinetics and the risk
assessment process for methylene chloride. Tox/co/. Appl. Phannacol. 87:185-205.
Previous
TOC
-------
542 C. M. THOMPSON ET AL
Andersen, M. E., and DennisonJ. E. 2001. Mode of action and tissue dosimetry in current and future risk assessments. Sci. Total Environ.
274:3-14.
Andersen, M. E., Clewell, H. J., Gearhart, J., Allen, B. C., and Barton, H. A. 1997. Pharmacodynamic model of the rat estr us cycle in rela-
tion to endocrine disruption.). Toxicol. Environ. Health 52:1 89-209.
Andersen, M., Sarangapani, R., Gentry, R., Clewell, H., Covington, T., and Frederick, C. B. 1999. Application of a hybrid CFD-PBPK
nasal dosimetry model in an inhalation risk assessment: an example with acrylic acid. Tox/co/. Sci. 57:312-325.
Andersen, M. E., Green, T., Frederick, C. B., and Bogdanffy, M. S. 2002. Physiologically based pharmacokinetic (PBPK) models for nasal
tissue dosimetry of organic esters: Assessing the state-of-knowledge and risk assessment applications with methyl methacrylate and
vinyl acetate. Regul. Toxicol. Pharmacol. 36:234-245.
Andersen, M. E., and Jarabek, A. M. 2001. Nasal tissue dosimetry-issues and approaches for "Category 1" gases: A report on a meeting
held in Research Triangle Park, NC, February 11-12, 1998. Inhal. Toxicol. 13:415-435.
Armitage, P., and Doll, R. 1957. A two-stage theory of carcinogenesis in relation to the age distribution of human cancer. Br. ]. Cancer
11:161-169.
Asgharian, B., Wood, R., and Schlesinger, R. B., 1995. Empirical modeling of particle deposition in the alveolar region of the lungs: A
basis for interspecies extrapolation. Fundam. Appl. Toxicol. 27:232-238.
Aylward, L. L., Hays, S. M., Karch, N. J., and Paustenbach, D. J. 1996. Relative susceptibility of animals and humans to the cancer hazard
posed by 2,3,7,8-tetrachlorodibenzo-p-dioxin using internal measures of dose. Environ. Sci. Techno/. 30:3534-3543.
Barton, H. A., Deisinger, P. J., English, J. C., Gearhart, J. N., Faber, W. D., Tyler, T. R., Banton, M. I., Teeguarden, J., and Andersen, M. E.
2000. Family approach for estimating reference concentrations/doses for series of related organic chemicals. Toxicol. Sci. 54:251-261.
Barton, H. A., Chiu, W. A., Setzer, R. W., Andersen, M. E., Bailer, A. J., Bois, F. Y., Dewoskin, R. S., Hays, S., Johanson, G., Jones, N.,
Loizou, G., Macphail, R. C., Portier, C. J., Spendiff, M., Tan, Y. M. 2007. Characterizing uncertainty and variability in physiologically-
based pharmacokinetic (PBPK) models: State of the science and needs for research and implementation. Toxicol. Sci. 99:395-402.
Benet, L. Z., Kroetz, D. L., and Sheiner, L. B. 1996. Pharmacokinetics: the dynamics of drug absorption, distribution, metabolism, and
elimination. In Goodman and Oilman's The pharmacological basis of therapeutics, eds. J. G. Hardman and L. E. Limbird, pp. 3-27.
New York: McGraw-Hill.
Benignus, V. A., Boyes, W. K., and Bushnell, P. J. 1 998. A dosimetric analysis of behavioral effects of acute toluene exposure in rats and
humans. Toxicol. Sci. 43:186-195.
Bogdanffy, M. S., Sarangapani, R., Plowchalk, D. R., Jarabek, A., and Andersen, M. E. 1999. A biologically risk assessment for vinyl
acetate-induced cancer and noncancer inhalation toxicity. Toxicol. Sci. 51:1 9-35.
Bogdanffy, M. S., Plowchalk, D. R., Sarangapani, R., Starr, T. B., Andersen, M. E. 2001. Mode-of-action-based dosimeters for interspe-
cies extrapolation on vinyl acetate inhalation risk. Inhal. Toxicol. 1 3:377-396.
Bogdanffy, M. S., and Sarangapani, R. 2003. Physiologically-based kinetic modeling of vapours toxic to the respiratory tract. Toxicol. Lett.
138:103-117.
Bouchard, M., Brunei, R. C., Droz, P. O., and Carrier, G. 2001. A biologically based dynamic model for predicting the disposition of
methanol and its metabolites in animals and humans. Toxicol. Sci. 64:169-184.
Boyes, W. K., Bushnell, P. J., Crofton, K. M., Evans, M., and Simmons, J. E. 2000. Neurotoxic and pharmacokinetic responses to trichlo-
roethylene as a function of exposure scenario. Environ. Health Perspect. 108(Suppl. 2):31 7-322.
Brown, R. P., Delp, M. D., Lindstedt, S. L., Rhomberg, L. R., and Beliles, R. P. 1997. Physiological parameter values for physiologically
based pharmacokinetic models. Toxicol. Ind. Health 13:407-484.
Bruckner, J. V., Keys, D. A., and Fisher, J. W. 2004. The Acute Exposure Guideline Level (AEGL) program: Applications of physiologically
based pharmacokinetic modeling. ]. Toxicol. Environ. Health A 67:621-634.
Bush, M. L., Frederick, C. B., Kimbell, J. S., and Ultman, J. S. 1998. A CFD-PBPK hybrid model for simulating gas and vapor uptake in
the rat nose. Toxicol. Appl. Pharmacol. 1 50:133-145.
Bushnell, P. J. 1997. Concentration-time relationships for the effects of inhaled trichloroethylene on signal detection behavior in rats.
Fundam. Appl. Toxicol. 36:30-38.
Canuel, G., Viau, C., Krishnan, K., 2000. A modeling framework for back-calculating ambient concentrations from data on biomarkers.
International Conference on Health Sciences Simulation. Proc. International Conference on Health Sciences Simulation, pp. 97-102.
San Diego, CA.
Casanova, M., Conolly, R. B., Heck, and H. A. 1 996. DNA-protein cross-links (DPX) and cell proliferation in B6C3F., mice but not Syrian
golden hamsters exposed to dichloromethane: Pharmacokinetics and risk assessment with DPX as dosimeter. Fundam. Appl.
Toxicol. 31:103-116.
Chen, C. W. 1993. Armitage-Doll two-stage model: Implications and extension. Risk Anal. 13:273-279.
Chiu, W. A., Barton, H. A., DeWoskin, R. S., Schlosser, P., Thompson, C. M., Sonawane, B., Lipscomb, J. C., and Krishnan, K. 2007.
Evaluation of physiologically based pharmacokinetic models for use in risk assessment. ]. Appl. Toxicol. 27:21 8-237.
Clark, L. H., Setzer, R. W., and Barton, H. A. 2004. Framework for evaluation of physiologically-based pharmacokinetic models for use
in safety or risk assessment. Risk Anal. 24:1697-1 71 7.
Clarke, D. O., DuignanJ. M., and Welsch, F. 1992. 2-Methoxyacetic acid dosimetry-teratogenicity relationships in CD-1 mice exposed
to 2-methoxyethanol. Toxicol. Appl. Pharmacol. 114:77.
Clarke, D. O., Elswick, B. A., Welsch, F., and Conolly, R. B. 1993. Pharmacokinetics of 2-methoxyethanol and 2-methoxyacetic acid in
the pregnant mouse: A physiologically-based mathematical model. Toxicol. Appl. Pharmacol. 1 21:239-252.
Clewell, H. J. Ill, and Andersen, M. E. 1 985. Risk assessment extrapolations and physiological modeling. Toxicol. Ind. Health 1:111-131.
Clewell, H. J. Ill, and Andersen, M. E. 1 987. Dose, species and route extrapolation using physiologically-based pharmacokinetic models.
Drink. Water Health 8:159-182.
Previous I TOC
-------
APPLICATIONS OF PBPK MODELS IN RISK ASSESSMENT 543
Clewell, H. J. Ill, and Andersen, M. E. 2004. Applying mode-of-action and pharmacokinetic considerations in contemporary cancer risk
assessments: An example with trichloroethylene. Crit. Rev. Toxicol. 34:385-445.
Clewell, H. J. Ill, and Jarnot, B. M. 1994. Incorporation of pharmacokinetics in noncancer risk assessment: Example with chloropentaflu-
orobenzene. Risk Ana!. 14:265-276.
Clewell, H. J. Ill, Andersen, M. E., Willis, R. J., and Latriano, L. 1997. A physiologically based pharmacokinetic model for retinoic acid
and its metabolites.). Acad Dermatol. 36:S77-S85.
Clewell, H. J. Ill, Gentry, P. R., Covington, T. R., and Gearhart, J. M. 2000. Development of a physiologically based pharmacokinetic
model of trichloroethylene and its metabolites for use in risk assessment. Environ. Health Perspect. 108:283-305.
Clewell, H. J., Gentry, P. R., Gearhart, J. M., Allen, B. C, and Andersen, M. E. 2001. Comparison of cancer risk estimates for vinyl chlo-
ride using animal and human data with a PBPK model. Sci. Total Environ. 274:37-66.
Clewell, H. J. Ill, Andersen, M. E., and Barton, H. A. 2002a. A consistent approach for the application of pharmacokinetic modeling in
cancer and noncancer risk assessment. Environ. Health Perspect. 110:85-93.
Clewell, H.J., Teeguarden, J., McDonald, T., Sarangapani, R., Lawrence, G., Covington, T., Gentry, R., and Shipp, A. 2002b. Review and
evaluation of the potential impact of age- and gender-specific pharmacokinetic differences on tissue dosimetry. Crit. Rev. Toxicol.
32:329-389.
Clewell, H. J., Gentry, P. R., Covington, T. R., Sarangapani, R., and Teeguarden, J. G. 2004. Evaluation of the potential impact of age-
and gender-specific pharmacokinetic differences on tissue dosimetry. Toxicol. Sci. 79:381-393.
Cohen, S. M., and Ellwein, L. B. 1990. Cell proliferation in carcinogenesis. Science 249:1007-1011.
Collins, J. 1 987. Prospective predictions and validations in anti-cancer therapy, pp. 431-440. Washington, DC: National Academy Press.
Conolly, R. B., and Andersen, M. E. 1997. Hepatic foci in rats after diethyl-nitrosamine initiation and 2,3,7,8-tetrachlorodibenzo-p-
dioxin promotion: Evaluation of a quantitative two-cell model and of CYP 1A1/1A2 as a dosimeter. Toxicol. Appl. Pharmacol.
146:281-293.
Conolly, R. B., and Butterworth, B. E. 1995. Biologically based dose response model for hepatic toxicity: A mechanistically based
replacement for traditional estimates of noncancer risk. Toxicol. Lett. 82-83:901-906.
Conolly, R. B., and Kimbell, J. S. 1994. Computer simulation of cell growth governed by stochastic processes: Application to clonal
growth cancer models. Toxicol. Appl. Pharmacol. 124:284-295.
Conolly, R. B., Kimbell, J. S., Janszen, D., Schlosser, P. M., Kalisak, D., Preston, J., and Miller, F. J. 2003. Biologically motivated computa-
tional modeling of formaldehyde carcinogenicity in the F344 rat. Toxicol. Sci. 75:432-447.
Corley, R. A., Bormett, G. A., and Ghanayem, B. I. 1994. Physiologically based pharmacokinetics of 2-butoxyethanol and its major
metabolite, 2-butoxyacetic acid, in rats and humans. Toxicol. Appl. Pharmacol. 129:61-79.
Corley, R. A., Markham, D. A., Banks, C., Delorme, P., Masterman, A., and Houle, J. M. 1 997. Physiologically based pharmacokinetics
and the dermal absorption of 2-butoxyethanol vapor by humans. Fundam. Appl. Toxicol. 39:120-1 30.
Corley, R. A., Gordon, S. M., and Wallace, L. A. 2000. Physiologically based pharmacokinetic modeling of the temperature-dependent
dermal absorption of chloroform by humans following bath water exposures. Toxicol. Sci. 53:13-23.
Corley, R. A., Mast, T. J., Carney, E. W., Rogers, J. M., and Daston, G. P. 2003. Evaluation of physiologically based models of pregnancy
and lactation for their application in children's health risk assessments. Crit. Rev. Toxicol. 33:137-211.
Cruzan G., Carlson G. P., Johnson K. A., Andrews L. S., Banton M. I., Bevan C., and Cushman J. R. 2002. Styrene respiratory tract toxicity
and mouse lung tumors are mediated by CYP2F-generated metabolites. Regul. Toxicol. Pharmacol. 35:308-31 9.
Csanady, G. A., Kreuzer, P. E., Baur, C., and Filser, J. G. 1996. A physiological toxicokinetic model for 1,3-butadiene in rodents and
man: Blood concentrations of 1,3-butadiene, its metabolically formed epoxides, and of haemoglobin adducts—Relevance of
glutathione depletion. Toxicology 113:300-305.
DeWoskin, R. S., Lipscomb, J. C., Thompson, C. M., Chiu, W. A., Schlosser, P., Smallwood, C., Swartout, J., Teuschler, K., and Marcus,
A. 2007. Pharmacokinetic/physiologically based pharmacokinetic models in integrated risk information system assessments. In
Toxicokinetics and risk assessment, eds. J. C. Lipscomb and E. V. Ohnaian, pp. 301-348. New York: Informa Healthcare.
Dome, J. L., Walton, K., and Renwick, A. G. 2001a. Human variability in glucuronidation in relation to uncertainty factors for risk
assessment. Food Chem. Toxicol. 39:1153-11 73.
Dome, J. L., Walton, K., and Renwick, A. G. 2001 b. Uncertainty factors for chemical risk assessment, human variability in the pharmaco-
kinetics of CYP1A2 probe substrates. Food Chem. Toxicol. 39:681-696.
Dome, J. L., Walton, K., Slob, W., and Renwick, A. G. 2002. Human variability in polymorphic CYP2D6 metabolism: Is the kinetic
default uncertainty factor adequate? Food Chem. Toxicol. 40:1633-1656.
Droz, P. O., Berode, M., and Jang, J. Y. 1999. Biological monitoring of tetrahydrofuran: Contribution of a physiologically based pharma-
cokinetic model. Am. Ind. Hyg. Assoc. J. 60:243-248.
el-Masri, H. A., Bell, D. A., and Portier, C. J. 1999. Effects of glutathione transferase theta polymorphism on the risk estimates of dichlo-
romethane to humans. Toxicol. Appl. Pharmacol. 158:221-230.
Faustman, E. M., Lewandowski, T. A., Ponce, R. A., and Bartell, S. M. 1999 Biologically based dose-response models for developmental
toxicants: Lessons from methylmercury. Inhal. Toxicol. 11:559-572.
Fennell, T. R., Sumner, S. C., and Walker, V. E. 1 992. A model for the formation and removal of hemoglobin adducts. Cancer Epidemiol.
Biomarkers Prev. 1:213-219.
Fisher, J. W., and Allen, B. C. 1993. Evaluating the risk of liver cancer in human exposed to trichloroethylene using physiological models.
Risk Anal. 13:87-95.
Frederick, C. B., Potter, D. W., Chang-Mateu, M. I., and Andersen, M. E. 1992. A physiologically-based pharmacokinetic and pharmacody-
namic model to describe the oral dosing of rats with ethyl acrylate and its implications for risk assessment. Toxicol. Appl. Pharmacol.
114:246-260.
Previous I TOC
-------
544 C. M. THOMPSON ET AL
Gearhart, J. M., Jepson, G. W., Clewell, H. J. Ill, Andersen, M. E., and Conolly, R. B. 1 990. Physiologically based pharmacokinetic and
pharmacodynamic model for the inhibition of acetylcholinesterase by diisopropylfluorophosphate. Toxicol. Appl. Pharmacol.
106:295-310.
Gearhart, J. M., Jepson, G. W., Clewell, H. J., Andersen, M. E., and Conolly, R. B. 1994. Physiologically based pharmacokinetic model
for the inhibition of acetylcholinesterase by organophosphate esters. Environ. Health Perspect. 102(Suppl. 11):51-60.
Gearhart, J. M., Clewell, H. J., and Crump, K. S. 1 995. Pharmacokinetic dose estimates of mercury in children and dose-response curves
of performance tests in a large epidemiological study. Water Air Soil Pollut. 80:49-58.
Gentry, P. R., Covington, T. R., Andersen, M. E., and Clewell, H. J. 2002 Application of a physiologically based pharmacokinetic model
for isopropanol in the derivation of a reference dose and reference concentration. Regul. Toxicol. Pharmacol. 36:51-68.
Gentry, P. R., Covington, T. R., and Clewell, H. J. III. 2003. Evaluation of the potential impact of pharmacokinetic differences on tissue
dosimetry in offspring during pregnancy and lactation. Regul. Toxicol. Pharmacol. 38:1-16.
Gentry, P. R., Haber, L. T., McDonald, T. B., Zhao, Q., Covington, T., Nance, P., Clewell, H. J. Ill, Lipscomb, J. C., and Barton, H. A.
2004. Data for physiologically based pharmacokinetic modeling in neonatal animals: Physiological parameters in mice and
Sprague-Dawley rats.j. Child. Health 2:363-411.
Gerrity, P. G., Henry, C. J., and Birnbaum, L. 1990. Principles of route-route extrapolation for risk assessment. New York: Elsevier.
Ginsberg, G., Hattis, D., Russ, A., and Sonawane, B. 2004. Physiologically based pharmacokinetic (PBPK) modeling of caffeine and theo-
phylline in neonates and adults: Implications for assessing children's risks from environmental agents.). Toxicol. Environ. Health A
67:297-329.
Haddad, S., Beliveau, M., Tardif, R., and Krishnan, K., 2001 a. A PBPK modeling-based approach to account for interactions in the health
risk assessment of chemical mixtures. Toxicol. Sci. 63:1 25-131.
Haddad, S., Restieri, C., and Krishnan, K. 2001 b. Characterization of age-related changes in body weight and organ weights from birth to
adolescence in humans. J. Toxicol. Environ. Health A 64:453-464.
Hanna, L. M., Lou, S. R., Su, S., and Jarabek, A. M. 2001. Mass transport analysis: Inhalation RfC methods framework for interspecies
dosimetric adjustment. Inhal. Toxicol. 1 3:437-463.
Himmelstein, K. J., and Lutz, R. J. 1979. A review of the application of physiologically based pharmacokinetic modeling. ]. Pharm.
Biopharm. 7:127-145.
Hurst, C. H., DeVito, M. J., Setzer, R. W., Birnbaum, L. S. 2000. Acute administration of 2,3,7,8-tetrachlorodibenzo-p-dioxin (TCDD) in
pregnant Long Evans rats: Association of measured tissue concentrations with developmental effects. Toxicol. Sci. 53:411-420.
IPCS. 2005. Chemical-specific adjustment factors (CSAEs) for interspecies differences and human variability in dose/concentration-
response assessment: Guidance document for the use of data in dose/concentration-response assessment. Geneva: World Health
Organization.
Jarabek, A. M. 1995a. Interspecies extrapolation based on mechanistic determinants of chemical disposition. Hum. Ecol. Risk Assess.
1:641-662.
Jarabek, A. M. 1995b. The application of dosimetry models to identify key processes and parameters for default dose-response assess-
ment approaches. Toxicol. Lett. 79:171-184.
Jarabek, A. M., Fisher, J. W., Rubenstein, R., Lipscomb, J. C., Williams, R. J., Vinegar, A., and McDougal, J. N. 1994. Mechanistic insights
aid the search for CFC substitutes: Risk assessment of HCFC-123 as an example. Risk Anal. 14:231-250.
Jonsson, F., and Johanson, G. 2001. A Bayesian analysis of the influence of GSTT1 polymorphism on the cancer risk estimate for dichlo-
romethane. Toxicol. Appl. Pharmacol. 1 74:99-11 2.
Kenyon, E. M., Kraichely, R. E., Hudson, K. T., and Medinsky, M. A. 1996. Differences in rates of benzene metabolism correlate with
observed genotoxicity. Toxicol. Appl. Pharmacol. 136:49-56.
Kim, A. H., Kohn, M. C., Portier, C. J., and Walker, N. J. 2002. Impact of physiologically based pharmacokinetic modeling on benchmark
dose calculations for TCDD-induced biochemical responses. Regul. Toxicol. Pharmacol. 36:287-296.
Kimbell, J. S., Gross, E. A., Joyner, D. R., Godo, M. N., and Morgan, K. T. 1993. Application of computational fluid dynamics to regional
dosimetry of inhaled chemicals in the upper respiratory tract of the rat. Toxicol. Appl. Pharmacol. 121:253-263.
Kirman, C. R., Hays, S. M., Kedderis, G. L., Gargas, M. L., and Strother, D. E. 2000. Improving cancer dose-response characterization by
using physiologically based pharmacokinetic modeling: An analysis of pooled data for acrylonitrile-induced brain tumors to assess
cancer potency in the rat. Risk Anal. 20:1 35-151.
Kohn, M. C., and Melnick, R. L. 1999. A physiological model for ligand-induced accumulation of alpha 2u globulin in male rat kidney:
Roles of protein synthesis and lysosomal degradation in the renal dosimetry of 2,4,4-trimethyl-2-pentanol. Toxicology 136:89-105.
Krishnan, K., and Andersen, M. E. 2007. Physiologically based pharmacokinetic modeling in toxicology. In Principles and methods of
toxicology, ed. A. W. Hayes. Philadelphia: Taylor & Francis, pp. 231-292.
Krishnan, K., Andersen, M.E., 2001. Physiologically based pharmacokinetic modeling in toxicology. In Principles and methods of toxicol-
ogy, ed. A. W. Hayes, pp. 193-241. Philadelphia: Taylor & Francis.
Krishnan, K., Gargas, M. L., Fennell, T. R., and Andersen, M. E. 1 992. A physiologically based description of ethylene oxide dosimetry in
the rat. Toxicol. Ind. Health 8:1 21-140.
Kumagai, S., and Matsunaga, I. 1995. Physiologically based pharmacokinetic model for acetone. Occup. Environ. Med. 52:344-352.
Leroux, B.C., Leisenring, W.M., Moolgavkar, S.H., Faustman, E.M., 1996 A biologically-based dose-response model for development
toxicology. Risk Anal. 16:449-458.
Leung, H. W. 1991. Development and utilization of physiologically based pharmacokinetic models for toxicological applications.). Toxicol.
Environ. Health 32:247-267.
Leung, H. W. 1 992. Use of physiologically based pharmacokinetic models to establish biological exposure indexes. Am. Ind. Hyg. Assoc.
7.53:369-374.
Previous I TOC
-------
APPLICATIONS OF PBPK MODELS IN RISK ASSESSMENT 545
Leung, H. W., and Paustenbach, D. J. 1990. Cancer risk assessment for dioxane based upon a physiologically-based pharmacokinetic
modeling approach. Toxicol. Lett. 51:147-1 62.
Levesque, B., Ayotte, P., Tardif, R., Ferron, L., Gingras, S., Schlouch, E., Gingras, G., Levallois, P., and Dewailly, E. 2002. Cancer risk
associated with household exposure to chloroform.). Toxicol. Environ. Health A 65:489-502.
Liao, K. H., Dobrev, I. D., Dennison, J. E., Jr., Andersen, M. E., Reisfeld, B., Reardon, K. F., Campain, J. A., Wei, W., Klein, M. T., Quann,
R. J., and Yang, R. S. 2002. Application of biologically based computer modeling to simple or complex mixtures. Environ. Health
Perspect. 110(Suppl. 6):957-963.
Lipscomb, J. C. 2004. Evaluating the relationship between variance in enzyme expression and toxicant concentration in health risk
assessment. Hum. Ecol. Risk Assess. 10:39-55.
Lipscomb, J. C., Fisher, J. W., Confer, P. D., and Byczkowski, J. Z. 1998. In vitro to in vivo extrapolation for trichloroethylene metabolism
in humans. Toxicol. Appl. Pharmacol. 152:376-387.
Lipscomb, J. C., Teuschler, L. K., Swartout, J., Popken, D., Cox, T., and Kedderis, G. L. 2003. The impact of cytochrome P450 2E1-
dependent metabolic variance on a risk-relevant pharmacokinetic outcome in humans. Risk Anal. 23:1221-1 238.
Loizou, G., Spendiff, M., Barton, H. A., Bessems, J., Bois, F. Y., DTvoire, M. B., Buist, H., Clewell, H. J. 3rd, Meek, B., Gundert-Remy, U.,
Goerlitz, G., and Schmitt, W. 2008. Development of good modelling practice for physiological based pharmacokinetic models for
use in risk assessment: The first steps. Regul Toxicol Pharmacol. 50:400-411.
Luebeck, E. G., Moolgavkar, S. H., Buchmann, A., and Schwartz, M. 1991. Effects of polychlorinated biphenyls in rat liver: quantitative
analysis of enzyme-altered foci. Toxicol. Appl. Pharmacol. 111:469-484.
Luecke, R. H., Wosilait, W. D., and Young, J. F. 1997. Mathematical analysis for teratogenic sensitivity. Teratology 55:373-380.
MacDonald, A. J., Rostami-Hodjegan, A., Tucker, G. T., and Linkens, D. A. 2002. Analysis of solvent central nervous system toxicity and eth-
anol interactions using a human population physiologically based kinetic and dynamic model. Regul. Toxicol. Pharmacol. 35:165-1 76.
Martonen, T. B., Zhang, Z., Yu, G., and Musante, C. J. 2001. Three-dimensional computer modeling of the human upper respiratory
tract. Cell Biochem. Biophys. 35:255-261.
Meek, M. E., Beauchamp, R., Long, G., Moir, D., Turner, L., and Walker, M. 2002 Chloroform: Exposure estimation, hazard character-
ization, and exposure-response analysis.). Toxicol. Environ. Health B 5:283-334.
Melnick, R. L., and Kohn, M. C. 2000. Dose-response analyses of experimental cancer data. Drug Metab. Rev. 32:193-209.
Miller, F. J., Schlosser, P. M., and Janszen, D. B. 2000. Haber's rule: A special case in a family of curves relating concentration and
duration of exposure to a fixed level of response for a given endpoint. Toxicology 149:21-34.
Monro, A. 1994. Drug toxicokinetics: Scope and limitations that arise from species differences in pharmacodynamic and carcinogenic
responses.). Pharmacokinet. Biopharm. 22:41-57.
Moolgavkar, S., and Knudson, A. 1981. Mutation and cancer: A model for human carcinogenesis.) Nat/. Cancer Inst. 66:1037-1052.
Moolgavkar, S. H., and Luebeck, G. 1990. Two-event model for carcinogenesis: biological, mathematical, and statistical considerations.
Risk Anal. 10:323-341.
Moolgavkar, S., and Venzon, D. 1979. Two-event models for carcinogenesis: Incidence curve for childhood and adult tumors. Math.
Biosci. 47:55-77.
Nong, A., McCarver, D. G., Mines, R.N., and Krishnan, K. 2006. Modeling interchild differences in pharmacokinetics on the basis of
subject-specific data on physiology and hepatic CYP2E1 levels: a case study with toluene. Toxicol. Appl. Pharmacol. 214:78-87.
O'Flaherty, E. J. 1981. Toxicants and drugs: Kinetics and dynamics. New York: John Wiley & Sons.
O'Flaherty, E. J. 1994. Physiologic changes during growth and development. Environ. Health Perspect. 102(Suppl. 11): 103-106.
Overton, J. H. 2001. Dosimetry modeling of highly soluble reactive gases in the respiratory tract. Inhal. Toxicol. 13:347-357.
Page, N. P., Singh, D. V., Farland, W., Goodman, J. I., Conolly, R. B., Andersen, M. E., Clewell, H.J., Frederick, C. B., Yamasaki, H., and
Lucier, G. 1997. Implementation of EPA revised cancer assessment guidelines: Incorporation of mechanistic and pharmacokinetic
data. Eundam. Appl. Toxicol. 37:1 6-36.
Pauluhn, J. 2003. Issues of dosimetry in inhalation toxicity. Toxicol. Lett. 140-141:229-238.
Paustenbach, D. J. 2000. The practice of exposure assessment: A state-of-the-art review.). Toxicol. Environ. Health B 3:1 79-291.
Perbellini, L., Mozzo, P., Olivato, D., and Brugnone, F. 1990. "Dynamic" biological exposure indexes for n-hexane and 2,5-hexanedione,
suggested by a physiologically based pharmacokinetic model. Am. Ind. Hyg. Assoc. ]. 51:356-362.
Pierce, C. H., Dills, R. L., Morgan, M. S., Vicini, P., and Kalman, D. A. 1998. Biological monitoring of controlled toluene exposure. Int.
Arch. Occup. Environ. Health 71:433-444.
Poet, T. S., Soelberg, J. J., Weitz, K. K., Mast, T. J., Miller, R. A., Thrall, B. D., and Corley, R. A. 2003. Mode of action and pharmacoki-
netic studies of 2-butoxyethanol in the mouse with an emphasis on forestomach dosimetry. Toxicol. Sci. 71:1 76-1 89.
Poulin, P., and Krishnan, K. 1996a. A mechanistic algorithm for predicting blood:air partition coefficients of organic chemicals with the
consideration of reversible binding in hemoglobin. Toxicol. Appl. Pharmacol. 1 36:1 31-137.
Poulin, P., and Krishnan, K. 1996b. A tissue composition-based algorithm for predicting tissue:air partition coefficients of organic chemi-
cals. Toxicol. Appl. Pharmacol. 136:126-130.
Portier, C., Tritscher, A., Kohn, M., Sewall, C., Clark, G., Edler, L., Hoel, D., and Lucier, G. 1993. Ligand/receptor binding for 2,3,7,8-
TCDD: Implications for risk assessment. Fundam. Appl. Toxicol. 20:48-56.
Portier, C. J., and Kaplan, N. L. 1 989. Variability of safe dose estimates when using complicated models of the carcinogenic process. A
case study: Methylene chloride. Eundam. Appl. Toxicol. 1 3:533-544.
Price, K., Haddad, S., and Krishnan, K. 2003. Physiological modeling of age-specific changes in the pharmacokinetics of organic chemi-
cals in children.). Toxicol. Environ. Health A 66:41 7-433.
Rao, H. V., and Ginsberg, G. L. 1997. A physiologically-based pharmacokinetic model assessment of methyl t-butyl ether in groundwater
for a bathing and showering determination. Risk Anal. 1 7:583-598.
Previous I TOC
-------
546 C. M. THOMPSON ET AL
Reddy, C. B., Yang, R. S. H., Clewell, H. J., and Andersen, M. E., eds. 2005. Physiologically based phannacokinetic modeling. Hoboken,
NJ: Wiley & Sons.
Reitz, R. H., McDougal, J. N., Himmelstein, M. W., Nolan, R. J., and Schumann, A. M. 1988a. Physiologically-based pharmacokinetic
modeling with methyl chloroform: Implications for interspecies, high-low dose and dose-route extrapolations. Toxicol. Appl.
Phannacol. 95:185-199.
Reitz, R. H., Mendrala, A. L, Park, C. N., Andersen, M. E., and Guengerich, F. P. 1988b. Incorporation of in vitro enzyme data into the phys-
iologically-based pharmacokinetic (PB-PK) model for methylene chloride: Implications for risk assessment. Tox/co/. Lett. 43:97-116.
Reitz, R. H., Mendrala, A. L., and Guengerich, F. P. 1989. In vitro metabolism of methylene chloride in human and animal tissues: Use
in physiologically based pharmacokinetic models. Toxicol. Appl. Phannacol. 97:230-246.
Reitz, R. H., Mendrala, A. L., Corley, R. A., Quasi, J. F., Gargas, M. L., Andersen, M. E., Staats, D. A., and Conolly R. B. 1990a. Estimating
the risk of liver cancer associated with human exposures to chloroform using physiologically-based pharmacokinetic modeling.
Toxicol. Appl. Pharmacol. 105:443-459.
Reitz, R. H., McCroskey, P. S., Park, C. N., Andersen, M. E., and Gargas, M. L. 1990b. Development of a physiologically-based pharma-
cokinetic model for risk assessment with 1,4-dioxane. Toxicol. Appl. Pharmacol. 105:37-54.
Reitz, R. H., Gargas, M. L., Andersen, M. E., Provan, W. M., Green, T. L. 1996. Predicting cancer risk from vinyl chloride exposure with
a physiologically based pharmacokinetic model. Toxicol. Appl. Pharmacol. 137:253-267.
Renwick, A. G. 2001. Toxicokinetics—Pharmacokinetics in toxicology. In Principles and methods of toxicology, ed. A. W. Hayes,
pp. 137-192. Philadelphia: Taylor & Francis.
Rogers, J. M., Mole, M. L., Chernoff, N., Barbee, B. D., Turner, C. I., Logsdon, T. R., and Kavlock, R. J. 1 993. The developmental toxicity
of inhaled methanol in the CD-1 mouse, with quantitative dose-response modeling for estimation of benchmark doses. Teratology
47:175-188.
Roth, W. L., Freeman, R. A., and Wilson, A. G. 1993. A physiologically based model for gastrointestinal absorption and excretion of
chemicals carried by lipids. Risk Anal. 13:531-543.
Rowland, M. 1985. Physiologic pharmacokinetic models and interanimal species scaling. Pharmacol. Ther. 29:49-68.
Roy, A., and Georgopoulos, P. G. 1 998. Reconstructing week-long exposures to volatile organic compounds using physiologically based
pharmacokinetic models.). Expos. Anal. Environ. Epidemiol. 8:407-422.
Roy, A., Weisel, C. P., Lioy, P. J., and Georgopoulos, P. G. 1996. A distributed parameter physiologically-based pharmacokinetic model
for dermal and inhalation exposure to volatile organic compounds. Risk Anal. 16:147-160.
Santostefano, M. J., Wang, X., Richardson, V. M., Ross, D. G., DeVito, M. J., and Birnbaum, L. S. 1 998. A pharmacodynamic analysis of
TCDD-induced cytochrome P450 gene expression in multiple tissues: Dose- and time-dependent effects. Toxicol. Appl. Pharmacol.
151:294-310.
Sarangapani, R., Gentry, P. R., Covington, T. R., Teeguarden, J. G., and Clewell, H. J. III. 2003. Evaluation of the potential impact of age-
and gender-specific lung morphology and ventilation rate on the dosimetry of vapors. Inhal. Toxicol. 15:987-1016.
Schlosser, P. M., Lilly, P. D., Conolly, R. B., Janszen, D. B., and Kimbell, J. S. 2003. Benchmark dose risk assessment for formaldehyde
using airflow modeling and a single-compartment, DNA-protein cross-link dosimetry model to estimate human equivalent doses.
RiskAnal. 23:473-487.
Simmons, J. E., Evans, M. V., and Boyes, W. K. 2005. Moving from external exposure concentration to internal dose: Duration extrapo-
lation based on physiologically based pharmacokinetic derived estimates of internal dose. J. Toxicol. Environ. Health A 68:927-950.
Slikker, W., Jr, Andersen, M. E., Bogdanffy, M. S., Bus, J. S., Cohen, S. D., Conolly, R. B., David, R. M., Doerrer, N. G., Dorman, D. C.,
Gaylor, D. W., Hattis, D., Rogers, J. M., Setzer, R. W., Swenberg, J. A., and Wallace, K. 2004. Dose-dependent transitions in mech-
anisms of toxicity: Case studies. Toxicol. Appl. Pharmacol. 201:226-294.
Starr, T. B., and Festa, J. L. 2003. A proposed inhalation reference concentration for methanol. Regul. Toxicol. Pharmacol. 38:224-231.
Steinbach, K. H., Raffler, H., Pabst, G., and Fliedner, T. M. 1980. A mathematical model of canine granulocytopoiesis. ]. Math. Biol.
10:1-12.
Sweeney, L. M., Tyler, T. R., Kirman, C. R., Corley, R. A., Reitz, R. H., Paustenbach, D. J., Holson, J. F., Whorton, M. D., Thompson, K.
M., and Gargas, M. L. 2001. Proposed occupational exposure limits for select ethylene glycol ethers using PBPK models and Monte
Carlo simulations. Toxicol. Sci. 62:124-139.
Tan, Y. M., Butterworth, B. E., Gargas, M. L., and Conolly, R. B. 2003. Biologically motivated computational modeling of chloroform
cytolethality and regenerative cellular proliferation. Toxicol. Sci. 75:192-200.
Tan, Y. M., Liao, K. H., Conolly, R. B., Blount, B. C., Mason, A. M., and Clewell, H. J. 2006. Use of a physiologically based pharma-
cokinetic model to identify exposures consistent with human biomonitoring data for chloroform.). Toxicol. Environ. Health A 69:
1727-1756.
Tardif, R., Charest-Tardif, G., Brodeur, J., and Krishnan, K. 1997. Physiologically based pharmacokinetic modeling of a ternary mixture of
alkyl benzenes in rats and humans. Toxicol. Appl. Pharmacol. 144:1 20-134.
Thomas, R. S., Lytle, W. E., Keefe, T. J., Constan, A. A., and Yang, R. S. 1 996. Incorporating Monte Carlo simulation into physiologically
based pharmacokinetic models using advanced continuous simulation language (ACSL): A computational method. Fundam. Appl.
Toxicol. 31:19-28.
Thomas, R. S., Conolly, R. B., Gustafson, D. L., Long, M. E., Benjamin, S. A., and Yang, R. S. 2000. A physiologically based pharmacody-
namic analysis of hepatic foci within a medium-term liver bioassay using pentachlorobenzene as a promoter and diethylnitrosamine
as an initiator. Toxicol. Appl. Pharmacol. 166:128-137.
Timchalk, C., Nolan, R. J., Mendrala, A. L., Dittenber, D. A., Brzak, K. A., and Mattsson, J. L. 2002. A Physiologically based pharmacoki-
netic and pharmacodynamic (PBPK/PD) model for the organophosphate insecticide chlorpyrifos in rats and humans. Toxicol. Sci.
66:34-53.
Previous I TOC
-------
APPLICATIONS OF PBPK MODELS IN RISK ASSESSMENT 547
Timchalk, C, Poet, T. S., Kousba, A. A., Campbell, J. A., and Lin, Y. 2004. Noninvasive biomonitoring approaches to determine dosime-
try and risk following acute chemical exposure: Analysis of lead or organophosphate insecticide in saliva. J. Toxicol. Environ. Health
A 67:635-650.
Timchalk, C., Poet, T. S., Lin, Y., Weitz, K. K., Zhao, R., and Thrall, K. D. 2001. Development of an integrated microanalytical system for
analysis of lead in saliva and linkage to a physiologically based pharmacokinetic model describing lead saliva secretion. Am. \nd.
Hyg. Assoc. J. 62:295-302.
Tran, C. L., Jones, A. D., Cullen, R. T., and Donaldson, K. 1999. Mathematical modeling of the retention and clearance of low-toxicity
particles in the lung. Inhal. Toxicol. 11:1059-1076.
U.S. Environmental Protection Agency. 1994. Methods for derivation of inhalation reference concentrations and application of inhalation
dosimetry. EPA/600/8-90/066F. Washington, DC: Office of Health and Environmental Assessment, http://cfpub.epa.gov/ncea/cfm/
record isplay.cfm?deid = 71 993
U.S. Environmental Protection Agency. 1999a. Extrapolation of the benzene inhalation unit risk estimate to the oral route of exposure (draft).
NCEA-W-0517. Washington, DC: National Center for Environmental Assessment, http://cfpub.epa.gov/ncea/cfm/recordisplay.
cfm?deid = 12140
U.S. Environmental Protection Agency. 1999b. Toxicological review of ethylene glycol monobutyl ether. In support of summary informa-
tion on the IRIS. CAS No. 777-76-2. Washington, DC: National Center for Environmental Assessment, http://www.epa.gov/iris/
toxreviews/0500-tr.pdf
U.S. Environmental Protection Agency. 2000a. Benchmark dose technical guidance document (external review draft). EPA/630/R-00/001.
Washington, DC: Risk Assessment Forum. http://www.epa.gov/ncea/pdfs/bmds/BMD-ExternaM 0_13_2000.pdf
U.S. Environmental Protection Agency. 2000b. Toxicological review of vinyl chloride. In support of summary information on the IRIS. EPA/
635/R-00/004. Washington, DC: National Center for Environmental Assessment. http://www.epa.gOV/iris/toxreviews/1001 -tr.pdf
U.S. Environmental Protection Agency. 2002. A review of the reference dose and reference concentration processes. EPA/630/P-02/002 F.
Washington, DC: Risk Assessment Forum, http://cfpub.epa.gov/ncea/cfm/recordisplay.cfm?deid = 55365
U.S. Environmental Protection Agency. 2004. Air quality criteria for particulate matter. EPA/600/P-99/002aF. Research Triangle Park, NC:
National Center for Environmental Assessment, http://cfpub.epa.gov/ncea/cfm/recordisplay.cfm?deid = 87903
U.S. Environmental Protection Agency. 2005a. Guidelines for carcinogen risk assessment. EPA/630/P-03/001 B. Washington, DC: Risk
Assessment Forum, http://cfpub.epa.gov/ncea/cfm/recordisplay .cfm?deid = 116283&CFID = 128481 5&CFTOKEN = 771 S6427&
jsessionid = 663067650c6a464245b4TR60306630c230
U.S. Environmental Protection Agency. 2005b. Supplemental guidance for assessing susceptibility from early-life exposure to carcinogens.
EPA/630/R-03/003F. Washington, DC: Risk Assessment Forum, http://cfpub.epa.gov/ncea/cfm/recordisplay.cfm?deid = 1 60003
U.S. Environmental Protection Agency. 2005c. Toxicological review of boron and compounds. In support of summary information on the
IRIS. CASRN 7440-42-8. Washington, DC: National Center for Environmental Assessment, http://www.epa.gov/iris/toxreviews/
0410-tr.pdf
U.S. Environmental Protection Agency. 2006a. Approaches for the application of physiologically based pharmacokinetic (PBPK) models
and supporting data in risk assessment. EPA/600/R-05/043F. Washington, DC: National Center for Environmental Assessment, http:/
/cfpub.epa.gov/ncea/cfm/recordisplay.cfm?deid = 157668
U.S. Environmental Protection Agency. 2006b. A framework for assessing health risks of environmental exposures to children. EPA/600/R-
05/093F. Washington, DC: National Center for Environmental Assessment, http://cfpub.epa.gov/ncea/cfm/recordisplay.
cfm?deid = 158363
U.S. Environmental Protection Agency. 2006c. Use of physiologically based pharmacokinetic models to quantify the impact of human age
and interindividual differences in physiology and biochemistry pertinent to risk. EPA/600/R-06/014A. Washington, DC: National
Center for Environmental Assessment, http://cfpub.epa.gov/ncea/cfm/recordisplay.cfm?deid = 157668
Van Asperen, J., Rijcken, W. R. P., and Lammers, J. H. C. M. 2003. Application of physiologically based toxicokinetic modelling to study
the impact of the exposure scenario on the toxicokinetics and the behavioural effects of toluene in rats. Toxicol. Lett. 1 38:51-68.
Vinegar, A., and Jepson, G. W. 1996. Cardiac sensitization thresholds of halon replacement chemicals predicted in humans by physiolog-
ically based pharmacokinetic modeling. Risk Anal. 16:571-579.
Voisin, E. M., Ruthsatz, M., Collins,]. M., and Hoyle, P. C. 1990. Extrapolation of animal toxicity to humans: Interspecies comparisons in
drug development. Regul. Toxicol. Pharmacol. 12:107-116.
Wagner, J. G. 1981. History of pharmacokinetics. Pharmacol. Ther. 1 2:537-562.
Welsch, F., Blumenthal, G. M., and Conolly, R. B. 1995. Physiologically based pharmacokinetic models applicable to organogenesis:
Extrapolation between species and potential use in prenatal toxicity risk assessments. Toxicol. Lett. 82-83:539-547.
Previous I TOC
-------
This article was downloaded by: [US EPA Environmental Protection Agency]
On: 2 September2009
Access details: /Access Details: [subscription number 789514190]
Publisher Taylor & Francis
Informa Ltd Registered in England and Wales Registered Number: 1072954 Registered office: Mortimer House,
37-41 Mortimer Street, London W1T 3JH, UK
JOURNAL of
TOXICOLOGY and
ENVIRONMENTAL
HEALTH J
Journal of Toxicology and Environmental Health, Part A
Publication details, including instructions for authors and subscription information:
http://www. info rmaworld. co m/smpp/title~content=t713667303
Bayesian Calibration of a Physiologically Based
Pharmacokinetic/Pharmacodynamic Model of Carbaryl Cholinesterase
Inhibition
Andy Nong a; Yu-Mei Tan a; Michael E. Krolskib; Jiansuo Wang a; Curt Lunchick c; Rory B. Conolly a; Harvey
J. Clewell III'
a The Hamner Institutes for Health Sciences, Research Triangle Park, North Carolina, USA b Bayer
CropScience, Stilwell, Kansas, USA c Bayer CropScience, Research Triangle Park, North Carolina, USA
Online Publication Date: 01 January 2008
To cite this Article Nong, Andy, Tan, Yu-Mei, Krolski, Michael E., Wang, Jiansuo, Lunchick, Curt, Conolly, Rory B. and Clewell
Harvey J.(2008)'Bayesian Calibration of a Physiologically Based Pharmacokinetic/Pharmacodynamic Model of Carbaryl
Cholinesterase Inhibition', Journal of Toxicology and Environmental Health, Part A,71:20,1363 — 1381
To link to this Article: DOI: 10.1080/15287390802271608
URL: http://dx.doi.org/10.1080/15287390802271608
PLEASE SCROLL DOWN FOR ARTICLE
Full terms and conditions of use: http://www.informaworld.com/terms-and-conditions-of-access.pdf
This article may be used for research, teaching and private study purposes. Any substantial or
systematic reproduction, re-distribution, re-selling, loan or sub-licensing, systematic supply or
distribution in any form to anyone is expressly forbidden.
The publisher does not give any warranty express or implied or make any representation that the contents
will be complete or accurate or up to date. The accuracy of any instructions, formulae and drug doses
should be independently verified with primary sources. The publisher shall not be liable for any loss,
actions, claims, proceedings, demand or costs or damages whatsoever or howsoever caused arising directly
or indirectly in connection with or arising out of the use of this material.
Previous
-------
Journal of Toxicology and Environmental Health, Part A, 71: 1363-1381, 2008
Copyright © Taylor & Francis Group, LLC
ISSN: 1528-7394 print/ 1087-2620 online
DOI: 10.1080/15287390802271608
Taylor & Francis
Bayesian Calibration of a Physiologically Based
Pharmacokinetic/Pharmacodynamic Model of Carbaryl
Cholinesterase Inhibition
Andy Nong1, Yu-Mei Tan1, Michael E. Krolski2, Jiansuo Wang1, Curt Lunchick3,
Rory B. Conolly1, and Harvey J. Clewell III1
1 The Hamner Institutes for Health Sciences, Research Triangle Park, North Carolina, 2Bayer
CropScience, Stihvell, Kansas, and ~:'Bayer CropScience, Research Triangle Park, North Carolina, USA
Carbaryl, an jV-methyl carbamate (NMC), is a common
insecticide that reversibly inhibits neuronal cholinesterase
activity. The objective of this work was to use a hierarchical
Bayesian approach to estimate the parameters in a physiologi-
cally based pharmacokinetic and pharmacodynamic (PBPK/
PD) model from experimental measurements of carbaryl in
rats. A PBPK/PD model was developed to describe the tissue
dosimetry of carbaryl and its metabolites (1-naphthol and
"other hydroxylated metabolites") and subsequently to predict
the carbaryl-induced inhibition of cholinesterase activity, in
particular in the brain and blood. In support of the model
parameterization, kinetic tracer studies were undertaken to
determine total radioactive tissue levels of carbaryl and metab-
olites in rats exposed by oral or intravenous routes at doses
ranging from 0.8 to 9.2 mg/kg body weight. Inhibition of
cholinesterase activity in blood and brain was also measured
from the exposed rats. Markov Chain Monte Carlo (MCMC)
calibration of the rat model parameters was implemented using
prior information from literature for physiological parameter
distributions together with kinetic and inhibition data on car-
baryl. The posterior estimates of the parameters displayed at
most a twofold deviation from the mean. Monte Carlo simula-
tions of the PBPK/PD model with the posterior distribution
estimates predicted a 95% credible interval of tissue doses for
carbaryl and 1-naphthol within the range of observed data.
Similar prediction results were achieved for cholinesterase inhi-
bition by carbaryl. This initial model will be used to determine
the experimental studies that may provide the highest added
value for model refinement. The Bayesian PBPK/PD modeling
Received 4 September 2007; accepted 15 April 2008.
The authors acknowledge Drs. Melvin E. Andersen, R. Woodrow
Setzer, Miyoung Yoon, and Jerry Campbell for their valuable reviews
and comments. This work was supported by the Bayer CropScience
and the American Chemistry Council Long-Range Research Initiative.
Current address for Rory B. Conolly is U.S. EPA, Research
Triangle Park, NC 27011, USA.
Address correspondence to Yu-Mei Tan, The Hamner Institutes
for Health Sciences, 6 Davis Drive, Research Triangle Park, NC
27709, USA. E-mail: ctan@thehamner.org
approach developed here will serve as a prototype for developing
mechanism-based risk models for the other NMCs.
Carbaryl (1-Naphthylenyl methylcarbamate) is a registered
insecticide used in a variety of fruits, vegetables, field crops,
ornamentals, and turf. Carbaryl (CAS no. [63-25-2]) belongs to
the pesticide family of TV-methyl carbamates (NMCs). As with
the other NMCs, carbaryl inhibits neuronal cholinesterase
activity, which is reversible upon discontinuation of exposure
(Sogorb & Vilanova, 2002). Increased liver and kidney
weights were observed in rats exposed chronically with high
oral doses of carbaryl (Carpenter et al., 1961). Presently, the
U.S. Environmental Protection Agency (EPA) has established an
oral reference dose for chronic exposure (RfD) of 0.1 mg/kg-d
for liver and kidney toxicity in rats (U.S. EPA, 1999). No RfD
was established for inhalation or dermal exposure to carbaryl,
although exposure by these routes to carbaryl during gardening
may be significant.
The kinetics of carbaryl biotransformation were studied in
animals (Knaak et al., 1965; Declume & Benard, 1977;
Strother & Wheeler, 1980; Tanaka et al., 1980; Mount et al.,
1981; Knight et al., 1987; McCraken et al., 1993) and humans
(Knaak et al., 1965; May et al., 1992; Ward et al., 1998). The
metabolism of carbaryl by hepatocytes was reported to produce
metabolites from three major pathways: aromatic hydroxyla-
tion, aliphatic hydroxylation, and hydrolysis (Figure 1). Both
aromatic and aliphatic hydroxylation reactions are catalyzed by
cytochromes P-450 (Tang et al., 2002). Esterase hydrolysis of
carbaryl, which also takes place in plasma, produces 1-naphthol
(McCraken et al., 1993). The resulting metabolites are subse-
quently conjugated with sulfate or glucuronic acid and
excreted via urine andfeces (IPCS, 1994).
In addition to the kinetics of carbaryl, the neuronal response
has been associated with the inhibition and binding of acetyl-
cholinesterase (AChE) in red blood cells (RBC) and brain
1363
Previous
TOC
-------
1364
A. NONG ET AL.
CARBARYL
Oxidation,
Epoxide
Synthetase,
Conjugation
OTHER
• COMPOSITE
METABOLITES
hydrolysis
1-NAPHTHOL
1
OSO,H
1-NAPHTHOL SULFATE
FIG. 1. Suggested depiction of the metabolic pathway of carbaryl.
1-Naphthol and 1-naphthol sulfate are described as the major metabolites in
the PBPK carbaryl model. All other composite metabolites are described
within a single kinetic substructure of the model.
(Banerjee et al., 1991; Dawson, 1994; Rao et al., 1994;
Mortensen et al., 1998). Binding of butyrylcholinesterase
(BChE) and binding of carboxylesterase (CaE) in plasma were
also determined in animal and human using in vitro assays.
One in vivo study in rats measured both the pharmacokinetics
of carbaryl and the inhibition of the plasma AChE following
intravenous (iv) administration (Fernandez et al., 1982).
In the present study, tissue concentrations (in total radioactive
residues) of carbaryl and its metabolites such as 1-naphthol,
together with cholinesterase activities, were collected from
male Sprague-Dawley (SD) rats exposed to [14C]carbaryl at
two dose levels following iv or oral exposures. The purpose of
the present study was to develop and calibrate a physiologi-
cally based pharmacokinetic/pharmacodynamic (PBPK/PD)
model with the available tissue-specific time course profiles. In
contrast to traditional model fitting with mean values, the
PBPK/PD model parameters were calibrated with a hierarchi-
cal Bayesian approach (Bernillon & Bois, 2000), permitting
the estimation of the distribution of parameters on the basis of
individual experimental data and prior information. Model
parameters lacking experimental support or previously fitted to
mean data were subjected to Markov Chain Monte Carlo
(MCMC) updating to estimate the posterior distributions of
these parameters. The calibrated model will be used to identify
key experimental data needs for refining the model prior to its
use for human dosimetry extrapolation. Further, as the first
instance of a PBPK model for an NMC, this Bayesian PBPK/PD
modeling approach will serve as a prototype for developing
mechanism-based risk models for other NMC.
MATERIALS AND METHODS
Chemicals
The [naphthyl-l-14C]carbaryl and the [naphthyl-4a,5,6,
7,8,8a-14C]carbaryl were prepared by Bayer CropScience
(Stilwell, KS). The [naphthyl-l-14C]carbaryl had a radiochemi-
cal purity of 100% and a specific activity of 21.33 mCi/mmol.
The [naphthyl-4a,5,6,7,8,8a-14C]carbaryl had a radiochemical
purity of 98.6% and a specific activity of 105.7 mCi/mmol.
Reference standards were obtained from Bayer CropScience
(Frankfurt, Germany). All other solvents and reagent chemi-
cals used were obtained from commercial suppliers without
additional purification.
Animals
Male SD rats were approximately 7 wk of age and weighed
200 g when obtained from Charles River Laboratories Inc.
(Kingston, NY). These rats were acclimated for 7 d prior to
dosing. During the acclimation period, rats were housed in indi-
vidual cages and maintained on a 12-h photocycle at 22 + 2°C
temperature and 70 + 5% relative humidity. The rats were allowed
access to food (Rodent Diet, PMI Nutrition International, Inc., St.
Louis, MO) and municipal tap water ad libitum. Immediately prior
to dosing, the rats were fasted for approximately 10 h.
Pharmacokinetic and Pharmacodynamic Experiments
To determine the pharmacokinetic behavior of carbaryl,
[14C]carbaryl was administered to rats at two dose levels
through oral or intravenous routes (Table 1). Rats were
treated orally at 1.08 mg/kg BW (O-LDE) or 8.45 mg/kg BW
(O-HDE) or intravenously at 0.8 mg/kg BW (IV-LDE) or 9.2
mg/kg BW (IV-HDE). [Naphthyl-l-14C]carbaryl was
administered for all high-dose experiments and [naphthyl-
4a,5,6,7,8,8a-14C] carbaryl was used for low-dose experiments.
Stock aqueous solutions were evaporated under a gentle
stream of dry nitrogen. For oral dosing solutions, the residue
was taken up in 20 ml of an aqueous solution containing 0.5%
(w/v) carboxymethylcellulose and 1% (w/v) Tween 80. For iv
dosing solutions, the residue was taken up in 9.6 ml polyeth-
ylene glycol with an average molecular mass of 200 Da (PEG
200). The resulting solution was sonicated for 2 min and
followed by mixing for 2 min. For high dosing experiments,
an aliquot (80 mg) of nonlabelled carbaryl was also added
into the solutions. Aliquots of the dosing solution at various
time of exposure were analyzed by high-performance liquid
chromatography (HPLC) to verify dosing solution stability.
Each rat chosen for the experiment was either dosed by oral
gavage with 0.5 ml of the dosing solution or with 0.2 ml of
Previous
TOC
-------
PBPK/PD MODELING OF CARBARYL
1365
TABLE 1
Summary of the In Vivo Pharmacokinetic and Pharmacodynamics Studies Conducted in Male
Sprague-Dawley Rats Exposed to Carbaryl
Experiment
Routes
Average dose rate
(mg/kg body weight)
Measured endpoints
O-LDE
O-HDE
IV-LDE
IV-HDE
Oral gavage
Oral gavage
Intravenous
Intravenous
1.08
8.45
0.80
9.20
• TRRa in blood, plasma, RBCfe, brain
• TRRa in blood, plasma, RBCfe, brain, liver, and fat
1 Carbaryl in brain
1 1-Naphthol in plasma and brain
1 Naphthol sulfate in plasmac
17V-Hydroxy carbaryl in braind
1 Cholinesterase activities in brain, plasma, and RBCfe
• TRRa in blood, plasma, RBCfe, brain
• TRRa in blood, plasma, RBCfe, brain, liver, and fat
1 Carbaryl in plasma, brain, liver and fat
1 1-Naphthol in plasma, brain, liver and fat
1 Naphthol sulfate in plasmac
17V-Hydroxy carbaryl in braind
1 Cholinesterase activities in brain, plasma, and RBCfe
Note. Four rats per experiment.
TRR: total radioactive residues.
*RBC: red blood cells.
"Simulated as the eliminated 1-naphthol in the model.
^Included in the "other composite metabolites" in the model.
the dosing solution introduced through a previously
implanted jugular cannula.
Groups of 4 animals were sacrificed at various time points
up to 24 h (oral: 15, 30 min, 1, 2, 4, 6, 12, or 24 h; iv: 5, 10, 20,
30 min, 1, 2, 4, or 8 h) following dose administration to collect
blood and tissues including brain, liver, and fat. At each time
point in each experiment, individual rats were anesthetized
with halothane (Aldrich Chemical Co. Inc., Milwaukee, WI).
Brain tissue and whole blood were collected in the low-dose
experiments, while portions of other tissues, including brain
and blood, in the high-dose experiments were collected and
processed for Cholinesterase experiments. Collected whole
blood was cooled and centrifuged to separate plasma from red
blood cells (RBC).
Total radioactive residue (TRR) levels were determined for
whole blood, plasma, RBC, brain, liver, and fat from individu-
ally rats at predetermined time points for high-dose groups in
each of the dosing routes. Liver and fat samples were not ana-
lyzed for the lower dose groups. Triplicate or quadruplicate tis-
sue or blood samples were used for radioassay. All liquid
samples and oxidized solid samples were radioassayed either
by HPLC with a radioactivity detector (Raytest RAMONA,
Pittsburg, PA) or with a liquid scintillation counter (Beckman,
model LS 6000LL, Irvine, CA). Samples with residues below
the HPLC system's detection limit of quantitation for the
radiodetector (9-15 ppb) were quantified by liquid scintillation
counting (detection limit of 0.48-0.79 ppb).
Metabolites and the parent compound carbaryl in plasma,
brain, liver, and fat were identified and quantified by liquid
chromatography/electron-spray ionization-mass spectrometry
(LC/ESI-MS). Samples from the high-dose exposure were
separated in half: samples for TRR measurements and samples
analyzed for metabolic composition. The samples for meta-
bolic composition were pooled together for a single measure-
ment at each time point. Among the known carbaryl metabolites,
1-naphthol, 1-naphthol sulfate, and 7V-hydroxymethyl carbaryl
were detected in plasma and the tissues were assayed under the
current experimental conditions.
Cholinesterase activities in brain, plasma, and RBC from
high-dose experiment groups were determined spectrophoto-
metrically with acetylcholine as a substrate using a modified
Ellman method designed to minimize the reactivation of
Cholinesterase during measurement (Nostrandt et al., 1993).
The absorbance change resulting from acetylcholine hydroly-
sis was proportional to time and protein concentration under
the assay conditions. Cholinesterase inhibition by carbaryl as
percent depression was calculated with the following
equation:
%ChEdepression= 100([ChE]contol - [ChE]exposed)/[ChE]contol (1)
where [ChE] represents the Cholinesterase concentration in
control or exposed rats.
Previous
TOC
-------
1366
A. NONG ET AL.
Model Structure
The PBPK/PD model for carbaryl was developed based on
the structure of the organophosphate (OP) models for diisopro-
pylfluorophosphate (DFP) and parathion (Gearhart et al., 1990,
1994). Since the experiments mainly consisted of individual
TRR values that involved radioactivity of carbaryl and all its
metabolic derivatives together, the model described the
combine molar mass burden of the carbaryl and its composite
metabolites. Therefore, three individual PBPK models inter-
connected by carbaryl metabolic processes described the phar-
macokinetics carbaryl, 1-naphthol (as major metabolite), and
other composite metabolites derived (Figure 2). A PD model
was then added to simulate the reversible cholinesterase inhibi-
tion in blood and brain based on carbaryl concentrations in
blood (plasma and RBC). The main differences between our
carbaryl model and the OP models include: (1) diffusion lim-
ited tissue compartments for carbaryl; (2) the addition of PBPK
models for carbaryl metabolites; and (3) reversible cholinest-
erase binding. A detailed description of the PBPK/PD carbaryl
model is also found in the appendix.
PBPKModel
Three PBPK submodels describe the time course of absorp-
tion, distribution, metabolism, and elimination of carbaryl, and
distribution and elimination of 1-naphthol and the "other com-
posite metabolites" (Figure 2). Michaelis-Menten metabolism
of carbaryl to 1-naphthol is used to describe the elimination by
esterase of the parent compound in liver and blood. 1-Naphthol
CARBARYL
Q
o
o
m
w
O
Ul
/
I/
BRAIN
FAT
REST OF THE 1
BODY J
LIVER
A|
II
Gl TRACT
t
ORAL
+
Q 0
0 0
3 3
ARTERIAL B
VENOUS B
METABOLISM
1-NAPHTHOL
BRAIN
FAT
REST OF THE
BODY
w
A
LIVER
i
Gl TRACT
!
f \
TERIAL BLOOD
NOUS BLOOD
< >
Manhthnl-
\ * sulfate
\
\
\
OTHER
COMPOSITE
METABOLITES
BRAIN
FAT
REST OF THE
BODY
J
r
LIVER
Gl TRACT
x--*-^
\
\
V
i
c»
Q
O
3
m
Ul
?IN
FECES
.,'""
r^ii °
'Synthesis py" "^fl'
\ Inhibition
JFree cholinesterase Carbamoylated cholinesterase j
Regeneration
^Degradation
V
FIG. 2. Schematic representation of the PBPK/PD model structure for carbaryl. Carbaryl, 1-naphthol, and other composite metabolites are presented, with each
its own PBPK structure where they are coupled by liver and blood metabolism. The metabolites are excreted out into the urine and feces by first-order rates. The
dotted lines for carbaryl tissue compartments represent the diffusion-limited distribution of carbaryl, while the tissue compartments in metabolite models are
flow-limited. The PD model of cholinesterase inhibition by carbaryl is introduced in the brain and blood compartments of the carbaryl PBPK model. Activities of
acetylcholinesterase and butyrylcholinesterase are influenced by the carbaryl-cholinesterase interaction.
Previous
TOC
-------
PBPK/PD MODELING OF CARBARYL
1367
is modeled to be subsequently conjugated to naphthol sulfate in
a first-order formation process. Naphthol sulfate is then elimi-
nated at a first-order rate. The submodel for the "other composite
metabolites" regroups the generic hydroxylated metabolites
(e.g., 4-hydroxycarbaryl, 5-hydroxycarbaryl, 7V-hydroxymeth-
ylcarbaryl). The other metabolites originate from the process-
ing of carbaryl in the liver compartment. In the model, the
CYP450-mediated metabolism of carbaryl to hydroxylated
metabolites in the liver is described as a common Michaelis-
Menten process.
The tissue compartments in the carbaryl model include
liver, fat, brain, and the rest of the body. Tissue distribution is
allowed to be diffusion limited. Originally, flow-limited
descriptions were attempted for carbaryl distribution, but the
failure to capture slow elimination of carbaryl in liver and fat
prompted the change to a diffusion-limited model for the
parent compound. However, the two metabolite models
(1-naphthol and "the others") were assumed to be flow-limited.
In the carbaryl model, each tissue compartment was divided
into blood and tissue subcompartments. While extraction of
carbaryl between the tissue and arterial blood is defined by par-
tition coefficients, the tissues compartments for carbaryl are
unevenly mixed. The transport of carbaryl into tissue was
assumed to be limited by cell membrane permeability rather
than blood flow rate [as described in Eqs. (2) and (3)]. For
example, the rate of change of carbaryl amount in the liver
tissue (Aliver) was expressed as:
dAllver
dt
C
liver
—
V Aliver
V xC1
max naphthol In
^m_naphthol ~*~^liv
)
y
Jer , max others
•er m others
x Doral
xQ.
+ cllver
(2)
where Cliver (|jmol/L) is the carbaryl concentration in liver tissue,
CVliver (|imol/L) is the carbaryl concentration in liver blood, PAL
is the liver permeability area constant, Pliver is the liver tissue to
blood partition coefficient, Vmax (|imol/h) is the maximum meta-
bolic velocity for 1-naphthol or the other composite metabolites,
Km (|jmol/L) is the Michaelis-Menten affinity constant for
1-naphthol or the other composite metabolites, kblle (h"1) is the
rate constant for bile excretion, ka (h"1) is the rate constant for
oral absorption, and Doral (|j,mol/L) is the oral gavage dose.
(Units for parameters are given in Table 2.) The amount of
carbaryl in the liver blood (Avliver) is then calculated as:
(3)
To simulate the in vivo radioactive residue data obtained
from the oral gavage study, the model incorporated a one-
compartment gut description for carbaryl absorption following
oral administration. The oral absorption of carbaryl is
described as a first order process from the gut to the liver.
Intravenous delivery of carbaryl is assumed as a bolus dose
directly in the venous blood.
PD Model
A description of the pharmacodynamic interaction of cho-
linesterase (AChE and BChE) with carbaryl was coupled to the
PBPK model (Figure 2). Tissue-specific bimolecular inhibition
of cholinesterase activity (Gearhart et al., 1990) was described
based on the levels of carbaryl in the brain, plasma, and RBC.
The tissue-specific cholinesterase activities were dependent
upon (1) the synthesis and degradation of the free esterase as
well as (2) the carbamolyation and decarbamolyation of cho-
linesterase. For example, the differential equations describing
the amount of acetylcholinesterase in the brain are:
7
^BrAChEl
dt
(4)
^'Brain ^ ™ AChE
_ i •
~ K'
AChE
where ABrAchE, and ABrAchEI are the amount of free and bound
(inhibited) brain acetylcholinesterase, KsAChE (|imol/h) is the
rate of esterase synthesis, kd (IT1) is the rate constant of
esterase degradation, ki (\\M~l x h"1) and kr (IT1) are the rate
constants of carbamolyation and decarbamolyation of cho-
linesterase, and CBrain (|imol/L) is the concentration of carbaryl
in the brain. The rate of synthesis of AChE was calculated as:
_[ ™Tissue,AChE X ^Tissue
—
tr
•'AChE
(6)
AChE
where BTissueAChE is the basal acetylcholinesterase level (in the
brain or blood), VTissue is the volume of the tissue (brain or
blood), and trAChE is the basal acetylcholinesterase turnover
rate. Similar calculations were also made with BChE. In addi-
tion, the rate of synthesis of AChE in red blood cells in the
whole body (KsAcheirbc) was calculated as:
B,
'brain.AChE
* blood '
v Hood
tr
AChE
(7)
where Ca (|imol/L) is the concentration of carbaryl in the
arterial blood, and Qliv (L/h) is the liver blood flow.
where rsblood and rsbrain are the fraction of AChE in red blood cells
in blood and the fraction of AChE in red blood cells in brain.
Previous
TOC
-------
1368
A. NONG ET AL.
TABLE 2
Physiological, Biochemical, and Pharmacodynamic Priors for the Carbaryl PBPK/PD Model
Parameters
Mean (GM) Variability (GSD) Uncertainty
Source
Physiological
Body weight (kg)
Fractional tissue volumes"
Brain
Liver
Fat
Blood
Rest of the body6
Hematocrit (HCT)
Cardiac output" (L/h/kg)
Fractional tissue blood flows
Brain
Liver
Fat
Rest of the body6
Pharmacodynamic
Acetylcholinesterase
Basal enzyme activity
Brain6 (|imol/h/kg tissue)
Plasma6 (|imol/h/kg tissue)
Turnover rate6 (IT1)
RBC enzyme activity
Fraction in brain6 (U/kg)
Fraction in blood (U/kg)
Cholinesterase rate constants
Inhibition (jjAT1 x h"1)
Degradation6 (h'1)
Regeneration (h"1)
Butyrylcholinesterase
Enzyme activity
Brain6 (nmol/h/kg)
Plasma6 Oimol/h/kg)
Turnover rate6 (h'1)
Cholinesterase rate constants
Inhibition QjAT1 x h'1)
Degradation6 (h'1)
Regeneration (h"1)
Chemical specific
Carbaryl
Partition coefficients'7
0.255
1.5
0.006
0.034
0.07
0.074
0.716
0.45
14.1
0.051
0.183
0.07
0.698
1.05
1.05
1.05
1.05
—
1.3
1.5
1.05
1.05
1.05
—
1.3
1.3
1.3
1.3
—
1.3
1.3
1.3
1.3
1.3
—
440000 —
13200 —
1.17X107 —
11848 —
1398 1.5
0.4 1.5
0.03 —
1.8 1.5
—
—
—
1.3
1.3
—
1.3
46800
15600
3.66 xlO6
0.04
0.03
0.1
—
—
—
1.5
—
1.5
—
—
—
1.3
—
1.3
Measured
Brown etal., 1997
Brown etal., 1997
Brown etal., 1997
Brown etal., 1997
Difference
Suckow et al., 2005
Arms and Travis, 1988
Brown etal., 1997
Brown etal., 1997
Brown etal., 1997
Difference
Maxwell etal., 1987
Maxwell etal., 1987
Wang and Murphy., 1982
Timchalk et al., 2002
Timchalk et al., 2002
Fitted (originally Gearhart et al., 1990)
Gearhart et al., 1990
Fitted (originally Gearhart et al., 1990)
Maxwell etal., 1987
Maxwell etal., 1987
Main etal., 1972
Fitted (originally Gearhart et al., 1990)
Gearhart et al., 1990
Fitted (originally Gearhart et al., 1990)
Brain
Liver
Fat
Rest of the body
1.4
8
6.5
1.34
1.5
1.5
1.5
1.5
1.3
1.3
1.3
1.3
Calculated
Calculated
Calculated
Calculated
(Continued)
Previous
TOC
-------
PBPK/PD MODELING OF CARBARYL
1369
TABLE 2
(Conitinued)
Parameters
Mean (GM) Variability (GSD) Uncertainty
Source
Tissue permeability -area constants"
Brain
Liver
Fat
Rest of the body
Gastro-intestinal rate constants
Oral absorption (IT1)
Bile excretion (IT1)
(L/h/kg)
0.052
0.55
2.5
3.98
3.6
0.01
.5
.5
.5
.5
.5
.5
1.3
1.3
1.3
1.3
1.3
1.3
Fitted
Fitted
Fitted
Fitted
Fitted
Fitted
1-Naphthol
Partition coefficients'7
Brain
Liver
Fat
Rest of the body
Elimination rate constants
0.36
1.5
0.37
0.003
1.5
1.5
1.5
1.5
Calculated
Calculated
Calculated
Calculated
Liver VmaxCa (|jmol/h/kg)
Liver Km, naphthol (|imol/kg)
Blood VmaxCa (nmol/h/kg)
Blood Km,blood naphthol (|imol/L)
Bile excretion (h :)
Blood first-order elimination (h"1)
Naphthol sulfate formation (h"1)
Naphthol sulfate elimination (h"1)
Other composite metabolites
Partition coefficients'7
Brain
Liver
Fat
Rest of the body
Elimination rate constants
Liver VmaxCa Otmol/h/kg)
Liver Km, others (|j,mol/L)
Bile excretion (h"1)
Blood 1st order elimination (h"1)
13
20
27
27
15
0.001
20
5
1.1
1.32
25
1.6
20
100
0.003
0.001
.5
.5
.5
.5
.5
.5
.5
.5
.5
.5
.5
.5
.5
.5
.5
.5
1.5
1.5
1.5
1.5
1.5
1.5
1.5
1.5
1.5
1.5
1.5
1.5
1.5
1.5
1.5
1.5
Fitted (originally McCracken et al.,
Fitted (originally McCracken et al.,
Fitted (originally McCracken et al.,
Fitted (originally McCracken et al.,
Fitted
Fitted
Fitted
Fitted
Fitted
Fitted
Fitted
Fitted
Fitted (originally Tang et al., 2002)
Fitted (originally Tang et al., 2002)
Fitted
Fitted
1993)
1993)
1993)
1993)
Note. Parameters that are indicated as "fitted" were either estimated directly from in vivo pharmacokinetic data measured in the present study
or adjusted from previously reported values by fitting the model to the data. Distributions are all lognormal in shape and described using geo-
metric mean (GM), geometric standard deviations (GSD) of variability and uncertainty. Sources refers to mean values.
"Cardiac output (Qc) is scaled to BW° 75 (e.g., Qc = Qcc x BW° 75) and tissue volumes are scaled to BW. Vmax and permeability-area constants
are scaled to BW075.
*Not subject to the MCMC analysis.
"Calculated from Poulin and Krishnan (1996).
Model Parameters and Prior Distributions
Prior distribution parameter values of the PBPK/PD models
were obtained from the literature, experimentally calculated or
estimated by fitting with the mean values of the PK/PD experi-
ments (Table 2). The values of the physiological parameters
were obtained from the literature (Arms & Travis, 1988;
Brown et al., 1997) or were measured data undertaken from the
present study (e.g., body weight). Partition coefficients for
carbaryl and 1-naphthol were estimated using a published
quantitative structure-property relationship (QSPR) algorithm
(Poulin & Krishnan, 1996). Cholinesterase-related parameters
were either obtained from studies of organophosphate
Previous
TOC
-------
1370
A. NONG ET AL.
pesticides (Gearhart et al, 1990; Timchalk et al., 2002) or esti-
mated using the cholinesterase inhibition data obtained from
the current study.
The remaining parameters unavailable from literature or
experimental source (e.g., partition coefficients of the metabo-
lites, carbaryl-cholinesterase inhibition rate constants, oral
absorption constants, and metabolic rate constants) were
estimated by fitting to the values of the composite tissue
measurements (carbaryl, 1-naphthol, 1-naphthol sulfate, and
jV-hydroxymethyl carbaryl) from the IV-HDE and O-HDE
experiments. Optimization of the prior estimates of the PBPK7
PD model with the mean data was undertaken in ACSL (Aegis,
Huntsville, AL).
For some of the parameters, means of the prior distributions
were set to values previously estimated by fitting mean values
of some of the data used in the MCMC analysis. Technically,
this practice violates the Bayesian division between prior infor-
mation and data used to refine the priors. However, from a
pragmatic viewpoint this approach, which has been used previ-
ously (Bois 2000), was necessary to obtain convergence.
For the MCMC analysis, while some distributions of inter-
individual difference were available in the literature (Table 2),
most of the distributions were unknown. The standard devia-
tions for the parameters with unknown distributions were
arbitrary given values as described from previous Bayesian
PBPK modeling studies (Bois et al., 1996; Jonsson et al.,
2001). For the carbaryl specific distributions, a geometric stan-
dard deviation (GSD) of 1.5 was assigned, yielding an
expected ratio of standard deviation to mean equal to 50%. For
all parameters, the uncertainty of the parameter distribution
was given a GSD of 1.3 or 1.5 (Table 2). Such a prior distribu-
tion reflected a "reasonably vague specification" as suggested
by Carlin and Louis (2000). The assignment of these uncertain-
ties was necessarily subjective, given the limited availability of
literature data. Each prior was also bounded by an upper and
lower value of threefold the GSD, which is not shown in Table
2. The distributions of the parameters were assumed to be log-
normal in shape with a normal error distribution.
Parameter Characterization of Credible Interval
A hierarchical approach was used to characterize the upper
and lower bound (variability) of parameters and their level of
confidence (uncertainty) in the PBPK/PD model with the
MCMC technique (Bernillon & Bois, 2000; Hack, 2006).
Physiological parameters known to possess high variability
and model parameters with estimated prior mean values were
re-estimated with the individual data sets in the MCMC simu-
lations (Table 2). Other parameters, such as the degradation
rate of free cholinesterase, were held constant from the MCMC
analysis to keep the integrity of these values during the simula-
tion and to generate posteriors related to these literature values.
The MCMC simulations were performed with MCSim soft-
ware (version 5.0.0; http://toxi.ineris.fr ). A different random
seed value to initiate the MCMC simulation was set for each
repeated simulation. The first 1000 iterations of the MCMC
process served as burn-in to generate stable initial values. The
simulation was then continued until the model parameters were
able to converge into stable estimates of their statistical distri-
bution. Specifically, half of the iterations from three different
random MCMC simulations (also known as chains) were com-
pared by visual inspection as well as by the Brooks, German,
and Rubin convergence test (Smith, 2005). Statistical analysis
on these posterior estimates was performed with the Bayesian
Output Analysis Program (BOA; Smith, 2005) package for R
and S-Plus. These posterior distributions were then applied in a
Monte Carlo simulation to characterize the credible interval of
model outputs, including both pharmacokinetics (i.e., tissue
dosimetry) and pharmacodynamics (i.e., cholinesterase
activities).
In addition, a visual inspection of the carbaryl PBPK/PD
model-simulated blood concentration and plasma AChE levels
was compared to experimental data observed in Fernandez
etal. (1982) (Figure 8). The comparison of the kinetic time
profiles demonstrates the fit between the mean posterior values
obtained from the MCMC analysis of the PBPK/PD model
parameters with a different data set.
RESULTS
Pharmacokinetic Studies—Radioactive Residue
Measurements
Uptake, metabolic degradation, and clearance of [14C] carbaryl
by rats were rapid and complete for both oral and iv routes of
administration (examples in Table 3). Peak TRR levels in all
tissues analyzed following a single oral dose at 1.05 mg/kg
body weight (BW) were reached within 15 min; peak levels
TABLE 3a
Average Total Radioactive Residue (TRR) Levels (ppm) in
Brain, Red Blood Cells (RBC), Liver, and Fat From Rats
Exposed to Carbaryl Orally
O-LDE
(1.08 mg/kg)
Time (h)
0.25
0.5
1
2
4
6
12
24
Brain
0.13
0.06
0.03
0.03
0.02
0.01
0.01
0.003
RBC
0.44
0.32
0.18
0.1
0.11
0.07
0.02
0.01
Brain
1.97
1.15
0.62
0.21
0.11
0.1
0.06
0.01
O-HDE
(8.45 mg/kg)
RBC
2.56
2.59
2.24
0.61
0.36
0.41
0.15
0.04
Liver
20.9
13.5
8.1
2.6
1.6
1.9
0.7
0.1
Fat
3.6
3.4
5.3
1.3
0.3
0.2
0.1
0.0
Note. Oral low and high dose experiment (O-LDE; O-HDE).
Previous
TOC
-------
PBPK/PD MODELING OF CARBARYL
1371
TABLE 3b
Average Total Radioactive Residue (TRR) Levels (ppm) in
Brain, Red Blood Cells (RBC), Liver, and Fat From Rats
Exposed to Carbaryl Intravenously
IV-LDE
(0.8 mg/kg)
Time (h)
0.08
0.17
0.33
0.5
1
2
4
8
Brain
0.74
0.41
0.23
0.14
0.06
0.03
0.01
0.01
RBC
1.06
0.91
0.67
0.49
0.3
0.15
0.1
0.05
Brain
13.2
10.7
7.9
6.7
2.4
1.1
0.3
0.2
IV-HDE
(9.2 mg/kg)
RBC
10.2
9.1
7.3
5.5
3.2
2.5
1
0.7
Liver
24.7
27.7
25.5
19.5
13.1
7.2
2.7
1.6
Fat
12.1
15.6
21.2
28.5
16.2
8.5
1.3
0.2
Note. Intravenous low and high dose experiment (IV-LDE;
IV-HDE).
following a single oral dose at 8.45 mg/kg BW were attained
within 30 min. Peak TRR levels in whole blood, RBC, and
brain following an iv dose at either 0.8 mg/kg BW or 9.2 mg/kg
BW were found at 5 min; peak levels in liver were reached at
10 min and in fat at 30 min.
Figure 3 illustrates the time-dependent concentration pro-
files of carbaryl, 1-naphthol, and 1-naphthol sulfate in blood
and brain from high-dose iv experiments. A single measure-
ment was sampled for every time point, since the compounds
were pooled together. Concentration profiles for other tissues,
other metabolites, or oral dosing experiments were not
included in this figure, but are described later in this article.
Carbaryl and metabolites were only detected in the high-dose
iv and oral experiments. Carbaryl and 1-naphthol were
detected in plasma, brain, liver, and fat following an iv dose at
9.2 mg/kg BW. Peak carbaryl levels in plasma, brain, and liver
were reached at 5 min; peak level in fat was attained at 30 min.
Peak 1-naphthol levels in plasma, brain, and liver were reached
later than carbaryl, but peak 1-naphthol level in fat was
reached earlier than for carbaryl. 1-Naphthol sulfate and
(A)
4 6
Time (hours)
(C)
Time (hours)
= 4-
Time (hours)
FIG. 3. Comparison of the PBPK model-simulated (A) carbaryl, (B) 1-naphthol, and (C) naphthol sulfate radioactive residue levels against average
experimental data. Blood levels (solid lines: predictions; D: data) and brain levels (dotted lines: predictions; A: data) were obtained from rats exposed to a single
iv dose of 9.2 mg/kg body weight (BW).
Previous
TOC
-------
1372
A. NONG ET AL.
7V-hydroxycarbaryl were detected in plasma following an oral
dose of 8.45 mg/kg BW or an iv dose at 9.2 mg/kg BW, but
these metabolites were not found in other tissues.
Pharmacodynamic Studies—Cholinesterase Inhibition
Cholinesterase inhibition by carbaryl was measured for
each of the two routes in only the high-dose exposures. The
depression of Cholinesterase activities in the brain and blood
reduced as the carbaryl concentrations decreased with time
(Table 4). The half-life of the inhibition effect was less than 2
h following a single oral or a single iv exposure (i.e., cho-
linesterase activity fell by half after 2 h following exposure).
Model Parameters Characterization
For all of the physiological parameters, the carbaryl-specific
kinetic constants (including tissue permeability and metabolic
TABLE 4a
Average and Standard Deviations of Cholinesterase Activity
Inhibition Levels" in Brain and Red Blood Cells (RBC) From
Rats Exposed to Oral High Doses (HDE) of Carbaryl
Brain
Time (h)
0.25
0.5
1
2
4
6
12
24
Average
57
52
45
27
10
13
2
1
SD
8
7
6
10
6
6
10
8
RBC
Average
59
41
76
29
50
20
3
<0.01
SD
9
31
26
25
21
37
46
50
"Expressed as percent Cholinesterase depression.
TABLE 4b
Average and Standard Deviations of Cholinesterase Activity
Inhibition Levels" in Brain and Red Blood Cells (RBC) From
Rats Exposed to IV High Doses (HDE) of Carbaryl
Brain
Time (h)
0.083
0.167
0.333
0.5
1
2
4
8
Average
83.2
79.6
79.7
75.8
64.4
44.7
17.8
1.9
SD
2.4
2.7
2.4
2.8
5.5
9.0
9.7
7.7
RBC
Average
78
85
74
71
65
45
38
29
SD
20
8
12
8
28
25
7
7
"Expressed as percent Cholinesterase depression.
constants), and various pharmacodynamic parameters, conver-
gence of the Markov Chains was obtained after 25,000 itera-
tions based on the Brooks, Gelman, and Rubin convergence
test. The resulting posterior distributions had at most a geomet-
ric standard deviation (GSD) of 2 (Table 5) or a coefficient of
variation of 100%. Similar significant posterior estimates were
also obtained with a larger prior GSD for the variability of the
parameters (results not shown): A geometric standard devia-
tion of 2 was used for the metabolite-specific parameters given
the limited available literature data.
PBPK/PD Model Simulation
The marginal posterior distributions from the MCMC anal-
ysis were used in the PBPK/PD model to simulate the tissue
dosimetry and Cholinesterase inhibition. The correlation of the
model posterior distributions was not considered in these sim-
ulations. Time-course predictions of the carbaryl, 1-naphthol,
and naphthol sulfate levels in rat blood and brain tissue were
consistent within less than twofold difference with the
measured data (Figure 3). With only the geometric mean pos-
terior estimates of the parameter, the individual measurements
of carbaryl and its components were compared to model simu-
lations. The predicted tissue dosimetry was then compared to
the total radioactive residue population data for all dose levels
and routes of exposure (Figure 4). Values close to the identity
line of the plot represent consistency between predictions of
TRR and the observed measurements. Over 80% of the
predicted tissue concentrations are consistent within twofold
of the experimental data. Although prediction of the tissue
concentrations was more consistent, the model predictions of
Cholinesterase inhibition exhibited considerable variability as
the distribution of residuals is spread out away from the iden-
tity line (Figure 5). In this case, over 60% of the model predic-
tions are consistent within 0.5-fold of the Cholinesterase
inhibition data.
With the mean and standard deviations of the marginal pos-
terior distributions for the parameters from the MCMC analy-
sis, a Monte Carlo simulation was conducted to generate
distributions of model outputs. The observed TRR values in
our study are within the 95% credible interval of the Monte
Carlo prediction with the PBPK/PD model (example for blood
and brain in Figure 6). Similar results were also obtained when
comparing the model predictions with the data from low dose
exposures (results not shown). The concentration of TRR in
brain for oral dose experiment was overpredicted by the model.
In general, the Cholinesterase depression by carbaryl measured
in the experiments was also within the 95% credible interval of
the simulated values (Figure 7). As observed with the pharma-
cokinetic data, the prediction of Cholinesterase depression from
the model was more consistent with the iv dosing data than
with the oral dosing data. The MCMC analysis was unable to
determine an oral absorption rate that was consistent with both
the kinetic tissue concentrations and the Cholinesterase
Previous
TOC
-------
PBPK/PD MODELING OF CARBARYL
1373
TABLE 5
Posterior Distribution Estimates for the Carbaryl PBPK/PD Mode
Parameters
Physiological
Fractional tissue volumes (kg"1)
Brain
Liver
Fat
Blood
Hematocrit (HCT)
Cardiac output" (L/h/kg)
Fractional tissue blood flows
Brain
Liver
Fat
Pharmacodynamic
Acetylcholinesterase
RBC enzyme activity in blood (U/kg)
Cholinesterase rate constants
Inhibition (jjAT1 x h"1)
Regeneration (h"1)
Butyrylcholinesterase
Cholinesterase rate constants
Inhibition QjAT1 x h'1)
Regeneration (h"1)
Chemical specific
Carbaryl
Partition coefficients
Brain
Liver
Fata
Rest of the body
Tissue permeability -area constants (L/h/kg)
Brain
Liver
Fat
Rest of the body
Gastrointestinal rate constants
Oral absorption (h"1)
Bile excretion (IT1)
1-Naphthol
Partition coefficients
Brain
Liver
Fata
Rest of the body
Mean (GM)
0.0069
0.033
0.080
0.065
0.45
16.58
0.051
0.196
0.067
824
0.69
6.0
0.043
0.20
1.04
2.53
5.44
0.93
0.13
1.82
0.83
6.29
1.09
0.011
0.33
2.45
0.30
0.0033
Variability (GSD)
1.3
1.3
1.4
1.3
1.4
1.4
1.3
1.3
1.3
1.6
1.8
2.0
1.4
1.8
1.6
2.1
1.5
1.7
1.9
2.1
2.0
1.7
1.7
1.6
1.5
1.8
1.6
1.6
Uncertainty
.2
.2
.2
.2
.2
.2
.2
.2
.2
1.2
1.2
1.2
1.3
1.2
1.2
1.2
1.2
1.2
1.2
1.2
1.2
1.2
1.2
1.3
1.3
1.2
1.3
1.3
(Continued)
Previous
TOC
-------
1374
A. NONG ET AL.
TABLE 5
(Continued)
Parameters
Mean (GM)
Variability (GSD)
Uncertainty
Elimination rate constants
Liver VmaxC (^mol/h/kg)
Liver Km, naphthol (jimol/L)
Blood VmaxC Oimol/h/kg)
22.4
59.2
21.2
1.7
2.0
1.6
1.2
1.2
1.3
Blood Km,blood (|jmol/L)
Bile excretion (IT1)
First-order blood elimination (IT1)
Naphthol sulfate formation (IT1)
Naphthol sulfate elimination (IT1)
Other composite metabolites
Partition coefficients
Brain
Liver
Fat
Rest of the body
Elimination rate constants
Liver VmaxC (^mol/h/kg)
Liver Km,others (|jmol/L)
Bile excretion (IT1)
1st order blood elimination (IT1)
78.0
12.6
0.0011
15.0
6.2
0.46
7.96
0.57
5.16
22.13
215
0.0032
0.0011
2.0
1.6
1.6
1.6
1.6
2.0
2.0
1.9
2.1
1.5
1.8
1.6
1.6
1.2
1.3
1.3
1.3
1.3
1.2
1.2
1.2
1.2
1.2
1.2
1.2
1.3
Note. Distributions are described using geometric mean (GM) and geometric standard deviations (GSD).
Uncertainties are also presented as geometric standard deviations. Cardiac output (Qc) is scaled to BW° 75 (e.g.,
Qc = Qcc x BW° 75) and tissue volumes are scaled BW. Vmax and permeability-area constants are scaled to BW° 7;
10 15 20 25
Observed 14C level (mg/kg)
10 15 20 25
Observed 14C level (mg/kg)
30
FIG. 4. Comparison of predicted and observed radioactive residue levels from rats exposed to carbaryl at various doses and routes (D: iv, •: oral). The identity
line marks similarity between prediction and experimental results.
Previous
TOC
-------
PBPK/PD MODELING OF CARBARYL
1375
100
80 -
•o
w
^
o
60 -
3 O O O
40 -
20 -
oo o
o o
o o o o
100
20 40 60
Observed % ChE depression
100
80 -
60
40-
20 -
20 40 60 ;
Observed % ChE depression
100
FIG. 5. Comparison of predicted and observed cholinesterase (ChE) activity depression from rats exposed to carbaryl at various doses and routes
(O: iv, •:oral). The identity line marks similarity between prediction and experimental results.
inhibition by carbaryl. This difficulty is reflected in the uncer-
tainty of the posterior estimates.
The model performance was also evaluated by fitting to pre-
viously reported carbaryl concentration in blood and AChE
inhibition in plasma (Fernandez et al., 1982) (Figure 8). The
model was able to predict both the carbaryl concentration and
the extent of AChE inhibition over time, providing some sup-
port for the validity of the calibrated model.
DISCUSSION
The pharmacokinetics of carbaryl was previously
described with either a noncompartmental model or an empir-
ical model with limited virtual compartments (Houston et al.
1974; Cambon et al. 1980, 1981; Lechner & Abdel-Rahman
1986). In the current study, a PBPK/PD model for carbaryl
was developed to describe the mechanisms that link exposure
with target tissue dose and cholinesterase inhibition. The
model incorporated key pharmacokinetic characteristics of
carbaryl, including rapid uptake, rapid metabolic degradation,
and rapid elimination for all routes of administration, and rea-
sonably simulated the concentration of the parent compound
and its metabolites in several tissues from two different
dosing methods, i.e., iv and oral, and the resulting cholinest-
erase inhibition. Experience gained in developing this model
for carbaryl is expected to provide a basis for the more
efficient development of models for other NMCs, similar to
the seminal impact of the PBPK model for DFP (Gearhart et
al., 1990) on the development of models of other OPs. Subse-
quent to the DFP model, PBPK models were developed for a
number of other OPs, including parathion (Gearhart et al.,
1994), chlorpyrifos (Timchalk et al., 2002), and diazinon
(Poet et al., 2004); all of these models made use of data and
approaches from the work on DFP.
The refinement of the PBPK/PD model parameterization,
including variability and uncertainty, in the present study was
performed with an MCMC technique on a large set of individ-
ual tissue concentrations, radioactive residue levels, and cho-
linesterase inhibition data from male SD rats. While the prior
estimates of some parameters were obtained by fitting with
mean rat experimental values (e.g., mean [14C]carbaryl brain
tissue concentration of all the rats together in the IV-HDE), the
MCMC analysis made use of the individual measurements
from the experiments (e.g., TRR brain tissue concentration of
rat number 1, rat number 2, rat number 3, and rat number 4
from the IV-HDE). The present Bayesian estimation of poste-
rior parameter distributions is more robust by allowing for the
physiological and experimental variability of the kinetics and
the inhibition response to carbaryl. In addition, because of the
complex nature of the population data on total radioactive resi-
dues, the model description involved the estimation of the
combined kinetics and responses of the parent compound as
well as the kinetics of the metabolites. Such large and compli-
cated calibration of the parameters is more sensibly applied
with a Bayesian technique like MCMC, which is beyond the
capability of conventional modeling with average tissue
concentration time-course data (Hack, 2006).
The present modeling work quantified the parameters in
terms of variability (difference between individual rats) and
uncertainty (level of confidence in estimates) at the popula-
tion level, assuming that all the rats of our study are part of
the same population. The MCMC analysis with the individual
data reduced uncertainty of the estimates from 0.5 to 0.3 fold.
While some parameter priors were subjective, the posteriors
Previous
TOC
-------
1376
A. NONG ET AL.
(A)
8 12 16
Time (hours)
20
24
345
Time (hours)
7-
6-
D)
^T 4-
g
I 3-1
S»H
o
1 -
12 16
Time (hours)
20
24
2345
Time (hours)
FIG. 6. Monte Carlo simulated distributions of the total radioactive residue level in the brain (A) and in blood (D) using MCMC posterior estimates. Simulated
results are compared with the high-dose exposure (HDE) data for (A) oral (8.45 mg/kg BW) and (B) iv (9.2 mg/kg BW). Dotted line represents median PK
profile, while the upper 95th and lower 5th credible intervals are presented in solid lines.
were calibrated with the individual experimental data, provid-
ing a greater level of confidence in the estimates. In general,
the model predictions using the posteriors were consistent
with the experimental iv data in this study as well as with the
study by Fernandez et al. (1982). Consistency between tissue
concentrations and cholinesterase inhibition for oral adminis-
tration were less well achieved in comparison. These route-
specific differences in response have been observed for other
pesticides such as carbolsulfan and methylparathion (Renzi &
Krieger, 1986; Kramer et al., 2002). Discrepancies in kinetics
and responses from the oral route suggest that the delivery of
carbaryl by the gut is more complicated than the current first-
order absorption. Further analysis of the dose and route
dependencies will be evaluated in a subsequent modeling
effort.
Previous studies quantified cholinesterase inhibition in the
rat brain and blood based on a single route of exposure (450,
800, or 1200 mg/kg orally in Mount et al., 1981; 20 mg/kg iv in
Fernandez et al., 1982). In the current study, cholinesterase
inhibition was investigated from two different routes of expo-
sure to carbaryl, oral and iv. The responsiveness of the model
parameters toward the predicted blood concentration and cho-
linesterase depression varied according to the exposure routes
(data not shown). Following iv exposure, changes of the pre-
dicted tissue concentrations are most likely to occur from
changes to blood flows such as blood flow to the fat or muscle,
while the blood:brain partition has a greater impact than tissue
blood flow on predicted carbaryl level and cholinesterase
activity in the brain. Following oral exposures, the simulated
tissue concentrations are most sensitive to the oral absorption
Previous
TOC
-------
PBPK/PD MODELING OF CARBARYL
1377
100 n
60 4
-------
1378
(A)
100,
D)
.
o
O
10:
1 :
0.1
A. NONG ET AL.
(B)
100n
8 12 16
Time (hours)
20
24
•F 60-
8
ra
-------
PBPK/PD MODELING OF CARBARYL
1379
Banerjee, J., Ghosh, P., Mitra, S., Ghosh, N., and Bhattacharya, S. 1991.
Inhibition of human fetal brain acetylcholinesterase: marker effect of
neurotoxicity. J. Toxicol. Environ. Health 33:283-290.
Bernillon, P., and Bois, F. Y. 2000. Statistical issues in toxicokinetic modeling:
A Bayesian perspective. Environ. HealthPerspect. 108:883-893.
Bois, F. 2000. Statistical analysis of Clewell et al. PBPK model of trichloroet-
hylene kinetics. Environ. Health. Perspect. 108 (Suppl. 2):307-316.
Bois, F. Y., Jackson, E. T., Pekari, K., and Smith, M. T. 1996. Population
toxicokinetics of benzene. Environ. Health. Perspect. 104:1405-1411.
Brown, R. P., Delp, M. D., Lindstedt, S. L., Rhomberg, L. R., and Beliles, R. P.
1997. Physiological parameter values for physiologically based pharmaco-
kinetic models. Toxicol. Ind. Health 13:407-484.
Cambon, C., Fernandez, Y., Falzon, M., and Mitjavila, S. 1981. Variations of
the digestive absorption kinetics of carbaryl with the nature of the vehicle.
Toxicology 22:45-51.
Cambon, C., Declume, C., and Derache, R. 1980. Fetal and maternal rat brain
acetylcholinesterase: Isoenzymes changes following insecticidal carbamate
derivatives poisoning. Arch. Toxicol. 45:257-262.
Carlin, B. P., and Louis, T. A. 2000. Bayes and empirical Bayes methods for
data analysis, 2nd ed. Boca Raton, FL: Chapman & Hall/CRC.
Carpenter, C. P., Weil, C. W., Palm, P. E., Woodside, M. W., Nair, J. H. Ill,
and Smyth, H. F., Jr. 1961. Mammalian toxicity of 1-naphthyl-W-methyl-
carbamate (Sevin insecticide). J. Agric. FoodChem. 9:30-39.
Dawson, R. 1994. Rate constants of carbamylation and decarbamylation of
acetylcholinesterase for physostigmine and carbaryl in the presence of an
oxime. Neurochem. Int. 24:173-182.
Declume, C., and Benard, P. 1977. Fetal accumulation of [14C] carbaryl in rats
and mice. Autoradiographic study. Toxicology 8:95-105.
Fernandez, Y., Falzon, M., Cambon-Gros, C., and Mitjavila, S. 1982. Carbaryl
tricompartmental toxicokinetics and anticholinesterase activity. Toxicol.
Lett. 13:253-258.
Fillmore, C., and Lessenger, J. 1993. A cholinesterase testing program for
pesticide applicators. J. Occup. Med. 35:61-70.
Gearhart, J., Jepson, G, Clewell, H., Andersen, M., and Conolly, R. 1990. A
physiologically-based pharmacokinetic and pharmacodynamic model for
the inhibition of acetylcholinesterase by disopropylfluorophosphate.
Toxicol. Appl. Pharmacol. 106:295-310.
Gearhart, J., Jepson, G., Clewell, H., Andersen, M., and Conolly, R. 1994.
Physiologically-based pharmacokinetic and pharmacodynamic model for
the inhibition of acetylcholinesterase by organophosphate esters. Environ.
Health Perspect. 102:51-60.
Hack, C. E. 2006. Bayesian analysis of physiologically based toxicokinetic and
toxicodynamic models. Toxicology. 221:241-248.
Houston, J. B., Upshall, D. G., and Bridges, J. W. 1975. Pharmacokinetics and
metabolism of two carbamate insecticides, carbaryl and landrin, in the rat.
Xenobiotica 5:637-648.
International Programme on Chemical Safety. 1994. Carbaryl. EHC 153.
Geneva: World Health Organization.
Jonsson, F., Bois, F., and Johanson, G. 2001. Physiologically based pharmaco-
kinetic modeling of inhalation exposure of humans to dichloromethane
during moderate to heavy exercise. Toxicol. Sci. 59:209-218.
Knaak, J. B., Tallant, M. J., Bartley, W. J., and Sullivan, L. J. 1965. The
metabolism of carbaryl in the rat, guinea pig, and man. J. Agric. Food
Chem. 13:537-543.
Knight, E. V., Alvares, A. P., and Chin, B. H. 1987. Effects of phenobarbital
pretreatment on the in vivo metabolism of carbaryl in rats. Bull. Environ.
Contain. Toxicol. 39:815-821.
Lechner, D. W., and Abdel-Rahman, M. S. 1985. Alterations in rat liver
microsomal enzymes following exposure to carbaryl and malathion in
combination. Arch. Environ. Contain. Toxicol. 14:451-457.
Lockridge, O., and Masson, P. 2000. Pesticides and susceptible populations:
People with butyrylcholinesterase genetic variants may be at risk. Neuro-
toxicology. 21:113-126.
Main, A. R., Tarkan, E., Aull, J. L., and Soucie, W. G. 1972. Purification of
horse serum cholinesterase by preparative polyacrylamide gel electro-
phoresis. J. Biol. Chem. 247:566-571.
Maxwell, D. M., Lenz, D. E., Groff, W. A., Kaminskis, A., and Frochlich,
H. L. 1987. The effects of blood flow and detoxification on in vivo
chloinesterase inhibition by soman in rats. Toxicol. Appl. Pharmacol.
88:66-76.
May, D., Naukam, R., Kambam, R., and Branch, R. 1992. Cimetidine-carbaryl
interaction in humans: Evidence for an active metabolite of carbaryl.
J. Pharmacol. Exp. Ther. 262:1057-1061.
McCracken, N. W., Blain, P. G., and Williams, F. M. 1993. Nature and role of
xenobiotic metabolizing esterases in rat liver, lung, skin and blood.
Biochem. Pharmacol. 45:31-36.
McDougal, J. N., Jepson, G. W., Clewell, H. J. 3rd, Gargas, M. L., and Ander-
sen, M. E. 1990. Dermal absorption of organic chemical vapors in rats and
humans. Fundam. Appl. Toxicol. 14:299-308.
Mirfazaelian, A., Kim, K. B., Anand, S. S., Kim, H. J., Tornero-Velez, R.,
Bruckner, J. V., and Fisher, J. W. 2006. Development of a physiologically
based pharmacokinetic model for deltamethrin in the adult male Sprague-
Dawleyrat. Toxicol. Sci. 93:432^142.
Mortensen, S. R., Hooper, M. J., and Padilla, S. 1998. Rat brain acetylcho-
linesterase activity: Developmental profile and maturational sensitivity to
carbamate and organophosphorus inhibitors. Toxicology 125:13-19.
Mount, M. E., Dayton, A. D., and Oehme, F. W. 1981. Carbaryl residues in
tissues and cholinesterase activities in brain and blood of rats receiving
carbaryl. Toxicol. Appl. Pharmacol 58:282-296.
Nostrandt, A. C., Duncan, J. A., and Padilla, S. 1993. A modified spectropho-
tometric method appropriate for measuring cholinesterase activity in tissue
from carbaryl-treated animals. Fundam. Appl. Toxicol. 21:196-203.
Poet, T. S., Kousba, A. A., Dennison, S. L., and Timchalk, C. 2004. Physiolog-
ically based pharmacokinetic/pharmacodynamic model for the organo-
phosphorus pesticide diazinon. Neurotoxicology 25:1013-1030.
Poulin, P., and Krishnan, K. 1996. A mechanistic algorithm for predicting
blood:air partition coefficients of organic chemicals with the consideration of
reversible binding in hemoglobin. Toxicol. Appl. Pharmacol. 136:131-137.
Rao, P. S., Roberts, G. H., Pope, C. N., and Ferguson, P. W. 1994. Compara-
tive inhibition of rodent and human erythrocyte acetylcholinesterase by
carbofuran and carbaryl. Pestic. Biochem. Physiol. 48:79-84.
Renzi, B. E., and Krieger, R. I. 1986. Sublethal acute toxicity of carbosulfan
[2,3-dihydro-2,2-dimethyl-7-benzofuranyl(di-«-butylaminosulfe-
nyl)(methyl)carbamate] in the rat after intravenous and oral exposures.
Fundam. Appl. Toxicol. 6:7-15.
Shealy, D. B., Barr, J. R., Ashley, D. L., Patterson, D. G., Camann, D. E., and
Bond, A. E. 1997. Correlation of environmental carbaryl measurements
with serum and urinary 1-naphthol measurements in a farmer applicator
and his family. Environ. Health Perspect. 105:510-513.
Smith, B. J. 2005. Bayesian Output Analysis program (BOA), version 1.1.5.
Ames: University of Iowa.
Sogorb, M. A., and Vilanova, E. 2002. Enzymes involved in the detoxification
of organophosphorus, carbamate and pyrethroid insecticides through
hydrolysis. Toxicol. Lett. 128:215-228.
Strother, A., and Wheeler, L. 1980. Excretion and disposition of [14C]carbaryl
in pregnant, non-pregnant and fetal tissues of the rat after acute administra-
tion. Xenobiotica 10:113-124.
Suckow, M. A., Weisbroth, S. H., and Franklin, C. L., eds. 2005. The labora-
tory rat. Boston: Elsevier Academic.
Tanaka, R., Fujisawa, S., Nakai, K., and Minagawa, K. 1980. Distribution and
biliary excretion of carbaryl, dieldrin and paraquat in rats: Effect of diets.
J. Toxicol. Sci. 5:151-162.
Tang, J., Cao, Y., Rose, R. L., and Hodgson, E. 2002. In vitro metabolism of
carbaryl by human cytochrome P450 and its inhibition by chlorpyrifos.
Chem. Biol. Interact. 141:229-241.
Thrall, K. D., and Woodstock, A. D. 2002. Evaluation of the dermal absorption
of aqueous toluene in F344 rats using real-time breath analysis and physio-
logically based pharmacokinetic modeling. J. Toxicol. Environ. Health A
65:2087-2100.
Timchalk, C., Nolan, R. J., Mendrala, A. L., Dittenber, D. A., Brzak, K. A.,
and Mattsson, J. L. 2002. A physiologically based pharmacokinetic and
pharmacodynamic (PBPK/PD) model for the organophosphate insecticide
chlorpyrifos in rats and humans. Toxicol. Sci. 66:34-53.
U.S. Environmental Protection Agency. 1999. Integrated Risk Information
System (IRIS) on carbaryl. Washington, DC: National Center for Environ-
mental Assessment, Office of Research and Development.
Previous
TOC
-------
1380
A. NONG ET AL.
Wang, C., and Murphy, S. D. 1982. The role of non-critical binding proteins in
the sensitivity of acetylcholinesterase from different species to diisopropyl
fluorophosphate (DFP) in vitro. Life Sci. 31:139-149.
Ward, S., May, D., Heath, A., and Branch, R. 1988. Carbaryl metabolism is
inhibited by cimetidine in the isolated perfused rat liver and in man. Clin.
Toxicol. 26:269-281.
Whittaker, M. 1986. Cholinesterase. Monographs in Human Genetics, Vol. 11.
Basel: Karger.
APPENDIX
MODEL PARAMETER ABBREVIATIONS
Doral: the oral dose
RIV: the intravenous injection rate
Qt: blood flow (Qc: cardiac output)
Hot: hematocrit
Vt: tissue volume
PAt: tissue permeability constant
Pt: tissue :blood partition cofffiecient
Ct: tissue concentration (a: arterial, v: venous)
A tissue'- tissue amount
AVtissue: tissue blood circulatory amount
T: cholinesterase tissue amount
7: inhibited cholinesterase tissue amount
ka: oral absorption rate constant
kmle: biliary excretion rate constant
RAM: rate of metabolism from carbaryl to 1-naphthol or other
composite metabolites
Vmax: maximal metabolic velocity
Km: Mechealis Menton affinity constant
kf, ke: rate constants of naphthol sulfate formation and elimination
RABrninh: rate of brain cholinesterase inhibition by carbaryl
RABldinh : rate of blood cholinesterase inhibition by carbaryl
kd, kr, ki: rate constants of cholinesterase degradation, regeneration
and inhibition
Ks: rate of cholinesterase synthesis
Note: Ks- was calculated from enzyme activity constants as
described in the methods and with the values in Table 2 and 5.
Tissue volumes given in Table 2 are the sum of true tissue
volume (95% of the reported value) and the tissue blood
volume (5% of the reported value).
PHARMACOKINETIC MODEL EQUATIONS
Carbaryl tissue kinetics
Liver tissue concentration:
dt
^- = Qliv(Ca-CVlwer
C
liver
dt
CVllver --^- \~RAMnapmol -RAMothers
"liver .
Cllver=Allver/(0.95xVllver)
RAMmetabolite —
max,metabolite '
K -4-C
r^m,metabolite ~"~ ^"liver
Fat and remaining body tissues concentration:
dt
dAti,,,
p
tissue
= PAt\CVasau—f
*t
dt
CVtissue=AVtJSSUJ(Q.Q5xVtissue)
Brain tissue concentration:
dAV>
dt
^ = Qbrn(Ca-CVbram) + PABr
I D
V brain
^^- = PABr\
dt
brain ~ brain/{• brain /
CVbrain=AVbmm/(0.05xVbmm)
Arterial and venous blood concentration (Ca, Cv):
dAh
^blood
dt
= Qc (CV - Ca) - RABldmh - RAMl
blood,naphthol
Cv = (Qbrn x Cvbram + Qfat x Cvfat + Qliv x Cvllver
+ Qbod*Cvbody+RIV)IQc
Ca = AUood/VUood
Cplasma = Ca(l - Hct)
Crbc = CaxHct
RAM,
-KMexAllver+kaxDoml
blood, naphthol
Previous
TOC
-------
PBPK/PD MODELING OF CARBARYL
1381
1 -Naphthol and other composite metabolites tissue
kinetics
Fat, brain and remaining body tissue concentration (includ-
ing liver for other metabolites):
fV = AY IV
tissue tissue I tissue
Cvxtissue=CXtissue/PXtissue
Liver tissue concentration (1-naphthol):
deliver,naphthol
dt
— QHv(Canaphthol Cvliver^naphthol ) + RAMnaphthol
aphthol
"•Bile,naphthol X ^liver,naphthol V X^liver,naphthol
C = A IV
liver,naphthol liver, naphthol / liver
Cv =C IP
liver,naphthol liver, naphthol / liver, naphthol
AA
"^liver, naphthol- sulfate
iphtho
~dT
utjau! _ ,f ,
— ft/ A s±nver ina
inaphthol
-kexA,
naphthol—sulfate
Arterial and venous blood concentration (Ca, Cv):
dA
blood, naphthol
~dt
Wood,naphthol
X ^-blood,naphthol + \*^-"^blood,naphthol )
Cv = (Qbrn x Cvbmm + Qfat x Cvfat +
QHvxCvllver+QbodxCvbody)/Qc
Ca = Ablood /Vhlood
Pharmacodynamic model equations
Brain, plasma or red blood cells acetyl- and butyryl-
cholinesterase (AChE, BChE) inhibition:
~ ^ X ^tissue,AChE X^-tissue
"Atissue,AChE
dt
X ^tissue,AChE X ^tissue + ^AChE X •"-tissue,AChEl
dt
v C —fa- v A
A <^Br MjichE A Atissue,
ltissue,BChE _ ^^
dt
•ue,BChEI
dt
X ^tissue,BChE X ^Br + ^BChE X ^tissue,BChEI
X -"-tissue,BChE X ^tissue
~ ™BChE X ^tissue,BChEl
Previous
TOC
-------
Bayesian Meta-Analysis
of Genetic Association Studies
with Different Sets of Markers
Claudio Verzilli,2-10 Tina Shah,1-10 Juan P. Casas,2 Juliet Chapman,2 Manjinder Sandhu,3
Sally L. Debenham,3 Matthijs S. Boekholdt,4 Kay Tee Khaw,3 Nicholas J. Wareham,5 Richard Judson,6
Emelia J. Benjamin,7 Sekar Kathiresan,7 Martin G. Larson,7 Jian Rong,7 Reecha Sofat,1
Steve E. Humphries,8 Liam Smeeth,2 Gianpiero Cavalleri,9 John C. Whittaker,2-*
and Aroon D. Hingorani1
Robust assessment of genetic effects on quantitative traits or complex-disease risk requires synthesis of evidence from multiple studies.
Frequently, studies have genotyped partially overlapping sets of SNPs within a gene or region of interest, hampering attempts to com-
bine all the available data. By using the example of C-reactive protein (CRP) as a quantitative trait, we show how linkage disequilibrium
in and around its gene facilitates use of Bayesian hierarchical models to integrate informative data from all available genetic association
studies of this trait, irrespective of the SNP typed. A variable selection scheme, followed by contextualization of SNPs exhibiting inde-
pendent associations within the haplotype structure of the gene, enhanced our ability to infer likely causal variants in this region with
population-scale data. This strategy, based on data from a literature based systematic review and substantial new genotyping, facilitated
the most comprehensive evaluation to date of the role of variants governing CRP levels, providing important information on the
minimal subset of SNPs necessary for comprehensive evaluation of the likely causal relevance of elevated CRP levels for coronary-
heart-disease risk by Mendelian randomization. The same method could be applied to evidence synthesis of other quantitative traits,
whenever the typed SNPs vary among studies, and to assist fine mapping of causal variants.
Introduction
Genetic effects underlying complex traits and disorders are
small, and their detection requires comprehensive typing
of single nucleotide polymorphisms (SNPs) in large sam-
ples.1'2 Many previous genetic association studies have
been underpowered,3'4 and even very large biobanks5
may not individually provide conclusive results for certain
outcomes. Quantitative synthesis of evidence from avail-
able studies remains vital,6"8 even in the era of genome-
wide analyses.9"11 However, a major obstacle is that studies
of the same gene, region, or even the genome as a whole
may type a different repertoire of SNPs, thereby yielding
partially overlapping genotypic data. Moreover, often
only single SNP summary data, for instance genotype
means at each SNP, is reported.
The meta-analysis of results from each marker in isola-
tion would exclude those studies that did not type the
marker in question, with a potential loss of power; more-
over, multiple single-SNP analyses are difficult to interpret.
Instead, it would be useful to be able to combine data with
information from all sites, adjusting any association at
each site for the possible correlation with the remaining
variants. One could then disentangle effects at causal sites
from those at sites that are in LD with a causal variant(s)
and also borrow information across studies. With focus
on a quantitative trait, we develop a Bayesian hierarchical
linear regression that models linear transformations of the
study-specific genotype-group-specific phenotypic means
and that uses pairwise LD measurements between markers
to make posterior inference on adjusted effects. Informa-
tion on pairwise marker LD is often provided by the indi-
vidual studies as part of the results reported. Alternatively,
for markers that are not considered jointly in any of the
study at hand, it can often be obtained from public data-
bases. This information is then used to specify informative
priors in our Bayesian framework. Specifically, the be-
tween-marker correlations are modeled by introduction
of spatially correlated random effects having a conditional
autoregressive distribution (CAR).12'13 The between-study
variability is then accommodated with a random intercept
term across studies.
Our approach is motivated by the meta-analysis of stud-
ies assessing the effect of variants in the C-reactive protein
(CRP [MIM 123260]) gene region on plasma CRP levels.
CRP is a circulating monomorphic hepatic acute-phase
protein that indexes and may mediate aspects of the in-
flammatory response.14 Aside from acute-phase elevations,
JCentre for Clinical Pharmacology, University College London, London WC1E 6JF, UK; 2Department of Epidemiology and Population Health, London
School of Hygiene and Tropical Medicine, London WC1E 7HT, UK; 3Department of Public Health and Primary Care, University of Cambridge, Cambridge
CB1 8RN, UK; 4Department of Cardiology, Academic Medical Center, Amsterdam 1100 DD, Netherlands; 5MRC Epidemiology Unit, University of Cam-
bridge, Cambridge CB2 OQQ, UK; 6Genaissance Pharmaceuticals, New Haven, CT 06511, USA; 7Framingham Heart Study, Framingham, MA 01702-
5827, USA; 8Centre for Cardiovascular Genetics, University College London, London WC1E 6JF, UK; 'Molecular and Cellular Therapeutics, RCSI Research
Institute Royal College of Surgeons in Ireland, Dublin 2, Ireland
1 "These authors contributed equally to this work.
•Correspondence: ]ohn.whittaker@lshtm.ac.uk
DOI 10.1016/].a]hg.2008.01.016. ©2008 by The American Society of Human Genetics. All rights reserved.
The American Journal of Human Genetics 82, 859-872, April 2008 859
Previous
TOC
-------
Omrviiw of Ctrl
1S7.W6K 157.9WK 1S7.VS1K
!™
.F-H
blood concentrations of CRP show similar within-individ-
ual variability to serum cholesterol, and like cholesterol,
CRP has been shown to be associated with future coronary
heart disease (CHD) risk in observational studies.15 How-
ever, the etiological relevance of this potentially important
and highly studied link with CHD is uncertain because
CRP may simply be a marker for established risk factors
or for subclinical atheroma.16'17 Common SNPs that are
in the gene encoding CRP and that influence its level
may help provide insight on the link because, unlike
CRP itself, genotype is fixed and unaffected by subclinical
disease and the naturally randomized allocation of alleles
at conception balances the distribution of potential con-
founding factors among genotypic classes. Genetic associ-
ations are therefore less prone to biases that limit causal
inference from observational studies, and genetic studies
possess properties of a randomized intervention trial.16"18
Therefore, identification of CRP-gene variants (HGNC:
2637; Iq21-q23) that influence its concentration is funda-
mental to evaluating the causal relevance of CRP with the
principle of Mendelian randomization.19
In the absence of hepatic stores of CRP, and given its
constant rate of clearance, gene transcription provides
the major point of regulation.14 Transcription may be
modified by regulatory SNPs because concentrations of
CRP show strong concordance among monozygotic twins
and family studies suggest substantial heritability.20 In
populations of European descent, there are 11 common
SNPs with minor allele frequency >5% within 6 kb of
the CRP gene, but extensive linkage disequilibrium (LD)
means that four major haplotypes account for 94% of
chromosomes (see Web Resources).21'22 Individual reports
evaluating associations of CRP SNPs with CRP concentra-
tion have either typed single SNPs or a subset of SNPs
(sometimes tag SNPs) in this region (see Table SI available
online). However, the SNPs have varied across studies,
thereby limiting the ability to pool all available data. We
therefore developed a new integrative approach to evi-
dence synthesis of genetic association studies that allows
for this complexity.
Methods for combining data from genome-wide scans
with nonoverlapping sets of SNPs with individual-level
genotyping data have been recently proposed by March-
ini et al.23 Here, because individual-level data are not
available for most of the studies on CRP, we develop
Figure 1. Location of the Eight CRP
SNPs Typed Directly in the 26 Data Sets
Included in This Study
The upper track shows chromosomal Lo-
cation; the middle track shows SNP Lo-
cation and Log(P) for the per-allele
random-effect meta-analysis (from Left
to right, the SNPs are ordered as foLLows:
rs3093077, rs!205, rsl!30864, rs!800947,
rs!417938, rs3091244, rs2794521, and
rs3093059); and the Lower track shows the
intron/exon structure of the CRP gene.
a method that allows the synthesis of studies providing
only summary data. Also, we are mainly concerned with
the synthesis of SNP data in regions of interest for fine
mapping, where the number of markers typed is small
and interest is on disentangling independent effects
using a variable selection scheme, for which the Marchini
approach is not suitable.
Material and Methods
We first conducted a literature-based systematic review of all rele-
vant studies (irrespective of the SNP typed). A total of 23 published
data sets identified by systematic review evaluated associations
of eight SNPs (rs3093059; rs2794521; rs3091244; rs!417938;
rs!800947; rsl!30864; rs!205; and rs3093077) in the CRP gene
with CRP concentration (Figure 1). With data from SeattleSNPs,
a combination of three SNPs (rsl!30864; rs!205; and rs3093077)
was identified as haplotype tag SNPs with the haplotype r2 method
in European subjects. These tag SNPs were typed in three addi-
tional population-based studies, thereby giving an aggregate of
26 studies including 32,802 subjects. No SNP was typed in every
study, but there was partial overlap of SNP typing across several
studies (see Appendix A and Table SI).
Bayesian Hierarchical Model
We indicate with Yf the continuous trait of interest for subject
/e {!,.. .,ns} and study s e {!,.. .,5}. If all studies have genotyped in-
dividuals at all m marker locations, and these data are available for
all individuals (individual patient data [IPD]), a sensible approach
to pool information across studies would be the random-effect
model
Ys ^T (f*s n , -I 2f \ (-1 \
^ IN llj p -f- Insjti5,(7 lns j \L)
where Cs is the rf x (m + 1) design matrix coding for the chosen
genetic model (e.g., for an additive models, 0, 1, and 2 for homo-
zygous wild-type, heterozygous, or homozygous mutant geno-
types, respectively) and the intercept term, fis ~N(0, a2) is
a study-specific random intercept term, !„* is the ns x 1 vector
of ones, Ins is the ns x ns identity matrix, and p1 = (fi0,fii,...,fim)'
is the (m + 1) x 1 vector of regression coefficients of interest mea-
suring the effect of genotype group on Y. One could then assess the
relative importance of each marker by using a variable selection
scheme; we use a reversible jump algorithm on the space of possi-
ble models as part of the MCMC scheme as described later in the
text.24'25
However, studies will rarely consider all m markers together;
rather, ms < m will have been typed in study s corresponding to
860 The American Journal of Human Genetics 82, 859-872, April 2008
Previous
TOC
-------
a subset Ls of columns of the complete design matrix Cs, Xs say of
size ns x (ms + 1). Also, complete individual patient data for all
studies are rarely available. Instead, we have the summary statis-
tics reported in each study as in the case of the CRP studies. Typi-
cally, data will consist of means, variances, and numbers of indi-
viduals for each genotype groups and each marker. These are
denoted by y^, vg, and n^, respectively, for genotype group
g = \,...,Gj of marker; e [Ls] in study s. The notation allows for
marker-specific numbers of genotype groups and thus the possibil-
ity of having a mixture of biallelic and triallelic markers, as in the
application to the CRP data, or different genetic models.
Our approach uses Equation (1) as the building block but
models the linear transformations Xs = Xs Ys as multivariate
normally distributed across studies
MVNm
(2)
where Xs' indicates the transpose of Xs. All entries of the vector Xs
can be obtained from the available data summaries. For instance,
the first element corresponding to the intercept term is the overall
sum of the y values, and any other entry can be obtained similarly
from the genotype-group-specific phenotype means and counts
Vs. ns.
'«" V
However, the new design matrix Ws = Xs Cs is only partially ob-
served. In particular, only the dot products involving the columns
of Xs with themselves or the intercept term can be derived from
the observed genotype-group counts. The remaining entries are re-
placed by their expected values under Hardy-Weinberg equilib-
rium (HWE) and the known pairwise LD patterns. Specifically, in-
dicating with wu, h 3= I, a generic such entry, we first obtain an
estimate of the joint bivariate genotype distribution from the
known marginal allele frequencies and the pairwise measure of
LD.26 For example, if both markers are biallelic, this involves esti-
mation of the 3x3 matrix of the genotype distribution, and this
estimation is then multiplied by the study size to give expected
counts. Finally, we obtained wu by summing the appropriate en-
tries of the resulting matrix of expected counts multiplied by the
values used to code the genotype groups in the design matrices.
Notice that the vector of coefficients (3 retains the same interpreta-
tion and scale of the original model in Equation (1) (in the exam-
ple below, additive effect of variants on log CRP plasma levels) be-
cause it is derived from a linear transformation of the variables
therein.
As well as in the derivation of the new design matrix Ws, prior
information of between-marker LD patterns is also incorporated
in the specification of the (partially unobserved) variance-covari-
ance matrix in Equation (2). Specifically, we partition CT2XSXS
into a spatially structured component and a residual, unstruc-
tured, component. We obtained the former by introducing
marker-specific random effects having a zero-mean conditional
autoregressive distribution
U
(3)
where U is a vector of size m, the number of unique markers across
studies, R is a matrix of weights reflecting spatial associations be-
tween the elements of Xs, and M is a diagonal matrix.12'27 Thus the
covariance matrix in Equation (2) becomes
-------
Linkage Disequilibrium
234567
Figure 2. Graphical Representation of Equation (4)
Solid and dotted Lines represent stochastic and deterministic
dependencies, respectively.
distribution on the number of regression terms k included in each
model.25'30-31 The simulation study in the next section includes
a sensitivity analysis of this choice. A graphical representation of
the hierarchical model (4) is given in Figure 2.
Single-Marker Random-Effect Meta-Analysis
Results from the multilocus model are compared to those obtained
from a more traditional single-locus random-effects meta-analysis
in both simulation studies and with real data from the CRP-gene
region. For the latter, a per-allele effect (95% CI) of individual
SNPs on CRP concentration was derived from each individual
study. The individual-study linear trend (additive effect) per cate-
gory increase in genotype with mean data was calculated by
simple linear regression, with genotypes coded as 0, 1, and 2 for
homozygous common allele, heterozygous, and homozygous rare-
allele, respectively, with the least-square linear-trend-coefficient
formula, which only depends on the mean values and its standard
deviations. A sensitivity analysis restricted to studies with more
than 500 subjects, healthy at time of blood sampling, or to studies
that reported all the required standard deviations was also con-
ducted (Table S2). Subsequently, the study-specific linear trend
and its standard error were pooled with random-effect models.
Subsidiary analyses included pairwise comparisons within each
polymorphism. The DerSimonian and Laird Q test, and the I2
test,32 were used for evaluating the degree of heterogeneity
between studies.
Results
Simulation Studies
We considered various scenarios differing in the number of
studies and, for the multilocus approach, in the priors on
the model space. Data were obtained as follows: We first
simulated a pool of 4000 haplotypes at seven biallelic
markers. Pairwise LD measures (r) between the seven
0.0418
0.0055
0.0394
-0.0300
0.0056
-0.1941
0.0124
0.0148
-0.0784
0.3697
0.0260
0.0154
-0.0567
0.2737
0.0144
0.0141
-0.0471
0.2443
Marker 2
Figure 3. Pairwise LD Measures between Markers Used in the
Simulation Study
Pairwise LD Measures are r values.
SNPs are shown in Figure 3, with high LD only between
the last three markers. SNP 6 is assumed to be the single
causal site in the region and is retained in all subsequent
analyses. Given the high LD between SNP 5, 6, and 7, we
expect the results from the univariate analyses to be less
conclusive than those from the multiple marker approach
that adjusts for the between-marker correlations. The study
size ns was drawn from a normal distribution with mean
600 and variance 100, rounded to the nearest integer.
Then, for subject ie{l,...,ns} and study 5, a continuous
phenotype /,• is simulated as
yl = 00
+ Ms
(6)
where gi6 denotes the genotype of subject / at marker site 6
(0, 1, or 2 for homozygous wild-type, heterozygous, or ho-
mozygous mutant, respectively), (P0, Pe) = (1, 2), ^s ~ N(0,
1), and s ~ N(0,1). To reflect the fact that not all markers are
typed in every study, we select at random ms markers out of
the possible seven for each study. Thus, in most cases the
univariate analyses are based on fewer than the maximum
total of S studies. For each simulated data set, we also esti-
mated the unadjusted univariate additive effects and their
standard errors at each SNP site; the additive effects are
then combined in the univariate random-effect analy-
ses.33'34 Tables 1 and 2 present the results from the multi-
ple-marker meta-analyses. The number of studies consid-
ered was 10, 20, or 40. In each case, the tables report the
results obtained with Poisson priors on the model size in
the reversible jump algorithm with different means (1 or
2 for priors a and b, respectively) or a uniform prior on
the model space (prior c). Notice that the Poisson priors
give more weight to the null model and may in general
be a more reasonable choice in this setting. For example,
862 The American Journal of Human Genetics 82, 859-872, April 2008
Previous
-------
Table 1. Bayesian Muttilocus Meta-Analysis
Parameter
03
06
Number of Studies
Prior
True
Post proba
a Meana
BCI length
Post prob
10 b Mean
BCI length
Post prob
c Mean
BCI length
Post prob
a Mean
BCI length
Post prob
20 b Mean
BCI length
Post prob
c Mean
BCI length
0.01
(0.01)
-0.01
(0.05)
0.19
0.02
(0.01)
-0.03
(0.04)
0.24
0.05
(0.02)
-0.03
(0.04)
0.24
0.01
(0.01)
-0.04
(0.03)
0.16
0.02
(0.02)
-0.04
(0.03)
0.16
0.06
(0.05)
-0.05
(0.03)
0.18
0.01
(0.004)
-0.004
(0.04)
0.20
0.01
(0.004)
-0.01
(0.03)
0.22
0.04
(0.02)
-0.01
(0.04)
0.21
0.01
(0.002)
-0.02
(0.03)
0.14
0.01
(0.01)
-0.02
(0.03)
0.15
0.03
(0.01)
-0.02
(0.02)
0.15
0.01
(0.004)
0.03
(0.03)
0.15
0.01
(0.01)
0.03
(0.03)
0.16
0.04
(0.02)
0.02
(0.03)
0.17
0.01
(0.003)
0.03
(0.02)
0.11
0.01
(0.01)
0.02
(0.03)
0.12
0.03
(0.01)
0.02
(0.02)
0.12
0.005
(0.002)
0.01
(0.03)
0.14
0.01
(0.004)
0.02
(0.03)
0.15
0.03
(0.01)
0.01
(0.03)
0.15
0.004
(0.002)
0.01
(0.02)
0.10
0.01
(0.003)
0.01
(0.02)
0.10
0.02
(0.01)
0.01
(0.02)
0.11
0.01
(0.01)
-0.01
(0.07)
0.35
0.02
(0.01)
-0.01
(0.08)
0.36
0.07
(0.03)
-0.01
(0.07)
0.35
0.01
(0.003)
-0.01
(0.05)
0.23
0.02
(0.01)
-0.02
(0.05)
0.24
0.06
(0.05)
-0.01
(0.06)
0.26
1.00
(0.00)
2.00
(0.04)
0.19
1.00
(0.001)
2.00
(0.05)
0.23
1.00
(0.001)
1.99
(0.04)
0.33
1.00
(0.00)
1.99
(0.04)
0.12
1.00
(0.001)
1.99
(0.03)
0.17
1.00
(0.002)
1.99
(0.03)
0.23
0.03
(0.02)
0.16
(0.14)
0.76
0.07
(0.08)
0.15
(0.16)
0.76
0.15
(0.05)
0.13
(0.12)
0.76
0.03
(0.08)
0.09
(0.10)
0.56
0.05
(0.04)
0.13
(0.112)
0.57
0.13
(0.08)
0.10
(0.108)
0.59
1.19
(0.01)
1.08
(0.02)
1.12
(0.01)
1.11
(0.01)
1.06
(0.01)
1.09
(0.01)
1.25
(0.61)
1.13
(0.34)
1.42
(0.46)
1.10
(0.26)
1.08
(0.17)
0.99
(0.16)
Results are averages (std) over 100 replicated data sets. Mean posterior estimates and credible intervals are conditional on the SNP being included in a model.
prior a is Poisson(l) and assigns a probability of -0.26 of
having more that one associated site (-0.59 for b). The
values shown are averages over 100 replicates. For each sce-
nario, we report the marginal posterior probability of se-
lecting each SNP and the mean and 95% credible intervals
of the posterior distributions of each additive effect, condi-
tional on the SNP being selected.35'36 Note that posterior
distributions can be reliably estimated only for markers
with relatively high posterior probability of inclusion
(e.g., >0.5), and results in the table should be interpreted
Table 2. Bayesian Multilocus Meta-Analysis
Parameter
0i
03
05
06
Number of Studies
Prior
True
Post prob
a Mean
BCI length
Post prob
40 b Mean
BCI length
Post prob
c Mean
BCI length
0.01
(0.15)
-0.05
(0.02)
0.12
0.03
(0.03)
-0.05
(0.02)
0.12
0.07
(0.06)
-0.05
(0.02)
0.12
0.01
(0.003)
-0.03
(0.02)
0.10
0.01
(0.01)
-0.02
(0.02)
0.11
0.03
(0.01)
-0.03
(0.02)
0.11
0.004
(0.002)
0.02
(0.02)
0.08
0.01
(0.00)
0.02
(0.02)
0.08
0.02
(0.01)
0.02
(0.02)
0.10
0.003
(0.001)
0.01
(0.01)
0.07
0.01
(0.002)
0.01
(0.02)
0.07
0.01
(0.01)
0.01
(0.01)
0.077
0.01
(0.004)
-0.02
(0.04)
0.17
0.02
(0.01)
-0.03
(0.03)
0.19
0.04
(0.02)
-0.01
(0.04)
0.19
1(0)
1.99
(0.02)
0.10
1(0)
1.99
(0.02)
0.12
1(0)
1.99
(0.02)
0.18
0.025
(0.28)
0.10
(0.08)
0.46
0.04
(0.03)
0.12
(0.07)
0.46
0.11
(0.06)
0.10
(0.10)
0.48
1.04
(0.01)
1.03
(0.01)
1.03
(0.01)
1.09
(0.14)
0.97
(0.13)
1.01
(0.11)
Results are averages (std) over 100 replicated data sets. Mean posterior estimates and credible intervals are conditional on the SNP being included in a model.
The American Journal of Human Genetics 82, 859-872, April 2008 863
Previous
TOC
-------
Table 3. Single-Locus Random-Effects Meta-Analysis
Number of Studies
10
20
40
Results are averages
SNPID
True
Mean (std)
Mean BCI length
Mean (std)
Mean BCI length
Mean (std)
Mean BCI length
over 100 replicated data sets.
1
0
-0.033
(0.028)
0.385
-0.03
(0.018)
0.258
-0.032
(0.014)
0.183
2
0
-0.023
(0.024)
0.379
-0.019
(0.018)
0.253
-0.021
(0.012)
0.179
3
0
0.069
(0.026)
0.367
0.072
(0.017)
0.245
0.067
(0.013)
0.175
4
0
0.312
(0.026)
0.365
0.314
(0.017)
0.244
0.311
(0.015)
0.175
5
0
1.158
(0.039)
0.414
1.169
(0.025)
0.269
1.165
(0.02)
0.194
6
2
1.990
(0.040)
0.434
1.999
(0.026)
0.286
1.998
(0.017)
0.203
7
0
1.945
(0.036)
0.482
1.942
(0.03)
0.327
1.94
(0.021)
0.23
with this in mind. The marginal probability of selecting the
causal site is 1 independently of the prior used and even
when considering as few as ten studies, with almost no var-
iability across replicates. Notably, all other markers have
posterior inclusion probabilities close to zero and would
therefore not be selected if we were to use the traditional
threshold of 0.5. All conditional mean additive effects are
very close to the true values with a minor bias only for the
effect at SNP 7, which is the SNP in highest LD with the
causal site. The choice of prior distribution on model space
does not have a large effect on the results, with possibly nar-
rower credible intervals and slightly larger posterior proba-
bility of including SNP 7 under prior c compared to priors
a and b. This is to be expected because the Poisson priors
favor models with few terms, whereas the uniform prior
gives equal weight to all models. The tables also report the
results for the variance terms cr^ and cr^, which have
posterior estimates close to the true values in both cases. In-
creasing the number of studies has little or no effect on both
marginal posterior probabilities and posterior estimate bias
but does lead to narrower credible intervals as expected.
The univariate analyses on the other hand fail to unam-
biguously identify the causal site at position 6 (Table 3). On
the basis of results reported therein, although SNP 6 shows
the highest association with the phenotype, SNP 7 could
still be considered causal if no prior information is avail-
able to discriminate between the two. Even markers 4
and 5 would be selected on the basis of posterior credible
intervals; paradoxically, increasing the number of studies
only exacerbates the problem because credible intervals
become narrower.
The previous simulation study assumed the same LD
pattern across studies because study-specific genotype
data are simulated from a common haplotype pool. To
mimic a more realistic scenario, we further considered
study-specific LD patterns by simulating genotype counts
from study-specific haplotype pools characterized by
slightly different LD structures. The multilocus analysis
then uses the average LD table shown in Figure Dl (in
which we also report the standard deviations of the pair-
wise r2 values across studies in brackets). Results are re-
ported in Table 4 for replicates with 20 studies. The method
Table 4. Bayesian Multilocus Meta-Analysis
Number
of studies
20
Parameter
Prior True
Post prob
a Mean
BCI length
Post prob
b Mean
BCI length
Post prob
c Mean
BCI length
0i
0
0.01
(0.01)
-0.02
(0.04)
0.16
0.01
(0.01)
-0.01
(0.04)
0.2
0.03
(0.01)
-0.02
(0.03)
0.17
when the LD Structure Is Allowed to Vary
02
0
0.01
(0.01)
-0.02
(0.03)
0.15
0.01
(0.01)
0.01
(0.04)
0.12
0.03
(0.01)
-0.02
(0.02)
0.15
03
0
0.03
(0.02)
0.07
(0.01)
0.10
0.02
(0.01)
0.06
(0.02)
0.15
0.11
(0.13)
0.05
(0.03)
0.12
04
0
0.03
(0.06)
0.04
(0.04)
0.10
0.01
(0.01)
-0.02
(0.04)
0.15
0.10
(0.08)
0.04
(0.04)
0.10
05
0
0.01
(0.01)
-0.01
(0.06)
0.23
0.02
(0.01)
-0.01
(0.07)
0.32
0.04
(0.05)
0.01
(0.04)
0.22
across Studies
06
2
1.00
(0.00)
1.97
(0.01)
0.14
1.00
(0.00)
1.98
(0.04)
0.29
1.00
(0.00)
1.99
(0.03)
0.24
07
0
0.03
(0.03)
0.13
(0.11)
0.58
0.08
(0.05)
0.14
(0.28)
0.67
0.12
(0.04)
0.15
(0.05)
0.58
"I
1
1.06
(0.02)
1.08
(0.02)
1.04
(0.01)
4
1
1.07
(0.23)
1.03
(0.13)
1.07
(0.12)
Results are averages (std) over 100 replicated data sets. Mean posterior estimates and credible intervals are conditional on the SNP being included in
a model. See Figure Dl.
864 The American Journal of Human Genetics 82, 859-872, April 2008
Previous
TOC
-------
CRP levels by CRP gene variant
Additive model
CRP Gene variant No. Studies
(individuals)
rs1800947(+1059G-*-C)* 10(11045)
rs1205(+2302G-»A) 10(16942)
rs1417938(+194A--T) 5(7460)
rs2794521(-717A--G) 6(5803)
rs1130864(+1444C-*T) 19(21674)
rs3091244(-286C-»T/A) 7 (7786)
rs3093077(«!899T-»G) 4 (661
rs3093059(-757T-*C) 3 (3475)
15) ~f-
1
)
) —
4)
)
)
)
-0.38 (-0.50, -0.25)
-0.35 (-0.41
— H 0.17(0.03
— 1 0.12 (-0.10
H 019(0.14
B 0.20(0.15
1 0.48 ( 0.33
H 0 58 ( 0 37
-0.28)
0.31)
0.35)
025)
0.25)
0.62)
0 78)
Figure 4. Summary Effect from Tradi-
Traditional meta-analyses Bayesian model tional Meta-Analysis and Bayesian Mul-
Estimate Estimate tiple-SNP Hierarchical Linear Model of
(95% confidence intervals) (95% credible intervals) thp Elflht SNPs in the CRP Gene
Values shown are additive genetic effects
on (log) CRP levels with 95% confidence
intervals or credible intervals for tradi-
tional and Bayesian analyses, respectively.
For the Bayesian analysis, results are
shown only for those markers that appear
to be strongly associated after variable
selection (see Figure 5). N/A refers to
SNPs excluded from the model. The asterisk
indicates the dominant model. Negative
N/A values indicate the variant allele is associ-
ated with a lower CRP concentration.
0.59 (0.37, 0.95)
-0.26 (-0.46,-0.11)
-0.52 (-0.68, -0.36)
N /A
N /A
0.34(0.20 1.17)
-0.7 -0.5 -0.3 -0.1
appears to be fairly robust to minor deviations in LD pat-
terns across studies (similar to those observed for the real
data in the next section); large differences in LD structures
across studies would necessarily invalidate the meta-
analytical approach because there would be little informa-
tion to borrow for variable selection.
Finally, we considered reducing the effect at the causal
site to 1.5 or placing it at marker position 2, which is in
linkage equilibrium with the other sites: In both cases,
the causal site is selected with high posterior probability
(>0.8, results not shown).
The WinBUGS code used to fit the model is given in
Appendix C.
A Meta-Analysis of CRP Studies
The traditional single-locus meta-analyses require that the
available data be partitioned into groups of studies in
which the same SNP was typed directly. In these analyses,
seven SNPs were associated with a codominant effect on
CRP concentration (Figure 4) with the per-allele effect
in the range of 0.19-0.58 mg/L (absolute p values:
rs!800947 = 4.35 x 10~9; rs!205 = 7.76 x 10~26;
rs!417938 = 1.77 x KT2; rsl!30864 = 2.73 x KT11;
rs3091244 = 4.50 x 10~ls; rs3093077 = 5.03 x KT11;
and rs3093059 = 2.27 x 10~8), corresponding to -0.3-
0.8 SD of the population distribution of CRP 37. The
main effect estimates were robust to analyses limited to
studies of >500 subjects (Table S2), providing strong evi-
dence for an association at this locus. However, because
pooled analyses of this type are limited to individual
SNPs, it is unclear which of these SNPs have independent
effects and which are associated because of correlation
with other observed or unobserved SNPs, including the
true causal variant(s). This can be overcome by incorporat-
ing available information on pairwise LD in the region (Ta-
ble S3) within a Bayesian multilocus model as described
above. Bayesian model selection can then facilitate identi-
fication of variants showing the strongest independent
association with CRP concentration (Figure 5 and Table
5). The approach yields posterior model probabilities
TS3093059
I-S2794521
rs3091244
rs1417938
-1.0 -0.5 0.0 0.5
rs1800947
-0.5 0.0 0.5 1,0
rs1130864
0,5
0.5
1
-1.0 -0.5 0.0 0.5 1.0 -1.0 -0.5 0.0 0.5 1.0
rs1205 TS3093077
0.5
1
0.5
1
-1.0 -0.5 0.0 0.5 1.0 -1.0 -0.5 0.0 0.5 1.0
Figure 5. Results from the Multiple-SNP Meta-Analysis using
the Bayesian Hierarchical Linear Model
The shaded bars show the posterior probability that each SNP is
included in a model, calculated from the posterior sample of models.
The x axis indicates the additive effects of each SNP on log CRP
plasma levels, conditionalon thatSNP being included in the model,
and the y axis indicates the corresponding posterior density. The
curves can thus be interpreted as smoothed histograms representing
the probability that the SNP effects take the values on the x axis.
Also shown are the densities, medians (A), and 95% credible inter-
vals ( ) for the additive effects of each SNP on log CRP levels.
The American Journal of Human Genetics 82, 859-872, April 2008 865
Previous
TOC
-------
Table 5. Application to the Meta-Analysis of CRP Studies
Prob
SNP included
rs3093059
rs2794521
rs3091244
rs!417938
rs!800947
rsl 130864
rs!205
rs3093077
0.22
0.12
0.10
0.07
0.06
Models with more than 2% posterior probability are shown. Results assume a Poisson(2) prior on model size in the reversible jump algorithm.
conditional on the observed data from which marginal
probabilities of association for each SNP can be readily
obtained. Of the markers considered, SNPs rsl!30864,
rs!205, and rs3093077, all in the 3 UTR, retain the stron-
gest independent association with CRP concentration.
An additional synonymous SNP in exon 2 (rs!800947) ap-
pears to be important, although its posterior probability of
association is sensitive to the prior on the model space, and
becomes unimportant if a more restrictive prior on the
number of associated markers in the region is used (results
not shown). These four SNPs yield the model with the
highest posterior probability (Figures 4 and 5 and Table
5). Again, the models were not materially altered when
analyses were limited to studies of >500 subjects (results
not shown).
Notably, SNPs rsl!30864, rs!205, andrs3093077 formed
the trio of tag SNPs. Because each tag SNP marks a different
haplotype, the Bayesian model implies the presence of at
least three functional SNPs regulating CRP level (Figure 6).
Using HapMap, we found that there were 11 SNPs in strong
LD with rs!205, (five with pairwise r2 = 1) within an asso-
ciated interval of -100 kb. There were 11 SNPs in strong LD
with rs3093077 (nine with pairwise r2 = 1), within a larger
associated interval of -300 kb. A total of 22 SNPs lay in an
associated interval of 100 kb encompassing rsl 130864
(nine with pairwise r2 = 1) (Figure 7). Because tightly
linked SNPs were identified in the associated intervals,
a careful assessment of potential functionality for each of
these SNPs is now required.
As mentioned in the previous section, in order to accom-
modate outliers and heavy tails, we assumed the distribu-
tion of the between-studies random effects fis to be a mix-
ture of normals. In particular, inspection of the residuals
from a model fitted without the between-studies random
effect appears to suggest the use of a two-component
mixture, see Figure 8. The graph plots a sample of the
quantities
r? = :r?-W-.(t)/?(t)-[7.(t) (7)
for current values of the spatial random effects U and
model at iteration t.28 The posterior distribution of a.\
and a.2 had means of —0.014 and 3.356, respectively (Fig-
ure 8), whereas TT had posterior median estimate of
0.879. By monitoring the mixture component assign-
ments of each study, we found that outlying studies were
mostly assigned to the second component as expected
(results not shown).
Discussion
With only small genetic effects expected to contribute to
most complex diseases, the meta-analysis of studies that
consider variants in the same genetic region is a promising
™o
~
r*1M1K5 nt7i»7
eofis3091244n
rsn» 1*2027471
HMlO
Figure 6. A Reduced Median Network
Constructed with HapMap CEPH Data for
a 20 kb Region Containing the CRP Gene
Yellow circles indicate haplotypes. The size
of each circle is proportional to the fre-
quency of that haplotype in the HapMap
CEPH population. Non-HapMap SNPs (indi-
cated in italics) were placed on the net-
work with information from other CEPH
populations.
866 The American Journal of Human Genetics 82, 859-872, April 2008
Previous
TOC
-------
Overview of Chrl
B
1.U
0.8
0.6.
0.4.
0.2.
0.0
1 II
08
0.6.
0.4.
0.2
0.0
1 0
08
0.6
0.4.
0.2.
0.0
1 n
08
0.6.
0.4.
0.2.
0.0
-S18C0947
WI13MM
,.l. .L.JJ..I
•CM
,. ,1 1, ,
111,1
rrfOMOT?
ill
Illl ill III, [LI .
|
,
||
,
!
, |
1 lllll
Illl
J ill j
,
Ml
i II
,
1
||
I
1 ,
|
,
1
|
,
, ,.
,1,, ,1,
,1 1
1
||
JlL 1
1,1,
, *
111.
, '1'
,\ 1 ,
-J UW
1
!
1
,
,
t
i
1 .lUllMl 1 . , ....... L... J , m
1
,11 III ,, L........
11,1.1111 , ,,L, ..L, , .
,,,,,,, l 1 ,,,,,,,,, 1
• kJLii - ,
Figure 7. Genomic Context for CRP Gene
(A) Ideogram depicting the chromosome and region in which the CRP gene Lies (red Line).
(B) Gene diagram with introns and exons depicted as horizontaL and vertical blue lines, respectively.
(C) Pairwise i2 LD values between independently associating SNPs from Bayesian analysis (identified in top left of window, position
indicated by red arrow) and all other HapMap SNPs in the region (release 20, build 35, red = r2 > 0.8, yellow = 0.5 < i2 < 0.8, gray =
0.3 < i2 < 0.5, blue = 0.2 < r2 < 0.3, and dark gray = missing data).
strategy to increase our chances of finding any associa-
tions. Recognizing the importance of this approach, sev-
eral coordinated efforts have been initiated to ensure that
results from the individual studies follow agreed guidelines
and can be combined more easily.7
Most of the meta-analyses conducted so far have consid-
ered each marker in isolation, ignoring the possible corre-
lation between markers due to linkage disequilibrium that
reduces efficiency and that compromises the identification
of any causal site. In this paper, we have presented a multi-
marker approach that yields estimates of effect at each site
adjusted for the effects of other variants, as in multiple re-
gression. In both the simulation study and the application
to the CRP data, we assumed an additive genetic model.
Other choices are possible and would only involve changes
in the entries of the matrices W and X'X.
The methods borrow from the spatial data literature and
incorporate the prior knowledge of marker pairwise LD in
a fully Bayesian framework. For example, similar hierarchi-
cal models with spatial random effects are used extensively
in the analysis of spatial epidemiological data. A conve-
nient feature of the joint specification (Equation [3]) is
that it allows incorporation of the required correlation
structure as prior information in an explicit way.13'37 In ad-
dition, a reversible jump algorithm on the space of possible
model structures enables the selection of the most promis-
ing associations. The proposed approach assumes data on
a continuous phenotype. However, it could be extended
to the case of discrete outcomes, say case-control status,
by introducing a further set of continuous latent variables
related to the discrete outcome as in probit regression. Ex-
tensions to include metaregression are straightforward and
only involve introduction of a further hierarchy for the
vector of coefficients |3 in Equation (4) with means that
would then depend on study-specific covariates. Work on
these extensions is currently in progress.
When applied to the meta-analysis of studies in the CRP-
gene region, results provide evidence for three CRP modi-
fying alleles distributed over three of four common haplo-
types in Europeans. These alleles could account for the
The American Journal of Human Genetics 82, 859-872, April 2008 867
Previous
-------
• :,_,.•:'• • .• .:• :2...-; • •- .',;:: -
50e3
60e3
70e3 80e3
Iteration
90e3
100e3
Figure 8. Posterior Sample of Residuals from the Hierarchical
Model of Material and Methods Fitted without the Between-
Study Random-Effect Term ns
strong association with CRP of each of the three SNPs that
are chosen for their ability to tag others and that mark the
different haplotypes. The associated interval for each inde-
pendently associating SNP extended at least 100 kb from
either side of the open reading frame with a very sharp
boundary of LD for at least two of these. Within each inter-
val were a number of additional candidate causal SNPs in
complete LD with the index SNP from the Bayesian analy-
sis, any of which could, in theory, regulate CRP. Although
the A and T alleles of the triallelic SNP rs3091244 appeared
to exhibit functionality in previous reporter-gene studies
in vitro,21 this SNP was not retained within the Bayesian
model. Experimental studies of this type may be biased to-
ward the study of potential regulatory SNPs in the immedi-
ate vicinity against those located remotely from the gene
of interest because of size constraints on reporter-gene con-
structs. This might explain why results of such reporter
studies are, at times, discordant with the findings of associ-
ation analyses in populations38 or alternative experimen-
tal approaches to assessing functionality.39 Irrespective of
the true causal sites, the three tag SNPs adequately capture
functional variation at this locus for large-scale gene-
disease association studies. Although the naive expecta-
tion would be of narrower limits of error around the point
estimates of SNP effects with a Bayesian approach that
includes all studies simultaneously, this was not observed.
This is because unlike the traditional meta-analyses, the
Bayesian analyses were corrected for the effect of other
SNPs; that is, uncertainty about which SNPs are directly
associated with the trait was properly incorporated in
the analyses. However, the simultaneous use of all data
strengthens evidence for an association at the gene level;
the null model does not appear at all in the posterior
sample of models, reflecting virtual certainty of an effect
on CRP at this gene.
Our approach facilitates the integration of data from
studies that have genotyped different SNPs across the
same gene or region utilizing prior information on LD. It
has a number of favorable attributes and potential applica-
tions. By increasing the available data set of information
on any SNP, the efficiency of evidence synthesis is en-
hanced and the reliability of any identified associations is
increased. Further, the variable selection procedure allows
inference on the relative magnitude of any marker-pheno-
type association and identifies those SNPs that show the
strongest association with the phenotype, either because
they are the functional site(s) or because they exhibit the
strongest allelic association with (unobserved) functional
sites. IPD (where available) can also be incorporated readily
into the analysis because the regression parameters mea-
suring the effect of variants retain the same interpretation
when considering aggregate data (i.e., phenotype means
by genotype groups as with CRP studies) or IPD (see Mate-
rial and Methods). Moreover, where a robust evidence
based on genetic association with a quantitative trait al-
ready exists (as it does for many blood measures, e.g.,
HDL cholesterol, triglycerides, and others), the methods
described could be used to add and integrate partially over-
lapping SNP data from new genome-wide analyses,
thereby harnessing existing data for both replication, and
to gain insight into likely causal sites in a gene or region.
The methods we describe, which use the freely available
software WinBUGS, are likely to be of substantial value
both to the emerging networks of investigators engaged
in synthesis of evidence on genetic associations of com-
plex quantitative traits and disorders7 and to those apply-
ing and extending findings from genome-wide association
studies.
Appendix A
Systematic Review
Two electronic databases (PubMed Medline and EMBASE)
were searched with the text words, which were also
MeSH terms, polymorphism(s), mutation(s), gene(s), ge-
netic, variant(s), and SNP(s) in combination with C-reac-
tive protein and CRP. The literature search was limited to
human and to the English language. Any additional stud-
ies in the references of all identified publications were
also searched. For inclusion, studies had to have an analyt-
ical design (case control, prospective, or cross sectional)
and examine the association between any polymorphisms
in the CRP gene and low-chronic CRP concentrations in
individuals of European descent. Studies measuring CRP
only during acute phase of an inflammatory response (e.g.,
acute ischemia or infection stimuli) were excluded. In areas
where more than one polymorphism had been studied, in-
formation about the LD between them was extracted
where available. If relevant information was not reported
868 The American Journal of Human Genetics 82, 859-872, April 2008
Previous
-------
(mean CRP levels, standard deviations, genotype numbers,
or linkage disequilibrium data), or it was not reported strat-
ified by ethnicity, the authors were contacted in several oc-
casions to obtain the information. A total of four potential
studies (n = 2614) in European subjects were excluded be-
cause of unavailability of data in the appropriate form (Flex
2004, n = 471; Obisesan 2004, n = 63; Zee 2004, n = 260;
Carlson 2005, n = 1820) (see Table SI).
New Data Sets
NPHS II is a prospective study of 3012 healthy white Euro-
pean middle-aged men, of which a total of 2479 with CRP
genetic data and CRP concentrations were included in this
report. Recruitment in the study commenced in 198940
in nine general practices. None of the participants had
a clinical history of unstable angina, myocardial infarction
(including silent infarction), coronary surgery, other car-
diovascular diseases, aspirin or anticoagulant use, or malig-
nant disease (except skin cancer other than melanoma) at
the time of recruitment. The Ely Study is a prospective pop-
ulation-based cohort study of the etiology and pathogene-
sis of type 2 diabetes and related metabolic disorders in
1122 individuals recruited in 1990 in Ely, Cambridge-
shire.41 Complete data on biochemical and anthropomet-
ric variables were available in 839 participants, and a total
of 548 individuals with data on the CRP genotypes and
CRP levels were included in this analysis. The EPIC-Norfolk
study is a population-based cohort study, recruiting partic-
ipants from general practices in Norfolk.42 For the present
report, only control participants from a nested case-control
study in coronary heart disease were included, providing
a total of 2196 participants with both data on CRP genetic
variants and CRP concentrations.
New Genotyping
Polymorphisms in the human CRP gene (HGNC: 2637;
Iq21-q23) were identified by reference to public-domain
databases of human sequence variation. We used this in-
formation to generate a consensus map of polymorphic
sites. By using validated genotype data (minor allele fre-
quency >5%) from subjects of European descent from
the SeattleSNPs database and the human HapMap database
(see Web Resources), we examined the pattern of linkage
disequilibrium across the CRP gene. We then used the
haplotype LD r2 method to select a set of tagging (t)SNPs
capable of capturing maximum haplotype diversity among
subjects of European descent by using the program TagIT
(see Web Resources).
LD
Public domain databases (see Web Resources) and individ-
ual publications were examined for information on the
LD structure in the CRP gene. Both D' and r2 values were
recorded, but r2 values were utilized in Bayesian modeling.
If more than one r2 value for a given pairwise was reported,
a weighted mean r2 was obtained.
Appendix B
Recovering the Joint Distribution of Multiallelic Sites
from Allele Frequencies and Marginal Diallelic r2
Values
We consider two loci, the first locus having G\ — three
alleles and the second locus having G2 = two alleles. The
joint probability of the haplotypes at these two loci can
be represented in a 3 x 2 table of the form:
Table Bl. Full 3x2 Table of Haplotype Probabilities
Allele at Locus 2
Allolo at-
Locus
1
2
3
1 1
Pii
P2i
Ql - Pll - P21
Qi
2
Pi
P2
(1
(P2
(1
-Pii
-P2i
- Qi) - (Pi - pii) -
-P2l)
-Qi)
Pi
P2
(i-
i
Pi - P2)
ij denotes the joint probability of allele /atlocus 1 and
allele/ at locus 2, Pt denotes the probability of allele / at locus
1, and Qj denotes the probability of allele 1 at locus 2.
The internal cells of this table are not observed. Our prob-
lem is to derive this table of probabilities on the basis of infor-
mation from the margins of this table (Pi, P2 and QJ) and
pair-wise correlation within two marginal tables of the form:
Table B2. First Marginal 2x2 Haplotype Table
Allele at Locus 2
Allele at Locus 1 1
1
3
Pll
Q'l - P'II
Qi
2
Pi
(1
(1
-Pii
- Qi) - (Pi - Py
-Qi)
Pi
(i
i
-Pi)
and
Table B3. Second Marginal 2x2 Haplotype Table
Allele at Locus 2
Allele at Locus 1 1
2
3
P'2i
Q'i - P'k
Q'i
2
(P"2 - P'k)
(1 -
(1-
Q'i) -
Q'i)
(P"2
- P"2i)
P"2
(i -
i
P"2)
In the first of these tables, p'n denotes the joint probability
of allele 1 at locus 1 and allele 1 at locus 2, but now this
probability is conditional upon the allele at locus 1 having
either a 1 or 3 allele. Similarly, p2\ denotes the probability
of allele 2 at locus 1 and allele 1 at locus 2 conditional upon
the allele at locus 1 being either a 2 or a 3.
We do not observe the two tables above but only the ap-
propriate deviations from linkage disequilibrium, §i3 and
8'b, defined by
and
The American Journal of Human Genetics 82, 859-872, April 2008 869
Previous
TOC
-------
We know that the probability of each pair-wise haplo-
type in Table 2 is equal to the corresponding probability
of that pairwise haplotype in Table 1, divided by the prob-
ability that the allele at the first locus is either equal to 1 or
3, (1 - PZ>- This means that:
and
Therefore,
(Ql-.Pll-.P2l)
Pll
and
(Qi-p2
Pu =
Following a similar argument, we also find that
and
By writing p'\\ and p"2\ in terms of 8'13 and
that:
= (Pi*Ql-ai1)(l-P2)
Pi (Qi-foi)
-
and
we find
P21 = j
= (F2* 01'-^) (I-PI)
P2 (Qi-pn)
(I-PI) (I-PI)
(I-PI
This means that we have two equations in two un-
knowns, pn and p2i, so that by substituting the second
equation for p2l into the first equation for pllt we can
then solve this equation in terms of pn. Substituting the
expression for p^2 into that for pn gives:
1
and rearranging in terms of pn results in the equation:
We may then write p2\ in the form
We are now able to calculate the probability of every cell
of Table Bl in terms of pn, pi2, P\, PI, and Qx.
Note that 8'n and §21 can be obtained from the relevant
r2 values with the formulae:
<5' =
and
Care must be taken when choosing which sign to assign
these 8 values because they must be consistent with the
margins of the Tables B2 and B3.
Appendix C
WinBUGS Code for the Model Described in Material
and Methods
model {
# likelihood
for(j in 1:Q) {# where Q = ^s (ms + 1)
T[j] ~dnorm(theta[j],tauy[j])
tauylj] <- tau.y/XsXs[j] # XsXs[j] in Equation (4)
theta[j] <- psi[j]+sumXis[j]*mu[study[j]]+U[marker[j]] #
linear predictor in Equation (4)
# pooled variances
for(i in 1:L) {# where L — J]s ms
scalep] <— tau.y/2
shapep] <- pooled[i,2]/2
pooled[i,l] ~dgamma(shape[i],scale[i]) # uses the gamma
parameterization
#reversible jump part as detailed in Lunn et al.2s
psi[l:Q] <- jump.lin.pred(W[l:Q,l:m],K,tau.beta)
id <— jump.model.id(psi[l:Q])
pred[l:(m+l)] <— jump.lin.pred.pred(psi[l:Q],X.pred[l:
for(i in l:m){
X.pred[i,i] <- 1
for(j in l:(i-l)) {X.pred[i,j] <- 0}
for(j in (i+l):m) {X.pred[i,j] <- 0}
X.pred[(m+l),i] <- 0
effectp] <— pred[i] -pred[m+l]
}
# mixture distribution for study effects
for(s in l:nstudies) {
mu[s] ~dnorm(mumu[s],tau.mu)
870 The American Journal of Human Genetics 82, 859-872, April 2008
Previous
TOC
-------
mumu[s] <— alpha[comp[s]]
comp[s] ~dcat(phi[])
}
phi[2] - l-phi[l]
alpha[2] <- (-phi[l]*alp
# prior distributions
U[l:m] ~car.proper(thetaU[],M[],adj[],num[],m[],prec,l)
# thetaU vector of zeros of length m (number of unique
markers)
# M is weighted average of the XS/XS matrices
# details on vectors adj, num and m are given in the
manual for GeoBUGS
prec ~dgamma(0.5,0.0005)
tau.y ~dgamma(0.001,0.001)
tau.mu ~dgamma(0.001,0.001)
tau.beta ~dgamma(0.0001,0.0001)
phi[l] ~dbeta(l,l)
alpha[l] ~dnorm(0.0,1.0E-6)
K ~dpois(l) # scenario (a)
}
The MCMC chain was run for 1,000,000 iterations with
a burn-in of 500,000 and thinning of 100 iterations, which
took -30 min of CPU time on an Itel Xeon 2.80 GHz with 2
GB of RAM. Convergence was checked by visual inspection
of posterior traces and by running chains with different
initial values.36
Appendix D
Linkage Disequilibrium
4 5
0.0684
(0.0486)
r
0.0063
(0.0161)
0.0287
(0.0559)
0.0051
(0.0143)
-0.0021
(0.0264)
-0.1594
(0.271)
0.0100
(0.0137)
0.0090
(0.0142)
-0.0454
(0.1243)
0.4017
(0.2558)
0.0089
(0.0167)
0.0149
(0.0144)
-0.0266
(0.0935)
0.2812
(0.1842)
0.7078
(0.1256)
-0.0054
(0.0147)
0.0013
(0.0151)
-0.0312
(0.0816)
0.2396
(0.1567)
0.6098
(0.1423)
0.8610
(0.1762)
Figure Dl. Mean Pairwise LD Measures between Markers Used
in the Simulation Study when Allowing LD Patterns to Vary
across Studies
Supplemental Data
Three tables are available at http://www.ajhg.org/.
Acknowledgments
Dongliang Ge provided help with Figures 1 and 7. Tabular data
were kindly provided by Jos E Krieger, Per Tornvall, and Moniek
P. M. de Maat. This work was supported by Medical Research
Council Research Grant G0600580. T.S. was supported by the Brit-
ish Heart Foundation (PhD Studentship FS/02/086/14760), R.S.
was supported by a British Heart Foundation Shillingford Training
Fellowship (FS/07/011), S.E.H. was supported by a British Heart
Foundation Programme Grant (PG2000/015), A.D.H. was sup-
ported by a British Heart Foundation Senior Fellowship (FS/05/
125), and L.S. was supported by a Wellcome Trust Senior Clinical
fellowship (082178). A.D.H. acknowledges the generous support
of the Rosetrees Trust. J.C. acknowledges the support of the Well-
come Trust (GR076024). C.J.V. is supported by a Research Council
UK Fellowship. The EPIC-Norfolk study is supported by the Med-
ical Research Council UK, Cancer Research UK, and Stroke Asso-
ciation and Research Into Ageing. The Framingham Heart Study
is funded by N01-HC 25195, and Framingham inflammation
research is funded by HL076784, AG028321.
Received: September 25, 2007
Revised: November 29, 2007
Accepted: January 22, 2008
Published online: April 3, 2008
Web Resources
The URLs for data presented herein are as follows:
CRP:C-reactive protein, pentraxin-related, http://pga.gs.washington.
edu/data/crp/
HapMap homepage, http://www.hapmap.org/
Online Mendelian Inheritance in Man (OMIM), http://www.ncbi.
nlm.nih.gov/Omim
TagIT, http://popgen.biol.ucl.ac.uk/software.html
WinBUGS software, http://www.mrc-bsu.cam.ac.uk/bugs/winbugs/
con tents, shtml
References
1. Cardon, L., and Bell, J. (2001). Association study designs for
complex diseases. Nat. Rev. Genet. 2, 91-99.
2. Colhoun, H., McKeigue, P., and Davey Smith, G. (2003).
Problems of reporting genetic associations with complex
outcomes. Lancet 361, 865-872.
3. Clayton, D., and McKeigue, P. (2001). Epidemiological
methods for studying genes and environmental factors in
complex diseases. Lancet 358, 1356-1360.
4. Zeggini, E., Rayner, W., Morris, A.P., Hattersley, A.T., Walker,
M., Hitman, G.A., Deloukas, P., Cardon, L.R., and McCarthy,
M.I. (2005). An evaluation of hapmap sample size and tagging
snp performance in large-scale empirical and simulated data
sets. Nat. Genet. 37, 1320-1322.
5. Cambon-Thomsen, A. (2003). Assessing the impact of
biobanks. Nat. Genet. 34, 25-26.
6. Little, J., Bradley, L., Bray, M., Clyne, M., Dorman, J., Ells-
worth, D., Hanson, J., Khoury, M., Lau, J., O'Brien, T, et al.
(2002). Reporting, appraising, and integrating data on geno-
type prevalence and gene-disease associations. Am. J. Epide-
miol. 156, 300-310.
The American Journal of Human Genetics 82, 859-872, April 2008 871
Previous
-------
7. loannidis, J.P., Gwinn, M., Little, J., Higgins, J.P., Bernstein,
J.L., Boffetta, P., Bondy, M., Bray, M.S., Brenchley, P.E., Buffler,
P.A., et al. (2006). A road map for efficient and reliable human
genome epidemiology. Nat. Genet. 38, 3-5.
8. Seminara, D., Khoury, M., O'Brien, T., Manolio, T., Gwinn, M.,
Little,]., Higgins,]., Bernstein,]., Boffetta, P., Bondy, M., et al.
(2007). The emergence of networks in human genome epide-
miology: Challenges and opportunities. Epidemiology 18,1-8.
9. Scott, L., Mohlke, K., Bonnycastle, L.,Wilier, C., Li,Y., Duren,W.,
Erdos, M., Stringham, H., Chines, P., Jackson, A., et al. (2007).
A genome-wide association study of type 2 diabetes in finns
detects multiple susceptibility variants. Science 316, 1341-
1345.
10. Dina, C., Meyre, D., Gallina, S., Durand, E., Krner, A., Jacob-
son, P., Carlsson, L., Kiess, W., Vatin, V., Lecoeur, C., et al.
(2007). Variation in fto contributes to childhood obesity and
severe adult obesity. Nat. Genet. 39, 724-726.
11. The Wellcome Trust Case - Control Consortium (2007). Ge-
nome - wide association study of 14,000 cases of seven com-
mon diseases and 3,000 shared controls. Nature 447, 661-678.
12. Besag, J., and Kooperberg, C.L. (1995). On conditional and
intrinsic autoregressions. Biometrika 82, 733-746.
13. Lawson, A.B. (2001). Statistical Methods in Spatial Epidemiol-
ogy (Chichester, UK: John Wiley).
14. Hirschfield, G., and Pepys, M. (2003). C-reactive protein and
cardiovascular disease: new insights from an old molecule.
OJM 96, 793-807.
15. Danesh, J., Wheeler, J., Hirschfield, G., Eda, S., Eiriksdottir, G.,
Rumley, A., Lowe, G., Pepys, M., and Gudnason, V. (2004). C-
reactive protein and other circulating markers of inflamma-
tion in the prediction of coronary heart disease. N. Engl. J.
Med. 350, 1387-1397.
16. Hingorani, A., and Humphries, S. (2005). Nature's randomised
trials. Lancet 366, 1906-1908.
17. Hingorani, A., Shah, T., and Casas, J. (2006). Linking observa-
tional and genetic approaches to determine the role of c-reac-
tive protein in heart disease risk. Eur. Heart J. 27, 1261-1263.
18. DaveySmith, G., andEbrahim, S. (2003). Mendelianrandomiza-
tion: Can genetic epidemiology contribute to understanding en-
vironmental determinants of disease? Int. J. Epidemiol. 32,1-22.
19. Davey Smith, G., and Ebrahim, S. (2004). Mendelian random-
ization: Prospects, potentials and limitations. Int. J. Epide-
miol. 33, 30-42.
20. MacGregor, A., Gallimore, J., Spector, T, and Pepys, M. (2004).
Genetic effects on baseline values of c-reactive protein and
serum amyloid a protein: A comparison of monozygotic
and dizygotic twins. Clin. Chem. 50, 130-134.
21. Carlson, C., Aldred, S., Lee, P., Tracy, R., Schwartz, S., Rieder,
M., Liu, K., Williams, O., Iribarren, C., Lewis, E., et al.
(2005). Polymorphisms within the c-reactive protein (crp)
promoter region are associated with plasma crp levels. Am. J.
Hum. Genet. 77, 64-77.
22. Kardys, L, de Maat, M., Uitterlinden, A., Hofman, A., and Wit-
teman, J. (2006). C-reactive protein gene haplotypes and risk
of coronary heart disease: The rotterdam study. Eur. Heart J.
27, 1331-1337.
23. Marchini, J., Howie, B., Myers, S., McVean, G., and Donnelly, P.
(2007). A new multipoint method for genome-wide association
studies by imputation of genotypes. Nat. Genet. 39, 906-913.
24. Green, P. (1995). Reversible jump markov chain monte carlo
computation and bayesian model determination. Biometrika
82, 711-732.
25. Lunn, D., Whittaker, J., and Best, N. (2006). A bayesian toolkit
for genetic association studies. Genet. Epidemiol. 30,
231-247.
26. Thomas, D. (2004). Statistical Methods in Genetic Epidemiol-
ogy (New York: Oxford University Press).
27. Besag, J., York, J., and Mollie, A. (1991). Bayesian image
restoration, with two applications in spatial statistics (with
discussion). Ann. Inst. Stat. Math. 43, 1-59.
28. Dominici, R, Parmigiani, G., Wolpert, R.L., and Hasselblad, V.
(1999). Meta-analysis of migraine headache treatments:
Combining information from heterogeneous designs. J. Am.
Stat. Assoc. 94, 16-28.
29. Kelsall, J., and Wakefield, J.C. (1999). Comment on Bayesian
models for spatially correlated disease and exposure data. In
Bayesian Statistics 6 Proceedings of the Sixth Valencia Interna-
tional Meeting (Oxford: Clarendon Press).
30. Denison, D.G.T., Holmes, C.C., Mallick, B.K., and Smith,
A.EM. (2002). Bayesian Methods for Nonlinear Classification
and Regression (Chichester, UK: John Wiley).
31. Verzilli, C.J., Whittaker, J.C., and Stallard, N. (2005). A hierar-
chical bayesian model for predicting the functional conse-
quences of amino acid polymorphisms. J. R. Stat. Soc. Ser. C
Appl. Stat. 54, 191-207.
32. Higgins, J., Thompson, S., Deeks, J., and Altman, D. (2003).
Measuring inconsistency in meta-analyses. BMJ 327, 557-
560.
33. DuMouchel, WH. (1990). Bayesian meta-analysis. In Statisti-
cal Methodology in the Pharmaceutical Sciences (New York:
Marcel Dekker).
34. Normand, S.L.T. (1999). Meta-analysis: Formulating, evaluat-
ing, combining and reporting. Stat. Med. 18, 321-359.
35. O'Hagan, A., and Forster, J. (2004). Bayesian Inference. In Ken-
dall's Advance Theory of Statistics, Vol 2B (London: Arnold).
36. Carlin, B.P., and Louis, T.A. (2002). Bayes and Empirical Bayes
Methods for Data Analysis (Boca Raton: Chapman & Hall).
37. Parent, O., and Riou, S. (2005). Bayesian analysis of know-
ledge spillovers in european regions. J. Reg. Sci. 45, 747-775.
38. loannidis, J., and Kawoura, F. (2006). Concordance of func-
tional in vitro data and epidemiological associations in com-
plex disease genetics. Genet. Med. 8, 583-593.
39. Cirulli, E., and Goldstein, D. (2007). In vitro assays fail to
predict in vivo effects of regulatory polymorphisms. Hum.
Mol. Genet. 16, 1931-1939.
40. Herbert, A., Lenburg, M., Ulrich, D., Gerry, N., Schlauch, K.,
and Christman, M. (2007). Open-access database of candidate
associations from a genome-wide snp scan of the framingham
heart study. Nat. Genet. 39, 135-136.
41. Wareham, N., Hennings, S., Prentice, A., and Day, N. (1997).
Feasibility of heart-rate monitoring to estimate total level
and pattern of energy expenditure in a population-based
epidemiological study: The ely young cohort feasibility study.
Br. J. Nutr. 78, 889-900.
42. Day, N., Oakes, S., Luben, R., Khaw, K., Bingham, S., Welch, A.,
and Wareham, N. (1999). Epic-norfolk: Study design and char-
acteristics of the cohort, european prospective investigation of
cancer. Br. J. Cancer 80, 95-103.
872 The American Journal of Human Genetics 82, 859-872, April 2008
Prtvfoys
TOG
-------
Regulatory Toxicology and Pharmacology 51 (2008) S27-S36
Contents lists available at ScienceDirect
Regulatory Toxicology and Pharmacology
journal homepage: www.elsevier.com/locate/yrtph
Biomonitoring Equivalents (BE) dossier for toluene (CAS No. 108-88-3)
Lesa L Aylward3, Hugh A. Barton b, Sean M. Haysc'*
a Summit Toxicology, LIP, 6343 Carolyn Drive, Falls Church, VA 22044, USA
bNational Center for Computational Toxicology U.S. EPA, B205-01, 109 TW Alexander Drive, Research Triangle Park, NC 27711, USA
c Summit Toxicology, LLP, 165 Valley Road, Lyons, CO 80540, USA
ARTICLE INFO
Article history:
Received 15 January 2008
Available online 22 May 2008
Keywords:
Biomonitoring
Biomonitoring Equivalents
BEs
Toluene
Pharmacokinetics
ABSTRACT
Recent efforts by the US Centers for Disease Control and Prevention and other researchers have resulted
in a growing database of measured concentrations of chemical substances in blood or urine samples
taken from the general population. However, few tools exist to assist in the interpretation of the mea-
sured values in a health risk context. Biomonitoring Equivalents (BEs) are defined as the concentration
or range of concentrations of a chemical or its metabolite in a biological medium (blood, urine, or other
medium) that is consistent with an existing health-based exposure guideline. This document reviews
available pharmacokinetic data and models for toluene and applies these data and models to existing
health-based exposure guidance values from the US Environmental Protection Agency, the Agency for
Toxic Substances and Disease Registry, Health Canada, and the World Health Organization, to estimate
corresponding BE values for toluene in blood. These values can be used as screening tools for evaluation
of biomonitoring data for toluene in the context of existing risk assessments for toluene and for prioriti-
zation of the potential need for additional risk assessment efforts for toluene.
© 2008 Elsevier Inc. All rights reserved.
1. Introduction
Measurements of environmental chemicals in air, water, or
other media can be compared to health-based exposure guidelines
to identify chemical exposures that may be of concern, or to iden-
tify chemicals for which a wide margin of safety appears to be
present. Interpretation of human biomonitoring data for environ-
mental compounds is hampered by a lack of similar screening cri-
teria applicable to measurements of chemicals in biological media
such as blood or urine. Such screening criteria would ideally be
based upon data from robust epidemiological studies that evaluate
a comprehensive set of health endpoints in relationship to mea-
sured levels of chemicals in biological media. However, develop-
ment of such epidemiologically based screening values is a
resource- and time-intensive effort, and appropriate data for such
values may never be available for many compounds. As an interim
effort, the development of Biomonitoring Equivalents (BEs) has
been proposed (Hays et al., 2007).
A Biomonitoring Equivalent (BE) is defined as the concentration
or range of concentrations of chemical in a biological medium
(blood, urine, or other medium) that is consistent with an existing
health-based exposure guideline. Existing chemical-specific phar-
macokinetic data are used to estimate biomarker concentrations
associated with the Point of Departure (PODs; such as No Observed
* Corresponding author.
E-mail address: shays@summittoxicology.com (S.M. Hays).
0273-2300/S - see front matter © 2008 Elsevier Inc. All rights reserved.
doi: 10.1016/j.yrtph.2008.05.009
Effect Levels [NOELs], Lowest Observed Effect Levels [LOELs], or
Benchmark Doses [BMDs]) used as the basis for the exposure guid-
ance value and to estimate biomarker concentrations that are con-
sistent with the guidance value. BEs can be estimated using
available human or animal pharmacokinetic data. Guidelines for
the derivation and communication of BEs are available (Hays
et al., 2008; LaKind et al., 2008). BEs are designed to be screening
tools to gauge which chemicals have large, small or no margin of
safety compared to existing health-based exposure guidelines,
and are designed to provide a basis for prioritization of chemicals
for risk assessment follow-up. BEs are only as robust as the under-
lying health-based exposure guidelines that they are based upon
and the underlying animal and/or human pharmacokinetic data
used to derive the BEs. BEs are not designed to be diagnostic for
potential health effects in humans, either individually or among
a population.
Toluene is used as a solvent in numerous products including
industrial paints, adhesives, coatings, inks, and cleaning products.
Toluene is also added to aviation fuel to improve octane ratings
and as a raw material for the manufacture of polymers used to
make nylon, plastic soda bottles, and polyurethanes. It is also used
in processes for manufacture of Pharmaceuticals, dyes, cosmetic
nail products, and in the synthesis of organic chemicals including
benzene. According to the US Environmental Protection Agency
(USEPA), the primary pathway for exposure to toluene is inhalation
from ambient and indoor air, although ingestion may also occur
through trace amounts of toluene that may occur in food or water.
Previous
-------
S28
LI. Ay/word et al/Regulatory Toxicology and Pharmacology 51 (2008,) S27-S36
Intentional inhalant abuse can result in high exposure to toluene
vapors. Additional general information regarding toluene can be
found at http://www.epa.gov/ttn/atw/hlthef/toluene.html.
This dossier describes the scientific basis for and derivation of
BE values for toluene and discusses issues that are important for
the interpretation of biomonitoring data using biomonitoring
equivalents. This BE dossier is not designed to be a comprehensive
compilation of the available hazard, dose-response or risk assess-
ment information for toluene.
1.1. Current health-based exposure guidance values
The primary effects of toluene in both humans and animals after
either acute or chronic exposure are effects on the central nervous
system (CNS). Acute exposure to high concentrations of toluene
causes symptoms including fatigue, sleepiness, headaches, and
nausea. Chronic exposure to toluene at levels above current occu-
pational exposure guidelines has been associated with subtle
changes in sensory function including reduced color vision (re-
viewed in USEPA, 2005). High level exposures through intentional
inhalant abuse or accidental or intentional ingestion of toluene
have also been reported to cause effects on the liver, kidneys,
and other organ systems. With respect to carcinogenicity, both
the International Agency for Research on Cancer (IARC) and the
US Environmental Protection Agency (USEPA) consider toluene as
"not classifiable" as to human carcinogenicity (Groups 3 and D,
respectively) (IARC, 1999; USEPA, 2005).
Health-based exposure guidelines and toxicity values have been
established for many chemicals for the general population by the
USEPA (Reference Doses or Reference Concentrations [RfDs or
RfCs]), the Agency for Toxic Substances and Disease Registry (ATS-
DR) (Minimal Risk Levels or MRLs), and Health Canada and the
World Health Organization (WHO) (Tolerable Daily Intakes or
TDIs). Although these health-based exposure guidance values have
different labels and slightly different definitions, they all generally
describe an approximation of daily intake rates (or air concentra-
tions) for a chemical expected to be without adverse effects in
the general population, including sensitive subpopulations.1. For
chemicals considered to be carcinogenic, the USEPA also establishes
estimates of cancer potency by assigning a quantitative estimate of
the upper bound of potential increased cancer risk associated with
a unit of intake or air concentration (unit cancer risks, or UCRs). Fi-
nally, several organizations set chemical-specific air concentrations
that are considered to be safe for workers in the occupational envi-
ronment (for example, Threshold Limit Values [TLVs], Permissible
Exposure Limits [PELs], and Maximum Air Concentrations [MAKs]).
These values are generally not appropriate for application to the gen-
eral population on a chronic basis, but can provide perspective for
evaluating non-workplace environmental exposures.
Several health-based exposure guidelines and toxicity values
are available for toluene including guidelines for both inhalation
and oral exposures. These values are summarized in Table 1. As
discussed above, toluene is generally not considered to be carcino-
genic so no cancer potency estimates for toluene are available. In
addition, biological monitoring values for toluene in blood or tolu-
ene metabolites in urine in occupationally exposed individuals
(Biological Exposure Indices (BEIs) and Biological Tolerance Values
(BATs)) have also been established (ACGIH, 2001; Angerer et al.,
1998). As discussed above, these are not appropriate for applica-
tion to the general population.
1 See the definition of RfD at http://www.epa.gov/NCEA/iris/help_gloss.htm#r;
definitions for ATSDR MRLs are included in ATSDR Toxicological Profiles at http://
www.atsdr.cdc.gov/toxpro2.html. Definition of the TDI is available at http://
ptcl.chem.ox.ac.uk/MSDS/glossary/tolerable_daily_intake.html.
1.2. Pharmacokinetics
The pharmacokinetics of toluene have been studied extensively
in human volunteers and persons occupationally exposed as well
as in laboratory animals. Toluene is well absorbed following inha-
lation and oral exposure. Toluene undergoes metabolism, princi-
pally via CYP2E1, and metabolites are excreted in urine. Toluene
is also eliminated as parent compound in urine and exhaled air.
The recent USEPA IRIS review of toluene includes a detailed
description of the metabolic pathways for toluene (USEPA, 2005).
Detailed physiologically based pharmacokinetic (PBPK) models
for toluene in humans and laboratory rats have been developed
by several groups of researchers and can accurately predict blood
levels associated with a variety of inhalation exposure regimens
(human and rat models by Tardif et al., 1993,1995; human models
by Jang, 1996; Pierce et al., 1998).
1.3. Biomarkers
The objective of using BEs is to provide a human health risk
framework for screening-level evaluation of human biomonitoring
data. The choice of the biomarker (analyte and medium) should be
optimized to facilitate this objective. The key criterion for the
choice of a biomarker is that it be as closely related to the appro-
priate dose to the target tissue as possible and that it be practical
for collection in a biomonitoring study. This, in turn, means that
the biomarker should be (i) the compound that causes the toxicity
(parent or metabolite), or (ii) should be just upstream on the met-
abolic pathway from the toxic compound, and (iii) as closely re-
lated to the target tissue as possible.
Several potential biomarkers are available for assessing inter-
nal exposure to toluene (Table 2). Toluene is excreted unchanged
in exhaled air. However, as a quantitative biomarker, toluene in
exhaled breath is relatively insensitive and it is difficult to ob-
tain reliable, reproducible measurements. In the occupational
setting, exposure to toluene has been monitored through mea-
surement of metabolites in urine and parent compound in blood
(ACGIH, 2001). However, use of urinary metabolites of toluene as
markers for assessing exposure in persons in the general popula-
tion is of limited utility because neither marker measured, hip-
puric acid or ortho-cresol, is specific to toluene exposure.
Instead, each can be observed as metabolites of numerous parent
compounds (Dossing et al., 1983). Hippuric acid levels in urine
are relatively poorly correlated with exposure even under occu-
pational exposure conditions (Truchon et al., 1999). Under condi-
tions of higher occupational exposure levels, elevated ortho-
cresol levels are closely correlated with inhalation exposures,
but at environmental exposure concentrations ortho-cresol levels
are non-specific. For instance, ortho-cresol is present in cigarette
smoke. Thus, it cannot serve as a specific marker for toluene
exposure at low environmental levels (Dossing et al., 1983).
Two other urinary metabolites, S-p-toluylmercapturic acid and
S-benzylmercapturic acid, could potentially serve as specific
markers for toluene exposure (Angerer et al., 1998; Inoue
et al., 2004). However, current analytical techniques are probably
not sensitive enough to quantitate concentrations following envi-
ronmental exposures, and insufficient data on quantitative rela-
tionships between these metabolites and toluene exposure or
blood levels are currently available. Finally, unchanged toluene
in urine has also been proposed as a biological marker for expo-
sure to toluene in the occupational setting (Fustinoni et al.,
2007; Kawai et al., 2008). There is relatively little literature
relating toluene in urine to external exposures, and none of
the current models for toluene pharmacokinetics explicitly
include this pathway to allow quantitative prediction of elimina-
tion in urine under differing exposure conditions. Thus, although
Previous
-------
LI. Ay/ward et al./Regulatory Toxicology and Pharmacology 51 (2008) S27-S36
S29
Table 1
Health-based exposure guidance values for toluene from various agencies
Organization, criterion, and year of
evaluation
Study description
Critical endpoint and dose
Uncertainty factors
Value
Inhalation exposure guidelines
USEPA RfC (USEPA, 2005)
Health Canada TDI-inhalation
(Health Canada, 1996)
World Health Organization Air
Quality Guideline (WHO, 2005)
Multiple studies of human
occupationally exposed
populations
Study of human
occupationally exposed
populations
Occupationally exposed
workers with mean
exposure at 332 mg/m3
Transient and persistent neurological effects
NOAEL (average): 34 ppm (128 mg/m3)
NOAEL (adjusted for 24 h, 7 day/week
exposure): 46 mg/m3
Nervous system effects in humans
NOAEL: 150 mg/m3
NOAEL (adjusted by a factor of 6/24 for 24 h,
7 day/week exposure): 38 mg/m3
Impacts on neurological performance tests
LOAEL of 332 mg/m3, duration adjusted to
80 mg/m3
10- total
10— interindividual
10- total
10— interindividual
300-total
10— interindividual
10-use of a LOAEL
5 mg/m3
(1.3 ppm)
3.8 mg/m3
(1 ppm)
0.26 mg/m3
(0.07 ppm)
ATSDR acute inhalation MRL (1-14 Human volunteers,
days exposure) (ATSDR, 2000)
ATSDR chronic inhalation MRL (>1
year) (ATSDR, 2000)
Oral exposure guidelines
USEPA RfD (USEPA, 2005)
Health Canada TDI-oral (Health
Canada, 1996)
World Health Organization TDI
(WHO, 2004)
ATSDR acute oral MRL (1-14 days
exposure) (ATSDR, 2000)
ATSDR intermediate oral MRL (up to
1 year) (ATSDR, 2000)
exposure to 10, 40, or
100 ppm, 6 h per day for 4
days
Occupationally exposed
workers with mean
exposure at 35 ppm
Rat gavage, 13 weeks (NTP,
1990)
Rat gavage, 13 weeks (NTP,
1990)
Mouse gavage, 13 weeks
(NTP, 1990)
Rats exposed by single dose
corn oil gavage to 0, 250,
500, or 1000 mg/kg day
Mice exposed by drinking
water for 28 days to 5,22, or
105 mg/kg day
Trend of decreased neurological performance,
with NOAEL at 40 ppm. Duration adjusted
Decreased color vision after adjustment for age
and alcohol use, with LOAEL at 35 ppm.
Duration adjusted
Kidney weight changes as a precursor to
kidney toxicity at higher doses
NOAEL: 223 mg/kg day
LOAEL: 446 mg/kg day (duration adjustment
applied)
BMDL: 228 mg/kg day
Liver and kidney weight changes
NOAEL: 223 mg/kg day
LOAEL: 446 mg/kg day (duration adjustment
applied)
Marginal hepatotoxic effects at the lowest dose
Decreased flash-evoked potential at all dose
levels (no dose-response trend)
Changes in neurotransmitter levels
3—potential sensitivity
of developing CNS
10-total
10—interindividual
3 mg/m3
(0.8 ppm)
100-total 0.3 mg/m3
10-minimal LOAEL to NOAEL (0.08 ppm)
10—interindividual
3000-total
10—interspecies
10—interindividual
10—subchronic to chronic
3—database uncertainties
1000-total
10—interspecies
10—interindividual
10—subchronic to chronic
1000-total
10—interspecies
10—interindividual
10—subchronic to chronic
and use of a LOAEL
300-total
3-minimal LOAEL to NOAEL
10—interspecies
10—interindividual
300-total
3-minimal LOAEL to NOAEL
10—interspecies
10—interindividual
0.08 mg/kg day
0.22 mg/kg day
0.2 mg/kg day
0.8 mg/kg day
0.02 mg/kg day
toluene in urine may prove useful as a specific biomarker for
environmental exposure to toluene, the current data are not
sufficient to rely on toluene in urine for the Biomonitoring
Equivalent process. Thus, no urinary biomarker for exposure to
toluene is currently useful for assessing general environmental
exposures.
Toluene has also been measured in blood and correlated with
inhalation exposure levels in persons exposed occupationally
(see, for example, Neubert et al., 2001), in volunteers under condi-
tions of controlled exposure (see, for example, Pierce et al., 1998),
and in the general population (see, for example, Sexton et al.,
2005). Toluene in blood has also been identified as a useful bio-
marker in the occupational setting (ACGIH, 2001).
Identification of relevant dose metrics depends upon the health
endpoints that are the bases of the health-based screening values.
The available health-based criteria presented in Table 1 focus on
two health endpoints. The USEPA oral RfD is based on subtle
kidney toxicity following oral gavage dosing in rats, while the
ATSDR acute and intermediate MRL values are based on changes
in neurological endpoints in rats and mice. Inhalation criteria from
all agencies are based on subtle neurological effects observed in
humans after acute and chronic exposure to toluene.
The mechanisms of the renal toxicity observed in rats following
subchronic oral gavage in the National Toxicology Program study
are unknown, but recent in vitro studies by Al-Ghamdi et al.
(2003a,b) in proximal tubule cell cultures suggest that the toxicity
may be attributable to benzyl alcohol, a toluene metabolite pro-
duced via CYP2E1. Al-Ghamdi et al. (2003a,b) showed that inhibit-
ing CYP2E1 activity prevented toxicity in cell culture following
toluene exposure. Renal toxicity has also been observed in humans
following intentional or accidental ingestion of large amounts of
toluene and following chronic inhalation abuse (Stengel et al.,
1998). However, such toxicity has not been reported in occupa-
tional populations exposed to more moderate air concentrations.
For example, Stengel et al. (1998) reported a no-observed-ad-
verse-effect-level (NOAEL) at the TLV of 50 ppm (188 mg/m3) for
renal function changes in a chronically exposed occupational co-
hort. Renal toxicity is most likely a phenomenon associated with
Previous
-------
S30
LL Ay/ward et al/Regulatory Toxicology and Pharmacology 51 ("2008) S27-S36
Table 2
Potential biomarkers of exposure to toluene
Analyte
Medium
Advantages
Disadvantages
Toluene
Hippuric acid
ort/io-Cresol
S-p-Toluylmercapturic
acid
S-Benzylmercapturic
acid
Blood Sensitive and specific; highly relevant to target
tissue concentrations
Urine Sample easily obtained; specific biomarker
Exhaled air Sample easily obtained; specific biomarker
Urine Sample easily obtained
Urine Sample easily obtained
Urine Sample easily obtained; specific biomarker
Urine Sample easily obtained; specific biomarker
Requires blood draw
Lack of robust data set or model to quantify relationship between
exposure and observed levels; not directly relevant to target tissue
concentrations
Insensitive, difficult to obtain reproducible results
Non-specific metabolite; not directly relevant to target tissue
concentrations
Non-specific metabolite at environmental exposure levels; not directly
relevant to target tissue concentrations
Lack of robust data set or model to quantify relationship between
exposure and observed levels; not directly relevant to target tissue
concentrations; analytical sensitivity may not be sufficient
Lack of robust data set or model to quantify relationship between
exposure and observed levels; not directly relevant to target tissue
concentrations; analytical sensitivity may not be sufficient
high peak toluene blood levels resulting in high rates of metabo-
lism and subsequent activity of metabolites in the kidney (Al-Gha-
mdi et al., 2003a,b).
Neurological responses following inhalation exposure to tolu-
ene in humans or oral exposure in rats and mice are likely to be re-
lated directly to brain concentrations of toluene, which in turn are
directly related to blood concentrations (Benignus et al., 2007;
Bushnell et al., 2007). The RfC for toluene derived by USEPA is
based on evaluation of potential neurotoxicity, which appears to
be the most sensitive endpoint identified in numerous studies of
long-term occupationally exposed populations (see USEPA, 2005,
for a complete description of these studies and populations). These
studies are characterized by long-term exposure with monitored
air exposure concentrations. The mechanisms underlying the ob-
served neurotoxicity are not fully understood, but appear to be re-
lated to concentrations of the parent compound (rather than
metabolites) reaching the brain (van Asperen et al., 2003; Benignus
et al., 2007; Bushnell et al., 2007). However, there are insufficient
data to conclusively identify whether peak or average toluene con-
centration in blood is the most appropriate dose metric for various
neurological responses. The direct correlation between toluene
blood concentration and neurological responses supports use of
blood concentration of toluene as a biomarker, and under chronic
exposure conditions, average blood concentration should be di-
rectly relevant.
2. BE derivation
2.1. Methods
2.1.1. Urine
As discussed above, data do not support the use of urinary
markers for toluene exposure at environmental exposure levels
at this time, although, as discussed above, selected specific metab-
olites or unchanged toluene in urine might serve as reliable bio-
markers if more data can be developed and analytical techniques
for those markers become sufficiently sensitive. No urinary BE val-
ues were derived for toluene exposure.
2.1.2. Blood
In order to estimate human blood levels associated with expo-
sure to toluene at the various health-based inhalation and oral
exposure guidelines detailed in Table 1, the human PBPK model
for toluene developed by Tardif et al. (1993, 1995) was imple-
mented. Models by Pierce et al. (1996,1998) and Jang (1996) were
also available. Each is similar to the model developed by Tardif
et al. (1995), but of the three, the Tardif et al. (1995) model has
been used the most extensively and was therefore chosen for use
in this BE derivation. The rat PBPK model by Tardif et al. (1993)
was also used to estimate blood concentrations in rats at the dose
levels used as the point of departure for derivation of the oral RfD
and oral MRL values. These blood concentrations at the point of
departure for risk assessment can provide additional context for
interpretation of measured blood concentrations in humans in
the general population.
Both the human and the rat PBPK models required minor mod-
ifications to incorporate the oral route of exposure. These additions
are described below.
2.1.2.1. Rat model. The PBPK model of Tardif et al. (1993) for toluene
inhalation exposure in rats was implemented in MS Excel® with
physiological and physicochemical parameters as described in Ta-
bles 1 and 2 of that publication. The model was modified to incorpo-
rate oral dosing by adding a virtual gastrointestinal tract
compartment with a first-order absorption process to the liver. To
parameterize the absorption rate from oral dosing, the gastrointesti-
nal absorption rate was calibrated visually against graphical data
from Sullivan and Conolly (1988) for the time course of blood tolu-
ene concentrations following oral gavage at four different dose levels
in Sprague-Dawley rats. The absorption rate from the rat gastroin-
testinal tract was adjusted to result in peak blood concentrations be-
tween 2 and 2.5 h post-gavage, as reflected in the Sullivan and
Conolly (1988) data set. All other parameters were retained as re-
ported by Tardif et al. (1993). The parameters used in the rat oral
and inhalation toluene PBPK model are presented in Table 3.
2.1.2.2. Adult human model. The human PBPK model of Tardif et al.
(1995) with parameters as reported by Nong et al. (2006, Table 1)
for toluene inhalation exposure was similarly implemented in MS
Excel®. The model was able to accurately reproduce the central
tendency of the measured blood and exhaled air concentrations
in an independent data set for volunteers exposed to 50 ppm tolu-
ene for 2 h from Pierce et al. (1998; results not shown).
An oral dose route was also added to the human PBPK model.
Addition of this dose route required parameterization of an oral
absorption rate constant. An oral absorption rate was calibrated
against the time course to peak exhaled air concentrations follow-
ing administration of toluene at measured drip rates for specified
time periods to human volunteers via nasal-gastric tube (Baelum
et al., 1993). The exhaled air concentrations peaked approximately
15-30 min following cessation of exposure. The full set of model
parameters for the adult human model is included in Table 3.
Previous
-------
LL Ay/ward et al/Regulatory Toxicology and Pharmacology 51 (2008,) S27-S36
S31
Table 3
Model parameters used in the rat and human
Parameter
Physiological parameters
Body weight (kg)
Tissue volumes (L)
Liver
Fat
Richly perfused
Poorly perfused
Cardiac output (L/h)
Alveolar ventilation (L/h)
GI Tract emptying ratec (tr1)
Tissue blood flow rates (L/h)
Liver
Fat
Richly perfused
Poorly perfused
Partition coefficients
Blood:air
Livenblood
Fat:blood
Richly perfused:blood
Poorly perfused :blood
Metabolic constants
Vmax (mg/h)
Km (mg/L)
a From Tardif et al. (1993).
b From Tardif et al. (1995) and Nong et al.
c Fit to data sets as described in text.
PBPK models
Adultb
70
1.82
13.3
3.5
43.4
418
418
0.69
109
21
184
104
15.6
2.98
65.8
2.66
1.37
116.2
0.55
(2006).
Rata
0.25
0.0123
0.0225
0.0125
0.18
5.3
5.3
0.23
1.33
0.48
2.70
0.80
18
4.64
56.7
4.64
1.54
1.7
0.55
External
Doss
Relevant
Internal
Dose
Monitored
Biomarker
2A.2.3. Evaluation of BEs for oral exposure guidelines. The general
approach for the derivation of BE values for the oral exposure
guidelines is presented in Figs. 1 and 2.
For those exposure guidance values derived based on rat toxicity
study data, the rat and human PBPK models were used in combina-
tion to derive BE values. Briefly, the process (Fig. 1) is as follows:
• Step 1: Calculate relevant animal internal dose at POD. In this case,
effects on liver and/or kidney following chronic gavage adminis-
tration of toluene are the most sensitive effects and serve as the
basis for the derivation of oral exposure guidelines. These effects
are likely related to production of metabolites in these organs.
Monitored
Biomarker
Animal
Human
Avg. blood
cone.
X Q
u. u_
13 Z>
Fig. 1. Schematic of approach used to estimate BE values for toluene in humans
corresponding to oral exposure guidance values based on rat toxicity data. NOAELatij
POD: Point of departure, adjusted for duration and LOAEL to NOAEL, as appropriate;
UFA_PD: component of interspecies uncertainty factor for pharmacodynamic sensi-
tivity; UFA-PIO component of interspecies uncertainty factor for pharmacokinetic
sensitivity; UFn: intraspecies uncertainty factor; UFo: uncertainty factor component
for database uncertainties, where applicable. See text for discussion.
Animal
Human
Human average blood concentration
Fig. 2. Schematic of approach used to estimate BE values for toluene in humans
corresponding to oral exposure guidance values based on mouse toxicity data.
UFH-PD: component of intraspecies uncertainty factor for pharmacodynamic
sensitivity.
Production of these metabolites is likely to be proportional to
area under the curve of toluene in these organs. Thus, toluene
area under the curve for kidney (modeled as richly perfused tis-
sue) or liver (modeled explicitly) was selected as the relevant
internal dose metric, and an estimate of the target organ AUC
at the duration- and LOAEL-to-NOAEL adjusted POD was made
for each of the oral exposure guidelines. The average blood con-
centration in the animals at the POD (BEPOD_animai) was also esti-
mated using the PBPK model.
• Step 2: Interspecies extrapolation. Interspecies extrapolation of
this relevant internal dose metric to a corresponding human tar-
get organ AUC by application of an interspecies uncertainty factor
for pharmacodynamic differences. An interspecies factor for phar-
macokinetic differences was also applied. This factor accounts for
unknown differences between humans and the experimental ani-
mals of interest in the pharmacokinetics of the metabolites
believed to be responsible for the organ-specific toxicity.
• Step 3: Calculate BEPOD. Application of the human pharmacoki-
netic model to identify an average blood concentration corre-
sponding to the relevant target organ internal dose measure
identified above (human equivalent BEPOD)-
• Step 4: Calculate BE. Application of relevant intraspecies uncer-
tainty factor(s) and any additional applicable uncertainty factors
identified by the organizations that derived the oral exposure
guidelines initially (for example, database uncertainty factors
sometimes applied by USEPA). Because the measured biomarker
is directly related to the internal dose metric of interest, direct
measurement of this biomarker concentration replaces applica-
tion of the pharmacokinetic component of the intraspecies
uncertainty factor in derivation of the BE values (Hays et al.,
2008); only the pharmacodynamic factor is appropriate on an
internal dose basis in this case.
For those oral exposure guidance values derived based on
mouse toxicity data (the WHO TDI and the ATSDR intermediate
MRL), a modified process outlined in Fig. 2 was used because of
the lack of a mouse PBPK model. In this approach, the interspecies
extrapolation is conducted on an external dose basis to obtain the
human equivalent external dose POD. The human average blood
concentrations associated with this POD were then estimated
using the human PBPK model to obtain the human equivalent
Previous
-------
S32
LI. Ay/word et al/Regulatory Toxicology and Pharmacology 51 (2008,) S27-S36
2.1.2.4. Evaluation ofBEsfor inhalation exposure guidelines. The gen-
eral approach for the derivation of BE values for the inhalation
exposure guidelines is presented in Fig. 3. Each of the applicable
guidelines is derived based on human data. Thus, the derivation
process does not involve an interspecies extrapolation. Briefly,
the process is as follows:
• Step 1: Calculate the BEPOD. The steady-state blood concentra-
tions in humans exposed at the duration- and LOAEL-to-NOAEL
adjusted PODs (based on human study data) were modeled
using the PBPK model described above. Because blood concen-
tration has been identified as a directly relevant dose metric
for neurological effects, the relevant internal dose metric and
the monitored biomarker concentration are the same. Thus,
these modeled blood concentrations are the BEPOD values used
in the derivation of the BEs for the inhalation exposure
guidelines.
• Step 2: Calculate the BE. Application of relevant intraspecies
uncertainty factor(s) and any additional applicable uncertainty
factors identified by the organizations that derived the oral
exposure guidelines initially (for example, database uncertainty
factors sometimes applied by USEPA). As above, because the
measured biomarker is directly related to the internal dose met-
ric of interest (they are the same for this set of exposure guide-
lines, blood toluene concentration) direct measurement of this
biomarker concentration replaces application of the pharmaco-
kinetic component of the intraspecies uncertainty factor in der-
ivation of the BE values (Hays et al., 2008); only the
pharmacodynamic factor is appropriate on an internal dose
basis in this case.
2.2. Results of modeling and identification of BE values
2.2.1. Urine
As discussed above, no specific and useful urinary markers for
toluene exposure at environmental exposure levels currently exist.
No urinary BE values were derived for toluene exposure.
2.2.2. Blood—oral exposure
All of the available chronic oral exposure guidelines are based
upon extrapolation from the same study of subchronic (13 weeks)
administration of toluene by gavage to rats or mice at duration-ad-
justed doses of 223, 446, 893, 1786, or 3571 mg/kg day (NTP,
1990). Two organizations, the USEPA and Health Canada, based
External
Dose
Relevant
Internal
Dose
Monitored
Biomarker
Animal
Human
Human
NOAELad(
POD
N
Human \
PK Model /
V
Human average blood concentration
u?5
Fig. 3. Schematic of approach used to estimate BE values for toluene corresponding
to inhalation exposure guidance values. See text for discussion.
their guidance values on liver or kidney toxicity observed in the
rat gavage study, while the WHO based its oral guidance value
on results from the mouse study. The ATSDR also derived exposure
guidance values for acute (1 to 14 day) and intermediate (up to 1
year) exposures. The acute MRL was derived based on a single dose
rat gavage study, while the intermediate MRL was derived based
on a 28-day mouse drinking water study.
Table 4 presents the modeling results and BE derivation for the
oral exposure guidance values based on rat toxicity data using the
general approach outlined in Fig. 1. Table 5 presents the corre-
sponding results and BE derivation for those oral exposure guid-
ance values derived from mouse toxicity data according to the
approach in Fig. 2.
2.2.3. Blood—inhalation exposure
All of the available inhalation exposure guidance values are
based on studies of human occupationally exposed populations
with a focus on a range of potential neurological effects. Several
studies in human occupational cohorts provide LOAEL or NOAEL
exposure estimates for all studied neurological endpoints includ-
ing both transient and persistent effects, as well as for a wide range
of other biochemical and health effect endpoints. Different agen-
cies have made slightly different choices in their selection of points
of departure for derivation of exposure guidance values, as sum-
marized in Table 1. Selected occupational exposure levels were ad-
justed to an equivalent continuous exposure concentration from
intermittent exposures experienced in the workplace. This adjust-
ment is applied to account for the presumption that the general
public could be continuously exposed in air. Note that this adjust-
ment implicitly assumes that the average concentration (or area
under the curve) is the critical dose metric. However, it is possible
that peak concentrations or time above a threshold level is as—or
more—important than average concentration in producing neuro-
toxic effects. Estimated peak blood concentrations following expo-
sure under actual occupational exposure concentrations are
approximately 3-fold higher than the duration-adjusted average
blood concentrations (modeling not shown). The ATSDR has also
derived an acute duration MRL (1-14 day exposure) based on
dose-response for neurological effects observed in a volunteer
study.
The results of PBPK modeling and BE derivation for the inhala-
tion exposure guidance values are presented in Table 6. The BE val-
ues from different agencies differ substantially due to different
judgments regarding whether selected occupational exposure lev-
els represent NOAELs or LOAELs. The BE values corresponding to
the USEPA RfC and the Health Canada inhalation TDI are higher
than the human equivalent BEPOD values derived from the WHO
and ATSDR exposure guidance values.
2.3. Discussion of sources of variability and uncertainty
2.3.1. Model uncertainty
The PBPK model used here has been used extensively to eval-
uate data sets for human and rat inhalation exposure and can
reproduce the observed blood concentration vs. time behavior
from independent data sets. The model incorporates understand-
ing of the physiological, physicochemical, and metabolic deter-
minants of toluene pharmacokinetics. However, as discussed
above, its application to oral exposures introduces some
additional uncertainty due to the behavior of rapidly eliminated
volatile compounds and the uncertainties associated with esti-
mation of peak blood concentrations associated with bolus oral
dosing. At very high oral exposures, saturation of metabolism
may become an issue resulting in non-linear relationships be-
tween external doses and resulting blood concentrations. How-
ever, at environmentally relevant exposures, such saturation is
Previous
-------
LL Ay/ward et al/Regulatory Toxicology and Pharmacology 51 (2008,) S27-S36
S33
Table 4
Estimated internal dose metrics and average human blood concentrations consistent with the derivation of oral exposure guidance values for toluene based on rat toxicity data
(see Fig. 1)
BE derivation step
USEPA chronic RfD
Health Canada chronic oral TDI
ATSDR acute MRL
Target organ
Administered dose regimen
Kidney
321 mgkg"1 day"1, rat gavage, 5 day/
week, 13 weeks (NOAEL)
Kidney, liver
321 mg kg"1 day"1, rat gavage, 5 day/
week, 13 weeks (NOAEL)
Brain
250 mgkg"1 rat single dose
gavage (LOAEL)
LOAEL-to-NOAEL adjustment
Duration adjustment and/or benchmark dose
modeling
Subchronic to chronic adjustment
POD, mgkg"1 day"1
BEpoD animal, Hg L"1 (Corresponding animal avg. blood
cone, from PBPK model)
Animal avg. target organ cone, from PBPK model, |ig L"1
Interspecies uncertainty factors
Pharmacodynamic
Pharmacokinetic
Human equivalent target organ avg. cone., j^gL"1
Human equivalent BEpoo, i^g L"1 (corresponding human
avg. blood cone, from PBPK model)
Intraspecies uncertainty factors
Pharmacodynamic
Pharmacokinetic
Other uncertainty factors
BE value, jig L"1
Confidence ratingd
None
Adjust for 5/7 day/week; benchmark
dose modeling
10
23
90
390a
io°-5
io°-5
39
16
io°-5
lc
3— database uncertainties
2
Medium
None
Adjust for 5/7 day/week
10
22
90
390a-450b
io°-5
io°-5
39-45
12-16
io°-5
lc
NA
3-5
Medium
3
None
NA
83
830
3650a
io°-5
io°-5
365
150
io°-5
lc
NA
50
Medium
a Average daily toluene concentration in kidney resulting from once daily bolus dosing at the NOAELadJ POD as estimated from the richly perfused compartment of the PBPK
model.
b Average daily toluene concentration in liver resulting from once daily bolus dosing at the NOAELadJ POD as estimated from the liver compartment of the PBPK model.
c Measurement of a biomarker that is directly relevant to the internal dose metric of interest replaces the default uncertainty factor for pharmacokinetic sensitivity. See
text for discussion.
d See text for discussion.
Table 5
Derivation of BEs for oral exposure guidance values for toluene based on mouse toxicity data (see Fig. 2)
BE derivation step
ATSDR intermediate MRL
WHO chronic TDI
Target organ
Administered dose regimen
LOAEL-to-NOAEL adjustment
Duration adjustment and/or benchmark dose modeling
Subchronic to chronic adjustment
POD, mgkg"1 day"1
Interspecies uncertainty factors
Pharmacodynamic
Pharmacokinetic
Human equivalent POD, mgkg"1 day"1
Human equivalent BEPOD, Hg L"1 (corresponding human avg. blood
cone, from PBPK model)
Intraspecies uncertainty factors
Pharmacodynamic
Pharmacokinetic
Other uncertainty factors
BE value, jig L"1
Confidence rating15
Brain
5 mgkg"1 day"1 mouse drinking water,
28 day (LOAEL)
3
NA
NA
10°-5
io°-5
0.17
1.7
io°-5
la
NA
0.5
Medium
Liver
321 mg kg"1 day"1, mouse gavage, 5 day/week,
13 weeks (LOAEL)
io°-5
Adjust for 5/7 day/week
10°-5
22
10°-5
io°-5
2.2
23
io°-5
la
NA
7
Medium
a Measurement of a biomarker that is directly relevant to the internal dose metric of interest replaces the default uncertainty factor for pharmacokinetic sensitivity. See
text for discussion.
b See text for discussion.
unlikely to occur. Thus, as a tool for predicting the central
tendency of blood concentrations associated with inhalation
exposure to toluene, the model uncertainty is low, while uncer-
tainty is somewhat higher for estimating peak concentrations
following oral exposures. Another potential uncertainty relates
to the modeling of kidney concentrations. The existing published
PBPK models used here do not include an explicit kidney
compartment. The use of the "richly perfused" compartment
for simulation of kidney concentrations is appropriate, but inclu-
sion of an explicit kidney compartment could be considered if
future data support this refinement of the model.
2.3.2. Analytical
The analytical methods for measuring toluene in blood are well
established (ACGIH, 2001). The variability due to analytical issues
should be minor in the context of the BE values.
2.3.3. Interindividual variations in pharmacokinetics
Differences in body composition (body fat levels, etc.), level of
physical activity, metabolic capability, and other factors can lead
to variations in blood concentrations of toluene associated with a
given exposure level. Data sets from controlled exposure experi-
ments show variations in blood levels in individuals exposed to
Previous
-------
S34
LI. Ay/word et al/Regulatory Toxicology and Pharmacology 51 (2008,) S27-S36
Table 6
Estimated internal dose metrics and average human blood concentrations consistent with the derivation of inhalation exposure guidance values for toluene based on human
toxicity data for central nervous system (CNS) effects (see Fig. 3)
BE derivation step
Target organ
Administered dose regimen
LOAEL-to-NOAEL adj.
Duration adjustment
Subchronic to chronic adjustment
POD, mg irr3 continuous
Human equivalent BEpoo, i^g L"1
(corresponding human avg. blood
cone, from PBPK model)
Intraspecies uncertainty factors
Pharmacodynamic
Pharmacokinetic
Other uncertainty factors
BE value, ng L"1
Confidence rating15
USEPA chronic RfC
CNS effects
34ppm (128 mgnr3)
NOAEL, occupational
exposure
NA
Adjust to continuous
exposure
None
46
170
io°-5
la
NA
50
High
Health Canada chronic
inhalation TDI
CNS effects
40ppm (150 mgm~3)
NOAEL, occupational
exposure
NA
Adjust to continuous
exposure
None
38
135
io°-5
la
NA
40
High
WHO air quality
guideline
CNS effects
SSppm (332 mgnr3)
LOAEL, occupational
exposure
10
Adjust to continuous
exposure
None
8
30
io°-5
la
3— potential
sensitivity of
developing CNS
3
High
ATSDR chronic
inhalation MRL
CNS effects
35ppm (132 mgirr3)
LOAEL, occupational
exposure
10
Adjust to continuous
exposure
NA
3
10
io°-5
la
NA
3
High
ATSDR acute MRL
CNS effects
40 ppm (150 mg irr3)
NOAEL, volunteers
exposed 6 h/day, 4 day
NA
Adjust to continuous
exposure
NA
30
100
io°-5
la
NA
30
High
a Measurement of a biomarker that is directly relevant to the internal dose metric of interest replaces the default uncertainty factor for pharmokinetic sensitivity. See text
for discussion.
b See text for discussion.
the same external air concentrations. Pierce et al. (1998) exposed
individuals to controlled concentrations of toluene in air for 2 h
and followed blood concentrations for approximately 100 h after
exposure ceased. At each time point, variations of a factor of two
to three from the mean were observed among the individuals, con-
sistent with a default assumption that interindividual pharmacoki-
netic differences could account for 3-fold variations from the
mean. These experimental results are consistent with the
variations predicted when physiological variability is incorporated
into PBPK modeling using the Tardif et al. models (Tardif et al.,
2002).
2.3.4. Gender and age
Nong et al. (2006) conducted a modeling study to evaluate the
impact of the development of CYP2E1 metabolic capability in in-
fants and children on the predicted blood concentration of toluene
following inhalation exposure. Nong et al. (2006) used data on the
concentration of hepatic CYP2E1 protein (Johnsrud et al., 2003) and
the change in liver tissue volume as a function of age to estimate
total CYP2E1 metabolic capability as a fraction of adult capability.
Using these data with age-specific physiologic parameters in the
Tardif et al. (1995) PBPK model, Nong et al. predicted that blood
levels in newborn infants could be as much as 3 times higher than
blood levels predicted in adults at the same air exposure level. Pre-
dicted blood concentrations in older children and adolescents were
more similar to those predicted in adults. Nong et al. (2006) noted
that this degree of variability was consistent with the pharmacoki-
netic component of the interindividual uncertainty factor used in
the derivation of the RfC. No data on the impact of gender on the
pharmacokinetics of toluene were identified, other than those
resulting from physiological variability, which can be accounted
for using the PBPK model. Varying bodyweight and other physio-
logical parameters in the model to account for female vs. male
physiology does not result in marked changes in predicted blood
concentrations (variations generally less than about 10 percent)
(Pelekis et al., 2001).
2.3.5. Smoking, drugs, alcohol co-exposures
Ethanol can inhibit metabolism of toluene through competition
for CYP2E1 (Baelum, 1991). Thus, co-exposure to these or other com-
pounds that are substrates for or inhibitors of CYP2E1 may result in
prolonged elevation of toluene blood concentrations compared to
those resulting from exposure to toluene alone. Smoking is a source
of toluene exposure and smokers consistently demonstrate higher
blood concentrations of toluene than non-smokers (see, for example,
Churchill et al., 2001). However, no information is available regard-
ing the impact of smoking on elimination rates of toluene.
2.3.6. Polymorphisms in enzymes or other factors
Researchers are beginning to identify polymorphisms in genes
coding for key metabolic enzymes and examine the impact of such
polymorphisms on potential responses. Polymorphisms in several
of the enzymes known to be involved in the metabolism of toluene,
including CYP2E1, have been identified. However, researchers have
focused on correlating the occurrence of such polymorphisms with
susceptibility to various conditions (see, for example, Heuser et al.,
2007; Kezic et al., 2006) rather than directly assessing the effects of
these polymorphisms on metabolic capability. Some studies have
indicated an impact of such polymorphisms on enzymatic activity,
but data are limited to date (reviewed in Gemma et al., 2006). Thus,
we cannot draw any conclusions regarding the impact of such
polymorphisms on predicted blood concentrations in individuals
exposed at an exposure guideline. Such polymorphisms may
account for some of the variability in blood levels observed among
individuals after controlled exposures (see above).
2.4. Confidence assessment
Guidelines for derivation of BE values (Hays et al., 2008) specify
consideration of two main elements in the assessment of confi-
dence in the derived BE values: robustness of the available phar-
macokinetic models and data, and understanding of the
relationship between the measured biomarker and the critical or
relevant target tissue dose metric.
2.4A. Confidence in BE values based on oral exposure guidelines
As discussed above, the oral exposure route introduces addi-
tional uncertainties in estimating blood BE values corresponding
to the oral exposure guidelines. These uncertainties stem from
uncertainty regarding the appropriate dose metric (for example,
Previous
-------
LL Ay/ward et al/Regulatory Toxicology and Pharmacology 51 (2008,) S27-S36
S35
area under the curve vs. peak target organ concentrations) and
uncertainty in the active metabolite responsible for liver or kid-
ney toxicity. In this assessment we have relied on area under the
curve of the parent compound in the target organ of interest as
the relevant dose metric. However, if peak concentration (or
peak metabolite production) is more relevant, then uncertainty
regarding the appropriate oral absorption rate value for the
models (which impacts estimates of peak, but not average, blood
concentrations), and uncertainty regarding the appropriate dos-
ing regimen to assume for exposure at the health-based expo-
sure guidelines (once per day bolus vs. divided dose; again,
impacting peak but not average blood concentrations) affects
confidence in the BE values. Blood concentration as a biomarker
for toluene should be directly related to average target organ
concentration, but may be less informative regarding peak target
organ concentration. For this reason, confidence in the BE values
associated with the oral dosing route is lower than that for the
inhalation exposure route.
In summary, the confidence ratings for BE values for oral expo-
sure guidance values are:
• Relevance of biomarker to relevant dose metrics: MEDIUM.
• Robustness of pharmacokinetic data/models: MEDIUM.
2.4.2. Confidence in BE values based on inhalation exposure guidelines
Blood toluene concentrations are directly related to target tis-
sue concentrations in the brain. The available PBPK models are
well-validated and have been extensively used in combination
with occupational data sets in humans. The BE values for inhala-
tion exposure guidance values for toluene based on blood concen-
tration as the biomarker thus have HIGH confidence for both
aspects.
In summary, the confidence ratings for BE values for inhalation
exposure guidance values are:
• Relevance of biomarker to relevant dose metrics: HIGH.
• Robustness of pharmacokinetic data/models: HIGH.
The summary confidence ratings are presented in Tables 4-6.
3. Discussion and interpretation of BE values
The BE values presented here represent the concentrations of
toluene in blood that are consistent with exposure at the exposure
guideline values that have been established by various agencies
(Table 1). These BE values should be regarded as interim values
that can be updated or replaced if the exposure guideline values
are updated or if the scientific and regulatory communities develop
additional data on acceptable or tolerable concentrations in human
biological media based directly on epidemiological data.
The BE values presented here are screening values and can be
used to provide a screening-level assessment of measured blood
concentrations of toluene in population- or cohort-based studies.
Comparison of measured values to the values presented here can
provide an initial evaluation of whether the measured values in a
given study are of low, medium, or high priority for risk assess-
ment follow-up. Fig. 4 illustrates the presentation of the BE value
corresponding to the Health Canada inhalation TDI. Measured
biomarker values in excess of the human equivalent BEPOD indi-
cate a high priority for risk assessment follow-up. Values below
the BEpoo but above the BE suggest a medium priority for risk
assessment follow-up, while those below the BE values suggest
low priority for risk assessment follow-up. Based on the results
of such comparisons, an evaluation can be made of the need for
135
40
High
Medium
Low
POD
BE
Fig. 4. Example presentation of the BE value corresponding to the Health Canada
inhalation TDI. The BEPOD corresponds to the average blood concentrations
estimated at the identified human no-observed-adverse-effect-level used as point
of departure for the guideline derivation. The BE value presents the blood
concentration consistent with the TDI (see Table 6 and the text for details on the
derivation). Similar graphs could be prepared for the BE values derived for each of
the available exposure guidance values.
additional studies on exposure pathways, potential health effects,
other aspects affecting exposure or risk, or other risk manage-
ment activities.
Numerous exposure guideline values and thus BEs exist for
interpreting human biomonitoring data for toluene. Selecting the
most appropriate BE (and BEPOD) for interpreting biomonitoring
data may depend on several factors including: the year the expo-
sure guidance value was established (and thus potentially reflects
advancement in understanding of toluene toxicity, mechanism of
action, or available studies for deriving an exposure guideline va-
lue); whether the exposure guidance value was based upon animal
or human toxicity data; the route of exposure from which the
exposure guideline value was derived (and thus potentially reflects
the more predominant pathway for exposure in the environment);
the degree of uncertainty involved in the derivation of the expo-
sure guidance value; and other judgments regarding the reliability
of the underlying exposure guidance value.
BE values do not represent diagnostic criteria and cannot be
used to evaluate the likelihood of an adverse health effect in an
individual or even among a population. In the case of toluene, BE
values corresponding to exposure guidance values from different
agencies differ widely, and interpretation of biomonitoring data re-
sults may depend upon which guidance value is regarded as most
reliable and appropriate for a given situation. Further discussion of
interpretation and communications aspects of the BE values is pre-
sented in LaKind et al. (2008).
Disclaimer
This work was reviewed by EPA and approved for publication,
but does not necessarily reflect official Agency policy. Mention of
trade names or commercial products does not constitute endorse-
ment or recommendation by EPA for use.
Conflict of interest disclosure statement
The authors declare that they have no conflicts of interest.
Previous
-------
S36
LI. Ay/ward et al./Regulatory Toxicology and Pharmacology 51 (2008) S27-S36
Acknowledgments
Funding for this project was provided by the US Environmental
Protection Agency, Health Canada, the American Chemistry Coun-
cil, the American Petroleum Institute, Responsible Industry for a
Sound Environment, and the Soap and Detergent Association. The
authors thank the BE Steering Committee for their advice and guid-
ance: John H. Duffus, Monty Eberhart, Bruce Fowler (advisor),
George Johnson, Mike Kaplan, Bette Meek, David Moir, David J.
Miller, Larry L. Needham (advisor), and Bob Sonawane. Finally,
the authors thank Michael J. Bartels, Peter J. Boogaard, and Kannan
Krishnan for their careful review and comments on this
manuscript.
References
Agency for Toxic Substances and Disease Registry (ATSDR), 2000. Toxicological
profile for toluene. Available from: .
Al-Ghamdi, S.S., Raftery, M.J., Yaqoob, M.M., 2003a. Acute solvent exposure induced
activation of cytochrome P4502E1 causes proximal tubular cell necrosis by
oxidative stress. Toxicol. In Vitro 17, 335-341.
Al-Ghamdi, S.S., Raftery, M.J., Yaqoob, M.M., 2003b. Organic solvent-induced
proximal tubular cell toxicity via caspase-3 activation. J. Toxicol. Clin. Toxicol.
41, 941-945.
American Conference of Governmental and Industrial Hygienists (ACGIH), 2001.
Documentation of toluene BEI. Available from: .
Angerer, J., Schildbach, M., Kramer, A, 1998. S-p-Toluylmercapturic acid in the urine
of workers exposed to toluene: a new biomarker for toluene exposure. Arch.
Toxicol. 72, 119-123.
Baelum, J., 1991. Human solvent exposure. Factors influencing the
pharmacokinetics and acute toxicity. Pharmacol. Toxicol. 68 (Suppl. 1), 1-36.
Baelum, J., Molhave, L, Honore Hansen, S., Dossing, M., 1993. Hepatic metabolism of
toluene after gastrointestinal uptake in humans. Scand. J. Work Environ. Health
19, 55-62.
Benignus, V.A., Boyes, W.K., Kenyon, E.M., Bushnell, P.J., 2007. Quantitative
comparisons of the acute neurotoxicity of toluene in rats and humans.
Toxicol. Sci. 100, 146-155.
Bushnell, P.J., Oshiro, W.M., Samsam, T.E., Benignus, V.A., Krantz, Q.T., Kenyon, E.M.,
2007. A dosimetric analysis of the acute behavioral effects of inhaled toluene in
rats. Toxicol. Sci. 99,181-189.
Churchill, J.E., Ashley, D.L, Kaye, W.E., 2001. Recent chemical exposures and blood
volatile organic compound levels in a large population-based sample. Arch.
Environ. Health 56,157-166.
Dossing, M., Aelum, J.B., Hansen, S.H., Lundqvist, G.R., Andersen, N.T., 1983. Urinary
hippuric acid and orthocresol excretion in man during experimental exposure
to toluene. Br. J. Ind. Med. 40, 470-473.
Fustinoni, S., Mercadante, R., Campo, L, Scibetta, L, Valla, C, Consonni, D., Foa, V.,
2007. Comparison between urinary o-cresol and toluene as biomarkers of
toluene exposure.]. Occup. Environ. Hyg. 4,1-9.
Gemma, S., Vichi, S., Testai, E., 2006. Individual susceptibility and alcohol effects:
biochemical and genetic aspects. Ann. 1st. Super. Sanita 42, 8-16.
Hays, S.M., Becker, R.A., Leung, H.W., Aylward, LL, Pyatt, D.W., 2007. Biomonitoring
equivalents: a screening approach for interpreting biomonitoring results from a
public health risk perspective. Regul. Toxicol. Pharmacol. 47, 96-109.
Hays, S.M., Aylward, LL, LaKind, J.S., Bartels, M.J., Barton, H.A., Boogaard, P.J., Brunk,
P.J., Dizio, S., Dourson, M., Goldstein, D.A., Lipscomb, J., Kilpatrick, M.E., Krewski,
D., Krishnan, K., Nordberg, M., Okino, M., Tan, Y.-M., Viau, C, Yager, J.W., 2008.
Guidelines for the derivation of Biomonitoring Equivalents: Report from the
Biomonitoring Equivalents Expert Workshop. Regul. Toxicol. Pharmacol. 51, S4-
S15.
Health Canada, 1996. Health-based tolerable daily intakes/concentrations and
tumorigenic doses/concentrations for priority substances. Environmental
Health Directorate. Available from: .
Heuser, V.D., Erdtmann, B., Kvitko, K., Rohr, P., da Silva, J., 2007. Evaluation of
genetic damage in Brazilian footwear-workers: biomarkers of exposure, effect,
and susceptibility. Toxicology 232, 235-247.
Inoue, 0., Kanno, E., Kasai, K., Ukai, H., Okamoto, S., Ikeda, M., 2004.
Benzylmercapturic acid is superior to hippuric acid and o-cresol as a
urinary marker of occupational exposure to toluene. Toxicol. Lett. 147,
177-186.
International Agency for Research on Cancer (IARC), 1999. IARC Monographs on the
Evaluation of Carcinogenic Risks to Humans. Vol. 71: Re-Evaluation of Some
Organic Chemicals, Hydrazine and Hydrogen Peroxide. Summary of Data
Reported and Evaluation. Available from: (accessed 12/2007).
Jang, J.-Y., 1996. Simulation of the toluene in venous blood with a physiologically
based pharmacokinetic model: its application to biological exposure index
development. Appl. Occup. Environ. Hyg. 11, 1092-1095.
Johnsrud, E.K., Koukouritaki, S.B., Divakaran, K., Brunengraber, LL, Hines, R.N.,
McCarver, D.G., 2003. Human hepatic CYP2E1 expression during development.
J. Pharmacol. Exp. Ther. 307, 402-437.
Kawai, T., Ukai, H., Inoue, 0., Maejima, Y., Fukui, Y., Ohashi, F., Okamoto, S., Takada,
S., Sakurai, H., Ikeda, M., 2008. Evaluation of biomarkers of occupational
exposure to toluene at low levels. Int. Arch. Occup. Environ. Health 81,253-262.
Kezic, S., Calkoen, F., Wenker, M.A., Jacobs, J.J., Verberk, M.M., 2006. Genetic
polymorphism of metabolic enzymes modifies the risk of chronic solvent-
induced encephalopathy. Toxicol. Ind. Health 22, 281-289.
LaKind, J.S., Aylward, LL, Brunk, C, DiZio, S., Dourson, M., Goldstein, D.A., Kilpatrick,
M.E., Krewski, D., Bartels, M., Barton, H.A., Boogaard, P.J., Lipscomb, J., Krishnan,
K., Nordberg, M., Okino, M., Tan, Y.-M., Viau, C, Yager, J.W., Hays, S.M., 2008.
Guidelines for the communication of Biomonitoring Equivalents: Report from
the Biomonitoring Equivalents Expert Workshop. Regul. Toxicol. Pharmacol. 51,
S16-S26.
National Toxicology Program (NTP), 1990. Toxicology and carcinogenesis studies of
toluene (CAS #108-88-3) in F344/N rats and B6C3F1 mice (inhalation studies).
NTP Technical Report Series No. 371. U.S. Department of Health and Human
Services.
Neubert, D., Gericke, C., Hanke, B., Beckmann, G., Baltes, M.M., Kiihl, K.P., Bochert, G.,
Hartmann, J.Toluene Field Study Group, 2001. Multicenter field trial on possible
health effects of toluene: II. Cross-sectional evaluation of acute low-level
exposure. Toxicology 168,158-183.
Nong, A., McCarver, D.G., Hines, R.N., Krishnan, K., 2006. Modeling interchild
differences in pharmacokinetics on the basis of subject-specific data on
physiology and hepatic CYP2E1 levels: a case study with toluene. Toxicol.
Appl. Pharmacol. 214, 78-87.
Pelekis, M., Gephart, L.A., Lerman, S.E., 2001. Physiological-model-based derivation
of the adult and child pharmacokinetic intraspecies uncertainty factors for
volatile organic compounds. Regul. Toxicol. Pharmacol. 33,12-20.
Pierce, C.H., Dills, R.L, Morgan, M.S., Nothstein, G.L, Shen, D.D., Kalman, D.A., 1996.
Interindividual differences in 2H8-toluene toxicokinetics assessed by
semiempirical physiologically based model. Toxicol. Appl. Pharmacol. 139,
49-61.
Pierce, C.H., Dills, R.L, Morgan, M.S., Vicini, P., Kalman, D.A., 1998. Biological
monitoring of controlled toluene exposure. Int. Arch. Occup. Environ. Health 71,
433-444.
Sexton, K., Adgate, J.L, Church, T.R., Ashley, D.L, Needham, LL, Ramachandran, G.,
Fredrickson, AL, Ryan, A.D., 2005. Children's exposure to volatile organic
compounds as determined by longitudinal measurements in blood. Environ.
Health Perspect. 113, 342-349.
Stengel, B., Cenee, S., Limasset, J.C., Diebold, F., Michard, D., Druet, P., Hemon, D.,
1998. Immunologic and renal markers among photogravure printers exposed to
toluene. Scand. J. Work Environ. Health 24, 276-284.
Sullivan, M.J., Conolly, R.B., 1988. Comparison of blood toluene levels after
inhalation and oral administration. Environ. Res. 45, 64-70.
Tardif, R., Lapare, S., Krishnan, K., Brodeur, J., 1993. Physiologically based modeling
of the toxicokinetic interaction between toluene and m-xylene in the rat.
Toxicol. Appl. Pharmacol. 120, 266-273.
Tardif, R., Droz, P.O., Charest-Tardif, G., Pierrehumbert, G., Truchon, G., 2002. Impact
of human variability on the biological monitoring of exposure to toluene: I.
Physiologically based toxicokinetic modelling. Toxicol. Lett. 134,155-163.
Tardif, R., Lapare, S., Charest-Tardif, G., Brodeur, J., Krishnan, K., 1995.
Physiologically based modeling of a mixture of toluene and xylene in
humans. Risk Anal. 15, 335-342.
Truchon, G., Tardif, R., Brodeur, J., 1999. o-Cresol: a good indicator of exposure to
low levels of toluene. Appl. Occup. Environ. Hyg. 14, 677-681.
United States Environmental Protection Agency (USEPA), 2005. Toxicological
Review of Toluene (CAS No. 108-88-3) In Support of Summary Information on
the Integrated Risk Information System (IRIS). EPA/635/R-05/004.
van Asperen, J., Rijcken, W.R., Lammers, J.H., 2003. Application of physiologically
based toxicokinetic modelling to study the impact of the exposure scenario on
the toxicokinetics and the behavioural effects of toluene in rats. Toxicol. Lett.
138, 51-62.
World Health Organization (WHO), 2004. Guidelines for Drinking-Water Quality,
vol. 1. Recommendations, third ed. World Health Organization, Geneva.
World Health Organization (WHO), 2005. Air Quality Guidelines for Europe, second
ed. WHO Regional Office for Europe Copenhagen. Available from: .
Previous
-------
TOXICOLOGICAL SCIENCES 108(1), 207-221 (2009)
doi:10.1093/toxsci/kfp005
Advance Access publication January 27, 2009
Comparative Microarray Analysis and Pulmonary Changes in Brown
Norway Rats Exposed to Ovalbumin and Concentrated Air Particulates
Brooke L. Heidenfelder,* David M. Reif,t Jack R. Harkema,^ Elaine A. Cohen Hubal,t Edward E. Hudgens,* Lori A. Bramble,^
James G. Wagner,^ Masako Morishita,§ Gerald J. Keeler,§ Stephen W. Edwards,^f and Jane E. Gallagher*'1
*Mail Drop 58 C Human Studies Division, National Health and Environmental Effects Research Laboratory, Office of Research and Development, US
Environmental Protection Agency, Research Triangle Park, North Carolina 27711; ^Mail Drop D343-03 and Mail Drop 205-01, National Center for
Computational Toxicology, Office of Research and Development, US Environmental Protection Agency, Research Triangle Park, North Carolina 27711;
^Department of Pathobiology and Diagnostic Investigation, Michigan State University, East Lansing, Michigan 48824; §Ai'r Quality Laboratory, University
of Michigan, Ann Arbor, Michigan 48109; and \Mail Drop B305-01 Office of the Associate Director of Health, National Health and Environmental
Effects Research Laboratory Immediate Office, US Environmental Protection Agency, Office of Research and Development, Research Triangle Park,
North Carolina 27711
Received October 7, 2008; accepted December 23, 2008
The interaction between air particulates and genetic suscepti-
bility has been implicated in the pathogenesis of asthma. The
overall objective of this study was to determine the effects of
inhalation exposure to environmentally relevant concentrated air
particulates (CAPs) on the lungs of ovalbumin (ova) sensitized
and challenged Brown Norway rats. Changes in gene expression
were compared with lung tissue histopathology, morphometry,
and biochemical and cellular parameters in bronchoalveolar
lavage fluid (BALF). Ova challenge was responsible for the
preponderance of gene expression changes, related largely to
inflammation. CAPs exposure alone resulted in no significant gene
expression changes, but CAPs and ova-exposed rodents exhibited
an enhanced effect relative to ova alone with differentially
expressed genes primarily related to inflammation and airway
remodeling. Gene expression data was consistent with the
biochemical and cellular analyses of the BALF, the pulmonary
pathology, and morphometric changes when comparing the CAPs-
ova group to the air-saline or CAPs-saline group. However, the
gene expression data were more sensitive than the BALF cell type
and number for assessing the effects of CAPs and ova versus the
ova challenge alone. In addition, the gene expression results
provided some additional insight into the TGF-p-mediated
molecular processes underlying these changes. The broad-based
histopathology and functional genomic analyses demonstrate that
exposure to CAPs exacerbates rodents with allergic inflammation
induced by an allergen and suggests that asthmatics may be at
increased risk for air pollution effects.
Key Words: ovalbumin; allergen; asthma; particulate matter;
remodeling; inflammation; microarray.
1 To whom correspondence should be addressed at Mail Drop 58C National
Health and Environmental Effects Laboratory, US Environmental Protection
Agency, Office of Research and Development, Research Triangle Park, NC
27711. Express mail: 104 Mason Farm Rd. Human Study Facility, Chapel Hill,
NC 27514. Fax: (919) 966-0655. E-mail: gallagher.jane@epa.gov.
Published by Oxford University Press 2009.
The prevalence of asthma has increased in recent years,
particularly in industrialized countries, making it an important
public health concern. The air pollution and specific allergens
that accompany urbanization are thought to be partly re-
sponsible for this increase (Busse and Mitchell, 2007). Diesel
exhaust (DE) is a major contributor to particulate matter (PM)
related air pollution, especially in urban areas. As DE
emissions have increased globally, so has asthma prevalence
(Keller and Lowenstein, 2002). Airborne PM has a number of
detrimental health effects, particularly in asthmatics (Riedl and
Diaz-Sanchez, 2005). Genetics are also a factor; they determine
the susceptibility to asthma or other respiratory diseases that
can be influenced by PM (Moller et al, 2007).
PM from air pollution has several mechanisms of action for
generating inflammation, damaging cells, and exacerbating
airway hyperresponsiveness (Gilmour et al., 2006). hi par-
ticular, PM derived from a variety of sources induces the
production of reactive oxygen species in inflammatory cells
(Becker et al., 2002) and oxidative DNA damage in human
airway epithelial cells (Prahalad et al., 2001). PM exposure
also has specific allergy-related effects. For example, the
polyaromatic hydrocarbons in DE particles have been shown to
increase IgE production by human B cells (Takenaka et al.,
1995; Tsien et al., 1997). Increases in coarse PM correlate with
increases in circulating eosinophils, serum triglycerides, and
decreased heart rate variability in asthmatics (Yeatts et al.,
2007). PM has also been shown to act as an adjuvant in
ovalbumin (ova)-induced allergic reactions in mice and Brown
Norway (BN) rats (Dong et al, 2005; Harkema et al, 2004a;
Matsumoto et al, 2006; Miyabara et al, 1998; Takano et al,
1997). BN rats challenged with ova have greater PM particle
retention and eosinophilia in the lungs, compared with
unchallenged BN rats (Morishita et al, 2004). Humans with
allergies have increased Th2 cytokines and ragweed-specific
IgE expression after coexposure to ragweed allergen and DE
Previous
TOG
Next
-------
208
HEIDENFELDER ET AL.
participates (Diaz-Sanchez et al., 1997). The mechanism for
the effect of PM on asthma remains unclear, due to its complex
interactions with the immune system and other metabolic
pathways (McCunney, 2005).
Examining the complex interaction between environmental
exposures and genetic factors can provide vital information
needed to better understand and treat respiratory diseases
(Kleeberger and Peden, 2005). Because asthma genetics are
complex and likely to involve many genes that vary by
population, the use of emerging genomics tools such as
expression profiling and pathway analyses are needed (Ober
and Hoffjan, 2006). Using a rodent model of airway allergic
inflammation or pulmonary allergy minimizes genetic variation
so that allergy- and exposure-induced gene expression changes
can be isolated and investigated. Gene expression profiling in
animal models has yielded large numbers of differentially
expressed genes, from which genes related to asthma
susceptibility and pathogenesis have been distilled (Follettie
et al, 2006; Izuhara and Saito, 2006; Kuperman et al, 2005;
Walker et al, 2006). However, there is little information
regarding the precise mechanisms of gene-environment
interactions in relation to asthma (London, 2007).
BN rats sensitized with ova are thought to resemble the
clinically significant features of allergic asthma (Salmon et al,
1999; Tarayre et al, 1992; Underwood et al, 1995) and often
used as an animal model for pulmonary allergic conditions
such as asthma because they have a high capacity for IgE
production and exhibit airway hyperresponsiveness following
exposure to inhaled allergens (Abadie and Prouvost-Danon,
1980; Pauwels et al, 1979). A few studies have examined
global gene expression profiles in rodents exposed to
environmental agents, including concentrated ambient partic-
ulates (CAPs) (Gunnison and Chen, 2005; Sigaud et al, 2007),
environmental tobacco smoke (Izzotti et al, 2005; Nadadur
et al, 2002), ozone (Leikauf et al, 2001; Park et al, 2004;
Williams et al, 2007), and PM (Kooter et al, 2005; Nadadur
and Kodavanti, 2002; Sato et al, 1999; Wise et al, 2006). BN
rats coexposed to both lipopolysaccharide (LPS) and tobacco
smoke (Meng et al, 2006) or mice coexposed to LPS and DE
(Yanagisawa et al, 2004) have also been examined via micro-
array, as models of chronic obstructive pulmonary disorder and
acute lung injury, respectively.
To our knowledge, this study is the first to use gene
expression array profiling and traditional toxicological ap-
proaches to decipher and evaluate how the combination of
susceptibility and exposure to environmentally relevant CAPS
act together to perturb biological pathways that may be
important to the exacerbation of asthma.
MATERIALS AND METHODS
Animals and exposure regimen. Figure 1 shows the experimental design
for the four treatment groups BN rodents (Charles River, Indianapolis, IN) Each
group consisted of eight male rats, aged 10-12 weeks. Rats were free of
pathogens and respiratory disease, and used in accordance with guidelines set
forth by the Institutional Animal Care and Use Committee at Michigan State
University. All four groups were sensitized in the animal care facility, for three
consecutive days to ova (0.5% in saline, intranasally). Two weeks later, two of
the groups were then challenged with saline vehicle, the other two with ova, by
intranasal instillation (0.5% ova in saline, 150 ul/nasal passage) for 3 con-
secutive days. Starting the day after the last challenge, two groups of rats (one
each of the saline and ova-challenged groups) were exposed to CAPs from
Grand Rapids, MI (8 h/day for 13 days). There was a second ova or saline
challenge 9 days following the first ova challenge. Rats were sacrificed 24 h
after the last CAPs exposure.
Treatment group
Sensitization
1st challenge
Exposures
2nd challenge
End
Air-
Saline
Air-
ova
CAPs-
Saline
CAPs-
ova
OVa Saline Saline
OVa ova ova
Ova Saline Saline
OVa OVa ova
14 16 17
Time (days)
26
29 30
FIG. 1. Experimental design and exposure regimen. The initial ova sensitizations and first ova or saline challenges were administered intranasally over three
consecutive days (0-2 and 14-16, respectively). Animals were exposed to CAPs or ambient filtered air on days 17-29. A second ova or saline challenge was
administered intranasally on day 26. Lung tissue was harvested on day 30.
Previous
TOC
-------
COMPARATIVE MICROARRAY ANALYSIS AND PULMONARY CHANGES IN BROWN NORWAY RATS
209
The CAPs exposure and ova challenge occurred via the mobile air research
laboratory, AirCARE 1 (Harkema et al., 2004b). AirCARE 1 contained whole
body inhalation chambers with a Harvard/EPA ambient fine particle
concentrator, a biomedical lab, an inhalation exposure lab, and an atmospheric
monitoring lab (Harkema et al., 2004b; Keeler et al., 2007). The fine particle
concentrator was a three-stage aerosol concentrator that utilizes virtual
impactors to increase the concentration of particles (size range 0.1-2.5 um)
by an approximate factor of 30 (Sioutas et al., 1997). The remaining two groups
were exposed to HEPA-filtered ambient air, also in AirCARE 1.
Characterization of CAPs. CAPs were collected during each 8-h exposure
period. The mass concentrations of CAPs were determined by placing 47-mm
Teflon filters (Gelman Sciences, Inc., Ann Arbor, MI) in Teflon filter packs
attached to the back of the animal exposure chambers at flow rates of 3 LPM.
After the gravimetric determination, CAPs samples on Teflon filters were
extracted in 10% nitric acid and analyzed for a suite of trace elements
(including crustal/urban dust related elements (Fe, Si, Ca, Al) using inductively
coupled plasma-mass spectrometry (ELEMENT 2, Thermo Finnigan, San Jose,
CA). Pre-baked quartz filters (Gelman Sciences, Inc., Ann Arbor, MI) were also
placed in Teflon filter packs mounted on the exposure chambers and were
analyzed for carbonaceous (organic and elemental) aerosols by a thermal-
optical analyzer (Sunset Labs, Forest Grove, OR). Annular denuder/filter pack
samples were also collected and analyzed for acid gases and ion species by ion
chromatography (Model DX-600, DIONEX, Sunnyvale, CA).
RNA isolation. Four animals from each group were randomly chosen for
gene expression analysis. Total RNA was isolated from the right cranial lung
lobe, using RNeasy Mini kits (Qiagen Inc., Valencia, CA) with DNase
treatment. RNA quality was checked using an Agilent Bioanalyzer (Agilent
Technologies, Palo Alto, CA). RNA was quantified on a NanoDrop ND-1000
spectrophotometer (NanoDrop Technologies, Wilmington, DE), and then 3.8 ug
of each sample was sent to Expression Analysis (Durham, NC) for cDNA target
generation and hybridization to rat R230 2.0 whole genome arrays (Affymetrix,
Inc., Santa Clara, CA). RNA from one rat in the CAPs-saline exposure group
failed to generate enough target to hybridize to a microarray chip.
Gene array data analysis. Invariant probe signals were removed using the
REDI (reduction of invariant probes) method, proprietary to Expression
Analysis. Array data were normalized using robust multiple-array averaging
(RMA), and compared by group using the permutation analysis for differential
expression (FADE, http://www.expressionanalysis.com/pdf/PADE_TechNote_
2005.pdf) algorithm. Noise from the FADE was removed by setting probe
signals that were < 128 to 128 (the level of noise) to prevent artificially high
fold change differences, and fold changes were recalculated. Absolute fold
changes < 1.5 and values which exceeded accumulated false discovery rates
(FDRs) of < 0.05 were removed. The resulting significant probesets from each
FADE analysis are shown in Supplemental Table S1. Probesets were annotated
using NetAffx (Affymetrix Inc., Santa Clara, CA). Only three FADE analyses
had a list of significant genes that met the cutoff values. The probeset lists
generated by this procedure were uploaded into GeneSpring 7 (Agilent
Technologies, Palo Alto, CA) for principal components and clustering analyses.
Pathways analysis with metacore. Probeset lists created using an FDR of
<0.2 (Supplemental Table S2) rather than <0.05 were uploaded into MetaCore
4.5 (GeneGo, Inc., St Joseph, MI, http://www.genego.com/metacore.php) for
functional and pathways analysis. Each gene identifier was mapped to its
corresponding gene object in the MetaCore database. The genes were then
compared with both gene ontology (GO) processes and GeneGo maps to
determine processes or pathways which were significantly overrepresented in
the differentially expressed gene list. The p values for maps and processes were
calculated using a hypergeometric distribution.
Pathways analysis with ingenuity pathways analysis. Probeset lists
created using an FDR of <0.2 (Supplemental Table S2) rather than <0.05
were analyzed through the use of Ingenuity Pathways Analysis (Ingenuity
Systems, www.ingenuity.com). Each gene identifier was mapped to its
corresponding gene object in the Ingenuity Pathways Knowledge Base. These
genes were overlaid onto a global molecular network developed from
information contained in the Ingenuity Pathways Knowledge Base. Networks
of these focus genes were then algorithmically generated based on their
connectivity. The Functional Analysis of the top scoring network identified the
biological functions and/or diseases that were most significant to the genes in
the network. The network genes associated with biological functions and/or
diseases in the Ingenuity Pathways Knowledge Base were considered for the
analysis. Fisher's exact test was used to calculate a p value determining the
probability that each biological function and/or disease assigned to that network
is due to chance alone.
Confirmatory quantitative PCR. Total RNA was used to generate cDNA
for confirmatory quantitative PCR (qPCR). Total RNA (350 ng) was reverse
transcribed in a buffer containing 1X reverse transcriptase polymerase chain
reaction buffer, 25uM random hexamers, 5mM dithiotreitol, SOOuM deoxnu-
cleotide triphosphates, 20 U RNase OUT, and 200 U Superscript III (all from
Invitrogen Corp., Carlsbad, CA) for 10 min at 25°C, 60 min at 50°C, and 15 min
at 75°C. Six genes of interest and one endogenous control (p-actin) were
analyzed by confirmatory qPCR. TaqMan MGB probes (Applied Biosystems,
Foster City, CA) for transforming growth factor (33 (TGF-J33), Fc receptor IgE
high affinity I alpha polypeptide (Peerla), complement component 4 binding
protein (C4BP), vascular endothelial growth factor C (VEGF-C), chitinase
3-like 1 (CM3L1), metallothionein la (Mtla), and the p-actin endogenous control
were used according to the recommended procedure on a 7500 Real-Time PCR
machine (Applied Biosystems, Foster City, CA). Briefly, 5 ul of cDNA was
mixed with 2.5 ulof 20X TaqMan MGB probe mix, 17.5 ul of water, and 25 ul
of 2X TaqMan Universal PCR Master Mix for a total volume of 50 ul per well
in a 96-well optical plate (Applied Biosystems, Foster City, CA). Reactions for
each sample were performed in duplicate. The plate was run for 2 min at 50°C,
then 10 min at 90°C, followed by 40 cycles of 15 s at 95°C and 1 min at 60°C.
The data were analyzed using the Auto CT method to generate a standard curve,
and the duplicate relative concentration measurements for each sample were
averaged. Data were normalized to the control gene concentration by dividing
the mean relative concentration for a gene of interest by the mean relative
concentration of the p-actin control. Normalized values were then averaged
among rats in each treatment group.
Statistical analysis of qPCR results. Data from qPCR (n = 4) were
expressed as the mean group value ± the standard error of the mean. The data
were subjected to ANOVA for factors of inhalation exposure type (e.g., CAPs
or filtered air) and airway sensitization/challenge (ova or saline). Significant
differences between experimental groups were identified using the Tukey
honest significant difference post hoc test. The criterion for statistical
significance was p < 0.05.
Lung tissue section and bronchoalveolar lavage. All animal sacrifices
were conducted in the laboratory of Dr. Harkema at Michigan State University.
At the time of sacrifice, animals were deeply anesthetized, 5 ml of whole blood
was collected from the ascending vena cava, and the animal was exsanguinated.
The trachea was cannulated, and the heart/lung block was removed from the
thoracic cavity. The right extrapulmonary bronchus was ligated with suture and
the right lung lobes were removed. The entire right cranial lobe was processed
for isolation of total RNA (see above). The left lung lobe was lavaged with
saline and the recovered lavage fluids (bronchoalveolar lavage fluid; BALF)
from each rat were analyzed for total and differential cell counts, total and ova-
specific IgE and secreted mucins by enzyme-linked immunosorbent assay
(ELISA). After bronchoalveolar lavage the left lung lobe was perfusion-fixed
with 10% neutral buffered formalin via the trachea at a constant pressure of
30 cm of fixative. After 2 h of intratracheal fixation, the trachea was ligated and
the lung lobe was immersed in a large volume of the same fixative for at least
24 h before the left lung lobe was further processed for light microscopy as
described below.
Analysis of BALF. Total cells recovered by bronchoalveolar lavages were
determined manually using a hemacytometer. Cytospin preparations from eight
rodents per treatment group were made with a cytocentrifuge and stained with
Diff-Quick (IMEB, Inc., San Marcos, CA). Differential cell counts
Previous
TOC
Next
-------
210
HEIDENFELDER ET AL.
(e.g., neutrophils, macrophages, eosinophils, lymphocytes) were determined by
counting 200 cells per animal. The BALE was centrifuged to remove cells and
debris, and the supernatant was stored at —80°C. Cell-free supernatant was
assayed for protein content by the bicinchoninic acid method (#23255, Pierce
Chemical Co., Rockford, IL). Mucin glycoprotein SAC in BALE was analyzed
by ELISA using a monoclonal antibody (mucin5AC Ab-1; Neomarkers,
Fremont, CA), a peroxidase-conjugated avidin/biotin complex (ABC Reagent;
Vector Laboratories, Burlingame, CA), and a fluorescent substrate (Quanta-
Blue; Pierce Chemical, Rockford, IL). Total and ova-specific IgE were
determined in ELISA (colormetric capture assays) by coating plates with anti-
rat IgE (Pharmingen, San Diego, CA) or ova, respectively. After incubation
with BALE, bound samples were detected with biotinylated anti-rat IgE and
quantified with a peroxidase system (Vector Laboratories, Burlingame, CA)
using a Bio Tek Elx 808 plate reader.
Lung tissue collection and epithelial morphometry. The intrapulmonary
airways of the fixed left lung lobe from each of eight rodents were
microdissected. Beginning at the lobar bronchus, airways were split down
the long axis of the largest daughter branches (i.e., main axial airway; large
diameter conducting airway) through the 12th airway generation. Tissue blocks
that transverse the entire lung lobe at the level of the fifth and eleventh airway
generation of the main axial airway were excised and processed for light
microscopy and morphometric analyses. The lung tissue blocks were embedded
in paraffin, and 5- to 6-um sections from the proximal face of each block were
cut and placed on charged slides (Probe-on-plus; Fischer Scientific, Pittsburgh,
PA). Tissue sections were histochemically stained with (1) hematoxylin and
eosin (H&E) for evaluation of epithelial morphology and quantitation of
epithelial cell numeric densities, or (2) Alcian Blue (pH 2.5)/Periodic Acid
Schiff's sequence (AB/PAS) to detect acidic and neutral mucosubstances for
quantitation of stored mucosubstances within the airway epithelium.
Morphometric Analysis of intraepithelial mucosubstances in air-
way. The volume density (Vs) of stored intraepithelial mucosubstances
(IMs) in the surface epithelium lining the proximal and distal pulmonary axial
airways (generations 5 and 11, respectively) were determined using image
analysis and standard morphometric techniques previously described in detail
(Harkema et al., 1997a, b). The quantity of stored mucosubstances per unit area
were determined as described by Harkema et al. (1987a, b) and expressed as the
mean volume (nl) of ip mucosubstances (IM)/mm of basal lamina ± standard
error of the mean (Fig. 8D).
Statistical analysis of BALE and morphometric endpoint. Data de-
scribing the type and magnitude of the pulmonary inflammatory response in
BALF and morphometric changes in mucosubstances in airway epithelium
(n = 8) were expressed as the mean group value ± the SEM. The data were
subjected to ANOVA for factors of inhalation exposure type (e.g., CAPs or
filtered air) and airway sensitization/challenge (ova or saline). Significant
differences between experimental groups were identified using appropriate
post hoc tests (Tukey's omega procedure). Transformation of data (e.g., log or
arcsin^1) was performed if needed to render variances homogeneous. The
criterion for statistical significance was p < 0.05.
RESULTS
Characterization of CAPs
The average CAPs concentration during the 13-day exposure
period was 493 ±391 Hg/m3 (Table 1). As shown in Table 1,
over half of the CAPs collected in Grand Rapids were organic
and elemental carbon. Although the 13-day averaged concen-
tration of sulfate was only about 10% of the CAPs, during the
first week of the exposure study a large amount of secondary
particles composed mostly of sulfate and organic carbon was
observed (Fig. 2). The Hybrid Single Particle Lagrangian
TABLE 1
Average Mass and Major Composition of Concentrated Air
Particles during 13-Day Exposure Period
Mass
Organic carbon
Elemental carbon
Sulfate
Nitrate
Ammonium
Urban dust"
493 ± 391
244 ± 144
10 ±4
79 ± 131
39 ± 67
39 ±59
18 ±6
Note. All values are presented as mean ± SD (ug/m ).
"Crustal urban dust was estimated from Fe, Al, Ca, and Si.
Integrated Trajectories model (HYSPLIT, National Oceanic
and Atmospheric Administration) indicated regional transport
of an air mass that passed through Missouri, Illinois, and
northwestern Indiana (Fig. 2). Cities in northwest Indiana,
including Gary and East Chicago, have been home to heavy
industry for the last century.
Differentially Expressed Genes
Criteria for the pairwise analyses were FDR < 0.05 and fold
change > 1.5 for expression over background. The centroid
plot in Figure 3A is a visual representation of the relative
7/17 7/18 7/19 7/20 7/21 7/22 7/23 7/24 7/25 7/26 7/27 7/28 7/29
FIG. 2. Air mass contributing to the CAPs used for exposure. The
HYSPLIT model (National Oceanic and Atmospheric Administration)-
generated 48-h backward trajectories showing the history of an air mass
arriving in Grand Rapids at noon on July 21. (Data are from National Oceanic
and Atmospheric Administration.)
Previous
TOC
-------
COMPARATIVE MICROARRAY ANALYSIS AND PULMONARY CHANGES IN BROWN NORWAY RATS
211
A
Air-Ova
Air-Saline
CAPs-Ova
CAPs-Saline
B
Saline vs air ova Air saline vs caps ova
Caps-saline vs Caps Ova
Col1a1
Fndcl
Cdh11
Ama4
TgfbS
Ms4a2
Fbnt
lgG-2a
IgsflO
Cpa3
Fcerla
Scd1
Aqp4
Vegfc
Tpm3
Slc34a1
Abp1
Col5a1
Eln
Baspl
Cnn1
CreW2
Actg2
Rrm2
EII2
Gpr176
DnajcS
Ada
PqlcS
Tmem97
Pdia4
Cars
Lamps
Srd5a1
Lrp2
Pcbdl
Bmp3 —
PipSklb
Hkr3
Nritp
LOC259224 —
Kcne2
Lbp
C5 —
Cfi
C4bpa
Mt1a
Lcn2
ChiSM
FIG. 3. Probesets with an absolute fold change > 1.5 and a FDR < 0.05 from the three permutation analyses for differential expression comparisons were
considered significant, (total of 87 differentially expressed probesets). (A) Centroid plot of significant genes. The bars represent expression values relative to the
centroid value for each gene considering all groups. Bars extending to the right of the vertical line in each treatment group represent elevated expression of that
gene: bars extending to the left represent lower expression of that gene. Genes are plotted in descending order based on expression by the CAPs-ova treatment
group. Red = air-ova, green = air-saline, dark blue = CAPs-ova, light blue = CAPs-saline. (B) Venn diagram of all 87 significant probesets.
class-wise expression of genes that met these criteria, not includ-
ing probesets of limited or ambiguous annotation (for the full list
of genes see Supplemental Table SI). The shrunken centroids
were calculated using the method described in (Tibshirani
et al., 2002). It is clear in this centroid plot that the relative
gene expression patterns for the CAPs-ova treatment group
were opposite that of the air-saline and CAPs-saline treatment
groups and more like that seen in the air-ova treatment group.
Figure 3B shows a Venn diagram indicating the amount of
overlap between the significant probesets from pairwise
comparisons. The first pairwise comparison group studied
was the air-exposed, saline-challenged rats compared with the
air-exposed, ova-challenged rats (air-saline vs. air-ova). The 28
differentially expressed probesets in this comparison group
were all upregulated; the greatest gene upregulation was 3.39-
fold for immunoglobulin heavy chain gamma 2a (IgG-2a). The
expression of 39 probesets was significantly changed in
the CAPs-exposed, saline-challenged rats compared with the
CAPs-exposed, ova-challenged rats (CAPs-saline vs. CAPs-
ova). The greatest increase in expression was 4.95-fold for
stearoyl-coenzyme A desaturated 1 (Scdl); the greatest
decrease was —4.23—fold for complement component 5
(C5). A total of 32 probesets were differentially expressed in
the air-saline rats compared with the CAPs-exposed, ova-
challenged rats (air-saline vs. CAPs-ova). The greatest
upregulation was 3.42-fold for IgG-2a, and the greatest down-
regulation was 2.49-fold for Mtla. No significant differentially
expressed probesets were found by comparing the air-saline
and air-ova groups to the CAPs-saline and CAPs-ova,
respectively (data not shown).
The relative upregulation/downregulation of genes was
confirmed using qPCR (Fig. 4). Specifically, the PADE
analysis indicated that TGF-p3 is expressed 1.6- and 2.2-fold
higher in CAPs-ova versus air-saline and versus CAPs-saline
exposed rodents, respectively; Fcerla was expressed 2.1-fold
higher in CAPs-ova versus air-saline rodents; C4BP was
expressed 2.8-fold lower in CAPs-ova versus CAPs-saline
rodents; VEGF-C was expressed 1.6-fold higher in CAPs-ova
Previous
-------
212
HEIDENFELDER ET AL.
qPCR Confirmation of Gene Expression Changes
i
c
o
o
13
Q)
N
O
TGFB3 FCER1 C4BP VEGF.C CM3L1
Gene Symbol
Mt1a
FIG. 4. Gene expression by confirmatory qPCR. The average relative
concentration of each target gene (run in duplicate), divided by the
corresponding fi-actin endogenous control per rat, then averaged with all rats
of the same treatment group. Error bars show the standard error of the mean.
The relative expression of each gene by treatment group corresponds with
microarray data outputs (compare to Supplemental Table 1). *Significantly
different from air-saline treatment (p < 0.05); +significantly different from
CAPs-saline treatment (p < 0.05); there were no significant differences between
the air-ova & CAPs-ova groups.
versus air-saline rodents; CM3L1 was expressed 3.5-fold lower
in CAPs-ova versus CAPS-saline rodents; and Mtla was
expressed 3.3-fold lower in CAPs-ova versus CAPs-saline
rodents and 2.5-fold lower in air-ova versus air-saline rodents.
Due to small sample size, the changes for three of the six genes
were not significantly different, illustrating the danger in
relying upon measurements of single genes in studies like this.
Microarray studies allow the simultaneous consideration of
groups of functionally related genes, allowing smaller changes
to be detected (as discussed below).
Principal Components and Cluster Analysis
PCA by treatment conditions was performed on array data
using the list of 87 differently expressed probesets described
above (Fig. 5A). The saline-challenged animals were separated
from the ova-challenged animals by the first principal com-
ponent, which accounts for 55.92% of the variance between
samples. The second principal component (accounting for
15.21% of variance) separated the ova-challenged rats between
those exposed to laboratory air and those exposed to CAPs
with one exception. The air-saline and the CAPs-saline groups
were indistinguishable in the first two principal components.
Hierarchical clustering of all samples by condition, based on
the list of 87 differentially expressed probesets, separated the
animals into groups of either saline-challenged or ova-
challenged rodents (Fig. 5B). Individual probesets were nor-
malized to the median value for coloring, and clustering
indicated two main branches. The bottom branch showed lower
levels in six of the eight ova-challenged samples versus higher
A '•'-
— 0.8-
i
.2 06 -
(5
S? 0.4 -
I
£ 02-
CM
^ n n
(^ U.U
e
§. -0.2 -
0 -0.4 -
O
Q- -0.6 -
.nn
•
g
o
* •
0
0
0
• AR-OVA
O CUP-OVA
T W-SWJNE
A i CW-SALINE
A
T T
a
T
T
-15 -1.0 -0.5 0.0 0.5 10
PCA component 1 (55.9%variance)
ff**ff«18S*»S*t«
SSSSSfffsIIsis
»iv>vtv»OL«»525**fji^f;
i«o,aa<«AB f *? A f i f
»*slfii***a ?»*? *^
I! I
co v> tr>
> i
1 5
I
I '
I
'• ••
.
I
I •
I •
I
I i
!
FIG. 5. Principal components analysis and hierarchical clustering of
differentially expressed probesets. The list of 87 differentially expressed
probesets was used for the analyses. (A) Two dimensional principal
components analysis by conditions of treatment, x-axis: component 1,
accounting for 55.92% of variance; y-axis: component 2, accounting for
15.21% of variance. Component 1 strongly separates the ova-challenged from
the saline-challenged rats. (B) Hierarchical clustering of treatment conditions
and genes. Expression levels were derived from REDI-adjusted and RMA-
normalized signal values, which were then normalized to the per gene median
in order to achieve the color scale (red indicates a high level of expression,
green indicates a low level). Clustering shows that the rats fall into two main
groups: ova-challenged and saline-challenged. The blue bracket indicates
probesets in a subset of the top branch that are highly expressed in all of the
CAPs-ova exposed rats and in one rat each from the air-ova and the CAPs-
saline exposure groups.
expression levels in all but one of the saline-challenged
animals. The top branch was the reverse: it showed mostly
higher expression in the ova-exposed animals and lower
expression values in all but one saline-exposed animal. One
gene, Scdl, is shown between these two groups.
Previous
TOC
-------
COMPARATIVE MICROARRAY ANALYSIS AND PULMONARY CHANGES IN BROWN NORWAY RATS
213
Functional Analysis
Functional analysis to screen the differentially expressed
genes in the three group comparisons was performed in
Metacore using a larger number of genes (air-saline vs. air-ova =
75, CAPs-saline vs. CAPs-ova = 93, and air-saline vs. CAPs-
ova = 200) obtained via a relaxed FDR cutoff of <0.2
(Supplemental Table S2). The most significant Metacore maps
(corresponding to known biological pathways curated by
GeneGo) of the air-saline versus air-ova comparison were split
into two main groups: nucleotide metabolism (maps 1-2, p =
1.279 X 10~2, 2.445 X 10~2, respectively) and immune
response/histamine metabolism (maps 3, 4, 6, 8, p value
range = 3.227 X 10~2 to 4.501 X 10~2) (data not shown).
Maps 1 and 5 for the CAPs-saline vs. CAPs-ova comparison
were related to cell adhesion and extracellular matrix (ECM)
remodeling (p = 2.726 X 10~4, 8.206 X 10~3, respectively),
whereas maps 2-A were related to immune response via
complement pathways (p value range = 8.324 X 10~4 to
6.367 X 10~3) (data not shown). The top 10 maps for the air-
saline versus CAPs-ova comparison also broke into two main
groups. Maps 1-3 were related to immune response via
complement pathways (p value range = 3.229 X 10~5 to 9.242 X
10~5), and maps 4-9 related to remodeling and cell adhesion
(p value range = 1.841 X 10~4 to 3.405 X 10~3) (data not
shown). GO processes for this comparison broke into similar
groups, with the first and second processes relating to lung
development (p = 4.0489 X 10~7 and 4.5838 X 10~7) and
processes 3-7 relating to the inflammatory response (p value
range = 6.0503 X 10~7 to 2.2709 X 10~5) (data not shown).
The most statistically significant gene ontology processes
across all genelists are shown in Figure 6. The top three relate to
tissue development and the fourth to the defense response (Fig.
6). It is of note that the top four categories were much more
significant in the presence of CAPs than with ova treatment
alone. Air-saline versus CAPs-ova had the most significant
relationship to lung/respiratory tube development and defense
response indicating both inflammatory responses as well as
remodeling in response to this treatment. The CAPs-saline
versus CAPs-ova comparison had the most significant relation-
ship with skeletal development and the differentially expressed
genes in the skeletal development category were predominantly
regulated via TGF-p. This, along with the significant
enrichment for the "regulation of TGF-p receptor signaling"
category (Fig. 6), suggests that changes in TGF-p signaling are
particularly significant when comparing these two groups. The
expression changes for TGF-p3 (Fig. 3) may explain this effect
in that transcription is reduced by CAPs exposure in the
absence of ova challenge but not in the presence of ovalbumin.
12
15
-log(pValue)
1. skeletal development
2. lung development
3. respiratory tube development
4. defense response
5. embryonic neurocranium morphogenesis
6. epithelial to mesenchymal transition
7. regulation of transforming growth
factor beta receptor signaling
pathway
8. negative regulation of phagocytosis
9. negative regulation of release of
sequestered calcium ion into cytosol
10. defense response to fungus,
incompatible interaction
• Processes
I^^Bi = air-saline vs. air-ova
^f^m = CAPs-saline vs. CAPs-ova
^^^m = air-saline vs. CAPs-ova
FIG. 6. Overrepresented gene ontology categories from Metacore for all three treatment comparisons, y-axis shows the top 10 GO processes sorted by the
lowest p value for any single treatment comparison. Longer bars correspond to lower p values for enrichment of genes in a GO category within the differentially
expressed genes in each treatment contrast with p values ranging from 1CT3 to 1CF15. Categories 1 and 7 were most significantly enriched in the CAPs-saline
versus CAPs-ova treatment. Categories 2-4, 6, and 8-10 were most significantly enriched in the air-saline versus CAPs-ova treatment. Category 5 was most
significantly enriched in the air-saline versus air-ova treatment. Top, air-saline versus air-ova, middle—CAPs-saline versus CAPs-ova, bottom—air-saline versus
CAPs-ova. p Values reported by MetaCore are not corrected for multiple testing.
Previous
-------
214
HEIDENFELDER ET AL.
The only genes that overlap between the skeletal and lung/
respiratory tube development processes that are found in these
two genelists were TGF-pl and TGF-p3, however several of
the other genes from the skeletal development process (Fbnl,
CollAl, CollA2, and Cdhll) could be involved in lung/
respiratory tube development as well, p Values reported by
MetaCore are not corrected for multiple testing.
Network Analysis
The CAPs-ova synergy was investigated further with
pathways analysis using Ingenuity IPA. The highest-scoring
network generated for air-saline versus CAPs-ova had top
functions of cellular movement, immune response, cell-to-cell
signaling and interaction (Fig. 7) and matched the results seen
with the MetaCore analysis (Fig. 6). The genes in this network
can be categorized as either related to remodeling (orange
circle), inflammation (purple circle) or both. Overlaying data
relating only to air-saline versus air-ova onto this network
show three genes (FcerlA, Ms4a2, Bcl6b) related to in-
flammation (Albrecht et al, 2004; Galli et al, 2008), two genes
(cadherin 11 [Cdhll] and TGF-p3) that fall under both
inflammation and remodeling (Broide, 2008), and one gene
(elastin [Eln]) that is consistent with remodeling only (Shi
et al., 2007). Overlaying data relating only to the air-saline
versus CAPs-ova comparison show a dramatic increase in the
Remodeling
Air+ova
CAPs+ova
1
9
Both
2
13
Inf
3
10
Inflammat
FIG. 7. Gene network derived from FADE data from the air-saline versus CAPs-ova comparison was uploaded into Ingenuity IPA. Cutoffs were set at > 1.5-fold
for fold change and < 0.2 for FDR. This network represents the highest-scoring network (score = 61), with genes involved in either inflammatory (purple circle) or
remodeling (orange circle) mechanisms. For nodes, red = upregulated expression and green = downregulated expression in the CAPs-ova versus air-saline
comparison. Bars adjacent to each node represent statistically significant expression (upregulated or downregulated) for air-saline versus air-ova (left) and air-saline
versus CAPs-ova (right). The table (inset) shows the number of genes falling in each class showing the enhancement of effects by coexposure to CAPs. Note that three
nodes (Sod, Vegf, G-protein beta) in the network correspond to gene families rather than genes and were not counted in the totals, (see Supplemental Table 4).
Previous
TOC
-------
COMPARATIVE MICROARRAY ANALYSIS AND PULMONARY CHANGES IN BROWN NORWAY RATS
215
number of the genes falling into inflammation, remodeling, or
both (Fig. 7, inset).
The genes in the remodeling category that do not overlap with
inflammation are predominantly upregulated, indicating an
increase in pathways that impact remodeling. Expression levels
of the remaining genes in the remodeling category and those in
the inflammation category are mixed. For example, the IgE
receptors FCER1A and MS4A2/FCER1B and the cytokine ILIA
are upregulated as expected, whereas genes (Mmpl2 and CCL5/
RANTES) that have been shown to be upregulated in asthma are
downregulated in this study. This is possibly due to mechanisms
such as feedback inhibitory signaling at the level of transcription
and may not reflect a decrease in the associated protein. The
inflammatory and remodeling genes PLAU, PLAUR, and Mmp9
were found in a screen for genes related to COPD (Wang et al.,
2007), and both PLAU and MMP9 have been implicated in
asthma (Begin et al., 2007; Sampsonas et al., 2007). Thus, these
three genes were included in the network even though their fold
changes and/or FDRs did not meet cutoff values. PLAU came
very close to significance, with an FDR of 0.146 and a fold
MucSAC
Total Protein
Air-saline
CAPs-
saline
Air-ova CAPs-ova
B
«
50
45
40-
35-
30-
25-
20-
15-
10-
B-
BALF Cell Summary
JIT
Air-saline CAPs-
saline
Air-ova CAPs -ova
| • Eoslnopnils a Neulrophils o Lymphocytes
BALF Cell Summary
Gene Expression Function by
Treatment
Air-saline CAPs.
saline
Air-ova CAPs-ova
CAPs-saline Air-Ova
CAPs- Ova
I Macrophages & Monocytes
D Total Cells
I Remodeling
D Inflammation
Airway Morphometric Determinations of
Volume Density
7 -
air-saline CAPs- Air-Ova CAPS-ova
saline
[• Proximal (Generation 5) D Distal (Generation 11) |
FIG. 8. Effect of ova sensitization and challenge (with and without coexposure to CAPs) on biochemical and cellular parameters in the BALF and gene
expression and airway morphometry in lung tissue. (A) IgE Mucin SAC and total Protein measured in BALF; (B) BALF cell differential; (C) number of genes
significantly changed in each treatment group (n = 4) by functional category. Data represent the group mean (n = 8) and SEM in each treatment group for (A, B,
D). *Significantly different from air-saline group (p < 0.05); +significantly different from air-ova group (p < 0.05). (D) Morphometric determinations of the
volume density of AB/PAS-stained IMs lining the axial airways of the left lobe (Proximal airway generation 5 and distal airway generation 11). ^Significantly
different from CAPs-saline group (p < 0.05). Volume density (Vs; nl/mm2).
Previous
TOC
-------
216
HEIDENFELDER ET AL.
change of — 1.3 in air-saline versus CAPs-ova group. Overall, the
network analysis shows a perturbation of inflammatory genes in
rodents challenged with allergen. Moreover, genes involved in
the remodeling of lung tissue are altered by CAPs exposure only
in the presence of allergen challenge.
BALF
Figure 8 contains the results of the biochemical and cellular
analyses of the BALF and gene expression and lung morpho-
metric determinations from each of the exposure groups
(Supplemental Table S3 contains the BALF data in table form).
Exposure to CAPs for 13 consecutive days produced no changes
in BALF proteins or cellularity in saline-challenged rats.
Sensitization and challenge with ova caused increases in total
cells (fourfold), neutrophils (15-fold), lymphocytes (16-fold),
eosinophils (fourfold) (Fig. 8B), and mucin glycoprotein
(mucSAC) (threefold) (Fig. 8A). CAPs exposure did not affect
ova-induced increases in BALF cellularity, but enhanced ova-
induced increase in mucin glycoprotein by 50% (Fig. 8A).
Furthermore, only ova-challenged rats exposed to CAPs had
significant elevations in total BALF protein. CAPs exposure also
enhanced ova-induced increases in ova-specific IgE in BALF
(Fig. 8A). The gene expression changes reflected this same trend
(Fig. 8C) with the number of inflammatory genes increasing
from 5 (OVA alone) to 23 (CAPS plus OVA) and the number of
remodeling genes increasing from 3 (OVA alone) to 22 (CAPS
and OVA). Total IgE in BALF was not significantly increased by
CAPs exposure or ova challenge. The small sample size for this
study did not provide the power required to explicitly test for
correlation, but the cellular & protein changes are consistent
with the gene expression results (Figs. 7 and 8C) with the caveat
that the cellular changes are less sensitive with regard to the
enhancement of inflammation and remodeling by CAPs.
The number of eosinophils in the BALFs from our ova-
sensitized and saline-challenged BN rats were higher than
routinely seen in other strains/stocks of rats (e.g., F344,
Sprague-Dawley), but not uncommon for the BN strain of rat,
and well below previously published reports in non-infected
BN rats that were similarly ova-sensitized and saline
challenged (Noritake et al., 2007). The levels of both
eosinophils and neutrophils in the BALFs of our ova-
challenged rats were markedly and significantly elevated
compared with our saline-challenged controls, indicative of
an allergen-induced hypersensitivity response.
Pulmonary Pathology IMorphometry
Light microscopic examination of the selected lung sections
from all the rats in this study, substantiated and correlated well
with the analytical observations of the BALF from these same
rats. Furthermore, significant morphometric differences among
the groups, described below, correlated with our findings in the
BALF and histopathology. The lungs of control rats (ova-
sensitized and saline-challenged rats) had a minimal influx of
eosinophils widely scattered along with a few mononuclear
leukocytes (e.g., lymphocytes, plasma cells) in the interstitium
surrounding some of the bronchiolar airways, In addition, there
were occasional small foci of eosinophils and mononuclear
cells (macrophages, monocytes, and lymphocytes) in the
alveolar parenchyma of these control rats. These background
findings are consistent with previous reports on the normal
pulmonary morphology of pathogen-free BN rats (Noritake
et al., 2007). In contrast, the lungs of BN rats sensitized and
challenged with ova had conspicuous allergic bronchiolitis and
alveolitis (allergic bronchopneumonia). There was a noticeable
proximal to distal decrease in the severity of the bronchiolitis in
the left lung lobe of these rodents with more inflammatory
lesions in the proximal lung section (G5 axial airway level;
closest to the hilus of the lung lobe) compared with the more
distal section (Gil axial airway level). Ova-induced inflamma-
tory and epithelial lesions in the conducting airways involved
the large diameter, proximal axial airways and the small
diameter, distal pre-terminal and terminal airways. Inflamma-
tory and epithelial lesions were usually more severe in the more
proximal axial bronchioles compared with those in the more
distal pre-terminal and terminal bronchioles. Ova-induced
bronchiolitis was characterized by peribronchiolar edema
associated with a mixed inflammatory cell influx of eosinophils,
lymphocytes, plasma cells, and occasional neutrophils (Fig. 9).
Bronchiole-associated lymphoid tissues in the air-ova group
airways were also enlarged due to lymphoid hyperplasia
relative to the air-saline group. Perivascular interstitial
accumulation of a similar mixture of eosinophils and mono-
nuclear cells, along with perivascular edema, were also present
in the lungs of air-ova rats (i.e., surrounding pulmonary arteries
adjacent to bronchioles and pulmonary veins scattered
throughout the alveolar parenchyma).
Air-ova rats had a mild-marked epithelial hypertrophy and
mucous cell metaplasia/hyperplasia (MCM) with increased
amounts of AB/P AS-stained mucosubstances in the mucous cells
within the surface epithelium (i.e., IMs) lining the affected large
diameter bronchioles, including the proximal and distal axial
airways (Fig. 9). Saline-challenged rats (the air-saline group) had
mucous cells with less IM compared with the ova-challenged (air-
ova) rats (Fig. 10). There was no significant difference in the
amounts of IM between air-saline and CAPs-saline rats (Fig. 8D).
In addition to the perivascular and peribronchiolar lesions,
there were varying sized focal areas of allergic alveolitis in the
lung parenchyma of the air-ova rats. These alveolar lesions
were characterized by accumulations of large numbers of
alveolar macrophages, epithelioid cells, and eosinophils, with
lesser numbers of lymphocytes, monocytes, and plasma cells,
in the alveolar airspace. Often the alveolar septa in these areas
of alveolitis were thickened due to type II pneumocyte
hyperplasia and hypertrophy, intracapillary accumulation of
inflammatory cells, and capillary congestion.
Ova-challenged rats exposed to CAPs (the CAPs-ova group)
had a more severe allergic bronchopneumonia than the air-ova
group. This was reflected both in the severity and distribution of
Previous
TOG
Next
-------
COMPARATIVE MICROARRAY ANALYSIS AND PULMONARY CHANGES IN BROWN NORWAY RATS
A B
217
n ••:-• '
—^r>^ v
FIG. 9. Light photomicrographs of the respiratory epithelium (e) lining the proximal axial airway (generation 5) in the left lung lobe of rats exposed to (A) air/
saline (control), (B) CAPs/saline, (C) air/ova, or (D) CAPs/ova. Compared with the normal airway in the control rats (A), significant morphologic changes are
present only in rats challenged with ova (C, D). The most prominent histologic changes due to ova challenge included a thickened, hypertrophic, respiratory
epithelium with increased numbers of mucous goblet cells, and a mixed inflammatory cell infiltrate (asterisks) consisting mainly of lymphocytes, plasma cells and
widely scattered eosinophils (arrows) in the interstitium (i) of the airway wall. These histologic airway changes are slightly greater in the CAPs/ova rats
(D) compared with those in the air/ova rats (C). All tissues stained with H&E. Sm, smooth muscle.
the allergic bronchiliolitis and alveolitis. Air-ova rats had a mild
to moderate allergic bronchopneumonia with the inflammatory
and epithelial lesions in approximately one fourth to one third of
the lung lobe. In contrast, CAPs-ova rats had a moderate to
marked bronchopneumonia with lesions in approximately one
half or more of the lung lobe. CAPs-ova exposed rats also had
more severe MCM in the epithelium lining the large diameter
axial airways compared with the air-ova group.
Thirteen consecutive days of CAPs exposure in saline-
challenged rats did not cause changes in the morpheme trie ally
measured amounts of stored mucus in airway epithelium in
either proximal or distal airways (Fig. 8D). The only significant
increases in IMs were found in CAPs-ova rats (Fig. 8D). This
CAPs-induced enhancement of ova-induced IM correlated with
the increased amounts of muc5AC in BALF from these animals
(Fig. 8A). There was a trend (though not statistically different)
for a CAPs-related exacerbation of ova-induced increases in the
stored intraepithelial AB/PAS-stained mucosubstances in both
the proximal and distal axial airways of the left lung lobe
(quantitative estimate of the severity of MCM).
DISCUSSION
Asthma is a complex airway disease involving gene-
environment interactions. This study used comparative micro-
array analysis to investigate the effect that CAPs have on a BN
rodent model of airway allergic inflammation. An enhanced
effect on the type and magnitude of gene expression changes
was observed with ova-challenged BN rats coexposed with
CAPs. The genomic data were consistent with the observed
histopathology. This combination of allergen challenge and
CAPs exposure led to differential expression of genes related to
airway remodeling that was not observed in rodents treated
with either allergen or CAPs alone.
In agreement with the data presented here, air particulates
have been shown to have an adjuvant effect on allergic
reactions as determined by changes in histopathology. For
example, BN rats exposed to both CAPs and ova showed
adjuvant effects which resulted in increased airway mucus,
mucin glycoprotein mucSAC, pulmonary inflammation, and
airway epithelial remodeling (e.g., MCM) (Harkema et al.,
2004a). Other studies point to an adjuvant effect of diesel
particulates on ova-challenged rodents which resulted in
enhanced airway hyperresponsiveness and pulmonary inflam-
mation (Dong et al., 2005; Matsumoto et al., 2006; Miyabara
et al., 1998; Takano et al., 1997). At the molecular level,
exposure to diesel particulates in ragweed-sensitized humans
resulted in an increase in mRNA transcripts of asthma-related
cytokines (Diaz-Sanchez et al., 1997).
CAPs exposure alone had no significant effect on gene
expression in the lung in this study. ApoE/LDL double
Previous
TOC
-------
218
HEIDENFELDER ET AL.
e
sm
B
e -
sm
50um
/
•
4 •*?***
sm
sm
<
FIG. 10. Light photomicrographs of AB/PAS-stained mucosubstances (arrows; dark magenta stain) in mucous goblet cells of the respiratory epithelium (e)
lining the proximal axial airway (generation 5) in the left lung lobe of rats exposed to (A) air/saline (control), (B) CAPs/saline, (C) air/ova, or (D) CAPs/ova.
Compared with the normal airway in the control rat(A), significant increases in the amount of IMs are present only in rats challenged with ova (C, D). This ova-
induced intraepithelial change is slightly greater in the CAPs/ova rats (D) compared with that in the air/ova rats (C). All tissues stained with AB (pH 2.5)/PAS
sequence to detect acidic and neutral mucosubstances. Sm, smooth muscle; I, interstitial layer of the airway wall.
knockout mice exposed to a longer period of inhaled CAPs
resulted in no significant changes in gene expression in lung
tissue (Gunnison and Chen, 2005). Another study in which
mice were intranasally instilled with CAPs showed increased
gene expression in several cytokines and an increase in
polymorphonuclear cells in BALF (Sigaud et al., 2007),
however the authors note the high dose of particles used for
instillation was a useful for proof-of-principle but not applicable
to realistic exposures. A recent toxicogenomics study found
differential gene expression upon inhalation exposure of
BALB/c mice to DE (Stevens et al., 2008) in the presence or
absence of ova. The genes identified in that study were involved
in immune function, cell signaling, and metabolic and oxidative
stress response. Taken together, this suggests that the response
is dependent on the source of air pollutants, individual
susceptibility and possibly route and duration of exposure.
Ova challenge alone resulted in the largest changes in
magnitude of gene expression, albeit for a limited number of
genes, in the present study (Fig. 5B). Two prominent genes
upregulated by ova challenge were Fcerla and Ms4a2/Fcerlb;
receptors for IgE that factor into allergy mediated events, In the
present study and in previous studies, cytological, and
immunohistochemical methods for assessing ova-induced
allergic airway inflammation in BN rats showed responses
common to asthma such as increased levels of airway mucous
glycoproteins, MCM in bronchiolar epithelium, increased
numbers of eosinophils, neutrophils, and lymphocytes in both
lung tissue and BALF, and increased IgE in the serum (Salmon
et al., 1999; Tarayre et al., 1992; Underwood et al., 1995).
Gene expression studies in ova-challenged mice have impli-
cated a variety of genes that relate to inflammation, airway
hyperresponsiveness, atopy, or mucus production and may
impact asthma pathogenesis (Follettie et al., 2006; Izuhara and
Saito, 2006; Kuperman et al, 2005; Walker et al, 2006). Our
study is consistent with these previous reports in that they all
implicate genes involved in inflammation.
In contrast to the results seen with ova alone, the effects
specifically due to CAPs/ova coexposure were more subtle
though much more widespread with regard to number of genes
effected In the current study, CAPs exposure in the ova-
challenged rats led to expression changes in remodeling genes,
and a larger number of genes related to inflammation compared
with ova challenge alone. Inflammation and remodeling
pathways centered predominantly on TGF-pl (Fig. 7). Two
other growth factors, TGF-p3 and VEGF-C, were also
identified in the network. Members of both the TGF family
and the VEGF family are thought to be important for airway
remodeling and inflammation (Lloyd and Robinson, 2007).
There were 23 TGF-p related inflammatory genes significantly
changed with CAPs-ova coexposure versus five with ova
exposure alone (Figs. 7 and 8C). This corresponded with
a significant increase in the amount of ova-specific IgE found
in the BALF from CAPs-ova rats in our study when compared
with either the air-saline or air-ova groups (Fig. 8A). This is
Previous
-------
COMPARATIVE MICROARRAY ANALYSIS AND PULMONARY CHANGES IN BROWN NORWAY RATS
219
further evidence for an enhancement of ova challenge by CAPs
via increased allergy mediators.
Airway remodeling is a hallmark of asthma referring to
changes in structural cells, including the thickening of airway
walls (Lloyd and Robinson, 2007). The findings of both
bronchiolar and alveolar wall thickening in air-ova and CAPs-
ova rats in the present study (due to epithelial hypertrophy and
MCM) are characteristic histopathologic features of allergic
airway disease that mimic those reported in asthmatic airways
of humans. The MCM in the epithelium lining the large
diameter axial airways was more severe in the CAPs-ova rats
compared with the air-ova rats, consistent with the gene
expression data implicating remodeling events in the CAPs-ova
exposure group. Significant increases in IM and total BALF
protein and elevated mucSAC were apparent only in the CAPs-
ova group. Taken together, pulmonary histopathology (sub-
jective assessment of inflammatory and epithelial responses),
morphometric measurements (quantitative assessment of airway
epithelial remodeling), and BALF analyses (quantitative
assessment of inflammatory and mucosecretory responses)
consistently demonstrated that CAPs exposure enhanced the
allergic airway disease. Typically, inflammation and remodel-
ing processes are viewed separately, however, there is emerging
data that suggest that smooth muscle changes can contribute
directly to proinflammatory changes that may perpetuate airway
inflammation and the development of airway remodeling
(Broide, 2008). Collectively, our data support this conclusion.
In total, 22 TGF-p related genes implicated in remodeling were
significantly up- or downregulated in response to CAPs-ova
coexposure (Figs. 7 and 8C). Several of these are components of
the ECM—fibrinogen, vitronectin, nidogen 1, fibrillin 1, elastin,
and matrix Gla protein, and three genes are related to smooth
muscle and/or the cytoskeleton—calponin 1, alpha 2 actin, and
tropomyosin 3. Altered ECM protein profiles in asthmatic lungs
result in an increase in the proliferation rate of airway smooth
muscle (ASM) cells, a feature of remodeling that causes the
thickening of airway walls (Johnson et al., 2004). TGF-pl can
increase the proliferation of smooth muscle cells and the deposition
of ECM proteins, whereas the smooth muscle cells aid in the
downregulation of ECM-degrading matrix metalloproteinases
such as Mmp9 and Mmpl2 (Parameswaran et al., 2006).
Because inflammatory cells in the BALF were not
significantly different between the CAPs-ova and CAPs-saline
groups, the changes in gene expression are most likely due to
an intracellular transcriptional response from either the in-
flammatory cells, the resident alveolar cells, or both. Our
results cannot distinguish whether the transcriptional changes
shown in Figure 7 represent signaling in a single cell type or
integrated responses across multiple cell types, but it does
provide clues to the signaling changes taking place in response
to ova and CAPs coexposure. TGF-pl and TGF-p2 have
previously been causally linked to allergen-induced prolifera-
tion of ASM cells and mucus-producing goblet cell hyperplasia
(Lloyd and Robinson, 2007; McMillan et al., 2005). TGF-p2
treatment of primary bronchial epithelial cells in culture
produced significant increases in MUCSAC mRNA and protein
levels as well as elevated mucin as measured by AB/PAS
staining (Chu et al., 2004). The results from the current study
show more pronounced changes in TGF-p3 transcript levels
suggesting that TGF-p3 may also play a key role in lung
remodeling. This is supported by reports of delays in
pulmonary development in TGF-p3 knockout mice (Kaartinen
et al., 1995). This study supports previous findings of TGF-p
involvement in lung remodeling, implicates a third member of
this gene family in the process, provides clues as to the
downstream mediators of this process, and shows an
enhancement of these effects upon coexposure to CAPs.
A concern from an air regulatory point of view is whether
allergic individuals are more sensitive to adverse effects from
exposure to air PM than nonallergic individuals. Application of
new genomic technologies together with more traditional
lexicological approaches provide a means to begin to address
links among inhalation exposures to fine particles, genetic and
metabolic changes in the lung cells in response to these
exposures, and aggravation of the symptoms of asthma. The
integration of these approaches can add important new in-
formation to the knowledge base regarding specific asthma
mechanisms at more environmentally relevant doses of PM2.5 to
help ensure adequate protection against adverse health effects,
particularly for susceptible individuals.
Asthma is a disease that results from a variety of environmental
factors acting on a background of genetic factors. There are likely
subtypes of asthmatics with varying susceptibility to a wide array
of environmental exposures. Using a rodent model of allergic
airway disease, this study explores the application of emerging
toxicogenomic tools in conjunction with bronchoalveolar lavage,
pulmonary pathology and morphometric analyses to investigate
the multifactorial etiology of allergy induced airway inflamma-
tion using ova-sensitized and challenged BN rats exposed to
environmentally relevant CAPs. By using a sensitive model such
as this, the mechanistic basis for the environmental influence can
be more easily characterized. Although the contribution of
genetics remains uncharacterized, this study provides a frame-
work for interpreting human studies which incorporate the
genetic variability most relevant for risk assessment.
SUPPLEMENTARY DATA
Supplementary data are available online at http://toxsci.
oxfordjournals .org/.
FUNDING
The research described in this paper was funded wholly by
the United States Environmental Protection Agency
(EPD07069 to jh and ED-D-07-017 for Expression Analysis).
Previous
TOG
Next
-------
220
HEIDENFELDER ET AL.
ACKNOWLEDGMENTS
We would like to thank Drs Susan Hester, Ian Gilmour, and
Markey Johnson for their helpful comments and careful review
of this manuscript. We also acknowledge Expression Analysis,
Durham, NC, for chip hybridization and preliminary analyses.
This research was subjected to review by the National Health
and Environmental Effects Research Laboratory and approved
for publication. Approval does not signify that the contents
necessarily reflect the views and policies of the Agency nor
does mention of trade names or commercial products constitute
endorsement or recommendation for use.
REFERENCES
Abadie, A., and Prouvost-Danon, A. (1980). Specific and total IgE responses to
antigenic stimuli in Brown-Norway, Lewis and Sprague-Dawley rats.
Immunology 39, 561-569.
Albrecht, E. A., Chinnaiyan, A. M., Varambally, S., Kumar-Sinha, C, Barrette, T. R.,
Sarma, J. V., and Ward, P. A. (2004). C5a-induced gene expression in
human umbilical vein endothelial cells. Am. J. Pathol. 164, 849-859.
Becker, S., Soukup, J. M., and Gallagher, J. E. (2002). Differential particulate
air pollution induced oxidant stress in human granulocytes, monocytes and
alveolar macrophages. Toxicol. In Vitro 16, 209-218.
Begin, P., Tremblay, K., Daley, D., Lemire, M., Claveau, S., Salesse, C.,
Kacel, S., Montpetit, A., Becker, A., Chan-Yeung, M., et al. (2007).
Association of urokinase-type plasminogen activator with asthma and atopy.
Am. J. Respir. Crit. Care Med. 175, 1109-1116.
Broide, D. H. (2008). Immunologic and inflammatory mechanisms that drive
asthma progression to remodeling. /. Allergy Clin. Immunol. 121, 560- 570;
quiz 571-572.
Busse, W. W., and Mitchell, H. (2007). Addressing issues of asthma in inner-
city children. /. Allergy Clin. Immunol. 119, 43^9.
Chu, H. W., Balzar, S., Seedorf, G. J., Westcott, J. Y., Trudeau, J. B., Silkoff, P.,
and Wenzel, S. E. (2004). Transforming growth factor-beta2 induces bronchial
epithelial mucin expression in asthma. Am. J. Pathol. 165, 1097-1106.
Diaz-Sanchez, D., Tsien, A., Fleming, J., and Saxon, A. (1997). Combined
diesel exhaust particulate and ragweed allergen challenge markedly enhances
human in vivo nasal ragweed-specific IgE and skews cytokine production to
a T helper cell 2-type pattern. /. Immunol. 158, 2406-2413.
Dong, C. C., Yin, X. J., Ma, J. Y., Millecchia, L., Wu, Z. X., Barger, M. W.,
Roberts, J. R., Antonini, J. M., Dey, R. D., and Ma, J. K. (2005). Effect of
diesel exhaust particles on allergic reactions and airway responsiveness in
ovalbumin-sensitized brown Norway rats. Toxicol. Sci. 88, 202-212.
Follettie, M. T., Ellis, D. K., Donaldson, D. D., Hill, A. A., Diesl, V.,
DeClercq, C., Sypek, J. P., Dorner, A. J., and Wills-Karp, M. (2006). Gene
expression analysis in a murine model of allergic asthma reveals overlapping
disease and therapy dependent pathways in the lung. Pharmacogenomics J.
6, 141-152.
Galli, S. J., Tsai, M., and Piliponsky, A. M. (2008). The development of
allergic inflammation. Nature 454, 445^-54.
Gilmour, M. L, Jaakkola, M. S., London, S. J., Nel, A. E., and Rogers, C. A.
(2006). How exposure to environmental tobacco smoke, outdoor air
pollutants, and increased pollen burdens influences the incidence of asthma.
Environ. Health Perspect. 114, 627-633.
Gunnison, A., and Chen, L. C. (2005). Effects of subchronic exposures to
concentrated ambient particles (CAPs) in mice. VI. Gene expression in heart
and lung tissue. Inhal. Toxicol. 17, 225-233.
Harkema, J. R., Barr, E. B., and Hotchkiss, J. A. (1997a). Responses of rat nasal
epithelium to short- and long-term exposures of ozone: Image analysis of
epithelial injury, adaptation and repair. Microsc. Res. Tech. 36, 276-286.
Harkema, J. R., Hotchkiss, J. A., and Griffith, W. C. (1997b). Mucous cell
metaplasia in rat nasal epithelium after a 20-month exposure to ozone:
A morphometric study of epithelial differentiation. Am. J. Respir. Cell. Mol.
Biol. 16, 521-530.
Harkema, J. R., Keeler, G., Wagner, J., Morishita, M., Timm, E., Hotchkiss, J.,
Marsik, F., Dvonch, T., Kaminski, N., and Barr, E. (2004a). Effects of
concentrated ambient particles on normal and hypersecretory airways in rats.
Res. Rep. Health Eff. Inst. 1- 68; discussion 69-79.
Harkema, J. R., Keeler, G., Wagner, J., Morishita, M., Timm, E., Hotchkiss, J.,
Marsik, F., Dvonch, T., Kaminski, N., and Barr, E. (2004b). HEI Research
Report. Effects of inhaled urban air particulates on normal and hyper-
secretory airways in rats. Health Effects Institute, Cambridge, MA.
Harkema, J. R., Plopper, C. G., Hyde, D. M., and St George, J. A. (1987a).
Regional differences in quantities of histochemically detectable mucosub-
stances in nasal, paranasal, and nasopharyngeal epithelium of the bonnet
monkey. /. Histochem. Cytochem. 35, 279-286.
Harkema, J. R., Plopper, C. G., Hyde, D. M., St George, J. A., and Dungworth, D. L.
(1987b). Effects of an ambient level of ozone on primate nasal epithelial
mucosubstances. Quantitative histochemistry. Am. J. Pathol. 127, 90-96.
Izuhara, K., and Saito, H. (2006). Microarray-based identification of novel
biomarkers in asthma. Allergol. Int. 55, 361-367.
Izzotti, A., Bagnasco, M., Cartiglia, C., Longobardi, M., Balansky, R. M.,
Merello, A., Lubet, R. A., and De Flora, S. (2005). Chemoprevention of
genome, transcriptome, and proteome alterations induced by cigarette smoke
in rat lung. Eur. J. Cancer 41, 1864-1874.
Johnson, P. R., Burgess, J. K., Underwood, P. A., Au, W., Poniris, M. H.,
Tamm, M., Ge, Q., Roth, M., and Black, J. L. (2004). Extracellular matrix
proteins modulate asthmatic airway smooth muscle cell proliferation via an
autocrine mechanism. /. Allergy Clin. Immunol. 113, 690-696.
Kaartinen, V., Voncken, J. W., Shuler, C., Warburton, D., Bu, D.,
Heisterkamp, N., and Groffen, J. (1995). Abnormal lung development and
cleft palate in mice lacking TGF-beta 3 indicates defects of epithelial-
mesenchymal interaction. Nat. Genet. 11, 415^-21.
Keeler, G. J., Morishita, M., Wagner, J. G., and Harkema, J. R. (2007).
Characterization of urban atmospheres during inhalation exposure studies in
Detroit and Grand Rapids, Michigan. Toxicol. Pathol. 35, 15-22.
Keller, M. B., and Lowenstein, S. R. (2002). Epidemiology of asthma. Semin.
Respir. Crit. Care Med. 23, 317-329.
Kleeberger, S. R., and Peden, D. (2005). Gene-environment interactions in
asthma and other respiratory diseases. Annu. Rev. Med. 56, 383^1-00.
Kooter, L, Pennings, J., Opperhuizen, A., and Cassee, F. (2005). Gene
expression pattern in spontaneously hypertensive rats exposed to urban
particulate matter (EHC-93). Inhal. Toxicol. 17, 53-65.
Kuperman, D. A., Lewis, C. C., Woodruff, P. G., Rodriguez, M. W.,
Yang, Y. H., Dolganov, G. M., Fahy, J. V., and Erie, D. J. (2005). Dissecting
asthma using focused transgenic modeling and functional genomics.
/. Allergy Clin. Immunol. 116, 305-311.
Leikauf, G. D., McDowell, S. A., Wesselkamper, S. C., Miller, C. R.,
Hardie, W. D., Gammon, K., Biswas, P. P., Korfhagen, T. R.,
Bachurski, C. J., Wiest, J. S., et al. (2001). Pathogenomic mechanisms for
particulate matter induction of acute lung injury and inflammation in mice.
Res. Rep. Health Eff. Inst. 5- 58; discussion 59-71.
Lloyd, C. M., and Robinson, D. S. (2007). Allergen-induced airway
remodelling. Eur. Respir. J. 29, 1020-1032.
London, S. J. (2007). Gene-air pollution interactions in asthma. Proc. Am.
Thome. Soc. 4, 217-220.
Matsumoto, A., Hiramatsu, K., Li, Y., Azuma, A., Kudoh, S., Takizawa, H.,
and Sugawara, I. (2006). Repeated exposure to low-dose diesel exhaust after
Previous
TOC
Next
-------
COMPARATIVE MICROARRAY ANALYSIS AND PULMONARY CHANGES IN BROWN NORWAY RATS
221
allergen challenge exaggerates asthmatic responses in mice. din. Immunol.
121, 227-235.
McCunney, R. J. (2005). Asthma, genes, and air pollution. /. Occup. Environ.
Med. 47, 1285-1291.
McMillan, S. J., Xanthou, G., and Lloyd, C. M. (2005). Manipulation of allergen-
induced airway remodeling by treatment with anti-TGF-beta antibody: Effect
on the Smad signaling pathway. /. Immunol. 174, 5774—5780.
Meng, Q. R., Gideon, K. M., Harbo, S. J., Renne, R. A., Lee, M. K.,
Brys, A. M., and Jones, R. (2006). Gene expression profiling in lung tissues
from mice exposed to cigarette smoke, lipopolysaccharide, or smoke plus
lipopolysaccharide by inhalation. Inhal. Toxicol. 18, 555-568.
Miyabara, Y., Takano, H., Ichinose, T., Lim, H. B., and Sagai, M. (1998).
Diesel exhaust enhances allergic airway inflammation and hyperresponsive-
ness in mice. Am. J. Respir. Crit. Care Med. 157, 1138-1144.
Moller, M., Gravenor, M. B., Roberts, S. E., Sun, D., Gao, P., and Hopkin, J. M.
(2007). Genetic haplotypes of Th-2 immune signalling link allergy to enhanced
protection to parasitic worms. Hum. Mol. Genet. 16, 1828-1836.
Morishita, M., Keeler, G., Wagner, J., Marsik, F., Timm, E., Dvonch, J., and
Harkema, J. (2004). Pulmonary retention of particulate matter is associated
with airway inflammation in allergic rats exposed to air pollution in urban
Detroit. Inhal. Toxicol. 16, 663-674.
Nadadur, S. S., and Kodavanti, U. P. (2002). Altered gene expression profiles
of rat lung in response to an emission particulate and its metal constituents.
/. Toxicol. Environ. Health A 65, 1333-1350.
Nadadur, S. S., Pinkerton, K. E., and Kodavanti, U. P. (2002). Pulmonary gene
expression profiles of spontaneously hypertensive rats exposed to environ-
mental tobacco smoke. Chest 121, 83S-84S.
Noritake, S., Ogawa, K., Suzuki, G., Ozawa, K., and Ikeda, T. (2007).
Pulmonary inflammation in brown Norway rats: Possible association of
environmental particles in the animal room environment. Exp. Anim. 56,
319-327.
Ober, C., and Hoffjan, S. (2006). Asthma genetics 2006: The long and winding
road to gene discovery. Genes Immun. 7, 95-100.
Parameswaran, K., Willems-Widyastuti, A., Alagappan, V. K., Radford, K.,
Kranenburg, A. R., and Sharma, H. S. (2006). Role of extracellular matrix
and its regulators in human airway smooth muscle biology. Cell Biochem.
Biophys. 44, 139-146.
Park, J. W., Taube, C., Swasey, C., Kodama, T., Joetham, A., Balhorn, A.,
Takeda, K., Miyahara, N., Allen, C. B., Dakhama, A., et al. (2004).
Interleukin-1 receptor antagonist attenuates airway hyperresponsiveness
following exposure to ozone. Am. J. Respir. Cell. Mol. Biol. 30, 830-836.
Pauwels, R., Bazin, H., Platteau, B., and Van Der Straeten, M. (1979). Relation
between total serum IgE levels and IgE antibody production in rats. Int.
Arch. Allergy Appl. Immunol. 58, 351-357.
Prahalad, A. K., Inmon, J., Dailey, L. A., Madden, M. C., Ohio, A. J., and
Gallagher, J. E. (2001). Air pollution particles mediated oxidative DNA base
damage in a cell free system and in human airway epithelial cells in relation
to particulate metal content and bioreactivity. Chem. Res. Toxicol. 14,
879-887.
Riedl, M., and Diaz-Sanchez, D. (2005). Biology of diesel exhaust effects on
respiratory function. /. Allergy Clin. Immunol. 115, 221- 228; quiz 229.
Salmon, M., Walsh, D. A., Koto, H., Barnes, P. J., and Chung, K. F. (1999).
Repeated allergen exposure of sensitized Brown-Norway rats induces airway
cell DNA synthesis and remodelling. Eur. Respir. J. 14, 633-641.
Sampsonas, F., Kaparianos, A., Lykouras, D., Karkoulias, K., and
Spiropoulos, K. (2007). DNA sequence variations of metalloproteinases:
Their role in asthma and COPD. Postgrad. Med. J. 83, 244-250.
Sato, H., Sagai, M., Suzuki, K. T., and Aoki, Y. (1999). Identification, by
cDNA microarray, of A-raf and proliferating cell nuclear antigen as genes
induced in rat lung by exposure to diesel exhaust. Res. Commun. Mol.
Pathol. Pharmacol. 105, 77-86.
Shi, W., Bellusci, S., and Warburton, D. (2007). Lung development and adult
lung diseases. Chest 132, 651-656.
Sigaud, S., Goldsmith, C. A., Zhou, H., Yang, Z., Fedulov, A., Imrich, A., and
Kobzik, L. (2007). Air pollution particles diminish bacterial clearance in the
primed lungs of mice. Toxicol. Appl. Pharmacol. 223, 1-9.
Sioutas, C., Koutrakis, P., Godleski, J. J., Ferguson, S. T., Kim, C. S., and
Burton, R. M. (1997). Fine particle concentrators for inhalation exposures—
Effect of particle size and composition. /. Aerosol Sci. 28, 1057-1071.
Stevens, T., Krantz, Q. T., Linak, W. P., Hester, S., and Gilmour, M. I. (2008).
Increased transcription of immune and metabolic pathways in naive and
allergic mice exposed to diesel exhaust. Toxicol. Sci. 102, 359-370.
Takano, H., Yoshikawa, T., Ichinose, T., Miyabara, Y., Imaoka, K., and
Sagai, M. (1997). Diesel exhaust particles enhance antigen-induced airway
inflammation and local cytokine expression in mice. Am. J. Respir. Crit.
Care Med. 156, 36-42.
Takenaka, H., Zhang, K., Diaz-Sanchez, D., Tsien, A., and Saxon, A. (1995).
Enhanced human IgE production results from exposure to the aromatic
hydrocarbons from diesel exhaust: Direct effects on B-cell IgE production. /.
Allergy Clin. Immunol. 95, 103-115.
Tarayre, J. P., Aliaga, M., Barbara, M., Tisseyre, N., Vieu, S., and Tisne-
Versailles, J. (1992). Model of bronchial allergic inflammation in the brown
Norway rat. Pharmacological modulation. Int. J. Immunopharmacol. 14,
847-855.
Tibshirani, R., Hastie, T., Narasimhan, B., and Chu, G. (2002). Diagnosis of
multiple cancer types by shrunken centroids of gene expression. Proc. Natl.
Acad. Sci. U. S. A. 99, 6567-6572.
Tsien, A., Diaz-Sanchez, D., Ma, J., and Saxon, A. (1997). The organic
component of diesel exhaust particles and phenanthrene, a major poly-
aromatic hydrocarbon constituent, enhances IgE production by IgE-secreting
EBV-transformed human B cells in vitro. Toxicol. Appl. Pharmacol. 142,
256-263.
Underwood, S. L., Kemeny, D. M., Lee, T. H., Raeburn, D., and Karlsson, J. A.
(1995). IgE production, antigen-induced airway inflammation and airway
hyperreactivity in the brown Norway rat: The effects of ricin. Immunology
85, 256-261.
Walker, J. K., Ahumada, A., Frank, B., Gaspard, R., Berman, K.,
Quackenbush, J., and Schwartz, D. A. (2006). Multistrain genetic
comparisons reveal CCR5 as a receptor involved in airway hyperresponsive-
ness. Am. J. Respir. Cell. Mol. Biol. 34, 711-718.
Wang, I. M., Stepaniants, S., Boie, Y., Mortimer, J. R., Kennedy, B.,
Elliott, M., Hayashi, S., Loy, L., Coulter, S., Cervino, S., et al. (2007). Gene
expression profiling in patients with chronic obstructive pulmonary disease
and lung cancer. Am. J. Respir. Crit. Care Med.
Williams, A. S., Issa, R., Leung, S. Y., Nath, P., Ferguson, G. D.,
Bennett, B. L., Adcock, I. M., and Chung, K. F. (2007). Attenuation of
ozone-induced airway inflammation and hyper-responsiveness by c-Jun
NH2 terminal kinase inhibitor SP600125. /. Pharmacol. Exp. Ther. 322,
351-359.
Wise, H., Balharry, D., Reynolds, L. J., Sexton, K., and Richards, R. J. (2006).
Conventional and toxicogenomic assessment of the acute pulmonary damage
induced by the instillation of Cardiff PM10 into the rat lung. Sci. Total
Environ. 360, 60-67.
Yanagisawa, R., Takano, H., Inoue, K., Ichinose, T., Yoshida, S.,
Sadakane, K., Takeda, K., Yoshino, S., Yamaki, K., Kumagai, Y., et al.
(2004). Complementary DNA microarray analysis in acute lung injury
induced by lipopolysaccharide and diesel exhaust particles. Exp. Biol. Med.
(Maywood) 229, 1081-1087.
Yeatts, K., Svendsen, E., Creason, J., Alexis, N., Herbst, M., Scott, J., Kupper, L.,
Williams, R., Neas, L., Cascio, W., et al. (2007). Coarse particulate matter
(PM2.5-10) affects heart rate variability, blood lipids, and circulating
eosinophils in adults with asthma. Environ. Health Perspect. 115, 709-714.
TOC
Next
-------
J Pharmacokinet Pharmacodyn (2008) 35:683-712
DOI 10.1007/sl0928-008-9108-2
Comparing models for perfluorooctanoic acid
pharmacokinetics using Bayesian analysis
John F. Wambaugh • Hugh A. Barton •
R. Woodrow Setzer
Received: 13 May 2008/Accepted: 8 December 2008 / Published online: 8 January 2009
© Springer Science+Business Media, LLC 2008
Abstract Selecting the appropriate pharmacokinetic (PK) model given the
available data is investigated for perfluorooctanoic acid (PFOA), which has been
widely analyzed with an empirical, one-compartment model. This research
examined the results of experiments [Kemper R. A., DuPont Haskell Laboratories,
USEPA Administrative Record AR-226.1499 (2003)] that administered single oral
or iv doses of PFOA to adult male and female rats. PFOA concentration was
observed over time; in plasma for some animals and in fecal and urinary excretion
for others. There were four rats per dose group, for a total of 36 males and 36
females. Assuming that the PK parameters for each individual within a gender
were drawn from the same, biologically varying population, plasma and excretion
data were jointly analyzed using a hierarchical framework to separate uncertainty
due to measurement error from actual biological variability. Bayesian analysis
using Markov Chain Monte Carlo (MCMC) provides tools to perform such an
analysis as well as quantitative diagnostics to evaluate and discriminate between
models. Starting from a one-compartment PK model with separate clearances to
urine and feces, the model was incrementally expanded using Bayesian measures
to assess if the expansion was supported by the data. PFOA excretion is sexually
dimorphic in rats; male rats have bi-phasic elimination that is roughly 40 times
slower than that of the females, which appear to have a single elimination phase.
The male and female data were analyzed separately, keeping only the parameters
describing the measurement process in common. For male rats, including
Electronic supplementary material The online version of this article (doi:
10.1007/sl0928-008-9108-2) contains supplementary material, which is available to authorized users.
J. F. Wambaugh (El) • H. A. Barton • R. W. Setzer
National Center for Computational Toxicology, US EPA, Research Triangle Park, NC 27711, USA
e-mail: wambaugh.john@epa.gov
4y Springer
Previous
-------
684 J Pharmacokinet Pharmacodyn (2008) 35:683-712
excretion data initially decreased certainty in the one-compartment parameter
estimates compared to an analysis using plasma data only. Allowing a third,
unspecified clearance improved agreement and increased certainty when all the
data was used, however a significant amount of eliminated PFOA was estimated
to be missing from the excretion data. Adding an additional PK compartment
reduced the unaccounted-for elimination to amounts comparable to the cage wash.
For both sexes, an MCMC estimate of the appropriateness of a model for a given
data type, the Deviance Information Criterion, indicated that this two-compart-
ment model was better suited to describing PFOA PK. The median estimate was
142.1 ± 37.6 ml/kg for the volume of the primary compartment and
1.24 ± 1.1 ml/kg/h for the clearances of male rats and 166.4 ± 46.8 ml/kg and
30.3 ± 13.2 ml/kg/h, respectively for female rats. The estimates for the second
compartment differed greatly with gender—volume 311.8 ± 453.9 ml/kg with
clearance 3.2 ± 6.2 for males and 1400 ± 2507.5 ml/kg and 4.3 ± 2.2 ml/kg/h
for females. The median estimated clearance was 12 ± 6% to feces and 85 ± 7%
to urine for male rats and 8 ± 6% and 77 ± 9% for female rats. We conclude that
the available data may support more models for PFOA PK beyond two-com-
partments and that the methods employed here will be generally useful for more
complicated, including PBPK, models.
Keyword Bayesian analysis • Pharmacokinetics • Model comparison •
Hierarchical models • WinBUGS • Perfluorooctanoic acid • Sprague-Dawley rats •
Markov Chain Monte Carlo • Risk assessment • Sexually dimorphic •
Deviance Information Criterion • Population models
Introduction
Compartmental pharmacokinetic models range in complexity from first order, one
compartment descriptions to physiologically based pharmacokinetic (PBPK) models
of the absorption, distribution, metabolism, and excretion of a given compound by
an organism using biologically relevant compartments connected by biologically
based flows. The concentration of the toxicant in each compartment is described by
differential equations that contain parameters which may be biologically based or
phenomenological and about which there may or may not be prior experimental
knowledge [1]. Calibrating a pharmacokinetic model to describe a particular
organism may therefore require simultaneously determining a large number of
parameters from a sparse data set. For this reason the choice of the specific
pharmacokinetic model used for a particular toxicant in a particular organism
depends both on the behavior of the toxicant and the relevant data available [2-6].
Bayesian techniques are particularly well-suited to the task of evaluating
pharmacokinetic models: distributions of model parameters can be determined for
complicated model structures and these distributions can remain broad if the data
does not inform the model; knowledge about model parameters can be incorporated
in the form of prior distributions; and diverse model structures can be quantitatively
compared with each other [7-9].
4y Springer
Previous
-------
J Pharmacokinet Pharmacodyn (2008) 35:683-712 685
We are investigating perfluorooctanoic acid (PFOA), a perfluorinated fatty acid
analog that is of concern in part because it has been detected in the general
population and wildlife around the world [10]. PFOA is a persistent chemical with a
human half-life that has been calculated to be 4 years [11]. Because it is wide-
spread and has a long half-life, PFOA is of interest to toxicology and environmental
risk assessment [11-14]. PFOA is not metabolized [15], and at first glance seems
ideally suited to simple, empirical pharmacokinetic approaches that model
organisms as one or two well-mixed compartments. The half-life of PFOA is
much more rapid in animals than humans, ranging from hours in female rats to days
in male rats to weeks in monkeys [10]. These differences lead to estimated
exposures that vary by several orders of magnitude between species. PFOA
excretion is well known to be sexually dimorphic in rats and there is evidence that
testosterone-dependent expression of specific transporters is responsible [16-18].
Surveys of human plasma concentration have found no clear gender differences [13].
If the pharmacokinetics of PFOA is driven by transporters, then saturation of these
transporters at high concentrations may be a further complication [12].
Determining the level of detail sufficient to characterize the relationship between
exposure and response for PFOA is an ongoing question. There is evidence in
studies on monkeys for saturable kinetics resulting in non-linear dose-dependence at
high concentrations [12], however serial measures of PFOA serum concentration in
occupationally exposed humans are consistent with linear kinetics [11]. Since the
human kinetics are presumed to be at steady-state and the observed PFOA
concentrations in the general population of the United States are even lower than the
occupationally exposed [13], it may be hard to provide the information needed to
support complicated pharmacokinetic models. There is limited evidence for dose-
dependent pharmacokinetics in rats [19], but much of the serial rat pharmacoki-
netics data [20] considered by the US EPA draft risk assessment appears linear with
dose. Some, but not all, of the time-courses indicate bi-phasic elimination behavior.
Because the choice of model is not clear by inspection, and since the human data
appear linear, a one-compartment model for PFOA has been used by the draft risk
assessment and elsewhere [14, 21]. Given the limited human data, care must be
taken if a more complicated model—whether it be two-compartments, a biolog-
ically based model like that of Andersen et al. [12], or even a full PBPK model—is
to be used to for PFOA pharmacokinetics.
This research has two goals. First, we wish to examine the tradeoff between
making models more sophisticated and introducing more unknowns that could
potentially increase uncertainty. We do this to help the development of verifiable
methodologies for understanding when a model is appropriate for the available data
[7, 22]. We used empirical pharmacokinetic models with very few parameters to
describe data for rats exposed to PFOA. Because analytic solutions exist for these
models they can be rapidly solved, allowing Bayesian analysis with little prior
information via Markov Chain Monte Carlo (MCMC). The methodology of
Bayesian statistics allows us to make model comparisons and determine which
parameters can be estimated and the extent to which the data inform these estimates.
Because not all parameters in empirical pharmacokinetic models have direct
biological analogs, we made relatively uninformative assumptions about the
4y Springer
Previous I TOC
-------
686 J Pharmacokinet Pharmacodyn (2008) 35:683-712
parameters to investigate whether the data truly contains information about the
parameters for our assumed models. This lets us assess the usefulness of these
techniques for estimating pharmacokinetic parameters in the hypothetical situation
of no prior estimates. When we find that the available data does not inform a model
parameter, we then learn that we must simplify the model, make an assumption
about the parameter value, or acquire more data.
Our second goal is to characterize the empirical pharmacokinetics of PFOA in
rats. Since a one-compartment model has been widely used, quantitatively
determining the relative advantages of a two-compartment model is both
immediately useful and instructive for future model development. We assumed a
population model for the parameters of individual rats allowing us to attempt to
separate measurement uncertainty from biological variability. This approach allows
more accurate assessment of the information that would be gained by additional
experiments as well as greater predictive power [23]. Since a population model is
one method for allowing different types of measurements to inform each other, we
jointly analyzed plasma and excretion data to determine, for instance, the
elimination of PFOA from the plasma and the fractions of that elimination that is
excreted in urine and feces. Reconciling data of different types requires careful
model construction—for a poorer model, additional data may actually increase
apparent uncertainty for parameter estimates.
Since there is ample reason to expect that PFOA pharmacokinetics are more
complicated than can be described by a one compartment model, PFOA provides an
opportunity to develop quantitative arguments for more complicated models that are
both immediately useful to PFOA risk assessment and generally useful for data-
driving modeling.
Methods
WinBUGS version 1.4.2 [24] was used to describe and analyze population
pharmacokinetic models for PFOA in rats. WinBUGS uses MCMC techniques to
make Bayesian statistical inferences about the probability distribution of the
parameters in the model. Analyses were performed on dual processor 3 GHz
Pentium-D desktop computer.
The PKBUGS [25] pharmacokinetic interface for WinBUGS was not used, in
order to allow the examination of slightly modified compartmental models with two
elimination rates instead of one. The freely available R statistical software package
[26] and the R2WinBUGS [27] library were used to parameterize our model and run
WinBUGS. To increase processing speed, compiled functions were created to
calculate pharmacokinetic properties using WBDEV, the WinBUGS developer
package [28], and the freely available Blackbox Component Pascal language [29] in
which WinBUGS is written.
Following the example of Gelman et al. [2] we identify the following three
components of our model; the individual pharmacokinetic model, the population
model, and the measurement model.
4y Springer
Previous
-------
J Pharmacokinet Pharmacodyn (2008) 35:683-712 687
Individual pharmacokinetic model
One-compartment model
The one-compartment model consisted of a volume of distribution Vj (ml/kg) from
which there is clearance CL (ml/kg/h). Although physiologically the volume of
distribution and the clearances are independent of each other, the mathematical form
of the time-dependence of the concentration depends on the ratio of the two, i.e. the
elimination rates k = ^ . For the actual MCMC implementation we sampled
elimination rates rather than clearances because the rate could be changed
independently of the volume of distribution and therefore allowed more rapid
convergence.
Because we have separate data for urinary and fecal clearance, two clearances, Cluri
and Clfec, are distinguishable. However, initial attempts to model excretion using a
total clearance that is the sum of the separate clearances, Cl — Cluri + Clfec failed.
Instead, we had to parameterize using a total clearance Cl and two fractions FRACuri
and FRACfec, where Clx — FRACX x ClwAFRACuri + FRACfec
-------
J Pharmacokinet Pharmacodyn (2008) 35:683-712
occurs with a clearance Clj. The parameter vector now contains these two additional
parameters:
9 = (Vd, Vt, 0, FRACfec, FRACuri, Cld, ka,f).
For the two-compartment model the differential equations for the concentration
in the two compartments are:
dCi f • doseor _kat Cl Cld
- --
(C,-0). (4)
These equations can also be solved to give an analytic solution for the primary
compartment (we do not need the solution for the secondary compartment), C\ (Q,i).
For both the one- and two-compartment models the concentration and integrated
concentration were each implemented in WinBUGS as analytic functions using
Blackbox Component Pascal [29].
Population model
A hierarchical statistical framework that assumes that each individual has its own
pharmacokinetic parameters that have been drawn from the same population
distribution was used. We distinguish between parameters — which lead to predictions
for the observed data — and the hyper-parameters that characterize the distribution of
the parameters. Because of the sexually dimorphic excretion of PFOA, we assumed
separate populations for male and female pharmacokinetic parameters.
For most of the pharmacokinetic model parameters we assume that there are two
population hyper-parameters that generally describe the mean (v) and standard
deviation (r\) of a log-normally distributed population of individual parameter
values. Achieving convergence for the estimated distributions of population
standard deviations was the slowest of all parameter types, necessitating long
MCMC run times. For this reason, cross-correlations were not included, as they
would be even less informed by the available data.
Since we have excretion data indicating that the bioavailability was likely less
than one, we felt comfortable assuming that non-radiolabeled PFOA in the mass-
balance experiments was not a significant factor. For the bioavailability to be less
than or equal to one and the excretion fractions to sum to one, we assume that those
parameters are Dirichlet distributed. The forms of the distributions are given in
Table 1. For the one-compartment model:
Vc ~ TLN(Vd • pop • fi,Vd • pop • T, —oo, 8)
ktot ~ TLN(ktot • pop • n, ktot • pop • T, -oo, -0.5)
ka ~ TLN(ka • pop • fi,ka • pop • T, —4, 4)
FRACx~D(FRAChyper)
f~D(BIOAVAILhyper)
The log-normal distribution constrains the individual parameters to positive
values. In order to speed convergence the truncated log-normal distribution, which
4y Springer
Previous I TOC
-------
J Pharmacokinet Pharmacodyn (2008) 35:683-712
Table 1 Definition of probability distributions
truncated log-normal
Dirichlet
Notation TLN(x, /j.,t,a, b)
Random variable Univariate x, e" < x < eb
Hyper-parameters Location /i and precision T of log*
Form
D(x, a)
Multivariate x, Y%=\ *; = 1 , 0 < *, < 1
a is a k element vector
*
is both constrained to elower< x < euPPer ancj normalized to the interval [lower,
upper}, was used to sample individual values. By preventing extreme individual
parameter values, the likelihood is more sensitive to changes in parameter value.
With this approximation, the posterior parameter value distributions of all
individuals must be inspected to ensure that mass has not accumulated near a
truncation boundary.
There are two additional parameter distributions for the two compartment model:
Vt ~ TLN(Vt • pop • n,Vt- pop • T, -oo, 8)
ki2 ~ TLN(k\2 • pop • jj., k\2 • pop • T, —oo, —1)
where fe12 — CIJVC. We used the WinBUGS Markov chain Monte Carlo package to
determine a Bayesian posterior probability distribution [30] for all parameters and
hyper-parameters. We explain this process in detail in the section "Bayesian
Analysis." The number of iterations necessary for convergence was greatly
decreased by working with the logarithm of the parameters when creating new
samples. By estimating the logarithm of the parameter, changes in parameter value
are of a proportional size, rather than fixed, speeding up convergence.
A Bayesian analysis requires prior probability distributions for the hyper-
parameters to characterize previous knowledge about their values. We specified
distributions that were uniform over an arbitrary interval that was hopefully large
enough to encompass the true value of each parameter. An important check of this
assumption was to ensure that there is little probability mass near these arbitrary
limits in the converged posterior distributions.
We chose to use priors on the actual parameter values, as opposed to the log-
transformed parameters, so that the units of the prior distributions were the same as
parameters themselves. We believe that using uniform priors, rather than more
numerically efficient conjugate priors, permits the upper and lower limits of the
parameter values to be more accessible to a general biological modeling audience.
We assumed the prior distributions indicated by Table 2, which were the same
for the analyses of all data and both one- and two-compartment models, with the
exception that different ka priors were used for males and females. In most cases the
mean and standard deviation hyper-parameters are drawn from the uniform
distribution U(x, min, max) = mca1_min , min < x < max [31]. We made use of
distributions that gave the logarithms of parameters but were uniform on the
original scale.
Previous
TOC
& Springer
-------
690 J Pharmacokinet Pharmacodyn (2008) 35:683-712
Table 2 Population priors for pharmacokinetic hyper-parameters
Parameter
vc
k,0,
Y%=\ FRAChyper
Y?i=i BIOAVAILhyper
ka
V,
kl2
Units
ml/kg
1/h
1/h
ml/kg
1/h
Mean
Min
10
10"4
1
1
IO"2
Max
IO3
1
50
50
20
SD
Min
10
10"3
10"2
Location Scale
Max Min Max Min
2 x IO3
2
60 -4 5.5 IO"4
0 8 IO"4
-25 -2 IO"1
Max
15
IO3
25
In addition to the use of log-transformed variables, we also found that sampling
speed could be improved if, for some parameters, we sampled the precision and not
the standard deviation. For instance, if the standard deviation is large it is insensitive
to small perturbations to its value. Since the precision is the reciprocal of the
squared standard deviation, it is a very small number for large standard deviations
and much more sensitive to changes by the sampler.
Since we wished to use a uniform prior for the mean v and standard deviation
r\, it was necessary to convert to log-normal location /x = log(v) — |log(l +^r)
and scale r = Wlog(l +^r) to draw log-normally distributed individual param-
eters. Problems estimating scale parameters in hierarchical models are well known
[31, 32]. In our case, because the log-normally transformed variables /x and i each
depend on both v and a, varying the mean and standard deviation independently
greatly slows the approach to convergence. For example, if all individuals have a
parameter value near 100, then since log(lOO) x 4.6 we know that fj, x 4.6. If
the standard deviation is 10, then a mean of 100 gives fj, x 4.6. If the standard
deviation is 100, however, the likelihood of large parameter values grows, so that
for fi to be centered where the individuals are located, the mean must actually be
larger than before, xl27. If the standard deviation is larger still, say 1,000, then
the mean needed to describe a log-normal distribution located at fi x 4.6 is
«325. In short, for large standard deviations the choice of mean depends on the
choice of standard deviation.
For the female rats, the clearance occurred so rapidly that only the lower limit
of absorption was sharply defined (the absorption must be sufficiently fast to
explain the presence of PFOA in the plasma). Because data was collected at the
same time points for males and females, we effectively have less female data.
With few data points, it is hard to set an upper limit since the concentration
primarily depends upon e~ka —for ka greater than 10 this term is effectively zero.
For this reason, the standard deviation can be large for the female rats and we
found that we could not achieve convergence by using the same parameterization
as for the other variables. Instead we estimated the log-normal parameters fi and T,
and specified uniform priors on the transformed variables. This amounts to a
4y Springer
Previous I TOC
-------
J Pharmacokinet Pharmacodyn (2008) 35:683-712 691
non-uniform prior on the mean and standard deviation for ka, but analyses of male
rat data using uniform mean and standard deviation or uniform log-normal
location and scale gave similar results. For both male and female rats, we found
we also had to use uniform priors on the log-normal location and scale for the
inter-compartmental clearance rate fe12.
Measurement model
We assume a single measurement model with the same parameters for both male
and female rats.
For each individual we either have a series of observations of plasma
concentration y(tl) at several discrete times t1, or data for the fraction of the PFOA
dose recovered separately in urine and feces at times t2, yuri(t2) and yfec(t2).
To evaluate the likelihood that a given set of pharmacokinetic parameters for an
individual, 9, gave rise to observations made for that individual, we model the
measurements as normally distributed about the model values for those parameters:
),y.',(Q,ti)) (5)
where N(x, v, d) = \f^f~^-x~^ and the precision y.x of each observation y is
modeled as:
~ (6)
Note that this precision depends both on the concentration and a constant offset £c.
Thus large concentrations are considered to be less precise, but the precision for
small concentrations does not become arbitrarily large. This is similar to the
approach of Rocke and Lorenzato [33] except that both the concentration-dependent
and independent contributions to error are lumped together and are normally
distributed in order to reduce the number of calculations necessary to find the error.
The corresponding standard deviation y.v is:
))c+ec. (7)
When observations were below the limit of detection or quantitation, we modeled
a censored observation by requiring that the modeled value lie below the smallest
observed quantity for that dose-level because we did not have the actual limit of
detection for each observation dose group. We allowed the measurement threshold
to depend on dose-level in case different dilutions were used for the different doses.
Observations of the urinary elimination are similarly modeled as:
yUri(ti) ~N(x, elimuri(9, ti),yuri.i;(9, ?,-)) (8)
where the precision has the same form as for the concentration but different
parameter values:
yuri-^d, t) = (e"uri x elimuri(9, t+e (9)
and the elimination over the time interval (f,_i, ?,) is:
4y Springer
Previous I TOC
-------
692 J Pharmacokinet Pharmacodyn (2008) 35:683-712
elimuri(Q, ti) = - - - - C(6, t')dt'. (10)
doseor
For the fecal elimination observations we note two features that need
modification of our one- and two-compartment analyses. First, there is a
conspicuously large fraction of the PFOA that is eliminated in the first one or
two observed eliminations that does not seem well-described by an exponential
process. This might be due to a bioavailability-like issue in which some fraction of
the initial dose passes straight through to feces. We include a fraction of 1 —/of
oral doses in the first sample of the feces.
yfec(ti) ~N(x, elimfec(9, t^y^.^O, f,-)) (11)
with the precision is given by:
elimfec(9, t+e^ (12)
and the elimination is calculated as:
Clfec
elimfec(9,ti) =
doseor
A \
C(9,t>) + (1 -/) d°Se°r - ,5(0 }dt>
doseor + doseiv J
where 8(t) is the Dirac <5-function defined as J^cSt/lX)) = /(0).
Note that times earlier than the initial dosing time — assumed to be t — 0 here —
must be truncated when solving for concentrations. This is handled by the Blackbox
functions.
Another aspect of the measurements of PFOA in feces and urine is that not all of
the urine and feces is recovered until the cages are washed at the end of the study.
Usually, around 1 % of the total dose was found in the cage and residual feed. In one
instance, however, the amount recovered was 16%. Because we do not expect the
contribution to cage wash to be consistent from measurement-to-measurement, the
contribution to total cage wash would have to be estimated for each elimination data
point. For the sake of model simplicity we have not done this, and as a result we
have potentially increased estimates of ECfec and Ecuri. When we later model a third
route of elimination, in addition to feces and urine, we can interpret this unknown
fraction as an estimated average loss to cage wash.
The models for precision introduce several additional parameters which must be
estimated with the MCMC analysis. For each of the parameters we must specify a
prior distribution reflecting our assumptions about that parameter. The priors
indicated in Table 3 were found to be sufficiently general that for all types of data
analyzed and both the one- and two-compartment models that there was minimal
posterior probability mass near the limits of the priors.
4y Springer
Previous I TOC
-------
J Pharmacokinet Pharmacodyn (2008) 35:683-712
693
Table 3 Prior distributions assumed for error model parameters
Distribution
E" Uniform
Eb Normal
EC Uniform
ffec Uniform
ffec Normal
ffec Uniform
Eauri Uniform
Eburi Normal
furi Uniform
Minimum Maximum Mean
10~2 10
1
10~8 5 x 10~3
10"2 20
1
10~6 0.5
10"2 5
1
10~6 0.5
SD
1
1
1
Bayesian analysis
Markov Chain Monte Carlo assists Bayesian statistical analysis by constructing
samples from the posterior distribution of the model parameters using the priors and
likelihood to construct a Markov Chain, a kind of random walk through the space
defined by the parameters: each new step in the walk being a new full set of
parameter values. The chain is constructed so that, in the long run, the joint
probability distribution of the collection of steps converges to the posterior
distribution [30]. Generally, parameters in the posterior distribution are not
independent of each other; entire sets of parameter values must be drawn from the
chain at once to preserve inter-parameter correlations. Multiple chains can be
created from different initial conditions to determine if similar posterior distribu-
tions are reached [30].
Typically, priors for model parameters would be based on the results of previous
relevant experiments, and physical constraints relevant to the system being
modeled. However, there are few very stringent physical constraints on the
parameters of the one- or two-compartment models used here, and we chose not to
base priors on previous experimental work in order to evaluate the effectiveness of
Bayesian analysis for a novel chemical. Priors for most parameters were instead
minimally informative proper uniform prior distributions.
To estimate the posterior distributions of the individual and hyper parameters,
two Markov chains were initialized using values drawn from the prior distributions
and several hundred thousand iterations of the WinBUGS sampler were typically
run for both chains simultaneously on separate processors. Parameter values in
consecutive samples are often very similar; to minimize this auto-correlation a
thinning interval n was selected such that only every wth sample was recorded and
the intervening samples were discarded. Depending on the total number of
iterations, the thinning interval was varied so that 8,000 evenly spaced sampler
states were recorded. The first half of each chain was treated as a "burn in" period
during which the chain converges to the posterior distribution. The final 4,000
sampled states were then taken as the estimate of the Bayesian posterior distribution
[30].
4y Springer
Previous
TOC
-------
694 J Pharmacokinet Pharmacodyn (2008) 35:683-712
Estimating the posterior is itself an iterative process in which we ran the sampler
for two chains, evaluated their convergence and then, if the chains did not pass our
convergence tests, either modified our model or re-initialized using draws from the
chains and ran the chains for more iterations.
When using MCMC to perform a Bayesian analysis a primary concern is whether
a chain is sufficiently long to fully estimate the posterior distribution. Since it is
possible to obtain physically plausible posterior distributions from unconverged
chains the utmost care must be taken in assessing convergence [34, 35]. Though
convergence can never be proven, we apply the following four methods to estimate
chain convergence.
We first visually inspected the traces of parameter value with iterations to detect
transients dependent upon the initial conditions and obvious auto-correlation. We
then used the Convergence Diagnosis and Output Analysis (CODA) package [36] as
implemented for R [37]. We use Gelman's f quantity [38] that compares the
variance from multiple chains to the variance within each chain. Values close to
unity indicate convergence, and we typically run until f < 1.02 for every parameter.
We also applied the Raftery and Lewis criteria for estimating sufficient chain length
[39]. Following Vicini and Dodds MCMC approach to Bayesian analysis of
hierarchical pharmacokinetics models, we used less restrictive values for the
Raftery and Lewis convergence criteria than the CODA default [40]. For a target
quantile of 0.025 to 0.975 and a target accuracy of ±0.02 we estimated the chain
length sufficient for a 90% probability of attaining our target accuracy. These
parameters are appropriate when the parameter probability distributions do not have
"fat" tails [39].
We found that at the least both the Raftery and Lewis criteria and the Gelman f
were necessary to determine convergence—insufficiently long chains could pass
one without passing the other.
Comparing prior to posterior distributions helps us recognize parameters that are
poorly identified by the available data, allowing us to either simplify the model or
choose different prior distributions. Additionally, analysis of the posterior
distributions provides a fourth criteria for assessing parameter estimates: large
probability mass near a boundary of the prior distribution indicates that we have not
used a sufficiently vague prior. If we observed evidence of an over-restrictive prior,
we widened the boundaries on the prior by roughly an order of magnitude and reran
the sampler.
Data analyzed
The data set analyzed includes the results from several different experiments [20] on
adult male and female Sprague-Dawley rats using radioactively labeled PFOA. The
mean and standard deviation of the body weights were 229.9 ± 22.9 g for males
and 191.1 ± 12.4 g for females. Serial plasma concentration time course samples
were collected for doses of approximately 0.1, 1, 5 and 25 mg/kg of PFOA, that
were administered by oral gavage in cohorts of four each of males and females. For
the male rats, samples were collected 0.25, 0.5, 1, 2, 4, 8, 12, 16, and 24 h after
dosing; then every 24 h through the 8th day; and then every 48 h through the 22 day
4y Springer
Previous
-------
J Pharmacokinet Pharmacodyn (2008) 35:683-712
695
for a total of 23 data points. For the female rats, for whom clearance was observed
to be much faster, samples were collected 0.25, 0.5, 1, 2, 4, 8, 12, 16, 24, 36, 48, 72,
and 96 h after dosing for a total of 13 data points. Plasma concentration was also
monitored at the same time intervals for additional cohorts of four male and four
female rats that were administered approximately 1 mg/kg intravenously. An
additional, extended time course cohort of four male and four female rats was
administered 0.1 mg/kg, by oral gavage. The males in the extended time course
cohort were monitored for 2,016 h (84 days) and the females were monitored for
312 h (13 days).
For three additional cohorts of four male and four female rats each, PFOA was
administered by oral gavage at doses of approximately 1, 5 and 25 mg/kg. For the
male rats, feces and urine were collected after 4, 8, 12, and 24 h; then after 24-h
intervals through the 14 day; and then after 48-h intervals through the 28th day,
giving a total of 24 data points per male rat. For female rats the same time points
were collected only until the 7th day, giving a total of 10 data points. Radioactivity
was monitored in liquid samples using liquid scintillation counting (LSC) and in
solid sample using combustion followed by LSC. The limit of quantitation was
1 ppb. Because PFOA is not known to be metabolized, radioactivity from
metabolites was not considered as an influence on the results of the mass-balance
experiments.
The entire data set includes 36 male and 36 female rats. The investigators fed the
rats PMI Nutrition International, LLC Certified Rodent Lab Diet 5002. The rats
were fasted overnight prior to dosing and for approximately 24 h afterwards.
Haskell Animal Welfare Committee guidelines were followed.
Results
The medians of the distributions for the population means and standard deviations
of the pharmacokinetic parameters for the two-compartment model are summarized
in Table 4. Only a portion of the two-compartment results are reported in detail
Table 4 Means and standard deviations of two compartment model parameter distributions
Population parameter
vc
v,
ci,0,
FRACfec
FRACuri
Cld
Ka
Unavailability /
Units
ml/kg
ml/kg
ml/kg/h
ml/kg/h
1/h
Male
Mean
142.10
311.80
1.24
0.12
0.85
3.17
5.18
0.97
SD
37.61
453.85
1.08
0.06
0.07
6.20
29.45
0.05
Female
Mean
166.40
1400.00
30.33
0.08
0.77
4.30
3.01
0.96
SD
46.80
2507.50
13.15
0.06
0.09
2.21
2.93
0.03
Previous
TOC
4y Springer
-------
696
J Pharmacokinet Pharmacodyn (2008) 35:683-712
here. Complete results for both the one- and two-compartment model analyses,
including simulated experiments, are available in on-line supplementary material.
Two compartment pharmacokinetics for PFOA
To generate our final, converged chains we used 960,000 iterations, which required
over 63 h per chain for the plasma, urine and feces data jointly.
The distribution of predictions for the two-compartment model is shown for male
rats in Fig. 1. The addition of a second compartment leads to plasma concentration
predictions that better capture the higher PFOA plasma concentration and more
rapid elimination at early times than the single-compartment model. The scatter of
the predictions is slightly broader than with the simpler model, but simulated
experiments using means of four simulated animals at each dose look more like
the experimental observations. Although the two-compartment model allows
1x10'
2x10'
5x10'
I I I I I
0 100 200 300 400 500
1 x10 '
2x10~'
5x10"'
1 mg/kg Oral
i i i i i
0 100 200 300 400 500
D)
2x10~'
• 1 mg/kg Intravenous
fffi
1
1)
1
'(
'l
I
:.
•(
1
(
1
1
, 1
J [
1 1
1 :
1 i
1 i
i
i j
| <
1 i
t
J ii
1 x10
2x10~'
5x10~'
5x 10~4 -I I Tl 1 x 10~;
0 100 200 300 400 500
5 mg/kg Oral
Mil
0 100 200 300 400 500
2x10'
.1 mg/kg Oral (Extended)
5x10'
0 100 200 300 400 500 0
Time (h)
500 1000 1500 2000
Fig. 1 Distribution of predicted PFOA plasma concentrations using a two-compartment model for male
rats. We compare the means (D) and 95% confidence interval (vertical lines) of 500 simulated
experiments to the observations. The mean observations (•) from cohorts of four animals for each dosing
fall mostly within the confidence intervals
4y Springer
Previous
TOC
-------
J Pharmacokinet Pharmacodyn (2008) 35:683-712
697
two-phases of clearance, the male extended time course data are still underesti-
mated, possibly indicating the need for an additional phase of even slower re-
distribution and clearance.
For the male rats the predicted urine excretion, shown in Fig. 2, better matches
the observations than the one-compartment model predictions. The predicted
excretion to feces, also shown in Fig. 2, is generally higher than the observations for
the 5 mg/kg dose, although the predictions for both 1 and 25 mg/kg seem
appropriate.
Though the excretion data also indicates a time lag—PFOA is observed in the
plasma and urine roughly 6 h before appearing in the feces—we had little success in
estimating this parameter using simple modifications to one- and two-compartment
pharmacokinetics so we did not include a fecal time lag in our final model.
Because of the rapid clearance of PFOA from female rats, the predicted plasma
pharmacokinetics shown in Fig. 3 do not appear especially different from the
1 x10
1 x10 '
1 x10 '
1 mg/kg - Feces
1x10
1x10 '
1x10 '
1 mg/kg - Urine
ttfffff
0 100 200 300 400 500 600
n i i i i i i
0 100 200 300 400 500 600
S.
T3
CD
§ 1x10-
O
Q
x10"
5 mg/kg - Feces
1x10"' -
1x10"
1x10"
5 mg/kg - Urine
0 100 200 300 400 500 600
0 100 200 300 400 500 600
ixio
1 x10'
1 x10'
i i i i r
0 100 200 300 400 500 600
1x10
1x10 J -
1x10 '
Time (h)
25 mg/kg - Urine
0 100 200 300 400 500 600
Fig. 2 Distribution of predicted fecal and urinary PFOA excretion using a two-compartment model for
male rats. We compare the means (D) and 95% confidence interval (vertical lines) of 500 simulated
experiments to the observations. The mean observations (•) from cohorts of four animals for each dosing
fall mostly within the confidence intervals
4y Springer
Previous
TOC
-------
J Pharmacokinet Pharmacodyn (2008) 35:683-712
1x10~
1x10~
1x10~
"
0.1
*
mg/kg oral
•
1 x10~
IxKT
1 mg/kg oral
f* ?
10 15 20
0 5 10 15 20
o
LL
Q_
1x10~
1x10'
1x10~
1x10
1x10 '
1x10 '
s
0 5
f f
1 mg/kg iv
1 t i
M t
i i i
10 15 20
25 mg/kg oral
* * t i
1
1 x10 '
1 x10 '
5 mg/kg oral
10 15 20
~
1 x10~3 -
IxKT4 -
1 x10~5 -
0.1 mg/kg oral (Extended)
ff f
Mi
t l
10 15 20
Time (h)
10 15 20
Fig. 3 Distribution of predicted PFOA plasma concentrations using a two-compartment model for
female rats. We compare the means (D) and 95% confidence interval (vertical lines) of 500 simulated
experiments to the observations. The mean observations (•) from cohorts of four animals for each dosing
fall mostly within the confidence intervals Although plasma sample were collected after 30 h, the
concentrations were below the limit of quantitation
one-compartment case. One clear difference between the models, however, is that
the estimated measurement error is reduced for the two-compartment model,
resulting in much better agreement at long times in the extended time-course data.
The mean excretion for female rats predicted using the two-compartment estimates,
shown in Fig. 4, is also very similar to the one-compartment case. However, the
uncertainty about these predictions is significantly reduced and the estimated
measurement error seems more similar to the observed point-to-point variation.
In Fig. 5 we plot the mean and 95% quantiles for the posterior distribution for the
parameters for every male rat. All the parameter distributions are too broad to
demonstrate systematic dose-dependence. However, for the four male rats receiving
the lowest dose (0.1 mg/kg) but not monitored over extended time, the distribution
of Vc is centered noticeably higher than for all other male rats, including the
extended time-course low dose category. There is no similar change with dose for
0 Springer
Previous
TOC
-------
J Pharmacokinet Pharmacodyn (2008) 35:683-712
1 x10
1 x10 '
1 x 1CT
[ i 1 mg/kg - Feces
II
i
i
i
r
4
1
litiij
1x10"'
1x10"'
1 mg/kg - Urine
11111
50
100 150
50 100
150
t 1x10-1 -
0
Q_
T3
0
0
g 1x10'3-
o
0
ce
0
ID
I
5 mg/kg - Feces
• i
1
1
1
m
T I
1 T T t 1
1x10~3 -
1x10~5 -
[I
P
f
5
t
mg/kg
t
t
Urine
t
t +
50
100 150
50 100
150
c
O
O 1 x10
(3
1 x10 •
1 x10 '
[ i 25 mg/kg - Feces
"I
|
k
tttui
1x10 1 -
1x10~3 -
1x10~5 -
L 25 mg/kg - Urine
n
f f 1 1 1 1
iii iii
50 100 150 50 100 150
Time (h)
Fig. 4 Distribution of predicted fecal and urinary PFOA excretion using a two-compartment model for
female rats. We compare the means (D) and 95% confidence interval (vertical lines) of 500 simulated
experiments to the observations. The mean observations (•) from cohorts of four animals for each dosing
fall mostly within the confidence intervals
the volume of distribution of the second compartment. One possible explanation for
the differences in Vc is that an error in the dosing of those animals caused them to
receive less than the desired 0.1 mg/kg. Separate parameter estimations including an
additional factor multiplying the doses received by those four animals were
performed, resulting in an estimate that those animals may have received only 0.67
of the reported dose. Since there is no further evidence of this hypothesis, it was not
included in the final analyses. For the male excretion data, qualitatively different fits
are found for the 1 mg/kg dose group and one of the 5 mg/kg animals than are found
for the higher-dose excretion data and plasma data. The total clearance, Cltot, is
substantially higher for these animals. Variation in the female parameter estimates
was less pronounced. The individual fits to the data for these animals indicate that
the feces data is better fit than the urine data. For the higher dose excretion data the
urine data is better fit than the feces. There is no discernable dose-dependence in the
means of the inter-compartmental clearance Clj, but the density of the distribution
0 Springer
Previous
TOC
-------
700
J Pharmacokinet Pharmacodyn (2008) 35:683-712
O)
.^
S-
0.1
\ I I
1 5 25
O
0.1
1 5 25
O
1 5 25
8-
>, °>-
-S1 o
o
in
in
o
0.1
25
0.1
25
Dose (mg/kg)
Fig. 5 Dose dependence of two-compartment model parameters for male rats. The parameters for the
individual animals give an estimate of parameter dependence on dose and study type. Individuals (indicated
by tone) are grouped into four study types (indicated by symbol): orally dosed, plasma time-course (O),
orally dosed, plasma extended time-course (D), iv-dosed, plasma time-course (O), and orally dosed, urine
and feces time-course (A). Each symbol indicates the mean parameter values and upper and lower error bars
indicate the 0.975 and 0.025 quantiles respectively. The x-axis is not continuous—individuals have been
offset from their actual dose value for clarity
for the lowest dose is much broader than for the higher doses. For the two-
compartment model there is no obvious dose dependence of fraction of PFOA
excreted to feces or urine.
As a reality check on our pharmacokinetic estimates we examined the
dependence of certainty about individual parameter estimates on the type of data
available for and dosing administered to that animal, as indicated by symbol in
Fig. 5. As expected, while the estimated means do not significantly vary, the breadth
of the distributions depend strongly upon dose-type. For the intravenously dosed
animals the volume of the central compartment Vc is well known since it is not
confounded by the rate of absorption ka. Correspondingly, the probability density
for ka is quite broad for iv-dosed animals since this parameter is unrelated to the
4y Springer
Previous
TOC
-------
J Pharmacokinet Pharmacodyn (2008) 35:683-712
701
experimental observations and is only informed by the population distribution. Also
as expected, for all of the parameters the distributions for animals where excretion
data was available are much broader.
The predicted exposure, as quantified by the integrated area under the plasma
concentration curve, is lower for the male rats when a two-compartment model is
used. The distribution of the ratio of the two-compartment AUC to the one-
compartment AUC is centered near 2/3, indicating a reduced predicted exposure.
This lower ratio is driven by the excretion data—when using the plasma data
alone the difference between the one- and two-compartment models is much less.
The ratio is closer to unity for the female rats, but is still peaked at slightly less
than one.
Measurement model parameter estimates
Using 500 draws from our posterior parameter distributions, we calculated a
distribution of predictions for every observed time point and subtracted each
prediction from the observed value to obtain a distribution of residuals. We scaled
the residuals by the standard deviation predicted using the measurement model. In
Fig. 6 the median of these standardized residuals is plotted against the median
prediction for each male rat observation. For the majority of data points the
predicted residuals are evenly distributed about zero and within three standard
deviations a wide range of values and experiment types. Some model predictions,
however, deviate significantly from observations. Additionally, the predicted
standard deviations appear to be over-estimated since the standardized residuals
are clustered near zero.
-------
702 J Pharmacokinet Pharmacodyn (2008) 35:683-712
Table 5 The medians of the estimated measurement parameters
Data
Plasma
Urine
Feces
£a
0.14
2.11
7.65
f
0.94
1.85
1.67
EC
5.62 x 10~8
2.36 x 10~3
2.05 x 10~4
Heteroskedasticity
0.646
0.001
0.068
As was clear in Fig. 1, at low concentrations (~10 5 mg/1) the male plasma
extended time-course observations are significantly underestimated and correspond-
ingly some of the residuals in Fig. 6 are as large as 100 standard deviations from the
actual observation and are beyond the plotted region. For the male excretion time
courses, large residuals are predicted for a few of the large feces measurements,
corresponding in Fig. 2 to the observation of the large initial quantity of PFOA in
feces that we have approximated as being excreted in the first time point, but
actually was occasionally distributed over the first few time points.
In Table 5 we list the estimated measurement parameters for the plasma, urine
and feces data. To give a sense of heteroskedasticity—how the relative magnitude
of the error varies with concentration—we also indicate the ratio of error for a
concentration of 10~2 to the error for concentration 10~5. Constant error would be
indicated by a ratio of 1.0, but there was pronounced variation in the magnitude of
the measurement error for all data sets, especially the excretion data.
Model comparison
To compare the one- and two-compartment models and assess the additional
information gained by jointly analyzing the plasma concentration and excretion data
we used the posterior distributions to predict simulated data, investigated obvious
discrepancies between model predictions and data, and then examined general
model appropriateness criteria for Bayesian analyses. The posterior predictive
distribution analysis is available in supplementary material on-line.
Unknown elimination in one-compartment model
We performed separate analyses for two data sets, one of just the 24 plasma
concentration curves for each gender, and one including both the plasma
concentration time-courses as well as the mass-balance time courses for the 12
additional animals of each gender for whom the amount of PFOA in feces and urine
was measured. When including the mass-balance data, separate clearances Clfec and
dun can be estimated for the feces and urine excretion.
We first attempted to estimate the clearance Cl as the sum of separate clearances
Clfec and Cluri, but had difficulty creating converged chains. The number of degrees
of freedom pD estimated by WinBUGS became negative, indicating major problems
reconciling the data with the model [41, 42] (see the section "Model appropriate-
ness given the available data" for further discussion of pD). Especially of concern
was that, for a one compartment model in which excretion to urine and feces
4y Springer
Previous I TOC
-------
J Pharmacokinet Pharmacodyn (2008) 35:683-712
703
O d
c
-------
704
J Pharmacokinet Pharmacodyn (2008) 35:683-712
R-
2
CL
t °^
o
t
I °H
Q
2
CL
0.0 0.2 0.4 0.6 0.8 1.0
Fraction of Eliminated PFOA in Excretion
0.0 0.2 0.4 0.6 0.8 1.0
Fraction of Eliminated PFOA in Excretion
Fig. 8 Distribution of the estimated fraction of PFOA eliminated from the plasma that is found in urine
and feces for male rats. By allowing a third, unknown route of excretion the total PFOA found in urine
and feces can be less than the estimated elimination from the plasma. The parameters estimated for the
one compartment model (left) are unable to account for nearly a third of the eliminated PFOA, while for
the two compartment model (right) only a small amount, if any, is missing
two-compartment model, the discrepancy between unknown elimination and cage
wash is resolved.
Somewhat of the reverse is found for the female case: For the one-compartment
model, all of the eliminated PFOA is almost certainly accounted for by excretion.
When we use the two-compartment model, however, we find that the uncertainty
has actually increased. This seems to indicate that using a two-compartment model
for female rats may introduce unnecessary parameters.
Model appropriateness given the available data
Model parameter estimates were qualitatively assessed by examining the difference
between the prior and posterior parameter distributions to determine whether model
parameter estimates have been informed by the available data. For example,
because of the rapid elimination of PFOA from the plasma of female rats the role of
a second compartment is ambiguous. Estimating the breadth of the population
distribution for second compartment parameters was especially difficult. When only
plasma data was used, the posterior distribution of the precision of the volume of
distribution for the second compartment is essentially identical to the assumed prior
distribution. Only when excretion data was considered was there sufficient
information to produce a posterior parameter distribution that differed from the
prior. Visually inspecting the posterior distributions sometimes allowed rapid
determination of the appropriateness of some model parameters.
As a measure of quantitative model consistency, Spiegelhalter et al. [41]
introduced a Bayesian deviance describing the fit of the data to the posterior
distribution of model parameters. They then took the difference between this
posterior deviance and the deviance of the data when only the means of the posterior
distribution are used to estimate the complexity of the data, as characterized by the
estimated degrees of freedom, pD.
0 Springer
Previous
TOC
-------
J Pharmacokinet Pharmacodyn (2008) 35:683-712 705
Each parameter in the model contributes 1 estimated degree of freedom if it is
completely free, 0 if it is completely constrained, and some fraction between if it is
correlated with other parameters. For the one compartment model with plasma data
only we have four parameters per rat, eight hyper-parameters and three measure-
ment model parameters, so for 24 rats we have 107 total parameters. The pD is
48.37 for the males and 46.29 for the females, indicating roughly two free
parameters per individual animal. This is unsurprising since many of the parameters
are interrelated, such as the bioavailability and the volume of distribution.
When we jointly analyzed excretion and plasma data with the one-compartment
model, there are roughly 37 additional degrees of freedom for the males and 27 for
the females—more than two for each of the 12 additional animals. This indicates
that including the excretion data does provide a more informative picture of the
pharmacokinetics.
We also found that the female feces data alone was insufficient to estimate both
pharmacokinetic and error model parameters (resulting in negative estimated
degrees of freedom). A major advantage of analyzing the male and female data set
jointly is that the information about the error model parameters in the male feces
data allowed the female feces data to be used.
When we examined the estimated free parameters for the two compartment
model, we found that for both male and female plasma data that nearly one
additional degree of freedom per individual animal is available for bi-phasic fits. It
is interesting to note that when we perform a joint analysis using the two-
compartment model, that the feces data provides a significant number of the degrees
of freedom and actually constrains the fit to the plasma data. This would seem to
indicate that both fecal and plasma data are important for characterizing PFOA
pharmacokinetics.
One method for evaluating the goodness of fit of a Bayesian analysis of a
particular model is the Deviance Information Criterion (DIG). Like the Akaike
Information Criterion or the Bayesian Information Criterion, the DIG is intended to
allow model comparison by characterizing the additional information gained by
including additional parameters. The DIG is specifically designed for Bayesian
analyses of hierarchical models. Roughly, the DIG is the difference between the
deviance of the observations from the predictions for each individual and the
deviance of the observations from the predictions for the population means. If a
model includes additional parameters that simply improve the "fit" of the mean
trends of the data, then the DIG is higher. If the additional parameters actually
describe experimental variation, then the DIG is lower. The DIG is meant to
discriminate between models, with the better model having a lower value [41].
For all combinations of data examined, listed in Table 6, the two-compartment
model has a lower value of DIG. This suggests that the additional complexity of a
second compartment is actually supported by the data. Note that all of the DIC's
calculated here are negative. This is because we examined parameters with
probability densities much greater than one, resulting in negative, but still
appropriate, values [41].
The success of the models was also examined by comparing the different
distributions predicted for plasma concentration integrated over time from the origin
4y Springer
Previous I TOC
-------
706
J Pharmacokinet Pharmacodyn (2008) 35:683-712
Table 6 Comparison of Deviance Information Criterion for one- and two-compartment model
Data
Male plasma
Male urine
Male feces
Male total
Female plasma
Female urine
Female feces
Female total
One comp.
Plasma
-8295
-8295
-2387
-2387
Joint
-8294
-2113
-2196
-12603
-2387
-618
-891
-3896
Two comp.
Plasma
-8745
-8745
-2655
-2655
Joint
-8808
-2049
-2511
-13368
-2689
-676
-945
-4310
to infinity—the AUC. The AUC was normalized across doses using the adminis-
tered dose and the volume of the primary compartment. For the one-compartment
model, the normalized AUC is equivalent to the clearance. As shown in Fig. 9, for
the male rats the one-compartment model (in the top row), the AUC is more
precisely estimated when the excretion data is used than when plasma data is used
alone. The means of the two distributions are roughly similar, however. For the two
compartment model (in the bottom row) the addition of excretion data shifts the
oo
q -
ci
O ci
•*- o
o q -
-S1 d
'co 0
100 200 300
oo
q
ci
o
q -
ci
100 200 300
Q
.•51 oo
CD
JD
O
sj
ci
o
q -
ci
1
100 200 300
oo
q -
ci
sj
ci
o
q -
ci
0
100 200 300
Normalized Area Under the Curve (h)
Fig. 9 Distribution of integrated plasma concentration in male rats for combinations of models and data.
We integrate the plasma concentration from zero to infinity (AUC) for one- (top) and two- (bottom)
compartment models for Bayesian estimations conducted with the plasma data alone (left) or plasma,
feces and urine data jointly (right). The AUC is calculated for a 25 mg/kg oral dose and is then
normalized by 25 mg/kg and the volume of the primary compartment
4y Springer
Previous
TOC
-------
J Pharmacokinet Pharmacodyn (2008) 35:683-712
707
o q _
>, o
in
ci
q
ci
1
0 2 4 6 8 10
0 2 4 6 8 10
Q q ,
ro
2 o
D. ^
in
ci
q
ci
1
02468 10 02468 10
Normalized Area Under the Curve (h)
Fig. 10 Distribution of integrated plasma concentration in female rats for combinations of models and
data. We integrate the plasma concentration from zero to infinity (AUC) for one- (top) and two- (bottom)
compartment models for Bayesian estimations conducted with the plasma data alone (left) or plasma,
feces and urine data jointly (right). The AUC is calculated for a 25 mg/kg oral dose and is then
normalized by 25 mg/kg and the volume of the primary compartment. Note that the x- and y-axis scalings
are changed from Fig. 9 to better show the details of the distributions
AUC to lower values. This indicates that the predicted pharmacokinetics are
somewhat different for our most elaborate analysis. Since the DIG is the lowest for
this analysis, it is likely that the male PFOA AUC is lower than would be estimated
from the one-compartment model and/or plasma data alone.
For the AUCs for the female rats shown in Fig. 10, the modes of the distributions
of AUC are very similar for all models and combinations of data. However, it is
clear that for plasma data alone that there is still some possibility of AUCs much
smaller than the mode. The inclusion of excretion data appears to rule out the
possibility of much smaller AUCs for female rats.
Discussion
We have performed Bayesian estimation of the distribution of parameters for one-
and two-compartment pharmacokinetics of PFOA in male and female rats. We
found that estimates could be obtained by analyzing 24 rats of each gender for
which plasma concentration data is available alone, and that also including
excretion data from twelve additional rats of each gender improved these estimates.
We conducted this research with two goals. We wished to not only characterize
the pharmacokinetics of PFOA in rats, but also to assess the benefits of Bayesian
0 Springer
Previous
TOC
-------
708 J Pharmacokinet Pharmacodyn (2008) 35:683-712
analysis of pharmacokinetic models in general as Bayesian methods are likely to be
of great value for more complex models such as PBPK [43].
The pharmacokinetics of PFOA
For both male and female rats we find that pharmacokinetic parameter estimates are
improved when plasma data is carefully combined with excretion data. Our initial
attempts to model elimination as a sum of the two types of excretion resulted in
parameter estimates where uncertainty increased as additional data was used. Since
additional data made the parameter estimates for the initial model seem more
unlikely, we modified the model; at first by allowing a third, unspecified route of
elimination. With this change uncertainty was reduced by including additional data
but the estimated amount of PFOA eliminated by this unknown route was quite
large.
What appeared to be a third elimination route for the one-compartment model
was consistent with the distribution to a second, deep-tissue compartment: for a
two-compartment model the estimated amount of PFOA cleared by the unknown
route was negligible. Thus, analyzing the uncertainty about one-compartment model
parameter estimates and jointly considering the plasma and excretion data forced a
more complicated, though still relatively simple, model. This is the main reason we
believe that the two-compartment model is more appropriate for PFOA pharma-
cokinetics in the male rat.
A quantitative measure of model fitness, the DIG, also indicates that the two-
compartment model is better supported by available data. This means that the
additional uncertainty from estimating two more parameters per individual appears
to be offset by better matching the data. For the female data, however, the plasma
AUC does not seem to differ significantly between the one- or two-compartment
model. This suggests that for some applications simple, empirical models may be
adequate to describe PFOA pharmacokinetics in female rats.
For male rats we estimate a population mean clearance of 1.24 ± 1.08 ml/kg/h
and a total volume of distribution of 453.9 ± 455.41 ml/kg for male Sprague-
Dawley, compared to 2.1 ± 0.6 ml/kg/h and 345.6 ± 57.3 ml/kg in Kudo et al.'s
[16] studies of Wistar rats. For female Sprague-Dawley rats we estimate a clearance
of 30.33 ± 13.15 ml/kg/h and a total volume of distribution of 1566.40 ±
2507.94 ml/kg compared to 93.06 ± 33.54 ml/kg/h and 211.2 ± 28.2 ml/kg for
the Wistar rats of Kudo et al. [16]. Because of the rapid clearance of PFOA from
female rats, the larger volume of distribution is likely due to the diminished
influence of, and thus greater uncertainty about, the second compartment for the
female time-courses—since we report the medians of parameter distributions, broad
distributions will have large median values.
PFOA clearance is strikingly faster in female than in male rats, complicating
extrapolation of rat exposures to human. In both cases the clearance is much more
rapid than would be consistent with the 3.8 year human half-life [11] and
extrapolation is further complicated by the lack of a sexual dimorphism in human
clearance [13]. It has been hypothesized that renal transporters may act on PFOA,
inhibiting renal excretion [12]. The function of these transporters have been shown
4y Springer
Previous I TOC
-------
J Pharmacokinet Pharmacodyn (2008) 35:683-712 709
to be modulated by sex hormones in rats, possibly explaining the difference between
rat genders [17]. However, in vitro studies of the affinity of these transporters for
PFOA have not found significant differences between rats and humans, possibly
indicating that expression of the transporters may drive the interspecies differences
[18].
In studies of Sprague-Dawley rats by Vanden Heuvel et al. [15] 91% of PFOA
was eliminated through urine within 24 h of dosing female rats. In male rats, the
cumulative excretion of PFOA was 36.4% to urine and 35.1% to feces after 28 days
(672 h). Using a one-compartment model and the data of Kudo et al. [16], Harada
et al.[44] calculated that 5 h after dosing 91.4 ± 50.2% of PFOA that was cleared
from plasma had been excreted through urine in males, while 47.2 ± 22.1% of
cleared PFOA was excreted through urine in females. Using the two compartment
model we estimated that overall, 77 ± 9% is excreted to urine and 8 ± 6% is
excreted to feces in female Sprague-Dawley rats in the Kemper data [20]. For male
Sprague-Dawley rats, we estimate that 85 ± 7% and 12 ± 6% of the PFOA
eliminated from plasma is excreted to urine and feces respectively.
The presence of residual PFOA in the animal carcass, confirmed by the
experimenters [20], indicates that more complicated models that include additional,
deeper compartments may be needed to accurately describe PFOA pharmacokinet-
ics. Research by Kudo et al. [45] has found that small (0.041 mg/kg) and large
(16.56 mg/kg) doses of PFOA distribute very differently to liver in Wistar rats,
making it very likely that the PFOA kinetics are concentration-dependent. We did
not observe trends among the two-compartment parameter estimates receiving
different doses, but since we assumed that all of our parameters were dose-
independent only very large trends should be observable.
Concentration-dependent models for PFOA PK, possibly including differential
levels of transporter expression, should be investigated in order to develop inter-
species extrapolations. For such models it will not be possible to use analytic
pharmacokinetic solutions, since the kinetics will change as the PFOA is distributed
and cleared. In order to perform a Bayesian analysis using MCMC it will be
necessary to solve differential equations for the relevant parameters for each
iteration of the Markov chains—greatly increasing the time per iteration. If new
models will allow better interspecies extrapolation then this is a computational
hurdle worth overcoming.
Bayesian analysis of pharmacokinetic models
Using fairly uninformative, though bounded, uniform priors we can obtain estimates
of some parameters both for specific test animals and the population from which the
test animal were drawn. We specify three models, one each for the pharmacoki-
netics, the distribution of parameters within the population of rats, and the
measurements performed on each test animal. A clear benefit of the Bayesian
approach is the requirement for explicit, quantitative descriptions of the pharma-
cokinetic model and prior knowledge of the parameter distributions. Estimates
obtained given a fully described Bayesian analysis can be replicated because no
further "goodness of fit" assumptions are made about the values of the parameters.
4y Springer
Previous I TOC
-------
710 J Pharmacokinet Pharmacodyn (2008) 35:683-712
If the Markov chains have converged, then a representative sample of the possible
combinations of parameters has been examined with respect to the likeliness of
having produced the observations.
Convergence is a primary drawback of Bayesian analysis performed with
MCMC, especially when using uninformative priors. For the empirical models
studied here, chain lengths on the order of a million iterations were required to
determine posterior distributions. These models have analytic solutions, so
determining the predictions based on the new parameters for each iteration is
relatively rapid. If more complicated models require numerically solving systems of
differential equations, then the time for each iteration may increase greatly. Part of
the need for large numbers of iterations is driven, however, by the necessity of
vague prior distributions on empirical pharmacokinetic parameters. One advantage
of more complicated, biologically based models may be that there should be a trade-
off between the time needed per iteration to solve more sophisticated models and
the additional available information about physiological parameter values that can
reduce the number of iterations needed to converge.
There are many advantages of having performed a Bayesian analysis of
pharmacokinetic data. Using sets of parameters drawn from the posterior
distribution, we are able to easily calculate probability distributions for quantities
that depend upon the parameter, e.g. the Area Under the Curve for the plasma
concentration as a measure of tissue exposure. By analyzing a hierarchical
statistical model we also found distributions for the population hyper-parameters,
allowing us to simulate new individuals under various dose conditions. The
hierarchical model also allowed us to analyze the gains and uncertainty in using
different types of data—plasma concentration and excretion—to inform the
estimates of different parameters that are not identifiable with just one type of the
data alone.
Another, very useful, outcome of Bayesian analysis is that we identify certain
parameters that, while possibly important to the underlying biology or pharmaco-
kinetics, simply are not sufficiently informed by the data to be estimated. When the
estimated posterior distribution is too similar to the assumed prior distribution we
know that our assumptions, and not the data, are driving the parameter estimation.
Recognizing when these situations might occur could inform the design of potential
future experiments.
Problems with estimating parameters also inform model design. For instance,
when we found that we could not constrain the total clearance to be the sum of the
fecal and urinary excretion, we changed to estimating excretion as a fraction of the
total clearance and learned that there may be some fraction of the elimination of
PFOA from the plasma for which we do not account.
Finally, Markov-chain Monte Carlo approaches to Bayesian analysis allow direct
comparison of different models. As in our comparison of one and two-compartment
empirical models, we can generally use the DIG to evaluate whether additional
information is gained from the data by the inclusion of additional model parameters.
In this way we can determine whether to expand or contract a model. This technique
should be extremely useful in constructing and comparing physiologically based
pharmacokinetic models.
4y Springer
Previous I TOC
-------
J Pharmacokinet Pharmacodyn (2008) 35:683-712 711
Acknowledgments and Disclaimer The United States Environmental Protection Agency through its
Office of Research and Development funded and managed the research described here. Interagency
Agreement RW-75-92207501 with the National Toxicology Program at the National Institute for
Environmental Health Science was a partial source of funding. This research has been subjected to
Agency review and approved for publication.
References
1. Andersen ME, Clewell HJ, Frederick CB (1995) Applying simulation modeling to problems in
toxicology and risk assessment—a short perspective. Toxicol Appl Pharmacol 133:181-187
2. Gelman A, Bois FY, Jiang J (1996) Physiological pharmacokinetic analysis using population mod-
eling and informative prior distributions. J Am Stat Assoc 91:1400-1412
3. Parham FM, Matthews HB, Portier CJ (2002) A physiologically based pharmacokinetic model of
/?,/?'-dichlorodiphenylsulfone. Toxicol Appl Pharmacol 181:153-163
4. Keys DA, Wallace DG, Kepler TB, Conolly RB (1999) Quantitative evaluation of alternative
mechanisms of blood and testes disposition of di(2-ethylhexyl) phthalate and mono(2-ethylhexyl)
phthalate in rats. Toxicol Sci 49:172-185
5. Collins AS, Sumner SCJ, Borghoff SJ, Medinsky MA (1999) A physiological model for tert-amyl
methyl ether and tert-amyl alcohol: hypothesis testing of model structures. Toxicol Sci 49:15-28
6. Cobelli C, DiStefano JJ (1980) Parameter and structural identifiability concepts and ambiguities: a
critical review and analysis. Am J Physiol Regul, Integr Comp Physiol 239:7-24
7. Bernillon P, Bois FY (2000) Statistical issues in toxicokinetic modeling: a Bayesian perspective.
Environ Health Perspect 108:883-893
8. Mezzetti M, Ibrahim JG, Bois FY, Ryan LM, Ngo L, Smith TJ (2003) A Bayesian compartmental
model for the evaluation of 1,3-butadiene metabolism. J R Stat Soc Ser C-Appl Stat 52:291-305
9. Gueorguieva I, Aarons L, Rowland M (2006) Diazepam pharamacokinetics from preclinical to phase
I using a Bayesian population physiologically based pharmacokinetic model with informative prior
distributions in Winbugs. J Pharmacokinet Pharmacodyn 33:571-594
10. Lau C, Anitole K, Hodes C, Lai D, Pfahles-Hutchens A, Seed J (2007) Perfluoroalkyl acids: a review
of monitoring and toxicological findings. Toxicol Sci 99:366-394
11. Olsen GW, Burris JM, Ehresman DJ, Froehlich JW, Seacat AM, Butenhoff JL, Zobel LR (2007)
Half-life of serum elimination of perfluorooctanesulfonate, perfluorohexanesulfonate, and perflu-
orooctanoate in retired fluorochemical production workers. Environ Health Perspect 115:1298-1305
12. Andersen ME, Clewell HJ, Tan Y-M, Butenhoff JL, Olsen GW (2006) Pharmacokinetic modeling of
saturable, renal resorption of perfluoroalkylacids in monkeys-probing the determinants of long
plasma half-lives. Toxicology 227:156-164
13. Calafat AM, Wong L-Y, Kuklenyik Z, Reidy JA, Needham LL (2007) Polyfluoroalkyl chemicals in
the US population: data from the National Health and Nutrition Examination Survey (NHANES)
2003-2004 and comparisons with NHANES 1999-2000. Environ Health Perspect 115:1596-1602
14. Washburn ST, Bingman TS, Braithwaite SK, Buck RC, Buxton LW, Clewell HJ, Haroun LA,
Kester JE, Rickard RW, Shipp AM (2005) Exposure assessment and risk characterization for
perfluorooctanoate in selected consumer articles. Environ Sci Technol 39:3904-3910
15. Heuvel JPV, Kuslikis BI, Rafelghem MJV, Peterson RE (1991) Tissue distribution, metabolism, and
elimination of perfluorooctanoic acid in male and female rats. J Biochem Toxicol 6:83-92
16. Kudo N, Katakura M, Sato Y, Kawashima Y (2002) Sex hormone-regulated renal transport of
perfluorooctanoic acid. Chemico-Biol Interact 139:301-316
17. Katakura M, Kudo N, Tsuda T, Hibino Y, Mitsumoto A, Kawashima Y (2007) Rat organic anion
transporter 3 and organic anion transporting polypeptide 1 mediate perfluorooctanoic acid transport. J
Health Sci 53:77-83
18. Nakagawa H et al. (2007) Roles of organic anion transporters in the renal excretion of perfluoroc-
tanoic acid. Basic Clin Pharmacol Toxicol 103:1-8
19. Loveless SE, Finlay C, Everds NE, Frame SR, Gillies PJ, O'Connor JC, Powley CR, Kennedy GL
(2006) Comparative responses of rats and mice exposed to linear/branched, linear, or branched
ammonium perfluorooctanoate (apfo). Toxicology 220:203-217
20. Kemper RA (2003) Perfluorooctanoic acid: toxicokinetics in the rat, DuPont Haskell Laboratories,
Laboratory Project ID: DuPont-7473. USEPA Administrative Record AR-226.1499
4y Springer
Previous I TOC
-------
712 J Pharmacokinet Pharmacodyn (2008) 35:683-712
21. Butenhoff JL, Gaylor DW, Moore JA, Olsen GW, Rodricks J, Mandel JH, Zobel LR (2004) Char-
acterization of risk for general population exposure to perfluorooctanoate. Regul Toxicol Pharmacol
39:363-380
22. Portier CJ, Lyles CM (1996) Practicing safe modeling: GLP for biologically based mechanistic
models. Environ Health Perspect 104:806
23. Sheiner LB, Beal SL (1980) Evaluation of methods for estimating population pharmacokinetic
parameters. I. Michaelis-Menten model: routine clinical pharmacokinetic data. J Pharmacokinet
Pharmacodyn 8:553-571
24. Lunn DJ, Thomas A, Best N, Spiegelhalter D (2000) WinBUGS—a Bayesian modeling framework:
concepts, structure and extensibility. Stat Comput 10:325-337
25. Lunn DJ, Best NG, Thomas A, Wakefield J, Spiegelhalter D (2002) Bayesian analysis of population
PK7PD models: general concepts and software. J Pharmacokinet Pharmacodyn 29:271-307
26. Ihaka R, Gentleman R (1996) R: a language for data analysis and graphics. J Comput Graph Stat
5:299-314
27. Sturtz S, Ligges U, Gelman A (2005) R2WinBUGS: a package for running WinBUGS from R. J Stat
Software, 12
28. Lunn DJ (2003) WinBUGS development interface (WBDev). ISBA Bull 10:10-11
29. Oberon Microsystems Inc (2001) Component pascal language report
30. Gelman A, Carlin JB, Stern HS, Rubin DB (2004) Bayesian data analysis, 2nd edn. Chapman and
Hall/CRC
31. Gelman A (2006) Prior distributions for variance parameters in hierarchical models. Bayesian Anal
1:515-533
32. Lambert PC, Sutton AJ, Burton PR, Abrams KR, Jones DR (2005) How vague is vague? A simu-
lation study of the impact of the use of vague prior distributions in MCMC using WinBUGS. Stat
Med 25:2401-2428
33. Rocke DM, Lorenzato S (1995) A two-component model for measurment error in analytic chemistry.
Technometrics 37:176-184
34. Bois FY, Gelman A, Jiang J, Maszle D, Zeise L, Alexeef G (1996) Population toxicokinetics of
tetrachloroethylene. Arch Toxicol 70:347-355
35. Chiu WA, Bois FY (2006) Revisiting the population toxicokinetics of tertracholoroethylene. Arch
Toxicol 80:382-385
36. Best NG, Cowles MK, Vines K (1995) CODA: convergence diagnosis and output analysis for Gibbs
sampling output
37. R Development Core Team (2008) R: a language and environment for statistical computing, ISBN
3-900051-07-0, URL http://www.R-project.org
38. Gelman A, Rubin DB (1992) Inferences from iterative simulation using multiple sequences. Stat Sci
7:457-472
39. Raftery AE, Lewis SM (1992) Comment: one long run with diagnostics: implementation strategies
for Markov Chain Monte Carlo. Stat Sci 7:493^1-97
40. Dodds MG, Vicini P (2004) Assessing convergence of Markov Chain Monte Carlo simulations in
hierarchical Bayesian models for populations pharmacokinetics. Ann Biomed Eng 32:1300-1313
41. Spiegelhalter DJ, Best NG, Carlin BP (2002) Bayesian measures of model complexity and fit. J R Stat
Soc B 64:583-639
42. Spiegelhalter DJ, Best NG, Carlin BP (1998) Bayesian deviance, the effective number of parameters,
and the comparison of arbitrarily complex models. Technical Report, Medical Research Council
Biostatistics Unit, Cambridge, UK
43. Barton HA et al. (2007) Characterizing uncertainty and variability in physiologically based phar-
macokinetic models: state of the science and needs for research and implementation. Toxicol Sci
99:395-402
44. Harada K, Inoue K, Morikawa A, Yoshinaga T, Saito N, Koizumi A (2005) Renal clearance of
perfluorooctane sulfonate and perfluorooctanoate in humans and their species-specific excretion.
Environ Res 99:253-261
45. Kudo N, Sakai A, Mitsumoto A, Hibino Y, Tsuda T, Kawashima Y (2007) Tissue distribution and
hepatic subcellular distribution of perfluorooctanoic acid at low dose are different from those at high
dose in rats. Biol Pharm Bull 30:1535-1540
46. Stone CJ, Hansen M, Kooperberg C, Truong YK (1997) The use of polynomial splines and their
tensor products in extended linear modeling (with discussion). Ann Stat 25:1371-1470
4y Springer
Previous I TOC
-------
Environ. Sci. Technol. 2008, 42, 934-939
te
aid
ELAINE A. COHEN HUBAL,*' +
MARCIA G. NISHIOKA,*
WILLIAM A. IVANCIC.*
MICHELE MO KARA,* AND
PETER P. EGEGHY§
National Center for Computational Toxicology, U.S. EPA,
Research Triangle Park, North Carolina 27711, Battelle
Memorial Institute, Columbus, Ohio 43201, and National
Exposure Research Laboratory, U.S. EPA,
Research Triangle Park, North Carolina 27711
Received July 6, 2007. "Revised manuscript received October
31, 2007. Accepted November 5, 2007.
Transfer of chemicals from contaminated surfaces such as
foliage, floors, and furniture is a potentially significant source
of both occupational exposure and children's residential exposure,
Increased understanding of relevant factors influencing
transfers from contaminated surfaces to skin and resulting dermal-
loading will reduce uncertainty in exposure assessment. In a
previously reported study, a fluorescence imaging system was
developed, tested, and used to measure transfer of riboflavin
residues from surfaces to hands. Parameters evaluated included
surface type, surface loading, contact motion, pressure,
duration, and skin condition. Results of the initial study indicated
that contact duration and pressure were not significant for
the range of values tested, butthat there are potentially significant
differences in transfer efficiencies of different compounds. In
the study reported here, experimental methods were refined and
additional transfer data were collected. A second fluorescent
tracer, Uvitex OB, with very different physicochemical
properties than riboflavin, was also evaluated to better
characterize the range of transfers that may be expected for
a variety of compounds. Fluorescent tracers were applied
individually to surfaces and transfers to skin were measured
after repeated hand contacts with the surface. Additional trials
were conducted to compare transfer of tracers and co-
applied pesticide residues. Results of this study indicate that
dermal loadings of both tracers increase through the seventh
brief contact. Dermal loading of Uvitex tends to increase at
a higher rate than dermal loadings of riboflavin. Measurement
of co-applied tracer and pesticide suggest results for these
two tracers may provide reasonable bounding estimates
of pesticide transfer.
Introduction
Although monitoring for surface contamination in work with
radioactive materials and dermal monitoring of pesticide
exposure to agricultural workers have been standard practice
* Corresponding author phone: (919) 541-4077; e-mail:
hubal.elaine@epa.gov.
f National Center for Computational Toxicology, U.S. EPA.
* Battelle Memorial Institute.
§ National Exposure Research Laboratory, U.S. EPA.
934 « ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 42, NO. 3, 2008
for 50 years, regular surface sampling and dermal monitoring
methods have been applied to industrial and residential
contamination only since the 1980s. In recent years, there
have been significant advances in tools available to measure
and assess dermal exposures resulting from contacts with
contaminated surfaces (1). However, due to the complexity
of this system, there are still important gaps in our under-
standing of determinants of dermal transfer and how best
to measure and assess resulting exposure. To identify major
uncertainties associated with quantifying dermal exposures
resulting from contact with contaminated surfaces it is useful
to consider pathways and mechanisms for these exposures.
Transfer of contaminants from a contaminated surface to
skin is a function of (1) contaminant form (residue, particle,
formulation, age, physicochemical properties); (2) surface
characteristics (hard, plush, porous, surface loading, previous
transfer); (3) nature of interaction between contaminant and
surface; (4) skin characteristics (moisture, age, loading); (5)
contact mechanics (pressure, duration, smudge, repetition);
and (6) environmental conditions (temperature, relative
humidity). Currently, it is not clear which of these many
factors will drive transfer and under what conditions.
Increased understanding of the most significant factors for
influencing transfers from contaminated surfaces to skin and
resulting dermal loading is required to reduce uncertainty
in exposure assessment (2).
Transfer of pesticide residues has been studied previously
by hand press to pesticide spiked surfaces (3-6). Currently,
there are no direct methods for measuring pesticide residues
on hands. As such, in each of these studies the hand was
wiped or rinsed with 2-propanol to collect transferred
pesticide for quantification. In addition to uncertainty
introduced by use of a rinse or wipe, conducting studies
with direclcdpes (icicle contact to children is clearly unethical.
Video fluorescent imaging is one approach that has been
used successfully in bo th occupational and residential settings
to explore dermal exposure mechanisms and mitigation
strategies (7-9). Application of nontoxic fluorescent tracers
provides an opportunity to design studies that address
limitations of pesticide transfer studies and lend insight on
occupational and residential pesticide exposures to both
children and adults.
In a previously reported study, a fluorescence imaging
system was developed, evaluated (10), and used to measure
transfer of riboflavin (Vitamin B2) residues from surfaces to
hands for multiple contacts (II). Results of this initial study
indicated that surface loading and skin condition were
important parameters affecting residue transfer of riboflavin.
Contact duration and pressure were not significant for the
range of values tested, and surface type was not significant
after the first contact. Preliminary results also suggested
potentially significant differences in transfer efficiencies of
different compounds. Limitations of this study included
surface loadings that were relatively high compared with
contaminant loadings expected in residential environments
using current crack and crevice application methods. In
addition, only one fluorescent surrogate was evaluated
limiting our ability to evaluate the effect of compound
properties on transfer efficiency and to extrapolate results
to a range of current use pesticides.
In the study presented here, experimental methods
developed previously were refined and dermal transfer data
were collected using both .Uvitex OB (lipophilic) andriboflavin
(hydrophilic). Residue transfers to three types of dislodgeable
residue sampling tools were also collected using these two
fluorescent tracers and live different organophosphatc (OP)
10.1021/es071668h CCC: $40.75
© 2008 American Chemical Society
Published on Web 12/19/2007
Previoys
TOG
Next
-------
TABLE 1. Hani Contact Trials3
Trial Number
parameter
surface type
surface loading
contact type
skin condition
tracer
1
a
b
c
d
e
2
A
B
c
d
e
3
A
b
C
D
E
4
a
B
C
D
E
5
A
b
C
d
e
6
a
B
C
d
e
7
a
b
c
D
E
8
A
B
c
D
E
9
A
b
c
d
E
10
a
B
c
d
E
11
a
b
C
D
e
12
A
B
C
D
e
13
A
b
c
D
e
14
a
B
c
D
e
15
a
b
C
d
E
16
A
B
C
d
E
3 Key: Surface, skin, and contact parameters. A = carpet; a = laminate. B = low surface loading (0.2 g/cm2); b = high
surface loading (2 g/cm2). C = uniform press; c = smudge. D = dry hand; d = moist hand. E = Riboflavin; e = Uvitex,
and pyrethroid pesticides. There were two main objectives
of this study: evaluate impact of parameters related to
compound, surface, and skin on transfer efficiency; and relate
transfer of tracers to transfer of representative pesticides with
similar physicochemical properties.
and
Study Design. This study was performed using both riboflavin
and Uvitex OB as surrogates for pesticide residues, and video-
imaging technology to quantify dermal loadings of fluorescent
tracers following contact with tracer-treated surfaces. The
approach was as follows: (1) Apply fluorescent tracer to test
surfaces. (2) Conduct controlled hand-transfer experiments
varying selected parameters. (3) Measure mass of tracer
transferred and estimate surface and dermal contact areas.
(4) Assess relative transfers of tracers and pesticides using
transferable residue sampling techniques.
A number of fluorescent tracers were considered, espe-
cially those used in previous studies. Safety was the overriding
concern in choosing tracers. Two tracers were selected having
physicochemical properties that bound properties of several
pesticides of interest: Uvitex OB (Ciba Specialty Chemicals}
and riboflavin. Pesticides selected areofcurrentinterestdue
to widespread use in the United States for residential and
agricultural applications. Chlorpyrifos and diazinon were OPs
used most extensively in the indoor residential market and
are still being measured in U.S. homes. Pyrethroids are now
the dominant residential-use insecticides. Q's- and trans-
permethrin as well as esfenvalerate are commonly found in
homes at measurable levels.
Parameters evaluated in this study included tracer, surface
type, surface loading, contact motion, and skin condition.
Eight experiments or trials involving contact withriboflavin-
treated surfaces, and 8 experiments involving Uvitex-treated
surfaces (Table 1) were conducted; each experiment was
repeated in triplicate. Because riboflavin can be washed from
hands, 3 subjects were recruited for riboflavin experiments;
each person completed all 8 experiments. In contrast, because
Uvitex cannot be washed from hands, 24 subjects were
recruited to gather triplicate data for each of the 8 Uvitex
experiments.
As described previously (2), the Youden ruggedness test
(.12) was used to select parameter combinations for each
trial. By using this design, more than one parameter could
be varied at a time minimizing the number of trials required
to test for main effects of all parameters. The experimental
plan used here is a l/2fractional replication of a3 x 25factorial
(Table 1). Tested parameter values for this study and the
previous study are summarized in Table 2. Available data for
octanol/water partition coefficient, vapor pressure, and water
solubility for tested pesticides and tracers are listed in Table
3. Study design, protocols, and consentforms wereapproved
by the Battelle Memorial Institute IRB for use of human
subjects, and subsequently reviewed by the EPA administrator
for human subjects experiments.
Application of Tracers to Test Surfaces. General protocols
for spray applications to surfaces have been discussed
TABLE 2.
parameter
tracer
skin condition
surface type
surface loading
contact motion
contact duration
contact pressure 0.1 or 1 psi
contact number multiple
initial experiments refined experiments"
riboflavina riboflavinfa or Uvitex"
dry, moist, or sticky dry or moist
carpet or laminate carpet or laminate
0.2 or 2 ag/cm2
press or smudge
2sd
0.1 psid
multiple
2 or 10 ag/cm2
press or smudge
2 or 20 s
'"* Refined experiments added Uvitex, reduced loading
levels, and reduced number of parameters tested, b Relatively
water soluble. c Relatively water insoluble. rf Parameter was
not varied in study.
previously (2). Detail of differences specific to this study are
presented in the Supporting Information (Section SI). In
general, an improved spray system was used to deliver
smaller, finer droplets to test surfaces than in the previous
study. Measured variability in loading across a platform was
25% with riboflavin solution and 14% with Uvitex OB solution.
Each application surface (platform), 60 cm x 180 crn, was
platted into 3 rows of 11 blocks. One block per row was
allocated to a deposition coupon, 10 blocks were used for
dermal contact. Each block was used only once.
Contact Trials and Transfer Off Protocols. For each
experiment, the subject was instructed on contact motion and
skin condition for surface contact. Contact duration was held
at 2 s and contact pressure was held at approximately 0.1 psi
for all experiments. For dry condition, hands were washed, dried,
and then held up in room air for 30 s. For moist condition,
hands were washed, dried and then held 8-10 cm away from
outlet of CoolMist vaporizer for 20 s. To familiarize subjects
with the feel of 0.1 psi contact, subjects practiced 10 presses on
a scale prior to initiating an experiment. The subject contacted
surface in an unused area (block), had the hand imaged, and
then repeated contact motion in a new area. A series of seven
sequential contacts with surface using the same motion
constituted one experiment.
Measurement of Dermal Loading. The fluorescent imag-
ing system described in Ivancic et al. (1) was used to monitor
and measure fluorescence on hands following contact with
tracer-treated surface. Details of the fluorescent lamp
configurations and wavelength settings for each of the tracers
are presented in the Supporting Information (Section S2).
For tests with riboflavin, a full calibration curve of
riboflavin (different points) was obtained with each subject.
Different amounts of a lOO^g/mL aqueous riboflavin solution
were deposited on the hand to simulate different riboflavin
loadings. In contrast, for Uvitex, each subject completed one
experiment with his right hand, and a series of 3-6 different
calibration curve points was imaged on his left hand. Uvitex
OB calibration curve solution was prepared as an aqueous
solution of Uvitex in 0.1% Pluronic (ernulsiiier to suspend
Previoys
VOL. 42, NO. 3, 2008 / ENVIRONMENTAL SCIENCE & TECHNOLOGY » 935
TOG
Next
-------
TABLE 3. Properties of Testei Pesticiies and Tracers
analyte
diazinon
chlorpyrifos
tech. permethrinc
c/s-permethrin
trans-permethrin
esfenvalerate
riboflavin
uvitex OB
aRef (73). fcRef (74). c Technical
octanol/water
partition
6,400°
50,000^
3,160,000d
naf
na
1,660,000^
Q.035g
not available
grade. d Ref (15), ° Ref (16).
fapor pressure
(mPa)
0,097 @ 20 °Cb
2.5 @ 25 "C^
0.0013 @ 20°Ce
0,0025 @ 20 °C
0,0015 @ 20 °C
0.067 @ 25 °Cb
negligible
0,000003 @ 20 °C''
fNot available. g Ref (IT).
water solubility
(mg/L)
40 @ 20 °Cb
2 @ 25 cCb
0.2 ca 20 CC&
na
na
<0.3 @ 25 °C&
150h
negligible'
ft Determined here. '' Ref ( 78).
Uvitex in water) to accurately simulate spectral interferences
that m ight arise from contact with surfaces that also contai ned
Pluronic.
Subject-specific calibration curves for riboflavin were not
significantly different. As such, the three individual riboflavin
curves were merged into one universal calibration curve and
applied to all data. Similarly, Uvitex calibration data from all
subjects were merged into one calibration curve and applied
to all Uvitex measurements. Details of the calibration
approach are presented in the Supporting Information
(Section S2).
Coapplied Pesticide and Tracer Transferable Residue
Sampling. Detailed application and sampling methods are
presented in the Supporting Information (Section S3).Two
separate mixtures of five pesticides and selected tracer were
applied to carpet and laminate platforms. One mixture
consisted of distilled/ deionized water doped with an aqueous
solution of riboflavin, concentrated acetonitrile solutions of
diazinon, cis- and rrans-permethrin, and esfenvalerate, and
a commercial "ready-mix" aqueous formulation of 6%
chlorpyrifos (Spectracide brand). Cis- and frans-permethrin
doping solution was prepared from neat material in 1:4 ratio
of cisl trans isomers. The second mixture consisted of
distilled/deionized water with 0.1% Pluronic F68 doped with
concentrated acetone solution of Uvitex OB and the same
pesticide solutions. Application loading of analytes, except
cis-permethrin, was 0.2 «g/cm2 for all of these tests; loading
of cis-permethrin was 0.04 ag/cm2. Three methods were used
to collect transferable residue samples from test surfaces:
aqueous wipe, CIS Empore (3M) Press Disk, and PUF roller.
In addition, deposition coupons were used to verify surface
loading.
Detailed analytical methods are presented in the Sup-
porting Information (Section S3). Pesticide QC spike samples
had recoveries as follows: 85-105% (depending on pesticide)
for deposition coupons (10/^g), 87-106% for aqueous wipes
(10 «g), 20-46% for PUF roller sleeves (1 /.«g), and 89-103%
for C18 disks (25 ng). PUF Roller data were surprising, as
previous testing of method showed recoveries of 120-130%.
For these QC spike samples analyzed with field samples,
when coapplied with riboflavin, recoveries of pcs ticides from
PUF were 45 ± 1, 36 ± 3, 30 ± 3, 31 ± 5, and 35 ± 4% for
diazinon, chlorpyrifos, cis-permethrin, rrans-permethrin, and
esfenvalerate, respectively. When coapplied with Uvitex,
recoveries from PUF were 18 ± 3, 20 ± 5, 16 ± 4,17 ± 1, and
17 ± 2% for diazinon, chlorpyrifos, cis-permethrin, trans-
permethrin, and esfenvalerate. PUF Roller sample data were
corrected by individual spike recoveries. Pesticides were not
detected in field matrix blanks.
Uvitex OB QC; spike samples had recoveries as follows:
89% for deposition coupons (lO^g), 69% for aqueous wipes
(lO^g), 94% for PUF Roller sleeves (1 ;tg), and 127% for CIS
disks (25 ng). Uvitex OB was not detected in matrix blanks
with exception of low levels in PUF Roller sleeve and C18
disks; corrections for these were made to data. Riboflavin
QC spike samples had recoveries as follows: 98% for
deposition coupons (10,«g), 91% for aqueous wipes (10,«g),
90% for PUF Roller sleeves (1 «g), and an anomalous 468%
for 25 ng applied to CIS disks. Earlier studies showed
recoveries of 85% for this method, so this was assumed to
be an outlier. Riboflavin was not detected in matrix blanks
with exception of low levels in CIS press disks sample extracts.
Statistical Methods. Descriptive statistics (mean and
standard deviation) were calculated for measured loading
on and percent transferred to hands. Multifactorial ANOVA
tests were performed to determine which experimental
parameters were statistically significant predictors of cither
hand loading or transfer efficiency using data only from the
initial contact between hand and surfaces. Parameters
evaluated in these two models were surface type, surface
loading, contact type, and skin condition. A linear mixed-
effects regression test was used to evaluate the above
parameters, with addition of contact number, on repeated
hand loading measurement data. Differences between tracers
in the rate of transfer to skin were evaluated with an
interaction term in the mixed-effects model.
Results
Hand Contact Trials. A summary of results for individual
experiments is presented in Table SI (Supporting Informa-
tion). Averages for the 3 subjects are shown for initial contact
and seventh contact in terms of loading on hand (ag/cm2)
and overall percent transfer. The range of dermal loading
after first contact was 0.01-0.62 ag/cm2, and the range of
loading after seventh contact was 0.02-1.69 «g/cm2. Overall
percent transfer ranged between 0.8 and 45.5% for the first
contact, and 0.6-19.4% for the seventh contact. Overall
percent transfer takes into account total area contacted over
the multiple presses, defined previously (2).
Results for this study compare well with previous data
reported in the literature. Camman ct al. (8) studied press
transfer of pesticide residues to moistened hands. Resulting
transfer efficiencies ranged from 1 to 5%. Hsu et al. (5)
evaluated transfer of thirteen different pesticides from
aluminum foil to the hand heel following a series of 10 presses.
Both dry and moist hands were studied; two different press
times (1 and 5 s per press) and two different hand motions
were also tested. The effect of varying these parameters was
not evaluated. Mean pesticide transfer efficiencies were
between 5 and 16% with a relative standard deviation greater
than 30%. Mean dermal recovery did not differ significantly
among evaluated pesticides.
Hand loading by contact number is compared for the two
tracers (riboflavin or Uvitex) at the two surface loadings (2
(Kg/cm2 or 0.2 ag/cm2) and presented in Figure 1. Results of
this study indicate that median dermal loadings of both
tracers increase in a near linear fashion through the seventh
contact. Dermal loading of Uvitex tends to increase at a higher
rate than dermal loading of riboflavin: 0.13 versus 0.069^/,g/
cm2/contact, respectively, at the higher surface loading, and
936 m ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 42, NO. 3, 2008
Previoys
TOG
Next
-------
Loading by Contact No., Follow—Up Experiment
Loading = high
Loading by Contact No., Follow-Up Experiment
Loading = high
2400-
2200-
2000-
1800-
1600-
1400-
1200-
1000-
800-
600-
400-
200-
0-
-200-
Analyte=Riboflavm 2400-
2200-
2000-
-,- 1800-
£• 1600-
E
^ 1400-
— i— ^
~T ^ 1200-
01
1 '1 1000-
— j— [-1— | 0
1 , ° 800-
1 1 I—1—! , C
1 _^_ , — in 600-
r~i li 400-
1 1 1 1 — 1 — 1 — —
0-
i i i i i i i -200-
0
-T
R
-1-
1234567 1
Contact Number
•
bd
2
Ana lyte= Uvitex
•
i
-L
m
m
•
~r
I
I 1 1 -L
34567
Contact Number
Loading by Contact No., Follow—Up Experiment
Loading = low
Loading by Contact No., Follow-Up Experiment
Loading = low
1300-
1200-
1100-
1DOO-
^ 900-
CN
| 800-
g> 700-
o> 600-
| 500-
.c 400-
1/1 300-
200-
100-
0-
-100-
Analyte=Riboflavin
1300-
1200-
1100-
1000-
^ 900-
IN
U 800'
a. 700-
o g> 600 -
a 500-
—r- ,i 400-
1/1 300-
^ ™ 0-
i i i i i i i — 10CH
1234567
Ana lyte= Uvitex
•
T
1
1
Coniac-t Number
1
b
E
2
I
3
I
4
Contact
FIGURE 1. Hand loading by contact number using riboflavin (left panels) or Uvitex (right
0.2 / 0,1
loading (//,g/crn2) p<0.01
surface type surface loading contact motion skin condition contact number
first contact (ANOVA)
p<0.05 p<0.01 p<0.1 p>0,1 b
p<0.05 p = 0.001 p< 0.001 p>0.1
repeated contact (Mixed-Effects Model)
p>0.1 p< 0.001 p< 0.001 p<0.05 p< 0.001
3 Bold text indicates parameter is statistically significant at p < 0.05. b Contact number only relevant to repeated contact
evaluation.
0.059 versus 0.017/^g/cm2/contact, respectively, at the lower
surface loading. Rates of change are significantly different (p
< 0.0001) in each condition. Atthe0.2,Mg/ cm2 surface loading
(low loading), Uvitex transfers more efficiently than riboflavin.
Results of statistical analysis to identify significant pa-
rameters for characterizing residue transfers are presented
in Table 4. These results show that effects of surface loading
on transfer from surfaces to hands under study conditions
is significant at alpha = 0.05. Surface type is significant only
with initial contact and skin condition is significant only
with repeated contacts. Comparison of "first contact" to
"repeated contact" results suggests that effect of surface type
appears to diminish with repeated contact while effect of
skin condition appears to increase with repeated contact.
Although effect estimates are similar for surface type: 0.126
(initial) versus 0.127 (repeated) there is loss of significance
VOL. 42, NO. 3, 2008 / ENVIRONMENTAL SCIENCE & TECHNOLOGY » 937
Previews
TOG
Next
-------
from from from from from
laminate carpet iaminate carpet laminate carpeJ
Transfer to Aqueous Transfer to PUF Roller Transfer to C1S Press
Wipe Disk (20 sec}
Transfer Efficiency (%Transfer) for Pesticides and Uvitex OB
and Riboflavin
O DiaziTOfi
O CMorpy rites
Dcis-Permelhrin
e trans -Permeihrtrfe
oEsfenvaierale
; "|_ m Ribaflavtrj
'"flj
1'iU
from
carpeJ
8 Press
TABLE 5. Evidence of
Surface-to-Skin Transfer
Importance of
Experiments3
parameter initial experiments
tracer
skin condition
surface type
surface loading
contact motion
contact duration
contact pressure
contact number
•o
oo
•o
•o
oo
oo
••
Factors Tested across
refined experiments
•O
•O
•O
••
•o
__
-
••
10 ',
s '
CL '
0.01 .U A
^!&"i
%= n
\l '•
v^ ( ^;
<=, -
,, =
fti I :l
HI
s I ~ '•
* • , " :
^ -.
H
^ : ' ?
-------
pesticides shortly after application. Compounds in other
forms (e.g., particle bound) may transfer differently.
On the whole, data developed in these studies will reduce
uncertainty in screening-level exposure assessments that are
based on limited default assumptions. In particular, these
results are currently being used with the SHEDS model to
improve estimates of exposures resulting from hand-to-
mouth behavior (21,22). However, the importance of multiple
contacts for characterizing residue transfers to skin and the
need to link dermal loading with absorption to characterize
dose suggest that measurement and modeling approaches
incorporating important temporal aspects of the system need
to be adapted for use in assessing exposures resulting from
dermal contact with contaminated surfaces.
Acknowledgments
The authors acknowledge laboratory assistance of K.Andrews,
J. Sowry, M. McCauley, A. Gregg, and C. Lukuch of Battelle.
We also acknowledge Tom McCurdy of U.S. EPA for helpful
review of this manuscript. The United States Environmental
Protection Agency through its Office of Research and
Development funded the research described here under
contract 68-D-99-011 to Battelle. It has been subjected to
Agency review and approved for publication.
Supporting Information Awailable
Detailed methods and summary results for the transfer
experiments (both tracer and pesticide). This information is
available free of charge via the Internet at http://pubs.acs.org.
Literature
(1) Fenske, R. A. Dermal exposure: a decade of real progress. Ann.
Occup. Hyg. 2000, 44 (7), 489-191.
(2) Cohen Hubal, E. A.; Sheldon, L. S.; Zufall, M. J.; Burke, J. M.;
Thomas, K. W. The challenge of assessing children's exposure
to pesticides./. Exposure Anal. Environ. EpidemioL 2000, 10,
638-649.
(3) Hsu, J. P.; Camann, D. E.; Schatterberg, H., 111.; Wheeler, B.;
Villalobos, K.; Garza, M.; Millard, P.; Lewis, R. G. New dermal
exposure sampling technique. In EPA/AWMA International
Symposium: Measurement of Toxic and Related Air Pollutants,
Raleigh, NC, 1990.
(4) U.S. EPA Protocol for Dermal Assessment: A Technical Report,
EPA/600/X-93/005;Office of Research and Development: Wash-
ington, DC, 1993.
(5) Geno, P. W.; Camann, D. E. Handwipe sampling and analysis
procedure for the measurement of dermal contact with
pesticides. Arch. Environ. Contam. Toxicol 1996, 30 (I), 132-8.
(6) Camann, D.; Harding, H. J.; Geno, P. W.; Agrawal, S. R.
Comparison of Methods to Determine Dislodgeable Residue
Transfer from Hoors;EPA/600/R-96/089; Office of Research and
Development: Washington, DC, 1996.
(7) Fenske, H. A. Nonuniform dermal deposition patterns during
occupational exposure to pesticides. Arch, Environ, Contam.
Toxicol 1990, 19, 332-7.
(8) Black, K. G.; Fenske, R. A. Dislodgeability of chlorpyrifos and
fluorescent tracer residues on turf: comparison of wipe and
foliar wash samplingtechniques. Arch, Environ. Contam. Toxicol,
1996, 31, 563-70.
(9) Fenske, R. A; Birnbaum, S. G. Second generation video imaging
technique for assessing dermal exposure (VITAE System). Am.
Ind. Hyg. Assoc. J. 1997, 58, 636-45.
(10) Ivancic, W. A.; Nishioka, M. G.; Barnes, H. H.; Cohen Hubal,
E. A. Development and evaluation of a quantitative video
fluorescence imaging system and fluorescent tracer for mea-
suring transfer of pesticide residues from surfaces to hands
with repeated contacts. Ann. Occup. Hyg. 2004, 48, 519-532.
(11) Cohen Hubal, E. A; Suggs, J. C.; Nishioka, M. G.; Ivancic, W. A.
Characterizing residue transfer efficiencies using a fluorescent
imaging technique. /. Exposure Anal. Environ. EpidemioL 2005,
15 (3), 261-270.
(12) Cochran,W. G.; Cox, G. M. Experimental Designs, 2nd ed.; John
Wiley and Sons: New York, 1957; Vol. 53, pp 2-540.
(13) Montgomery, J. H. Agrochemicals Desk Reference; CRC Press:
Boca Raton, FL, 1997.
(14) Kamrin, M. A. Pesticide Profiles; CRC Press: Boca Raton, FL,
1997.
(15) International Programme on Chemical Safety. Chemical Safety
Information from Intergovernmental Organizations; www.
inchem.org/documents.
(16) British Crop Protection Council. The Pesticide Manual; British
Crop Protection Council: Hampshire, U.K., 1997.
(17) SyracuseResearchCorporation.interkowjwww.syrres.com/esc/.
(18) Ciba Specialty Chemicals. Uvitex OB Fact Sheet; June. 1998.
(19) U.S. EPA. Important Exposure Factors for Children: An Analysis
of Laboratory and Observational Field Data Characterizing
Cumulative Exposure to Pesft'cofes;EPA600/R-07/013; Oficc of
Research and Development: Washington, DC, 2007, p 63; http://
www.epa.gov/nerl/research/data/ (accessed 06/01/07).
(20) Cohen Hubal, E. A; Egeghy, P; Leovic, K; Akland, G. Measuring
potential dermal transfer of a pesticide to children in a daycare
center. Environ. Health Perspect. 2006, 114 (2), 264-269.
(21) Ozkaynak, H. Personal communication. October 26, 2007.
(22) Zartarian, V. G.; Ozkaynak, H.; Burke, J. M.; Zufall, M. J.; Higas,
M. L.; Furtaw, E. J. A modeling framework for estimating
children's residential exposure and dose to chlorpyrifos via
dermal residue contact and non-dietary ingestion. Environ,
Health Perspect. 2000, 108 (6), 505-514.
ES071668H
Previous
TOC
VOL. 42, NO. 3, 2008 / ENVIRONMENTAL SCIENCE & TECHNOLOGY » 939
Next
-------
Commentary
Computational Molecular Modeling for Evaluating the Toxicity
of Environmental Chemicals: Prioritizing Bioassay Requirements
James R. Rabinowitz, Michael-Rock Goldsmith, Stephen B. Little, and Melissa A. Pasquinelli
National Center for Computational Toxicology, U.S. Environmental Protection Agency, Research Triangle Park, North Carolina, USA
BACKGROUND: The human health risk from exposure to environmental chemicals often must be
evaluated when relevant elements of the preferred data are unavailable. Therefore, strategies are
needed that can predict this information and prioritize the outstanding data requirements for the
risk evaluation. Many modes of molecular toxicity require the chemical or one of its biotransforma-
tion products to interact with specific biologic macromolecules (i.e., proteins and DNA). Molecular
modeling approaches may be adapted to study the interactions of environmental chemicals with
biomolecular targets.
OBJECTIVE: In this commentary we provide an overview of the challenges that arise from applying
molecular modeling tools developed and commonly used for pharmaceutical discovery to the prob-
lem of predicting the potential toxicities of environmental chemicals.
DISCUSSION: The use of molecular modeling tools to predict the unintended health and environ-
mental consequences of environmental chemicals differs strategically from the use of the same tools
in the pharmaceutical discovery process in terms of the goals and potential applications. It also
requires consideration of the greater diversity of chemical space and binding affinity domains than
is covered by pharmaceuticals.
CONCLUSION: Molecular modeling methods offer one of several complementary approaches to evalu-
ate the risk to human health and the environment as a result of exposure to environmental chemicals.
These tools can streamline the hazard assessment process by simulating possible modes of action and
providing virtual screening tools that can help prioritize bioassay requirements. Tailoring these strate-
gies to the particular challenges presented by environmental chemical interactions make them even
more effective.
KEY WORDS: computational toxicology, docking, enrichment, false negatives, high-throughput screen-
ing, molecular modeling, prioritizing bioassays, virtual screening. Environ Health Perspect 116:573—577
(2008). doi:10.1289/ehp.H077 available via http:lldx.doi.orgl [Online 1 February 2008]
A diverse spectrum of anthropogenic mole-
cules is found in the environment, including
chemicals introduced deliberately as well as
unintended by-products of human activity.
Through diligent monitoring, we are learn-
ing the identity, distribution, extent, and
environmental persistence of these chemicals.
To provide a reliable evaluation of the risk
presented by these compounds, information
about the specific molecules is required. This
includes knowledge of the interaction of the
chemicals with the environment and the
effects of the chemicals or their successors on
human health and ecologic systems.
The health and environmental effects of a
chemical derive from a continuum of processes
that proceed from the source of a chemical or
its predecessors to a set of outcomes. However,
it is often convenient to consider each process
in the continuum as a discrete entity [U.S.
Environmental Protection Agency (EPA)
2003]. Ideally, a risk assessment uses informa-
tion relative to the specific chemical being
considered. However, often the potential
effects of a chemical must be evaluated when
some relevant elements of the preferred data
matrix are missing. In these situations, an esti-
mate is derived by extrapolating from existing
information.
Various approaches, including computa-
tional methods, have been developed to model
these discrete steps in the source-to-outcome
paradigm. These models provide approxima-
tions of the missing experimental information
and a measure of the impact of specific miss-
ing data on the evaluation of risk. The mod-
els use existing information and can suggest
new experiments. As a result, the source-to-
outcome continuum becomes populated with
information that includes experimental data,
model-derived data, and connection models.
The toxicant—target paradigm is a computa-
tional approach that employs molecular model-
ing methods to estimate relevant interactions
and to populate the outcomes side of the
source-to-outcome continuum.
The Toxicant-Target Paradigm
The differential step in many mechanisms of
toxicity may be generalized as the interaction
between a small molecule (a toxicant) and
one or more macromolecular targets. Targets
include genetic material, receptors, transport
molecules, and enzymes. In addition, other
targets for toxicity could conceptually be
described. The difference in activity observed
between chemicals acting through the same
biologic mode of action may then be under-
stood as differences between their interactions
with putative targets.
Some molecular modeling methods have
been developed specifically to study interactions
of this type and are commonly employed for
the discovery of novel pharmaceutical agents
(Coupez and Lewis 2006; Sousa et al. 2006).
These methods can estimate the capacity of a
chemical to interact with a specific target and
cause a biologic effect. In the context of esti-
mating chemical toxicity, this approach can
yield predictions of the potential biologic
activity. These molecular modeling tools can
inform testing strategies or provide elements
in a scheme for estimating toxicity that also
include experimental results.
Molecular Modeling in
Computational Toxicology:
Probing Toxicant-Target
Structure
The toxicant—target paradigm can be used to
develop models for predicting chemical toxic-
ity. These models are composed of approxi-
mate mathematical descriptions of the
underlying physics and chemistry governing
the behavior of the interacting molecules.
These descriptions and their computational
implementations construct a bridge between
the information domains of experimental bio-
molecular structure and biologic effects.
Figure 1 depicts how molecular modeling can
be used to estimate chemical toxicity via the
toxicant—target paradigm.
Experimental information is used to pro-
vide a putative list of potential macromolecu-
lar targets related to chemical toxicity. For
some of these targets, structural information
is available or may be inferred from similar
structures via homology modeling (Hillisch
et al. 2004). The specific interactions
between potential toxicants and the struc-
tures of known targets may be modeled via
Address correspondence to M.A. Pasquinelli, North
Carolina State University, Fiber and Polymer
Science/Department of TECS, Campus Box 8301,
2401 Research Dr., Raleigh, NC 27695 USA.
Telephone: (919) 515-9426. Fax: (919) 515-6532.
E-mail: Melissa_Pasquinelli@ncsu.edu
During a portion of this work, M-R.G. was sup-
ported by National Health and Environmental Effects
Research Laboratory-Department of Environmental
Sciences and Engineering Training Agreement EPA
CT829471.
This work was reviewed by the U.S. Environmental
Protection Agency and approved for publication but
does not necessarily reflect official agency policy.
The authors declare they have no competing
financial interests.
Received 16 November 2007; accepted 1 February
2008.
Environmental Health Perspectives • VOLUME 1161 NUMBER 5 I May 2008
573
Previous
TOC
-------
Rabinowitz et al.
"docking" molecular modeling formalisms
(Kuntzl992).
In the absence of specific structural infor-
mation about the targets, an alternative is to
employ a ligand-based, cheminformatics strat-
egy. This method derives relationships among
various attributes of a database of ligands and
known target-based activities. The attributes
of the ligand may be simple or complex struc-
tural descriptions and properties that are
either measured or derived computationally
(Tong et al. 1997; Waller et al. 1996). Note
that these cheminformatics methods have also
been applied to predict chemical toxicity
without direct consideration of a target
(Prathipati et al. 2007), but methods of this
type are not the primary subject of this report.
With both the structural bioinformatics
and cheminformatics approaches, predictive
models are developed and tested with experi-
ments. A feedback process may be used to
improve the quality of the predictions. In
addition, these prediction tools can be used to
identify important missing experimental infor-
mation and relevant bioassays or properties
that are currently unavailable. The underlying
mechanism of action determines the range of
applicability of the model. In order to use this
approach as an element in a toxicity screen or
for developing bioassay strategies, a number of
choices must be made.
To a large extent, the pharmaceutical
industry has driven recent advances in the
design of molecular modeling tools for study-
ing the interactions between a small molecule
and a complex macromolecule (Jorgensen
2004). One approach for the discovery of
leads for developing novel pharmaceutical
agents employs computational "docking" of
each member of a chemical library to macro-
molecular targets that are chosen for potential
therapeutic benefit. Molecular docking is
designed to simulate the binding feasibility
and affinity of small molecules to protein tar-
gets (Abagyan and Totrov 2001; Halperin
et al. 2002). A docking calculation generates a
variety of poses of a small molecule within a
"binding region" of the macromolecular tar-
get, and typically includes ligand flexibility
(Sousa et al. 2006). At times, some form of
macromolecular flexibility (Carlson 2002) is
also included. An important component of the
docking simulation is to identify the potential
binding sites within a macromolecular target.
These sites could be an interior pocket or an
indentation on the macromolecular surface
(Huang and Schroeder 2006).
The calculation of a score assesses the
potential relevance of each docking pose.
Functions used for scoring poses typically
take into account geometric shape comple-
mentarity as well as the physicochemical
interactions between the small molecule and
the macromolecular target (Coupez and Lewis
2006; Sousa et al. 2006). The docking score
can be construed as a surrogate for the energy
Structural bioinformatics
Analytics
numeric and visual}
Experiment
kinetic parameters
thermodynamic parameter*
structural parameters
functional insight
molecular mode of action
biophysical interactions
screening
>
••MCnowledge
Figure 1. An overview of molecular modeling in computational toxicology. Abbreviations: QSAR, quantita-
tive structure-activity relationship; QSPR, quantitative structure-property relationship. After the identifica-
tion of a putative toxicant and target complexes (yellow sphere), the target structure (red spheres) is
either experimentally determined or modeled based on structures with known sequence identity.
Cheminformatics approaches and molecular docking (green spheres) can be used to obtain information
about the putative toxicant (overlap of red and green spheres) and predict the desired properties, such as
target-specific binding affinity and molecular modes of binding. Mathematical and visual analytics, such as
hierarchical clustered heat maps or target-specific linkage maps, can yield knowledge that is chemical-
class specific or target specific. Experimental guidance (blue arrow) optimizes this virtual screening
approach.
of interaction between the target and the
small molecule, and in some cases is provided
in terms of measures such as the log of the
dissociation constant for inhibitor binding
(Kj) or kilocalories per mole that may be
directly compared with binding experiments.
Comparison of these scores or computed
interaction energies for a library of chemicals
provides a means for ordering the molecules
by their capacity to interact with the macro-
molecular target. Chemicals with the best
scores are most likely to interact with the tar-
get and are selected as subjects for further
study. It is important to consider more than
just a single pose with the best score because
there are likely several local minimum energy
poses in the interaction profile and a variety
of highly ranked poses (Coupez and Lewis
2006; Sousa et al. 2006).
As is the case for the design of novel
pharmaceutical agents, the successful applica-
tion of docking methods to problems in
chemical toxicity depends on the identification
and availability of the crystal structures of the
macromolecular targets or similar proteins. A
variety of structures are available for macro-
molecular targets that are known to be linked
to the adverse effects of environmental chemi-
cals, and their number is continually increasing
(Hu et al. 2005; Wang et al. 2005). However,
simulations of the interaction between small
molecules and a macromolecular target for the
purposes of drug discovery versus toxicity
screening have distinct differences and, thus,
present distinct challenges: a) the focus on dif-
ferent (yet overlapping) regions of chemical
space; U) the strength of interaction between a
small molecule and macromolecular targets;
and c) the ultimate purpose of the virtual
screening results.
Figure 2 is an approximate depiction of the
chemical space for nonpharmaceutical com-
mercial chemicals versus druglike chemicals in
three selected dimensions of physicochemical
characteristics. Viable drug candidates are typi-
cally those that have a strong interaction with a
specific target, have good bioavailability, and
are readily metabolized to inactive compounds
and cleared from the system, in other words,
compounds that have specific absorption, dis-
tribution, metabolism, excretion, and toxicity
(ADMET) profiles and prescribed chemical
properties. In contrast, environmental chemi-
cals span a considerably larger chemical space
and tread into "undesirable" property space
from an ADMET perspective (too small, too
insoluble, too reactive, etc.). They can also
elicit adverse biologic effects from both strong
and weak interactions with targets and in both
a specific and nonspecific manner. Weak inter-
actions and nonspecificity are also important
aspects of pharmaceutical development because
some side effects might arise from unintended
binding to secondary targets (Ekins 2004;
574
VOLUME 116 I NUMBER 5 I May 2008 • Environmental Health Perspectives
Previous
TOC
-------
Molecular modeling for prioritizing bioassays
Ji et al. 2006). In addition, some environmen-
tal chemicals are produced and disposed of in
significantly larger quantities than are pharma-
ceuticals and, hence, may present inadvertent
human hazards over a long-term, low-dose
exposure scenario. This is particularly the case
if they are more chemically stable and persis-
tent (i.e., resistant to metabolism), are poten-
tially as bioavailable as drug candidates, or act
through common pathways (thus posing
cumulative effects) even if their individual tar-
get-specific interactions are much weaker than
drugs or endogenous chemicals. Hence, evalu-
ating the relative effectiveness of chemicals that
bind more weakly or to multiple targets less
specifically presents a greater challenge experi-
mentally and computationally than does the
discovery of novel pharmaceutical leads.
Scoring functions in molecular modeling
methods are typically optimized to identify
chemicals that bind best to the target.
Another significant difference between
pharmaceutical optimization and assessing the
chemical toxicity of environmental chemicals
is the purpose of an initial screen of a chemical
library. For the pharmaceutical industry, the
purpose of the initial screen for finding new
drug candidates is to limit the number of
chemicals that proceed to the next (more
expensive) phase of testing while increasing
the ratio of chemicals likely to become drugs
to those likely to be inactive (i.e., increasing
the "hit rate"). As long as the hit rate becomes
significantly improved by this process, the
Percent halogen
exclusion of some active chemicals is a reason-
able cost. In contrast, the purpose of an initial
screen of environmental chemicals is to maxi-
mize the chance that active chemicals advance
to the next phase of testing while eliminating
as many inactive chemicals as possible. Given
this objective and the corresponding uncer-
tainties in assessing "potency" or activity based
solely on computed scoring functions, the goal
is to discover all or almost all of the agents
having the potential to interact with the tar-
get, even those in significantly lower binding
affinity domains than the endogenous or puta-
tive cognate ligand for the receptor. Thus,
minimizing the number of false negatives is
critical when screening environmental chemi-
cals because the expectation is that positive
chemicals will be tested later in an experimen-
tal protocol. A toxicity screen should not reject
a compound (i.e., classify as inactive or safe)
that has a weak affinity for a target or multiple
targets without considering its ADMET prop-
erties, persistence, and chance of exposure.
Obtaining activity signatures from receptor
affinity profiles of compounds not intended
for therapeutic application may become an
important aspect of multilevel screening pro-
grams that include measured biologic proper-
ties, such as ToxCast (Dix et al. 2007).
Enrichment and False
Negatives
Figure 3 shows two hypothetical data scenar-
ios derived from computational docking
Log P(octanol/water
Topologic
polar surface area (A3)
Registered Pharmaceuticals
High-production-volume chemicals
Figure 2. Plot of environmental anthropogenic compounds and registered Pharmaceuticals subject to a
Lipinski druglike filter. The axes represent three physicochemical characteristics for each compound: total
polar surface area, partition coefficient (log P) between octanol and water, and fraction halogenated. The
environmental compounds are the high-production-volume chemicals (Wolf et al. 2006), and the registered
Pharmaceuticals are the FDAMDD [FDA (Food and Drug Administration) maximum (recommended) daily
dose] set from the DSSTox (Distributed Structure-Searchable Toxicity) database (Matthews et al. 2005).
experiments using the same library of chemi-
cals against a model target. The difference
between the two sets arises either from choos-
ing different docking score thresholds between
predicted active and inactive chemicals or
from using different scoring functions. For
this example, definitive (ideal) experimental
tests determine that 5% of the chemicals are
active and 95% are inactive relative to the
macromolecular target of interest. Scenario A
has 89% of the chemicals classified correctly,
whereas scenario B has only 55% of the chem-
icals classified correctly. The enrichment factor
for scenario A is 4 because 20% of the chemi-
cals selected for further testing (i.e., screened
positive) will prove to be positive, whereas the
enrichment factor for scenario B is only 2.
However, the type II error for scenario A is
0.6, whereas it is 0.0 for scenario B.
The screening method used for scenario A
appears to be better by many measures and is
an appropriate approach if the goal is to dis-
cover novel pharmaceutical leads. On the
other hand, the screening method used for
scenario B is more appropriate when screening
chemicals for potential toxicity. Scenario B
will carry many more chemicals to the next
phase of testing, but the negatives are true
negatives. Chemicals identified by this pre-
screen as negative will have a lower priority for
continued testing and perhaps will not be
tested in any other manner for effects at this
particular target.
This discussion addresses the challenges in
using current docking methods for assessing
chemical toxicity. The methods that are cur-
rently available for computational molecular
docking were developed for drug discovery
and therefore are optimized to screen large
chemical databases to find the most active
molecules and increase the enrichment factor.
Scenario B
55% predictive
accuracy
Experiment
Enrichment factor = 4 Enrichment factor = 2
Type II error = 0.6 Type II error = 0
Figures. Illustration of type II errors and enrich-
ment factors in chemical screening. The statistical
"type II error" is the ratio of the number of false
negatives to the sum of false negatives and true
positives. The "enrichment factor" is the ratio of
the true positive rate of the screen (the number of
true positives divided by the number of true posi-
tives plus false positives) to the ideal positive rate
of the chemical library (the number of positive
chemicals in the library divided by the number of
chemicals in the library).
Environmental Health Perspectives • VOLUME 1161 NUMBER 5 I May 2008
575
Previous
TOC
-------
Rabinowitz et al.
Some false negatives are not an important
concern as long as the enrichment rate is sig-
nificantly increased. In contrast, a screen for
assessing chemicals for potential toxicity often
deals with a smaller database of chemicals (the
chemicals encountered in the environment)
and must be capable of identifying chemicals
with much lower affinities than the natural
ligands. Therefore, scoring functions and/or
methods for delineating active chemicals from
inactive chemicals must be explored and bet-
ter understood in the context of environmen-
tal chemicals, and may involve computational
methods that are more accurate but computa-
tionally intensive.
Virtual Screening of Chemicals
The usual approach for virtual screening of
chemicals for toxicity is to screen a database of
chemicals for each chemical's capacity to inter-
act with a single macromolecular target and
initiate a single mode of chemical toxicity. A
virtual screen that is receptor specific pro-
duces a score vector where each element rep-
resents the interaction of that receptor with a
different chemical entity. Inverting the prob-
lem so that the vector now contains elements
that represent the capacity of a single chemi-
cal to interact with each of a series of targets
allows the most likely targets and, therefore,
the most likely modes of toxicity for a specific
chemical to be identified. A matrix is pro-
duced by interrogating a library of targets
with a database of chemicals. The relation-
ships among the elements in this matrix have
the potential to yield additional insights, such
as receptor cross-talk or multiple modes of
biologic potency (Macchiarulo et al. 2004),
and modes of sequestration (Perry et al.
2004). A combination of these computation-
ally derived data and experimentally derived
data can be data-mined to extract patterns and
associations. These associations can provide
additional knowledge for assessing the hazards
of chemicals and chemical mixtures or be used
to improve prediction tools in the context of
toxicity such as scoring functions for molecu-
lar docking calculations.
For some targets in the library, other inter-
actions in addition to those included in dock-
ing algorithms must be considered. For
instance, modes of toxicity have been identified
that require covalent interactions between the
toxicant and the target (Zhou et al. 2005) or
that necessitate the redistribution of charge in
both the toxicant and the target (Pardo et al.
1993). These interactions involve the elec-
tronic structure of both the putative toxicant
and target molecules and, hence, require some
level of quantum chemistry. However, most
current docking methods include only classical
interactions. One approach is to use molecular
docking to determine the structure of com-
plexes, and then to calculate the short-range
interactions with quantum chemistry methods.
A few attempts have been made in recent
years to build essential quantum effects
directly into molecular docking calculations,
such as quantum polarized ligand docking
(Choetal. 2005).
In addition, to take into account the
known or hypothesized biotransformation
products during the molecular docking calcu-
lations, each constituent must be included as
separate chemical entities in the docking cal-
culations. Computational tools already exist
for predicting metabolites (Jolivette and Ekins
2007), so docking calculations could be
improved by networking with metabolism
prediction models.
Another application for virtual screening is
predicting prospective targets for a particular
chemical and its metabolites using inverse
docking strategies. In drug discovery, this
approach can identify potential alternate uses
for drug candidates or predict side effects of
pharmaceuticals that might arise from unin-
tended interactions with other targets (i.e., off-
target effects), thus producing adverse
outcomes. As an element in a toxicity screen
for environmental chemicals, inverse docking
tools can be used to guide experimental test-
ing. Inverse docking can help focus efforts and
lead to a reduction in the use of resources as
well as the time required for a hazard or risk
assessment. Some attempts at inverse docking
methods have arisen in recent years (Ekins
2004; Ji et al. 2006), although these methods
still face some limitations that prevent their
more general use in virtual screenings. Inverse
docking strategies could become a more viable
resource as further target crystal structures
become available and molecular docking
methods are improved, and in conjunction
with systems biology methods such as pro-
teomics and genomics (Loging et al. 2007).
Conclusions
Computational molecular modeling methods
aid the risk assessment process by providing a
rational approach for some extrapolations in
the evaluation of chemical hazard. For
instance, when elements of a data set required
for evaluating the potential hazard of a chemi-
cal are unavailable and inferences can be made
based on interactions with putative targets,
molecular modeling can be used to simulate
the relevant missing information. Both ligand-
and structure-based molecular modeling
methods used in pharmaceutical discovery can
be adapted to provide this type of simulated
data. However, because of the greater diversity
of chemical space and binding affinity
domains being considered and the differences
in the strategic application of the results (the
need to minimize false negatives), these
molecular modeling strategies require addi-
tional considerations when assessing chemical
hazards. Molecular docking of potential envi-
ronmental chemicals to putative macromolecu-
lar targets for toxicity provides a measure of
their capacity to interact and hence is an aid in
the (pre)screening process for specific modes of
toxicity. These results provide a rationale for
developing further, more complete testing
strategies.
REFERENCES
Abagyan R, Totrov M. 2001. High-throughput docking for lead
generation. Curr Opin Chem Biol 5(41:375-382.
Carlson HA. 2002. Protein flexibility and drug design: how to hit
a moving target. Curr Opin Chem Biol 6(41:447-452.
Cho AE, Guallar V, Berne B, Friesner RA. 2005. Importance of
accurate charges in molecular docking: quantum mechani-
cal/molecular mechanical (QM/MM) approach. J Comput
Chem 26:915-931.
Coupez B, Lewis RA. 2006. Docking and scoring: theoretically
easy, practically impossible? Curr Med Chem 13:2995-3003.
Dix DJ, Houck KA, Martin MT, Richard AM, Setzer RW,
Kavlock RJ. 2007. The ToxCast program for prioritizing toxi-
city testing of environmental chemicals. Toxicol Sci
95(1):5-12.
Ekins S. 2004. Predicting undesirable drug interactions with
promiscuous proteins in silico. Drug Discov Today
9(61:276-285.
Halperin I, Ma B, Wolfson H, Nussinov R. 2002. Principles of
docking: an overview of search algorithms and a guide to
scoring functions. Proteins 47(41:409-443.
Hillisch A, Pineda LF, Hilgenfeld R. 2004. Utility of homology
models in the drug discovery process. Drug Discov Today
9(151:659-669.
Hu L, Benson ML, Smith RD, Lerner MG, Carlson HA. 2005. Binding
MOAD (mother of all databases). Proteins 60(31:333-340.
Huang B, Schroeder M. 2006. LIGSITECSC: predicting ligand bind-
ing sites using the Connolly surface and degree of conser-
vation. BMC Struct Biol 6:19; doi:10.1186/1472-6807-6-19
[Online 24 September 2006].
Ji ZL, Wang Y, Yu L, Han LY, Zheng CJ, Chen YZ. 2006. In silico
search of putative adverse drug reaction related proteins
as a potential tool for facilitating drug adverse effect pre-
diction. Toxicol Lett 164(21:104-112.
Jolivette LJ, Ekins S. 2007. Methods for predicting human drug
metabolism. Adv Clin Chem 43:131-176.
Jorgensen WL. 2004. The many roles of computation in drug
discovery. Science 303(56651:1813-1818.
Kuntz ID. 1992. Structure-based strategies for drug design and
discovery. Science 257(50731:1078-1082.
Loging W, Harland L.Williams-Jones B. 2007. High-throughput
electronic biology: mining information for drug discovery.
Nat Rev Drug Disc 6(31:220-230.
Macchiarulo A, Nobeli I, Thornton JM. 2004. Ligand selectivity
and competition between enzymes in silico. Nat
Biotechnol 22(81:1039-1045.
Matthews EJ, Kruhlak NL, Benz RD, Contrera JF, Rogers BA,
Wolf MA, et al. 2005. DSSTox FDA Maximum (Recom-
mended) Daily Dose Database (FDAMDD): SDF Files and
Documentation. Updated version FDAMDD_v2b_1217_
10Apr2006. Available: http://www.epa.gov/ncct/dsstox/
[accessed 1 September 2007].
Pardo L, Osman R, Weinstein H, Rabinowitz JR. 1993.
Mechanisms of nucleophilic addition to activated double
bonds: 1,2- and 1,4-Michael addition of ammonia. J Am
ChemSoc 115:8263-8269.
Perry JL, Goldsmith MR, Peterson MA, Beratan DN, WozniakG,
Rueker F, et al. 2004. Structure of the ochratoxin A binding
site within human serum albumin. J Phys Chem B
108(431:16960-16964.
Prathipati P, Dixit A, Saxena AK. 2007. Computer-aided drug
design: integration of structure-based and ligand-based
approaches in drug design. Curr Comp Aid Drug Des
3(21:133-148.
Sousa SF, Fernandes PA, Ramos MJ. 2006. Protein-ligand
docking: current status and future challenges. Proteins
Struct Funct Bioinform 65:15-26.
Tong W, Perkins R, Xing L, Welsh WJ, Sheehan DM. 1997.
QSAR models for binding of estrogenic compounds to
estrogen receptor alpha and beta subtypes. Endocrinology
138(91:4022-4025.
576
VOLUME 116 I NUMBER 5 I May 2008 • Environmental Health Perspectives
Previous
TOC
-------
Molecular modeling for prioritizing bioassays
U.S. EPA. 2003. A Framework for a Computational Toxicology
Research Program in ORD. EPA/600/R-03/065. Washington,
DC:U.S. Environmental Protection Agency.
Waller CL, Oprea Tl, Chae K, Park HK, Korach KS, Laws SC,
et al. 1996. Ligand-based identification of environmental
estrogens. Chem Res Toxicol 9(81:1240-1248.
Wang R, Fang X, Lu Y, Yang CY, Wang S. 2005. The PDBbind
database: methodologies and updates. J Med Chem
48(121:4111-4119.
Wolf MA, Burch J, Richard AM. 2006. DSSTox EPA High
Production Volume Challenge Program Structure-Index
File: SDF File and Documentation. Updated version
HPVCSI_v1b_3548_040ct2006. Available: http://www.
epa.gov/ncct/dsstox/[accessed 1 September 2007].
Zhou S, Chan E, Duan W, Huang M, Chen Y-Z. 2005. Drug bio-
activation, covalent binding to target proteins and toxicity
relevance. Drug Metab Rev 37(11:41-213.
Environmental Health Perspectives • VOLUME 1161 NUMBER 5 I May 2008
Previous
577
TOC
-------
TOXICOLOGICAL SCIENCES 103(1), 14-27 (2008)
doi: 10.1093/toxsci/kfm297
Advance Access publication December 7, 2007
REVIEW
Computational Toxicology—A State of the Science Mini Review
Robert J. Kavlock,*'1 Gerald Ankley,t Jerry Blancato,* Michael Breen,$ Rory Conolly,* David Dix,* Keith Houck,*
Elaine Hubal,* Richard Judson,* James Rabinowitz,* Ann Richard,* R. Woodrow Setzer,* Imran Shah,*
Daniel Villeneuve,t and Eric Weberj:
*National Center for Computational Toxicology; ^National Health and Environmental Effects Research Laboratory; and ^.National Exposure Research
Laboratory, Office of Research and Development, U.S. Environmental Protection Agency, Research Triangle Park, North Carolina
Received October 5, 2007; accepted December 5, 2007
Advances in computer sciences and hardware combined with
equally significant developments in molecular biology and
chemistry are providing toxicology with a powerful new tool
box. This tool box of computational models promises to increase
the efficiency and the effectiveness by which the hazards and risks
of environmental chemicals are determined. Computational toxi-
cology focuses on applying these tools across many scales, in-
cluding vastly increasing the numbers of chemicals and the types
of biological interactions that can be evaluated. In addition,
knowledge of toxicity pathways gathered within the tool box will
be directly applicable to the study of the biological responses
across a range of dose levels, including those more likely to be
representative of exposures to the human population. Progress in
this field will facilitate the transformative shift called for in the
recent report on toxicology in the 21st century by the National
Research Council. This review surveys the state of the art in many
areas of computational toxicology and points to several hurdles
that will be important to overcome as the field moves forward.
Proof-of-concept studies need to clearly demonstrate the addi-
tional predictive power gained from these tools. More researchers
need to become comfortable working with both the data gener-
ating tools and the computational modeling capabilities, and
regulatory authorities must show a willingness to the embrace new
approaches as they gain scientific acceptance. The next few years
should witness the early fruits of these efforts, but as the National
Research Council indicates, the paradigm shift will take a long
term investment and commitment to reach full potential.
1 To whom correspondence should be addressed at B-205-01, National
Center for Computational Toxicology, Office of Research and Development,
U.S. Environmental Protection Agency, Research Triangle Park, NC 27711.
Fax: 919-541-1194. E-mail: kavlock.robert@epa.gov.
This mini review is based on presentations and discussions at the
International Science Forum on Computational Toxicology that was
sponsored by the Office of Research and Development of the U.S.
Environmental Protection Agency and held in Research Triangle Park, NC
on May 21-23, 2007. The complete agenda and copies of the individual
presentations from the Forum are available on the Internet (www.epa.gov/ncct/
sciforum).
Published by Oxford University Press 2007.
Key Words: bioinformatics; biological modeling; QSAR;
systems biology; cheminformatics; high throughput screening;
toxicity pathways.
Computational toxicology is a growing research area that is
melding advances in molecular biology and chemistry with
modeling and computational science in order to increase the
predictive power of the field of toxicology. The U.S.
Environmental Protection Agency (U.S. EPA) defines compu-
tational toxicology as the "integration of modern computing
and information technology with molecular biology to improve
Agency prioritization of data requirements and risk assessment
of chemicals" (U.S. EPA, 2003). Success in this area would
translate to greater efficiency and effectiveness in determining
the hazards of the many environmental stressors that must be
dealt with, and deciding what types of information are most
needed to decrease uncertainties in the protection of human
health and the environment. Computational toxicology differs
from traditional toxicology in many aspects, but perhaps the
most important is that of scale. Scale in the numbers of
chemicals that are studied, breadth of endpoints and pathways
covered, levels of biological organization examined, range of
exposure conditions considered, and in the coverage of life
stages, genders, and species. It will take considerable progress
in all these areas to make toxicology a broadly predictive
science. Key advances leading the field include construction
and curation of large-scale data repositories necessary to
anchor the interpretation of information from new technolo-
gies; the introduction of virtual and laboratory-based high-
throughput assays on hundreds to thousands of chemicals per
day and high-content assays with hundreds to thousands of
biological endpoints per sample for the identification of
toxicity pathways; and the latest advances in computational
modeling that are providing the tools needed to integrate in-
formation across multiple levels of biological organization for
characterization of chemical hazard and risk to individuals and
Previous
-------
COMPUTATIONAL TOXICOLOGY
15
TABLE 1
Tasks Identified by the National Research Council (2007) in Each Main Topic Area that are Necessary to Transform Toxicity Testing
from the Current Animal-Model Based Approach to One that is more Reliant on In Vitro Systems to Detect and Characterize Toxicity
Pathways of Concern
Population-based and human-exposure data
Chemical characterization
Toxicity pathway characterization
Targeted testing
Dose-response and extrapolation modeling
Develop novel approaches to gather exposure data needed for making
hazard ID and risk assessment decisions.
Environmental chemicals would be first characterized for a number of properties
related to environmental distribution, exposure risk, physicochemical properties.
Toxicity pathways describe the key details of modes and mechanisms at a molecular level.
By characterizing these and developing relevant in vitro assays, one can make definitive
statements about the potential hazards posed by chemicals being tested.
In many cases, even when it is known what toxicity pathways are activated by a chemical,
it will be necessary to perform specialized or targeted tests, for instance to determine
dose-response relationships. The targeted testing phase may continue to use animal models.
Increasingly accurate and predictive computer models need to be developed to make use
of the information derived from the earlier phases and to aid in making regulator decisions.
populations. Collectively, these advances reflect the wave of
change that is sweeping and reinvigorating toxicology, just in
time to facilitate the vision of toxicology in the 21st century
that was recently released by the National Research Council
(NRC) of the National Academy of Science (National Research
Council, 2007). The NRC report's overall objective is to foster
a transformative paradigm shift in toxicology based largely on
the use of in vitro systems that will (1) provide broad coverage
of chemicals, chemical mixtures, outcomes, and life stages; (2)
reduce the cost and time of testing; (3) use fewer animals and
cause minimal suffering in the animals used; and (4) develop
a more robust scientific base for assessing health effects of
environmental agents. The report describes this effort as one
that will require the involvement of multiple organizations in
government, academia, industry, and the public. This mini
review describes advances that are now occurring in many of
the areas that are contributing to computational toxicology, and
is organized along the dimensions outlined by the National
Research Council (2007). The principle tasks outlined in the
NRC report are presented in Table 1, and each relevant aspect
of computational toxicology is discussed accordingly.
CHEMICAL CHARACTERIZATION
Chemical characterization involves the compilation of data
on physical and chemical properties, uses, environmental
surveillance, fate and transport, and properties that relate to the
potential for exposure, bioaccumulation, and toxicity (National
Research Council, 2007).
Predicting the Environmental Fate and Transport of
Chemical Contaminants
The ability to conduct chemical exposure and risk assess-
ments is dependent on tools and models capable of predicting
environmental concentrations. As the size (currently > 80,000
chemicals) and diversity of the regulated chemical universe
continues to increase, so does the need for more sophisticated
tools and models for calculating the physical-chemical prop-
erties necessary for predicting environmental fate and transport.
This need is further driven by the increasingly complex array of
exposure and risk assessments necessary to develop scientif-
ically defensible regulations. As this modeling capability in-
creases in complexity and scale, so must the data inputs. These
new predictive models will require huge arrays of input data,
and many of the required inputs are neither available nor easily
measured.
Currently, the Estimation Program Interface Suite (EPI
Suite) is the primary modeling system utilized within U.S. EPA
for providing estimates of the common physical-chemical
properties necessary for predicting chemical fate and transport
such as octanol/water partition coefficients, water solubility,
hydrolysis rate constants, and Henry's law constants (http://
www.epa.gov/oppt/exposure/pubs/episuite.htm). The EPI Suite
calculators are based primarily on a fragment constant ap-
proach that has been validated with an independent set of
chemicals. In general, the EPI Suite predicts physical-chemical
properties within an order of magnitude, which is normally
sufficient for screening level regulatory assessments.
The limitations of the EPI Suite calculators (e.g., inability to
calculate ionization constants (pKas) and transformation rates
constants beyond hydrolysis) require the use of other compu-
tational methods for meeting data needs. SPARC Performs
Automated Reasoning in Chemistry (SPARC) uses computa-
tional algorithms based on fundamental chemical structure
theory (i.e., a blending of linear free energy [LEER] to compute
thermodynamic properties and PMO theory to describe quan-
tum effects) to estimate numerous physical-chemical proper-
ties (Hilal et al., 2005; Whiteside et al., 2006). The power of
the tool box is its ability to couple whole molecule and site-
specific chemistry to calculate new properties. For example,
pKa and property models are coupled to calculate tautomeric
Previous
TOC
-------
16
KAVLOCK ET AL.
equilibrium constants; and pKa, hydrolysis, and property
models are coupled to calculate complex macro pKa's where
ionization, hydrolysis, and tautomerization may couple to yield
very complex apparent pKa's. This capability is essential for
calculating physical-chemical properties of organic chemicals
with complex chemical structures that contain multiple ioniz-
able functional moieties, such as many of the pharmaceuticals
that are being detected in the effluents of many waste water
treatment plants.
In addition to the more traditional computational approaches
such as the fragment constant approach and LFER, quantum
mechanical calculators coupled with aqueous solvation models
are also finding increasing applications in predicting physical-
chemical properties for predicting chemical reactivity (Lewis
et al., 2004) and for investigating reaction mechanisms for
transformation processes of interest such as reductive trans-
formations (Arnold et al., 2002).
Tools for predicting transformation kinetics and pathways
are quite limited, particularly with respect to biological pro-
cesses. The EPI Suite and SPARC calculators have limited
capability for the calculation of hydrolysis rate constants, and
currently have no ability to calculate biodegradation rate
constants. CATABOL is an expert system that begins to fill this
gap by predicting biotransformation pathways and calculating
probabilities of individual transformations (Jaworska et al.,
2002). The core of CATABOL is a degradation simulator,
which includes a library of hierarchically ordered individual
transformations (abiotic and enzymatic reactions). It also
provides the magnitude and chemical properties of the stable
daughter products resulting from biodegradation.
The future development of models for predicting the environ-
mental fate and transport of chemical contaminants is driven pri-
marily by the need for multimedia and multipathway assessments
over broad spatial and temporal scales. Geographic information
system-based technologies will be required for accessing,
retrieving, and processing data contained in a wide range of
national databases maintained by various government agencies.
Toxico-Cheminformatics
The term "Toxico-Cheminformatics" encompasses activities
designed to harness, systematize, and integrate the disparate
and largely textual information available on the toxicology and
biological activity of chemicals. These data exist in corporate
archives, published literature, public data compilations, and in
the files of U.S. government organizations such as the National
Toxicology Program (NTP), U.S. EPA, and the U.S. Food and
Drug Administration. Data mining approaches and predictive
toxicity models that can advance our ability to effectively screen
and prioritize large lists of chemicals are dependent upon the
ability to effectively access and employ such data resources.
The National Center for Biotechnology Information
(NCBI)'s PubChem project (http://pubchem.ncbi.nlm.nih.gov/)
is a large, public chemical data repository and open search/
retrieval system that links chemical structures to bioassay data.
PubChem has become an indispensable resource for chemists
and biologists due to its wide coverage of chemical space (> 10
million structures) and biological space (> 500 bioassays),
structure-searching and analysis tools, and linkages to the large
suite of NCBI databases (http://www.ncbi.nlm.nih.gov). Pub-
Chem includes data for the NCI 60 cell line panel, used by the
NCI Developmental Therapeutics Program to screen more than
100,000 compounds and natural products for anticancer
activity and providing a rich data resource for a comprehen-
sively characterized set of cells. Weinstein (2006) has incor-
porated these data into a fully relational, public resource titled
"CellMiner," and coined the term "integromics" to convey the
highly flexible functionality of this system for chemical/
biological profiling, spanning genomics, high-throughput
screening (HTS), and chemical information domains. Contrib-
uting to efforts in data standardization and access, U.S. EPA is
creating a large relational data warehouse for chemical and
toxicity data from various public resources. This Aggregated
Computational Toxicology Resource is designed to support
flexible data mining and modeling efforts across a wide range
of biological information domains and the new U.S. EPA
ToxCast program (Dix et al, 2007).
With HTS approaches being increasingly applied to tox-
icology data sets, such as represented by the NTP High-
Throughput Testing Program (National Toxicology Program
High-Throughput Screening Program, 2006), come challenges
to determine the most effective means for employing such data
to improve toxicity prediction models. Anchoring large ma-
trices of HTS activity data to relatively sparse phenotypic
endpoint data across chemical compound space presents a fun-
damental challenge. Yang (2007) has demonstrated the value
of linking bioassays with toxicity endpoints via the structural
feature dimension, rather than the compound level, generating
matrices to determine correlation of bioassays with toxicity.
This paradigm addresses the practical problem of the sparse
data space and allows quantitative multivariate analysis.
These toxico-cheminformatics tools and public resources are
evolving in tandem with increasing legislative pressures within
the United States, Europe, and Canada to prioritize large lists of
existing chemicals for testing and/or assessment. Health
Canada has been the first to fully implement a tiered Hazard
ID and Exposure Assessment evaluation process relying upon
weight-of-evidence consideration of existing data and results
of toxicity prediction models, and structure-analog inferences
(Health Canada, 2007). The approach is pragmatic and
transparent, relying upon existing capabilities and technolo-
gies, and was successfully employed to prioritize the Domestic
Substance List inventory of 23,000 chemicals by the
legislatively mandated deadline under the Canadian Environ-
mental Protection Act of September 2006. This approach will
greatly benefit from advances in toxico-cheminformatics, and
will influence other governmental agencies as they straggle
with similar mandates for prioritizing large lists of chemicals.
Previous
TOC
-------
COMPUTATIONAL TOXICOLOGY
17
Molecular Modeling Methods as a Virtual Screening Tool for
the Assessment of Chemical Toxicity
Molecular modeling methods provide an approach for
estimating chemical activity when the relevant data is not
available. When used in this way it becomes an important tool
for screening chemicals for toxicity and hazard identification.
Computational molecular methods may also be applied to
model toxicity pathways when some of the relevant experi-
mental data are unavailable. As noted above, some of these
methods have been used to estimate various physical and
chemical properties of the molecules relevant to environmental
fate and transport. Other molecular modeling methods may be
applied to simulate critical processes in specific mechanisms of
action involved in toxicity. An initial and often differential step
in many of these mechanisms of action requires the interaction
of the molecular environmental contaminant, or one of its
descendants, with a (macro)molecular target. An element of
a virtual screen for potential toxicity may be developed from the
characterization of these toxicant-target interactions. One large
and important subset of target-toxicant interactions is the
interaction of chemicals with proteins. Many computational
approaches for screening libraries of molecules for pharmaceu-
tical application have been developed. These methods also may
be applied to screen environmental chemicals for toxicity, but the
differing requirements of these two similar problems must be
considered. For example, screening of environmental chemicals
requires minimizing false negatives, whereas drug discovery only
requires the identification of some of the most potent chemicals,
which can yield a significant number of false negatives.
Molecular modeling methods that incorporate both the
structure of the protein target and/or that of known ligands
have been used to investigate nuclear receptor and cytochrome
P450 targets, In addition to the ligand binding site, features on
the protein surface, such as the Activation Function 2 site or
other coactivator and corepressor regions of the Human
Pregnane X Receptor, are potential sites for interference by
environmental chemicals (Wang et al., 2007). Methods that
map the binding of functional groups from chemicals to protein
surfaces and binding sites have been developed (Kaya et al.,
2006; Sheu et al., 2005). These maps of the favorable positions
of molecular substructures provide fragment libraries to
which chemicals may be fitted and their suitability for binding
evaluated. Current studies have demonstrated the importance of
the motion of the target for ligand binding, protein function,
and subunit assembly. Local motion of the amino acids in the
binding site provides the flexibility to allow the potential ligand
to sculpt the ligand binding domain. Concepts that incorporate
protein flexibility to identify binding modes of lexicological
interest are being developed (Lill et al., 2006; Vedani et al.,
2006). This technology combines structure-based molecular
docking with multidimensional quantitative structure activity
relationships. Global modes of protein motion have been found
to influence protein function by affecting binding and subunit
assembly (Wang et al., 2007). Metabolizing enzymes present
potential targets for clearance of chemicals as well as activation
that could result in toxicity. Understanding the relationship
between structure and function for P450 serves to illuminate both
of these issues that are relevant for assessing the effects of chem-
icals. Pharmacophores and quantitative structure activity relation-
ships have been developed for the various CYPs (Jolivette and
Ekins, 2007), and machine learning methods have been developed
to predict metabolic routes (Ekins et al., 2006). These approaches
will allow relatively rapid and comprehensive coverage of the
interaction of chemicals with multiple macromolecules, thus
complementing results from HTS assays (see below).
TOXICITY PATHWAYS
Toxicity pathways represent the normal cellular responses
that are expected to result in adverse health effects when
sufficiently perturbed by chemical exposure (National Research
Council, 2007). A wide variety of in vitro and in vivo tools are
being developed to identify critical toxicity pathways.
Application of Drug Discovery Technologies in
Environmental Chemical Prioritization
Strategies for investigating the toxicity of environmental
chemicals have changed little over many years and continue to
heavily rely on animal testing. However, recent advances in
molecular biology, genomics, bioinformatics, systems biology,
and computational toxicology have led to the application of
innovative methods toward more informative in vitro
approaches. The application of quantitative, HTS assays is
a key method. Originally developed for use in drug discovery
by the pharmaceutical industry, these assays quantify molecular
target-, signaling pathway-, and cellular phenotype-focused
endpoints with capacity to evaluate thousands of chemicals in
concentration-response format. As an example, National
Institutes of Health (NIH) Chemical Genomics Center has
built an infrastructure for robust, quantitative, HTS assays
(Inglese et al., 2006) that is currently being used to screen
thousands of environmental chemicals for a variety of
toxicology-related endpoints. This project utilizes data provided
by the NTP's HTS Initiative (http://ntp.niehs.nih.gov/index.
cfm?objectid=05F80E15-FlF6-975E-77DDEDBDF3B941CD)
and U.S. EPA's ToxCast Program (Dix et al., 2007).
HTS using cellular assays offers perhaps the greatest hope
for transformation of the current toxicity testing paradigm.
Such systems incorporate comprehensive, functioning, cellular
signaling pathways, the disturbance of which by environmental
chemicals would suggest a potential for toxicity. Development
of high-content screening (HCS) platforms consisting of
automated, fluorescence microscope imaging instruments and
image analysis algorithms greatly facilitated quantitation of
chemical perturbations of cell signaling pathways and vital
organelle function on a single cell basis. As an illustration of
Previous
TOC
-------
18
KAVLOCK ET AL.
the utility of this approach, human liver toxicants with a variety
of mechanisms of action were detected with both good sen-
sitivity and specificity through screening multiple endpoints
such as nuclear area and cell proliferation in a human liver cell
line (O'Brien et al., 2006). This approach also is useful in
examining effects of new classes of chemicals (e.g., nano-
materials) for potential toxicity by reporting effects on toxicity-
associated endpoints and allowing visual appreciation for
novel, and perhaps unexpected, effects on cellular morphology
and function (Ding et al., 2005). With an eye toward repro-
ducing normal physiology in vitro to the greatest extent
possible, Berg et al. (2006) established coculture systems of
primary human cells and developed assays that measure many
endpoints encompassing a wide variety of signaling pathways.
Screening of pharmacological probes in these assays demon-
strated similar behavior of chemicals related by mechanism
of action, thus providing a system potentially useful for
understanding mechanisms of toxicity. Although HCS was not
used in this application, the marriage of complex, primary
human cell cultures with HCS analysis is a likely, and highly
valuable, development in the field of toxicity screening. HTS
approaches do have imposing hurdles to overcome, however,
including volatile or aqueous insoluble environmental chem-
icals, need for inclusion of biotransformation capacity in the
in vitro test systems, the myriad of potential toxicity pathways
that must be covered, the likelihood of cell-type dependent
activity, and the probability of dependence of some mecha-
nisms of toxicity on higher level interactions not found in cell
culture systems (Houck and Kavlock, 2007).
The HTS and HCS methods described are all data-intensive
and require computational approaches to analyze and properly
interpret. The high dimensionality of the data may require novel
statistical approaches. Results are likely to be used in building
models that predict the potential for toxicity for new chemicals
based on their behavior in in vitro assays, In addition, screening
results integrated into systems biology models should lead to
insights into mechanisms of action that will be invaluable for
risk assessment. Validation and harmonization of protocols at
the international level should result in a much more efficient and
comprehensive safety net for hazardous chemical protection,
and greatly reduce the number of laboratory animals needed to
accomplish this (Hartung, 2006).
Using Genomics to Predict Potential Toxicity
Transcriptomics is a useful approach for understanding the
interactions of chemicals with biological targets, and can
complement the HTS assays used for bioactivity profiling.
Using bioactivity profiles to accurately predict toxicity and
prioritize chemicals for further testing would allow for the
focusing of resources on greater potential hazards or risks.
Prioritization efforts to which genomics data might contribute
include U.S. EPA's voluntary high production volume (HPV)
program, wherein chemicals manufactured in large amounts are
identified and hazard characterized according to chemical
category. Genomics is being developed as part of a suite of
tools to help confirm the category groupings of HPV chemicals,
and identify which chemicals or chemical categories may
present greater hazard or risk. The U.S. EPA is actively
developing the methods, policies, and infrastructure for using
genomics data in such a regulatory context (Dix et al., 2006).
In vitro toxicogenomics methods are being developed and
evaluated for toxicity prediction and for addressing fundamen-
tal questions about the ability to identify toxicity pathways for
large numbers of chemicals in a number of research programs
in the United States, Europe, and Asia. The throughput,
molecular specificity, and applicability of this approach to
human cell systems are highly consistent with the goals and
directions described in the NRC report on the future of toxicity
testing (National Research Council, 2007).
Genomic signatures predictive of toxicological outcomes
have been derived from in vivo studies, and the evaluation and
application of these signatures to hazard identification and
risk assessment is an area of active research. Perhaps most
significantly, genomic signatures predicting tumor incidence in
2-year rodent cancer bioassays have the potential to provide
shorter-term tests as an alternative to the expensive two-year
rodent bioassay. The ability to predict chemically induced
increases in lung tumor incidence based on gene expression
biomarkers has been demonstrated in microarray studies
performed on mice exposed for 90 days to chemicals that
were previously tested by the National Toxicology Program
(Thomas et al., 2007). In an even shorter 5-day study design,
liver gene expression data from rats treated with structurally
and mechanistically diverse chemicals was used to derive
a genomic signature that predicted nongenotoxic liver tumor-
igenicity in the 2-year bioassay (Fielden et al., 2007). In both
of these studies, sensitivity and specificity of the genomic
signatures was high, and the signatures provided accurate
predictions and identified plausible modes of action. Both the
Thomas et al. and the Fielden et al. data sets are being utilized
in the Microarray Quality Control assessment of best practices
in developing and validating predictive genomic signatures (http://
www.fda.gov/nctr/science/centers/toxicoinformatics/maqc/).
Success in developing predictive genomic signatures from
in vitro studies has been more modest, to date, than what has
been accomplished using in vivo data. Gene expression profiles
for more than 100 reference compounds in isolated rat
hepatocytes have been used to derive predictive signatures
identifying potential mitochondria! damage, phospholipidosis,
microvesicular steatosis, and peroxisome proliferation, with
a high degree of sensitivity and specificity (Yang et al., 2006).
A large European Union program project entitled carcinoGE-
NOMICS (http://www.carcinogenomics.eu/) was initiated in
2006 to develop genomics-based in vitro screens predictive of
genotoxicity and carcinogenicity in the liver, kidneys, and
lungs. In vitro toxicogenomics is also part of U.S. EPA's
ToxCast research program, which is being designed to forecast
Previous
TOC
-------
COMPUTATIONAL TOXICOLOGY
19
toxicity based on genomic and HTS bioactivity profiles (Dix
et al., 2007; http://www.epa.gov/comptox/toxcast/). The initial
goal of these in vitro toxicogenomic efforts is hazard prediction
and chemical prioritization for subsequent in vivo testing, but
the ultimate goal goes beyond refinement to actually replacing
in vivo testing. This will require a sustained, systematic, and
substantial effort on the part of government, academic,
industry, and nongovernmental organization partners.
Signaling as a Determinant for Systems Behavior
Understanding processes at the molecular, cellular, and
tissue levels is an ongoing challenge in toxicology. Central to
this hierarchy of biological complexity is the field of signal
transduction, which deals with the biochemical mechanisms
and pathways by which cells respond to external stimuli.
Computational systems approaches are critical for mechanistic
modeling of environmental chemicals to predict adverse out-
comes in humans at low doses.
For decades, computational modeling has complemented
laboratory-based biology with in silico experiments to generate
and test mechanistic hypotheses. Computational approaches
have been used to model biological networks as dynamical
systems in which the quantitative variation of molecular en-
tities are elucidated by the solution of differential equations
(Aldridge et al., 2006). Such models of signaling networks
have been used to predict the dynamic response at molecular
(Behar et al., 2007), cellular (Sasagawa et al., 2005), and tissue
levels (Schneider and Haugh, 2006). Postgenomic, large-scale
biological assays present new challenges and opportunities for
modeling signaling networks. Though large-scale data provide
a global view of a biological system, they remain difficult
to utilize directly in traditional dynamic models. This has
stimulated research on alternative formalisms for modeling
pathways (Faure et al., 2006). In addition, concurrent measure-
ments on thousands of proteins, genes, and metabolites in
response to stimuli, or in different disease states, enable the
"reverse-engineering" of biological networks from data using
empirical methods (D'haeseleer et al., 2000).
Synthesizing disparate information into coherent mechanis-
tic hypotheses is an important challenge for modeling toxicity
pathways. Knowledge-based approaches (Karp, 2001) provide
an avenue for efficiently managing the magnitude and com-
plexity of such information. Through such techniques, large-
scale biological interaction data can be algorithmically searched
to infer signaling pathways (Scott et al., 2006), to extrapolate
between species, or to signify mechanistic gaps. Some of these
gaps may be filled by literature mining (Krallinger et al., 2005)
and others will require additional experiments. Moreover,
intelligent computational techniques will aid in designing such
experiments by using biological knowledge to infer testable
hypotheses about novel mechanisms (Nguyen and Ho, 2006).
Computational predictive modeling of cellular signaling
systems will aid risk assessment in two important ways. First,
knowledge-based and data-driven approaches will aid in orga-
nizing and refining biological insight on perturbations leading
to adverse outcomes. Second, dynamic simulation of these
mechanisms will help in predicting dose-dependent response.
This will reduce the scope of animal testing and the time
required for understanding the risk of toxic effects due to
environmental chemicals.
Systems Biology Models of the HPG Axis
Over the past decade, there has been a focused international
effort to identify possible adverse effects of endocrine dis-
rupting chemicals (EDCs) on humans and wildlife. Scientists
have identified alterations in the concentration dynamics of
specific hormones as risk factors for common cancers such as
breast cancer (estrogen, progesterone), endometrial cancer
(estrogen), and prostate cancer (estrogen, testosterone) in
humans (Portier, 2002). Chemicals capable of acting as EDCs
include pesticides, pharmaceuticals, and industrial chemicals.
Ecological exposures to EDCs are primarily from industrial
and waste water treatment effluents, whereas human exposures
are mainly through the food chain. There is convincing evi-
dence that fish are being affected by EDCs both at the
individual and population levels.
As many of the adverse effects have been related to
alterations in the function of the hypothalamus-pituitary-
gonadal (HPG) axis, the development of computational system
biology models that describe the biological perturbations at the
biochemical level and integrate information toward higher
levels of biological organization will be useful in predicting
dose-response behaviors at the whole organism and population
levels. For example, a mechanistic computational model of the
intraovarian metabolic network has been developed to predict
the synthesis and secretion of testosterone and estradiol and
their responses to the EDC, fadrozole (Breen et al., 2007).
Physiologically based pharmacokinetic (PBPK) models cou-
pled with pharmacodynamic models that include the regulatory
feedback of the HPG axis also can be used to predict the
biological response to EDCs in whole organisms (Plowchalk
and Teeguarden, 2002; Watanabe et al., 2006). In addition,
these computational models can be developed for fish and other
wildlife. They can be used to identify biomarkers of exposure
to EDCs that are indicative of the ecologically relevant effects
at the individual and population levels in support of predictive
environmental risk assessments (Rose et al., 2003).
Because the mechanism of action of EDCs is generally
understood, there has been a considerable emphasis on the
development of screening tools for use in hazard identification,
and the involvement of feedback loops in physiological re-
gulation of hormone function has provided a foundation upon
which to build computational models of the relevant biology.
Hence, EDCs represent a prime example of how toxicity
pathway elucidation and characterization can be applied to
hazard and risk assessment as envisioned by the National
Previous
TOC
-------
20
KAVLOCK ET AL.
Research Council (2007). Of course, additional research is
needed in this area to bring a higher level of involvement of
cell based screening assays, especially those which incorporate
human cells or receptors, and to employ the computational
models of response.
DOSE-RESPONSE AND EXTRAPOLATION MODELS
Dose-response is the combination of the relationship between
exposure and a relevant measure of internal dose (pharmaco-
kinetics), and the relationship between internal dose and the
toxic effect (pharmacodynamics). They are intended to reliably
predict the consequences of exposure at other dose levels and
life stages, in other species, or in susceptible individuals.
Dose—Response and Uncertainty
Risk analysis for environmental exposures involves expo-
sure assessment (factoring in various routes such as drinking
water, food, air, and skin exposure) and the effects of those
exposures on individuals (dose-response assessment), In
modern exposure assessments, exposure may well be charac-
terized by a distribution of possible exposure levels over a
population, with confidence intervals on the quantiles of that
distribution (e.g., specifying the 99th percentile of the exposure
distribution and its 95 percent confidence bounds), and
a sophisticated analysis of the components of variability and
uncertainty (e.g., Cullen and Frey, 1999; U.S. EPA, 1997). In
contrast, standard approaches to dose-response analysis treat
the uncertainties surrounding dose-response metrics simplisti-
cally, using standard factors to extrapolate across species and
to quantify variability among exposed people. Probabilistic
dose-response assessment methods allow a more complete
characterization of uncertainty and variability in dose-response
analysis (Evans et al, 2001; Hattis et al, 2002; Slob and
Pieters, 1998), and are naturally compatible with probabilistic
exposure assessments (van der Voet and Slob, 2007). Dose-
response analysis is divided into the analysis of the delivery of
toxic substances to target tissues (pharmacokinetics), and the
action of toxic substances at their targets (pharmacodynamics).
Much progress has been made in understanding pharmaco-
kinetics and in building models (PBPK models) that quantify
that understanding. Such models may be used to quantify the
relationship between potency in animals and humans, human
variability for internal dose, and the overall uncertainty of such
predictions (Barton et al., 2007). Hierarchical Bayesian
techniques are useful for characterizing the uncertainty of
model outputs (Hack et al., 2006). Monte-Carlo methods allow
uncertainty and variability in model parameters to be translated
into distributions of internal doses in a human population with
attendant uncertainty (Allen et al., 1996; Clewell et al., 1999).
Ideally, pharmacodynamic relationships also would be
modeled based on mechanistic understanding (Setzer et al.,
2001). In practice, however, dose-response evaluations are
based on empirical dose-response modeling of animal
toxicology data. Typically, many empirical curves may fit a
given dataset, reflecting real uncertainty about the "true" dose-
response relationship. Wheeler and Bailer (2007) have de-
veloped a method using model averaging that approximates the
uncertainty in our understanding of a given dose-response
relationship.
Uncertainty in a risk assessment may be reduced by the
collection of further information, and sensitivity analysis
(Saltelli et al., 2000) can help to quantify the contribution of
individual sources of uncertainty and their interactions to that
of the overall risk analysis. Frey and Patil (2002) and Mokhtari
et al. (2006) have compared the utility of different sensitivity
analysis methods in a probabilistic risk assessment. Mokhtari
and Frey (2005) have recommended how sensitivity analysis
can be used and applied to aid in addressing risk management
and research planning questions. These approaches provide
considerable information to the risk manager for making
decisions about the exposure levels needed to protect target
populations.
Genetic Variation, Gene-Environment Interactions, and
Environmental Risk Assessment
Understanding relationships between environmental expo-
sures and complex disease requires consideration of multiple
factors, both extrinsic (e.g., chemical exposure) and intrinsic
(e.g., genetic variation). This information must be integrated to
evaluate gene-environment interactions to identify vulnerable
populations and characterize life-stage risks. Although the
association between genetic and environmental factors in de-
velopment of disease has long been recognized, tools for large-
scale characterization of human genetic variation have only
recently become available (The International HapMap Con-
sortium, 2005).
It is well known that different species, and individuals within
species, react differently to identical exposures to pharmaceut-
icals or environmental chemicals. This is, in part, driven by
genetic variation in multiple pathways affecting multiple pro-
cesses such as adsorption, metabolism and signaling. Recent
advances in our understanding of the pattern of human
molecular genetic variation have opened the door to genome-
wide genetic variation studies (Gibbs and Singleton, 2006).
Pharmacogenetics is a well-developed field studying the
interaction between human genetic variation and differential
response to pharmaceutical compounds (Wilke, 2007). Many
of the insights developed in these studies have direct relevance
to environmental chemicals. Pharmacogenetic studies increas-
ingly analyze both pharmacokinetics and pharmacodynamics
pathways. Emphasis is shifting from a focus on individual
markers, such as single-nucleotide polymorphisms (SNPs), to
multi-SNP and multigene haplotypes.
Gene-drug interaction studies have provided many insights
for understanding the effects of chemical exposure in
Previous
TOC
-------
COMPUTATIONAL TOXICOLOGY
21
genetically heterogeneous populations. For example, inves-
tigators in the NIH Pharmacogenetics Research Network are
examining multiple approaches to correlate drug response with
genetic variation. Data from this program is stored and
annotated in a publicly accessible knowledge base (Giacomini
et al., 2007). Lessons learned from these and related studies are
being incorporated into drug development and governmental
regulation, and are models for approaches to identify vul-
nerable populations in the context of environmental exposure.
Although genetic variation plays a major role in gene-
environment interactions, recent work has shown that epige-
netic effects also are important. This complicates the picture
because the effects of exposure can lead to multigenerational
effects even in the absence of genetic mutations. Epidemio-
logical evidence increasingly suggests that environmental ex-
posures early in development have a role in susceptibility to
disease in later life, and that some of these effects are passed on
through a second generation. Epigenetic modifications provide
a plausible link between the environment and alterations in
gene expression that might lead to disease phenotypes. For
example, a potential mechanism underpinning early life pro-
gramming is that of exposure to excess stress steroid hormones
(glucocorticoids) in early life. It has recently been shown that
the programming effects of glucocorticoids can be transmitted
to a second generation. This information provides a basis for
understanding the inherited association between low birth
weight and cardiovascular disease risk later in life (Drake et al.,
2005).
It is becoming increasingly clear that specific genetic
variants modulate individual vulnerability to many diseases.
A major challenge for future toxicogenomics research is to link
exposure, internal dose, genetic variation, disease, and gene-
chemical interactions (Schwartz and Collins, 2007). This effort
should yield improved dosimetry models that will reduce
uncertainties associated with the assumption that populations
are homogeneous in their response to toxic chemicals. Ex-
posure information on par with available toxicogenomic infor-
mation will improve our ability to identify vulnerable
populations, classify exposure in studies of complex disease,
and elucidate important gene-environment interactions.
The study of genetic variation intersects with several issues
discussed in the NRC report. At one end, genetic variation
provides a handle for investigating mechanism of action of
chemicals and for elucidating toxicity pathways. Gene
knockout strains in many species provide a standard tool for
delineating pathways (Wijnhoven et al., 2007), but less severe
changes in the form of genetic polymorphisms are also useful
and potentially more relevant to the understanding human
health effects. By testing a chemical in a panel of animals
with polymorphic, but well-characterized genetic backgrounds
(Roberts et al., 2007), one can generate valuable information
on what pathways are being modulated by the chemical
(Ginsburg, 2005). At the other end of the spectrum, it is
possible in some cases to understand in detail how genetic
differences alter dose-response relationships, and from there to
develop specific risk assessment recommendations which take
into account genetic variation in human populations. The
primary examples of this approach to risk assessment involve
chemical metabolism (Dome, 2007), which is also the most
well studied area in the field of pharmacogenetics. In summary,
there is an ever growing body of knowledge about the effects
and uses of genetic variation in many species, and the field of
predictive computational toxicology will be able to increas-
ingly benefit from these advances.
Computational Tools for Ecological Risk Assessment
Ecological systems pose some unique challenges for quan-
titative risk assessment. Human health risk assessment requires
extrapolation from effects in well-characterized animal models
to well-studied human biology, with the aim of protecting
individuals, In contrast, ecological risk assessment requires
extrapolation among widely divergent taxonomic groups of
relatively understudied organisms, with the intent of protecting
populations and critical functional processes within ecological
communities.
Modern computational capabilities and tools for conducting
high-content biological analyses (e.g., transcriptomics, proteo-
mics, and metabolomics) have the potential to significantly
enhance our ability to predict or evaluate ecological risks. For
example, high-content assays that provide multivariate results
can be used to quantitatively classify individual organisms
(sentinels) or communities of organisms (e.g., microbial
communities) as within or deviated from a normal operating
range (Kersting, 1984; van Straalen and Roelofs, 2006). As a
key advantage, these general profiling and multivariate con-
cepts can be applied to species that lack a well-characterized
genome (van Straalen and Roelofs, 2006). Beyond profiling
approaches, high-content biological analyses provide powerful
tools for examining system-wide responses to stressors.
Through iterations of system-oriented hypothesis generation,
testing, and gradual refinement of biologically based models, it
should be feasible to establish a credible scientific foundation
for predicting adverse effects based on chemical mode of
action and/or extrapolating effects among species with well
conserved biological pathways (Villeneuve et al., 2007).
However, even with the ability to conduct high-content
analyses, high quality data sets for parameterizing computa-
tional models, particularly those that bridge from effects on
individual model animals to predicted effects on wildlife
populations, are likely to remain rare (e.g., Bennett and
Etterson, 2007). Consequently, strategies for making the best
possible use of laboratory toxicity data to forecast/project
population-level risks will remain critical (Bennett and
Etterson, 2007). Additionally, alternative computational ap-
proaches will have an important role to play. For example,
computational methods that examine overall network topology
may be used as a way to deduce system function, control
Previous
TOC
-------
22
KAVLOCK ET AL.
properties, and robustness of biological networks to stressors.
Such approaches can be applied at many scales of biological
organization, from gene regulatory networks within a single
cell to trophic interactions and food webs at the ecosystem
level (Proulx et al., 2005). Similarly, there is an increasingly
important role for models, simulation, and landscape level
spatial forecasting related to the overlapping impacts of mul-
tiple stressors (e.g., chemicals, climate change, habitat loss,
exotic species). There are many examples of creative uses of
geographic information systems and remote sensing technol-
ogies for this purpose (e.g., Haltuch et al., 2000; Kehler and
Rahel, 1996; Kooistra et al., 2001; Leuven and Poudevigne,
2002, McCormick, 1999; Tong, 2001). Thus, although the
challenge of ecological risk assessment and balancing
environmental protection against the demands of human
commerce and activities remains daunting, ecotoxicologists,
"stress ecologists" (van Straalen, 2003), and risk assess-
ment professionals have increasingly powerful tools at their
disposal.
Virtual Tissues—The Next Big Step for Computational
Biology
To date, biologically motivated computational modeling in
toxicology has consisted largely of dosimetry models (PBPK
and respiratory tract airway models) and, to a lesser extent,
biologically based dose-response models that combine dosim-
etry with descriptions of one or more modes of action (Clewell
et al., 2005; Conolly et al., 2004). PBPK models are usually
highly lumped and contain little spatial information. Early
models of the lung were one-dimensional, though more re-
cently, three-dimensional descriptions of both the nasal and
pulmonary airways have been developed (Kimbell et al., 2001;
Timchalk et al., 2001). Thus, for the most part, current
biologically motivated modeling in toxicology involves
significant abstraction of biological structure.
Ongoing developments in high-throughput technologies,
systems biology, and computer hardware and software are
creating the opportunity for "multiscale" modeling of bi-
ological systems (Hunter et al., 2006; Kitano, 2002). These
models incorporate structural and functional information at
multiple scales of biological organization. For example,
Bottino et al. (2006) studied cardiac effects of drugs using
a hierarchical set of models extending from ion channels to
cells to the tissue level. They showed how such models can be
developed for multiple species and how in silico experiments
can be conducted where drugs are used to perturb the cardiac
system. An additional important aspect of this kind of
modeling is that one can superimpose certain risk factors,
such as hypokalemia and ischemia, in order to make clinical
predictions prior to the actual use of the drug in the clinic. A
conceptually similar approach is being taken in the HepatoSys
project (HepatoSys, 2007), where a suite of models describing
various aspects of the functional biology of hepatocytes is
under development. The overall aim of the HepatoSys project
is to arrive at a holistic understanding of hepatocyte biology
and to be able to present and make these processes accessible
in silico.
A "virtual liver" is being developed at U.S. EPA's National
Center for Computational Toxicology. The overall goal of this
project is to develop a multiscale, computational model of the
liver that incorporates anatomical and biochemical information
relevant to toxicological mechanisms and responses. As model
development progresses, integration of within-cell descriptions
and cell-to-cell communication will evolve into a computational
description of the liver. The approach will be to first describe
normal biological processes, such as energy and oxygen me-
tabolism, and then describe how perturbations of these
processes by chemicals lead to toxic effects. In the longer
run, the project also will provide an opportunity to develop
descriptions of diseases, such as diabetes, and to examine how
such diseases influence susceptibility to environmental stressors.
Virtual tissues are being developed not only in the context of
computational toxicology, but also in clinical and translational
research. Thus, there is an increasing emphasis on systematic
integration of scientific data, visualization, and transparent
computing that creates easily accessible and customizable
workflows for users. This integration of basic research and
clinical data has created the demand for more streamlined tools
and necessary resources for on demand investigation and
modeling of pressing biological problems, and subsequent
validation of in silico predictions through further clinical and
environmental observations, In response to this need, the
National Biomedical Computation Resource (NBCR; http://
nbcr.sdsc.edu/) and their collaborators are developing tools
such as Continuity, which describes molecular interactions,
diffusion, and electrostatics in the human heart. Continuity is
capable of transparently accessing remote computational
resources from an end user's desktop environment. Develop-
ment of middleware at the NCBR, such as the Opal toolkit,
makes such transparent access possible.
The potential payoffs from development of virtual tissues in
toxicology are significant. Virtual tissues will build on current
successes with PBPK modeling and take the development of
quantitative descriptions of biological mechanisms to a new
level of complexity. Virtual tissues will have much greater
capabilities than PBPK models for providing insights into
dose-response and time course behaviors, and will promote
inclusion of larger amounts of integrated biological data into
risk assessment.
With adequate development, virtual tissues will also become
capable of providing capabilities necessary for a full imple-
mentation the National Research Council (2007) report.
Development of in vitro assays of toxicity pathways will
require validation studies that can at present only be conducted
in vivo, In the future, sufficiently mature virtual tissues will
provide an in silico alternative for at least some aspects of
in vivo testing. The continuing and probably increasing
Previous
TOC
-------
COMPUTATIONAL TOXICOLOGY
23
pressure to reduce animal use for toxicity testing will only
encourage this trend.
Finally, it must be noted that success in development of
virtual tissues will depend not only on coordination of
computational modeling with targeted data collection but also,
perhaps even more importantly, on the appropriate training of
a new generation of computational toxicologists. These
individuals will have expertise in computational tools,
mathematics, and biology, and will be able to move seamlessly
between the laboratory and the computer. It is likely that this
vision applies not only to development of virtual tissues but
also, more broadly, to research and development in toxicology
and risk assessment.
SUMMARY AND CONCLUSION
The field of toxicology is rapidly approaching what could be
a golden era. Spurred on by-far reaching advances in biology,
chemistry, and computer sciences, the tools needed to open the
veritable black boxes that have prevented significant achieve-
ments in predictive power are being witnessed. We have
highlighted many of the topic areas that have demonstrated
advances in the state of the science, and from which more
advances are expected in the near future. Although the new
paradigm suggested by the NRC its Toxicity Testing in the
Twenty First Century: A Vision and a Strategy (National
Research Council, 2007) departs somewhat from the traditional
risk assessment approach exposed by the National Research
Council (1983), the two approaches can be mapped together,
and the tools of computational toxicology can provide outputs
that will help close gaps in many of the areas (Table 2). Some
aspects of computational toxicology discussed here, such as the
use of fate and transport models, the development of curated
and widely accessible databases, physiological based pharma-
cokinetic models, and characterizing uncertainty in models are
already being used in evaluating chemical risks, although
continued development is necessary to address emerging issues
such as nanomaterials. Other aspects, such as HTS and
toxicogenomics are witnessing extensive development and
application efforts in toxicology but have yet to become part of
mainstream data generation. Still others, like the assessment of
gene-environment interactions and development of virtual
tissues are really only beginning to be tested for applicability,
although these areas offer significant potential for improved
understanding of susceptibility and for extrapolating responses
across life stages, genders, and species.
Much of the high-throughput and genomics technology
beginning to be applied to toxicology was developed by the
pharmaceutical industry for use in drug discovery. Environ-
mental chemicals differ from drug candidates in a number of
important ways. For example, drugs are developed with
discrete targets in mind, conform to physicochemical properties
that assist in absorption, distribution, metabolism, and
•3
a _
o M
1 I
§ S,
'•g s
^7 O
u U
rl
W
^H CS
.-a* oo
° —
'S o
| •§
•§ £
a — i
J! <8
1 §
S "*"*
W H_*
(A ^
(D K
u 1 1
on fl
u
'1
G
QJ
O
•° B- §
a S y
"3 ;S
0 -o
1 1
§1
f2 =2
.3
2 I" "8
[2 _§ 6
>, 1 —
'o^o
.-H CO O
l|i
en"
y °°
^ S s ^
'S 1 .§> -
< 1 "1 ^
^ s & 5-
*
*
*
4, *
* * *
*
* * ^
* *
-X-
*
*
* * Jt
* * *
! 1 *
* *
* * *
* *
* *
1 *
* *
t *
-X-
I s
.2 Jf.f
?3 fr lj =3 ^^
y y T3 "^ rf
'g & M"2 2 &!
s * .s § B a ^
rt bo -H -^ ,, c ~—^ ^
S.S|^S|o^3
1 1 &-s ai 1 a
g^'oSSg'Su
1 1 1 §" S | S |
^gHHgS^S
U H Of*
2 S|
S S S
O *•* N
S|S § 1
1 a | a «
•^ OJ c« SH 3
'H T 1 ^ "
1 § l a -^
K Q W '(2
^j
1
1
^
"^
Q
S
O
o
<
OJ
•S
Si
CO
S
^§
Previous
-------
24
KAVLOCK ET AL.
excretion, have well understood metabolic profiles, and have
use patterns that are known and quantified. In contrast,
environmental chemicals generally are not designed with
biological activity as a goal, cover extremely diverse chemical
space, have poorly understood kinetic profiles, and are
generally evaluated at exposures levels well in excess of likely
real world situations. The challenge to successfully employ
these screening technologies for broader goals in toxicology
will be considerable, given that they have yet to yield the
significant increase in the pace of drug discovery that was
expected. On the other hand, whereas the goal of drug
discovery is to find the "needle in the haystack" using targeted
screening tools, the goal of predictive toxicology is to use these
tools more broadly to discern patterns of activity with regard
chemical impacts on biological systems and hence may be
more achievable. It will take a concerted effort on the part of
government, academia, and industry to achieve the trans-
formation of "Toxicity Testing in the 21st Century" that is so
eagerly awaited. Success will depend on building a robust
chemo-informatics infrastructure to support the field, on
conducting large-scale proof-of-concept studies that integrate
diverse data sources and types into more complete understand-
ing of biological activity, on developing a cadre of scientists
comfortable with both molecular tools and mathematical
modeling languages, and on convincing risk managers in
regulatory agencies that the uncertainties inherent in the new
approaches are sufficiently smaller or better characterized than
in traditional approaches. The rewards from such a success
would be significant. More chemicals will be evaluated by
more powerful and broad based tools, animals will be used
more efficiently and effectively in the bioassays designed to
answer specific questions rather than to fill in a checklist, and
the effects of mixtures of chemicals will be better understood
by employing system-level approaches that encompass the
underlying biological pathways whose interactions determine
the responses of the individual and joint effect of components
of mixtures. Clearly this will not happen soon, or without
significant investment. The National Research Council (2007)
estimates a 10- to 20-year effort at about $100 million per year
will be required for the paradigm shift they envisioned. This is
probably several-fold more than is being invested currently in
the area and, in most cases, those funds have not been
specifically guided by an overarching strategic vision such as
put forth by the NRC. Nonetheless, there are pockets of
progress occurring and the first success will likely be seen in
the ability to detect and quantify the interactions of chemicals
with key identifiable biological targets (e.g., nuclear receptors,
transporters, kinases, ion channels) and to be able to map these
potentials to toxicity pathways and phenotypic outcomes using
computational tools. Later successes will be seen in modeling
responses that require ever greater understanding of system-
level functioning that will ultimately take us to the un-
derstanding of susceptibility factors (be they for the individual,
life-stage, gender or species). All of these new methods,
capabilities, and advances offer great promise for the predictive
discipline of toxicology.
FUNDING
The Office of Research and Development of the United
States Environmental Protection Agency.
ACKNOWLEDGMENTS
The authors wish to recognize the contributions to the
International Science Forum on Computational Toxicology of
the session co-chairs (Steve Bryant, Richard Corley, Sean
Ekins, Tim Elston, Wout Slob, Rusty Thomas, Donald Tillit,
Raymond Tice, and Karen Watanabe), and presenters (Ellen
Berg, Robert Boethling, Steve Bryant, Lionel Carreira, Fanqing
Frank Chen, Harvey Clewell, Richard Corley, Christopher
Cramer, Amanda Drake, Sean Ekins, Tim Elston, Matthew
Etterson, H. Mark Fielden, Christopher Frey, Anna Georgieva,
Thomas Hartung, Jason Haugh, Kate Johnson, Jun Kanno,
Shinya Kuroda, Wildred Li, Markus Lill, Bette Meek, Ovanes
Mekenyan, John Petterson, Steve Proulx, Matt Redinbo,
Matthias Reuss, Kenneth Rose, Phil Sayre, Wout Slob, Roland
Somogyi, Clay Stephens, Justin Teeguarden, Rusty Thomas,
Raymond Tice, Sandor Vajda, Nico van Straalen, Chihae
Yang, Jeff Waring, Karen Watanabe, Richard Weinshilboum,
John Weinstein, and Matt Wheeler) all of whom were instru-
mental in bringing the state of the science of toxicology to the
International Science Forum on Computational Toxicology.
REFERENCES
Aldridge, B. B., Burke, J. M., Lauffenburger, D. A., and Sorger, P. K. (2006).
Physicochemical modeling of cell signaling pathways. Nat. Cell Biol. 8,
1195-1203; Available at: http://dx.doi.org/10.1038/ncbl497 (accessed June
27, 2007).
Allen, B. C., Covington, T. R., and Clewell, H. J. (1996). Investigation of the
impact of pharmacokinetic variability and uncertainty on risks predicted with
a pharmacokinetic model for chloroform. Toxicology 111, 289-303.
Arnold, W., Winget, P., and Cramer, C. J. (2002). Reductive dechlorination of
1,1,2,2-tetrachloroethane. Environ. Sci. Technol. 36, 3536.
Barton, H. A., Chiu, W. A., Setzer, R. W., Andersen, M. E., Bailer, A. J.,
Bois, F. Y., Dewoskin, R. S., Hays, S., Johanson, G., Jones, N., et al. (2007).
Characterizing uncertainty and variability in physiologically-based pharma-
cokinetic (PBPK) models: State of the science and needs for research and
implementation. Toxicol. Sci. 4. [Epub ahead of print].
Behar, M., Hao, N., Dohlman, H. G., and Elston, T. C. (2007). Mathematical
and computational analysis of adaptation via feedback inhibition in signal
transduction pathways. Biophys. J. 93, 806-821.
Bennett, R. S., and Etterson, M. A. (2007). Incorporating results of avian
toxicity tests into a model of annual reproductive success. Integr. Environ.
Assess. Monitor. 3(4), 498-507.
Berg, E. L., Kunkel, E. J., Hytopoulos, E., and Plavec, I. (2006).
Characterization of compound mechanisms and secondary activities by
BioMAP analysis. /. Pharmacol. Toxicol. Methods. 53, 67-74.
Previous
TOC
-------
COMPUTATIONAL TOXICOLOGY
25
Bottino, D., Penland, R. C., Stamps, A., Traebert, M., Dumotier, B.,
Georgieva, A., Helmlinger, G., and Lett, G. S. (2006). Preclinical cardiac
safety assessment of pharmaceutical compounds using an integrated systems-
based computer model of the heart. Prog. Biophys. Mol. Biol. 1-3, 414^143.
Breen, M. S., Villeneuve, D. L., Breen, M., Ankley, G. T., and Conolly, R. B.
(2007). Mechanistic computational model of ovarian steroidogenesis to
predict biochemical responses to endocrine active compounds. Ann. Biomed.
Eng. 35, 970-981.
Clewell, H. J., Gearhart, J. M., Gentry, P. R., Covington, T. R.,
VanLandingham, C. B., Crump, K. S., and Shipp, A. M. (1999). Evaluation
of the uncertainty in an oral Reference Dose for methylmercury due to
interindividual variability in pharmacokinetics. Risk Anal. 19, 547-558.
Clewell, H. J., Genty, P. R., Kester, J. E., and Andersen, M. E. (2005).
Evaluation of physiologically based pharmacokinetic models in risk assess-
ment: An example with perchloroethylene. Crit. Rev. Toxicol. 35, 413^33.
Conolly, R. B., Kimbell, J. S., Janszen, D. J., Schlosser, P. M., Kalisak, D.,
Preston, J., and Miller, F. J. (2004). Human respiratory tract cancer risks of
inhaled formaldehyde: Dose-response predictions derived from biologically-
motivated computational modeling of a combined rodent and human dataset.
Toxicol. Sci. 82, 279-296.
Cullen, A. C., and Frey, H. C. (1999). The Use of Probabilistic Techniques in
Exposure Assessment: A Handbook for Dealing with Variability and
Uncertainty in Models and Inputs. Plenum, New York.
D'haeseleer, P., Liang, S., and Somogyi, S. (2000). Genetic network inference:
From co-expression clustering to reverse engineering. Bioinformatics
(Oxford, England) 16, 707-726.
Ding, L., Stilwell, J., Zhang, T., Elboudwarej, O., Jiang, H., Selegue, J. P.,
Cooke, P. A., Gray, J. W., and Chen, F. F. (2005). Molecular
characterization of the cytotoxic mechanism of multiwall carbon nanotubes
and nano-onions on human skin fibroblast. Nano Lett. 5, 2448-2464.
Dix, D. J., Gallagher, K., Benson, W. H., Groskinsky, B. L., McClintock, J. T.,
Dearfield, K. L., and Farland, W. H. (2006). A framework for the use of
genomics data at the EPA. Nat. Biotechnol. 24, 1108-1111.
Dix, D. J., Houck, K. A., Martin, M. T., Richard, A. M., Setzer, R. W., and
Kavlock, R. J. (2007). The ToxCast program for prioritizing toxicity testing
of environmental chemicals. Toxicol. Sci. (Forum) 95, 5-12.
Dome, J. L. (2007). Human variability in hepatic and renal elimination:
Implications for risk assessment. /. Appl. Toxicol. 27, 411^-20.
Drake, A. J., Walker, B. R., and Seckl, J. R. (2005). Intergenerational
consequences of fetal programming by in utero exposure to glucocorticoids
in rats. Am. J. Physiol. Regul. Integr. Comp. Physiol. 288(1), R34-R38.
Ekins, S., Andreyev, S., Ryabov, A., Kirillov, E., Rakhmatulin, E. A.,
Sorokina, S., Bugrim, A., and Nikolskaya, T. (2006). A combined approach
to drug metabolism and toxicity assessment. Drug Metab. Dispos. 34,
495-503.
Evans, J. S., Rhomberg, L. R., Williams, P. L., Wilson, A. M., and Baird, S. J.
(2001). Reproductive and developmental risks from ethylene oxide: A
probabilistic characterization of possible regulatory thresholds. Risk Anal.
21, 697-717.
Faure, A., Naldi, A., Chaouiya, C., and Thieffry, D. (2006). Dynamical analysis
of a generic Boolean model for the control of the mammalian cell cycle.
Bioinformatics (Oxford, England) 22(14), el24-e!31.
Fielden, M. R., Brennan, R., and Gollub, J. (2007). A gene expression
biomarker provides early prediction and mechanistic assessment of hepatic
tumor induction by non-genotoxic chemicals. Toxicol. Sci. 99(1), 90-100.
Frey, H. C., and Patil, S. R. (2002). Identification and review of sensitivity
analysis methods. Risk Anal. 22, 553-578.
Giacomini, K. M., Brett, C. M., Altaian, R. B., Benowitz, N. L., Dolan, M. E.,
Flockhart, D. A., Johnson, J. A., Hayes, D. F., Klein, T., Krauss, R. M., et al.
(2007). The pharmacogenetics research network: From SNP discovery to
clinical drug response, din. Pharmacol. Ther. 81, 328-345 (Review).
Gibbs, J. R., and Singleton, A. (2006). Application of genome-wide single
nucleotide polymorphism typing: Simple association and beyond. PLoS
Genet. 2(10), e!50.
Ginsburg, G. (2005). Identifying novel genetic determinants of hemostatic
balance. /. Thromb. Haemost. 3, 1561-1568.
Hack, C. E., Chiu, W. A., Jay Zhao, Q., and Clewell, H. J. (2006). Bayesian
population analysis of a harmonized physiologically based pharmacokinetic
model of trichloroethylene and its metabolites. Regul. Toxicol. Pharmacol.
46, 63-83. [Epub 2006 Aug 4].
Haltuch, M. A., Berkman, P. A., and Garton, D. W. (2000). Geographic
information system (GIS) analysis of ecosystem invasion: Exotic mussels in
Lake Erie. Limnol. Oceanogr. 45, 1778-1787.
Hartung, T. (2006). ECVAM's progress in implementing the 3Rs in Europe.
ALTEX 23( Suppl.), 21-28.
Hattis, D., Baird, S., and Goble, R. (2002). A straw man proposal for
a quantitative definition of the RfD. Drug Chem. Toxicol. 25, 403^-36.
Health Canada. (2007). An integrated framework for the health-related
components of categorization of the Domestic Substances List under CEPA
1999. Available at: http://www.hc-sc.gc.ca/exsd. Accessed December 24,
2007.
HepatoSys. (2007). HepatoSys competence network: Systems biology of
hepatocytes. Available at: http://www.systembiologie.de/en/ (accessed July
10, 2007).
Hilal, S. H., Karickhoff, S. W., Carreira, L. A., and Shresth, B. P. (2005).
Hydration equilibrium constants of aldehydes, ketones, and quinazolione.
QSAR Comb. Sci. 24, 631.
Houck, K. A., and Kavlock, R. J. (2007). Understanding mechanisms of
toxicity: Insights from drug discovery research. Toxicol. Appl. Pharmacol.
in press.
Hunter, P. J. and Borg, T. K. (2003). Integration from proteins to organs: the
Physiome Project. Nat. Rev. Mol. Cell Biol. 4(3), 237-243.
Inglese, J., Auld, D. S., Jadhav, A., Johnson, R. L., Simeonov, A., Yasgar, A.,
Zheng, W., and Austin, C. P. (2006). Quantitative high-throughput
screening: A titration-based approach that efficiently identifies biological
activities in large chemical libraries. Proc. Natl. Acad. Sci. U.S.A. 103,
11473-11478.
Jaworska, J., Dimitrov, S., Nikolova, N., and Mekenyan, O. (2002).
Probabilistic assessment of biodegradability based on metabolic pathways:
CATABOL system. SAR QSAR Environ. Res. 13, 307-323.
Jolivette, L. J., and Ekins, S. (2007). Methods for predicting human drug
metabolism. Adv. Clin. Chem. 43, 131-176.
Karp, P. D. (2001). Pathway databases: A case study in computational
symbolic theories. Science 293, 2040-2044.
Kaya, T., Mohr, S. C., Waxman, D. J., and Vajda, S. (2006). Computational
screening of phthalate monoesters for binding to PPARgamma. Chem. Res.
Toxicol. 19, 999-1009.
Kehler, C. J., and Rahel, F. J. (1996). Thermal limits to salmonid distributions
in the Rocky Mountain region and potential habitat loss due to global
warming: A geographic information system (GIS) approach. Trans. Am.
Fish. Soc. 125, 1-13.
Kersting, K. (1984). Normalized ecosystem strain a system parameter for the
analysis of toxic stress in micro-ecosystems. Ecol. Bull. 36, 150-153.
Kimbell, J. S., Subramaniam, R. P., Gross, E. A., Schlosser, P. M., and
Morgan, K. T. (2001). Dosimetry modeling of inhaled formaldehyde:
Comparisons of local flux predictions in the rat, monkey, and human nasal
passages. Toxicol. Sci. 64, 100-110.
Kitano, H. (2002). Systems biology: A brief overview. Science 295,
1662-1664.
Kooistra, L., Leuven, R. S. E. W., Nienhuis, P. H., Wehrens, R., and
Buydens, L. M. C. (2001). A procedure for incorporating spatial variability
Previous
-------
26
KAVLOCK ET AL.
in ecological risk assessment of Dutch river flood plains. Environ. Manage.
28, 359-373.
Krallinger, M., Erhardt, R. A., and Valencia, A. (2005). Text-mining
approaches in molecular biology and biomedicine. Drug Discov. Today.
10, 439^45.
Leuven, R. S. E. W., and Poudevigne, I. (2002). Riverine landscape dynamics
and ecological risk assessment. Freshw. Biol. 47, 845-856.
Lewis, A., Bumpus, J. A., Truhlar, D. G., and Cramer, C. J. (2004). Molecular
modeling of environmentally important processes: Reduction potentials. /.
Chem. Educ. 81, 596.
Lill, M. A., Dobler, M., and Vedani, A. (2006). Prediction of small-molecule
binding to cytochrome P450 3A4: Flexible docking combined with
multidimensional QSAR. Chem. Med. Chem. 103(1), 14-27.
McCormick, C. M. (1999). Mapping exotic vegetation in the Everglades from
large-scale aerial photographs. Photogramm. Eng. Remote Sens. 65,
179-184.
Mokhtari, A., and Frey, H. C. (2005). Recommended practice regarding
selection of sensitivity analysis methods applied to microbial food safety
process risk models. Hum. Ecol. Risk Assess. 11, 591-605.
Mokhtari, A., Frey, C., and Zheng, J. (2006). Evaluation and recommendation
of sensitivity analysis methods for application to stochastic human exposure
and dose simulation models. /. Exposure Sci. Environ. Epidemiol. 16,
91-506.
National Research Council. (1983). Risk Assessment in the Federal Govern-
ment: Managing the Process. National Academies Press, Washington D. C.
National Research Council. (2007). Toxicity testing in the twenty-first century:
A vision and strategy. Available at: http://dels.nas.edu/dels/reportDetail.
php?link_id=4286&;session_id. Accessed December 24, 2007.
National Toxicology Program High-Throughput Screening Program. (2006).
Available at: http://ntp.niehs.nih.gov/ntpweb/index.cfm?objectid=05F80E15-
F1F6-975E-77DDEDBDF3B941CD. Accessed December 24, 2007.
Nguyen, T. P., and Ho, T. B. (2006). Discovering signal transduction
networks using signaling domain-domain interactions. Genome Inform.
17(2), 35^5.
O'Brien, P. J., Irwin, W., Diaz, D., Howard-Cofield, E., Krejsa, C. M.,
Slaughter, M. R., Gao, B., Kaludercic, N., Angeline, A., Bernardi, P., et al.
(2006). High concordance of drug-induced human hepatotoxicity with in
vitro cytotoxicity measured in a novel cell-based model using high content
screening. Arch. Toxicol. 80, 580-604.
Plowchalk, D. R., and Teeguarden, J. (2002). Development of a physiologically
based pharmacokinetic model for estradiol in rats and humans: A
biologically motivated quantitative framework for evaluating responses to
estradiol and other endocrine-active compounds. Toxicol. Sci. 69, 60-78.
Portier, C. J. (2002). Endocrine dismodulation and cancer. Neuro Endocrinol.
Lett. 23(Suppl. 2), 43^7.
Proulx, S. R., Promislow, D. E. L., and Phillips, P. C. (2005). Network thinking
in ecology and evolution. Trends Ecol. Evol. 20, 345-353.
Roberts, A., Pardo-Manuel de Villena, F., Wang, W., McMillan, L., and
Threadgill, D. W. (2007). The polymorphism architecture of mouse genetic
resources elucidated using genome-wide resequencing data: Implications for
QTL discovery and systems genetics. Mamm. Genome. 18, 473^-81.
Rose, K. A., Murphy, C. A., Diamond, S. L., Fuiman, L. A., and Thomas, P.
(2003). Using nested models and laboratory data for predicting population
effects of contaminants on fish: A step towards a bottom-up approach for
establishing causality in field studies. Hum. Ecol. Risk Assess. 9, 231-257.
Saltelli, A., Chan, K., and Scott, E. M. (2000). Mathematical and Statistical
Methods: Sensitivity Analysis. Wiley, Chichester.
Sasagawa, S., Ozaki, Y.-L, Fujita, K., and Kuroda, S. (2005). Prediction and
validation of the distinct dynamics of transient and sustained ERK activation.
Nat. Cell Biol. 7, 365-373.
Schneider, I. C., and Haugh, J. M. (2006). Mechanisms of gradient sensing and
chemotaxis: Conserved pathways, diverse regulation. Cell Cycle (George-
town, Tex.) 5, 1130-1134.
Schwartz, D., and Collins, F. (2007). Environmental biology and human
disease. Science 316, 695-696.
Scott, J., Ideker, T., Karp, R. M., and Sharan, R. (2006). Efficient algorithms
for detecting signaling pathways in protein interaction networks. /. Comput.
Biol. 13, 133-144.
Setzer, R. W., Lau, C., Mole, L. M., Copeland, M. F., Rogers, J. M., and
Kavlock, R. J. (2001). Toward a biologically based dose-response model for
developmental toxicity of 5-fluorouracil in the rat: A mathematical construct.
Toxicol. Sci. 59, 49-58.
Sheu, S. H., Kaya, T., Waxman, D. J., and Vajda, S. (2005). Exploring the
binding site structure of the PPAR-y ligand binding domain by computa-
tional solvent mapping. Biochemistry 44, 1193-1209.
Slob, W., and Pieters, M. N. (1998). A probabilistic approach for deriving
acceptable human intake limits and human health risks from toxicological
studies: General framework. Risk Anal. 18, 787-798.
The International HapMap Consortium. (2005). A haplotype map of the human
genome. Nature 437, 1299-1320.
Thomas, R. S., Pluta, L., Yang, L., and Halsey, T. A. (2007). Application of
genomic biomarkers to predict increased lung tumor incidence in 2-year
rodent cancer bioassays. Toxicol. Sci. 97, 55-64.
Timchalk, C., Trease, H. E., Trease, L. L., Minard, K. R., and Corely, R. A.
(2001). Potential technology for studying dosimetry and response to
airborne chemical and biological pollutants. Toxicol. Ind. Health. 17,
270-276.
Tong, S. T. Y. (2001). An integrated exploratory approach to examining the
relationships of environmental stressors and fish responses. /. Aquat. Ecosys.
Stress Recov. 9, 1-19.
U.S. EPA. (2003). A framework for a computational toxicology research
program. Washington, D.C. EPA600/R-03/65.
U.S. EPA. (1997). Guiding principles for Monte Carlo analysis. U.S. Envi-
ronmental Protection Agency, Risk Assessment Forum, Washington, D.C.
EPA/630/R-97/001.
van der Voet, H., and Slob, W. (2007). Integration of probabilistic exposure
assessment and probabilistic hazard characterization. Risk Anal. 27, 351-371.
van Straalen, N. M. (2003). Ecotoxicology becomes stress ecology. Environ.
Sci. Technol. 37, 324A-330A.
van Straalen, N. M., and Roelofs, D. (2006). Introduction to Ecological
Genomics. Oxford University Press, New York.
Vedani, A., Dobler, M., and Lill, M. A. (2006). The challenge of predicting
drug toxicity in silico. Basic Clin. Pharmacol. Toxicol. 99, 195-208.
Villeneuve, D. L., Larkin, P., Knoebl, L, Miracle, A. L., Kahl, M. D.,
Jensen, K. M., Makynen, E. A., Durhan, E. J., Carter, B. J., Denslow, N. D.,
et al. (2007). A graphical systems model to facilitate hypothesis-driven
ecotoxicogenomics research on the teleost brain-pituitary-gonadal axis.
Environ. Sci. Technol. 41, 321-330.
Wang, H., Huang, H., Li, H., Teotico, D. G., Sinz, M., Baker, S. D.,
Staudinger, J., Kalpana, G., Redinbo, M. R., and Mani, S. (2007). Activated
pregnenolone X-receptor is a target for ketoconazole and its analogs. Clin.
Cancer Res. 13, 2488-2495.
Watanabe, K. H., Jensen, K. M., Villeneuve, D. L., Nichols, J., Zhenhong, L.,
Breen, M. S., Bencic, D. C., Collette, T., Denslow, N. D., and Ankley, G. T.
(2006). A physiologically based model of the hypothalamus-pituitary-
gonadal axis in female fathead minnows. 27th Annual Society of
Environmental Toxicology and Chemistry (SETAC) North America Meeting.
Montreal, Quebec, Canada, November 2006 (Abstract).
Weinstein, J. (2006). Spotlight on molecular profiling: 'Integromic' analysis of
the NCI-60 cancer cell lines. Mol. Cancer Ther. 5, 2601-2605.
Previous
TOC
-------
COMPUTATIONAL TOXICOLOGY
27
Wheeler, M. W., and Bailer, A. J. (2007). Properties of model-averaged
BMDLs: A study of model averaging in dichotomous response risk
estimation. Risk Anal. 27, 659-670.
Whiteside, T. S., Hilal, S. H., and Carreira, L. A. (2006). Estimation of
phosphate ester hydrolysis rate constants—Alkaline hydrolysis. QSAR
Comb. Sci. 25, 123-133.
Wijnhoven, S. W., Hoogervorst, E. M., de Waard, H., van der Horst, G. T., and
van Steeg, H. (2007). Tissue specific mutagenic and carcinogenic responses
in NER defective mouse models. Mutat. Res. 3, 77-94.
Wilke, R. A., Lin, D. W., Roden, D. M., Watkins, P. B., Flockhart, D.,
Zineh, I., Giacomini, K. M., and Krauss, R. M. (2007). Identifying genetic
risk factors for serious adverse drug reactions: current progress and
challenges. Nat Rev Drug Discov. 6(11), 904-916.
Yang, C. (2007). Understanding toxicity through chemical and biological
fingerprints. Int. Sci. Forum Comput. Toxicol. 21-23; (Abstract).
Yang, Y., Abel, S. J., Ciurlionis, R., and Waring, J. F. (2006). Development of
a toxicogenomics in vitro assay for the efficient characterization of
compounds. Pharmacogenomics 7, 177-186.
Previous
-------
This article was downloaded by: [US EPA Environmental Protection Agency]
On: 2 September2009
Access details: /Access Details: [subscription number 789514190]
Publisher Taylor & Francis
Informa Ltd Registered in England and Wales Registered Number: 1072954 Registered office: Mortimer House,
37-41 Mortimer Street, London W1T 3JH, UK
JOURNAL OF
TOXICOLOGY AND
ENVIRONMENTAL
HEALTH
Journal of Toxicology and Environmental Health, Part B
Publication details, including instructions for authors and subscription information:
http://www. info rmaworld. co m/smpp/title~content=t713667286
Database for Physiologically Based Pharmacokinetic (PBPK) Modeling:
Physiological Data for Healthy and Health-Impaired Elderly
Chad M. Thompson a; Douglas O. Johns b; Babasaheb Sonawane a; Hugh A. Barton c; Dale Hattis d; Robert
Tardife; Kannan Krishnan e
* National Center for Environmental Assessment, Office of Research and Development, U.S. Environmental
Protection Agency, Washington, DC, USA b National Center for Environmental Assessment, Office of
Research and Development, U.S. Environmental Protection Agency, Research Triangle Park, North Carolina,
USA c Pfizer Inc., Pharmacokinetic/Pharmacodynamic Modeling, Groton, CT, USA d Marsh Institute Center for
Technology, Environment and Development, Clark University, Worcester, Massacusetts, USA e Groupe de
recherche interdisciplinaire en sante et Departement de sante environnementale et sante au travail,
Universite de Montreal, Montreal, Canada
Online Publication Date: 01 January 2009
To cite this Article Thompson, Chad M., Johns, Douglas O., Sonawane, Babasaheb, Barton, Hugh A., Hattis, Dale, Tardif, Robert and
Krishnan, Kannan(2009)'Database for Physiologically Based Pharmacokinetic (PBPK) Modeling: Physiological Data for Healthy and
Health-Impaired Elderly',Journal of Toxicology and Environmental Health, Part 6,12:1,1 —24
To link to this Article: DOI: 10.1080/10937400802545060
URL: http://dx.doi.org/10.1080/10937400802545060
PLEASE SCROLL DOWN FOR ARTICLE
Full terms and conditions of use: http://www.informaworld.com/terms-and-conditions-of-access.pdf
This article may be used for research, teaching and private study purposes. Any substantial or
systematic reproduction, re-distribution, re-selling, loan or sub-licensing, systematic supply or
distribution in any form to anyone is expressly forbidden.
The publisher does not give any warranty express or implied or make any representation that the contents
will be complete or accurate or up to date. The accuracy of any instructions, formulae and drug doses
should be independently verified with primary sources. The publisher shall not be liable for any loss,
actions, claims, proceedings, demand or costs or damages whatsoever or howsoever caused arising directly
or indirectly in connection with or arising out of the use of this material.
Previous
-------
Journal of Toxicology and Environmental Health, PartB, 12:1-24, 2009
ISSN: 1093-7404 print /1521-6950 online
DOI: 10.1080/10937400802545060
j Taylor £t Francis
DATABASE FOR PHYSIOLOGICALLY BASED PHARMACOKINETIC (PBPK)
MODELING: PHYSIOLOGICAL DATA FOR HEALTHY
AND HEALTH-IMPAIRED ELDERLY
Chad M. Thompson1, Douglas O. Johns2, Babasaheb Sonawane1, Hugh A. Barton3,
Dale Hattis4, Robert Tardir, Kannan Krishnan5
1 National Center for Environmental Assessment, Office of Research and Development, U.S.
Environmental Protection Agency, Washington, DC, USA, 2National Center for Environmental
Assessment, Office of Research and Development, U.S. Environmental Protection Agency,
Research Triangle Park, North Carolina, USA, 3Pfizer Inc., Pharmacokinetic/Pharmacodynamic
Modeling, Groton, CT, USA, 4Marsh Institute Center for Technology, Environment and
Development, Clark University, Worcester, Massacusetts, USA, and 5Groupe de recherche
interdisciplinaire en sante et Departement de sante environnementale et sante au travail,
Universite de Montreal, Montreal, Canada
Physiologically based pharmacokinetic (PBPK) models have increasingly been employed in chemical health risk
assessments. By incorporating individual variability conferred by genetic polymorphisms, health conditions, and
physiological changes during development and aging, PBPK models are ideal for predicting chemical disposition in
various subpopulations of interest. In order to improve the parameterization of PBPK models for healthy and health-
impaired elderly (herein defined as those aged 65 yr and older), physiological parameter values were obtained from
the peer-reviewed literature, evaluated, and entered into a Microsoft ACCESS database. Database records include
values for key age-specific model inputs such as ventilation rates, organ volumes and blood flows, glomerular filtration
rates, and other clearance-related processes. In total, 528 publications were screened for relevant data, resulting in
the inclusion of 155 publications comprising 1051 data records for healthy elderly adults and 115 data records for
elderly with conditions such as diabetes, chronic obstructive pulmonary disease (COPD), obesity, heart disease, and
renal disease. There are no consistent trends across parameters or their associated variance with age; the gross
variance in body weight decreased with advancing age, whereas there was no change in variance for brain weight.
The database contains some information to inform ethnic and gender differences in parameters; however, the
majority of the published data pertain to Asian (mostly Japanese) and Caucasian males. As expected, the number of
records tends to decrease with advancing age. In addition to a general lack of data for parameters in the elderly with
various health conditions, there is also a dearth of information on blood and tissue composition in all elderly groups.
Importantly, there are relatively few records for alveolar ventilation rate; therefore, the relationship between this
parameter and cardiac output (usually assumed to be 1:1) in the elderly is not well informed by the database.
Despite these limitations, the database represents a potentially useful resource for parameterizing PBPK models for
the elderly to facilitate the prediction of dose metrics in older populations for application in risk assessment.
Physiologically based pharmacokinetic (PBPK) models and biologically based dose-response
(BBDR) models have increasingly been employed in chemical health risk assessments, and their
evaluation and application in a regulatory context has recently received much attention (Clark et al.,
2004; Barton et al., 2007; Chiu et al., 2007; Thompson et al., 2008). These models utilize biological
information in order to predict the disposition of chemicals for which limited human data exist. It is
widely recognized that, in addition to genetic variation, differences in life stage and health status
affect the disposition of environmental toxicants. When pharmacokinetic models are developed for
predicting the disposition of environmental toxicants in humans, the models frequently represent a
The authors extend their gratitude to Drs. Andrew Geller and Linda Birnbaum of the National Health and Environmental Effects
Research Laboratory for their insightful assistance in the development of this database. This database was developed under a contract
(RFQ-DC-03-00328) funded by the U.S. Environmental Protection Agency.
The views expressed in this article are those of the authors and do not necessarily reflect the views or policies of the U.S. Environ-
mental Protection Agency.
Address correspondence to Chad M. Thompson, National Center for Environmental Assessment, Office of Research and Development,
U.S. Environmental Protection Agency, 1200 Pennsylvania Avenue, NW, Washington, DC, 20460, USA. E-mail: Thompson.Chad@epa.gov
Previous
TOC
-------
C. M. THOMPSON ET AL
standard healthy adult, in part because the human data used to develop such models are often
obtained from younger adults. Thus children, the elderly, and health-impaired individuals represent
subpopulations that may benefit from specific consideration of susceptibility by modeling internal
dosimetry via PBPK modeling.
In order to construct models for these subpopulations, it is important to have as much information
as possible for the physiological parameter values that most influence disposition. Such data include
alveolar ventilation, cardiac output, organ and tissue weights/volumes and corresponding blood
flows, clearance parameters (e.g., glomerular filtration rate, liver enzyme content), and body com-
position. In developing PBPK models, analysts often rely on a variety of resources for obtaining
these data; unfortunately, this can add to the variability and uncertainty in and among models. In
an attempt to alleviate these problems, peer-reviewed compilations of physiological parameters
were developed for children, laboratory animals, and adult humans (Davies & Morris, 1993; Brown
et al., 1997; Price et al., 2003a, 2003b; Gentry et al., 2004); however, no such compilations exist
for empirical measures of the quantitative changes in physiological parameters in healthy and
health-impaired older adults. Recognizing that aging may impart various sensitivities to environ-
mental exposures as well as increased variability in responses (Geller & Zenick, 2005; Brown et al.,
2005), it is anticipated that the development of such information will improve health risk assess-
ments by better characterizing risk to older populations and perhaps reducing the uncertainty
inherent in risk assessment.
The primary objectives of this study were to review and collect relevant physiological data for
the elderly (i.e., individuals >65 yr of age), and to integrate these data within a free downloadable
relational database. In addition, an analysis of the influence of age on selected physiological param-
eters is presented, and data gaps in physiological parameters required for PBPK modeling in these
populations are identified.
LITERATURE REVIEW AND DATABASE DESCRIPTION
Literature Review
This work was conducted between 2005 and 2006 using published reports in the peer-reviewed
literature, and searched via PUBMED using the terms aged, elderly, old, age-dependent, aging, ageing,
or geriatric, in combination with the terms related to body mass index (BMI), body composition,
cardiac output, breathing rate, tissue volumes, tissue blood flows, tissue composition, and clearance
parameters. Initially, about 3000 to 24,000 references were obtained every time a search was con-
ducted using two terms in combination (e.g., cardiac output and age-dependent). Further, the use
of aged, old, age-dependent, aging, ageing, and geriatric redundantly produced several thousand
papers that were all either previously captured by the use of the term "elderly" or irrelevant to the
present study. On the basis of these observations and feasibility issues, the search strategy was
refined as follows. Searches were conducted using elderly as the sole key word in combination with
each of the physiological terms and by specifying certain limits during the search. These limits
resulted in literature that (1) contained both search terms in either the title or the abstract, (2) inves-
tigated human subjects, and (3) collected data in subjects aged 65 yr or older. The searches were
then repeated with the addition of a term representative of the pathophysiological condition of
interest, i.e., diabetes, COPD, obesity, angina pectoris, and renal disease (Table 1). There is no
claim that 100% of publications on quantitative physiological information in elderly were identified
during this process, but it is believed that these approaches yielded a reasonably unbiased repre-
sentation of the available literature.
Following the literature search, the abstract of each publication (title when the abstract is not
available in PUBMED) was evaluated to see whether it might contain relevant data. In this process,
clinical and in vivo pharmacokinetic studies with various drugs were excluded except when the
study used a probe substance for an enzyme of interest, measured renal clearance parameters, or
included measurements of organ weights or blood flows. In total, 528 peer-reviewed publications
were retrieved and reviewed to identify the specific parameter evaluated as well as the experimental
Previous I TOC
-------
PHYSIOLOGICAL PARAMETERS IN THE ELDERLY
TABLE 1. Health Conditions Included in the Database
Health condition
Obesity3
Diabetes3
COPD3
Heart disease3
Liver disease
Renal disease3
Reason for inclusion
Prevalent among elderly; about 26% men aged 65-74 and greater percentage of women
aged between 65 and 74 are obese
1 8% of all people over 60 have diabetes
Fourth largest cause of death in the United States and ranks as the second major Social
Security-compensated disability
Has a major incidence (30%) and is the leading causing of death in persons aged >65 yr
Influence on hepatic clearance
75% of elderly persons have renal failure at presentation with 20% requiring dialysis
Records
31
25
30
14
10
5
3Beers (2005).
design/protocol, sampling method, number of samples, statistical analysis, precision/variability, and
any bias. A publication was excluded from further consideration if it reported data only in figures,
data not relevant to the database, or data without appropriate or interpretable units. Studies reporting
physiological parameters without body weight or cardiac output measurement were retained, as
were studies that did not identify the ethnicity (frequent) or gender (rare) of the subjects; however,
these studies reported physiological data (e.g., tissue blood flow) along with age or some other
parameter relevant for further analysis (e.g., tissue weight, body mass index). Designation of study
subjects' disease states by the original investigator was accepted as such, and no reevaluations of
these classifications were attempted. Generally, the medication history of the study subjects was not
reported; however, based on the summary information available, studies with individuals (or patients)
on treatment/medication affecting the physiological parameter of interest were excluded from the
present study. In all, 155 publications were retained as sources of data for inclusion in the database.
Although some relevant studies may have been overlooked, the current structure of the database
will permit the future addition of new data and updating of existing physiological information in
older adults.
Description of the Database
A general description of the structure of the database can be found in Gentry et al. (2004), as
well as at the following U.S. Environmental Protection Agency (EPA) web site, where the database is
freely available for download (http://cfpub.epa.gov/ncea/cfm/recordisplay.cfm?deid = 188288). This
Microsoft ACCESS database is an extension of a similar database for early human life stages developed
in cooperation with the International Life Sciences Institute. (The early life stage database is available
at http://rsi.ilsi.org/Projects/Physio_Parameters_db.htm). In brief, this database contains three tables
linked by a study identification number (Figure 1). Following the entry of information into the study
table, the information on the study subjects was entered into the subject characteristics and PBPK
tables. Figure 1 shows the fields in the each table. Some of the fields in the PBPK table include
those relating to specific parameter evaluated, value of parameter, parameter units, group or indi-
vidual data, parameter variability, type of variability, number evaluated, body weight, method used,
and comments.
Each data record contained the numerical value along with variability for each parameter (if
applicable and available), recognizing that for grouped data the values capture some elements of
biological variability and measurement-related uncertainty. Multiple data records were created for
studies reporting values for more than one parameter. Similarly, multiple records were created for
studies containing values in more than one study group (e.g., diabetic group and control group).
Although the initial intent was to collect parameters in older adults of specific health statuses
(healthy, obese, diabetic, chronic obstructive pulmonary disease [COPD], renal disease, or cardiac
disease), data were also collected in those designated as patients but not specifically representative
of the aforementioned categories. This was done, in part, because the literature review indicated
that a number of studies reported data collected in patients often admitted for diagnosis or
treatment of other symptoms. Therefore, in an effort to include such data, the term "patients" was
Previous I TOC
-------
C. M. THOMPSON ET AL
Study
f DM0
ReferenceGtaison
Laborataryl^fsapallnvestsgatof
Published
RtflD
Sufaje tf Characterises
* IDNO
f SSNO
S«
AgeCategof>'
Age
AgeUnrts
EthrwGroup
1DNO
SSNO
NumberEyaJuated
BodyWeight
BWUnits
BWVarType
BWVar
BMI
BMIUnits
BMIVarType
EMIVar
BSA
BSAUnits
BSAVarType
BSAVaf
Specif lePajameterEyaSuMed
MethodsUsedtoMeasufePaTametef
SemrtivityofMethod
GraupData
ParametefUnrts
ParameterVarTYpe
TypeofDistributionfcHMeasurement
ParameterVar
Digit iieafijyte
Comments
LinfctoData
UneertaintyinParameter
Multiplexes
QuestionabieRciiabiiityofMetnods
J
FIGURE 1. Database structure. Database contains three tables linked by a study identification number. Abbreviations: IDNO, unique
identification number for each study; RefID, reference manager database identification number; SSNO, subject-specific number
assigned to each individual or group within a given study (i.e., it is a subfield under IDNO); BW, body weight; Var, variability; BMI, body
mass index; BSA, body surface area.
introduced into the database. It is likely a consideration of the end users of this database as to
whether individuals identified as patients should be included as healthy subjects, as it is probable
that many older adults are affected by similar yet undiagnosed conditions. Some studies reported
physiological parameter values along with data on BMI and body surface area (BSA), but did not
include data on body weight. Thus, additional fields were added for BMI and BSA to potentially
allow for normalization of physiological parameters (e.g., cardiac output divided by BSA) to facili-
tate comparison across studies.
The large majority of the studies were cross-sectional in design, comparing individuals or popu-
lations of different ages at the same time. There were also a few longitudinal studies that reported
measures of a parameter obtained more than once during a period of several years. Such data
corresponding to each sampling time were entered as individual records. In total, 1166 data
records were entered and in several cases a single record consisted of several useful observations
(e.g., body weight, BMI, BSA, blood flow rate). For quality assurance, 10% of the data entered into
the database were independently checked by a second scientist, who compared the values entered
into in the database with that in the original study. Studies reporting parameter values for individuals
between 50 and 64 yr of age were included when there were limited or no data in elderly, as these
data can provide additional support of changes, if any, in physiological parameter values as a
function of age.
PHYSIOLOGICAL DATA IN HEALTHY ELDERLY
Gross Weight
Body Weight Body weight is a basic anthropometric parameter reported in most studies included
in the database. Most records in the database containing a numerical value for a physiological
Previous
TOC
-------
PHYSIOLOGICAL PARAMETERS IN THE ELDERLY
parameter also have data on the body weight, BSA, or BMI measured in the same individual or
study group. Four major reports used in populating the database provide data on body weight.
Inoue and Otsu (1987) published data based on 1067 serial autopsies (493 women and 507 men)
performed from June 1972 to March 1977 at a general hospital in Tokyo. Puggard et al. (2002)
reported body weight data for Danish women (65 yr of age (n = 22), 75 yr of age (n = 26), and
85 yr of age (n = 31)). Galloway et al. (1965) reported mean body weights from 400 autopsies of
people in their twenties through nineties performed at a Veteran's Administration Hospital in
Wisconsin. Unfortunately, the latter study did not provide data on the number of subjects in each
of their 10-yr age groups, the standard deviations of the measurements, ethnicity, or gender. It was
presumed that these data represent male subjects only, since they were collected before large numbers
of women veterans became eligible for medical services from this agency. Unlike the previous three
studies, the Third National Health and Nutrition Examination Survey (NHANES III)* contains body
weight data for 5679 living individuals (2742 males and 2937 females). Table 2 provides an illustra-
tion of the data on body weight as a function of age in Japanese males and females based on
autopsy data from Inoue and Otsu (1987). These data indicate an overall decline in body weights
with age, and perhaps a modest tendency toward narrowing of the distribution, as indicated by the
lower gross coefficient of variation (CV) values in older age groups for both genders. Comparison of
gross mean male body weights in the Japanese and U.S. populations (Figure 2) indicates that the
U.S. subjects studied by Galloway et al. (1965) are much heavier on average than the male subjects
as reported by Inoue and Otsu (1987). Therefore it is important to judge the rates of body-weight
decline indicated for the two populations in relative terms. When the slope of each of the regres-
sion lines in Figure 2 is normalized to the mean body weights in the 60-69 yr age group, it can be
seen that both populations show body-weight declines of approximately 0.4% per year in the eld-
erly age range.
Body Mass Index, Fat Mass, and Fat-Free Mass BMI or Quetlet's index is defined as body
weight (kg) divided by height in square meters (m2) (Quetelet, 1869). The database contains numer-
ous entries for BMI, as this index is computed fairly routinely in clinical studies, particularly in those
with elderly subjects. Data for BMI were entered into the database along with body-weight infor-
mation or data on specific physiological parameters for each individual or study group, where avail-
able. Figure 3 depicts changes in body weight and BMI as a function of age based on the data from
the NHANES III study. It can be seen that despite the larger body weights in adult males, mean BMI
TABLE 2. Elderly Body weight (kg) in Japanese Population
Age range Number of cases Gross mean SD Gross CV Standardized mean Standardized SD Standardized CV
Males
60-69
70-79
80-89
90-99
100+
Total
Females
60-69
70-79
80-89
90-99
100+
Total
104
234
158
11
507
73
198
174
44
4
493
45.1
44.0
41.8
39.6
39.2
38.4
35.1
33.9
34.1
9.4
8.8
7.9
4.4
8.6
8.8
7.3
6.6
5.9
0.208
0.200
0.189
0.111
0.219
0.229
0.208
0.195
0.173
46.0
44.3
42.5
39.5
40.5
38.8
36
34.5
31.9
9.0
9.0
8.3
7.5
9.5
9.8
7.3
7.3
6.7
0.196
0.203
0.195
0.190
0.235
0.253
0.203
0.212
0.210
Note. Data from Inoue and Otsu (1 987).
*NHANES III is based on a nationally representative sample of the U.S. civilian non-institutionalized population between the ages
of 2 months and 90 yr.
Previous
TOC
-------
C. M. THOMPSON ET AL
70
gi
'33
50-
40-
30
0.340; y = 84.2-0.236x RA2 = 0.824 D U.S. Males
0.415;y = 57.6-0.187x RA2 = 0.980 • Japanese Males
60 70 80 90
Midpoint of Age Range (Yr)
100
FIGURE 2. Age-related declines in male body weight. Data for U.S. (Galloway et al., 1965) and Japanese (Inoue & Otsu, 1987) autopsies.
100
80-
=- 60-
gi
I
c
s
40-
20-
0 10 20 30 40 50 60 70 80 90
Age (yrs)
0 10 20 30 40 50 60 70 80 90
Age (yrs)
FIGURE 3. Population-weighted mean body weights (A) and BMI (B). Data are derived from NHANES III.
values are almost identical between the genders with the possible exception of a short period in
early adulthood. It is also evident, by comparing Figures 2 and 3A, that body weight has increased
in U.S. males during the 30 yr between the Galloway et al. (1965) and NHANES III studies.
One reason for calculating BMI is that it shows a strong relationship with body fat content. Even
though there are sophisticated methods for the quantification of body fat, the BMI has been used
frequently for this purpose (Deurenberg et al., 1991). "Body fat" in this context refers to organic
compounds constituting the esters of glycerol and fatty acids and their associated organic groups,
which serve as a reserve of energy for the body. Body fat reported in the literature represents both
essential and nonessential fat. "Essential" fat refers to the lipid constituents of cells and represents
Previous
TOC
-------
PHYSIOLOGICAL PARAMETERS IN THE ELDERLY
about 2 to 5% of the lean body mass, whereas the "nonessential" fat is contained in adipose tissue,
which occurs principally in subcutaneous tissue, yellow bone marrow, and the abdominal cavity—
genital, perirenal, mesenteric, and omental compartments (ICRP, 1975). It is worth noting that in
PBPK modeling, "essential" fat is included in all tissue compartments, whereas the "nonessential"
fat is often represented as a separate compartment with its own tissue volume and separate blood
flow. Several hundred values of fat mass (in kilograms and as percentages) in older adults have been
entered into the database. The fat mass is a function of a number of factors: height, weight, age, gen-
der, diet, and physical activity (Sitar, 1998). Generally, women have a higher proportion of body fat
than men with the same BMI, and in both genders the fat percentage increases with age. In women,
there appears to be a postmenopausal acceleration of this trend. In general, the maximum age for
fat accumulation is around the sixth decade of life, with a plateau phase and a subsequent reduction
in the amount and proportion of body fat occurring in the seventh and eighth decades of life (Sitar,
1998). The data for older adults included in the database are consistent with this general trend.
Although the fat-free mass refers to body mass devoid of all physically extractable fat, the lean body
mass additionally includes the essential fat. These data were entered along with body weight or BMI
information from each study, thus facilitating further analysis, particularly with respect to tissue
weights.
Cardiopulmonary Parameters
Breathing Rate and Pulmonary Function Parameters Vital capacity is the most widely mea-
sured respiratory parameter and information is available from both cross-sectional and longitudinal
studies. Previous analyses suggest a vital capacity decline of about 36 ml/yr in men and about
21 ml/yr in women after age 40 (Goldman & Becklake, 1959; Knudson, 1991; Crapo, 1993). On
the other hand, cross-sectional and longitudinal studies are suggestive of an age-dependent increase
in dead air space; i.e., the volume of the conducting airways where gas exchange does not occur
(Pierce & Ebert, 1958; Anthonisen etal., 1994). Seven records on alveolar ventilation rate, obtained
from three studies (Morris et al., 1956, 1964; Miller & Tenney, 1956), are included in the database.
Table 3 contains data for most of the cardiopulmonary entries in the database, including those for
individuals with impaired health (note that not all fields in the database are shown in the table).
Cardiac Output and Cardiac Index Cardiac output refers to the volume of blood ejected
from each ventricle of the heart per unit time. Cardiac output normalized to BSA is referred to as
cardiac index. Some studies suggested a reduction in cardiac output with increasing age whereas
others did not (Lauson et al., 1944; Bradfonbrener et al., 1955; Julius et al., 1967; Collis et al.,
2001; Katori, 1979; Luisada etal., 1980; Crean etal., 1986; Lakatta, 1990; Okazaki etal., 1996).
This might indicate that factors other than age exert important influences on cardiac output and that
these factors may have differed among the individuals in the various studies.
Tissue Weights (Volumes)
Liver Weight A number of studies showed that liver weight decreases with the advancement
of age. Boyd (1933) reported a decrease in liver weight between ages 20 and 80 yr in men and
women, based on measurements in 1582 subjects following accidental death. This study and others
are suggestive of an approximate 20% decrease in liver weight between the second or third decade
and the eighth or ninth decade (Swift et al.,1978; Marchesini et al., 1988; Bach et al., 1981). The
database contains 33 records on liver volume from the following studies: Galloway et al. (1965),
Sato et al. (1970), Swift et al. (1978), Bach et al. (1981), Inoue and Otsu (1987), Wynne et al.
(1989), Puggaard et al. (2002), and Chouker et al. (2004). Tissue volumes for liver weight and other
organs/compartments are shown in Table 4. Comparison of the gross liver weights in the U.S. and
Japanese male subjects indicates a more rapid decline with age in the U.S subjects (1.17 vs. 0.97%
per year) (Figure 4A).
Brain Weight The database contains 16 records of brain weight in older adults, collected from
four studies (Gordan, 1956; Galloway et al., 1965; Inoue & Otsu, 1987; Puggaard et al., 2002).
Based on data from Inoue and Otsu (1987), brain weight exhibits the smallest interindividual variation
Previous I TOC
-------
C. M. THOMPSON ET AL
TABLE 3. Sample Query Data for Cardiopulmonary Function in Healthy and Health-Impaired Elderly
Parameter
Alveolar ventilation3
Alveolar ventilation6
Alveolar ventilation0
Alveolar ventilation6
Alveolar ventilation0
Alveolar ventilation6
Alveolar ventilation0
Cardiac indexd
Cardiac indexd
Cardiac index6
Cardiac indexf
Cardiac indexf
Cardiac index6
Cardiac index6
Cardiac index6
Cardiac index11
Cardiac index11
Cardiac index11
Cardiac index11
Cardiac index11
Cardiac index11
Cardiac index11
Cardiac index11
Cardiac index11
Cardiac index11
Cardiac index11
Cardiac index11
Cardiac index11
Cardiac index11
Cardiac index11
Cardiac index11
Cardiac index11
Cardiac index11
Cardiac index11
Cardiac index11
Cardiac index'
Cardiac index'
Cardiac index'
Cardiac index6
Cardiac index'
Cardiac index6
Cardiac index6
Cardiac index
Cardiac index'
Cardiac index"1
Cardiac index"1
Cardiac index"1
Cardiac index"1
Cardiac index"
Cardiac index"
Cardiac index"
Cardiac index"
Cardiac index0
Ventilation rate01
Ventilation rateP
Ventilation rate*3
Vital capacity^
Vital capacity01
Health
Healthy
NS
NS
NS
NS
NS
NS
COPD
COPD
COPD
Diabetes
Diabetes
Healthy
Healthy
Healthy
Healthy
Healthy
Healthy
Healthy
Healthy
Healthy
Healthy
Healthy
Healthy
Healthy
Healthy
Healthy
Healthy
Healthy
Healthy
Healthy
Healthy
Healthy
Healthy
Healthy
Healthy
Healthy
Healthy
Healthy
Healthy
Healthy
Healthy
Heart disease
Heart disease
Heart disease
Heart disease
Heart disease
Heart disease
Patients
Patients
Patients
Patients
Patients
Healthy
Hypertensive
Hypertensive
COPD
COPD
Gender
M
M
M
M
M
M
M
F/M
F/M
F/M
F/M
F/M
F
F
F
F/M
F/M
F/M
F/M
F/M
F/M
F/M
F/M
F/M
F/M
F/M
F/M
F/M
F/M
F/M
F/M
F/M
F/M
F/M
F/M
M
M
M
M
M
M
M
F/M
F/M
M
F
M
M
F
M
M
M
NS
M
F
M
F/M
F/M
Age
68-89
60-69
60-69
70-79
70-79
80-89
80-89
42-72
42-72
65 ±8
29-70
29-70
60-69
70-79
80-89
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
65.4
73.3
82
60-69
63 ±2
70-79
80-89
27-67
>66
60
60
60
68
64
63
63
65
90-97
62
61
71
42-72
42-72
Values
4.86
3.80
4.34
3.89
4.37
3.87
4.19
3.5
3.09
2.8
3.18
4.03
2.59
2.56
2.60
3.49
3.47
3.44
3.42
3.39
3.37
3.35
3.32
3.30
3.28
3.26
3.24
3.21
3.19
3.17
3.15
3.12
3.10
3.08
3.05
2.58
2.54
2.36
2.83
2.43
2.86
2.39
1.5-3.7
1.75
3.3
2.4
3.3
1.6
1.96
1.55
2.40
1.85
2.72
4.90
6.45
3.72
3.5
2.5
Units
L/min
L/min
L/min
L/min
L/min
L/min
L/min
L/min/sq.m
L/min/sq.m
L/min/sq.m
L/min/sq.m
L/min/sq.m
L/min/sq.m
L/min/sq.m
L/min/sq.m
L/min/sq.m
L/min/sq.m
L/min/sq.m
L/min/sq.m
L/min/sq.m
L/min/sq.m
L/min/sq.m
L/min/sq.m
L/min/sq.m
L/min/sq.m
L/min/sq.m
L/min/sq.m
L/min/sq.m
L/min/sq.m
L/min/sq.m
L/min/sq.m
L/min/sq.m
L/min/sq.m
L/min/sq.m
L/min/sq.m
L/min/sq.m
L/min/sq.m
L/min/sq.m
L/min/sq.m
L/min/sq.m
L/min/sq.m
L/min/sq.m
ml/sq.m
L/min/sq.m
L/min/sq.m
L/min/sq.m
L/min/sq.m
L/min/sq.m
ml/min/sq.m
ml/min/sq.m
L/min/sq.m
L/min/sq.m
L/min/sq.m
L/min
L/min
L/min
L
L
Var Type BW Units Var Type n
0.8 SD 66.4 kg 9.7 SD 18
71
39
54
38
8
21
0.7 SD 22
0.5 SD 9
0.3 SD 57 kg 15 SD 11
0.2 SE 19
0.2 SE 25
0.6 142 Ib 12
0.3 142 Ib 5
0.2 121 Ib 7
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
0.2 SD 10
0.2 SD 9
0.2 SD 7
0.7 160 Ib 25
0.1 SE 79.8 kg 2.3 SE 14
0.4 165 Ib 15
0.3 148 Ib 13
Range 20
0.5 SD 41
1
1
1
1
1
1
1
1
0.6 9
70.5 kg 1
63.7 kg 1
58.8 kg 1
0.6 SD 22
0.4 SD 9
(Continued)
Previous
TOC
-------
PHYSIOLOGICAL PARAMETERS IN THE ELDERLY
TABLES. (Continued)
Parameter Health Gender Age Values Units Var Type BW Units Var Type n
Vital capacity'
Vital capacity'
Vital capacity6
Vital capacity6
Vital capacity6
COPD
COPD
NS
NS
NS
M
M
M
M
M
67.7 ±6.8
71 .5 ±8.8
60-69
70-79
80-89
2.81
2.15
4.18
3.87
3.38
L
L
L
L
L
0.8
0.7
0.8
0.7
0.7
SD
SD
SD
SD
SD
55.3
43.8
kg
kg
6.4 SD 13
6.2 SD 11
71
55
8
Note. For clarity, only select fields are shown in this table (and those that follow). More detailed queries and output can be generated
by users after downloading the database. NS, not specified; SD, standard deviation; SE, standard error; Var, Variability; sq.m, square
meters.
aMiller and Tenney et al. (1956).
6Norrisetal. (1964).
cNorris et al. (1964); Shock and Yiengst (1955).
dSeibold etal. (1988).
eCapderouetal. (2000).
fjermendy etal. (1986).
sLuisada etal. (1980).
hKatorietal. (1979).
'Brandfonbrener et al. (1955).
'Dinenno etal. (2001).
^Creanetal. (1986).
'Cody etal. (1988).
mBenchimol etal. (1968).
"Lauson etal. (1944).
°Okazaki et al. (1996).
^Bolomeyetal. (1948).
'Nishimuraetal. (1995).
among all organ weights covered in the database; however, there is no clear trend of gross coeffi-
cient of variation (CV) values with age. Given that neurotoxicity is of particular concern in the eld-
erly (Ginsberg et al., 2005; Brown et al., 2005; Geller & Zenick, 2005), it may be important to
include this organ specifically in PBPK models for older adults. Figure 4B shows that the age-related
decline in brain weights in elderly is identical between Japanese and U.S. males.
Heart Weight Twenty-two records in the database relate to heart weight in healthy older adults.
With the exception of 2 records that contain data collected in mixed groups of adults (>60 yr) and
elderly, all other records contain values of heart weight from studies that investigated elderly adults
aged 65 to over 100 yr (Galloway et al., 1965; Inoue & Otsu, 1987; Olivetti et al., 1995; Puggaard
etal., 2002).
Kidney Weight The database contains 36 records on kidney weight in healthy elderly,
obtained primarily from 5 studies (Galloway et al., 1965; Tauchi etal., 1971; Inoue & Otsu 1987;
Schmitz et al., 1990; Puggaard et al., 2002). Although some records contain data collected in mixed
groups of adults over the age of 60, there are enough data to ascertain kidney weight as a function
of age in adults from 65 to 100 yr of age. Because some studies reported whole kidney weight while
others reported right and left kidney weights, database entries can be found for "kidney weight,"
"left kidney weight," and "right kidney weight."
Spleen Weight There are 16 records for spleen weight in the database, based on the studies
of Galloway et al. (1965), Inoue and Otsu (1987), and Puggaard et al. (2002). Of all the organ weights
entered into the database, the largest variability is seen for the spleen. This may be the result of
upregulation of the immune system leading to increases in functional cells to fight infections on the
one hand, and possible vulnerability to stress-hormone related decreases in organ size on the other.
Other Tissue Weights The database also contains several entries on weights of left lung, right
lung, testes, thyroid, pituitary, pancreas, prostate and adrenal glands reported for three or four sub-
groups of the elderly (average age: 65, 73, 84, or 92 yr), on the basis of 400 autopsies performed by
Previous I TOC
-------
10
C. M. THOMPSON ET AL
TABLE 4. Sample Query Data for Weight of Selected Tissues in Elderly
Tissue
Adrenal3
Adrenal3
Adrenal3
Brain6
Brain6
Brain6
Brainc
Brainc
Brainc
Brainc
Brainc
Brain3
Brain3
Brain3
Brainc
Brainc
Brainc
Brainc
Heart6
Heart6
Heart6
Hearf
Hearf
Hearf
Hearf
Hearf
Heart3
Heart3
Heart3
Heart3
Hearf
Hearf
Hearf
Hearf
Kidney6
Kidney6
Kidney6
Kidneyd
Kidneyd
Kidney6
Kidneyd
Kidneyd
Kidneyd
Kidneyd
Left kidneyc
Left kidneyc
Left kidneyc
Left kidneyc
Left kidneyc
Left kidney3
Left kidney3
Left kidney3
Left kidney3
Left kidneyc
Left kidneyc
Left kidneyc
Left kidneyc
Left lung3
Gender
F/M
F/M
F/M
F
F
F
F
F
F
F
F
F/M
F/M
F/M
M
M
M
M
F
F
F
F
F
F
F
F
F/M
F/M
F/M
F/M
M
M
M
M
F
F
F
F/M
F/M
F/M
F/M
F/M
F/M
F/M
F
F
F
F
F
F/M
F/M
F/M
F/M
M
M
M
M
F/M
Age
64.5
73.3
83.8
64.3
74
83.9
60-69
70-79
80-89
90-99
>99
64.5
73.3
83.8
60-69
70-79
80-89
90-99
64.3
74
83.9
60-69
70-79
80-89
90-99
>99
64.5
73.3
83.8
92
60-69
70-79
80-89
90-99
64.3
74
83.9
60-69
60-69
61 ±9
70-79
70-79
>81
>81
60-69
70-79
80-89
90-99
>99
64.5
73.3
83.8
92
60-69
70-79
80-89
90-99
64.5
Value
16.7
19.6
14.6
1321
1271
1236
1220.1
1158.5
1139
1092.1
1107.5
1300
1270
1198
1321.9
1294
1242.1
1165.5
344
342
337
287.9
321.9
308.4
307.6
306.3
444
463
369
390
345.9
343.3
327.1
285
252
213
200
355
249.5
141
327
224.5
294.9
183.6
119.2
112.2
100
95
93.8
177
149
145
110
144.7
145.1
123.4
96.2
575
Units
g
g
g
g
g
g
g
g
g
g
g
g
g
g
g
g
g
g
g
g
g
g
g
g
g
g
g
g
g
g
g
g
g
g
g
g
g
g
g
g/1 .73 m2
g
g
g
g
g
g
g
g
g
g
g
g
g
g
g
g
g
g
Var
126
104
112
93.3
108.5
101.2
69.9
63.4
124.4
127.1
116
77.2
76
64
64
99.9
96.7
77.1
76.7
83.2
116.3
96.3
79
42.4
47
39
40
13.67
17.7
24
10.89
9.5
18.38
7.33
40.9
34.8
29.2
27.4
31.5
38.7
106.5
33.4
18.2
Type
SD
SD
SD
SD
SD
SD
SD
SD
SD
SD
SD
SD
SD
SD
SD
SD
SD
SD
SD
SD
SD
SD
SD
SD
SD
SD
SD
SD
SD
SD
SD
SD
SD
SD
SD
SD
SD
SD
SD
SD
SD
SD
SD
BW
153
143
145
61.7
60
56.2
39.2
38.4
35.1
35.1
34.1
153
143
145
45.1
44
41.8
39.6
61.7
60
56.2
39.2
38.4
35.1
35.1
34.1
153
143
145
135
45.1
44
41.8
39.6
61.7
60
56.2
39.2
38.4
35.1
35.1
34.1
153
143
145
135
45.1
44
41.8
39.6
153
Units
Ib
Ib
Ib
kg
kg
kg
kg
kg
kg
kg
kg
Ib
Ib
Ib
kg
kg
kg
kg
kg
kg
kg
kg
kg
kg
kg
kg
Ib
Ib
Ib
Ib
kg
kg
kg
kg
kg
kg
kg
kg
kg
kg
kg
kg
Ib
Ib
Ib
Ibs
kg
kg
kg
kg
Ib
Var
13.3
11.8
11.6
8.6
8.8
7.3
7.3
5.9
9.4
8.8
7.9
4.4
13.3
11.8
11.6
8.6
8.8
7.3
7.3
5.9
9.4
8.8
7.9
4.4
13.3
11.8
11.6
8.6
8.8
7.3
7.3
5.9
9.4
8.8
7.9
4.4
Type
SD
SD
SD
SD
SD
SD
SD
SD
SD
SD
SD
SD
SD
SD
SD
SD
SD
SD
SD
SD
SD
SD
SD
SD
SD
SD
SD
SD
SD
SD
SD
SD
SD
SD
SD
SD
n
400
61
48
19
70
194
172
43
4
400
99
225
150
11
61
48
19
73
196
174
44
4
400
103
225
150
11
61
48
19
35
19
14
32
11
22
14
73
196
172
44
4
400
102
225
150
11
400
(Continued)
Previous
TOC
-------
PHYSIOLOGICAL PARAMETERS IN THE ELDERLY
11
TABLE 4. (Continued)
Tissue
Left lung3
Left lung3
Left lung3
Liver6
Liver6
Liver6
Liver0
Liverf
Liver
Liverf
Liver0
Liver0
Liver0
Liver0
Liver3
Liver3
Liver3
Liver3
Liver0
Liver6
Liver6
Liver6
Liver6
Liverf
Liverf
Liverf
Liver0
Liver6
Liver0
Liver6
Liver0
Pancreas3
Pancreas3
Pancreas3
Pancreas3
Pituitary3
Pituitary3
Pituitary3
Prostate3
Prostate3
Prostate3
Right kidneyc
Right kidney0
Right kidney0
Right kidney0
Right kidney0
Right kidney3
Right kidney3
Right kidney3
Right kidney3
Right kidney0
Right kidney0
Right kidney0
Right kidney0
Right lung3
Right lung3
Right lung3
Right lung3
Gender
F/M
F/M
F/M
F
F
F
F
F
F
F
F
F
F
F
F/M
F/M
F/M
F/M
M
M
M
M
M
M
M
M
M
M
M
M
M
F/M
F/M
F/M
F/M
F/M
F/M
F/M
F/M
F/M
F/M
F
F
F
F
F
F/M
F/M
F/M
F/M
M
M
M
M
F/M
F/M
F/M
F/M
Age
73.3
83.8
92
64.3
74
83.9
60-69
61-70
61-70
61-70
70-79
80-89
90-99
>99
64.5
73.3
83.8
92
60-69
60-69
60-69
60-69
60-69
61-70
61-70
61-70
70-79
70-79
80-89
>81
90-99
64.5
73.3
83.8
92
64.5
73.3
83.8
64.5
73.3
83.8
60-69
70-79
80-89
90-99
>99
64.5
73.3
83.8
92
60-69
70-79
80-89
90-99
64.5
73.3
83.8
92
Value
539
612
690
1425
1295
1178
1006.9
1581.81
1720
1825
984
803.6
718.8
565
1569
1398
1273
1000
1137.1
1410.1
1487.8
1479.5
1004.9
1953
2042
1832.23
1064.4
886.6
879.2
771.8
833
112
105
103
110
0.72
0.69
0.7
40.6
52.9
50.4
115.6
111.6
97.4
88.3
81.3
173
146
138
105
139.9
126.5
114.6
93.7
706
640
722
690
Units
g
g
g
g
g
g
g
g
g
g
g
g
g
g
g
g
g
g
g
g
g
g
g
g
g
g
g
g
g
g
g
g
g
g
g
g
g
g
g
g
g
g
g
g
g
g
g
g
g
g
g
g
g
g
g
g
g
g
Var
305
241
289
405.1
381.1
364
264
482
260.2
200.2
102.5
389.3
53.93
46.1
54.57
32.59
377
432
452
417.5
26.4
252.1
38.53
121.2
37.5
32.8
31
22.7
22.5
48.4
36.1
32.7
14.2
Type
SD
SD
SD
SD
SD
SD
SD
SD
SD
SD
SD
SD
SD
SD
SD
SD
SD
SD
SD
SD
SD
SD
SD
SD
SD
SD
SD
SD
SD
SD
SD
SD
SD
BW
143
145
135
61.7
60
56.2
39.2
68.48
38.4
35.1
35.1
34.1
153
143
145
135
45.1
80.03
44
41.8
39.6
153
143
145
135
153
143
145
153
143
145
93.2
38.4
35.1
35.1
34.1
153
143
145
135
45.1
44
41.8
39.6
153
143
145
135
Units
Ib
Ib
Ib
kg
kg
kg
kg
kg
kg
kg
kg
kg
Ib
Ib
Ib
Ib
kg
kg
kg
kg
kg
Ib
Ib
Ib
Ib
Ib
Ib
Ib
Ib
Ib
Ib
kg
kg
kg
kg
kg
Ib
Ib
Ib
Ib
kg
kg
kg
kg
Ib
Ib
Ib
Ib
Var
13.3
11.8
11.6
8.6
15.49
8.8
7.3
7.3
5.9
9.4
17.21
8.8
7.9
4.4
8.6
8.8
7.3
7.3
5.9
9.4
8.8
7.9
4.4
Type
SD
SD
SD
SD
SD
SD
SD
SD
SD
SD
SD
SD
SD
SD
SD
SD
SD
SD
SD
SD
SD
SD
SD
n
61
48
19
72
48
16
11
196
173
44
4
400
103
214
214
214
231
34
29
101
225
231
150
231
11
400
400
400
72
196
171
43
4
400
104
225
150
11
400
(Continued)
Previous
TOC
-------
12
C. M. THOMPSON ET AL
TABLE 4. (Continued)
Tissue
Gender Age
Value
Units
Var
Type
BW
Units Var
Type n
Testes3
Testes3
Testes3
Thyroid3
Thyroid3
M
M
M
F/M
F/M
64.5
73.3
83.8
73.3
83.8
42.1
45.3
32.9
26.2
29.5
g
g
g
g
g
153
143
145
143
145
Ib
Ib
Ib
Ib
Ib
400
Note. Data are for individuals with health characterized as "Healthy" or "Unspecified" in the database.
3Callowayetal. (1965).
6Puggaard etal. (2002).
clnoueetal. (1987).
dTauchi etal. (1971).
"Schmitzetal. (1990).
thouker etal. (2004).
sSatoetal. (1970).
1800
1600-
u>
2 1400 •
O)
'5
oi 1200 •
1000 •
800
1.17;y = 2776-18.3x RA2 = 0.974 n U.S.
0.97; y = 1856 -11 .Ox RA2 = 0.947 • Japanese
O)
I
c
're
m
1400
1300-
1200-
60 70 80 90
Midpoint of Age Range (Yr)
100
1100
0.394 y = 1673-5.21x RA2 = 0.958 • Japanese
0.392 y = 1638-5.10x RA2 = 0.947 n U.S.
60 70 80 90
Midpoint of Age Range (Yr)
100
FIGURE 4. Age-related declines in male organ weights. Comparison of age-dependent declines in liver (A) and brain (B) weight. Data are
from U.S. (Galloway et al., 1965) and Japanese (Inoue & Otsu, 1987) autopsies. Note that ethnicity was not reported in the Galloway
etal. (1965) study; thus, these data may not be representative of a mixed U.S. population.
Galloway et al. (1965). For each tissue, there are three or four records in the database—each corre-
sponding to one age group as reported in the original study. For several of these tissues, there is no
significant age-related change in absolute weight. Data on blood volume obtained from Lauson
etal. (1944), Fulop et al. (1985), Jermendy et al. (1986), Kenney and Zappe (1994), and Kenney
and Ho (1995) are also included in the database.
Tissue Blood Flows
Liver (Hepatic and Splanchnic Blood Flow) Blood flow to liver consists of hepatic arterial
flow and portal venous flow. The portal flow represents the blood collected from the alimentary
organs such as stomach, spleen, pancreas and intestine. Total hepatic blood flow and portal
blood flow measurements in older adults were reported in Dencker et al. (1972); and Zoli et al.
(1989, 1999). The records based on Dencker et al. (1972) contain subject-specific data for 5 indi-
viduals aged 61, 66, 66, 67, and 74 yr, whereas the data from Zoli et al. (1989, 1999) are presented
as age-specific group means and standard deviations. These data are presented in 20 records in the
database. Of these, 15 records contain flow rates exclusively for those 65 to 75 yr of age, whereas
Previous
TOC
-------
PHYSIOLOGICAL PARAMETERS IN THE ELDERLY 13
the 5 other records contain data collected in mixed groups of elderly and adults over the age of 56.
The hepatic blood flow specified in PBPK models frequently represents the sum of the hepatic
artery flow and the portal vein flow. Hepatic artery flow rises to about 25% of total hepatic flow
after age 45, and then remains relatively stable thereafter (Zoli et al., 1999). However, the portal
vein flow decreases significantly, resulting in a 29% reduction of total hepatic flow in adults 75 yr of
age and older as compared to those 45 yr of age and younger (Zoli et al., 1989, 1999).
Splanchnic blood flow refers to the blood flow to the abdominal organs such as liver, spleen,
stomach, pancreas and intestine. However, when the splanchnic blood flow is estimated on the
basis of clearance of substances that are extracted primarily by liver (e.g., indocyanine green and
bromsulfalein), the resulting data essentially reflect liver blood flow (Bender, 1965; Williams & Leg-
gett, 1989). Eleven records on splanchnic blood flow from three studies in older adults have been
included in the current database (Sherlock et al. 1950; Kenney & Ho 1995; Ho etal. 1997). Tissue
blood flows for liver and other organs/compartments are shown in Table 5.
Cerebral Blood Flow Blood flow to the brain in older adults has been measured more fre-
quently than the blood flow to other tissues. Even though the data are generally suggestive of a
decline in absolute blood flow to the brain with advancing age (Williams & Leggett, 1989), several
of the earlier studies reported brain blood flow without additional information such as body weight,
BSA, BMI, cardiac output, brain weight, gender, race, or health status. Thus, the older data on brain
blood flow were included in the database only when, at the very least, the age and health status of
individuals were known. In total, 39 records on cerebral blood flow in elderly were included
(Schieve & Wilson, 1953; Fazekas et al., 1952, 1953, 1960; Gordan, 1956; Scheinberg et al.,
1950; Shenkin et al., 1953; Meltzer et al., 2000; Slosman et al., 2001; Kamper et al., 2004). Of
these records, 34 relate to data collected in individuals aged 65 and greater only. Four records
relate to data on the blood flow to the grey and white matter, reported separately (Frackowiak
etal., 1980). The mean values for the gray and white matter can be combined to calculate the
mean of cerebral blood flow, on the basis of volume ratio of these components (3:2) (Frackowiak
etal., 1980).
Renal Blood Flow Renal blood flow refers to the volume of blood perfusing the excretory
tissue (kidney) per unit time. It is understood that renal blood flow decreases with increasing age,
and the age-related decrease in perfusion holds even after correcting for the kidney volume
(Hollenberg et al., 1974). There are 40 data records on renal blood flow in the database, reflecting
data from 5 studies (Lauson et al., 1944; Bolomey et al., 1949; Davies & Shock, 1950; Kenney &
Ho, 1995; Ho et al., 1997). Most of the records correspond to individual data obtained in people
between 61 and 89 yr of age. Additionally, the data on the clearance of p-aminohippurate (PAH) in
individuals from 60 to 88 yr of age were included (McDonald et al., 1951; Miller et al., 1951;
Lindeman et al., 1966; Slack & Wilson, 1976). These data can be used to calculate renal plasma
flow, and converted to renal blood flow on the basis of hematocrit levels (Williams & Leggett,
1989).
Muscle and Skin Blood Flows There are 17 records in the database relating to muscle
and skin blood flow in older adults. Because the data on skin blood flow in the literature often
are measured/reported as forearm or leg blood flow, not all records will be found under "skin
blood flow." As described by Williams and Leggett (1989), the blood flow to forearm, calf,
and foot may be used to estimate the skin blood flow rate for the entire body. It should be
noted that several of the studies included in the database reported muscle blood flow for a
mixed group of younger adults and elderly (e.g., 7 records cover 59 to 79 yr old), but the
results were reported on the basis of milligrams per minute per 100 g (or 100 me) tissue. All
values on muscle blood flow rate included in the database represent means and variability for
a group, since none of the studies reported subject-specific muscle blood flow rates (Hellon &
Clarke, 1959; Lassen et al., 1964; Amery et al., 1969; Dwyer and Howe, 1995; Kenney & Ho,
1995; Ho et al., 1997; Dinenno et al., 2001). Several of the studies just mentioned as well as
other reports not included in the database investigated leg and forearm blood flow rates in
older adults during exercise; however, this is not an aspect specifically covered in the data-
base at this point.
Previous I TOC
-------
14
C. M. THOMPSON ET AL
TABLE 5. Sample Query for Data on Blood Flow to Tissues in the Elderly
Tissue
Adipose3
Adipose3
Adipose3
Adipose3
Adipose3
Adipose3
Adipose3
Calf6
Cerebral0
Cerebral
Cerebral6
Cerebral6
Cerebral6
Cerebral6
Cerebral0
Cerebral'
Cerebral'
Cerebral6
Cerebral6
Cerebral6
Cerebral6
Cerebral6
Cerebral6
Cerebral6
Cerebral6
Cerebral6
Cerebral6
Cerebral6
Cerebral6
Cerebral6
Cerebral11
Cerebral'
Cerebral'
Cerebral'
Cerebral'
Cerebral'
Cerebral'
Cerebral'
Cerebral'
Cerebral'
Cerebral'
Foot6
Forearm*
Forearm*
Forearm'
Forearm"1
Forearm"
Leg0
Liverp
Liverp
Liver'
Liver'
LiverP
Liverp
Muscle'
Gender
F
F
M
M
M
M
M
M
F
F/M
M
M
M
M
M
M
M
NS
NS
NS
NS
NS
NS
NS
NS
NS
NS
NS
NS
NS
NS
NS
NS
NS
NS
NS
NS
NS
NS
NS
NS
M
F/M
F/M
M
M
M
M
F
F
F/M
F/M
M
M
M
Age
66
77
64
66
70
74
74
70-82
50-71
69.8 ±5.4
65
69
74
79
50-71
78 ± 6.6
78 ± 6.6
72
72
77
78
78
78
78
79
81
86
91
91
91
50-76
66-92
66-92
66-92
66-92
66-92
66-92
66-92
66-92
66-92
90-1 02
70-82
60-79
60-79
38-73
59-71
65 ±1
43-68
>66
>66
61-75
>76
>66
>66
38-73
Value
0.453
0.491
0.202
0.538
0.602
0.397
0.387
0.7-3.1
36.4
62
56
55
50
40
38.5
612
567
41.1
42.7
27.8
32.0
37.0
34.2
34.7
57.9
34.7
30.6
36.1
42.4
44.1
55
17.2
33.7
66.4
38.8
32.5
18.7
24.6
44.9
40.0
39.3
1 .0-6.7
4.4
4.1
4.01
1.38
4.13
0.34
0.95
869
1297
1020
1211
1.07
2.8
Units
L/min
L/min
L/min
L/min
L/min
L/min
L/min
ml/100g/min
ml/min/100g
ml/min/100 g
ml/100g/min
ml/100g/min
ml/100g/min
ml/100g/min
ml/min/100g
ml/min
ml/min
ml/min/100 g
ml/min/100g
ml/min/100g
ml/min/100g
ml/min/100 g
ml/min/100 g
ml/min/100 g
ml/min/100g
ml/min/100g
ml/min/100g
ml/min/100 g
ml/min/100 g
ml/min/100 g
ml/min/100g
ml/100g/min
ml/100g/min
ml/100g/min
ml/100g/min
ml/100g/min
ml/100g/min
ml/100g/min
ml/100g/min
ml/100g/min
ml/100g/min
ml/100g/min
ml/min/100g
ml/min/100g
ml/100g/min
ml/min/100 g
ml/min
L/min
ml/min/g tissue
ml/min
ml/min
ml/min
ml/min
ml/min/g tissue
ml/100g/min
Var
6.4
4.4
5.3
34
21
9
1.9
1.8
0.35
0.11
0.61
0.04
0.04
62
253
148
66
0.04
0.48
Type
SD
SD
SD
SE
SE
SD
SD
SD
SE
SE
SE
SE
SE
SE
SD
SD
SE
SE
SE
BW Units Var Type n
62.8 kg 1
74.7 kg 1
57.3 kg 1
78.9 kg 1
78.1 kg 1
56.9 kg 1
77.1 kg 1
20
3
9
1
1
1
1
18
7
7
1
1
1
1
1
1
1
1
1
1
1
1
1
7
1
1
1
1
1
1
1
1
1
22
20
70 kg 5 SD 10
71 kg 7 SD 11
25
74.1 kg 5.2 SE 6
89 kg 2 SE 4
10
11
11
10
10
11
11
25
(Continued)
Previous
TOC
-------
PHYSIOLOGICAL PARAMETERS IN THE ELDERLY
15
TABLES. (Continued)
Tissue
Muscler
Muscles'
PAH clearance'
PAH clearance"
PAH clearance'
PAH clearance"
Portal""
Portal""
Portal""
Portal""
Renalx
Renalx
Renalx
Renal^
Renalx
Renalx
Renalx
Renalx
Renalx
Renalx
Renalx
Renalx
Renalx
Renalx
Renalx
Renalx
Renalx
Renalx
Renalx
Renalx
Renalx
Renalx
Renalx
Renalx
Renalx
Renalx
Renalx
Renalx
Renalx
Renalx
Renalx
Renalx
Renal"1
Renalx
Renal"
Renalx
Renalx
Skin'
Skinz
Splanchnic33
Splanchnic33
Splanchnic33
Splanchnic33
Splanchnic33
Splanchnic33
Gender
M
NS
F/M
F/M
F/M
M
M
M
M
M
M
M
M
M
M
M
M
M
M
M
M
M
M
M
M
M
M
M
M
M
M
M
M
M
M
M
M
M
M
M
M
M
M
M
M
M
M
M
M
M
M
M
M
M
M
Age
50-75
>51
60-69
>61
70-79
77-88
56-70
56-70
>72
>72
61
62
62
62
64
65
66
66
67
68
69
70
71
71
71
72
72
74
77
78
80
80
80
80
81
83
85
86
86
87
87
89
59-71
61-68
65 ±1
70-78
80-89
38-73
66-82
60
60
62
63
64
65
Value
3.15
1.96
439
431
386
283
656
394
595
361
806
941.4
853
903
733.6
931.2
673.6
837
453.2
845.8
672
411
728.7
425.9
603.6
568
844.6
474.4
627.3
587.8
460
284.1
731.5
680.8
360
498.6
411.1
506
570.6
412.2
237
553.8
1004
774.7
895
589
475.4
25.6
2.99
765
660
890
500
920
1020
Units Var Type
ml/min/1 00 g 0.55 SD
ml/100g/min 0.56 SD
ml/min/1 .73 sq.m
ml/min/1 .73 sq.m
ml/min/1 .73 sq.m
ml/min 27 SE
ml/min 175 SD
ml/min/sq.m 101 SD
ml/min 106 SD
ml/min/sq.m 53 SD
ml/min/1 .73 sq.m
ml/min/1 .73 sq.m
ml/min/1 .73 sq.m
ml/min
ml/min/1 .73 sq.m
ml/min/1 .73 sq.m
ml/min/1 .73 sq.m
ml/min/1 .73 sq.m
ml/min/1 .73 sq.m
ml/min/1 .73 sq.m
ml/min/1 .73 sq.m
ml/min/1 .73 sq.m
ml/min/1 .73 sq.m
ml/min/1 .73 sq.m
ml/min/1 .73 sq.m
ml/min/1 .73 sq.m
ml/min/1 .73 sq.m
ml/min/1 .73 sq.m
ml/min/1 .73 sq.m
ml/min/1 .73 sq.m
ml/min/1 .73 sq.m
ml/min/1 .73 sq.m
ml/min/1 .73 sq.m
ml/min/1 .73 sq.m
ml/min/1 .73 sq.m
ml/min/1 .73 sq.m
ml/min/1 .73 sq.m
ml/min/1 .73 sq.m
ml/min/1 .73 sq.m
ml/min/1 .73 sq.m
ml/min/1 .73 sq.m
ml/min/1 .73 sq.m
ml/min 71 SE
ml/min/1 .73 sq.m 139 SD
ml/min 69 SE
ml/min/1 .73 sq.m 133 SD
ml/min/1 .73 sq.m 141 SD
ml/100g/min 5.05 SE
ml/100g/min 0.46 SD
ml/min/sq.m
ml/min/sq.m
ml/min/sq.m
ml/min/sq.m
ml/min/sq.m
ml/min/sq.m
BW Units Var Type n
10
42
5
4
3
61 kg 7
60
60
60
60
1
1
1
70.5 kg 1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
74.1 kg 5.2 SE 6
10
89 kg 2 SE 4
9
12
25
74.9 kg 5.3 SD 7
60 kg 1
49 kg 1
55 kg 1
57 kg 1
64 kg 1
48 kg 1
(Continued)
Previous
TOC
-------
16
C. M. THOMPSON ET AL
TABLES. (Continued)
Tissue
Gender Age
Value
Units
Var Type
BW
Units Var Type n
Splanchnic33
Splanchnic33
Splanchnic33
Splanchnic"1
Splanchnic"
M
M
M
M
M
67
68
75
59-71
65 ±1
795
723
578
1302.00
1050
ml/min/sq.m
ml/min/sq.m
ml/min/sq.m
ml/min
ml/min
109
158
SE
SE
48
46
50
74.1
89
kg
kg
kg
kg
kg
5.2
2
SE
SE
1
1
1
6
4
Note. Data are for individuals with health characterized as "Healthy" or "Unspecified" in the database. NS, not specified; SD,
standard deviation; SE, standard error; sq.m, square meters.
3Lesseretal. (1967).
'Allwood etal. (1958).
cSlosmanetal. (2001).
dMeltzeretal. (2000).
eScheinbergetal. (1950).
fKamperetal. (2004).
gFazekas et al. (1952).
Achieve et al. (1953).
'Fazekasetal. (1960).
'Fazekasetal. (1953).
^Dwyeretal. (1995).
'Mellon etal. (1959).
"Tenneyand Ho (1995).
"Ho etal. (1997).
"Thomson etal. (1988).
^Wynneetal. (1989).
'Zolietal. (1999).
rAmery etal. (1969).
sLassenetal. (1964).
'Miller etal. (1951).
"Slack etal. (1976).
'Lindeman et al. (1966).
"Zoli etal. (1989).
xDaviesand Shock (1950).
xBolomey et al. (1948).
zRookeetal. (1994).
33Sherlocketal. (1950).
Adipose Tissue Blood Flow Published reports on blood flow to adipose tissues are scant.
During the literature review, only one study on the measurement of blood flow to adipose tissue
was identified (Lesser & Deutsch, 1967). In this study, blood flow to adipose tissue was reported for
7 healthy individuals (5 males and 2 females) aged 64 to 74 yr. These individual data were included
in the database along with subject-specific body weight and adipose tissue volume.
Blood Flow to Other Organs The literature search did not yield any data on blood flow to
other organs in subjects aged 65 yr or older.
Metabolic Clearance
The data on metabolic clearance included in the database relate to the activities (measured
with a probe substrate) and levels of enzymes responsible for phase I biotransformation, as well as
the levels of cofactors and activities of enzymes responsible for various phase II conjugation reac-
tions. It should be noted that several seminal studies on drug metabolism in older adults reported
results either in terms of percent activity relative to controls or to some specific age group (usually
young adults). Further, many of these studies report data in graphical form, which were not
included in this current iteration of the database. For many environmental chemicals, the lack of
such data is not crucial since the decline in metabolic activity in elderly individuals might be more a
Previous
TOC
-------
PHYSIOLOGICAL PARAMETERS IN THE ELDERLY 1 7
consequence of the decrease in blood flow to liver and reduction in hepatic mass than a change in
protein concentration or enzyme activity. A compilation and analysis of age-related differences for
in vivo metabolism for a large number of individual drugs is reported by Ginsberg et al. (2005) and
made accessible at http://www2.clarku.edu/faculty/dhattis.
Phase I Enzymes
Schmucker et al. (1990) analyzed liver samples from individuals aged 9 to 89 yr (= 54) for
the levels of microsomal protein, cytochrome P-450 (CYP), epoxide hydrolase, and NADPH
cytochrome c reductase. They did not find any change with age in enzyme activity (assessed with
13 substrates and 9 procarcinogens), enzyme content, or cytochrome P-450 reductase levels
expressed on volume basis (Schmucker et al., 1990). Shimada et al. (1994) analyzed liver samples
from 60 individuals aged 12 to 73 yr (30 Japanese and 30 Caucasian patients) for the content
and activities related to CYP 1A2, 2A6, 2B6, 2C, 2D6, 2E1, and 3A. These authors and others
(Woodhouse et al., 1984; Wynne et al., 1988; Hunt et al., 1990) could not detect any apparent
age-related changes in CYP content or activity. Furthermore, Wynne et al. (1988) reported that
there was no correlation between age and apparent affinity (KJ of either high- or low-affinity com-
ponents of 7-ethoxycoumarin O-deethylase. Blanco et al. (2000) did not find any significant age-
related differences in the oxidation of substrates reflective of the activities of various isoforms of
CYP in 37 samples (age: 6 mo to 93 yr): ethoxyresorufin (CYP1A2), ethoxycoumarin (CYP2E1), teni-
poside and midazolam (CYP3A4/5), pacilitaxel (CYP2C8), and tolbutamide (CYP2C9).
In contrast to the studies just mentioned, some studies reported age-dependent changes in
phase I enzymes. Sotaniemi et al. (1997) reported a 29% reduction in antipyrine clearance and
33% reduction in liver P-450 content in subjects >70 yr of age relative to individuals 20-29 yr of
age. Parkinson et al. (2004) reported that CYP activities varied widely in liver microsomes from
donors of all ages and that there were only few statistically significant differences. Based on few
samples per age group, these authors reported that activities associated with CYP1A2, CYP2B6,
CYP2C19, CYP2D6, and CYP2E1 appeared to decrease with age, whereas CYP2A6 and CYP4A11
activities appeared to increase with age. An in vivo study by Bebia et al. (2004) reported some
trends consistent with Parkinson et al. (2004) (decreased CYP2C19 activity), but also reported
contrasting data (increased CYP2E1 activity and no change in CYP2D6). The changes reported in
these studies appear to be relatively small compared to the extent of interindividual differences and
consequence of decreased liver weight in elderly.
Data on alcohol dehydrogenase class I (ADHI) activity included in this database were obtained
form Seitz et al. (1993), who reported that elderly men had lower gastric ADH activity compared to
young men. Elderly women, however, exhibited ADH activity levels comparable to younger
women.
Phase II Enzymes The limited available data suggest that there is no age-related difference in
phase II metabolism of xenobiotics in older adults (e.g., acetylation of isoniazid and glucuronidation
of temazepam) (Farah et al., 1977). Data indicate no significant difference in the number of fast
acetylators among the young (< 35 yr) and elderly (>65 yr), on the basis of the half-lives of isoniazid
(Farah et al., 1977). Further, Woodhouse et al. (1984) reported that hepatic glutathione concentra-
tions are not affected by aging. Limited data on extrahepatic phase II metabolism (e.g., sulfotrans-
ferase activity in kidney and lungs) included in this database do not indicate any significant change
or loss of the activity in elderly compared to younger adults (Pacific! et al., 1988).
Renal Clearance Kidney function in general is known to be impaired by aging Gassal &
Oreopoulos, 1998). A decrease in glomerular filtration rate (GFR), as reflected by creatinine clear-
ance (CLcr), was documented in a number of studies in the elderly. More recent analyses appear
to suggest that the decrease in GFR in elderly, as reflected by CLcr, may be due to increased
incidence of renal disease rather than a true age-related effect (McLean & LeCouteur, 2004).
Further, the estimates of CLcr have been questioned on the basis of the fact that the rate of produc-
tion of this endogenous substrate is highly variable (e.g., decrease in serum creatinine due to
decrease in muscle mass) and the possibility of tubular secretion of creatinine. Most reliable
estimates of GFR have been made with the urinary clearance of inulin, a starch-like polymer of
Previous I TOC
-------
18 C. M. THOMPSON ET AL
fructose. GFR estimates were also obtained using ethylenediamine tetraacetic acid (EDTA) and the
results compare well with CLcr (Groth et al., 1989). The database contains 91 records relating to
measurements of GFR, based on clearance of creatinine, inulin, and EDTA in the elderly. The
subjects in these studies were not on medication known to affect renal function.
Body, Blood, and Tissue Composition The data under this category specifically refer to the
neutral lipids (triglycerides, diglycerides, monoglycerides, cholesterol, and other nonpolar lipids),
phospholipids (phosphatidylcholine, phosphatidylethanolamine, phosphatidylserine, sphingomyelin,
and other lipids containing phosphoric acid esterified at the 1-position of the glycerol molecule),
proteins (albumin, hemoglobin, gamma-globulin, alpha-1 acid glycoprotein), and water content in
blood and tissues. These data are potentially useful in estimating the volume of distribution and
partition coefficients essential for development of PBPK models for organic chemicals. In general,
the observations included in the database indicate that serum cholesterol, high-density lipoproteins,
and triglycerides tend to increase with age but decrease in individuals over 90 yr of age (Tietz et al.,
1992).
In regard to serum albumin level, a decline with older age groups was observed (Greenblatt,
1979). Data on albumin levels entered into the database are consistent with the pharmaceutical
literature, which suggests that the reduced binding of a number of drugs in elderly (compared to
younger subjects) is not a consequence of an age-related change in binding affinity, but rather due
to decreased serum albumin levels leading to a reduction of the total number of binding sites.
Body water content is generally considered as a measure of fat-free mass. For all ages, percent
body water is less in women than in men, and it declines with increasing age, coinciding with the
decline in muscle mass in both genders (Sitar, 1998). The database contains several entries for body
water content as well as extracellular and intracellular water content in older adults. Further, data
on the composition of adipose tissue were included in the database; however, data on the compo-
sition of other tissues were not obtained during the literature review. Indications are that hepatic
triglyceride and cholesterol levels rise with aging while phospholipid content remains unchanged
(Schneeman & Richter, 1993).
PHYSIOLOGICAL DATA IN HEALTH-IMPAIRED ELDERLY
It is well established that health conditions affect the disposition of pharmaceuticals and
environmental toxicants (Timbrell, 2000). There is little doubt that fat mass is increased in obese
individuals, or that alveolar ventilation rate is altered in those with chronic obstructive pulmonary
disease (COPD). Overall, there are only 115 data records in the database that relate to physiological
parameters in health-impaired older adults. Some of these data were collected in mixed groups of
middle-aged and older adults. Additional data supporting the alteration of physiological parameters
in specific disease conditions are available in the literature but relate to young adults and therefore
are not relevant to the present database effort. Reasons for including certain disorders and disease
conditions are briefly summarized in Table 1.
Obesity
Obesity is characterized by excessive adipose tissue deposition, which is primarily the result of
increased caloric intake above the maintenance energy requirements of the body (Crandall &
DiGirolamo, 1990). Obesity is characterized as a condition of volume expansion, accompanied
by increased blood volume and elevated cardiac output (Alexander et al., 1962-1963). According
to the Centers for Disease Control and Prevention, individuals with a BMI exceeding 30 are
considered obese (http://www.cdc.gov/nccdphp/dnpa/obesity/defining.htm); alternatively, those
who are at or above 120% of their ideal body weight are considered obese (Beers, 2005).
In extreme obesity, adipose tissue blood flow can represent up to one half of the entire cardiac
output, emphasizing the importance of expanding adipose tissue mass on the hemodynamic
changes (Alexander et al., 1962-1963). Lesser and Deutsch (1967) concluded that adipose tissue
perfusion rose with increasing degrees of obesity. Kjellbergand Reizenstein (1970) reported that the
blood volume increases with body weight in obese individuals (38 to 42 yr old). Data relating to all
Previous I TOC
-------
PHYSIOLOGICAL PARAMETERS IN THE ELDERLY 19
of the above parameters were included in the database. Further, Lucas et al. (1998) and Wang et al.
(2003), using chlorzoxazone as a substrate, found that hepatic CYP2E1 activity increased in obese
type 2 diabetes; similarly, O'Shea et al. (1994) reported enhanced clearance of chlorzoxazone in
obese individuals aged 24 to 48 yr. None of these data were obtained specifically in obese elderly,
and therefore were not included in the database. There are also data that suggest increases in
glucuronidation and sulfation in obese individuals with no changes in other phase II reactions
(glycine conjugation and acetylation) in comparison with lean individuals (Blouin & Warren, 1999).
Data on GFR in obese individuals are mixed. Some studies indicate no change, whereas others
indicate an increase or decrease in obese women in GFR. The discrepancy among the various stud-
ies might be due to the difference in the extent of obesity and/or associated renal pathology (Blouin
& Warren, 1999) in the samples of people studied. In all, the database contains 31 records on phys-
iological parameters in obese older adults.
Diabetes
Non-insulin-dependent diabetes mellitus (NIDDM or type II diabetes) is characterized by
persistent hyperglycemia but rarely leads to ketoacidosis. Type II diabetes generally manifests
after age 40 and is often a result of genetic defects causing both insulin resistance and insulin
deficiency. There is also a strong correlation between obesity and onset of type II diabetes
(http://web.indstate.edu/theme/mwking/diabetes.html). Although reduced GFR is associated with
diabetes (Knobler et al., 2004), little is known about possible alterations in other physiological
parameters. The database contains a total of 25 records on GFR, cardiac output, fat mass, and
serum triglyceride levels in elderly with type II diabetes. Data on hepatic CYP2E1 activity and
CYP2E1 mRNA in obese type II diabetics were also included (Wang et al., 2003). It should be
noted that in these studies the treatment history is not always known or reported, even though the
diagnostic information is routinely provided.
Chronic Obstructive Pulmonary Disease (COPD)
COPD, characterized by obstruction of airflow, is projected to be the third leading cause of
mortality and fifth leading cause of disability worldwide by the year 2020 (Sandstrom et al., 2003).
In patients with COPD, decreased airway diameter limits the airflow rates (Mahler et al., 1986). The
ventilation-perfusion relationships are also altered in COPD patients. Mahler et al. (1985) showed
that the relationship between cardiac index and pulmonary arterial pressure was increased in
COPD patients compared to normal individuals. The literature search yielded 8 articles (30 records)
for the alteration of cardiac output, alveolar ventilation rate, and physiological dead space in aged
COPD patients. The search also identified two studies on the fat mass and BMI in COPD patients,
which were entered into the database.
Heart, Kidney, and Liver Disease
Limited data were retrieved from the literature on values of cardiac output, cerebral blood flow,
leg blood flow, and GFR in older adults with heart disease. In total, 14 records on the elderly with
heart disease were included in the database. In regards to kidney disease, the only data found dur-
ing the literature search relate to GFR in patients with renal disease; thus, there are no data entries
for renal function in individuals with other health conditions. There are 10 records in the database
from individuals characterized as having liver disease or acute liver failure.
DATA GAPS AND CONCLUDING REMARKS
The data availability for the various physiological parameters in elderly are summarized in
Figure 5. A lack of data is particularly marked in those over the age of 85. However, the overall data
gap across the age groups in healthy older adults relates to tissue composition data (neutral lipid,
phospholipids, and protein levels), which are necessary for estimating volume of distribution and
partition coefficients for PBPK models. The blood composition data found in the database are also
very limited and need to be expanded. There is some information, albeit limited, on the age-specific
Previous I TOC
-------
20
C. M. THOMPSON ET AL
.
CD ra
Tissue Volume Tissue Blood Flow
I 1
\
Card
Renal
M
Tis
Brain
Liver
Kidney
SPT
Fat
Other
Brain
Liver
Kidney
SPT
Blood
^— Other
/entilation
ac Output
Clearance
etabolism
sue Comp
r i i i i i
1 1 1 1 1 I
S I I 1 I I
r3 | | | | |
1 * >! ''•
r* III
^ ] 1 \
m=^ ' i I \
" " " i ! \
""""""" """"""— ' i i i
D 50-64
D 65-74
• 75-84
• >85
25
50
75
Number of Records
100
125
150
FIGURE 5. Database summary. Note that collection of data for the 50-64 yr age range was not the focus of this work; thus the record
numbers are not directly comparable to the other age ranges. SPT, slowly perfused tissue; Comp, composition.
variation between ethnic groups and genders. Most of the information in the database, however,
relates to values collected in Asian and Caucasian males. The physiological measurements in health-
impaired older adults are also highly limited and represent a major data gap; although there is some
information regarding the direction of change in physiological parameters in specific pathophysio-
logical conditions, most of these data are from middle-aged adults or those 60-65 yr of age.
There are several ways to utilize the physiological parameter values in this database for risk
assessment purposes. For instance, the data can be analyzed statistically to characterize the impact
of age, gender, health, and ethnicity on specific parameters (e.g., ventilation or CYP3A content)
for informing dosimetric adjustment in risk assessment, particularly when PBPK models are not
available for a specific chemical. In addition to such dosimetric adjustments, the database can be
useful for PBPK modeling in three ways. First, the database can be viewed as a collection of
peer-reviewed literature containing relevant data for lifestage PBPK model development. Second,
the database can be used to develop age-, gender-, health-, or ethnic-specific point estimates based
on multiple sources within the database and subsequently to use these point estimates as inputs
into deterministic PBPK models. Finally, the database can be used together with statistical analyses
to replace point estimates of PBPK model inputs with population distributions (or plausible ranges)
for probabilistic modeling (e.g., Monte Carlo simulation).
This database, in its current form, may not yet be sufficient for obtaining or deriving physiological
parameters for all elderly age groups and conditions of interest. However, it should be recognized that
the current database can aid researchers and risk assessors in the development of PBPK models, but
also that the database can be viewed as a starting point for further physiological parameter data collec-
tion and analysis.
Overall, the literature review conducted in this study resulted in the identification of 528 publi-
cations potentially useful for populating the database; of these, 155 publications contained relevant
physiological data that resulted in 1051 and 115 data records for healthy and diseased elderly,
respectively. These data provide a scientific basis for developing distributions and reference values
of several key physiological parameters for PBPK modeling in these populations.
Previous
TOC
-------
PHYSIOLOGICAL PARAMETERS IN THE ELDERLY 21
REFERENCES
Alexander, J. K., Dennis, E. W., Smith, W. G., Mad, K. H., Duncan, W. C, and Austin, R. C. 1 962-1963. Blood volume, cardiac output,
and distribution of systemic blood flow in extreme obesity. Cardiovasc. Res. Cent. Bull. 1:39-44.
Allwood, M. J. 1958. Blood flow in the foot and calf in the elderly; A comparison with that in young adults. Clin. Sci. (Lone/.) 1 7:331-338.
Amery, A., Bossaert, H., Verstraete, M., and Belgium, L. 1 969. Muscle blood flow in normal and hypertensive subjects: Influence of age,
exercise, and body position. Am. Heart]. 78:211-216.
Anthonisen, N. R., Connett, J. E., and Kiley, J. P. 1994. Effects of smoking intervention and the use of an inhaled anticholineergic
bronchodilator on the rate of decline of FEV1 .J. Am. Med.Assoc. 272:1497-1505.
Bach, B., Hansen, J. M., Kampmann, J. P., Rasmussen, S. N., and Skovsted, L. 1981. Disposition of antipyrine and phenytoin correlated
with age and liver volume in man. Clin. Pharmacokinet. 6:389-396.
Barton, H. A., Chiu, W. A., Setzer, R. W., Andersen, M. E., Bailer, A. J., Bois, F. Y., DeWoskin, R. S., Hays, S., Johanson, G., Jones, N.,
Loizou, G., MacPhail, R. C., Portier, C. J., Spendiff, M., and Tan, Y.-M. 2007. Characterizing uncertainty and variability in
physiologically based pharmacokinetic models: State of the science and needs for research and implementation. Toxicol. Sci.
99:395-402.
Bebia, Z., Buch, S. C., Wilson, J. W., Frye, R. F., Romkes, M., Cecchetti, A., Chaves-Gnecco, D., and Branch, R. A. 2004. Bioequivalence
revisited: Influence of age and sex on CYP enzymes. Clin. Pharmacol. Jher. 76:61 8-627.
Beers, M. H. 2005. The Merck manual of geriatrics, 3rd ed. Rahway, NJ: Merck & Co.
Benchimol, A., Maroko, P. R., Pedraza, A., Brener, L., and Buxbaum, A. 1968. Left ventricular end-diastolic pressure and cardiac output
at rest and during exercise in patients with angina pectoris. Cardiology 53:261-279.
Bender, A. D. 1965. The effect of increasing age on the distribution of peripheral blood flow in man.). Am. Geriatr. Soc. 13:192-198.
Blanco, J. G., Harrison, P. L., Evans, W. E., and Relling, M. V. 2000. Human cytochrome P450 maximal activities in pediatric versus adult
liver. DrugMetab. Dispos. 28:379-382.
Blouin, R. A., and Warren, G. W. 1999. Pharmacokinetic considerations in obesity.). Pharm. Sci. 88:1-7.
Bolomey, A. A., Michie, A. J., Michie, C., Breed, E. S., Schreiner, G. E., and Lauson, H. D. 1 948. Simultaneous measurement of effective
renal blood flow and cardiac output in resting normal subjects and patients with essential hypertension. J. Clin. Invest. 28:10-1 7.
Boyd, E. 1933. Normal variability in weight of the adult human liver and spleen. Arch. Pathol. 16:350-372.
Brandfonbrener, M., Landowne, M., and Shock, N. W. 1955. Changes in cardiac output with age. Circulation 12:557-566.
Brown, R. P., Delp, M. D., Lindstedt, S. L., Rhomberg, L. R., and Beliles, R. P. 1997. Physiological parameter values for physiologically
based pharmacokinetic models. Tox/co/. Ind. Health 13:407-484.
Brown, R. P., Lockwood, A. H., and Sonawane, B. R. 2005. Neurodegenerative diseases: An overview of environmental risk factors.
Environ. Health Perspect. 113:1250-1256.
Galloway, N. O., Foley, C. F., and Lagerbloom, P. 1965. Uncertainties in geriatric data. II. Organ size.). Am. Geriatr. Soc. 13:20-28.
Capderou, A., Aurengo, A., Derenne, J. P., Similowski, T., and Zelter, M. 2000. Pulmonary blood flow distribution in stage 1 chronic
obstructive pulmonary disease. Am. ]. Respir. Crit. Care Med. 162:2073-2078.
Chiu, W. A., Barton, H. A., DeWoskin, R. S., Schlosser, P., Thompson, C. M., Sonawane, B., Lipscomb, J. C., and Krishnan, K. 2007.
Evaluation of physiologically based pharmacokinetic models for use in risk assessment. ]. Appl. Tox/co/. 27:21 8-237.
Chouker, A., Martignoni, A., Dugas, M., Eisenmenger, W., Schauer, R., Kaufmann, I., Schelling, G., Lohe, F., Jauch, K. W., Peter, K., and
Thiel, M. 2004. Estimation of liver size for liver transplantation: The impact of age and gender. Liver Transplant. 10:678-685.
Clark, L. H., Setzer, R. W., and Barton, H. A. 2004. Framework for evaluation of physiologically-based pharmacokinetic models for use
in safety or risk assessment. Risk Anal. 24:1697-1 71 7.
Cody, R. J., Ljungman, S., Covit, A. B., Kubo, S. H., Sealey, J. E., Pondolfino, K., Clark, M., James, G., and Laragh, J. H. 1988. Regulation
of glomerular filtration rate in chronic congestive heart failure patients. Kidney Int. 34:361-367.
Collis, T., Devereux, R. B., Roman, M. J., de, S. G., Yeh, J., Howard, B. V., Fabsitz, R. R., and Welty, T. K. 2001. Relations of stroke
volume and cardiac output to body composition: The strong heart study. Circulation 103:820-825.
Crandall, D. L., and DiGirolamo, M. 1990. Hemodynamic and metabolic correlates in adipose tissue: pathophysiologic considerations.
FASEBJ. 4:141-147.
Crapo, R. O. 1993. The aging lung: Pulmonary disease in the elderly patients, ed. D. A. Mahler, pp. 1-25. New York: Dekker.
Crean, P. A., Pratt, T., Davies, G. J., Myers, M., Lavender, P., and Maseri, A. 1986. The fractional distribution of the cardiac output in
man using microspheres labelled with technetium 99m. Br. J. Radiol. 59:209-215.
Davies, B., and Morris, T. 1993. Physiological parameters in laboratory animals and humans. Pharm. Res. 10:1093-1095.
Davies, D., and Shock, N. 1950. Age changes in glomerular filtration rate, effective renal plasma flow, and tubular excretory capacity in
adult males.). Clin. Invest. 29:496-507.
Dencker, H., GothlinJ., Olin, T., and Tibblin, S. 1972. Portal circulation in humans studied by a dye-dilution technique. Eur. Surg. Res. 4:81-89.
Dinenno, F. A., Seals, D. R., DeSouza, C. A., and Tanaka, H. 2001. Age-related decreases in basal limb blood flow in humans: Time
course, determinants and habitual exercise effects.). Physiol. 531:573-579.
Duerenberg, P., Weststrate, J. A., and Siedell, J. C. 1 991. Body mass index as a measure of body fatness: Age and sex specific prediction
formulas. Br.]. Nutr. 65:105-114.
Dwyer, R., and Howe, J. 1995. Peripheral blood flow in the elderly during inhalational anaesthesia. Acta Anaesthesiol. Scand. 39:939-944.
Farah, F., Taylor, W., Rawlins, M. D., and James, O. 1977. Hepatic drug acetylation and oxidation: Effects of aging in man. Br. Med. ].
2:155-156.
Fazekas, J., Thomas, A.Johnson, J., and Young, W. 1960. Effect of arterenol (norepinephrine) and epinephrine in cerebral hemodynamics
and metabolism. AMA Arch. Neurol. 2:435-438.
Previous I TOC
-------
22 C. M. THOMPSON ET AL
Fazekas, J., Kleh, J., and Witkin, L. 1 953. Cerebral hemodynamics and metabolism in subjects over 90 years of age. ]. Am. Geriatr. Soc.
1:836-839.
Fazekas, J., Alman, R., and Bessman, A. 1952. Cerebral physiology of the aged. Am. J. Med. Sci. 223:245-257.
Frackowiak, R. S., Lenzi, G. L., Jones, T., and Heather, J. D. 1 980. Quantitative measurement of regional cerebral blood flow and oxygen
metabolism in man using 15O and positron emission tomography: theory, procedure, and normal values.). Comput. Assist. Tomogr.
4:727-736.
Fulop, T., Worum, I., Csongor, J., Foris, G., and Leovey, A. 1 985. Body composition in elderly people. I. Determination of body compo-
sition by multiisotope method and the elimination kinetics of these isotopes in healthy elderly subjects. Geronto/ogy 1 7:6-14.
Geller, A. M., and Zenick, H. 2005. Aging and the environment: A research framework. Environ. Health Perspect. 113:1 257-1 262.
Gentry, P. R., Haber, L. T., McDonald, T. B., Zhao, Q., Covington, T., Nance, P., Clewell, J. J. III., and Lipscomb, J. C. 2004. Data for
physiologically based pharmacokinetic modeling in neonatal animals: Physiological parameters in mice and Sprague-Dawley rats.
J. Child. Health 2:363-411.
Ginsberg, G., Hattis, D., Russ, A., and Sonawane, B. 2005. Pharmacokinetic and pharmacodynamic factors that can affect sensitivity to
neurotoxic sequelae in the elderly. Environ. Health Perspect. 113:1243-1 249.
Goldman, H. I., and Becklake, M. R. 1959. Respiratory function tests: Normal values of median altitudes and the prediction of normal
results. Am. Rev. Tuberc. 79:457-467.
Gordan, G. S. 1956. II. Hormones and metabolism. Influence of steroids on cerebral metabolism in man. Recent Prog. Horm. Res.
12:153-174.
Greenblatt, D. J. 1979. Reduced serum albumin concentration in the elderly: A report from the Boston Collaborative Drug Surveillance
Program.). Am. Geriatr. Soc. 27:20-22
Groth, S., Assted, M., and Vestergaard, B. 1989. Screening of kidney function by plasma creatinine and single-sample 51Cr-EDTA
clearance determination—A comparison. Scand. J. Clin. Lab. Invest. 49:707-710.
Hellon, R. F, and Clarke, R. S. J. 1959. Changes in forearm blood flow with age. Clin. Sci. 18:1-7.
Ho, C. W., Beard, J. L., Farrell, P. A., Minson, C. T., and Kenney, W. L. 1997. Age, fitness, and regional blood flow during exercise in the
heat.;. Appl. Physio/ 82:1126-1135.
Hollenberg, N. K., Adams, D. F., Solomon, H. S., Rashid, A., Abrams, H. L., and Merrill,]. P. 1974. Senecence and the renal vasculature
in normal man. Circ. Res. 34:309-316.
Hunt, C. M., Strater, S., and Stave, G. M. 1990. Effect of normal aging on the activity of human hepatic cytochrome P450IIE1. Biochem.
Pharmacol. 40:1666-1669.
ICRP. 1 975. Report of the Task Group on Reference Man. Publication 23. Oxford: Pergamon Press.
Inoue, T., and Otsu, S. 1987. Statistical analysis of the organ weights in 1,000 autopsy cases of Japanese aged over 60 years. Acta Pathol.
Jpn. 37:343-359.
Jassal, S. V., and Oreopoulos, D. 1998. The aging kidney. Geriatr. Nephrol. Urol. 8:141-147.
Jermendy, G., Istvanffy, M., Kammerer, L., Koltai, M., and Pogatsa, G. 1 986. Circulating blood volumes in diabetic patients. Exp. Clin.
Endocrinol. 88:123-125.
Julius, S., Amery, A., Whitlock, L., and Conway, J. 1967. Influence of age on the hemodynamic response to exercise. Circulation 36:222-230.
Kamper, A. M., Spilt, A., de Craen, A. J., van Buchem, M. A., Westendorp, R. G., and Blauw, G. J. 2004. Basal cerebral blood flow is
dependent on the nitric oxide pathway in elderly but not in young healthy men. Exp. Gerontol. 39:1245-1248.
Katori, R. 1979. Normal cardiac output in relation to age and body size. Tohoku J. Exp. Med. 128:377-387.
Kenney, W. L., and Zappe, D. H. 1 994. Effect of age on renal blood flow during exercise. Aging (Milano) 6:293-302.
Kenney, W. L., and Ho, C. W. 1995. Age alters regional distribution of blood flow during moderate-intensity exercise. ]. Appl. Physiol.
79:1112-1119.
Kjellberg, J., and Retzenstein, P. 1970. Body composition in obesity. Acta Med. Scand. 1 88:161-169.
Knobler, H., Zornitzki, T., Vered, S., Oettinger, M., Levy, R., Caspi, A., Faraggi, D., and Livschitz, S. 2004. Reduced glomerular filtration
rate in asymptomatic diabetic patients: Predictor of increased risk for cardiac events independent of albuminuria. J. Am. Coll.
Card/o/. 44:2142-2148.
Knudson, R. J. 1991. Physiology of the aging lung. In The lung, Scientific foundations, ed. R. G.Crystal, pp. 1 749-1 759. New York: Raven
Press.
Lakatta, E. G. 1990. Changes in cardiovascular function with aging. Eur. Heart]. 11(Suppl. C):22-29.
Lassen, N., Lindbjerg, J., and Munck, O. 1964. Measurements of blood-flow through skeletal muscle by intramuscular injection of
xenon-1 33. Lancet 1 5:686-689.
Lauson, H. D., Bradley, S. E., Cournand, A., and Andrews, V. V. 1944. The renal circulation in shock.). Clin. Invest. 23:381-402.
Lesser, G. T., and Deutsch, S. 1967. Measurement of adipose tissue blood flow and perfusion in man by uptake of Kr1. ]. Appl. Physiol
23:621-630.
Lindeman, R. D., Lee, T. D., Yiengst, M. J., and Shock N. W. 1966. Influence of age, renal disease, hypertension, diuretics, and calcium
on the antidiuretic responses to suboptimal infusions of vasopressin.j. Lab. Clin. Med. 68:206-223.
Lucas, D., Farez, C., Bardou, L. G., Vaisse, J., Attali, J. R., and Valensi, P. 1998. Cytochrome P450 2E1 activity in diabetic and obese
patients as assessed by chlorzoxazone hydroxylation. Fundam. Clin. Pharmacol. 12:553-558.
Luisada, A. A., Bhat, B. A., and Bioeng, V. K. 1980. Changes of cardiac output caused by aging: An impedance cardiographic study.
Angiology 31:75-81.
Mahler, D. A., Matthay, R. A., Snyder, P. E., Neff, R. K., and Loke, J. 1 985. Determination of cardiac output at rest and during exercise
by carbon dioxide rebreathing method in obstructive airway disease. Am. Rev. Respir. Dis. 131:73-78.
Mahler, D. A., Barlow, P. B., and Matthay, R. A. 1986. Chronic obstructive pulmonary disease. Clin. Geriatr. Med. 2:285-312.
Previous I TOC
-------
PHYSIOLOGICAL PARAMETERS IN THE ELDERLY 23
Marchesini, G., Bua, V., Brunori, A., Bianchi, G., Pisi, P., Fabbri, A., Zoli, M., and Pisi, E. 1988. Galactose elimination capacity and liver
volume in aging man. Hepato/ogy 8:1079-1083.
McDonald, R. K., Solomon, D. H., and Shock, N. W. 1951. Aging as a factor in the renal hemodynamic changes induced by a standardized
pyrogen. J. Clin. Invest. 30:457-462.
McLean, A. J., and Le Couteur, D. G. 2004. Aging biology and geriatric clinical pharmacology. Phannacol. Rev. 56:1 63-184.
Meltzer, C. C, Cantwell, M. N., Greer, P.J., Ben-Eliezer, D., Smith, G., Frank, G., Kaye, W. H., Houck, P. R., and Price,]. C. 2000. Does
cerebral blood flow decline in healthy aging? A PET study with partial-volume correction.). Nucl. Med. 41:1842-1848.
Miller, J. H., McDonald, R. K., and Shock, N. W. 1951. The renal extraction of p-aminohippurate in the aged individual. J. Gerontol.
3:213-216.
Miller, R. M., and Tenney, S. M. 1956. Dead space ventilation in old age. J. Appl. Physio!. 9:321-327.
NHANES III. 1997. National Health and Nutrition Examination Survey, http://www.cdc.gov/nchs/about/major/nhanes/nh3data.htm
Nishimura, Y., Tsutsumi, M., Nakata, H., Tsunenari, T., Maeda, H., and Yokoyama, M. 1995. Relationship between respiratory muscle
strength and lean body mass in men with COPD. Chest 107:1232-1 236.
Norris, A. H., Mittman, C., and Shock, N. W. 1964. Changes in ventilation with age. In Aging of the lung perspectives, eds. L. Gander and
J. H. Moyer, pp. 311-317. New York. Grune and Stratton.
Okazaki, K., Hirano, M., Suga, T., and Ohara, Y. 1996. [Cardiac output in patients in their 90s]. Masui 45:352-355.
Olivetti, G., Giordano, G., Corradi, D., Melissari, M., Lagrasta, C., Gambert, S. R., and Anversa, P. 1995. Gender differences and aging:
effects on the human heart.). Am. Co!!. Cardiol. 26:1068-1079.
O'Shea, D. O., Davis, S. N., Kim, R. B., and Wilkinson, G. R. 1994. Effect of fasting and obesity in humans on the 6-hydroxylation of
chlozoxazone: A putative probe of CYP2E1 activity. Clin. Pharmacol. Ther. 56:359-367.
Pacific!, G. M., Franchi, M., Colizzi, C., Giuliani, L., and Rane, A. 1988. Sulfotransferase in humans: Development and tissue distribution.
Pharmacology 36:411-419.
Parkinson, A., Mudra, D. R., Johnson, C., Dwyer, A., and Carroll, K. M. 2004. The effects of gender, age, ethnicity, and liver cirrhosis on
cytochrome P450 enzyme activity in human liver microsomes and inducibility in cultured human hepatocytes. Tox/co/. Appl.
Pharmacol. 199:193-209.
Pierce, J. A., and Ebert, R. V. 1958. The elastic properties of the lungs in the aged. J. Lab. Clin. Med. 51:63-71.
Price, K., Haddad, S., and Krishnan, K. 2003a Physiological modeling of age-specific changes in the pharmacokinetics of organic chemicals
in children. J. Tox/co/. Environ. Health A 66:41 7-433.
Price, P. S., Conolly, R. B., Chaisson, C. F., Gross, E. A., Young, J. S., Mathis, E. T., and Tedder, D. R. 2003b Modeling interindividual
variation in physiological factors used in PBPK models of humans. Cr/t. Rev. Tox/co/. 33:469-503.
Puggaard, L., Bjornsbo, K. S., Kock, K., Luders, K., Thobo-Carlsen, B., and Lammert, O. 2002. Age-related decrease in energy expenditure
at rest parallels reductions in mass of internal organs. Am. J. Hum. Biol. 14:486-493.
Quetelet, L. A. 1869. Physique socia/e, Vol. 2, p. 92. Brussels: C. Muquardt.
Rooke, G. A., Savage, M. V., and Brengelmann, G. L. 1994. Maximal skin blood flow is decreased in elderly men. J. Appl. Physiol. 77:11 -14.
Sandstrom, T., Frew, A. J., Svartengren, M., and Viegi, G. 2003. The need for a focus on air pollution research in the elderly, Eur. Respir. ].
21:92S-95S.
Sato, T., Miwa, T., and Tauchi, H. 1 970. Age changes in the human liver of the different races. Geronto/og/a 1 6:368-380.
Scheinberg, P., Blackburn, I., Rich, M., and Saslaw, M. 1950. Effects of aging on cerebral circulation and metabolism.Am. J. Med. 8:77-85.
Schieve, J. F., and Wilson, W. P. 1953. The influence of age, anesthesia and cerebral arteriosclerosis on cerebral vascular activity to CO2.
Am.]. Med. 15:171-174.
Schmitz, A., Nyengaard, J. R., and Bendtsen, T. F. 1990. Glomerular volume in type 2 (noninsulin-dependent) diabetes estimated by a
direct and unbiased stereologic method. Lab. Invest. 62:108-113.
Schmucker, D. L., Woodhouse, K. W., Wang, R. K., Wynne, H., James, O. F., McManus, M., and Kremers, P. 1990. Effects of age and
gender on in vitro properties of human liver microsomal monooxygenases. Clin. Pharmacol. Ther. 48:365-374.
Schneeman, B. O., and Richter, D. 1993. Changes in plasma and hepatic lipids, small intestinal histology and pancreatic enzyme activity
due to aging and dietary fiber in rats.). Nutr. 123:1328-1337.
Seibold, H., Wieshammer, S., and Kress, P. 1988. Relation of pressure and flow of pulmonary circulation in patients with chronic
obstructive pulmonary disease. Clin. Physiol Biochem. 6:29-35.
Seitz, H. K., Egerer, G., Simanowski, U. A., Waldherr, R., Eckey, R., Agarwal, D. P., Goedde, H. W., and von Wartburg, J. P. 1993.
Human gastric alcohol dehydrogenase activity: Effect of age, sex, and alcoholism. Gut 34:1433-1437.
Shenkin, H. A., Novak, P., Goluboff, B., Soffe, A. M., Bortin, L., Golden, D., and Batson, P. 1953. The effects of aging, arteriosclerosis,
and hypertension upon the cerebral circulation.). Clin. Invest. 32:465.
Sherlock, S., Beam, A. G., Billing, B. H., and Paterson, J. C. 1950. Splanchnic blood flow in man by the bromsulfalein method: The relation
of peripheral plasma bromsulfalein level to the calculated flow.). Lab. Clin. Med. 35:923-932.
Shimada, T., Yamazaki, H., Mimura, M., Inui, Y., and Guengerich, F. P. 1994. Interindividual variations in human liver cytochrome
P-450 enzymes involved in the oxidation of drugs, carcinogens and toxic chemicals: Studies with liver microsomes of 30 Japanese
and 30 Caucasians.). Pharmacol. Exp. Ther. 270:414-423.
Shock, N. W. and Yiengst, M. J. 1955. Age changes in basal respiratory measurements and metabolism in males.). Gerontol. 10:31-40.
Sitar, D. S. 1 998. Geriatric clinical pharmacology. In Principles of medical pharmacology, eds. H. Kalant and W. H. E. Roschlau, 6th ed.,
pp. 820-829. New York: Oxford University Press.
Slack, T. K., and Wilson, D. M. 1976. Normal renal function. Mayo Clin. Proc. 51:296-300.
Slosman, D. O., Chicherio, C., Ludwig, C., Genton, L., de, R. S., Hans, D., Pichard, C., Mayer, E., Annoni, J. M., and de, R. A. 2001.
(1 33)Xe SPECT cerebral blood flow study in a healthy population: Determination of T-scores.). Nud. Med. 42:864-870.
Previous I TOC
-------
24
C. M. THOMPSON ET AL
Sotaniemi, E. A., Arranto, A. J., Pelkonen, O., and Pasanen, M. 1997. Age and cytochrome P450-linked drug metabolism in humans:
An analysis of 226 subjects with equal histopathologic conditions. Clin. Pharmacol. Ther. 61:331-339.
Swift, C. G., Homeida, M., Halliwell, M., and Roberts, C. J. 1978. Antipyrine disposition and liver size in the elderly. Eur. J. Clin. Pharmacol.
14:149-152.
Tauchi, H., Tsuboi, K., and Okutomi, J. 1971. Age changes in the human kidney of the different races. Geronto/og/a 1 7:87-97.
Thomson, A., Fletcher, P. J., Harris, P. J., Freedman, B., and Kelly, D. T. 1988. Regional distribution of cardiac output at rest and during
exercise in patients with exertional angina pectoris before and after nifedipine therapy.). Am. Coll. Cardiol. 11:837-842.
Thompson, C. M., Sonawane, B., Barton, H. A., DeWoskin, R. S., Lipscomb, J. C., Schlosser, P., Chiu, W. A., and Krishnan, K. 2008.
Approaches for applications of physiologically based pharmacokinetic models in risk sssessment. J. Toxicol. Environ. Health B
11:1-29.
Tietz, N. W., Shuey, D. F., and Wekstein, D. R. 1992. Laboratory values in fit aging individuals—Sexagenarians through centenarians.
Clin. Chem. 38:1167-1185.
Timbrell, J. 2000. Principles of biochemical toxicology, 3rd ed. Philadelphia: Taylor & Francis.
Wang, Z., Hall, S. D., Maya, J. F., Li, L., Asghar, A., and Gorski, J. C. 2003. Diabetes mellitus increases the in vivo activity of cytochrome
P450 2E1 in humans. Br.J. Clin. Pharmacol. 55:77-85.
Williams, L., and Leggett, R. 1989. Reference values for resting blood flow to organs of man. Clin. Phys. Physiol. Meas. 10:187-217.
Woodhouse, K. W., Mutch, E., Williams, F. M., Rawlins, M. D., and James, O. F. 1984. The effect of age on pathways of drug metabolism
in human liver. Age Ageing 13:328-334.
Wynne, H. A., Mutch, E., James, O. F., Wright, P., Rawlins, M. D., and Woodhouse, K. W. 1988. The effect of age upon the affinity of
microsomal mono-oxygenase enzymes for substrate in human liver. Age Ageing 1 7:401-405.
Wynne, H. A., Cope, L. H., Mutch, E., Rawlins, M. D., Woodhouse, K. W., and James, O. F. 1989. The effect of age upon liver volume
and apparent liver blood flow in healthy man. Hepatology 9:297-301.
Zoli, M., lervese, T., Abbati, S., Bianchi, G. P., Marchesini, G., and Pisi, E. 1989. Portal blood velocity and flow in aging man. Gerontology
35:61-65.
Zoli, M., Magalotti, D., Bianchi, G., Gueli, C., Orlandini, C., Grimaldi, M., and Marchesini, G. 1999. Total and functional hepatic blood
flow decrease in parallel with ageing. Age Ageing 28:29-33.
Previous
TOC
-------
ELSEVIER
Available online at www.sciencedirect.com
ScienceDirect
Regulatory Toxicology and Pharmacology 50 (2008) 400-411
Regulatory
Toxicology and
Pharmacology
www. elsevier. com/locate/yrtph
Development of good modelling practice for physiologically
based pharmacokinetic models for use in risk assessment: The first steps
George Loizoua'*, Martin Spendiffa, Hugh A. Bartonb, Jos Bessemsc,
Frederic Y. Bois d, Michel Bouvier d'Yvoire e, Harrie Buistf, Harvey J. Clewell IIIg,
Bette Meekh, Ursula Gundert-Remy \ Gerhard Goerlitzj, Walter Schmittj
a Health and Safety Laboratory, Harpur Hill, Buxton, SK17 9JN, UK
b National Centre for Computational Toxicology, Office of Research and Development, U.S. Environmental Protection Agency,
Research Triangle Park, NC 27711, USA
c Centre for Substances and Integrated Risk Assessment (SIR) National Institute for Public Health and the Environment (RIVM),
P. O. Box 1, Bilthoven, 3720 BA, The Netherlands
d INERIS, Pare Technologique Alata BP2, 5 rue Taffanel, Verneuil-en-Halatte 60550, France
° ECVAM, Institute for Health and Consumer Protection, EC-Joint Research Centre, via E. Fermi 1, 1-21020 Ispra (VA), Italy
{TNO, P.O. Box 360, Zeist, 3700 AJ, The Netherlands
eThe Hamner Institutes for Health Sciences, P.O. Box 12137, 6 Davis Drive, Research Triangle Park, NC, USA
h McLaughlin Institute, University of Ottawa, 1 Stewart Street, Ottawa, Ont., Canada KIN 6N5
1 BfR, Federal Institute for Risk Assessment, Thielallee 88-92, Berlin D-14195, Germany
1 BayerCropScience AG, Building 6690, Monheim 40765, Germany
Received 10 January 2008
Available online 1 February 2008
Abstract
The increasing use of tissue dosimetry estimated using pharmacokinetic models in chemical risk assessments in various jurisdictions
necessitates the development of internationally recognized good modelling practice (GMP). These practices would facilitate sharing of
models and model evaluations and consistent applications in risk assessments. Clear descriptions of good practices for (1) model devel-
opment i.e., research and analysis activities, (2) model characterization i.e., methods to describe how consistent the model is with biology
and the strengths and limitations of available models and data, such as sensitivity analyses, (3) model documentation, and (4) model
evaluation i.e., independent review that will assist risk assessors in their decisions of whether and how to use the models, and also model
developers to understand expectations for various purposes e.g., research versus application in risk assessment. Next steps in the devel-
opment of guidance for GMP and research to improve the scientific basis of the models are described based on a review of the current
status of the application of physiologically based pharmacokinetic (PBPK) models in risk assessments in Europe, Canada, and the Uni-
ted States at the International Workshop on the Development of GMP for PBPK Models in Greece on April 27-29, 2007.
Crown copyright © 2008 Published by Elsevier Inc. All rights reserved.
Keywords: Good modelling practice; PBPK; Risk assessment
1. Introduction
The increasing use of tissue dosimetry estimated
using pharmacokinetic models in chemical risk assess-
Corresponding author.
E-mail address: George.loizou@hsl.gov.uk (G. Loizou).
ments in a number of countries necessitates the need
to develop internationally recognized good modelling
practices. These practices would facilitate sharing of
models and model evaluations and consistent applica-
tions in risk assessments. Clear descriptions of good
practices for:
0273-2300/S - see front matter Crown copyright © 2008 Published by Elsevier Inc. All rights reserved.
doi:10.1016/j.yrtph.2008.01.011
Previous
-------
G. Loizou et al. I Regulatory Toxicology and Pharmacology 50 (2008) 400-411
401
1. Model development i.e., research and analysis activities,
2. Model characterization i.e., methods to describe how
consistent the model is in capturing the relevant biolog-
ical events with respect to mode of action and the
strengths and limitations of available model and data,
e.g., sensitivity analyses,
3. Model documentation, and
4. Model evaluation i.e., independent review,will assist risk
assessors in their decisions of whether and how to use
the models, and assist model developers to meet various
expectations (e.g., research versus application in risk
assessment) (Cobelli et al., 1984; Portier and Lyles,
1996; Rescigno and Beck, 1987).
For risk assessors, good modelling practice would pro-
vide guidance as a basis to evaluate the potential for a
pharmacokinetic model, particularly a physiologically
based pharmacokinetic (PBPK) model, to contribute to a
risk assessment. PBPK models represent part of a contin-
uum of increasingly data-informed approaches to dose-
response characterization that increasingly incorporate
more information and as such, contribute to better under-
standing and precision in estimating risks. These
approaches range from default ("presumed protective")
to more "biologically-based predictive" (Meek et al.,
2001). Default approaches are based on empirical observa-
tions from broad databases of information that are not
group, species or chemical specific; pharmacokinetics and
dynamics are not explicitly addressed. "Categorical" and
"species-specific" approaches incorporate category or
group specific information and increasingly along the
continuum, chemical-specific data are incorporated. This
includes development of chemical specific adjustment
factors (CSAF) incorporating compound-related or
chemical-specific pharmacokinetic (including PBPK
models) or pharmacodynamic data (Gundert-Remy and
Sonich-Mullin, 2002; IPCS, 2005) When appropriate, fully
data-derived, chemical-specific, biologically-based
dose-response risk assessment methods can be employed
for chemicals of high concern or with high economic
impacts thus entailing fuller quantitative characterization
of toxicokinetic and toxicodynamic aspects.
Increasingly, data-derived approaches to dose-response
assessment are based on weight of evidence descriptions of
known or hypothesized modes of action, the latter being a
description of the key events leading to toxicity rather than
a full mechanistic understanding. A framework for orga-
nizing and evaluating the weight of evidence supporting
modes of action in animals and their relevance to humans
has been developed which is applicable to all toxicity end-
points. This framework has evolved from consideration of
the weight of evidence of an animal mode of action
(Sonich-Mullin et al., 2001) to extension to human rele-
vance (Boobis et al., 2006; Meek et al., 2003b) and from
cancer (Boobis et al., 2006; Meek et al., 2003b) to non can-
cer endpoints (Boobis et al., 2008; Seed et al., 2005). This
framework which is now widely used in assessments
nationally and internationally continues to evolve, cur-
rently being extended to integrate dose-response analysis.
As such, it provides a transparent basis for defining the suf-
ficiency of data on mode of action that is needed to inform
the use of physiologically based pharmacokinetic models in
risk assessment.
For modellers, GMP is important to delineate the nat-
ure of model characterization and documentation that is
optimal for application in risk assessment. The initial cre-
ation of models, along with needed laboratory experimen-
tation, can be a creative and unpredictable process that
will be minimally altered by GMP. However, even at this
very early stage, awareness of GMP can be valuable,
including recommendations regarding transparency for
publication of the models in the peer reviewed literature
(Andersen et al., 1995). For example, modellers often try
several alternative structures as they attempt to reconcile
the available data and the description of the biology in
the model. While documentation to the same degree as a
model proposed for use in risk assessment is unnecessary,
understanding of the alternatives considered is important
in supporting the model structure eventually selected (Bar-
ton et al., 2007).
The International Workshop on the Development of
GMP for PBPK models1 was convened with two principal
themes:
1. The selection and evaluation of an appropriate determin-
istic^ model structure.
2. Increasing the understanding of regulators and risk
assessors through increased transparency and accessibil-
ity to user-friendly modelling techniques.
This was the first forum dedicated to promotion of best
practice in deterministic PBPK model development and
parameterisation, including consideration of transparency
in documentation with clear audit trails for model compo-
nents. Increase in consistency and transparency of support-
ing documentation is expected to facilitate dialogue and
understanding between PBPK practitioners, risk assessors
and regulators. By bringing together PBPK modellers,
mathematicians, statisticians, risk assessors, regulators
and laboratory scientists, the sponsors of this workshop
seek increased implementation of PBPK modelling in risk
assessment internationally, which GMP for PBPK should
1 April 26-April 28, 2007, at the Mediterranean Agronomic Institute of
Chania, Crete, Greece. Presentations, and discussion papers are at http://
www.hsl.gov.uk/news/news_pbpk.htm. Additional information is avail-
able at www.pbpk.org.
2 A "deterministic" model is the mathematical representation of the
biological/chemical system (e.g., PBPK model and metabolic scheme) as
opposed to a "non-deterministic" model, which is the mathematical/
statistical representation of the uncertainty, variability, and covariance of
the data and parameters of the deterministic model (e.g., statistical model
for measurement errors and population variability). Non-deterministic
modelling was a focus of the International Workshop on Uncertainty and
Variability in PBPK Models, 2006, North Carolina, US http://www.epa.
gov/ncct/uvpkm/ (Barton et al., 2007).
Previous
-------
402
G. Loizou et al. I Regulatory Toxicology and Pharmacology 50 (2008) 400^tll
facilitate. This paper presents the results and conclusions of
the GMP workshop.
2. Current practice—where do we stand?
The structure of PBPK models may differ to reflect the
requirements of the application, e.g., research (hypothesis
testing) and risk assessment. Appropriate practice for
these different uses and various stages of model develop-
ment are desirable. Past efforts to develop GMP for
other types of models applied in environmental regulation
are informative in terms of their form, content, and
application.
3. Value of PBPK models in risk assessments
The need for increasing incorporation of kinetic data
in the current risk assessment paradigm is due to an
increasing demand from risk assessors and regulators
for higher precision of risk estimates, a greater under-
standing of uncertainty and variability (Allen et al.,
1996; Barton et al., 1996; Clewell et al., 1999, 2002b;
Cox, 1996; Delic et al., 2000), more informed means of
extrapolating across species, routes, doses and time (Cle-
well and Andersen, 1987), the need for a more meaning-
ful interpretation of biological monitoring data
(Georgopoulos et al., 1994; Hays et al., 2007) and reduc-
tion in the reliance on animal testing (Barratt et al., 1995;
Blaauboer et al., 1996, 1999; DeJongh et al., 1999).
Incorporating PBPK modelling into the risk assessment
process can advance all of these objectives. Further, the
increasing trend to cost-benefit analysis should also
increase the utility of biologically based approaches in
the support of risk management decisions by regulatory
agencies (US EPA, 2006).
In addition, increasingly, testing and risk assessment is
being driven by considerations of mode of action and
resulting in more data-informed approaches to character-
ization of dose response, which should facilitate the incor-
poration of PBPK modelling. These approaches are
increasingly being adopted by risk assessment and regula-
tory communities, based on, for example, international ini-
tiatives such as the IPCS harmonization initiative for the
risk assessment of chemicals (Sonich-Mullin et al., 2001).
The latter initiative seeks to improve methods and to
increase understanding and acceptance through the pursuit
of common principles and approaches by drawing on glo-
bal expertise, leading ultimately to greater consistency
and convergence which will permit the sharing of assess-
ments and avoid duplication. Potential areas of conver-
gence for which analytical frameworks, guidance and
associated training materials have been developed through
this initiative include weight of evidence for mode of
action, CSAFs and more recently PBPK modelling (Boobis
et al., 2006, 2008; IPCS, 2005; Meek et al., 2001, 2002,
2003a,b; Meek and Renwick, 2006; Sonich-Mullin et al.,
2001).
4. Current status of implementation of PBPK models in risk
assessments
One of the first PBPK models to be adopted in regula-
tory risk assessment was that for methylene chloride,
whose evolution involved an iterative hypothesis testing
process for the pharmacokinetics and glutathione transfer-
ase-mediated mode of action leading to cancers in rodents.
The mathematical model gave a quantitative form to the
researcher's conception of the biological system, permitting
the development of a testable, quantitative hypothesis, the
design of informative experiments and the ability to recog-
nize inconsistencies between theory (model) and data. The
explicit description of model parameters also led to the
ability to study and quantify uncertainty. The model has
been widely applied in risk assessments by the US Con-
sumer Products Safety Commission (Babich, 1998), by
the US Occupational Safety and Health Administration
for establishing the permissible exposure level including
use of Bayesian statistical parameter estimation and char-
acterization of uncertainty and variability (OSHA, 1997),
by the US Environmental Protection Agency in the Inte-
grated Risk Information System (IRIS) assessment for
inhalation cancer risk (Dewoskin, 2007; US EPA, 1987)
and Health Canada in their assessment for the general pop-
ulation under the Canadian Environmental Protection Act
(Government of Canada, 1993).
An overview of the use of PBPK modelling by various risk
assessment/regulatory authorities is presented in Table 1.
The results of this limited analysis presented in Table
1 indicate that PBPK models are increasingly being
adopted in risk assessment by regulatory agencies in
Europe and North America, most often to date, as a
basis for quantitatively considering interspecies differ-
ences as a basis to replace the default approach. The
extent of documentation of the rationale for accepting
or rejecting the use of particular models varies consider-
ably and is likely to be dependent upon access to rele-
vant expertise. In most cases, lack of adoption of
particular models within risk assessment has been a
function of insufficient weight of evidence of the under-
lying hypothesized mode of action and/or the lack of a
standardized procedure for the evaluation of PBPK
models and their output.
5. What can we learn from other similar modelling
experiences?
While the use of quantitative modelling in human
health risk assessment has been more limited, particularly
biologically-based dose-response analyses, modelling for
environmental fate and transport has gained increased
acceptance since the 1990s and is now widely accepted
in European, Canadian, and US regulatory contexts.
Today in Europe modelling endpoints for groundwater
are decisive in the registration of pesticides. In North
America and Europe, risk assessments for specific con-
Previous
-------
G. Loizou et al. I Regulatory Toxicology and Pharmacology 50 (2008) 400-411
403
Table 1
The use of PBPK modelling by various risk assessment/regulatory authorities
Assessments
Use of PBPK Models
Impact/Rationale
Random selection of 80/141 EU Existing Substances Mentioned in 8/80
Reports (1996-2007) (European Chemicals
Bureau)
Adopted in 418
(vinyl acetate, 2-butoxyethanol, propylene
methyl glycol,styrene)
Not used in 418 (benzene,
acrylic acid,
UK Health and Safety Executive
cyclohexane, methyl methacrylate,)
Formaldehyde
2-butoxyethanol
French Agency for Environmental and Occupational Consideration in setting reference values
Health for reproductive health
Health Canada Priority Substances under the Considered inadequate for quantifying
Canadian Environmental Protection Act (n = 44 interspecies differences for
on the first Priority Substances List (PSL 1) and tetrachloroethylene, styrene and
25 on PSL 2 (1989-1994) (Health Canada Priority diethylhexylphthalate (PSL 1)
Substances Assessment Program)
Adopted for
Cadmium (PSL 1)
Formaldehyde
Chloroform
2-butoxyethanol
US FDA
US EPA (IRIS)
Trans retinoic acid
Not applied for
Acetone
Chloroform
Methyl ethyl ketone
Adopted for
Dichloromethane
Ethylene glycol monobutyl ether
Vinyl chloride
Xylene
Reduction of uncertainty factor for interspecies
differences or reduction of classification category
Mode of action judged to vary between humans and
animals
In vitro activity of enzymes in one tissue as
surrogates of in vivo activity in another tissue judged
to be implausible
No reason provided
Lack of biological plausibility of association with
leukaemia (Franks, 2005)
Consideration of validity of a biomarker and
robustness of past regulatory decisions (Delic et al.,
2000, Franks et al., 2006)
(INERIS, 2007)
Quantification of human variability
Quantification of interspecies differences in
biologically motivated case specific model
Quantification of interspecies differences
Quantification of interspecies differences
Consideration of potential risk of dermal application
(Clewell et al., 1997, Rowland et al., 2004)
Lack of necessary exposure route in model.
Lack of model parameterization in species with
critical effect.
Lack of sufficient supporting data for model and
demonstration of predictive capability.
Quantification of interspecies differences.
Quantification of interspecies differences.
Quantification of interspecies differences in PK and
demonstration of interspecies similarities in cancer
PD. Route-to-route extrapolations to derive point of
departure.
Comparison to default RfCa
a The concentration of a chemical in air that is very unlikely to have adverse effects if inhaled continuously over a lifetime (http://cfpub.epa.gov/ncea/
cfm/recordisplay.cfm?deid=55365).
laminated sites or permitting of industrial facilities also
rely heavily on often complex models for exposure path-
ways including food chains (US EPA, 1989). More
recently, in view of the introduction of demanding man-
dates to consider much larger numbers of existing chem-
ical substances (e.g., categorization and screening of the
Domestic Substances List (DSL) (Health Canada
Domestic Substances List) in Canada and the Registra-
tion, Evaluation and Authorization of Chemical Sub-
stances (REACH) (European Community Regulation
REACH) in Europe, there is increasing development of
quantitative structure activity relationship (QSAR) mod-
els particularly for application in human health risk
assessment and associated GMP. These experiences pro-
vide perspectives that are potentially useful for the devel-
opment of GMP for PBPK modelling.
Previous
-------
404
G. Loizou et al. I Regulatory Toxicology and Pharmacology 50 (2008) 400^tll
6. Environmental modelling-achieving acceptance in the
regulatory world
The development of good practice in environmental fate
modelling may provide a relevant perspective for the devel-
opment of GMP for PBPK modelling. GMP for environ-
mental fate modelling evolved in Europe as a result of
two issues; firstly EU legislation in the late 1980s set a max-
imum pesticide residue concentration of 0.1 (ig L"1 in both
drinking and ground water and secondly lysimeter3 studies
which took between three to four years leading to long
delays on decisions on the use of critical products in agri-
culture while avoiding contamination of groundwater
resources. Environmental fate modelling was recognized
as a promising approach to address these issues but ques-
tions were raised concerning whether model predictions
were sufficiently reliable and how to ensure the integrity
of model calculations.
Clear divisions in attitudes among environmental fate
modellers, regulators, and registrants emerged following ini-
tial discussions. Researchers used the models for the investi-
gation of processes and systems, requiring flexibility and
adaptability while maintaining full control of processes
and algorithms in the models. Regulators and registrants
wanted to predict exceedence or adherence to a regulatory
limit. They required scientific and legal certainty and pre-
ferred models for which the code could not be altered, and
had complete documentation with clear audit trails for cal-
culations. Further conflicts arose because version control
and documentation of research models was rudimentary at
best, no guidance on the selection of appropriate input
parameters was available, and it was rarely properly estab-
lished whether a model design was suitable for regulatory
purposes. These issues reflected the variations in objectives
of models developed for research versus regulatory applica-
tion with the former being intended for use by a specialist
with specific and intensive training, which at the time was
almost totally lacking in regulatory agencies and companies
assessing the environmental behaviour of plant protection
products. As a consequence, results for different modellers
using the same models in similar applications varied.
Initially, software packages comprising models with a
user-friendly graphical interface and pre-configured scenar-
ios were developed. However, non-expert users still pro-
duced poor results for two main reasons (i) model
processes, algorithms and standard parameters did not
appropriately reflect substance properties, and (ii) sub-
stance data from standard environmental fate studies were
conceptually different from these required for model imple-
mentation. This led to proposals from regulatory agencies
to apply good laboratory practice (GLP) for modelling to
ensure that all data could be 'verified'. Also, GLP had just
been successfully transferred from toxicology to metabo-
lism, environmental fate and residue analysis laboratories.
3 The measurement of the water percolating through soils and the
determination of the materials dissolved in the water.
On the other hand, measurements are never perfectly
reproducible (especially not for living systems) whereas
simulations are and GLP is difficult to apply to electronic
data systems and calculations.
This led to the development of a short document by the
Federal Biological Research Centre for Agriculture and For-
estry (BBA), the Federal Environmental Agency (UBA), the
Fraunhofer Institute for Environmental Chemistry and Eco-
toxicology (FhG IUCT) and the German Agrochemical
Industry (IVA) entitled, "Rules for the correct performance
and evaluation of model calculations for simulation of the
environmental behaviour of pesticides" (Gorlitz, 1993).
Later referenced as the 'Codex' this document outlined gen-
eral principles of GMP, rather than prescriptive guidance. It
focused on leaching models but was generally applicable to
other simulation models and addressed the following topics:
selection of models, documentation of models, validation,
support, official recognition and version control, selection
and treatment of input data, consistency of input data and
models, documentation of simulations, reporting and inter-
pretation. The Codex led to regulatory acceptance of simula-
tion models on a national scale in Germany, as well as
providing a basis to address the requirements of the Euro-
pean directive 91/414 (European Community Regulation
Council Directive 91/414).
After several informal meetings between modellers, reg-
ulators and registrants, the F<9rum for the Coordination of
pesticide fate models and their t/Se (FOCUS) was created
in 1993 through an initiative of the European Commission
(European Commission FOCUS). The steering committee
of FOCUS met under the auspices of the EU Directorate
General for Health and Consumer Affairs (DG SANCO)
for the first time in 1993 and approved two research area
themes on models for groundwater and surface water.
FOCUS has an equal representation of regulators,
researchers and industry that operate by consensus and
offer technical support to the EU registration process 91/
414. It has no administrative infrastructure but DG
SANCO provides funds for attendance at meetings for reg-
ulatory experts and researchers. The FOCUS committee
meets approximately four times per year and has two per-
manent institutions. The FOCUS website (European Com-
mission FOCUS) provides all the reports of past FOCUS
projects, the actual recommended versions of models as
well as essential scenario data. Members of the supporting
technical Version Control Group are model developers/
supporters. This group approves new model versions, and
the content of the website, by correspondence.
Currently, FOCUS reports figure prominently in expo-
sure assessments for the registration of plant protection
products in the EU. This is best illustrated by the fact that
the present draft of the revision of the EU directive 91/414
on the authorization of plant protection products refer-
ences directly FOCUS reports as guidance on important
decision points. FOCUS outputs are also widely adopted
as guidance by member states in their exposure and risk
assessments.
Previous
-------
G. Loizou et al. I Regulatory Toxicology and Pharmacology 50 (2008) 400-411
405
7. Evolving acceptance of QSAR modelling
Evolving advancements internationally in the documen-
tation and implementation of quantitative structure activ-
ity relationship (QSAR) models to meet demanding
mandates to consider much larger numbers of existing
substances may also contribute in the development of
GMP for PBPK models. These include principles for ver-
ification of QSAR model output (OECD, 2007) and pro-
posed templates for QSAR development, prediction and
reporting (http://ecb.jrc.it/qsar/) (European Chemicals
Bureau RIP 3.3). Whilst the documentation is still evolv-
ing internationally, information on the training domain,
internal validation, cross validation and external valida-
tion requirements has been proposed to be included in a
'development template' whereas substance-specific infor-
mation is proposed to be included in a 'prediction
template'.
8. Future directions—where do we need to go?
The following sections briefly summarize some of the
major issues considered and recommendations from the
workshop designed to facilitate the development of GMP
for PBPK modelling as well as to identify research
priorities.
8.1. Risk assessors needs and their role in the process
Two possible paradigms were proposed for the involve-
ment of the risk assessor throughout the modelling pro-
cess: (1) issues raised by the risk assessor are included
during model development, and (2) at appropriate times,
the model would be evaluated for fitness for regulatory
use. To the extent that it is possible, the former process
is clearly preferred and necessitates involvement of an
interdisciplinary team in model development and charac-
terization (Barton et al., 2007), whereas the latter process
is more typical for models that have already been
published.
Risk assessors have important roles to play in mode of
action and dosimetry-based risk assessments utilizing
PBPK models. These include transparently assessing the
weight of evidence of hypothesized modes of action as a
basis for clearly delineating the goals for using the model
in the risk assessment (Clewell et al., 2002a; US EPA,
2006) and participating in a transparent process that
brings together appropriate interdisciplinary expertise to
evaluate the model and its proposed risk assessment appli-
cations (Chiu et al., 2007; Clark et al., 2004). Further-
more, risk assessors play pivotal roles in organizing the
information on mode of action and dose-response e.g.,
critical studies and endpoints that form the context for
applying a dosimetry model. Transparent frameworks
developed for this purpose (Boobis et al., 2008; IPCS,
2005) may assist the risk assessor in assimilation of this
information. Determining whether a PBPK model is
parameterised for the chemical(s), including metabolites,
species and life stages, exposure routes and matrices in
the toxicity studies used in dose-response analysis or the
human exposures relevant for the risk assessment, can
be accomplished by non-modellers. Identifying the dose
metrics relevant to the modes of action under consider-
ation and evaluation of the biology captured by the model
often requires communication among risk assessors, toxi-
cologists, and modellers. Evaluation of the mathematical
and computer implementation as well as characterization
of its consistency with available data and the model's
strengths and weaknesses for the proposed risk assessment
applications will generally require involving those with
appropriate mathematical, statistical and computational
expertise. However, to ensure a transparent process, com-
munications describing the review process and its conclu-
sions need to be clear and comprehensible to all parties.
8.2. Model development practices
Model standardization can facilitate intra- and inter-dis-
ciplinary communication but creates challenges of adapting
to a variety of software used to produce a range of model
structures necessary to describe different kinetic behaviours
and address varying model purposes. There are significant
benefits to the use of generic model structures, including
the establishment of standard abbreviations or parameter
nomenclature and glossary, which would facilitate efficient
communication of models and avoid confusion in seman-
tics that can hinder understanding. In addition, the need
to justify selected aspects of the model could be eliminated
as is currently done by citing existing literature. To be truly
generic, however, a model would have to encompass a wide
range of physiological compartments and all useful dose
metrics.
Standard methodology for model building might be a
more viable alternative than a fixed model form (Cobelli
et al., 1984). Moreover, the use of a hybrid approach
whereby a simple standard model is used as a starting point
and refinements during the modelling workflow are con-
ducted utilising a standardized model building methodol-
ogy may be a viable compromise. In discussing the issues
associated with model code that is specific to a particular
solver package, it was agreed by the workshop delegates
that the use of a standard representation similar to Systems
Biology Mark-up Language (SBML) or Cell Mark-up Lan-
guage (cellML)4 would improve communication between
modellers and risk assessors. Mark-up Language (ML) is
a type of representation that gives a structured description
of the conceptual model, free of mathematical equations
and confusing syntax. The provision of an intuitive graph-
ical interface such as MEGen5 could make such standard
formats more accessible to non-modellers by allowing
rapid generation of this 'PBPKML' representation.
4 http://sbml.org; www.cellml.org
5 www.opentox.com/megen
Previous
-------
406
G. Loizou et al. I Regulatory Toxicology and Pharmacology 50 (2008) 400^tll
8.3. Model verification
Models can be analysed to demonstrate that they are
mathematically and computationally free of errors and
that the behaviour of the model in the region of parameter
space that is biologically plausible, reasonably approxi-
mates the available data (Barton et al., 2007; Oreskes,
1998). Demonstration that a model is mathematically
and computationally correctly implemented can involve
checks incorporated in the model, e.g., mass balance
checks, rigorous manual checking of the equations and
computer code, and independent receding of the model
using another software environment. The ease of imple-
menting these options varies with the particular software
used. A PBPK model code generator tool such as
MEGen5 could facilitate these checks by permitting rapid
receding of models.
8.3.1. Roles and methods of sensitivity analysis
Sensitivity analysis is a tool for model characterization
that can address a number of issues frequently raised con-
cerning PBPK models.
Sensitivity analyses can be implemented in model devel-
opment, characterization and evaluation to address several
aspects including the following:
1. Characterizing parameters that are well determined by
available data.
2. Iterating with experiments and evaluating the sensitivity
of parameters to new data that will be collected (Cho
et al., 2003; Gueorguieva et al., 2006; Nestorov et al.,
1998).
3. For dose-response analysis predictions, evaluating the
sensitivity of dose metrics predicted under the conditions
relevant to the toxicity studies (or epidemiological stud-
ies) to the parameters in the model.
4. For risk assessment, evaluating the predicted dose met-
rics in humans under relevant environmental exposure
conditions to characterize their sensitivity with respect
to the model parameters.
The many existing sensitivity analysis methods can be
grouped into two categories: (1) local methods that con-
sider sensitivities close to a specific set of input param-
eter values, and (2) global methods that calculate the
contribution of a parameter over the set of all possible
input parameters. Currently, gaining insight into a
model often involves the adjustment of individual model
parameters and observation of the predicted changes in
model output, either at a single time or throughout a
time course. This useful practice can be supplemented
by examining the time-dependent global sensitivities of
the chosen dose-metric for dominant parameters. When
trying to establish the contribution of a parameter to
model predictions, local sensitivity analysis techniques
are fairly rapid and simple to implement but can give
somewhat misleading results if there are substantial
interactions among multiple parameters.
Global sensitivity analysis using the Extended Fourier
Amplitude Sensitivity Test (FAST) is a variance-based
method that is independent of any assumptions about the
model structure and is effective for monotonic, exclusively
increasing or decreasing predictions, and non-monotonic
models (Campolongo and Saltelli, 1997). The FAST is
preferable over other global methods due to its computa-
tional efficiency and capability to consider parameter inter-
actions as well as main effects. Since PBPK models are
likely to become increasingly complex as more pertinent
data become more readily available more robust sensitivity
analysis techniques will be required and FAST appears to
satisfy these criteria.
8.4. Model documentation
Suggestions for documenting models in publications
have been presented previously (Andersen et al.,
1995). As noted therein, model documentation must
address a diverse readership. Recommendations from
this workshop were to develop a standard, brief model
description summary for the broad risk assessment
audience and more detailed documentation for special-
ists. The summary would contain at least seven elements
including:
1. Introduction including problem formulation (applicabil-
ity of model).
2. A text description of the model (species, routes, etc) with
schematic diagram, and an overview of the information
and data supporting the model structure.
3. Metabolic pathways for the chemical and an overview of
the supporting information and data.
4. Relationship to mode of action including dose metric
predictions and supporting information.
5. Distributional predictions of model outputs and their
implications (e.g., Monte Carlo simulation of human
variability).
6. Overview of uncertainty and sensitivity analyses.
7. Source of complete information (e.g., citation).
Further recommendations for more complete model
documentation could include the possibility of utilising
hyperlinked documents that facilitate easy access to sup-
porting materials, including calculations done to convert
published scientific information into the form used in
the model. This extended model documentation would
be utilized by subject experts in the model evaluation pro-
cess and would ideally be publicly accessible via the Inter-
net. The documentation would strive for transparency
through the integration of diagrams of model structure
and metabolic pathways, tables of model state variables
and parameters and mathematical equations and model
code.
Previous
-------
G. Loizou et al. I Regulatory Toxicology and Pharmacology 50 (2008) 400-411
407
8.5. Model evaluation
Best practices allow efficient evaluation of models
through standardization, documentation, and transpar-
ency. The six-step process of assessment of model purpose,
assessment of model structure and biological characteriza-
tions, assessment of mathematical descriptions, assessment
of computer implementation, parameter analysis and
assessment of model fit and assessment of any specialized
analyses described by Clark et al. (2004) and extended by
providing more detail by Chiu et al. (2007) provides a use-
ful framework for model evaluation. Further, specification
of criteria that would assist reviewers in determining the
strengths and limitation of a specific model and a process
for implementation of model evaluation, which must be
transparent and involve independent review, would be
valuable.
Development of a robust model evaluation process must
take into account the need for external review since while
involvement of risk assessors and modellers throughout
the steps leading from model development to application
in risk assessment is valuable, it can impact on the percep-
tion of the model evaluation as an independent process. An
independent review is essential to identify and correct mis-
takes and to make judgments on the adequacy of the model
and its supporting scientific database. Such reviews present
a challenge internationally, not least because of the limited
PBPK modelling expertise globally. For this reason, it
would be valuable to be able to share model evaluations
among countries, by agreeing upon a common framework
and process even if the final decisions concerning model use
might be different, for example due to risk assessment
needs.
A major challenge of model evaluation is to provide per-
spective on the scientific uncertainties identified by a model
and its supporting scientific database. Models allow char-
acterization of uncertainty in a way that default analyses
cannot: for example, a default value of 10 is commonly
applied for interspecies extrapolation, but the uncertainty
for any specific chemical with regard to the toxicity it
causes in animals ranges from close to zero (the effect only
occurs in the animals) to a much larger value (the effect
only occurs in humans). While the factor of 10 represents
a judgment concerning the general tendency across many
chemicals, it cannot describe the uncertainties for a specific
chemical whereas this is possible using biologically based-
modelling. However, this creates a challenge for consider-
ing whether the model adequately captures the science
and thus, should be implemented in the risk assessment.
8.6. Improving the scientific basis supporting models
Efforts to use PBPK models more broadly have also
resulted in a range of scientific issues that require addi-
tional research. These include improving methods for using
in vitro data in order to limit controlled animal and human
studies, for model development by extrapolating from
widely studied chemicals to those with limited information
and for better characterizing uncertainty and variability in
PBPK models.
8.6.1. In vitro to in vivo extrapolations
Ideally, in vitro data should be used in PBPK models
because they can limit the need for in vivo studies in ani-
mals or humans. However, limitations of models to predict
in vivo rat data using metabolic parameters estimated from
in vitro studies have been noted (Csanady and Filser, 2007;
Faller et al., 2001; Lee et al., 2005; Osterman-Golkar et al.,
2003).
In vitro to in vivo extrapolation, particularly with
regards to metabolism, requires further detailed study (Bla-
auboer et al., 1999, 1996; DeJongh et al., 1999; Gulden and
Seibert, 2003; Houston, 1994; Kedderis, 1997; Lipscomb
et al., 1998; Miners et al., 1994; Rostami-Hodjegan and
Tucker, 2004; Verwei et al., 2006; Wilson et al., 2003).
The importance of protein and non-specific binding and
partitioning of substrates are fundamental to improving
the utility of in vitro systems and the use of such data in
PBPK models. While there are initiatives underway to
assist in addressing many of these issues6 and encouraging
results have recently been reported (Acutetox Newsletter
July, 2007), the limitations of in vitro metabolism data must
be borne in mind until and unless they can be demonstrated
to be reliable surrogates.
8.6.2. Cross chemical extrapolation
Risk assessors are increasingly having to address prior-
itisation and assessment for the large numbers of chemicals
in commerce, notably the REACH legislation in Europe or
the Categorization and Screening of the Domestic Sub-
stances List under the Canadian Environmental Protection
Act (1999)7. Methods to develop initial PBPK models for
chemicals using cross-chemical prediction methods would
be valuable and efforts to date have primarily been directed
at predicting tissue:blood or tissue: air partition coefficients
(Beliveau et al., 2005), though in vitro to in vivo extrapola-
tion for metabolism and other aspects of pharmacokinetics
are also receiving attention.
8.6.3. Uncertainty and variability in PBPK models
Much of the focus in the development of PBPK models
has been to identify and capture the average behaviour of
the key biological processes controlling the pharmacoki-
netics of a chemical. These models have successfully
assisted in evaluating biological hypotheses for mode of
action e.g., methylene chloride carcinogenesis described
previously, as well as identifying previously unrecognised
pharmacokinetic behaviours. The increasing application
of PBPK models in risk assessment has led to a range
of efforts to better characterize the relationship between
6 (http://www.acutetox.org/)
7 (http://www.ec.gc.ca/substances)
Previous
-------
408
G. Loizou et al. I Regulatory Toxicology and Pharmacology 50 (2008) 400^tll
the model and supporting data and quantify uncertainty
and variability.
Improved computing power is essential to more wide-
spread use of distributional analyses to characterize human
variability with Monte Carlo sampling techniques and
methods of parameter estimation ranging from optimisa-
tion of selected chemical specific parameters (e.g., meta-
bolic rates) to global parameter estimation using
Bayesian statistical characterization of uncertainty and
variability. Priorities for research and implementation of
concepts of uncertainty and variability in risk assessments
using PBPK models have been previously described (Bar-
ton et al., 2007).
8.7. Good modelling practices for PBPK models: developing
a description, case studies and training materials
The International Programme on Chemical Safety
(IPCS) steering group of the World Health Organization
(WHO) identified PBPK modelling as an important com-
ponent of chemical risk assessment that merits interna-
tional harmonization8. The ability to review a PBPK
model according to accepted criteria would greatly facili-
tate widespread acceptance, in particular amongst regula-
tors. While agreement amongst PBPK model developers
is paramount for the development of GMP, the guidelines
must also be acceptable to regulators and risk assessors.
Development of guidelines for GMP is best achieved
through a cross-disciplinary exchange of experience and
ideas among laboratory scientists, PBPK modellers, regula-
tors and risk assessors.
The adequacy of the GMP description can be evaluated
using case studies that in turn could form the basis for
training materials on GMP. Some recommendations were
proposed for case studies:
• Comparing a dose metric for which data were directly
available versus one where they were not.
• Examples where PBPK models were accepted and used
by regulatory Agencies and ones where they were
rejected to ensure appropriate documentation.
• Comparisons of data-rich chemicals with data-limited
chemicals including not just comparison of pharmacoki-
netic or metabolic data, but also mode of action data
such as toxicogenomic or metabolomic data.
• Illustrations of different risk assessment applications.
Potential chemicals to use as case studies would
include those for which PBPK models had been consid-
ered or applied in risk assessments in Europe, Canada,
and the United States. Other chemicals could include iso-
propanol (with acetone metabolite sub model) for
non-cancer endpoints, styrene as an example of an inac-
cessible dose metric, acrylamide as an example of great
8 http://www.who.int/ipcs/methods/harmonization/areas/pbpk/en/
index.html
current regulatory interest with multiple proposed modes
of action and target sites and 1,3-butadiene due to the
substantial animal modelling and uncertainty in human
metabolism resulting in assessment based upon
epidemiology.
Finally, development of training materials and hiring of
personnel with the required expertise will be essential to
facilitate implementation of mode of action and dosime-
try-based risk assessment by regulatory Agencies. Training
materials are needed so that risk assessors and managers
with diverse expertise can successfully interact with model-
lers to implement PBPK models in risk assessment. Train-
ing will also be important for modellers to learn about
newer methodologies for characterizing uncertainty and
variability in PBPK models or implementing local and
global sensitivity analyses at appropriate stages of model
maturation. A longer-term strategy would be to include
a more quantitative, computationally based study of toxi-
cology in university courses. The adaptation of a PBPK
model generator tool such as MEGen as a teaching tool
would be very useful in demonstrating to students how
biological knowledge can be applied to solve real-world
problems.
9. Conclusions and recommendations
The use of PBPK modelling in risk assessment is
increasing in various jurisdictions but would benefit from
development of principles and guidance for GMP to
assist modellers during design and verification and risk
assessors in evaluation for application. Experience in
development of similar guidance in other areas such as
environmental fate modelling and more recently evolving
principles and documentation prototypes for QSAR
response modelling can inform this process. Recommen-
dations for aspects to be addressed in GMP for model
development, characterization, documentation and evalu-
ation were based on an international workshop and
include:
1. Transparency of model documentation and the weight
of evidence for the underlying hypothesized mode of
action is needed to aid transition of models from devel-
opers to evaluators and users.
2. Independent review of models is essential to evaluate
documentation and implementation quality for applica-
tions in risk assessment.
3. Consistent model evaluation approaches would facilitate
international sharing of the analyses that would then
form the basis for decisions appropriate to different reg-
ulatory applications. An international committee would
be valuable to further this goal.
4. Successful development, evaluation and application of a
PBPK model requires multidisciplinary skills through-
out the process. Regulatory agencies need to develop
access to those who can provide those skills through
training, hiring or other approaches.
Previous
-------
G. Loizou et al. I Regulatory Toxicology and Pharmacology 50 (2008) 400-411
409
5. Training in toxicology at the university and professional
levels needs to recognize that quantitative risk assess-
ment applications are major drivers for interest in toxi-
cological data by providing more quantitative,
computationally-based studies.
These recommendations will be considered further in the
development of relevant guidance in an ongoing initiative
of the IPCS harmonization project.
Disclaimer
The views expressed in this article are those of the
authors and do not necessarily represent the views or pol-
icies of the UK Health and Safety Executive, US Environ-
mental Protection Agency or INERIS. Mention of trade
names or commercial products does not constitute endorse-
ment or recommendation for use.
The present document does not represent an official
position of the European Commission.
Acknowledgments
The Health and Safety Executive (UK), the European
Chemical Industry Council (CEFIC), Health Canada and
the European Centre for the Validation of Alternative Meth-
ods (ECVAM) provided funds for the support of the work-
shop. The scientific committee and organizers of the
workshop thank the speakers; Woodrow Setzer (US EPA),
Gerhard Goerlitz (Bayer CropScience, Germany), Mel
Andersen (The Hamner Institutes for Health Research,
USA), Bette Meek (McLaughlin Institute, University of
Ottowa, Canada), Martin Spendiff (HSL, UK), Gyorgy
Csanady (GSF-Institute of Toxicology, Germany) and
Ursula Gundert-Remy (BfR, Germany) for her impromptu
presentation, the breakout session chairs; Hugh Barton
(US EPA), George Loizou (HSL, UK), Martin Spendiff
(HSL, UK) and Kannan Krishnan (University of Montreal,
Canada), the rapporteurs; Harvey Clewell (The Hamner
Institutes for Health Research, USA) and Rob DeWoskin
(US EPA) and the recorders; Cecilia Tan (The Hamner
Institutes for Health Research, USA), Tammie Covington
(Henry Jackson Foundation, USA), Jos Bessems (RIVM,
The Netherlands) and Marco Zeilmaker (RIVM, The
Netherlands). The scientific committee and organizers
also thank Katerina Karapataki and her staff at the
Mediterranean Agronomic Institute at Chania, Crete,
Greece for their invaluable logistical help and advice.
References
Acutetox (Newsletter July 2007). .
Allen, B.C., Covington, T.R., Clewell, H.J., 1996. Investigation of the
impact of pharmacokinetic variability and uncertainty on risks
predicted with a pharmacokinetic model for chloroform. Toxicology
111, 289-303.
Andersen, M.E., Clewell III, H.J., Frederick, C.B., 1995. Applying
simulation modeling to problems in toxicology and risk assessment—a
short perspective. Toxicology and Applied Pharmacology 133, 181-
187.
Babich, M.A., 1998. Risk assessment of low-level chemical exposures from
consumer products under the u.s. consumer product safety commission
chronic hazard guidelines. Environmental Health Perspectives 106
(Suppl. 1), 387-390.
Barratt, M.D., Castell, J.V., Chamberlain, M., Combes, R.D., Dearden,
J.C., Fentem, J.H., Gerner, I., Giuliani, A., Gray, T.J.B., Livingstone,
D.J., Provan, W.M.L., Rutten, F., Verhaar, H.J.M., Zbinden, P., 1995.
The integrated use of alternative approaches for predicting toxic
hazard—the report and recommendations of Ecvam workshop-8.
ATLA-Alternatives To Laboratory Animals 23, 410-429.
Barton, H.A., Chiu, W.A., Setzer, R.W., Andersen, M.E., Bailer, A.J.,
Bois, F.Y., Dewoskin, R.S., Hays, S., Johanson, G., Jones, N., Loizou,
G., Macphail, R.C., Portier, C.J., Spendiff, M., Tan, Y.M., 2007.
Characterizing uncertainty and variability in physiologically-based
pharmacokinetic (PBPK) models: state of the science and needs for
research and implementation. Toxicological Sciences 99, 395-402.
Barton, H.A., Flemming, C.D., Lipscomb, J.C., 1996. Evaluating human
variability in chemical risk assessment: hazard identification and dose-
response assessment for noncancer oral toxicity of trichloroethylene.
Toxicology 111, 271-287.
Beliveau, M., Lipscomb, J., Tardif, R., Krishnan, K., 2005. Quantitative
structure-property relationships for interspecies extrapolation of the
inhalation pharmacokinetics of organic chemicals. Chemical Research
in Toxicology 18, 475^185.
Blaauboer, B.J., Barratt, M.D., Houston, B.J., 1999. The integrated use of
alternative methods in toxicological risk evaluation: ECVAM inte-
grated testing strategies task force report 1. ATLA-Alternatives To
Laboratory Animals 27, 229-237.
Blaauboer, B.J., Bayliss, M.K., Castell, J.V., Evelo, C.T.A., Frazier, J.M.,
Groen, K., Gulden, M., Guillouzo, A., Hissink, A.M., Houston, B.J.,
Johanson, G., de Jongh, J., Kedderis, G.L., Reinhardt, C.A., van de
Sandt, J.J.M., Semino, G., 1996. The use of biokinetics and in vitro
methods in toxicological risk evaluation. ATLA-Alternatives To
Laboratory Animals 24, 473^197.
Boobis, A.R., Cohen, S.M., Dellarco, V., McGregor, D., Meek, M.E.,
Vickers, C., Willcocks, D., Farland, W., 2006. IPCS framework for
analyzing the relevance of a cancer mode of action for humans. Critical
Reviews in Toxicology 36, 781-792.
Boobis, A.R., Doe, J.E., Heinrich-Hirsch, B., Meek, M.E., Munn, S.,
Ruchirawat, M., Schlatter, J., Seed, J., Vickers, C., 2008. IPCS
framework for analysing the relevance of a non-cancer mode of action
for humans. Critical Reviews in Toxicology 38, 87-96.
Campolongo, F., Saltelli, A., 1997. Sensitivity analysis of an environmen-
tal model; a worked application of different analysis methods.
Reliability Engineering and System Safety, 49-69.
Chiu, W.A., Barton, H.A., Dewoskin, R.S., Schlosser, P., Thompson,
C.M., Sonawane, B., Lipscomb, J.C., Krishnan, K., 2007. Evaluation
of physiologically based pharmacokinetic models for use in risk
assessment. Journal of Applied Toxicology 27, 218-237.
Cho, K.-H., Shin, S.-Y., Kolch, W., Wolkenhauer, O., 2003. Experimental
design in systems biology, based on parameter sensitivity analysis
using a Monte Carlo method: a case study for the TNFalpha-mediated
NF-kappaB signal transduction pathway. Simulation 79, 726-739.
Clark, L.H., Setzer, R.W., Barton, H.A., 2004. Framework for evaluation
of physiologically-based pharmacokinetic models for use in safety or
risk assessment. Risk Analysis 24, 1697-1717.
Clewell 3rd, H.J., Andersen, M.E., Barton, H.A., 2002a. A consistent
approach for the application of pharmacokinetic modeling in cancer
and noncancer risk assessment. Environmental Health Perspectives
110, 85-93.
Clewell 3rd, H.J., Andersen, M.E., Wills, R.J., Latriano, L., 1997. A
physiologically based pharmacokinetic model for retinoic acid and its
metabolites. Journal of the American Academy of Dermatology 36,
S77-S85.
Previous
-------
410
G. Loizou et al. I Regulatory Toxicology and Pharmacology 50 (2008) 400^tll
Clewell, H.J., Gearhart, J.M., Gentry, P.R., Covington, T.R., VanLand-
ingham, C.B., Crump, K.S., Shipp, A.M., 1999. Evaluation of the
uncertainty in an oral reference dose for methylmercury due to
interindividual variability in pharmacokinetics. Risk Analysis 19, 547-
558.
Clewell, H.J., Teeguarden, J., McDonald, T., Sarangapani, R., Lawrence,
G., Covington, T., Gentry, R., Shipp, A., 2002b. Review and
evaluation of the potential impact of age- and gender-specific
pharmacokinetic differences on tissue dosimetry. Critical Reviews in
Toxicology 32, 329-389.
Clewell III, H.J., Andersen, M.E., 1987. Dose, species and route
extrapolation using physiologically-based pharmacokinetic modeling.
Drinking Water and Health 8, 159-184.
Cobelli, C, Carson, E.R., Finkelstein, L., Leaning, M.S., 1984. Validation
of simple and complex models in physiology and medicine. American
Journal of Physiology 246, R259-R266.
Cox Jr., L.A., 1996. Reassessing benzene risks using internal doses and
Monte-Carlo uncertainty analysis. Environmental Health Perspectives
104 (Suppl. 6), 1413-1429.
Csanady, G.A., Filser, J.G., 2007. A physiological toxicokinetic model for
inhaled propylene oxide in rat and human with special emphasis on the
nose. Toxicological Sciences 95, 37-62.
DeJongh, J., Forsby, A., Houston, J.B., Beckman, M., Combes, R.,
Blaauboer, B.J., 1999. An integrated approach to the prediction of
systemic toxicity using computer-based biokinetic models and biolog-
ical in vitro test methods: overview of a prevalidation study based on
the ECITTS project. Toxicology In Vitro 13, 549-554.
Delic, J.I., Lilly, P.D., MacDonald, A.J., Loizou, G.D., 2000. The utility
of PBPK in the safety assessment of chloroform and carbon
tetrachloride. Regulatory Toxicology and Pharmacology 32, 144-155.
Dewoskin, R.S., 2007. PBPK models in risk assessment-A focus on
chloroprene. Chemico-Biological Interactions 166, 352-359.
European Chemicals Bureau. Existing Substances Lists. .
European Chemicals Bureau (RIP 3.3). .
European Commission (FOCUS), .
European Community Regulation (Council Directive 91/414). .
European Community Regulation (REACH), .
Faller, T.H., Csanady, G.A., Kreuzer, P.E., Baur, C.M., Filser, J.G., 2001.
Kinetics of propylene oxide metabolism in microsomes and cytosol of
different organs from mouse, rat, and humans. Toxicology and
Applied Pharmacology 172, 62-74.
Franks, S.J., 2005. A mathematical model for the absorption and
metabolism of formaldehyde vapour by humans. Toxicology and
Applied Pharmacology 206, 309-320.
Franks, S.J., Spendiff, M.K., Cocker, J., Loizou, G.D., 2006. Physiolog-
ically based pharmacokinetic modelling of human exposure to 2-
butoxyethanol. Toxicology Letters 162, 164-173.
Georgopoulos, P.G., Roy, A., Gallo, M.A., 1994. Reconstruction of
short-term multi-route exposure to volatile organic compounds using
physiologically based pharmacokinetic models. Journal of Exposure
Analysis and Environmental Epidemiology 4, 309-328.
Gorlitz, G., 1993. Rules for the correct performance and evaluation of
model calculations for simulation of the environmental behaviour of
pesticides (Biologische Bundesanstalt fur Land- und Forstwirtschaft,
Fraunhofer Institut fur Umweltchemie und Okotoxikologie and
Arbeitsgruppe "Simulationsmodelle" in Industrieverband Agrar and
the Umweltbundesamt, eds.). BayerCropScience AG, Internal
Report.
Government of Canada, 1993. Priority Substances Assessment Report
on Dichloromethane. .
Gueorguieva, I., Aarons, L., Ogungbenro, K., Jorga, K.M., Rodgers, T.,
Rowland, M., 2006. Optimal design for multivariate response phar-
macokinetic models. Journal of Pharmacokinetics and Pharmacody-
namics 33, 97-124.
Gulden, M., Seibert, H., 2003. In vitro-in vivo extrapolation: estimation
of human serum concentrations of chemicals equivalent to cytotoxic
concentrations in vitro. Toxicology 189, 211-222.
Gundert-Remy, U., Sonich-Mullin, C., 2002. The use of toxicokinetic and
toxicodynamic data in risk assessment: an international perspective.
Science of the Total Environment 288, 3-11.
Hays, S.M., Becker, R.A., Leung, H.W., Aylward, L.L., Pyatt, D.W.,
2007. Biomonitoring equivalents: a screening approach for interpreting
biomonitoring results from a public health risk perspective. Regulatory
Toxicology and Pharmacology 47, 96-109.
Health Canada (Domestic Substances List), .
Health Canada (Priority Substances Assessment Program). .
Houston, J.B., 1994. Relevance of in vitro kinetic parameters to
in CTuometabolism of xenobiotics. Toxicology In Vitro 8, 507-
512.
INERIS, 2007. Reprotoxicity of Ethylene Glycol Ethyl Ether (EGEE) in
Humans—Development of a Dose-Response Relationship.
IPCS, 2005. Chemical-Specific Adjustment Factors (CSAFs) for Interspe-
cies Differences and Human Variability: Guidance Document for Use of
Data in Dose/Concentration-Response Assessment, WHO, Geneva, p.
96. .
Kedderis, G.L., 1997. Extrapolation of in vitro enzyme induction data to
humans in vivo. Chemical and Biological Interactions 107, 109-121.
Lee, M.S., Faller, T.H., Kreuzer, P.E., Kessler, W., Csanady, G.A., Putz,
C., Rios-Blanco, M.N., Pottenger, L.H., Segerback, D., Osterman-
Golkar, S., Swenberg, J.A., Filser, J.G., 2005. Propylene oxide in
blood and soluble nonprotein thiols in nasal mucosa and other tissues
of male Fischer 344/N rats exposed to propylene oxide vapors—
relevance of glutathione depletion for propylene oxide-induced rat
nasal tumors. Toxicological Sciences 83, 177-189.
Lipscomb, J.C., Fisher, J.W., Confer, P.D., Byczkowski, J.Z., 1998.
In vitro to in vivo extrapolation for trichloroethylene metabolism
in humans. Toxicology and Applied Pharmacology 152,
376-387.
Meek, B., Renwick, A., Sonich-Mullin, C., 2003a. Practical application of
kinetic data in risk assessment—an IPCS initiative. Toxicology Letters
138, 151-160.
Meek, M.E., Bucher, J.R., Cohen, S.M., Dellarco, V., Hill, R.N.,
Lehman-McKeeman, L.D., Longfellow, D.G., Pastoor, T., Seed, J.,
Patton, D.E., 2003b. A framework for human relevance analysis of
information on carcinogenic modes of action. Critical Reviews in
Toxicology 33, 591-653.
Meek, M.E., Renwick, A. (Eds.), 2006. Guidance for the development of
chemical specific adjustment factors—integration with mode of action
frameworks. Informa Healthcare, New York.
Meek, M.E., Renwick, A., Ohanian, E., Dourson, M., Lake, B.,
Naumann, B.D., Vu, V., 2001. Guidelines for application of chemical
specific adjustment factors (CSAF) in dose/concentration response
assessment. Comments on Toxicology 7, 575-590.
Meek, M.E., Renwick, A., Ohanian, E., Dourson, M., Lake, B.,
Naumann, B.D., Vu, V., 2002. Guidelines for application of chemi-
cal-specific adjustment factors in dose/concentration-response assess-
ment. Toxicology 181-182, 115-120.
Miners, J.O., Veronese, M.E., Birkett, D.J., 1994. In vitro approaches for
the prediction of human drug metabolism. Annual Reports in
Medicinal Chemistry 29, 307-316.
Nestorov, I.A., Aarons, L.J., Arundel, P.A., Rowland, M., 1998. Lumping
of whole-body physiologically based pharmacokinetic models. Journal
of Pharmacokinetics and Biopharmaceutics 26, 21-46.
OECD, 2007. Guidance on the Validation of (Quantitative) Structure
Activity Relationship [(Q)SAR] Modelling, Paris. In: Environment
Previous
-------
G. Loizou et al. I Regulatory Toxicology and Pharmacology 50 (2008) 400-411
411
and Health Publications Series on Testing and Assessment. Organiza-
tion for Economic Cooperation and Development, Paris.
Oreskes, N., 1998. Evaluation (not validation) of quantitative models.
Environmental Health Perspectives 106 (Suppl. 6), 1453-1460.
OSHA, 1997. Occupational Exposure to Methylene Chloride; Final Rule
20 CFR Parts 1910, 1915 and 1926 70FR1493-1619. (Accessed August 1, 2007).
Osterman-Golkar, S., Czene, K., Lee, M.S., Faller, T.H., Csanady, G.A.,
Kessler, W., Perez, H.L., Filser, J.G., Segerback, D., 2003. Dosimetry
by means of DNA and hemoglobin adducts in propylene oxide-
exposed rats. Toxicology and Applied Pharmacology 191, 245-254.
Portier, C.J., Lyles, C.M., 1996. Practicing safe modeling: GLP for
biologically based mechanistic models. Environmental Health Per-
spectives 104, 806.
Rescigno, A., Beck, J.S., 1987. The use and abuse of models. Journal of
Pharmacokinetics and Biopharmaceutics 15, 327-344.
Rostami-Hodjegan, A., Tucker, G.T., 2004. 'In silico' simulations to
assess the 'in vivo' consequences of 'in vitro' metabolic drug-drug
interactions. Drug Discovery Today: Technologies 1, 441-448.
Rowland, M., Balant, L., Peck, C., 2004. Physiologically based pharma-
cokinetics in drug development and regulatory science: a workshop
report (Georgetown University, Washington, DC, May 29-30, 2002).
American Association of Pharmaceutical Scientists 6, 1-12.
Seed, J., Carney, E.W., Corley, R.A., Crofton, K.M., DeSesso, J.M.,
Foster, P.M., Kavlock, R., Kimmel, G., Klaunig, J., Meek, M.E.,
Preston, R.J., Slikker Jr., W., Tabacova, S., Williams, G.M., Wiltse, J.,
Zoeller, Fenner-Crisp, P., Patton, D.E., 2005. Overview: using mode of
action and life stage information to evaluate the human relevance of
animal toxicity data. Critical Reviews in Toxicology 35, 664-672.
Sonich-Mullin, C., Fielder, R., Wiltse, J., Baetcke, K., Dempsey, J.,
Fenner-Crisp, P., Grant, D., Hartley, M., Knaap, A., Kroese, D.,
Mangelsdorf, I., Meek, E., Rice, J.M., Younes, M., 2001. IPCS
conceptual framework for evaluating a mode of action for chemical
carcinogenesis. Regulatory Toxicology and Pharmacology 34, 146-
152.
US EPA, 1987. Update to the Health Assessment Document and
Addendum for Dichloromethane (Methylene Chloride): Pharmacoki-
netics, Mechanism of Action, and Epidemiology. External Review
Draft, EPA/600/8-87/030A.
US EPA, 1989. Risk Assessment Guidance for Superfund, vol 1. Human
Health Evaluation Manual (Part A), EPA/540/1-89/002.
US EPA, 2006. Approaches for the Application of Physiologically Based
Pharmacokinetic (PBPK) Models and Supporting Data in Risk
Assessment (Final Report). EPA/600/R-05/043A.
Verwei, M., van Burgsteden, J.A., Krul, C.A., van de Sandt, J.J., Freidig,
A.P., 2006. Prediction of in vivo embryotoxic effect levels with a
combination of in vitro studies and PBPK modelling. Toxicology
Letters.
Wilson, Z.E., Rostami-Hodjegan, A., Burn, J.L., Tooley, A., Boyle, J.,
Ellis, S.W., Tucker, G.T., 2003. Inter-individual variability in levels of
human microsomal protein and hepatocellularity per gram of liver.
British Journal of Clinical Pharmacology 56, 433^140.
Previous
-------
APPLICATIONS NOTE
Vol. 25 no. 5 2009, pages 692-694
doi: 10.1093/bioinformatics/btp042
Databases and ontologies
DSSTox chemical-index files for exposure-related experiments
in ArrayExpress and Gene Expression Omnibus: enabling
toxico-chemogenomics data linkages
ClarLynda R. Williams-DeVane1'*, Maritja A. Wolf2 and Ann M. Richard1
1 National Center for Computational Toxicology, Office of Research and Development, US EPA and
2Lockheed Martin, Research Triangle Park, NC 27711, USA
Received on October 31, 2008; revised on January 12, 2009; accepted on January 18, 2009
Advance Access publication January 21, 2009
Associate Editor: Alex Bateman
ABSTRACT
Summary: The Distributed Structure-Searchable Toxicity (DSSTox)
ARYEXP and GEOGSE files are newly published, structure-annotated
files of the chemical-associated and chemical exposure-related
summary experimental content contained in the ArrayExpress
Repository and Gene Expression Omnibus (GEO) Series (based on
data extracted on September 20, 2008). ARYEXP and GEOGSE
contain 887 and 1064 unique chemical substances mapped to
1835 and 2381 chemical exposure-related experiment accession
IDs, respectively. The standardized files allow one to assess,
compare and search the chemical content in each resource, in the
context of the larger DSSTox toxicology data network, as well as
across large public cheminformatics resources such as PubChem
(http://pubchem.ncbi.nlm.nih.gov).
Availability: Data files and documentation may be accessed online
at http://epa.gov/ncct/dsstox/.
Contact: williams.clarlynda@epa.gov
Supplementary information: Supplementary data are available at
Bioinformatics online.
1 INTRODUCTION
In recent years, the number of publicly available gene expression,
toxicology and cheminformatics resources with the potential to
support toxicogenomics investigation has grown considerably
(http://www.microarryaworld.com/DatabasePage.html; Richard
et al., 2008; C.R.Williams-DeVane et al., manuscript submitted).
These trends are encouraging aggregation and use of data in a
much broader context, spanning domains of inquiry in relation to
toxicology, chemistry and genomics (Waters et al., 2008).
The European Bioinformatics Institute's (EBI) ArrayExpress
Repository (http://www.ebi.ac.uk/microarray-as/ae/) and the National
Center for Biotechnology Information's (NCBI) Gene Expression
Omnibus (GEO) (http://www.ncbi.nlm.nih.gov/geo/) are the two main
public repositories for gene expression experiments associated
with the published scientific literature. Although they each support
MIAME-compliant submissions (i.e. adhering to guidelines
for Minimum Information about Microarray Experiments;
http://www.mged.org/Workgroups/MTAME/miame.html), neither
*To whom correspondence should be addressed.
resource has standardized requirements for reporting of chemical
information associated with submitter-deposited microarray
experiments. As a result, not only has it been difficult to assess the
chemical-related content within these resources, but also microarray
data have been effectively isolated from rapidly growing public
sources of chemically indexed information pertaining to toxicology
(Richard et al, 2006).
We report here the publication of chemical-index files for
experimental content in the ArrayExpress Repository and GEO
Series (data extracted on September 20, 2008), in association
with the Environmental Protection Agency's (EPA) Distributed
Structure-Searchable Toxicity (DSSTox) Data Network project
(http://www.epa.gov/ncct/dsstox/) (Supplementary References).
2 DATABASE METHODS AND COMPONENTS
2.1 DSSTox
The DSSTox project publishes high-quality, standardized chemical
structure toxicity data files pertaining to high-interest chemicals
for environmental toxicology and of potential use for structure-
activity relationship (SAR) modeling. The DSSTox website
offers documentation, freely downloadable structure data
files (SDF) and tabular data files (.xls) for each published
Data File (http://www.epa.gov/ncct/dsstox/DataFiles.html). A
unique aspect of this effort is the quality annotation, review
and representation of chemical information both in terms of
a unique mapping to a curated chemical structure, as well
as at the generic test substance level (similar to Chemical
Abstracts Service (CAS) Registry Number distinctions)—see
http://www.epa.gov/ncct/dsstox/MoreonStandardChemFields.html.
The current DSSTox inventory contains over 8000 unique
chemicals and has been incorporated into the online DSSTox
Structure-Browser (http://www.epa.gov/dsstox_structurebrowser/),
the NCBI PubChem (http://pubchem.ncbi.nlm.nih.gov/) inventory
containing millions of searchable chemical structures and
thousands of bioassays, ChemSpider (http://chemspider.com/)
containing millions more chemical structures, properties and
linkages and the new EPA Aggregated Computational Toxicology
Resource (ACToR) database (http://www.epa.gov/actor/) providing
searchability and comparative read-across for over 200 chemical
inventories specifically pertaining to environmental toxicology.
692 © The Author 2009. Published by Oxford University Press. All rights reserved. For Permissions, please email: journals.permissions@oxfordjournals.org
Previous
TOC
-------
Chemical-indexing of gene expression experiments
2.2 ArrayExpress
Since the public launch of ArrayExpress in 2002, the repository
has grown to more than 6500 experiments. Public microarray
data in ArrayExpress are available for browsing and querying
across a wide range of experiment properties, including array
type, submitter, species, MIAME score, etc. and complete datasets
or subsets can be retrieved (http://www.ebi.ac.uk/microarray-
as/aer/entry). Although the TOXM label is available to designate
toxicogenomics experiments, it is rarely used and represents a
very small portion of the total ArrayExpress chemical-experiment
inventory. Chemical involvement is primarily indicated by non-
standard, error-prone chemical names or abbreviations included
in free-text user description fields, and these are very rarely
accompanied by chemical identifiers such as CAS or Chemical
Entities of Biological Interest (ChEBI) numbers. Where ChEBI
identifiers are used (~20 instances), these have been recently cross-
referenced within the ChEBI system (http://www.ebi.ac.uk/chebi/).
2.3 GEO series
A GSE is a GEO Series Accession ID (e.g. GSE5594) that defines a
set of related samples considered to be part of a single experiment
and, for present purposes, most closely (or precisely) corresponds to
an ArrayExpress Accession ID (e.g. E-GEOD-5594). GEO currently
contains over 9900 Series Accession entries and more than 2000
curated GEO Datasets. However, fewer than 6500 of the Series
Accession entries could be programmatically extracted due to a
backlog in the GEO curation process at the time of annotation.
Chemical information is most often located in GEO Series records
as chemical names or abbreviations in the Summary field (a user-
submitted, free-text description field), but in some cases this
information only was provided in the 'Title' or 'Samples' field.
A handful of GEO records contained chemical identifiers such as
CAS, but the quality of chemical annotation across GEO, in general,
was poor and in some cases absent entirely.
2.4 Methods: ARYEXP_Aux and GEOGSE_Aux
Recent additions to ArrayExpress include new portals for
programmatic access where users can query and download data
in a systematic manner from the ArrayExpress FTP site (http://
www.ebi.ac.uk/microarray/doc/help/programmatic_access.html).
To create the ARYEXP auxiliary file (ARYEXP_Aux), content
was extracted using Perl scripts from the programmatic access
FTP site in XML format. Similarly, the creation of GEOGSE_Aux
involved first programmatically accessing GEO to retrieve all
current experimental descriptions associated with Series records
(http://www.ncbi.nlm.nih.gov/projects/geo/info/geo_paccess.html).
Entrez tools were used to generate an XML document containing
a summary of each of the GEO Series experiments currently
curated in the GEO Data Sets Accession ID (GDS) system.
A series of Perl scripts were developed to parse both the
ArrayExpress and GEO XML documents and to retrieve
records with probable chemical association. The resulting
experiment descriptions were evaluated and verified manually, and
chemical information was extracted and subsequently underwent
stringent review and annotation according to DSSTox procedures
(http://www.epa.gov/ncct/dsstox/ChemicalInfQAProcedures.html).
Each of the resulting DSSTox files, ARYEXP_Aux and
GEOGSE_Aux, is a chemical-experiment pair index (one record
per chemical per experiment). Each file also contains the full
complement of 20 DSSTox Standard Chemical Fields and an
additional 14 Standard Genomics Fields (including URL field
linking to experiment accession IDs) to allow cross comparisons
of ArrayExpress and GEO content similarly indexed by DSSTox
(Supplementary Tables 1 and 2). In addition, ARYEXP_Aux
contains an additional set of 30 experimental description
fields specific to ArrayExpress (Supplementary Table 3), and
GEOGSE_Aux contains an additional set of four experimental
description fields specific to GEO Series (Supplementary Table 4).
Further details of chemical-indexing and data extraction methods,
comparison of ArrayExpress and GEO chemical-experimental
summary content and assessment of lexicologically relevant
chemical content are provided elsewhere (C.R.Williams-Devane,
manuscript submitted; Supplementary References).
2.5 Methods: ARYEXP and GEOGSE
Of greatest toxicogenomics and SAR interest are those experiments
for which a chemical treatment and the resulting gene expression
changes are the primary focus of the experiment. We introduced
the Standard Genomics Field, Chemical_StudyType, to annotate the
purpose of the chemical in the experiment, which could include
uses such as 'Treatment', 'Vehicle', 'Reference', 'Media', etc. The
main DSSTox files, ARYEXP and GEOGSE, pertain only to the
'Treatment' category of ArrayExpress and GEO Series experiments
and contain one record per unique chemical substance. In ARYEXP
and GEOGSE, one chemical substance can map to one or more
experiments in GEO or ArrayExpress. Hence, unlike the Auxilliary
files, where records map to individual experiments, the main
DSSTox structure-index files do not contain summary details of
particular experiments. Rather, these files contain DSSTox Standard
Chemical Fields, Chemical_StudyType (which can be Treatment
AND other conditions), one or more Experiment Accession IDs and
the corresponding URLs to ArrayExpress or GEO Series Experiment
Summary pages.
3 CONCLUSIONS AND PERSPECTIVES
The ARYEXP_Aux file contains a total of 2365 chemical-
experiment records (with 44 total source fields), corresponding
to 1011 unique chemical substances. Of these 2365 chemical-
experiment pairs, 1835 were identified as 'Treatment' and these map
to 887 unique chemical records in the ARYEXP file. Similarly, the
GEOGSE_Aux file contains a total of 2381 chemical-experiment
records (with 18 total source fields), corresponding to 1064 unique
chemical substances. Of these 2381 chemical-experiment pairs,
2134 were identified as 'Treatment' and these map to 1014
unique chemical records in the GEOGSI file. These numbers
indicate that the exposure-related experimental content in these
two public resources covers a significant range of chemicals of
potential interest and utility for toxicogenomics investigations. All
four files are available for download from the DSSTox website
(http://www.epa.gov/ncct/dsstox/).
The ARYEXP and GEOGSE chemical-index files, with
associated ArrayExpress Experiment Accession ID URLs, have
been incorporated into the DSSTox Structure-Browser, ACToR and
PubChem. This enables ArrayExpress and GEO Series experiments
to be structure located from these public resources, creating the
693
Previous
TOC
-------
C.R.Williams-DeVane et al.
Chemical Search Paradigm
Urerlptton {Olft£ta|)
Acetaminophen
•"V.
E-TABM-131 Custom Array, Rat
E-MEXP-82: Affymetrix; Rat
E-TOXM-18. Agilent; Mouse
E-TOXM-31.Custom Array, Human
SEPA DSSTox
CPDBAS EPAFHIJ
FDAMDD NTPBSI
CEBS
Rodent carzinogenicsty
Fish acute toxscrty
Genetic toxicity
Cell viability toxicity
JZJChemSpider
Publgphem
200005594 Aoj lent. Rat
200005593 Agilent, Rat
200005595 AgilenlRat
200000633: Custom Array, Rat
200005$52;A9iient. Rat
2000WS74:Custom. Mouse
200005860:Agi!ent Rat
200008858 GE Healthcare.Rat
Acetaminophen
M eta-Data Set
Fig. 1. Illustration showing linkages from DSSTox Structure-Browser to
experiments from ARYEXP and GEOGSE, linked to various chemically
indexed resources of bioassay and toxicity data for constructing a meta-
dataset for acetaminophen; links to actual microarray data are provided by
GEO or ArrayExpress (dashed arrows indicate future linkages).
microarray data or annotation files. To encompass these data
types, integration of ArrayExpress and GEO into the Chemical
Effects in Biological Systems (CEBS; http://cebs.niehs.nih.gov/)
toxicogenomics database is (Waters et al., 2008). An automated
process of porting microarray data and annotation files directly
from ArrayExpress and GEO to CEBS is to be implemented with
chemical annotation handled in collaboration with the DSSTox
project. Recommendations for chemical standards for microarray
experiments are being forwarded to the MIBBI (Minimum
Information for Biological and Biomedical Investigations) project
(http://www.mibbi.org/).
Better coordination betweenNCBI's GEO and PubChem projects
is needed. Likewise, EBI's ChEBI and ArrayExpress projects
have recently improved their coordination. The present effort has
chemically annotated a large portion of the current inventories of
ArrayExpress Repository and GEO Series, although more efficient
mechanisms for updates must be instituted. The preferred solution
is to incorporate standard chemical reporting requirements into
these resources directly. The present effort charts a path forward
and will be used to encourage implementation of these changes.
ArrayExpress currently has the ability to adequately capture most
of the required chemical information; however, depositors are
not doing so. We recommend that GEO and ArrayExpress, in
coordination with projects such as MIBBI and DSSTox, each
adopt formal requirements for a minimum level of chemical
annotation in relation to experiments (e.g. valid chemical name,
Chemical_StudyType). Once repositories and depositors recognize
the importance and enhanced capabilities to be gained from chemical
annotation and indexing, and make it a priority, the possibility to
more fully integrate and utilize existing data in toxicogenomics
studies can be realized. Updates to DSSTox GEOGSE and ARYEXP
files by current means are scheduled for February 2009.
capability to query multiple domains of data by chemical structure.
Figure 1 illustrates this capability with a chemical structure search
of acetaminophen. Summary toxicology and microarray results are
provided or located through the DSSTox structure search, and
other domain results are provided through linkages with PubChem,
ACToR and ChemSpider. With these new capabilities, a meta-dataset
on a particular chemical or family of structurally similar chemicals
could be constructed for further analysis.
The DSSTox ARYEXP and GEOGSE chemical-index files
have been deposited within PubChem such that chemicals can
be located by keyword search under 'PubChem Substance' on
the main search page (e.g. ARYEXP, ArrayExpress, etc.). From
the PubChem Substance results page, a user can link directly
to the chemical-associated experimental accession ID summary
pages in GEO and ArrayExpress. Likewise, through chemical
linkages, these microarray data could be placed in a much larger
data and chemical context (including linkage to data for structurally
and biologically similar chemicals) within PubChem. ARYEXP
and GEOGSE auxiliary data files contain summary experimental
factors (Supplementary Tables 3 and 4), but do not contain actual
ACKNOWLEDGEMENTS
This manuscript was approved by the US EPA's National Center
for Computational Toxicology for publication; the contents do not
necessarily reflect the views and policies of the EPA and mention of
trade names or commercial products does not constitute endorsement
or recommendation for use.
Funding: NCSU/EPA Cooperative Training Program in
Environmental Sciences Research, Training Agreement
CT833235-01-0 with North Carolina State University (to C.R.W.).
Conflict of Interest: none declared.
REFERENCES
Richard,A. et al. (2008) Toxicity data informatics: supporting a new paradigm for
toxicity prediction. Tox. Mech Meth, 18, 103-118.
Richard,A.M. et al. (2006) Chemical structure indexing of toxicity data on the Internet.
Curr. Opin. DrugDiscov. Dev., 9, 314-325.
Waters,M. et al. (2008) CEBS—chemical effects in biological systems: a public data
repository integrating study design and toxicity data with microarray and proteomics
data. Nucleic Acids Res., 36, D892-D900.
694
Previous
TOC
-------
Bioinformation
www.bioinformation.net
by Biomedical Informatics Publishing Group
open access
Hypothesis
Effect of single nucleotide polymorphisms on
Affymetrix® match-mismatch probe pairs
Eric Christian Rouchka1' *, Abhijit Waman Phatak1 and Amar Vir Singh 2'3
'Department of Computer Engineering and Computer Science, University of Louisville, Louisville, KY, USA; 2Department of Molecular, Cellular,
and Craniofacial Biology, University of Louisville, Louisville, KY, USA; 'Department of Botany and Industrial Microbiology, JV College, CCS
University, Meerut, UP, India; Eric Rouchka* - E-mail: eric.rouchka@louisville.edu; Phone: 502-852-1695; Fax: 502-852-4713;
* Corresponding author
received January 06, 2008; revised June 17, 2008; accepted July 03, 2008; published July 14, 2008
Abstract:
Microarrays provide a means of studying expression level of tens of thousands of genes by providing one or more oligonucleotide
probe(s) for each transcript studied. Affymetrix® GeneChip™ platforms historically pair each 25-base perfect match (PM) probe
with a mismatch probe (MM) differing by a complementary base located in the 13th position to quantify and deflate effects of cross-
hybridization. Analytical routines for analyzing these arrays take into account difference in expression levels of MM and PM probes
to determine which ones are useful for further study. If a single nucleotide polymorphism (SNP) occurs at the 13* base, a probe
with a higher MM expression level may be incorrectly omitted. In order to examine SNP affects on PM and MM expression levels,
known human SNPs from dbSNP were mapped to probe sets within the Affymetrix® HG-U133A platform. Probe sets containing
one or more probe pairs with a single SNP at the 13th position were extracted. A set of twelve microarray experiments were
analyzed for the PM and MM expression levels for these probe sets. Over 6,000,000 human SNPs and their flanking regions were
extracted from dbSNP. These sequences were aligned against each of the 247,965 probe pair sequences from the Affymetrix® HG-
U133A platform. A total of 915 probe sets containing a single probe sequence with a SNP mapped to the 13th base were extracted.
A subset containing 166 probe sets result in complementary base SNPs. Comparison of gene expression levels for the SNP to non-
SNP PM and MM probes does not yield a significant difference using %2 analysis. Thus, omission of probes with MM expression
levels higher than PM expression levels does not appear to result in a loss of information concerning SNPs for these regions.
Keywords: Affymetrix HG-U133A; single nucleotide polymorphism; microarray; probe; mismatch
Background:
Microarray technology
Technological breakthroughs within the past couple of
decades have changed the face of molecular biology by
allowing researchers to generate large volumes of biologically
relevant data in a short period of time. These advances have
led to the "omics" era of research [1] marked by genomics
(study of genomes), proteomics (study of protein expression),
cellomics (study of the cell), and transcriptomics (study of
transcribed regions). Transcriptomics has been aided by the
invention of the microarray [2] which allows researchers to
study patterns of gene expression across tens of thousands of
genes simultaneously. Major companies providing
commercial solutions include Affymetrix®, Agilent®, and
CodeLink®. Each of these approaches provides one or more
short oligonucleotide probe(s) sequence complementary to the
product of the transcript of interest.
The Affymetrix® oligonucleotide platforms are constructed to
allow multiple oligonucleotide probes per probe set, where
each probe set represents a single gene or transcript. For the
HU-133A platform for studying human transcripts, there are
over 22,000 different probe sets represented, with each non-
control probe set containing 11 25-base oligonucleotide
sequence probes [3]. In order to help quantify and control the
ISSN 0973-2063 405
Bioinformation 2(9): 405-411 (2008)
effects of cross-hybridization, the Affymetrix® approach
groups probes into pairs consisting of a perfect match probe
(PM) and a mismatch probe (MM). The perfect match probe
is a 25 base oligonucleotide complementary to the transcript
and the mismatch probe is the same as the PM with the
exception that the 13th base is complementary to the
corresponding position in the PM set. For example, one of the
eleven probe pairs for the 206055_s_at probe set is as follows:
PM GCACAGCTTGCAAAGGATATTGCCA
MM GCACAGCTTGCATAGGATATTGCCA
Figure 1 shows an example of the expression levels of the
MM and PM probes for three Affymetrix® probe sets found
within the HU-133A platform.
Since mismatch data allows for detection of cross-
hybridization, a probe set could be selected for inclusion or
exclusion based on the corresponding match/mismatch values.
For the probe set 206055_s_at in Figure 1, each of the probe
pairs could be used since the expression values of the match is
consistently higher than the value of the corresponding
mismatch probe located directly below. However, for probe
set 219820_at in Figure 1, the fifth match/mismatch pair from
Bioinformation, an open access forum
© 2008 Biomedical Informatics Publishing Group
Previous
TOC
-------
Bioinformation
www.bioinformation.net
by Biomedical Informatics Publishing Group
open access
Hypothesis
the left is potentially excluded since the mismatch expression
value is much greater than the match expression value. This
probe pair would be excluded since the resulting differences
in expression level is thought to be due to cross hybridization.
HflTCH
niSHHTCH
Affymetrix ID: 2Q6Q55_s_at
NOTCH
MIEMflTCH
Affymetrix ID: 208913_at
NOTCH
MISMATCH
Affymetrix ID: 219820_at
Figure 1: Affymetrix probe set pair expression levels. Shown are the expression levels for three separate Affymetrix probe sets
represented on a 0-255 color scale. Each probe set contains eleven probe pairs, with each pair represented by a match and
mismatch sequence.
Single nucleotide polymorphisms (SNPs)
Single nucleotide polymorphisms (SNPs, often pronounced as
snips) are single nucleotide base differences in a specific
position of genomic DNA among two different individuals of
the same species. SNPs are the most common form of genetic
variation that helps to differentiate individuals in a
population. A number of diseases and abnormalities,
including sickle cell anemia [4], cystic fibrosis [5], muscular
dystrophy [6], type II diabetes [7], and migraine headaches
[8] are influenced by the presence of SNPs occurring within
gene coding regions.
The rate of occurrence of SNPs in the human genome is
around one every 100 to 300 base pairs. The National Center
for Biotechnology Information (NCBI) maintains a publicly
available database of annotated human SNPs, known as
dbSNP [9]. The current build of dbSNP (build 127) contains
nearly 12 million annotated human SNPs.
While SNPs are important in disease association studies, their
presence becomes problematic for genome wide analysis. As
an example, one of the difficulties with the Affymetrix®
microarray platforms is that each of the chips are designed to
be representative for all individuals within an organism-level
classification. However, with the high frequency of SNPs, it is
ISSN 0973-2063 406
Bioinformation 2(9): 405-411 (2008)
possible that a SNP locus is found within a particular probe
sequence. This becomes especially problematic if the locus
corresponds to the 13th base pair, and the SNP variant is the
complementary base. Such a case would result in a higher
hybridization rate for the mismatch probe as opposed to the
match probe.
The independent and dependent effect of both SNPs and copy
number variants (CNVs) on gene expression has been known
to be an issue when studying microarrays [10]. The
development of SNP chips [11] has made it possible to
genotype SNPs and has led to the real possibility of Whole
Genome Association Studies (WGAS). However, a large
number of gene expression studies using microarray probe
technology exist that might label certain probes for exclusion
due to higher MM hybridization rates that is actually due to
the presence of complementary SNPs.
In order to test for the effect of SNPs on probe hybridization,
we looked at all 247,965 match probes within the
Affymetrix® HU-133A platform and compared them against
dbSNP to see which probes contained SNP loci within them.
For those where the sole SNP loci was found at the 13th base,
we compared the expression levels of the PM probe to the
MM probe for each of the probes within the probe set. We
Bioinformation, an open access forum
© 2008 Biomedical Informatics Publishing Group
Previous
TOC
-------
Bioinformation
www.bioinformation.net
by Biomedical Informatics Publishing Group
open access
Hypothesis
specifically wanted to see if the MM probe with the SNP at
the 13th base varied more than those probes that did not
contain any SNP loci.
Methodology:
Data acquisition
For the purposes of this study, three key data components
were required: human genomic data, human SNP data, and
probe sequence data. Human genomic sequence was obtained
from the University of California-Santa Cruz's goldenpath
web site (http://goldenpath.ucsc.edu) [12, 13] for the hg!7
build of the human genome. The resulting data was contained
in 27 files. Human SNPs were downloaded from build 124 of
the dbSNP database [14] maintained by the National Center
for Biotechnology Information (NCBI). This data is itself
based on build 33 of the NCBI Human genome. Probe
sequence data representing the 247,965 twenty-five base
perfect match oligomers from the HG-U133A microarray
manufactured by Affymetrix® were downloaded from the
netaffx utility on the Affymetrix® web site
http://www.affymetrix.com/.
Expression levels of probe sequences containing SNPs were
compared within a set of twelve samples for the GEO [15]
record GDS1758. This dataset originates from a study on the
developmental pathway involved in pterygium, an ocular
surface disorder within a sample set of Chinese patients [16].
Rather than focus on a large dataset with a mixture of patients
from different ethnicities, this smaller dataset was chosen
from a single ethnic group so that ethnic specific major and
minor allele frequencies could be determined. The individual
CEL files used are labeled GSM48026.CEL,
GSM48027.CEL, GSM48028.CEL, GSM48029.CEL,
GSM48030.CEL, GSM48031.CEL, GSM48032.CEL,
GSM48033.CEL, GSM48034.CEL, GSM48035.CEL,
GSM48036.CEL, and GSM48037.CEL.
Preprocessing
Perfect match probe sequences for the HG-U133A platform
were stored in a tab-delimited format with information
concerning the probe set name, probe x position, probe y
position, interrogation position, probe sequence, and
strandedness (Table 1 under supplementary material). The
resulting tab-delimited files were parsed using perl scripts to
reconstruct sequence files in FASTA format for sequence
comparison.
Sequences originating from the dbSNP database each
represent a single instance of a known SNP denoted by the
standard IUPAC-IUB code [17]. dbSNP sequences typically
range from a few hundred to a few thousand bases in length.
For the purpose of our study we were only interested in the
sequence immediately surrounding the SNP for alignment
with the twenty-five base oligomer sequence from the HG-
U133A microarray. A perl script was created to extract a
forty-nine base segment from each dbSNP sequence, spanning
twenty-four bases upstream and downstream of the SNP
location, when available. In some cases, the allele position or
ISSN 0973-2063 407
Bioinformation 2(9): 405-411 (2008)
length of the original sequence did not allow for all of the
bases to be extracted.
The original downloaded dbSNP sequences were soft-masked
for low-complexity regions and tandem repeats. While this
can be beneficial in order to remove regions of low
significance and to avoid spurious sequence hits, our study
required us to unmask the source data in order to produce
exact alignments with microarray probe sequences. Sequences
were thus restored to their original format for sequence
alignment purposes. The resulting unmasked data was verified
by comparison between the original and truncated data.
Sequence Alignment
Alignments between the microarray oligomer probes and the
dbSNP sequences were performed using the nucleotide-
nucleotide comparison tool wublastn from the WU-BLAST
2.0 suite of programs [18, 19]. The dbSNP database was
formatted into a BLASTable database using the xdformat
utility, leaving the microarray probes as the query sequences.
Since the sequences were expected to be exact matches with
the exception of any SNPs present, ungapped alignments were
performed. This has the additional benefit of decreasing
search time. A word size of eight was used to allow for
alignments with up to two mismatches within a 25 base
alignment. A score cutoff of 95 was used to allow for a
combination of two mismatches/gaps within a 25 base
alignment using a scoring scheme of +5/-4 for
matches/mismatches and -10 for gap open penalty. The
remaining parameters were set at the default values. In
summary, the blast command line was as follows: blastn
-nogaps -S=95 -W=8. To further
maintain the focus of the project, the parameterized wublastn
results were filtered through a Perl script to only store those
alignments that were at least 22 bases in length.
Parsing and storing the results
The wublastn searches were conducted chromosome-wise,
keeping the structure of the source data intact, wublastn
output was piped through Perl scripts to filter out the basic
statistical information required for a database table. A Perl
script incorporating BPlite [20] was used to further parse the
output to store alignments of 22 or more bases with the
following information stored in plain text files are as follows:
1. Reference numbers of the query and target
sequences.
2. Sequence locations on the microarray and on the
dbSNP and genomic databases.
3. Location of the SNP within a dbSNP segment.
4. Lengths of the query and target sequences.
5. Start and end positions of the alignments found.
6. Aligned segment pairs.
7. Alignment string itself, which is a means of
depicting the matched and mismatched base pairs
within a sequence. Matching pairs have a '|'
between them, mismatches remain blank and where
a base in the query sequence matches any one of the
possible variations of the SNP, a plus sign ('+') is
used to show this 'partial' match.
Bioinformation, an open access forum
© 2008 Biomedical Informatics Publishing Group
Previous
TOC
-------
Bioinformation
www.bioinformation.net
by Biomedical Informatics Publishing Group
open access
Hypothesis
8. Length of the alignment found.
9. Number of matches within the sequence.
10. Percent identity of the matches (number of matches
divided by the alignment length).
11. Raw score of the alignment from the standard
scoring scheme of wublastn.
Percentage of observations of MM > PM
—0— ALL-NoSNP
—•— ALL-SNP
—_— COUPLEMEf/TARY MoSNP
COMPLEMENTARY SNP
0 1
45678
Number of MM > PM
10 11 12
Figure 2: Percentage of observations of mismatch probe expression greater than perfect match probe expression.
A MySQL database was created to store the parsed results.
The database schema consists of six tables. One of these
tables captures the database information, and a second is used
for the organism records. The remaining four tables captured
alignment data: one to hold the identification for the
microarray probes; a second to hold the identification
information for the segments from the dbSNP files; a third to
store the alignment data such as the source and target
alignment strings; and the fourth to store the statistical data
corresponding to the alignments. A shared key field was
generated for each of the tables using the chromosome and
alignment number.
Discussion:
SNP and probe alignments
Over six million ungapped alignments were found between
microarray probes and SNP segments. These resulted in a
total of 45,984 perfect match sequences between the probes
and SNP segments. An additional of 1,656 probe sequence
alignments result in a mismatch nucleotide in the 13th base.
Further filtering yields 915 alignments where the probe
sequence contains only a single SNP, and that probe is the
only one within its corresponding probe group to contain any
known SNPs. Of these 915, a subset of 166 results in a
ISSN 0973-2063 408
Bioinformation 2(9): 405-411 (2008)
complementary base mismatch. Bearing in mind that
Affymetrix® microarray s have pairs of probes where one half
of the pair has the complement of the other's 13th base, these
probes were marked for further analysis. A total of 58,505
sequences with a single mismatch were detected, but not
included in our analysis.
The 166 alignments resulting in a single, complementary base
mismatch originate from unique probe sets. Those probe pairs
containing a region where a SNP is present have the potential
to have higher expression level in the mismatch probe than in
the match probe depending upon the individual's genotype. In
order to test if this was the case, each of the eleven probe
pairs within each corresponding probe sets were compared to
see how frequently the match expression level was greater
than the mismatch expression both in those probe pairs
without a SNP and those probe pairs where a SNP is mapped
to the 13th position. Expression data was obtained using CEL
files for 12 different experiments as discussed in the
Methodology section. The resulting data set yields 1670 non-
SNP containing probes, and 166 SNP containing probes. For
the complete set of 1836 probes, the number of times that the
mismatch expression data was higher than the match
expression data was reported. Table 2 (see supplementary
Bioinformation, an open access forum
© 2008 Biomedical Informatics Publishing Group
Previous
TOC
-------
Bioinformation
www.bioinformation.net
by Biomedical Informatics Publishing Group
open access
Hypothesis
material) is constructed from this dataset, noting the number
of times that the mismatch probe expression level is observed
to be greater than the match probe expression level.
Analysis
Table 2 (see supplementary material) indicates the number of
times the MM probe is greater than the PM probe occurs with
less frequency in the probes with a SNP in the 13th base than
it does in probes without SNPs, debunking our hypothesis. It
is observed that about 52% of the time in probes without
SNPs, the perfect match probe expression level is always
greater than the mismatch probe expression level, while this
occurs at approximately the same rate in probes with a SNP in
the 13th base. One interesting piece of information is that the
mismatch probe expression level is always greater than the
match expression level in both instances around 9% of the
time. A graph of the frequency of these events is shown in
Figure 2. The number of observations from the SNP and
nonSNP data was compared using %2 analysis. The resulting
j2 value of 6.0108 with 12 degrees of freedom has a p-value
of 0.9155, thus rejecting the alternative hypothesis that the
SNP and nonSNP data are significantly different.
Minor allele frequencies
The 166 unique complementary SNPs have been mapped
according to known SNPs from the dbSNP database which
contains SNPs for all population types. However, since the
experiments selected are focused on the HapMap Han
Chinese in Beijing (HCB) population, it is possible the
observed results are skewed according to population-specific
SNPs. Each of the 166 dbSNP references were searched
against HapMap using HapMart release 23a. Sixty-four of the
166 SNP containing probes have been genotyped for the HCB
group by the HapMap project, each with between 72 and 90
allelic observations. However, only 20 of these have a minor
allele frequency of 5% or greater (Table 3 in supplementary
material). The SNP to nonSNP probe groups for this set of 20
SNPs were compared as previously discussed. A comparison
of the number of times the MM probes was greater than the
PM probes is given in Table 4 (shown under supplementary
material) for these 20 SNPs. Since the number of observations
is low, Fisher's exact test was performed on the SNP to
nonSNP group, resulting in ap-value of 0.04113. The p-value
is much lower than before, and indicates a significant
difference in the distributions when a p-value threshold of
0.05 is considered. This indicates that perhaps with more
observations, it could be possible to differentiate between
differences in MM and PM arising due to allelic variations
and those from cross hybridizations.
Conclusion:
The recent publications of the complete diploid genome of
two individual humans indicate that the rate of SNP variation
within an individual is much larger than previously expected
[21]. The higher rate of variation in the zygosity presents an
issue when looking at gene expression. Our hypothesis states
that we would expect to see higher hybridization rates for
mismatch probes in regions where a SNP is found in the 13th
base of a probe sequence. However, initial results on twelve
ISSN 0973-2063 409
Bioinformation 2(9): 405-411 (2008)
microarray experiments illustrate this is not the case, and in
fact, the opposite is true. Further analysis of the samples used,
including genotyping information, would be useful in
determining if these discrepancies result due to the frequency
of certain haplotypes within a population.
When known haplotype frequencies are considered, it is still
difficult to differentiate between true SNPs and cross
hybridization although the distributions are more distinct. Part
of this inability may be due to low number of SNPs (20)
falling into this category. As more haplotype frequency
information becomes available for all 166 candidate SNPs
through the HapMap project, it may become plausible to
differentiate between cross-hybridization. Additional
haplotype information for the other HapMap populations may
result in additional alleles with higher minor allele
frequencies.
The ability to discern between cross-hybridization and
infrequent SNPs based on PM and MM data is difficult at
best. SNPs remain a tricky issue when microarray probe
design is considered. It is our conclusion that information is
not lost when these probes are discarded, since the source of
the discrepancy cannot be consistently determined.
Acknowledgment:
Support for this project was provided by NIH-NCRR grant
P20RR16481 and NIH-NIEHS grant P30ES014443. The
contents of this manuscript are solely the responsibility of the
authors and may not represent the official views of the
National Center for Research Resources, the National Institute
for Environmental and Health Science, or the National
Institutes of Health. ECR and AWP contributed equally to this
project. The authors would like to thank the University of
Louisville Bioinformatics Research Group (BRG) and the
University of Louisville Bioinformatics Laboratory for
numerous fruitful discussions.
References:
[01] B. Palsson, Nat.BiotechnoL, 20: 649 (2002) [PMID:
12089538]
[02] M. Schena et al, Science, 21Q: 467 (1995) [PMID:
7569999]
[03] http://www.affymetrix.com/
[04] J. C. Chang and Y. W. Kan, Lancet, 2: 1127 (1981)
[PMID: 6118575]
[05] E. Mateu et al, Am. J. Hum. Genet, 68: 103 (2001)
[PMID: 11104661]
[06] M. Koenig et al, Am. J. Hum. Genet., 45: 498 (1989)
[PMID: 2491009]
[07] N. Vionnet et al, Nature, 356: 721 (1992) [PMID:
1570017]
[08] M. Wessman et al, Am. J. Hum. Genet., 70: 652
(2002) [PMID: 11836652]
[09] S. T. Sherry et al, Nucleic Acids Res., 29: 308 (2001)
[PMID: 11125122]
[10] B. E. Stranger et al, Science, 315: 848 (2007) [PMID:
17289997]
Bioinformation, an open access forum
© 2008 Biomedical Informatics Publishing Group
Previous
TOC
-------
Bioinformation
www.bioinformation.net
by Biomedical Informatics Publishing Group
open access
Hypothesis
[11] J. B. Fan et al, Genome Res., 10: 853 (2000) [PMID: [17]
10854416]
[12] W. J. Kent and D. Haussler, Genome Res., 11: 1541
(2001) [PMID: 11544197] [18]
[13] W. J. Kent et al, Genome Res., 12: 996 (2002)
[PMID: 12045153] [19]
[14] S. T. Sherry et al., Genome Res., 9: 677 (1999) [20]
[PMID: 10447503] [21]
[15] T. Barrett et al., Nucleic Acids Res., 35: D760 (2007)
[PMID: 17099226]
[16] Y. W. Wong et al., Br. J. Ophthalmol, 90: 769 (2006)
[PMID: 16488932]
Edited by S. Datta
Citation: Rouchka etaL, Bioinformation 2(9): 405-411 (2008)
License statement: This is an open-access article, which permits unrestricted use, distribution, and reproduction in
any medium, for non-commercial purposes, provided the original author and source are credited.
IUPAC-IUB commission on biochemical
nomenclature (CBN), J. Mol. Biol., 55: 299 (1971)
[PMID: 5551389]
S. F. Altschul et al., J. Mol. Biol., 215: 403 (1990)
[PMID: 2231712]
E. C. Rouchka, Conversation with: W. Gish (2004)
E. C. Rouchka, Conversation with: I. Korf (1999)
S. Levy et al., PLoS.Biol., 5: e254 (2007) [PMID:
17803354]
Supplementary material
Probe set
name
1007_s_at
1007_s_at
1007_s_at
1007_s_at
1007_s_at
1007_s_at
1007_s_at
1007_s_at
1007_s_at
1007_s_at
1007_s_at
1007_s_at
1007_s_at
Probe X
467
531
86
365
207
593
425
552
680
532
143
285
383
Probe Y
181
299
557
115
605
599
607
101
607
139
709
623
479
Probe
interrogation
position
3330
3443
3512
3563
3570
3576
3583
3589
3615
3713
3786
3793
3799
Probe sequence target
CACCCAGCTGGTCCTGTGGATGGGA
GCCCCACTGGACAACACTGATTCCT
TGGACCCCACTGGCTGAGAATCTGG
AAATGTTTCCTTGTGCCTGCTCCTG
TCCTTGTGCCTGCTCCTGTACTTGT
TGCCTGCTCCTGTACTTGTCCTCAG
TCCTGTACTTGTCCTCAGCTTGGGC
ACTTGTCCTCAGCTTGGGCTTCTTC
TCCTCCATCACCTGAAACACTGGAC
AAGCCTATACGTTTCTGTGGAGTAA
TTGGACATCTCTAGTGTAGCTGCCA
TCTCTAGTGTAGCTGCCACATTGAT
GTGTAGCTGCCACATTGATTTTTCT
Strandedness
Antisense
Antisense
Antisense
Antisense
Antisense
Antisense
Antisense
Antisense
Antisense
Antisense
Antisense
Antisense
Antisense
Table 1: Sample source data from the HG-U133Amicroarray in tab-delimited format.
# observations
MM>PM
0
1
2
3
4
5
6
7
8
9
10
11
12
Probes without SNP
No. of occurrence
873
123
92
64
54
60
46
34
48
39
37
54
146
% Occurrence
52.3%
7.4%
5.5%
3.8%
3.2%
3.6%
2.8%
2.0%
2.9%
2.3%
2.2%
3.2%
8.7%
Probes with SNP
No. of occurrence
88
14
8
7
2
5
5
4
3
4
6
3
17
% Occurrence
53.0%
8.4%
4.8%
4.2%
1.2%
3.0%
3.0%
2.4%
1.8%
2.4%
3.6%
1.8%
10.2%
Table 2: Mismatch to match expression level results.
ISSN 0973-2063
Bioinformation 2(9): 405-411 (2008)
410
Bioinformation, an open access forum
© 2008 Biomedical Informatics Publishing Group
Previous
TOC
-------
Bioinformation
www.bioinformation.net
by Biomedical Informatics Publishing Group
open access
Hypothesis
Probe ID
202192_s_at
20761 l_at
203680_at
201678_x_at
206529_x_at
210732_s_at
219502_at
215261_at
206226_at
21681 l_at
215986_at
221344_at
214836_x_at
209313_at
210618_at
217530_at
219093_at
216463_at
207075_at
219424_at
dbSNP
reference
Rs9545
Rs200485
Rs257378
Rsl0712
Rs272679
Rs2273865
Rsl055677
Rsl2198616
Rsl 042464
Rsl 1009339
Rs2 19307
Rsl011985
Rs232230
Rs8731
Rs4654973
Rs7447593
Rs3755302
Rsl6849300
Rsl0754558
Rs6613
Major
allele
G
G
C
G
C
T
C
G
T
C
G
G
G
G
C
C
T
G
G
A
Frequency
0.932
0.911
0.9
0.875
0.872
0.852
0.849
0.844
0.756
0.714
0.689
0.659
0.655
0.633
0.622
0.622
0.589
0.578
0.5
0.5
Minor
allele
C
C
G
C
G
A
G
C
A
G
C
C
C
C
G
G
A
C
C
T
Frequency
0.068
0.089
0.1
0.125
0.128
0.148
0.151
0.156
0.244
0.286
0.311
0.341
0.345
0.367
0.378
0.378
0.411
0.422
0.5
0.5
Table 3: Complementary SNP probes with Minor Allele Frequency > 5%.
#MM>PM Expected % of SNPs* Observed Observed
SNP probes (%) nonSNP probes (%)
0
1
2
3
4
5
6
7
8
9
10
11
12
5%
35%
10%
30%
5%
10%
0%
0%
0%
0%
0%
0%
0%
6 (30%)
0 (0%)
0 (0%)
1 (5%)
1 (5%)
2(10%)
1 (5%)
1 (5%)
0 (0%)
0 (0%)
5 (25%)
1 (5%)
2 (10%)
77 (38.3%)
23(11.4%)
10(5.0%)
12 (6.0%)
9 (4.5%)
10(5%)
5 (2.5%)
8 (4.0%)
11 (5.5%)
3 (1.5%)
6 (3%)
9 (4.5%)
18 (9%)
Table 4: Probes with MM > PM for SNPs with Minor Allele Frequency > 5%.
ISSN 0973-2063
Bioinformation 2(9): 405-411 (2008)
411
Previous
TOC
Bioinformation, an open access forum
© 2008 Biomedical Informatics Publishing Group
-------
Aquatic Toxicology 92 (2009) 168-178
Contents lists available at ScienceDirect
Aquatic Toxicology
journal homepage: www.elsevier.com/locate/aquatox
Endocrine disrupting chemicals in fish: Developing exposure indicators and
predictive models of effects based on mechanism of action
Gerald T. Ankley3'*, David C. Bencicb, Michael S. Breenc, Timothy W. Colletted, Rory B. Conollyc,
Nancy D. Denslow6, Stephen W. Edwardsf, Drew R. Ekmand, Natalia Garcia-Reyero6'1,
Kathleen M. Jensen3, James M. Lazorchakb, Dalma Martinovic3'2, David H. Millerg,
Edward J. Perkins11, Edward F. Orlando1, Daniel L Villeneuve3, Rong-Lin Wangb, Karen H. WatanabeJ
' USEPA, National Health and Environmental Effects Research Lab, Duluth, MN, United States
b USEPA, National Exposure Research Lab, Cincinnati, OH, United States
c USEPA, National Center for Computational Toxicology, RTP, JVC, United States
d USEPA, National Exposure Research Lab, Athens, GA, United States
e University of Florida, Gainesville, EL, United States
1 USEPA, National Health and Environmental Effects Research Lab, RTP, JVC, United States
z USEPA, National Health and Environmental Effects Research Lab, Grosse lie, MI, United States
h US Engineer Research and Development Center, Vicksburg, MS, United States
1 University of Maryland, College Park, MD, United States
J Oregon Health and Science University, Beaverton, OR, United States
ARTICLE INFO
Article history:
Received 28 August 2008
Received in revised form 28 January 2009
Accepted 31 January 2009
Keywords:
Fish
EDC
Toxic ity
MOA
Model
Genomics
ABSTRACT
Knowledge of possible toxic mechanisms (or modes) of action (MOA) of chemicals can provide valuable
insights as to appropriate methods for assessing exposure and effects, thereby reducing uncertainties
related to extrapolation across species, endpoints and chemical structure. However, MOA-based testing
seldom has been used for assessing the ecological risk of chemicals. This is in part because past reg-
ulatory mandates have focused more on adverse effects of chemicals (reductions in survival, growth
or reproduction) than the pathways through which these effects are elicited. A recent departure from
this involves endocrine-disrupting chemicals (EDCs), where there is a need to understand both MOA
and adverse outcomes. To achieve this understanding, advances in predictive approaches are required
whereby mechanistic changes caused by chemicals at the molecular level can be translated into apical
responses meaningful to ecological risk assessment. In this paper we provide an overview and illustrative
results from a large, integrated project that assesses the effects of EDCs on two small fish models, the fat-
head minnow (Pimephales promelas) and zebrafish (Danio rerio). For this work a systems-based approach
is being used to delineate toxicity pathways for 12 model EDCs with different known or hypothesized
toxic MOA. The studies employ a combination of state-of-the-art genomic (transcriptomic, proteomic,
metabolomic), bioinformatic and modeling approaches, in conjunction with whole animal testing, to
develop response linkages across biological levels of organization. This understanding forms the basis for
predictive approaches for species, endpoint and chemical extrapolation. Although our project is focused
specifically on EDCs in fish, we believe that the basic conceptual approach has utility for systematically
assessing exposure and effects of chemicals with other MOA across a variety of biological systems.
Published by Elsevier B.V.
1. Background
Prospective ecological risk assessments of most chemicals typi-
cally are conducted with little consideration for toxic mechanisms
* Corresponding author.
E-mail address: ankley.gerald@epa.gov (G.T. Ankley).
1 Current affiliation: Jackson State University, Jackson, MS, United States.
2 Current affiliation: University of St. Thomas, St. Paul, MN, United States.
0166-445X/S - see front matter. Published by Elsevier B.V.
doi:10.1016/j.aquatox.2009.01.013
(or modes) of action (MOA). Testing for ecological effects usually
includes a wide array of species and endpoints, with a focus primar-
ily on apical responses. When little is known about the properties
of a test chemical, this is a pragmatic approach; however, substan-
tial benefits can be realized by basing testing and subsequent risk
management decisions on known or probable MOA. For example, a
priori knowledge of MOA can lead to identification of mechanism-
based (and, hence, stressor-specific) molecular indicators that can
potentially be linked to environmental concentrations and used to
inform exposure assessments. Furthermore, knowledge of MOA can
Previous
TOC
-------
C.T. Ankley et al/Aquatic Toxicology 92 (2009) 168-178
169
Compartment
ftiitlrnseii / Estrosen
Responsive Tissues
Chemical "Probes"
Fipronil (-)
Muscimol (+)
Apomorphine (+)
Haloperidol (-)
Trilostane (-)
Ketoconazole (-)
/ / yfl Fadrozole (-)
/ /\/fl Prochloraz (-,-)
Vinclozolin (-)
Flutamide (-)
17(3-Trenbolone (+)
17a Ethinylestradiol (+)
(e.g.. liver, falpad. gonads)
Fig. 1. Overview of the fish hypothalamic-pituitary-gonadal (HPG) axis, and experimental chemical "probes" with different mechanisms of action. The "+" or "-" shown in
parentheses indicate, respectively, stimulation or inhibition of a particular target (enzyme or receptor) by the test chemical. See text for further details. Depiction of the HPG
axis is adapted from Villeneuve et al. (2007b).
serve as a basis for effective extrapolation of biological effects across
species, biological levels of organization, and chemical structures.
This information can help identify potentially sensitive responses,
and even species prior to extensive testing, thereby optimizing time
and resource use (Bradbury et al., 2004).
Endocrine-disrupting chemicals (EDCs) represent a compara-
tively recent departure from past regulatory activities with toxic
compounds in that there is a need to know both MOA and potential
adverse effects. There have been several definitions of EDCs from a
MOA perspective, ranging from (in the most limited sense) chemi-
cals which are estrogenic (specifically, estrogen receptor agonists)
to (in the broadest sense) "an exogenous agent that interferes with
the production, release, transport, metabolism, binding, action, or
elimination of natural hormones in the body responsible for the
maintenance of homeostasis and the regulation of developmental
processes" (Kavlock et al., 1996). From a regulatory perspective, the
definition currently most widely used for EDCs encompasses agents
that cause alterations in reproduction or development through
direct effects on the vertebrate hypothalamic-pituitary-thyroidal
or hypothalamic-pituitary-gonadal (HPG) axes (USEPA, 1998).
Due to the emphasis on MOA, consideration of EDCs in current
testing and regulatory frameworks has been challenging. For exam-
ple, it is important that tests include responses other than apical
endpoints if the assays are to be indicative of specific MOA. How-
ever, mechanism-specific endpoints are not necessarily predictive
of an adverse biological outcome, and it is problematic to inten-
sively regulate a chemical that does not cause adverse effects even
if it does, for example, activate the estrogen receptor (ER). A com-
mon (and logical) approach to addressing this seeming dilemma
has been the development of tiered testing frameworks that use
short-term assays first to identify chemicals as possessing a MOA
of concern before proceeding with longer-term tests better suited
to quantifying adverse effects (e.g., USEPA, 1998). However, even
relatively efficient tiered testing programs for EDCs may not be
sustainable in terms of resources (or timeliness) if hundreds or
thousands of chemicals need to be assessed using long-term assays.
The efficiency of EDC testing programs could be enhanced
through the use of emerging technologies in the areas of genomics
and computational biology to provide mechanistic insights as to
exposures and possible adverse effects in animals, such as fish (e.g.,
Ankley et al., 2006; Hoffmann et al., 2006, 2008; Hook et al., 2006;
Samuelsson et al., 2006; Filby et al., 2007; Martyniuk et al., 2007).
This type of approach is consistent with recent recommendations
from the National Research Council (NRC, 2007), who suggest a
Previous
TOC
-------
170
C.T. Ankley et al/Aquatic Toxicology 92 (2009) 168-178
Increasing Diagnostic (Screening) Utility Increasing Ecological Relevance
Levels <
Biologic
Organization
Oman
Functional and
(ructural ch
(Patholoi
Individual Population
Decreased
Computational
modeling
Poorly char
genoi
high ecolo
regulatory re
PhaseS. I Phase 1.
Fathead minnow 21 d reproduction test
riptomlcs I (e g fecundity, histology, vitellogenin, sex
.-.abolomics I steroids)
Population
modeling
Fig. 2. Conceptual linkages across biological levels of organization for effects of endocrine-disrupting chemicals in fish. Different research phases of the project are aligned
to reflect where/how they address these linkages and arrows reflect the flow of information. See text for further details.
shift toward greater use of short-term (e.g., in vitro) assays and
predictive toxicology tools for assessment of human health risks of
chemicals. In this paper we describe a research effort to support the
development of approaches for assessing chemicals with the poten-
tial to impact the HPG axis offish. These approaches, which could
encompass techniques ranging from computational models to in
vitro assays and short-term in vivo tests, would help provide regu-
latory agencies throughout the world with cost-effective, predictive
tools for monitoring and testing EDCs.
This is a large, highly-integrated project that includes govern-
ment, academic and industry scientists from several laboratories
across North America. In this paper, we describe the conceptual
basis of the approach we have employed and present illustrative
results. The information provided herein is necessarily brief; for
further detail on methods and results, interested readers should
consult the indicated citations or contact us directly, as many
aspects of the data collection/analyses are ongoing.
2. Experimental overview
The basic approach used for our work involves perturbation
of the HPG axis with chemical probes known or hypothesized
to impact different key control points, ranging from neurotrans-
mitter receptors in the brain to steroid hormone receptors in
gonads (Fig. 1). Following perturbation of the axis by chemicals
with different MOA, information is collected at multiple biological
levels of organization, ranging from molecular changes to api-
cal responses (i.e., reproductive success), and even (via modeling)
to likely population-level effects (Fig. 2). This type of integrated
analysis facilitates a mechanistic understanding of the effects of
HPG-active chemicals from the molecular to whole-organism levels
from a toxicity pathway perspective (Bradbury et al., 2004).
2.1. Organisms
The experimental organisms for this research are small fish.
There are several different EDC testing programs being imple-
mented throughout the world, and most include fish assays (Ankley
and Johnson, 2004). A pragmatic reason for this is that there are
clearly documented adverse impacts of EDCs on fish populations
in the field; this differs from the situation in humans where expo-
sure to, and subsequent effects of environmental EDCs tend to be
more uncertain (WHO, 2002). In addition, in terms of animal avail-
ability (e.g., generation of large numbers of high-quality organisms
at suitable life-stages), chemical exposure dynamics and biological
flexibility, small fish species are well suited for mechanistic stud-
ies with chemicals such as EDCs (Stoskopf, 2001, and references
therein; Ankley and Johnson, 2004). Significantly, although there
are some unique aspects of fish reproductive endocrinology, the
basic structure and function of the HPG axis across all vertebrates
tends to be well conserved. Hence, the results of fish studies with
EDCs potentially can serve as the basis for effective cross-species
extrapolation of potential effects.
Our research utilizes the zebrafish (Danio rerio) and fathead
minnow (Pimephales promelas), two small cyprinids that have com-
plementary attributes that make them useful for this work. The
genome of the zebrafish is fully sequenced, thus reducing bioinfor-
matic challenges when evaluating alterations in gene and protein
expression (Hill et al., 2005). As such, the zebrafish is a useful
model for exploratory or hypothesis-generating work focused on
the effects of EDCs with different MOA on response profiles of genes
and proteins (Hoffmann et al., 2006, 2008). In contrast to zebrafish,
the fathead minnow has a rich history of use in regulatory pro-
grams in the US, including testing for EDCs (Ankley and Villeneuve,
2006). In addition to its relevance to regulatory activities, a fair
Previous
TOC
-------
C.T. Ankley et al/Aquatic Toxicology 92 (2009) 168-178
171
amount is known about basic reproductive biology in the fathead
minnow, thus providing a basis for "anchoring" observed alterations
in gene, protein or metabolite expression caused by test chemicals
to phenotypic changes in gonad histology and reproductive success.
2.2. Test chemicals
Test chemicals used for the work include those that (could)
impact HPG function relatively "high" in the axis, such as mus-
cimol (a pharmaceutical) and fipronil (an insecticide) which act,
respectively, as an agonist and antagonist of specific GABA (gamma-
amino butyric acid) receptors (Fig. 1). The drugs apomorphine
and haloperidol act as an agonist and antagonist, respectively, of
dopamine receptors (D2) involved in the release of gonadotrophic
hormones from the pituitary. The fungicides ketoconazole and
prochloraz, and the Pharmaceuticals trilostane and fadrozole
inhibit one or more enzymes involved in steroid biosynthesis in the
gonad, including reactions catalyzed by 3(3-hydroxysteroid dehy-
drogenase (3|3HSD) and different cytochromes P450 (CYPs). Finally,
several chemicals that directly impact hormone receptors located
in the gonad and other steroid-responsive tissues are being tested,
including 17a-ethinylestradiol and 17(3-trenbolone, potent syn-
thetic steroidal agonists of the ER and androgen receptor (AR),
respectively, and vinclozolin (a fungicide) and flutamide (a phar-
maceutical), which antagonize the AR (Fig. 1). Although several of
these chemicals do occur in the environment as contaminants (e.g.,
the pesticides and synthetic steroids), others are less likely to do
so (some of the drugs). Overall, our strategy in selection of test
chemicals was not necessarily to focus on known environmental
contaminants, but to perturb HPG pathways of known (or potential)
biological relevance.
2.3. Phased testing
Sexual development, including gonad differentiation, during lar-
val and juvenile life-stages, and reproduction in mature adults offer
two "windows" of enhanced sensitivity of fish to EDCs (Ankley
and Johnson, 2004). For this research, the adult life-stage was
chosen because we felt that the substantial alterations in gene
and protein expression and metabolite profiles that occur during
early development might complicate understanding of the effects
produced by the test chemicals. However, due to the known sen-
sitivity of fish to EDCs during sexual development, studies of the
type described herein encompassing this life-stage also would be
desirable.
For our research, three different types of studies—termed Phases
1, 2 and 3—are conducted on adult fish exposed to the various HPG-
active chemicals (Fig. 2). In Phase 1 studies, each chemical is tested
in a standardized 21-d reproduction assay with the fathead min-
now using flow-through (water) exposures and measured chemical
concentrations to produce a high-quality exposure/effects dataset.
Endpoints measured in the Phase 1 studies span a wide range
of biological levels of organization, including determination of
plasma concentrations of sex steroids (testosterone, 17(3-estradiol,
ll-ketotestosterone)andvitellogenin(Vtg; egg yolk protein precur-
sor), gonad size and histopathology, secondary sex characteristics,
reproductive behavior, fecundity, fertility and hatchability (Ankley
et al., 2001). In addition, the Phase 1 studies incorporate analyses of
a small complement of genes (measured via quantitative real-time
polymerase chain reaction; PCR) known to be involved in HPG
function/control, and hypothesized to be impacted by the chemical
exposure (Villeneuve et al., 2007a,b). Information from the Phase
1 studies is subsequently used for three primary purposes, to:
(a) aid in design (e.g., selection of test chemical concentrations)
for subsequent, shorter-term Phase 2 and 3 assays, (b) generate
information for systems and population modeling, and (c) provide
a robust phenotypic dataset for anchoring the various genomic
responses collected in subsequent testing (Fig. 2).
Phase 2 tests are short-term assays conducted with zebrafish in
which samples from multiple tissues (gonad, liver and brain) are
collected after 1, 2 and 4 d of exposure to the test chemicals. The
samples are used for genomic measurements, with an emphasis
on gene expression determined using commercially-available
22,000 or 4 x 44,000 gene microarrays (Agilent, Palo Alto, CA, USA;
Wang et al., 2008a,b) as well as hypothesis-driven and microarray-
confirmatory PCR analyses. A subset of the zebrafish samples
also are analyzed for alterations in protein expression using two-
dimensional (2-D) Fluorescence Difference Gel Electrophoresis
(Ettan™ DICE) technology (G.E. Healthcare Bio-Sciences Corp.,
Piscataway, NJ, USA). Overall, information from the Phase 2 studies
provides insights on relationships between gene and protein
expression, and helps identify candidate indicators/markers of
EDC exposure and effects for subsequent evaluation in Phase 3
fathead minnow studies focused on temporal changes in the HPG
axis (Fig. 2).
The HPG axis is a highly dynamic system capable of respond-
ing to environmental stressors, including contaminants, through
various feedback mechanisms to maintain conditions conducive to
reproduction. These types of compensatory responses can occur
both during exposure to the stressor, and after the stressor has been
removed. This, coupled with the fact that changes in some end-
points (e.g., gene expression) can be rapid and/or transitory, dictates
a need for temporal studies to develop robust exposure indicators
and predictive models. The Phase 3 studies in our research address
this through systematic time-course experiments with the fathead
minnow. In these studies, animals are sampled after 1, 2, 4 and 8
d of exposure to the various test chemicals, as well as 1, 2, 4 and 8
d after cessation of exposure. A variety of endpoints are examined,
including a subset of those considered in the Phase 1 studies such
as plasma steroid and Vtg concentrations, gonad histopathology
and secondary sexual characteristics. Gene expression is evaluated
using both targeted (PCR) assays (e.g., Villeneuve et al., 2007a),
and through microarray analysis, using custom microarrays con-
structed on an Agilent platform (N. Denslow, unpublished data).
Among other uses, the time intensive microarray analyses in Phase
3 studies provide data for reverse engineering of transcriptional
networks within the HPG axis (di Bernardo et al., 2005). Changes
in biological networks can be particularly useful in discerning MOA
and mechanisms of response and compensation. Finally, Phase 3
samples are used for examination of changes in protein expression
(based on targets from the Phase 2 studies), and metabolomic anal-
yses via nuclear-magnetic resonance (NMR) and mass spectroscopy
(MS) techniques (Ekman et al., 2007, 2008, in press). Information
from Phase 3 serves a number of overall purposes, including (a)
directly identifying linkages between changes in gene, protein and
endogenous metabolite profiles; (b) relating these genomic changes
to apical endpoints such as histopathology; (c) evaluating the con-
sistency in responses to EDCs across species (zebrafish, fathead
minnows) exposed under the same conditions; (d) providing infor-
mation as to temporal alterations in a stressed (and unstressed
or recovering) system as a basis for modeling HPG axis function;
(e) evaluating the rapidity and persistence of potential indicator
responses identified in earlier phases of testing; and (f) developing
dynamic models to understand feedback control and compensation
for stress (Fig. 2).
3. Insights from experimental work
3.1. Phase 1
Table 1 summarizes the fathead minnow 21 -d reproduction tests
that have been conducted to date and, where available, provides ref-
Previous
TOC
-------
172
C.T. Ankley et al/Aquatic Toxicology 92 (2009) 168-178
Table 1
Overview of reproductive toxicity to the fathead minnow of chemicals with differing MOA in the hypothalamic-pituitary-gonadal (HPG) axis.
Test chemical
Presumptive HPG target(s)3
Reproduction LOECb
Reference
Fipronil
Muscimol
Apomorphine
Haloperidol
Trilostane
Ketoconazole
Fadrozole
Prochloraz
Vinclozolin
Flutamide
Trenbolone
Ethinylestradiol
GABA receptor antagonist
GABA receptor agonist
D2 receptor agonist
D2 receptor antagonist
3(3HSD inhibitor
CYP11A/CYP17 inhibitor
CYP19 inhibitor
CYP17/19 inhibitor
AR antagonist
AR antagonist
AR agonist
ER agonist
>5
NCC
NC
>20
1500
25
2
100
60
500
0.05
NC
Kahl et al. (2007)
Villeneuve et al. (in preparation)
Villeneuve et al. (2008)
Ankley etal. (2007)
Ankley etal. (2002)
Ankley etal. (2005)
Martinovic et al. (2008)
Jensen et al. (2004)
Ankley etal. (2003)
' Abbreviations used: GABA, gamma-amino butyric acid; D2, dopamine; 3(3HSD, 3(3-hydroxysteroid dehydrogenase; CYP11A, cytochrome P450scc (side-chain-cleavage);
CYP17, cytochrome P450cl7,20-lyase; CYP19, cytochrome P450 aromatase; AR, androgen receptor; ER, estrogen receptor.
b Lowest-observable effect concentration (LOEC) for egg production in 21-d tests. Values are nominal water concentrations provided in jxg/L
c NC, not conducted/completed.
erence information for the completed studies. In terms of exposure
concentrations that cause impacts on reproductive health, the test
chemicals span a wide range of potency and efficacy, ranging from
trenbolone which significantly decreased egg production at a water
concentration of 0.05 |jig/L (Ankley et al., 2003), to trilostane which
affected egg production at a concentration of 1500 |jig/L (Villeneuve
et al., 2008). Some of the test chemicals (e.g., fipronil, haloperidol)
did not cause marked effects on reproductive endocrine function,
even when tested at concentrations at maximum water solubil-
ity, or within a factor of five of those that produced toxicity in
short-term range-finding assays. When effects were observed, bio-
chemical and apical responses in the 21-d test generally reflected
the anticipated MOA of the test chemicals. For example, consistent
with activation of the AR, trenbolone caused morphological mas-
culinization of female fathead minnows, while the AR antagonist,
vinclozolin, demasculinized males (Ankley et al., 2003; Martinovic
et al., 2008). Although different enzymes were affected, inhibitors
of steroidogenesis (fadrozole, prochloraz, trilostane) all decreased
Vtg concentrations in female fish due to a depression in synthesis
of estradiol (Ankley et al., 2002, 2005; Villeneuve et al., 2008).
A critical role of the Phase 1 studies in the overall project is
delineation of hypothesized toxicity pathways across biological lev-
els of organization, such that the Phase 2 and 3 transcriptomic,
proteomic and metabolomic data can be mechanistically linked to
higher-level apical responses. The fadrozole data provide a partic-
ularly good example of how this is achieved (Fig. 3). Fadrozole was
developed to treat breast cancer as a relatively specific inhibitor of
CYP19 aromatase, the enzyme that catalyzes conversion of testos-
terone to estradiol. The pharmaceutical decreases brain and ovarian
aromatase activity in vitro and in vivo in the fathead minnow, and
produces a corresponding decrease in circulating plasma estradiol
concentrations in female fish (Ankley et al., 2002; Villeneuve et
al., 2006). This, in turn, translates into a decreased circulating con-
centration of Vtg (which is produced in the liver via activation of
the ER) in the females and, ultimately, decreased deposition of the
lipoprotein in the developing oocytes. This corresponds with signif-
icant reductions in fecundity of fadrozole-exposed fish, resulting in
complete cessation of egg production at higher fadrozole exposure
concentrations (Fig. 3). As described in greater detail below, these
laboratory fecundity data can then be taken, via modeling, one step
further to predict likely population-level responses of fish exposed
to HPG-active chemicals (Fig. 2).
In addition to providing baseline effects and toxicity pathway
data for the various chemicals, several other significant observa-
tions have been made in the Phase 1 studies. One of these involves
indirect changes in the HPG axis in response to certain EDCs. For
example, evidence for at least some degree of compensation within
the axis comes from 21-d tests with three of the test chemicals,
trenbolone, vinclozolin and ketoconazole, with the latter providing
the most complete demonstration of the phenomenon (Ankley et
al., 2007). Ketoconazole is a pharmaceutical that decreases fungal
growth through inhibition of an ergosterol (cell wall component)
biosynthesis step catalyzed by CYP51. However, ketoconazole is not
particularly specific to CYP51, and can inhibit a variety of vertebrate
CYPs involved in xenobiotic metabolism and steroid biosynthesis.
In fact, the fungicide is considered a model inhibitor of testosterone
production in mammals (Feldman, 1986). In the fathead minnow,
ketoconazole decreased fecundity in 21 -d tests and, consistent with
its anticipated MOA, inhibited testosterone production by gonad
tissue from both males and females. However, after a continuous
21-d exposure, this inhibition was not translated into decreased
circulating testosterone (or estradiol) concentrations in vivo in the
fish, suggesting that the animals were somehow able to compensate
for effects of the fungicide. This response was manifested in several
(B)
(C)
.1-
liJ
(D)
Fadrozole (pg/l)
Fadrozole (pg/l)
(E)
Exposure (d)
Fig. 3. Example of linkage of effects across biological levels of organization for a model endocrine-disrupting chemical, the aromatase inhibitor fadrozole (2, 10, 50jxg/L
water), tested in a 21-d reproduction assay with the fathead minnow. Panels (from left to right) depict (A) inhibition of aromatase activity in male and female fish, (B) decrease
in plasma estradiol (E2) concentrations in female fish, (C) depression in plasma vitellogenin (Vtg) concentrations in exposed females, (D) decreased Vtg deposition in the
ovary (compare the amount of dark staining material in the top [control] versus bottom [treated] fish), and (E) reduction in cumulative egg production in the fish. Results
from Ankley et al. (2002).
Previous
TOC
-------
C.T. Ankley et al/Aquatic Toxicology 92 (2009) 168-178
173
different ways (Ankley et al., 2007). For example, in males there
was more than a two-fold increase in relative gonad weight, accom-
panied by a proliferation of testicular Leydig cells (responsible for
steroid production), and up-regulation of genes coding for two key
steroidogenic enzymes, CYP11A and CYP17, both of which could be
specific targets of ketoconazole. The net result of these alterations
was that circulating steroid concentrations in the fish did not differ
from controls after 21 d of exposure to the fungicide. Understand-
ing the basis of possible compensatory responses within the HPG
axis clearly is needed to identify reliable exposure indicators and
develop approaches to predict adverse effects of EDCs; achieving
this understanding is a critical component of the Phase 3 studies of
the overall project.
3.2. Phase 2
Phase 2 zebrafish exposures and subsequent microarray mea-
surements have been completed for all the chemicals shown in
Fig. 1. Initial analysis of the microarray data has been described
for three of the chemicals: ethinylestradiol, trenbolone and fadro-
zole (Wang et al., 2008a,b). One goal of the Phase 2 research was to
determine a flexible and efficient microarray experimental design
to (1) characterize the zebrafish transcriptome, and (2) identify
an optimal combination of gene feature selection/class prediction
algorithms for evaluating gene expression changes caused by EDCs
with different MOA. An unbalanced, incomplete block microarray
experimental design was tested using various tissues of individ-
ual zebrafish exposed to fadrozole, trenbolone, or ethinylestradiol
(Wang et al., 2008a). Based on the high microarray reproducibil-
ity/low variability, low gene-specific dye bias, and good similarity
between microarray and PCR profiles observed, the design appears
well suited to these and other ecotoxicogenomic studies. Hyper-
spectral imaging identified a cyanine 3-background contaminant,
and correction of this fluorescence contamination reduced the
variability of weakly expressed genes, which constitute a signifi-
cant portion of the zebrafish transcriptome (Wang et al., 2008a).
Evaluation of several methods for gene classifier (indicator) dis-
covery determined that the optimal gene feature selection method
(of those tested) for reducing the dimensionality of microarrays
was via a genetic algorithm (GA), with the best prediction algo-
rithm of those evaluated, support vector machine (SVM; Wang
et al., 2008b). These algorithms are being applied in subsequent
microarray experiments to identify multi-gene expression profiles
(classifiers) capable of discriminating exposures to EDCs acting
through varying MOA based on microarray responses. As an exam-
ple, the preliminary analysis with the three chemicals identified
classifiers that discriminated exposures to fadrozole, trenbolone,
and ethinylestradiol, with the first two chemicals clustering more
closely to one another as chemicals that depress rather than elevate
plasma estrogen activity (Wang et al., 2008b).
Beyond identification of effective microarray experimental
design and analysis approaches, the Phase 2 zebrafish experiments
fulfill two critical roles. First, they provide a means to compre-
hensively interrogate the large number of transcripts that code for
proteins known or hypothesized to play key roles in the regulation
of the teleost HPG axis. As opposed to real-time PCR and similar
approaches that target a single or relative handful of genes, microar-
rays can be used to survey hundreds or thousands of components of
a biological system (represented by transcripts) and simultaneously
evaluate their response to various stressors. This type of approach
provides the basis for conducting a hypothesis-driven investigation
of the response of an entire system and a means to test and refine
biologically-based systems models that may ultimately be applied
to predictive risk assessment (e.g., Villeneuve et al., 2007b).
The second critical role of the zebrafish experiments is discovery.
Whereas Phase 1 studies examine only those endpoints selected
by the investigators based on some prior knowledge or hypothe-
ses, the microarray and proteomic analyses conducted as part of
the Phase 2 experiments are unsupervised. Because of the ability
to screen hundreds (in the case of proteomics) or thousands (in
the case of microarrays) of endpoints/targets, data from Phase 2
studies can be used to identify novel responses to the stressors
examined. By examining gene ontologies and pathways associ-
ated with differentially expressed genes or proteins, it is possible
to identify a broad spectrum of processes and/or targets that are
impacted either directly or indirectly by the chemical stressor. Such
knowledge can lead to an improved understanding of the overall
biological impact of the stressor, and may also aid the identifica-
tion of novel indicators (biomarkers) of exposure and/or effects.
Hypotheses and putative indicators that emerge from the Phase 2
analyses are being tested in a supervised fashion in the subsequent
Phase 3 experiments, examining the robustness of the observa-
tions both between experiments and among species. Through the
combination of hypothesis- and discovery-driven analyses, Phase 2
experiments test our overall systems model, expand on the analyses
conducted in Phase 1, and provide a foundation for novel hypothesis
testing in Phase 3.
3.3. Phase 3
Phase 3 tests have been conducted with six chemicals to
date: fadrozole, trenbolone, prochloraz, vinclozolin, flutamide and
trilostane. In addition, a preliminary Phase 3 like exposure with
ethinylestradiol using fewer sampling times and with a primary
focus on metabolomic measurements has been completed (Ekman
et al., 2008). Although much of the information associated with
these studies (e.g., gene expression) is still being assembled and
analyzed, some intriguing observations have already emerged.
Compared to transcriptomic measurements, metabolomic anal-
yses have received less attention in the field of ecotoxicology
(Lin et al., 2006). However, knowledge of profiles of endogenous
metabolites can provide important information concerning chem-
ical MOA, thereby helping to identify exposure indicators and
define toxicity pathways. In addition, compared to transcriptomics
and proteomics, metabolomic analyses are relatively inexpensive
and amenable to high-throughput, which enables a comparatively
large number of samples to be processed. This is an important
attribute for time-course studies, such as the Phase 3 work. Ini-
tial metabolomic studies by Ekman et al. (2007) demonstrated the
feasibility of NMR-based analyses of urine samples from the fat-
head minnow to assess impacts of the anti-androgen vinclozolin
on metabolite profiles. The use of urine in such studies not only
allows one to assess important metabolic endpoints, but also pro-
vides the potential for non-invasive and repeated sampling from
individual fish over time. In more recent work, Ekman et al. (2008,
in press) demonstrated the potential for metabolomic measure-
ments to provide novel insights about responses of the fish HPG
axis to chemical stressors. Adult fathead minnows of both sexes
were exposed to two different concentrations of ethinylestradiol,
and animals were sampled after 1, 4 and 8 d of exposure, and
8 d after termination of the exposure. NMR evaluation of polar
metabolites in livers of the fish revealed a greater impact of the
estrogen on males than females; in addition, the metabolite pro-
file in exposed males reflected a "feminization" response, in that
the profile assumed similarities to that of female fathead minnows
(2008). Assessment of the metabolomic data using partial least-
squares discriminant analysis revealed that response trajectories
in the males showed evidence of compensation of the fish dur-
ing the ethinylestradiol exposure, as well as a marked recovery
after cessation of exposure to the estrogen (Fig. 4). Evaluation of
other more traditional endpoints in the fish (changes in plasma Vtg
concentrations and secondary sex characteristics) indicated fern-
Previous
TOC
-------
174
C.T. Ankley et al/Aquatic Toxicology 92 (2009) 168-178
0.4
0.2-
CN
OT
o-
-0.2-
-0.4
8d (post-exp)
8d
-0.6
-0.4
-0.2 0
PLS1
0.2
0.4
Fig. 4. Exposure response trajectory plots for male fathead minnows exposed to
17a-ethinylestradiol for 1, 4, or 8 d, followed by 8 d of depuration (i.e., "post-
exp") during which the fish were maintained in water without test chemical.
Exposures were conducted using two ethinylestradiol concentrations (either 10 or
100 ng/L) delivered via a continuous flow-through system. The scores plot shown
was generated using the first two components (i.e., PLS1 and PLS2) of a validated
partial-least squares discriminant analysis (PLS-DA) model built using NMR spectral
data acquired from the livers of these fish. Each point represents the average score
value for a given class (n = 7 or 8), shown with its associated standard error. Note:
the controls across all time points showed relatively little variation and thus were
modeled as a single class. Results from Ekman et al. (2008).
inization of the males, as well as (in the case of secondary sex
characteristics) recovery following termination of exposure, con-
firming that alterations observed in metabolite profiles are a robust
indicator of the physiological state of the animals exposed to the
estrogen. Ekman et al. (in press) also evaluated the non-polar frac-
tion of hepatic metabolites from the ethinylestradiol study, and
noted a number of alterations in lipid profiles associated with expo-
sure to the estrogen. Ongoing MS-based metabolomic studies are
focused on assessing changes in sex steroids and steroid precursors
in fish exposed to EDCs for differing periods of time. Overall, the
types of temporally-intensive data collected from NMR- and MS-
based metabolomic analyses can be used to better understand how
exposure parameters—such as chemical concentration, frequency,
and duration—influence adverse outcomes. This new understand-
ing can help regulators differentiate chemical exposures that have
a lasting and detrimental biological effect from those that are either
not effective, or those to which an organism can adapt (albeit with
some potential cost to the organism).
Data from the fadrozole Phase 3 study also provide insights
as to compensatory responses of the HPG axis (Villeneuve et al.,
in press). As would be expected based on the MOA of fadrozole
(described above), exposure to the drug caused rapid (within 1
d), concentration-dependent reductions in estradiol production in
ex vivo assays with ovary tissue held in culture (detailed meth-
ods for the ex vivo assay can be found in Ankley et al. (2007) and
Martinovic et al. (2008)), and plasma concentrations of both estra-
diol and Vtg in female fish (Fig. 5). However, by the eighth day
of the exposure period, ex vivo estradiol production had returned
to control levels, and plasma estradiol concentrations had also
recovered to control levels in the fish exposed to 3 |jig fadro-
zole/L, albeit not in those exposed to 30 |Jig/L (Fig. 5). This apparent
compensation coincided with significant concentration-dependent
increases in the abundance of mRNA transcripts coding for aro-
matase (CYP19A isoform), CYP11A, steroidogenic acute regulatory
protein and follicle stimulating hormone receptor (Villeneuve et
al., in press). Shortly after cessation of the fadrozole exposure,
there was a rapid recovery of plasma estradiol concentrations in
the fish, and a gradual recovery of plasma Vtg concentrations, even
0 2 4 o a iu n it ID
Day
• Exposure period . . Recovery period .
Fig. 5. (A) Ex vivo estradiol (E2) production, (B) plasma E2, and (C) plasma
vitellogenin (Vtg) measured in female fathead minnows exposed to 0, 3, or
30 jxg fadrozole/L and sampled after 1, 2, 4, or 8 d of exposure or 1, 2, 4, or 8 d after
cessation of exposure (days 9, 10, 12, 16, respectively; recovery period). Data are
expressed as fold change (log 2) relative to the control mean measured on a given
day. Error bars indicate standard error. The * and # indicate statistically significant
difference from the control for the 3 and 30 jxg/L treatments, respectively (p<0.05).
Results from Villeneuve et al. (in press).
in the 30 |Jig/L group. In fact, in the 3 |Jig/L treatment, there was a
brief period of elevated plasma estradiol accompanied by a seem-
ing over-production of estradiol, ex vivo, relative to the control
group (Fig. 5). These data are consistent with the idea that estra-
diol production rates had been increased as part of a compensatory
response to the stressor. Of over a dozen transcript-level responses
examined, expression of mRNAs coding for follicle-stimulating hor-
mone receptor appeared to have the greatest potential utility as
an indicator of reproductive dysfunction mediated through the
estradiol synthesis-disruption toxicity pathway, based on the rapid-
ity, persistence, and concentration-dependence of the response
(Villeneuve et al., in press). Thus, based on the preliminary Phase
3 experiments substantively analyzed to date, the results have
shown excellent promise for identifying potentially useful expo-
sure indicators, detailing toxicity pathway characterization, and
improving our understanding of compensatory responses of the
HPG-axis to chemical stressors, all of which should enhance the
Previous
TOC
-------
C.T. Ankley et al/Aquatic Toxicology 92 (2009) 168-178
175
CYP11A1 CYP17H CYP17L
CHOL ^^^PREGknumiilMHPREG bzz±H DHEA
Fig. 6. Steroidogenesis model forthe female fathead minnow gonad based on in vitro data from control and fadrozole-treated fish. The model consists of two compartments,
medium and ovary tissue. Transport processes (black arrows) occur between the medium and ovary. Irreversible metabolic reactions (arrows with each pattern representing
a unique enzyme) occur in the ovary. Six enzymes labeled in italics next to reactions they catalyze are: cytochrome P450scc (side-chain-cleavage) (CYP11A1), cytochrome
P450cl7ahydroxylase (CYP17H), cytochrome P450cl7, 20-lyase (CYP17L), 3(3-hydroxy-dehydrogenase (3(3HSD), 17-beta-hydroxy-dehydrogenase (17(3HSD), and cytochrome
P450 aromatase (CYP19). Steroids and their precursors are: cholesterol (CHOL), pregnenolone (PREG), 17a-hydroxypregnenolone (HPREG), dehydroepiandrosterone (DHEA),
progesterone (PROG), 17a-hydroxyprogesterone (HPROG), androstenedione (AD), testosterone (T), estrone (El) and 17(3-estradiol (E2). Fadrozole is depicted as an inhibitor
of CYP19. The steroidogenic metabolic pathway encompasses two ovarian cell types: theca cells and granulosa cells. In theca cells, cholesterol is converted to AD and T. In
granulosa cells, AD and T are converted to El and E2. Model from Breen et al. (2007).
ability to generate predictive models with utility for ecological risk
assessment.
4. Integrating the data: predictive modeling
To help design the Phase 1, 2 and 3 studies and subsequently
interpret and integrate the large amounts of data collected, we
are using a systems biology/toxicology approach. Villeneuve et
al. (2007b) described development of a graphical systems model
focused on defining the HPG axis of teleost fish, which enables
consideration of the interactive nature of the system at multiple
levels of biological organization, ranging from changes in gene, pro-
tein and metabolite expression profiles to effects in cells/tissues
that directly influence reproductive success. The model plays a
role both in terms of designing our studies (e.g., deciding where
to perturb the system), and interpreting the sometimes seemingly
disparate biological responses observed (e.g., those associated with
compensation), both from hypothesis- and discovery-driven per-
spectives. The model also enables consideration of the HPG axis in
an integrated manner, such that effects of mixtures of chemicals
with similar or dissimilar MOA can be more directly evaluated. The
overall framework, which is written in open-source code (SBML;
Systems Biology Markup Language) is not intended to be static
but, rather, to evolve as this project (and the many other stud-
ies on EDCs and fish reproductive endocrinology being conducted
throughout the world) generate mechanistic data to better inform
the model. Although intended to support prediction of the effects of
HPG-active chemicals with different MOA on reproductive function
in fish, the model does not do so from a quantitative ("computa-
tional") perspective. Rather, the model described by Villeneuve et
al. (2007b) provides a framework for incorporation of more focused
computational models into an integrated assessment of the poten-
tial ecological risk of EDCs. Several of these types of computational
models associated with our current effort are discussed further
below.
Steroid hormones are critical to maintenance of HPG axis func-
tion, and feedback controls on the system are achieved largely
through alterations in steroid production. In addition, several estab-
lished EDCs exert adverse effects through their ability to directly
modulate (generally inhibit) different enzymes involved in steroid
synthesis (Figs. 1 and 6). Despite the importance of steroid produc-
tion, until recently there had been no mechanistic computational
models for describing baseline and/or chemically-perturbed con-
ditions in vertebrates. As part of our effort, Breen et al. (2007)
developed a steady-state model to predict synthesis and release
of testosterone and estradiol by ovarian tissue, and evaluated the
model using data generated from the fathead minnow (Fig. 6).
Model-predicted concentrations of the two steroids over time cor-
responded well with both baseline (control) data, and information
from experiments in which estradiol synthesis was blocked by
fadrozole. A sensitivity analysis of the model identified specific
processes that most influenced production of testosterone and
estradiol, thereby lending insights as to potential points of control
in the HPG axis. We have further developed predictive capabilities
for understanding Steroidogenesis by integration of the graphical
model of Villeneuve et al. (2007b), with the steady state model of
Breen et al. (2007), and the Hao et al. (2006) model of G protein,
protein kinase A, and steroid acute regulatory protein activation, to
examine effects of chemicals on steroid production and regulation
(Shoemaker et al., 2008). In this expanded model, we examined
the role of local regulation (within the ovary) and global regula-
tion (between components of the HPG axis) in maintaining control
of steroid synthesis. Incorporation of gene expression data into
the Shoemaker et al. (2008) model suggests that local regulation
reacts to fadrozole to increase gene expression of Steroidogene-
sis enzymes. Higher enzymatic capability is then coupled with
increased cholesterol transport due to testosterone and estradiol
feedback regulation via the pituitary and hypothalamus. The fat-
head minnow appears to react locally in the ovary to increase
steroidogenic enzymes and inter-organ signaling reacts to increase
cholesterol pools available to the enzymes. From this model we can
formulate additional, testable hypotheses by which feedback reg-
ulation, combined with local gene expression, could compensate
for low-dose chemical exposure. In addition, these mechanism-
based models should facilitate the study of the effects of mixtures of
EDCs with different MOA, by predicting inhibition constants of each
Previous
TOC
-------
176
C.T. Ankley et al./Aquatic Toxicology 92 (2009) 168-178
steroidogenic enzyme for the individual chemicals. As information
from additional EDCs are used for parameterization and valida-
tion, the computational models will serve as an effective "module"
within the broader systems framework for making quantitative pre-
dictions of the impacts on endocrine function of either direct or
indirect chemical effects on steroidogenesis.
Documenting alterations in function of the HPG axis at molecu-
lar and biochemical levels is useful to regulatory decision-making
only if these changes can be translated into effects at higher biolog-
ical levels of organization (Fig. 2). To achieve this we are developing
two predictive models: one for male and another for female fathead
minnows. Both models are physiologically-based computational
models that link exposure to EDCs with changes in measured repro-
ductive endpoints such as plasma concentrations of sex steroids and
Vtg. The models are parameterized using information both from
large-scale "control" datasets (Watanabe et al., 2007), and data from
Phase 1 and 3 studies with EDCs with differing MOA. In males, our
model currently accounts for endocrine responses to estradiol or
ethinylestradiol based on their relative binding affinities to ERs and
salient downstream effects of ER activation, such as induction of
Vtg production (Watanabe et al., in press). The model is formulated
in such a way that other estrogenic chemicals (or mixtures of chem-
icals) can be simulated with only minor modifications. Our model
for female fathead minnows incorporates biological processes anal-
ogous to the male model, but extends the male model by including
the AR, and interactions of the receptor with agonists and antago-
nists (K.H.W., unpublished data). Both models yield predictions of
plasma concentrations of sex steroids and Vtg which fit measured
data from unexposed and exposed fathead minnows. To connect the
physiologically-based model at the individual organism level with
a population model for predictions at a higher level of biological
organization, predictions of changes in plasma Vtg concentrations
can be used as input into a dynamic population model (Murphy et
al., 2005).
The inclusion of microarray, proteomics, and metabolomics
measurements in these studies also allows the reverse engineering
of molecular networks (Schadt and Lum, 2006) which can then be
compared with the measured endpoints. The use of this approach
for disease characterization (Loscalzo et al., 2007) and subsequent
target discovery for drug development (Chen et al., 2008) has been
described, and application to environmental risk assessment has
been proposed (Edwards and Preston, 2008). In this project, the
molecular networks derived from genome-wide measurements
provide an unbiased assessment of the physiologically-based mod-
els discussed above. This aids in the interpretation of the existing
models since the completeness of each model can be estimated
based on the percentage of variation in the molecular network
explained by the descriptive model. It also aids in further devel-
opment of the physiologically-based models by providing clues as
to the missing components of each model.
The final component of the project involves development of tools
for the prediction of population-level effects of EDCs. Except for
instances in which threatened or endangered species are involved,
most ecological assessments of the risk of contaminants ultimately
are concerned with potential population-level responses. Kidd et
al. (2007) evaluated the effects of ethinylestradiol on fish popu-
lations in dosing studies with a whole-lake ecosystem; however,
the opportunity to conduct a controlled study of this magnitude is
rare, so the only practical way to routinely link the effects of EDCs
in individuals to population-level impacts is via modeling (Gleason
and Nacci, 2001; Brown et al., 2003; Hurley et al., 2004; Gurney,
2006). Miller and Ankley (2004) describe a modeling approach for
predicting the status of fathead minnow populations exposed to
trenbolone, based on fecundity data from the 21-d Phase 1 study
design. The basic model employs a Leslie matrix in conjunction with
the logistic equation (to account for density dependence) to trans-
late laboratory toxicity information into prediction of population
trajectories. Miller et al. (2007) expanded on this effort by first
relating changes in Vtg to fecundity in female fathead minnows,
and then using this relationship in the population model to predict
population status in fish exposed to EDCs which inhibit produc-
tion of the egg yolk protein, most notably compounds that depress
steroid synthesis (e.g., fadrozole, prochloraz, trenbolone; Fig. 1).
That analysis is unique in that it focuses on a biochemical endpoint,
female Vtg, that reflects both toxic MOA of EDCs and has a functional
relationship to reproductive success (formation of eggs). As such,
within the overall systems framework for the project, the computa-
tional model described by Miller et al. (2007) and Miller and Ankley
(2004) can serve as the basis via which genomic information can
be quantitatively translated to responses in populations.
5. Prospectus
In this paper we describe a MOA/systems-based research effort
with HPG-active chemicals that will help provide the technical basis
for development of predictive toxicology tools (models, in vitro
and short-term in vivo assays) which could improve the efficiency
of current testing and monitoring programs for EDCs. As we con-
template informational needs for chemical risk assessments in the
coming years, it is clear that historical toxicology approaches which
focus mostly on generating empirical data cannot solely suffice.
Toxicologists and risk assessors are being asked to do more with
fewer resources, in a sociopolitical environment that emphasizes
reduced animal testing. Examples of new testing mandates that
promise to require additional toxicity data for a large number of
chemicals include the REACH (registration, evaluation, authoriza-
tion and restriction of chemicals) program in Europe, and the high
production volume challenge program in the US, in addition to a
variety of EDC testing efforts throughout the world. In recognition
of these informational needs, the NRC (2007) proposed a greater
emphasis on predictive toxicology tools to support human health
assessments. There is an analogous requirement for advanced pre-
dictive methods in ecotoxicology, and many of the tools discussed
in the NRC report are applicable to ecological risk assessments.
However, there are added challenges in ecological assessments; for
example, in contrast to human health toxicology, ecological assess-
ments need to extrapolate toxicity from a few (sometimes one)
species to many (sometimes thousands), and require an under-
standing of impacts of chemicals at the population (rather than
individual) level. We feel that the research approach presented
herein provides a broad conceptual framework for developing
mechanism-based, predictive approaches for effectively assessing
the ecological risk of chemicals with a variety MOA, in addition to
EDCs.
Acknowledgements
We thank our many colleagues who have been involved in dif-
ferent aspects of this work, including L Blake, J. Brodin, J. Cavallin, E.
Durhan, K. Greene, M. Kahl, A. Linnum, E. Makynen and N. Mueller
from the Duluth EPA lab; M. Henderson and Q, Teng from the Athens
EPA lab; A. Biales, M. Kostich, D. Lattier and G. Toth from the Cincin-
nati EPA lab; X. Guan, C. Warner, L. Escalon, Y. Deng and S. Brasfield
from the US Army Engineer Research and Development Center; J.
Shoemaker, K. Gayen, and F.J. Doyle III from the University of Califor-
nia at Santa Barbara; K. Kroll and C. Martyniuk from the University
of Florida; and Z. Li from Oregon Health Sciences University.
We also thank A. Miracle, who was involved in initial concep-
tualization of this project, and R. Kavlock and D. Hoff for helpful
comments on an earlier version of the manuscript.
This work was supported in part by the USEPA National Cen-
ter for Computational Toxicology, a grant from the USEPA National
Previous
TOC
-------
G.T. Ankley et al/Aquatic Toxicology 92 (2009) 168-178
177
Center for Environmental Research to the University of Florida,
and the US Army Environmental Quality Installations Program. The
manuscript has been reviewed in accordance with USEPA guide-
lines; however, the views expressed are those of the authors and
do not necessarily reflect USEPA policy. Permission was granted by
the Chief of Engineers to publish this information.
References
Ankley, G.T., Johnson, R.D., 2004. Small fish models for identifying and assessing the
effects of endocrine-disrupting chemicals. Inst. Lab. Anim. Res. J. 45,469-483.
Ankley, G.T., Villeneuve, D.L., 2006. The fathead minnow in aquatic toxicology: past,
present and future. Aquat. Toxicol. 78,91-102.
Ankley, G.T., Jensen, K.M., Kahl, M.D., Korte, J.J., Makynen, E.A., 2001. Description
and evaluation of a short-term reproduction test with the fathead minnow
(Pimephales promeias). Environ. Toxicol. Chem. 20,1276-1290.
Ankley, G.T., Kahl, M.D., Jensen, K.M., Hornung, M.W., Korte, J.J., Makynen, E.A., Leino,
R.L., 2002. Evaluation of the aromatase inhibitor fadrozole in a short-term repro-
duction assay with the fathead minnow (Pimephales promeias). Toxicol. Sci. 67,
121-130.
Ankley, G.T., Jensen, K.M., Makynen, E.A., Kahl, M.D., Korte, J.J., Hornung, M.W., Henry,
T.R., Denny, J.S., Leino, R.L., Wilson, V.S., Cardon, M.C., Hartig, P.C., Gray, L.E.,
2003. Effects of the androgenic growth promoter 17(3-trenbolone on fecundity
and reproductive endocrinology of the fathead minnow (Pimephales promeias).
Environ. Toxicol. Chem. 22,1350-1360.
Ankley, G.T., Jensen, K.M., Durhan, E.J., Makynen, E.A., Butterworth, B.C., Kahl, M.D.,
Villeneuve, D.L., Linnum, A., Gray, L.E., Cardon, M., Wilson, V.S., 2005. Effects
of two fungicides with multiple modes of action on reproductive endocrine
function in the fathead minnow (Pimephales promeias). Toxicol. Sci. 86, 300-
308.
Ankley, G.T., Daston, G.P., Degitz, S.J., Denslow, N.D., Hoke, R.A., Kennedy, S.W.,
Miracle, A.L., Perkins, E.J., Snape, J., Tillitt, D.E., Tyler, C.R., Versteeg, D., 2006. Tox-
icogenomics in regulatory ecotoxicology. Environ. Sci. Technol. 40,4055-4065.
Ankley, G.T, Jensen, K.M., Kahl, M.D., Makynen, E.A., Blake, L.S., Greene, K.J., Johnson,
R.D., Villeneuve, D.L, 2007. Ketoconazole in the fathead minnow (Pimephales
promeias): reproductive toxicity and biological compensation. Environ. Toxicol.
Chem. 26,1214-1223.
Bradbury, S.P., Feijtel, T.C.J., Van Leeuwen, T.C.J., 2004. Meeting the scientific needs
of ecological risk assessment in a regulatory context. Environ. Sci. Technol. 38,
463A-470A.
Breen, M.S., Villeneuve, D.L., Breen, M., Ankley, G.T., Conolly, R.B., 2007. Mecha-
nistic computational model of ovarian steroidogenesis to predict biochemical
responses to endocrine active compounds. Ann. Biomed. Eng. 35, 970-
981.
Brown, A.R., Riddle, A.M., Cunningham, N.L., Kedwards, T.J., Shillabeer, N.S., Hutchin-
son, T.H., 2003. Predicting the effects of endocrine disrupting chemicals on fish
populations. Hum. Ecol. Risk Assess. 9, 761-788.
Chen, Y., Zhu, J., Lum, P.Y., Yang, X., Pinto, S., MacNeil, D.J., Zhang, C, Lamb, J., Edwards,
S., et al., 2008. Variations in DNA elucidate molecular networks that cause dis-
ease. Nature 452, 429-435.
di Bernardo, D., Thompson, M.J., Gardner, T.S., Chobot, S.E., Eastwood, EX., Wojtovich,
A.P., Elliott, S.J., Schaus, S.E., Collins, J.J., 2005. Chemogenomic profiling on a
genome-wide scale using reverse-engineered gene networks. Nat. Biotechnol.
23,377-383.
Edwards, S.W., Preston, R.J., 2008. Systems biology and mode of action based risk
assessment. Toxicol. Sci. 106,312-318.
Ekman, D.R., Teng, Q., Jensen, K.M., Martinovic, D., Villeneuve, D.L, Ankley, G.T,
Collette, T.W., 2007. NMR analysis of fathead minnow urinary metabolites: a
potential approach for studying impacts of chemical exposures. Aquat. Toxicol.
85,104-112.
Ekman, D.R., Teng, Q, Villeneuve, D.L., Kahl, M.D., Jensen, K.M., Durhan, E.J., Ankley,
G.T., Collette, T.W., 2008. Investigating compensation and recovery of fathead
minnow (Pimephales promeias) exposed to 17a-ethynylestradiol with metabolite
profiling. Environ. Sci. Technol 42,4188-4194.
Ekman, D.R., Teng, Q., Villeneuve, D.L., Kahl, D.L., Jensen, K.M., Durhan, E.J., Ankley,
G.T., Collette, T.W., in press. Profiling lipid metabolites yields unique informa-
tion on gender- and time-dependent responses of fathead minnows (Pimephales
promeias) exposed to 17a-ethynylestradiol. Metabolomics.
Feldman, D., 1986. Ketoconazole and other imidazole derivatives as inhibitors of
steroidogenesis. Endocr. Rev. 7,409-420.
Filby, A.L., Thorpe, K.L., Maack, G., Tyler, C.R., 2007. Gene expression profiles revealing
the mechanisms of anti-androgen- and estrogen-induced feminization in fish.
Aquat. Toxicol. 81, 219-231.
Gleason, T.R., Nacci, D.E., 2001. Risks of endocrine-disrupting compounds to wildlife:
extrapolating from effects on individuals to population response. Human Ecol.
Risk Assess. 7,1027-1042.
Gurney, W.S.C., 2006. Modeling the demographic effects of endocrine disrupters.
Environ. Health Perspect. 114 (Suppl. 1), 122-126.
Hao, H., Zak, D.E., Sauter, T., Schwaber, J., Ogunnaike, B.A., 2006. Modeling the VPAC2-
Activated cAMP/PKA signaling pathway: from receptor to circadian clock gene
induction. Biophys.J. 90,1560-1571.
Hill, A.J., Teraoka, H., Heideman, W., Peterson, R.E., 2005. Zebrafish as a model ver-
tebrate for investigating chemical toxicity. Toxicol. Sci. 86, 9-16.
Hoffmann, J.L., Torontali, S.P., Thomason, R.G., Lee, D.M., Brill, J.L., Price, B.B.,
Carr, G.J., Versteeg, D.J., 2006. Hepatic gene expression profiling using
GeneChips in zebrafish exposed to 17a-ethynylestradiol. Aquat. Toxicol. 79,
233-246.
Hoffmann, J.L., Thomason, R.G., Lee, D.M., Brill, J.L., Price, B.B., Carr, G.J., Versteeg, D.J.,
2008. Hepatic gene expression profiling using GeneChips in zebrafish exposed
to 17a-methyldihydrotestosterone. Aquat. Toxicol. 87,69-80.
Hook, S.E., Skillman, A.D., Small, J.A., Schultz, I.R., 2006. Gene expression patterns
in rainbow trout, Oncorhynchus mykiss, exposed to a suite of model toxicants.
Aquat. Toxicol. 77,372-385.
Hurley, M.A., Matthiessen, P., Pickering, A.D., 2004. A model for environmental sex
reversal in fish. J. Theoret. Biol. 227,159-165.
Jensen, K.M., Kahl, M.D., Makynen, E.A., Korte, J.J., Leino, R.L., Butterworth, B.C., Ank-
ley, G.T., 2004. Characterization of responses to the antiandrogen flutamide in
a short-term reproduction assay with the fathead minnow. Aquat. Toxicol. 70,
99-110.
Kahl, M.D., Bencic, D.C., Blake, L.S., Brodin, J.D., Durhan, E.J., Jensen, K.M., Ankley, G.T.,
2007. Evaluation of a novel mechanism of endocrine disruption in the fathead
minnow. In: Abstracts, 28th Annual Meeting of the Society of Environmental
Toxicology and Chemistry, Milwaukee, WI.
Kavlock, R.J., Daston, G.P., DeRosa, C, Fenner-Crisp, P., Gray, L.E., Kaattari, S., et al.,
1996. Research needs forthe risk assessment of health and environmental effects
of endocrine disrupters: a report of the U.S. EPA-sponsored workshop. Environ.
Health Perspect. 104 (Suppl. 4), 715-740.
Kidd, K.A., Blanchfield, P.J., Mills, K.H., Palace, V.P., Evans, R.E., Lazorchak, J.L., Flick,
R.W., 2007. Collapse of a fish population after exposure to a synthetic estrogen.
Proc. Natl. Acad. Sci. 21, 8897-8901.
Lin, C.Y., Viant, M.R., Tjeerdema, R.S., 2006. Metabolomics: methodologies and appli-
cations in the environmental sciences. J. Pestic. Sci. 31, 245-251.
Loscalzo, J., Kohane, I., Barabasi, A.L., 2007. Human disease classification in the
postgenomic era: a complex systems approach to human pathobiology. Mol.
Syst. Biol. 3,124.
Martinovic, D., Blake, L.S., Durhan, E.J., Greene, K.J., Kahl, M.D., Jensen, K.M., Makynen,
E.A., Villeneuve, D.L, Ankley, G.T., 2008. Reproductive toxicity of vinclozolin in
the fathead minnow: confirming an anti-androgenic mode of action. Environ.
Toxicol. Chem. 27,478-488.
Martyniuk, C.J., Gerrie, E.R., Popesku, J.T., Ekker, M., Trudeau, V.L, 2007. Microarray
analysis in the zebrafish (Danio rerio) liver and telencephalon after exposure to
low concentration of 17a-ethinylestradiol. Aquat. Toxicol 84,38-49.
Miller, D.H., Ankley, G.T, 2004. Modeling impacts on populations: fathead minnow
(Pimephales promeias) exposure to the endocrine disrupter 17(3-trenbolone as a
case study. Ecotoxicol. Environ. Safety 59,1-9.
Miller, D.H., Jensen, K.M., Villeneuve, D.L, Kahl, M.D., Makynen, E.A., Durhan, E.J.,
Ankley, G.T., 2007. Linkage of biochemical responses to population-level effects:
a case study with vitellogenin in the fathead minnow (Pimephales promeias).
Environ. Toxicol. Chem. 26,521-527.
Murphy, C.A., Rose, K.A., Thomas, P., 2005. Modeling vitellogenesis in female fish
exposed to environmental stressors: predicting the effects of endocrine dis-
turbance due to exposure to a PCB mixture and cadmium. Repro. Toxicol. 19,
395-409.
National Research Council (NRC), 2007. Toxicity Testing in the 21st century: A Vision
and a Strategy. National Academies Press, Washington, DC.
Samuelsson, L.M., Forlin, L, Karlsson, G., Adolfsson-Erici, M., Larsson, D.G.J., 2006.
Using NMR metabolomics to identify responses of an environmental estrogen in
blood plasma offish. Aquat. Toxicol. 78,341-349.
Schadt, E.E., Lum, P.Y., 2006. Thematic review series: Systems biology approaches
to metabolic and cardiovascular disorders. Reverse engineering gene networks
to identify key drivers of complex disease phenotypes. J. Lipid Res. 47, 2601-
2613.
Shoemaker, J.E., Gayen, 1C, Garcia-Reyero, N., Perkins, E.J., Doyle III, F.J., 2008. Fathead
minnow steroidogenesis—in vitro modeling and experimentation reveals global
regulation of sex hormone synthesis. In: International Conference on Systems
Biology, Gothenburg, Sweden, August 23-27.
Stoskopf, M.K., 2001. Introduction. Inst. Lab. Anim. Res. J. 42, 271-273.
US Environmental Protection Agency. 1998. Endocrine disrupter screening and test-
ing advisory committee (EDSTAC) final report. Office of Prevention, Pesticides
and Toxic Substances, US Environmental Protection Agency, Washington, DC.
http://www.epa.gov/scipoly/oscpendo/pubs/edspoverview/finalrpt.htm.
Villeneuve, D.L, Knoebl, I., Kahl, M.D., Jensen, K.M., Hammermeister, D.E., Greene,
K.J., Blake, L.S., Ankley, G.T., 2006. Relationship between brain and ovary aro-
matase activity and isoform-specific aromatase mRNA expression in the fathead
minnow (Pimephales promeias). Aquat. Toxicol. 76,353-368.
Villeneuve, D.L, Blake, L.S., Brodin, J.D., Greene, K.J., Knoebl, I., Miracle, A.L, Mar-
tinovic, D., Ankley, G.T., 2007a. Transcription of key genes regulating gonadal
steroidogenesis in control and ketoconazole- or vinclozolin-exposed fathead
minnows. Toxicol. Sci. 98,395-407.
Villeneuve, D.L, Larkin, P., Knoebl, I., Miracle, A.L, Kahl, M.D., Jensen, K.M., Makynen,
E.A., Durhan, E.J., Carter, B.J., Denslow, N.D., Ankley, G.T., 2007b. A graphical sys-
tems model to facilitate hypothesis-driven ecotoxicogenomics research on the
teleost brain-pituitary-gonadal axis. Environ. Sci. Technol. 41,321-330.
Villeneuve, D.L, Blake, LS., Brodin, J.D., Cavallin, J.E., Durhan, E.J., Jensen, K.M., Kahl,
M.D., Makynen, E.A., Martinovic, D., Mueller, N.D., Ankley, G.T., 2008. Effects of a
3(3-hydroxysteroid dehydrogenase inhibitor, trilostane, on the fathead minnow
reproductive axis. Toxicol. Sci. 104,113-123.
Villeneuve, D.L, Mueller, N.D., Martinovic, D., Makynen, E.A., Kahl, M.D.,Jensen, K.M.,
Durhan, E.J., Cavallin, J.E., Bencic, D., Ankley, G.T, in press. Direct effects, com-
Previous
TOC
-------
178
C.T. Ankley et al./Aquatic Toxicology 92 (2009) 168-178
pensation and recovery in female fathead minnows exposed to the aromatase
inhibitor fadrozole. Environ. Health Perspect.
Wang, R.-L., Biales, A., Bencic, D., Lattier, D., Kostich, M., Villeneuve, D., Ankley,
G.T., Lazorchak, J., Toth, G., 2008a. DNA microarray application in ecotoxicology:
experimental design, microarray scanning, and factors affecting transcriptional
profiles in a small fish species. Environ. Toxicol. Chem. 27,652-663.
Wang, R.-L., Bencic, D., Biales, A., Lattier, D., Kostich, M., Villeneuve, D., Ank-
ley, G.T., Lazorchak, J., Toth, G., 2008b. DNA microarray-based ecotoxicological
biomarker discovery in a small fish model species. Environ. Toxicol. Chem. 27,
664-675.
Watanabe, K.H., Jensen, K.M., Orlando, E.F., Ankley, G.T., 2007. What is normal? A
characterization of the values and variability in reproductive endpoints of the
fathead minnow, Pimephales promelas. Comp. Biochem. Physiol. 146,348-356.
Watanabe K.H., Li, Z., Kroll, It, Villeneuve, D.L., Garcia-Reyero, N., Orlando, E.F.,
Sepulveda, M.S., Collette, T.W, Ekman, D.R., Ankley, G.T., Denslow, N.D., in press.
A computational model of the hypothalamic-pituitary-gonadal axis in male
fathead minnows. Toxicol. Sci.
World Health Organization (WHO), 2002. IPCS Global Assessment of the State-
of-the-Science of Endocrine Disruptors. International Programme on Chemical
Safety, WHO/PCS/EDC/02.2.
-------
Commentary
Exposure as Part of a Systems Approach for Assessing Risk
Linda S. Sheldon1 and Elaine A. Cohen Hubal2
1National Exposure Research Laboratory, and 2National Center for Computational Toxicology, U.S. Environmental Protection Agency,
Research Triangle Park, North Carolina, USA
BACKGROUND: The U.S. Environmental Protection Agency is facing large challenges in managing
environmental chemicals with increasingly complex requirements for assessing risk that push
the limits of our current approaches. To address some of these challenges, the National Research
Council (NRC) developed a new vision for toxicity testing. Although the report focused only on
toxicity testing, it recognized that exposure science will play a crucial role in a new risk-based
framework.
OBJECTIVE: In this commentary we expand on tlie important role of exposure science in a fully inte-
grated system for risk assessment. \^e also elaborate on tlie exposure research needed to achieve this
vision.
DISCUSSION: Exposure science, when applied in an integrated systems approach for risk assessment,
can be used to inform and prioritize toxicity testing, describe risks, and verify the outcomes of
testing. Exposure research in several areas will be needed to achieve the NRC vision. For example,
models are needed to screen chemicals based on exposure. Exposure, dose—response, and biological
pathway models must be developed and linked. Advanced computational approaches are required
for dose reconstruction. Monitoring methods are needed that easily measure exposure, internal
dose, susceptibility, and biological outcome. Finally, population monitoring studies are needed to
interpret toxicity test results in terms of real-world risk.
CONCLUSION: This commentary is a call for the exposure community to step up to the challenge by
developing a predictive science with the knowledge and tools for moving into the 21st century.
KEY WORDS: computational biology, exposure science, modeling, risk assessment, systems biology.
Environ Health Perspect 117:1181-1184 (2009). doi:10.1289/ehp.0800407 available via http://
dx.doi.org/ [Online 8 April 2009]
The U.S. Environmental Protection Agency
(EPA) and other regulatory agencies are respon-
sible for managing large numbers of environ-
mental chemicals. Although current regulatory
decisions are based on a wide range of tools and
information that represent the best available
science, often limited or no exposure or toxic-
ity data are available for making these decisions
(Judson et al. 2009). Recent statutory changes
require increasingly complex approaches for
evaluating the impact of life-stage vulnerabil-
ity, genetic susceptibility, varying exposure sce-
narios, and exposures to multiple stressors on
environmental health risks. These new require-
ments push the limits of our current tools and
scientific understanding. Fortunately, the rapid
explosion of new computational, physical, and
biological science tools have the potential to
address these challenges and to transform the
ways in which exposure and toxicity testing
come together to assess health risks.
Because of the number of chemicals
involved and the increasing complexity
of future assessments, new approaches are
needed. To examine and address these limita-
tions, the National Research Council (NRC)
evaluated the issues and developed a frame-
work for toxicity testing as it could be applied
to risk assessment. The report, Toxicity Testing
in the 21st Century: A Vision and Strategy
(NRC 2007), articulates a long-range vision
that applies systems biology, rapid assay tech-
nologies, and bioinformatic tools to improve
toxicity testing. Although the focus was
intentionally on toxicity testing, the NRC
recognized that exposure must be a key com-
ponent if the intended goal is to evaluate risks
and inform public health decisions. Exposure
science, when incorporated throughout the
entire framework, will increase the efficiency
of the testing process, help inform toxicity
testing, describe risks, and verify the out-
comes of new risk assessment approaches. The
NRC recommended that exposure science be
considered at every step in the new testing
and risk assessment strategy.
In this commentary, we establish a stra-
tegic framework for the exposure research
needed to achieve a new approach for risk
assessment. Crucial to this vision is the appli-
cation of a systems approach that fully inte-
grates exposure and toxicity information in
a holistic framework for improved public
health decision making. We also elaborate on
the exposure research needed to achieve this
new vision.
A Systems Approach for
Assessing Risk
The authors of Toxicity Testing in the 21st
Century (NRC 2007) proposed to use sys-
tems biology to serve as the basis for a new
toxicity-testing paradigm. The fundamen-
tal construct is to develop in vitro tests to
characterize toxicity pathway perturbations
and then predict health impacts that could
result from these perturbations. If we broaden
this vision, systems theory also provides the
required conceptual framework for linking
exposure science and toxicology in order to
study, characterize, and predict the complex
interactions between humans and environ-
mental chemicals that lead to health risks.
Toxicity pathways, as articulated by the
NRC, are normal pathways for maintain-
ing cellular functions that, when sufficiently
perturbed, will lead to an adverse health out-
come. The consequences of a perturbation
depend on its magnitude, which is related to
dose at the cellular level, the timing and dura-
tion of the perturbation, and the susceptibility
and life stage of the host. Exposure science
provides information on the magnitude, tim-
ing, and duration of individual exposure as
well as the resulting dose at the tissue, cellular,
and even molecular level (Cohen Hubal et al.
2008). Importantly, exposure information will
determine whether toxicity pathways can be
perturbed and whether there is a risk.
A fully integrated systems approach will
reduce many of the uncertainties with current
risk assessment approaches. Understanding
the mechanisms of toxicity pathways will
reduce uncertainties associated with using
animal data to predict human risk. When
integrated with exposure and dose informa-
tion, it also affords the opportunity to reduce
uncertainties associated with using high doses
to predict risk at lower environmental expo-
sures, predict cumulative risks, and predict
risk to susceptible populations.
Exposure Science for the
21st Century
Because of the complex nature of the human
system, health risk predictions associated with
chemical exposures will be only as good as the
least resolved or least understood component.
Advanced tools are available to rapidly exam-
ine toxicity pathways at a depth and breadth
not previously possible. For a fully integrated
system, a comparable set of advanced exposure
Address correspondence to L.S. Sheldon, U.S. EPA,
National Exposure Research Laboratory, Mail Drop
D305-01, 109 Alexander Dr., Research Triangle
Park, NC 27711 USA. Telephone: (919) 541-2205.
Fax: (919) 541-0445. E-mail: sheldon.linda@epa.gov
We thank L. Reiter, R. Kavlock, H. Zenick, and
J. Blancato for valuable discussions and suggestions.
This work was reviewed by the U.S. EPA and
approved for publication, but does not necessarily
reflect official agency policy.
The authors declare they have no competing
financial interests.
Received 18 November 2008; accepted 8 April
2009.
Environmental Health Perspectives • VOLUME 117 I NUMBER 81 August 2009
Previous
TOC
1181
-------
Sheldon and Cohen Hubal
tools must be developed; these tools must be
rapid, efficient, and predictive.
Exposure science provides the linkages
between what is present in the environment
and the internal dose that individuals and
populations receive. A strategic long-term
program for exposure research must develop
predictive computation tools based on a
mechanistic understanding of important (i.e.,
rate-limiting) exposure processes and determi-
nants. High-priority research needs include
the development and application of
• integrated modeling approaches to reliably
predict exposure and dose
• highly efficient screening tools for chemical
prioritization
• easily accessible exposure databases aligned
with toxicity databases
• efficient and affordable tools for generating
new exposure and dose data.
Integrated modeling approaches for pre-
dicting exposure and dose. Computational
models that can be efficiently integrated to
predict exposure and dose at the toxicity path-
way are fundamental to the new risk assess-
ment vision. These models, in turn, should be
integrated with dose-response and biological
pathways models to describe the entire source
to outcome continuum.
Exposure models estimate concentrations of
chemicals in environmental media and describe
activities that bring individuals into contact
with the contaminated media. Several mod-
els have been developed and applied for this
purpose (Williams et al., in press). The U.S.
EPA Stochastic Human Exposure and Dose
SHEDS
D)
=
.as
V)
c
•£
F
<
E
Q_
Q_
•^ o
15 2
— 0>
o
D)
•E
cc
o
1x10-6
5x10-7
0
1 X 1 0~6
8x10-7
4x10-7
2x10-7
c
0.0006
0.0004
0.0002
o
C
^ Dermal <*
exposure *^^**
|
^
500 1,000 1,50
Time (min)
* Inhalation n t
exposure |:|
II I
r^.»J!U
i
„.,.», ,.„ J
500 1,000 1.5C
Time (min)
• Oral exposure 1
I i
J
*1
,«, „_,.. ,,«JiJ
500 1,000 1,50
Time (min)
Simulation (SHEDS) model (U.S. EPA2009a)
can track activities minute by minute through-
out the day and link these activities to environ-
mental concentrations to estimate exposures
by specific route and pathways (Zartarian et al.
2006). The longitudinal aspect of the model
provides the ability to estimate not only the
magnitude, but also the frequency and dura-
tion of exposure over the same time period.
When SHEDS model outputs are linked to a
physiologically based pharmacokinetic (PBPK)
model such as U.S. EPA's Exposure Related
Dose Estimating Model (ERDEM) (U.S. EPA
2007; Zhang et al. 2007), the magnitude, fre-
quency, and duration of internal dose can also
be predicted. Figure 1 illustrates this linkage
for methyl tert-buty\ ether exposures.
Integrated exposure/PBPK models can
be used in several ways. Outputs can be used
directly to inform toxicity testing as well as to
conduct quantitative risk assessments. Linked
models can simulate dose for multiple routes
(inhalation, ingestion, and dermal) and mul-
tiple chemicals simultaneously, thus provid-
ing the ability to evaluate cumulative risks.
Integrated models also provide the ability to
evaluate risks to susceptible populations by
considering differential activities that could
change exposures or differential physiology
that could affect adsorption, distribution,
metabolism, or elimination characteristics.
Finally, the models can be used in reverse for
dose reconstruction as an alternative approach
for comparing toxicity testing results to popu-
lation exposures (Georgopoulos et al. 2008;
Tan etal. 2007).
ERDEM
500 1,000
Time (min)
1,500
Figure 1. Illustration of linked exposure (SHEDS) and dose (ERDEM) models for methyl fert-butyl ether.
Realizing the potential for integrated
modeling approaches requires a coordinated
and sustained research effort. PBPK models
need to be extended to allow dose estimation
at the cellular and molecular level. Integrated
exposure/PBPK models must be enhanced
to provide distributional outputs along with
uncertainty and variability. Developing sys-
tems that are efficient and generalizable must
be a part of this effort. New data and new
approaches are needed for exposure recon-
struction in order to reduce uncertainties
with current approaches. Current efforts
(Georgopoulos and Lioy 2006; Rosenbaum
et al. 2007) to provide models that use a com-
mon platform and/or common programming
language must continue.
At the same time, research is required
to develop approaches for estimating model
inputs and parameters without resource-inten-
sive and burdensome studies. For example,
environmental informatics, quantitative struc-
ture—activity relationships (QSARs), and com-
putational chemistry approaches should be
developed to predict and quantify behaviors
such as environmental fate and transport or
metabolism. Development of metabolic pre-
dictors or simulators that can address single
chemicals, multiple chemicals, and the inter-
action among chemicals should be accelerated
(Mekenyan et al. 2005). Novel statistical and
informatic approaches should be applied to
extant exposure data to facilitate the identifica-
tion of critical metrics that represent personal
exposure through time, place, life stage, life-
style, or behavior.
Exposure screening tools for accelerated
chemical prioritization. Current risk assess-
ment approaches cannot meet demands for
the large number of chemicals that must be
evaluated. Screening tools are needed that
reliably identify those chemicals that will
require more comprehensive risk assessments.
Chemical prioritization should consider
both exposure and hazard. The U.S. EPA,
through its ToxCast program (U.S. EPA
2009b), is developing rapid in vitro assays to
screen chemicals for further testing based on
toxicity (Dix et al. 2007). Innovative rapid-
screening tools based on exposure are also
needed. Predictive approaches for estimat-
ing important parameters for screening need
to be developed. Ideally, these tools should
account for chemical use, physical and chemi-
cal properties, occurrence and co-occurrence
of chemicals, potential exposure scenarios,
routes of exposure, and various exposure fac-
tors. This will include developing approaches
that describe a chemical's behavior in the
environment as well as approaches to identify
important human activities that will impact
exposure. Exposure prioritization approaches
will require easily accessible databases, as
described below.
1182
VOLUME 1171 NUMBER 8 I August 2009 • Environmental Health Perspectives
Previous
-------
Exposure and computational toxicology
One plausible approach may be to formu-
late an exposure classification index based on
a limited set of metrics designed to efficiently
cover exposure potential (Cohen Hubal et al.
2008). As a first step, innovative approaches
for chemical prioritization (e.g., Arnot and
MacKay 2008; Hays et al. 2007) as well as
indexing approaches from other fields should
be reviewed and mined. This index could be
"trained" on data-rich chemicals and products
and then validated on a representative set of
chemicals for which little exposure data are
available. In this way, a limited set of criti-
cal metrics could be identified for efficient
screening of new chemicals. Finally, because
consumer products often incorporate mul-
tiple chemicals in a variety of forms, rapid
experimental screening protocols that measure
the potential for availability or release of these
compounds into exposure media are under
early development and should be pursued fur-
ther (Little and Cohen Hubal, in press).
Significant research and model develop-
ment activities are currently under develop-
ment within the U.S. EPA (U.S. EPA 2005)
as well as in Canada and Europe (Bridges
2007; Environment Canada 1999; Van der
Wielen 2007). Partnerships with these groups
should be fostered to leverage and establish
collaborative exposure science research for
future chemical screening and prioritization.
Exposure databases. Easily accessible
exposure databases that can be linked to each
other and with toxicity databases can and
should be developed immediately. Data on
chemical manufacture, product use, environ-
mental fate, media concentrations, biomarker
levels, and metabolism should be identified.
International standards for exposure data rep-
resentation should be discussed. Approaches
for improving access to human exposure data
and for facilitating links between exposure
and toxicity data should be implemented.
Existing tools and platforms that are cur-
rently being implemented with environmen-
tal toxicity information should be adapted
for exposure information to provide the most
useful links to existing toxicity data. Chemical
structure annotation of exposure-related data,
such as could be provided by the Distributed
Structure Searchable Toxicity (DSSTox) data-
base (Richard et al. 2006, 2008; U.S. EPA
2009c), and incorporation of such data into
the new Aggregate Computational Toxicology
Resource (ACToR) (Judson et al. 2008) will
greatly enhance linkages between these data
and toxicity-related human health end points.
For maximum impact, this activity should
be conducted in collaboration with inter-
national partners working to achieve similar
goals (Environment Canada 1999; Van der
Wielen 2007).
Efficient monitoring methods for assessing
risk. Population-based and surveillance studies
will provide the ability to link the results from
toxicity testing to the real world and to track
our progress in protecting public health. To be
feasible, new low-cost, low-burden methods
and approaches for conducting these studies
will be needed. New technologies need to be
applied to develop a toolbox of methods for
assessing exposure, susceptibility, and biologi-
cal response in large surveillance studies. New
sensor technologies, applications of nano-
technology, geographic information systems,
and genomics assays need to be developed and
put into use for this purpose.
Emerging tools in molecular biology
provide the potential to develop cellular and
molecular indicators of exposure and bio-
logical response. Better understanding of
genomic expression may also provide insight
into factors impacting differences in suscep-
tibility to chemical exposure in the human
population (Oberemm et al. 2005). "Omics"
should be explored as a way to identify
expression patterns associated with exposure
to individual chemicals or chemical mixtures.
Such technologies could then provide link-
ages between exposure and health outcomes
in population studies. Development of envi-
ronmental and/or molecular indicators of
exposure combined with development of
novel sensor-based monitoring tools will pres-
ent the opportunity for simultaneous, near-
real-time measurement of exposure and dose
to multiple real-world stressors in mixtures
(Weis et al. 2005). A strong and immedi-
ate research effort is required for novel tech-
nologies that will generate the data required
for risk assessments and decisions that truly
protect public health.
Conclusions
Exposure science is crucial for addressing many
of our important and complex environmen-
tal health issues. As discussed here, exposure
science is essential for toxicity testing to be
valuable in public health protection. A systems
approach is required that fully integrates expo-
sure and toxicity into a holistic framework for
risk assessment.
The exposure community must step up
to the challenge to develop a robust and pre-
dictive science that can be used to address
the complex problems in the 21st century. A
research program that provides the necessary
exposure data and tools within an integrated
framework will need to be multidisciplinary
and take advantage of collaborative opportuni-
ties. Key to this work are strong collaborations
within the exposure community and with
those researchers who are developing infor-
mation on toxicity pathways and conducting
toxicity testing. Multiple collaborations are
needed to ensure that
• chemical prioritization considers both
exposure and toxicity
• databases are developed
• analysis is conducted using these databases
to understand exposures, doses, and toxicity
• new information on biological interactions
and pathways is used to develop the appro-
priate indicators for exposure and surveillance
studies
• models on exposure and dose are linked for
extrapolations
• feedback loops are developed to inform
future planned research.
Collaborations should include partnerships
between governmental and nongovernmental
research groups as well as academia and indus-
try for developing and applying new exposure
tools. Just as a new vision and initiatives have
been developed for toxicity testing, it is now
time for the exposure community to dedicate
itself to engaging in similar activities to move
our science into the 21st century.
REFERENCES
Arnot JA, MacKay D. 2008. Policies for chemical hazard and
risk priority setting: can persistence, bioaccumulation,
toxicity, and quantity information be combined? Environ
SciTechnol 42(131:4648-4654.
Bridges J. 2007. REACH epilogue. J Expo Sci Environ Epidemiol
17(suppl):S101-S104.
Cohen Hubal EA, Richard AM, Imran S, Gallagher J, Kavlock R,
Blancato J, et al. 2008. Exposure science and the U.S. EPA
National Center for Computational Toxicology. J Expo
Sci Environ Epidemiol; doi: 10.1038/jes.2008.70 [Online
5 November 2008].
Dix D, Houck K, Martin M, Richard A, Setzer R, Kavlock R. 2007.
The ToxCast program for prioritizing toxicity testing of
environmental chemicals. Toxicol Sci 95(11:5-12.
Environment Canada. 1999. Domestic Substances List:
Categorization and Screening Program. Gatineau, Quebec.
Available: http://www.ec.gc.ca/substances/ese/eng/dsl/
dslprog.cfm [accessed 23 June 2009].
Georgopoulos P, Lioy P. 2006. Theoretical aspects of human
exposure and dose assessment to computational model
implementation: the Modeling ENvironment for TOtal Risk
Studies (MENTOR]. J Toxicol Environ Health Part B Crit
Rev 9(61:457-483.
Georgopoulos P, Sasso A, Isukapalli S, Lioy P, Vallero D,
Okino M, et al. 2008. Reconstructing population exposures to
environmental chemicals from biomarkers: challenges and
opportunities. J Expo Sci Environ Epidemiol; doi:10.1m038/
jes.2008.9 [Online 26 March 2008].
Hays SM, Becker R, Leung H, Aylward L, Pyatt D. 2007.
Biomonitoring equivalents: a screening approach for inter-
preting biomonitoring results from a public health risk
perspective. Regul Toxicol Pharmacol 47(11:96-109.
Judson R, Richard A, Dix D, Houck K, Elloumi F, Martin M,
et al. 2008. ACToR—Aggregated Computational Toxicology
Resource. Toxicol Appl Pharmacol 233(11:7-13.
Judson R, Richard A, Dix DJ, Houck K, Martin M, Kavlock R,
et al. 2009. The toxicity data landscape for environmental
chemicals. Environ Health Perspect 117:685-695.
Little J, Cohen Hubal E. In press. Rapid screening to estimate
health risks of materials and products. In: Proceedings
of the 11th International Conference on Indoor Air
Quality and Climate, 17-22 August 2008, Copenhagen,
Denmark.
Mekenyan 0, Jones J, Schmieder P, Kotov S, Pavlov T,
Dimitrov S. 2005. Performance, reliability, and improvement
of a tissue-specific metabolic simulator. In: Abstract Book.
SETAC North America 26th Annual Meeting "Environmental
Science in a Global Society: SETAC's Role in the Next
25 Years," 13-17 November, Baltimore, MD. Pensacola,
FL:SETAC North America—Society of Environmental
Toxicology and Chemistry.
NRC (National Research Council]. 2007. Toxicity Testing in
the 21st Century: A Vision and a Strategy. Washington,
DC:National Academies Press.
Environmental Health Perspectives • VOLUME 117 I NUMBER 81 August 2009
Previous
TOC
1183
-------
Sheldon and Cohen Hubal
Oberemm A, Onyon L, Gundert-Remy U. 2005. How can toxico-
genomics inform risk assessment? ToxicolAppI Pharmacol
207(suppl 21:592-598.
Richard AM, Gold LS, Nicklaus MC. 2006. Chemical structure
indexing of toxicity data on the internet: moving toward a
flat world. CurrOpin Drug Discov Devel 9(31:314-325.
Richard A, Yang C, Judson R. 2008. Toxicity data informatics:
supporting a new paradigm for toxicity prediction. Toxicol
Mech Methods 18:103-118.
Rosenbaum R, Margni M, Jolliet 0. 2007. A flexible algebra
framework for the multimedia multipathway modeling of
emissions to impacts. Environ Int 33:624-634.
Tan YM, Lios KH, Clewell HJ III. 2007. Reverse dosimetry: inter-
preting trihalomethanes biomonitoring data using physi-
ologically based pharmacokinetic modeling. J Expo Sci
Environ Epidemiol 17(71:591-603.
U.S. EPA (U.S. Environmental Protection Agency]. 2005. Chemical
Assessment and Management Program (ChAMP). Available:
http://www.epa.gov/champ/ [accessed 23 June 2009].
U.S. EPA (U.S. Environmental Protection Agency]. 2007.
Exposure Related Dose Estimating Model (ERDEM).
Available: http://www.epa.gov/heasd/products/erdem/
erdem.htm [accessed 23 June 2009].
U.S. EPA (U.S. Environmental Protection Agency]. 2009a.
SHEDS-Multimedia. Details of SHEDS-Multimedia ver-
sion 3: ORD/NERL's Model to Estimate Aggregate and
Cumulative Exposures to Chemicals. Available: http://
www.epa.gov/heasd/products/sheds_multimedia/sheds_
mm.html [accessed 23 June 2009].
U.S. EPA (U.S. Environmental Protection Agency]. 2009b.
ToxCase™ Program: Predicting Hazard, Characterizing
Toxicity Pathways, and Prioritizing the Toxicity Testing of
Environmental Chemicals. Available: http://www.epa.gov/
ncct/toxcasV [accessed 23 June 2009].
U.S. EPA (U.S. Environmental Protection Agency]. 2009c.
DSSTox: Distributed Structure-Searchable Toxicity
(DSSTox] Database Network. Available: http://www.epa.
gov/NCCT/dsstox/ [accessed 23 June 2009].
Van der Wielen A. 2007. REACH: next step to a sound chemi-
cals management. J Expo Sci Environ Epidemiol 17(supp1|:
S2-S6.
Weis BK, Balshaw J, Barr D, Brown A, Ellisman M, Lioy P. 2005.
Personalized exposure assessment: promising approaches
for human environmental health research. Environ Health
Perspect 113:840-848.
Williams PRO, Hubbell BJ, Weber E, Fehrenbacher C, Hrdy D,
Zartarian V. In press. An overview of exposure assess-
ment models used by the U.S. Environmental Protection
Agency. J Expo Sci Environ Epidemiol.
1184
VOLUME 1171 NUMBER 8 I August 2009 • Environmental Health Perspectives
Previous
-------
Genetic Basis for Adverse Events after Smallpox
Vaccination
David M. Reif,13a Brett A. McKinney,7 Alison A. Motsinger,31' Stephen J. Chanock,8 Kathryn M. Edwards,5
Michael T. Rock,5 Jason H. Moore,1-2 and James E. Crowe, Jr.4-5-6
'Computational Genetics Laboratory and department of Genetics, Dartmouth Medical School, Lebanon, New Hampshire; 3Center for Human
Genetics Research, Vanderbilt University, and 4Program in Vaccine Sciences and Departments of 5Pediatrics and 6Microbiology and Immunology,
Vanderbilt University Medical Center, Nashville, Tennessee; 'Department of Genetics, University of Alabama School of Medicine, Birmingham;
and 8Center for Cancer Research and Division of Cancer Epidemiology and Genetics, National Cancer Institute, National Institutes of Health,
Bethesda, Maryland
(See the editorial commentary by Relman, on pages 4-5.)
Identifying genetic factors associated with the development of adverse events might allow screening before vac-
cinia virus administration. Two independent clinical trials of the smallpox vaccine (Aventis Pasteur) were con-
ducted in healthy, vaccinia virus-naive adult volunteers. Volunteers were assessed repeatedly for local and sys-
temic adverse events (AEs) associated with the receipt of vaccine and underwent genotyping for 1442 single-
nucleotide polymorphisms (SNPs). In the first study, 36 SNPs in 26 genes were associated with systemic AEs
(P ^ .05); these 26 genes were tested in the second study. In the final analysis, 3 SNPs were consistently associated
with AEs in both studies. The presence of a nonsynonymous SNP in the methylenetetrahydrofolate reductase
(MTHFR) gene was associated with the risk of AE in both trials (odds ratio [OR], 2.3 [95% confidence interval {CI j,
1.1-5.2] [P = .04] and OR, 4.1 [95% CI, 1.4-11.4] [P<.01]). Two SNPs in the interferon regulatory factor-1
(IRF1) gene were associated with the risk of AE in both sample sets (OR, 3.2 [95% CI, 1.1-9.8] [P = .03] and OR,
3.0 [95% CI, 1.1-8.3] [P = .03]). Genetic polymorphisms in genes expressing an enzyme previously associated
with adverse reactions to a variety of pharmacologic agents (MTHFR) and an immunological transcription factor
(IRF1) were associated with AEs after smallpox vaccination in 2 independent study samples.
Although reactions occurring after inoculation of
vaccinia virus were commonly observed in recent
population-wide vaccination programs [1], the biologi-
cal basis for these adverse events (AEs) is not well under-
stood. Performance of 2 independent clinical studies of a
single vaccinia virus vaccine at our study site afforded us
Received 14 September 2007; accepted 28 December 2007; electronically
published 2 May 2008.
Potential conflicts of interest: J.E.C. received research funding from Sanofi-
Aventis and Vaxgen and a joint Small Business Technology Transfer award (with
Mapp Pharmaceuticals). He has consulted for Medlmmune, Vaxin, Evogenix,
Symphogen, and Syngenta. K.M.E. received research funding from Sanofi-Aventis,
Medlmmune, Vaxgen, Merck, and Wyeth. She has also consulted for Medlmmune
and Wyeth. All other authors: no conflicts.
Financial support: Vaccine Trials and Evaluation Unit, National Institutes of
Health (Mini/National Institute of Allergy and Infectious Diseases (NIAID) (contract
N01-AI-25462 for Division of Microbiology and Infectious Diseases (DMID) studies
02-054 and 03-044); NIH/NIAID (grants K25-AI-64625, R21-AI-59365, and R01-AI-
59694); and NIH/National Institute of General Medical Sciences (grant R01
GM-62758). The clinical study was supported in part by the National Center for
Research Resources, NIH (grant M01 RR-00095).
The Journal of Infectious Diseases 2008; 198:16-22
© 2008 by the Infectious Diseases Society of America. All rights reserved.
0022-1899/2008/19801 -0006$15.00
DOI: 10.1086/588670
the unique opportunity to assess genetic factors that
might predict systemic AEs. All of the vaccinia virus-
naive subjects who were enrolled in the study developed
pock formation at the vaccination site, and a subset ex-
perienced systemic reactions that included fever, rash, or
regional lymphadenopathy. Because poxviruses have
developed multiple mechanisms by which to evade host
immune responses, such as targeting of primary innate
immunity and manipulation of intracellular signal
transduction pathways [2], we questioned whether sub-
jects experiencing AEs exhibited unique genetic poly-
morphisms in these pathways that made them more sus-
ceptible to these reactions.
The funding organizations played no role in the design and conduct of the study;
collection, management, analysis, and interpretation of the data; or in the prep-
aration, review, or approval of the manuscript.
a Present affiliation: National Center for Computational Toxicology, US Envi-
ronmental Protection Agency, Research Triangle Park, North Carolina.
b Present affiliation: Bioinformatics Research Center, Department of Statistics,
North Carolina State University, Raleigh.
Reprints or correspondence: Dr. James E. Crowe, Jr., T2220 Medical Center N,
1161 21st Ave. S, Nashville, TN 37232-2905 (james.crowe@vanderbilt.edu).
16 • JID 2008:198 (1 July) • Reifetal.
Previous
-------
In earlier studies, we characterized humoral and cellular im-
mune responses and outlined patterns of systemic cytokine ex-
pression after smallpox vaccination [3-8]. In the present study,
we utilized data collected during 2 independent studies to iden-
tify stable genetic factors associated with AEs. Because there is
failure to replicate the results of many genetic association studies
during subsequent studies, we sought to repeat the assessment in
an additional study group [9,10]. The fact that the results of our
first study were independently replicated in the second study
strengthens the plausibility of these genetic associations. An
identical panel of candidate single-nucleotide polymorphisms
(SNPs) was evaluated in each of the studies. Subjects with sys-
temic AEs, including fever, lymphadenopathy, or generalized ac-
neiform rash, were compared with subjects who did not experi-
ence these reactions. The data used in both studies were the
genotypes at 1442 SNPs across at least 386 candidate genes. The
present investigation provides, for 2 independent data sets, im-
portant preliminary findings addressing the contribution of
common genetic variants to a complex clinical phenotype,
which also is of substantial importance with respect to public
health.
Study subjects. The vaccines, study subjects, and study design
used in both of the clinical trials have been previously described
in detail. Both trials were conducted at the National Institutes of
Health (NIH)-funded Vaccine and Treatment Evaluation Unit
at Vanderbilt University (Nashville, Tennessee) [4, 8, 11]. The
first study [7] enrolled 85 healthy, vaccinia virus-naive adults in
genotyping studies, and the second study [11] enrolled 46
healthy, vaccinia virus-naive adults. In both studies, individuals
were asked to self-identify their ethnic background. Both studies
complied with the policies of the internal review boards of
Vanderbilt University and the NIH, and written informed con-
sent was obtained from all individuals.
Clinical assessments. For both studies, the same team of
trained physicians and nurses used the same forms to obtain a
medical history and to record local and systemic AEs occurring
after vaccination. Subjects were examined at regular intervals
(on days 3-5, 6-8, 9-11, 12-15, and 26-30 after vaccination).
Local and systemic AEs were recorded. Subjects who had an oral
temperature >38.3°C at any time during the study, generalized
skin eruptions on areas noncontiguous to the site of vaccination
[11], or enlarged or tender regional lymph nodes associated with
vaccination were defined as subjects experiencing systemic AEs.
Identification of genetic polymorphisms. We used a previ-
ously described custom SNP panel based on the National Cancer
Institute (NCI) SNP500 Cancer Project [12]; this panel targets in-
vestigation of soluble-factor mediators and signaling pathways,
many of which have known immunological significance [13]. In
this panel, there is heavy weighting toward nonsynonymous SNPs
(i.e., SNPs that result in an amino acid substitution). Genotyping
for SNPs was performed using DNA directly amplified from Ep-
stein Barr virus (EBV)-transformed B cells generated from periph-
eral blood samples collected from each subject. Genotyping was
performed at the Core Genotyping Facility of the NCI in Gaithers-
burg, Maryland. Genotypes were generated using Illumina Golden-
Gate assay technology. Of the 1536 SNPs assayed, a total of 1442
genotypes passed quality control filters for both the first and second
sample sets.
Statistical analysis. The clinical and demographic charac-
teristics, including age, sex, and race, noted in the first and sec-
ond studies were compared using Student's t test (for age) and
2-sample tests of proportions (for AE status and for sex and
race). Allele frequencies were estimated by dividing the total
number of copies of individual alleles by the total number of
alleles in the sample, and the frequencies noted in the 2 studies
were compared using a 2-sample test of proportions. Deviations
in the fitness for Hardy-Weinberg proportion were evaluated
using the exact test as described by Wigginton et al. [ 14].
We chose a 2-stage design for identifying and replicating
genetic associations in the independent clinical trials. This
study design was selected with the goal of minimizing type I
errors (false-positive results). For comparison, we also per-
formed genetic association analysis in a single pooled sample.
In the first study, we tested for potential associations between
each of the 1442 SNPs passing quality control filters, as well as
for the occurrence of AEs, by use of logistic regression. For
each SNP in the first sample set, we recorded the odds ratio
(OR) estimate and P value of the likelihood ratio test for a
univariate logistic model. No correction for multiple com-
parisons was made in the first sample set, because we reserved
the second study sample set for determination of probable
true-positive results. In the second sample set, we tested only
those SNPs that had an AE-associated P value of ss.05 in the
first study. A significant SNP association in the first study was
considered to have been replicated if it met the following
criteria in the second study: (1) an OR that consistently asso-
ciated the risk of AE with the same genotypes and (2) a P value
=£0.05. To obtain an empirical probability of meeting our
replication criteria purely by chance, we generated 1000 sim-
ulated data sets from both study sample sets by permuting
case-control labels. An additional association for which
P = .06 is discussed below because of its high biological plausi-
bility.
Patterns of linkage disequilibrium (LD) between replicated
SNPs on the same chromosome were assessed using Haploview
software (version 3.32; Broad Institute) [15]. Haplotypes were
inferred for SNPs in high LD, by use of the iterative approach
described by Lake et al. [16]. The resulting haplotypes were
tested for association with AEs by use of univariate logistic mod-
els. Statistical analyses and simulations were performed using
software that we wrote using the R language (version 2.5.1; R
Genetic Basis for Adverse Events • JID 2008:198 (1 July) • 17
Previews
TOG
Next
-------
Table 1. Summary of data on adverse event (AE) status and
subject age, sex, and race in both studies.
Characteristic
AE statusb
Age, mean ± SD,
years
Sexc
Raced
First study
(n = 85)
16/69
23.2 ± 3.9
40/45
84/0/1
Second study
(n = 46)
24/22
24.2 ± 3.8
27/19
44/1/1
pa
-------
Table 3. Distribution of genotypes at single-nucleotide poly-
morphisms (SNPs) in MTHFR, IRF1, and IL4.
Gene, SNPa
MTHFR, 1801133
IRF1
9282763
839
IL4
2070874
2243268
2243290
SNP
location,11
bp Genotype
6393745 CC
CT
TT
34237146 AA
AG
GG
34234139 GG
AA
AG
34424723 CC
CT
TT
34428976 AA
AC
CC
34433182 CC
AA
AC
No. (%) of
subjects with
genotype,
by study
First
(n = 85)
36(42)
39(46)
10(12)
39(46)
43(51)
3(4)
39(46)
43(51)
3(4)
52(62)
28(33)
4(5)
52(62)
27(32)
5(6)
53(62)
26(31)
6(7)
Second
(n = 46)
18(39)
21(46)
7(15)
17(37)
24(52)
5(11)
17(37)
24(52)
5(11)
34(74)
12(26)
0(0)
34(74)
12(26)
0(0)
34(74)
12(26)
0(0)
a By rs number.
b As determined according to dbSNP (Human Genome Build 36.1; National
Center for Biotechnology Information).
tion with 5?1 functional variants in that region. Because of the
close physical proximity of the associated variants in the 2 genes,
Haploview software (version 3.32) [15] was used to examine the
patterns of LD among those variants in each sample. Figure 1
shows that the LD plots for SNPs in the 2 genes follow the same
pattern in each study sample. Although there is strong LD be-
tween the SNPs within the 2 genes, there is little evidence for LD
between the 2 genes, indicating that the associations for each
gene are statistically separate signals.
This region of chromosome 5q31 contains discrete haplotype
blocks [22]. Accordingly, haplotypes were inferred for AE-
associated SNPs in IRF1 (rs839 and rs9282763) and IL4
(rs2070874, rs2243268, and rs2243290). In both studies, 2 IRF1
haplotypes accounted for all subjects. The common IRF1 haplo-
type listed in table 4 was found in 71% of the first sample set and
63% of the second sample set. The rare IRF1 haplotype was sig-
nificantly associated with AEs in both studies (P = .03). Across
both studies, 2 different 3-SNP haplotypes in IL4 were found
among 99% of subjects. The common IL4 haplotype shown in
table 4 was found in 78% of the first sample set and 87% of the
second sample set. The rare IL4 haplotype was significantly as-
sociated with the risk of AEs in the first study (P = .05); the
association was similar in the second study (P = .06).
DISCUSSION
MTHFR and IRF1. The candidate genes noted to have the
strongest association with AEs in both studies include a metab-
olism gene previously associated with adverse reactions to a va-
riety of pharmacologic agents (MTHFR) and an immunological
transcription factor (IRF1) gene. The statistical results from
these studies have strong biological plausibility and are in agree-
ment with previous work on the immune response to poxvi-
ruses.
An SNP in the 5,10-methylenetetrahydrofolate reductase gene
(MTHFR; rslSOl 133) was associated strongly with the risk of AE
in both studies. This nonsynonymous SNP in exon 5 causes
an amino acid change from alanine to valine, and functional
characterization of this SNP demonstrated that it is thermola-
bile and affects both the quantity and activity of the MTHFR
enzyme [23]. The enzyme catalyzes the conversion of 5,10-
methylenetetrahydrofolate to 5-methyltetrahydrofolate, which
is a cosubstrate for homocysteine remethylation to methionine.
MTHFR function provides pools of methyl groups that are cru-
cial for the control of DNA synthesis and repair mechanisms
[24]. MTHFR is a key enzyme in homocysteine metabolism,
which plays a major role in regulating endothelial function. In
the future, it may be of interest to examine the association of
genetic variation in this gene with the rare cardiac events that
occur after vaccination.
Genetic variation of MTHFR has been associated with a range
of clinical outcomes, including altered cardiovascular function,
organ transplantation, toxicity of immunosuppressive drugs,
and systemic inflammation [25-28]. Elevated plasma levels of
homocysteine stimulate endothelial inflammatory responses,
which could contribute to the development of systemic AEs. Al-
ternatively, because vaccination elicits immune responses in-
volving the rapid proliferation of cells, demand for DNA synthe-
sis metabolites would be elevated, and alterations in the level or
activity of the MTHFR enzyme may exert significant influence
over this process.
Interferon (IFN) regulatory factor-1 (IRF-1). The IRF1
gene is part of the immunological gene cluster on chromosome
5q31. We found 2 SNPs in IRF1 that were significantly associated
with AEs in both study samples. The IRF1 gene encodes an im-
portant member of the IFN regulatory transcription factor (IRF)
family. The IRF family regulates IFNs and IFN-inducible genes.
IRF1 activates transcription of type 1 IFN-a and IFN-/3, as well
as genes induced by type 2 IFN-y [29]. Many viruses target IRFs
to evade host immune responses by binding to cellular IRFs and
blocking transcriptional activation of IRF targets [30].
Polymorphisms in the gene coding for a transcription fac-
tor with such far-reaching effects as IRF1 could have pro-
Genetic Basis for Adverse Events • JID 2008:198 (1 July) • 19
Previous
TOC
-------
*•>
S
Block 1 (3kb) Block 2 (8kb)
.Ill
Block 1 (3kb) Block 2 (8kb)
B
Figure 1. Haploview plot of single-nucleotide polymorphisms (SNPs) at chromosome 5q31.1. A, First study. B, Second study. Squares are shaded to
denote the strength of evidence of linkage disequilibrium (LD) between the pairwise markers. Black squares, strong evidence of LD (r2 >0.90); light-gray
squares, weak evidence of LD (r2 <0.]0)', and white squares, no evidence of LD (r2 <0.0). The same 2 LD blocks are apparent in both studies,
encompassing SNPs in IRF1 (rs839 and rs9282763) or IL4 (rs2070874, rs2243268, and rs2243290).
found effects on the proper immune response to and clear-
ance of vaccinia virus. Mice deficient in IFN receptors are
especially susceptible to vaccinia virus infection, suggesting
an important role for these molecules in controlling vaccinia
virus infection [31]. Vaccinia virus dedicates several host-
modifying genes to counteracting IFNs. For example, the vi-
ral gene B18R encodes a protein that serves as a viral IFN-a/
/3-binding protein that binds IFNs from several species [32].
This protein also can bind to the cell surface after secretion,
thus preventing host IFN from binding to cellular IFN recep-
tors [33]. Although the SNPs identified in IRF1 and IL4 do
not change amino acids in the encoded proteins, recent evi-
dence suggests that synonymous SNPs, such as rs839, can
alter regulation of mRNA or splice junctions [34,35]. It is also
plausible that one or both SNPs are in LD with the causal
variant not tested in this study.
Interleukin-4. Genetic polymorphisms in this major cyto-
kine gene involved in adaptive immunity to viruses also may be
associated with AEs, albeit with a P value of .06 in our relatively
small replication study. We found 3 SNPs in IL4 that may be
associated with AEs in both studies. There was high intragenic
LD (r2 > 0.9) between the tested SNPs within each gene (IRF1
and IL4) and haplotypes inferred separately for each of these
genes mirrored the significant risk patterns of the SNPs observed
individually. Thus, the fact that multiple SNPs that were in high
LD were identified in regions oflRFl and IL4 strongly suggests
that there are additional markers in LD, several of which could
functionally contribute to the risk of AEs.
Table 4. Haplotypes inferred for adverse event (AE)-associated single-nucleotide
polymorphisms (SNPs) in //?F/(rs839 and rs9282763) and /W(rs2070874, rs2243268, and
rs2243290).
Haplotype
Risk
First study
Second study
Gene, SNPa at baseline11 haplotypec OR (95% Cl)d Pe OR (95% Cl)d Pe
IRF1
9282763
839
IL4
2070874
2243268
2243290
C
A
C
G
A
T
C
A
3.2(1.0-10.2) .03 3.0(1.0-9.0) .03
2.4(1.0-5.7) .05 3.8(1.0-14.4) .06
NOTE. Cl, confidence interval; OR, odds ratio.
a By rs number.
b Most common haplotype, considering 2 SNPs in IRF1 or 3 SNPs in IL4.
= Rare (variant) haplotype, considering 2 SNPs in IRF1 or 3 SNPs in IL4.
d Estimated OR comparing the risk haplotype with the haplotype at baseline (95% Cl).
3 By likelihood ratio x2 test with 1 df.
20 • JID 2008:198 (1 July) • Reifetal.
Previous
-------
The IIA gene encodes a pleiotropic cytokine produced by a
variety of immune cells, especially activated T cells. IL4 con-
trols humoral immune responses, isotype switching, and sup-
pression of cytotoxic T cell function and expansion. Thus,
genetic polymorphisms related to inappropriate regulation of
IIA expression and/or activity of IL-4 cytokine could be asso-
ciated with overstimulated inflammatory responses leading
to the development of clinical AEs. Previous studies of the
role of IL4 in the pathogenesis of poxvirus have shown that
IL4 has a central role in altering the adaptive immune re-
sponse. Overexpression of IL4 during infection with recom-
binant poxviruses encoding IL4 suppresses the induction of
cytotoxic T cell activity by inhibiting CDS ' T cell prolifera-
tion, which increased the pathogenicity of such recombinant
viruses even in previously immunized animals [36]. 1L4 also
plays a role in preventing optimal innate immune responses
to poxviruses. Secretion of IL-4 during vaccinia virus infec-
tion in individuals with atopic dermatitis alters the cytokine
milieu, resulting in blocking of production of the antimicro-
bial peptide LL-37; this accounts, in part, for the increased
risk of vaccinia virus infection in subjects with atopic derma-
titis [37].
Model of pathogenesis. Because the outcome of interest
here was the aggregation of specific AEs, it is logical that > 1 gene
may be involved. The genes with variants for which we discov-
ered an association with AEs are all potentially involved in path-
ways that are in line with our previously hypothesized mecha-
nism of AEs involving excess stimulation of inflammatory
pathways and the imbalance of tissue damage repair pathways.
This model was developed from studies of circulating cytokines
and relevant immunological effector cells [3-5]. For subjects ex-
periencing AEs, vaccination appears to trigger an acute inflam-
matory response that is excessive. Antigen presentation to T cells
in the dermis leads to the release of T cell cytokines that trigger a
cascade of cytokines and chemokines whose release enhances the
inflammatory response by (1) promoting the migration of
monocytes into the lesion and their maturation into macro-
phagesand (2) further attracting T cells [38, 39]. Taken together,
these findings suggest that systemic AEs occurring after small-
pox vaccination maybe consistent with low-grade macrophage
activation syndrome caused by virus replication and vigorous
tissue injury and repair.
There are limitations to the present study. The numbers of
subjects are too small for a genetic association study of low-
penetrance, high-frequency alleles. The association between
IL4 variations and AEs was weaker than that between varia-
tions in other genes and AEs. Nevertheless, the observation of
the same variants in 2 independent clinical trials, the high
biological plausibility of these associations in light of what is
known about the biological profile of poxvirus, and the po-
tential public health significance suggest that the findings are
of interest.
Conclusions and future directions. These data provide
the rare opportunity to (1) study 2 independent cohorts of
smallpox vaccine recipients and (2) attempt to identify asso-
ciations between common genetic variations and the occur-
rence of AEs after vaccination. Statistical analysis of the re-
sults of the first study revealed potentially significant
associations between SNPs in biologically interesting candi-
date genes. Of the AE-associated genes identified in the first
study, 2 were replicated in an independent study, with one
additional candidate gene having results just beyond the cut-
off value used to denote statistical significance but neverthe-
less having a high level of biological plausibility. It is possible
that our findings could be due to chance, but we avoided
multiple testing issues by testing only the most promising
results in the validation sample. Although all SNPs were
tested in the first study, only those SNPs that were signifi-
cantly associated with AEs were tested in the second study,
and our empirically derived probability of replication by
chance alone was <0.1%. The association of SNPs in 3 genes
across both studies and the biologically plausibility that these
SNPs were associated with the development of AEs lend cre-
dence to the reproducibility of these associations.
As with any statistical association, follow-up studies in ad-
ditional populations are needed to identify the particular ge-
netic susceptibility variants and examine the functional con-
sequences of polymorphisms in the AE-associated genes. The
polymorphisms that were identified show consistently high
heterozygosity across Hispanic, African, African-American,
Asian, and white population samples [40]. Therefore, al-
though future population samples may reveal population-
specific differences in allele frequencies that require analytical
consideration, the variability in these SNPs makes them rea-
sonable candidates for association studies in more racially
diverse populations. Because we found multiple AE-
associated SNPs in regions of IRt'l and IL4, focused studies
should be undertaken to characterize the genetic variability in
these candidate regions. Indeed, haplotypes in IRF and II.4
displayed altered susceptibility to a specific systemic AE (i.e.,
fever) after smallpox vaccination [41]. Although the associa-
tion between AEs and a nonsynonymous polymorphism in
the gene for MTHFR points toward functional significance of
this SNP, fine mapping of this locus should determine
whether this is the case. For all 3 candidate genes, both
follow-up replication and functional studies are needed to
establish the plausibility of the association of common ge-
netic polymorphisms with the hypothesized etiological path-
ways.
Acknowledgments
We thank Jennifer Hicks, Karen Adkins (Vanderbilt Pediatric Clinical
Research Office, Vanderbilt University Medical Center, Nashville, Ten-
Genetic Basis for Adverse Events • JID 2008:198 (1 July) • 21
Previous
TOG
Next
-------
nessee), and the staff at the Vaoderbilt General Clinical Research Center
(Vanderbilt University Medical Center), for nursing support.
References
1. Keraper AR, Davis MM, Freed GL. Expected adverse events in a mass
smallpox vaccination campaign. EffClin Pract 2002; 5:84-90.
2. Seet BT, Johnston JB, Brunetti CR, et al. Poxviruses and immune eva-
sion. Annu Rev Immunol 2003; 21:377 423.
3. McKinncy BA, Reif DM, Rock MT, et al. Cytokinc expression patterns
associated with systemic adverse events following smallpox immuniza-
tion. J Infect Dis 2006; 194:444-53.
4. Rock MT, Yoder SM, Talbol TR, Edwards KM, Crowe JE Jr. Adverse
events after smallpox immunizations are associated with alterations in
systemic cytokine levels. J Infect Dis 2004; 189:1401-10.
5. Rock MT, Yoder SM, Wright PF, Talbot TR, Edwards KM, Crowe JE Jr.
Differential regulation of granzyme and perform in effector and mem-
ory T cells following smallpox immunization. J Immunol 2005; 174:
3757-64.
6. Rock MT, Yoder SM, Talbot TR, Edwards KM, Crowe JE Jr. Cellular
immune responses to diluted and undiluted Aventis Pasteur smallpox
vaccine. J Infect Dis 2006; 194:435-43.
7. Shaklee JF, Talbot TR, Muldowney JA III, et al. Smallpox vaccination
does not: elevate systemic levels of prothrombotic proteins associated
with ischernic cardiac events. J Infect Dis 2005; 191:724-30.
8. Talbot TR, Stapleton JT, Brady RC, et al. Vaccination success rate and
reaction profile with diluted and undiluted smallpox vaccine: a random-
ized controlled trial. JAMA 2004; 292:1205 12.
9. Hirschhorn JN, Lohmueller K, Byrne E, Hirschhorn K. A comprehensive
review of genetic association studies. Genet Med 2002; 4:45-61.
10. Lohmueller KE, Pearce CL, Pike M, Lander ES, Hirschhorn JN. Meta-
analysis of genetic association studies supports a contribution of com-
mon variants to susceptibility to common disease. Nat Genet 2003; 33:
177-82.
11. Talbot TR, Bredenberg I IK, Smith M, LaFleur BJ, Boyd A, Edwards KM.
Focal and generalized folliculitis following smallpox vaccination among
vaccinia-naive recipients. JAMA 2003; 289:3290-4.
12. Garcia-Closas M, Malats N, Real FX, et al. Large-scale evaluation of
candidate genes identifies associations between VEGF polymorphisms
and bladder cancer risk. PLoS Genet 2007; 3:e29.
13. Packer BR, Yeager M, Burdett L, et al. SNPSOOCancer a public resource for
sequence validation, assay development, and frequency analysis for genetic
variation in candidate genes. Nucleic Acids Res 2006; 34:L*617—21.
14. Wigginton JE, Cutler DJ, Abecasis GR. A note on exact tests of Hardy-
Weinberg equilibrium. Am J Hum Genet 2005; 76:887-93.
15. Barrett JC, Fry B, Mailer J, Daly MJ. Haploview: analysis and visualiza-
tion of LD and haplotype maps. Bioinformatics 2005; 21:263 5.
16. Lake SL, Lyon H, Tantisira K, et al. Estimation and tests of haplotype-
environment interaction when linkage phase is ambiguous. Hum Hered
2003; 55:56-65.
17. Ihaka R, Gentleman R. R: a language for data analysis and graphics.
Journal of Computational and Graphical Statistics 1996; 5:299 314.
18. R Development Core Team. R: a language and environment for statisti-
cal computing. R Foundation for Statistical Computing. Available at:
http://www.R-project.org. Accessed 31 August 2007.
19. Howell MD, Gallo RL, Boguniewicz M, et al. Cytokine milieu of atopic
dermatitis skin subverts the innate immune response to vaccinia virus.
Immunity 2006; 24:341-8.
20. Jackson RJ, Ramsay AJ, Christensen CD, Beaton S, Hall DF, Ramshaw
I A. Expression of mouse interleukin-4 by a recombinant ectromelia vi-
rus suppresses cylolytic lymphocyte responses and overcomes genetic
resistance to mousepox. J Virol 2001; 75:1205—10.
21. KcrrPJ, Perkins HD, InglisB, ctal. Expression of rabbit IL-4 by recom-
binant myxoma viruses enhances virulence and overcomes genetic re-
sistance to myxomatosis. Virology 2004; 324:117-28.
22. Daly MJ, Rioux JD, Schaffner SF, Hudson TJ, Lander ES. Iligh-
resolulion haplotype structure in the human genome. Nat Genet 2001;
29:229—32.
23. Martin YN, Salavaggione OE, Eckloff BW, Wieben ED, Schaid DJ,
Weinshilboum RM. Human methylenetetra.hydrofola.tereducta.se phar-
macogenomics: gene resequencing and functional genomics. Pharma-
cogenet Genomics 2006; 16:265 77.
24. Friso S, Girelli D, Trabetti E, et al. The MTHFR U9SA>C polymor-
phism and genomic DNA methylation in human lymphocytes. Cancer
Epidemiol Biomarkcrs Prcv 2005; 14:938- 43.
25. Dedoussis GV, Panagiotakos DB, Pitsavos C, et al. An association be-
tween the methylenetetrahydrofolate reductase (MTHFR) C677T mu
tation and inflammation markers related to cardiovascular disease. Int
J Cardiol 2005; 100:409-14.
26. Lim U, Peng K, Shane B, et al. Polymorphisms in cytoplasmic serine
hydroxymethyltransferase and methylenetetrahydrofolate reductase
affect the risk of cardiovascular disease in men. J Nutr 2005; 135:
1989-94.
27. Murphy N, Diviney M, Szer J, et al. Donor methylenetetrahydrofolale
reductase genotype is associated with graft-vcrsus-host disease in herna-
topoietic stem cell transplant patients treated with methotrexate. Bone
Marrow Transplant 2006; 37:773—9.
28. Urano W, Taniguchi A, Yamanaka II, et al. Polymorphisms in the meth-
ylenelelrahydrofolate reduclase gene were associated with both the effi-
cacy and the toxicity of methotrexate used for the treatment of rheuma-
toid arthritis, as evidenced by single locus and haplotype analyses.
Pharmacogenetics 2002; 12:183-90.
29. Harada II, Fujita T, Miyamoto M, et al. Structurally similar but func-
tionally distinct factors, IRF 1 and IRF 2, bind to the same regulatory
elements of IFN and IFN-inducible genes. Cell 1989; 58:729-39.
30. Goodbourn S, Didcock L, Randall RE. Interferons: cell signalling, im-
mune modulation, antiviral response and virus countermeasures. J Gen
Virol 2000; 81:2341-64.
31. van den Broek MF, Mtiller U, Huang S, Aguet M, Zinkernagel RM.
Antiviral defense in mice lacking both alpha/beta and gamma interferon
receptors. J Virol 1995; 69:4792 6.
32. Symons JA, Alcami A, Smith GL. Vaccinia virus encodes a soluble type I
interferon receptor of novel structure and broad species specificity. Cell
1995; 81:551 60.
33. Alcami A, Syrnons JA, Smith GL. The vaccinia virus soluble alpha/beta
interferon (IFN) receptor binds to the cell surface and protects cells from
the antiviral effects of IFN. J Virol 2000; 74:11230-9.
Crawford DC, Nickerson DA. Definition and clinical importance of
haplotypes. Annu Rev Med 2005; 56:303 20.
Duan J, Wainwright MS, Comeron JM, et al. Synonymous mutations in
the human dopamine receptor D2 (DRD2) affect rriRNA stability and
synthesis of the receptor. Hum Mol Genet 2003; 12:205 16.
36. Jackson RJ, Ramsay AJ, Christensen CD, Beaton S, Hall DF, Ramshaw
LA. Expression of mouse interleukin 4 by a recombinant ectromelia vi
rus suppresses cytolytic lymphocyte responses and overcomes genetic
resistance to mousepox. J Virol 2001; 75:1205-10.
37. Howell MD, Gallo RL, Boguniewicz M, et al. Cytokinc milieu of atopic
dermatitis skin subverts the innate immune response to vaccinia virus.
Immunity 2006; 24:341-8.
38. Fong TA, Mosmann TR. The role of IFN-gamma in delayed-type
hypersensilivily mediated by Till clones. J Immunol 1989; 143:
2887-93.
39. Grom AA, Passo M. Macrophage activation syndrome in systemic juve-
nile rheumatoid arthritis. J Pediarr 1996; 129:630-2.
40. Sherry ST, Ward Mil, Kholodov M, et al. dbSNP: the NCBI database of
genetic variation. Nucleic Acids Res 2001; 29:308-11.
41. Stanley SL Jr, Frey SL, Taillon-Miller P, et al. Immunogcnctics of small-
pox vaccination. J Infect Dis 2007; 196:212-9.
22 • JID 2008:198 (1 July) • Reifetal.
Previous
TOG
Next
-------
Hepatol Int (2008) 2:39^9
DOI 10.1007M2072-007-9025-2
Genome-wide transcriptome expression in the liver of a mouse
model of high carbohydrate diet-induced liver steatosis and its
significance for the disease
Ion V. Deaciuc • Zhenyuan Song • Xuejun Peng • Shirish S. Barve •
Ming Song • Qiang He • Thomas B. Knudsen • Amar V. Singh •
Craig J. McClain
Received: 11 April 2007/Accepted: 8 August 2007/Published online: 27 November 2007
© Asian Pacific Association for the Study of the Liver 2007
Abstract
Purpose To perform a large-scale gene profiling of the
liver in a mouse model of fatty liver induced by high
carbohydrate (sucrose) diet (HCD) to gain a deeper insight
into potential mechanisms of diet-induced hepatic steatosis.
Methods C57BL/6 male mice were fed either a purified,
control diet or a HCD for 16 weeks. HCD feeding led to
marked liver steatosis without inflammation or necrosis.
The expression of 42,500 genes/sequences was assessed.
Results A number of genes (471) underwent significant
expression changes in HCD- as compared to standard diet-
fed mice (n — 5/group; P < 0.01). Of these genes, 211 were
down- and 260 up-regulated. The latter group includes 20
genes encoding enzymes involved in carbohydrate conver-
sion to fat. The genes that underwent expression changes
perform a large variety of molecular functions, and the vast
majority of these have never been tested before in non-
alcoholic fatty liver of nutritional origin. They reveal novel
aspects of the disease and allow identification of candidate
genes that may underlie the initiation of hepatic steatosis and
progression to non-alcoholic steatohepatitis.
Conclusions HCD-fed laboratory animals provide a
model of early non-alcoholic fatty liver disease resembling
the disease in humans. The genome wide gene profiling of
Electronic supplementary material Supplementary material is
available for this article at http://dx.doi.org/10.1007/sl2072-007-
9025-2 and is accessible for authorized users.
Ion V. Deaciuc and Zhenyuan Song are contributed equally to this
study.
I. V. Deaciuc (El) • Z. Song • S. S. Barve • M. Song •
C. J. McClain
Division of Gastroenterology/Hepatology, Department
of Medicine, University of Louisville School of Medicine,
550 S. Jackson Street, ACB Bldg., Third Floor, Louisville,
KY 40202, USA
e-mail: ion.deaciuc@louisville.edu
Z. Song
e-mail: zOsongOz@louisville.edu
S. S. Barve
e-mail: shirish.barve@louisville.edu
M. Song
e-mail: mOsong03@louisville.edu
C. J. McClain
e-mail: craig.mcclain@louisville.edu
X. Peng
Biometrics and Data Management Department, Takeda Global
Research and Development Center, Inc., Lincolnshire, IL 60069,
USA
e-mail: xpeng@tgrd.com
S. S. Barve • C. J. McClain
Department of Pharmacology and Toxicology, University of
Louisville School of Medicine, Louisville, KY 40202, USA
Q. He
Department of Biochemistry and Molecular Biology, University
of Louisville School of Medicine, Louisville, KY 40202, USA
e-mail: qOhe0002@louisville.edu
T. B. Knudsen • A. V. Singh
Department of Molecular, Cellular and Craniofacial Biology,
University of Louisville, Louisville, KY 40202, USA
T. B. Knudsen
e-mail: thomas.knudsen@louisville.edu
A. V. Singh
e-mail: avsing01@louisville.edu
C. J. McClain
Louisville Veterans Administration Medical Center, Louisville,
KY 40202, USA
4y Springer
Previous
-------
40
Hepatol Int (2008) 2:39^9
the liver reveals the complexity of the disease, unravels
novel aspects of HCD-induced hepatic steatosis, and helps
elucidate its nature and mechanisms.
Keywords Fatty liver • Gene profiling •
High carbohydrate diet • Mouse
Abbreviations
ALT Alanine-2-oxoglutarate amino transferase (EC
2.6.1.2)
HCD High-carbohydrate diet
NAFLD Nonalcoholic fatty liver disease
NASH Nonalcoholic steatohepatitis
TEARS Thiobarbituric acid reactive substances
Introduction
Nonalcoholic fatty liver disease (NAFLD) is a spectrum of
pathology ranging from simple steatosis to nonalcoholic
steatohepatitis (NASH), and in some instances progressing
to cirrhosis and even hepatocellular carcinoma [1]. NAFLD
is by far the most frequent cause of abnormal liver enzymes
in the United States. Therefore, there is great interest in the
potential pathogenesis, prevention, and/or treatment of this
disease. Multiple factors have been considered and iden-
tified as causes of hepatic steatosis, and they can be
classified in two major groups: exogenous and endogenous.
Among exogenous factors, hepatotoxic drugs, hepatitis C
infection, malnutrition, and, perhaps the most frequently
encountered factor, composition and amount of food, have
been confirmed as causes of nonalcoholic steatosis.
Because of its high prevalence, the disease has become a
focus of intensive research and, although some 1,000
studies dealing with the disease have been published in the
last 25 years, the mechanisms underlying its occurrence
and progression to NASH are not fully elucidated.
Experimental models based on voluntary food intake
have been developed to induce hepatic steatosis in labo-
ratory animals [2-8]. The nutritional models using
complete diets resemble the human condition in that they
contain amounts of lipids or carbohydrates that exceed the
energy needs of the body. As a consequence, the body
processes and deposits the excessive nutrients as fat
regardless of their original chemical nature. No matter the
source, the body deposits the fat preferentially in subcu-
taneous and visceral areas and, to a lesser extent, in the
liver. The biochemical pathways involved in fat processing
and deposition differ as a function of the origin of fat. The
deposition of the excessive dietary fat involves several
biochemical events, including digestion, reconstitution in the
intestinal epithelium, assembling, transport, and deposition
[9]. However, excessive amounts of ingested carbohydrates,
mainly hexoses, can be stored only as glycogen and, in sig-
nificant amounts, only in skeletal muscle and the liver. The
glycogen content of these organs at the saturating levels is
around 5-6% of their mass. Therefore, after such levels are
attained, the excessive carbohydrates cannot be stored further
as glycogen; instead the body converts them into fat. The
metabolic pathways accomplishing carbohydrate conversion
into fat are well known and they were well characterized in
earlier studies [10, 11].
During the last 15 years it has become increasingly
apparent that macronutrients such as long-chain fatty acids
and glucose act as signaling molecules leading to changes
in gene expression. Therefore, gene profiling of organs as
affected by macronutrients may provide important infor-
mation on the mechanisms underlying disturbances such as
liver steatosis, overweight, obesity, insulin resistance, and
others. To our knowledge, no comprehensive, genome-
wide gene profiling of hepatic steatosis induced by a high-
carbohydrate diet (HCD), without the complications of
steatohepatitis, has been reported in either animals or
humans. Therefore, this study was undertaken to (i) gain a
deeper insight into the biochemical and cell physiological
mechanisms associated with HCD-induced liver steatosis
and (ii) identify potential "hidden" genes/pathways that
may contribute to the progression of liver steatosis to
NASH.
The model of HCD-induced liver steatosis used in this
study consists of long-term (16 weeks) feeding of an HCD
to mice, and resembles the disease in humans. Human
clinical investigations have demonstrated that a diet low in
fat and rich in carbohydrates (closely resembling the HCD
used in our mouse study), even when administered for short
periods of time, for example, 5 or 25 days, can lead to
occurrence of uncomplicated fatty liver [12-15]. Thus, this
mouse model is a highly relevant means of investigating
mechanisms of hepatic steatosis.
Comprehensive gene profiling of the liver was per-
formed using the microarray DNA technology, which
allows simultaneous assessment of the expression of
42,500 genes/sequences. A number of genes that under-
went significant changes were classified according to their
function and selected genes were analyzed from the
viewpoint of their potential participation in various cellular
processes related to nonalcoholic hepatic steatosis.
Materials and methods
Animals and their treatment
The animals were treated in accordance with the Guide for
the Care and Use of Laboratory Animals (National
4y Springer
Previous
-------
Hepatol Int (2008) 2:39^9
41
Research Council, USA, 1996) as approved by the Insti-
tutional Animal Care and Use Committee of the University
of Louisville (Louisville, KY). Male C57BL/6 mice (Har-
lan, Indianapolis, IN), weighing 23.5 ± 0.8 g were
maintained under standard conditions for 7 days before
initiation of study diets. Thereafter, the mice were divided
in two groups of 10 individuals each and started on two
different diets: high-carbohydrate diet (HCD) and a puri-
fied, control diet named herein standard diet (SD; both
from Harlan Teklad, Madison, WI). The composition of the
HCD was identical to that used by Feldstein et al. [8], that
is (in g kg"1), 650 sucrose, 200 casein, 50 corn oil, 40
mineral mixture (AIN-93G-MX), 10 vitamin mixture
(AIN-93-VX), 2.5 choline bitartrate, 3.0 DL-methionine,
and 10 cellulose.
Animal killing and tissue sampling
After 16 weeks of feeding HCD or SD, the mice were
fasted from 2,200 to 8,000 h, anesthetized with urethane
(100 mg kg"1 body weight, intraperitoneally) and the
abdominal cavity opened. Blood (0.5-0.7 mL) was drawn
from the inferior vena cava with citrate-containing syrin-
ges, immediately centrifuged, and the plasma was
collected. The liver was perfused through the portal vein
with 3-5 mL of ice-chilled phosphate-buffered saline
(pH 7.4), with the inferior vena cava severed to remove the
blood. The left one-third of the left lobe was immersed in
formalin while the rest was placed immediately in liquid
nitrogen.
Liver histology
Liver sections were stained with hematoxylin-eosin and
analyzed for the presence of fat, polymorphonuclear infil-
tration, and necrotic areas.
Biochemical assays
The following assays were performed using commercial
kits; in plasma: glucose, triacylglycerols (TAG), free fatty
acids, alanine-2-oxoglutarate aminotransferase (ALT)
(Infinity, Thermo Electron Corp., Melbourne, Australia),
adiponectin and TNF-a (R&D Systems, Minneapolis, MN),
and insulin (Crystal Chem. Inc., Downers Grove, IL); in
the liver: TEARS according to Quintanilha et al. [16] and
TAG as above. Total RNA was extracted from the liver
using a kit (Ambion, Austin, TX, Cat. No. 1924) and
purified with Qiagen minicolumns (Qiagen, Valencia, CA;
Cat. No. 74104). RNA quality was assessed using Agilent
2100 bioanalyzer and reagents supplied by the manufac-
turer (Agilent Technologies, Inc., Palo Alto, CA). About
10 (ig of total RNA were processed for mRNA expression
using the Affymetrix GeneChip MGU 430 2.0 Array
(Affymetrix, Santa Clara, CA) and Affymetrix technology.
This chip allows testing 42,500 transcripts for their
expression.
Quantitative real-time PCR
The amount of mRNA for 10 randomly selected genes was
measured by quantitative real-time (RT) PCR for five mice
in each group. The Taqman Gold RT-PCR kit (Applied
Biosystems, Inc., Foster City, CA) was used for all of the
reactions and the manufacturer's protocol for GAPDH
control was followed. For each mouse, 2ug of total RNA
and random hexamer primers were used in the initial
reverse transcription reaction. Each gene was detected by a
revalidated Taqman Gene Expression Assay probe set that
was labeled with 6FAM and the amplification step was
done in triplicate for each gene in two variants: (i) template
and reverse transcriptase present, (ii) no template present,
and (iii) no reverse transcriptase present. The PCR ampli-
fication was analyzed by an iCycler iQ Real-Time
Detection System (BioRad Laboratories, Inc., Hercules,
CA) and the resulting expression ratios were calculated by
the 2
-AACt
method as described in the Technical Bulletin of
Gene Expression (Applied Biosystems, 2002).
Assay of protein abundance
The total liver protein extraction, gel electrophoresis,
immunoblotting and band visualization were performed
using reagents and technology provided by Santa Cruz
Biotechnology, Inc. (Santa Cruz, CA). The following
antibodies were used: GCK (H-88)—sc7908 (Santa Cruz),
EGFR (Cell Signaling Technology, Inc., Cat. No. 2232,
Danvers, MA), and cytochrome P450 reductase (Abeam,
Cat. No. ab!3513, Cambridge, MA).
Gene data processing and statistics
Gene data were analyzed with the Affymetrix Microarray
Suite 5.0 algorithm to generate signal value and detection
label. Only genes that generated 5 present calls in each
group (n — 5, in each group, in a one chip—one animal
design) were taken into consideration for further statistics
and classification. Also, a false discovery rate was set at
10% and calculated for all probe sets. The genes whose
expression was changed 1.5-fold or greater in either
4y Springer
Previous
-------
42
Hepatol Int (2008) 2:39^9
Fig. 1 Liver sections of
standard diet- (a) and high
carbohydrate diet- (b) fed mice.
Note the normal appearance of
the liver in A and fat
accumulation in B. Note that fat
accumulation in the hepatocyte
has the shape of macrovacuole
(arrows). Magnification: 200 x
a b
direction (up or down), in the HCD-fed mice as compared
to the SD-fed mice, were given priority in ascribing a
potential significance. Finally, the comparison between SD
and HCD-fed mice was made on the basis of a P-value of
0.01. A detailed presentation of the gene statistics proce-
dure used in this study was given in an earlier publication
from our laboratory [17].
Results
Body weight
At the end of the feeding period, the body weight of the
HCD-fed mice was 29% greater than that of the SD-fed
mice (P < 0.05). SD-fed mice gained 28% body weight
over the initial time point while the HCD-fed mice gained
65% (P < 0.05; Table 1).
Blood biochemistry
The data in Table 1 show that, at the time of killing, the
HCD-fed animals had increased levels of glucose, choles-
terol, and insulin in plasma. Free fatty acids in plasma were
significantly lower than in control, SD-fed mice. No sig-
nificant changes were observed in plasma adiponectin,
TNF-a, TAG, and ALT levels.
Liver histology
The histological appearance of the livers was normal in
SD-fed mice while in HCD-fed mice the liver displayed fat
infiltration with no inflammation or necrosis. The fat
infiltration score was estimated to be 60% (3+, according to
the method of Jarvelainen et al. [18] (Fig. la and b).
Liver biochemistry
TAG content in the liver of HCD-fed mice was signifi-
cantly higher (P < 0.001) than in the livers of SD-fed
mice. TEARS were also increased (P < 0.01; Table 1).
All these changes demonstrate that the HCD used in
these experiments induced typical hepatic steatosis and, for
the feeding period employed, the liver did not display
markers of hepatitis (inflammation, necrosis, and others).
Some of these data resemble, in part, the results reported by
Feldstein et al. using HCD feeding [8]. Whether feeding
Table 1 Body weight and
biochemical parameters of the
plasma and liver in SD- and
HCD-fed mice at the end of the
feeding period
* Means ± SEM were
calculated for 5 animals in each
group
** The initial body weight was
23.5 ± 0.8 g (n = 10)
Parameter/marker
Body weight (g)
Glucose (mM)
ALT (mU mL"1)
Free fatty acids (plasma, mEq dL~!)
Cholesterol (plasma, mg dL~!)
Insulin (pg mL~ )
Triacylglycerols (liver, mg g~!)
Triacylglycerols (plasma, mg dL~ )
TEARS (nmol g"1 wet weight)
TNF-a (pg mL"1)
Adiponectin (ng mL^1)
Mean ± SEM*
Standard diet
30.1 ± 0.3**
9.3 ± 0.25
28.3 ± 3.0
0.50 ± 0.02
36.4 ± 0.86
582 ± 54
60.2 ± 6.7
101.0 ± 9.8
90.4 ± 14.5
40.3 ± 13.4
23.7 ± 0.48
HCD
38.8 ± 1.2**
14.0 ± 0.75
31.2 ± 2.9
0.33 ± 0.021
67.4 ± 2.6
2,037 ± 360
159.4 ± 14.6
92.9 ± 8.3
170.9 ± 16.8
37.6 ± 18.7
23.2 ± 2.0
P
<0.05
<0.001
NS
<0.001
<0.001
<0.001
<0.001
NS
<0.010
NS
NS
4y Springer
Previous
-------
Hepatol Int (2008) 2:39^9
43
this diet for longer periods of time may lead to NASH
remains to be established.
Liver genomics
A numerical account of the genomics data obtained in our
experiments is given in the self-explanatory diagram of
Fig. 2. Selected genes that underwent a change in expres-
sion of 1.5-fold or more in either direction are presented in
Tables 2 and 3. The genes in Table 2 were classified
according to Bulera et al. [19], with slight modifications
[20]. Also a group of genes was selected and tabulated,
comprising several glutathione S-transferases, because of
their potential participation in the progression of liver
steatosis to NASH (Table 4). Two Tables I and II are
presented in Microsoft Excel as supplementary material.
Table I presents all gene changed by HCD feeding while
Table I contains an expanded listing of genes involved in
carbohydrate and fat metabolism processed using the
DAVID Functional Annotation Chart for gene ontology
(GO) and Kegg.
The QRT-PCR data (Table 5) confirmed changes in
gene expression (for 10 genes in each experimental group)
obtained with microarray technology. Only minor quanti-
tative differences were observed between the two methods.
Such differences are routinely observed in many studies.
Protein abundance
The gel images in Fig. 3 show that three proteins randomly
selected to be tested for their abundance—glucokinase
42.449 unique Gene bar* IDs
on me chip
260 genes UP-tegulated
In HCD
211 genes DOWN-
regulaied
in HCD
Fig. 2 Flow chart of numeric distribution of genes and sequences
detected in the liver. We propose that the number of genes in the dark
ovals should be taken into consideration for future analysis of the
genomics in the experimental model used in this study
(BC011139.1), cytochrome-P450 oxidoreductase
(NM_008898.1), and Emr4 (AY032690.1; EGF-like mod-
ule containing mucin-like, hormone receptor-like sequence
4)—changed in the same direction as their transcriptome. A
fourth protein, MCP-1 (AF128196.1), tested using both the
Western blot method and ELISA, could not be detected.
This cytokine, however, was not assayed in plasma.
Interpretation of genomic data is mainly based on changes
in transcriptome expression rather than in protein abun-
dance. It is generally assumed that changes in protein
abundance parallels changes in gene expression. However,
if a gene or a set of genes are studied closely, the assess-
ment of their protein products must be performed before
functional significance is ascribed to expression changes.
Discussion
In this study, 471 genes (Fig. 2) were identified whose
expression was changed in the livers of HCD-fed animals.
These genes include the ones that encode the enzymes,
approximately 20, involved in glucose and fructose
metabolism and their conversion to fat [10, 12, 15]
(Table 3 and Table I of supplementary material, and
Fig. 4). The large number of genes that underwent changes
in expression, taken together with the direction of change,
and with the functional diversity they belong to, demon-
strate that fat accumulation in the hepatocyte in HCD-
induced liver steatosis is associated with alterations of a
much wider spectrum of biochemical and molecular pro-
cesses than expected on the basis of the data available thus
far. Such a conclusion could have emerged only from the
large-scale gene profiling data and supports the usefulness
of this tool in unraveling multifaceted mechanisms of
disease.
Importantly, a large number of genes involved in car-
bohydrate conversion to fat were upregulated by the HCD.
A group of genes that are of particular interest for under-
standing potential mechanisms of HCD-induced hepatic
steatosis are presented in Table 3. The data in this table
demonstrate that, as expected, HCD upregulates several
genes directly involved in carbohydrate conversion to fat
[11, 21, 22]. These genes have been displayed within the
context of the metabolic pathways to which they belong
(Fig. 4) in an attempt to facilitate understanding of their
role(s) in HCD-induced liver steatosis. In addition, one
gene, sterol regulatory element-binding protein (SREBP)-
Ic, which controls the transcription of genes involved in
fatty acid synthesis [23], and whose transcription is regu-
lated by insulin [24], was upregulated. Another gene,
peroxisome proliferator-activated receptor (PPAR)-a, also
a transcription factor, but encoding enzymes involved in
fatty acid oxidation [25], was likewise upregulated.
4y Springer
Previous
-------
44
Hepatol Int (2008) 2:39^9
Table 2 Selected genes that underwent 2-fold or higher change in their expression in either direction (up or down) in the liver of HCD-fed mice
as compared to the liver of SD-fed mice
Functional group and gene name
Apoptosis
Phosphatidylserine receptor
Cell motility
Eps8 (Epidermal growth factor receptor pathway substrate 8
or EGF receptor kinase substrate 8)
Cell proliferation
Cold inducible RNA binding protein
Channels/Transporters
Amiloride-sensitive cation channel 5, intestinal
Lipocalin 13 (precursor) (retinoid carrier protein)
Complex lipid metabolism
Galactocerebrosidase (galactosylceramidase; precursor) (EC 3.2.1.46).
Cytochromes P450
Cytochrome P450, family 17, subfamily a, polypeptide 1
(catalyzes 17-a hydroxylase and 17,20-lyase activities) (EC 1.14.99.9)
P450 (cytochrome) oxidoreductase (EC 1.6.2.4)
Cytokines/Cytokine receptors
Emr4 (EGF-like module containing, mucin-like, hormone receptor like sequence 4)
Chemokine (C-C) motif ligand 9 (CCL-9; small inducible cytokine A9;
macrophage inflammatory protein 1-gamma
Chemokine (C-C) motif ligand 2
Macrophage inflammatory protein-related protein-2 (MRP-2)
Glutathione metabolism
Glutathione 5-transferase, n 3 (EC 2.5.1.18)
Glutathione 5-transferase, alpha 2 (Yc2) (EC 2.5.1.18)
Nucleic acid metabolism
Deoxyribonuclease II alpha (EC 3.1.22.1). A role in DNA degradation in apoptosis.
Nucleotide metabolism
Cytidine deaminase (EC 3.5.4.5)
Protein metabolism
Ubiquitin specific protease 18
Serine (or cysteine) proteinase inhibitor, clade B, member la
Secretory products
Intestinal trefoil factor (TFF3 precursor)
Signaling/signal transduction
Membrane anchored glycoprotein RECK
(inhibitor of tumour invasion, regulator of MMP-9)
Heat shock protein 1
Guanine nucleotide binding proptein, alpha 14
Macrophage expressed gene 1 (shares a distant ancestry to perform)
Calcium/calmodulin-dependent kinase II gamma (Camk2g)
LIM homebox protein 2 (Lhx2)
Methyl-CpG binding domain protein 1
Retinoic acid early transcript 1, alpha (Rae-1 alpha)
(early mammalian embryogenesis)
STAT-induced STAT inhibitor-2 mRNA
DNAJ (Hsp40) homolog, subfamily B, member l(Heat shock 40 kDa protein 1)
Gene bank Fold change
accession
code UP Down
AK017622.1 2.5
NM_007945 2.0
NM_007705.1 2.5
NM_02 1370.1 2.4
BC027556.1 2.4
NM_008079.1 3.1
NM_007809.1 2.6
NM_008898.1 26
AY032690.1 2.9
AF128196.1 2.2
AF065933.1 10.2
NM_011338 2.2
JO3953.1 3.7
NM_008182 8.3
NM_010062.1 3.4
AK008793.1 2.2
NM_01 1909.1 2.5
AB030426 3.4
NM_01 1575.1 4.5
NM_016678.1 4.5
NM_01 3560.1 2.1
NM_008137.1 2.3
L20315.1 2.5
BC025597.1 2.1
NMJH0710.1 2.1
AK007371.1 4.5
NM_009016. 1 3.5
BB244736 2.3
NM_018808.1 3.7
Expression*
489 ± 40
67 ± 12
328 ± 19
182 ±6
256 ± 24
71 ± 6
326 ± 23
4,896 ± 306
85 ±20
2,034 ± 130
218 ± 26
4,897 ± 231
5,970 ± 490
9,268 ± 966
8,015 ± 529
475 ± 41
332 ± 27
520 ± 122
256 ± 36
187 ± 23
5,538 ± 995
187 ± 27
1,453 ± 79
133 ± 6
63 ±5
385 ± 37
2,613 ± 460
1545 ± 87
8175 ± 314
4y Springer
Previous
-------
Hepatol Int (2008) 2:39^9
45
Table 2 continued
Functional group and gene name
SH3-binding kinase 1 (Sbk)
Inhibin E (INHBE)
Xenobiotic metabolism
Sulfotransferase 1C1 (cytosolic)
Gene bank
accession
code
BC025837.1
NM_007945.1
NM_026935.1
Fold change
Up Down
2.3
2.7
2.1
Expression*
137 ± 9
67 ± 11
214 ± 19
* Expressed in intensity reading units (Mean ± SEM, for five animals in each group), for the SD-fed mice. To find the absolute value for the high
carbohydrate-diet fed mice, the value given in this column will be multiplied by the value in the column Up or divided by the value in the column
Down. The difference between the two groups with regard to gene expression is significant at P < 0.01. A complete Table I) comprising all 451
genes that underwent changes of 1.5-fold or more is available as supplemental material to this article
Table 3 Selected genes that likely have direct relevance to the biochemical mechanisms underlying HCD-induced fatty liver in the mouse
Functional group and gene name
Gene bank
accession code
Fold change
Up
Expression*
Carbohydrate metabolism
Solute carrier family 2 (facilitated glucose transporter, member 5) NM_019741.1 2.7 165 ± 8
Glucokinase (Hexokinase D, EC 2.7.1.1) BC011139.1 2.4 1,614 ± 202
Phosphoglucomutase 3 (EC 5.4.2.6) AK013402.1 1.6 492 ± 53
UDP-glucose pyrophosphorylase 2 (EC 2.7.7.9) AF424698.1 1.6 10,426 ± 455
Ketohexokinase (EC 2.7.1.3) BC013464.1 1.7 11,278 ± 326
Malic enzyme, supernatant (cytosolic) (EC 1.1.1.40) NM_008615.1 1.8 4,292 ± 440
Glucose phosphate isomerase (EC 5.3.1.9) NM_008155.1 1.8 3,742 ± 147
Pyruvate dehydrogenase kinase, isoenzyme 4 (EC 2.7.1.99) NM_013743.1 1.8 472 ± 24
Pyruvate dehydrogenase kinase, isoenzyme 1 (EC 2.7.1.99) BC027196.1 1.7 1,261 ± 61
Glycerolphosphate dehydrogenase 2, mitochondrial (EC 1.1.1.8) NM_010274.1 1.6 2,412 ± 240
Dihydrolipoamide S-acetyltransferase (E2 component of pyruvate BC026680.1 1.6 1,815 ± 126
dehydrogenase complex) (EC 2.3.1.12)
Glucose 6-phosphate dehydrogenase X-linked (EC 1.1.1.49) NM_008062.1 1.8 265 ± 38
Fatty acid and complex lipid metabolism
ATP citrate lyase (EC 4.1.3.8) BI456232 1.9 9,127 ± 1072
Fatty acid desaturase 2 (EC 1.14.99.6) BB430611 1.9 5,419 ± 390
Glycerol-3-phosphate acyltransferase, mitochondrial (EC 2.3.1.15) NM_008149.1 1.9 8,022 ± 395
Monoglyceride lipase (EC 3.1.1.23) NM_011844.2 1.6 2,362 ± 128
Peroxisome proliferator-activated receptor alpha BC016892.1 1.7 6,062 ± 714
Stearoyl-CoA desaturase (EC 1.14.99.5) NM_009127.1 1.9 9,793 ± 1,908
Fatty acid desaturase 2 (Delta-6 desaturase) (EC 1.14.99.5) NMJM9699.1 2.5 6,486 ± 270
Sterol regulatory element binding factor 1 (SREBP-1) AI326423 1.6 5,679 ± 257
Lipocalin 13 (precursor) (retinoid carrier protein) BC027556.1 2.4 256 ± 24
Fatty acid binding protein 5, epidermal BC002008.1 2.5 3,714 ± 448
Adiponutrin (a triacylglycerol lipase and acylglycerol O-acyltransferase) NM_054088.1 5.6 144 ± 8
(EC 3.1.1.3, and EC 2.3.1.-)
ELOVL family member 6 (elongation of very long chain fatty acids; NM_130450.1 2.4 2,787 ± 185
a lipogenic enzyme regulated by SREBPs)
* Expressed in intensity reading units (Mean ± SEM, for five animals in each group), for the SD-fed mice. To find the absolute value for the
high carbohydrate-diet fed mice, the value given in this column should be multiplied by the value in the column Up. The difference between the
two groups with regard to gene expression is significant at P < 0.01. An expanded list of genes involved in carbohydrate and fat metabolism is
given in Table I as supplementary material
4y Springer
Previous
-------
46
Hepatol Int (2008) 2:39^9
Table 4 Down-regulation of glutathione 5-transferases in the liver of
mice fed a high carbohydrate diet
Gene name
Glutathione
5-transferase fj, I
Glutathione
5-transferase 9 3
Glutathione
5-transferase /j, 3
Glutathione
5-transferase, a 2 (Yc2)
Glutathione
5-transferase a 4
Access code
J03952.1
BC003903.1
J03953.1
NM_008 182.1
NM_010357.1
Change (-fold)
in HCD group
Down 1.8
Down 2.0
Down 3.7
Down 8.0
Down 3.4
Of interest, the vast majority of genes identified in this
study (Tables 2 and 5, and Table I of supplementary
material) cannot be directly linked to the biochemical or
molecular processes leading to fatty liver. Changes in
expression of these genes likely reflect alterations in cel-
lular processes caused by, rather than leading to, hepatic
steatosis. Owing to space limitations, these genes will not
be discussed in any detail.
NASH is thought to evolve through a 2-hit process in
which the first hit is steatosis. The second hit or hits include
multiple factors such as oxidative stress, proinflammatory
cytokines, mitochondrial dysfunction, insulin resistance,
and even industrial exposures [1]. An important issue is
whether the data reported herein provide insights into
genes that may predispose the liver to a "second hit," thus
leading to NASH. Since no changes in transcriptome
expression of proinflammatory or profibrotic cytokines
(e.g., TNF-a, IL-1/3, IL-18, TGF-j3, and others), classically
thought to mediate liver injury including fibrosis, were
identified in the HCD-fed mouse liver, it seems that these
classic proinflammatory cytokines, at least those secreted
in the liver, may not be critically involved in the early
aspects of this disease. The lack of TNF-a participation in
dietary-induced NASH was recently suggested by Deng
et al. [7], who demonstrated that knocking out the TNF-a
receptor 1 does not prevent NASH induced by force-
feeding a fat-enriched diet. Similar studies by Dela Pena
et al. [26] showed that TNFR1 knockout mice still develop
hepatic steatosis when fed a methionine-restricted, choline-
deficient diet. Moreover, studies in children with NAFLD
demonstrate normal serum TNF but decreased adiponectin
as early events [27]. Our study does show that the tran-
scriptome of two macrophage inflammatory proteins,
MCP-1 and MCP-2 (Table 2), were upregulated, which
may predict potential facilitation of extrahepatic cell
infiltration into the liver and the initiation of inflammation.
Several genes, other than proinflammatory cytokines, may
be considered as plausible candidates for a potential role in
the progression to NASH. One of these is the macrophage-
expressed gene 1 (L20315.1), a relative of perforin (gran-
zyme B), which was upregulated 2.5-fold (Table 2).
Another gene is methyl-CpG-binding domain protein 1
(AK007371.1, known as MBD1), a member of a family of
five mammalian methyl-CpG-recognizing proteins, which
plays a key role in maintaining a transcriptionally inactive
state of methylated promoters [28, 29]. This gene was
downregulated 4.5-fold. Its downregulation may facilitate
expression of genes that otherwise would be in a state of
restricted transcription.
Another group of genes of interest for potential pro-
gression of the steatotic liver to NASH is represented by
glutathione 5-transferases (Table 5), which were down-
regulated in the steatotic liver. Downregulation of these
enzymes may lead to a decreased capacity of the liver to
detoxify xenobiotics, thereby increasing the susceptibility
Table 5 Comparison of gene expression changes for control and HCD-fed mice as determined by quantitative RT-PCR and cDNA microarray
Gene code
AK003441.1
AW489168
AF065933.1
NM_009998.1
NM_010062.1
NM_018808.1
NM_007945
NM_008182
U72881.1
NM_008898.1
Applied Biosystems
assay identification
Mm00614943_ml
Mm00519268_ml
Mm00441242_ml
Mm00456591_ml
Mm00438463_ml
Mm00444519._ml
Mm00514752_ml
Mm00833353_ml
Mm00803317_ml
Mm00435876_ml
Gene name
Ankyrin repeat and KH Domain containing 1
Bcl-2 binding component 3
Chemokine (C-C motif) ligand 2
Cytochrome P450, family 2, subfamily b, poly-peptide 10
Deoxyribonuclease II alpha
DnaJ (HSP40) Homolog, subfamily b, member 1
Eps8 (Epidermal growth factor receptor pathway substrate 8)
Glutathione 5-transferase a 2
Regulator of G protein signaling 16
P450 Cytochrome oxidoreductase
Change in
Microarray
2.5 T
2.1 T
10.2J
20.0J
3.41
3.71
2.01
8.31
7.3 T
26.0J
expression
Q-RT-PCR
2.6 T
2.8 t
2.71
ND*
2.1|
5.8|
1.81
12.51
7.2 t
ND*
; ND, not detected. The relative expression for Q-RT-PCR was normalized to the 18S rRNA copy level. Arrows indicate the direction of change
4y Springer
Previous
-------
Hepatol Int (2008) 2:39^9
47
GCK
EGFR
Cyp450 reductase
p-actin
SD
HCD
Fig. 3 Gel images illustrating the Western blot assay of protein
abundance. The following genes were tested for their protein
abundance: GCK, glucokinase; EGFR, epidermal growth factor-like
module containing mucine-like, hormone receptor-like receptor
sequence 4 (or Emr4); Cyp450 reducatse, and /f-actin. Other
abbreviations: SD, standard diet; HCD, high carbohydrate diet. The
following values apply for the HCD-to-SD ratio: GCK, 2.00 ±0.18;
Cyp450 reductase, 0.16 ± 0.08; EGFR, 1.23 ± 0.28
of the liver to undergo pathologic changes including
necrosis. Such changes may be triggered by chemical
agents from the environment, and there are well-docu-
mented examples of industrial NASH, such as that caused
by petrochemical exposure [30, 31].
Lastly, the increased circulating levels of both glucose
and insulin recorded in this study suggest the existence of a
certain degree of insulin resistance. These findings raise the
question of whether the steatotic liver induced by HCD
feeding is insulin resistant. The concept of an obligatory
association of the steatotic liver with insulin resistance
has been challenged by experimental and clinical data.
Thus, it has been demonstrated that NASH can occur in the
absence of overt insulin resistance [32-36]. On the basis of
the data presented in this and other studies, we surmise
that, in the model of hepatic steatosis used in this study,
and at the moment of animal killing, the liver is not insulin
resistant. Thus, (i) the insulin response element-binding
protein (IREBP-1) [37, 38], a target of insulin signal
transduction downstream of the PI-3K/protein kinase B
(Akt) pathway, regulates the expression of many enzymes
involved in carbohydrate conversion to fat; this factor can
only be active in the presence of an intact insulin signaling
cascade, (ii) the enzymes involved in carbohydrate con-
version to fat, including glucokinase (EC 2.7.1.1),
ketohexokinase (EC 2.7.1.3), glucose-6-phosphate dehy-
drogenase (EC 1.1.1.49), and others, were upregulated in
the liver of the mouse model used in our study (Table 3);
Extracellular space
The hepatocyte
Fig. 4 Schematic representation of the metabolic pathways involved
in carbohydrate conversion to fat in the liver. Represented are
glycolysis, part of the citric acid cycle, citrate cleavage enzyme, fatty
acid synthase, fatty acyl-CoA desaturase, pentosephosphate pathway
and triacylglycerol synthesis. The following nonstandard abbrevia-
tions are used: Fru, fructose; Glue, glucose; Fru-lP, fructose 1-
phosphate; Gluc-6-P, glucose 6-phosphate; TrP, triosephosphates;
OxAc, oxaloacetate; Citr, citrate; Mai, malonyl-; CAC, citric acid
cycle; Facyl-, fatty acyl; — (C=C)—, monounsaturated, long-chain fatty
acid; TAG, triacylglycerol; PPP, pentosephosphate pathway. Red
triangles denote enzymes or other proteins whose gene expression
was upregulated at least 1.5-fold, and they are as follows: 1, glucose
transporter 5; 2, ketohexokinase; 3, glucokinase; 4, components of
pyruvate dehydrogenase complex (pyruvate dehydrogenase kinase
isoenzymes); 5, citrate cleavage enzyme; 6, glucose 6-phosphate
dehydrogenase; 7, stearoyl (fatty acyl)-CoA desaturase; 8, acylglyc-
erol O-acyltransferase; 9, malic enzyme (cytosolic and NADP-
dependent). The pentosephosphate pathway (PPP) is represented here
at the lower right side of the figure by the reaction catalyzed by
glucose 6-phosphate dehydrogenase, which was upregulated in HCD-
fed animals. Genes encoding enzymes involved in fatty acid /?-
oxidation, a pathway that may contribute to triacylglycerol accumu-
lation in the liver, were not found to be changed. Also, some of the
genes listed in Table 4 are not represented in the figure in order to
keep a certain degree of simplicity. Enzyme classification (EC) for the
enzymes in the map is given in Table 3
4y Springer
Previous
-------
48
Hepatol Int (2008) 2:39^9
such an upregulation could not be accomplished otherwise
than through an adequate response of the liver cells to
insulin, (iii) nonenzymatic factors involved in lipid syn-
thesis, such as SREBP-1, are also under the control of
insulin [24, 39]; the expression of this factor that, in turn,
controls several major enzymes involved in fatty acid
synthesis [40], was upregulated 1.6-fold (Table 3). Taken
together, the gene expression data of this and of cited
studies are not compatible with an insulin-resistant liver
during the phase of NAFLD in which our animals were
killed. Further research is required to study the evolution of
the fatty liver, including potential progression to steato-
hepatitis, in the model used in this study.
In conclusion, our study (i) demonstrates the usefulness
of the mouse model of HCD-induced hepatic steatosis for
the study of the fatty liver of nutritional origin, (ii)
emphasizes the importance of using large-scale gene pro-
filing of the liver in identifying potential causes and
understanding the mechanisms underlying the disease, and
(iii) offers a database for further investigation of the
mechanisms underlying the hepatic steatosis of dietary
origin and its potential progression to NASH.
Acknowledgements The work reported in this study was supported
by National Institutes of Health grants (to I.V.D., Z.S., S.S.B., T.B.K.,
A.V.S., and C.J.M.) and a Department of Veterans Affairs grant (to
C.J.M.).
References
1. McClain CJ, Mokshagundam PL, Barve SS, Song Z, Hill DB,
Chen T, Deaciuc I. Mechanisms of non-alcoholic steatohepatitis.
Alcohol 2004;34:67-79.
2. Den Boer M, Voshol PG, Kuipers F, Havekes LM, Romijn JA.
Hepatic steatosis: a mediator of the metabolic syndrome. Lessons
from animal models. Artherioscl Thromb Vase Biol
2004;24:644-9.
3. Koteish A, Diehl AM. Animal models of steatosis. Semin Liver
Dis 2001;21:89-104.
4. Lieber CS, Leo MA, Mak KM, Xu Y, Cao Q, Ren C, et al.
Model of non-alcoholic steatohepatitis. Am J Clin Nutr 2004;79:
502-9.
5. Nanji AA. Animal models of nonalcoholic fatty liver disease and
steatohepatitis. Clin Liver Dis 2004;8:559-74.
6. Portincasa P, Grattagliano I, Palmieri VO, Palasciano G. Nonal-
coholic steatohepatitis: recent advances from experimental
models to clinical management. Clin Biochem 2005;38:203-17.
7. Deng QG, She H, Cheng JH, French SW, Koop DR, Xion S, et al.
Steatohepatitis induced by intragastric overfeeding in mice.
Hepatology 2005;42:905-14.
8. Feldstein AE, Canbay A, Guicciardi ME, Higuchi H, Bronk SF,
Gores GJ. Diet associated steatosis sensitizes to Fas mediated
liver injury in mice. J Hepatol 2003;39:978-83.
9. Friedman HA, Nylund B. Intestinal fat digestion, absorption and
transport. Am J Clin Nutr 1980;33:1108-39.
10. Foufelle F, Girard J, Ferre P. Regulation of lipogenic enzyme
expression by glucose in liver and adipose tissue: a review of the
potential cellular and molecular mechanisms. Adv Enzyme Regul
1996;36:199-226.
11. Flatt JP. Conversion of carbohydrate to fat in the adipose tissue:
an energy-yielding and, therefore, self-limiting process. J Lip Res
12. Basciano H, Federico L, Adeli K. Fructose, insulin resistance,
and metabolic dyslipidemia. Nutr Metab 2005 ;2:5. doi:
10.1186/1743-7075-2-5.
13. Diraison F, Yankah V, Letexier D, Dusserre E, Jones P, Beylot
M. Differences in the regulation of adipose tissue and liver
lipogenesis by carbohydrates in humans. J Lipid Res
2003;44:846-53.
14. Hudgins L, Hellerstein M, Seidman C, Seidman C, Neese R,
Diakun J, et al. Human fatty acid synthesis is stimulated by eu-
caloric low fat, high carbohydrate diet. J Clin Invest
1996;97:2081-91.
15. Schwartz JM, Linfoot P, Dare D, Aghajanian K. Hepatic de novo
lipogenesis in normoinsulinemic and hyperinsulinemic subjects
consuming high-fat, low carbohydrate and low-fat, high carbo-
hydrate isoenergetic diets. Am J Clin Nutr 2003;77:43-50.
16. Quintanilha AT, Packer L, Davies JM, Racanelli TM, Davies KJ.
Membrane effects of vitamin E deficiency: bioenergetic and
surface charge density studies of skeletal muscle and liver
mitochondria. Ann NY Acad Sci 1982;393:32-47.
17. Deaciuc IV, Peng X, D'Souza NB, Shedlofsky SI, Burikhanov R.
Microarray gene analysis of the liver in a rat model of chronic,
voluntary alcohol uptake. Alcohol 2004;32: 113-27.
18. Jarvelainen HA, Fang C, Ingelman-Sundberg M, Lindros KO.
Effect of chronic administration of endotoxin and ethanol on rat
liver pathology and proinflammatory and antiinflammatory
cytokines. Hepatology 1999;29:1602-14.
19. Bulera SJ, Eddy SM, Ferguson E, Jatkoe TA, Reindel JF,
Bleavins MR, et al. RNA expression in the early characterization
of hepatotoxicants in Wistar rats by high-density DNA micro-
arrays. Hepatology 2002;33: 1239-58.
20. Deaciuc IV, Doherty DE, Burikhanov R, Lee EY, Stromberg AJ,
Peng X, et al. Large-scale gene profiling of the liver in a mouse
model of chronic, intragastric ethanol infusion. J Hepatol
2004;40:219-27.
21. Spencer AF, Lowenstein JM. The supply of precursors for the
synthesis of fatty acids. J Biol Chem 1962;237:3640-8.
22. Wise EM, Ball EG. Malic enzyme and lipogenesis. Proc Natl
Acad Sci USA 1964;52: 1255-63.
23. Miiller-Wieland D, Kotzka J. SREBP-1: gene regulatory key to
syndrome X? Ann NY Acad Sci 2002;967:19-27.
24. Eberle D, Hegarty B, Bossard P, Ferre P, Foufelle F. SREBP
transcription factors: master regulators of lipid homeostasis.
Biochimie 2004;86:839^18.
25. Tan NS, Michalik L, Desvregne B, Wahli W. Multiple expression
control mechanisms of peroxisome proliferator-activated recep-
tors and their target genes. J Steroid Biochem Mol Biol
2005;93:99-105.
26. Dela Pena A, Leclerq I, Field J, George J, Jones B, Farrell G. NF-
kappaB activation, rather than TNF, mediates hepatic inflam-
mation in a murine dietary model of steatohepetitis.
Gastroenterology 2005 ; 1 29: 1 663-74.
27. Louthan MV, Barve S, McClain CJ, Joshi-Barve S. Decreased
serum adiponectin: an early event in pediatric non-alcoholic fatty
liver disease. J Pediatr 2005;147:835-8.
28. Ballestar E, Wolffe AP. Methyl-CpG-binding proteins. Targeting
specific gene repression. Eur J Biochem 2001 ;268: 1-6.
29. El-Osta A, Wolffe AP. DNA methylation and histone deacety-
lation in the control of gene expression: basic biochemistry to
human development and disease. Gene Expr 2000;9:63-75.
30. Mehlman MA. Dangerous and cancer-causing properties of
products and chemicals in the oil refining and petrochemical
industry. VIII. Health effects of motor fuels: carcinogenicity of
gasoline — scientific update. Environ Res 1992;59(l):238^-9.
4y Springer
Previous
-------
Hepatol Int (2008) 2:39^9
49
31. Cave M, Deaciuc IV, Mendez C, Song Z, Joshi-Barve S, Barve S,
et al. Nonalcoholic fatty liver disease: predisposing factors and
the role of nutrition. J Nutr Biochem 2007;18:184-95.
32. Adams LA, Angulo P, Lindor KD. Nonalcoholic fatty liver dis-
ease. Can Med Assoc J 2005;172:899-905.
33. Browning JD, Horton JD. Molecular mediators of hepatic stea-
tosis and liver injury. J Clin Invest 2004; 114:147-52.
34. Browning JD, Szczepaniak LS, Dobbins R, Nurember P, Horton
JD, Cohen JC, et al. Prevalence of hepatic steatosis in an urban
population in the United States: impact of ethnicity. Hepatology
2004;40:1387-95.
35. Evans RM, Barish GD, Wang YX. PPARs and the complex
journey to obesity. Nat Med 2004;10:l-7.
36. Hamaguchi M, Kojima T, Takeda N, Taniguchi H, Fujii K,
Omatsu T, et al. The metabolic syndrome as a predictor of non-
alcoholic fatty liver disease. Ann Intern Med 2005; 143:722-8.
37. Villafuerte BC, Phillips LS, Rane MJ, Zhao W. Insulin-response
element-binding protein 1. A novel Akt substrate involved in
transcriptional action of insulin. J Biol Chem 2004;279:36650-9.
38. Villafuerte BC, Kaytor EN. An insulin-response element-binding
protein that ameliorates hyperglycemia in diabetes. J Biol Chem
2005;280:20010-20.
39. Foufelle F, Ferre P. Regulation of carbohydrate metabolism by
insulin: role of transcription factor SREBP-lc in the hepatic
transcriptional effects of the hormone. J Soc Biol 2001; 195:243-
8.
40. Griffin MJ, Sul HS. Insulin regulation of fatty acid synthase gene
transcription: roles of USF and SREBP-lc. Life 2004;56:595-
600.
4y Springer
Previous
-------
TOXICOLOGICAL SCIENCES 110(2), 449-462 (2009)
doi:10.1093/toxsci/kfp098
Advance Access publication May 7, 2009
Mode of Action for Reproductive and Hepatic Toxicity Inferred from
a Genomic Study of Triazole Antifungals
Amber K. Goetz*'t and David J. Dix*'1
*National Center for Computational Toxicology, Office of Research and Development, U.S. Environmental Protection Agency, Research Triangle Park, North
Carolina 27711; and ^Department of Environmental and Molecular Toxicology, North Carolina State University, Raleigh, North Carolina 27695
Received January 16, 2009; accepted April 21, 2009
The mode of action for the reproductive toxicity of some
triazole antifungals has been characterized as an increase in serum
testosterone and hepatic response, and reduced insemination and
fertility indices. In order to refine our mechanistic understanding
of these potential modes of action, gene expression profiling was
conducted on liver and testis from male Wistar Han IGS rats
exposed to myclobutanil (500, 2000 ppm), propiconazole (500,
2500 ppm), or triadimefon (500,1800 ppm) from gestation day six
to postnatal day 92. Gene expression profiles indicated that all
three triazoles significantly perturbed the fatty acid, steroid, and
xenobiotic metabolism pathways in the male rat liver. In addition,
triadimefon modulated expression of genes in the liver from the
sterol biosynthesis pathway. Although expression of individual
genes were affected, there were no common pathways modulated
by all three triazoles in the testis. The pathways identified in the
liver included numerous genes involved in phase I-III metabolism
(Aldhlal, Cyplal, Cyp2b2, CypSal, Cyp3a2, Slcola4, Udpgtr2),
fatty acid metabolism (Cyp4alO, PCX, Ppap2b), and steroid
metabolism (Ugtlal, Ugt2al) for which expression was altered
by the triazoles. These differentially expressed genes form part of
a network involving lipid, sterol, and steroid homeostatic path-
ways regulated by the constitutive androstane (CAR), pregnane X
(PXR), peroxisome proliferator-activated alpha, and other
nuclear receptors in liver. These relatively high dose and long-
term exposures to triazole antiningals appeared to perturb fatty
acid and steroid metabolism in the male rat liver predominantly
through the CAR and PXR signaling pathways. These toxicoge-
nomic effects describe a plausible series of key events contributing
to the disruption in steroid homeostasis and reproductive toxicity
of select triazole antifungals.
Key Words: myclobutanil; propiconazole; triadimefon;
toxicogenomics; steroid metabolism.
Disclaimer: The United States Environmental Protection Agency through its
Office of Research and Development funded and managed the research
described here. It has been subjected to Agency administrative review and
approved for publication.
1 To whom correspondence should be addressed at National Center for
Computational Toxicology, Mail Drop D343-03, U.S. Environmental
Protection Agency, Research Triangle Park, NC 27711. Fax: (919) 541-1194.
E-mail: dix.david@epa.gov.
Published by Oxford University Press 2009.
The ability of triazole antifungals to bind and inhibit fungal
lanosterol-14a-demethylase activity (cypSl) makes this class
of compounds an effective tool in controlling many species and
strains of fungi (Ghannoum and Rice, 1999). Disruption in
ergosterol biosynthesis leads to a build up of toxic intermediate
sterols in the fungal cell membrane, increasing membrane
permeability and inhibition of fungal growth (Vanden Bossche
et al., 1990). Hence, triazole antifungals have been used for
their target pesticidal mode of action on fungal cypSl
inhibition and have proven to be valuable in the control
against multiple types of fungal disease.
Cytochrome P450 51 (CypSl) is a conserved gene in fungi,
protists, mammals, and plants required for sterol biosynthesis
in all eukaryotic systems. The ability of triazoles to bind to the
heme protein and inhibit CYP-dependent enzymes raises
concerns over triazole effects on hormone synthesis and drug
metabolism (Barton et al., 2006; Goetz et al., 2007; Hester
et al., 2006; Sun et al., 2007; Tully et al., 2006; Wolf et al.,
2006). In rodents, reproductive toxicity has been reported
following administration of myclobutanil or triadimefon, but
not propiconazole; and carcinogenicity following administra-
tion of propiconazole or triadimefon, but not myclobutanil
(Goetz et al, 2007; U.S. EPA, 1995, 1996, 2001, 2005a,b,c,
2006). These data prompted interest in gaining a better
understanding of the modes and mechanisms of action for
triazole related reproductive toxicity and whether there are
common modes of actions for triazole fungicides.
In a 14-day oral (gavage) toxicity study, adult male Sprague
Dawley rats were administered fluconazole (0, 2, 25, or 50 mg/
kg/day), myclobutanil (0, 10, 75, or 150 mg/kg/day), propico-
nazole (0, 10, 75, or 150 mg/kg/day) or triadimefon (0,10, 50, or
115 mg/kg/day). Only myclobutanil (150 mg/kg/day) produced
a statistically significant increase in serum testosterone levels
(Tully et al., 2006). In contrast results from a reproduction and
fertility study examining developmental and adult reproductive
effects in Wistar Han rats exposed via feed to myclobutanil (ca.
6.1, 32.9, or 133.9 mg/kg/day), propiconazole (ca. 6.7, 31.9, or
169.7 mg/kg/day) or triadimefon (ca. 6.5, 33.1, or 139.1 mg/kg/
day) from gestation day 6 to postnatal day 92, demonstrated that
all three triazoles caused a significant increase in serum
Previous
TOC
-------
450
GOETZ AND DIX
testosterone levels. This evidence combined with reduced
fertility, hepatomegaly, and changes in pituitary weights
(myclobutanil only) suggested a disruption in testosterone
homeostasis was a key mode of action in the reproductive
toxicity (Goetz et al., 2007). However, the modes of triazole
related toxicity to the testis and liver, disruption in testosterone
homeostasis, and reduced fertility were unclear.
Previous research on triazoles has focused on the metabolic,
hepatic and thyroid response to myclobutanil, propiconazole
and triadimefon following short term (4 days) to subchronic
(90 days) exposure in the adult rat and/or mouse (Allen et al.,
2006; Barton et al., 2006; Chen et al., 2009; Goetz et al., 2006;
Hester and Nesnow, 2008; Hester et al., 2006; Sun et al., 2007;
Tully et al, 2006; Ward et al, 2006; Wolf et al, 2006). This
work focused on delineating common modes of triazole
toxicity in the adult animal, with an emphasis on defining
biological pathways relevant for both carcinogenic and
noncarcinogenic effects. Toxicogenomics studies have been
conducted in liver, thyroid, and testis following 4 days to
subchronic exposures to these three triazoles. The goal of the
present study was to gain a better understanding of the modes
and mechanisms of action for the triazole reproductive toxicity
observed following gestation to adulthood exposure (Goetz
et al, 2007). The specific aim was to identify key biological
pathways affected by triazoles in the liver and testis following
exposure from gestation through adulthood, in order to better
understand the disruptions in testosterone homeostasis.
Using tissue samples from the previously published study
(Goetz et al, 2007), gene expression changes were assessed in
the male rat liver and testis following exposure to myclobu-
tanil, propiconazole, or triadimefon from gestational day (GD)
6 through postnatal day (PND) 92 in order to test the
hypothesis that triazoles disrupt testosterone homeostasis by
increasing expression of genes involved in testosterone
synthesis in the testis and decrease expression of genes
involved in testosterone metabolism within the liver. The
PND92 time point was chosen for this study due to increased
testosterone levels and compromised fertility in young adult
males at this time point following exposure to triazole
antifungals (Goetz et al, 2007). We also hypothesized that if
triazoles share common modes of action, exposure to these
various triazoles would result in similar expression profiles of
steroidogenic and steroid metabolism related genes. The gene
expression profiling in this study, in addition to prior
toxicological assessments, was expected to provide mechanis-
tic insights into potentially shared modes of action for
myclobutanil, propiconazole, and triadimefon.
MATERIALS AND METHODS
Animal husbandry and dosing regimen. A full description of animal
husbandry and dosing is described in Goetz et al. (2007). Briefly, animal care,
handling, and treatment were conducted in an American Association for
Accreditation of Laboratory Animal Care-International accredited facility,
and all procedures were approved by the U.S. Environmental Protection
Agency (EPA) National Human and Environmental Effects Research
Laboratories Institutional Animal Care and Use Committee. Timed pregnant
Wistar Han IGS rats were received from Charles River Laboratories (Raleigh,
NC) on GD1-3; single housed, and allowed to acclimate for 3 days prior to
the start of the treatment. Dams delivered naturally with day of delivery
designated as PNDO for the FI generation. FI offspring were housed with their
respective mothers until weaning on PND23. Males were removed from the
dams and housed by treatment group in pairs until PND50. Males were single
housed after PND50.
Feed was prepared by Bayer CropScience (Kansas City, MO) as part of
a Materials Cooperative Research and Development Agreement between the
U.S. EPA and the U.S. Triazole Task Force. Control animals were fed 5002
Certified Rodent Diet with acetone vehicle added. Treatment groups used in
this study received feed containing a dietary concentration of myclobutanil
(M: 500 or 2000 ppm), propiconazole (P: 500 or 2500 ppm), or triadimefon
(T: 500 or 1800 ppm). Dose levels used for this study were selected to match
dose levels used in regulatory studies for registering these triazoles with the
U.S. EPA. Dams began treated feeds on GD6, feed intake and body weights
were measured weekly. The Ft generation continued on the same treated feed
diets; feed intake and body weights were measured weekly until necropsy.
Refer to Goetz et al. (2007) for achieved dose levels on a week by week basis.
One male from each litter was necropsied on PND92 for transcriptional
profiling analysis.
RNA isolation. Total RNA was extracted from PND92 liver and testis of
control and treatment groups using TRI Reagent (Molecular Research Center
Inc., Cincinnati, OH) according to the manufacturer's protocol and subjected to
quality control measures before application to microarrays. For quality control,
RNA A2eo^280 ratios were assessed via NanoDrop Fluorospectrometer
(NanoDrop Technologies, Inc., Wilmington, DE). RNA absorbance readings
with a range 1.8-2.1 were followed with DNase treatment, Total RNA Cleanup
(Qiagen RNeasy), and checked for RNA quality using the model 2100
Bioanalyzer (Agilent Technologies, Inc., Palo Alto, CA). Samples with a ratio
of 28S: 18S rRNA > 1.6 were accepted for subsequent use in DNA microarrays.
RNA was stored at — 80°C until labeling for microarray hybridization.
Microarray hybridization and scanning. Microarrays and reagents were
provided by Affymetrix as part of a Materials Cooperative Research and
Development Agreement. Microarray processing was conducted for EPA by
Expression Analysis Inc. (Durham, NC). Five micrograms of purified total
RNA from each liver or testis of three to seven individual rats per treatment
group was hybridized to Affymetrix GeneChip Rat Genome 230 2.0 plus
microarrays according to the Affymetrix GeneChip Expression Analysis
Technical Manual (www.affymetrix.com).
Microarray and probe set analysis. To minimize nonbiological factors,
for example, total amount of target hybridized to each array, signal values from
each microarray were multiplied by a scaling factor to achieve a mean intensity
equal to 500. Converted .eel files were loaded into the JMP Genomics program
(SAS, Inc., Gary, NC), Log2 transformed, normalized using interquartile
normalization, and analyzed for significant changes in transcript levels through
row-by-row modeling using one-way ANOVA. For initial exploratory analysis,
principle component analysis (PCA) was applied using JMP Genomics.
Comparisons were made between controls and each treatment group with
statistical cut-offs applied at a p value adjusted false discovery rate (FDR) of
10% for liver (p < 0.000724) or FDR of 25% for testis (p < 0.000229), and an
absolute difference of 11.21 or greater. Probe sets representing transcribed loci,
unknown genes, and image clones were removed from the final list of each
analysis; probe sets with predicted annotations were kept in the analysis. The
Affymetrix .eel files can be accessed through Gene Expression Omnibus
(www.ncbi.nkn.nih.gov/geo); series accession numbers GSE10411 and
GSE10412.
Pathway analysis. Ingenuity Pathways Analysis (IPA; Ingenuity Systems,
www.ingenuity.com) was used for initial pathway level analysis. Genes from
Previous
TOC
-------
TRIAZOLE TOXICOGENOMICS
451
the data set that met the absolute difference cut-off of 11.21, p value cut-off
based on the data set p value adjusted FDR, and associated with a canonical
pathway in the Ingenuity Pathways Knowledge Base (IPKB) were considered
for the IPA-based analysis. Canonical pathways were identified from the IPA
library that were impacted most significantly by the triazoles. The significance
of the association between the data set and the canonical pathway was
measured using the ratio of genes from the data set that mapped to the pathways
divided by the total number of genes that mapped to the canonical pathway.
Significance was calculated using the right-tailed Fisher's Exact Test by
comparing the number of focus genes that participated in a given pathway,
relative to the total number of occurrences of these genes in all path-
way annotations in the IPKB. Using this methodology, over-represented path-
ways were identified containing more focus genes than expected by chance.
Further analysis of a broader set of genes and pathways included those
identified by IPA, in combination with relevant pathways from the Kyoto
Encyclopedia of Genes and Genomes (http://www.genome.jp/kegg/pathway.html)
and other references to associate genes with the most complete biological
pathways possible. For Figure 1 and the fatty acid metabolism pathway, additional
references included Coleman etal. (2000), Nordlie and Foster (1999), and Sul and
Wang (1998). The sterol biosynthesis pathway in Figure 2 was supplemented with
information from Shibata et al. (2001) and Tansey and Shechter (2000). Figure 3,
the cholesterol and bile acid biosynthetic pathway utilized Pandak et al. (2002),
Russell (1999), Schwarz et al. (2001), Staudinger et al. (2001), and Wang et al.
(2005). Figures 4 and 5 on nuclear receptor signaling pathways depended on
Baldan et al. (2006), Dixit et al. (2005), Guzelian et al. (2006), Jigorel et al.
(2005, 2006), Kretschmer and Baldwin (2005), Maglich et al. (2002), Shenoy
et al. (2004), Yoshikawa et al. (2003), and You (2004).
Quantitative PCR. TaqMan-based quantitative RT-PCR was used to
determine the relative levels of Abcbl, Cyplal, Cyp2bl/Cyp2b2, CypSal,
Cyp3a2, Cyp4al, and Ugtlal mRNA in the samples from each treatment group.
Primer/probe sets specific for each gene were utilized from Applied Biosystems
(Foster City, CA) for Abcbl (Rn00561753_ml), Cyplal (Custom assay),
Cyp2bl/2 (Custom assay), Cyp3al (Rn01640761_gl), Cyp3a2 (Rn00756461_ml),
Cyp4al (Rn00598510_ml), and Ugtlal (Rn00754947_ml). The exception to
this was for Cyp2bl and Cyp2b2, for which the primer/probe set could detect
either gene transcript-that is why these results are hereafter referred to as
Cyp2bl/2. A two-step RT-PCR process was performed by initial reverse
transcription of ca. 200 ng of total RNA in a 60-ul reaction using the High
Capacity cDNA Archive Kit (Applied Biosystems, Foster City, CA), followed
by quantitative PCR amplification with isoforms-specific primer/probe sets on
2 ul of each reverse transcribed cDNA. The reactions were characterized by the
point during PCR amplification at which fluorescence of the product crossed
a defined threshold (Or, automatically determined by the PE Applied Biosystems
ABI 7900HT Sequencer software), and inspected to ensure all CT values were
within the linear phase (log scale) of exponential growth for all targets. CT values
were determined for target CYP genes and an endogenous control gene, p-actin.
Each sample was normalized to both the P-actin control and to a vehicle control.
A difference of one CT was considered equivalent to a twofold difference in gene
expression (exponential relationship, i.e., RQ = 2~DDCt). Sample means for each
replicate were determined along with the standard error of the mean if
appropriate and percent of adjusted positive control. Relative fold changes in
mRNA content were analyzed using the Kruskal-Wallis nonparametric ANOVA
with Dunn's multiple comparisons post-test, measures with p < 0.05 were
considered significant.
Glucose
1
Glucose-6-phosphate
I
**•"
***
uiyceraiaenae-j-pno
3hosphoenol- „ Pklr
. •* Pyr
glucose mediated ind
decreased by PU
Pdk
1 RSI RS
x- — ^
i Co
Fatty Acid >J
(3 Oxidation -J^
M&?
ogt
we Up-regulated
R5 Down-regulate
jction,
<=A
Pyr
, PCV |
in
Ac<
> Ac
/^>
/ Fatty Acid Ma
' Synthesis
Cyp3a3
a-Hydroxy FA-* P
1 183 II \Cvp4aW
I I I %9% \Cvp4a12
I R5I I t$ACvp4a1
acid | |j^
IS?) 1
(s)-3-Hydroxy-3-
methylglutaryl-Co/
|Hmgcs2
Acetoacetyl-CoA
etyl-CoA A
J/Acafr
X\cac6 A
I \Echdc1
lonvl-CoA I I
Fasn i
almitate HNS
1 m
ACSI4
SJ VA
J 1 53
Gpam
Monoacylgly
3-phosphe
:erol
ate
I Dga(2
Triacylglycerol
Sterol
^ Biosynthes s
*®
Triacylglycero
|Dgaf2
'\ ,2-Diacylglycerol
t Ppap2b
\ $$ W 1
acid
4 Agpat4
Phospholipid
FIG. 1. Effects of three triazoles on the expression of genes in rat liver from the fatty acid metabolism pathway. Key in the lower left corner indicates the order
of presentation for the six treatment groups, and up- or downregulated genes.
Previous
TOC
-------
452
GOETZ AND DIX
Acetoacetyl-CoA HmgCs2 > (s)-3-Hydroxy-3-
methylglutaryl-CoA
Hmgcr
:*• Mevalonate
Mvk
err
-* Mevalonate-5P
HMG CoA reductase
Mevalonate-5PP
Mvd
LLL
Sqle
Squalene
Fdftl
TTTIgg
Presqualene-PP ±:
Fdftl
Farnesyl-PP
(S)-Squalene-
2,3-epoxide
I
24,25-dihydro-
lanosterol
Lathosterol
Fdps
Sc5d
Isopentenyl-PP
7-dehydro-
cholesterol
Lanosterol
Cyp51
Lanosterol 14a-
demethylase
FF-MAS—»• T-MAS —> Zymosterol —
Sterol-A14- Sterol-4,4- 1 Ebp; sterol-
& Up-regulated
Down-regulated
reductase
Tm7sf2
X
demethylase
A8,A7-isomerase
Sterol-C5-
desaturase
Sterol-A7-
reductase
DhcrJ
Desmosterol —> CHOLESTEROL
Steml-A24- *
reductase
FIG. 2. Triadimefon effects on the expression of genes in rat liver from the sterol biosynthesis pathway. Key in the lower left corner indicates the order of
presentation for the six treatment groups, and up- or downregulated genes.
RESULTS
GeneChip Quality Analysis
Microarrays of poor quality, with a scaling factor of greater
than 15.0 were removed from the analysis. Three GeneChips
from the liver and three GeneChips from the testis data sets
were removed prior to normalization and statistical analysis of
the two separate data sets. Each treatment group had three to
seven GeneChips available for robust analysis following
removal of microarray chips with weak intensity readings.
Probe Set Analysis in the Liver
PCA applied to the normalized data set grouped microarrays
by treatment group, with largest variation occurring within the
controls, M500, and T1800 groups. It is unlikely this variation
is due to sample size (Controls: 7; M500: 4; T1800: 3
microarray chips). Gene expression changes were determined
using one-way ANOVA and a FDR of 10% as the multiple
testing correction method to control the familywise error rate.
The FDR of 10% (a = 0.10) generated a p value cut-off of
7.24E—4 and was considered an ideal cut-off in order to obtain
a subset of 1,043 differentially expressed probe sets for use in
pathway and gene-level analyses. Of those probe sets, 455 had
an absolute difference of 11.21 or greater. Removing probe sets
that interrogated unknown genes or transcribed loci, the final
list of probe sets identified 308 genes up- or downregulated by
myclobutanil, propiconazole, or triadimefon (Table 1). The 16
genes differentially expressed in response to all three triazoles,
as detected by microarray, are listed in Table 2. Although four
isoforms of Cyp2B (2b2, 2b3, 2bl3, 2bl5) were assessed using
the Affymetrix Rat 230 2.0 GeneChip, only Cyp2b2 was
induced in all the triazole treatment groups. The majority of
genes modulated by all three triazoles function in lipid or fatty
acid metabolism, transporter, and xenobiotic metabolism
pathways; thus a pathway based approach was the focus of
subsequent analysis and interpretation.
Pathway Analysis in the Liver
Pathway analysis identified common biological pathways
and processes affected by the three triazoles in rat liver. For
initial analysis of pathways the entire liver data set (31,099
probe sets) was uploaded into IPA and the absolute difference
11.21 or greater and p value < 7.24E—4 was used to identify
differentially expressed genes, In the liver data set, 180 of the
308 significant probe sets mapped to the IPKB. These focus
genes were overlaid onto a molecular network developed from
information within the IPKB. Table 3 shows the pathways
identified by IPA as being affected by the three triazoles.
Several metabolic pathways are common across the three
triazoles including androgen and estrogen, arachidonic acid,
Previous
TOC
-------
TRIAZOLE TOXICOGENOMICS
453
C21-Steroid ,-^ Bile Acid -x
Hormone > /l)J' Biosynthesis ^ \
/"~ ^\C
/ CHOLESTEROL -, N
Cypfta7\
20a, 22p-Dihydroxy-
cholesterol
\Cyp11a1
Cyp1
V Mitochondria /
Estriol 16-
Glucuronide
VVV
~-^\r
yp7a1 7a.Hvdroxy- . .. SrdSaf __ 7a,12a-Dihvdroxv-
i ..I
A-p27af
x^ 27-Hydroxy- Cyp7b1
cholesterol | | | [^ |
r r- II. 1 n
7a-Hydroxylated
Oxysterol *" *" Primary
Cyp7b1 is regulated by Choi, and B.A.
Decrease in Choi. = decrease in Cyp7b1
Hsd3b
791 , 17«-HyHrnvy- CyP17a1 ^ Dehvdro_ HSd3b1 . 3p
pregnenolone
I M W Wuat1a1
\ M m \ \Ugtla1 E3
I M m MUdpgtrt
Hsd17b7
I 583 K8 I22 Uat1a1
-S-jScv^ I 9& 9K I \Ugt2a1
*&£/
-------
454
GOETZ AND DIX
i Triazole
Fatty Acids Oxysterols Bile Acids
s"
Aip
TAhr
Hspca '
i i ^ i
| Hspca 1
*
V
r—1 i |
p
TH
u ^ ^ik
[ I ! I I ] I I f~^~\ \ S^^S C jn^tLJ*
[LXRo] [FXR
INr1h3l lNr1h4F
I I
i 1
Arnt Ncor^ i 1, 1 , II 1 j 1
1 Ahr 1 \ [PparalRXRj [LXRqlRXRj |_ FXR
n I / 1 I! 1 INr1h3l1 1 lNr1h4f
]
_^S
ERXRJ
ARE II * PPREl I RE I I I RE I
Cyp1a1
j J J
j | | m m Cyp4alO | jSj I Alasi 1 1 1 1 1 W
Cvp4ai2 i i i Ka«a i ces2 i m m w
Cyp4a1 I ISS I I Rvl Me1 | | | ffi gg
Me3 I ! ! RSfSJ I
Scd2 I I I I I m
Cyp7h1l I I RSI I I
J
Triazole |
Cytoplasm
^
JGAKJ *
I Nr1 3 I
J^— Gripl
Nucleus
fCARTRXR
Ncoal ->
— XREMI )•
: - -
; Triazole :
/\l 88 I I I
\ / DnajcY
) " ICAR
I \JI VI \
Hspca
CAR specific CAR/PXR PXR specific _!_
Phase I Phase II IPX!
Cvo1a1 Fl 0n?3 Gstm4 \\\VW\ Aldhlal I W W W L,^!
Cvo2b2 I03i%£t%£££0 Udpgtr2 I W W, W, Cyp3a1 [_W W W .
Cyp2b1/2L W Ml M Ugt1a1 \ VA YA W Cyp3a2 \ W W \ I
Ahr I 83 ggj I I Ugt2a1 \ YA YA \ \ *
Gadd45bmJJ^ Phase III fPXR
-LSLo ^ AbCbla PS "iNrliZj
^s
R 1
~ Ncorf
IRXRJ
1 ™'"~ ••'""•• AbccS 1 I I BB Bl —XREMI
1 ^D2b 1 [j H H Slco1a4 I VA VOW |
J8t®|?
I I I I ]
82 Up-regulated
S Down-regulated
Alasl 1 1 1 1 B-J Alasi : heme svnthesis:
Por 1 1 1 1 JSJ Por cnfica/ tor all Uyp
reacf/ons
4 1
FIG. 4. Impact of triazoles on nuclear receptor regulated gene expression in the rat liver. Key in the lower left corner indicates the order of presentation for the
six treatment groups, and up- or downregulated genes.
Previous
TOC
-------
TRIAZOLE TOXICOGENOMICS
455
Androstanes
Xenobiotics
Bilirubin
Acat2 ((VLDL)
assembly!.
secretion)
Sqle, Lss, Cyp51
Fatty Acid
(3-oxidation
PC
Ppap2b
Aldh1a1
Xenobiotics
PPARa
RXR
Phase I Phase II Phase III
Aldh1a1 Gstm4 AbccS Alasl
Cyplal Ugtlal Abcbla For
Cyp2b1/2 Ugt2a1 Slco1a4
Cyp3a1/2 Udpgtr2
Apolipoproteins
Lipoprotein Upases
FA transport proteins
Peroxisomal FA MEs
Mitochondrial FA MEs
Cell Growth [
Hepatomegaly
Detoxification
Elimination
control plasma
lipid transport
Cd36
(FA translocase)
FIG. 5. Nuclear receptor regulation of genes, enzymes, pathways, and processes in the liver. Genes listed represent effects of triazoles on expression in rat
liver from this study. Expression of these genes is regulated by the nuclear receptors CAR, PXR, and PPAR-a, and the activity of these receptors, or transcription
factors, is modulated by various endobiotics and Xenobiotics. Perturbations of CAR, PXR, and PPAR-a signaling pathways can alter lipid and steroid homeostasis
and promote hepatotoxicity.
treatment, respectively. The two triazoles target different gene
transcripts, however both enzymes' activities are involved in
primary bile acid synthesis. Propiconazole also decreased the
transcript levels of inositol 1,4,5-triphosphate receptor 2 (Itpr2)
which is involved in intracellular calcium homeostasis. The
transcript levels for several uridine diphosphate glucuronosyl-
transferases (Ugtlal, Ugt2al, Udpgtr2) were increased by
all three triazoles, indicating increased metabolism of steroids
and Xenobiotics. Increased expression of hydroxysteroid
17p-dehydrogenase (Hsdl7b) by triadimefon also indicated
elevated androgen and estrogen metabolism in the liver.
Nuclear receptor regulated genes. Figure 4 shows the
impact of triazoles on nuclear receptor regulated genes within
the rat liver. There were several genes differentially expressed
that are regulated by the constitutive androstane receptor (CAR),
pregnane X receptor (PXR), aryl hydrocarbon receptor (AhR),
as well as the peroxisome proliferator-activated receptor alpha
(PPAR-a) and liver X receptor (LXR). The genes regulated by
CAR and PXR are phase I, II, and III enzymes that are part of
fatty acid and xenobiotic metabolism, sterol biosynthesis,
steroid metabolism, and cell cycle pathways. Increased
transcript levels of most of these CAR/PXR-regulated genes
indicate activation of CAR and/or PXR. Also consistent with
CAR activation, the genes PCX and Ppap2b were downregulated.
Results for Cyp2b2 from the arrays and for Cyp2bl/Cyp2b2
from PCR are both listed (Table 4). CAR-specific, PXR-
specific, and genes coregulated by CAR and PXR are identified
in Figure 4. Additional genes with overlapping regulation by
multiple receptors, modulated by triazoles in rat liver; included
Alasl expression regulated by CAR, PXR, and LXR, and
Cyplal expression regulated by AhR and CAR. Effects on
nuclear receptor regulated genes included increased expression
of steroid and xenobiotic metabolism genes aldehyde de-
hydrogenase (Aldhlal), Cyplal, and Cyp2b2, the previously
mentioned glucuronide and glucoside conjugation genes
Ugtlal, Ugt2al, and Udpgtr2, glutathione conjugator Gstm4
and phase III transporter Abcc3; all changes were likely to have
altered metabolism and excretion of steroids and Xenobiotics.
A more integrated representation of the genes, enzymes,
pathways and processes regulated by nuclear receptors in the
liver is presented in Figure 5. Various agonists and antagonists
of the CAR, PXR, and PPAR-a receptors are indicated, as well
as the regulated genes which in this study were differentially
Previous
TOC
Next
-------
456
GOETZ AND DIX
TABLE 1
Number of Affymetrix Probe Sets Signaling Significant
Treatment-Related Gene Expression Changes in Rat Liver and
Testis
Dose
level
Triazole (ppm) (mg/kg/day)
Myclobutanil
Liver
Liver
Testis
Testis
Propiconazole
Liver
Liver
Testis
Testis
Triadimefon
Liver
Liver
Testis
Testis
500 (32.9)
2000 (133.9)
500 (32.9)
2000 (133.9)
500 (31.9)
2500 (169.7)
500 (31.9)
2500 (169.7)
500 (33.1)
1800 (139.1)
500 (33.1)
1800 (139.1)
Downregulated
probe
sets"
4
64
0
0
2
44
0
1
46
23
1
0
Up
regulated
probe sets"
1
9
16
0
5
45
0
49
8
154
4
6
Total
number
of probe sets
5
73
16
0
7
89
0
50
54
177
5
6
"Probe sets significantly changed with a fold change greater than 11.21.
expressed in response to triazole exposure. This included up
regulation of the antiapoptotic genes Gadd45b and Mdm2.
Fatty acid metabolism, sterol biosynthesis, steroid metabolism,
bile acid metabolism, and cellular growth pathways are all
coordinately regulated by these nuclear receptors. Based on
these findings, it is postulated that alterations in these pathways
from exposure to xenobiotics like the triazoles, mediated by
these nuclear receptors, resulted in effects on biological
processes including hepatomegaly, detoxification and elimina-
tion, and plasma lipid transport.
Quantitative PCR
To further examine and confirm changes in liver gene
expression, a set of genes modulated by all three triazoles, as
well as additional CAR regulated genes, were analyzed by
PCR. Quantitative PCR confirmed the increased expression of
Cyplal and Cyp2bl/Cyp2b2, and yielded more definitive
results for Cyp3al and Ugtlal (Table 4). Based on the PCR
results, Cyp3al is the 17th gene for which expression is
modulated by all three triazoles (see Table 2 for other 16).
Cyp3a2 was not represented on the microarray. PCR detected
an increase in Cyp3a2 mRNA in response to myclobutanil and
propiconazole, and with the increased Cyp3al mRNA content
by all three triazoles, suggests PXR activation by triazoles in
the rat liver. The magnitude of increased Cyp2bl/Cyp2b2
indicated a strong activation of CAR consistent with a malad-
aptive toxic response as a result of long term triazole exposure.
The one clear case of discordance between microarray and
PCR results, for Abcbl and triadimefon, may be due to the
sequences used for probes in the separate assays.
Probe Set Analysis in the Testis
Gene expression changes were determined using a one-way
ANOVA and a FDR of 25% (a = 0.25) which generated a
p value cut-off of 2.29E—4 yielding 169 differentially expressed
probe sets. Of those probe sets, 77 had an absolute difference of
11.21 or greater. Removal of probe sets interrogating unknown or
transcribed loci, the final list of probe sets equaled 70 (Table 1).
An ANOVA analysis using a FDR of 10% defined one unknown
TABLE 2
Expression Changes Common to all Three Triazoles for 16 Genes in Rat Liver, as Detected by Microarray
Accession
number
NMJH7161
BM392091
NM_022407
NM_012737
BM986220
NMJ33586
AI137640
BF564195
AI454613
BG378579
BM390462
AI010233
BI296089
U95011
NM_031741
M13506
Gene
symbol
Adora2b
Ahctfl
Aldhlal
Apoa4
App
Ces2
Cldnl
Crem
Cyp2b2
Htatip2
RGD1310209_pre
Ccdcl26
RGD1562101_pre
Slcola4
Slc2a5
Udpgtr2
Myclobutanil
(32.9 mg/kg/day)
-1.24302
-1.12196
1.60289
-1.09288
1.08894
1.159467
-1.172
-1.20309
2.728842
1.047174
-1.1009
-1.29462
-1.11631
1.212303
-1.0778
1.688198
Myclobutanil
(133.9 mg/kg/day)
-1.29105
-1.30861
1.999272
-1.87156
1.471356
1.650327
-1.36187
-1.40519
3.292202
1.232627
-1.25043
-1.38855
-1.22294
1.954614
-1.22666
2.04723
Propiconazole
(3 1.9 mg/kg/day)
-1.20012
-1.13063
1.310263
-1.45817
1.359173
1.347478
-1.20889
-1.20393
2.793866
1.15738
-1.18184
-1.28179
-1.11001
1.616718
-1.16297
1.678868
Propiconazole
(169.7 mg/kg/day)
-1.30677
-1.23386
2.880486
-1.99054
1.753045
2.317871
-1.42721
-1.32225
4.672197
1.369729
-1.32877
-1.35011
-1.27771
2.129518
-1.23688
2.385029
Triadimefon
(33.1 mg/kg/day)
-1.16624
-1.14095
1.532281
-1.66115
1.256155
1.325125
-1.32804
-1.1299
2.810611
1.120655
-1.37249
-1.24059
-1.14867
2.048135
-1.16947
1.614087
Triadimefon
(139.1 mg/kg/day)
-1.293
-1.26832
4.841574
-2.79911
1.986423
3.361062
-1.38697
-1.39254
7.625469
1.344915
-1.30658
-1.30313
-1.27933
1.973023
-1.18458
3.023706
Note. Values given as fold change relative to control. Bold: transcript level changes were significant. Suffix _pre represents probe sets with predicted annotation.
Previous
-------
TRIAZOLE TOXICOGENOMICS
457
TABLE 3
Biological Pathways Containing Significant Changes in Rat Liver Gene Expression following Exposure to Triazoles
Myclobutanil
Propiconazole
Triadimefon
Aminosugars metabolism
Androgen and estrogen metabolism
Arachidonic acid metabolism
Ascorbate and aldarate metabolism
Fatty acid metabolism
Fructose and mannose metabolism
Galactose metabolism
Glycerolipid metabolism
Glycolysis/gluconeogenesis
Linoleic acid metabolism
Metabolism of xenobiotics by cytochrome
P450
p38 MAPK signaling
Pentose and glucuronate interconversions
Retinol metabolism
Starch and Sucrose metabolism
Tryptophan metabolism
Wnt/p-catenin signaling
Xenobiotic metabolism signaling
Amyloid processing
Androgen and estrogen metabolism
Arachidonic acid metabolism
Arginine and proline metabolism
Ascorbate and aldarate metabolism
Fatty acid metabolism
Fc Epsilon RI signaling
Glutamate metabolism
Glutathione metabolism
Glycerolipid metabolism
Glycerophospholipid metabolism
Linoleic acid metabolism
Metabolism of xenobiotics by cytochrome
P450
Pentose and glucuronate interconversions
Phospholipid degradation
Pyruvate metabolism
Sphingolipid metabolism
Starch and sucrose metabolism
Tryptophan metabolism
Xenobiotic metabolism signaling
Androgen and estrogen metabolism
Antigen presentation pathway
Arachidonic acid metabolism
Butanoate metabolism
Cyanoamino acid metabolism
Cysteine metabolism
Fatty acid metabolism
Glutathione metabolism
Glycine, Serine and Threonine metabolism
Glycolysis/gluconeogenesis
Histidine metabolism
IL-6 signaling
Keratin sulfate biosynthesis
Linoleic acid metabolism
Metabolism of xenobiotics by cytochrome
p450
N-Glycan degradation
Nitrogen metabolism
Propanoate metabolism
Pyruvate metabolism
Sterol Biosynthesis
Toll-like receptor signaling
Tryptophan metabolism
Xenobiotic metabolism signaling
Note. Pathways listed were affected by mid and/or high dose of each triazole. Bold: pathways affected by two or more triazoles.
gene; a FDR of 15 or 20% defined a list of 10 genes. The liberal
cut-off was used in order to obtain a subset of genes for pathway
analysis. It is clear, however, from the ANOVA that there were
either large variations within treatment groups or very small
changes in gene expression within the testis.
Pathway Analysis in the Testis
IPA was used to identify common pathways affected by the
three triazoles. From the entire data set (31,099 probe sets), the
absolute difference of 11.21 or greater andp value < 2.29E—4 was
used to identify genes whose expression was differentially
regulated. In the testis data set, 50 of the 72 significant probe sets
mapped to the IPKB. There were no common pathways affected
by all three triazoles, however, there were five pathways
modulated by at least two triazoles (Table 5). The IPA-based
analysis of altered pathways in the testis did identify 11
potentially significant matches to pathways affected by triazoles
in the liver. Many of these pathways common to testis and liver
TABLE 4
Comparisons between Microarray and Quantitative PCR Measurement of Gene Expression in Rat Liver following Triazole Exposure
Abcbl
Cyplal
Cyp2b2
CypSal
Cyp3a2
Cyp4al
Ugtlal
Treatment
ng/kg/day Array qPCR Array qPCR Array" qPCR Array qPCR Array6 qPCR Arrayc qPCR Array qPCR
Myclobutanil
Propiconazole
Triadimefon
134
170
139
1.24
1.48
1.41
-2.37
1.64
-4.99
1.76
3.02
5.60
5.82
21.31
79.99
3.29
4.67
7.63
64.57
132.12
63.13
-1.12
1.08
1.98
2.05
2.04
18.20
3.22
4.01
1.57
-1.73
-1.59
-1.30
-1.91
-1.53
-2.70
1.50
1.82
1.78
1.44
1.32
7.51
Note. Values given as fold change relative to control. Bold: significant transcript level or fold changes.
"Probe set representing Cyp2b2 on GeneChip.
6No representative probe set on GeneChip.
cProbe set representing Cyp4alO on GeneChip.
Previous
TOC
-------
458
GOETZ AND DIX
TABLE 5
Biological Pathways Containing Significant Changes in Rat Testis Gene Expression following Exposure to Triazoles
Myclobutanil
Propiconazole
Triadimefon
Androgen and estrogen metabolism
C21-Steroid hormone metabolism
Nitrogen metabolism
Propanoate metabolism
Sterol biosynthesis
Valine, leucine, and isoleucine
degradation
fJ-Alanine metabolism
Androgen and estrogen metabolism
Arachidonic acid metabolism
Complement and Coagulation cascades
Cysteine metabolism
Fatty acid metabolism
Linoleic acid metabolism
Metabolism of xenobiotics by cytochrome
P450
Methionine metabolism
Pentose and glucuronate interconversions
Phenylalanine metabolism
Retinol metabolism
Selenoamino acid metabolism
Starch and sucrose metabolism
Taurine and hypotaurine metabolism
Tryptophan metabolism
Urea cycle and metabolism of amino groups
Xenobiotic metabolism signaling
Fatty acid metabolism
Neuregulin signaling
Propanoate metabolism
Valine, leucine, and isoleucine
degradation
fJ-Alanine metabolism
Note. Pathways listed were affected by mid and/or high dose of each triazole. Bold: pathways affected by two or more triazoles.
are critical to reproduction, including androgen and estrogen
metabolism, C21-Steroid hormone metabolism, and sterol bio-
synthesis. The other common pathways are significant in relation
to how testis and liver respond to triazole exposures: metabolism
of xenobiotics by cytochrome P450, and xenobiotic metabolism
signaling. Several additional pathways recognized in the testis,
were common and robust responders to all three triazoles in the
liver: arachidonic acid metabolism, fatty acid metabolism, linoleic
acid metabolism, and tryptophan metabolism.
Expanding on the pathways highlighted by the IPA analysis,
the larger set of 72 differentially expressed genes were
examined based on relevant biological processes or pathways.
No one major biological process or pathway stood out in this
analysis. This set of genes did map to the lipid, fatty acid and
C21-steroid hormone metabolism, inflammatory response,
intracellular signaling, or xenobiotic metabolism pathways.
However, due to the liberal threshold (FDR of 25%) used
during the ANOVA, and the limited number of genes
differentially expressed, it was difficult to have confidence in
these results for defining specific modes of triazole related
toxicity within the testis. This was exacerbated by the fact that
propiconazole, which demonstrated no overt reproductive
toxicity, caused a greater number of differentially expressed
genes in the testis compared with the two reproductive
toxicants, myclobutanil and triadimefon.
DISCUSSION
Toxicological endpoints from our previous assessment of the
reproductive toxicity of myclobutanil, propiconazole, and
triadimefon identified a potential mode of action for the
reproductive toxicity of triazole antifungals (Goetz et al.,
2007). The combination of increased serum testosterone levels
by all three triazoles, increased anogenital distance and testis
weight, hepatomegaly, and decreased insemination and fertility
indices strongly suggested disruption in testosterone homeo-
stasis as a mode of action for triazole toxicity. However, the
molecular mechanisms underlying these effects remained
indeterminate. This study was designed to test the hypothesis
that disruption in testosterone homeostasis was a result of
changes in gene expression leading to increased steroidogen-
esis in the testis and decreased steroid metabolism in the liver.
Furthermore, it was designed to identify whether this putative
mode of action was common to the triazoles, and if so, to gain
mechanistic understanding of the common biological pathways
perturbed by triazoles that lead to toxicity.
Analyses based on biological pathways was used to interpret
the significant gene expression changes in the liver and testis
and to provide context for interpreting these changes.
Numerous common gene transcripts and biological pathways
were identified for triazoles in the liver defining common
biological processes modulated by all three triazoles and
supporting the interpretation of a common mode of action. In
contrast, the small number of differentially expressed genes
and affected pathways in the testis indicated that the observed
reproductive effects were not due to modulations of gene
expression within the testis, and that the testis was not a target
organ for triazole reproductive toxicity.
Common metabolic pathways for all three triazoles included
androgen and estrogen, arachidonic acid, fatty acid, glycerolipid,
linoleic acid, tryptophan, and xenobiotic. Several of the
Previous
TOC
-------
TRIAZOLE TOXICOGENOMICS
459
perturbed pathways common to the triazoles formed a large
interconnected network between glycolysis and fatty acid
catabolism, sterol biosynthesis and bile acid biosynthesis or
steroid metabolism. All three triazoles had a significant impact
on lipid metabolism, including fatty acid and steroid metab-
olism, as well as lipid transport. The genomic data in this study
demonstrates triazole disruption of pathways of key biological
functions in the liver; including energy homeostasis, biological
membrane fluidity, and CYP and other metabolic activities.
Moreover, these pathways are critical to liver-mediated steroid
homeostasis, and we propose that it is the perturbation of these
critical pathways that leads to the observed reproductive and
hepatic toxicity of the triazoles.
Hepatic Fatty Acid Metabolism
Many phase I, II, and III metabolic genes perturbed by
triazole exposures are regulated by the nuclear receptors PPAR,
CAR, PXR, LXR, FXR, and AhR. Nuclear receptors within the
liver regulate specific and overlapping subsets of genes
(Honkakoski and Negishi, 2000; Wei et al, 2002; Yoshinari
et al., 2008), and respond to a variety of endogenous
metabolites. As an example, fatty acids activate PPAR-a
through a ligand-induced conformational structure change.
Down regulation of Cyp4alO and Cyp4al, and up regulation of
Cyp4al2 indicates changes in fatty acid levels and PPAR-a
modulation by the triazoles (Gonzalez and Shah, 2008).
Triazoles affected multiple fatty acid metabolic genes, such
as Acsl3 and AcslS, which encode enzymes residing in
different subcellular locations and regulating different steps
in the metabolic pathway (Lewin et al., 2001). Overall, there
appeared to be a shift from insulin-stimulated glucose
metabolism and fatty acid synthesis and storage (lipogenesis),
over to fatty acid oxidation—similar to what has been reported
for PPAR-a agonists (Xu et al., 2006). The triazoles also
seemed to modulate LXR regulated genes and pathways, the up
regulation of Alasl, Ces2, and Scdl expression caused by
triazoles indicated oxysterol activation of LXRa (Chu et al.,
2006) and promotion of bile acid biosynthesis and secretion.
Constitutive Androstance Receptor
The most robust nuclear receptor mediated response to
triazoles in the rat liver is the induction of Cyp2b2, which is
regulated by CAR (Wei et al, 2000). Cyp2b2 was the one gene
differentially expressed in all the triazole treatment groups in
the rat liver. Expression levels of Cyp2b, like Cyp4a, are
increased by ketone bodies and fatty acids, and decrease
following mitochondrial p-oxidation due to decreased in-
tercellular fatty acids. Cyp2b2 catalyzes the oxidation of
testosterone, arachidonic acid, lauric acid, and numerous
environmental agents. Of the five Cyp2b isoforms assessed
by microarray (2b2, 2b3, 2bl3, 2bl5) or PCR (2bl and 2b2) in
the present study, only Cyp2b2 was clearly and strongly
induced by all three triazoles. The biological functions of
Cyp2b3 and Cyp2bl3 have not been determined; however
Cyp2b3 is not phenobarbital inducible (Jean et al., 1994). Cyp2b2
and Cyp2bl5 are both induced by phenobarbital and regulated
by CAR, yet only the array probe set designed for Cyp2b2 was
consistently positive.
PCR confirmed that Cyp2b2 was highly induced by the
triazoles, and the sensitivity of Cyp2b2 to xenobiotics, elevated
levels of testosterone and fatty acids suggests multiple
functions across several metabolic pathways. It should be
noted that triazoles may not be metabolized by rat Cyp2bl or
human CYP2B6 (Barton et al, 2006). However, Barton et al
did not test rat Cyp2b2 metabolism of triazoles to determine if
it might be directly involved in triazole biotransformation. It is
likely that Cyp2b2 ability to metabolize androgens and its
impact on related metabolic pathways in the liver is more
critical to understanding triazole reproductive and hepatic
toxicity.
CAR regulates multiple metabolic enzyme and transporter
genes modulated by triazoles, including Alasl, Cyplal,
Cyp2b2, Lss, Abcc3, Slcola4, PCX, and Ugtlal. Results for
these and other genes are consistent with CAR activation by
triazoles, demonstrating the multifunctionality of this receptor
and a direct or indirect responsiveness to triazoles. Other CAR
activators, like triazoles, also induce hepatomegaly, hepatocyte
hypertrophy, and induction of CYP and other xenobiotic
metabolizing enzymes in rodent liver, In the case of this
triazole study in rats, repeated exposures to triazoles also led to
disruption of steroid homeostasis and infertility. Chronic, high-
dose exposures to these and other triazoles can also lead to
hepatic tumors and carcinogenesis in rodents, similar to what
has been reported for other CAR activators (Huang et al,
2005). The mode of action behind these CAR mediated tumor
and cancer outcomes appears to be increased cell proliferation
and suppression of apoptosis (Huang et al, 2005). Up
regulation of cell growth and antiapoptotic genes, as well as
well established CAR regulated genes such as Cyp2b2,
following the triazole treatments in this study suggest
activation of rat CAR is a key event in the observed
hepatomegaly and other hepatotoxicity reported (Goetz et al,
2007). Using CAR knockout mice, Yamamoto et al. (2004)
have demonstrated that CAR is essential for at least some cases
of mouse hepatotoxicity and tumor formation, and additional
studies have proven these CAR-dependent mechanisms
relevant to at least cyproconazole (Peffer et al, 2007). It is
worth noting that the antiapoptotic gene Gadd45b, which was
up regulated by triadimefon in the present study, has recently
been reported as a CAR coactivator (Yamamoto and Negishi,
2008). It appears that triazole modulation of Gadd45b is
dependent on CAR (Peffer et al, 2007).
Pregnane X Receptor
Activation of genes regulated by PXR was not as robust as
the effects on CAR genes, but it was another common effect of
triazoles in rat liver. Cyp3al and Cyp3a2, Aldhlal, as well as
Previous
-------
460
GOETZ AND DIX
some of the genes coregulated by CAR or other nuclear
receptors were induced by the triazoles. Rat Cyp3al and
Cyp3a2 both metabolize myclobutanil, and perhaps triadime-
fon also (Barton et al, 2006). Human CYP3A4 appears to do
the same, so this PXR mediated induction of Cyp3ais likely
enhancing triazole biotransformation. However, because of the
many ligands shared by PXR and CAR (Moore et al., 2000),
overlapping regulation of hepatic metabolism by these
receptors (Maglich et al., 2002; Tien and Negishi, 2006), and
the promiscuity of PXR binding to many xenobiotics (Orans
et al., 2005), it was difficult to determine the specificity and
significance of PXR in the liver following triazole exposure.
PXR regulates numerous metabolic pathways and genes,
including several altered by triazoles in the present study.
Results with transgenic mice have provided evidence that PXR
response to xenobiotics is an important determinant in both
rodent and human hepatocarcinogenesis (Ma et al., 2007).
Future studies will need to clearly distinguish the functions of
PXR from those of CAR and other nuclear receptors in
regulating genes and pathways. These studies will be crucial in
defining triazole modes of action relative to hepatic and
reproductive toxicity, and determining the relevance of these
mechanistic insights to human health risk.
Hepatic Steroid Metabolism
Many of the genes in the steroid metabolism pathway
perturbed by triazoles in the present study are regulated by
CAR and PXR. This genomic response indicated an attempt by
the rat liver to respond to increased serum testosterone levels
following long term and relatively high-dose exposures to
triazoles. Many of these triazole-induced changes are likely to
have altered metabolism and excretion of steroids and xeno-
biotics. As an example, changes in Hsdl7b were likely to be an
adaptive response to the elevated circulating testosterone seen
with all three triazoles; part of a hepatic attempt to regain
steroid homeostasis (Mustonen et al., 1997). In contrast, down
regulation of SrdSal, which helps eliminate excess androgens
and is typically positively regulated by testosterone and
dihydrotestosterone (Torres and Ortega, 2003), appears to be
a maladaptative response to triazole exposure.
Conclusions
The molecular events measured in this toxicogenomic
study demonstrate that myclobutanil, propiconazole, and
triadimefon perturb common biological pathways, many of
which are regulated by nuclear receptors. By defining the
common changes in gene expression and their associated
biological pathways, strong inferences can be made about the
causative factors leading up to a disruption in testosterone
homeostasis and associated reproductive toxicity of triazoles:
triazoles increased fatty acid catabolism, reduced bile acid
biosynthesis, induced cholesterol biosynthesis, and impaired
steroid metabolism. The induction of CAR and PXR nuclear
receptors drove many of these changes in gene expression,
and subsequent changes in fatty acid, steroid and xenobiotic
metabolism. It is likely that modulation of CAR and PXR by
the triazoles in this long term exposure study led to the
observed hepatomegaly.
The observed disruption in testosterone homeostasis by
triazoles was not due to modulation of steroidogenic genes in
the testis. Instead, there appeared to be disruption of normal
hepatic testosterone metabolism, leading to increased expres-
sion of genes in the steroid metabolism and sterol biosynthesis
pathways as an adaptive response. For reasons not currently
understood, negative feedback mechanisms in the hypothala-
mus-pituitary-gonadal axis did not compensate for these
increases in circulating steroids, and changes in hepatic
metabolism were not able to maintain steroid homeostasis. In
this rat model where exposure started gestationally and
continued to adulthood, disruption of systemic steroid
homeostasis was accompanied by reproductive toxicity and
infertility. The gene expression profiles in this study have
provided strong support for a mode of action for reproductive
and hepatic toxicity, mediated through the CAR and PXR
signaling pathways that is common to the triazole antifungals.
FUNDING
EPA/North Carolina State University Cooperative Training
Agreement (#CT826512010) supported A.K.G.
ACKNOWLEDGMENTS
We thank Drs Hongzu Ren and Indira Thillainadarajah
(EPA) for excellent technical support. We also thank
Dr Douglas Wolf (EPA) for technical review of this manuscript.
Microarrays and reagents for a portion of this study were
provided by Affymetrix as part of a Materials Cooperative
Research and Development Agreement with EPA.
REFERENCES
Allen, J. W., Wolf, D. C., George, M. H., Hester, S. D., Sun, G., Thai, S.-F.,
Delker, D. A., Moore, T., Jones, C., Nelson, G., et al. (2006). Toxicity
profiles in mice treated with hepatotumorigenic and non-hepatotumorigenic
triazole conazoles fungicides: Propiconazole, triadimefon, and myclobutanil.
Toxicol. Pathol. 34, 853-862.
Baldan, A., Tarr, P., Lee, R., and Edwards, P. A. (2006). ATP-binding cassette
transporter Gl and lipid homeostasis. Curr. Opin. Lipidol. 17, 227-232.
Barton, H. A., Tang, J., Sey, Y. M., Stanko, J. P., Murrell, R. N, Rockett, J. C.,
and Dix, D. J. (2006). Metabolism of myclobutanil and triadimefon by
human and rat cytochrome P450 enzymes and liver microsomes. Xenobiotica
36, 793-806.
Chen, P.-J., Padgett, W. T., Moore, T., Winnik, W., Lambert, G. R., Thai, S.-F.,
Hester, S. D., and Nesnow, S. (2009). Three conazoles increase hepatic
Previous
TOC
-------
TRIAZOLE TOXICOGENOMICS
461
microsomal retinoic acid metabolism and decrease mouse hepatic retinoic
acid levels in vivo. Toxicol. Appl. Pharmacol. 234, 143-155.
Chu, K., Miyazaki, M., Man, W. C., and Ntambi, J. M. (2006). Stearoyl-
coenzyme A desaturase 1 deficiency protects against hypertriglyceridemia
and increases plasma high-density lipoprotein cholesterol induced by liver X
receptor activation. Mol Cell Biol. 26, 6786-6798.
Coleman, R. S., Lewin, T. M., and Muoio, D. M. (2000). Physiological and
nutritional regulation of enzymes of triacylglycerol synthesis. Annu. Rev.
Nutr. 20, 77-103.
Dixit, S. G., Tirana, R. G., and Kim, R. B. (2005). Beyond CAR and PXR.
Curr. Drug Metab. 6, 385-397.
Ghannoum, M. A., and Rice, L. B. (1999). Antifungal agents: Mode of action,
mechanisms of resistance, and correlation of these mechanisms with bacterial
resistance. Clin. Microbiol. Rev. 12, 501-517.
Goetz, A. K., Bao, W., Ren, H., Schmid, J. E., Tully, D. B., Wood, C.,
Rockett, J. C., Narotsky, M. G., Sun, G., Lambert, G. R., etal. (2006). Gene
expression profiling in the liver of CD-I mice to characterize the
hepatotoxicity of triazole fungicides. Toxicol. Appl. Pharmacol. 215,
274-284.
Goetz, A. K., Ren, H., Schmid, J. E., Blystone, C. R., Thillainadarajah, L,
Best, D. S., Nichols, H. P., Strader, L. F., Wolf, D. C., Narotsky, M. G., etal.
(2007). Disruption of testosterone homeostasis as a mode of action for the
reproductive toxicity of triazole fungicides in the male rat. Toxicol. Sci. 95,
227-239.
Gonzalez, F. J., and Shah, Y. M. (2008). PPARa: Mechanism of species
differences and hepatocarcinogenesis of peroxisome proliferators. Toxicol-
ogy 246, 2-8.
Guzelian, J., Barwick, J. L., Hunter, L., Phang, T. L., Quattrochi, L. C., and
Guxelian, P. S. (2006). Identification of genes controlled by the pregnane X
receptor by microarray analysis of mRNAs from pregnenolone 16alpha-
carbonitrole-treated rats. Toxicol. Sci. 94, 379-387.
Hester, S. D., and Nesnow, S. (2008). Transcriptional responses in thyroid
tissues from rats treated with a tumorigenic and a non-tumorigenic conazoles
fungicide. Toxicol. Appl. Pharmacol. 227, 357-369.
Hester, S. D., Wolf, D. C., Nesnow, S., and Thai, S. F. (2006). Transcriptional
profiles in liver from rats treated with tumorigenic and non-tumorigenic
triazoles conazole fungicides: Propiconazole, triadimefon, and myclobutanil.
Toxicol. Pathol. 34, 879-894.
Honkakoski, P., and Negishi, M. (2000). Regulation of cytochrome P450
(CYP) genes by nuclear receptors. Biochem. J. 347, 321-337.
Huang, W., Zhang, J., Washington, M., Liu, J., Parant, J. M., Lozano, G., and
Moore, D. D. (2005). Xenobiotic stress induces hepatomegaly and liver
tumors via the nuclear receptor constitutive androstane receptor. Mol.
Endocrinol. 19, 1646-1653.
Jean, A., Reiss, A., Desrochers, M., Dubois, S., Trottier, E., Trottier, Y.,
Wirtanen, L., Adesnik, M., Waxman, D. J., and Anderson, A. (1994). Rat
liver cytochrome P450 2B3: Structure of the CYP2B3 gene and
immunological identification of a constitutive P450 2B3-like protein in rat
liver. DNA Cell Biol. 13, 781-792.
Jigorel, E., Le Vee, M., Boursier-Neyret, C., Bertrand, M., and Fardel, O.
(2005). Functional expression of sinusoidal drug transporters in primary
human and rat hepatocytes. Drug Metab. Dispos. 33, 1418-1422.
Jigorel, E., Le Vee, M., Boursier-Neyret, C., Bertrand, M., and Fardel, O.
(2006). Differential regulation of sinusoidal and canalicular hepatic drug
transporter expression by xenobiotics activating drug-sensing receptors in
primary human hepatocytes. Drug Metab. Dispos. 34, 1756-1763.
Kretschmer, X. C., and Baldwin, W. S. (2005). CAR and PXR: Xenosensors of
endocrine disrupters? Chem. Biol. Interact. 155, 111-128.
Lewin, T. M., Kim, J. H., Granger, D. A., Vance, J. E., and Coleman, R. A.
(2001). Acyl-CoA synthetase isoforms 1,4, and 5 are present in different
subcellular membranes in rat liver and can be inhibited independently.
/. Biol. Chem. 276, 24674-24679.
Ma, X., Shah, Y., Cheung, C., Guo, G. L., Feigenbaum, L., Krausz, K. W.,
Idle, J. R., and Gonzalez, F. J. (2007). The pregnane X receptor gene-
humanized mouse: A model for investigating drug-drug-interactions
mediated by cytochromes P450 3A. Drug Metab. Dispos. 35, 194-200.
Maglich, J. M., Stoltz, C. M., Goodwin, B., Hawkins-Brown, D., Moore, J. T.,
and Kliewer, S. A. (2002). Nuclear pregnane X receptor and constitutive
androstane receptor regulate overlapping but distinct sets of genes involved
in xenobiotic detoxification. Mol. Pharmacol. 62, 638-646.
Moore, L. B., Parks, D. J., Jones, S. A., Bledsoe, R. K., Consler, T. G.,
Stimmel, J. B., Goodwin, B., Liddle, C., Blanchard, S. G., Willson, T. M.,
et al. (2000). Orphan nuclear receptors constitutive androstane receptor and
pregnane X receptor share xenobiotic and steroid ligands. /. Biol. Chem. 275,
15122-15127.
Mustonen, M. V. J., Poutanen, M. H., Isomaa, V. V., and Vihko, R. K. (1997).
Cloning of mouse l?p-hydroxysteroid dehydrogenase type 2, and analyzing
expression of the mRNAs for types 1, 2, 3, 4 and 5 in mouse embryos and
adult tissues. /. Biochem. 325, 199-205.
Nordlie, R. C., and Foster, J. D. (1999). Regulation of glucose production by
the liver. Annu. Rev. Nutr. 19, 379^06.
Orans, J., Teotico, D. G., and Redinbo, M. R. (2005). The nuclear xenobiotic
receptor pregnane X receptor: Recent insights and new challenges. Mol.
Endocrinol. 19, 2891-2900.
Pandak, W. M., Hylemon, P. B., Ren, S., Marques, D., Gil, G., Redford, K.,
Mallonee, D., and Valhcevic, Z. R. (2002). Regulation of oxysterol 7alpha-
hydroxylase (CYP7B1) in primary cultures of rat hepatocytes. Hepatology
35, 1400-1408.
Peffer, R. C., Moggs, J. G., Pastoor, T., Currie, R. A., Wright, J., Milburn, G.,
Waechter, F., and Rusyn, I. (2007). Mouse liver effects of cyproconazole,
a triazole fungicide: Role of the constitutive androstane receptor. Toxicol.
Sci. 99,315-325.
Russell, D. W. (1999). Nuclear orphan receptors control cholesterol catabolism.
Cell 97, 539-542.
Schwarz, M., Russell, D. W., Dietschy, J. M., and Turley, S. D. (2001).
Alternate pathways of bile acid synthesis in the cholesterol 7alpha-
hydroxylase knockout mouse are not upregulated by either cholesterol or
cholestyramine feeding. /. Lipid Res. 42, 1594-1603.
Shenoy, S. D., Spencer, T. A., Mercer-Haines, N. A., Alipour, M.,
Gargano, M. D., Runge-Morris, M., and Kocarek, T. A. (2004). CYP3A
Induction by liver X receptor ligands in primary cultured rat and mouse
hepatocytes is mediated by the pregnane X receptor. Drug Metab. Dispos.
32, 66-71.
Shibata, N., Arita, M., Misaki, Y., Dohmae, N., Takio, K., Ono, T., Inoue, K.,
and Arai, H. (2001). Supernatant protein factor, which stimulates the
conversion of squalene to lanosterol, is a cytosolic squalene transfer protein
and enhances cholesterol biosynthesis. Proc. Nad. Acad. Sci. U. S. A. 98,
2244-2249.
Staudinger, J., Liu, Y., Madan, A., Habeebu, S., and Klaassen, C. D. (2001).
Coordinate regulation of xenobiotic and bile acid homeostasis by pregnane X
receptor. Drug Metab. Dispos. 29, 1467-1472.
Sul, H. S., and Wang, D. (1998). Nutritional and hormonal regulation of
enzymes in fat synthesis: Studies of fatty acid synthase and mitochondrial
glycerol-3-phosphate acyltransferase gene transcription. Annu. Rev. Nutr. 18,
331-351.
Sun, G., Grindstaff, R. D., Thai, S. F., Lambert, G. R., Tully, D. B., Dix, D. J.,
and Nesnow, S. (2007). Induction of cytochrome P450 enzymes in rat liver
by two conazoles, myclobutanil and triadimefon. Xenobiotica 37, 180-193.
Tansey, T. R., and Shechter, I. (2000). Structure and regulation of mammalian
squalene synthase. Biochim. Biophys. Acta 1529, 49-62.
Previous
TOC
-------
462
GOETZ AND DIX
Tien, E. S., and Negishi, M. (2006). Nuclear receptors CAR and PXR in the
regulation of hepatic metabolism. Xenobiotica 36, 1152-1163.
Torres, J. M., and Ortega, E. (2003). Precise quantitation of 5a-reductase type 1
mRNA by RT-PCR in rat liver and its positive regulation by testosterone and
dihydrotestosterone. Biochem. Biophys. Res. Commun. 308, 469-473.
Tully, D. B., Bao, W., Goetz, A. K., Blystone, C. R., Ren, H., Schmid, J. E.,
Strader, L. F., Wood, C. R., Best, D. S., Narotsky, M. G., et al. (2006). Gene
expression profiling in liver and testis of rats to characterize the toxicity of
triazole fungicides. Toxicol. Appl. Pharmacol. 215, 260-273.
U.S. Environmental Protection Agency, Office of Prevention, Pesticides and
Toxic Substances. (1995). Myclobutanil; Pesticide tolerances. Fed. Regist.
60, 40500-40503.
U.S. Environmental Protection Agency, Office of Prevention, Pesticides and
Toxic Substances. (1996). Carcinogenicity Peer Review of Bayleton. U.S.
EPA, Washington, DC.
U.S. Environmental Protection Agency, Office of Prevention, Pesticides and
Toxic Substances. (2001). Myclobutanil, pesticide tolerance for emergency
exemptions. Fed. Regist. 66, 298-306.
U.S. Environmental Protection Agency, Office of Prevention, Pesticides and
Toxic Substances. (2005a). Myclobutanil; pesticide tolerances for emergency
exemptions. Fed. Regist. 70, 49499^9507.
U.S. Environmental Protection Agency, Office of Prevention, Pesticides and
Toxic Substances. (2005b). Propiconazole; pesticide tolerances for emer-
gency exemptions. Fed. Regist. 70, 43284^-3292.
U.S. Environmental Protection Agency, Office of Prevention, Pesticides and
Toxic Substances. (2005c). Memorandum. Triadimefon: Occupational and
Residential Exposure Assessment for the Registration Eligibility Decision
Document. PC Code 109901, DP Barcode 314814, 315040. Office of
Prevention, Pesticides, and Toxic Substances, Washington, DC.
U.S. Environmental Protection Agency, Office of Prevention, Pesticides and
Toxic Substances. (2006). Triadimefon. Preliminary Human Health Risk
Assessment revised. Office of Prevention, Pesticides, and Toxic Substances,
Washington, DC.
Vanden Bossche, H., Marichal, P., Gorrens, J., and Coene, M.-C. (1990).
Biochemical basis for the activity and selectivity of oral antifungal drugs.
Br. J. Clin. Pract. Suppl. 71, 41^6.
Wang, X. J., Chamberlain, M., Vassieva, O., Henderson, C. J., and Wolf, C. R.
(2005). Relationship between hepatic phenotype and changes in gene
expression in cytochrome P450 reductase (FOR) null mice. Biochem. J. 388,
857-867.
Ward, W. O., Delker, D. A., Hester, S. D., Thai, S.-F., Wolf, D. C.,
Allen, J. W., and Nesnow, S. (2006). Transcriptional profiles in liver from
mice treated with hepatotumorigenic and nonhepatotumorigenic triazole
conazoles fungicides: Propiconazole, triadimefon, and myclobutanil.
Toxicol. Pathol. 34, 863-878.
Wei, P., Zhang, J., Egan-Hafley, M., Liang, S., and Moore, D. D. (2000). The
nuclear receptor CAR mediates specific xenobiotic induction of drug
metabolism. Nature 407, 920-923.
Wei, P., Zhang, J., Dowhan, D. H., Han, Y., and Moore, D. D. (2002). Specific
and overlapping functions of the nuclear hormone receptors CAR and PXR
in xenobiotic response. Pharmacogenomics J. 2, 117-126.
Wolf, D. C., Allen, J. W., George, M. H., Hester, S. D., Sun, G., Moore, T.,
Thai, S. F., Delker, D., Winkfield, E., Leavitt, S., et al. (2006). Toxicity
profiles in rats treated with tumorigenic and nontumorigenic triazole
conazole fungicides: Propiconazole, triadimefon, and myclobutanil. Toxicol.
Pathol. 34, 895-902.
Xu, J., Christian, B., and Jump, D. B. (2006). Regulation of rat hepatic
L-pyruvate kinase promoter composition and activity by glucose, n-3
polyunsaturated fatty acids, and peroxisome proliferator-activated receptor-a
agonist. /. Biol. Chem. 281, 18351-18362.
Yamamoto, Y., Moore, R., Goldsworthy, T. L., Negishi, M., and
Maronpot, R. R. (2004). The orphan nuclear receptor constitutive active/
androstane receptor is essential for liver tumor promotion by phenobarbital in
mice. Cancer Res. 64, 7197-7200.
Yamamoto, Y., and Negishi, M. (2008). The antiapoptotic factor growth arrest
and DNA-damage-inducible 45 beta regulates the nuclear receptor
constitutive active/androstane receptor-mediated transcription. Drug Metab.
Dispos. 36, 1189-1193.
Yoshikawa, T., Ide, T., Shimano, H., Yahagi, N, Amemiya-Kudo, M.,
Matsuzaka, T., Yatoh, S., Kitamine, T., Okazaki, H., Tamura, Y., et al.
(2003). Cross-talk between peroxisome proliferator-activated receptor
(PPAR) a and liver x receptor (LXR) in nutritional regulation of fatty acid
metabolism. I. PPARs suppress sterol regulatory element binding protein-lc
promoter through inhibition of LXR signaling. Mol. Endocrinol. 17,
1240-1254.
Yoshinari, K., Tien, E., Negishi, M., and Honkakoski, P. (2008). Receptor-
mediated regulation of cytochromes P450. In Cytochromes P450: Role in the
Metabolism and Toxicity of Drugs and Other Xenobiotics (C. loannides,
Ed.), pp. 417^48. RSC Publishing, Cambridge.
You, L. (2004). Steroid hormone biotransformation and xenobiotic induction of
hepatic steroid metabolizing enzymes. Chem. Biol. Interact. 147, 233-246.
Previous
TOC
-------
TOXICOLOGICAL SCIENCES 107(2), 331-341 (2009)
doi:10.1093/toxsci/kfh234
Advance Access publication November 12, 2008
Modeling Single and Repeated Dose Pharmacokinetics of PFOA in Mice
Inchio Lou,* John F. Wambaugh,* Christopher Lau,t'! Roger G. Hanson,t Andrew B. Lindstrom,^ Mark J. Strynar,^
R. Dan Zehr,t R. Woodrow Setzer,* and Hugh A. Barton*'2
*National Center for Computational Toxicology; ^Reproductive Toxicology Division, National Health and Environmental Effects Research Laboratory; and
$Human Exposure and Atmospheric Science Division, National Exposure Research Laboratory, Office of Research and Development, U.S. Environmental
Protection Agency, Research Triangle Park, North Carolina 27711
Received July 28, 2008; accepted October 31, 2008
Perfluorooctanoic acid (PFOA) displays complicated pharma-
cokinetics in that serum concentrations indicate long half-lives
despite which steady state appears to be achieved rapidly. In this
study, serum and tissue concentration time-courses were obtained
for male and female CD1 mice after single, oral doses of 1 and
10 mg/kg of PFOA. When using one- and two-compartment models,
the pharmacokinetics for these two dosages are not consistent
with serum time-course data from female CD1 mice administered
60 mg/kg, or with serum concentrations following repeated daily
doses of 20 mg/kg PFOA. Some consistency between dose regimens
could be achieved using the saturable resorption model of Andersen
et al. In this model PFOA is cleared from the serum into a filtrate
compartment from which it is either excreted or resorbed
into the serum by a process presumed transporter mediated with
a Michaelis-Menten form. Maximum likelihood estimation found
a transport maximum of Tm = 860.9 (1298.3) mg/l/h and half-
maximum concentration of KT = 0.0015 (0.0022) mg/1 where the
estimated standard errors (in parentheses) indicated large un-
certainty. The estimated rate of flow into and out of the filtrate
compartment, 0.6830 (1.0131) 1/h was too large to be consistent with
a biological interpretation. For these model parameters a single dose
greater than 40 mg/kg, or a daily dose in excess of 5 mg/kg were
necessary to observe nonlinear pharmacokinetics for PFOA in
female CD1 mice. These data and modeling analyses more fully
characterize PFOA in mice for purposes of estimating internal
exposure for use in risk assessment.
Key Words: perfluorooctanoic acid (PFOA); compartment
model; resorption model; pharmacokinetic parameters; statistical
analysis; CD1 mice.
Perfluorooctanoic acid (PFOA) and related compounds are
used primarily as surface-active agents in the production of
various fluoropolymers and fluoroelastomers (Kudo and
Kawashima, 2003). Because of the strength of the carbon-
fluorine bond, PFOA is stable to metabolic and environmental
1 To whom correspondence should be addressed at Mail Drop 67, U.S.
Environmental Protection Agency, Research Triangle Park, NC 27711. Fax:
(919) 541-4017. E-mail: lau.christopher@epa.gov.
2 Current address: Pfizer, Inc., PDM PK/PD Modeling, Eastern Point Rd.,
MS 8220-4328, Groton, CT 06340.
Published by Oxford University Press 2008.
degradation (Butenhoff et al., 2004). PFOA is widespread in
wildlife and humans—from polar bears living in Greenland
(Dietz et al, 2008), to giant pandas in China (Dai et al, 2006),
from the general population to occupationally exposed workers
(Betts, 2007; Olsen et al, 2007). Average blood levels from the
general population in the United States are approximately 4-5
parts per billion (Calafat et al, 2007).
The toxicology of PFOA has been extensively reviewed
(Andersen et al, 2008; Lau et al, 2007; Kennedy et al, 2004).
PFOA is associated with liver enlargement in rodents and
nonhuman primates. Hepatocellular adenomas, Ley dig cell
tumors, and pancreatic acinar cell tumors occurred in rats
(Biegel et al, 2001; Cook et al, 1992). Exposure to a high
dose of PFOA (20 mg/kg) for two days late in gestation was
sufficient to produce neonatal mortality and birth weight
reduction in mice (Wolf et al, 2007). Further investigations
showed the daily PFOA treatment with 5 mg/kg and lower
doses during gestation was associated with effects (White et al,
2007; Abbott et al, 2007).
PFOA is found in human blood and breast milk from the
general population in countries worldwide (Butenhoff et al,
2004). Workers occupationally exposed to fluorochemicals
have serum levels of PFOA approximately one order of
magnitude higher than those reported in the general
population. The PFOA serum elimination half-life in work-
ers was estimated as 3.8 years (Olsen et al, 2007). This
is much longer than in laboratory animals, for example,
hours for the female rat to days for the male rat to weeks for
the monkey (Lau et al, 2007). Gender differences are
particularly notable in rats, with limited differences in other
animals (Kudo and Kawashima, 2003). The basis for the
species and gender differences in elimination of PFOA is
still not well understood PFOA is high bound to plasma
proteins and this does not appear to differ substantially
across species (Kudo and Kawashima, 2003). Differential
expression of transporter proteins in the kidney may be one
explanation and is clearly a major factor in the sex difference
observed in rats (Kudo et al, 2002). Transporter activity has
been confirmed in rat, in which organic anion transporters
1 and 3, organic anion transporting polypeptide 1, and
Previous
TOG
Next
-------
332
LOU ET AL.
perhaps others mediate PFOA cross-membrane transport
(Kudo et al, 2002; Nakagawa et al, 2008). Recent studies
with expressed human and rat organic anion transporters
1 and 2 found similar activity (Nakagawa et al., 2008). In
addition, liver distribution in rats is dose-dependent (Kudo
et al., 2007) and transporter-dependent (Han et al., 2008),
whereas studies in mice have demonstrated that both uptake
and efflux transporters in liver are regulated by PFOA
(Cheng and Klaassen, 2008; Maher et al., 2008). These kinds
of effects presumably underlie the observation of the need to
incorporate time-dependent changes in pharmacokinetic
modeling for PFOA and PFOS (Harris and Barton, 2008;
Tan et al., 2008)
A margin of exposure approach was used in the U.S. EPA's
PFOA preliminary risk assessment, which compared mea-
sured human blood levels with laboratory animal blood levels
associated with toxic effects (U.S. EPA, 2005). The area
under the blood concentration-time curve (AUC), concentra-
tion at steady state (Css), or peak concentration (Cmax) were
dose metrics for evaluating effects in this draft assessment.
The cross-species pharmacokinetic extrapolation using AUC
or Css (e.g., Csshuman/Cssmouse) and a one-compartment
pharmacokinetic model is estimated by the ratio of half-lives
assuming the volume of distribution is a similar fraction of
body weight across species, for example, Css = dose rate
(mg/kg/day)/[volume of distribution (I/kg) X elimination rate
constant (I/day)], where half-life equals In 2/elimination rate
constant. To date, the one-compartment model has been used
for PFOA pharmacokinetic analysis in rat and monkey
(Harada et al., 2005; U.S. EPA, 2005; Washburn et al.,
2005). However, the half-life estimated in humans and
animals may not be constant and animal half-lives estimated
by following blood levels after a single dose may not be
comparable with estimates from humans who have had
chronic exposure. Monkey and rat data have been interpreted
as indicating that the volume of distribution changes with
concentration (Trudel et al., 2008; Washburn et al, 2005).
Large volumes of distribution do not seem likely, however,
because PFOA is known to rapidly achieve quasi-steady-state
in blood (Andersen et al, 2006; Lau et al, 2006).
Alternatively, monkey data suggests that excretion is
concentration dependent (faster elimination rate at higher
concentrations), that is, half-lives are not constant (Andersen
et al, 2006). These issues increase the difficulty in
extrapolating from one species to another. A recently
developed biologically motivated pharmacokinetic model of
saturable, renal resorption that depends on kinetic factors for
transport successfully described the monkey data (Andersen
et al, 2006). The difference in apparent elimination rates with
increasing dose indicated that capacity-limited, saturable
transport processes may be involved in the kinetic behavior
of PFOA.
In this study, one- and two-compartment models with first
order absorption and clearance were statistically analyzed for
PFOA time course data to estimate the pharmacokinetic
parameters—volume of distribution, absorption rate, and
elimination rate—for female and male CD-I mice based
upon PFOA concentrations following single and repeated
doses. The saturable resorption model, which elaborates the
description of elimination, also was applied to investigate the
kinetic behaviors of PFOA in mice. These analyses charac-
terize models for mice that provide initial estimates of
dosimetry that could be applied in risk assessment, though
they may also be considered intermediate steps in the
development of a more complete physiologically based
pharmacokinetic model that would better characterize PFOA
tissue distribution, which is not directly addressed by any of
these current models.
MATERIALS AND METHODS
PFOA (ammonium salt; >98% pure) was purchased from Fluka Chemical
(Steinheim, Switzerland). Nuclear magnetic resonance analysis kindly provided
by 3M Company (St Paul, MN) indicated that approximately 98.9% of the
chemical was straight-chain and the remaining 1.1% was branched isomers.
[1,2-13C]-PFOA was purchased from Perkin-Elmer (Wellesley, MA) and used
as an internal standard in the quantitative analysis. For all studies, PFOA dosing
solutions were dissolved in deionized water and prepared fresh daily.
Complete data tables are available as an online supplement.
Animal treatment. All animal studies were conducted in accordance with
the Institutional Animal Care and Use Committee guidelines established by
the U.S. Environmental Protection Agency's Office of Research and
Development/National Health and Environmental Effects Research Labora-
tory. Procedures and facilities were consistent with the recommendations
of the 1996 National Research Council's "Guide for the Care and Use of
Laboratory Animals," the Animal Welfare Act, and Public Health Service
Policy on the Humane Care and Use of Laboratory Animals. Animal facilities
were controlled for temperature (20-24°C) and relative humidity (40-60%)
and kept under a 12-h light-dark cycle. Mature male and female CD-I mice
(70-80 days of age) were purchased from Charles River Laboratories
(Raleigh, NC) and shipped by truck to our facilities, with a transit time of
less than one hour. Animals were segregated by sex, housed in polypropylene
cages (three per cage), and provided pellet chow (LabDiet 5001, PMI
Nutrition International, St. Louis, MO) and tap water ad libitum. Mice were
allowed several days for acclimation and randomly assigned to treatment
groups. Several studies were undertaken involving single or repeated
dosing. Two studies with very similar designs were carried out in which
mice were given a single oral gavage treatment of either 1 mg/kg or 10 mg/kg
PFOA. In the first study (PK1), three males and three females from each dose
group were sacrificed by decapitation at the following time intervals: 4, 8, or
12h, and 1,3,6,9,13,20,27,34,42, or48 days. Trunk blood was collected for
serum preparation and stored at —20°C; liver and kidney were dissected, flash-
frozen on dry-ice and stored at —80°C until being processed for analysis. For
the second study (PK2), the evaluation time points were extended to include
55, 62,70, and 80 days. Serum, liver and kidneys were collected and stored as
described previously. Based upon initial modeling efforts, a study at a higher
dose was carried out in which female mice were given 60 mg/kg PFOA (6 mg/
ml dosing solution, 10 ml/kg dosing volume) and three mice were sacrificed at
each of the following time intervals: 2,4,6, 8,12,24,36 h, or 2,4,6, 8,11,14,
21 days, 28 days. Only serum was analyzed for these animals. Finally,
a repeated dose study was carried out in which five animals received 20 mg/kg/
day for 17 days and serum was obtained 24 h after the final dose as previously
described (Lau et al, 2006).
Previous
TOG
Next
-------
SINGLE AND REPEATED DOSE PHARMACOKINETICS OF PFOA
333
PFOA determination. Serum samples were thawed and mixed well by
vortexing; an aliquot (25-100 ul) was removed for analysis. The volume of
serum assayed was varied to optimize detection of PFOA because levels were
very high at early time points and very low at the latest. Liver and kidney
were thawed, weighed and homogenized (polytron) in 5 volumes of
deionized, distilled water. Analysis of PFOA in serum and tissues was
performed using a modification of a method originally developed by Hansen
et al. (2001). Briefly, 25-100 ul of serum or 25 ul of tissue homogenate was
combined with 1 ml of 0.5M tetrabutylammonium hydrogen sulfate (pH 10)
and 2 ml of 0.25M sodium carbonate and then vortexed for 20 min in a 15 ml
of polypropylene tube. Three hundred microliters of this mixture was then
transferred to a fresh 15 ml of polypropylene tube and 25 ul of a 1 ng/ul
solution of 13C-PFOA was added as an internal standard. Five milliliters of
methyl tert-butyl ether (MTBE) was then added and vortexed again for
20 min. The sample was centrifuged at 2000 X g for 3 min to separate the
aqueous and organic phases, and 1 ml of the MTBE layer was transferred
to a 5-ml polypropylene tube where it was evaporated to dryness at 45°C
under a gentle stream of dry nitrogen. The residue was then solubilized in
400 ul of a 2mM ammonium acetate/acetonitrile (1:1 by vol) solution and
transferred to a polypropylene autosampler vial. No pH adjustments were
made for this solution. Extracts were analyzed using an Agilent 1100 high-
performance liquid chromatograph (Agilent Technology, Palo Alto, CA)
coupled with an API 3000 triple quadrupole mass spectrometer (Applied
Biosystems, Foster City, CA) (LC/MS/MS). Ten microliters of the extract was
injected onto a Luna C18(2) 3 X 50 mm, 5-um column (Phenomenex,
Torrance, CA) using an isocratic mobile phase consisting of 30% 2mM
ammonium acetate solution and 70% acetonitrile at a flow rate of 200 ul/min.
PFOA and 13C2-PFOA were monitored using parent and daughter ion
transitions of 413 —> 369 and 415 —> 370, respectively. Peak integrations and
areas were determined using Analyst software (Applied Biosystems Version
1.4.2, Foster City, CA). For each analytical batch, matrix-matched calibration
curves were prepared as described above using mouse serum spiked with
varying levels of PFOA. For quality control, check standards were prepared
by spiking large volumes of mouse serum at several arbitrary levels. These
check standards were stored frozen and aliquots analyzed with each analytical
set. Different preparations of standards were used for each experimental
study; the concentration of each newly prepared standard was compared with
the previous batch to ensure consistency. In addition, control mouse serum
samples were fortified at two or three levels in duplicate with known
quantities of PFOA during the preparation of each analytical set. Duplicate
fortified and several check standards were run in each analytical batch to
assess precision and accuracy. The limit of quantification (LOQ) was set as
the lowest calibration point on the standard curve. Analytical batches were
considered to be acceptable if: matrix and reagent blanks had no significant
PFOA peaks approaching the LOQ, the standard curve had a correlation
coefficient > 0.98, and all standard curve points, fortified, and check samples
were within 70-130% of the theoretical and previously determined values,
respectively.
One- and two-compartment pharmacokinetic analysis of 1 and 10 mg/kg
data. The PFOA single oral dose time course data at the two lower doses
included three tissues (blood sera, liver and kidney), two genders (female and
male mice), and two doses (1 and 10 mg/kg) collected in two experimental
blocks (PK1 and PK2). Thus, there are 24 data sets in all. We estimated
parameters using R (version 2.4.1, R Development Core Team, 2007). One-
and two-compartment models were fit to blood sera, liver, and kidney time-
course data for each gender, dose, and block.
The one- and two-compartment models with first order absorption and first
order elimination can be described as:
kj>
(>
c(t)=-
kj)
One-compartment model, where D, dose; V&, volume of distribution;
adsorption rate constant; &e, elimination rate constant.
Two-compartment model, central compartment, where D and &a are as above
and Vi, volume of central compartment; fc12, rate constant for transfer from
compartment 1 to compartment 2; k2i, rate constant for transfer from
compartment 2 to compartment 1; a, agglomerate rate constant representing
net loss from the central compartment during the distribution phase;
P, agglomerate rate constant representing net loss from the central compartment
after the distributional phase is complete.
The models were fit by using generalized nonlinear least square (gnls) using
the R function gnls in package nlme (Pinheiro et al., 2007), to estimate the
parameters. The likelihood ratio test was applied to compare the one- and two-
compartment models to determine which model better described the data.
Initially, one-compartment models were fit with a separate parameter value
for each gender, dose-level, and experimental block in each tissue. Conditional
F-tests (Pinheiro and Bates, 2000) on linear combinations of the dose-block-
gender specific parameters were then used to determine the extent to which
each parameter could be simplified. An orthogonal series of contrasts,
analogous to those in multi-factorial analysis of variance, was developed, so
that interaction terms were first tested (in order of decreasing complexity)
followed by main effects terms. That is, we first tested for a given parameter
type (e.g., volume of distribution) whether there was a significant three-way
interaction (dose X block X gender), which was followed by (if the three-way
interaction was not significant) dose X block, dose X gender, block X gender,
and then gender, block and dose. Using the results of these tests, a new
statistical model was constructed by collapsing over the effects that were not
significant. For example, if only gender effects remained significant for volume
of distribution, a new model would be constructed in which volume of
distribution was allowed to vary among genders, but not across blocks or dose
levels. When block was found to be significant, it was incorporated as a random
effect in a nonlinear mixed-effects model, fit using the function nlme.
One- and two-compartment pharmacokinetic analysis of 60 mg/kg
data. Subsequent to the analysis of 1 and 10 mg/kg data, an oral time course
in serum of female mice exposed to 60 mg/kg was collected based upon
preliminary model predictions that the time course should be biphasic. This
data was also evaluated for one- and two-compartment model fits.
The data differ from that collected for 1 and 10 mg/kg doses in having
replicate values for about half the measurements, so a hierarchical statistical
model was fit to the data, with yd (in the one-compartment model) and Vi (in
the two-compartment model) varying among subjects. ks and &a were found to
not be statistically identifiable as individually varying parameters. In this
model, the distribution of the log of the compartment volume is assumed to be
Gaussian, and the population mean and variance are additional parameters to be
estimated. Estimation was via the method of Lindstrom and Bates (1990), as
implemented in the package nlme (Pinheiro et al., 2007) for R (R Development
Core Team, 2007).
Saturable resorption model analysis. Our saturable resorption model was
adapted with minor modifications from Andersen et al. (2006), and
implemented using Matlab (version R2007a, The Mathworks, Natick, MA)
(see Appendix) to simulate and predict the single and repeated oral dose data
for blood sera in female mice. Solutions were obtained using a stiff solver
(ode23s) that implemented the modified Rosenbrock (2,3) pair approach
(Shampine and Reichelt, 1997). All the simulations were run on a computer
equipped with 3-GHz Dual Core Pentium 4 processor and the Windows XP
operating system.
The salient feature of the Andersen et al. (2006) model is that the free
concentration of PFOA in the central compartment (given by free* Ci) is
cleared to a filtrate compartment where it is either excreted or resorbed via
a saturable process with a Michaelis-Menten form. We examined the original
three compartment (two body, one filtrate) model with eight model parameters
Previous
TOG
Next
-------
334
LOU ET AL.
i Second Compartment
! (vb c2)
Oral Dose
(D)
Central Compartment
(Vc, C1; free)
Qfil'
,
Tm,KT
Filtrate Compartment
(Vffl, C3)
FIG. 1. A schematic for the renal saturable resorption pharmacokinetic
model.
as well as a simplified, two-compartment (one body, one filtrate) model with six
parameters (Fig. 1). Oral uptake was assumed to be first order with the same
rate we determined for the one-compartment model.
We determined the model parameters for female mice via Maximum
Likelihood Estimation. Values of the likelihood function were calculated for
eight sets of observations in sera: the two blocks each of 1 mg/kg and 10 mg/kg
single doses, the 60 mg/kg single dose, the 17-day repeated 20 mg/kg/day dose
observations, and the repeated dose data from Lau et al. (2006) for 7 and
17 days also at 20 mg/kg/day. We allowed the coefficient of variation to be
different for each of these eight sets of observations so that our likelihood
function depended upon either six or eight model parameters and eight variance
parameters. The contribution to the likelihood of for each animal was
calculated—for a few animals where replicate measures had been performed,
the results were averaged. We used a Nelder-Mead optimizer (Lagarias et al.,
1998) to find the combination of parameters that maximized the likelihood. We
numerically approximated the second derivative (D'Errico, 2007) of the
likelihood function at the optimized parameter values to obtain the standard
error for each parameter estimate.
RESULTS
Experimental Data
Serum, and in some cases tissue concentrations, of PFOA in
mice were obtained in several studies including time course
data following a single dose at three different dosages and
single time points following two durations of repeated dosing
(Table 1).
The blood sera, liver and kidney concentration time-courses
after single oral doses of 1 and 10 mg/kg are plotted in Figure 2
for the male and female mice. Male and female mice fairly
rapidly absorbed PFOA, as judged by the time of maximum
observed concentration (4 or 8 h). The liver concentrations
were often higher than those in sera, whereas both were
substantially higher than the kidney concentrations. The data
from sera, liver, and kidney and plots for all the male and
female mice dosed with 1 and 10 mg/kg are presented in the
Supplemental Materials.
For single doses of 1 and 10 mg/kg the pharmacokinetics are
essentially linear, as illustrated in Figure 3 in which the female
serum time courses collapse, when scaled by dose, onto
approximately the same line on a semilogarithmic plot. The
pharmacokinetics were quite different at the highest dose, 60
mg/kg, appearing roughly bi-exponential with low concen-
trations, as a fraction of total dose, achieved more rapidly.
Statistical Analysis of Compartmental Pharmacokinetic
Models
Pharmacokinetic data are routinely analyzed using classical
one and two-compartmental models and serum concentration data
(Table 2). To compare clearance and apparent volume of
distribution, as reflected by liver and kidney concentrations, data
for each of these tissues were also fitted using the compartmental
models. For the 1 and 10 mg/kg data, all kinetic parameters were
identified and estimated in the one-compartment model. With the
two-compartment model, it was possible to estimate parameters in
only six of the twenty-four 1 and 10 mg/kg data sets for blood
sera, liver and kidney, due to failures of convergence. These
failures are likely to be related to the inability to uniquely estimate
some of the parameters. We compared the one- and two-
compartment models for estimating the available corresponding
identified parameters using the likelihood ratio test and found that
none of the results were significant (p > 0.05), that is, adding
a second compartment did not significantly improve the ability of
the model to account for the data. Thus, we at first focused on the
one-compartment model for further parameter estimation studies
with the 1 and 10 mg/kg data.
yd and £e differed significantly between males and females
for all three tissues, although the differences generally are not
large (Table 2). Vd differed between doses in kidney. Vd also
varied significantly between the two blocks in sera, which was
therefore included in the estimation of experimental error (all
p < 0.05). The absorption rate constant, £a, was marginally
estimable with these data, so it was estimated as a single value
across all data sets for each tissue.
TABLE 1
Pharmacokinetic Studies of PFOA in Mice
Dose
Sex
Tissues
Sampling
Single dose (PK 1)
Single dose (PK 2)
Single dose
Repeated dose
Repeated dose; Lau et al. (2006)
1, 10 mg/kg
1, 10 mg/kg
60 mg/kg
20 mg/kg, 17 days
20 mg/kg, 7 and 17 days
Male, female
Male, female
Female
Female
Female
Serum, liver, kidney
Serum, liver, kidney
Serum
Serum
Serum
Time course
Time course
Time course
24 h after final dose
24 h after final dose
Previous
TOG
Next
-------
SINGLE AND REPEATED DOSE PHARMACOKINETICS OF PFOA
335
c
o
O
<
O
LL.
CL
10o
o
o
**.. 1
• 1 mg/kg Plasma T 1 mg/kg Liver ^ 1 mg/kg Kidney
0 10 mg/kg Plasma v 10 mg/kg
* 5 $ *
f * 5
f*ft. .
Liver > 10 mg/kg Kidney
•H I 1 ! 1 1 f * * *
ft * • •
0
•
Female
Male
ttf
v
V V
20
40
60
20
40
80 0
Day
FIG. 2. Experimental data plots for different doses, genders, and blocks.
60
80
The mean and 95% confidence intervals for each parameter
are shown in Table 2. Because the initial sampling time point
4 h, is too late to capture the absorption processes, the 95%
confidence intervals for ka for blood sera and liver are quite
wide. We cannot identify £a for kidney from our data. To solve
this problem, we first explored the sensitivities of estimates of
yd and ke to values of £a by estimating Vd and ke while fixing £a
at 0.3, 0.5, 1, and 1.5/h. The estimates and 95% confidence
intervals were insensitive to the particular value chosen for £a,
so £a for kidney was set to the mean (0.527/h) of £a values in
10.0-
"3
|5-°-
0
CO
€ 2.0-
2"
^. 1.0-
o
0. 0.5-
E
'„-
0.1 -
I
\
• 1 mg/kg
»„
> •
o
<><>
) <
y
0 1 0 mg/kg
O 60 mg/kg
>
II
<> ^
> O
O
o
i 1
III V
^ 1 .
0
CD
T
1
<
0
i i
20 40 60 8
Day
FIG. 3. Serum concentrations scaled by dose for females administered single
doses of 1, 10, and 60 mg/kg. Points are means, error bars are 95% confidence
intervals for the means. 1 and 10 mg/kg dose groups are largely superimposed and
linear in time on this semi-log suggesting linear first-order kinetics at these doses.
The 60 mg/kg group has a substantially different shape and time course.
blood sera and liver. As shown in Table 2, female mice have
lower values of Vd in blood sera and kidney, but higher values
in liver, than male mice, hi all the three tissues, ke values are
modestly higher in female mice than in male mice.
The one-compartment model was successful in describing
the 1 and 10 mg/kg single dose serum data sets with estimated
values of Vd (= 0.135 I/kg) and ke (= 0.00185/h) for female
mice. These same two parameter values fail to predict serum
concentrations for both the higher, 60 mg/kg single dose and
repeated dose (20 mg/kg) blood sera data.
The serum concentrations resulting from doses of 10 and
60 mg/kg, converge after roughly a day, indicating that linear
pharmacokinetic models—including both one and two-
compartment models—cannot reproduce the observed phar-
macokinetics. Though the 60 mg/kg single dose data was
well-described by a two-compartment analysis (Table 3), as is
shown in Figures 4a and 4b, predictions made using the two-
compartment parameter estimates from the 60 mg/kg data
were not consistent with the 1 and 10 mg/kg blood sera data.
Jointly analyzing all the available data to optimize the two-
compartment model, also shown in Figure 4, resulted in pre-
dictions that did not reproduce any of the dose regimens.
After 1 and 17 days of dosing female CD1 mice with 20 mg/kg
PFOA, Lau ef al. (2006) found that the measured serum levels were
approximately equal (176 ± 56 vs. 172 ±34 mg/1, respectively). We
use recalculated values based on the original, individual mouse
data that are somewhat different from what was reported by Lau
ef al. (2006). Additional 17 day, 20 mg/kg repeated dose
experiments were performed for five female CD1 mice and
a serum PFOA concentration of 130 ± 23 mg/1 was measured.
The one-compartment model only fit the repeated dose data
if the elimination rate ke was increased from 0.00185/h to
0.025 5/h, that is, the half-life of PFOA in blood sera decreased
from 15 to 1.2 days. This results in the contradiction that differ-
ent kinetics were observed when using the one-compartment
Previous
TOC
-------
336
LOU ET AL.
TABLE 2
One compartment model parameters for 1 and 10 mg/kg doses of PFOA
Female (95% confidence interval)
Male (95% confidence interval)
Blood Sera
Liver
Kidney
Vd (L/kg)
ka (1/h)
ke (1/h)
>t>/2 (day)
Vd (L/kg)
ka (1/h)
ke (1/h)
Vd
-------
SINGLE AND REPEATED DOSE PHARMACOKINETICS OF PFOA
337
O)
E
<
O
LL
0-
E
O
CO
10.00 -
5.00-
2.00-
1.00-
0.50-
0.20-
0.10-
0.05-
1 mg/kg
2 cmpt: all
• • • 2 cmpt: 60 mg/kg
resorption model
I
20
I
40
r
60
r
80
800-
600-
400-
200-
20 mg/kg repeated _
10
12
14
16
200.0
100.0
50.0
20.0
10.0
5.0
2.0-
1.0-
0.5-
200-
100-
50-
20-
20
I
40
r
60
80
60 mg/kg
10
15
20
25
Days
FIG. 4. Comparing predictions for the two-compartment model when fit to all the available data (dashed line) with a fit to just the 60 mg/kg data (dotted line).
Neither model does a good job of describing all of the data, whereas the saturable resorption model (solid line) is more consistent between doses.
(0.6830 1/h) is roughly two thirds of the total output. For
comparison, in monkeys Andersen et al. (2006) assumed that
2fli was 10% of cardiac output to the kidney, corresponding to
gfli = 0.0943 1/h. For a standard body weight of 0.025 kg,
glomerular filtration for female C57BL/6J mice is much
smaller, 0.00945 1/h (Qi et al., 2004), whereas in adult male
CD1 mice the urinary flow rate is even smaller—0.000076 1/h
(Luippold et al., 2002). Thus, our optimized value for Q^\
does not appear to have a direct physiologic analog.
Using our calibrated saturable resorption model, we in-
vestigated what doses of PFOA are predicted to show the two
different kinetic behaviors. For selected dose levels, we
predicted changes in kinetics from low dose to high dose
(Fig. 7). Only one phase was predicted at low doses (< 40 mg/kg),
100
I
10
02 1
0.1
iCU,
O. ..
0.1
• 1 mg/kg
0 10 mg/kg
O 60 mg/kg
10
100
300
250
200
150
100
50
0
10
15
Day
FIG. 5. Predictions and quantiles for the saturable resorption model when optimized using all available data. The predictions for the maximum likelihood
estimated parameters are indicated by a solid line, with open squares indicating where model predictions should be compared with observations for the repeated
dose data. Dashed lines indicate the 95% upper and lower quantiles using the estimated parameter uncertainty. The Lau et al. (2006) 7- and 17-day observations as
well as our new 17-day observations are indicated by solid triangles.
Previous
TOC
-------
338
LOU ET AL.
TABLE 4
Assumed and Optimized PFOA Resorption Model Parameters
Parameter
Value (SE)
Source
Body weight (BW)
Cardiac output
Absorption rate (&a)
Volume of distribution of
the central compartment (Vc)
Volume of renal filtrate (yfil)
Renal blood filtrate rate (<2m)
Volume of distribution of
second body
compartment (Vt)
Intercompartmental
clearance (Qd)
Transport maximum (rm)
Transport affinity
constant (/fT)
Proportion of PFOA
free in serum (free)
25 g
16.5 1/h for mice
0.537 /h
0.0027 (0.0002) 1
0.01 1
0.6830(1.0131) 1/h
0.0545 (0.0151) 1
0.00059 (0.00037) 1/h
860.9 (1298.3) mg/l/h
0.0015 (0.0022) mg/1
0.02
Assumed standard value
Barbee et al (1992)
Estimated from single
dose data using one-
compartment model
Optimized
Assumed,
Andersen et al. (2006)
Optimized
Optimized
Optimized
Optimized
Optimized
Assumed,
Andersen et al. (2006)
100
"a
-§
2 0.01
Q_
£
2
W 0.0001
0.01
0.1
10
100
Day
FIG. 6. Saturable resorption model predictions using parameters obtained
with a maximum likelihood estimate show that the concentration in the filtrate
compartment (dashed) spikes early on allowing the concentration in the primary
compartment (solid) to rapidly converge for doses of 10 and 60 mg/kg.
whereas two phases occurred at the high doses (> 40 mg/kg)
with a fast initial elimination rate giving way to a much slower rate
after roughly one day. Simulations using an earlier version of the
model were the basis for selecting the 60 mg/kg dose, which did
demonstrate biphasic behavior predicted by the saturable re-
sorption model but not previously observed in the serum time
course data with lower doses. For repeated doses, daily doses of
0.01, 0.1, and 1 mg/kg saturated after about two weeks, whereas
for 5 mg/kg the serum concentration quickly saturated. Within
a day of daily doses of 50 and 500 mg/kg, serum concentration
saturated at the same concentration as with 5 mg/kg.
Normalized sensitivity coefficients, defined as (change of
output/output)/(change of input/input), were used to test the
parameter sensitivity at different days after a single dose of
20 mg/kg (Fig. 8). After 1 day, the most sensitive parameters
were Qju, Tm, and KI, that is, the kinetics of the resorption
process, because they dictate clearance, are the most important
for predicting long-term concentrations.
DISCUSSION
For 1 and 10 mg/kg single doses, kinetic parameters differ
significantly between genders but the magnitude of the
differences are small indicating PFOA pharmacokinetic
behaviors are similar in female and male mice in contrast
to rats. The values of the parameter £a were not well
estimated, and the 95% confidence intervals were wide. This
is because the PFOA absorption in mice was fairly rapid, and
the absorption was almost finished before the initial
sampling time point (4 h). Due to the uncertain estimation
of fca values, we used only one fca for female and male mice
for each tissue.
100
< 10
o
E
0.1
0
100
10
- 0.1
Day
''
'-j i
f--f w^i
0
10
20
FIG. 7. Delineation of predictions for the PFOA concentration (mg/1) in the central compartment. For the single dose (top) solid lines depict doses of 0.1, 1,
10, 100, and 1000 mg/kg. The dashed line indicates a dose of 40 mg/kg which is roughly where the onset of nonlinearity occurs. For the repeated dose (bottom)
solid lines depict repeated daily doses of 0.001, 0.1, 1, 50, and 500 mg/kg. The dashed line indicates a daily dose of 5 mg/kg.
Previous
-------
SINGLE AND REPEATED DOSE PHARMACOKINETICS OF PFOA
339
c
o>
'o
^
Qfii
* ka
* Vfil
» free
'*»>,
10
Day
15
20
FIG. 8. Analysis of the parameter sensitivity by increasing each parameter
in turn by 1% and comparing predicted concentrations for a 20 mg/kg single
dose with those for the optimized/assumed value. Note that plot points for
several parameters are on top of each other near zero.
The one-compartment model is successful in describing the
1 and 10 mg/kg single dose data sets with the estimated values
of Vd, ka, and ke, but fails in predicting the higher, 60 mg/kg
single dose and the repeated dose data. Similarly, although the
60 mg/kg data can be described by a two-compartment model,
for the optimized parameters that model overestimates the 1 and
10 mg/kg single dose data. Neither model predicts the repeated
dose observations without changing some model parameters
drastically from the single dose case.
The saturable resorption model of Andersen et al. (2006)
reconciles the lower two doses with the high single dose by
allowing the clearance to change for different exposure levels in
place of the first order or proportional clearance in the previous 1
and 2 compartment models. At low dose or the early period of
repeated dose for our data, the PFOA concentration in the filtrate
compartment is low and is proportional to dose, which has a low
net urine elimination rate, whereas at high dose (including
pseudosteady state of repeated dose for our data), the PFOA
concentration in the filtrate compartment is high and resorption
is saturated, which results in a high net urine elimination rate.
The saturable resorption model does not, however, com-
pletely reconcile the single dose data with the repeated dose
concentrations. Though parameters can be estimated such that
most data is with the 95% confidence intervals, the concen-
trations observed after repeated doses by Lau et al. (2006) seem
to be systematically higher than predicted. This may reflect
experimental variability in light of the repeated dose data
reported here, which was lower than that previously measured.
The modeling analyses presented here can be used to
estimate initial internal dose metrics for toxicity studies carried
out in adult mouse. Characterization of the uncertainties in the
parameter estimates permits some description of the uncertain-
ties in predicted dose metrics. However, none of these models
describe tissue dosimetry, which would require a physiologically
based pharmacokinetic model structure. The recent demonstra-
tions that PFOA exposures in mice alter transporter expression
in at least the liver (Cheng and Klaassen, 2008; Maher et al.,
2008), raise further questions about the time course and dose-
response for these changes and how they affect serum, liver, or
other tissue concentrations. Additional experimental work
would also benefit from quantifying the fecal and urinary
elimination from mice because current analyses assume nearly
complete absorption and do not distinguish elimination routes.
Models that are more physiologically based would potentially
need to explicitly address these issues.
FUNDING
Interagency Agreement (RW-75-92207501) with the
National Toxicology Program at the National Institute for
Environmental Health Science was a source of partial funding.
ACKNOWLEDGMENTS
The United States Environmental Protection Agency through
its Office of Research and Development funded and managed
the research described here. This research has been subjected to
Agency's administrative review and approved for peer review.
We appreciate technical assistance from Kaberi Das.
APPENDIX
function [dCdt] =pfoa_ode_new2compab(t,C,P)
% From one-compartment analysis of 1 and 10 mg/kg data:
ka=0.537;
% From Andersen et al. (2006):
Vfil = 0.01;
free = 0.02;
% Parse the parameter vector P
Vc = P(l);
Vt = P(2);
kd = P(3);
Qd = kd*Vc;
Tm = P(4);
KT = P(5);
Ml = P(6);
Qfil = kfil*Vc;
% Note that gut compartment has different units:
dCdt=zeros(4,l);
dCdt(l) = ka/Vc*C(4)-Qd/Vc*free*C(l)+Qd/
Vc*C(2)-Qfil/Vc*C(l)*free+
Tm*C(3)/(KT+C(3));
dCdt(2) = l/Vt*(free*Qd*C(l) - Qd*C(2));
dCdt(3) = l/Vnl*(Qfil*C(l)*free-Vc*Tm*C(3)/
(KT+C(3))-Qfil*C(3));
dCdt(4) = -ka*C(4);
%[Vfil] = L
%[free] = 1
% [Vc] = L
% [Vt] = L
%[kd] = 1/h
%[Qd] = L/h
%[Tm] = mg/L/h
%[KT] = mg/L
%[kfil] = 1/h
%[Qfil] = L/h
%[C(1)] = mg/L
%[C(2)] = mg/L
%[C(3)] = mg/L
%[C(3)] = mg
Previous
TOG
Next
-------
340
LOU ET AL.
REFERENCES
Abbott, B. D., Wolf, C. J., Schmid, J. E., Das, K. P., Zehr, R. D., Helfant, L.,
Nakayama, S., Lindstrom, A. B., Strynar, M. J., and Lau, C. S. (2007).
Perfluorooctanoic acid (PFOA)-induced developmental toxicity in the mouse
is dependent on expression of peroxisome proliferator activated receptor-
alpha (PPARa). Toxicol. Sci. 98, 571-581.
Akaike, H. (1974). A new look at the statistical model identification. IEEE
Trans. Automat. Contr. 19, 716-723.
Andersen, M. E., Butenhoff, J. L., Chang, S. C., Farrar, D. G., Kennedy, G. L.,
Jr., Lau, C., Olsen, G. W., Seed, J., and Wallace, K. B. (2008).
Perfluoroalkyl acids and related chemistries—Toxicokinetics and modes of
action. Toxicol. Sci. 102, 3-14.
Andersen, M. E., Clewell, H. J., Tan, Y., Butenhoff, J. L., and Olsen, G. W.
(2006). Pharmacokinetic modeling of saturable, renal resorption of
perfluoroalkylacids in monkeys—Probing the determinants of long plasma
half-lives. Toxicology 227, 156-164.
Barbee, R. W., Perry, B. D., Re, R. N., and Murgo, J. P. (1992). Microsphere
and dilution techniques for the determination of blood flows and volumes in
conscious mice. Am. J. Physiol. 263(3 Pt 2), R728-R733.
Belts, K. S. (2007). Perfluoroalkyl acids: what is the evidence telling us?
Environ. Health Perspect. 115, A250-A256.
Biegel, L. B., Hurtt, M. E., Frame, S. R., O'Connor, J. C., and Cook, J. C.
(2001). Mechanisms of extrahepatic tumor induction by peroxisome
proliferators in male CD rats. Toxicol. Sci. 60, 44-55.
Butenhoff, J. L., Gaylor, D. W., Moore, J. A., Olsen, G. W., Rodricks, J.,
Mandel, J. H., and Zobel, L. R. (2004). Characterization of risk for general
population exposure to perfluorooctanoate. Regul. Toxicol. Pharmacol. 39,
363-380.
Calafat, A. M., Wong, L.-Y., Kuklenyik, Z., Reidy, J. A., and Needham, L. L.
(2007). Polyfluoroalkyl chemicals in the U.S. population: Data from the
National Health and Nutrition Examination Survey (NHANES) 2003-2004
and comparisons with NHANES 1999-2000. Environ. Health Perspect. 115,
1596-1602.
Cheng, X., and Klaassen, C. D. (2008). Critical role of PPARj alpha) in
perfluorooctanoic acid- and perfluorodecanoic acid-induced down-regulation
of Oatp uptake transporters in mouse livers. Toxicol. Sci. 106, 37-45.
Cook, J. C., Murray, S. M., Frame, S. R., and Hurtt, M. E. (1992). Induction of
Leydig cell adenomas by ammonium perfluorooctanoate: A possible
endocrine-related mechanism. Toxicol. Appl. Pharmacol. 113, 209-217.
Dai, J., Li, M., Jin, Y., Saito, N., Xu, M., and Wei, F. (2006).
Perfluorooctanesulfonate and periluorooctanoate in red panda and giant
panda from China. Environ. Sci. Technol. 40, 5647-5652.
D'Errico, J. R. (2007). Automatic numerical differentiation, MatlabCentral.
Available from: http://www.mathworks.com/matlabcentral/fileexchange/loadFile.
do?objectld='3490. Accessed January 31,2008.
Dietz, R., Bossi, R., Riget, F. F., Sonne, C., and Born, E. W. (2008). Increasing
perfluoroalkyl contaminants in east Greenland polar bears (Ursus maritimus):
A new toxic threat to the Arctic bears. Environ. Sci. Technol. 42, 2701-2707.
Han, X., Yang, C. H., Snajdr, S. L, Nabb, D. L., and Mingoia, R. T. (2008).
Uptake of perfluorooctanoate in freshly isolated hepatocytes from male and
female rats. Toxicol. Lett. 181, 81-86.
Harada, K., Inoue, K., Morikawa, A., Yoshinaga, T., Saito, N., and
Koizumi, A. (2005). Renal clearance of perfluorooctane sulfonate and
perfluorooctanoate in humans and their species-specific excretion. Environ.
Res. 99, 253-261.
Hansen, K. J., Clemen, L. A., Ellefson, M. E., and Johnson, H. O. (2001).
Compound-Specific, Quantitative Characterization of Organic Fluorochem-
icals in Biological Matrices. Environ. Sci. Technol. 35, 779-770.
Harris, L. A., and Barton, H. A. (2008). Comparing single and repeated dosimetry
data for perfluorooctane sulfonate in rats. Toxicol. Lett. 181, 148-156.
Kennedy, G. L., Butenhoff, J. L., Olsen, G. W., O'Connor, J. C., Seacat, A. M.,
Perkins, R. G., Biegel, L. B., Murphy, S. R., and Farrar, D. G. (2004). The
toxicology of perfluorooctanate. Crit. Rev. Toxicol. 34, 351-384.
Kudo, N., Katakura, M., Sato, Y., and Kawashima, Y. (2002). Sex hormone-
regulated renal transport of perfluorooctanoic acid. Chem. Biol. Interact. 139,
301-316.
Kudo, N., and Kawashima, Y. (2003). Toxicity and toxicokinetics of
perfluorooctanoic acid in humans and animals. /. Toxicol. Sci. 28, 49-57.
Kudo, N., Sakai, A., Mitsumoto, A., Hibino, Y., Tsuda, T., and Kawashima, Y.
(2007). Tissue distribution and hepatic subcellular distribution of perfluor-
ooctanoic acid at low dose are different from those at high dose in rats. Biol.
Pharm. Bull. 30, 1535-1540.
Lagarias, J. C., Reeds, J. A., Wright, M. H., and Wright, P. E. (1998).
Convergence properties of the Nelder-Mead Simplex Method in Low
Dimensions. SIAM J. Optim. 9, 112-147.
Lau, C., Anitole, K., Hodes, C., Lai, D., Pfahles-Hutchens, A., and Seed, J.
(2007). Perfluoroalkyl acids: A review of monitoring and toxicological
findings. Toxicol. Sci. 99, 366-394.
Lau, C., Thibodeaux, J. R., Hanson, R. G., Narotsky, M. G., Rogers, J. M.,
Lindstrom, A. B., and Strynar, M. J. (2006). Effects of perfluorooctanoic acid
exposure during pregnancy in the mouse. Toxicol. Sci. 90, 510-518.
Lindstrom, M. J., and Bates, D. M. (1990). Nonlinear mixed effects models for
repeated measures data. Biometrics 46, 673-687.
Luippold, G., Pech, B., Schneider, S., Osswald, H., and Muhlbauer, B. (2002).
Age dependency of renal function in CD-I mice. Am. J. Physiol. Renal
Physiol. 282, 886-890.
Maher, J. M., Aleksunes, L. M., Dieter, M. Z., Tanaka, Y., Peters, J. M.,
Manautou, J. E., and Klaassen, C. D. (2008). Nrf2 and PPARjalpha)-
Mediated Regulation of Hepatic Mrp Transporters after Exposure to
Perfluorooctanoic Acid and Perfluorodecanoic Acid. Toxicol. Sci 106,
319-328.
Nakagawa, H., Hirata, T., Terada, T., Jutabha, P., Miura, D., Harada, K. H.,
Inoue, K., Anzai, N., Endou, H., Inui, K., et al. (2008). Roles of organic
anion transporters in the renal excretion of perfluorooctanoic acid. Basic
Clin. Pharmacol. Toxicol. 103, 1-8.
Olsen, G. W., Burris, J. M., Ehresman, D. J., Froehlich, J. W., Seacat, A. M.,
Butenhoff, J. L., and Zobel, L. R. (2007). Half-life of serum elimination of
perfluorooctanesulfonate, perfluorohexanesulfonate, and perfluooctanoate in
retired fluorochemical production workers. Environ. Health Perspect. 115,
1298-1305.
Pinheiro, J. C., and Bates, D. M. (2000). In Mixed-Effects Models in S and
S-PLUS. Springer, New York.
Pinheiro, J. C., Bates, D. M., DebRoy, S., Sarkar, D., and R Core Team
(2007). In Nlme: Linear and Nonlinear Mixed Effects Models. R package
version 3, pp. 1-84. The R Foundation for Statistical Computing, Vienna,
Austria.
Qi, Z., Whitt, L, Mehta, A., Jin, J., Zhao, M., Harris, R. C., Fogo, A. B., and
Breyer, M. D. (2004). Serial determination of glomerular filtration rate in
conscious mice using FITC-insulin clearance. Am. J. Physiol. Renal Physiol.
286, 509-596.
R Development Core Team. (2007). R: A Language and Environment for
Statistical Computing. R Foundation for Statistical Computing, Vienna,
Austria. ISBN 3-900051-07-0. http://www.R-project.org.
Shampine, L. F., and Reichelt, M. W. (1997). The Matlab ODE suite. SIAM J.
Sci. Comput. 18(1), 1-22.
Tan, Y. M., Clewell, H. J., 3rd, and Andersen, M. E. (2008). Time
dependencies in perfluorooctylacids disposition in rat and monkeys: A
kinetic analysis. Toxicol. Lett. Ill, 38^-7.
Trudel, D., Horowitz, L., Wormuth, M., Scheringer, M., Cousins, I. T., and
Hungerbiihler, K. (2008). Estimating consumer exposure to PFOS and
PFOA. Risk Anal. 28, 251-269.
Previous
TOG
Next
-------
SINGLE AND REPEATED DOSE PHARMACOKINETICS OF PFOA
341
U.S. EPA. (2005). Draft risk assessment of the potential human health effects
associated with exposure to perfluorooctanoic acid and its salts (PFOA).
OPPT, Washington, DC. Available from: http://www.epa.gov/oppt/pfoa/
pubs/pfoarisk.pdf. Accessed September 28, 2008.
Washburn, S. T., Bingman, T. S., Braithwaite, S. K., Buck, R. C.,
Buxton, L. W., Clewell, H. J., Haroun, L. A., Kester, J. E., Rickard, R. W.,
and Shipp, A. M. (2005). Exposure assessment and risk characterization for
perfluorooctanoate in selected consumer articles. Environ. Sci. Technol. 39,
3904-3910.
White, S. S., Calafat, A. M., Kuklenyik,Z., Villanueva,L.,Zehr,R. D., Helfant,L.,
Strynar, M. J., Lindstrom, A. B., Thibodeaux, J. R., Wood, C., et al. (2007).
Gestational PFOA exposure of mice is associated with altered mammary gland
development in dams and female offspring. Toxicol. Sci. 96, 133-144.
Wolf, C. J., Fenton, S. E., Schmid, J. E., Calafat, A. M., Kuklenyik, Z.,
Bryant, X. A., Thibodeaux, J. R., Das, K. P., White, S. S., Lau, C. S., et al.
(2007). Developmental toxicity of perfluorooctanoic acid in the CD-I mouse
after cross-foster and restricted gestational exposure. Toxicol. Sci. 95,
461^73.
Previous
TOC
Next
-------
Modulation of TLR2 Protein Expression by miR-105 in Human
Oral Keratinocytes*
Received for publication, January 28,2009, and in revised form, May 22,2009 PublishedJBC Papers in Press, June 9,2009,001 10.1074/jbc.M109.013862
Manjunatha R. Benakanakere*, Qiyan Li*, Mehmet A. Eskan*, Amar V. Singh§\ Jiawei Zhao*, Johnah C. Galicia*,
Panagiota Stathopoulou*, Thomas B. Knudsen§", and Denis F. Kinane*1
From the ^Center for Oral Health and Systemic Disease and the § Department of Molecular, Cellular, and Craniofacial Biology,
University of Louisville School of Dentistry, Louisville, Kentucky 40202, the ^National Center for Computational Toxicology,
Environmental Protection Agency, Research Triangle Park, North Carolina 2777 7, and ^Lockheed Martin,
Research Triangle Park, North Carolina 2777 7
Mammalian biological processes such as inflammation,
involve regulation of hundreds of genes controlling onset and
termination. MicroRNAs (miRNAs) can translationally repress
target mRNAs and regulate innate immune responses. Our
model system comprised primary human keratinocytes, which
exhibited robust differences in inflammatory cytokine produc-
tion (interleukin-6 and tumor necrosis factor-«) following spe-
cific Toll-like receptor 2 and 4 (TLR-2/TLR-4) agonist chal-
lenge. We challenged these primary cells with Porphyromonas
gingivalis (a Gram-negative bacterium that triggers TLR-2 and
TLR-4) and performed miRNA expression profiling. We identi-
fied miRNA (miR)-105 as a modulator of TLR-2 protein trans-
lation in human gingival keratinocytes. There was a strong
inverse correlation between cells that had high cytokine
responses following TLR-2 agonist challenge and miR-105 lev-
els. Knock-in and knock-down of miR-105 confirmed this
inverse relationship. In silica analysis predicted that miR-105
had complementarity for TLR-2 mRNA, and the luciferase
reporter assay verified this. Further understanding of the role of
miRNA in host responses may elucidate disease susceptibility
and suggest new anti-inflammatory therapeutics.
The innate immune response is a crucial first line of
defense against pathogens. Host detection of microbes
occurs through pattern recognition receptors, including
Toll-like receptors (TLRs)2 that are expressed on many cells,
including macrophages, monocytes (1), and keratinocytes
(2). To date, 11 TLRs have been identified in humans, recog-
nizing a range of distinct and conserved microbial molecules
(3). TLRs responding to particular pathogens may activate
complex networks of pathways and interactions, positive
and negative feedback loops, and multifunctional transcrip-
* This work was supported, in whole or in part, by National Institutes of Health
Grant DE017384 (to D. F. K.) from the United States Public Health Service,
NIDCR.
1 To whom correspondence should be addressed: University of Louisville
School of Dentistry, 501 South Preston St., Rm. 204, Louisville, KY 40202.
Tel.: 502-852-3175; Fax: 502-852-5572; E-mail: dfkina01@louisville.edu.
2 The abbreviations used are: TLR, Toll-like receptor; HGEC, human gingival
epithelial cell; IL-6, interleukin 6; TNF-a, tumor necrosis factor a; miRNA,
microRNA; hsa-miR, Homo sapiens microRNA; pre-miR, precursor
microRNA; UTR, untranslated region; SOCS3, suppressor of cytokine signal-
ing 3; FSL-1, Pam2CGDPKHPKSF (a synthetic diacylated lipoprotein and a
specific ligand for TLR-2).
tional responses (4). Among the key downstream targets of
these networks are NF-KB, mitogen-activated protein
kinases, and members of the IRF family (5). Proper regula-
tion of the gene products comprising these networks by tran-
scriptional and post-transcriptional processing is not only
important for selective pathogen elimination but also for pre-
venting excessive accumulation of cytokines such as interfer-
on-/3, interferon-y, IL-6, and TNF-a that initiate the host
defense against microbial attack (6). Deregulated expression of
these cytokines has been implicated in cancer, autoimmunity,
and hyper-inflammatory states (7-9).
MicroRNAs (miRNAs) have been implicated in pathway-
level regulation of complex biological processes (10). The role
of miRNA-based regulation of the innate immune responses is
a current topic of investigation (11). Mammalian miRNAs are a
class of conserved, small noncoding RNA oligonucleotides that
function as negative regulators of translation for multiple target
transcripts (12). As many as 5000 distinct miRNAs maybe tran-
scribed and processed in mammalian cells (13-17). Mature
miRNAs bind to specific cognate sequences in the 3'-UTRs of
target transcripts, resulting in either mRNA degradation or
inhibition of translation (12).
In mammalian cells, the miRNAs provide a key level of bio-
logical regulation in developmental and differentiation path-
ways (18). Deregulation of specific miRNA abundance has been
associated with malignancies in the colon, breast, and lung (19,
20). Recently, miRNAs have been shown to modulate the
NF-KB pathway (miR-146a) (21) and negatively regulate
TRAF6, IRAKI (miR-155) (22), or SOCS3 (miR-203) (23). It is
presently unclear how miRNAs regulate cellular pathways in
innate and inflammatory processes, where precise control of
complex networks is needed to engage an appropriate response
to microbes that avoids a cytokine storm.
Periodontitis is a common chronic inflammatory condition
affecting 50% of humans that results in loss of bone and teeth
(24). This disease is initiated by dental plaque, a microbial bio-
film composed mainly of Gram-negative anaerobic bacilli (25,
26), including the pathogen Porphyromonas gingivalis. Individ-
ual human variability in susceptibility to periodontitis is recog-
nized (27) and may involve individual variation in the immune
response (25, 26). We identified innate immune variations
within a bank of over 30 primary human gingival cell cultures
(25) based on variations in cytokine response following TLR
agonist challenge.
a
o
a.
a>
a.
o
cr
p
b
<3
W
a>
T£
a>
3
cr
a>
K)
o
AUGUST 21, 2009-VOLUME 784-NUMRFR 34
Previous
IAL OF BIOLOGICAL CHEMISTRY 23107
-------
miR- 705 in Human Oral Keratinocytes
The present study tested the hypothesis that differential
expression of miRNAs may account for some of the variability
in innate immunity. To test this hypothesis, we selected three
"normal" and three "diminished" cytokine-response pheno-
types. We subjected the corresponding primary human gingival
cell cultures to TLR agonist challenge and profiled the expres-
sion of 600 miRNAs. We found strong up-regulation of hsa-
miR-105 specifically in the diminished cytokine-response phe-
notype and furthermore showed that TLR-2 protein levels were
depressed. This implies a concordant logic circuit in which
miR-105 inversely regulates TLR-2 function. A computational
(in silico) search of the miRNA database revealed that TLR-2
transcript is a potential target for miR-105 regulation at the
3'-UTR. This binding was confirmed using a linked luciferase
reporter gene, and through small interference RNA and inhib-
itor (antagomir) studies a functional association with cell sur-
face TLR-2 expression. We also confirmed this complementa-
rity. We conclude that cell surface TLR-2 expression is
inversely regulated by miR-105 expression in human gingival
epithelial cells. This mechanism may reduce inflammatory
cytokine production and provide a novel target for therapeutic
intervention.
EXPERIMENTAL PROCEDURES
Cell Culture and Challenge Assays—A total of 13 human gin-
gival epithelial cells (keratinocytes), with University of Louis-
ville IRB approval, were obtained from healthy patients after
third molar extraction. They were grown as previously
described (28) to sub-confluence, sub-cultured, and challenged
as described (2, 29, 30). At confluence, they were challenged
with heat-inactivated P. gingivalis (strain 33277) or 1 /xg/ml
FSL-1 (Pam2CGDPKHPKSF, a synthetic diacylated lipoprotein
and a specific ligand for TLR-2) (InvivoGen, CA). Cells were
challenged for 24 h, and culture supernatants were subjected to
IL-6 and TNF-a cytokine levels were measured by enzyme-
linked immunosorbent assay (BD Biosciences). The transcrip-
tion factor NF-KB assay was performed using a modified elec-
trophoretic mobility shift assay technique with TransAM™
NF-KB enzyme-linked immunosorbent assay kit from Active-
Motif (Carlsbad, CA) according to the manufacturer's instruc-
tions. HEK-293 (ATCC number: CRL-1573) cells were cultured
following ATCC protocol. Briefly, the cell monolayer was
washed and incubated with 2-3 ml of trypsin-EDTA solution to
the flask and neutralized with trypsin inhibitor after 5 min. The
cells were centrifuged and suspended in ATCC-formulated
Eagle's minimum essential medium (catalogue no. 30-2003)
with 10% fetal bovine serum (complete medium). The cells
were propagated in complete medium until they were ready for
transfection.
miRNA Array Profiling/Analysis—Total RNA was collected
by the TRIzol method and purified with a Qiagen purification
kit (Qiagen), and total RNA quality was analyzed using a Bio-
analyzer 2100 (Agilent). Equal amounts of each sample were
used to generate a reference pool. For each array to be hybrid-
ized, 2 /j,g of total RNA from each sample, and the reference
pool were labeled with Hy3™ and Hy5™ fluorescent label,
respectively, using the miRCURY™ LNA Array labeling kit
(Exiqon, Denmark) following the manufacturer's instructions.
The Hy3™-labeled sample and the Hy5™-labeled reference
pool RNA were mixed and hybridized to the miRCURY™ LNA
array version 8.1 (Exiqon). The hybridization was performed
according to the miRCURY™ LNA array manual using a
Tecan HS4800 hybridization station (Tecan, Austria). The
miRCURY™ LNA array microarray slides were scanned by a
ScanArray 4000 XL scanner (Packard Biochip Technologies),
and the image analysis was carried out using the ImaGene 7.0
software (BioDiscovery, Inc.). Expression ratios were deter-
mined for microarray data by computing the background-cor-
rected fluorescent signal from the query sample (Q)/reference
sample (R). Ratiometric data were transformed to log 2 to pro-
duce a continuous spectrum of up- and down-regulated values.
Data were normalized by plotting the difference, log 2(Q/R),
against the average, (1/2) log 2(Q*R) followed by the application
of locally weighted regression (lowness) to smooth intensity-
dependent ratios. The log 2(Hy3/Hy5) intensity data were
uploaded into GeneSpring v7.3 for two-way analysis of variance
(factors = cell-line and second treatment), using a parametric
test with variances assumed equal, a cutoff (p = 0.05) to gen-
erate a heat-map through bi-directional hierarchical clustering
(31, 32).
TLR-2 mRNA and miR-105 Real-time PCR—Total RNA was
extracted from cultured cells using TRIzol reagent (Invitrogen).
The isolated total RNA samples were used for first strand
cDNA synthesis with specific miR-105 hairpin loop primers
(Applied Biosystems, Foster City, CA). Real-time PCR was per-
formed by using 1 ng of cDNA with an miR-105-specific primer
and probe on an ABI 7500 system (Applied Biosystems) in the
presence of TaqMan DNA polymerase. The data were analyzed
by normalizing miRNA level to miRNA RNU48 (small nucleo-
lar RNA used as internal control, which has least variability
across the cell types and challenges). For TLR-2 mRNA quan-
tification, the total RNA was converted to single-stranded
cDNA using a cDNA archive kit (Applied Biosystems) and 100
ng of cDNA to quantify TLR-2 mRNA using the TaqMan
method (Applied Biosystems). Glyceraldehyde-3-phosphate
dehydrogenase was the internal control, and -fold increase was
calculated as described (33).
Transfection of miRNA—Epithelial cells were transfected
with 100 pmol of miR-105 mimic (UCAAAUGCUCAGACUC-
CUGUGGU) and miR-105 inhibitor (AGTTTACGAGTCT-
GAGGACACCA) (Dharmacon, CA) and co-transfected with
100 pmol of small interference RNA control, labeled with 6-car-
boxyfluorescein to monitor transfection efficiency. The trans-
fection reaction was performed using FuGENE 6 reagent
(Roche Applied Science). Cells were challenged with P. gingiva-
/zs/FSL-1 for 24 h following transfection.
Immunohistochemistry—The cells were seeded onto collag-
en-coated chamber glass slides (Lab-Tek™ II Chamber Slide®,
Nalgene Nunc International, Rochester, NY). At 50-60% con-
fluence, the cells were transfected either with miR-105 mimic
or miR-105 inhibitor or with scrambled small interference RNA
using FuGENE 6 transfection reagent as described above. The
transfection reaction was performed for up to 24 h and replaced
with fresh medium. The challenge assay was performed after
48 h of transfection with FSL-1 (0.5 /Ag/ml, InvitroGen), they
were fixed in 4% paraformaldehyde, permeabilized, and stained
a
o
a.
a>
a.
o
cr
p
b
<3
w
a>
T£
a>
3
cr
a>
23108 JOURNAL OF BIOLOGICAL
Previous
1MB 284-NUMBER 34-AUGUST 21, 2009
-------
miR-W5in Human Oral Keratinocytes
CD
m
£
,wj
i
6
B
hsa-miR-105
11006 (ISH*2B)
19530 (h»W7i)
1SOM(W#7c)
1~JJ(l»W7j)
177«(HsH*7t)
10975 (HSH*182)
Normal type
Stem loop of pre-miR-105
s^^Si^r^SS^S^^^S
V
Diminished type
J
13171 (hsw*«J|
11W7(hsw*«4)
ikm
V
Mature miR-105 UCAAAUGCUCAGACUCCUGUGGU
miR-105 UGUCCUCA- - GACUCGUAAAC
i ii 11 ii i mi 1111
3' UTR of TLR2 ATAAGAGTGGCATAGTATTTG
Antagomir-105 AGTTTACGAGTCTGAGG ACACCA
FIGURE 1. Normal and diminished cytokine response cells were challenged with heat-inactivated P. gingivalis (strain 33277) at 100 multiplicity of
infection for 24 h, and RNA was miRNA microarray profiled using the miRCURYLNA Array (Exiqon). The heat map shows two-way hierarchical clustering
of genes and samples (rows = miRNA, columns = sample). The color scale indicates relative expression:ye//ow, above mean; blue, below mean; and black, below
background. Global microarray expression revealed distinct profiles for 109 miRNAs expressed among these cells (p = 0.05) out of which, 26 well annotated
miRNAs revealed distinct patterns (p = 0.0037), and miR-105 is represented with a rectangular dotted box (A). This signal is up-regulated in the diminished
cytokine response phenotype after challenging with heat-inactivated P. glnglvalls and down-regulated in normal response cells after challenge (p = 0.0017) (6).
The stem loop of miR-105 with complimentarity toTLR-2 mRNAand sequence of antagomir used for the present study are shown (Q.
with anti-human TLR-2 antibody overnight at 4 °C followed by
Alexa Fluor® 488 anti-mouse IgG in 3% bovine serum albumin
(1/500, InvitroGen) for 1 h at room temperature and SYTO®
83-orange for 15 min. The stained cultures were photographed
using a Confocal Laser Scanning Microscope (FV500, Olym-
pus, Melville, NY).
Western Blotting— Total protein was extracted from cells
using radioimmune precipitation assay buffer after 24 h of chal-
lenge. The Western blot was performed by loading 25 jug of
total proteins on to each lane. Blotted membranes were blocked
using 5% nonfat milk and incubated at 4 °C overnight in anti-
TLR-2 antibody (Cell Signaling Technology, Danvers, MA).
The membranes were washed and incubated in anti-mouse IgG
conjugated with horseradish peroxidase (Cell Signaling Tech-
nology) secondary antibody and signal was developed using
ECL plus™ Western blotting detection reagent (Amersham
Biosciences). The ratiometric analysis of band intensity was cal-
culated using FluorChem HD software (Alpha Innotech).
Luciferase Reporter Assay—-The putative miRNA-105 target
site of 52 bp within the 3'-UTR region of human TLR-2 mRNA
(Ensembl transcript ID: ENST 00000260010) were synthesized
with flanking Spel and Hindlll restriction enzyme sites. In addi-
tion, the primers with their putative binding site mutated, were
also synthesized from Integrated DNA Technologies Inc.
(sense primers: 5'-gctgactagtCATAGATGATCAAGTCCCT-
TATAAGAGTGGCATAGTATTTGCATATAACaagcttggac-
3'; antisense primer: 5'-gtccaagcttGTTATATGCAAATACT-
ATGCCACTCTTATAAGGGACTTGATCATCTATGactag-
tcagc-3'; mutated sense primer: 5'-gctgactagtCATAGATGA-
TCAAGTCCCTTATAAGAGTGGCATAGTCATATAACaa-
gcttggac-3'; and mutated antisense primer: 5'-gtccaagcttGTT-
ATATGACTATGCCACTCTTATAAGGGACTTGATCAT-
CTATGactagtcagc-3')- The sense and antisense strands of the
oligonucleotides were annealed (34). The annealed oligonu-
cleotides were digested with Spel and Hindlll and ligated into
the multiple cloning site of the pMIRREPORT Luciferase vec-
tor (Ambion, Inc.). The post-transcriptional regulation of
pMIRREPORT luciferase vector was potentially regulated by
miRNA interactions with the TLR-2 3'-UTR. We then trans-
fected cultured HEK293 cells with each of these reporter con-
structs (pMIR-TLR2 or pMIR-mutTLR-2), as well as co-trans-
fecting them with pMIF-cGFP-Zeo-miR-105 (pMIF-miR-105)
AUGUST 21, 2009-VOLUME 284-NUMREBJ4
TOC
a.
3
cr
p
b
<3
.
w
a>
T£
a>
cr
a>
-------
miR-105 in Human Oral Keratinocytes
A
K
B
a
£
<
150i
E 100
d 50-
Normal type
Diminished type
Normal type
to
Diminished type
Normal type
Diminished type
FIGURE 2. TLR-2 gene, IL-6, and TNF-a cytokine expression in both normal type and diminished cytokine response cells. Human gingival keratinocytes
were challenged with heat-inactivated P. gingivalis (100 MOI) or FSL-1 (1 /ng/ml) for 24 h. Total RNA was isolated for real-time PCR. The TLR-2 receptor mRNA
abundance increased upon TLR-2 ligand challenge in normal cells and was unchanged in diminished response cells after challenge (A), compared with control
(p< 0.001). An enzyme-linked immunosorbent assay was performed on the supernatants, and IL-6 (6) and TNF-a (Q were found to be up-regulated in normal
cells and remained unchanged in diminished response cells after challenge (p< 0.001). Results are mean ± S.D.fn = 3) using the same primary cells. Statistical
comparisons are shown by horizontal bars with asterisks above them (*** indicates p < 0.001 ;NS = no significant difference).
a
o
^_
o
B)
a.
a>
a.
3
cr
p
b
<3
.
cr
3
W
a>
plasmid (System Biosciences) following transfection with
FuGENE 6 as noted above. Luciferase expression was assessed
by confocal microscopy 24 h after transfection by use of anti-
Luciferase antibody (Abeam). The protein extract from the
transfected cells was collected using radioimmune precipita-
tion assay buffer and equal amounts of protein were tested by a
Luciferase activity assay kit following the manufacturer's
instructions (Stratagene).
Statistical Analysis—The microarray statistical analysis is
detailed under "miRNA Array Profiling/Analysis" above. The
mRNA -fold increase data were calculated according to the
AACT method (33). Cytokine data were evaluated by analysis of
variance using the InStat program (GraphPad, San Diego, CA)
with Bonferroni corrections applied. Statistical differences
were declared significant at/? < 0.05 level. Statistically signifi-
23110 JOURNAL OF BIOLOGICAL CHFMI^TRY
Previous
cant data are indicated by asterisks (p < 0.05 (*), p < 0.01 (**),
and p< 0.001 (***)).
RESULTS
MiR-105 Is Up-regulated in Human Gingival Keratinocytes
with a Diminished Inflammatory Response—We hypothesized
that miRNAs may play a role in innate immune response vari-
ation (25) and could differentiate periodontitis disease-suscep-
tible and disease-resistant subjects. From a bank of over 30 pri-
mary cell cultures of human gingival epithelial cells (HGECs)
(2) we selected cultures having a diminished cytokine response
type and "normal" response type as reported previously. Briefly,
the rule specification for the latter selected cells that up-regu-
lated IL-6 and TNF-a production by at least 2-fold after 4-h
challenge with TLR agonists, and the rule for the former
ME 284-NUMBER 34-AUGUST 21, 2009
-------
miR-W5in Human Oral Keratinocytes
Diminished type Normal type
H
95kd
45kd
P-9
"* TLR-2
• (1-Actin
Diminished type Normal type
H h
1.5-,
Nortnal type + P.g Diminished type * P.g
FIGURE 3. miR-105 and TLR2 expression in normal and diminished cytokine response cells. The cells were
subjected to P. gingivalis and FSL-1 treatment for 24 h and quantitated the miR-105 expression and Western
blot for TLR-2. Total RNA was amplified with specific miR-105 hairpin loop primers and subjected to real-time
PCR. miR-105 up-regulated in diminished cytokine response cells after challenge. The miR-105 expression
showed minimal or no change in normal cells (XI) (p = 0.05). Three normal and three diminished cell types were
compared after P. gingivalis challenge for miR-105 gene expression (6) and TLR-2 protein expression, using
Western blot (Q with ratio metric analysis (/3-Actin/TLR-2) (D). Statistical comparisons are shown by bars with
asterisks above them (* indicates p < 0.05; NS = no significant difference).
(diminished) phenotype was no significant increase in pro-in-
flammatory cytokine production after challenge. We thus
tested in depth the 3 diminished cytokine response primary
cultures against 3 representatives of the normal HGEC type
chosen from the median of the range of the 13 available cultures
(2).
We performed chip-based miRNA profiling from normal
and diminished cytokine response cells challenged with heat-
inactivated P. gingivalis, a TLR-2 and TLR-4 agonist. The data
were deposited in the Gene Expression Omnibus data base
under the platform GPL7423 and series GSE13042
(www.ncbi.nlm.nih.gov/geo). In the preliminary statistical
analysis, the normalized miRNA data were analyzed by two-
way analysis of variance (phenotype, treatment, and interac-
tion) for all miRNA chips. Of the 600 miRNA species tested on
the platform, 95 miRNAs were significantly different between
phenotypes and 45 between treatments. Among the 109
miRNAs that were differentially expressed only 26 were well
annotated. These 26 were significantly altered by challenge
with heat-inactivated P. gingivalis when compared with
unstimulated cells (p = 0.0038). The miR-105 signal was mark-
edly down-regulated in normal versus diminished cells after
TLR agonist challenge (Fig. L4) (p = 0.0017).
In silico analyses revealed 1060 hits for miR-105 targets
and predicted complimentarity to TLR-2 (Fig. 1C). Because
we noted miR-105 up-regulation
in the diminished cytokine re-
sponse cells, we constructed a
computational matrix for miR-105
targets, which revealed 37 poten-
tial mRNA targets across species.
In matrix, hsa-miR-577 and hsa-
miR-19b were also differentially
expressed between normal and
diminished cells but to a lesser
degree than miR-105 (see dotted box
in Fig. I A). Numerical expression
values were plotted to show the dif-
ference between normal and dimin-
ished cell types (Fig. 15). For this
reason, we focused on miR-105.
TLR-2 mRNA Lip-regulation Cor-
relates with Pro-inflammatory
Cytokines in Gingival Keratinocytes—
Because P. gingivalis signals
through both TLR-2 and TLR-4 (2,
35) we evaluated TLR-2 mRNA by
quantitative real-time PCR after
challenging both cell types with
FSL-1, a specific agonist to TLR-2.
Analysis of TLR-2 mRNA abun-
dance between phenotypes revealed
a 2.5-fold up-regulation with P. gin-
givalis challenge and a 4.5-fold up-
regulation with FSL-1 challenge rel-
ative to the unstimulated control
(p = 0.001) (Fig. 2A). In striking
contrast, diminished cytokine
response cells did not show this up-regulation (Fig. 2A). TLR-2
induction upon P. gingivalis or FSL-1 stimulation is consistent
with TLR-2 recognition of P. gingivalis (36, 37). Furthermore,
IL-6 cytokine protein levels corresponded to corresponding
mRNA levels. Normal response cells up-regulated their IL-6
(Fig. 25) and TNF-a (Fig. 2C) production followingP.gingivalis
or FSL-1 challenge, and again this response was not evident in
diminished response cells (Fig. 2, 5 and C). We also explored
IL-12p40 secretion in the gingival epithelial cells after P. gingi-
valis and FSL-1 challenge. The primary gingival epithelial cells
do not secret IL-12p40.3 This has to be further verified and
hence not included in the present study.
Modulation of TLR-2 Protein Expression by miR-105—Con-
firmation of the differential expression of miR-105 was sought
by a non-array method. The real-time PCR data indicated an
8-fold up-regulation of miR-105 following P. gingivalis chal-
lenge, and an 11-fold increase following FSL-1 challenge, in
diminished-response cells relative to an internal benchmark,
miRNA RNU48 (Fig. 3, A and 5). Consistent with the microar-
ray data, the normal response cells did not show significant
up-regulation of miR-105 (Fig. 3).
' M. R. Benakanakere, Q. Li, M. A. Eskan, A. V. Singh, J. Zhao, J. C. Galicia, P.
Stathopoulou, T. B. Knudsen, and D. F. Kinane, unpublished data.
a
o
a.
a>
a.
o
cr
p
b
<3
W
a>
•g.
a>
3
cr
a>
AUGUST 21, 2009-VOLUME 284-NUMREBJ4
TOC
>AL OF BIOLOGICAL CHEMISTRY 23111
-------
miR- 705 in Human Oral Keratinocytes
95kd
45kd
Mock
Antagomir
P-9
FSL-1
300i
200-
100-
TLR-2
[i-Actin
+ . . .
IL-6
FIGURE 4. The level of TLR-2 protein by Western blot after transfecting
miR-105 antagomir. The diminished type cells transfected with miR-105
antagomir up-regulated TLR-2 protein expression upon TLR-2 agonist chal-
lenge (A). The ratio metric analysis (/3-Actin/TLR-2) of Western blot intensity
showing TLR2 protein expression (B). Similarly, IL-6 (Q and TNF-a (D) were
up-regulated in diminished cell types when miR-105 antagomir was trans-
fected and challenged either with P. gingivalis or FSL-1. Results are mean ±
S.D. (n = 3) representative of two independent experiments. Statistical com-
parisons are shown by bars with asterisks above them (*** indicates p < 0.001;
NS = no significant difference).
Search of the open miRNA data base revealed over 1060
potential target transcripts for miR-105. These potential target
genes were loaded onto Ingenuity pathway analysis (Ingenuity
Systems Inc.) software to discover the most significant path-
ways associated with the global miR-105 target genes. This
analysis revealed a preponderance of immune diseases (data
not shown) and identified TLR-2 as one of the important tar-
gets. Because microarray profiling revealed miR-105 as the
most significant miRNA discriminating between normal and
diminished HGEP phenotypes, we tested for evidence that
miR-105 expression would down-regulate TLR-2 as well as pro-
inflammatory cytokines in the 3 selected cell types, as well as
across the broader panel of 13 HGECs available. This analysis
revealed a generalized, strong inverse correlation between
TLR2 protein and miR-105 gene expression (Fig. 3, B and C).
We next sought to determine whether miR-105 directly
modulated TLR-2 mRNA and/or protein expression in HGECs.
To test this, we transfected miR-105 mimic (same sequence as
mature miR-105) and miR-105 antagomir (sequence comple-
mentary to miR-105, which blocks its effect) to diminished
cytokine cell phenotypes. After miR-105 transfection, the cells
were challenged for 24 h with either heat-inactivated P. gingi-
valis or FSL-1. The diminished cytokine response cells trans-
fected with miR-105 antagomir up-regulated the TLR-2 protein
levels in either challenge (Fig. 4A) compared with mock trans-
fected cells. This indicates a role for miR-105 in modulating
TLR-2 protein expression and implies a post-transcriptional
repression of TLR-2 translation. However, the miR-105
antagomir control cells expressed higher TLR-2 protein versus
non-transfected cells. This implies a role for miR-105 in mod-
ulating basal receptor expression, an inference supported by
the correlation of TLR-2 and miR-105 (Fig. 3, B and C). Evi-
dence for a functional relationship was sought by transfecting
miR-105 antagomir into diminished cytokine response cells
challenged with P. gingivalis and FSL-1. This scenario induced
significantly higher IL-6 and TNF-a compared with mock con-
trol, confirming that miR-105 overexpression is biologically rel-
evant (Fig. 4, C and D). Because we previously showed that P.
gingivalis activates NF-KB (38) and induces cytokines (39), the
induction of IL-6 in miR-105 mimic and antagomir-transfected
cells was further verified by measuring NF-KB activity by mod-
ified electrophoretic mobility shift assay technique (FACE kit,
ActifMotif). The miR-105 inhibitor-transfected cells exhibited
increased NF-KB activity upon P. gingivalis and FSL-1 ligand
challenge (data not shown). The normal cell phenotype trans-
fected with miR-105 mimic down-regulated NF-KB activation
after P. gingivalis or FSL-1 challenge suggesting that miR-105
induction is dependant on NF-KB.
MiR-105 Modulated Surface TLR-2 Expression—The func-
tional relationship was further confirmed by Western blot for
TLR-2 and immunohistochemistry after inhibiting miR-105.
Overexpression of miR-105 mimic in normal cells challenged
with FSL-1 suppressed TLR-2 protein levels compared with
scrambled miRNA target or mock transfection (Fig. 5A). We
then sought to determine the effect of reducing the miR-105
levels by transfecting miR-105 antagomir into the cell type with
miR-105 up-regulation. Lysates of cells challenged with FSL-1
had increased TLR-2 protein (Fig. 55). The antagomir control
had increased TLR-2 protein level confirming our TLR-2
mRNA observations (Fig. 3A), which correlated with the
expression of miR-105.
Surface expression of TLR-2 immunoreactivity was con-
firmed by confocal microscopy. TLR-2 expression was down-
regulated in cells transfected with miR-105 mimic and chal-
lenged with FSL-1 (Fig. 6D). In contrast, TLR-2 expression was
retained in cells transfected with miR-105 antagomir (Fig. 6C).
This confirmed that miR-105 up-regulation suppressed TLR-2
protein expression on the cell surface. These data are consistent
with observations of another miRNA, let-7i, and the link
between surface TLR-4 expression, NF-KB activation, and cyto-
kine modulation.
To test if we had predicted correctly the binding site for miR-
105 on TLR-2, miR-105 binding to the 3'-UTR of TLR-2 was
assessed by cloning a putative cognate 22-bp fragment from 332
bp away from the stop codon (Ensembl transcript ID:
ENST00000260010) to the multiple cloning site located at the
3'-UTR of the Luciferase reporter gene in pMIRREPORTER. A
mutant vector was also constructed by deleting the predicted
AGTTTA binding site of miR-105 and cloning the fragment
into a multiple cloning site located in the 3'-UTR of the Lucif-
erase reporter gene. Co-transfection of HEK293 cells with
Luciferase construct (pMIR-TLR2) and a vector overexpressing
miR-105 precursor (pMIF-miR-105) inhibited Luciferase activ-
ity (Fig. 7A, panel C). The mutant Luciferase vector (pMIR-
mutTLR2) co-transfected with pMIF-miR-105 retained lucifer-
o
o
Q.
0>
Q.
O
CT
O
O
<3
W
(D
T£
(D
3
cr
o>
23112 JOURNAL OF BIOLOGICAL CHFMISTRY
Previous
UME 284-NUMBER 34-AUGUST 21, 2009
-------
miR-W5in Human Oral Keratinocytes
CD
m
£
,wj
i
6
*TLR2
•*• p-Actin
Mock
FSL-1
miR-105 mimic
miR-105 mimic+FSL-1
45kDa
Mock
FSL-1
miR-105 antagomir
miR-105 antagomir+FSL-1
diction of the binding site. We
may conclude that modulation of
TLR-2 expression by miR-105
occurs through binding to the
3'-UTR of TLR-2 mRNA, thus
inhibiting TLR-2 translation.
0.8-1
0.0
FIGURE 5. The expression of TLR-2in epithelial cells following miR-105 mimic and antagomir transfection.
The normal cells were transfected with miR-105 mimic, and diminished cells were transfected with miR-105
antagomir and challenged with FSL-1 (1 /ng/ml) for 24 h. Protein (20 /ng) was loaded onto each well and detected by
anti-TLR-2 antibody for chemiluminescence detection. The normal cells transfected with miR-105 down-regulated
TLR-2 protein expression (A), and diminished cell type transfected with miR-105 antagomir up-regulated TLR-2
protein level (6), ratios of Western blot intensity data (/3-Actin/TLR2) are represented in Cand D, respectively.
FIGURE 6. Surface expression of TLR-2 in gingival keratinocytes. Normal cells transfected with either miR-105
mimic or antagomir and stained with antiTLR-2 antibody-clone TL2.3 (eBiosciences) with proper Isotype control
(Mouse lgG2a).TLR-2 was detected by immunohistochemistryand photographed by confocal microscopy. Control
(A), FSL-1-treated cells increased TLR-2 expression (6) miR-105 inhibitor did not affect the surface expression of
TLR-2, but the expression was maximized after FSL-1 challenge (Q, miR-105 mimic suppressed TLR-2 surface expres-
sion even after challenging with FSL-1 as seen by confocal microscopy (D). Bright field overlay with merged SYTO®
83-orange-stained nucleus and Alexa Fluor® 488-stained TLR2 (/), SYTO® 83-orange (//), Alexa Fluor® 488 (Hi), and
merged image (iv).
ase activity (Fig. 7A, panel D). Cell lysates had significantly less
luciferase from samples co-transfected with pMIR-TLR2
and pMIF-miR-105 (Fig. 75). This confirms our in silico pre-
DISCUSSION
This study identified miR-105 as a
modulator of TLR-2 protein transla-
tion in human gingival keratinocytes.
There was a strong inverse correla-
tion between cells that naturally had
high cytokine responses following
TLR-2 agonist challenge and miR-
105 levels. Knock-in and knock-down
of miR-105 confirmed this inverse
relationship. In silico analysis pre-
dicted that miR-105 had complemen-
tarity for TLR-2 mRNA, and the lucif-
erase reporter assay verified this.
Recently, miRNAs have been
shown to fine-tune innate immune
responses (40). For example, miR-
146a/b was up-regulated in an NF-
KB-dependent manner (21). In
another study, IL-6 induced let-7a-
modulated apoptosis in cholan-
giocytes (41) and up-regulated miR-
19a and -19b while down-regulating
SOCS1 (suppressor of cytokine sig-
naling 1), a gene important in nega-
tive regulation of TLR signaling
(42). The present study adds miR-
105 to the panel of miRNA species
known to influence innate immune
function.
Human miR-105 is located on
the intronic region of GABRA3A
(y-aminobutyric acid receptor 3a),
which resides on the X chromosome.
Certain types of tumor cells have been
shown to transcribe miR-105 but lack
processing machinery in the nucleus
to form mature miRNA (43). It is still
unclear how miR-105 is processed
and exported out of the nucleus in
gingival epithelial cells. Perhaps it is
analogous to miR-155, which is pres-
ent on the B-cell integration cluster
transcript up-regulated with polyri-
boinosinicpolyribocytidylic acid or the
cytokine interferon-/3 challenge.
The up-regulated B-cell integration
cluster transcript with miR-155 pre-
cursor undergoes processing to export mature miR-155 out of
the nucleus, which suppresses the macrophage inflammatory
response via c-Jun NH2-terminal kinase pathway (21).
a
o
a.
a>
a.
o
cr
p
b
<3
W
a>
T£
a>
3
cr
a>
AUGUST 21, 2009-VOLUME 784-NUMRFR 34
Previous
-------
miR- 705 in Human Oral Keratinocytes
B
5 —| CMV | Luciferase | 3'UTR of TLR2
r
pMIR-TLR2 I-GUUAUAAGAGUGGCAUAGUAUUUG ^1
pMIR-mutTLR2 I-GUUAUAAGAGUGGCAUAG( —del—)-^l
v«-
FIGURE 7. The putative miRNA-105 target site within the 3[prime]-UTR of
human TLR-2 mRNA (Ensembl transcript ID: ENST00000260010) and a pre-
dicted binding site mutated primers were synthesized, annealed, digested
with Spel and Hindlll, and ligated into the multiple cloning site of the
pMIRREPORT Luciferase vector. Cultured HEK293 cells with each of these
reporter constructs were co-transfected with pMIF-cGFP-Zeo-miR-105 plasmid
and assessed for Luciferase expression by confocal microscopy. The positive con-
trol had only pMIRREPORTER Luciferase construct (A (A)), The negative control
had pMIRREPORTER /3-galactosidase co-transfected with pMIF-cGFP-Zeo-miR-
105 (A (B)), cells with pMIRREPORTER Luciferase co-transfected with pMIF-cGFP-
Zeo-miR-105 plasmid showing decreased Luciferase expression (A (Q), and cells
with a mutated vector pMIRREPORTER Luciferase co-transfected with pMIF-
cGFP-Zeo-miR-105 vector retained Luciferase expression (A (D)). A Luciferase
reporter vector with potential binding site is represented in 6. The activity of
Luciferase was measured using Luciferase assay kit (Stratagene). The Luciferase
activity was significantly down-regulated in cells co-transfected with pMIF-cGFP-
Zeo-miR-105 (pMIF-miR-105) and pMIRREPORTER-TLR2 (plasmid containing
putative miR-105 binding region of TLR2) but not in cells co-transfected with
pMIF-miR-105 and pMIRREPORTER-mutTLR2 (plasmid containing mutated miR-
105 binding region of TLR-2) (Q. Results are mean ± S.D. of triplicates and are
representative of three independent experiments. Statistical comparisons are
shown by bars with asterisks above them (** indicatesp < 0.01;N5 = no significant
difference).
High basal secretion of IL-6 seen in deficient cell types may
be explained by multiple factors such as activation of NF-KB at
the basal level or cAMP signaling. IL-6 production is not solely
dependent on NF-KB, and the IL-6 gene in epithelial cells con-
tains cAMP-responsive elements on the promoter that are
important for its transcriptional regulation (35, 44).
Agonist availability and receptor compartmentalization are
pivotal in regulating TLR signaling (45). TLR itself may be
degraded, preventing ligand activation, or its protein expres-
sion may be inhibited (45). It has been shown that expression of
TLR-4 is necessary for intestinal homeostasis (46), increases in
TLR-4 are pathogenic in lupus-like autoimmune disease (47),
and increased TLR-2 levels are associated with the response to
vaccinia virus (48). Taken together, the receptor surface levels
regulate pathogen recognition and inflammatory responses.
Up-regulation of miR-203 has been shown to inhibit the sup-
pressor of cytokine signaling 3 (SOCS3) involved in inflamma-
tory responses and keratinocyte function (23), repress the
expression of p63-promoting differentiation of epithelium (49),
and repress sternness by targeting DeltaNp63 (50). In contrast
to miR-105, the TLR ligands tested did not up-regulate miR-
155 in our epithelial cell model, although both miR-146 a/b
were up-regulated (data not shown). This suggests that miR-
155 is not involved in the down-regulation of epithelial cell
cytokines, but miR-146 a/b might have a role in inhibiting
IRAKI and TRAF6 protein expression as previously observed
(21).
It is unclear how miRNA-105 itself is regulated in gingival
epithelial cells, because its precursor resides on intron I of the
GABRA3A gene within the X chromosome and has a neuro-
transmitter function (51). Lithium has been used as a potent
drug against affective neurological disorders (52). By inhibiting
glycogen synthase kinase 3 (53), lithium invokes an anti-inflam-
matory response (54). The y-aminobutyric acid receptor has
been shown to inhibit immune responses of T cells, and mod-
ulation of this receptor influences T-cell responses and autoim-
mune diseases (55). These data also suggest that intronic
regions are involved in regulating cell function.
Our diminished cytokine response cell findings reflect the
observation made with TLR-2 knock-out mice (37) in that the
low inflammatory response correlates with the low level of
TLR-2 expression in these cells. These data also confirm other
studies (21,22,34) showing that miRNAs may play a crucial role
in modulating immune function. Although miR-105 targets
TLR-2 and can induce down-regulation of cytokine production
in primary epithelial cell cultures, it is unlikely that low cytokine
response is solely explained by miR-105 regulation. Instead,
miR-105 may play an important role in fine-tuning TLR-2
response to control excessive inflammation. Further under-
standing of the mechanism of miR-105 and other targets may
lead to a better understanding of variations in inflammatory
responses within the oral mucosa and to new anti-inflamma-
tory therapeutics.
REFERENCES
1. Akira, S., Uematsu, S., and Takeuchi, O. (2006) Cell 124, 783-801
2. Kinane, D. F., Shiba, H., Stathopoulou, P. G., Zhao, H., Lappin, D. F., Singh,
A., Eskan, M. A., Beckers, S., Waigel, S., Alpert, B., and Knudsen, T. B.
(2006) Genes Immun. 7,190-200
o
o
a.
a>
a.
o
cr
p
b
(3
W
a>
T£
a>
3
cr
a>
23114 JOURNAL OF BIOLOGICAL CHFMISTRY
Previous
ME 284-NUMBER 34-AUGUST 21, 2009
-------
miR-W5in Human Oral Keratinocytes
3. Akira, S., and Takeda, K. (2004) Nat. Rev. Immunol. 4, 499-511
4. Bartel, D. P. (2004) Cell 116, 281-297
5. Liew, F. Y., Xu, D., Brint, E. K., and O'Neill, L. A. (2005) Nat. Rev. Immunol.
5,446-458
6. Lynn, D.}., Winsor, G. L., Chan, C, Richard, N., Laird, M. R., Barsky, A.,
Gardy, J. L., Roche, F. M., Chan, T. H., Shah, N., Lo, R., Naseer, M., Que,}.,
Yau, M., Acab, M., Tulpan, D., Whiteside, M. D., Chikatamarla, A., Mah,
B., Munzner, T., Hokamp, K., Hancock, R. E., and Brinkman, F. S. (2008)
Mol. Syst. Biol. 4,218
7. Farh, K. K., Crimson, A., Jan, C., Lewis, B. P., Johnston, W. K., Lim, L. P.,
Burge, C. B., and Bartel, D. P. (2005) Science 310,1817-1821
8. Gregory, R. L, Yan, K. P., Amuthan, G., Chendrimada, T., Doratotaj, B.,
Cooch, N., and Shiekhattar, R. (2004) Nature 432, 235-240
9. Chendrimada, T. P., Gregory, R. L, Kumaraswamy, E., Norman, J., Cooch,
N., Nishikura, K., and Shiekhattar, R. (2005) Nature 436, 740-744
10. Ambros, V. (2004) Nature 431, 350-355
11. Fritz, J. H., Girardin, S. E., and Philpott, D. J. (2006) Sci STKE 2006, pe27
12. Lee, Y., Kim, M., Han,}., Yeom, K. H., Lee, S., Baek, S. H., and Kim, V. N.
(2004) EMBOJ. 23, 4051-4060
13. Fazi, F., Rosa, A., Fatica, A., Gelmetti, V., De Marchis, M. L., Nervi, C., and
Bozzoni, I. (2005) Cell 123, 819-831
14. O'Donnell, K. A., Wentzel, E. A., Zeller, K. I., Dang, C. V., and Mendell,
J. T. (2005) Nature 435, 839 - 843
15. Pasquinelli, A. E., Hunter, S., and Bracht, J. (2005) Curr. Opin. Genet. Dev.
15,200-205
16. Chen, C. Z., Li, L., Lodish, H. F., and Bartel, D. P. (2004) Science 303,
83-86
17. Monticelli, S., Ansel, K. M., Xiao, C., Socci, N. D., Krichevsky, A. M., Thai,
T. H., Rajewsky, N., Marks, D. S., Sander, C., Rajewsky, K., Rao, A., and
Kosik, K. S. (2005) Genome Biol. 6, R71
18. Esau, C., Kang, X., Peralta, E., Hanson, E., Marcusson, E. G., Ravichandran,
L. V., Sun, Y., Koo, S., Perera, R. J., Jain, R., Dean, N. M., Freier, S. M.,
Bennett, C. F., Lollo, B., and Griffey, R. (2004) /. Biol. Chem. 279,
52361-52365
19. Calin, G. A., Sevignani, C., Dumitru, C. D., Hyslop, T., Noch, E., Yen-
damuri, S., Shimizu, M., Rattan, S., Bullrich, F., Negrini, M., and Croce,
C. M. (2004) Proc. Natl. Acad. Sci. U.S.A. 101, 2999 -3004
20. Lu, L, Getz, G., Miska, E. A., Alvarez-Saavedra, E., Lamb, L, Peck, D.,
Sweet-Cordero, A., Ebert, B. L., Mak, R. H., Ferrando, A. A., Downing, J. R.,
Jacks, T., Horvitz, H. R., and Golub, T. R. (2005) Nature 435, 834-838
21. Taganov, K. D., Boldin, M. P., Chang, K. J., and Baltimore, D. (2006) Proc.
Natl. Acad. Sci. U.S.A. 103,12481-12486
22. O'Connell, R. M., Taganov, K. D., Boldin, M. P., Cheng, G., and Baltimore,
D. (2007) Proc. Natl. Acad. Sci. U.S.A. 104,1604-1609
23. Sonkoly, E., Wei, T., Janson, P. C., Saaf, A., Lundeberg, L., Tengvall-Linder,
M., Norstedt, G., Alenius, H., Homey, B., Scheynius, A., Stable, M., and
Pivarcsi, A. (2007) PLoS ONE 2, e610
24. Kinane, D. F. (2000) Ann. R. Australas Coll. Dent. Surg. 15, 42-50
25. Kinane, D. F., Galicia, J. C., Gorr, S. U., Stathopoulou, P. G., and Benaka-
nakere, M. (2008) Front. Biosci. 13, 966 -984
26. Spratt, D. (2003) in Medical Biofilms: Detection, Prevention, and Control
(Jass, J., Surman, S., and Walker, J., eds) pp. 175-198, Wiley and Sons, Ltd.,
London
27. Kinane, D. F., and Hart, T. C. (2003) Crit. Rev. Oral Biol. Med. 14,430-449
28. Shiba, H., Venkatesh, S. G., Gorr, S. U., Barbieri, G., Kurihara, H., and
Kinane, D. F. (2005) /. Periodontal Res. 40, 153-157
29. Eskan, M. A., Rose, B. G., Benakanakere, M. R., Zeng, Q., Fujioka, D.,
Martin, M. H., Lee, M. J., and Kinane, D. F. (2008) Eur. J. Immunol. 38,
1138-1147
30. Eskan, M. A., Benakanakere, M. R., Rose, B. G., Zhang, P., Zhao, J., Statho-
poulou, P., Fujioka, D., and Kinane, D. F. (2008) Infect. Immun. 76,
2080-2089
31. Eisen, M. B., Spellman, P. T., Brown, P. O., and Botstein, D. (1998) Proc.
Natl. Acad. Sci. U.S.A. 95,14863-14868
32. Singh, A. V., Green, M., States, J. C., and Knudsen, T. B. (2007) Birth
Defects Res. Part A. Clin. Mol. Teratol. 79, 319-350
33. Livak, K. J., and Schmittgen, T. D. (2001) Methods 25, 402-408
34. Chen, X. M., Splinter, P. L., O'Hara, S. P., and LaRusso, N. F. (2007) /. Biol.
Chem. 282, 28929-28938
35. Eskan, M. A., Hajishengallis, G., and Kinane, D. F. (2007) Infect Immun 75,
892-898
36. Darveau, R. P., Pham, T. T., Lemley, K., Reife, R. A., Bainbridge, B. W.,
Coats, S. R., Howald, W. N., Way, S. S., and Hajjar, A. M. (2004) Infect.
Immun. 72, 5041-5051
37. Burns, E., Bachrach, G., Shapira, L., and Nussbaum, G. (2006)/. Immunol.
177,8296-8300
38. Brozovic, S., Sahoo, R., Barve, S., Shiba, H., Uriarte, S., Blumberg, R. S., and
Kinane, D. F. (2006) Microbiology 152, 797-806
39. Sandros, J., Karlsson, C., Lappin, D. F., Madianos, P. N., Kinane, D. F., and
Papapanou, P. N. (2000) /. Dent. Res. 79,1808-1814
40. Can tier, M. P., Sadler, A. J., and Williams, B. R. (2007) Immunol. Cell Biol.
85,458-462
41. Meng, F., Henson, R., Wehbe-Janek, H., Smith, H., Ueno, Y., and Patel, T.
(2007)/. Biol. Chem. 282, 8256-8264
42. Pichiorri, F., Suh, S. S., Ladetto, M., Kuehl, M., Palumbo, T., Drandi, D.,
Tacdoli, C., Zanesi, N., Alder, H., Hagan, J. P., Munker, R., Volinia, S.,
Boccadoro, M., Garzon, R., Palumbo, A., Aqeilan, R. I., and Croce, C. M.
(2008) Proc. Natl. Acad. Sci. U.S.A. 105,12885-12890
43. Lee, E. J., Baek, M., Gusev, Y., Brackett, D. J., Nuovo, G. J., and Schmittgen,
T. D. (2008) RNA 14, 35-42
44. Krueger, J., Ray, A., Tamm, L, and Sehgal, P. B. (1991) /. Cell. Biochem. 45,
327-334
45. Miggin, S. M., and O'Neill, L. A. (2006) /. Leukoc. Biol. 80, 220 -226
46. Rakoff-Nahoum, S., Paglino, J., Eslami-Varzaneh, F., Edberg, S., and
Medzhitov, R. (2004) Cell 118, 229-241
47. Liu, B., Yang, Y., Dai, J., Medzhitov, R., Freudenberg, M. A., Zhang, P. L.,
and Li, Z. (2006) /. Immunol. 177, 6880 - 6888
48. Zhu, J., Martinez, J., Huang, X., and Yang, Y. (2007) Blood 109, 619-625
49. Yi, R., Poy, M. N., Stoffel, M., and Fuchs, E. (2008) Nature 452, 225-229
50. Lena, A. M., Shalom-Feuerstein, R., di Val Cervo, P., Aberdam, D., Knight,
R. A., Melino, G., and Candi, E. (2008) Cell Death Differ. 15,1187-1195
51. Bell, M.V., Bloomfield, J., McKinley, M., Patterson, M. N., Darlison, M. G.,
Barnard, E. A., and Davies, K. E. (1989) Am. J. Hum. Genet. 45, 883-888
52. Rowe, M. K., and Chuang, D. M. (2004) Expert Rev. Mol Med. 6,1-18
53. Phiel, C. J., and Klein, P. S. (2001) Annu. Rev. Pharmacol. Toxicol. 41,
789-813
54. Martin, M., Rehani, K., Jope, R. S., and Michalek, S. M. (2005) Nat. Immu-
nol. 6, 777-784
55. Tian, J., Chau, C, Hales, T. G., and Kaufman, D. L. (1999) /. Neuroimmu-
HO/. 96, 21-28
a
o
a.
a>
a.
o
cr
p
b
<3
W
a>
T£
a>
3
cr
a>
K)
o
AUGUST 21, 2009-VOLUME 784-NUMRFR 34
Previous
TOC
IAL OF BIOLOGICAL CHEMISTRY 23115
-------
Reproductive Toxicology 27 (2009) 373-386
I I N! Ml R
Contents lists available at ScienceDirect
Reproductive Toxicology
journal homepage: www.elsevier.com/locate/reprotox
Pharmacokinetic modeling of perfluorooctanoic acid during gestation and
lactation in the mouse1^
Chester E. Rodriguez*, R. Woodrow Setzer, Hugh A. Barton1
US Environmental Protection Agency, Office of Research and Development, National Center for Computational Toxicology, Research Triangle Park, NC27711, United States
ARTICLE INFO
Article history:
Received 14 November 2008
Received in revised form 3 February 2009
Accepted 19 February 2009
Available online 4 March 2009
Keywords:
Perfluorooctanoic acid
Pharmacokinetics
Modeling
Mice
Rats
Developmental toxicity
Gestation
Lactation
ABSTRACT
Perfluorooctanoic acid (PFOA) is a processing aid for the polymerization of commercially valuable flu-
oropolymers. Its widespread environmental distribution, presence in human blood, and adverse effects
in animal toxicity studies have triggered attention to its potential adverse effects to humans. PFOA is
not metabolized and exhibits dramatically different serum/plasma half-lives across species. Estimated
half-lives for humans, monkeys, mice, and female rats are 3-5 years, 20-30 days, 12-20 days, and 2-4 h,
respectively. Developmental toxicity is one of the most sensitive adverse effects associated with PFOA
exposure in rodents, but its interpretation for risk assessment is currently hampered by the lack of
understanding of the inter-species pharmacokinetics of PFOA. To address this uncertainty, a biologically
supported dynamic model was developed whereby a two-compartment system linked via placental blood
flow described gestation and milk production linked a lactating dam to a growing pup litter compart-
ment. Postnatal serum levels of PFOA for 12951/SvImJ mice at doses of 1 mg/kg or less were reasonably
simulated while prenatal and postnatal measurements for CD-I mice at doses of 1 mg/kg or greater were
simulated via the addition of a biologically based saturable renal resorption description. Our results sug-
gest that at low doses a linear model may suffice for describing the pharmacokinetics of PFOA while a
more complex model may be needed at higher doses. Although mice may appear more sensitive based
on administered dose of PFOA, the internal dose metrics estimated in this analysis indicate that they may
be equal or less sensitive than rats.
Published by Elsevier Inc.
1. Introduction
Perfluorooctanoic acid (PFOA) is a synthetic, fully fluorinated
alkyl acid that has been produced industrially for several decades
for use primarily as an emulsifier in the aqueous polymeriza-
tion of fluoropolymers such as polytetrafluoroethylene (Teflon®).
Fluoropolymers made through the use of PFOA exhibit valu-
able commercial properties that include water and oil repellency,
high stability, and inertness, and thus find extensive utilization
* The United States Environmental Protection Agency through its Office of
Research and Development funded and managed the research described here. It
has been subjected to Agency administrative review and approved for publication.
Approval does not signify that the contents reflect the views of the Agency, nor
does mention of trade names or commercial products constitute endorsement or
recommendation for use.
* Corresponding author at: The National Center for Computational Toxicology
(B205-1), Office of Research and Development, US Environmental Protection Agency,
109 T.W. Alexander Dr., Research Triangle Park, NC 27711, United States.
Tel.: +1 919 541 0447: fax: +1 919 541 1194.
E-mail addresses: rodriguez.chester@epa.gov (C.E. Rodriguez),
habarton@alum.mit.edu (H.A. Barton).
1 Current address: Pfizer, Inc., PDM PK/PD Modeling, Eastern Point Road,
MS 8220-4328, Groton, CT 06340, United States.
0890-6238/S - see front matter. Published by Elsevier Inc.
doi:10.1016/j.reprotox.2009.02.009
in numerous sectors including aerospace, automotive, build-
ing/construction, chemical processing, electrical and electronics,
semiconductor, textile, biomedical, and others. The manufacture
and use of PFOA have been accompanied by considerable direct
emissions of this chemical to the environment. Since 1951 when
PFOA began to be used industrially, it has been estimated that
cumulative global emissions are in the range of 2400-5200 tonnes
[1].
Emissions of PFOA may also be indirect in nature, through the
manufacture and use of precursor compounds that can be degraded
to PFOA under normal environmental conditions. One of these com-
pounds is 8-2 fluorotelomer alcohol (8:2-FTOH) which has been
produced for surface protection applications. 8:2-FTOH is volatile
and has been shown to readily undergo atmospheric oxidation to
PFOA [2,3]. The biotic degradation of 8:2-FTOH in aquatic media has
also been shown to yield PFOA as a terminal product [4].
The chemical properties exhibited by PFOA include excep-
tional stability, low volatility, and inertness which are ideal for
its intended commercial applications, but have also resulted in a
non-biodegradable and persistent environmental pollutant [5,6].
Recent reports of widespread environmental distribution, presence
in human blood and wildlife samples, and toxic effects in laboratory
animal studies have generated significant toxicological and regu-
Previous
TOC
-------
374
C.E. Rodriguez eta/./Reproductive Toxicology 27(2009)373-386
latory interest. Particularly concerning is the long serum half-life
observed in humans (estimated at ~3-5 years) for whom reported
serum levels range from low parts per billion for the general US pop-
ulation to low parts per million for occupationally exposed workers
and other highly exposed populations [7-10]. Thus, even though
human serum levels of PFOA maybe considered low, a long half-life
could be indicative of the tendency of this chemical to bioaccu-
mulate, potentially leading to higher body burdens and associated
long-term health risks.
The pharmacokinetics of PFOA is unusual in that serum/plasma
clearance can vary dramatically across species, and for some, across
gender. Among the species examined to date, humans exhibit
the longest plasma half-life at 3-5 years, followed by monkeys
at 21-30 days, and mice at 12-20 days [9-11]. The suggestion
of concentration-dependent changes in renal excretion, however,
means these half-life estimates may reflect clearance only at rela-
tively low plasma concentrations [12]. In the case of the rat, a very
dramatic gender difference in plasma clearance is observed. The
male adult rat clears PFOA with a plasma half-life of about 6 days
while a half-life that is more than 30-fold faster at 2-4 h is observed
for its female counterpart [9,10]. It should be noted that clearance
of PFOA likely reflects removal of only the parent compound since
the carbon-fluorine bond is too strong for mammalian metabolic
systems to cleave and no metabolites have been reported to date
[13]. The underlying mechanisms for the differential retention of
PFOA across species/sexes remain to be conclusively demonstrated
but evidence suggests a role for organic anion transporters in the
liver and kidneys [12,14,15].
Developmental toxicity has been demonstrated in rodent
species to be one of the most sensitive adverse effects associated
with PFOA exposure. When coupled to human studies reporting low
levels of PFOA in umbilical cord blood [16], neonatal blood collected
immediately after birth [17], and breast milk [18,19], the need to
understand the significance and relevance of animal developmen-
tal toxicity studies becomes more critical for doing human health
risk assessment. In particular, the implications of the differences in
plasma clearance on findings from developmental toxicity studies
carried out in rodents have been largely unaddressed and represent
a substantial source of uncertainty in risk analysis of PFOA. In the
rat, reported developmental effects from two generation reproduc-
tive toxicity studies at an administered daily oral gavage dose of
30mg/kg-d PFOA include deleterious effects in the Fl generation
in the form of increased post-weaning mortality, pre- and post-
weaning body weight (BW) deficits, and delayed sexual maturation.
Fewer developmental effects were reported at 10 mg/kg-d, and little
or no effects at 3 mg/kg-d [20]. In an inbred mouse strain, increased
neonatal mortality has been reported at an administered oral gav-
age dose as low as 0.6 mg/kg-d [21], and neonatal BW deficits and
eye opening delays are among the effects reported at 1 mg/kg-d
[21,22]. At higher administered doses, mammary gland develop-
ment alterations in both lactating dams and female pups have
been reported [23] along with prenatal effects that can range from
maternal weight gain deficits and delayed parturition to full-litter
resorption [22,24]. For all developmental endpoints examined,
mice were more sensitive than rats based on administered doses of
PFOA.
The interpretation of animal developmental toxicity studies
can greatly benefit from information on the pharmacokinetics of
PFOA during gestation and lactation particularly for reconciling the
disparities of the large interspecies differences in plasma clear-
ance. In the case of the female rat, for instance, it is unclear if
the higher administered doses required to induce developmental
toxicity as compared to mice are reflective of the faster plasma
clearance of PFOA in this species and/or lower inherent sensitivity.
Indeed, with a half-life of 2-4 h, the female rat being dosed daily
can be expected to clear most of the chemical by 24 h, resulting
in a repeated episodic pharmacokinetic profile that would con-
trast with the bioaccumulative profile expected for other species
with much slower plasma clearance. Consequently, toxicity findings
from rat studies are difficult to interpret especially for any type of
cross-species extrapolation. In contrast, the female mouse has been
proposed to be more amenable for interpretation since its pharma-
cokinetic profile should resemble that of other species including
humans in its bioaccumulative nature [22]. From a risk assessment
perspective, it is critical to select the most appropriate laboratory
species for cross-species extrapolation of toxic effects, and pharma-
cokinetic information can be very helpful to that end in the case of
PFOA.
Biologically based pharmacokinetic modeling may provide the
best approach for addressing the lack of understanding of the inter-
species pharmacokinetics of PFOA during gestation and lactation in
animal developmental toxicity studies. A pharmacokinetic model
that describes the critical physiological and anatomical changes
associated with gestation and lactation can be very useful for
estimating internal dose as a means of reconciling the large dif-
ferences in plasma clearance across species. Such a model can also
be used for estimating the dose to the offspring in order to eval-
uate the impact of the default risk assessment approach that is
based solely on the maternal dose, even when the adverse effects
are observed in the offspring. Since the critical window of expo-
sure for postnatally observed developmental effects (e.g., neonatal
BW deficits and neonatal lethality) may be during the prenatal
period, a pharmacokinetic model can also be used for estimat-
ing the appropriate dose metric (prenatal versus postnatal internal
dose) for a given adverse effect. At a minimum, a pharmacokinetic
model of gestation and lactation can be used to maximize the util-
ity of the limited information available and help prioritize research
needs.
This report describes the development and application of a
biologically based two-compartment model for describing the
pharmacokinetics of PFOA during gestation and lactation in the
mouse. The model is based on limited data available but incor-
porates the critical changes in growth, placental blood flow,
PFOA-partitioning, and milk production. Gestation was described
with a pregnant dam and a concept! compartment linked via
placental blood flow while milk production linked a lactating
dam to a growing pup litter compartment. The overall goal for
developing the model was to address some of the uncertain-
ties associated with the pharmacokinetics of PFOA in animal
developmental toxicity studies. The specific objectives were as
follows:
to explore the possibilities in model structure that can be
supported by serum information reported from mouse develop-
mental toxicity studies;
to provide estimates of internal dose for dam and offspring during
gestation and lactation;
to compare internal dose estimates to those reported for the rat as
a means for addressing differences in plasma clearance in these
two species;
to help identify research needs for the development of a more
elaborate pharmacokinetic model.
The results suggest that at lower doses of PFOA a linear model
may suffice for describing the pharmacokinetic behavior of PFOA
in gestation and lactation, while a renal resorption component may
be necessary to describe the non-linear behavior at higher doses.
The model was used to derive estimates of internal dosimetry for
dams as well as pups. Some of the implications of these findings as
well as the information gaps encountered are discussed.
Previous
-------
C.E. Rodriguez eta/./Reproductive Toxicology 27(2009)373-386
375
Table 1
Mathematical expressions describing the growth of maternal tissues during gestation.
Tissue
Uterus (vu)
Mammary (vmg)
Carcass fat (vf)
Liver (vl)
Pre-pregnancy fraction of BWa
0.002
0.01
0.07
0.04
Mathematical expression3'1"'0
pvu =vu x (1 +0.077 x (gd-3)1-6)
pvmg = vmg x ( 1 + 0.27 x gd)
pvf=vfx(l +0.0165 xgd)
pvl = vlx(l +0.0255 x(gd-6))
Duration
gd3-18
gdO-18
gdO-18
gd6-18
Taken from [25]; gd denotes gestation day.
The designation p denotes tissue weight during pregnancy.
vu, vmg, vf, and vl are the respective pre-pregnancy tissue weights.
2. Materials and methods
2.1. Model code
All pharmacokinetic models were coded and implemented in acslX (Version
2.04, Aegis Technologies, Huntsville, AL). The total simulation time was set to 39 full
days (or 936 h). Gestation constituted the first 18 full days of simulation with the first
24 h designating gestation day 1 (gdl). Parturition was set to take place at the end of
gd!8 with the subsequent 24 h constituting postnatal day 1 (pndl). The appropriate
24 h adjustment was accounted for in cases where experimental data was obtained
using the designation of gdO and/or pndO for the 24 h following mating and birth,
respectively.
3. Model structure
3.1. Gestation
Gestation was described as a two-compartment system consist-
ing of pregnant dam and concept! linked by placental plasma flow
(Qcon). The pregnant dam compartment accounted for maternal
tissues that increase in growth during gestation, namely uterus,
mammary tissue, carcass fat, and liver. Mathematical expressions
describing the growth of these tissues during gestation have been
previously reported and are listed along with their respective pre-
pregnancy fractions of BW in Table 1 [25]. BW values of 23.0 and
26.7 g were used for 12951/SvImJ and CD-I adult non-pregnant
female mice, respectively, based on information from respective
studies [21,24].
The concept! compartment consists of embryo/fetuses and all
of the associated tissues including placentas and its growth was
estimated as the difference between total pregnant female BW
and the contribution of maternal tissues beginning with the non-
pregnant adult female mouse BW. Raw BW measurements for
12951/SvImJ pregnant mice throughout gestation were kindly pro-
vided by Abbott et al. [21] and analyzed using the R statistical
software package [26]. The resulting mathematical expression that
describes the BW changes during gestation is listed in Table 2.
Similarly for CD-I mice, raw BW measurements kindly pro-
vided by Lau et al. [22] were analyzed for fit using R to the
expression listed in Table 2. BW information of only untreated
groups was considered for this analysis since maternal weight
gain deficits were not observed in 129Sl/SvImJ pregnant mice
[21] and only at the higher administered doses of PFOA in preg-
nant CD-I mice [22]. Furthermore, in modeling treated groups,
only those administered doses that did not affect litter size are
being considered, namely 0.1-1.0 and l-10mg/kg-d PFOA for
129Sl/SvImJ and CD-I mice, respectively [21,22]. Since gestation
is being described strictly as a flow-limited model, the concept!
was modeled beginning with the onset of placenta blood flow and
any growth before this time (gdO-5) was attributed to maternal
tissues.
The development of placental blood flow was described by a
mathematical expression beginning at gd6 as previously reported
[25]. Accounting for differences in gestation times, the predicted
profile was in agreement with a measurement reported in the
published literature of 1.26 ± 0.54 mL/min/g placenta for pregnant
Balb/c mice at gd!6 [27]. If the 21-day gestation period of Balb/c
mice is assumed to scale proportionally to the 18-day gestation
period used in this model for 129Sl/SvImJ and CD-I mice, gd!6
would be equivalent to the modeled gd!3. Such scaling has been
described previously for addressing differences in gestation times
for rodents [25]. Using a placental weight of 0.0858 g for gd!3 [28],
the reported placental blood flow measurement was expressed as
0.108±0.046mL/min and plotted along with the predicted profile
(Fig. 2B). Reported serum levels of PFOA were simulated assuming
a hematocrit fraction of 0.45 to obtain the corresponding plasma
flow rate.
3.2. Lactation
Lactation was described as a dynamic dam and pup litter com-
partment linked by a milk compartment (Fig. 3). BW information
for lactating dams was derived from a study with MF1 mice [29] and
implemented in the model by applying the same percent increase
to the predicted BW of pregnant 12951/SvImJ and CD-I mice at
the end of gestation (excluding concept!). This manipulation of the
data avoids discontinuities in the model that may result from strain
differences. The resulting BW values for 12951/SvlmJ and CD-I lac-
tating dams were analyzed for fit to the mathematical expressions
listed in Table 2 (Graphpad Prism, San Diego, CA).
Raw BW measurements for 12951/SvImJ and CD-I pups
throughout lactation were obtained from the respective toxicity
studies [21,22] and analyzed for fit to the corresponding expres-
sions listed in Table 2 using the R statistical software package [26].
Pup BW information of only the untreated group was considered
for this analysis since BW gain deficits are less than 25% at the
higher administered doses to the dam (i.e., 1 mg/kg-d and 5 mg/kg-
d or higher for 12951/SvImJ and CD-I mice, respectively) [21,22,24].
Moreover, it is unknown at this time if the observed BW gain deficits
of the nursing pups are reflective of lower milk consumption which
would decrease the lactational transfer of PFOA. In the absence of
Table 2
Mathematical expressions used to simulate body weight changes in pregnant/lactating dams and nursing pups.
Mathematical expression (g)
CD-I pregnant dam
CD-I lactating dam
CD-I pup
12951/SvImJ pregnant dam
12951/SvImJ lactating dam
12951/SvImJ pup
BW = 31.08 +1.07 x (gd-10) + 0.136 x (gd-10)2 +0.00582 x (gd-10)3 - 0.000379 x (gd-10)4
BW = 30.40 + 0.78 x pnd-0.028 x pnd2
BW= 1.235 + 0.22 x pnd + 0.0654 x pnd2 - 0.0055 x pnd3 + 0.000145 x pnd4
BW= 24.96 + 0.56 x (gd-10) + 0.0653 x (gd-10)2 + 0.000846 x (gd-10)3 - 0.000268 x (gd-10)4
BW= 25.2 + 0.548 x pnd + 0.00234 x pnd2 - 0.000963 x pnd3
BW= 1.42- 0.189 x pnd + 0.114 x pnd2 - 0.00673 x pnd3 +0.000129 x pnd4
Previous
TOC
-------
376
C.E. Rodriguez eta/./Reproductive Toxicology 27(2009)373-386
such information, pup BW changes are being modeled as those of
the untreated group for both mouse strains.
Mouse milk yield information for a litter size of 10 was obtained
from a report by Knight et al. [30], expressed on a per pup basis,
and fitted to the following cubic expression (Graphpad prism):
MY(pnd) = 0.484 + 0.124 x pnd - 0.00510 x pnd2
+ 0.0000570 xpnd3
MY and pnd denote milk yield in gram/day/pup and postnatal day,
respectively. In the absence of strain-specific MY information, the
expression of MY on a per pup basis is a reasonable way of mini-
mizing the impact of strain-dependent differences in litter size.
It was assumed that all the milk produced was consumed by the
pups without loss or delay. The rate of milk production by the dam
(and suckling rate of nursing pups) was assumed constant and mod-
eled as continuous without any circadian variation, as previously
modeled for rats [31 ].
3.3. Pup excreta recirculation
Neonatal rodents are reportedly unable to eliminate waste with-
out maternal stimulation which involves the dams consuming pup
urine [32,33]. This process was modeled as previously reported [31 ]
as an additional input to the dam whereby the amount of PFOA
eliminated by the pups during the first two weeks of life would be
transferred back to the lactating dam.
3.4. Renal resorption
The kidney resorption model was adapted from a report by
Russel et al. [34]. It was first implemented for 12951/SvlmJ adult
non-pregnant female mice assuming a constant BW of 21 g and the
physiological parameters listed in Table 3. The sum of glomerular
filtration rate and renal plasma flow rate equals the total kidney
plasma flow forthe mouse (assuming a hematocrit fraction of 0.45)
[35].
4. PFOA-speciflc pharmacokinetic parameters
Absorption and elimination of PFOA were described as first
order processes. PFOA is well-absorbed with a time of maximum
observed concentration of less than 4h in adult mice [11]. An
absorption rate constant of 0.537/h estimated for the adult CD-I
mouse was assumed to be the same for all pregnant and lactating
dams [11 ]. In the absence of any absorption information for nursing
pups, all of the PFOA contained in milk was assumed to be absorbed
without loss or delay.
A volume of distribution of 0.135 L/kg BW estimated for adult
non-pregnant CD-I mice [ 11 ] was assumed to also apply to all mice
including concept! and nursing pups.
Table 3
Physiological parameters used to describe renal resorption in the mouse.
Renal resorption parameter Value
Table 4
Embryo/fetus:maternal serum partition coefficients (Pe/f) for PFOAa.
Cardiac output (CCC, L/h/kg0-75)
Kidney blood flow (QRC, fraction of QC)
Glomerular filtration rate (GFR, L/h/kg BW
Urine flow rate (L/h/kg BW)
Volume of renal plasma (fraction of BW)
Volume of renal filtrate (fraction of BW)
16.5'
0.091 *
0.378b
0.000303C
0.00067d
0.000097d
* Obtained from [35].
b Obtained from [47].
c Obtained from [48].
d Estimated by BW-scaling of values reported forthe dog [34].
gd9
gdlO
gd!3
gd!5
gd!8
Pe/f
0.05
0.07
0.06
0.10
0.20
' Estimated from [36].
4.1. Gestation
The elimination rate constant for 129Sl/SvlmJ pregnant mice
was estimated to be 0.00233/h by optimization (acslX, Version 2.04,
Aegis Technologies) against serum levels of PFOA in non-lactating
dams (pregnant dams that did not give birth to viable pups or
pups died shortly after birth) at pnd22. These pregnant dams were
dosed daily from gdl-17 with 0.1, 0.3, 0.6, and l.Omg/kg PFOA
[21]. The estimated elimination rate constant was scaled allomet-
rically to account for BW changes associated with gestation and
lactation.
An elimination rate constant of 0.00185/h estimated for adult
CD-I mice [11] was also assumed to apply to CD-I nursing pups.
Disposition of PFOA during gestation was described using lin-
ear interpolation (TABLE function, acslX) of embryo/fetus:maternal
serum partition coefficients (designated as Pe/f) that were esti-
mated from serum information kindly provided by M. Henderson
[36] (Table 4). In the absence of any other information, the same
Pe/f value as gd9 was assumed for gd6-8.
4.2. Lactation
Lactational transfer was described as a first order process. The
first order rate constant was estimated as previously described for
rats [31] via the following expression:
Pm x Vm
klac =
24 x Vdam
where Pm refers to the millcmaternal serum partition coefficient
(assumed constant throughout lactation). Vm is the volume of milk
in liters produced in 24 h obtained from mouse milk yield informa-
tion already described [37]. Vdam is the volume of distribution of
PFOA in liters for the lactating dam assuming a value of 0.135 L/kg
BW [11 ]. The only unknown in the klac expression is Pm which, in
the absence of any milk-partitioning information, was estimated to
be 0.0285 by optimization against serum levels of lactating dams
and nursing pups [21]. The same value ofPm was assumed for CD-I
mice. A Pm value of 0.0285 is much lower than that estimated for
rats at 0.1 [38].
4.3. Renal resorption
The transport affinity constant (Kt) and transport maximum
(Tmc) for 129Sl/SvlmJ adult non-pregnant mice were estimated
by optimization (acslX, Version 2.04, Aegis Technologies) against
serum levels of PFOA for mice dosed daily from gdl-17 with 0.1,
0.3, 0.6, and l.Omg/kg PFOA but whose litters were fully resorbed
early in pregnancy. The pharmacokinetics of PFOA in these mice
can be considered as adult non-pregnant mice since they did not
undergo the BW changes associated with pregnancy and lactation.
Serum measurements were kindly provided by Abbott et al. [21].
The resulting values for Kt and Tmc are listed in Table 5. With the
exception of Kt and Tmc, the same parameter values were used
for modeling pregnant and lactating CD-I mice, but accounting for
the respective BW changes. Both Kt and Tmc were first optimized
against reported maternal serum levels of PFOA measured at term
corresponding to administered doses of 1,3, 5 and 10 mg/kg-d PFOA
[22]. Since postnatal serum measurements only included two doses,
namely 3 and 5 mg/kg-d [24], only Tmc was optimized and the same
Previous
-------
C.E. Rodriguez eta/./Reproductive Toxicology 27(2009)373-386
377
Table 5
Renal resorption parameters used to simulate serum levels of PFOA in 12951/SvlmJ
and CD-I mice.
Reference
Mouse strain
Serum sampling time
I
-------
378
C.E. Rodriguez eta/./Reproductive Toxicology 27(2009)373-386
Qcon
Oral gavage dose
(mg/kg)
Concept!
Vcon Ccon
Dam
Cdam Vdam
Qcon
ked ,
Urine
Pup excreta
recirculation
(from birth to PND14)
klac
Milk
Cmilk Vm
klac
Pups
Cpup Vpup
kep
Urine
Fig. 3. Pharmacokinetic model of gestation and lactation used in the analysis of 12951/SvlmJ mice.
Lactation was similarly described following parturition at the
end of gd!8 as dynamic lactating dam and pup litter compart-
ments, but linked to a milk compartment (Fig. 3, bottom portion).
Lactational exposure of PFOAto nursing pups was modeled assum-
ing that all of the milk was produced at a constant daily rate and
consumed by nursing pups without loss or delay, as previously
described for rats [31]. The BW changes for lactating dams and
nursing pups are shown in Fig. 4A-D, respectively, and the corre-
sponding mathematical expressions that describe them are listed
in Table 2. Since the inability of neonatal mice in the early postna-
tal period to eliminate waste without maternal stimulation likely
results in maternal re-exposure to PFOA [32], this phenomenon was
modeled postnatally as an additional dose to the lactating dam, as
previously described for rats (Fig. 3) [31 ].
§
40-|
:' >
-
'a> to
73
o
14-
12-
10-
8-
6-
4-
2-
0
(B)
Weaning
(D)
weaning
Abbott et al. 2007
0 2 4 6 8 10 12 14 16 18 20 22
Postnatal day
Q.
3
Q.
Q
O
•o
o
DO
14-
12-
10-
8-
6-
4-
2-
0
Lau et al. 2006
0 2 4 6 8 10 12 14 16 18 20 22 24
Postnatal day
Fig. 4. Simulation of body weight changes for pregnant/lactating dams and nursing pups.
Previous
TOC
-------
C.E. Rodriguez eta/./Reproductive Toxicology 27(2009)373-386
379
Non-pregnant
* Abbott etal. 2007
~~ Pregnant/non-lactating dam
A Abbott et al. 2007
(A) 1.0 mg/kg/day
~ ~ Pregnant/Lactating dam
0 Abbott et al. 2007
• • • Pup
• Abbott et al. 2007
(B) 0.6 mg/kg/day
100-,
60-1
40-
20-
Gestation Lactation
200
400 600
time (hr)
800 1000
(C) 0.3 mg/kg/day
0 200 400 600
time (hr)
(D) 0.1 mg/kg/day
800
1000
B)
a>
2.0-
89
0.0
Gestation
Lactation
200
400 600
time (hr)
800
1000
200
400 600
time (hr)
800
1000
Fig. 5. Simulation of serum levels of PFOA in 12951/SvlmJ mice.
5.2. Comparison of model-simulated profiles
Three mouse developmental toxicity studies have been reported
and some serum information is available to evaluate the pharma-
cokinetic model of gestation and lactation [21,22,24]. The studies
have been carried out with pregnant 129Sl/SvImJ and CD-I mice.
In both cases, the pregnant dam was dosed daily with different con-
centrations of PFOA from gdl-17 and serum levels of PFOA were
measured either prenatally after 1 day (i.e., at term) or postnatally at
weaning, (i.e., 23 days following the last treatment). No single study
provided both prenatal and postnatal serum measurements of
PFOA. Nonetheless, the study with 129Sl/SvlmJ pregnant mice [21]
provided the most thorough serum information for examining the
pharmacokinetic behavior of PFOA during gestation and lactation.
Reported serum measurements included mice that were non-
pregnant, pregnant but did not lactate (pregnant/non-lactating),
pregnant then lactating (pregnant/lactating), and corresponding
nursing pups. Pregnant/non-lactating mice were those that were
pregnant but whose pups were stillborn or died soon after birth,
and therefore, did not lactate or their lactation was not consid-
ered significant [21 ].Thus, although serum levels of PFOA were only
measured at only one time point, namely pnd22, in combination,
these limited data can be used to examine the pharmacoki-
netic behavior of PFOA during gestation and lactation within a
given model structure. Thus, the model structure in Fig. 3 which
accounts for the critical changes in BW, placental blood flow, PFOA-
partitioning, and milk production was evaluated for its ability to
simulate serum levels of PFOA reported in 129Sl/SvImJ mice [21].
As shown in Fig. 5A-D, the model reasonably simulated serum lev-
els of PFOA for all pregnant mice and nursing pups. Only serum
levels corresponding to doses of 0.1-1 mg/kg-d were available for
simulation since full-litter resorption was observed very early in
gestation at higher doses [21 ].
The serum information provided for non-pregnant female mice
is of significance because in addition to allowing comparison with
pregnant mice at doses <1 mg/kg-d, it also allows an examination of
the pharmacokinetic behavior of PFOA at higher doses (>1 mg/kg-
d) that would not be possible in pregnant mice because of the
induction of full litter resorption resulting from PFOA exposure
[21]. Thus, a linear one-compartment model based on a constant
BW of 21 g was initially evaluated for its ability to simulate serum
levels of PFOA collected for non-pregnant mice at 23 days follow-
ing the last treatment (the equivalent of pnd22), but this model
structure failed to simulate non-linear behavior of serum levels
which became more apparent at the higher doses (data not shown).
Extrapolating from previous findings that implicate transporters in
the renal clearance of PFOA [14] and modeling efforts that involve
renal resorption to explain the non-linear pharmacokinetic behav-
Previous
TOC
-------
380
C.E. Rodriguez eta/./Reproductive Toxicology 27(2009)373-386
Kidney
Urine
Fig. 6. Pharmacokinetic model used in the analysis of 12951/SvlmJ non-pregnant
adult mice.
ior of PFOA in monkeys [12], a biologically based saturable renal
resorption component was implemented in the one-compartment
model for 129S1 /SvlmJ non-pregnant mice. In this hypothetical kid-
ney compartment depicted in Fig. 6, the unbound fraction of PFOA
from the central compartment would get filtered by the glomeru-
lous into a filtrate compartment from which PFOA would be either
excreted through the urine or resorbed via saturable renal trans-
porter^), with a transport affinity Kt and transport maximum Tm,
into a renal plasma compartment for entry back into the systemic
circulation. This biologically based modeling of renal clearance has
been used previously to describe active tubular secretion in renal
2" 800n
"S>
< 600-
£
Q.
O 400-
I
.2 200-
|
5
co o
5.0 mg/kg/day
A Abbott et al. 2007
10.0 mg/kg/day
• Abbott et al. 2007
— 20.0 mg/kg/day
» Abbott et al. 2007
0 100 200 300 400 500 600 700 800 900
time (hr)
Fig. 7. Simulation of serum levels of PFOA in 12951/SvlmJ non-pregnant adult mice
at reported doses greater than 1 mg/kg-d.
clearance [34], but was adapted in this case to describe renal resorp-
tion of PFOA. This non-linear model reasonably simulated serum
levels for 12951/SvlmJ non-pregnant mice at all doses. Simulations
for doses of 0.1 -1.0 mg/kg-d are shown for comparison with preg-
nant 129S1/SvlmJ mice in Fig. 5A-D, and those for doses greater
than 1 mg/kg-d are shown in Fig. 7.
In comparison, 129S1/SvlmJ adult non-pregnant mice
achieved the highest serum concentrations of PFOA, followed
by pregnant/non-lactating, pregnant/lactating, and nursing pups
(Fig. 5A-D). Based on the magnitude of serum measurements,
gestation and lactation in combination act to lower maternal
serum levels of PFOA by about 3-fold, and thus represent impor-
tant clearance pathways for the dam and correspondingly major
sources of exposure for the offspring.
The ability of models to predict serum levels of PFOA was
quantitatively evaluated using Pearson residual analysis for each
dose modeled. As shown in Fig. 8A, the model predictions for
(A) Non-pregnant 129S1/SvlJ mice
(B) Pregnant/non-lactating 129S1/SvlJ mice
5
W 4.0-
1 2.0-
!
0. 0.0-
•a
£
= -2.0-
i
0.0-
.»
•5
£ -1.0-
l « -2.0-
i i i i i 5
0.2 0.4 0.6 0.8 1.0 3
(0 -3.0 ~
Q
Dose (mg/kg) ^ -4.0-
T
1
0.2
i
0.4 0
i
1
L
6 0.8 1.0
i
Dose (mg/kg}
Fig. 8. Evaluation of model predictions for 12951/SvImJ mice. The dots represent the residual values (measured - predicted) divided by the standard error of the mean (SEM)
for each dose and the lines depict the deviation from the measured value.
Previous
TOC
-------
C.E. Rodriguez eta/./Reproductive Toxicology 27(2009)373-386
381
12951/SvImJ adult non-pregnant mice can deviate from the serum
measurements by as much as 40% at the lowest and highest dose
of 0.1 and 20 mg/kg-d, respectively, but are less than 20% for all
other doses. In contrast, the predictions by the linear pharmacoki-
netic model of gestation and lactation model are generally better,
ranging from 35% to less than 10% (Fig. 8B-D).
The serum information from two developmental toxicity studies
carried out in CD-I mice is also very limited. In one of the studies,
serum levels of PFOA were measured at term, i.e., 24 h following
the last treatment [22], while the other study reported serum mea-
surements at pnd22 [24]. The lowest dose of PFOA examined in
CD-I mice was 1 mg/kg-d [22], reflecting an apparent lower sensi-
tivity based upon administered dose for this strain as compared
to 12951/SvlmJ [21,22]. In an initial effort to examine the phar-
macokinetics of PFOA in pregnant CD-I mice, the linear model
of gestation and lactation in Fig. 3 was parameterized for CD-I
mice and evaluated for its ability to simulate serum levels of PFOA
measured at term. Absorption and elimination rate constants were
those estimated for adult non-pregnant female CD-I mice [11 ]. As
shown in Fig. 9, even at the lowest dose reported, the linear model
over-predicted serum levels by nearly four-fold. The discrepancy
increased at higher doses, presumably indicating the inability of
the model to simulate non-linear behavior (data not shown). In
another effort, the renal resorption component previously used to
describe the non-linear pharmacokinetic behavior of PFOA in adult
non-pregnant 129Sl/SvlmJ mice was implemented for CD-I mice.
The resulting model structure is shown in Fig. 10. The saturable
renal resorption component significantly improved the simulation
of serum levels of PFOA at term for all doses reported (Fig. 11A-D)
[22]. No serum information for concept! or pups was reported, but
the model can predict the corresponding levels in this compartment
(Fig. HA-D).The renal resorption parameters used to simulate CD-
1 serum levels of PFOA are listed in Table 5 along with those used
with adult non-pregnant 129Sl/SvImJ mice.
Serum levels of PFOA measured in CD-I mice postnatally were
also simulated via the incorporation of renal resorption. These
serum measurements were performed at pnd22 and corresponded
Parturition
100-
80-
5"
B)
I 601
J2
a> 40-
I ^
CO
Simulation by linear model
Lau et al. (2006)
100
200
time (hr)
300
400
Fig. 9. Predicted serum profile of PFOA at 1 mg/kg-d in pregnant CD-I mice by linear
pharmacokinetic model of gestation and lactation.
to only two administered doses, namely 3 and 5 mg/kg-d, but
included dam and pups [24]. When compared to prenatal serum
measurements at term at the same administered doses [22], these
pnd22 measurements do not reflect any impact of parturition and
lactation in decreasing serum levels of PFOA. In fact, they are
only about 27% and 48% lower for 3 and 5 mg/kg-d, respectively
(40.50 ± 1.89 mg/L versus 29.47 ± 2.55 mg/L and 71.91 ±8.33mg/L
versus 36.90 ±4.75 mg/L). Thus, based solely on magnitude, a
slower serum clearance of PFOA (i.e., greater accumulation) would
be necessary to simulate the pnd22 serum measurements using
the same pharmacokinetic model in Fig. 10. As shown in Fig. 12,
maternal and pup serum levels of PFOA at pnd22 were success-
fully simulated, but required an approximate increase in Tmc of
10-fold (while maintaining Kt constant due to the absence of an
adequate range of doses). The predicted prenatal serum profiles of
PFOA were as much as 5-fold higher than those predicted based
on serum measurements at term (Figs. 11 and 12). The basis for
Urine
o\
*-<
&
a
Pup excreta
'' recirculation
(from birth to PND14)
Cmilk Vm
klac
4
Pups
Cpup Vpup
Jkep
Urine
Fig. 10. Renal resorption pharmacokinetic model of gestation and lactation used in the analysis of CD-I mice.
Previous
TOC
-------
382
C.E. Rodriguez eta/./Reproductive Toxicology 27(2009)373-386
(A) 1 mg/kg/day
40 -i
30 -
20 -
10 -
o>
OT
Gestation
Lactation •
Pregnant t Lactatng Dam
i lau etal 2006
Concept! / Pups
0 100 200 300 400 500 600 700 800 900
time (hr)
(B) 3 mg/kg/day
• Pregnant / Lactating Dam
A Lau el al 2006
— - Concept /Pup
150-,
0 100 200 300 400 500 600 700 800 900
time (hr)
(C) 5 mg/kg/day
Lactation •
"fi
v>
I
E
o
in
=. 100-
50 -
Gestation
Pregnant / Lactating Dam
A Lau et al 2006
- - • Concepti / Pup
o +—r-^7-
0 100 200 300 400 500 600 700 800 900
time (hr)
(D) 10 mg/kg/day
250 -i Gestation
Lactation -
— Pregnant / Lactating Datr
1 Lau et al 2006
Concept / /Pups
0 100 200 300 400 500 600 700 800 900
time (hr)
Fig. 11. Simulation of prenatal serum levels of PFOA in CD-I mice.
the large inconsistency in the predicted bioaccumulative prenatal
profiles is unclear since both studies were performed in the same
mouse strain using the same doses, dosing regimen, and supplier
of PFOA [22,24]. The profiles were not expected to vary by more
than the reported inter-laboratory analytical coefficient of varia-
tion of 20.3% for measurements of PFOA in serum samples [42].
With the exception of serum measurements at 3 mg/kg-d mea-
sured at term, model simulations were in good agreement with
reported measurements in CD-I mice, deviating at maximum 20%
from measurements (Fig. 13A-C).
E
»_
a>
w
300 -\
200-
100-
Gestation
3 mg/kg/day
Lactation
Pregnant / Lactating Dam
Wolf etal. 2007
Concepti / Pups
Wolf etal. 2007
100 200 300 400 500 600 700 800 900
time (hr)
500-]
400-
Gestation
5 mg/kg/day
Lactation
I
E 300-
tn
$
g 200
E
i
0>
OT
100-
Pregnant / Lactating Dam
• Wolf etal. 2007
— • Concepti / Pups
Wolf et al. 2007
— I ---- 1 - 1 - 1 - 1 - 1 - 1 - 1 - 1 —
0 100 200 300 400 500 600 700 800 900
time (hr)
Fig. 12. Simulation of postnatal serum levels of PFOA in CD-I mice.
6. Discussion
This modeling effort was aimed at addressing some of the
uncertainties associated with the pharmacokinetics of PFOA in
animal developmental toxicity studies. To this end, a phar-
macokinetic model of gestation and lactation was developed
for the mouse based on limited information available. A two-
compartment system linked by placental blood flow described
gestation and milk production sequentially linked a lactating dam
to a growing pup compartment. The model incorporated the crit-
ical changes in BW (for dam and offspring), placental blood flow,
embryo/fetus:maternal serum partition coefficients for PFOA, and
milk production. The ability of the model to simulate serum lev-
els of PFOA from developmental toxicity studies involving two
different strains of mice was evaluated. The results indicate that
a linear clearance description was sufficient to reasonably sim-
ulate serum levels of PFOA at administered doses of 1 mg/kg-d
or less for 12951/SvlmJ mice that were pregnant/non-lactating,
pregnant/lactating, and respective nursing pups. In contrast, the
non-linear behavior exhibited by serum levels of PFOA in CD-I mice
at the only reported doses of 1 mg/kg or greater was simulated by
the incorporation of a saturable renal resorption description. The
manipulation of renal clearance of PFOA in rats through hormones
and inhibitors of organic anion transporters served as precedence
for the renal resorption component [14]. Previous modeling efforts
have also made use of a different mathematical formulation for
renal resorption to explain the non-linear pharmacokinetic behav-
ior of PFOA in monkeys [12].
The Implementation of renal resorption also allowed simu-
lation of serum levels of PFOA for 129Sl/SvlmJ non-pregnant
Previous
TOC
-------
Table 6
Estimates of internal dose metrics for developmental toxicity studies in rodents.
Dose (mg/kg-d)
0.1
0.3
0.6
1.0
1.0
3.0
3.0
3.0
5.0
5.0
Species (strain)
Mouse (12951/SvImJ)
Mouse (12951/SvImJ)
Mouse (12951/SvImJ)
Mouse (12951/SvImJ)
Mouse (CD-I)
Rat (Sprague Dawley)
Mouse (CD-I)
Mouse (CD-I)
Mouse (CD-I)
Mouse (CD-I)
Life stage
Dam
Pups
Dam
Pups
Dam
Pups
Dam
Pups
Dam
Pups
Dam
Pups
Dam
Pups
Dam
Pups
Dam
Pups
Dam
Pups
Toxic effect
None
Liver hypertrophy
None
Liver hypertrophy
Neonatal lethality
Liver hypertrophy
Liver hypertrophy
Neonatal lethality
Neonatal BW deficit
Eye opening delay
Liver hypertrophy
Liver hypertrophy
Neonatal lethality
Neonatal BW deficit
Eye opening delay
Liver hypertrophy
Parturition delay
Neonatal BW deficit
Eye opening delay
None
Neonatal BW deficit
Post-weaning lethality
Sexual maturation delay
Post-weaning BW deficits
Liver hypertrophy
Parturition delay
Neonatal BW deficit
Eye opening delay
Neonatal lethality
Liver hypertrophy
Full litter resorption
Neonatal BW deficit
Eye opening delay
Liver hypertrophy
Neonatal lethality
Liver hypertrophy
Parturition delay
Full litter resorption
Neonatal BW deficit
Eye opening delay
Neonatal lethality
Liver hypertrophy
Full litter resorption
Neonatal BW deficit
Eye opening delay
Liver hypertrophy
Neonatal lethality
NOAEL or LOAEL
LOAEL
-
NOAEL
NOAEL
-
LOAEL
NOAEL
NOAEL
LOAEL
-
LOAEL
LOAEL
LOAEL
NOAEL
LOAEL
LOAEL
-
NOAEL
NOAEL
NOAEL
NOAEL
LOAEL
-
-
NOAEL
LOAEL
NOAEL
LOAEL
LOAEL
LOAEL
NOAEL
-
-
NOAEL
-
LOAEL
LOAEL
-
LOAEL
Reference
[21]
[21]
[21]
[21]
[21]
[21]
[21]
[21]
[22]
[22]
Draft risk assessment
Draft risk assessment
[22]
[22]
[24]
[24]
[22]
[22]
[24]
[24]
AUC average daily (mg/L-hr)
G
118
14
355
41
711
82
1185
137
548
55
-
N/A
1429
138
3811
447
2160
205
6120
707
G + L
94
25
283
76
566
153
943
254
296
48
83
N/A
755
115
2981
678
1127
165
4644
1033
Cmax (mg/L)
8
2
25
7
49
14
82
24
34
5
13
N/A
87
12
270
49
131
18
423
75
tmax (h)
390
498
390
498
390
498
390
498
388
390
2
N/A
388
390
390
458
388
390
390
458
o
rn
1
uez eta/.,
I
3
n.
^
a.
s
a.
8
1
tSJ
•s]
To
g
10
Lo
\
Lo
CO
Ol
'Abbreviations: G = gestation only; G + L=gestation and lactation; N/A = not available.
Previous
TOC
-------
384
C.E. Rodriguez eta/./Reproductive Toxicology 27(2009)373-386
2.0 -
5
LLJ
5°. o.o-
0)
•5
f -2.0 -
Q.
•o
3 -4.0 -
in
a
I
(A) Pregnant uu-1 mice
• . 9
i
2
1
i i i i i
4 6 8 10 12
Dose (mg/kg/day)
l
14 16 18 20
E
LU
CO
I
TJ
£
Q.
3
(A
ra
-6.0 J
2.0 n
(B) Pregnant/ lactating CD-1 Mice
5
LU
co
~ 1.0-
I
0
£ o.o -
a
•
•o
£
3 -1.0-
ra
a>
S
<
i
3.0 3.5 4.0 4.5 5.0
Dose (mg/kg/day)
-2.0 J
0.0
-1.0 -
-2.0 -J
(C) Nursing CD-1 pups
3.0
—J—
3.5
—J—
4.0
—|—
4.5
1
Dose (mg/kg/day)
Fig. 13. Evaluation of model predictions for CD-I mice. The dots represent the resid-
ual values (measured - predicted) divided by the standard error of the mean (SEM)
for each dose and the lines depict the deviation from the measured value.
adult mice which exhibited non-linear behavior at doses greater
than 1 mg/kg-d at which full litter resorption was observed in
the respective pregnant mice [21]. Modeling of 12951/SvlmJ non-
pregnant, pregnant/non-lactating, and pregnant/lactating mice
at the same administered doses of PFOA confirm that gesta-
tion and lactation serve as significant clearance pathways for
the dam (and correspondingly sources of exposure for the
offspring).
The pharmacokinetic behavior of PFOA at doses less than
1 mg/kg-d in pregnant CD-I mice is unclear. Nevertheless, the
analyses with 129Sl/SvlmJ mice is indicative that a linear pharma-
cokinetic model may be appropriate in the analysis of gestational
and lactational exposures to PFOA at low doses which may be more
relevant to environmental exposures.
Another objective of this modeling study was to use the respec-
tive pharmacokinetic models to derive initial estimates of internal
dose as a means for reconciling large interspecies differences in
plasma clearance. Although different model structures and param-
eter values were necessary in the analyses for PFOA, estimates
of internal dose can still be derived to make initial intra- and
inter-species comparisons and estimate the relative contribution
of gestational and lactational exposures. Intra-species comparisons
may be important given that a single set of parameter values failed
to describe different data sets even when they were obtained using
the same species and strain, doses, dosing regimen, and source of
PFOA. From a risk assessment perspective, interspecies compar-
isons are of particular interest since developmental toxicity studies
carried out in rats versus mice have yet to be reconciled given the
observed large difference in plasma clearance between these two
species (i.e., half-life of 2-4 h versus 12-20 days, respectively). Thus,
the average daily area under the serum concentration-time curve
(AUC) and the maximum concentration (Cmax) of PFOA for both dam
and offspring were selected as measures of internal dose and esti-
mated from the respective models. These estimates were compared
to those of the rat at an administered dose of 3 mg/kg-d estimated in
the US EPA draft risk assessment for PFOA [43]. As listed in Table 6,
at the lowest dose of 0.1 mg/kg-d reported for 129Sl/SvlmJ mice,
the AUC values of 118 and 94mgh/L for gestation only (G) and
gestation and lactation (G + L), respectively, for the dam are com-
parable to the value of 83 mg h/L estimated for the pregnant rat at
the 30-fold higher administered dose of 3 mg/kg-d. The estimated
value for Cmax of 8mg/L is also comparable to 13mg/L reported
for the rat. Since liver hypertrophy in 129Sl/SvlmJ nursing pups
was the only adverse effect reported and its relevance to humans
remains the subject of debate, the next higher dose of 0.3 mg/kg-d
represented the no adverse effect level (NOAEL) for neonatal lethal-
ity for 129Sl/SvlmJ nursing pups. The estimated G + L AUC value
of 283 mg h/L for 12951/SvlmJ dams does not reflect the 10-fold
decrease in administered dose as compared to the rat and is in fact
more than 3-fold higher. It should be noted that internal dose esti-
mates for pups are not part of the default risk assessment approach
for analyzing developmental toxicity studies and therefore were not
estimated in the draft risk assessment for PFOA. However, our esti-
mates for 12951/SvlmJ nursing pups can still be compared to those
for rat dams. At 0.3 mg/kg-d, the G + L AUC value of 76 mg h/L for
129S1 /SvlmJ pups is comparable to that of the rat dams at 83 mg h/L.
Based on these internal dose estimates, irrespective of model struc-
ture, 129Sl/SvlmJ mice do not appear to be more sensitive than
the rat and the seemingly greater sensitivity to PFOA may be due
to a much higher internal dose achieved for a given administered
dose.
Also listed in Table 6 are the AUC and Cmax estimates for CD-
1 mice. At the lowest administered dose of 1 mg/kg-d, the G + L
AUC for CD-I dams is about 3.6-fold greater than the value esti-
mated for the rat at 3 mg/kg-d. The G + L AUC estimates for the
corresponding pups are about 1.7-fold lower, but still do not reflect
the 3-fold decrease in administered dose of the rat. Also shown
are the estimates at doses of 3 and 5 mg/kg-d reported in both
studies with CD-I mice [22,24]. Although these doses and internal
dose estimates are probably too high for any type of risk assess-
ment applications, they are nonetheless useful for intraspecies and
interlaboratory comparisons. The AUC estimates for the reported
postnatal serum measurements [24] are about 2.8- and 4.5-fold
higher than the reported prenatal measurements for both dam and
pups at 3 and 5 mg/kg-d PFOA, respectively. The inconsistency may
be analytical in nature since similar adverse effects were observed
in both studies [22,24].
The predicted AUC estimates for 129Sl/SvImJ mice suggest that
the relative contribution of gestation and lactation as a source of
exposure for the offspring are comparable when exposure to the
dam occurs during gestation without continued exposure during
lactation.
Previous
TOC
-------
C.E. Rodriguez eta/./Reproductive Toxicology 27(2009)373-386
385
In the case of CD-I mice, the relative contributions vary
depending on the study analyzed. Based on the prenatal serum
measurements of PFOA [22], the contribution of gestation far out-
weighs that of lactation. The contribution of lactation to pup
exposure is predicted to increase significantly based on postna-
tal serum measurements [24]. It should be noted that windows
of susceptibility may render gestation more toxicologically impor-
tant for organs/tissues that undergo critical development before
birth. Results from cross fostering studies suggest that exposure to
PFOA in utero may be sufficient to produce postnatal BW deficits
and developmental delays in the pups, and a window of sus-
ceptibility for some of these effects may exist early in gestation
[24].
The results of this modeling effort are based, due to limited
data available, on several assumptions that need to be emphasized.
First, gestation was modeled strictly as a flow-limited process; any
diffusion-limited transfer of PFOA was not included. Second, the
millcmaternal serum partition coefficient was modeled as constant
throughout lactation which may not be accurate since milk com-
position changes as a function of postnatal day [44,45]. Third, in
the absence of data, the first order elimination rate constant for the
pups was either optimized based on pup serum levels (129Sl/SvlmJ
mice) or assumed to be the same as the respective dam (CD-I mice).
The identity and ontogeny of the renal transporters involved in the
renal clearance of PFOA remain an area of investigation and there-
fore cannot be modeled at this time. Fourth, our models are based
on a classical compartmental approach augmented with biological
information, and therefore are not meant to be used for any type of
extrapolation. Cross-species and high-to-low dose extrapolations
maybe dependent on additional biological processes such as dose-
dependent changes in liver distribution [46] and may need a whole
body PBPK model approach.
Several research needs were identified throughout the develop-
ment and evaluation of the models. First, the lack of longitudinal
prenatal and postnatal serum measurements of PFOA in the same
mouse strain within the same study. The available prenatal and
postnatal serum measurements for CD-I mice from different stud-
ies seem inconsistent and do not complement each other. Second,
there were no milk measurements of PFOA available in the mouse.
Ideally, such measurements should be carried out in parallel
with maternal serum measurements from the same dam so that
changes in partition can be properly monitored. In the absence
of such data, the millcmaternal serum partition coefficient for
PFOA was optimized based on maternal and pup serum informa-
tion [21] and assumed constant throughout lactation. Third, pup
BW deficits resulting from PFOA exposure are not being mod-
eled at this time since it is unclear whether or not they reflect
a lower consumption of milk which would decrease the lacta-
tional transfer of PFOA. Pup BW deficits can be as much as 20-
25% at the higher doses administered to the dam [21,22], and thus
are not likely to dramatically change our estimates of internal dose
metrics.
Although administered doses of PFOA less than 1 mg/kg-d may
exhibit little or no phenotypic toxicity in CD-I mice, serum informa-
tion for these low doses may further support a linear model for the
pharmacokinetics of PFOA at low doses that may be more reflective
of environmental exposures.
In conclusion, this effort at a minimum provides initial phar-
macokinetic model structures for further explorations of the
pharmacokinetic behavior of PFOA during gestation and lactation
relevant to developmental toxicity studies which involve differ-
ent exposures (e.g., in utero and lactational exposures) but whose
current analysis for risk are based solely on the maternal dose. Fur-
thermore, our estimates of internal dose suggest that the mouse
achieves much higher internal dose of PFOA and may not be
more sensitive than the rat to the developmental toxicity induced
by PFOA. These results may be important in selecting the most
appropriate laboratory species for risk assessment and/or further
investigations associated with PFOA.
Conflict of interest
The authors declare that there are no conflicts of interest.
Acknowledgements
The invaluable assistance through sharing of unpublished data
and discussions with Drs. Christopher Lau, Barbara Abbott, and
Suzanne Fenton are greatly appreciated by the authors throughout
this modeling effort. Input from Drs. Andrew Lindstrom and Mark
Strynar with regards to analytical chemistry issues is appreciated
along with comments from Dr. Jennifer Seed (internal reviewers).
This project was partially funded by Interagency Agreement RW-
75-92207501 with the National Toxicology Program at the National
Institute for Environmental Health Science.
References
[1] Prevedouros K, Cousins IT, Buck RC, Korzeniowski SH. Sources, fate and trans-
port of perfluorocarboxylates. Environ SciTechnol2006;40(January(l)):32-44.
[2] Wallington TJ, Hurley MD, Xia J, Wuebbles DJ, Sillman S, Ito A, et al. For-
mation of C7F15COOH (PFOA) and other perfluorocarboxylic acids during
the atmospheric oxidation of 8:2 fluorotelomer alcohol. Environ Sci Technol
2006;40(February (3)):924-30.
[3] Nabb DL, Szostek B, Himmelstein MW, Mawn MP, Gargas ML, Sweeney LM, et al.
In vitro metabolism of 8-2 fluorotelomer alcohol: interspecies comparisons and
metabolic pathway refinement. Toxicol Sci 2007;100(December(2)):333-44.
[4] Dinglasan MJ, Ye Y, Edwards EA, Mabury SA. Fluorotelomer alcohol biodegra-
dation yields poly- and perfluorinated acids. Environ Sci Technol 2004;38(May
(10)):2857-64.
[5] Hori H, Nagaoka Y, Yamamoto A, Sano T, Yamashita N, Taniyasu S, et al. Effi-
cient decomposition of environmentally persistent perfluorooctanesulfonate
and related fluorochemicals using zerovalent iron in subcritical water. Environ
Sci Technol 2006:40(February (3)): 1049-54.
[6] Key BD, Howell RD, Criddle CS. Fluorinated organics in the biosphere. Environ
Sci Technol 1997:31:2445-54.
[7] Calafat AM, Wong LY, Kuklenyik Z, Reidy JA, Needham LL. Polyfluoroalkyl
chemicals in the US. population: data from the National Health and Nutri-
tion Examination Survey (NHANES) 2003-2004 and comparisons with NHANES
1999-2000. Environ Health Perspect 2007:115(November(ll)):1596-602.
[8] Olsen GW, Burris JM, Ehresman DJ, Froehlich JW, Seacat AM, Butenhoff JL,
et al. Half-life of serum elimination of perfluorooctanesulfonate, perfluoro-
hexanesulfonate, and perfluorooctanoate in retired fluorochemical production
workers. Environ Health Perspect 2007:115(September(9)):1298-305.
[9[ Lau C, Anitole K, Hodes C, Lai D, Pfahles-Hutchens A, Seed J. Perfluoroalkyl acids:
a review of monitoring and toxicological findings. Toxicol Sci 2007:99(October
(2)):366-94.
[10] WambaughJF, Barton HA, Setzer RW. Comparing models for perfluorooctanoic
acid pharmacokinetics using Bayesian analysis.] Pharmacokinet Pharmacodyn
2008:35(December (6)):683-712.
[11] Lou I, WambaughJF, Lau C, Hanson RG, Lindstrom AB, Strynar MJ, et al. Mod-
eling single and repeated dose pharmacokinetics of PFOA in mice. Toxicol Sci
2009:107(February(2)):331-41.
[12[ Andersen ME, Clewell 3rd HJ, Tan YM, Butenhoff JL, Olsen GW. Pharma-
cokinetic modeling of saturable, renal resorption of perfluoroalkylacids in
monkeys—probing the determinants of long plasma half-lives. Toxicology
2006:227(October(l-2)):156-64.
[13] Kudo N, KawashimaY. Toxicity and toxicokinetics of perfluorooctanoic acid in
humans and animals. J Toxicol Sci 2003:28(May (2)):49-57.
[ 14] Kudo N, Katakura M, Sato Y, Kawashima Y. Sex hormone-regulated renal trans-
port of perfluorooctanoic acid. ChemBiol Interact 2002:139(March(3)):301-16.
[15] Han X, Yang CH, Snajdr SI, Nabb DL, Mingoia RT. Uptake of perfluorooc-
tanoate in freshly isolated hepatocytes from male and female rats. Toxicol Lett
2008:181(2):81-6.
[16[ Apelberg BJ, Witter FR, Herbstman JB, Calafat AM, Halden RU, Needham LL,
et al. Cord serum concentrations of perfluorooctane sulfonate (PFOS) and per-
fluorooctanoate (PFOA) in relation to weight and size at birth. Environ Health
Perspect 2007:115(November (11)): 1670-6.
[17[ Spliethoff HM, Tao L, Shaver SM, Aldous KM, Pass KA, Kannan K, et al. Use
of newborn screening program blood spots for exposure assessment: declin-
ing levels of perluorinated compounds in New York State infants. Environ Sci
Technol 2008;42(July (14)):5361-7.
[18] TaoL, Kannan K, Wong CM, Arcaro KF, Butenhoff JL. Perfluorinated compounds
in human milk from Massachusetts, U.S.A. Environ Sci Technol 2008:42(April
(8)):3096-101.
Previous
TOC
-------
386
C.E. Rodriguez eta/./Reproductive Toxicology 27(2009)373-386
[19] So MK, Yamashita N, Taniyasu S, Jiang Q, Giesy JP, Chen K, et al. Health risks in
infants associated with exposure to perfluorinated compounds in human breast
milk from Zhoushan, China. Environ Sci Technol 2006;40(May (9)):2924-9.
[20] Butenhoff JL, Kennedy Jr GL, Frame SR, O'Connor JC, York RG. The reproduc-
tive toxicology of ammonium perfluorooctanoate (APFO) in the rat. Toxicology
2004;196(March(l-2)):95-116.
[21] Abbott BD, Wolf CJ, Schmid JE, Das KP, Zehr RD, Helfant L, et al. Perfluo-
rooctanoic acid induced developmental toxicity in the mouse is dependent
on expression of peroxisome proliferator activated receptor-alpha. Toxicol Sci
2007;98(August (2)):571-81.
[22] Lau C, Thibodeaux JR, Hanson RG, Narotsky MG, Rogers JM, Lindstrom AB, et
al. Effects of perfluorooctanoic acid exposure during pregnancy in the mouse.
Toxicol Sci 2006;90(April (2)):510-8.
[23[ White SS, Calafat AM, Kuklenyik Z, Villanueva L, Zehr RD, Helfant L, et al.
Gestational PFOA exposure of mice is associated with altered mammary
gland development in dams and female offspring. Toxicol Sci 2007;96(March
(1)): 133-44.
[24[ Wolf CJ, Fenton SE, Schmid JE, Calafat AM, Kuklenyik Z, Bryant XA, et al. Develop-
mental toxicity of perfluorooctanoic acid in the CD-I mouse after cross-foster
and restrictedgestational exposures.Toxicol Sci 2007;95(February(2)):462-73.
[25] O'Flaherty EJ, Scott W, Schreiner C, Beliles RP. A physiologically based kinetic
model of rat and mouse gestation: disposition of a weak acid. Toxicol Appl
Pharmacol 1992;112(February (2)):245-56.
[26] R, Team DC. R: a language and environment for statistical computing. Vienna,
Austria: R Foundation for Statistical Computing: 2008. ISBN 3-900051-07-0.
http://www.R-project.org.
[27] Taillieu F, Salomon LJ, Siauve N, Clement O, Faye N, Balvay D, et al.
Placental perfusion and permeability: simultaneous assessment with dual-
echo contrast-enhanced MR imaging in mice. Radiology 2006:241(December
(3)):737-45.
[28] Iguchi T, Tani N, Sato T, Fukatsu N, Ohta Y. Developmental changes in mouse
placental cells from several stages of pregnancy in vivo and in vitro. Biol Reprod
1993:480anuary(l)):188-96.
[29] Johnson MS, Thomson SC, Speakman JR. Limits to sustained energy intake.
III. Effects of concurrent pregnancy and lactation in Mus musculus.] Exp Biol
2001:204(June(Pt ll)):1947-56.
[30] Knight CH, Maltz E, Docherty AH. Milkyield and composition in mice: effects of
littersize and lactation number. Comp Biochem Physiol A 1986:84(l):127-33.
[31] Yoon M, Barton HA. Predicting maternal rat and pup exposures: how different
are they? Toxicol Sci 2008:102(March(l)):15-32.
[32] Gubernick DJ, Alberts JR. Resource exchange inthe biparental California mouse
(Pemmyscus californicus): water transfer from pups to parents. J Comp Psychol
1987:101(December(4)):328-34.
[33] Henning SJ. Postnatal development: coordination of feeding, digestion, and
metabolism. Am J Physiol 1981:241(September(3)):G199-214.
[34] Russel FG, Wouterse AC, Van Ginneken CA. Physiologically based pharma-
cokinetic model for the renal clearance of iodopyracet and the interaction
with probenecid in the dog. Biopharm Drug Dispos 1989:10(March-April
(2)):137-52.
|35] Brown RP, Delp MD, Lindstedt SL, Rhomberg LR, Beliles RP. Physiological param-
eter values for physiologically based pharmacokinetic models. Toxicol Ind
Health 1997:13Guly-August (4)):407-84.
[36] Henderson WM, Smith MA. Perfluorooctanoic acid and perfluorononanoic acid
in fetal and neonatal mice following in utero exposure to 8-2 fluorotelomer
alcohol. Toxicol Sci 2007;95(February (2)):452-61.
[37] Knight CH, Maltz E, Docherty AH. Milk yield and composition in mice: effects
of litter size and lactation number. Comp Biochem Physiol A Comp Physiol
1986:84(l):127-33.
[38] Hinderliter PM, Mylchreest E, Gannon SA, Butenhoff JL, Kennedy Jr GL. Per-
fluorooctanoate: placental and lactational transport pharmacokinetics in rats.
Toxicology 2005:211(July (l-2)):139-48.
[39] McCullagh P, Nelder JA. Generalized linear models. 2nd ed. London xix + 511:
Chapman and Hall: 1989. p. 37.
[40] Clarke DO, Elswick BA, Welsch F, Conolly RB. Pharmacokinetics of
2-methoxyethanol and 2-methoxyacetic acid inthe pregnant mouse: a physio-
logically based mathematical model. Toxicol Appl Pharmacol 1993:121(August
(2)):239-52.
[41] Salomon LJ, Siauve N, Taillieu F, Balvay D, Vayssettes C, Frija G, et al. In vivo
dynamic MRI measurement of the noradrenaline-induced reduction in placen-
tal blood flow in mice. Placenta 2006:27(September-October (9-10)):1007-13.
[42] Longnecker MP, Smith CS, Kissling GE, Hoppin JA, Butenhoff JL, Decker E, et
al. An interlaboratory study of perfluorinated alkyl compound levels in human
plasma. Environ Res 2008:107(June (2)):152-9.
[43] USEPA. Draft Risk Assessment for the Potential Human Health Effects Associ-
ated with Exposure to Perfluorooctanoic Acid and Its Salts, http ://www.e pa.gov/
oppt/pfoa/pubs/pfoarisk. Access date: 11/01/08.
[44] Godbole VY, Grundleger ML, Pasquine TA, Thenen SW Composition of rat milk
from day 5 to 20 of lactation and milk intake of lean and preobese Zucker pups.
JNutrl981:lll(March(3)):480-7.
[45] McMullinTS, Lowe ER, Bartels MJ, Marty MS. Dynamic changes in lipids and pro-
teins of maternal, fetal, and pup blood and milk during perinatal development
in CD and Wistar rats. Toxicol Sci 2008:105(October (2)):260-74.
[46] Kudo N, Sakai A, Mitsumoto A, Hibino Y, TsudaT, Kawashima Y. Tissue distribu-
tion and hepatic subcellular distribution of perfluorooctanoic acid at low dose
are different from those at high dose in rats. Biol Pharm Bull 2007:30(August
(8)): 1535-40.
[47] Qi Z, Whitt I, Mehta A, Jin J, Zhao M, Harris RC, et al. Serial determination of
glomerular filtration rate in conscious mice using FITC-inulin clearance. Am J
Physiol Renal Physiol 2004:286(March (3)):F590-6.
[48] Luippold G, Pech B, Schneider S, Osswald H, Muhlbauer B. Age dependency
of renal function in CD-I mice. Am J Physiol Renal Physiol 2002:282(May
(5)):F886-90.
[49] Han X, Snow TA, Kemper RA, Jepson GW Binding of perfluorooctanoic acid to
rat and human plasma proteins. Chem Res Toxicol 2003;16(June (6)):775-81.
Previous
TOC
-------
Environ. Sci. Technol. 2009, 43, 2374-2380
te
from
lifl A
A. COHEN
AND
H U B A L ,
YING XU,1" ELAINE
PER A. CLAUSEN,5
JOHN C. LITTLE*'"1"
Department of Civil and Environmental Engineering, Virginia
Tech, Blacksburg, Virginia, National Center for
Computational Toxicology, Environmental Protection Agency,
Research Triangle Park, North Carolina, and New
Technologies Group, National Research Centre for the
Working Environment, Lers0 Parkalle 105, DK-2100
Copenhagen 0, Denmark
Received May 24, 2008, Revised manuscript
January 5, 2009. Accepted January 5, 2009.
received
A two-room model is developed to estimate the emission
rate of di-2-ethylhexyl phthalate (DEHP) from vinyl flooring and
the evolving gas-phase and adsorbed surface concentrations
in a realistic indoor environment. Because the DEHP emission
rate measured in a test chamber may be quite different from
the emission ratefromthe same material inthe indoor environment,
the model provides a convenient means to predict emissions
and transport in a more realistic setting. Adsorption isotherms for
phthalates and plasticizers on interior surfaces, such as
carpet, wood, dust, and human skin, are derived from previous
field and laboratory studies. Log-linear relationships between
equilibrium parameters and chemical vapor pressure are obtained.
The predicted indoor air DEHP concentration at steady state
is0.15^g/m3. Room 1 reaches steady state within about one year,
while the adjacent room reaches steady state about three
months later. Ventilation rate has a strong influence on DEHP
emission rate while total suspended particle concentration
has a substantial impact on gas-phase concentration. Exposure
to DEHP via inhalation, dermal absorption, and oral ingestion
of dust is evaluated. The model clarifies the mechanisms that
govern the release of DEHP from vinyl flooring and the
subsequent interactions with interior surfaces, airborne
particles, dust, and human skin. Although further model
development, parameter identification, and model validation
are needed, our preliminary model provides a mechanistic
framework that elucidates exposure pathways for phthalate
plasticizers, and can most likely be adapted to predict emissions
and transport of other semivolatile organic compounds, such
as brominated flame retardants and biocides, in a residential
environment.
Introduction
Since the 1930s, phthalates have been used as plasticizers to
enhance the flexibility of rigid polyvinylchloride (PVC)
* Corresponding author fax: (540) 231-7916; e-mail: jclCaVt.edu.
f Virginia Tech.
* Environmental Protection Agency.
§ National Research Centre for the Working Environment.
2374 « ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 43, NO. 7, 2009
products (1), with worldwide phthalate production exceeding
3.5 million tons/year (2). About 90% of phthalates are used
as plasticizers in polymers (e.g., PVC) and are found in a
wide range of consumer products including floor and wall
covering, toys, car interior trim, clothing, gloves, footwear,
and artificial leather (3). Di(2-ethylhexyl) phthalate (DEHP)
is most widely used and accounts for more than 50% of total
phthalalc production (3). The main use of DEHP is in PVC
products such as vinyl flooring, where it is typically present
at concentrations of ~20—40% (w/w) (4, 5). Other common
phthalates are dibutyl phthalate (DBF), benzyl butyl phthalate
(BBP), di-isononyl phthalate (DINP), and di-isodecyl ph-
thalate (DIDP).
Because phthalates are notchemically bound in polymers,
slow emission from the products to air or o ther media usually
occurs. Adverse health effects of phthalates are briefly
reviewed in the Supporting Information (SI). Phthalate esters
have been recognized as major indoor pollutants (3, 4, 6}. By
sampling in 120 homes and analyzing for 89 organic
chemicals, Rudel et al. (7) revealed that phthalates are one
of the most abundant contaminants in indoor air. In the
recent EPA-sponsored CTEPP (Children's Total Exposure to
Persistent Pesticides and Other Persistent Organic Pollutants)
study (ff), concentrations of over 50 target compounds were
measured in multimedia samples from the homes and
daycare centers of 260 preschool age children. The two
phthalates targeted in the CTEPP study were detected in
residential air and house dust, and on interior surfaces and
dermal wipe samples. As in Rudel et al.'s study, measured
phthalate concentrations were among the highest of any of
the targeted compounds, including pesticides, PA! Is, and
PCBs. Despite this, only a few studies of phthalate emission
characteristics and exposure are available. Uhde et al. (9)
measured emission of several phthalates from PVC-coated
wall-coverings in test chambers under standard room
conditions. Clausen et al. (5) measured emissions of DEHP
from vinyl flooring for more than a year in both the FLEC
(field and laboratory emission cell) and the CLIMPAQ
(chamber for laboratory investigations of materials, pollution,
and air quality). In addition, the effect of humidity on the
emission of DEHP from vinyl flooring was studied for one
year in the FLEC (10), emission of phthalates from different
types of plasticizcd product was studied for 150 days in the
CLIMPAQ (11), and emission of phthalates from different
types of plasticized materials was studied using a passive
flux sampler (12),
Based on the Clausen et al. (5) experiments, Xu and Little
(13) developed a model to predict the emission rate of
phthalates from polymer materials. Their analysis revealed
that emissions of the very low volatility semivolatile organic
compounds (SVOCs) (such as DEHP) are subject to "external"
control (partitioning into the gas phase, the convective mass-
transfer coefficient, and adsorption onto chamber surfaces).
The tendency of phthalates to adsorb strongly to surfaces is
most likely similar to that of other SVOCs. Gebefiigi (14)
showed that SVOCs were sorbed by cotton and Van Loy et
al. (15) found that more than 99% of recovered nicotine was
adsorbed to the walls of their stainless steel chamber.
Compared to these experimental chamber systems, the
indoor environment has many other types of surface that
will adsorb SVOCs such as DEHP to different extents. The
emission rate measured in a test chamber may therefore be
quite different from the emission rate from the same material
in the indoor environment.
The model developed by Xu and Little (13) provides a
convenient means to estimate the emissionratc and gas phase
10.1021/es8013E4f CCC: $40.75
© 2009 American Chemical Society
Published on Web 02/19/2009
Previoys
TOG
Next
-------
fa TSP, Q
p.**, c""«l
Room 1
Glass v, y,
y,, TSP
I ft
Vinyl Flooring
Ceiling .
Room 2 ' f
V, y, Glass
window
Q
-» PI, Wood
hT
Carptt
yj. TSP, (
"""•"""
FIGURE 1. Schematic representation of the two-room model.
TABLE 1. Conditions for Two-Room Model
volume (m x m x m)
ventilation rate (m3/h)
area of vinyl flooring, Av (m2)
area of carpet, Ac (m2)
area of glass window, Ag (m2)
area of furniture, A (m2)
area of ceiling and wall, >4CW (m2)
total suspended particles, TSP
room 1
! x3 xl
13.3
9
1.7a
41
20fi
room 2
! x3 x3
13.3
9
1.7a
20.3a
41
20fi
3 According to typical surface to volume ratio in
residences (76). b Typical TSP in residential environment
(77).
and adsorbed surface concentrations likely to occur in more
realistic indoor environments. In this paper we both simplify
and extend the Xu and Little model to investigate potential
emission and distribution of DEHP in a residential environ-
ment. Field data collected in the CTEPP study as well as
recent laboratory data are used to parameterize the extended
model. The model is then used to estimate the emission rate
and gas-phase DEHP concentration following the installation
of vinyl flooring in a room. Finally, we examine the influence
of two key parameters (air exchange rate and airborne particle
concentration) on DEHP emissions, and estimate the po-
tential exposure through inhalation, dermal absorption, and
oral ingestion of dust.
Two-Room Model
To better estimate DEHP emissions in a residential environ-
ment the SVOC emissions model (13) was extended from a
one-compartment description of an experimental chamber
to a two-compartment representation of two adjacent rooms
in a home (Figure 1). Vinyl flooring, the only source of DEHP
considered, is placed in room 1, while carpet and wooden
furniture are arranged in room 2. The room conditions are
provided in Table 1.
Mass Balance. With reference to Figure 1, the accumula-
tion of phthalate in room 1 obeys the following mass balance:
dyj(f) dCg(f) dCcw(f)
dt 1
dt
dt
dt
•+Av-m(t)-Q-y1(t)-QF1(t) (I)
where yi (/
-------
Somnlar\ layer
5
«
O
Boundary layer
Vfl, surf
Ailsorntion l
h Aftsnrnlmn hi airhornt* particle
FIGURE 2. Schematic of sorption process. Note that the four individual materials shown for illustrative purposes in Figure 2a do not
comprise a layered structure.
where Csurf is the surface concentration (/
-------
80-
60-
i 40-
c;
8 20
0)
ro 0
Raw data
1 Linear regression
95% Prediction interval
95% Confidence interval
R2= 0.68, p-value <2e-16
-9
CD
§
40-
20-
« 0-
Raw data
Linear regression
95% Prediction interval
95% Confidence interval
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7
Gas phase concentration, y (ng/m3)
3 A H1111 c V i n
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 H
Gas phase concentration, y
h Duct
FIGURE 3. Linear regression results for DBP.
E 6
•c
0-
logK>urf=-1.06logVii-3.30
R*= 0.97
DBP"
DEHP
DBP
-7.0
-6.5
-6.0 -55 -5.0
logV (mmHg)
-4.5
-4.0
t-li
an ciri
Vir
-7.0 -6.5 -6.0 -5.5 -5.0
log Vp (mmHg)
h Duct
-4.5
FIGURE 4. Linear regression between log (Vf) and log (Ksmf}.
TABLE 2. Partition Coefficients for DEHP
surface
furniture, wall and ceiling a
carpet b
glass °
skin d
airborne particles "
dustf
partition coefficient, K
2500(m)
1700 (m)
3800 (fig/m2)/ (fig/m3) '
9500 (m)
0.25 (m3/ug)
21100 (m3/g)
isotherm exponent, n
1.5
"Calculated using log Ksuri = -0.779 logl/p - 1.93, Figure S6. bCalculated using log /Csurf = -0.627 logl/p -1.08, Figure
S6. °Xu and Little fitted the Freundlich isotherm for glass. dCalculated using log /Csurf = -1.06 logl/p - 3.30, Figure 4.
"Calculated using log KVi particle = -0.860 log l/p - 4.67, eq S3. fRegression result of Figure S5.
we developed simple correlations between the equilibrium
parameters and the vapor pressure of the target chemicals.
Correlation of Equilibrium Parameters with Vapor
Pressure. Correlations were obtainedbetweenvaporpressure
(Vp) and sorption parameters (KSUIf) for different interior
surfaces, including settled dust. Linear relationships between
log (Vp) and log (KSUIf) were found (e.g., human skin and
dust) as shown in Figure 4. Results for all surfaces are shown
in Figure S6. Data for child hand wipe and adult hand wipe
were combined to get the relationship between the human
skin partition coefficient and vapor pressure. The partition
coefficient of BPA for dust did not conform well to this
relationship, thus Figure 4b only shows the relationship
between phthalate vapor pressures and partition coefficients
for dust.
While using only three chemicals does not provide a
conclusive relationship, the overall results suggest that it is
possible to relate the equilibrium partition coefficients to
vapor pressures. Finally, we used the new correlations to
obtain the partition coefficient for DEHP on different interior
surfaces, as shown in Table 2. The isotherm used for glass
is based on a previous study (13), although the nonlinear
nature of the isotherm may have been due to the data fitting
procedure. In the SI, the skin/air partition coefficient is
checked using a completely different procedure and shown
to be acceptable.
Mass Transfer Coefficient. The value of hm, the mass-
transfer coefficient for the boundary layer adjacent to the
various surfaces, was estimated using correlation equations
(21), which express hm as a function of Reynolds number
and Schmidt number. Huang et al. (22) measured air velocities
in a typical house in the United States. They found velocities
with a range of 0.01—0.16 m/s and showed that values near
the floor are higher than those in the center of the room. In
VOL. 43, NO. 7, 2009 / ENVIRONMENTAL SCIENCE & TECHNOLOGY • 2377
Previous
TOC
-------
Concentration (ng/m")
o o o o o
8O r*' -* N
Ul 0 01 0
Ccn-centratton in Room 1
— •— Concentration in Room 2
— C— Errtssiwt rate
Aw exchange
f ^f^ Air exchang*
//^Jp*" Air exchangs
rate= 0 25h '. TSP= 20ugftn '
fSte=0.5h'. TSP=20ua/m"'
jLjor*'^ At MCFwnae r»t«« O.Sh'1. TSP« aOusrtn1
|^^-T>--<>-0~OH^&-O-O-<>-O— O-O— O—O— O-O
Air exchange rsle= 025h', TS P= 40ug'(m '
Air exchange rste= 025h ' TSP=20Ltg/m
JBas*lm*iCondtcn)
* ^v W M U CJ
3 bi O in b en
Emission rate (M9/m2h)
Surface concentration (^9/m2)
POfrCRODOM-fcO
oS8S8SS8g
x
r*"*''
—9- Human skin
— T— Furniture
— •— Glass window
y — f~-^^W — T — V — * — T-
-i-A-t-fc-t-
0 200 400 600 800 1000 0 200 400 600 800
Time (days) Time (days)
0 h
4-1
1000
FIGURE 5. (a) Effect of air exchange rate and TSP concentration on DEHP concentration and emission rate, (b) Predicted
concentration on interior surfaces.
DEHP
TABLE 3. Concentration of Phthalates in Indoor Air and Household Dust Samples
DEHP
gas phase concn (fig/m3)
dust phase concn (fig/g)
references
(25)
(26)
(7)
(4)
(25)
(27)
(7)
(4) and (28)
n
40
125
102
59
12
600
101
30
mean
0.48
0.14
0.07
0.19
950
1200
340
776
max our study
1.6 0.15
1.0
0.4
3100 3000
3500
7700
1542
the CLIMPAQ chamber (5), which roughly approximates
conditions in a real room, the velocity at the test piece surface
was estimated to be 0.15 m/s and this value is used to estimate
fem. Odum et al. (23) measured mass transfer of PAHs and
others SVOCs to and from combustion aerosols at 25 °C, and
their result is used here as femp, the mass-transfer coefficient
for particles.
We now have estimates of all the partition coefficients for
DEHP between indoor air and interior surfaces (hardwood
floor, carpet, human skin, and particles), as well as the
associated mass transfer coefficients. We are therefore able
to use the two-room model to estimate the emission rate
and evolving gas-phase DEHP concentration following the
installation of vinyl flooring in room 1.
Because no other data for DEHP concentrations on real
interior surfaces are available, the CTEPP study provides the
only available data that we can use to estimate DEHP
adsorption isotherms. Even though the values of the partition
coefficient for DEHP on interior surfaces can only be
considered rough estimates, we showed in a sensitivity
analysis (24) that they do not have a strong influence on the
steady state indoor air DEHP concentration, which is the
basis for our exposure analysis (although they do influence
the time it takes to reach steady state). We therefore believe
that our emissions and transport model represents a reason-
able first step.
Results and Discussion
For baseline conditions (Table 1), the indoor air DEHP
concentration at steady state is 0.15 ,Mg/m3. As shown in Table
3, this value is similar to that measured within homes in
both the United States and Europe, although it should be
emphasized that vinyl flooring is the only source of DEHP
considered here. Room 1 reaches steady state within about
one year, while the adjacent room reaches steady state about
three months later. Airborne particles increase the rate at
which DEHP is transported between rooms by a factor of 5
relative to gas-phase transport. The boundary layer sur-
rounding the airborne particles is much thinner than the
boundary layer adjacent to the other indoor surfaces and the
suspended particles reach equilibrium with the gas-phase
much more rapidly than the larger surfaces. Suspended
particles are therefore very effective at transporting DEHP
from one room to another, because DEHP also desorbs very
rapidly from the particles.
In Figure 5a, the impact of air exchange rate and total
suspended particle concentration on the DEHP emission rate
and the DEHP concentration in rooms 1 and 2 is examined.
Increasing air exchange rate will increase the DEHP emission
rate from the vinyl flooring significantly while an increase in
the TSP concentration causes a substantial decrease in the
gas-phase concentration in both rooms, but increases the
emission rate in Room 1. An increase in the air exchange rate
was assumed to double the velocity of the air above the vinyl
flooring (from 0.15 to 0.30 m/s) and this higher value was
used to calculate the h^ associated with the flooring.
Figure 5b shows the predicted DEHP concentration
change with time on various interior surfaces in room 2. The
predicted DEHP concentration on human skin is 5—7 times
higher than on the other surfaces due to the high skin/air
partition coefficient for DEHP. The skin/air partition coef-
ficient was obtained from hand-wipe samples in the CTEPP
study. It is generally believed that these hand-wipe samples
are measuring chemicals transferred from indoor surfaces
onto the hands directly. However, the fact that the skin/air
isotherms determined for both adult and child are almost
identical for DBF, BBP, and BPA (see, for example, Figure
4a), suggests that SVOCs may be transferring directly from
the air to the skin, or that if large amounts are picked up by
direct dermal transfer, that some desorbs to re-establish
2378 • ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 43, NO. 7, 2009
Previous
TOC
-------
equilibrium with the air. Indeed, in a subsequent paper (24),
we show thai there is a strong correlation between the
concentrations of DBF, BBP, and BPA on skin and those in
the gas phase, but almost no correlation with those on interior
surfaces. This further suggests that certain SVOCs may reach
the skin through the gas phase, and not via dermal transfer
as is commonly suspected.
Estimating Exposure to DEHP from Vinyl Flooring.
Based on the model results, we are interested in evaluating
exposure to vapor phase DEHP in air, particle bound DEHP
in air, and DEHP in settled dust. The exposure pathways of
interest are inhalation of vapor, inhalation of particles, dermal
absorption of DEHP deposited on the skin, and oral ingestion
via household dust.
The detailed exposure calculations are shown in the SI.
For dermal exposure, the overall skin permeability coefficient,
P, is controlled by permeation through the skin (Akin;air) as
well as by permeation through the air boundary layer adjacent
to the skin (Patr), or:
1
(10)
where PSkjn/air (cm/hr) is vapor to skin permeability, and Pajr
(cm/hr) is permeability of the boundary layer. For low
volatility compounds, convective mass transfer through the
air boundary layer adjacent to the skin may become the rate
limiting factor, and this is the case for DEHP. As detailed in
the SI, the estimated value of Pis 580 cm/hr.
As shown in Table S8, the reference dose (RID) for DEHP
is 20 ag/kg/d according to the U.S. EPA. Airborne particles
contribute 80% of the inhalation exposure, although the
highest value of total inhalation exposure is less than 0.6
ag/kg/d, which is much lower than the RfD. For infants,
exposure through oral intake via dust is 1.6 times higher
than the RfD, although the estimate for dust intake rate of
10.3 mg/kg/d (29) may be high. Exposure via these two
pathways is similar to other study results (6, 30, 31}. Dermal
absorption of DEHP deposited on skin is greater than that
taken up through inhalation. For DEHP, the primary route
of exposure is oral ingestion of dust. Overall, children
experience 2™ 10 times higher exposure risk than adults based
on all exposure pathways.
Acknowledgments
Financial support was provided by the National Science
Foundation (CBET 0504167). We thank Peter Egeghy at EPA's
National Exposure Research Laboratory for his assistance
with the CTEPP database, and John Kissel, Linsey Marr, Bill
Nazaroff, and Charlie Weschler for their useful comments
on the draft manuscript. This paper has been subjected to
U.S. EPA Office of Research and Development review and
approved for publication,
Supporting Information Available
Further details on the regression results for the CTEPP data,
modification of surface partition coefficients, and detailed
exposure calculations. This material is available free of charge
via the Internet at http://pubs.acs.org.
Literature
(1) Latini, G.; Felice, C. D.; Vcrrotti, A. Plasticizers, infant nutrition
and reproductive health. RegrodJfoxJCQl- 2004, 1,9, 27-33.
(2) Cadogan, D. F.; Howick, C. J. Plaslicizers. Kirk-Othmer Ency-
clopedia of Chemical Technology' 1996, 19, 258-290.
(3) Bornehag, C.-G.; Lundgren. B.; Weschler, C. J.; Sigsgaard, T.;
Hagerhed-Engman, L.; Sundell, I. Phthalates in Indoor Dust
and Their Association with Building Characteristics. Environ^
Health Perspect. 2005, 11.9 (10), 1399-1404.
(4) Fromme, H.; Lahrz, T.; Piloty, M.; Gebhart, FL; Oddoy,A.; Ruden,
H. Occurrence of phthalates and musk fragrances in indoor air
and dust from apartments and kindergartens in Berlin (Ger-
many). Indoor Air 2004. 14, 188-195.
(5) Clausen, P. A.; Ilansen, V.; Gunnarsen, L.; Afshari, A.; Wolkoff,
P. Emission of di-2-ethylhexyi phthalate from PVC flooring into
air and uptake in dust: Emission and sorption experiments in
FLEC and CL1MPAQ. Environ. Set Technol 2004, 38 (9), 2531-
2537.
(6) Wensing, M.; Uhde, E.; Salthammer, T. Plastics additives in the
indoor environment—flame rctardants and plasticizcrs. Sd.
TotajEnmon. 2005, 339, 19-40.
(7) Rudel, R.; Camann, D.; Spengler,J.; Korn, I,.; Brody,J. Phthalates,
alkylphenols, pesticides, polybrominated diphenyl ethers, and
other endocrine-disrupting compounds in indoor air and dust.
Environ. Sci. Technol. 2003, 37(20), 4543-4553.
(8) U.S. EPA. A Pilot Study of Children's Total Exposure to Persistent
Pesticides and Other Persistent Organic Pollutants (CTEPP); 2005;
http://www.epa.gov/heasd/ctepp/index.htmCaccessedMayl3,
2008).
(9) Uhde, E.; Bednarek, M.; Fuhrmann, F.; Salthammer, T. Phthalic
Esters in the Indoor Environment -Test chamber studies on
PVC-coated wallcoverings. Indgo^Air 2001, 11, 150-155.
(10) Clausen, P. A.; Xu, Y.; Kofoed-S0rensen, V.; Little, J. C.; Wolkoff,
P. Hie influence of humidity on the emission of di-(2-ethylhexyl)
phthalate (DEHP) from vinyl flooring in the emission cell FFEC.
Atmos. Environ. 2007, 41, 3217-3224.
(11) Afshari, A.; Gunnarsen, L.; Clausen, P. A.; Hanson, V. Emission
of phthalates from PVC and other materials. Indoor Air 2004,
14, 120-128.
(12) Fujii, M.; Shinohara, N.; Lim, A.; Otake, T.; Kumagai, K.;
Yanagisawa, Y. A study on emission of phthalate esters from
plastic materials using a passive flux sampler. Atmos. Environ.
2003, 37, 5495-5504.
(13) Xu, Y.; Little, J. Predicting emissions of SVOCs from polymeric
materials and their interaction with airborne particles. Environ.t
Sd. Technol. 2006, 40 (2), 456-461.
(14) Gebefiigi, I. Chemical exposure in enclosed environments.
Toxicol Environ. Chem. 1989, 20- 21, 121-127.
(15) Van Loy, M. D.; Lee, V. C.; Gundel, L. A; Daisey, J. M.; Sextro,
R. G.; Nazaroff, W. W. Dynamic behavior of scmivolatilc organic
compounds in indoor air. 1. Nicotine in a stainless steel chamber.
Enwron.__Scl_Technol. 1997, 31, 2554-2561.
(16) Hodgson, A. T.; Ming, K. Y.; Singer, B. C. Quantifying Object and
Material Surface Areas in Residences; LBNL-56786; Lawrence
Berkeley National Laboratory, 2005.
(17) Weschler, C. J. Indoor/outdoor connections exemplified by
processes that depend on an organic compound's saturation
vapor pressure. Atmos. Environ. 2003, 37, 5455-5465.
(18) VanLoy, M.; Riley, W.; Daisey, J.; Nazaroff, W. Dynamic behavior
of semivolatile organic compounds in indoor air. 2. Nicotine
and phcnanthrcnc with carpet and wallboard. Environ. Set.
Technol. 2001, 35 (3), 560-567.
Won, D.; Corsi, R. L.; Ryncs, M. Sorptivc Interactions between
VOCs and IndoorMaterials. IndoorAtr200l, 11, 246-256.
Pankow, J. Application of common y-intcrccpt regression
parameters for log kp vsl /t for predicting gas particle partitioning
in the urban-environment. Atmos. Environ. Part A 1992,26(14),
2489-2497.
Axley, J. W. Adsorption modeling for building contaminant
dispersal analysis. Indoor Air 1991, 2, 147-171.
Huang, J. M.; Chen, Q.; Hibot, B.; Rivoalen, H. Modeling
contaminant exposure in a singlefamily house. Indoor Built
Environ, 2004, 13 (1), 5-19.
(23) Odum, J.; Yu, J.; Kamens, R. Modeling the mass transfer of
semivolatile organics in combustion aerosols. Environ. Set
Technol, 1994, 28, 2278-2285.
(24) Xu, Y.; Hubal, E. C.; Little,). C., Predicting Residential Exposure
to Phthalate PI asticizcr Emitted from Vinyl Flooring-Sensitivity,
Uncertainty, and Implications for Biomonitoring. Submitted.
(25) BATJCH. Anafyse und Bewertungder in Raumluft und Hausstaub
vorhandenen Konzentrationen der Weichmacherbestandteile
Diethylhexylphlhalat(DEIiP) und Dibulylphlhalat (DBF}, Berlin,
Germany, 1991.
(26) Sheldon, L.; Clayton, A.; Keever, J.; PerriU, R.; Whitaker, D.
PTEAM: Monitoring of phthalates and PAHs in indoor and
outdoor air. In Samples in Riverside California; Research Note
94-10; California Environmental Protection Agency, Air Re-
sources Board: Sacramento, CA, 1994.
(27) Mattulat, A. (SOFIA GmbH Berlin) (2002) Konzentration von
mittel-undschwerfliichtigenorganischenVerbindungeninStaub
aus Innenraumen - Belastungssituation im Jahr, 2001.
Previoys
VOL. 43, NO. 7, 2009 / ENVIRONMENTAL SCIENCE & TECHNOLOGY « 2379
TOG
Next
-------
(28) Weschler, C. J.; Salthammer, T.; Fromme, H. Partitioning of
phthalates among the gas phase, airborne particles and settled
dust in indoor environments. Atmos.Em>iron. 2008,42 (7), 1449-
1460.
(29) Stubenrauch, S.; Hempfling, R.; Doetsch, P. D. G. Vorschlage
zur Charakterisierung und Quantifizierang pfadiibergreifender
Schadstoffexposition. Z, Umweltchem. OkotoxikoL 1999, 11,
219-226.
(30) Fromme, H.; Gruber, L.; Schlummer, M.; Wolz, G.; Bohrner, S.;
Angerer, J.; Mayerd, R.; Liebld, B.; Bolte, G. Intake of phthalates
and di(2-ethylhexyi)adipate: Results of the Integrated Exposure
Assessment Survey based on duplicate diet samples and
biomonitoring data. Em/iron. Int. 2007, 33 (8), 1012-1020.
(31) Wormuth, M.; Scheringer, M.; Vollenweider, M.; Hungerbiihler,
K. What Are the Sources of Exposure to Eight Frequently Used
Phthalic Acid Esters in Europeans? Risk Anal 2006, 23 (3),
803-824.
ES801354F
2380 » ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 43, NO. 7, 2009
Previoys
TOG
Next
-------
Research
Profiling Chemicals Based on Chronic Toxicity Results from the U.S. EPA
ToxRef Database
Matthew T. Martin, Richard S. Judson, David M. Reif, Robert J. Kavlock, and David J. Dix
National Center for Computational Toxicology, Office of Research and Development, U.S. Environmental Protection Agency, Research
Triangle Park, North Carolina, USA
BACKGROUND: Thirty years of pesticide registration toxicity data have been historically stored as
hardcopy and scanned documents by the U.S. Environmental Protection Agency (EPA). A signifi-
cant portion of these data have now been processed into standardized and structured toxicity data
within the EPA's Toxicity Reference Database (ToxRefDB), including chronic, cancer, develop-
mental, and reproductive studies from laboratory animals. These data are now accessible and mine-
able within ToxRefDB and are serving as a primary source of validation for U.S. EPA's ToxCast
research program in predictive toxicology.
OBJECTIVES: We profiled in vivo toxicities across 310 chemicals as a model application of
ToxRefDB, meeting the need for detailed anchoring end points for development of ToxCast predic-
tive signatures.
METHODS: Using query and structured data-mining approaches, we generated toxicity profiles from
ToxRefDB based on long-term rodent bioassays. These chronic/cancer data were analyzed for suit-
ability as anchoring end points based on incidence, target organ, severity, potency, and significance.
RESULTS: Under conditions of the bioassays, we observed pathologies for 273 of 310 chemicals, witli
greater preponderance (> 90%) occurring in the liver, kidney, thyroid, lung, testis, and spleen. We
observed proliferative lesions for 225 chemicals, and 167 chemicals caused progression to cancer-
related pathologies.
CONCLUSIONS: Based on incidence, severity, and potency, we selected 26 primarily tissue-specific
pathology end points to uniformly classify the 310 chemicals. The resulting toxicity profile classifi-
cations demonstrate tlie utility of structuring legacy toxicity information and facilitating the com-
putation of these data within ToxRefDB for ToxCast and other applications.
KEY WORDS: cancer, chronic toxicity, pesticides, relational database, toxicity profile. Environ Health
Perspect 117:392-399 (2009). doi:10.1289/ehp.0800074 available via http://dx.doi.org/ [Online
20 October 2008]
The U.S. Environmental Protection Agency
(EPA) and other regulatory agencies are inves-
tigating novel approaches to predict chemical
toxicity, with the major goals being to enable
the rapid screening of thousands of chemi-
cals that have not previously been character-
ized, to increase mechanistic understanding of
chemical toxicity, and to reduce the number
of animals required for toxicity testing. All of
these goals initially require high-quality in vivo
toxicity data in order to test and validate these
new approaches. To support U.S. EPA's
ToxCast effort (Dix et al. 2007), we have
created the structured and curated Toxicity
Reference Database (ToxRefDB) to tabulate
information from guideline in vivo toxicity
studies. ToxRefDB and related databases will
help support computational analysis and mod-
eling of the links from molecular interactions
through cellular and organ phenotypes all the
way to whole-animal toxicity. This transfor-
mation of existing toxicity data will facilitate a
transition to the National Research Council's
(NRC) vision for Toxicity Testing in the 21st
Century (Collins et al. 2008; NRC 2007). The
NRC envisions a focus on toxicity pathways
that will link molecular assays to toxicity out-
comes in humans and ecological species.
Traditional toxicity testing for risk assess-
ment of single compounds or limited groups
of compounds can cost millions of dollars
per chemical and years of effort. Since 1970,
the U.S. EPA has accumulated a vast store of
high-quality regulatory toxicity information
on thousands of compounds, most of which
has been inaccessible for computational analy-
ses. The curation and structuring of chemical
toxicity information into the readily accessible
ToxRefDB have created a valuable resource for
both retrospective and prospective toxicologic
studies. ToxRefDB initially focused on captur-
ing developmental rat and rabbit, multigenera-
tion reproduction rat, and chronic/cancer rat
and cancer mouse studies. In addition to the
data model, we developed a detailed toxicity-
based controlled vocabulary for all the study
types spanning clinical chemistry, pathology,
reproductive, and developmental effects.
An important initial application of
ToxRefDB is to provide anchoring of in vivo
toxicity data for the U.S. EPA's ToxCast
research program, which has been designed to
address the agency's needs for chemical prior-
itization by using state-of-the-art approaches
in high-throughput screening (HTS) and
toxicogenomics (U.S. EPA 2008b). Nearly
all of the ToxCast phase I chemicals are food-
use pesticide active ingredients that have
undergone a full suite of mammalian toxic-
ity tests, creating an unparalleled reference
set of toxicologic information. The complete
and highly standardized data set provided by
ToxRefDB facilitates analysis of the ToxCast
phase I chemicals across chemical, study type,
species, target organ, and effect. Additionally,
ToxRefDB serves as a model for other efforts
to capture quantitative, tabular toxicology
data from legacy and new studies and to make
these data useful for cross-chemical computa-
tional toxicology analysis.
Methods
Data characteristics. We collected reviews
of registrant-submitted toxicity studies,
known as data evaluation records (DERs), for
roughly 400 chemicals from the U.S. EPA's
Office of Pesticide Programs (OPP) within
the Office of Pollution Prevention and Toxic
Substances (OPPTS). The file types of the
DERs include TIFF, Microsoft Word, Word
Perfect, and PDF formats, some of which are
not directly text-readable. We indexed every
DER file based on a file name convention
that consisted of the pesticide chemical (PC)
code, study identification number (MRID),
study type identification number [based on
870 series OPPTS harmonized health effect
guidelines (U.S. EPA 1996)], species code,
review identification number (TXR), and a
review version code. The latter code identi-
fied the review as a primary review, secondary
review, supplemental review, updated execu-
tive summary, or a deficient review.
For the initial build of ToxRefDB, we
collected and indexed a total of 4,620 DERs
from OPP. These included five types of studies
Address correspondence to M.T. Martin, U.S.
Environmental Protection Agency, MD D343-03,
109 TW Alexander Dr., Research Triangle Park, NC
27711 USA. Telephone: (919) 541-4104. Fax: (919)
685-3399. E-mail: Martin.Matt@epa.gov
Supplemental Material is available online at http://
www.ehponline.org/members/2008/0800074/
suppl.pdf
We thank the Office of Pesticide Programs for
contributions to the ToxRefDB project, including
access to toxicity data evaluation records, scientific
consultation, and review of the manuscript by V.
Dellarco. We also thank D. Corum, K. McLaurin,
L. Peck, D. Rotroff, and D. Scoville for excellent
work entering data into ToxRefDB.
The U.S. EPA, through its Office of Research and
Development funded and managed the research
described here. It has been subjected to agency
review and approved for publication.
The authors declare they have no competing
financial interests.
Received 6 August 2008; accepted 20 October 2008.
392
VOLUME 117 | NUMBER 3 | March 2009 • Environmental Health Perspectives
Previous
TOC
-------
Profiling toxicity of chemicals using ToxRefDB
from a variety of species: developmental in rat
and rabbit, reproductive in rat, subchronic in
mouse and rat, and chronic or cancer in rat
and mouse. Approximately 1,000 DERs pro-
vided chronic and cancer data, and we selected
a subset of these for curation into the data-
base to yield data on 310 unique chemicals:
rat chronic/cancer studies on 283 chemicals,
and mouse cancer studies on 267 chemicals.
Each study assessed a single technical-grade
chemical's toxicity potential in a single species
and study type. The first portion of the DER
outlines the test substance, purity, lot/batch
numbers, MRID, study citation, OPPTS test
guideline, and reviewers of the study. The
executive summary captures all of the basic
study design information, including species
and strain, doses, number of animals per
treatment group, and any deficiencies in study
protocol.
Dose levels are listed in parts per million
and through food consumption and body
weight calculation or standard conversion
as milligrams per kilogram body weight per
day. Where possible, dose levels were listed as
milligrams per kilogram body weight per day
in ToxRefDB. The executive summary also
describes adverse effects observed at all dose
levels in the study. No observed adverse effect
level (NOAEL) and lowest observed adverse
effect level (LOAEL) are established based
on adverse effects. The adverse effects used to
derive NOAEL and LOAEL are referred to
as "critical effects" in this article, regardless of
their role in establishing reference dose levels
in regulatory determinations for a chemical.
The body of the DERs provides detailed
test material, animal information, and full
dose—response data in text and tables for a
variety of "effect types", including mortality,
clinical signs, clinical chemistry, hematology,
urinalysis, gross pathology, nonneoplastic
pathology, and neoplastic pathology. For each
effect type, we also specified an "effect target"
(e.g., liver as target organ) and "effect descrip-
tion" (e.g., hypertrophy).
ToxCast phase I chemicals also included
nonpesticidal chemicals such as perfluorinated
compounds, phthalates, and other industrial
chemicals. Although DERs and pesticide regis-
tration studies were not available for these
chemicals, there were often high-quality and
standardized chronic and other types of toxicity
studies available from the National Toxicology
Program, peer-reviewed literature, or other
sources. We organized and evaluated data from
these study reports and publications consistent
with the information from the DERs.
Information on chemical identity and
structure was provided by the U.S. EPA
DSSTox (Distributed Structure-Searchable
Toxicity) program (U.S. EPA 2007).
ToxRefDB outputs are linked to informa-
tion from other sources through the U.S. EPA
ACToR (Aggregated Computational
Toxicology Resource) database (Judson et al.
2008b; U.S. EPA 2008a). ACToR will also
serve as the primary portal for public access to
ToxRefDB and related outputs. ACToR stores
the HTS data being generated by the ToxCast
program and will link these HTS data with
traditional toxicity data from ToxRefDB and
other sources.
Relational model. In the development of
ToxRefDB, a relational model approach was
taken with input from other toxicity data-
base standards, including ToxML (Yang et al.
2006). The resulting data model is semi-
hierarchical in nature: a single compound can
be tested in multiple studies, each study can
contain multiple treatment groups, and mul-
tiple effects can be observed in each treatment
group. The data model is organized from a
chemical-centric viewpoint to allow data inte-
gration and exchange with other data sources
and to facilitate the linkage of the reference
toxicity information to chemical-specific data
generated using in vitro technologies (i.e.,
ToxCast). The relational model was then
implemented into a table structure with estab-
lished relationships that ensure data integ-
rity, updateability, and standardization [see
Supplemental Material, Figure 1 (http://www.
ehponline.org/members/2008/0800074/
suppl.pdfj.
Development of a toxicity-based controlled
vocabulary. The development of a controlled
vocabulary within ToxRefDB was neces-
sary for the standardization of data captured
across various studies and study types per-
formed over roughly 30 years. The nonredun-
dant list of terms across various information
domains provided data integrity and search-
ability. We based study type terminology on
the unique study types harmonized by the
Organisation for Economic Co-operation and
Development and the OPPTS (U.S. EPA
1996). Specificstandardized terminology for
study design was established for species/strain,
method/route of administration, and units for
dose and dosing duration. Treatment group-
related vocabularies were developed to estab-
lish the generation, gender, and dosing period.
A primary goal in evaluating the registrant-
submitted toxicity studies is to establish
NOAEL and LOAEL values for a variety of
categorical end points, including systemic, off-
spring, maternal, parental, developmental, and
reproductive toxicity across the various study
types. These categorical end points are captured
and normalized across studies for each effect
responsible for deriving the NOAEL/LOAEL.
The development of a toxicologic effect
vocabulary was approached in a domain-
specific manner. For example, we derived
clinical pathology terms from OPPTS guide-
lines and collected clinical pathology labo-
ratories and organ pathology terms from
various public resources, including the
National Toxicology Program's Pathology
Code Tables (2007). The vocabulary under-
went further standardization by mapping
all synonymous terms to a single nonre-
dundant value. We took a taxonomical
approach for establishing the finalized effect
vocabulary based on a three-tiered hierarchi-
cal model, with the effect type at the top,
followed by effect target and then effect
description. Examples of effect type include
clinical chemistry, hematology, urinalysis,
body weight, mortality, gross pathology,
nonneoplastic pathology, neoplastic pathol-
ogy, and developmental and reproductive
effects. Subclasses of these types include spe-
cific target organs (e.g., liver, lung, spleen) or
measured analytes (e.g., alanine aminotrans-
ferase, aspartate aminotransferase, choles-
terol). The specific combinations of effect
type and target are then further subclassed
based on a nonredundant descriptive term
(e.g., increase, decrease, hypertrophy, atro-
phy). For organ pathology terms, each target
organ has a set of regions, zones, and cell
types that characterize the site of toxicity.
The full effect vocabulary is available on the
ToxRefDB home page (U.S. EPA 2008c).
Data input. The ToxRefDB Data Entry
Tool was developed with Microsoft Access
providing the user interface for all initial data
input and is also available at the ToxRefDB
home page (U.S. EPA 2008c). After the initial
quality control (QC) steps discussed below,
the data are migrated to ToxRefDB, which is
implemented using the open-source MySQL
platform. Data entry followed a series of pro-
tocols outlined in the ToxRefDB Standard
Operating Procedure (SOP) documents that
define mapping of toxicologic information
to standardized fields, use of a standardized
vocabulary, and extraction of biologically and
statistically significant treatment-related effects.
Data QC and management, QC con-
sisted of 100% cross-checking of studies,
systematic updates of ToxRefDB to ensure
consistency across the studies, expert review
of data outputs, and external review by stake-
holders. All data entered into ToxRefDB have
undergone cross-checking, which entailed a
second person validating each entered value
based on the source information (primarily
DERs). Systematic QC involved querying the
database for potential inconsistencies (e.g.,
male-only effects being assigned to female
treatment groups, or systemic LOAEL being
set at multiple dose levels) along with updat-
ing vocabularies and related records. Expert
review was performed on data outputs of the
chronic/cancer rat or mouse studies, includ-
ing all of the end points captured in the data
tables of this publication. In addition to inter-
nal QC, an ongoing process allowing stake-
holders the opportunity to review ToxRefDB
Environmental Health Perspectives • VOLUME 117 I NUMBER 3 I March 2009
Previous
393
-------
Martin et al.
records is in place. The companies or regis-
trants that sponsor the data or support the
registration of the chemical are reviewing the
accuracy of the data relative to DERs and
other risk assessment documents. To date,
studies on 235 chemicals have been reviewed
by registrants, and comments from these
reviews indicate greater than 99% accuracy
in capturing treatment-related effects from
DERs. The stakeholder review process has
facilitated additional information from addi-
tional studies, DERs, and other risk assess-
ment documents to be collected and entered
into ToxRefDB.
Data output and analysis. The structured
toxicity information stored within ToxRefDB
can be extracted in various formats using
MySQL queries. For the purpose of provid-
ing computable outputs, that is, quantita-
tive outputs amenable to statistical analysis,
we used a consistent data output. The cross-
tabulated data output consisted of rows of
chemical information (e.g., CAS registry
number and chemical name) and columns
of end points or effects, with the cross sec-
tion being the lowest dose at which the effect
or end point was observed, that is, lowest
effect level (LEL) in mg/kg/day. Even though
NOAEL/LOAEL values can be queried from
the database, the current analysis uses LELs,
which do not reflect the NOAEL/LOAEL
regulatory determinations derived from
the studies and refer only to the minimum
dose at which a specific effect or group of
effects occurs. We used administered dose
levels rather than molar concentrations to
represent the chemically induced effects and
end points, because of uncertainties in the
pharmacokinetics linking administered dose
to tissue concentrations reinforcing the fact
that molecular weight alone cannot substi-
tute for dosimetry. Additional transformation
of the dosing information was performed,
including log-based and binning methods for
potency. For example, we developed a bin-
ning method for illustrating relative potency
to provide information into the sensitivity of
the end point from the perspective of treat-
ment dose. To derive nonarbitrary dosing
intervals, LEL for body weight changes were
analyzed and separated into equivalent quin-
tile bins (data not shown). The resulting bins,
< 15, < 50, < 150, < 500, and > 500 mg/kg/
day, were then applied to all end points. For
instance, a chemical that caused liver hyper-
trophy at 5 mg/kg/day would be assigned a 5,
at 25 mg/kg/day a 4, and so on. If the effect
was not observed, then a zero was assigned.
Table 1. Summary statistics for chronic/cancer rat and mouse studies entered into ToxRefDB.
Study
Total chronic/cancer
Rat
Mouse
Chemicals
310
283
267
No. of
studies
577
298
279
Treatment
groups
7,340
4,228
3,059
Treatment groups
with effects
3,082
1,721
1,344
Effects3
19,537
12,215
7,416
Critical
effects*
3,119
1,816
1,303
aTotal number of effect type, target, and description combinations assigned to any treatment group. ^Effects that are cri-
teria for establishing the study-specific NOAEL/LOAEL.
Rat
• Low incidence of toxicty
nThyroid and liver toxicants
• Liver toxicants
n Kidneytoxicants
• Spleen and anemia toxicants
niesticulartoxicants
n Cholinesterase inhibitors
• Body weight decrease
Mouse
• Low incidence of toxicty
n Lung tumorigens
• Liver tumorigens
• Spleen and anemia toxicants
n Liver toxicants (general)
n Cholinesterase inhibitors
• Body weight decrease
Chemical
Chemical
' f - •-''-• I
f jr. -.
-i
&
Mouse
-i '
.
. M3
, _j
No effect 2,048 0.015625
mg/kg/day mg/kg/day
Lowest effect level (LEL): -log2(LEL)
Figure 1. Unsupervised two-way hierarchical clustering of 207 effects in rat (A) and 112 effects in
mouse (6) with incidence > 5, for 310 chemicals with chronic/cancer toxicity data in ToxRefDB. Specific
clusters or classes based on associated toxicities are indicated by the color-coded chemical dendrogram:
seven clusters for rat, and six for mouse.
Additionally, log-transformed potency values
were derived using -Iog2 of LEL. We used
Iog2 to reflect the minimal dose spacing, that
is, doubling, typically used for in viva toxi-
cology studies. A constant value of 12 was
then added to zero-center the data, allow-
ing for zero to represent no observed effect.
Therefore, a value of 1 would be equivalent
to an effect at 2,048 mg/kg/day and 18 would
be equivalent to 0.015625 mg/kg/day. The
resulting data formats are highly amenable to
statistical data analysis, including descriptive
and predictive data-mining algorithms.
We carried out unsupervised two-way hier-
archical clustering across all chemicals of all
effects with incidence greater than 5, as well as
selected end points, based on log-transformed
potency values using Pearson's dissimilarity
measure for both chemicals and effects. This
analysis used Ward's method for linkage (Ward
1963) and the agglomerative clustering method
as implemented in the Partek Discovery Suite
(Partek Inc., St. Louis, MO). In order to assess
statistically significant species concordance
across different effects, a permutation study
was carried out. For each effect, the associa-
tion between chemical and effect for the cor-
responding rat and mouse study was randomly
permuted 1,000 times. We recorded the cross-
species concordance for all simulations (per-
mutations) and compared it with the observed
concordance, thus giving an estimate of the
concordance due purely to chance. Analyses
were carried out using R version 2.6.1 (Ihaka
and Gentleman 1996).
An initial 10% incidence cutoff was used
to filter out individual and groups of effects
for potential use in predictive modeling. This
cutoff was chosen following the results of a
related simulation study that demonstrated
high levels of sensitivity and specificity for
various machine learning methods on data
with at least a 10% hit rate for predicted end
points (Judson et al. 2008a). For other appli-
cations, it may be useful to add less frequently
occurring effects and end points.
Results
Summary profiles of the ToxRefDB chronic/
cancer data set. To date, ToxRefDB has
captured in vivo mammalian toxicity study
information from DERs for 411 conventional
pesticide active ingredients. This present
analysis focuses on the systemic toxicity and
cancer end points culled from chronic/cancer
rat or mouse studies on 310 of the chemicals
entered into ToxRefDB. ToxRefDB enabled
analysis to be performed along lexicologically
related axes, including by chemical, study
type, species, and effect. Study duration, dos-
ing methods, data quality, guideline adher-
ence, and sex were additional parameters for
data filtering. In looking across all chronic/
cancer rat and mouse studies, we assigned
394
VOLUME 1171 NUMBER 3 I March 2009 • Environmental Health Perspectives
Previous
-------
Profiling toxicity of chemicals using ToxRefDB
19,537 effects to 3,082 different treatment
groups in a total of 577 studies on 310 chemi-
cals (Table 1). Effects are a combination of
study type, species, effect type, effect target,
and effect description for a given chemical,
for example, chronic/cancer, rat, neoplastic
pathology, liver, and adenoma. Across the
19,537 effects, 1,135 unique effects were
observed, of which 484 were deemed criti-
cal effects, that is, criteria for establishing
NOAEL/LOAEL, in at least a single study.
The ToxRefDB chronic/cancer data set on
310 chemicals contained approximately 20,000
observed effects in rat or mouse studies. We
achieved a high-level view of a subset of these
data, and the relationships among chemical,
effect, and potency, by unsupervised two-way
hierarchical clustering of 207 rat (Figure 1A)
and 112 mouse (Figure IB) effects. For the
rat, the 283 chemicals separated into seven dis-
tinct clusters or classes of the chemicals based
on these toxicity profiles. Approximately 70
chemicals formed a cluster with an overall low
incidence of toxicity, whereas the remaining
chemicals displayed a unique set of toxicologic
properties. More than 80 chemicals clustered
as hepatotoxicants, and a subset of these also
caused thyroid toxicity. Ten of the 15 conazole
fungicides analyzed were in this hepatoxicity
cluster. Clusters of chemicals exhibiting kid-
ney, spleen/anemia, or testicular toxicities were
not enriched for a specific chemical structural
class. Cholinesterase inhibitors clustered sepa-
rately from other chemicals and were enriched
for organophosphates. In mouse, the 267
chemicals included clusters of cholinesterase
inhibitors, spleen/anemia toxicants, and hepa-
totoxicants comparable with that observed for
rat. Of the 112 total effects clustered in the
mouse, 28 of these were liver toxicities, dem-
onstrating the predominance of the liver as a
target organ in the mouse. The unsupervised
clustering of rat and mouse effects identified
concentrations of effects and chemicals that
were emphasized in subsequent, expert-driven
approaches to chemical classification.
Toxicity-based classification of chemi-
cals. The distribution of effects across effect
types (Figure 2A) revealed that nonneoplastic
pathologies dominate determination of sys-
temic NOAEL/LOAEL, demonstrating the
potential importance of this class of effects
or end points to chemical regulation. The
percentage of chemicals positive for an end
point in both rat and mouse, over the total
positive for the same end point in only the
rat or mouse, was defined as "species concor-
dance." Species concordance for nonneoplastic
pathology was 68%. Of the 167 chemicals
that caused neoplastic lesions in rat or mouse
chronic/cancer studies, 35% caused neoplastic
lesions in both rat and mouse. We observed
one or more pathologies in 273 of the 310
chemicals. The incidence of pathologic
response, analyzed by target organ and species,
was used to identify target organs for further
investigation (Figure 2B). More than 90% of
those 273 chemicals caused pathologies in the
liver, kidney, thyroid, lung, testis, or spleen.
Whereas individual effects relating to
highly detailed pathologic outcomes would
provide classifications with the highest bio-
logical specificity, the limitations of classifying
chemicals based solely on specific individ-
ual effects was apparent early in the analysis
of ToxRefDB data. Only 11 specific, indi-
vidual pathologic effects were observed for
more than 10% of the chemicals (Table 2).
25 50 75 100
Percent chemicals with observed
effect type
25 50 75
Percent chemicals with observed
pathology by target organ
Liver hypertrophy is the only common effect
across both species based on a 10% inci-
dence cutoff. In addition to low incidences
of detailed pathologic effects, biases based on
study design and pathology nomenclature
limited the overall ability to compare chemi-
cal toxicities when we used individual effects.
Grouping or aggregating related or near-syn-
onymous terms, such as liver adenoma, com-
bined adenoma/carcinoma, and carcinoma,
resulted in more informative and statistically
powerful sets of effects. Thus, the limitations
of classifying chemicals based solely on spe-
cific individual effects were addressed by cre-
ating biologically related groupings of effects.
Grouping tumor end points and extending
to include proliferative lesions. This aggre-
gative approach was illustrated by creating
groups of neoplastic end points and the exten-
sion of these groups to include nonneoplastic
proliferative lesions. The aggregation of neo-
plastic effects for each target organ resulted
in an increase in the number of useful group-
ings beyond the individual mouse liver tumor
effects shown in Table 2. However, the end
points were still limited to mouse liver and rat
thyroid neoplasia, based on an initial > 10%
incidence cutoff. Associating the neoplastic
end points with proliferative lesions increased
the number of target organs to include liver,
kidney, thyroid, lung, and testes. In general,
only neoplastic lesions are considered indica-
tive of rodent carcinogenicity. However,
including nonneoplastic proliferative lesions
provides a conservative model for assessing
and predicting rodent tumorigenic poten-
tial, based on the assumption that prolonged
proliferative response leads to eventual tumor
formation. A simulation study was performed
to assess whether the concordance between rat
and mouse effects occurred at a rate greater
than chance across neoplastic and prolifera-
tive classifications. Extending tumorigenicity
groupings to include proliferative lesions
significantly increased species concordance
across numerous target organs, including the
liver and kidney [see Supplemental Material,
Table 2. Pathology observed for > 10% of ToxRefDB
chemicals in chronic/cancer rat and mouse studies.
Target organ
Effect
Percent
observed
Figure 2. ToxRefDB chronic/cancer incidence data summarized by effect type (A) and by target organ
pathology (6) for 310 chemicals with rat or mouse studies. Blue bars, total percentage of chemicals with
that observed effect; black bars, percentage of chemicals for which that effect was used to derive systemic
NOAEL/LOAEL levels.
Rat
Liver Hypertrophy 25
Kidney Nephropathy 14
Liver Vacuolization 12
Thyroid Adenoma 11
Thyroid Hyperplasia 11
Mouse
Liver Hypertrophy 25
Liver Adenoma 21
Liver Necrosis 16
Liver Adenoma/carcinoma combined 14
Liver Pigmentation 14
Liver Carcinoma 12
Environmental Health Perspectives • VOLUME 117 I NUMBER 3 I March 2009
Previous
TOC
395
-------
Martin et al.
Figure2 (http://www.ehponline.org/members/
2008/0800074/suppl.pdf)].
Mapping oftoxicity end points to a can-
cer progression schema. Relationships between
effects and the relative severity of those effects
are not inherent to the database structure.
Figure 3A presents a conceptualization of the
end point progression schema in which chemi-
cals were scored from 0 to 5 for each target
organ, based on the severity of the effect, rang-
ing from no observed pathology (0) to neoplas-
tic lesions (5). End-point progression scoring
reduced the possible chemical classifications
to a single ordinal score (i.e., scores 0—5) for
each target organ. Figure 3B presents the dis-
tribution of end-point progression scores for
rat and mouse, liver and kidney. Examples
of the impact of this scoring system include
resmethrin, which caused treatment-related
increases in a preneoplastic lesion (i.e., hyper-
plastic nodules) in the liver without progressing
to a tumor. In contrast, metaldehyde caused
treatment-related increases in liver tumors but
was not identified as causing any preneoplas-
tic lesions, even though preneoplastic lesions
can be assumed to have occurred as a precur-
sor event to liver tumor formation. Using the
end-point progression scoring system allowed
reasonable comparison of these two chemicals,
if desired, by linking the preneoplastic score
of 4 for resmethrin, to the neoplastic score of
5 for metaldehyde, along the continuum of
end-point progression. The incidence of liver
pathology between rats and mice was com-
parable when we grouped end-point progres-
sion scores. More than 50% of the chemicals
tested resulted in a range of nonneoplastic to
neoplastic lesions (i.e., scores 2-5). However,
the relative severity for liver pathologic pro-
gression in mice was higher than in rats: 25
chemicals caused rat liver tumors, whereas 80
chemicals caused mouse liver tumors.
Selected end points for predictive mod-
eling. In addition to end points specific to
various target organs, chemicals were classi-
fied with respect to multigender, multisite,
or multispecies tumorigenicity (Table 3). Of
the 310 chemicals in the chronic/cancer data
set for which 240 chemicals were tested in
both species, 167 chemicals were classified
as tumorigens; 109 of those chemicals were
multigender, multisite, or multispecies tumori-
gens. Of the 283 chemicals tested in the rat,
42 chemicals were classified as multigender
and multisite tumorigens. Of 267 chemicals
tested in the mouse, 57 and 25 chemicals were
classified as multigender and multisite tumori-
gens, respectively. Of the 240 chemicals tested
in both species, 49 chemicals were classified as
multispecies tumorigens. The distribution of
relative potency values indicated that the rat
was commonly more sensitive than the mouse
for multigender and multisite tumorigenicity.
In the rat, 38% of the multigender and 45%
of the multisite incidences were at < 50 mg/
kg/day (i.e., relative potency values of 4—5),
compared with 23% and 28% in the mouse.
Conversely, 39% multigender and 28% multi-
site tumorigenicity occurred in the mouse at
> 500 mg/kg/day (i.e., relative potency value
1), compared with 17% and 10% in the rat.
Multispecies tumorigenicity was not achieved
at doses < 15 mg/kg/day, and 41% of inci-
dences occurred at > 500 mg/kg/day.
End point progression
Mouse
Rat
100 150
No. of chemicals
200
250
300
Figure 3. (A) ToxRefDB systemic toxicity and cancer outcomes represented along an end-point progression
continuum. This schema was used to derive a severity score for each chemical based on the maximum
value within a target organ. (6) Based on end-point progression, 310 chemicals were scored for liver and
kidney pathology in rat and mouse chronic/cancer studies. Clinical chemistry used in this analysis is limited
to target-organ-specific analytes (e.g., alanine aminotransferase for liver, and urea nitrogen for kidney).
Unsupervised and expert-driven approaches
to end-point selection and subsequent chemical
classification yielded near identical sets of target
organs from which to select specific effects or
aggregated effects. Based on incidence, severity,
potency, and significance, 25 end points from
chronic/cancer rat and mouse studies were
selected for subsequent ToxCast predictive
modeling (Figure 4A). The addition of multi-
species tumorigens raised the total to 26 end
points, each caused by 20 or more chemicals.
Besides the multispecies tumorigen end point,
16 of the end points were from rat studies
and 9 end points were from mouse. The same
four end points were characterized in both rat
and mouse liver, affording direct comparisons
across species for tumors, proliferative lesions,
apoptosis/necrosis, and hypertrophy. The only
other frequent target organ common to both
species was the kidney. Frequent rat-specific
target organs included thyroid, testis, and
spleen, whereas the only target organ specific to
mouse was the lung. Unsupervised hierarchical
clustering of the 16 rat end points (Figure 4B)
and the 9 mouse end points (Figure 4C) dis-
played the relative distribution of the selected
end points and chemicals. Of the 283 chemi-
cals with a rat chronic/cancer study, 218 were
positive in at least one of the selected end
points, whereas 155 of 276 chemicals with a
mouse cancer study were positive in at least
one selected end point. Rat and mouse end
points clustered primarily by target organ, with
distinct clusters of thyroid, spleen, kidney, and
liver toxicants in the rat. The high incidence of
liver tumorigens in the mouse drives chemical
groupings. However, chemicals causing or not
causing liver hypertrophy and necrosis appear
to segregate into two large groups of liver toxi-
cants. In both species, the selected chronic/can-
cer end points represent the robust patterns of
toxicologicresponse shown in Figure 2A and B.
A full listing of the chronic/cancer end points
derived from ToxRefDB for ToxCast predic-
tive modeling, with their associated LELs,
log-transformed potency, and relative potency
values, are available on the ToxRefDB home
page (U.S. EPA 2008c).
Discussion
Advancing alternative testing methods for
assessing chemical safety requires an informed
transition from the current toxicity testing to
systems that are higher throughput, more pre-
dictive, and not as dependent on the extensive
use of animals. To support this transition,
we created ToxRefDB to capture a rich set
of existing in viva laboratory animal toxicity
data on a group of environmentally relevant,
well-studied chemicals. Pesticide active ingre-
dients have comprehensive toxicity profiles
that are opportune data sets for creating a
bridge from in vivo to in vitro toxicology.
ToxRefDB digitizes and stores toxicity data
396
VOLUME 1171 NUMBER 3 I March 2009 • Environmental Health Perspectives
Previous
-------
Profiling toxicity of chemicals using ToxRefDB
in a structured and searchable format, and
using structured data mining methods makes
these data a computable resource for predic-
tive toxicology efforts such as the U.S. EPA's
ToxCast program (U.S. EPA2008b).
Individual toxicity effects based on unique
type, target, and description yielded only a small
number of in vivo end points across a significant
number of chemicals supportive of robust pre-
dictive modeling. However, grouping effects by
effect type and target often collapsed hundreds
of individual effects into a single end point,
common to dozens of chemicals. The goal was
to strike a balance between maintaining biologi-
cal specificity across a group of related effects
while increasing total incidence for effects
across a critical mass of chemicals. For example,
extending tumor end points to include prolif-
erative lesions increased not only total incidence
but also species concordance and thus increased
confidence in characterizing a chemical's poten-
tial toxicity. Grouping proliferative lesions
also addressed other potential factors, such as
changes in pathology nomenclature over time
(Wolf and Mann 2005) and reporting incon-
sistencies. Deriving end points based on groups
of effects yielded organ- and species-specific
end points in the liver, kidney, thyroid, testis,
spleen, and lung in rats or mice with a high
enough incidence across ToxRefDB chemicals
to support predictive modeling.
Another approach for addressing the
limitations of profiling chemicals based on
Table 3. Multigender, multisite, and multispecies tumorigens in ToxRefDB.
Chemical
Carbaryl
Dipropyl
isocinchomeronate
Fentin
Dazomet
Clodinafop-propargyl
Lactofen
Dimethoate
Malathion
Diuron
Dacthal
Isoxaflutole
Spirodiclofen
Diclofop-methyl
Cinmethylin
Imazalil
Nitrapyrin
Propoxur
Daminozide
Thiacloprid
Vinclozolin
Di(2-ethylhexyl)phthalate
Folpet
MGK(octacide264)
Iprodione
Cacodylicacid
Propyzamide
Oxadiazon
Resmethrin
Pyrithiobac-sodium
Bentazone
Fluthiacet-methyl
Metaldehyde
Triflusulfuron-methyl
Fludioxonil
Prodiamine
Tepraloxydim
Clofencet-potassium
Isoxaben
Pymetrozine
Topramezone
Triadimefon
Oryzalin
Simazine
Tebufenpyrad
Dichloran
Dimethenamid
Prosulfuron
Acetochlor
Ametryn
Oxytetracycline HCI
Bifenthrin
Disulfoton
Metam-sodium
Quizalofop-ethyl
Rat
Multigender
2
1
5
5
4
3
4
4
2
2
2
4
2
4
3
1
5
2
1
1
4
4
4
3
3
3
3
2
1
Multisite
2
1
5
5
4
3
5
4
3
2
2
2
4
3
3
2
4
4
5
2
4
2
2
1
4
4
4
3
3
3
2
3
1
Mouse
Multigender Multisite
1 1
1 1
4
3
4
5
4 4
1
1
1
1
1
3
3
3 3
2 1
1
1
1
1 1
5
5
2
2
2
1
1
1
1
5 5
5 5
4 5
4 4
Multispecies
2
1
4
4
4
4
4
1
1
1
1
1
4
3
3
3
2
2
1
1
1
1
1
1
3
3
3
2
2
2
2
2
2
2
1
1
1
1
1
1
1
Chemical
Tribufos
Amitraz
Fenoxycarb
Spiroxamine
Tefluthrin
Permethrin
Trifloxystrobin
Chloridazon
Triforine
Dichlorvos
Pyraclostrobin
Alachlor
Captan
Maneb
Azafenidin
Lindane
Fluazinam
Paclobutrazol
Acephate
Linuron
Propanil
Triasulfuron
Fipronil
Thiabendazole
Boscalid
Pendimethalin
Pyrimethanil
5,5-Dimethylhydantoin
Cyazofamid
Chloropicrin
Fenamiphos
Molinate
Chlorpyrifos-methyl
Fluoxastrobin
Fenitrothion
Cyproconazole
Prochloraz
Thiamethoxam
Bispyribac-sodium
Piperonyl butoxide
Propiconazole
Acifluorfen-sodium
Difenoconazole
Primisulfuron-methyl
Pyraflufen-ethyl
Thiodicarb
Fenoxaprop-ethyl
Buprofezin
Propargite
Dichlobenil
Quintozene
Tralkoxydim
Benomyl
Cloprop
Thiophanate-methyl
Rat
Multigender
5
5
4
N
N
4
3
2
2
2
1
1
4
2
2
N
N
N
Multisite
5
5
3
N
N
5
5
5
4
1
3
N
N
N
Mouse
Multigender Multisite
3
3
3
3
3
2
2
1
1
N
N
N
3
2
5
4
4
3
2
2
2
1
1
1
1
1
N
N
N
N
3
2
1
4
3
3
3
3
2
2
1
1
N
N
N
3
2
4
2
N
N
N
N
Multispecies
N
N
N
N
N
4
4
3
3
2
2
2
1
N
N
N
N
N
N
N
Relative potency: 5, < 15mg/kg/day;4, <50 mg/kg/day;3, < 150mg/kg/day; 2, <500mg/kg/day; 1,> 500 mg/kg/day; N, not assessed (no study available).
Environmental Health Perspectives • VOLUME 117 I NUMBER 3 I March 2009
Previous
TOC
397
-------
Martin et al.
individual toxicity effects was to compare the
severity of these effects across a continuum of
pathophysiology. Because the progression to
cancer (Hanahan and Weinberg 2000) and
organ-specific progression to tumorigenicity
(Cohen and Arnold 2008) have been well
characterized, we created a five-point sever-
ity scoring system to encode this. Using this
approach, ToxRefDB provides a quantitative
value associated with the key events in the
progression to tumor formation and cancer.
Incorporating additional information on the
severity of in vivo effects in ToxRefDB may
be fruitful in future modeling and predictive
toxicology efforts. Additional data not cur-
rently in ToxRefDB, including incidence data,
would have to be added for more detailed
dose-response analyses and assessment of the
magnitude of change for specific effects.
Because many of the tumors caused by
chemical exposure in ToxRefDB occur at
high doses that are many orders of magnitude
removed from potential human exposures, it
is useful to also consider multigender, multi-
site, and multispecies tumorigenicity in the
course of evaluating chemicals. Current U.S.
EPA cancer risk assessments use multisite and
multispecies tumorigenicity as indicators of
increased significance for tumor findings (U.S.
EPA 2005). Thus, the tumorigenic end points
selected for ToxCast predictive modeling
included multigender, multisite, and multi-
species tumorigens. Additional analyses of
these multiplicities in the tumorigenicity data
of ToxRefDB are under way, with the goal of
improving hazard assessments, chronic/cancer
study protocols, and future data requirements.
Success in predicting target-organ—specific
effects in ToxCast will depend on numerous
factors, including the target, species, and dose
response of the effects that are being predicted.
In the present analysis of ToxRefDB, we iden-
tified effects in the liver, kidney, thyroid, tes-
tis, spleen, and lung in rats or mice that we
will now attempt to predict using in vitro data
from ToxCast. Because species concordance
of the in vivo effects in ToxRefDB was fairly
limited, success in predicting species-specific
versus multispecies effects will be an interest-
ing outcome of ToxCast. The dose responses
for selected end points are also provided by
ToxRefDB, including log-transformed
potency values conducive to computational
analysis, and relative potency values that facili-
tate comparisons across chemicals and end
points. These quantitative data should facili-
tate development of new in vitro and in silica
methods to predict in vivo chemical toxicity.
Although numerous studies have evalu-
ated the use of biochemical, cell-based, and
genomic assays to build predictive models of
toxicity, these efforts have usually been lim-
ited to only a partial view of the complex biol-
ogy underlying tissue, organ, or whole-animal
toxicity. By probing such a broad spectrum of
biology in the hundreds of ToxCast assays, the
"toxicity signatures" will be optimally pre-
dictive and representative of a broad range
of in vivo toxicity end points. A variety of
statistical techniques and machine learning
approaches will be used to mine this com-
plex data set for toxicity signatures with high
sensitivity and specificity. These include
linear discriminant analysis, support vector
machines, and neural networks. In addition
to these automated approaches, more hypoth-
esis-driven, biologically based signatures will
assist in filling the large gap between molecu-
lar and phenotypic end points. It is expected
that assays of multiple types, probing multiple
pathways, will be required to predict in vivo
toxicity across a wide range of chemicals—
this is the approach taken within ToxCast
and ToxRefDB.
ToxRefDB continues to develop, add-
ing toxicity end points from additional study
types, including multigeneration reproductive
and prenatal developmental tests, for predictive
modeling in the ToxCast research program.
Besides expanding toxicity coverage to other
study types, ToxRefDB will expand in chemi-
cal coverage to include more nonpesticide
chemicals. As each of these ToxRefDB data
sets pass through U.S. EPA quality and clear-
ance processes, they will be made publicly
available through peer-reviewed publications,
ToxRefDB home page, and ACToR. The con-
tents of the entire database will be viewable and
searchable in the future through a Web-based
Chemical
Rat
Liver tumors
Liver proliferative lesions
Liver apoptosis/necrosis
Liver hypertrophy
Kidney nephropathy
Kidney proliferative lesions
Thyroid proliferative
Thyroid tumors
Thyroid hyperplasia
Testiculartumors
Testicular atrophy
Spleen pathology
Cholinesterase inhibition
Tumorigen
Multisite tumorigen
Multigender tumorigen
Mouse
Liver tumors
Liver proliferative lesions
Liver apoptosis/necrosis
Liver hypertrophy
Kidney pathology
Lung tumors
Tumorigen
Multisite tumorigen
Multigender tumorigen
Thyroid hyperplasia
Thyroid tumors
Thyroid proliferative lesions
Testiculartumors
Testicular atrophy
Kidney proliferative lesions
Spleen pathology
Kidney nephropathy
Cholinesterase inhibition
Liver tumors
Liver proliferative lesions
Liver apoptosis/necrosis
Liver hypertrophy
Multigendertumorigen
Multisite tumorigen
Tumorigen
Kidney pathology
Liver necrosis/apoptosis
Liver hypertrophy
Lung tumors
Multisite tumorigen
Multigendertumorigen
Liver tumors
Liver proliferative lesions
Tumorigen
Percent chemicals with observed end point
(Rat: 283 chemicals; mouse: 267 chemicals)
Lowest effect level (LEL): -Log2(LEL)
No effect
2,048
mg/kg/day
0.015625
mg/kg/day
Figure 4. (A) The 16 rat and 9 mouse ToxRefDB end points from chronic/cancer studies selected for ToxCast predictive modeling. Two-way hierarchical clustering
of the rat (B) and mouse (C) end points based on log-transformed potency values. Dose and potency values for all chemicals relative to these 25 end points are
provided on the ToxRefDB home page (U.S. EPA 2008c).
398
VOLUME 117 I NUMBER 3 I March 2009 * Environmental Health Perspectives
Previous
TOC
-------
Profiling toxicity of chemicals using ToxRefDB
query tool located on the ToxRefDB website
(U.S.EPA2008c).
ToxRefDB offers unparalleled amounts
of legacy toxicity information on environ-
mental chemicals captured in a structured
format, providing a platform for repeated and
updated chemical characterizations. Creating
the ability to search and filter across 30 years'
worth of toxicity data required extensive
amounts of data normalization, annotation,
and curation and was made possible through
the development of a robust standardized
vocabulary for the fields and data elements
within ToxRefDB. In the present study, we
used chronic toxicity data in ToxRefDB to
derive toxicity profiles for the ToxCast phase I
chemicals, yielding a set of toxicity-based and
predictable end points. In future applications
of ToxRefDB, researchers, risk assessors, and
regulators will use the database for retrospec-
tive and modeling projects looking across a
large landscape of chemical and toxicity space.
REFERENCES
Cohen SM, Arnold LL. 2008. Cell proliferation and carcino-
genesis. J Toxicol Pathol 21:1-7.
Collins FS, Gray GM, Bucher JR. 2008. Transforming environ-
mental health protection. Science 319:906-907.
Dix DJ, Houck KA, Martin MT, Richard AM, Setzer RW, Kavlock
RJ. 2007. The ToxCast program for prioritizing toxicity test-
ing of environmental chemicals. Toxicol Sci 95:5-12.
Hanahan D, Weinberg RA. 2000. The hallmarks of cancer. Cell
100:57-70.
lhaka R, Gentleman R. 1996. R: a language for data analysis and
graphics. J Comput Graph Stat 5: 299-314.
Judson RS, Elloumi F, Setzer RW, Li Z, Shah I. 2008a. A com-
parison of machine learning algorithms for chemical toxic-
ity classification using a simulated multi-scale data model.
BMC Bioinformatics 9:241 doi:10.1186/1471-2105-9-241
[Online 19 May 2008].
Judson RS, Richard AM, Dix DJ, Houck K, Elloumi F, Martin
MT et al. 2008b. ACToR—Aggregated Computational
Toxicology Resource. Toxicol Appl Pharmacol doi:10.1016/j.
taap.2007.12.037 [Online 11 June 2008].
National Research Council. 2007. Toxicity Testing in the 21st
Century: A Vision and a Strategy. Washington, DC:National
Academies Press.
National Toxicology Program. 2007. Pathology Code Tables.
Available: http://jones.mehs.nih.gov:8080/pct/ [accessed
7 February 2008].
U.S. EPA (U.S. Environmental Protection Agency]. 1996. OPPTS
Harmonized Test Guidelines Available: http://www.epa.gov/
opptsfrs/publications/OPPTS_Harmonized/870_Health_
Effects_Test_Guidelines/[accessed 10 January 2008].
U.S. EPA. 2005. Guidelinesfor Carcinogen Risk Assessment. Risk
Assessment Forum. Washington, DC:U.S. Environmental
Protection Agency.
U.S. EPA (U.S. Environmental Protection Agency]. 2007.
Distributed Structure-Searchable Toxicity Database.
Available: http://www.epa.gov/ncct/dsstox/ [accessed
14 February 2008].
U.S. EPA (U.S. Environmental Protection Agency]. 2008a.
ACToR Home. Available: http://www.epa.gov/ncct/actor/
[accessed 29 September 2008].
U.S. EPA (U.S. Environmental Protection Agency]. 2008b.
ToxCast Program. Available: http://www.epa.gov/ncct/
toxcast/ [accessed 12 June 2008].
U.S. EPA (U.S. Environmental Protection Agency]. 2008c.
ToxRefDB Home. Available: http://www.epa.gov/ncct/
toxrefdb/ [accessed 6 October 2008].
Ward JH. 1963. Hierarchical grouping to optimize an objective
function. J Am Stat Assoc 58:236-244.
Wolf DC, Mann PC. 2005. Confounders in interpreting pathology
for safety and risk assessment. Toxicol Appl Pharmacol
202:302-308.
Yang C, Benz RD, Cheeseman MA. 2006. Landscape of current
toxicity databases and database standards. Curr Opin
Drug DiscovDev 9:124-133.
Environmental Health Perspectives • VOLUME 117 I NUMBER 3 I March 2009
Previous
399
-------
Reproductive Toxicology 28 (2009) 209-219
I I N! Ml R
Contents lists available at ScienceDirect
Reproductive Toxicology
journal homepage: www.elsevier.com/locate/reprotox
Profiling the activity of environmental chemicals in prenatal developmental
toxicity studies using the U.S. EPA's ToxRefDB^
Thomas B. Knudsen3'*, Matthew T. Martin3, Robert J. Kavlock3, Richard S. Judson3,
David J. Dix3, Amar V. Singhb
' National Center for Computational Toxicology (NCCT), Office of Research and Development, U.S. Environmental Protection Agency, Research Triangle Park, JVC 27711, United States
b Lockheed Martin, contractor to NCCT, Research Triangle Park, JVC, United States
ARTICLE INFO
Article history:
Received 20 March 2009
Received in revised form 31 March 2009
Accepted 31 March 2009
Available online 10 April 2009
Keywords:
Database
Environmental chemicals
Pesticides
Developmental toxicity
ABSTRACT
As the primary source for regulatory developmental toxicity information, prenatal studies characterize
maternal effects and fetal endpoints including malformations, resorptions, and fetal weight reduction.
Results from 383 rat and 368 rabbit prenatal studies on 387 chemicals, mostly pesticides, were entered
into the U.S. Environmental Protection Agency's (EPA) Toxicity Reference Database (ToxRefDB) using
harmonized terminology. An initial assessment of these data was performed with the goal of profiling
environmental chemicals based on maternal and fetal endpoints for anchoring in vitro data provided in the
EPA's ToxCast™ research program. Using 30 years worth of standard prenatal studies, maternal and fetal
effects were culled from the database and analyzed by target-description fields and lowest effect levels
(LELs). Focusing on inter-species comparison, the complexity of fetal target organ response to maternal
dosing with environmental chemicals during the period of major organogenesis revealed hierarchical
relationships. Of 283 chemicals tested in both species, 53 chemicals (18.7%) had LELs on development
(dLEL) that were either specific, with no maternal toxicity (mLEL), or sensitive (dLEL
-------
210
T.B. Knudsen et al. / Reproductive Toxicology 28 (2009) 209-219
hood. Standard practice for assessing disruptions in embryogenesis
involves testing pregnant laboratory animals of two species,
typically rats and rabbits, exposed during the period of major
organogenesis and evaluated just prior to term along with mon-
itoring maternal status throughout pregnancy. Under this design
the major manifestations of developmental toxicity may express as
one or more of a number of possible endpoints such as intrauterine
death, fetal growth retardation, structural variations and abnormal-
ities [5,6]. Predictive modeling of developmental toxicity requires a
computational framework that can integrate mechanistic data with
high-quality toxicity data from in vivo studies.
EPA's Toxicity Reference Database (ToxRefDB) has been imple-
mented with animal-based toxicity data from chronic/cancer rat
and cancer mouse studies [7], multi-generation reproduction rat
studies [8], and prenatal developmental toxicity studies in rats and
rabbits (described here). The data has been manually entered from
source documents representing EPA's Office of Pesticide Programs
(OPP) reviews of registrant-submitted guideline studies known
as Data Evaluation Records (DERs). The initial build of ToxRefDB
entered these data for 387 chemicals that include 280 chemicals
within Phase-I of EPA's ToxCast™ research program [4]. Here we
describe the implementation of ToxRefDB for prenatal developmen-
tal toxicity studies. Experimental protocols in general follow EPA
Health Effects Test Guidelines OPPTS 870.3700 [9] or the preceding
OPP 83-3 guideline [10], and are similar to the OECD guideline for
prenatal developmental toxicity testing [11].
Databases of birth defect registries [12,13], developmental toxi-
cology literature (http://rsi.ilsi.org/Projects/devtoxsar.htm) [14,15],
and animal studies (http://ntp.niehs.nih.gov/) [6,16] have generally
looked to identify relationships within classes of toxic agents, devel-
opmental outcomes, test species and human populations (see [17]).
ToxRefDB represents the first large-scale implementation of its kind
for profiling the activity of environmental chemicals based on a
comprehensive analysis of source data for a broad range of end-
points relevant to EPA risk assessments, including developmental
toxicity.
The present study describes the initial build of ToxRefDB for
prenatal developmental toxicity studies (herein referred to as
ToxRefDB_prenatal). Through this implementation, a detailed anal-
ysis is possible to link observational relationships within classes
of toxic agents and developmental outcomes in rats and rabbits.
Rather than an exhaustive analysis of chemical-endpoint linkages,
the present study was designed to identify and evaluate key hier-
archical relationships that represent the primary determinants of
developmental toxicity. Focusing on inter-species comparison, the
study goals were to evaluate: (1) the complexity of fetal target organ
response to maternal dosing with environmental chemicals dur-
ing the period of maj or organogenesis, e.g., how many times target
endpoints were affected by chemical; (2) the relative sensitivity
and specificity of maternal and fetal parameters in comparing these
responses between rat and rabbit test species; and (3) how many
times each chemical was counted by target endpoint, e.g., how
many chemicals in the set produce a certain effect. Profiling devel-
opmental toxicity in this manner has revealed a number of findings
that are consistent with previous database studies of developmen-
tal toxicology, some that differ with those studies, and some novel
relationships. The novel data model reported here is envisaged to
provide an important public resource for mechanistic modeling and
predictive understanding of developmental processes and toxici-
ties.
2. Implementation and methods
2.1. Data sources
The initial build of ToxRefDB [7] indexed 4618 DERs of different study types
(chronic/cancer, sub-chronic, reproductive, developmental) and test species (mostly
rat, mouse, and rabbit). The DERs were from EPA's Office of Pesticide Programs
(OPP) within the Office of Prevention, Pesticides and Toxic Substances (OPPTS).
Source DERs consisted of 3775 printed documents optically scanned to ".tiff flies;
837 WordPerfect documents; and 6 documents in other electronic file formats.
Each DER was indexed by filename convention of Pesticide Chemical (PC) Code,
Master Record Identification (MRID) number, study type identification number
(based on most relevant 870 series OPPTS harmonized health effect guidelines),
species code, and review identification number and version code [7,8]. Informa-
tion on chemical identity and structure was provided by the EPA DSSTox program
(www.epa.gov/ncct/dsstox/index.html). The work described here specifically covers
a subset of 1318 DERs indexed for 'prenatal developmental toxicity' denoted by the
filename extension 3700, and subsequently referred to here as ToxRefDB.prenatal.
2.2. Source vocabulary
The use of standardized nomenclature is essential for ToxRefDB operations. An
internationally harmonized terminology for developmental toxicology was estab-
lished in 1997 by the International Federation of Teratology Societies (IFTS) [18].
A subsequent series of workshops on terminology development eliminated certain
ambiguities and established working definitions for malformations and variations
[ 19-21 ]. The DevTox lexicon was downloaded from www.DevTox.org. An enhanced
annotation system was used by ToxRefDB in which 895 terms from the harmonized
nomenclature was joined with standardized terms from the OECD-OPPTS vocabu-
lary 111 ] to generate a thesaurus of 988 non-redundant terms that apply to maternal
and developmental endpoints. In the enhanced system, 'description' annotates the
particular apical endpoint or phenotype (observation) and 'target' annotates coarse
regional anatomy (localization). The description-target fields represent the basic
observational effects entered into ToxRefDB.prenatal.
2.3. Data entry and quality assurance
The data entry tool was developed in Microsoft Access® and implemented using
an open source MySQL™ platform [7]. The relational model took inputs from ToxML
[22] and included metrics for data integrity, quality, updateability, and standardiza-
tion. Quality control (QC) consisted of 100% cross-checking of studies, systematic
updates of ToxRefDB to ensure consistency across the studies, expert review of data
outputs, and external review by registrants. All data entered into ToxRefDB have
undergone cross-checking, which entailed a second person validating each entered
value based on the source information (primarily DERs). Systematic quality control
involved querying the database for potential inconsistencies (e.g., fetal effects being
assigned to the maternal treatment group) along with updating vocabularies and
related records [7[.
2.4. Source information
The 1318 DERs for prenatal developmental studies encompassed 4896 dose
groups. Doses were expressed as mg/kg-d where available since the studies were
conducted via gavage. Endpoint parameters entered as 'adult' included maternal
body weight gain, food and water consumption, fertility and pregnancy, and other
general maternal effects. Parameters entered as 'fetal' included fetal weight reduc-
tion, skeletal variations, malformations and other pathologies. Due to the two
annotation systems used to enter data into ToxRefDB (DevTox.org and OECD-OPPTS)
different expressions of fetal wastage cross-reference maternal and conceptal fea-
tures. For that reason, the present analysis lumped all expressions of fetal wastage,
including pre-implantation loss, implantation failure, resorptions, fetal death, preg-
nancy loss as a maternal feature under the category of'pregnancy-related losses'.
Any maternal or fetal outcome tagged as a'critical effect' inToxRefDB was reported in
the DER to have occurred at the minimum dose for which any specific effect or group
of effects had been observed (LEL, Lowest Effect Level); the next lowest dose being
the NEL (No Effect Level). Although some of these dose levels may represent the
NOAEL and LOAEL (No Observed Adverse Effect Level, and Lowest Observed Adverse
Effect Level) used for risk assessment purposes across different study types, the ter-
minology used here (NEL, LEL) is intended specifically to rank chemical endpoints
and endpoint-combinations on maternal and fetal parameters. As such, these terms
are used without regulatory implications.
2.5. ToxRefDB data extraction
Relational data were expressed using specific SQL™ queries and global data
dump to a sortable data grid having rows of exposure conditions and columns of
input/output criteria. Input source information included details on study design such
as unique study identifier (MRID), chemical CAS registry number (CASRN), route of
administration, exposure window, and dose level (mg/kg-d). Output endpoint effects
included details on evaluation criteria such as the biological compartment (adult,
fetus), type of effects (developmental, systemic), their localization and phenotype
(target, description), and any LEL noted (maternal mLEL, or developmental dLEL).
Because ToxRefDB entered data for description-target effects individually when
more than one effect may have occurred within the same fetus or litter, the data
grid replicated rows if more than one treatment-related effect was entered for the
same dose group in a particular study.
Previous
TOC
-------
T.B. Knudsen et al. / Reproductive Toxicology 28 (2009) 209-219
211
Table 1
Summary statistics for prenatal developmental toxicity studies entered into ToxRefDB (December31, 2008).
ToxRefDB_prenatal'
Rat
Rabbit
Ratio1
Normalized^
Species biash
A. Input source information
Number of studies entered (MRID)
> Studies passing acceptability criteriab
Number of chemicals represented (CASRN)
383
357
372
368
325
320
1.04
1.10
1.16
0.00
0.08
0.16
Number of dose groups (mg/kg-d) represented0
> Dose groups by oral administration"1
B. Output endpoint effects
Number of dose-effect groups recorded6
> Maternal endpoint effects (pregnancy)
> Reduced maternal weight gain
> Resorptions/fetal loss
> Fetal endpoint effects (developmental)
> Fetal weight reduction
> Developmental defects
2469
2463
5592
2429
596
262
1588
182
1383
2327
2307
4749
2462
482
498
716
95
611
1.06
1.07
1.18
0.99
1.24
0.53
2.22
1.92
2.26
0.03
0.04
0.15
-0.10
0.23
-1.00
1.06
0.86
1.09
Rabbit
Rabbit
Rat
Rat
Rat
Denotes prenatal developmental toxicity studies.
Acceptable guideline pre-1998, acceptable guideline post-1998, acceptable non-guideline.
Adult and fetal evaluation in the same study counts as separate dose groups.
Includes gavage, intubation, feed; residual routes of administration were dermal, subcutaneous, or not indicated.
Total number of endpoint effects for mother, fetus or pups recorded across all dose groups.
Rat to rabbit ratio.
: Inputs normalized to studies entered and outputs normalized to dose groups (mg/kg-d) represented and Iog2-transformed.
Rule for species bias = normalized ratio falls outside the range -0.02 < ratio <+0.46 based on 95% confidence interval on the mean.
2.6. Data analysis and visualization
Associations for exposure-effect and effect-effect were built across dose groups
for all prenatal developmental studies. Summary statistics for each treatment-
related effect extracted from ToxRefDB was based on target-description entity and
its higher level classification by embryological system. Dose values in mg/kg-d were
used to calculate potency for each chemical in rat and/or rabbit at the maternal,
developmental, and categorical LEL. For consistency the data transformation rule
used here was the same one applied in the chronic/cancer and multi-generation
reproduction studies [7,8]: LEL mg/kg-d extracted when available for maternal and
developmental effects, else blank; compute -log2(LEL mg/kg-d), else zero (LTD
present) or blank; and add constant = 12 to scale the data between ~0.0 (very low
LEL) and ~20 (very high LEL). Unsupervised two-way hierarchical clustering used
Pearson's dissimilarity measure for both chemicals and effects. This analysis used
Ward's method for linkage [23] and the agglomerative clustering method and was
implemented in R version 2.6.1 [24]. Clusters of chemicals were identified based on
a distance height cutoff of four.
3. Results
3.1. Input source information
Table 1A summarizes several fields of input source informa-
tion for ToxRefDB.prenatal. With few exceptions (<2% DERs) the
information entered derived from studies that evaluated chem-
ical effects in pregnant rats or rabbits. The number of studies
entered by MRID was 383 for rat and 368 for rabbit, with the few
exceptions being mouse, hamster and other species. Due to the pre-
dominance of rat and rabbit studies we focus here on these two
species. OPP acceptability criteria [10] designated 79.1% studies
as acceptable/guideline (pre-1998); 5.5% as acceptable/guideline
(post-1998); 6.3% as acceptable/non-guideline; 3.5% with deficient
evaluation; and 5.6% as unacceptable. No attempt was made to re-
evaluate the acceptability or deficiencies of the studies relative to
the different guidelines; as such, the results for all studies were
considered as presented in DERs.
The number of chemicals by CASRN represented across rat
or rabbit studies was 387. About 280 of these overlap with 320
ToxCast™ phase-I chemicals; therefore, a number of chemicals
entered into ToxRefDB.prenatal are not currently represented in
ToxCast™ and some ToxCast™ chemicals cannot draw on ToxRefDB
for direct information on developmental effects. Although the preg-
nant female was the usual exposure unit for these studies we
emphasize total number of dose groups (mg/kg-d) counted across
chemicals: 2469 for rat studies and 2307 for rabbit studies (Table 1).
As such, dose groups are replicated in adult and fetal evaluations
for the same study. The usual (>97.7%) route of exposure was oral
(gavage, intubation, feed).
Information about dosing interval (start, finish, duration) is sum-
marized schematically in Fig. 1. This plots the exposure design in
ToxRefDB.prenatal based on cumulative start and completion dates
dosing period
RAT
RABBIT
10 15
gestation days
25
30
Fig. 1. Exposure design summarized for all dose groups entered into ToxRefDB pre-
natal developmental toxicity studies. Gestational period for rat and rabbit mapped by
gestation days (CD) from fertilization (CD 0) through usual parturition, CD 21-22
in rat and CD 30-32 in rabbit. The gray shaded histogram graphs the cumulative
number of dosing groups across DERs for rat (2469) and rabbit (2327). The fuzzy
exposure window across ToxRefDB is indicated by the most frequent start and finish
days for dosing (e.g., CD 6-17 in rat and CD 6-20 in rabbit). Superimposed is the
period of dosing for ICH 4.1.3 Segment II study covering primitive streak formation
through palatal closure. The shaded arrowhead denotes the usual time of evaluation
inguideline rat (CD 20) and rabbit (CD 29) studies. A few ToxRefDB studies extended
into the postnatal period for rats (postnatal day 3, line) and rabbits (postnatal day
42, not indicated).
Previous
TOC
-------
212
T.B. Knudsen et al. / Reproductive Toxicology 28 (2009) 209-219
Table 2
Distribution of developmental effects across ToxRefDB dose groups.
ToxRefDB-prenatai
Number of dose groups (mg/kg-d) represented
Dose-effect groups with fetal (developmental) effects1
Skeletal defects
> Appendicular
> Axial
> Cranial
Orofacial defects
> Cleft lip/cleft palate
Neurosensory defects
> Brain
> Optic
Cardiovascular defects
> Heart
> Major vessels
Urogenital defects
> Renal
> Ureter
> Genital
Other visceral defects (splanchnic)
Body wall defects (somatic)
Rat
2469
1404
956
185
640
126
41
19
28
15
13
8
6
2
86
42
40
4
16
53
Rabbit
2327
709
366
77
241
48
18
5
22
13
9
10
5
5
5
2
2
1
23
14
Ratiob
1.06
1.98
2.61
2.40
2.66
2.63
2.28
3.80
1.27
1.15
1.44
0.80
1.20
0.40
17.20
21.00
20.00
4.00
0.70
3.79
Normalized^
-
0.90
1.30
1.18
1.32
1.31
1.10
1.84
0.26
0.12
0.45
-0.41
0.18
-1.41
4.02
4.31
4.24
1.92
-0.61
1.84
Species biasd
Rabbit
Rabbit
Rabbit
Rabbit
Rabbit
Rabbit
Rat
Rat
Rat
(Rat)
Rabbit
Description-target terms for developmental effects (DevTox) integrated by systems-based ontology.
Rat to rabbit ratio.
Normalized to number of dose groups (mg/kg-d) represented and Iog2-transformed.
Rule for species bias = normalized ratio falls outside the range 0.60 < ratio < 1.91 based on 95% confidence interval on the mean.
of oral dosing across gestation in 2469 dose groups for the rat and
2327 dose groups for the rabbit. In general, the dosing period for
OECD guidelines coincided with the ICH 4.1.3 Segment II study
guidelines to cover major events in morphogenesis and organogen-
esis for these species; however, a number of studies in rabbit had an
earlier onset than these guidelines and in both species a minority
of studies extended treatment to near term or postnatal days.
3.2. Output endpoint effects
Table IB summarizes several fields of treatment-related effects.
These are sorted by adult and fetal compartments (Table IB). The
endpoint effects recorded for the pregnant dam or doe included
general maternal parameters, food and water consumption, and
body weight gain as well as pregnancy-related indicators such as
increased resorptions and fetal wastage (pre- /post-implantation
losses, resorptions, intrauterine deaths). Fetal endpoints included
fetal weight reduction, structural abnormalities or variations, and
general fetal pathology. A small number of newborn observations
are not considered here. It should also be noted that the post-
treatment gestational interval was relatively longer in rabbits than
rats (Fig. 1); hence, any treatment-related effects that might have
been reversible or associated with developmental delay are more
likely to be detected in rats.
Summarizing endpoint effects by dose group is a logical way to
array the response between rat and rabbit across a large number
of conditions. In this approach, each endpoint effect is linked to
a discrete condition (dose, chemical, species). Counting the num-
ber of effect-condition linkages in a class or subclass of endpoints
provides a qualitative measure of that response by its representa-
tion across the chemical spectrum. It is important to recognize that
ToxRefDB data entry tracks individual target-description entities
for each discrete condition. As such, this read-out can artificially
amplify or de-emphasize specific classes of endpoints. ToxRefDB
registered 5592 effect-condition linkages in rat (average 2.3 effects
per non-zero dose group) and 4749 effect-linkages in rabbit (aver-
age 2.0 effects per non-zero dose group). Aggregating these effects
into higher level classifications revealed obvious species differ-
ences in the representation of endpoints sorted by adult and fetal
compartment (Table IB). Using a simple rule to compute over-
representation, resorptions and fetal losses were more prevalent
in rabbit, and fetal weight reduction and developmental defects in
the rat. A species bias could arise from complex factors such as bio-
logical variation in embryology, differences in maternal behavior
or physiology, sensitivity to various xenobiotic disturbances, or the
time between dosing and evaluation.
3.3. Developmental defects
To gain deeper insight into the species response we next exam-
ined the spectrum of developmental (fetal) effects across all studies.
Individual effect-condition linkages were counted for 988 features
in the enhanced DevTox thesaurus. This iterated 1404 and 709
developmental (fetal) effects across dose groups in rat and rab-
bit, respectively and covered 293 of 988 (29.7%) target-description
terms. Representation of individual effects and their occurrences
in ToxRefDB is dependent on the nature of the embryological sys-
tems from which the observation was originally made. Skeletal
defects, for example, are highly represented in part because most
bone elements are entered into the database as individual targets
(vertebrae, ribs, femur and so forth) and then further annotated
by a range of elementary descriptions (absent, incomplete ossifi-
cation, misshapen, bent and so forth). Other systems with isolated
occurrences of malformation such as the heart, brain or eye have
relatively low representation in part because they are annotated as
individual targets.
Given this caveat, we aggregated defects into specific embry-
ological systems and focused on this representation across species
(Table 2). Cross-species differences exist for some effects or groups
of effects aggregated by target system. Although we did not analyze
each skeletal element by abnormality or variation, aggregating the
individual occurrences into regional anatomy showed a similar dis-
tribution of response across species (axial > appendicular > cranial).
For regional orofacial defects (palate, jaw, hyoid) we find over-
representation of cleft palate in the rat. Urogenital defects (renal,
ureter, reproductive) are also highly over-represented in rat and
Previous
TOC
-------
T.B. Knudsen et al. / Reproductive Toxicology 28 (2009) 209-219
213
Table 3
Distribution of LEL effects across ToxRefDB dose groups.
ToxRefDB-prenatal
Dose-effect groups at mlEl or dlEl (total effects)
Dose-effect groups atmLEL (maternal)
Maternal body weight gain
Maternal-pregnancy losses
Embryo-fetal losses
Dose-effect groups atdlEl (fetal)
Fetal weight reduction
Variations and abnormalities
> Skeletal defects
> Appendicular
> Axial
> Cranial
> Urogenital defects
> Renal
> Ureter
> Genital
> Orofacial defects
> Cleft lip / cleft palate
> Neurosensory defects
> Brain
> Eye
> Cardiovascular defects
> Heart
> Major vessels
> Other visceral defects (splanchnic)
> Body wall defects (somatic)
Rat
1711
996
976
28
45
715
95
609
492
97
320
66
34
17
10
7
20
10
6
4
2
4
2
2
3
9
Rabbit
1363
1013
991
115
91
350
57
288
189
36
120
25
4
2
2
0
11
2
11
8
3
7
3
4
10
0
Ratio
1.26
0.98
0.98
0.24
0.49
1.62
1.32
1.68
2.07
2.14
2.12
2.10
6.75
6.75
3.97
>5.56
1.44
3.97
0.43
0.40
0.53
0.45
0.53
0.40
0.24
>7.14
Normalized1
-
-0.36
-0.36
-2.37
-1.35
0.70
0.40
0.75
1.05
1.10
1.08
1.07
2.75
2.75
1.99
>2.47
0.53
1.99
-1.21
-1.33
-0.92
-1.14
-0.92
-1.33
-2.07
>2.84
Species biasb
Rabbit
Rabbit
Rabbit
Rabbit
Rat
Rat
Rat
Rat
Rat
Rat
Rat
Rat
Rat
Rabbit
Rabbit
Rabbit
Rabbit
Rabbit
Rabbit
Rabbit
Rat
Normalized to number of dose groups with critical effects and Iog2-transformed.
iNoiiiianzeu LO nuiiiuei 01 uose gioups wan cinicai eiiecis anu logz-Liansioinieu.
Rule for species bias = normalized ratio falls outside the range -0.26 < ratio < 0.88 based on 95% confidence interval on the mean.
somatic body wall defects (e.g., umbilical hernia, diaphragmatic
hernia) nearly so. In contrast, neurosensory defects (brain, eye),
cardiovascular defects (heart, great vessels) and defects of the
splanchnic body wall (abdominal and thoracic viscera) are over-
represented in rabbit (Table 2). Incomplete ossification and missing
skeletal elements are the most frequent observational terms in both
species (not shown).
Would a similar discordance follow endpoint effects limited to the
dLEL? We looked at the defects that occurred at the LEL for any
effect versus all effects no matter what dose (above). ToxRefDB iter-
ated dLEL effects in 609 and 288 effect-conditions for rat and rabbit
fetuses, respectively and covered 212 of 988 (21.5%) description-
target terms. LELs were recorded for 202 of 383 (52.7%) chemicals
tested in rat (3.2 effects per chemical on average) and 193 of 368
(52.4%) chemicals tested in rabbit (2.3 effects per chemical on aver-
age). Although total occurrence of effects or groups of effects was
predictably lower at the dLEL than when all dose groups are con-
sidered, the species dissymmetry was evident (Table 3). Similar
to the analysis noted above, skeletal defects, cleft palate, urogen-
ital defects, and somatic body wall defects predominated in the
rat, whereas neurosensory defects, defects of the cardiovascular
system and splanchnic (visceral body wall) defects sided toward
rabbit. Effects on maternal body weight gain and fetal weight reduc-
tion were species-neutral although maternal-pregnancy losses and
embryo-fetal losses were more evident in rabbits (Table 3). We may
conclude from these findings that patterns of effects seen at the LEL
are also manifested at higher doses.
3.4. Developmental activity
Profiling chemicals by developmental toxicity is an important
output from ToxRefDB; however, this analysis must consider that
developmental effects may not be the most sensitive endpoint in
the database. For example, any particular chemical may be highly
ranked based on dLEL but express an even lower NOAEL/LOAEL in
other types of studies. Capacity for developmental activity does
not necessarily indicate that the endpoint identified in a prena-
tal study is the most sensitive endpoint in the database of for that
matter the most appropriate for the exposure scenarios being eval-
uated. Furthermore, developmental activity does not necessarily
equate to risk since this analysis does not take into considera-
tion exposure to the human population, a key element in the
determination of risk. To evaluate the developmental activity of
chemical responses, we applied a rules-based approach adapted
from ToxRefDB chronic/cancer and multi-generation reproduc-
tion studies [7,8]. A chemical response was ranked by LEL dose
level for mLEL and dLEL effects using the value, &, represent-
ing -Iog2(mg/kg-d) computed for each chemical, median-centered
and scaled (e.g., (9 = 1.0 when LEL = 2048 mg/kg-d and 0 = 18.0
when LEL = 0.015625 mg/kg-d). This derived parameter (0) was
useful as a general metric for representing chemical activity
in a computable form, based on the administered dose at the
LEL. Fig. 2 correlates mLEL and dLEL for within- and between-
species; 283 chemicals had computable ® based on mLEL and
dLEL. This implies any sort of treatment-related effect whether
developmental or not. We generally considered correlations inside
2-fold as concordant within studies (maternal versus fetal) and
10-fold between studies (rat versus rabbit). The correlation
ranged slightly toward the maternal field in rabbits (Fig. 2).
For a subset of chemicals, however, the LEL effects were sensi-
tive or specific for developmental endpoints and are described
below.
Which chemicals had LEL effects that were developmentally spe-
cific? A response was considered 'specific' for developmental
toxicity if an effect or class of effects was recorded at the dLEL
(@>2.0) without maternal toxicity. Benomyl, for example, was
Previous
TOC
-------
214
T.B. Knudsen et al / Reproductive Toxicology 28 (2009) 209-219
Rabbil
/
/.
/
•
/
>
/
t
..
*
/
«
•J
*
•
!
*
/*
•
'.•
:.
/
*
/
2'
/
j/
/-
2
/ *
k»/
i/«
«. /
*
y
/
*4
^
,
/
'•
X
' »
/
/
•
*
/
t
*
/'
/
/
/
/
/
/
.
/
/
/
Rat
/
0.03125
0.0625
0.125 r
0.25 1
0.5 g
J 1
4
8 ft
<
16 2.
32 f
64 "S-
M *-
128 1-
256 3
512
1024
2048
Maternal
sc
a:
J-1 —
n
f
'
;velopmental
/
/
$\
* •
.*
in/
7
/'
4
/
•
/
/
:
•
/
y
y
/ ,
• y
*''t *
£ '
;*^7y
t! y
2
^
7
•
/
,
/
*
'
•
/
.
,
/*
/
»
/
/
*
/
/
/
/
/
•
/
/
-,
/
Maternal
—* 0.03125
0.0625
— ; 0.125
- 0.5
1
&
— Mf
i
€.
— '^ — 00
U. k) — S O O
Lowest Effect Level (mg/kg/day)
NE-Nol Established
—
i u *•> ui
K N
Kt — p p p p p
'vi IO ^ U* —
Lowest Effect Level (mg/kg/day)"
—
"
Fig. 2. Chemical potency for maternal and developmental toxicity. Maternal and developmental lowest effect levels (LELs) extracted from ToxRefDB were compared for rat
and rabbit for 386 distinct chemicals. Data were graphed between studies to correlate rat and rabbit studies for maternal LEL (top left panel) and developmental LEL (top right
panel), as well as within studies to compare maternal and developmental LELs for rat studies (lower left panel) and rabbit studies (lower right panel). Points within amber
lines indicate less than 2-fold difference and points within orange lines indicate less than 10-fold difference. NE, not established due to lack of observation of effects.
flagged for specific developmental toxicity in the rat because
dLEL = 62.5mg/kg-d (© = 6.03) without maternal effects. Overall
this rule flagged 16 unique chemicals of 283 chemicals tested
(5.7%) in both species. In 11 of the 16 instances, the chemical pro-
duced maternal toxicity in the other test species. Benomyl had
no maternal toxicity in the rat but mLEL = 180 mg/kg-d in the rab-
bit.
Which chemicals had LEL effects that were developmental sen-
sitive? A response was considered 'sensitive' for developmental
toxicity if the dLEL dose for any fetal endpoint was lower than the
corresponding mLEL for the chemical within a species. Prodiamine,
for example, was flagged for developmental sensitivity in the rat
because dLEL= 100mg/kg-d (© = 5.36) versus mLEL = 300 mg/kg-d
(© = 3.77). Overall this rule flagged 38 of 283 chemicals (13.4%)
in rat (30) or rabbit (11). Only 4 of them matched between
species (dLEL mLEL) across species. Prodiamine showed greater potency
toward maternal effects in the rabbit (mLEL = 100 mg/kg-d).
In total: 53 of 283 chemicals (18.7%) had critical effects on
development that were either specific (no maternal toxicity) or
sensitive (dLEL< mLEL) to exposure in one species or another; in
43 cases a ©-value was computed for maternal and developmental
effects in both species (see Table 4). The complete list of m/d-LEL
and ©-values for all ToxRefDB chemicals can be downloaded from
http: //www.e pa.gov/ncct/toxre fdb/.
3.5. Chemical-phenotype linkages
Linkage classification clustered the LEL effects from 283 chem-
icals utilizing maternal and developmental potency scores (©), as
well as categorical LELs (cLEL) registered for each class of develop-
mental effect. Hierarchical clustering is shown in Fig. 3. The primary
division segmented 90 chemicals based on maternal toxicity. This
grouping included chemicals with high © scores for reduced mater-
nal body weight gain, pregnancy-related losses and resorptions.
Distinct responses in rabbit and rat further grouped these 90 chem-
icals into subclasses of 40 and 50 chemicals, respectively. There
were 9 chemicals without maternal or fetal effects. All remaining
chemicals shared relatively high ©(dLEL) scores: 80 chemicals pos-
itive for developmental toxicity in the rat, 50 chemicals positive
in the rabbit, and 54 chemicals with developmental toxicity across
species.
How well do the different endpoint effects align with develop-
mental potency? Most chemicals with higher ©(dLEL) scores had
high ©(mLEL) scores as well. The overlap between chemicals in
these clusters and the 43 chemicals flagged for developmental
activity is given in Table 4. Whereas reduction in maternal body
weight correlated with ©(mLEL), fetal weight reduction correlated
with ©(dLEL). We did not observe any correlation between weight
changes in the pregnant mother and developing fetus at term for
either species. It is clear from Fig. 3 that a second level of clustering is
Previous
TOC
-------
T.B. Knudsen et al. / Reproductive Toxicology 28 (2009) 209-219
215
Table 4
Chemicals (rat and rabbit data) tagged for developmental sensitivity or specificity.
CAS no.
Chemical name
RAT (mg/kg-d)
mlEl
RABBIT (mg/kg-d)
dLEL
mLEL
dLEL
RAT(0)
MAT
RABBIT (0)
DEV
MAT
DEV
CLUSTER-SO developmental toxicity in rat (subcluster of 17 chemicals)
34256-82-1
17804-35-2
3691-35-8
120116-88-3
83657-24-3
131341-86-1
103361-09-7
361377-29-9
85509-19-9
93-65-2
76738-62-0
86209-51-0
29091-21-2
28434-00-6
148477-71-8
149979-41-9
55219-65-3
Acetochlor
Benomyl
Chlorophacinone
Cyazofamid
Diniconazole
Fludioxonil
Flumioxazin
Fluoxastrobin
Flusilazole
MCPP acid
Paclobutrazol
Primisulfuron-methyl
Prodiamine
S-Bioallethrin
Spirodiclofen
Tepraloxydim
Triadimenol
600
0.1
20
1000
30
1000
50
300
195
360
15
150
62.5
0.0125
1000
1
100
10
300
0.4
125
40
500
100
50
1000
120
5
180
0.025
100
3000
400
35
75
125
300
100
300
300
180
125
180
0.025
35
125
2.77
0.00
15.32
0.00
7.68
2.03
7.09
2.03
6.36
0.00
0.00
0.00
3.77
4.39
0.00
3.51
8.09
4.77
6.03
18.32
2.03
12.00
5.36
8.68
3.77
13.32
5.03
6.68
3.03
5.36
6.36
2.03
5.09
9.68
0.00
4.51
17.32
0.00
0.00
5.36
0.45
3.36
6.87
5.77
5.03
3.77
5.36
3.77
3.77
4.51
5.03
0.00
4.51
17.32
0.00
0.00
0.00
0.00
0.00
6.87
0.00
5.03
0.00
0.00
0.00
0.00
0.00
0.00
CWSTER-54 developmental toxicity in rats and rabbits (subcluster of 17 chemicals)
1912-24-9
57966-95-7
85-00-7
79241-46-6
79622-59-6
117337-19-6
79983-71-4
36734-19-7
16484-77-8
141112-29-0
2212-67-1
123312-89-0
118134-30-8
4151-50-2
111988-49-9
210631-68-8
87820-88-0
Atrazine
Cymoxanil
Diquat dibromide
Fluazifop-P-butyl
Fluazinam
Fluthiacet-methyl
Hexaconazole
Iprodione
Mecoprop-P
Isoxaflutole
Molinate
Pymetrozine
Spiroxamine
Sulfluramid
Thiacloprid
Topramezone
Tralkoxydim
25
75
4
300
250
250
50
500
140
100
100
3.3
50
100
200
5
25
40
5
50
2.5
200
100
100
35
30
10
13.3
10
100
3
75
3
50
7
100
60
50
100
200
75
80
3
10
100
75
8
1
50
4
1000
50
200
20
5
200
75
80
0.3
10
50
100
7.36
5.77
10.00
3.77
4.03
0.00
4.03
0.00
6.36
3.03
4.87
5.36
5.36
10.28
6.36
5.36
4.36
9.68
7.36
6.68
9.68
6.36
0.00
10.68
4.36
5.36
5.36
6.87
7.09
8.68
8.27
8.68
5.36
10.42
5.77
0.00
10.42
6.36
9.19
0.00
5.36
6.09
6.36
5.36
4.36
5.77
5.68
10.42
8.68
0.00
5.36
5.77
9.00
12.00
6.36
10.00
2.03
6.36
4.36
7.68
9.68
4.36
5.77
5.68
13.74
8.68
6.36
5.36
CLUSTER-SO developmental toxicity in rabbits (subcluster of 9 chemicals)
61-82-5
99607-70-2
33629-47-9
82697-71-0
94361-06-5
142-59-6
82-68-8
43121-43-3
199119-58-9
3-Aminotriazole
AA 5-C-8-Q"
Butralin
Clofencet
Cyproconazole
EBC'
Quintozene
Triadimefon
Trifloxysulfuron-sodium
400
1250
12
75
25
1000
1000
400
500
1000
12
7.5
750
50
1000
80
300
135
500
50
32.8
125
120
250
80
60
45
500
10
2.62
250
50
100
0.00
3.36
1.71
0.00
8.42
5.77
0.00
7.36
2.03
2.03
3.36
3.03
2.03
8.42
9.09
2.45
6.36
2.03
5.68
3.77
4.92
3.03
6.36
6.96
5.03
5.09
4.03
5.68
6.09
6.51
3.03
8.68
10.61
4.03
6.36
5.36
Abbreviations: EBC: ethylenebisdithiocarbamate, disodium: AA 5-C-8-Q: acetic acid, {(5-chloro-8-quinolinyl)oxy}-,l-methylhexyl ester: potency score 0 = -log2(LEL):
clusters correspond to chemical numbers in Fig. 3. Blank cells mean no effects reported.
patterned by 6>(cLEL) scores for fetal weight reduction and skeletal
defects. The @(cLEL) for embryonic-fetal loss was linked between
species although based on critical effect counts this endpoint
was over-represented in rabbits. The strongest correlated variables
among embryonic targets overall were rat appendicular-cranial
skeleton (correlation coefficient = 0.811) and rat kidney-ureter (cor-
relation = 0.843). Fig. 4 plots the chemical counts for each endpoint
variable at the dLEL and across all dose groups, dLEL or higher. As can
be seen the dLEL is sufficient to pick up categorical effects for most
but not all chemicals within each test species. A perhaps interesting
association is the over-representation of somatic body wall defects
in rats and of splanchnic (visceral) body wall defects in rabbit.
4. Discussion
Mining 30 years worth of guideline animal studies using
ToxRefDB classified chemical-phenotype relationships for hun-
dreds of chemicals and endpoints related to pregnancy outcome.
The process of gathering, curating, and integrating these data con-
stitutes a considerable effort that now for the first time provides a
common data model for mining the prenatal developmental tox-
icity of environmental chemicals. This repurposes study reviews
from their original use in regulatory toxicology decisions to a novel
use to anchor high-throughput screening assays in ToxCast™ [4].
ToxRefDB derives maternal and fetal data comprehensively from
reviews of prenatal studies on pregnant animals. The present imple-
mentation captured data on 387 environmental chemicals, mostly
pesticides, from 751 studies in pregnant rats or rabbits. This imple-
mentation adds to the considerable body of reference toxicity data
for these chemicals for chronic/cancer rodent endpoints [7] and
multi-generation reproduction rat endpoints [8].
Focusing on inter-species comparison, the complexity of fetal
target organ response to maternal dosing with environmental
chemicals during the period of major organogenesis revealed hier-
archical relationships. There was a clear hierarchy to the sensitivity
and specificity of maternal and fetal LELs in comparing responses
between chemicals and inter-species, with rats being more sensi-
tive to developmental effects than rabbits. The dependence of any
Previous
TOC
-------
216
T.B. Knudsen et al / Reproductive Toxicology 28 (2009) 209-219
I 40 I J~50~|
Egn..rC7l jkrhikjr^ xSTftL
Fig. 3. Hierarchical relationship of ToxRefDB chemicals to developmental effects. Chemicals (columns) by effects (rows) for 283 chemicals with potency scores (0) in rat (A,
dark green) and rabbit (C, light green). Clustering by Pearson's and Ward linkage; color intensity represents potency score. Variables are maternal LEL (mLEL), developmental
LEL (dLEL), and categorical LELs. Other effects classes are grouped by system: cardiovascular (CV), general (GN), neurosensory (NS), orofacial (OF), pregnancy-related (PR),
skeletal (SK), trunk (TR) and urogenital (UG). Abbreviations for specific effects targets: reduced maternal body weight gain (MBW), fetal weight reduction (FWR), maternal-
pregnancy losses (PRL), embryo-fetal losses (RES), general fetal pathology (GRL); skeletal defects - axial (AXL), appendicular (APP), cranial (CRN); orofacial defects - cleft
lip/palate (CLP), altered jaw / hyoid bone (]W"H); defects of the fetal brain (BRN) and eye (EYE); renal (REN), ureteric (URT), genital (GEN); body wall defects (SOM) and
abnormal splanchnic viscera (SPL); heart (HRT) and major vessels (VAS).
LEL on the study design is clearly a matter of choice by the orig-
inal investigators. These dose selections may not have been ideal,
but merely reflect practice over the era in which the studies were
performed. The primary factors underlying developmental toxicity
were fetal weight reduction, skeletal variations and abnormalities,
and fetal urogenital defects in rats, and general pregnancy/fetal
losses and structural malformations to the visceral body wall and
CNS in rabbits. Many aspects were consistent with other database
studies indicating the key relationships are likely driven by the
biology of test species.
4.1. Chemical activity
The spectrum of chemical activity based on the administered
dose was primarily resolved by the relative values of the mLEL
and fetal dLEL dosages. Reduced maternal body weight gain dur-
ing gestation and fetal weight reduction were the most common
endpoints. Weight changes would have been expected to underlie
determination of the mLEL and dLEL for a large number of chem-
ical structures. This was evident for fetal weight reduction and
the dLEL although neither weight change correlated with mLEL. A
recent study [16] examined general relationships between maternal
and fetal toxicity using a dataset of 56 rat, 46 mouse and 25 rab-
bit studies compiled from the National Toxicology Program (NTP).
That study found weight changes to be the primary factors in deter-
mining levels of maternal and fetal toxicity and noted that the
degree of association between maternal and fetal weight changes
followed the rank order: mouse (91% concordance, P<0.001)>rat
(41%, P< 0.01 )> rabbit (24%, not significant). They attributed the
inter-species difference to time lapse between dosing and evalu-
ation (mouse < rat < rabbit) and amount of time that the fetus has
to recover [16]. Although guideline study designs used to build
ToxRefDB.prenatal had the same inter-species time lapse, there was
no correlation between doses that caused maternal and fetal weight
changes (correlation coefficient < 0.01) in spite of a modest inter-
species correlation for mLELs (correlation coefficient = 0.59). Data
localizing maternal body weight changes to specific gestational
stages may improve the correlation [16]; however, this informa-
tion cannot be obtained from the current build of ToxRefDB which
has term body weight information only.
The subset of 387 ToxRefDB chemicals perturbing fetal weight
was much higher in rats (35.7%) than rabbits (19.2%). Consistent
with that finding, a high incidence of fetal weight reduction was the
lone endpoint effect in defining fetal LOAELs for >71% NTP rodent
studies [16]. Although the fraction of ToxRefDB chemicals that pro-
duced fetal weight change at the dLEL (71.2%) was also consistent
with NTP rat studies, the high incidence of fetal weight changes
also holds for the rabbit (86.9%). This contrasts with the NTP study
where fewer than half of the rabbit studies where a fetal LOAEL was
determined involved fetal weight reduction [16]. Due to the preva-
lence of pesticides for the initial build of ToxRefDB, it may be that
these are more bioactive chemicals than the many industrial types
of chemicals that NTP tested. Another disparate finding is the minor
subset of ToxRefDB chemicals that produced fetal weight change as
Previous
TOC
-------
T.B. Knudsen et al. / Reproductive Toxicology 28 (2009) 209-219
217
244
4
SPL
VAS
HRT
EYE
1 9fifi URUU
5 RES
PRL
UG
TR
cv
NS
OF
SK
GN
PR
RABBIT 0 RAT
317 chemicals 350 chemicals
Fig. 4. Percentage of chemical counts distributed across phenotype classes. Each phenotype is plotted by a normalized bar that shows the %-distribution of the response
in each species. Hatched area (dLEL) relative to solid area (cLEL) indicates the number of chemical counts at the dLEL relative to any dose, dLEL or higher. Rat studies are
plotted from left to right and rabbit studies from right to left on the same scale. Total chemical hits are listed aside each species bar. Effects (19) and classes of effects (8) are
abbreviated as in Fig. 3.
the sole determinant of the dLEL (~6% versus ~70%). Collectively,
these data support the notion that fetal weight reduction is a key
parameter in profiling developmental toxicity [16]. At the same time
the data suggest that for a substantial number of environmental
chemicals the fetal weight change correlates more strongly with
malformations than maternal weight change. Indeed, reductions in
fetal weight in absence of maternal weight change have been found
in several NTP chemicals that are 'preferentially toxic to both the
developing embryo and the fetus' [16].
On the basis of mLEL and dLEL a subset of ToxRefDB chemicals
showed increased sensitivity or specificity of the developing fetus
as compared to the pregnant mother (dLEL< mLEL) or in one species
versus another. This list included 53 of 387 chemicals (13.7%) with
data in rat or rabbit, and 43 of 283 chemicals (15.2%) where data
existed in both species. Although these chemicals provide inter-
esting prototypes for mechanistic study, a number of them were
active only at very high doses as defined on an mg/kg-d basis. The
proportions are slightly higher than observed in the NTP data anal-
ysis, which reported adverse effects in the fetus at lower doses of
exposure as compared to the mother in 5/62 (8.1%) cases across
rat-rabbit studies; however, that study also noted adverse fetal
effects and overt maternal toxicity at the same dose level in 19/62
(30.6%) rat-rabbit studies [16]. Further analysis is needed to learn
if there were any developmental effects that were always occurring
only in the presence of a specific form and/or severity of maternal
toxicity.
Across the 43 chemicals flagged for developmental sensitivity or
specificity, only a subset of cases had fetal weight effects. Apart from
the consistencies with NTP studies [16], the role of maternal under-
nutrition as a primary determinant of fetal weight at term could
not be substantiated through ToxRefDB profiling of primarily pes-
ticidal chemicals for developmental toxicity. The broader analysis
with 387 chemicals on 751 studies suggests a chemical-phenotype
association with specific variations or malformations that is direct
(mechanism) versus indirect (maternal) factors, especially in preg-
nant rats.
4.2. Phenotype representation
In current practice, the guideline prenatal developmental tox-
icity studies are used to identify NOAELs and LOAELs based on
maternal and fetal endpoints, rather than to estimate specific
developmental phenotypes in humans [25]. Analysis of chemical
counts for most developmental endpoints inToxRefDB captured the
majority of actives at the dLEL and many of the toxicants altered
development at doses near the mLEL. The inter-relationships of
developmental toxicity endpoints may, however, provide useful
information that can be mined from guideline studies [16,26,27].
A comprehensive weight-of-evidence model for reproductive and
developmental toxicity hazard identification has been constructed
by the U.S. Food and Drug Administration (FDA) to predict toxic-
ities based on quantitative structure-activity relationship (QSAR)
across large blocks of chemicals and chemical classes [15]. That
database derives secondary data for many chemicals (2000) and
studies (10,000). In contrast, ToxRefDB structures data from origi-
nal guideline studies. This enables profiling developmental toxicity
from high-quality source data annotated by internationally harmo-
nized target-description effects [18].
Most if not all of the 988 possible DevTox endpoints might
be expected from a survey of 751 source studies observed; how-
ever, only 29.7% of these terms were represented in the initial
build of ToxRefDB. This proved to be sufficient to classify targets
into individual defect 'categories' and then analyze their distribu-
tion by chemical count by species. Skeletal defects, in correlation
with fetal weight reduction in rats were the strongest factors pro-
filing developmental toxicity. Incomplete ossification and missing
Previous
TOC
-------
218
T.B. Knudsen et al. / Reproductive Toxicology 28 (2009) 209-219
bone elements are easily recognized in fetal evaluation protocols.
It is therefore not surprising to see these common findings over-
represented in the ToxRefDB source data and to have the response
amplified by aggregating individual skeletal elements by system. On
the other hand, some relatively common defects such as hypospa-
dias and ocular colobomata may have been under-represented in
ToxRefDB. These phenotypes may be difficult to detect in a fetal
rat, or their under-representation may very well indicate that these
endpoints are not associated with the pesticides evaluated here;
hence, the need for a multi-generational study. Finally, the extent
to which low-frequency malformations induced by exposure are
captured as being present in ToxRefDB has not been examined here.
Mapping the aggregation of less frequent defects to the tar-
get organ can improve the statistical power of representation
analysis [12,13,15]. For ToxRefDB, the aggregation of fundamental
DevTox observations revealed chemical effects on 19 generalized
target classes and 8 higher level embryonic systems. Although
the percentage of active chemicals was low for individual cate-
gories of defects and observations (skeletal changes excluded), a
large number of environmental chemicals had significant findings
among the 19 categories. Similarly, results from the QSAR repro-
tox database [15] showed the majority of agent-induced structural
defects could be aggregated into a relatively small number (11) of
defects categories. For both databases the major class of malfor-
mations included cleft palate, CNS, craniofacial, eye, skeletal and
urogenital defects. Therefore, specific defects in addition to fetal
weight changes contribute to the phenotype spectrum for profiling
the developmental toxicity of agents in general and environmen-
tal chemicals in particular. Because the phenotype spectrum is
well-represented at the dLEL, and because the percentage of active
chemicals becomes only slightly broader at high-dose levels, we
speculate that chemical profiling for developmental toxicity has its
strongest predictive value of specific fetal target systems at the dLEL
It is axiomatic that a dLEL assignment in ToxRefDB was dependent
on the sensitivity of anatomical methods used to identify a fetal
change. This implies an underlying biology leading to apical end-
points that can serve as an in vivo anchor to the bioactivity profiles
generated in HTS in vitro and QSAR models [28].
4.3. Inter-species concordance
Although many environmental chemicals had significant find-
ings among the 19 categories of effects, the percentage of actives
for individual categories varied between species. The stronger
response of rats in terms of skeletal defects, fetal weight reduc-
tion and urogenital defects perhaps reflects an underlying biological
susceptibility to the chemical, or may be simply explained by
nuances in the examination techniques. Of particular importance
is the longer period between exposure and evaluation that can
allow rabbit fetuses more time to recover from transitory delays
than might be detected in the rat fetus [14]. Incomplete ossification
or fetal weight reduction, for example, may be less evident in the
rabbit due to a longer recovery interval.
On the other hand, study design factors are less compelling
explanations for some of the ToxRefDB findings. Rat renal defects
and resorptions in rabbits, for example, may be more directly
principled on species differences. FDA's weight-of-evidence QSAR
database concluded from among 936 chemicals that the rabbit
was about 6-fold less susceptible than rat to chemicals causing
fetal dysmorphogenesis [13]. Results from ToxRefDB indicate that
inter-species differences depend on the target organ since some
endpoints were over-represented in percentages for rabbits. Since
rabbit is noted as having more developmental variations than rat
it follows that more uncertainty can be anticipated in assessing
treatment-related malformations in this species; however, the His-
torical Control Database [29] did not reveal an inter-species bias for
either renal defects (rat) or eye defects (rabbit). This further implies
an underlying biology for ToxRefDB endpoints.
The importance of placental differences between rat and rab-
bit embryos as a potential reason for inter-species differences has
been emphasized [30]. Development of the chorioallantoic placenta
is precocious in rabbit versus rat; consequently, visceral yolk sac
expansion occurs relatively late in rabbits and the volume of exo-
coelomic fluid is much higher than in rat gestation. These factors
may influence the transport and concentration of chemical reach-
ing the embryo at critical times during organogenesis, which in turn
may account for some of the inter-species differences in suscep-
tibility. Because unique attributes of placentation in rabbits more
than rats closely resemble the human condition [30], testing in both
species has implications in estimating human risk [6].
Not all inter-species findings were consistent between ToxRefDB
and the weight-of-evidence QSAR database. A higher percentage of
ToxRefDB chemicals with significant activity on fetal death param-
eters in the rabbit, and the higher percentage of chemicals with
greater fetal weight reduction in the rat, were not noted in the QSAR
training set [15]. Again, ToxRefDB chemicals are likely to be more
bioactive in general because they are compiled of many pesticides.
Analysis of data for 54 potential developmental toxicants and 73
substances considered to be teratogenic in the rabbit and not the
rat showed generally similar sensitivity between species, although
for some chemicals the rat is more sensitive and others the rabbit
study is more sensitive [25]. Those authors suggested that differ-
ences between rat and rabbit studies in terms of classification of
developmental toxicity may reflect consequences of maternal toxi-
city between the species, rather than direct developmental toxicity
[31].
Aside from a relatively longer gestational period and the higher
frequency of developmental variation in rabbits, the doe is less tol-
erant of chemical treatment than the rat dam [15]. Clearly, some
ToxRefDB chemicals showing developmental activity in rats pro-
duced maternal toxicity in rabbits at the same (dLEL) dose level.
Among 91 substances with teratogenicity information reviewed [6]
a lack of concordance between rat and rabbit was observed in 18
of 91 (20%) compounds tested in both species. Chemical profiling
of 283 ToxRefDB chemicals with an evaluation of developmental
toxicity in both species identified clusters of about 130 chemicals
with developmental effects in either species alone; however, chem-
icals may have multiple effects on maternal and fetal parameters
and the interaction between mother-conceptus may differ across
species and chemicals. Selection of rabbit as a test species is primar-
ily driven by historical interest in thalidomide-induced limb defects
observed in humans, monkeys and rabbits, but not rats [32].
The present study shows that specific developmental effects
differ between species, and we know this to be true for the
comparison with the human condition as experience with some
chemicals shows [32]. The added value of rabbit studies for prena-
tal developmental toxicity evaluation has been recently questioned
based on NOAEL comparison and developmental outcomes [25]
and the weight-of-evidence QSAR database finding "no evidence
of trans-species tissue specific dysmorphogenic findings" [15]. Ret-
rospective analysis of several hundred Pharmaceuticals tested in
both rodent and non-rodent species for general toxicological end-
points showed an overall 71% concordance with true positives in
human populations; concordance was lower when non-rodents
(63%) and rodents (43%) were considered separately [33]. For devel-
opmental toxicity, rat studies alone predicted teratogenicity in
61% of chemicals that showed teratogenicity in rat, mouse or rab-
bit, whereas a rat study and a rabbit study together identified
teratogenicity in 100% of these chemicals [6]. Taking the devel-
opmental toxicity alone, without regarding maternal toxicity as
strictly causal and without extrapolating the nature of effects equiv-
alently between species, the question remains open whether the rat
Previous
TOC
-------
T.B. Knudsen et al. / Reproductive Toxicology 28 (2009) 209-219
219
as the only in vivo model would not detect almost all developmental
toxicants.
4.4. Conclusion
Results from analysis of 387 chemicals in the EPA ToxRefDB
support the value of a traditional two species paradigm for
identification of developmental toxicity. Manifestations of direct
(mechanism-based) developmental toxicity with or without indi-
rect (maternal-mediated) effects underscore the need for improved
methods of assessing the dynamical relationship between devel-
opmental processes and maternal health status. In the future,
data from alternative methods and HTS in vitro assays that enable
'pathway-based risk assessment' may increase confidence in test-
ing strategies while limiting required animal testing [1,34]. For this
to occur, public data models are needed that structure conventional
in vivo toxicity data into computable forms. ToxRefDB provides
such a novel data model for relational assessment of source data
from guideline (in vivo) prenatal developmental toxicity studies.
We envisage high value in these animal studies to anchor cross-
scale modeling and predictive understanding of developmental
processes and toxicities [17].
Conflict of interest
The authors declare they have no competing financial interest.
Acknowledgements
We thank the United States Environmental Protection Agency's
Office of Pesticide Programs (OPP) for contributions to the ToxRefDB
project, including access to toxicity data evaluation records, scien-
tific consultation, and the review of this manuscript (Dr. Elizabeth
Mendez and Dr. Vicki Dellarco). We also thank Daniel Corum,
Daniel Rotroff and Jeffrey Finn for excellent work entering data into
ToxRefDB.
References
[1] National Research Council. Toxicity testing in the 21st century: a vision and a
strategy. Washington, DC: The National Academies Press: 2007,196 pp.
[2] Collins FS, Gray CM, Bucher JR. Transforming environmental health protection.
Science 2008:319:906-7.
[3] Judson R, Richard A, Dix DJ, Houck K, Martin M, Kavlock RJ, et al. The Tox-
icity data landscape for environmental chemicals. Environ Health Perspect
2009:117:685-95.
[4] Dix DJ, Houck KA, Martin MT, Richard AM, Setzer RW, Kavlock RJ. The ToxCast
program for prioritizing toxicity testing of environmental chemicals. ToxicolSci
2007:95:5-12.
[5] Kimmel CA, Kimmel GL. Principles of developmental toxicology risk assess-
ment. In: Hood RD, editor. Handbook of developmental toxicology. Boca Raton:
CRC Press: 1996. p. 667-93.
[6] Hurtt ME, Cappon CD, Browning A. Proposal for a tiered approach to
developmental toxicity testing for veterinary pharmaceutical products for food-
producing animals. Food ChemToxicol 2003:41:611-9.
[7[ Martin MT, Judson RS, Reif DM, Kavlock RJ, Dix DJ. Profiling chemicals based
on chronic toxicity results from the U.S. EPA ToxRef Database. Environ Health
Perspect 2009:117:1-8.
[8[ Martin MT, Mendez E, Corum DC, Judson RS, Kavlock RJ, Rotroff DM, Dix
DJ. Profiling the Reproductive Toxicity of Chemicals from Multigeneration
Studies in the Toxicity Reference Database (ToxRefDB). Toxicol Sci 2009:
doi:10.1093/toxsci/kfp080.[online 10 April 2009[.
[9] U.S. Environmental Protection Agency. Health effects test guidelines
OPPTS 870.3700 prenatal developmental toxicity study. Washington, DC:
Office of Prevention, Pesticides and Toxic Substances. EPA Publication
712-C-98-207, 1998. Available at http://www.epa.gov/opptsfrs/publications/
OPPTS _Harmonized/870_Health_Effects_Test_Guidelines/Series/870-3700.pdf
[accessed 10 January 2009].
[10] U.S. Environmental Protection Agency. Guideline 83-3: teratogenicity study,
pesticide assessment guidelines, subdivision F, hazard evaluation: human and
domestic animals. Washington, DC: Office of Pesticides and Toxic Substances:
1982. EPA Publication 540.9-82-025.
[11[ Organisation for Economic Cooperation and Development. OECD guideline for
the testing of chemicals, No. 414: prenatal developmental toxicity study. Paris,
France: Organisation for Economic Cooperation and Development: 2001,11 pp.
[12] Scheuerle A, Tilson H. Birth defect classification by organ system: a novel
approach to heighten teratogenic signaling in a pregnancy registry. Pharma-
coepidemiol Drug Safety 2002:11:465-75.
|13] Correa-Villasenor A, Cragan J, Kucik J, O'Leary L, Siffel C, Williams L. The
metropolitan Atlanta congenital defects program: 35 years of birth defects
surveillance at the centers for disease control and prevention. Birth Def Res
(Part A) 2003:67:617-24.
[14] Julien E, Willhite CC, Richard AM, DeSesso JM. Challenges in constructing sta-
tistically based structure-activity models for developmental toxicity. Birth Def
Res (Part A) 2004:70:902-11.
[15] Matthews EJ, Kruhlak NL, Benz RD, Contrera JF. A comprehensive model
for reproductive and developmental toxicity hazard identification: I. Devel-
opment of a weight of evidence QSAR database. Regul Toxiol Pharmacol
2007:47:115-35.
[16] Chernoff NA, Rogers EA, Gage MI, Francis BM. The relationship of maternal and
fetal toxicity in developmental toxicology bioassays with notes on the bio-
logical significance of the "no observed adverse effect level". Reprod Toxicol
2008:25:192-202.
[17] Knudsen TB, Kavlock RJ. Comparative bioinformatics and computational
toxicology. In: Abbott B, Hansen D, editors. Developmental toxicology vol-
ume 3, target organ toxicology series. New York: Taylor and Francis: 2008.
p. 311-60.
118] Wise LD, Beck SL, Beltrame D, Beyer BK, Chahoud I, Clark RL, et al. Terminology
of developmental abnormalities in common laboratory mammals (version 1).
Teratology 1997:55:249-92.
[19] Chahoud I, Buschmann J, Clark R, Druga A, Falke H, Faqi A, et al. Classification
terms in developmental toxicology: need for harmonisation. Report of the sec-
ond workshop on the terminology in developmental toxicology. Reprod Toxicol
1999:13:77-82.
[20] Solecki R, Burgin H, Buschmann J, Clark R, Duverger M, Fialkowski O, et al.
Harmonisation of rat fetal skeletal terminology and classification. Report of
the third workshop on the terminology in developmental toxicology. Reprod
Toxicol 2001:15:713-21.
[21] Solecki R, Bergmann B, Burgin H, Buschmann J, Clark R, Druga A, et al. Har-
monization of rat fetal external and visceral terminology and classification.
Report of the fourth workshop on the terminology in developmental toxicology.
Reprod Toxicol 2003:17:625-37.
[22] Yang C, Benz RD, Cheeseman MA. Landscape of current toxicity databases and
database standards. Curr Opin Drug Discov Dev 2006:9:124-33.
[23] Ward JH. Hierarchical grouping to optimize an objective function. J Am Stat
Assoc 1963:58:236-44.
[24] Ihaka R, Gentleman RR. A language for data analysis and graphics. J Comput
Graph Stats 1996:5:299-314.
[25] Janer G, Slob W, Hakkert BC, Vermeire T, Piersma AH. A retrospective anal-
ysis of developmental toxicity studies in rat and rabbit: what is the added
value of the rabbit as an additional test species? Regulat Toxicol Pharmacol
2008:50:206-17.
[26] Guittina P, Elefant E, Saint-Salvic B. Hierarchization of animal teratology
findings for improving the human risk evaluation of drugs. Reprod Toxicol
2000:14:369-75.
|27] Piersma AH, Janer G, Wolterink G, Bessems JG, Hakkert BC, Slob W. Quanti-
tative extrapolation of in vitro whole embryo culture embryotoxicity data to
developmental toxicity in vivo using the Benchmark Dose approach. Toxicol Sci
2008:101:91-100.
[28] Judson RS, Richard AM, Dix DJ, Houck K, Elloumi F, Martin MT, et al.
ACToR—aggregated computational toxicology resource. Toxicol Appl Pharm
2008:233:7-13.
[29] Historical Control Database of Preclinical Developmental and Reproductive
Toxicology Parameters. A Joint Project of Middle Atlantic Reproduction and
Teratology Association (MARTA) & Midwest Teratology Association (MTA)
(http://www.hcd.org/) [accessed January 14, 2009].
|30] Carney EW, Tornesi B, Keller C, Findlay HA, Nowland WS, Marshall VA, et al.
Refinement of a morphological scoring system for postimplantation rabbit con-
ceptuses. Birth Def Res (Part B) 2007:80:213-22.
[31] Khera KS. Maternal toxicity: a possible etiological factor in embryo-fetal
deaths and fetal malformations of rodent-rabbit species. Teratology 1985:31:
129-53.
[32] Brent R, Holmes L. Clinical and basic science lessons from the thalidomide
tragedy: what have we learned about the cause of limb defects? Teratology
1998:38:241-51.
[33] Olson H, Betton G, Robinson D, Thomas K, Monro A, KolajaG, et al. Concordance
of the toxicity of Pharmaceuticals in humans and in animals. Regulat Toxicol
Pharmacol 2000:32:56-67.
[34] Bremer S, Pellizzer C, Hoffmann S, Seidle T, Hartung T. The development of new
concepts for assessing reproductive toxicity applicable to large scale toxicolog-
ical programmes. Curr Pharm Des 2007:13:3047-58.
Previous
TOC
-------
TOXICOLOGICAL SCIENCES 110(1), 181-190 (2009)
doi:10.1093/toxsci/kfp080
Advance Access publication April 10, 2009
Profiling the Reproductive Toxicity of Chemicals from Multigeneration
Studies in the Toxicity Reference Database
Matthew T. Martin,*'1 Elizabeth Mendez,t Daniel G. Corum,* Richard S. Judson,* Robert!. Kavlock,* Daniel M. Rotroff,* and
David J. Dix*
*National Center for Computational Toxicology, Office of Research and Development, U.S. Environmental Protection Agency, Research Triangle Park, North
Carolina 27711; and ^Health Effects Division, Office of Pesticide Programs, U.S. Environmental Protection Agency, Washington, District of Columbia 20460
Received January 26, 2009; accepted April 8, 2009
Multigeneration reproduction studies are used to characterize
parental and offspring systemic toxicity, as well as reproductive
toxicity of pesticides, industrial chemicals and Pharmaceuticals.
Results from 329 multigeneration studies on 316 chemicals have
been digitized into standardized and structured toxicity data
within the Toxicity Reference Database (ToxRefDB). An initial
assessment of data quality and consistency was performed prior to
profiling these environmental chemicals based on reproductive
toxicity and associated toxicity endpoints. The pattern of toxicity
across 75 effects for all 316 chemicals provided sets of chemicals
with similar in vivo toxicity for future predictive modeling.
Comparative analysis across the 329 studies identified chemicals
with sensitive reproductive effects, based on comparisons to
chronic and subchronic toxicity studies, as did the cross-
generational comparisons within the multigeneration study. The
general pattern of toxicity across all chemicals and the more
focused comparative analyses identified 19 parental, offspring and
reproductive effects with a high enough incidence to serve as
targets for predictive modeling that will eventually serve as
a chemical prioritization tool spanning reproductive toxicities.
These toxicity endpoints included specific reproductive perfor-
mance indices, male and female reproductive organ pathologies,
offspring viability, growth and maturation, and parental systemic
toxicities. Capturing this reproductive toxicity data in ToxRefDB
supports ongoing retrospective analyses, test guideline revisions,
and computational toxicology research.
Key Words: pesticides; relational database; reproductive
toxicity; toxicity profiling.
The U.S. Environmental Protection Agency (EPA) and other
regulatory agencies are investigating novel approaches for
Disclaimer: The United States Environmental Protection Agency through its
Office of Research and Development funded and managed the research
described here. It has been subjected to Agency administrative review and
approved for submission and peer review.
1 To whom correspondence should be addressed at National Center for
Computational Toxicology, Office of Research and Development, U.S.
Environmental Protection Agency, MD D343-03, 109 TW Alexander Drive,
Research Triangle Park, NC 27711. Fax: (919) 685-3399. E-mail:
martin.matt@epa.gov.
Published by Oxford University Press 2009.
predicting chemical toxicity, with the goal of rapidly screening
the thousands of environmental chemicals with limited toxicity
data (U.S. EPA, 2009). Building predictive models of chemical
toxicity requires high-quality in vivo toxicity data, in order to
develop and validate new in vitro and in silico approaches, hi
support of EPA's ToxCast predictive toxicology effort (Dix
et al., 2007), we have developed the Toxicity Reference
Database (ToxRefDB) for capturing information from in vivo
toxicity studies. ToxRefDB includes endpoints from multiple
study types, including chronic rat and mouse carcinogenicity
2-year bioassays that have been previously reported and made
publicly available (http://www.epa.gov/ncct/toxrefdb/) (Martin
et al., 2009). ToxRefDB is being used to build computational
models linking whole animal toxicity, and specific tissue and
cellular phenotypes, to specific chemical-biological interactions
detected by cellular, genomic and biochemical in vitro assays.
The in vivo toxicity data captured in ToxRefDB is facilitating
a transition to the National Research Council's vision for
"Toxicity Testing in the 21st century: A Vision and a Strategy"
(Collins et al., 2008; NRC, 2007), by linking toxicity endpoints
from animal studies to molecular targets and pathways relevant
to humans.
The multigeneration study data entered into ToxRefDB
provides anchoring in vivo reproductive toxicity data for the
EPA ToxCast research program (http://www.epa.gov/ncct/
toxcast/). Within the ToxCast program, bioactivity profiles
for hundreds of environmental chemicals are being derived
from hundreds of in vitro assays (Dix et al., 2007; Houck and
Kavlock, 2008). Phase I of ToxCast is focused on chemicals
with known in vivo toxicity data, supporting the development
of in vitro data signatures predictive of these in vivo outcomes
(Kavlock et al., 2008). It is worth noting that for environmental
chemicals, unlike pharmaceuticals, quantitative in vivo toxicity
data is essentially restricted to animal species. Nearly all of the
ToxCast Phase I chemicals are food-use pesticide active
ingredients that have undergone numerous mammalian toxicity
tests, including guideline multigeneration studies. This highly
standardized dataset provided in ToxRefDB facilitates profiling
ToxCast Phase I chemical toxicity based on parental, offspring
Previous
TOC
Next
-------
182
MARTIN ET AL.
and reproductive effects. It was hypothesized that, through the
use of ToxRefDB, key effects from the 329 multigeneration
reproduction studies would characterize the reproductive
toxicity potential of the 316 chemicals, and further differentiate
the chemicals and effects with regards to generational and life-
stage sensitivities. Subsequently, the well-characterized re-
productive effects could be used to phenotypically anchor
predictive toxicity models.
Traditional toxicity testing for the risk assessment of
environmental compounds or groups of compounds can cost
millions of dollars and take years of effort. Since 1970, EPA
has accumulated a vast store of high-quality regulatory toxicity
information on hundreds of compounds, most of which has
been inaccessible to computational analyses. The curation and
structuring of this chemical toxicity information into Tox-
RefDB has created a valuable resource for both retrospective
and prospective toxicological studies (Martin et al., 2009). In
addition to the chronic/cancer rat and cancer mouse studies
(Martin et al., 2009) and multigeneration studies reported here,
we are also extracting developmental toxicity studies in the rat
and rabbit. The multigeneration reproductive toxicity data
set—studies used by EPA in the pesticide registration process
to assess the performance and integrity of the male and female
reproductive systems (U.S. EPA, 1996) include assessment of
gonadal function, the estrous cycle, mating behavior, concep-
tion, gestation, parturition, lactation, weaning, and on the growth
and development of the offspring. The multigeneration study
also provides information about the effects of the test substance
on neonatal morbidity, mortality, target organs in the offspring,
and data on prenatal and postnatal developmental toxicity.
Two historical test guidelines have been used for the
multigeneration studies in ToxRefDB. Multigeneration studies
according to the 1982 Reproductive and Fertility Effects
guideline (U.S. EPA, 1982) on over 700 chemicals have been
conducted and submitted to EPA. Multigeneration studies
according to the newer 1998 guideline (U.S. EPA, 1998) on
over 90 chemicals have been conducted and submitted,
including 40 studies extracted into ToxRefDB. Information
on data submissions to EPA was drawn from the Office of
Pesticide Programs (OPP) Information Network—the OPPIN
database (http://www.epa.gov/pesticides/). The 1998 guideline
was harmonized by EPA's Office of Pollution Prevention &
Toxic Substances (OPPTS; http://www.epa.gov/opptsfrs/home/
guidelin.htm) to meet testing requirements of the EPA's Office
of Pollution Prevention and Toxics and OPP, as well as
international guidelines published by the Organization for
Economic Cooperation and Development. Both of these
guidelines call for a two-generation study in which continu-
ously treated male and female rats are mated to produce first
generation offspring, and in turn the adult offspring are mated
to produce a second generation. Continuing refinement of these
test guidelines has been proposed (Cooper et al., 2006), and
ToxRefDB is being used to test hypotheses concerning the
validity of these refinements.
MATERIALS AND METHODS
Data characteristics. Reviews of registrant-submitted multigeneration
reproductive toxicity studies, known as Data Evaluation Records (DER), were
collected for roughly 300 chemicals from EPA's OPP. File types of DER
include TIFF, Microsoft Word, Word Perfect and PDF formats, some of which
are not directly text-readable. Approximately 500 multigeneration reproductive
toxicity DER were reviewed, and based on data quality a subset of 329 was
selected for curation into ToxRefDB. The first portion of the DER outlines the
test substance, purity, lot/batch numbers, MRID (Master Record Identification),
study citation, OPPTS test guideline (U.S. EPA, 1982, 1998) and reviewers of
the study. The executive summary captures all of the basic study design
information, including species and strain, doses, number of animals per
treatment group and any deficiencies in study protocol. All dose levels were
stored in ToxRefDB as "mg/kg/day" and, where possible, recorded or
calculated from food consumption data as an average over the entire dosing
period. The executive summary also describes treatment-related effects
observed at various dose levels in the study. The body of the DER provides
detailed test material and animal information, and full dose response data in text
and tables for all measured and observed endpoints. All treatment-related
effects were captured for each study in ToxRefDB.
Multigeneration study DER contain all the information necessary to infer
lowest effect level (LEL) values for all treatment-related effects that were
statistically or biologically significant. Typically, the DER also designated
"critical" effects for each study, and lowest observed adverse effect level
(LOAEL) and no observed adverse effect level (NOAEL) for each study. If
provided by the DER, ToxRefDB captured these study-level NOAEL, LOAEL,
and critical effect data. However, it is important to note that the critical effects
used to establish NOAEL, LOAEL, and a reference dose (RfD) for
a conventional chemical pesticide active, and to make regulatory risk
assessment and management decisions, are based on a toxicological review
of multiple studies across many study types.
Treatment-related effects were further identified as either a "Parental,"
"Offspring," or "Reproductive" effect. Consistent with DER, "Parental"
endpoints were defined as systemic toxicity observed in the male or female
adult parents, and exclude effects directly related to reproduction (e.g.,
reproductive organ toxicity). "Offspring" endpoints were defined as systemic
toxicity observed in the preweaning and juvenile animals, and exclude birthing
indices up to postnatal day (PND) 4 (e.g., litter size and live birth index).
"Reproductive" endpoints were defined as observed effects on the reproductive
performance or capacity of the animals and included all reproductive organ
toxicities, effects on estrous cyclicity, sperm parameters, fertility, and mating,
and prenatal and early postnatal viability.
A small number of ToxCast Phase I chemicals were not pesticide active
chemicals, such as some perfluorinated compounds and phthalates. Though
DER and pesticide registration studies were not available for these chemicals,
there was often high-quality, standardized reproductive toxicity studies
available from the National Toxicology Program, peer-reviewed literature, or
other sources. When data from such studies were available, it was crated into
ToxRefDB consistent with information taken from DER.
Data model and quality control. The relational data model for ToxRefDB
was previously described (Martin et al., 2009) in a diagram showing the data
model and field-level. A Data Entry Tool was developed for database
population, including a controlled vocabulary for reproductive and other test
data (available for download at http://www.epa.gov/ncct/toxrefdb/). Additional
data entry and quality control procedures for ToxRefDB are described in
Martin et al. (2009), and on the ToxRefDB homepage.
Full descriptions of the available data and conclusions as to the potential for
the pesticides to cause harm to humans or the environment, risk mitigation
measures, and the regulation of pesticides can be found at U.S. EPA's OPP
websites: http://www.epa.gov/pesticides/regulating/index.htm; http://www.epa.
gov/pesticides/reregistration/status.htm; http://www.epa.gov/oppsrrdl/registration_
review/; http://www.epa.gov/oppsrrdl/reregistration/index.htm. The study-level
critical effects captured in ToxRefDB and taken from individual DER and studies
Previous
TOG
Next
-------
REPRODUCTIVE TOXICITY PROFILING FROM TOXREFDB
183
cannot be related directly to regulatory determinations or Reds without additional
information and analysis.
Data output and analysis. The structured toxicity information stored
within ToxRefDB can be extracted in various formats utilizing SQL queries.
For the purpose of providing computable outputs, that is, quantitative outputs
amenable to statistical analysis, a consistent data output was used. The cross-
tabulated data output consisted of rows of chemical information (e.g., Chemical
Abstracts Service Registry Number, chemical name), by columns of toxicity
endpoints with the value entered being the lowest dose at which the endpoint
was observed (i.e., LEL) in "mg/kg/day." Even though NOAEL/LOAEL
values for each study's "Parental," "Offspring," or "Reproductive" effect can
be queried from the database, the current analyses for ToxCast only utilized
LEL. Log-transformed potency values were derived using -Iog2 of LEL. A
constant value of 12 was then added to zero-center the data allowing for zero to
represent no observed effect. Therefore, a value of 1 would be equivalent to an
effect at 2048 mg/kg/day and 18 would be equivalent to 0.015625 mg/kg/day.
The log-transformed values are predominantly used in the current analysis.
However, mill molar concentrations (mol/kg/day) were calculated for each
endpoint using the molecular weight of tested chemical and the LEL in mg/kg/
day. These data tables are available on the ToxRefDB homepage: http://
www.epa.gov/ncct/toxrefdb/. It should be noted, however, that potency does
not necessarily equate to risk because this analysis does not take into
consideration levels of exposure, a key element in the determination of risk.
Moreover, the potency of a compound in this analysis does not represent that
the endpoints identified in the multigeneration toxicity study are the most
sensitive in the database or for that matter the most appropriate for the exposure
scenarios being evaluated.
Hierarchical clustering across all the chemicals and effects was carried out
based on log-transformed potency values. Effects were selected based on an
occurrence in five or more chemicals, which was level shown to have minimal
predictive capability based on a simulation study performed by Judson et al.
(2008a). The clustering analysis, implemented in R version 2.6.1 (Ithaca and
Gentleman, 1996), used Pearson's dissimilarity as the distance measure for both
chemicals and effects and Ward's method for linkage (Ward, 1963). The
chemicals were divided into six groups based on the percent of explained
variance (Thorn dike, 1953). The weight for each effect in deriving the
chemical groupings was calculated as the ratio of the number of positives
within the chemical grouping over the number of positives out of the chemical
grouping.
RESULTS
Summary Characterization of Multigeneration Study Results
This analysis focused on reproduction-related endpoints
culled from 329 multigeneration rat studies on 316 unique
chemicals entered into ToxRefDB. The vast majority of studies
(294 of 329) were performed using a two-generation protocol.
There were seven one-generation studies, for which four were
supplementary studies to longer-term two- or three-generation
studies. Of the 28 three-generation studies, only first and
second generation effects were used in subsequent analyses,
whereas third generation effects were excluded, In total, there
were 11 chemicals with more than one study in this dataset.
Four chemicals had an additional study run to satisfy study
guideline requirements. Two chemicals had an additional study
to test at additional dose levels. Five chemicals had two studies
performed at similar dose levels and the conclusions between
each pair of studies were similar.
TABLE 1
Distribution of Chemicals and Effects Across Life-Stage,
Endpoint Category and Generation for 316 Chemicals in
ToxRefDB with a Multigeneration Reproductive Study
Life stage
Endpoint category
Generation PI
Fl
F2
Adult
Parental"
275° (2935)6
275 (3265)
Adult
Reproductive^
100 (376)
129 (648)
Juvenile
Offspring"
255 (2274)
247 (1979)
"Number of chemicals with at least one effect observed at specified life-
stage, endpoint category and generation.
^Number of effects observed at specified life-stage, endpoint category, and
generation.
"Parental endpoints include adult body weight, mortality, clinical signs, and
target-organ weight and pathology effects.
^Reproductive endpoints include reproductive organ weight and pathology
and reproductive indices (e.g., fertility, mating, live birth index).
"Offspring endpoints include pup weight, offspring survival (e.g., viability
and lactation index), and juvenile target-organ weight and pathology, and
pubertal delay (e.g., PPS and VO) effects.
Across all studies and treatment groups 12,230 treatment-
related effects were observed, corresponding to 458 different,
unique types of effects. Each effect was tagged with specific
endpoint category, life-stage, and generational information.
The distribution of treatment-related effects and positive
chemicals across life-stage and generation provide insight into
the sensitivities of specific classes of endpoints (Table 1).
Parental effects were associated with 275 of the 316 chemicals
for both the PI and Fl generation, whereas reproductive effects
were associated with only 100 or 129 chemicals in the PI and
Fl generations, respectively. Besides more chemicals, there
were 73% more adult reproductive effects in the Fl generation,
than in the PI. A similar number of chemicals and offspring
effects were observed in the Fl and F2 generation. The relative
generational sensitivity among reproductive effects compared
to offspring effects prompted us to investigate the patterns of
specific reproductive and offspring toxicities across all
chemicals.
Patterns of Reproductive Toxicity
Identification of chemical groups with similar reproductive
toxicity profiles was achieved by hierarchical clustering of 75
target-level effects (Fig. 1). These were defined as target-level
effects because specific descriptive terms were aggregated to
the target organ (i.e., liver) or measured index (e.g., lactation
index), rather than all possible outcomes for each target
(hypertrophy, hyperplasia, degeneration, etc.). Six groups of
chemicals were identified based methods described above in
the Methods section. Each chemical grouping was described by
the effects that most heavily weighted the formation of the
chemical groupings in Figure 1 and does not mean that every
Previous
TOC
Next
-------
184
MARTIN ET AL.
Chemical Groupings
2048 0.015625
mg'kgday mgitg/day
Lowest Effect Level (LEL)
-Log,(LEL)
Chemicals
FIG. 1. Two-way hierarchical clustering of 75 treatment-related effects from multigeneration reproduction tests on 316 chemicals inToxRefDB. Six chemical
groups were identified based on their patterns of reproductive toxicity. Each chemical group description is derived from the mostly heavily weighted endpoints (see
Results and http://www.epa.gov/ncctAoxrefdb/ for details) and does not mean that every chemical in the group causes the endpoint.
chemical in the group causes the endpoint. Group 1 consists of
the 14 chemicals with no observed toxicities across the 75
effects in this analysis. Group 2 contains 115 chemicals for
which general systemic toxicities are driving the formation of
the group. Interestingly, this chemical grouping is also heavily
weighted with endpoints relating to sperm counts and
morphology, endocrine-related organ pathologies and weight
changes, and delays in sexual maturation. Of the 115
chemicals, all five phthalate compounds are found in this
group. Group 3 contains 63 chemicals with limited toxicity for
which parental and offspring body weight changes are driving
the formation of the chemical group. Group 4 formation is
heavily weighted with cholinesterase inhibition effects and is
comprised of 12 organophosphorus compounds. Groups 5 and
6 contain 48 and 64 chemicals, respectively, and the formation
of these groupings were heavily weighted with reproductive
toxicity endpoints, including testicular and epididymal pathol-
ogies in group 5 and decrements in offspring viability and
survival in group 6.
The complete listing of chemical groupings and endpoint
weights are available for download from the ToxRefDB
homepage (http://www.epa.gov/ncct/toxrefdb/). This analysis
clearly segmented the chemicals into distinct classes based on
their profile of systemic and reproductive toxicities. This
analysis also guides endpoint selection process by highlighting
groups of related chemicals or endpoints based on the entire
profile of toxicological activity rather than a single outcome.
Many of the associations between endpoints or chemicals were
expected, but others were not. For instance, reproductive
performance, reproductive organ and offspring viability effects
were segregated slightly from each other and to a greater extent
from parental systemic effects and even delays in sexual
maturation.
Comparative Analysis with Chronic and Subchronic Systemic
Toxicity
Parental, reproductive, and offspring potencies (i.e., inverse
log-transformed LEL) from the multigeneration studies were
compared to potency values for systemic toxicity from 2-year
chronic and 90-day subchronic studies in the rat (Fig. 2). For
this comparison, data were available in ToxRefDB for 254
chemicals tested in both multigeneration and 2-year chronic
studies, and 207 chemicals tested in both multigeneration and
90-day subchronic studies. The potency values compared
rarely correspond to the same treatment-related effect across
study type. For the majority of chemicals, potency values
between the multigeneration, chronic and subchronic studies
were comparable, with a general linear relationship falling
within ten-fold of each other. However, for four chemicals
(bisphenol A, deltamethrin, flucycloxuron, flufenpyr-ethyl) that
caused parental or reproductive effects in the multigeneration
study, there was no systemic toxicity observed in either the
chronic or subchronic studies. For another five chemicals
(cyprodinil, diethyltoluamide, difenoconazole, ethametsulfuron
methyl, thiamethoxam) potencies for the most sensitive
multigeneration endpoints were more than 10-fold greater than
Previous
TOC
-------
REPRODUCTIVE TOXICITY PROFILING FROM TOXREFDB
185
CHRONIC SYSTEMIC
SUBCHRONIC SYSTEMIC
C/3
1
Lowest Effect Level (mg/kg/day)
ME- Not Established
254 Chemicals with Multigeneration and 2-year Chronic Study
207 Chemicals with Multigeneration and 90-day Subchronic Study
Lowest Effect Level (mg/kg/day)
Within 2-Fold
FIG. 2. Parental, reproductive, and offspring LELs (inverse log transformed) from multigeneration rat studies were compared to systemic LEL from chronic/
cancer and subchronic rat studies for 254 and 207 chemicals, respectively. Points within gold lines indicate less than twofold difference between multigeneration
and chronic studies. Points within orange lines indicate less than 10-fold difference between multigeneration and chronic studies. "NE" stands for not established.
for the most sensitive effects in chronic studies. Of these five reproductive, or offspring endpoints, respectively. Of the seven
chemicals only thiamethoxam was more potent based solely on chemicals identified as twofold more potent reproductive
reproductive endpoints, that is, testicular atrophy. Decreasing toxicants, no reproductive organ toxicity was observed in the
the threshold from 10-fold to a 2-fold increase in potency rat chronic/cancer or subchronic studies for these chemicals—
resulted in 37, 7, and 20 chemicals more potent for parental, the multigeneration test detected reproductive toxicity that
Previous
TOC
-------
186
MARTIN ET AL.
could have been missed in chronic or subchronic studies.
Under the conditions of the 2-year chronic studies, the vast
majority of chemicals observed effects at lower doses than in
the multigeneration reproductive study. However, even in these
cases, the multigeneration test often identified selective
reproductive toxicants and endpoints not detected in the
chronic study.
Comparative Analysis of Parental, Reproductive, and
Offspring Endpoints
Chemicals with increased potency in the second generation
were identified by comparing PI and Fl, or Fl and F2 LEL
across parental, reproductive and offspring endpoint categories
for 316 chemicals (Fig. 3). Specific second generation effects
(i.e., Fl parental or reproductive, F2 offspring) not observed in
the first generation (i.e., PI parental or reproductive, Fl
offspring), or sensitive effects occurring at a lower LEL in the
second generation are provided for all 316 chemicals on the
ToxRefDB homepage (http://www.epa.gov/ncct/toxrefdb/).
For parental effects, 15 chemicals had specific effects in the
Fl versus PI, and another 48 were more sensitive in the Fl
versus PI based upon at least a twofold difference in LEL.
For reproductive toxicity endpoints, 52 chemicals had
specific effects in the Fl versus PI, and another 14 were more
sensitive in the Fl versus PI based upon at least a twofold
difference in LEL. For offspring toxicity endpoints, 14
chemicals had specific effects in the F2 versus Fl, and another
28 were more sensitive in the F2 versus Fl based upon at least
a twofold difference in LEL. Across reproductive and offspring
effects, a total of 94 chemicals displayed specificity or
sensitivity in the second generation. However, the Fl re-
productive or F2 offspring LEL was the most sensitive LEL
across all endpoint categories for only 16 of these 94 chemicals
(Table 2). Of the 16 second generation sensitive chemicals as
determined by specific LEL, only three of these chemicals had
reproductive or offspring LOAEL based on critical effects that
required mating of the Fl adults or were observed in the F2
offspring. Of these three, only fenarimol effects on F2 litter size
determined the chronic reference dose in the risk assessment.
This analysis in ToxRefDB has identified a subset of reference
chemicals for ToxCast predictive modeling that may be more
specific or potent reproductive toxicants. However, it is
important to note that these ToxRefDB values are LEL for
all treatment-related effects, and are in only a small minority of
cases critical effects being used for determination of NOAEL/
LOAEL.
Selected Multigeneration Study Endpoints for Predictive
Modeling
Figure 4 presents the incidence and distribution by
generation of effects on reproductive performance, reproduc-
tive organs, offspring viability, and parental systemic toxicities
selected as anchoring endpoints for ToxCast predictive
Fl
/
/, *.
/^
^
yf
/
,7
•
/
f
,.»
'
t
/
*
*
,
<*
/
'
*
x4
*
^
T
»
/
^
/'.
•
•
/
"•
» t
*
*
v.
*
/.*
X
,
/
t
KT
7
~'
**
*
'•
I
*
»ly
/
s
/
'i '
/
1
•
/
/
/
.* --/
*
/
*
4
/
X
PI
0.03125
0.0625
0.125
0.25
0.5
1
2
4
8
16
32
64
128
256
512
1024
2048
I
I
O
1
/
/
•
Fl
t
/
'
-•
/
• i
s
/
«
.
••
1
/
/
.
4,
/'
/
'/
'.
,'
'
/
/
/
A
•
/
f
/»
..
/:
/
/
/
•.
/
*
/
/
/
•
, •
V
'
t
/
*
*
/
*
/
••
•
/
/
/
«•
/
A
/
/
/
•
/
/
*
\
/
f
Jl
/
U.U130J3
0.125
*>
/
•
/
/
*
,
/
/
y
/
,
/
/
«,,
'
/
.
*
/
A
/
'
/
•'
• •
/
•
t
/
\
»
/
'.
*
/
•
,
-'
,
/'
»
/
»
>
/
/
,
*
'/
' •
*
/
/
.'
*
<
/
'
/
*
/
J
/
'
/
•
/
/
/
,-'
'
/
•
/'
*
'/
\
— p p
to —
Wi IO C> 'JJ —
w* to — ui
vy> to ON
Lowes! Effect Level (mg/kg/day)
ME - Not Established
Within 10-Fold
Wilhin 2-Fold
FIG. 3. Comparing LELs across generation and endpoint category. Points
within dark orange lines indicate less than twofold difference between
generations. Points within light orange lines indicate less than 10-fold
difference between generations. "NE" stands for not established.
Previous
-------
REPRODUCTIVE TOXICITY PROFILING FROM TOXREFDB
187
TABLE 2
Sixteen Chemicals with the Most Sensitive LELs from Fl Reproductive or F2 Offspring Toxicities
LEL (mg/kg/day)
Chemical name
Parental
Reproductive
Offspring
PI
Fl
PI
Fl
Fl
F2
Fl reproductive/F2 offspring sensitive effect
2,4-DB
Azoxystrobin
Bromuconazole
Carbaryl
Chlorethoxyfos
Clethodim
Desmedipham
Dicyclohexyl phthalate
Epoxiconazole
Fenarimol
Mepiquat chloride
Propetamphos
Stannane, tributylchloro-
TCMTB
Thiamethoxam
Triclosan
112
165
141
92.4
NE
263
20
402
31.9
NE
499
2.8
NE
NE
61.3
50
NE
165
141
92.4
0.78
263
20
89.9
22.1
NE
575
7.1
6.25
NE
61.3
150
112
NE
141
NE
NE
51
NE
NE
22.1
NE
499
5.5
NE
NE
NE
NE
NE
NE
NE
92.4
NE
1
110
17.8
22.1
1.2
575
0.3
0.25
NE
1.84
NE
112
165
141
92.4
0.6
NE
20
457
22.1
NE
575
5.5
1.25
NE
158
150
25
32.3
15.5
31.3
0.3
NE
4
457
0.85
NE
48.6
5.5
0.25
38.4
158
15
Kidney dilation6
Pup and liver weight changes6
Liver weight changes6
Offspring viability"'6
Pup weight decrease6
Prostate and seminal vesicle weight
Liver and kidney weight changes6
Prostate weight decrease"
Offspring viability6
Litter size decrease"'6
Eye opening delay6
Litter size decrease6
Pup, testis, and epididymis weight changes"
Pup weight decrease"'6
Testicular atrophy"
Pup weight decrease6
Note. Underline = parental, reproductive, or offspring LOAEL (study-level LOAEL). NE = not established (no observed effects).
"Study-level critical effect (Fl reproductive or F2 offspring).
6F1 mating required.
modeling. Toxicity profiles from multigeneration studies on
316 chemicals were based on a diverse set of 19 selected
effects or effect aggregations distributed in various combina-
tions across the PI, Fl, and F2 generations. A detailed table
listing all 19 of these endpoints for the 316 chemicals,
including endpoint descriptions and various transformations of
LEL values, is available for download from the ToxRefDB
homepage (http://www.epa.gov/ncct/toxrefdb/).
These 19 highly prevalent effects identified treatment-related
changes to reproductive performance including fertility,
mating, gestational interval, implantations, litter size, and live
birth index demonstrated effects at different stages of the
reproductive cycle. Besides effects of many chemicals on
offspring viability at PND4 and PND21 (viability and lactation
indices, respectively), pubertal delays were also recorded for
some chemicals. Pubertal delays were not part of the ToxCast
modeling dataset because only a small subset of chemicals and
studies assessed these endpoints. Effects on reproductive
performance and offspring viability were observed in 110
(35%) and 108 (34%) of the 316 tested chemicals, respectively.
Effects on reproductive organs, both organ weight and
pathology, were observed in 98 (31%) of the chemicals with
roughly 50% of those chemicals causing the effect only in the
second generation (Fl adult). Of the 98 chemicals, 31 caused
both male and female reproductive organ effects, 43 male only,
and 24 female only. Systemic target organ weight and
pathology endpoints were also selected, including the liver,
kidney and spleen, along with the endocrine-related adrenal,
pituitary, and thyroid glands.
The fairly restricted set of 19 effects characterized 151 of the
152 chemicals that demonstrated any reproductive toxicity.
Additionally, these 19 effects identified 229 of the 269
chemicals that caused any offspring toxicity. The remaining
40 chemicals not identified were predominantly affecting pup
weight only. This supports the hypothesis that we can extract
a small finite set of key reproductive effects from this dataset
for use in developing robust predictive signatures in the future
stages of ToxCast research as a prioritization tool spanning
reproductive toxicity.
DISCUSSION
ToxRefDB is being developed with several applications in
mind. One is to provide in vivo toxicity effects as targets for
ToxCast predictive models. In this fashion, ToxCast can be
established as a cost-effective rapid approach for screening and
prioritizing a large number of chemicals for further lexicolog-
ical testing (Dix et al., 2007). Using data from high-throughput
screening (HTS) bioassays developed in the pharmaceutical
industry, ToxCast is building computational models to forecast
the potential toxicity of chemicals. These hazard predictions
should provide EPA regulatory programs with science based
information helpful in prioritizing chemicals for more detailed
Previous
TOG
Next
-------
188
MARTIN ET AL.
Reproductive
Performance
Reproductive
Organ
Offspring
Parental
Fertility
Mating
Gestational Interval
Implantations
Litter Size
Live Birth Index (PND1)
Testis
Epididymis
Prostate
Ovary
Uterus
Viability Index (PND4)
Lactation Index (PND21)
Adrenal
Pituitary
Thyroid
Kidney
Liver
Spleen
20 40 60 80 100
No. Chemicals (Total: 316)
120
140
FIG. 4. Incidence and distribution, by generation, of the 19 endpoints selected for predictive modeling, including reproductive, offspring, and systemic
toxicity endpoints from the rat multigeneration reproduction study (see Results and http://www.epa.gov/ncctAoxrefdb/ for details). The light gray bar indicates
chemicals observing the endpoint only in the first generation, either PI adult or Fl juvenile. The medium gray bar indicates chemicals observing the endpoint in
both first and second generation treatment groups. The dark gray bar indicates chemicals observing the endpoint only in the second generation, either Fl adult or
F2 juvenile.
toxicological evaluations, and therefore lead to using fewer
animal tests. Target chemicals for such prioritization include
pesticidal inerts, antimicrobials, and the many industrial
chemicals with limited toxicity information (Judson et al.,
2008c). ToxCast is currently in the proof-of-concept phase,
wherein over 300 chemicals have been assayed in over 500
different HTS bioassays, creating bioactivity profiles being
used to derive signatures predicting the known toxicity for
these chemicals (Judson et al., 2009).
The Phase I chemicals are primarily conventional pesticide
actives that have been extensively evaluated using traditional
mammalian toxicity testing, and hence have known properties
representative of a number of toxicity outcomes (e.g.,
reproductive toxicity). Thus a critical component of ToxCast
is ToxRefDB, which is being populated with data from OPP for
pesticide active chemicals and being extracted from the
evaluations on these studies conducted by OPP scientists.
Comparable toxicity data from other toxicity sources (e.g.,
National Toxicology Program) are also being captured in
ToxRefDB. A broader and more diverse set of complementary
data on thousands of chemicals is being captured in EPA's
Aggregated Computational Toxicology Resource (http://actor.
epa.gov/actor; Judson et al., 2008b). Although pesticide
toxicity data currently predominates in ToxRefDB, the
database is being expanded to a broader range of chemicals,
both by category and use.
The underlying data represented in ToxRefDB has been
evaluated by EPA in prior pesticide registration decisions, and
the presence of effects in high-dose animal studies do not
translate directly into significant human risk stemming from
registered uses of the pesticide. One major issue to note is that
the current analysis of ToxRefDB is not limited to just the
critical effects leading to regulatory determinations of LOAEL
and NOAEL. In addition, it should be noted that the EPA uses
animal toxicology studies, like those entered into ToxRefDB,
as well as other sources of information such as effects on
wildlife populations, mechanisms of action, use patterns,
environmental fate and persistence, food residue levels, and
Previous
-------
REPRODUCTIVE TOXICITY PROFILING FROM TOXREFDB
189
human exposure potential in its determinations to register
pesticides, and to establish acceptable levels of pesticide
residues for uses in the United States (http://www.epa.gov/
pesticides/).
The toxicity data in ToxRefDB (www.epa.gov/ncct/toxrefdb)
and the HTS data generated in ToxCast (www.epa.gov/ncct/
toxcast) is being made publicly available through EPA
websites. The first component of ToxRefDB was recently
published (Martin et al., 2009), presenting endpoints to be used
for predictive modeling from two-year rodent bioassays on 310
chemicals. The analysis and release of developmental toxicity
endpoints on 383 chemicals from ToxRefDB will also provide
key endpoints to be used for predictive modeling (Knudsen,
2009). Multigeneration reproduction study data for 316
chemicals was entered into ToxRefDB making the vast library
of legacy data computable for the first time. The pattern of
reproductive toxicity across these chemicals resulted in group-
ings of similar chemicals that could be used to match up with
HTS bioactivity profiles, In the meantime, the analysis
corroborated the distinction between parental, offspring and
reproductive effects in downstream analyses based on the
distribution of endpoints across the chemical groups.
All 12,230 effects in the multigeneration study dataset were
placed into three major classes of effects; parental, reproductive
and offspring. The LEL for each class or category of effects
were used to identify sensitive or specific reproductive
toxicants based on comparisons to chronic and subchronic
study data and cross-generational comparisons within the
multigeneration reproductive test, In general, chemical expo-
sures under conditions of the multigeneration reproduction
study were less sensitive than under the conditions of the
2-year chronic study and comparable to the 90-day subchronic
study. The analysis did, however, identify a subset of 94
chemicals with sensitive or specific reproductive or offspring
toxicities when compared to systemic effects under longer
continuous exposure periods. Additional future analyses
comparing, for instance, maternal and fetal toxicity from
developmental toxicity studies (Knudsen, 2009) to parental and
offspring toxicity from reproductive toxicity studies, will
provide additional insight into the role of developmental
exposures in the manifestation of specific toxicities.
Similar insight can be gleaned from comparing endpoints
occurring at a lower dose or only in the second generation, that
is, second generation sensitive or specific effects, respectively.
Effects that occur in the first generation and are not
corroborated in the second generation can be questioned as
to its toxicological relevance. Conversely, effects with
consistent increases in second generation sensitivity or
specificity might reflect the need for reproductive or de-
velopmental exposure to occur. Comparisons across these
broad classes of endpoints honed in on specific effects for
which to characterize the chemical set. The primary set of
effects selected as anchoring endpoints for ToxCast predictive
modeling were reproductive indices, offspring viability, and
male and female reproductive organ effects, along with a set of
parental systemic organ toxicities.
The current study focused on providing endpoints for
predictive modeling as part of the ToxCast research program
(Dix et al., 2007), but also began to address the importance of
specific study design parameters, including differences across
generation, life-stage and various classes of endpoints. It has
recently been suggested that the reproductive test guidelines for
agrichemicals could be refined to make the second generation
optional based on results seen in the first generation (Cooper
et al., 2006). Consistent with results from Janer et al. (2007),
the current analysis of this ToxRefDB dataset supports the
hypothesis that the second, F2 generation in these 329 studies
would rarely impact either the qualitative or quantitative
evaluations of these studies. Of the sixteen second generation
sensitive chemicals, carbaryl, fenarimol, and TCMTB observed
second generation effects that would have required Fl mating.
However, of these three chemicals only fenarimol effects on F2
litter size determined the chronic reference dose determination
(U.S. EPA, 2006, 2007a,b). Additional analysis wiU be per-
formed on this dataset in collaboration with OPP and other
international chemical regulatory agencies to expound upon the
role of these and other study design parameters with respect to
chemical regulation and potential guideline study design
changes. For instance, 53 of the 73 chemicals proposed for
screening in the Endocrine Disrupter Screening Program (EDSP;
http://www.epa.gov/endo/pubs/prioritysetting/draftlist.htm) have
multigeneration studies entered into ToxRefDB. Where avail-
able, multigeneration study data for the remaining chemicals are
now being entered into ToxRefDB. A focused analysis of the
EDSP chemical set to assess the ability of the current and
previous guidelines to identify reproductive effects related to
endocrine disruption would be just one example of the utility
of ToxRefDB (Kavlock et al., 2009). The use of ToxRefDB to
address many research and regulatory science questions re-
garding in vivo mammalian toxicity not only provides trans-
parency, but also assists in guiding the next set of questions.
The diverse utility of ToxRefDB as a reference database for
research applications such as ToxCast demonstrates the power
of curating toxicity information into a relational database, In
the current analysis on the multigeneration reproductive
toxicity test, six chemical sets were derived and subsequently
nineteen specific endpoints were identified to serve anchoring
endpoints for eventual predictive modeling. These endpoints
are further defined by life-stage or generation, and fully
characterize the reproductive toxicity potential of the 316 in
this study. Capturing this reproductive toxicity data in
ToxRefDB supports ongoing retrospective analyses, test
guideline revisions, and computational toxicology research.
SUPPLEMENTARY DATA
Supplementary data are available online at http://toxsci.
oxfordjournals .org/.
Previous
TOG
Next
-------
190
MARTIN ET AL.
FUNDING
United States Environmental Protection Agency.
REFERENCES
Collins, F. S., Gray, G. M., and Bucher, J. R. (2008). Transforming
environmental health protection. Science 319, 906-907.
Cooper, R. L., Lamb, J. C., Barlow, S. M., Bentley, K., Brady, A. M.,
Doerrer, N. G., Eisenbrandt, D. L., Fenner-Crisp, P. A., Mines, R. N.,
Irvine, L. F., et al. (2006). A tiered approach to life stages testing for
agricultural chemical safety assessment. Crit. Rev. Toxicol. 36, 69-98.
Dix, D. J., Houck, K. A., Martin, M. T., Richard, A. M., Setzer, R. W., and
Kavlock, R. J. (2007). The ToxCast program for prioritizing toxicity testing
of environmental chemicals. Toxicol. Sci. 95, 5-12.
Houck, K. A., and Kavlock, R. J. (2008). Understanding mechanisms of toxicity:
Insights from drug discovery research. Toxicol. Appl. Pharmacol. 227,163-178.
Ihaka, R., and Gentleman, R. (1996). R: A language for data analysis and
graphics. /. Comput. Graph. Stat. 5, 299-314.
Janer, G., Hakkert, B. C., Slob, W., Vermeire, T., and Piersma, A. H. (2007). A
retrospective analysis of the two-generation study: What is the added value
of the second generation? Reprod. Toxicol. 24, 97-102.
Judson, R. S., Dix, D. J., Houck, K. A., Martin, M. T., and Kavlock, R. J.
(2009). Developing predictive signatures using in vitro data from the EPA
ToxCast program. Toxicologist 108, 79-80.
Judson, R. S., Elloumi, F., Setzer, R. W., Li, Z., and Shah, I. (2008a). A
comparison of machine learning algorithms for chemical toxicity classifica-
tion using a simulated multi-scale data model. BMC Bioinformatics 9, 241
doi:10.1186/1471-2105-9-241.
Judson, R. S., Richard, A. M., Dix, D. J., Houck, K., Elloumi, F., Martin, M. T.,
Carney, T., Transue, T. R., Spencer, R., and Wolf, M. (2008b). ACToR—
Aggregated computational toxicology resource. Toxicol. Appl. Pharmacol.
doi:10.1016/j.taap.2007.12.037.
Judson, R., Richard, A., Dix, D. J., Houck, K., Martin, M., Kavlock, R.,
Dellarco, V., Henry, T., Holderman, T., Sayre, P., et al. (2008c). The toxicity
data landscape for environmental chemicals. Environ. Health Perspect.doi:
10.1289/ehp.0800168.
Kavlock, R. J., Dix, D. J., Houck, K. A., Judson, R. S., Martin, M. T., and
Richard, A. M. (2008). ToxCast: Developing predictive signatures for
chemical toxicity. Altern. Anim. Test. Exp. 14(Special issue), 623-627.
Kavlock, R. J., Dix, D. J., Houck, K. A., Martin, M. T., and Judson, R. (2009).
Biological profiling of endocrine related effects of chemicals using ToxCast.
Toxicologist 108, 226.
Knudsen, T. B., Martin, M. T., Kavlock, R. J., Judson, R. S., Dix, D. J., and
Singh, A. V. (2009). Profiling the activity of environmental chemicals in
prenatal developmental toxicity studies using the U.S. EPA's ToxRefDB.
Reprod. Toxicol. doi: 10.1016/j.reprotox.2009.03.016.
Martin, M. T., Judson, R. S., Reif, D. M., Kavlock, R. J., and Dix, D. J. (2009).
Profiling chemicals based on chronic toxicity results from the U.S. EPA
ToxRef database. Environ. Health Perspect. 117, 392-399.
National Research Council (NRC). (2007). Toxicity Testing in the 21st
Century: A Vision and a Strategy. The National Academies Press,
Washington, DC.
Thorndike, R. L. (1953). Who belongs in the family? Psychometrika 18,
267-276.
U.S. Environmental Protection Agency (U.S. EPA). (1982). OPP Guideline
83—84: Reproductive and Fertility Effects. Pesticide Assessment Guide-
lines, Subdivision F, Hazard Evaluation: Human and Domestic Animals.
Office of Pesticides and Toxic Substances, Washington, DC, EPA-540/
9-82-025.
U.S. Environmental Protection Agency (U.S. EPA). (1996). Reproductive
toxicity risk assessment guidelines. Fed. Regist. 61, 56274-56322.
U.S. Environmental Protection Agency (U.S. EPA). (1998). Health Effects
Test Guidelines OPPTS 870.3800 Reproduction and Fertility Effects.
Office of Pesticides and Toxic Substances, Washington, DC, EPA 712-
C-98-208.
U.S. Environmental Protection Agency (U.S. EPA). (2006). Reregistration
Eligibility Decision for 2-(Thiocyanomethylthio)-benzothiazole (TCMTB).
Office of pesticides and toxic substances, Washington, DC, EPA 739-R-07-003
U.S. Environmental Protection Agency (U.S. EPA). (2007a). Fenarimol
Summary Document Registration Review: Initial Docket. Office of Pesticides
and Toxic Substances, Washington, DC, EPA-HQ-OPP-2006-0241.
U.S. Environmental Protection Agency (U.S. EPA). (2007b). Reregistration
Eligibility Decision (RED) for Carbaryl. Office of Pesticides and Toxic
Substances, Washington, DC, EPA 738-R-07-018.
U.S. Environmental Protection Agency (U.S. EPA). (2009). The U.S.
Environmental Protection Agency's Strategic Plan for Evaluating the
Toxicity of Chemicals. Office of the Science Advisor, Science Policy
Council, Washington, DC, EPA 100/K-09/001.
Ward, J. H. (1963). Hierarchical grouping to optimize an objective function.
/. Am. Stat. Assoc. 58, 236-244.
Previous
TOG
Next
-------
Guest Editorial The future of toxicity testing
The Future of Toxicity
Testing for Environmental
Contaminants
Toxicity testing and assessment sit on the cusp
of a transformational change brought about by
the rapid emergence of tools and capabilities
in molecular biology and computational and
informational sciences. This transformation
has the potential to dramatically reshape the philosophy and approaches
underlying toxicity testing and the assessment of human health risks
associated with exposure to environmental contaminants.
Such a transformation is especially significant for agencies that
are responsible for implementing congressionally mandated pro-
grams under which the risks of exposure to a wide variety of envi-
ronmental pollutants are assessed and regulated. Most often, such
regulatory decisions have relied on toxicity testing data obtained
nearly exclusively from experimental animal models. This approach,
however, presents challenges in accommodating the need for more
efficient and cost-effective means to screen and prioritize chemicals
for testing and addressing increasingly complex issues such as life-
stage susceptibility and genetic variations in the human population,
the risks of concurrent, cumulative exposure to multiple and diverse
chemicals, and, fundamental to all, improved understanding of the
mechanism through which toxicity occurs.
The U.S. Environmental Protection Agency (EPA) has recognized
the potential application of emerging science to improve toxicity test-
ing and risk assessment (U.S. EPA 2002, 2004), notably by taking the
lead in commissioning the National Research Council (NRC) in 2004
to review existing strategies (NRC 2006) and develop a long-range
vision for toxicity testing and risk assessment (NRC 2007). Beyond
EPA, other federal programs have also recognized the need for this
transformative shift, as reflected in the National Toxicology Program's
(NTP) A National Toxicology Program for the 21st Century: Roadmap
for the Future (NTP 2004) and the Food and Drug Administration's
FDA's Critical Path Initiative (FDA 2008).
To build on the NRC document, the U.S. EPA established
an internal, cross-agency workgroup that produced The U.S.
Environmental Protection Agency's Strategic Plan for Evaluating the
Toxicity of Chemicals (U.S. EPA 2009) to provide a framework for
EPA to comprehensively move forward to incorporate this new
scientific paradigm into future toxicity testing and risk assessment
practices. The strategy is centered on three interrelated issues: a) the
use of toxicity pathways information in screening and prioritization
of chemicals for further testing; b) the use of toxicity pathways infor-
mation in risk assessment; and c) organizational transition. The last
element explicitly recognizes that regulatory offices within EPA will
need to be actively involved in overseeing the significant transition to
this new paradigm and the translation of the attendant data for regu-
latory application.
Research to address the first issue will build on the efforts of
EPA's ToxCast program in identifying and developing simple, reli-
able screening models to predict chemical hazard (U.S. EPA 2008a).
The second effort will seek to apply the toxicity pathways concept
in a systems biology approach, to better delineate the molecular
and cellular changes that perturb normal homeostatic mechanisms
toward a given toxicity pathway or set of toxicity pathways. This
information should reduce the uncertainty currently associated with
dose-response models by increasing their biological plausibility.
Recognizing the necessity and benefits of collaboration to
achieve the NRC's vision, EPA recently signed a Memorandum of
Understanding with the NTP and the National Institutes of Health
Chemical Genomics Center (U.S. EPA(2008b) to advance the high
throughput screening and toxicity pathway profiling in risk assess-
ment. This "Tox21" consortium is now actively coordinating efforts
to identify chemicals, pathways, screening assays, and informatic
approaches to assess the effects of thousands of chemicals (Kavlock
et al. 2009). The U.S. EPA is also working with the European
Commission and the Organization for Economic Cooperation and
Development to facilitate global collaborations.
As recognized by the NRC (2007), the development and imple-
mentation of a transformational paradigm will require a major
commitment to new funding to sustain an iterative and long-term
process that changes institutional toxicity testing and risk assessment
practices. Regulators, stakeholders, and the public must be confident
that the new types of data can be used to effectively assess risk and
ultimately protect public health. As such, education and transpar-
ent communication will be critical. Ultimately, the testing paradigm
must be evaluated via a comprehensive development and review
process, involving public comment, expert peer review, and harmo-
nization with other agencies and international organizations. EPA's
Strategic Plan for Evaluating the Toxicity of Chemicals (U.S. EPA
2009) lays the framework upon which the development, implemen-
tation, acceptance, and application of this transformative paradigm
can be built.
The views expressed in this letter are those of the individual authors and
do not necessarily reflect the views of the U.S. EPA.
Melissa G. Kramer
Office of the Science Advisor
U.S. Environmental Protection Agency
Washington, DC
E-mail: Kramer.melissa@epa.gov
Michael Firestone
Office of Children's Health Protection and Environmental
Education
U.S. Environmental Protection Agency
Washington, DC
Robert Kavlock
Harold Zetiick
Office of Research and Development
U.S. Environmental Protection Agency
Research Triangle Park, NC
Melissa Kramer has worked at the U.S. EPA on science policy issues since
2002, when she was a AAAS Science and Technology Policy Fellow. She
currently works in the Office of Policy, Economics, and Innovation.
Environmental Health Perspectives • VOLUME 1171 NUMBER 7 I July 2009
Previous
A 283
-------
Guest Editorial The future of toxicity testingGuest
Michael Firestone is Science Director in the U.S. EPA Office of
Children's Health Protection and Environmental Education. His
primary focus is on the development of risk assessment guidance and
policy that explicitly considers the potential for greater early lifestage
susceptibility, including work on developing EPA's guidance on cancer
risk assessment, childhood age grouping, and probabilistic modeling. In
addition, he has helped develop EPA's Risk Assessment Portal (www.epa.
gov/risk) and was the agency's project officer for the recent NRC report
Toxicity Testing in the 21st Century.
Robert Kavlock is the Director of U.S. EPA National Center for
Computational Toxicology. He has more than 30years' experience in
developing screening tools and methodologies for noncancer risk assess-
ments.
Harold Zenick is Director of the National Health and Environmental
Effects Research Laboratory in the Office of Research and Development
in the U.S. EPA. His current interests are in integrating human health
and ecological risk assessment, strengthening the linkages between envi-
ronmental and public health agendas and agencies, and the application
of emerging computational, informational, and molecule sciences in
improving toxicity testing and risk assessment practices.
REFERENCES
FDA. 2008. FDA's Critical Path Initiative. Available: http://www.fda.gov/ScienceResearch/
SpecialTopics/CriticalPathlnitiative/default.htm [accessed 10 June 2009].
Kavlock RJ, Austin CP, Tice RR. 2009. Toxicity testing in the 21st century: implications for human
health risk assessment. Risk Anal 29(41:485-487.
NRC (National Research Council). 2006. Toxicity Testing for Assessment of Environmental
Agents. Washington, DC:National Academies Press.
NRC (National Research Council). 2007. Toxicity Testing in the 21st Century: A Vision and a
Strategy. Washington, DC:National Academies Press.
NTP. 2004. A National Toxicology Program for the 21st Century: A Roadmap for the Future.
Research Triangle Park, NC:National Toxicology Program.
U.S. EPA. 2002. Interim Genomics Policy. Washington, DC:U.S. Environmental Protection
Agency. Available: http://www.epa.gov/osa/spc/genomics.htmlaccessed 1 August2008].
U.S. EPA. 2004. Genomics White Paper. EPA 100/B-04/002. Washington, DC:U.S. Environmental
Protection Agency.
U.S. EPA. 2008a. ToxCast™ Program: Predicting Hazard, Characterizing Toxicity Pathways,
and Prioritizing the Toxicity Testing of Environmental Chemicals. Washington, DC:U.S.
Environmental Protection Agency. Available: http://www.epa.gov/nccVtoxcast/ [accessed
1 August 2008].
U.S. EPA. 2008b.Memorandum of Understanding. Washington, DC:U.S. Environmental Protection
Agency, National Center for Computational Toxicology. Available: http://www.epa.gov/
comptox/articles/comptox_mou.html [accessed 1 August 2008].
U.S. EPA. 2009. The U.S. Environmental Protection Agency's Strategic Plan for Evaluating the
Toxicity of Chemicals. EPA100/K-09/001. Washington, DC:U.S. Environmental Protection
Agency.
Benzene 2009: Health Effects and Mechanisms of Bone Marrow Toxicity. Implications for t-AML
and the Mode of Action Framework will be held September 7-11, 2009 at Technische Universitat
Munchen, Munich Germany. A satellite meeting to Eurotox 2009, Benzene 2009 will:
• Review the relationship between bone marrow toxicity and leukemia;
• Discuss recent studies on the epidemiology of benzene-induced diseases;
• Explore mechanistic studies of reactive metabolites and reactive oxygen species,
transgenics and signal transduction, DNA damage and repair;
• Evaluate exposure metrics and biomarkers; and
• Place the evidence in context using a mode-of-action approach to inform risk assessment.
Abstracts are invited for poster and plenary presentations (deadline for submitting: June 30, 2009). Travel
fellowships are available for students or postdoctoral researchers. For further information, please visit
www.tum-benzenesvmposium.de.
A 284
VOLUME 1171 NUMBER 7 I July 2009 • Environmental Health Perspectives
Previous
TOC
-------
Review
The Toxicity Data Landscape for Environmental Chemicals
Richard Judson,1 Ann Richard,1 David J. Dix,1 Keith Houck,1 Matthew Martin,1 Robert Kavlock,1 Vicki Dellarco,2
Tala Henry,3 Todd Holderman,3 Philip Sayre,3 Shirlee Tan,4 Thomas Carpenter,5 and Edwin Smith6
1National Center for Computational Toxicology, Office of Research and Development, U.S. Environmental Protection Agency,
Research Triangle Park, North Carolina, USA; 2Office of Pesticide Programs, Office of Prevention, Pesticides, and Toxic Substances,
U.S. Environmental Protection Agency, Arlington, Virginia, USA; 3Office of Pollution Prevention and Toxics and 4Office of Science
Coordination and Policy, Office of Prevention, Pesticides, and Toxic Substances, U.S. Environmental Protection Agency, Washington,
DC, USA; 5Office of Water, Office of Ground Water and Drinking Water, U.S. Environmental Protection Agency, Washington, DC, USA;
6Great Lakes National Program Office, U.S. Environmental Protection Agency, Chicago, Illinois, USA
OBJECTIVE: Thousands of chemicals are in common use, but only a portion of them have undergone
significant toxicologic evaluation, leading to the need to prioritize the remainder for targeted testing.
To address this issue, the U.S. Environmental Protection Agency (EPA) and other organizations are
developing chemical screening and prioritization programs. As part of tliese efforts, it is important
to catalog, from widely dispersed sources, the toxicology information that is available. The main
objective of this analysis is to define a list of environmental chemicals that are candidates for the
U.S. EPA screening and prioritization process, and to catalog the available toxicology information.
DATA SOURCES: We are developing ACToR (Aggregated Computational Toxicology Resource),
which combines information for hundreds of thousands of chemicals from > 200 public sources,
including the U.S. EPA, National Institutes of Heahh, Food and Drug Administration, correspond-
ing agencies in Canada, Europe, and Japan, and academic sources.
DATA EXTRACTION: ACToR contains chemical structure information; physical—chemical properties;
in vitro assay data; tabular in vivo data; summary toxicology calls (e.g., a statement that a chemical
is considered to be a human carcinogen); and links to online toxicology summaries. Here, we use
data from ACToR to assess the toxicity data landscape for environmental chemicals.
DATA SYNTHESIS: We show results for a set of 9,912 environmental chemicals being considered for
analysis as part of the U.S. EPA ToxCast screening and prioritization program. These include high-
and medium-production-volume chemicals, pesticide active and inert ingredients, and drinking
water contaminants.
CONCLUSIONS: Approximately two-thirds of these chemicals have at least limited toxicity sum-
maries available. About one-quarter have been assessed in at least one highly curated toxicology
evaluation database such as tlie U.S. EPA Toxicology Reference Database, U.S. EPA Integrated
Risk Information System, and the National Toxicology Program.
KEY WORDS: ACToR, carcinogenicity, database, developmental, hazard, HPV, MPV, pesticide,
reproductive, toxicity. Environ Health Perspect 117:685-695 (2009). doi:10.1289/ehp.0800168
available via http:lldx,doi,orgl [Online 22 December 2008]
The U.S. Environmental Protection Agency
(EPA) has a significant interest in develop-
ing more efficient and informative toxicity
determination approaches in part because
of the large number of chemicals under its
jurisdiction. Ultimately, it would be bene-
ficial to characterize the toxicologic profiles
of all chemicals in use in the United States.
However, the size of this chemical universe
[in excess of 75,000 chemicals, which is the
estimated number in the Toxic Substances
Control Act (TSCA 1976) inventory (U.S.
EPA 2004b) makes this goal too difficult
using current approaches to toxicity charac-
terization that rely on extensive animal test-
ing, cost millions of dollars, and can take
2—3 years per chemical. The International Life
Sciences Institute/Health and Environmental
Sciences Institute (ILSI/HESI) recently
released several reports describing a more
focused, tier-based approach for toxicity test-
ing of agricultural chemicals, which would
ultimately lead to the use of fewer animals
(Barton et al. 2006; Carmichael et al. 2006).
The National Research Council (NRC)
recently released a report titled Toxicity
Testing in the 21st Century: A Vision and a
Strategy that outlines a much more ambi-
tious and long-term vision for developing
novel in vitro approaches to chemical tox-
icity characterization and prediction (NRC
2007) that would largely eliminate animal
testing. The NRC report addresses several
concerns about the current testing methods,
specifically, the desire d) to reduce the num-
ber of animals used in testing, b) to reduce
the overall cost and time required to charac-
terize each chemical, and c) to increase the
level of mechanistic understanding of chemi-
cal toxicity. The U.S. EPA and the National
Institutes of Health (NIH) are actively pursu-
ing approaches to implement ideas outlined
in the NRC report (Collins et al. 2008).
Regardless of the level of quality of toxi-
cology data on environmental chemicals,
many chemicals lack significant amounts
of data. In the United States and Canada,
an estimated 30,000 chemicals are in
wide commercial use, based on U.S. EPA
and Environment Canada data (Muir and
Howard 2006). The European Union's
Registration, Evaluation, and Authorization
of Chemicals (REACH) program has recently
released its first set of registered substances,
which contains > 140,000 entries (REACH
2008). The exact number of chemicals in use
is, in a sense, unknowable because it depends
on where one sets the threshold of use and
because use changes over time. The major
point is that the number is relatively large
and that only a relatively small subset of these
chemicals have been sufficiently well charac-
terized for their potential to cause human or
ecologic toxicity to support regulatory action.
This "data gap" is well documented (Allanou
et al. 1999; Applegate and Baer 2006;
Birnbaum et al. 2003; Guth et al. 2005; NRC
2007; U.S. EPA 1998).
The high cost and lengthy times associ-
ated with the use of animal testing to deter-
mine a chemical's potential for toxicity make
this strategy impractical for evaluating tens
of thousands of chemicals, hence the large
inventories of existing chemicals for which
few or no test data are available. An alterna-
tive approach is to attempt to assess much
larger numbers of chemicals by employing
more efficient in vitro methods. One strategy
applies a broad spectrum of relatively inex-
pensive and rapid high-throughput screening
Address correspondence to R. Judson, U.S.
Environmental Protection Agency, 109 T.W.
Alexander Dr. (B205-01), Research Triangle Park,
NC 27711 USA. Telephone: (919) 541-3085. Fax:
(919) 541-1194. E-mail: judson.richard@epa.gov
We acknowledge significant contributions from
members of the U.S. EPA Aggregated Computational
Toxicology Resource (ACToR) development team:
T. Cathey, T. Transue, and R. Spencer of Lockheed
Martin, and F. Elloumi, D. Smith, J. Vail, and K.
Daniel. We also acknowledge the significant contribu-
tion of M. Wolf (Lockheed Martin) in relation to the
U.S. EPA's Distributed Structure-Searchable Toxicity
Data Network structure inventory incorporated into
ACToR.
This article has been reviewed by the U.S. EPA
and approved for publication. Approval does not
signify that the contents necessarily reflect the views
and policies of the agency, nor does mention of trade
names or commercial products constitute endorse-
ment or recommendation for use.
The authors declare they have no competing
financial interests.
Received 8 September 2008; accepted 22 December
2008.
Environmental Health Perspectives • VOLUME 117 I NUMBER 5 I May 2009
Previous
685
-------
Judson et al.
(HTS) assays to a large set of chemicals, fol-
lowed by the use of these results to prioritize
a much smaller subset of chemicals for more
detailed analysis. The "prioritization score"
for a chemical would be based on signatures,
or patterns extracted from the HTS data, that
are predictive of particular effects or modes
of chemical toxicity. A comprehensive priori-
tization approach will also require the use of
exposure and pharmacokinetic estimates, in
addition to the intrinsic hazard information
provided by in vitro assays. Chemicals of known
toxicity make up the training and validation
sets that are used to develop and validate these
predictive signatures. HTS assays that yield
data for the predictive signatures would then
be run on chemicals of unknown toxicity (the
test chemicals), and a prioritization score for
those chemicals would be produced. The U.S.
EPA has made a significant investment in this
approach through the ToxCast research pro-
gram (Dix et al. 2007). ToxCast is currently
screening hundreds, and eventually thousands,
of environmental chemicals using hundreds of
HTS assays with the goal to develop predictive
toxicity signatures, and is using these signa-
tures to prioritize chemicals for further test-
ing. In this context, the term "environmental
chemicals" refers primarily to pesticides and
industrial chemicals that are used or produced
in large enough quantities to pose potential for
human or ecologic exposure [largely the high-
production-volume (HPV) and medium-pro-
duction-volume (MPV) chemicals described
below]. However, a number of environmental
chemicals that are captured in our analysis are
food ingredients or naturally occurring human
metabolites. We included many of the former
because they are classified as inert ingredients
in pesticide products.
In this article we address two key aspects
of this chemical screening and prioritization
process. The first is the definition of a set of
chemicals of interest to a screening program,
based on their widespread use or other poten-
tial for significant human exposure, or the
current availability of toxicity information
that can be used in building screening mod-
els. Some widely used but as yet uncharacter-
ized chemicals may not be good candidates
for screening because their physical-chemical
properties make them impractical to test in
in vitro assays (e.g., insoluble or highly vola-
tile compounds), whereas other substances
that we define as environmental chemicals are
regarded to be safe under intended use situa-
tions and may not require further testing, but
can serve as negative controls. For instance, a
subset of pesticide inert ingredients are also
on the U.S. Food and Drug Administration
(FDA) Generally Recognized as Safe chemical
list. As a further example, some "chemicals"
that are listed as pesticide inert ingredients are
common foods, such as milk.
The second objective is the characterization
of the sources and amount of reliable in vivo
toxicology data that can be used for develop-
ing and validating screening models in pro-
grams such as ToxCast. A significant amount
of high-quality toxicity data are needed to
train and validate in vitro—based models for
predicting chemical hazard. Equally important
is the presence of both negative and positive
examples for each toxicity end point to be
modeled. In addition to the sets of environ-
mental chemicals described here, pharmaceuti-
cal compounds are another source of detailed
animal and human toxicology data.
The sets of chemicals on which we have
focused are the HPV and MPV chemicals
from the TSCA inventory, pesticide and anti-
microbial active and inert ingredients, known
drinking water contaminants, hazardous air
pollutants (HAPs) and certain defined classes of
chemicals of interest, including the U.S. EPA's
Toxics Release Inventory (TRI), Integrated Risk
Information System (IRIS), and the first set of
chemicals to be tested through the Endocrine
Disrupter Screening Program (EDSP). The
TRI, drinking water contaminant, and EDSP
chemicals are largely included in the TSCA
inventory and pesticide active and inert ingredi-
ent lists. By combining these sources, we define
a set of 9,912 chemicals. Below we describe in
detail the process we used to arrive at this num-
ber. At present, we have limited the scope of
in vivo toxicology data to that which is relevant
to human health, as opposed to ecotoxicity. An
equivalent analysis for the ecotoxicity data land-
scape will be carried out in the future.
To support a data-intensive analysis
of environmental chemicals, we have devel-
oped a system called ACToR (Aggregated
Computational Toxicology Resource) (Judson
et al. 2008; U.S. EPA 2008a), which is a data-
base holding essentially all publicly available
information on chemical identity, structure,
physical-chemical properties, in vitro assay
results, and in vivo toxicology data. All of the
data described in this article have been col-
lected in ACToR.
Target Chemicals for Analysis
The U.S. EPA has authority to review and/or
regulate a large number of chemicals under
a variety of statutes, including those govern-
ing the manufacture, import, sale, and use of
pesticides and industrial chemicals. The large
numbers of chemicals on various U.S. chemi-
cal inventories, and the limited toxicity infor-
mation for many of these, have already been
stated as the driver for the need to set priorities
for additional testing. Because this universe of
chemicals is so large, it is even necessary to
prioritize what goes into a science-based pri-
oritization approach such as ToxCast. In this
article we focus on chemicals that are of inter-
est because d) they are known to be bioactive
(e.g., pesticide active ingredients), £) they are
manufactured or used in large quantities (HPV
and MPV chemicals), or c) many people may
be exposed to them on a routine basis (e.g.,
drinking water contaminants). We include
both largely uncharacterized chemicals and
chemicals for which significant toxicology
information is already available (e.g., pesti-
cide active ingredients, IRIS chemicals, and
chemicals on the TRI). The well-characterized
chemical groups are important because these
allow us to develop and validate predictive
models for prioritization of the remaining,
largely uncharacterized chemicals.
Based on these criteria, we focused on sets
of chemicals that are defined in the remain-
der of this section. Some of these lists are not
static, so we have chosen versions available
as of a specific date. For each of the lists, we
describe the rules for inclusion and provide
the total number of chemicals used for the
current evaluation. "Official" versions of these
lists are updated and posted to the relevant
U.S. EPA websites only every 2 or more years,
so in several cases, we have extracted more
current snapshots of the lists from internal
U.S. EPA databases. Many of the chemicals
we included in this analysis are complex mix-
tures. Additionally, these lists have significant
overlap; for instance, some pesticide active
ingredients are also HPV chemicals. Finally, to
be included in the current ACToR inventory,
a chemical must be identified by a Chemical
Abstracts Service Registry Number (CASRN).
Possible later extensions of this analysis
could consider chemicals with lower produc-
tion volumes or lower exposure potential
than those considered presently. These would
include the Canadian Domestic Substances
List (DSL), which includes approximately
30,000 chemicals, and the large collection of
chemicals to be analyzed under the REACH
program. REACH is still in the process
of defining its target list, but an estimated
30,000 chemicals will be included. Many of
the Canadian DSL and REACH chemicals
have U.S. use and/or production levels below
the cutoffs used for the present analysis. Note,
however, that the Canadian DSL and the
chemicals we considered here significantly
overlap. Additionally, pharmaceutical com-
pounds will be included in the future because
of the corresponding wealth of both animal
and human toxicology data.
The TSCA Inventory and Inventory Update
Reporting (IUR). In 1977, the U.S. EPA pub-
lished a rule to assemble an inventory of chemi-
cal substances currently in commerce. This
inventory, commonly referred to as the TSCA
Inventory, is the basis for the U.S. EPA's
Existing Chemicals Program. Starting in 1986,
the Inventory was periodically updated using
the IUR regulation. The TSCA Inventory is
composed of approximately 85,000 chemical
686
VOLUME 117 I NUMBERS I May 2009 • Environmental Health Perspectives
Previous
TOC
-------
The toxicity data landscape for environmental chemicals
substances (U.S. EPA 2004b), including both
substances that are nonconfidential and those
claimed to be confidential business informa-
tion (CBI) under TSCA. Originally, the IUR
was updated on a 4-year cycle, but starting
with the 2006 IUR, it will be updated on a
5-year cycle. The IUR reporting requirements
depend on the volume of the chemical that is
produced as well as certain exemptions. Hence,
the IUR list is a subset of the larger TSCA
inventory. Before 2006, the IUR contained
organic chemicals manufactured or distributed
in the United States in amounts > 10,000 lb/
year. The 2006 IUR regulation requires manu-
facturers and importers of certain chemical
substances to report site and manufacturing
information for chemicals manufactured or
imported in amounts of > 25,000 lb at a single
site. Additional information on domestic pro-
cessing and use must be reported for chemicals
manufactured in amounts of > 300,000 lb at a
single site. The full inventory, including both
confidential and nonconfidential substances,
is maintained by U.S. EPA and Chemical
Abstract Service and is not available to the
public. The nonconfidential or "public" inven-
tory is published periodically, usually after each
IUR cycle. We have included the 2002 version
of the public TSCA inventory in our analy-
ses. This list is available from the U.S. EPA
Substance Registry System (U.S. EPA 2008q).
This list contains 65,513 chemicals indexed by
CASRN. Note that this number differs from
the 75,000 quoted elsewhere because this is the
publicly released list and excludes chemicals
added under the claim of CBI.
HPV chemicals. The U.S. HPV chemi-
cals are those manufactured in or imported
into the United States in amounts > 1 million
Ib/year. The U.S. EPA HPV list is fluid,
changing to some degree with each IUR
cycle. Our current list contains 2,539 chemi-
cals (U.S. EPA 1990). We also include two
important subsets of the HPV list.
U.S. EPA HPV Challenge. The HPV
Challenge Program chemical list consists of all
the HPV chemicals reported during the 1990
IUR reporting year. Inorganic chemicals and
polymers, except in special circumstances, were
not included in the HPV Challenge Program.
Our version of the HPV Challenge list con-
tains 1,973 chemicals (U.S. EPA 1990).
U.S. EPA HPV information system. These
are chemicals with data submitted under the
HPV Challenge Program for which "Robust
Summary" data have been entered into the
U.S. EPA HPV information system (HPVIS;
U.S. EPA 20081). There are 991 chemicals
from HPVIS with information in ACToR.
MPV chemicals. Another set of industrial
chemicals of interest are the non-HPV chemi-
cals included in the TSCA IUR list. These are
the chemicals exceeding a reporting threshold
of 10,000 Ib/year before 2006, and 25,000 lb in
2006 and beyond, but < 1 million Ib/year. The
2002 IUR list contains 5,375 MPV chemi-
cals (U.S. EPA 2004a, 2004b). The updated,
draft 2006 IUR list contains approximately
3,668 MPV chemicals that are not CBI; the
2006 IUR public list will be released by the
U.S. EPA in 2009.
Pesticides and antimicrobials. This
category covers a wide range of substances.
Chemicals regulated as part of the U.S. EPA
pesticide program are generally classified as
"active" or "inert." The active ingredients
are further classified by whether they are tar-
geted at microbes (antimicrobials) or complex
organisms (pesticides). Additionally, all pesti-
cide compounds (conventional actives, anti-
microbials, and inert ingredients) are classified
by whether or not they have food-use toler-
ances or tolerance exemptions. Finally, one
can classify these chemicals by whether or not
they are in use in significant quantities. Here
we rely on the Office of Pesticide Products
Information (OPPIN) system of the U.S.
EPA to extract lists of chemicals. OPPIN is
not publically accessible. From this, we have
drawn the following subsets:
• Conventional Pesticide Actives: (EPA
OPPIN pesticide active): active pesticide
ingredients (834 chemicals)
• Antimicrobial Actives (EPA OPPIN anti-
microbial active): active ingredients used
against microbes (337 chemicals)
• Pesticide inert ingredients: an inert ingre-
dient means any substance, other than
an active ingredient, that is intentionally
included in a pesticide product. Inert ingre-
dients have a number of uses, for instance,
as a solvent, as an aid in increasing the pes-
ticide product's shelf life, or as an agent
to protect the pesticide from degradation
due to exposure to sunlight. We used two
sources: a) U.S. EPA OPPIN inert ingredi-
ents (the complete OPPIN list containing
3,532 chemicals); and V) U.S. EPA inert
nonfood ingredients [a list of inert pesticide
ingredients classified by hazard potential,
not approved for food contact use, avail-
able from the U.S. EPA's Office of Pesticide
Programs (OPP) website (3,492 chemicals)
(U.S. EPA2008n)]
• Pesticide ingredients with food-use toler-
ances or tolerance exemptions (U.S. EPA
OPPIN food use) (1,320 chemicals)
U.S. EPA TRI. The Emergency Planning
and Community Right-to-Know Act of 1986
(EPCRA 1986) requires businesses to report the
locations and quantities of chemicals stored on-
site to state and local governments in order to
help communities prepare to respond to chem-
ical spills and similar emergencies. EPCRA
requires U.S. EPA and the states to annually
collect data on releases and transfers of certain
toxic chemicals from industrial facilities, and
to make the data available to the public in the
TRI. In 1990 Congress passed the Pollution
Prevention Act of 1990 (Pollution Prevention
Act 1990), which requires that additional data
on waste management and source reduction
activities be reported under the TRI. The U.S.
EPA compiles the TRI data each year and
makes these data available through several data
access tools, including their website (U.S. EPA
2008p). Our analysis includes 636 chemicals
from TRI.
Drinking water contaminants. The U.S.
EPA develops drinking water standards and
identifies lists of potential drinking water con-
taminants because they are anticipated to occur
in drinking water supplies and may have adverse
health effects. The lists tracked in the present
analysis are the U.S. EPA's Drinking Water
Standards and Health Advisory Chemicals
(DWSHA; 200 chemicals) and the Candidate
Chemical Lists [CCLs: U.S. EPA CCL1, U.S.
EPA CCL2, and U.S. EPA draft CCL3, which
include 47, 39, and 92 chemicals, respectively
(U.S. EPA 2008e)]. We also included the
Preliminary CCL (PCCL) listing of the 528
chemicals that the U.S. EPA evaluated during
the development of draft CCL3 (U.S. EPA
2008d). The U.S. EPA PCCL was derived from
a collection of approximately 6,000 chemicals
analyzed by the U.S. EPA's Office of Water,
and the PCCL was selected from these 6,000
chemicals based on available health effects and
occurrence data (U.S. EPA 2008c).
U.S. EPA Great Lakes National Program
Office. A set of 429 candidate persistent, bioac-
cumulative toxicants (PBTs) compiled by the
U.S. EPA Great Lakes National Program Office
(GLNPO) are included in the present analysis
(Muir and Howard 2006). These are designated
as U.S. EPA GLNPO PBT chemicals.
U.S. EPA HAPs. This is a list of chemi-
cals that are under review by the U.S. EPA
specified in the Clean Air Act Amendments of
1990. These chemicals include volatile organic
chemicals, chemicals used as pesticides and
herbicides, inorganic chemicals, and radionu-
clides. Many of these chemicals are used for a
variety of purposes in the United States today.
Other chemicals, although not in use today,
were used extensively in the past and may still
be found in the environment. We include a
total of 185 chemicals from this source.
EDSP chemicals. A variety of chemicals
have been found to disrupt the endocrine
systems of animals in laboratory studies, and
compelling evidence shows that endocrine
systems of certain fish and wildlife have been
affected by chemical contaminants, resulting
in developmental and reproductive problems.
Based on this and other evidence, Congress
passed the Food Quality Protection Act of
1996, which requires that the U.S. EPA test
for the potential estrogenic effects in humans.
Subsequently, a U.S. EPA advisory committee
recommended that this be expanded to include
Environmental Health Perspectives • VOLUME 117 I NUMBER 5 I May 2009
Previous
687
-------
Judson et al.
effects occurring via androgen and thyroid
mechanisms and potential for effects on eco-
logic species. We have included the 73 chemi-
cals that were listed to be screened under Tier 1
of the U.S. EPA EDSP (U.S. EPA 2007a).
ToxCast phase I chemicals, ToxCast is a
U.S. EPA program designed to apply HTS,
high-content screening and genomics tech-
niques to the screening and prioritization of
environmental chemicals (Dix et al. 2007).
Phase I of this program is screening 309
unique chemicals, most of which are pesti-
cide active ingredients. (One of the ToxCast
chemicals has no CASRN, so we do not
include it in the analyses below.) This chemi-
cal listing is available for download from the
ToxCast or U.S. EPA Distributed Structure-
Searchable Toxicity Data Network (DSSTox)
websites (U.S. EPA 2008k, 2008o).
Toxicology Reference Database, This is
a collection of summary in vivo toxicology
data, currently focused on pesticide active
ingredients. Data on pesticide actives is col-
lected and summarized from U.S. EPA OPP
data evaluation records (DERs), which are
summaries of guideline studies required
before approval of new pesticide active ingre-
dients. The Toxicology Reference Database
(ToxRefDB) provides the toxicology data
required to link in vitro assays from ToxCast
with in vivo toxicity end points (Martin et al.
2008). ToxRefDB will eventually contain
information on most of the pesticide active
chemicals of ToxCast phase I and will later
expand to include toxicity data on additional
pesticide and nonpesticide chemicals. The
current database contains information on 431
chemicals. In addition to data derived from
pesticide DERs, ToxRefDB will contain data
from other primary in vivo toxicology sources.
U. S. EPA Integrated Risk Information
System, The collection of chemicals subject to
evaluation by the U.S. EPA Integrated Risk
Information System (IRIS) program make up
three major lists: the main U.S. EPA IRIS set
(U.S. EPA 2008f), for which evaluations are
currently available (535 chemicals); the U.S.
EPA IRIS nominations (U.S. EPA 2008g;
currently 20 chemicals nominated for inclu-
sion); and the U.S. EPA IRIS queue (U.S.
EPA 2008h), which are chemicals in queue to
have IRIS reports written (68 chemicals).
Target collection summary. The total
number of chemicals (defined by unique
CASRN) in this set of collections comes to
9,912. Table 1 shows the overlap matrix
between these target chemical lists. The sum
of the number of chemicals in the individual
lists is 23,985. This number drops to 9,912
once we remove overlaps. For instance, 720
chemicals are on the U.S. EPA HPV and on
the U.S. EPA OPPIN inert ingredients lists.
From the U.S. EPA CCL3 list, 29 of 92 are
also HPV chemicals. Interestingly, in a few
cases no overlap occurs between pairs of lists.
Two instances are the lack of overlap between
the U.S. EPA CCL1 and CCL2 lists and the
U.S. EPA GLNPO PBT list.
Information Sources
The information that is available on the target
chemicals can be divided into several assay
categories. The sources for each of these types
of data are available online at http://www.epa.
gov/ncct/toxcast/.
Table 1. Numbers of chemicals that overlap between the screening target chemical collections.
EPACCL1
EPACCL2
EPAdraftCCL3
EPA PCCL
EPA DWSHA
EPA EDSP 73
EPAGLNPOPBT
EPA HAPs
EPA HPV
EPA HPV Challenge
EPA HPVIS
EPA IRIS
EPA IRIS nominations
EPA IRIS queue
EPA IUR (2002|
EPA OPPIN pesticide active
EPA OPPIN antimicrobial active
EPA OPPIN food use
EPA OPPIN inerts
EPA inerts nonfood
EPA TRI
ToxCast phase 1
ToxRefDB
47
39
92
528
200
73
429
185
2,539
1,973
992
535
20
68
5,375
834
337
1,320
3,532
3,492
636
308
431
EPACCL1
47
47
39
13
34
25
6
0
14
11
11
4
31
2
7
14
15
5
14
8
6
27
11
16
EPACCL2
39
39
39
13
28
19
5
0
12
10
10
4
25
1
5
13
13
4
13
7
5
22
10
15
EPAdraftCCL3
92
13
13
92
92
19
9
2
28
29
30
8
56
1
10
39
31
11
41
19
10
60
25
31
EPA PCCL
528
34
28
92
528
62
33
21
77
237
259
91
187
6
27
302
125
52
166
162
135
206
73
93
m
CO
1
200
25
19
19
62
200
29
4
69
61
60
26
176
6
40
77
63
23
69
55
33
130
43
59
00
CO
£
73
6
5
9
33
29
73
1
12
8
11
6
57
1
3
12
64
15
66
15
12
44
56
66
EPAGLNPOPBT
429
0
0
2
21
4
1
429
8
109
75
37
22
2
4
194
3
2
12
43
39
20
4
7
1
185
14
12
28
77
69
12
8
185
92
101
27
144
3
36
122
24
15
43
68
43
173
15
21
1
2,539
11
10
29
237
61
8
109
92
2,539
1,746
701
145
11
34
2,187
102
84
246
720
676
162
13
28
EPA HPV Challenge
1,973
11
10
30
259
60
11
75
101
1,746
1,973
703
147
10
37
1,759
77
60
212
612
567
166
11
25
EPA HPVIS
992
4
4
8
91
26
6
37
27
701
703
992
54
6
12
747
37
34
81
268
250
58
8
15
CO
en
535
31
25
56
187
176
57
22
144
145
147
54
535
10
50
183
179
42
187
115
75
290
122
147
•I
'E
CO
CC
£
20
2
1
1
6
6
1
2
3
11
10
6
10
20
0
13
2
2
4
6
5
9
1
2
O"
CO
CC
68
7
5
10
27
40
3
4
36
34
37
12
50
0
68
46
8
8
10
31
18
46
2
4
CC
£
5,375
14
13
39
302
77
12
194
122
2,187
1,759
747
183
13
46
5,375
151
140
378
1,195
1,126
230
23
47
EPA OPPIN pesticide acti\
834
15
13
31
125
63
64
3
24
102
77
37
179
2
8
151
834
217
484
178
169
175
272
363
CO
EPA OPPIN antimicrobial;
337
5
4
11
52
23
15
2
15
84
60
34
42
2
8
140
217
337
129
155
151
57
33
63
1
O
O
i
1,320
14
13
41
166
69
66
12
43
246
212
81
187
4
10
378
484
129
1,320
744
724
169
239
300
EPA OPPIN inerts
3,532
8
7
19
162
55
15
43
68
720
612
268
115
6
31
1,195
178
155
744
3,532
3,183
136
22
35
EPA inerts nonfood
3,492
6
5
10
135
33
12
39
43
676
567
250
75
5
18
1,126
169
151
724
3,183
3,492
92
15
26
CC
I —
£
636
27
22
60
206
130
44
20
173
162
166
58
290
9
46
230
175
57
169
136
92
636
112
144
CO
CO
to
I —
308
11
10
25
73
43
56
4
15
13
11
8
122
1
2
23
272
33
239
22
15
112
308
304
ToxRefDB
431
16
15
31
93
59
66
7
21
28
25
15
147
2
4
47
363
63
300
35
26
144
304
431
688
Previous
TOC
VOLUME 117 I NUMBERS I May 2009 • Environmental Health Perspectives
-------
The toxicity data landscape for environmental chemicals
Chemical structures. We have compiled
structures for most of the defined compounds
(as opposed to mixtures) in the target lists.
For subsets of chemicals, structures have been
hand curated and quality reviewed as part
of the U.S. EPA DSSTox program (Richard
et al. 2008). We took the remaining structures
from a variety of sources, including PubChem
[National Center for Biotechnology
Information (NCBI) 2008], the National
Cancer Institute's Chemical Structure Lookup
Service (National Cancer Institute 2008),
and the U.S. EPA Substance Registry System
inventory. In many cases, structures were
derived from Simplified Molecular Input Line
Entry Specification (SMILES) codes (Daylight
Chemical Information Systems, Inc. 2008).
At present, we have chemical structures for
7,099 of the 9,912 target chemicals. We lack
structure information for many chemicals
because many substances on these lists are
mixtures, sometimes relatively simple ones
for which representative structures could be
designated (e.g., "sulfuric acid, mono-C]4_]8-
alkyl esters, sodium salts"), and sometimes
very complex mixtures (agar, sesame oil).
Physical-chemical properties. We used
U.S. EPA's EPISuite (U.S. EPA 2007c) set
of programs to calculate physical—chemical
properties for a subset of chemicals. The
input to EPISuite is a list of SMILES codes.
Several EPISuite programs were used includ-
ing KOWWIN [estimates the logarithmic
octanol-water partition coefficient (logP,
also sometimes called log Kov) of organic
compounds (Meylan and Howard 1995)],
MPBPWIN [estimates the boiling point (at
760 mm Hg), melting point, and vapor pres-
sure of organic compounds (Stein and Brown
1994)], WATERNT (estimates the water solu-
bility of organic compounds at 25°C; Meylan
and Howard 1995), and WSKOWWIN
(estimates the water solubility of an organic
compound using the compounds log octanol-
water partition coefficient; Meylan et al.
1996). The properties we use are molecular
weight (MW), logP, boiling point, melting
point, vapor pressure, phase at 25°C, and
molar water solubility. EPISuite reports (and
we) use experimental values when available.
Biochemical (in vitro or cell-based) assay
data. For a subset of the chemicals of inter-
est, in vitro (biochemical) or cell-based assay
data are currently available. This can include
receptor binding, enzyme inhibition, or
cytotoxicity. The major sources of these data
are PubChem and the National Institute of
Mental Health's Psychoactive Drug Screening
Program Kt Database (Roth and Lopez 2008).
In vivo toxicology assay data (tabular). We
derived these data from guideline (or equiva-
lent) toxicology studies from which the pri-
mary or secondary data are available. For our
purposes, the main sources of this primary data
are the National Toxicology Program (NTP),
U.S. EPA OPP (through ToxRefDB; Martin
et al. 2008), the U.S. EPA's HPVIS, and the
FDA. The FDA data we used here came from
the following databases: a) FDA Generally
Recognized as Safe list; b) FDA Cumulative
Estimated Daily Intake/Acceptable Daily
Intake Database; c) FDA Everything Added
to Food in the United States database; and
the d) FDA List of "Indirect" Additives Used
in Food Contact Substances. We compiled
our tabular primary data largely through the
ToxRefDB database (Martin et al. 2008) and
the DSSTox programs (Richard et al. 2006).
HPVIS is a special case because it includes
both primary and secondary data, often pro-
vided in summary by sponsors, with data
derived either from the open literature or
from sponsor-derived study reports. The data-
base captures so-called "Robust Summaries."
Examples of secondary tabular in vivo tox-
icity data are the Carcinogenic Potency
Database (Gold et al. 2001), U.S. EPA IRIS
reports, National Library of Medicine (NLM)
TOXNET databases (Hazardous Substances
Data Bank and Chemical Carcinogenesis
Research Information System), and California
EPA. Data from a number of these secondary
sources have been tabulated and made avail-
able through the DSSTox program (Richard
et al. 2007). Types of tabular information that
are captured in the DSSTox program include
high-level summary results such as food-use
tolerances, LOAELs and NOAELs (lowest
and no observed adverse effect levels), and
reference doses, as well as highly detailed data
such as the per-animal or group-level results of
toxicology studies. Cell-based genotoxicity is
currently captured under this category because
it co-occurs with rodent carcinogenicity data
in current ACToR data sources.
In vivo toxicology text reports via URL,
Much of the publicly available in vivo toxi-
cology data are in the form of narrative
reports from which detailed tabular data may
or may not have been extracted. Examples
are the original NTP, IRIS, and Screening
Information Data Sets (SIDS) reports, the
latter from the Organization for Economic
Cooperation and Development (OECD)
HPV Programme. We also included the
International Agency for Research on Cancer
(IARC) and Agency for Toxic Substances and
Disease Registry (ATSDR) study reports in
this set. These reports contain quantitative and
categorical data, but for most of these sources,
the data provided are not easily extractable.
All of the studies we used here are accessible
via the Web. Information can be extracted
from these reports on a case-by-case basis.
In vivo toxicology summary calls. Several
sources have made definitive calls concerning
particular modes of toxicity, for instance, label-
ing chemicals as being human carcinogens or
developmental toxicants. These calls are made
by experts using data from the detailed toxicity
reports described previously. Although the calls
are subject to debate by experts, they provide
a useful source of data for training prioritiza-
tion models. This information is typically cate-
gorical. Examples of summary calls are cancer
potential determinations of the California EPA
(2008), the NTP Report on Carcinogens (NTP
2008b), NTP Center for the Evaluation of
Risks to Human Reproduction (NTP 2008a),
and the U.S. EPA OPP cancer classifications
(U.S.EPA2007b).
Regulatory listings. By law, the U.S. EPA
and some state agencies maintain a number
of lists of chemicals that are of toxicologic
concern. The presence of a chemical on one of
these lists indicates that toxicity data are avail-
able. For the present analysis, we derived these
lists from the U.S. EPA Substance Registry
System (U.S. EPA2008J).
Phenotypes, Above we have described the
information types of the data rather than the
disease or toxicology categories. Where pos-
sible, assays or data sources have also been
labeled by appropriate disease or toxicology
categories, and we label these categories as
"phenotypes." The set of phenotypes imple-
mented in ACToR span traditional toxicology
study areas. The subset of phenotypes we use
here are general hazard, carcinogenicity, geno-
toxicity, developmental toxicity, reproductive
toxicity, and chronic toxicity. Other toxic-
ity phenotypes are represented in ACToR,
but for small numbers of chemicals. Many
data sources, especially the toxicology sum-
mary reports, contain information on mul-
tiple types of toxicity or end points. In this
category, we have included only IRIS, NTP,
ToxRefDB, and U.S. EPA and OECD HPV
SIDS reports because they can be assumed to
have covered a defined standard set of areas
of toxicity for most chemicals. "Hazard" is
a very broad phenotype category that can
include assays derived from acute and sub-
chronic rodent studies at one end or material
safety data sheets at the other. We further
track information on food safety assessments,
as provided by the FDA (FDA 2006, 2007,
2008). In addition, the U.S. EPA sets food-
use tolerances (or tolerance exemptions) for
a subset of pesticide ingredients. There is a
significant overlap between chemicals regu-
lated by the U.S. EPA and those analyzed by
the FDA. It is obviously of great value to have
both positive and negative toxicity informa-
tion for all of the phenotypes, and both types
were captured where they were available.
Several reviews of the toxicology data land-
scape have described sources of data that are
included in ACToR. Yang et al. (2006a, 2006b)
have recently published two such reviews. In
2001 and 2002, several review papers were
published surveying the landscape of toxicity
Environmental Health Perspectives • VOLUME 117 I NUMBER 5 I May 2009
Previous
689
-------
Judson et al.
data available on the Internet (Brinkhuis 2001;
Felsot 2002; Junghans et al. 2002; Patterson
et al. 2002; Polifka and Faustman 2002; Poore
et al. 2001; Richard and Williams 2003;
Russom 2002; Winter 2002; Wolfgang and
Johnson 2002; Young 2002).
We provide a summary of the sources
of toxicology data we used in this analysis,
available online at http://www.epa.gov/ncct/
toxcast/. In the simplest case, each toxicology
source is a single assay in the ACToR database.
(There are multiple exceptions; e.g., DSSTox
and NTP each contribute multiple assays.) For
each assay, we list the short name, a descrip-
tion, the institutional source, the number of
chemicals covered, the types of information
provided, and a URL. There were 22 sources
from which target screening chemicals were
taken, 47 sources of toxicology data, and
48 lists of chemicals covered by regulations.
Data Collection and
Integration: ACToR
All of the data for this analysis are collected
in the ACToR system (Judson et al. 2008;
U.S. EPA 2008a). The organizing principles
for the design of the chemical/assay system
are largely derived from the PubChem proj-
ect, which captures chemical structure and
HTS information on millions of chemicals
in its role as the main data repository for the
NIH Molecular Libraries Roadmap (Austin
et al. 2004). PubChem characterizes data
in terms of "substances" (the actual chemi-
cal on which one performs an experiment as
defined by the data source), "compounds"
(the idealized structures of chemicals), and
assays (data generated on substances). ACToR
collects these same three main types of data:
substances, indexed by substance identifier
(called the SID); compounds (i.e., chemi-
cal structures) indexed by compound iden-
tifier (called the CID); and assays, indexed
by assay identifier (called the AID). A sub-
stance is a single chemical entity from one
data source and often corresponds to the
physical substance on which some experiment
was performed. A compound is a chemical
entity that corresponds to a unique chemical
structure. Because a substance is defined as
being specific to both data source and experi-
ment, many substances (SIDs) may map to a
single compound (CID). An assay, indexed
by AID, represents a specific type of test data
associated with one or more substances. In
ACToR, a substance is minimally charac-
terized by a data-collection-specific SID
and a chemical name. Most often, the sub-
stance will also have synonyms, a CASRN,
and several other parameters. A compound
always has an associated chemical structure
and a data-collection—specific CID, in addi-
tion to optional parameters derived directly
from chemical structures, such as SMILES
(Daylight Chemical Information Systems, Inc.
2008) and International Chemical Identifier
[International Union of Pure and Applied
Chemistry (IUPAC) 2008)] linear chemi-
cal structure representations and MW. Note
that because ACToR is in essence a "super-
aggregator," pulling in large external data col-
lections, it also stores the source-labeled SIDs
and CIDs from each independent collection
(e.g., PubChem CID, DSSTox CID).
In ACToR, as in DSSTox, data on chemi-
cals across data collections are aggregated
using the concept of a generic chemical.
Because most environmental chemicals, along
with their related toxicity data, are indexed by
CASRN, which can be thought of as a source-
independent test SID, ACToR aggregates
information based on this identifier. A generic
chemical is defined by a CASRN, a preferred
name (typically a common name rather than
an IUPAC or other systematic name), and
an optional ACToR CID. Some sources (in
particular, the FDA and NTP) have provided
CASRN-like identifiers for some compounds,
and these are used in ACToR in place of the
CASRN. All data on all substances sharing
a particular CASRN are attached to the cor-
responding generic chemical. In particular,
a generic chemical will inherit all names
attached to substances with the corresponding
CASRN as synonyms.
In ACToR, an assay is a generic collec-
tion of data values associated with a set of
substances and (potentially) compounds (i.e.,
chemical structures). An assay has a unique
AID, a name, an assay category, and, option-
ally, one or more "phenotypes." Table 2 lists
the assay categories (major types of assays).
Assay phenotypes are linked to high-level
classes of toxicity testing such as carcino-
genicity or reproductive or developmental
toxicology. This allows quick searching of the
database to find all assays that pertain to that
high-level toxicology concept. The concept of
an assay as implemented in ACToR is pur-
posely broad so as to capture any information
potentially relevant to understanding toxicity
and evaluating risk for environmental chemi-
cals. An assay can also have one or more com-
ponents, which are separate data fields that
naturally fall together into an assay (e.g., the
binding constant to a receptor at different
concentrations). Each component is defined
by an assay component identifier, the cor-
responding AID, a name, a description, units
(when applicable), and a data type (float, inte-
ger, categorical, text, Boolean, URL). The
actual data values are called assay results and
are linked to the assay, the assay component,
and the original data-collection—specific sub-
stance. All of the data for an assay can be rep-
resented as a table with one row per chemical
and one column per assay component.
To be included in ACToR, a data source
must meet several criteria: a) data must be
publicly available; b) information sources
must have a significant overlap with chemicals
of interest; c) information must be indexed by
chemical, that is, available on a chemical-by-
chemical basis; and d) information must be
indexed by CASRN (although data are also
included for substances having no assigned
CASRN). We do not require that data be peer
reviewed, although for the analysis we report
Table 2. Categories of assays in ACToR that are described in this analysis.
Assay category
Description
Examples
Physical-chemical
Biochemical
In wVotoxicology (tabular)
In wVotoxicology (study listing primary)
In wVotoxicology (summary calls)
In wVotoxicology (summary report via URL)
Regulatory
Physical and chemical properties (in vitro and/or in silica)
Biochemical (non-cell-based) (in vitro and/or in silica}
Tabulated results from primary or secondary animal-based
studies of chemical effect
Primary studies are available but have not been tabulated
Derived summary determinations of risk
Links to text reports on the Web for which specific data values
are not directly accessible in tabular form
Listings of chemicals that fall under specific environmental laws
or government mandates
LogP
Boiling point
Enzyme inhibition constants
Receptor binding constants
Clinical chemistry
Histopathology
Clinical chemistry
Histopathology
Developmental and reproductive assays
Chemicals determined to pose a
defined risk of human cancer
Reports from U.S. EPA IRIS or NTP
TSCA
690
VOLUME 117 I NUMBERS I May 2009 • Environmental Health Perspectives
Previous
TOC
-------
The toxicity data landscape for environmental chemicals
here, most of the data sources either have been
externally peer reviewed or, when from gov-
ernment agencies, have undergone extensive
internal review. Data entered into ACToR
undergo a limited quality control process.
Data are preferably taken from sources of
high-quality data, so our quality control is
limited to checking that the data are correctly
transferred from the source via a reformatting
and loading process into the ACToR data-
base. No checks are made on the correctness
of the data from the original source. Each data
set is manually spot-checked for gross issues
with reformatting. All CASRNs in the data-
base are checked to be sure that they have a
proper checksum (Chemical Abstracts Service
2008). (The checksum is the result of a par-
ticular formula performed on all but the final
digit of the CASRN. This result must match
the final digit.) All data-handling tasks are
documented in standard operating procedures
to ensure consistency.
The ACToR database is implemented using
MySQL. Software to preprocess and load data
is written in Perl, and the Web interfaces are
written in Java. The use of 100% open-source
software allows the entire system to be easily
distributed to other interested groups. We used
the ACToR database version 2008Q2d for
all of the analyses in this article. Of the sub-
sets of data sources in ACToR, only the ones
most relevant to toxicology are included in this
analysis and publication. ACToR is available
online (http://actor.epa.gov).
Results
In vivo toxicology data. This section describes
the overlap between the target chemical set
and the set of toxicity data sources. Table 3
summarizes the overlap matrix. Each cell
provides the number and the percentage of
the 9,912 for the chemicals that have infor-
mation for a specific category of data (e.g.,
tabular) and a particular phenotype (e.g.,
carcinogenicity). The last column gives the
number and percentage of chemicals for each
phenotype, regardless of the information cate-
gory. Chemicals are only counted once in any
cell, even if they have multiple data points or
sources of data. Cells that list 0 indicate that
there were no corresponding data from any
source. The available toxicity data almost all
derive from animal studies, because essentially
no experimental human data are available.
However, some of the data are in the form
of human reference doses, or summary calls
of the form "this chemical is considered to
be a human carcinogen." These data points
were, of course, derived by extrapolating from
primary data on animals. Chemical hazard
has been evaluated for 5,810 (58.6%) of these
chemicals. Carcinogenicity potential for 2,579
(26%) of these chemicals has been evaluated
by at least one source. The genotoxic poten-
tial of 2,724 (27.5%) of the chemicals has
been evaluated. A total of 2,862 (28.9%) of
the chemicals have their developmental toxic-
ity reported, and 1,081 (10.9%) have repro-
ductive toxicity data reported. Food safety
information (from one of the sources men-
tioned above) is available for 2,258 (22.8%)
of the chemicals. Chemicals count in this table
whether they have positive or negative data for
toxicity for a particular phenotype. To date,
we have not systematically tabulated the rela-
tive number of toxic and nontoxic indications
for all chemicals.
Table 4 provides overlaps of the chemicals
of interest with more general information and
biological assays of potential interest. One or
more in vitro biochemical assays are available
for 781 (7.9%) of the chemicals. Most of these
are in vitro cytotoxicity assays in PubChem,
but also include receptor binding and enzyme
inhibition data. A small number of the target
chemicals (234 or 2.4%) are naturally occur-
ring human metabolites, based on data from
the Human Metabolome Database (Wishart
et al. 2007).
The highest-quality toxicity assessments,
based on guideline studies or on extensive
review of the literature, are U.S. EPA OPP
reviews (which are captured in the ToxRefDB
database), U.S. EPA IRIS assessments, NTP
studies, OECD SIDS guideline studies of HPV
chemicals, studies in the U.S. EPA HPVIS,
and assessments by the ATSDR and I ARC.
From the current list, there are 431 (4.3%),
536 (5.4%), 1,168 (11.8%), 343 (3.5%),
992 (10%), 216 (2.2%), and 537 (5.4%)
TableS. Summary of overlap between the target chemical list and the set of assay components.
Assay
Tabular
Primary
study listing
Summary calls
Summary
report via URL
Any
Hazard
Carcinogenicity
Genotoxicity
Developmental toxicity
Reproductive toxicity
Food safety
4,454 (44.9)
1,211 (12.2)
2,496(25.2)
755(7.6)
734(7.4)
1,692(17.1)
0
401 (4.0)
1,102(11.1)
37(0.4)
0
0
255(2.6)
726 (7.3)
32 (0.3)
125(1.3)
31 (0.3)
533 (5.4)
4,767(48.1)
2,035(23.3)
1,047(10.6)
2,324(23.4)
396(4)
0
5,810(58.6)
2,579 (26)
2,724 (27.5)
2,862 (28.9)
1,081(10.9)
2,258 (22.8)
chemicals in these respective sets (Table 4).
Looking across all of these data sources, 2,767
(27.9%) are covered by one or more of these
high-quality toxicology sources. Finally, a total
of 4,641 (46.8%) are currently subject to one
or more U.S. EPA regulations. These regula-
tions are available online (http://www.epa.gov/
ncct/toxcast/).
Chemical categories. Both the U.S. HPV
Challenge and the OECD HPV programs
encourage the use of categories because of
the large number of chemicals being assessed.
Using a category approach, chemicals are evalu-
ated as a group, or category, rather than as indi-
vidual chemicals, and not every chemical needs
to be tested for every end point. The category
approach entails grouping chemicals with simi-
lar structures, physical-chemical properties, fate
parameters, and toxicologic properties in order
to extrapolate toxicologic information from
tested chemicals and end points to untested
chemicals and end points. For most categories,
the number of chemicals with toxicology data
that could be used for model building is much
smaller than the total number of chemicals
included within the category.
ACToR includes listings of chemical cat-
egories taken from the U.S. EPA HPVIS and
from the OECD HPV Programme. From these
lists, a total of 1,274 (12.9%) chemicals are in
at least one category, and there are 256 unique
categories that include at least one of the target
chemicals. However, most of the categories in
HPVIS represent "proposals," which are cur-
rently under review by the U.S. EPA, such that
the final number of categories and chemicals
assigned to them is subject to change. In addi-
tion, the U.S. EPA is currently using chemical
clustering techniques with the goal of creating
chemical categories to facilitate hazard assess-
ment of MPV chemicals. The outcome of these
efforts will be included in ACToR in the future.
Information will also flow in the opposite direc-
tion; that is, the data and information included
in ACToR will be useful in reviewing and refin-
ing the U.S. EPA's HPV and MPV categories.
Production volumes. An important com-
ponent of any prioritization program will be
Table 4. Coverage by specific data types and
sources.
Each cell provides the number and the percentage of the 9,912 for the chemicals that have information for a specific
category of data (e.g., tabular) and a particular phenotype (e.g., carcinogenicity). The last column gives the number and
percentage of chemicals for each phenotype, regardless of the information category. Chemicals are only counted once in
any cell, even if they have multiple data points or sources of data. Cells with 0 indicate that there were no corresponding
data from any source.
Name
Biochemical
Human-metabolite
ToxRefDB
IRIS
NTP
SIDS
HPVIS
ATSDR
IARC
ToxRefDB, IRIS, NTP,
SIDS, ATSDR,
and/or IARC
Regulation
Total
781
234
431
536
1,168
343
992
216
537
2,767
4,641
Percent coverage
7.9
2.4
4.3
5.4
11.8
3.5
10.0
2.2
5.4
27.9
46.8
Environmental Health Perspectives • VOLUME 117 I NUMBER 5 I May 2009
Previous
691
-------
Judson et al.
an assessment of potential for exposure. In
the absence of specific information and for
screening and prioritization purposes, produc-
tion volumes are often used as a surrogate for
exposure potential. Table 5 lists counts for
each of the production volume categories. A
total of 5,939 (59.9%) of the target chemicals
have production volume information in the
2002 IUR.
Properties related to chemical structure,
Physical-chemical properties were calculated
using the EPISuite collection of programs,
which use chemical structure (in the form of a
SMILES string) as input. Of the 7,099 chemi-
cals for which structures and SMILES data
were available, EPISuite was able to process
5,857. The chemicals for which calculations
could not be performed were mainly certain
types of salts, inorganic compounds, organo-
metallics, or chemicals with nonstandard
SMILES.
Several parameters will be useful for deter-
mining whether a compound can be bioavail-
able or whether it will be amenable to HTS
assays: MW, logP, solubility, and vapor pres-
sure. Typical ranges for properties for chemi-
cals that can be tested using HTS methods are
MW < 500 Da, logP between 0 and 6, and
vapor pressure < 10 mm (not volatile at room
temperature). Filtering the larger list against
this set of criteria yields a set of 3,060 com-
pounds that are candidates for HTS testing.
One could produce slightly different lists, of
course, by altering these threshold values. The
primary requirements for use in an HTS assay
are that chemicals be soluble in dimethyl sul-
foxide or water, that they be nonvolatile, and
that they be stable in solution.
Figures 1 and 2 show distributions of
MW and logP for the complete set of chemi-
cals with structures and for four representative
subsets of the larger data collection: HPV
chemicals, pesticide inert ingredients, pesti-
cide active ingredients, and the ToxCast phase
I collection. For MW, the main trend is that
the HPV and pesticide inert collections con-
tain significantly larger fractions of low-MW
chemicals (< 200 Da) than do the pesticide
active ingredients and the ToxCast chemi-
cals. Given that most ToxCast phase I chemi-
cals are pesticide active ingredients and that
Table 5. Production volumes from the 2002 IUR.
Production volume (Ib/year)
<10K
10K-500K
>500K-1M
>1M-10M
>10M-50M
>50M-100M
>100M-500M
>500M-1B
>1B
Total
Count
11
2,827
485
1,381
512
130
246
67
280
5,939
Percent coverage
0.1
29.0
4.9
14.0
5.0
1.0
2.0
0.7
3.0
60.0
this set was prefiltered for HTS suitability,
it is not surprising that this set has a smaller
fraction of high-MW chemicals (> 500 Da)
than do the other collections. Distributions
of logP are similar for all of the subsets except
for ToxCast, which is more tightly clustered,
with a peak between 0 and 2.
Discussion
In this article we describe and analyze a com-
pilation of chemical structures, physical-
chemical properties, in vitro biochemical assay
data, and in vivo toxicology data on a large
collection of chemicals of interest to the U.S.
EPA. Most of these data are currently pub-
licly available but have not been organized
previously in a unified manner that allows
for the analysis of large trends and simplified
review based on either chemical or assay axes.
The data we describe here are a subset of those
contained in the ACToR system being devel-
oped at the U.S. EPA to manage large collec-
tions of data on environmental chemicals.
We have used the ACToR database to
characterize the state of toxicologic knowl-
edge on a subset of environmental chemicals
that are on a variety of lists of interest to the
U.S. EPA. This analysis is used to address the
extent of the perceived data gap on potentially
toxic chemicals. Although the picture is com-
plicated, some summary observations are pos-
sible. About two-thirds of the chemicals have
some toxicology information. The unique set
of chemicals in Table 3 is 6,551 of 9,912
(66%). The alternative view is that many of
these chemicals remain largely uncharacter-
ized—a total of 3,361 (34%) chemicals have
no information in any of the data sources we
used in this analysis. On the other hand, more
than one-quarter (27.9%) have been analyzed
in one or more high-quality and/or systematic
evaluation programs (NTP, IRIS, ToxRefDB,
U.S. EPA HPV, OECD SIDS, IARC, and/
or ATSDR). Of the individual types of toxic-
ity (or end points) that have been tabulated,
carcinogenicity, genotoxicity, and develop-
mental and reproductive toxicity have been
most widely covered (26%, 27.5%, 28.9%,
and 10.9%, respectively).
One immediate application of this analysis
is to select compounds for further screening in
programs such as ToxCast. ToxCast phase I
is using a set of compounds (primarily pesti-
cide active ingredients) that are amenable to
HTS and that have rich toxicologic data. The
outcome of the phase I analyses will be a set of
"signatures" that use in vitro screening data as
inputs to predict in vivo toxicology phenotypes
with high enough sensitivity and specificity to
be useful for prioritization for more detailed
testing. Phase II needs to include compounds
that can be used to independently validate the
phase I signatures. Therefore, the phase II set of
chemicals should contain as many compounds
as possible with high-quality in vivo toxicol-
ogy data, have physical—chemical properties
that make them candidates for HTS, and be
drawn from a more diverse collection than the
phase I chemicals to help define the chemi-
cal domain of applicability of the signatures.
We calculated the intersection of the set of
2,767 chemicals that have data from one of
the high-quality and/or systematic toxicol-
ogy data sources (NTP, IRIS, HPVIS, OPP/
ToxRefDB, OECD SIDS, IARC, ATSDR)
with the set of 3,060 chemicals with reasonable
physicochemical properties. This yields a list of
1,308 candidate chemicals that have both high-
quality toxicity data and physicochemical prop-
erties very well suited for HTS. After removing
the ToxCast phase I chemicals, we arrived at
a list of 1,046 chemicals that are candidates
for inclusion in ToxCast phase II for use in
validating ToxCast phase I findings across a
0.45
0.40
0.35
I 0.30
0
r*
| 0.25
O
= 0.20
o
o
= 0.15
0.10
0.05
n
r-
n ALL
n HPV
_
L
1 .
~L
• Pesticide inerts
• Pesticide actives
m
• ToxCast
4
1 R^L n^ n^ ^^_ ^m
Abbreviations: B, billion; K, thousand; M, million.
MW
Figure 1. Distribution of MW for representative chemical sets. The sum of fractions for each data set equals 1.
692
VOLUME 117 I NUMBERS I May 2009 • Environmental Health Perspectives
Previous
-------
The toxicity data landscape for environmental chemicals
variety of end points. Many of these chemicals
are currently being analyzed in a series of HTS
assays at the NIH Chemical Genomic Center
(NCGC) as part of the Tox21 partnership
between U.S. EPA, NCGC, and NTP. These
Tox21 chemicals include an even broader
range of physicochemical properties, with a
MW range of 32 to 1,255 and a logP range
of-13.2 to 13.2. An important analysis that
is yet to be carried out is chemical structure
characterization and clustering for the ToxCast
phase I and II lists and the larger target list.
This will be important to help understand our
ability to extrapolate within and across chemical
structural classes.
ACToR is not alone in its goal of aggregat-
ing large sets of chemical structure and assay
data but is distinguished from other efforts
by its focus on toxicology and environmen-
tal chemicals and its goal of facilitating com-
putational analysis. PubChem (NCBI 2008)
is the largest effort currently available, with
information on more than 10 million unique
chemical compounds. ChemSpider (2008) is
an even larger chemical aggregation project but
does not house biological data or download-
able data sets. Another important compari-
son is with TOXNET, which is a collection of
multiple data sources covering many aspects of
chemical toxicity. TOXNET has a common
search engine that allows the user to easily find
data from multiple sources. However, it is a
closed system that does not allow a user to pull
together data sets that are useful for compu-
tational purposes. One unique aspect of the
ACToR system is that it aggregates the data
from PubChem (focused on chemical structure
and HTS in vitro assay data) and TOXNET
(NLM 2008) (focused on in vivo toxicology
data) and combines it in a way that it can be
used for computational analysis. eChemPortal
(OECD 2008) is an OECD effort very similar
0.20
0.15
to ACToR. It mainly aggregates information
on HPV chemicals and pesticides. eChem
Portal currently contains links to seven large
database systems, some of which contain what
in ACToR are multiple individual databases
(e.g., INCHEM contains 11 individual data-
bases; International Programme on Chemical
Safety 2008). Unlike eChemPortal, which pro-
vides links to Web pages for the component
databases, ACToR extracts tabular data from
a large number of sources and makes it search-
able by name, CASRN, or chemical structure.
A system called Vitic is being developed by
Lhasa Limited in collaboration between the
European Chemicals Agency's International
Uniform Chemical Information Database
(IUCLID 2008) project and a number of phar-
maceutical companies, with the goal of being
an international toxicology information center
(Judson et al. 2005). In addition, the European
Substances Information System provides links
to a number of databases, including U.S. EPA
HPV, IUCLID, and European Inventory of
Existing Commercial Chemical Substances.
Finally, the Chemical Effects in Biological
Systems project at the National Institute of
Environmental Health Sciences is constructing
a multidomain information repository to hold
the detailed results and summaries of in vivo
and in vitro toxicology experiments from NTP
studies, with particular emphasis on toxico-
genomics and microarray experiments (Waters
et al. 2008).
To adequately characterize the toxicology
of all environmental chemicals of potential
concern, we still face significant challenges.
Screening and prioritization approaches such as
ToxCast can make significant headway in ana-
lyzing small organic and organometallic com-
pounds, for which most HTS methods have
been developed for use in the pharmaceutical
industry. Because of solubility and volatility
0.05
Figure 2. Distribution of calculated logP for representative chemical sets. The sum of fractions for each
data set equals 1.
issues, however, many exceptionally high- and
low-MW environmental compounds or highly
lipophilic compounds may require new screen-
ing methods. Of special interest are nano-
materials, which will require new standards for
description (i.e., size, shape, composition, etc.)
and may require entirely new approaches to
thinking about cellular and organism-level tox-
icity (Maynard et al. 2006; Shaw et al. 2008).
One rarely has knowledge of metabolites that
can arise from a parent compound in vivo and
whether any of these metabolites are more or
less toxic than the parent. However, a number
of metabolic pathway databases and/or simula-
tors are currently available or under develop-
ment that could potentially be incorporated
into ACToR in the future. Finally, a large
number of known biological pathways (i.e.,
signaling, metabolism, etc.) have the potential
to lead to toxicity when significantly perturbed.
Many toxicity pathways have been implicated
in whole-animal end points, such as liver can-
cer, and most chemicals can perturb multiple
candidate toxicity pathways. Gaining a predic-
tive and mechanistic understanding of chemi-
cal toxicity will require the ability to predict
which set of toxicity pathways are triggered by
individual chemicals.
A significant amount of data on chemicals
is not currently accessible for modeling, either
because it is not publicly available or because
it is not yet extracted from primary reports
in a useful, tabular format. Several efforts are
under way at the U.S. EPA and other institu-
tions to extract, standardize, compile, and ana-
lyze such high-quality data (U.S. EPA 2008b).
We would welcome collaborations with other
groups producing such tabular data sets on
these important classes of chemicals.
Conclusions
In this article, we have described a process for
determining a set of environmental chemi-
cals with the highest need for hazard and risk
evaluation, which is based primarily on objec-
tive, simple measures of data availability. In
addition, we have collected information from
a large number of publicly available sources
to determine the state of our current knowl-
edge of these chemicals. The list we developed
includes HPV and MPV chemicals, pesticide
and antimicrobial active and inert ingredi-
ents, and potential air and drinking water
pollutants, in addition to chemicals already
being evaluated by the U.S. EPA IRIS and
ToxCast programs. Although the input lists
are developed from the perspective of regula-
tory and research needs of the U.S. EPA, we
believe that our overall conclusions will have
wide applicability. This process resulted in
a collection of 9,912 unique chemicals. We
have at least limited hazard information on
approximately two-thirds of these and detailed
toxicology information on approximately
Environmental Health Perspectives • VOLUME 117 I NUMBER 5 I May 2009
Previous
TOC
693
-------
Judson et al.
one-quarter. The combination of chemical
structure and in vivo data on this large range
of environmental chemicals in ACToR can
facilitate structure-activity relationship and
other types of trend analyses. These analyses
will have direct relevance to U.S. EPA pro-
grams such as HPV Challenge (U.S. EPA
2008m) and the Chemical Assessment and
Management Program (U.S. EPA 2008b).
The principal reason for the lack of more
complete toxicity information is the extremely
high cost for full evaluation using standard
guideline animal studies, which is millions of
dollars per chemical. This has prompted the
call for the use of more cost-effective HTS
methods for quickly screening and prioritiz-
ing chemicals for more detailed testing. The
analysis presented here is a first step in such
a screening and prioritization process being
carried out at the U.S. EPA as part of the
ToxCast program. ToxCast is using hun-
dreds of in vitro HTS assays to assess poten-
tial mechanisms through which chemicals
could cause toxicity. This hazard prediction is
just one of several axes along which potential
risk needs to be evaluated. Chemicals need
to be evaluated for exposure potential, and
for adsorption, distribution, metabolism,
and excretion (ADME) and pharmacokinet-
ics properties. Of special concern would be
compounds that are persistent or bioaccu-
mulative. Researchers at Health Canada have
demonstrated a process to evaluate exposure
for many of these chemicals (Health Canada
2006). Chemical structure analysis can be
used as part of the prioritization process,
for instance, in predicting bioaccumulation
potential (Meylan et al. 1999; Weisbrod
et al. 2007) and fractional absorption (Ekins
et al. 2007a, 2007b). Nonanimal experimen-
tal methods are available to approximate gut
absorption (Sun et al. 2008) and total hepatic
clearance (Naritomi et al. 2003). Reverse-
pharmacokinetic methods (Brightman et al.
2006) can be used to predict oral doses that
would be required to trigger molecular pro-
cesses, for instance, based on half maximal
inhibitory concentrations (ICjo) for recep-
tor binding from in vitro assays. These and
other related approaches are being considered
as part of the overall ToxCast screening and
prioritization process. Of special relevance to
the ToxCast program, we have identified a set
of 1,046 candidate chemicals that have reli-
able in vivo toxicology data and have physico-
chemical properties that make them suitable
for in vitro HTS analysis. These are candidates
for phase II of the ToxCast program, which
will be used to validate in vitro-to-in vivo tox-
icity predictions, which are one outcome of
phase I of this program.
Another important input to this process is
high-quality, tabular in vivo toxicity data. This
is required to anchor our in vitro-to-in vivo
prediction models, in both the model build-
ing and model validation phases. Initially,
we are making use of the results of guideline
toxicology studies for pesticide active ingredi-
ents, which are being collected into the U.S.
EPA ToxRefDB (Martin et al. 2008). We are
expanding this data collation effort in coordi-
nation with the ACToR project. As already
described, ACToR is a database consisting
of information on environmental chemicals
from a wide number of sources. However,
currently much of the high-quality toxicology
data indexed in ACToR still resides in text
reports and remains to be manually extracted
into tabular form.
An important aspect of this program is
openness and transparency. The ToxCast
program is making all of its data publicly
available. It has a large community of col-
laborators, from government labs, compa-
nies, and universities. Finally, important open
venues for learning about this program and
the Chemical Prioritization and Exposure
Communities of Practice are providing
input (U.S. EPA 20081). These are bringing
together representatives from U.S. EPA, state,
and other national environmental regulatory
organizations, academic labs, stakeholder
companies, and public interest groups, all of
whom are providing important input as we
collectively work to address this important
problem. All of these efforts are consistent
with achieving the goals and vision of the
recent NRC report Toxicity Testing in the 21st
Century (NRC 2007).
REFERENCES
Allanou R, Hansen B, van det Bill Y. 1999. Public Availability of
Data on EU High Production Volume Chemicals. Available:
http://ecb.jrc.it/documents/Existing-Chemicals/PU BLIC_
AVAILABILITY_OF_DATA/[accessed 8 August 2008].
Applegate J, Baer K. 2006. Strategies for Closing the Data Gap.
Available: http://www.progressivereform.org/articles/
Closing_Data_Gaps_602.pdf [accessed 8 August 2008].
Austin CP, Brady LS, Insel TR, Collins FS. 2004. NIH Molecular
Libraries Initiative. Science 306(56991:1138-1139.
Barton HA, Pastoor TP, Baetcke K, Chambers JE, Diliberto J,
Doerrer NG, et al. 2006. The acquisition and application
of absorption, distribution, metabolism, and excretion
(ADME) data in agricultural chemical safety assessments.
Crit Rev Toxicol 36(1]:9-35.
Birnbaum LS, Staskal DF, Diliberto JJ. 2003. Health effects of
polybrominated dibenzo-p-dioxins (PBDDs) and dibenzo-
furans(PBDFs). Environ Int29(61:855-860.
Brightman FA, Leahy DE, Searle GE, Thomas S. 2006.
Application of a generic physiologically based pharmaco-
kinetic model to the estimation of xenobiotic levels in rat
plasma. Drug Metab Dispos 34(11:84-93.
Brinkhuis RP. 2001. Toxicology information from US govern-
ment agencies. Toxicology 157(1-21:25-49.
California EPA. 2008. Chemicals Known to the State of California
to Cause Cancer. Available: http://www.oehha.ca.gov/
prop65/prop65_list/Newlist.html [accessed 8 August2008].
Carmichael NG, Barton HA, Boobis AR, Cooper RL, Dellarco VL,
Doerrer NG, et al. 2006. Agricultural chemical safety assess-
ment: a multisector approach to the modernization of human
safety requirements. Crit Rev Toxicol 36(1):1-7.
Chemical Abstracts Service. 2008. Check Digit Verification of
CAS Registry Numbers. Available: http://www.cas.org/
expertise/cascontent/registry/checkdig.html [accessed
8 August 2008].
ChemSpider. 2008. ChemSpider: Building a Structure Centric
Community for Chemists. Available: http://www.chemspider.
com/ [accessed 8 August 2008].
Clean Air Act Amendments of 1990.1990. Public Law 101-549.
Collins FS, Gray GM, Bucher JR. 2008. Toxicology. Transforming
environmental health protection. Science 319(58651:906-907.
Daylight Chemical Information Systems, Inc. 2008. SMILES
(Simplified Molecular Input Line Entry System). Available:
http://www.daylight.com/dayhtml/doc/theory/theory.
smiles.html [accessed 24 October 2008].
Dix DJ, Houck KA, Martin MT, Richard AM, Setzer RW, Kavlock RJ.
2007. The ToxCast program for prioritizing toxicity testing of
environmental chemicals. Toxicol Sci95(1):5-12.
Ekins S, Mestres J, Testa B. 2007a. In silico pharmacology
for drug discovery: applications to targets and beyond.
Br J Pharmacol 152(11:21-37.
Ekins S, Mestres J, Testa B. 2007b. In silico pharmacology for
drug discovery: methods for virtual ligand screening and
profiling. Br J Pharmacol 152(11:9-20.
EPCRA. 1986. Emergency Planning and Community Right-to-
Know Act of 1986. 42 USC §11001 et seq. Public Law 103-337.
FDA (Food and Drug Administration). 2006. Select Committee on
GRAS Substances (SCOGS) Database Overview. Available:
http://www.cfsan.fda.gov/~dms/opascogs.html [accessed
8 August 2008].
FDA (Food and Drug Administration). 2007. Cumulative
Estimated Daily Intake/Acceptable Daily Intake Database.
Available: http://www.cfsan.fda.gov/~dms/opa-edi.html
[accessed 8 August 2008].
FDA (Food and Drug Administration). 2008. EAFUS: A Food
Additive Database. Available: http://vm.cfsan.fda.
gov/~dms/eafus.html [accessed 8 August 2008].
Felsot AS. 2002. Web resources for pesticide toxicology,
environmental chemistry, and policy: a utilitarian per-
spective. Toxicology 173(1-21:153-166.
Food Quality Protection Act of 1996.1996. Public Law 104-170.
Gold LS, Manley NB, Slone TH, Ward JM. 2001. Compendium of
chemical carcinogens by target organ: results of chronic
bioassays in rats, mice, hamsters, dogs, and monkeys.
Toxicol Pathol 29(61:639-652.
Guth J, Denison R, Saas J. 2005. Background Paper for Reform
No. 5 of the Louisville Charter for Safer Chemicals:
Require Comprehensive Safety Data for All Chemicals.
Available: http://www.louisvillecharter.org/downloads/
CharterBkgrdPaper5.pdf [accessed 8 August 2008].
Health Canada. 2006. Categorization of Substances on the
Domestic Substances List. Available: http://www.hc-sc.
gc.ca/ewh-semVcontaminants/existsub/categor/index-eng.
php [accessed 1 August 2008].
IPCS (International Programme on Chemical Safety).
2008. INCHEM—Chemical Safety Information from
Intergovernmental Organizations. Available: http://www.
inchem.org/ [accessed 8 August 2008].
IUCLID. 2008. IUCLID —International Uniform Chemical
Information Database. Available: http://ecbwbiu5.jrc.it/
[accessed 8 August 2008].
IUPAC (International Union of Pure and Applied Chemistry).
2008. IUPAC—International Chemical Identifier. Available:
http://old.iupac.org/projects/2000/2000-025-1-800.html
[accessed 24 October 2008].
Judson PN, Cooke PA, Doerrer NG, Greene N, Hanzlik RP, Hardy
C, et al. 2005. Towards the creation of an international toxi-
cology information centre. Toxicology 213(1-2):117-128.
Judson R, Richard A, Dix D, Houck K, Elloumi F, Martin M,
et al. 2008. ACToR—Aggregated Computational Toxicology
Resource. Toxicol Appl Pharmacol 233:7-13.
Junghans TB, Sevin IF, lonin B, Seifried H. 2002. Cancer infor-
mation resources: digital and online sources. Toxicology
173(1-21:13-34.
Martin MT, Judson RS, Reif DM, Kavlock RJ, Dix DJ. 2009.
Profiling chemicals based on chronic toxicity results from
the U.S. EPA ToxRef Database. Environ Health Perspect
117:392-399.
Maynard AD, Aitken RJ, Butz T, Colvin V, Donaldson K,
Oberdorster G, et al. 2006. Safe handling of nanotechnol-
ogy. Nature 444(71171:267-269.
Meylan W, Howard P, Boethling R. 1996. Improved method for
estimating water solubility from octanol/water partition
coefficient. Environ Toxicol Chem 15:100-106.
Meylan W, Howard P, Boethling R. 1999. Improved method for
estimating bioconcentration/bioaccumulation factor from
octanol/water partition coefficient. Environ Toxicol Chem
18(41:664-672.
Meylan WM, Howard PH. 1995. Atom/fragment contribution
694
VOLUME 117 I NUMBERS I May 2009 • Environmental Health Perspectives
Previous
-------
The toxicity data landscape for environmental chemicals
method for estimating octanol-water partition coefficients.
J PharmSci 84(11:83-92.
Muir DC, Howard PH. 2006. Are there other persistent organic
pollutants? A challenge for environmental chemists.
Environ Sci Technol 40(231:7157-7166.
Naritomi Y, Terashita S, Kagayama A, Sugiyama Y. 2003. Utility
of hepatocytes in predicting drug metabolism: comparison
of hepatic intrinsic clearance in rats and humans in vivo
and in vitro. Drug Metab Dispos 31(51:580-588.
National Cancer Institute. 2008. Chemical Structure Lookup
Service. Available: http://cactus.nci.nih.gov/cgi-bin/
lookup/search [accessed 8 August 2008].
NCBI (National Center for Biotechnology Information). 2008.
PubChem. Available: http://pubchem.ncbi.nlm.nih.gov/
[accessed 8 August 2008].
NLM (National Library of Medicine). 2008. TOXNET—
Toxicology Data Network. Available: http://toxnet.nlm.nih.
gov/[accessed 8 August 2008].
NRC (National Research Council). 2007. Toxicity Testing in
the 21st Century: A Vision and a Strategy. Washington,
DC:National Academies Press.
NTP (National Toxicology Program). 2008a. Center for the
Evaluation of Risks to Human Reproduction (CERHR).
Available: http://cerhr.niehs.nih.gov/chemicals/index.html
[accessed 1 August 2008].
NTP (National Toxicology Program). 2008b. Report on
Carcinogens (RoC). Available: http://ntp.niehs.nih.
gov/index.cfm?objectid = 72016262-BDB7-CEBA-
FA60E922B18C2540 [accessed 8 August 2008].
OECD. 2008. eChemPortal—the Global Portal to Information on
Chemical Substances. Available: http://webnet3.oecd.org/
echemportal/ [accessed 8 August 2008].
Patterson J, Hakkinen PJ, Wullenweber AE. 2002. Human
health risk assessment: selected Internet and World Wide
Web resources. Toxicology 173(1-21:123-143.
Polifka JE, Faustman EM. 2002. Developmental toxicity: web
resources for evaluating risk in humans. Toxicology
173(1-21:35-65.
Pollution Prevention Act of 1990.1990. Public Law 106-40.
Poore LM, King G, Stefanik K. 2001. Toxicology information
resources at the Environmental Protection Agency.
Toxicology 157(1-21:11-23.
REACH (Registration, Evaluation, and Authorization of
Chemicals). 2008. European Chemical Agency—List of
Pre-registered Substances. Available: http://apps.echa.
europa.eu/preregistered/pre-registered-sub.aspx
[accessed 27 October 2008].
Richard A, Judson R, Yang C. 2008. Toxicity data informatics:
supporting a new paradigm for toxicity prediction. Toxicol
Mech Methods 18:103-108.
Richard A, Williams C. 2003. Public sources of mutagenicity
and carcinogenicity data: use in structure-activity rela-
tionship models. In: QSARS of Mutagens and Carcinogens
(Benigni R, ed). NewYork:CRC Press, 145-173.
Richard AM, Gold LS, Nicklaus MC. 2006. Chemical structure
indexing of toxicity data on the Internet: moving toward a
flat world. CurrOpin Drug Discov Devel 9(31:314-325.
Richard AM, Wolf MA, Burch J. 2007. DSSTox EPA Integrated
Risk Information System (IRIS) Toxicity Review Data: SDF
File and Documentation. Available: www.epa.gov/ncct/
dsstox/ [accessed 2008 8 August].
Roth B, Lopez E. 2008. PDSP Ki Database. Available: http://pdsp.
med.unc.edu/kidb.php [accessed 8 August 2008].
Russom CL. 2002. Mining environmental toxicology information:
Web resources. Toxicology 173(1-2):75-88.
Shaw SY, Westly EC, Pittet MJ, Subramanian A, Schreiber SL,
Weissleder R. 2008. Perturbational profiling of nanomaterial
biologic activity. Proc Natl Acad Sci USA 105(211:7387-7392.
Stein S, Brown R. 1994. Estimation of normal boiling points from
group contributions. J Chem Inf Comput Sci 34:1242-1250.
Sun H, Chow EC, Liu S, Du Y, Pang KS. 2008. The Caco-2 cell
monolayer: usefulness and limitations. Expert Opin Drug
Metab Toxicol 4(41:395-411.
TSCA (Toxic Substances Control Act of 1976). 1976. 15 USC
§2601 etseq. Public Law 94-469.
U.S. EPA (Environmental Protection Agency). 1990. The HPV
Voluntary Challenge Chemical List. Available: http://
www.epa.gov/hpv/pubs/update/hpvchmlt.htm [accessed
8 August 2008].
U.S. EPA (Environmental Protection Agency). 1998. Chemical
Hazard Data Availability Study. Available: http://www.epa.
gov/hpv/pubs/general/hazchem.pdf [accessed 8 August
2008].
U.S. EPA (Environmental Protection Agency). 2004a. Inventory
Update Reporting. Available: www.epa.gov/opptintr/iur/
[accessed 8 August2008].
U.S. EPA (Environmental Protection Agency). 2004b. What Is
the TSCA Chemical Substance Inventory? Available: www.
epa.gov/oppt/newchems/pubs/invntory.htm [accessed
8 August 2008].
U.S. EPA (Environmental Protection Agency). 2007a. EPA
Endocrine Disruptor Screening Program (EDSP). Available:
http://www.epa.gov/endo/Iaccessed 8 August2008].
U.S. EPA (Environmental Protection Agency). 2007b. EPA OPP
List of Chemicals Evaluated for Carcinogenic Potential.
Av a i I a b I e: h tt p ://www .epa.gov/pesticides/carlist/
[accessed 8 August2008].
U.S. EPA. 2007c. Estimation Programs Interface Suite
for Microsoft Windows, v3.20. Washington, DC U.S.
Environmental Protection Agency.
U.S. EPA (Environmental Protection Agency). 2008a.
ACToR Online (Aggregated Computational Toxicology
Resource). Available: http://actor.epa.gov/actor
[accessed 19 November 2008].
U.S. EPA (Environmental Protection Agency). 2008b. ChAMP—
EPA OPPT Chemical Assessment and Management
Program. Available: http://www.epa.gov/oppt/ar/2007-2008/
managing/champ.htm [accessed 8 August 2008].
U.S. EPA (Environmental Protection Agency). 2008c. Contaminant
Candidate List 3 Chemicals: Identifying the Universe.
Available: http://www.epa.gov/ogwdw/ccl/pdfs/report_
ccl3_chemicals_universe.pdf [accessed 8 August 2008].
U.S. EPA (Environmental Protection Agency). 2008d.
Contaminant Candidate List 3 Chemicals: Screening to a
PCCL. Available: http://www.epa.gov/ogwdw/ccl/pdfs/
report_ccl3_chemicals_screening.pdf [accessed 8 August
2008].
U.S. EPA (Environmental Protection Agency). 2008e. Drinking
Water Contaminant Candidate List and Regulatory
Determinations. Available: http://www.epa.gov/ogwdw/
ccl/index.html [accessed 8 August2008].
U.S. EPA (Environmental Protection Agency). 2008f. EPA
Integrated Risk Information System. Available: http://cfpub.
epa.gov/ncea/iris/ [accessed 24 October 2008].
U.S. EPA (Environmental Protection Agency). 2008g. EPA IRIS
Nominations. Available: http://www.epa.gov/fedrgstr/
EPA-RESEARCH/2007/December/Day-21/r24844.htm
[accessed 24 October 2008].
U.S. EPA (Environmental Protection Agency). 2008h. EPA IRIS
Queue. Available: http://www.epa.gov/fedrgstr/EPA-
RESEARCH/2007/December/Day-21/r24844.htm [accessed
24 October 2008].
U.S. EPA (Environmental Protection Agency). 2008L EPA NCCT
Communities of Practice. Available: http://www.epa.gov/
ncct/practice_community/ [accessed 27 October 2008].
U.S. EPA (Environmental Protection Agency). 2008J. EPA
Substance Registry System. Available: http://www.epa.
gov/srs/[accessed 8 August 2008].
U.S. EPA (Environmental Protection Agency). 2008k. EPA
ToxCast. Available: http://www.epa.gov/ncct/toxcast/
[accessed 8 August 2008].
U.S. EPA (Environmental Protection Agency). 2008I. High
Production Volume Information System (HPVIS). Available:
http://www.epa.gov/hpvis/index.html [accessed 8 August
2008].
U.S. EPA (Environmental Protection Agency). 2008m. HPV
Challenge. Available: http://www.epa.gov/HPV/ [accessed
19 November 2008].
U.S. EPA (Environmental Protection Agency). 2008n. Inert
Ingredients Permitted in Pesticide Product. Available:
http://www.epa.gov/opprd001/inerts/lists.html [accessed
8 August 2008].
U.S. EPA (Environmental Protection Agency). 2008o. TOXCST:
Research Chemical Inventory for EPA's ToxCast Program:
Structure-Index File. Available: http://www.epa.gov/ncct/
dsstox/sdfjoxcst.html [accessed 8 August 2008].
U.S. EPA (Environmental Protection Agency). 2008p. Toxics
Release Inventory Program. Available: http://www.epa.
gov/tri [accessed 8 August 2008].
U.S. EPA (Environmental Protection Agency). 2008q. TSCA
Inventory. Available: http://iaspub.epa.gov/srs/srs_proc_
qry.navigate?P_REG_AUTH_ID=1&P_DATA_ID=169&P_
VERSIONS [accessed 8 August 2008].
Waters M, Stasiewicz S, Merrick BA, Tomer K, Bushel P,
Paules R, et al. 2008. CEBS—Chemical Effects in Biological
Systems: a public data repository integrating study design
and toxicity data with microarray and proteomics data.
Nucleic Acids Res 36:0892-0900.
Weisbrod AV, Burkhard LP, Arnot J, Mekenyan 0, Howard PH,
Russom C, et al. 2007. Workgroup report: review offish
bioaccumulation databases used to identify persistent, bio-
accumulative, toxic substances. Environ Health Perspect
115:255-261.
Winter CK. 2002. Electronic information resources for food
toxicology. Toxicology 173(1-21:89-96.
Wishart DS, Tzur D, Knox C, Eisner R, Quo AC, Young N, et al.
2007. HMDS: the Human Metabolome Database. Nucleic
Acids Res 35:0521-0526.
Wolfgang GH, Johnson DE. 2002. Web resources for drug toxicity.
Toxicology 173(1-21:67-74.
Yang C, Benz RD, Cheeseman MA. 2006a. Landscape of current
toxicity databases and database standards. Curr Opin
Drug Discov Devel 9(11:124-133.
Yang C, Richard AM, Cross KP. 2006b. The art of data mining
the minefields of toxicity databases to link chemistry to
biology. Curr Comput Aided Drug Dis 2(21:135-150.
Young RR. 2002. Genetic toxicology: Web resources. Toxicology
173(1-21:103-121.
Environmental Health Perspectives • VOLUME 117 I NUMBER 5 I May 2009
Previous
TOC
695
-------
TOXICOLOGICAL SCIENCES 109(2), 358-371 (2009)
doi: 10.1093/toxsci/kfp061
Advance Access publication March 30, 2009
Toward a Public Toxicogenomics Capability for Supporting Predictive
Toxicology: Survey of Current Resources and Chemical Indexing of
Experiments in GEO and ArrayExpress
ClarLynda R. Williams-Devane,* Maritja A. Wolf,t and Ann M. Richard^'1
* U.S. EPA/Office of Research and Development (ORD)INational Health & Environmental Effects Research Laboratory (NHEERL), Research Triangle Park, NC
27519; ^Lockheed Martin (Contractor to U.S. EPA), Research Triangle Park, NC 27519; and $U.S. EPA/Office of Research and Development (ORD)INational
Center for Computational Toxicology (NCCT), Research Triangle Park, NC 27519
Received January 18, 2009; accepted March 23, 2009
A publicly available toxicogenomics capability for supporting
predictive toxicology and meta-analysis depends on availability of
gene expression data for chemical treatment scenarios, the ability
to locate and aggregate such information by chemical, and broad
data coverage within chemical, genomics, and toxicological
information domains. This capability also depends on common
genomics standards, protocol description, and functional linkages
of diverse public Internet data resources. We present a survey of
public genomics resources from these vantage points and conclude
that, despite progress in many areas, the current state of the
majority of public microarray databases is inadequate for support-
ing these objectives, particularly with regard to chemical indexing.
To begin to address these inadequacies, we focus chemical
annotation efforts on experimental content contained in the two
primary public genomic resources: ArrayExpress and Gene
Expression Omnibus. Automated scripts and extensive manual
review were employed to transform free-text experiment descrip-
tions into a standardized, chemically indexed inventory of experi-
ments in both resources. These flies, which include top-level
summary annotations, allow for identification of current chemical-
associated experimental content, as well as chemical-exposure-
related (or "Treatment") content of greatest potential value to
toxicogenomics investigation. With these chemical-index files, it is
possible for the first time to assess the breadth and overlap of
chemical study space represented in these databases, and to begin
to assess the sufficiency of data with shared protocols for chemical
similarity inferences. Chemical indexing of public genomics
databases is a first important step toward integrating chemical,
toxicological and genomics data into predictive toxicology.
Key Words: microarray; chemical; toxicogenomics; toxicity;
prediction.
Disclaimer: This manuscript was approved by the U.S. EPA's National
Center for Computational Toxicology for publication. However, the contents
do not necessarily reflect the views and policies of the EPA and mention of
trade names or commercial products does not constitute endorsement or
recommendation for use. Each of the authors declares no competing interests
pertaining to the present work.
1 To whom correspondence should be addressed at Mail Drop D343-03, 109
TW Alexander Dr., U.S. Environmental Protection Agency, Research Triangle
Park, NC 27711. Fax: (919) 685-3263. E-mail: richard.ann@epa.gov.
Published by Oxford University Press 2009.
Conventional toxicology investigates cellular and animal
responses to chemical treatment through domain-specific bio-
assay studies (e.g., chronic, developmental), typically mapping
a single chemical to a toxicological endpoint. Microarray
technologies, in contrast, detect genome-wide perturbations
resulting from a chemical treatment, and measure response
variables that probe a large number of genes and gene pathways
potentially underlying multiple toxicological endpoints. A
typical toxicogenomics experiment requires that linkages be
established between these technologies, focusing on treatment-
related effects of one or a few chemicals and attempting to relate
gene expression changes to a toxicological endpoint (Gomase
et al., 2008; Hamadeh et al., 2002; Hirabayashi and Inoue, 2002).
In silico toxicogenomic meta-analysis methods combine data
across existing toxicological and gene expression experiments to
generate new, and to confirm existing hypotheses of the effect of
a compound treatment. Such a capability depends upon the
availability of gene expression data derived from chemical
treatment scenarios, as well as anchoring toxicology data to
support predictive inferences.
The chemical nature of the problem requires a standardized,
chemical-centric view of data at all levels. Hence, a publicly
available toxicogenomics capability sufficiently robust for
mechanistic inferences and building predictive models requires
not only common data standards, protocols, and the ability to
query and aggregate common data types across resources, but
also broad data coverage within, and linkages across chemical,
genomics and toxicological information domains. These
requirements have, to varying degrees, informed development
of the major public microarray databases, and have been the
central design principle of specialized toxicogenomic resources
(Waters et al., 2008). In recent years, there have also been
significant advances in promoting toxicology standards and
data models (i.e., controlled vocabulary and hierarchical data
organization), quantitative high-throughput screening, and
chemically indexed bioassay data that, taken as a whole, have
Previous
-------
CHEMICAL INDEXING OF TOXICOGENOMICS RESOURCES
359
the potential to greatly enhance toxicogenomics capabilities in
the public domain (Dix et al, 2007; Martin et al, 2009;
Richard et al., 2008; Yang et al, 2006a, 2006b).
In the genomics field, the two largest public resources for
deposition of microarray data, approved by the Microarray
Gene Expression Data (MGED) Society (http://www.mged.
org/), are the European Bioinformatics Institute's (EBI)
ArrayExpress (http://www.ebi.ac.uk/arrayexpress) and the
National Center for Biotechnology Information's (NCBI) Gene
Expression Omnibus (GEO) (www.ncbi.nlm.nih.gov/geo).
Publishing requirements for the deposition of raw or processed
microarray data into these database repositories, coupled with
MIAME (Minimum Information About a Microarray Experi-
ment) standards for data reporting, are increasing the compara-
bility, utility and breadth of these resources (Ball et al., 2004).
Enhanced external programmatic access to the major public
microarray data repositories also allows third parties to
automatically extract and reformulate data to enhance informat-
ics and data mining capabilities (Boyle, 2005; Ivliev et al., 2008;
Zhu et al., 2008). Additional public efforts are aimed at
standardizing the description of experimental protocols (Taylor
et al., 2008), as well as improving toxicity data standards in
relation to toxicogenomics experiments (Burgoon, 2007; Fostel,
2008; Fostel et al. 2005, 2007). Largely neglected in the
genomics field, however, has been the standardization of
chemical information associated with the experimental data
when chemical treatment is a primary objective of the
experiment. Such annotation is essential for systematically
relating chemical property and effects information, irrespective
of whether the study has an explicit toxicological focus, across
the diverse data domains potentially contributing to toxicoge-
nomics. Furthermore, the ability to query, relate, and aggregate
information by chemical and across chemical space is essential
to the goal of chemical screening and toxicity assessment
(Dix et al., 2007; Richard et al., 2008; Yang et al., 2008).
In the remainder of this paper, we broadly survey the current
state of public microarray resources from the above vantage
points, focusing particularly on the two primary resources,
ArrayExpress and GEO. Although the latter resources are not
explicitly designed to meet the needs of the toxicogenomics
community, they currently serve as the two largest public
microarray data repositories of potential toxicogenomic
relevance and, as such, are potentially valuable sources of
data for toxicogenomics study. Despite progress in many areas,
we find the present state of public microarray repositories
inadequate for supporting interoperability and linkages across
diverse data domains in support of toxicogenomics. Particu-
larly noteworthy is the lack of minimal chemical annotation
and, as a result, the effective isolation of these resources and
associated data from the growing inventories of chemically
indexed bioassay information of potential relevance to
toxicology (Richard et al, 2006, 2008).
To begin to address these inadequacies, we propose and
implement a set of standard genomic fields for indexing of
experimental study records, aligned with current MIAME
guidelines, that enables cross-referencing and comparison of
total experimental content in GEO and ArrayExpress. In
addition, we implement a set of established chemical standards
for labeling experiments contained within ArrayExpress and
GEO, in collaboration with the U.S. Environmental Protection
Agency's (EPA) Distributed Structure-Searchable Toxicity
Database (DSSTox) project. We briefly describe the process
of annotation and creation of public-distribution DSSTox
chemical-index files for both GEO Series and Array Express
Repository. These files enable, for the first time, assessment of
the chemical scope, diversity, and coverage of experimental
content within GEO and ArrayExpress of potential use for
toxicogenomics study.
METHODS
For the purpose of assessing the relevance of public microarray resources to
toxicogenomics and predictive toxicology, we considered the current
annotation of experimental content pertaining to chemical treatment scenarios,
that is, cases in which study of gene expression changes induced by chemical
treatment constituted the primary goal of the experiment. As a measure of
interoperability between data resources, we examined the standardization of
terminology and data accessibility, as well as the formatting of data, paying
particular attention to specification of experimental protocols, such as animal/
tissue/cell treatment, RNA extraction, microarray preparation, data import/
export, and analysis. As a measure of chemical indexing, we examined the
degree of standardization and annotation pertaining to chemical-associated
experiments across public genomics resources and, particularly, whether the
chemical information was formally indexed, that is, contained in a separate,
searchable field.
For the purpose of chemically indexing experimental content in ArrayEx-
press and GEO, a chemical-exposure (or "Treatment") microarray experiment
is broadly defined by us as a study in which the cells, tissues, or whole
organisms were treated with a defined chemical, chemical mixture, or natural
substance (including biologies and proteins), DNA was extracted, and gene
expression changes resulting from this treatment were investigated with
microarray technologies. Whether the chemical to which the system was
exposed is a known toxicant, potential toxicant, natural substance, or
therapeutic agent need not be distinguished because the measured outcome is
the same, that is, treatment-related gene expression changes. However,
experiments in which chemical treatment was secondary to the primary
purpose of the experiment (e.g., treatment with prophylactic antibiotics for
maintaining tissue culture conditions) or where study of chemical-exposure-
induced effects was not the primary purpose of the experiment (e.g., treatment
with streptozocin to induce Diabetes Mellitus for investigating the effects of
diabetes) required further annotation and review. These cases of chemical-
experiment associations were labeled by us to indicate the role of the chemical
as other than "Treatment."
For initial inventory purposes, extraction of experimental description fields,
and locating chemical-associated experiments within ArrayExpress and GEO,
we used available web search options and programmatic access tools within
each system, as well as extensive manual review (a workflow diagram is
provided in Supplemental Fig. 1; additional details of the methodology
employed here are publicly available—see Acknowledgments). For the present
study, GEO Series provides the most complete inventory of current
experiments within GEO and these are also most closely aligned with
ArrayExpress Repository experiments. Hence, ArrayExpress Repository and
GEO Series experiments were the focus of the present chemical-indexing
efforts.
Previous
-------
360
WILLIAMS-DEVANE, WOLF, AND RICHARD
ArrayExpress. Due to its large size (over 6300 experiments at the time of
data extraction), limited and unstructured chemical annotation, and dynamic
content (updated regularly with new experiments), the review and annotation
process for ArrayExpress involved several iterative steps for identification and
characterization of chemical treatment experiments within the main database
repository. Initially, a bulk download of all data housed in the repository from
the main web site ((http://www.ebi.ac.uk/arrayexpress) was undertaken with
a wildcard query in the accession number query box (i.e., to retrieve all
experiments). The resulting records were individually reviewed and a pre-
liminary index of chemical information and indications of a Chemical
Exposure Record> was constructed. The latter field included any detail deemed
as potentially useful for discerning whether a record pertained to a chemical
treatment experiment, for example, designations in the ArrayExpress
field, such as "dose," "treatment," etc. This pre-
liminary chemical index was used to identify chemical-associated and chemical
"treatment" experiments, to infer the minimum information necessary to
identify such records from within ArrayExpress, and to build and test an
automated indexing capability using custom Perl scripts (http://www.perl.com).
Through an iterative process, scripts were refined to achieve better success at
detecting "true" chemical treatment experiments, verified by manual review
according to our definition above. Perl scripts for keyword text searching and
filtering were combined with manual curation methods to construct ArrayEx-
press Repository chemical-index files from website content downloaded on
September 20, 2008.
Gene Expression Omnibus. GEO contains user-deposited dynamic
content, and limited and unstructured chemical annotation. Hence, a manual
method similar to that employed in the review of ArrayExpress was initially
required. All data were downloaded from the GEO homepage in the GSE Series
format. Each of the GEO Series was manually reviewed for chemical content
and this information was used to construct an index of the chemical information
and indications of a Chemical Exposure Recordx As in ArrayExpress, the
indications of a Chemical Exposure Record> field contained details to aid in
discerning whether a record pertained to a chemical treatment experiment.
From this chemical-experiment index, the first chemical annotation of GEO
was completed. Similar to the annotation of ArrayExpress, this manually
curated chemical index was used to test and refine automated curation
approaches that were applied to subsequent versions of the GEO Series
inventory. Several automated methods were developed using NCBI Entrez
Programming Utilities (E-Utilities) (http://www.ncbi.nkn.nih.gov/entrez/query/
static/eutils_help.html), an XML version of the U.S. National Library of
Medicine's (NLM) chemical Medical Subject Headings (MeSH) library (http://
www.ncbi.nkn.nih.gov/sites/entrez7db = mesh), and a series of custom Perl
scripts to parse through a complete XML version of the GEO Series database.
The chemical index of GEO Series was completed using a series of Perl scripts
that call on E-Utilities, combined with manual curation methods, and was based
on content downloaded on September 20, 2008.
Chemical index. The main result of the above process was to produce
a static, preliminary chemical index for all chemical-associated microarray
experiments in ArrayExpress and GEO, in which the subset of chemical
"treatment" experiments were identified. These preliminary index files took the
form of a list of minimal chemical identifiers (most often chemical names only)
directly extracted from the user-deposited information in these two resources.
These chemical-experiment index files subsequently underwent a rigorous
cleanup and chemical quality review, using source (submitter)-provided
chemical information and contextual text descriptions to definitively identify
the chemical substance and its relationship to the experiment (e.g., treatment,
vehicle, reference). The generally poor quality and consistency of chemical
information contained in ArrayExpress and GEO submitter-supplied de-
scription fields, the high frequency of abbreviations and spelling errors, and
the lack of chemical identifiers such as Chemical Abstracts Service Registry
Numbers (CASRN; http://www.cas.org/) or EBI's Chemicals of Biological
Interest (ChEBI; http://www.ebi.ac.uk/chebi) identifiers, all prevented greater
application of automated text-mining and chemical name-to-structure conver-
sion capabilities. In addition, the need to accurately discern the role of the
chemical in the experiment (i.e., treatment, etc.) from the free-text description
prevented use of more efficient automated methods.
DSSTox Standard Chemical Fields were assigned to the chemical-index files
according to established procedures (http://www.epa.gov/ncct/dsstox/
ChemicalInfQAProcedures.html). These fields allow for standardized repre-
sentation of both the test substance ("TestSubstance" fields) and the chemical
structure ("STRUCTURE" fields) in relation to any chemical-associated
experiment record. DSSTox Standard Chemical Fields include chemical name,
CASRN (if available), and test substance description (e.g., single chemical
compound, macromolecule, mixture or formulation, etc). Where the test
substance is not overly large (> 1800 amu) and can be reasonably represented
by a single molecular structure, "STRUCTURE" fields are provided. These
include a public standard, "molfile" of the chemical structure (a two-
dimensional projection of the three-dimensional structure) assigned to the
substance, several fields automatically derived from the "molfile" structure
(i.e., molecular weight, formula, IUPAC name, SMILES, SMILES_Parent,
InChI, InChlKey), chemical type (i.e., defined organic, inorganic, organome-
tallic), and a field indicating the relationship of the STRUCTURE to the
TestSubstance (i.e., tested chemical, representative isomer in mixture, active
ingredient in a formulation, etc.) (for more information, see http://www.epa.
gov/ncct/dsstox/MoreonStandardChemFields.html). Assessment of chemical
overlap between GEO and ArrayExpress DSSTox chemical-index files was
determined on the basis of DSSTox TestSubstance identifiers.
RESULTS
Over 42 public Internet resources housing microarray data of
potential toxicogenomics relevance were initially identified
from various categories (Microarray World list of databases,
http://www.microarrayworld.com/DatabasePage.html). From
this list, we identified eight resources containing chemical-
exposure-related content, and divided these into two catego-
ries: primary and secondary genomics resources. Primary
genomics resources consist of the three MIAME-supportive,
MGED-approved gene expression repositories: NCBI's GEO,
EBI's ArrayExpress, and the Center for Information Biology
Gene Expression (CIBEX) database (see Table 1 for listing of
Sources, URLs, and references). Secondary genomics resour-
ces consist of five additional public genomics resources of
potential toxicogenomics relevance that contain data gathered
from chemical-exposure experiments in one or more laborato-
ries (see Table 1 for listing of Sources, URLs, and references).
A selection of public cheminformatics resources potentially
useful for supporting a public toxicogenomics capability are
listed in Supplemental Table 1. A brief description of survey
results are given for each data resource below, followed by
chemical-indexing results for the two major resources,
ArrayExpress and GEO.
ArrayExpress Repository
ArrayExpress is the largest user-depositor data repository
and MIAME-supportive public archive of microarray data in
Europe, consisting of two parts—ArrayExpress Repository and
the ArrayExpress Data Warehouse (Table 1). The ArrayEx-
press Repository currently exceeds 6900 experiments, and is
Previous
TOC
-------
CHEMICAL INDEXING OF TOXICOGENOMICS RESOURCES
361
TABLE 1
Primary and Secondary Genomics Data Resources with Content of Potential Use for Toxicogenomics
Primary genomic
resources
Database Source/URL
ArrayExpress European Bioinformatics Institute
(EBI); www.ebi.ac.uk/
microarray-as/ae/
GEO National Center for Biotechnology
Information (NCBI), National
Institutes of Health; www.ncbi.
References
Ball at al., 2004;
Brazma et al., 2006;
Parkinson et al., 2007;
Rustici et al., 2008
Barrett and Edgar, 2006;
Barrett et al., 2007;
Wheeler et al, 2008
Public data deposition Programmatic access
Yes Yes (XML)
Yes Yes (E-Utilities)
CIBEX
Secondary genomic EDGE
resources
CEBS
PEPR
dbZach
CTD
nkn.nih.gov/geo
DNA Data Bank of Japan (DDBJ),
National Institute of Genetics;
http://cibex.nig.ac.jp/
McArdle Laboratory for
Cancer Research, University of
Wisconsin-Madison; http://
edge.oncology.wisc.edu/edge3.php
National Institute of Environmental
Health Sciences (NIEHS); http://
cebs.niehs.nih.gov/cebs-browser/
Center for Genetic Medicine
Research; http://pepr.
cnmcresearch.org/
Department of Biochemistry &
Molecular Biology, Michigan
State University; http://
dbzach.fst.msu.edu
Mount Desert Island Biological
Laboratory; http://ctd.mdibl.org
Tateno and Ikeo, 2004
Hayes et al., 2005
Fostel et al., 2005;
Waters et al., 2003; 2008
Chen et al, 2004
Burgoon et al., 2006
Davis et al, 2009
Yes
No
Yes
No
No
No
No
No
No
No
No
No
indexed by Experiment Array Design and Protocol. Experi-
ments can be queried by Keyword, Experimental Accession
Number, Species, Experiment Type and Factors, Author,
Laboratory, and Publication information (http://www.ebi.
ac.uk/microarray-as/aer/entry). Repository data are cataloged,
assessed for completeness, and assigned a MIAME score that
represents the degree of MIAME compliance. The ArrayEx-
press Data Warehouse is based on more limited processed data
results from the ArrayExpress Repository, currently contains
740 Expression Profiles (website accessed on November 14,
2008), and allows users to browse curated datasets from both
a gene- and/or experiment-centric view. ArrayExpress has
incorporated significant experimental content from GEO,
which can be located within ArrayExpress by GEO Accession
Identifiers.
At the time of survey, ArrayExpress was not chemically
indexed, nor did it contain additional information about the
chemical tested other than the infrequently provided CASRN
or ChEBI number. Chemical information may be located in the
user-supplied protocols and free-text experimental description,
or can be searched with the advanced query tools from
ArrayExpress, including a keyword or text search in the
Description field in "Query for Experiments." These can also
be combined with specifications of =
"compound treatment" or "dose response," but these latter
annotations are optionally utilized and not consistently applied
by depositors to all chemical treatment experiments in the
database. Chemical information can also be embedded within
the ArrayExpress Sample-Data Relationship File (http://
tab2mage. s ourceforge .net/docs/sdrf .html).
In 2002, ArrayExpress introduced the Tox-MIAMExpress
data entry method, optionally employed by data submitters to
store toxicogenomics data in an effective manner (http://
www.ebi.ac.uk/miamexpress/). Tox-MIAMExpress was later
discontinued; however, the ArrayExpress Accession Number
Code, TOXM, designated to identify experiments for this
purpose, is still available for use when requested by data
submitters. Currently, the optional TOXM label is assigned to
fewer than 25 experiments, but in these cases, typically more
chemical identifier information, such as a CASRN and/or
a ChEBI number, is provided by the submitter along with
additional information recommended by the MIAME/Tox
initiative (http://www.ebi.ac.uk/tox-miamexpress).
Gene Expression Omnibus
GEO is the largest user-depositor data repository and
MIAME-supportive public archive of microarray data in the
U.S. (Table 1), containing data from approximately 10,000
Previous
-------
362
WILLIAMS-DEVANE, WOLF, AND RICHARD
experiments at the time of this writing. In GEO, raw and/or
processed data can be exported through the ftp website as well
as through the main GEO Series website. User information,
however, is entered using a free-text format that is sub-
sequently curated. GEO allows for a wide range of informed
queries with the Preview/Index window, where users can select
data based on choices for each attribute of the experiment. The
GEO repository has three key components: "Platform,"
"Sample," and "Series." "Platform" provides a description
of the array used in the experiment, as well as a data table
defining the array template. The data table contains hybridiza-
tion measurements for each element of the corresponding
platform. "Sample" provides a description of the biological
source and the experimental protocols. "Series" defines a set of
related samples considered to be part of a study and describes
the overall study aim and design. GEO has a complex,
hierarchical structure that works with the NCBI E-Utilities,
allowing one to query by submitter, organism, platform,
sample type, sample titles, and release date. Similar to
ArrayExpress, GEO hosts a smaller warehouse-type addition
named "GEO Datasets and Profiles" containing processed,
curated datasets that can be explored from both a gene- and/or
experiment-centric view.
Also, similar to Array Express, GEO is not chemically
indexed nor does it consistently contain information about the
chemical tested. Chemical names may be located in the
submitter-deposited GEO Data Series fields—Title, Summary,
Citation, or Samples—and are not consistently present in any
single field. Chemical names are provided by the submitter, are
rarely accompanied by CASRN or ChEBI identifiers, and do
not undergo curation or review. Hence, as is the case with
ArrayExpress, there is no easy or reliable way to identify
a chemical-exposure-related experiment, there is no central
listing of chemical content and, in both resources, we find that
the chemical names embedded within user-deposited descrip-
tion fields are highly variable, prone to errors and misspellings,
and frequently incorporate nonstandard abbreviations.
Center for Information Biology Gene Expression
The CIBEX database is a Japanese gene expression
MIAME-supportive, MGED-approved user-depositor system
(Table 1) that primarily serves experimenters from Asian
countries. It is included for completeness sake, but currently
does not contain significant chemical treatment content.
However, the experimental protocol and detail standardization
are noteworthy, with each record accompanied by a document
containing full MIAME details. There is also a high level of
curation and collaboration between CIBEX administrators and
depositors that allows for missing information to be identified
before publication, as well as for a high level of standardization
and accuracy.
At the time of this writing, CIBEX contains 32 experiments,
only one of which is a chemical-exposure experiment, with
CBX14 clearly labeled in the field
as "compound_treatment_design." Despite the high degree of
standardization of this resource, however, there is currently no
formal chemical annotation field accompanying a chemical
treatment experiment.
Environment, Drug, Gene Expression Database
The EDGE database (Table 1) is a closed (i.e., not open to
public user-deposits of data), curated system designed for the
comparison, analysis and distribution of toxicogenomics
information in a relational format. EDGE is chemical treatment
centric and chemically indexed, with a toxicological focus. All
experiments were performed in the Bradfield Laboratory using
a standardized protocol involving custom cDNA arrays of
minimally redundant hepatic clones, chosen through chemical-
exposure experiments with prototype hepatic toxicants: 2,3,7,8,
tetrachlorodibenzo-/? dioxin (TCDD), cobalt chloride, and
phenobarbital. The experimental conditions include 22 chem-
ical treatments, 4 control treatments, and 1 environmental
stressor (fasting) over 1 mutant (circadian wild-type control).
All chemical treatments were chosen for the express purpose of
investigating transcriptional profiles pertaining to hepatotoxic-
ity in mice.
Despite its small size and limited focus, EDGE incorporates
a high level of standardization and comparability across
species, array, experimental protocol, and experimental details,
and demonstrates how a fully relational database built on such
data can facilitate toxicogenomics investigation. However,
EDGE is not a user-depositor system and currently lacks the
tissue, species, and chemical diversity necessary for broader
toxicogenomics exploration.
Chemical Effects in Biological Systems
CEBS is a public user-depositor data repository with an
explicit toxicological and toxicogenomics focus (Table 1).
CEBS can accommodate study design, timeline, clinical
chemistry, and histopathology findings, as well as microarray
and proteomics data. Each experiment in CEBS pertains to
a chemical/environmental exposure or a genetic alteration in
reference to clinical or environmental studies. CEBS has
a complementary functional component known as the Bio-
medical Investigation Database (BID) (https://dir-apps.niehs.
nih.gov/arc/), which is a relational database used to load and
curate study data prior to exporting to public CEBS. BID also
aids in the capture and display of novel data, including PCR
and toxicogenomic-relevant fields, as used in Array Express's
TOXM designation.
CEBS is currently indexed by study and subject character-
istics, such as environmental, chemical, or genetic stressor and
stressor protocol, and includes observations on rat, mouse, and
C. elegans. CEBS is one of the few genomics resource profiled
in this survey, and the only resource with significant
toxicogenomics-relevant microarray content, that incorporates
formal chemical name annotation of experiments. At the time
Previous
TOC
-------
CHEMICAL INDEXING OF TOXICOGENOMICS RESOURCES
363
of this writing, CEBS lists an inventory of 136 chemical
names, or "chemical stressors," associated with experimental
content, along with a searchable CASRN field containing 121
entries. CEBS plans to incorporate additional chemical stand-
ards, including structure annotation, in collaboration with the
EPA DSSTox project.
Public Expression Profiling Resources
Similar to EDGE, the PEPR database (Table 1) is a closed,
curated system designed to serve as a public resource of gene
expression profile data generated in the same laboratory, using
the same chip type for three species, and subject to the same
quality and procedural controls. PEPR is aimed at providing
a standardized warehouse for the analysis of time-series data.
The high degree of standardization within PEPR grants users
comparability across arrays without laboratory and array bias,
much like EDGE. PEPR adheres to quality control and
standard operating procedures and is indexed by Principle
Investigator, Tissue type, Experiment, and Organism, but has
a very few chemical treatment-related experiments and lacks
relational searching capabilities. However, the time-series
query analysis tool (SGQT) enables the novel generation of
graphs and spreadsheets showing the action of any transcript of
interest over time. PEPR also differs from EDGE in the
extensive data export options that include raw image files
(.dat), processed image files (.eel) and interpretation files (.txt).
PEPR also has external links to GEO, where PEPR data are
mirrored through an automated export/import process.
In PEPR, chemical information is stored in free-text fields
such as the title, description, and array titles, similar to
ArrayExpress and GEO. At the time of this writing, PEPR
contains 72 experiments, of which 10 are determined to be
chemical/environmental exposure experiments. Hence, PEPR
currently covers very limited chemical space, but the SGQT
tool for analysis of time-series microarray data, as well as the
standardized chemical-exposure experiments are of potential
value for toxicogenomics studies.
DbZach
dbZach, a laboratory tool offered for local installation, is of
interest as a modular MIAME-compliant, toxicogenomic-
supportive relational database designed to facilitate data
integration, analysis, and sharing in support of mechanistic
toxicology and toxicogenomics studies (Table 1). dbZach
consists of several subsystems for the standardization of all
data elements of a toxicogenomics experiment as well as
traditional toxicological experiments and, additionally, has
built-in functionality for data import and export of both raw
data and processed data. Similar to EDGE, the dbZach project
has created a sophisticated relational data environment for
integrating and exploring many aspects of a toxicogenomics
study. However, also similar to EDGE, this project is very
narrowly focused in chemical space and primarily limited to
estrogen and estrogenic chemicals.
Comparative Toxicogenomics Database
The CTD is worthy of mention for its toxicogenomics
relevance, but is not a traditional genomics database (Table 1).
Rather, it is a database of curated relationships between
chemicals, genes, and diseases mined from journal articles.
CTD provides text-mineable access to the toxicogenomic
literature, but currently provides direct linkage to only one
secondary genomics resource, that is, EDGE.
Also worthy of note, CTD uses the chemical subset of the
NLM MESH vocabulary to provide formal chemical annota-
tion of its content and to link to various chemically indexed
toxicology resources (http://ctd.mdibl.org/resources.jsp7type =
chem). The present CTD inventory of over 4400 chemical
substances also has recently been deposited into the NCBI
PubChem resource (http://pubchem.ncbi.nlm.nih.gov/; Supple-
mental Table 1) to offer stracture-searchability and broader
access to chemically indexed resources.
Table 2 compares the above inventory of genomics resources
from the standpoint of being chemically indexed (i.e., chemical
identifiers are required and entered in standard fields),
MIAME-supportive, and standardized with respect to various
experimental descriptions. Additional details on the compari-
son of the primary and secondary genomics resources identified
in this study with respect to the types of gene expression data
stored, toxicological focus, formats of data available for
download (raw or processed), ability to query data, ability to
import or export experimental data, and programmatic access
are presented in Supplemental Table 2.
Web-based queries and programmatic access were used in
the present study to extract current experimental content from
ArrayExpress Repository and GEO Series, and to identify
corresponding experiment annotation fields (resulting from
adherence to MIAME guidelines in the two systems) that could
be mapped to common fields to enable comparisons across the
two inventories. We implemented a set of 14 Standard
Genomics Fields in Table 3 to serve this purpose and to
confer read-across capability between the two inventories. All
but two of these fields map to existing MIAME-compliant data
fields, which vary only slightly in name in GEO and
ArrayExpress (see expanded columns in Supplemental Table
3) and, thus, are straightforward to implement. One new field,
"Experiment_URL," contains a static URL link to enable
outside Internet access directly to the experiment accession
summary page in either ArrayExpress or GEO. The last field,
"Chemical_StudyType," has no corresponding field in either
ArrayExpress or GEO, and was introduced by us to begin to
address the currently missing chemical annotation layer for
gene expression experiments in both resources.
DSSTox chemical-index files for GEO and ArrayExpress
created by the above methods are publicly available for
download from the DSSTox website (http://epa.gov/ncct/
dsstox/). In addition to the main DSSTox chemical-index files
that include one record for each unique chemical (i.e., unique
Previous
TOC
-------
364
WILLIAMS-DEVANE, WOLF, AND RICHARD
TABLE 2
Standardization and Indexing of Genomics Data Resources
Standardized
Data resource"
Indexed by
chemical
MIAME-supportive
Species
Array
information
Experimental
protocol
Experimental
details
Allows relational
searching17
ArrayExpress
GEO
CIBEX
EDGE
CEBS
PEPR
dbZach
CTD
NA
NA
"Refer to Table 1 for full names, sources, URLs, and references associated with these data resources; feature present (+) or absent (—).
^Standardized entries refer to internal content adhering to controlled vocabularies, and represented in defined and required fields; NA, not applicable.
Relational searching refers to the ability to construct AND/OR-type queries across the content of defined fields.
test substance) in the "Treatment" category in each of the two
repositories, we have published Auxiliary files that include
DSSTox Standard Chemical Fields, Standard Genomics Fields
(14), and additional Source-specific experiment description
fields (33 for ArrayExpress, 4 for GEO) for the full chemical-
associated experiment inventories in the two resources (i.e.,
one record for each chemical-experiment pair). Detailed
descriptions of the content of these files and their incorporation
into the DSSTox Structure-Browser (http://www.epa.gov/
dsstox_stracturebrowser/) and PubChem, the results of which
enable structure-based Internet linkages directly to ArrayEx-
press and GEO experiment summary pages, are provided
elsewhere (Williams-Devane et al, 2009).
Table 4 provides a breakdown of the current chemical-
associated experimental content within ArrayExpress Reposi-
tory and GEO Series according to all Chemical_StudyType
TABLE 3
Standard Genomics Fields for Common Indexing of Experiments Contained in ArrayExpress Repository and GEO Series
Field name
Description
Experiment_Accession
Experiment_AlternativeAccession
ExperimentJdNumber
Experiment_Title
Experiment_Description
Experiment_URL
Experiment_PubMed_Information
Experiment_PublicationDate
Species
Number_S amples
Experiment_ArrayAccession
Experiment_ArrayType
Experiment_ArrayTitle
Chemical_StudyType:°
Reference
Treatment
Vehicle
Combination ^Treatment
Media
Not_Enough _Information
A unique combination of informative prefix and number used to identify each dataset.
An alternate accession number. Example: GEO files in ArrayExpress have GSE#### (GEO Series) secondary Accession
number for users
to find the same data in GEO).
A unique identification number for each experiment within each database.
The title of the experiment.
A free-text, user-submitted description of the experiment or dataset.
URL links to the Source Experimental Download Page.
A unique number that links users to each PubMed publication associated with each experiment or dataset.
Date indicating when the dataset was released to the public or published.
Species as listed by the user.
Number of samples used within a microarray experiment or dataset.
An accession number for each array design or platform.
Details about the platform used or details about data other than raw data that users have submitted.
The user-submitted title of the Array/Platform used in the experiment.
A designation of the role of the identified chemical in the given experiment. Allowed entries are listed as Subsections to
this field (e.g., Reference, Treatment, Vehicle, ...).
Chemical used to mimic a biological or environmental situation.
The primary focus of experiment or study is to understand the transcriptomic effects of the chemical.
Chemical used to aid the administration of the treatment to the organism, such as dimethyl sulfoxide.
Multiple chemicals used together for treatment purposes (see "Treatment" above).
Chemical used in maintenance of the tissue culture or sample conditions, such as phosphate buffered saline.
Sufficient information is not present in the experimental description to determine the role of the chemical.
"Subsections to the Chemical_StudyType field have allowed entries: Reference, Treatment, etc., with linkage text "AND" used for combinations (e.g.,
TreatmentANDReference).
Previous
-------
CHEMICAL INDEXING OF TOXICOGENOMICS RESOURCES
365
TABLE 4
Classification of Chemically Indexed Genomics Experiments in ArrayExpress and GEO by "Chemical_StudyType"
Database"
ArrayExpress Repository
GEO Series
Total no. of
Experiments6
6346
9957
Total no. of
Chemical-
Experiment
Records'7
2365
2381
Chemical_StudyType Classification6
Breakdown of Total no. of Chemical-Experiment Records (Unique Chemicals/
Total no. of
Unique Chemicals^
1011
1064
Treatment*
1609 (f
1951 (f
510)
538)
Reference
266 (157)
152 (60)
Vehicle
138 (26)
81 (48)
Media
111 (68)
14 (14)
Combination
Treatment*
109 (91)
72 (38)
Multiple
Classifications
118 (83)
111 (67)
Other
14 (10)
0(0)
"All numbers relate to database content extracted on September 20, 2008.
6Total number of experiments contained in the public resource (also corresponds to the number of unique Accession IDs).
^Number of Chemical-Experiment pairs extracted from the Total no. of Experiments prior to determination of the Chemical_StudyType Classification, where
some experiments in the Total no. of Experiments map to no chemicals, and some experiments involving multiple chemicals map to more than one Chemical-
Experiment record.
rfTotal number of unique chemical test substances (i.e., no chemical test substance identity is duplicated) identified in the total group of Chemical-Experiment
records, irrespective of Chemical_StudyType Classification.
^Definitions of Chemical_StudyType Classifications are provided in Table 3.
^Number of Chemical-Experiment Records corresponding to each Chemical_StudyType category (with corresponding number of unique chemicals in
parentheses), where for the purposes of this table one record is assigned to one category and if the chemical is used for different purposes within one experiment
(e.g., TreatmentANDReference), it is assigned to the "Multiple Classifications" category.
^Number of Chemical-Experiment Records (with corresponding number of unique chemicals in parentheses) out of the total group of Chemical-Experiment
Records that are associated with the ' 'Treatment'' category according to the criteria for a chemical-exposure scenario set forth in this paper; any record labeled as
"Treatment" or "CombinationTreatment" (alone or in combination with other Chemical_StudyType labels, e.g., TreatmentANDReference), are included in the
final DSSTox chemical-index file.
categories (all counts correspond to data extracted on September
20, 2008). Of the 6346 total ArrayExpress experimental
descriptions downloaded, more than a third (2365) were
identified as chemical-associated experiments by the procedures
outlined in the Methods section, corresponding to 1011 unique
chemical test substances. Similarly, of the 9957 GEO Series
experimental descriptions downloaded, nearly a quarter (2381)
were identified as chemical-associated experiments, correspond-
ing to a total of 1064 unique chemical test substances (Table 4).
Table 5 provides a breakdown of the "Treatment"-
associated experimental content within the ArrayExpress
Repository and GEO Series according to DSSTox chemical
classification categories. Of the 1835 total "Treatment"-
associated experiments in the ArrayExpress Repository, 1282
experiments (or 70% of the total) are associated with a "defined
organic" chemical test substance (note that multiple experi-
ments can map to the same chemical). GEO Series contains
a similarly high percentage of "Treatment" experiments
associated with a defined organic chemical test substance, that
is, 1544/2134, or 72%. The above indicators give a rough sense
of the size of the inventory of microarray experiments
associated with defined organics in the public domain.
Table 5 also provides indications of the size of the chemical
space associated with these "Treatment" experiments. Of the
total number of unique chemical test substances associated
with the "Treatment" category of experiments in ArrayExpress
Repository, 628/887, or 71% correspond to defined organics.
Although these include some drugs, small peptides and
biologies with molecular weights ranging 600-1700 amu, the
majority (> 90%) are small molecular weight (< 500 amu)
organic chemicals for which a chemical structure can be
assigned and that tend to be of greatest interest for environ-
mental toxicology and structure-activity relationship models and
inferences. A similar percentage applies to GEO, that is, 751/
1014, or 74% of unique chemical test substances associated with
"Treatment" experiments correspond to defined organics.
Hence, both resources span a relatively large number of unique
defined organic chemicals, which implies a broad chemical
diversity associated with public microarray experiments. Within
ArrayExpress, the chemical that maps to the largest number of
chemical-experiments is "estradiol," occurring in 53 experi-
ments, 44 of which are classified as "Treatment" experiments.
Comparison of GEO and ArrayExpress Experimental and
Chemical Content
Application of DSSTox Standard Chemical Fields and the
set of 14 Standard Genomics Fields enable direct comparison
of GEO Series and ArrayExpress Repository experimental and
chemical content. In addition, the DSSTox Auxiliary files for
ArrayExpress include a number of easily extracted field
characteristics affiliated with each experiment, including
Array/Platform type, Species, and the MIAME Score and its
five subcategories: Array or Platform information, Factor
information, Raw Data information, Processed Data informa-
tion, and Protocol information. The latter annotations are
particularly valuable for assessing the sufficiency of the
experimental data for reanalysis.
The distribution of ArrayExpress "Treatment" chemical-
experiments assigned to these various categories of experimental
description is provided in Table 6. The distribution of MIAME
Previous
-------
366
WILLIAMS-DEVANE, WOLF, AND RICHARD
TABLE 5
Classification of Chemically Indexed "Treatment" Genomics Experiments in ArrayExpress and GEO by DSSTox Chemical
Classification
Database"
ArrayExpress
repository
GEO series
Total no. of
Chemical-
Experiment
Records6
2365
2381
Total no. of
"Treatment"
Chemical-
Experiment
Records'7
1835
2134
DSSTox Chemical Classification6
Breakdown for "Treatment" Chemical-Experiment
Records (Unique Chemicals/
Total no. ot
Unique Chemicals'* No structure*
887 373 (179)
1014 346 (173)
Defined organic Inorganic Organometallic
1282 (628) 153 (60) 27 (20)
1544 (751) 210 (71) 34 (19)
"All numbers relate to database content extracted on September 20, 2008.
6See Table 4.
cTotal number of Chemical-Experiment records assigned to any "Treatment" Chemical_StudyType (e.g., Treatment, CombinationTreatment,
Treatment&Reference, etc.) according to the criteria for a chemical-exposure scenario set forth in this paper.
''Total number of unique chemical test substances (i.e., no chemical test substance identity is duplicated) identified in the total group of "Treatment" Chemical-
Experiment Records.
eRefers to DSSTox Standard Chemical Field Definition and allowed entries for STRUCTURE_ChemicalType (http://www.epa.gov/ncct/dsstox/
CentralFieldDef.html#STRUCTURE_ChemicalType).
^Number of "Treatment" Chemical-Experiment Records corresponding to each Chemical Classification category (with corresponding number of unique
chemical substances in parentheses), where each record maps to a single chemical classification and the list of unique chemicals for this "Treatment" subset of
experiments constitutes the final DSSTox structure-index file.
^Number of "Treatment" Chemical-Experiment Records (with corresponding # of unique chemicals) where the chemical test substance is identified, but not
assigned to a DSSTox chemical structure, for example, this can be an undefined mixture, polymer, or macromolecule.
scores is particularly illuminating. A total MIAME Score of
5 indicates that all components of the MIAME compliance
criteria have been included by the submitter. Only 18% (or 216)
of the chemical treatment experiments in the ArrayExpress
Repository have all five components of information, whereas
50% (or 596) have four components of information. Most
noteworthy for this subset of "Treatment" experiments, how-
ever, raw data information is missing for 29% (or 347), pro-
cessed data is missing for 11% (or 131), and protocol is missing
for 21% (or 291) (Table 6). Given that these are essential
experimental components for the reanalysis of gene expression
data, these numbers limit the number of chemical treatment
experiments within ArrayExpress that are potentially useful for
broader toxicogenomics investigation.
The ArrayExpress Repository has experienced steep growth in
the past few years, largely as a result of the integration of GEO
experimental content (approximately 4500 experiments were
added from January 2007 to January 2008). ArrayExpress files
with E-GEOD-XXXX accession numbers mirror GEO Series
entries and currently represent more than 50% of the chemical-
exposure, or "Treatment" experiments in the ArrayExpress
Repository (Fig. 1). Figure 1 also shows that the total number of
chemical-experiment pairs (a pair being a 1:1 mapping of
chemical to experiment) and total number of "Treatment"
chemical-experiment pairs identified in the current study are
comparable between ArrayExpress and GEO, with greater than
50% overlap of chemical-experiment pairs in all categories.
Unlike ArrayExpress, GEO Series currently provides no
MIAME scoring of content. However, because a significant
portion of the "Treatment" experiments represented in GEO
Series has been incorporated into the ArrayExpress Repository,
it was possible to create a table summarizing these experimental
factors for the subset of GEO chemical "Treatment" experi-
ments contained within ArrayExpress (Table 6). Only 11% (or
81) of the GEO records in ArrayExpress are assigned a MIAME
Score of 5; however, 56% (or 415) have a MIAME score of 4. A
much greater percentage, 45% (or 335) of GEO records in
ArrayExpress, have corresponding Raw Data, whereas 100% (or
745) have Processed Data (most likely a precondition for
inclusion of GEO experiments in ArrayExpress).
Figure 2 presents overlap of the unique chemical content
pertaining to the "Treatment" chemical-experiment category.
Assessment of chemical overlap between GEO and ArrayExpress
DSSTox files was determined on the basis of DSSTox
"TestSubstance" identifiers. The steroids, estradiol and dexa-
methasone, are associated with the largest numbers of microarray
experiments in both cases, and the largest number of shared
experiments as well. Other test substances most commonly
associated with experiments in either GEO or ArrayExpress
include Ethanol, 2,3,7,8-TCDD, Retinoic Acid, and Trichostatin,
each of which is of broad lexicological interest.
Assessing Toxicological relevance of GEO and ArrayExpress
Chemical Content
DSSTox chemical structure annotation enables, for the first
time, an examination of the chemical diversity and coverage of
GEO Series and ArrayExpress Repository experiments. We
Previous
TOC
-------
CHEMICAL INDEXING OF TOXICOGENOMICS RESOURCES
367
TABLE 6
Characteristics of the ArrayExpress Repository pertaining to
"Treatment" Experiments (based on Data Extracted on
September 20, 2008)
GEO series from
ArrayExpress ArrayExpress
Characteristics
Major Number (%) of Number (%) of
characteristic "treatment" "treatment"
value experiments'* experiments'*
Array/platform
Species
MIAMEScore_Total6
MIAMEScore_Arrayc
MIAMEScore_Factor'J
MIAMEScore_RawDatae
MIAMEScore_
ProcessedData^
MIAMEScore_ProtocoP
Affymetrix
Agilent
Other
Homo sapiens
Mus musculus
Rattus
Arabidopsis
Saccharomyces
cerevisiae
Other
5
4
3
2
1
0
1
0
1
0
1
0
1
0
1
861 (73%)
82 (7%)
238 (20%)
377 (32%)
317 (27%)
173 (15%)
159 (13%)
55 (5%)
100 (8%)
216 (18%)
595 (50%)
309 (26%)
55 (5%)
6 (1%)
78 (7%)
1103 (93%)
551 (47%)
630 (53%)
347 (29%)
834 (71%)
131 (11%)
1050 (89%)
291 (21%)
890 (79%)
691(93%)
54 (7%)
0 (0%)
264 (35%)
220 (30%)
126 (17%)
76 (10%)
15 (2%)
44 (6%)
81(11%)
415 (56%)
211 (28%)
38 (5%)
0 (0%)
0 (0%)
745 (100%)
421 (57%)
324(43%)
335 (45%)
410 (55%)
0 (0%)
745 (100%)
195 (26%)
550 (74%)
"Note that the total number of "Treatment" experiments (or studies) will be
less than the total number of ' 'Treatment'' chemical-experiment pairs in Table
5 due to inclusion of experiments/studies that have tested multiple chemicals
(and/or used multiple platforms, etc.).
6The Total MIAME score ranges from 0 to 5 and is a sum of the independent
scores of the five subcomponent scores, each of which takes on the value of
either 0 or 1 (absent or present).
cSpecific information about the design of the array or the platform used was
submitted (1) or not submitted (0) with the experiment by the submitter.
Included Array information is assigned an Array Accession number (see
Experimental_Accession, Table 3) within ArrayExpress.
"A list of experimental factors was submitted (1) or not submitted (0) with
the experiment by the submitter; factors might include information on the cell
line or particular compounds and dose information used in the experiment.
eRaw data was submitted (1) or not submitted (0) with the experiment by the
submitter.
^Processed data was submitted (1) or not submitted (0) with the experiment
by the submitter.
^Specific information about the experimental protocols used in the
experiment was submitted (1) or not submitted (0) with the experiment by the
submitter. Included Protocol information is assigned a Protocol Accession
number within ArrayExpress.
found significant numbers of experiments in both resources
mapped to families of similar chemicals, as well as to a broad
diversity of chemical structures, spanning a wide range of
lexicologically relevant chemical functional hierarchies and
classes (Supplemental Fig. 2).
A further metric of toxicological relevance is provided by
the overlap of unique "Treatment" chemical substances in
GEO and ArrayExpress with the current published DSSTox
inventory, which includes more than 10,000 unique chemical
substances, and spans a variety of environmentally and
lexicologically relevant chemical inventories and dala sels
from various sources, including EPA, Ihe National Toxicology
Program, and Ihe U.S. Food and Drug Association (Richard
et al., 2008). Al Ihe lime of Ihis survey, more lhan 550 unique
chemical subslances in Ihe DSSTox GEO and/or ArrayExpress
files (GEOGSE and ARYEXP) corresponding to "Trealmenl"
experimenls are conlained wilhin one or more of Ihe 11
previously published DSSTox Dala Files (http://www.epa.gov/
nccl/dsslox/DalaFiles.hlml), and there are a lolal of 1294
overlapping instances (i.e., some chemicals occur in multiple
DSSTox Dala Files). Of Ihese overlapping instances, Ihree
chemical subslances (Bisphenol A, di(2-elhylhexyl) phlhalale
and dibulylphthalale) occur in eighl DSSTox Dala Files, and
a lolal of 309 chemical subslances occur in Iwo or more
DSSTox Dala Files. These numbers indicate lhal significanl
numbers of GEO and Array Express "Trealmenl" chemical-
experimenls correspond to chemicals of potential toxicological
concern, for which additional in vitro or in vitro dala may exist
DISCUSSION
The term "chemogenomics" has been proposed to more
generally encompass Ihe overlap of genomics technologies
with Irealmenl-relaled chemical effecls on biological systems,
including both toxicily-relaled and Iherapeulic effecls (Fielden
and Kolaja, 2006). Chemogenomics adds a lop-mosl chemical
layer to dala organization, wilh broad chemical coverage of
slandardized-prolocol experimenls a key requiremenl for
discerning activity patterns lhal can be confidenlly extrapolated
across chemical space. This approach and ils implementation
are perhaps besl exemplified by Ihe Iconix DrugMalrixR
database and applications (Ganler et al., 2005). The Iconix
database consisls of dala generated for a single species (ral),
treated by more lhan 600 compounds in seven tissue types,
representing upwards of 3200 differenl drug-dose-lime-lissue
combinations. The database covers five differenl domains of
dala: microarray, clinical chemislry, hemalology, organ weighl,
and hislopalhology, and was buill using a common microarray
platform and srringenl experimental protocols and standards for
dala generation and processing.
Whereas Ihe Iconix database represenls an ideal, practically
speaking, il is far removed from Ihe reality of a public
microarray resource, upon which mosl public toxicogenomics
investigations musl rely. In Iheir role as primary repositories of
dala associated wilh Ihe published scientific lileralure, public
microarray dala repositories such as GEO and ArrayExpress
Previous
TOC
-------
368
WILLIAMS-DEVANE, WOLF, AND RICHARD
10000
9000
8000
7000
6000
5000
4000
3000
2000
1000
9957'
GEO Series
ArrayExpress Repository
| Overlapping Content
6346
3713
2500
2000
1500
1000
500
2381
2365
1374
2500
2000
1500
1000
500
2134
1075
Total Experiments
Total
Chemical-Experiment Pairs
Total "Treatment"
Chemical-Experiment Pairs
FIG. 1. Comparison of numbers of GEO Series and ArrayExpress Repository experiments, chemical-experiment pairs, and ' 'Treatment'' chemical-experiment
pairs, also showing overlapping content between the two systems; refer to totals and legends in Tables 4 and 5 (based on data extracted September 20, 2008).
cannot limit their content to include only experiments adhering
to strict common protocol standards and traditional model
organisms. A public data resource can, however, strive for
completeness and accuracy of experimental annotations and to
provide user-access to raw data for reanalysis. Similarly, the
accurate identification of a chemical in relation to an
experiment, particularly where the primary purpose of the
experiment is to discern effects of the chemical on a biological
system, should be considered as primary experimental
annotation and absolutely essential to experimental reproduc-
ibility. Whereas standardization and chemical indexing of
microarray experiments at the time of data deposition and
publication is the ideal, if minimally sufficient information (i.e.,
a valid chemical name, along with specification of the purpose
FIG. 2. Comparison of the total sets of unique chemicals pertaining to Treatment Chemical-Experiment pairs in ArrayExpress Repository and GEO Series
from the DSSTox data files; shown in each section are the chemicals mapping to the largest number of "Treatment" Chemical-Experiments in each case, with the
number of experiments shown in parentheses (GEO/ArrayExpress) (based on data extracted September 20, 2008).
Previous
TOC
-------
CHEMICAL INDEXING OF TOXICOGENOMICS RESOURCES
369
of the chemical in relation to the experiment) is collected at the
time of data deposition in required data fields, formal chemical
indexing with structure annotation and quality review can be
performed efficiently with the appropriate chemical expertise in
collaboration with public efforts such as DSSTox and ChEBI
(Supplemental Table 1).
As the present survey has shown, although a number of
public microarray resources have the potential to support
toxicogenomics investigations, these resources currently rep-
resent a patchwork of disconnected or loosely connected
inventories and capabilities (Larsson and Sandberg, 2006),
having different goals, degrees of standardization, public data
accessibility, data mining ability, and utility for toxicogenom-
ics investigation (Table 2; Supplemental Table 2). Primary
genomics resources (Table 1) serve as official MGED-
sanctioned repositories for public gene expression data
associated with the scientific literature (Mattes et al., 2004;
Salter, 2005) with GEO and ArrayExpress, by far, the largest
and most important resources, currently. They both are
MIAME-supportive databases, meaning that they accept all
information about an experiment set forth by the MIAME
guidelines; however, they do not actually require this in-
formation. In addition, there is insufficient standardization
currently within GEO or ArrayExpress pertaining to protocol
or experimental description to fully support exploration within
and across these resources (Table 2). Secondary genomics
resources contain genomic-related data but generally have
more limited content and are designed for more specialized
purposes and applications (Table 1).
With its specific focus on toxicogenomics, attention to
chemical indexing, and addition of the BID system, CEBS is
worthy of special mention, having incorporated many elements
of an ideal toxicogenomics resource. To support robust
relational searching for toxicogenomics, CEBS has the added
task of capturing and systematizing user-deposited data
pertaining to a study. CEBS bridges the gap between an
open-access, user-depositor system and a relational, curated
database by instituting a high degree of standardization and
data controls that extend beyond MIAME guidelines (Fostel
et al., 2005). CEBS is striving for much larger coverage of
chemical space in relation to chemical treatment experiments.
In collaboration with DSSTox, and building on current
annotation efforts of GEO and ArrayExpress, CEBS will
provide structure-searching capabilities and chemical linkages
to external public resources, such as PubChem. In addition,
CEBS will provide direct access to GEO and ArrayExpress
"Treatment" chemical-experiment content, as well as auto-
mated secondary deposition of CEBS content to GEO.
A most noteworthy deficiency of most secondary genomics
resources and the two primary genomics resources—ArrayEx-
press and GEO—highlighted in the present study, is the
complete lack of incorporation of chemical annotation and
standards that would allow aggregation of data for the same or
similar chemicals, and linkage to growing lists of chemically
indexed resources (Supplemental Table 2). Due to the lack of
chemical-reporting standards, the process for identifying
chemical treatment-related experiments in ArrayExpress Re-
pository and GEO Series in this study was time-consuming and
difficult to automate (Supplemental Fig. 1 and Supplemental
Example 1). Present efforts serve to highlight deficiencies in
microarray experiment data deposition requirements and
standards with regard to chemistry and chemical treatment-
related experiments that, if better addressed, could greatly
facilitate chemical annotation and data integration efforts in the
future. With formal chemical annotation, it becomes possible to
assess the chemical coverage of public gene expression
databases, to link data for common or similar chemicals across
information domains, including toxicology, as well as to gather
data from comparable experiments, possibly performed in
different labs and species, that can begin to serve as the basis
for meta-analysis or structure-activity hypotheses. Furthermore,
the proposed set of Standard Genomics Fields, most of which
map to existing fields from both GEO and ArrayExpress, serve
to bridge the two resources and facilitate comparisons and
incorporation of their content into other resources in a stan-
dardized way.
CONCLUSION
It is hoped that the current exercise to create, publish, and
link chemical-index files for GEO Series and ArrayExpress
Repository has had two primary impacts: (1) to highlight
deficiencies in the current chemical annotation and curation
methods within ArrayExpress and GEO that particularly impact
toxicogenomics applications of these resources; and (2) to
show the way forward in terms of the potential benefits that can
be derived by incorporating robust chemical annotation and
linkages of chemical treatment-related content to these public
resources. Recently improved coordination of the EBI
ArrayExpress and ChEBI projects, whereby ChEBI provides
link-outs from chemical structure to particular ArrayExpress
experiments (currently only provided for a handful of experi-
ments for which ArrayExpress data submitters provided ChEBI
identifiers), is a significant step forward and should immedi-
ately benefit from incorporation of the DSSTox ArrayExpress
chemical-experiment index file, as well as the addition of the
corresponding DSSTox GEO index file. However, as is
apparent from past failures, it is not sufficient to recommend
that users add accurate chemical information at the time of data
submission unless more stringent efforts to require this
information are instituted. In addition, we strongly recommend
adoption of the "Chemical_StudyType" categories, or some-
thing comparable, for each chemical-associated study or
experiment deposited into GEO and ArrayExpress. Finally,
recognizing that GEO and ArrayExpress are not designed
primarily as toxicogenomics resources, submitters of explicit
toxicogenomic study data should be strongly encouraged to
Previous
TOC
-------
370
WILLIAMS-DEVANE, WOLF, AND RICHARD
initially deposit studies into CEBS as a way to ensure capture
of sufficient toxicogenomics experimental description, utilizing
the automated deposition capabilities of CEBS to secondarily
deposit well annotated, chemically indexed data into GEO.
Postscript: All of the initial published DSSTox chemical
files and results reported here were based on data extracted
from the ArrayExpress Repository and GEO Series on
September 20, 2008. Subsequent updates of both ArrayExpress
Repository and GEO Series chemical-index files, based on data
extracted on January 20, 2009 and February 2, 2009,
respectively, have been published on the DSSTox website
and incorporated into PubChem as of March 2009; these
updated files do not change the overall trends or conclusions of
the present study.
SUPPLEMENTARY DATA
Supplementary data are available online at http://toxsci.
oxfordjournals.org/.
FUNDING
NCSU/EPA Cooperative Training Program in Environmen-
tal Sciences Research, Training Agreement (CT833235-01-0)
with North Carolina State University supported C.R.W.
ACKNOWLEDGMENTS
We would like to thank Drs Jennifer Fostel (CEBS), Chihae
Yang (FDA Center for Food Safety and Nutrition), David Dix
(EPA), and William Ward (EPA) for helpful comments and
suggestions in review of this manuscript. This work was
carried out by C.R.W. as part of a graduate research project
within the Bioinformatics Program at North Carolina State
University; thesis is publicly accessible at http://www.lib.
ncsu.edu/theses/available/etd-12112008-214342/.
REFERENCES
Ball, C., Brazma, A., Causton, H., Chervitz, S., Edgar, R., Hingamp, P.,
Matese, J. C., Icahn, C., Parkinson, H., Quackenbush, J., at al. (2004).
Microarray Gene Expression Data (MGED) Society. Standards for micro-
array data: An open letter. Environ. Health Perspect. 112, A666-A667.
Barrett, T., Troup, D. B., Wilhite, S. E., Ledoux, P., Rudnev, D.,
Evangelista, C., Kim, I. F., Soboleva, A., Tomashevsky, M., and
Edgar, R. (2007). NCBI GEO: Mining tens of millions of expression
profiles—Database and tools update. Nucleic Acids Res. 35(Database issue),
D760-D765.
Barrett, T., and Edgar, R. (2006). Gene expression omnibus: Microarray data
storage, submission, retrieval, and analysis. Methods Enzymol. 411,
352-369.
Boyle, J. (2005). Gene-Expression Omnibus integration and clustering tools in
SeqExpress. Bioinformatics 21, 2550-2551.
Brazma, A., Kapushesky, M., Parkinson, H., Sarkans, U., and Shojatalab, M.
(2006). Data storage and analysis in ArrayExpress. Methods Enzymol. 411,
370-386.
Burgoon, L. D. (2007). Clearing the standards landscape: The semantics of
terminology and their impact on toxicogenomics. Toxicol. Sci. 99, 403^1-12.
Burgoon, L. D., Boutros, P. C., Dere, E., and Zacharewski, T. R. (2006).
dbZach: A MIAME-compliant toxicogenomic supportive relational database.
Toxicol. Sci. 90, 558-568.
Chen, J., Zhao, P., Massaro, D., Clerch, L. B., Almon, R. R., DuBois, D. C.,
Jusko, W. J., and Hoffman, E. P. (2004). The PEPR GeneChip data
warehouse, and implementation of a dynamic time series query tool (SGQT)
with graphical interface. Nucleic Acids Res. 32(Database issue), D578-D581.
Davis, A. P., Murphy, C. G., Saraceni-Richards, C. A., Rosenstein, M. C.,
Wiegers, T. C., and Mattingly, C. J. (2009). Comparative toxicogenomics
database: A knowledgebase and discovery tool for chemical-gene-disease
networks. Nucleic Acids Res. 37(Database issue), D786-D792.
Dix, D. J., Houck, K. A., Martin, M. T., Richard, A. M., Setzer, R. W., and
Kavlock, R. J. (2007). The ToxCast program for prioritizing toxicity testing
of environmental chemicals. Toxicol. Sci. 95, 5-12.
Fielden, M. R., and Kolaja, K. L. (2006). The state-of-the-art in predictive
toxicogenomics. Curr. Opin. Drug Discov. Dev. 9, 84-91.
Fostel, J. M. (2008). Towards standards for data exchange and integration and
their impact on a public database such as CEBS (Chemical Effects in
Biological Systems). Toxicol. Appl. Pharmacol. 233, 54-62.
Fostel, J. M., Burgoon, L., Zwickl, C., Lord, P., Gorton, J. C., Bushel, P. R.,
Cunningham, M., Fan, L., Edwards, S. W., Hester, S., et al. (2007). Toward
a checklist for exchange and interpretation of data from a toxicology study.
Toxicol. Sci. 99, 26-34.
Fostel, J., Choi, D., Zwickl, C., Morrison, N., Rashid, A., Hasan, A., Bao, W.,
Richard, A., Tong, W., Bushel, P., et al. (2005). Chemical Effects in
Biological Systems—Data dictionary (CEBS-DD): A compendium of terms
for the capture and integration of biological study design description,
conventional phenotypes, and 'omics data. Toxicol. Sci. 88, 585-601.
Ganter, B., Tugendreich, S., Pearson, C. L, Ayanoglu, E., Baumhueter, S.,
Bostian, K. A., Brady, L., Browne, L. J., Calvin, J. T., Day, G. J., et al.
(2005). Development of a large-scale chemogenomics database to improve
drug candidate selection and to understand mechanisms of chemical toxicity
and action. /. Biotechnol. 119, 219-244.
Gomase, V. S., Tagore, S., and Kale, K. V. (2008). Microarray: An approach
for current drug targets. Curr. Drug Metab. 9, 221-231.
Hamadeh, H. K., Amin, R. P., Paules, R. S., and Afshari, C. A. (2002). An
overview of toxicogenomics. Curr. Issues Mol. Biol. 4, 45-56.
Hayes, K. R., Vollrath, A. L., Zastrow, G. M., McMillan, B. J., Craven, M.,
Jovanovich, S., Rank, D. R., Penn, S., Walisser, J. A., Reddy, J. K., et al.
(2005). EDGE: A centralized resource for the comparison, analysis, and
distribution of toxicogenomic information. Mol. Pharmacol. 67, 1360-1368.
Hirabayashi, Y., and Inoue, T. (2002). Toxicogenomics—A new paradigm of
toxicology and birth of reverse toxicology. Kokuritsu lyakuhin Shokuhin
Eisei Kenkyusho Hokoku 120, 39-52.
Ivliev, A. E., 't Hoen, P. A., Villerius, M. P., den Dunnen, J. T., and Brandt, B. W.
(2008). Microarray retriever: A web-based tool for searching and large scale
retrieval of public microarray data. Nucleic Acids Res. 36, W327-W331.
Larsson, O., and Sandberg, R. (2006). Lack of correct data format and
comparability limits future integrative microarray research. Nat. Biotechnol.
24, 1322-1323.
Martin, M. T., Judson, R. S., Reif, D. M., and Dix, D. J. (2009). Profiling
chemicals based on chronic toxicity results from the U. S. EPA ToxRef
Database. Environ. Health Perspect. 117, 392-399.
Mattes, W. B., Pettit, S. D., Sansone, S. A., Bushel, P. R., and Waters, M. D.
(2004). Database development in toxicogenomics: Issues and efforts.
Environ. Health Perspect. 112, 495-505.
Previous
TOC
-------
CHEMICAL INDEXING OF TOXICOGENOMICS RESOURCES
371
Parkinson, H., Kapushesky, M., Shojatalab, M., Abeygunawardena, N.,
Coulson, R., Fame, A., Holloway, E., Kolesnykov, N., Lilja, P., Lukk, M.,
et al. (2007). ArrayExpress—A public database of microarray experiments
and gene expression profiles. Nucleic Acids Res. 35(Database issue),
D747-D750.
Richard, A., Yang, C., and Judson, R. (2008). Toxicity data informatics:
Supporting a new paradigm for toxicity prediction. Toxicol. Mech. Methods
18, 103-118.
Richard, A. M., Gold, L. S., and Nicklaus, M. C. (2006). Chemical structure
indexing of toxicity data on the internet: Moving toward a flat world. Curr.
Opin. Drug Discov. Dev. 9, 314-325.
Rustici, G., Kapushesky, M., Kolesnikov, N., Parkinson, H., Sarkans, U., and
Brazma, A. (2008). Data storage and analysis in ArrayExpress and
expression profiler. Curr. Protoc. Bioinformatics. Chap. 7, Unit 7.13.
Salter, A. H. (2005). Large-scale databases in toxicogenomics. Pharmacoge-
nomics 6, 749-754.
Tateno, Y., and Ikeo, K. (2004). International public gene expression database
(CIBEX) and data submission. Tanpakushitsu Kakusan Koso 49,
2678-2683.
Taylor, C. F., Field, D., Sansone, S. A., Aerts, J., Apweiler, R., Ashburner, M.,
Ball, C. A., Binz, P. A., Bogue, M., Booth, T., et al. (2008). Promoting
coherent minimum reporting guidelines for biological and biomedical
investigations: The MIBBI project. Nat. Biotechnol 26, 889-896.
Waters, M., Boorman, G., Bushel, P., Cunningham, M., Irwin, R., Merrick, A.,
Olden, K., Paules, R., Selkirk, J., Stasiewicz, S., et al. (2003). Systems
toxicology and the chemical effects in biological systems (CEBS) knowledge
base. Environ. Health Perspect. Toxicogenomics 111, 15-28.
Waters, M., Stasiewicz, S., Merrick, B. A., Tomer, K., Bushel, P., Paules, R.,
Stegman, N., Nehls, G., Yost, K. J., Johnson, C. H., et al. (2008).
CEBS—Chemical effects in biological systems: A public data repository
integrating study design and toxicity data with microarray and proteomics
data. Nucleic Acids Res. 36(Database issue), D892-D900.
Wheeler, D. L., Barrett, T., Benson, D. A., Bryant, S. H., Canese, K.,
Chetvernin, V., Church, D. M., Dicuccio, M., Edgar, R., Federhen, S., et al.
(2008). Database resources of the National Center for Biotechnology
Information. Nucleic Acids Res. 36(Database issue), D13-D21.
Williams-Devane, C. R., Wolf, M. A., and Richard, A. M. (2009). DSSTox
Chemical-Index files for exposure-related experiments in ArrayExpress and
Gene Expression Omnibus: Enabling toxico-chemogenomics data linkages.
Bioinformatics 25, 692-694.
Yang, C., Benz, R. D., and Cheeseman, M. A. (2006a). Landscape of current
toxicity databases and database standards. Curr. Opin. Drug Discov. Dev. 9,
124-133.
Yang, C., Hasselgren, C. H., Boyer, S., Arvidson, K., Aveston, S., Diekes, P.,
Benigni, R., Benz, R. D., Contrera, J., Kruhlak, N. L., et al. (2008).
Understanding genetic toxicity through data mining: The process of building
knowledge by integrating multiple genetic toxicity databases. Toxicol. Mech.
Methods 18, 277-295.
Yang, C., Richard, A. M., and Cross, K. P. (2006b). The art of data mining the
minefields of toxicity databases to link chemistry to biology. Curr. Comput.
Aided Drug Design 2, 135-150.
Zhu, Y., Zhu, Y., and Xu, W. (2008). EzArray: A web-based highly automated
Affymetrix expression array data management and analysis system. BMC
Bioinformatics 9, 46.
Previous
TOC
-------
Toxicology Mechanisms and Methods, 18:103-118, 2 008
Copyright© Informa Healthcare USA, Inc.
ISSN: 1537-6516 print; 1537-6524 online
DOI: 10.1080/15376510701857452
informa
healthcare
REVIEW
Toxicity Data Informatics: Supporting a New Paradigm
for Toxicity Prediction
Ann M. Richard
National Center for Computational
Toxicology, U.S. Environmental
Protection Agency, Research
Triangle Park, NC 27711
Chihae Yang
Leadscope, Inc., Columbus,
OH 43235
Richard S. Judson
National Center for Computational
Toxicology, U.S. Environmental
Protection AgencyResearch Triangle
Park, NC 27711
Received 9 November 2008;
accepted 21 November 2008.
One of the authors (AR) thanks Maritja
Wolf and Thomas Transue (Lockheed
Martin, Contractors to the U.S. EPA) for
major technical support of the DSSTox
project and structure-browser, and Lois
Gold for continuing to expand the
exceedingly valuable, publicly available
Carcinogenic Potency Database. We wish
to further acknowledge the significant
efforts of Matthew Martin and David Dix
in development of the EPA's ToxRef DB
database, and Elizabeth Julien for her
leadership of the ILSI Developmental
Toxicity Database Project. Lastly, we
thank David DeMarini and Matthew
Martin for contributing valuable
suggestions and references in review of
the manuscript.
Disclaimer. This manuscript has been
reviewed by the U.S. EPA's National
Center for Computational Toxicology
and approved for publication. Approval
does not signify that the contents
necessarily reflect the views and policies
of the agency, nor does mention of trade
names or commercial products constitute
endorsement or recommendation for
use.
Address Correspondence to Ann Richard,
109 TW Alexander Dr., Mail Drop
D343-03, U.S. EPA, RTP, NC 27711. E-mail:
richard.ann@epa.gov
ABSTRACT Chemical toxicity data at all levels of description, from
treatment-level dose response data to a high-level summarized toxicity
"endpoint," effectively circumscribe, enable, and limit predictive toxicology
approaches and capabilities. Several new and evolving public data initiatives
focused on the world of chemical toxicity information—as represented here
by ToxML (Toxicology XML standard), DSSTox (Distributed Structure-
Searchable Toxicity Database Network), and ACToR (Aggregated Computa-
tional Toxicology Resource)—are contributing to the creation of a more unified,
mineable, and modelable landscape of public toxicity data. These projects
address different layers in the spectrum of toxicological data representation
and detail and, additionally, span diverse domains of toxicology and chemistry
in relation to industry and environmental regulatory concerns. For each of the
three projects, data standards are the key to enabling "read-across" in relation
to toxicity data and chemical-indexed information. In turn, "read-across"
capability enables flexible data mining, as well as meaningful aggregation
of lower levels of toxicity information to summarized, modelable endpoints
spanning sufficient areas of chemical space for building predictive models.
By means of shared data standards and transparent and flexible rules for data
aggregation, these and related public data initiatives are effectively spanning
the divides among experimental toxicologists, computational modelers, and
the world of chemically indexed, publicly available toxicity information.
KEYWORDS ACToR; Data Models; DSSTox; Predictive Toxicology; SAR; Structure-Activity
Relationships; Toxicoinformatics; ToxML
INTRODUCTION
Computational predictive toxicology methods applied to screening chemicals for
potential toxicity must be built on the scaffolding of existing data. The goal is to glean
sufficiently predictive patterns and inferences to be able to extrapolate from data-rich
chemicals to chemicals and classes that are data poor. If a deductive predictive method is
fully computational (or in silico; i.e., it requires no a priori biological measurement data),
then toxicity endpoints are predicted based solely on properties derived from the molecular
structure of a chemical of concern. On the other hand, if it is possible and practical to
generate intermediate biological measurements at low cost relative to a benchmark in vivo
experiment, then these biological measures can be used to augment chemical structure in a
more encompassing predictive computational toxicology approach (Richard 2006). In either
case, a defined toxicological "endpoint" must be chosen for which sufficient existing data are
available for training and testing of candidate prediction models. These legacy data, which
103
Previous
TOC
-------
often include supporting layers of experimental detail and
biological description, serve as bookends to the problem of
prediction. On the one hand, models must be anchored to
a toxicological "endpoint" with relevance to product safety
or environmental regulatory concerns. On the other hand,
less summarized toxicological data can provide more nuanced
descriptions of activity that can better align with underlying
biological mechanisms and productively inform and guide
model development.
Issues pertaining to toxicity data availability, quality, com-
parability, and representation profoundly impact our ability to
derive and apply useful prediction models and have slowed
the development of robust prediction models in many areas of
toxicology. In this paper, we will consider three public initiatives
that are directly addressing the "data" arm of predictive
toxicology: toxicity data models as represented by the ToxML
(Toxicology XML standard) publicly available schema, with
ontology-based fields and controlled vocabulary (ToxML 2007);
the U.S. Environmental Protection Agency's (EPA's) DSSTox
(Distributed Structure-Searchable Toxicity) Database Network
(EPA DSSTox 2007) that publishes summarized structure-
annotated toxicity data files for use in structure-activity rela-
tionship (SAR) modeling, and structure-index files for public
toxicity data inventories; and EPA's new ACToR (Aggregated
Computational Toxicology Resource) data integration system
slated for public release in 2008 (Judson et al. submitted) that
will broadly index and house existing chemical toxicity data as
well as new data being generated in EPA's ToxCast program
(Dix et al. 2007; EPA ToxCast 2007). These initiatives are
focusing on four fundamental and interrelated problems: (1)
the electronic capture and standardization of existing or legacy
toxicology experimental data down to the dose-treatment level
(i.e., the domain of experimental toxicologists); (2) meaningful
aggregation and representation of toxicity data into forms
that are easily understood and employed by modelers; (3) the
creation of bidirectional linkages between previously isolated or
"siloed" sources of toxicity data and the larger world of data in-
formatics; and (4) the central indexing of chemicals and publicly
available toxicity data, particularly in relation to environmental
regulations and industrial use, across the multitude of potential
sources on the Internet. Additionally, these three initiatives
are attempting to address the questions: "How are toxicity
data best organized to facilitate exploration and data mining?";
"How do we engage toxicologists (i.e., the domain experts), in
these efforts?"; "How do we meaningfully aggregate data and
interface toxicologists and modelers?"; "Where are the data?";
"Which chemicals should be of highest regulatory interest?";
and, finally, "Which chemicals have the richest foundation of
data for anchoring new predictive technologies?"
TOXICITY DATA MODELS: WHAT'S
IN IT FOR MODELERS?
In a recent review, Yang and coauthors (Yang et al.
2006a) surveyed the "landscape of current toxicity databases
and database standards." These authors noted that what is
generally perceived as a "lack of data" problem encompasses the
fragmented, disparate state of existing data, much of which do
not exist in electronic form or exist in diverse nonexchangeable
data formats. These problems span both public (e.g., literature,
government regulatory agencies) and nonpublic (i.e., corporate
A. M. Richard et al.
or proprietary) toxicity data sources. The authors argued that the
path forward must involve the development and adoption of
shared public standards for toxicity databases. These standards
should include controlled vocabulary and hierarchical data
relationships or ontologies (i.e., using the same terminology to
describe the same things, and incorporating the layered relation-
ships of different terms to one another), should be derived from
close working knowledge of the toxicity study domain (e.g.,
carcinogenicity, developmental toxicity, immunotoxicity, neu-
rotoxicity, etc.), and should be inclusive of chemical structure.
The authors cited a number of public efforts that are moving in
this direction. The "Birth Defects System Manager," is an exem-
plary example of a well-designed computational-bioinformatics
infrastructure for standardizing available toxicity data, spanning
the biological interaction spectrum, to enable exploration of
these data from a mechanistic and systems biology perspective
(Singh et al. 2005). The public ToxML initiative (ToxML 2007),
on the other hand, has as its clear ultimate objective the
enrichment and enlargement of chemical structure-mineable in-
formation in relation to broadly encompassing and meaningful
representations of toxicology experiments.
A second paper (Yang et al. 2006b) elaborated on the state
of the existing public toxicity databases and data sources,
particularly in relation to their suitability and preparedness
for relational "read-across" data mining and structure-based
modeling. (The term "relational read-across" refers to the ability
to broadly query a database using controlled vocabulary and
standardized search terms, and to combine search terms in
such a way as to define conditional statements [i.e., using
AND, OR, NOT].) This paper laid out in greater detail the
requirements for such a capability, and a general strategy to data
mine structure-integrated toxicity databases. This was illustrated
with an example that focused on target organ specificity of
chronic toxicity results, which involved iterative probing of the
chemical domain with preindexed toxicity endpoint descriptors,
and complementary probing of the biological domain with
chemical descriptors. A third paper, published elsewhere in this
issue (Yang et al., 2007), extends these themes into the realm
of practical application, employing the commercially available
Leadscope SAR-ready Genetic Toxicity Database (Leadscope,
Inc. 2007), built using the public ToxML data model schema,
to illustrate how new insights into the nature of chemical-
biological domains for various genetic toxicity endpoints can
be extracted from a well-constructed, populated data model.
ToxML was an early and influential entrant to the field of
toxicity data models, encompassing a wide domain of toxicity
experiments, including genetic toxicity, chronic/subchronic,
and reproductive and developmental studies. Under a cooper-
ative research and development agreement with the U.S. Food
and Drug Administration (FDA), Leadscope has built databases
enriched with pharmaceuticals from publicly available data
within the FDA's Center for Drug Evaluation and Research
(CDER) and food ingredients from the Center for Food Safety
and Applied Nutrition (CFSAN). At the time of this writing,
eight such databases are commercially available (Leadscope,
Inc. 2007). The ToxML methodology, as implemented within
Leadscope, also includes aggregation criteria to construct higher-
level (i.e., more summarized) endpoints for data mining and
SAR analysis. Figure 1 shows an aggregated list of studies
within Leadscope for the compound "Atrazine," CASRN
(Chemical Abstracts Service Registry Number) [1912-24-9],
spanning several Study Types, Sources, Species, etc.
104
Previous
TOC
-------
• F3 Study Table
Add|Remove Columns Sort Export Table
Each row represents a Test (part of a Study).
CA5RN Study Type Study
Source
1 1912-24-9 bacterial mutagenesis ntp
1912-24-9 bacterial mutagenesis ntp
1912-24-9 bacterial mutagenesis ntp
1912-24-9 bacterial mutagenesis ntp
1912-24-9 bacterial mutagenesis ntp
1912-24-9 bacterial mutagenesis ntp
1912-24-9 bacterial mutagenesis ntp
1912-24-9 carcinogenicity dsscpdb
1912-24-9 bacterial mutagenesis dsscpdb
1912-24-9 carcinogenicity dsscpdb
1912-24-9 carcinogenicity dsscpdb
1912-24-9 carcinogenicity dsscpdb
1912-24-9 irritation rtecs
1912-24-9 irritation rtecs
1912-24-9 acute toxicity rtecs
1912-24-9 multiple dose rtecs
1 191 2-24-9 acute toxicity rtecs
1912-24-9 RTECS mutation rtecs
:= (
Print Print Preview Print PDF ,
Species Strain
Salmonella typhimunum TA98
Salmonella typhimunum TA97
Salmonella typhimurium TA 1 537
Salmonella typhimurium TA98
Salmonella typhimurium TA98
Salmonella typhimurium TA98
Salmonella typhimurium TA98
Rat
Salmonella typhimurium
Rat
Mouse
Mouse
Mammal - species unsp, , ,
Rabbit
Rat
Rat
Mouse
Mouse
,,,,L^ &..•,»
1 ' F-inH •• i ' " " - ' t
59 Type Sax
male Sprague Dawley rat liver 59
male Sprague Dawley rat liver 59
male Sprague Dawley rat liver 59
male Syrian hamster liver S9
male Syrian hamster liver 59
male Sprague Dawley rat liver 59
Male
Female
Male
Female
• - i dose
Route Of All Doses
Exposure
10.0; 33.0; ... _«j
33.0; 100,0;...
33.0; 100,0;...
10.0; 33,0; ...
33.0; 100.0;... i
100,0; 333,0.. ,_|
10,0; 33,0; ...
31.7mg/Ng
31,7mg/kg
Eyes lOO.Omg
Eyes 6.32 mg
Inhalation 5,2 micro-g/mL
Inhalation 5.494SOS2E-,,,
Intraperitoneal 626,0 mg/Jag
Intraperitoneal 500,0 mg/kg j*J
Ready |~~
FIGURE 1 Sample view of Leadscope Study-level aggregation of data across all component databases for Atrazine, built on the ToxM
data model.
Several additional public efforts are under way to con-
struct toxicity data models for different purposes and areas
of toxicology, with a goal to make both the data and
schema publicly available. The International Life Sciences
Institute's (ILSI's) Developmental Toxicity Database Project
(http://rsi.ilsi.org/devtoxsar.htm) evolved from efforts to eval-
uate the state of SAR prediction models for developmental
toxicity (Julien et al. 2004). That evaluation concluded that
SAR models could be substantially improved by refining the
ways in which toxicity data are used to train the models
(e.g., use data on specific endpoints, which can be combined
into biologically meaningful categories). To begin to address
this need, a follow-up effort brought together prominent
developmental toxicologists and modelers to create a public
data model directed toward capturing data from the disparate
developmental toxicology literature.
Within EPA, the ToxRefDB data model (Martin et al. 2007)
has been constructed to house reference in vivo toxicology
data across a variety of toxicity domains (including acute,
chronic, subchronic, developmental, and reproductive toxicity)
in support of the ToxCast Program (Dix et al. 2007). ToxCast is
an EPA research initiative with the goal of building predictive
toxicity models based on a combination of in vitro and in silico
approaches. The program is generating a large data set that
will include results from hundreds of in vitro high-throughput
screening (HTS) assays as well as whole-genome transcription
profiles of several hundred chemicals for which detailed in vivo
toxicology data are available (EPA ToxCast 2007). ToxRefDB is
housing these reference in vivo data, which are being extracted
primarily from EPA Pesticide Data Evaluation Records (DERs)
(EPA Regulating Pesticides 2007) for hundreds of registered
pesticidal active ingredients (Martin et al. 2007). The DERs
summarize data from standardized guideline studies that are
required to be run for all new food-use pesticides. Within the
ToxCast Program, these data will serve as phenotypic anchors
for deriving predictive "signatures" and models from chemical
structure properties and newly generated HTS data (Dix et al.
2007; EPA ToxCast 2007).
Both the ILSI Developmental Toxicity Database Project and
ToxRefDB have incorporated elements of the public ToxML
data model and are striving for broad chemical coverage in
their content, which is a key requirement for structure-based
data mining and modeling applications. In addition, efforts are
under way to ensure that the final data models resulting from
these various efforts (ToxML, ILSI, ToxRefDB), despite having
somewhat different target data and objectives, have sufficient
compatibility for interoperability and data merging, both of
which are absolutely essential for meeting the goal of "read-
across" in the larger data world.
These new toxicity data models provide the means, and
the data entry tools provide the mechanism, for migrating
previously inaccessible data and new data into a standard-
ized, relational format. Since these toxicity data models are
designed to capture experimentally relevant details pertaining
to dose treatment and effects at the study level, they have
the potential to engage and be used by toxicologists, unlike
the highly (some would say overly) summarized toxicity data
representations typically used in SAR modeling (e.g., yes
or no calls for carcinogenicity, teratogenicity, genotoxicity,
etc.). SAR modelers, on the other hand, have difficulty
processing this level of experimental detail. Instead, they are
more interested in obtaining statistically sufficient chemical
coverage in relation to a defined toxicity endpoint, which
is a requirement that most often mandates high levels of
toxicity data aggregation and summarization. Hence, modelers
generally rely upon toxicity domain experts and regulators to
105
Toxicity Data Informatics
Previous
TOC
-------
CompoundSummaryCall I Multiple data sources: BacterialMutagenesis
L— • " All Strains (w/or w/o S9)
-------
STRUCTURE
STRUCTURE Shown
DSSTox_CID
STRUCTURE Formula
- tested chemical,
- genera! form of chemical,
- active ingredient of formulation,
- representative isomer in mixture,
- representative comporient in mixture,
- monomer of polymer
, simplified to parent
STRUCTURE_MoiecularWeight
STRUCTURE_ChemicalType
STRUCTURE_TestedForm_D
efinedOrganic
DSSTox Generic SID
TestSubstance ChemicalName
TestSubstance CASRN
TestSubstance_Description
• single chemical compound
• rnacrornolecule
mixture or formulation
unspecified or multiple forms
defined organic
inorganic
organometallic
ChemicalNote
STRUCTURE_ChemicalN
ameJUPAC
STRUCTURE_SMILES
STRUCTURE_Parent_SMILES
STRUCTURE InChl
- parent,
- salt Na, Cl, etc
- complex HCI, H20,
mesylate, etc
description of mixture
nature & CAS of components
tautorners
stereochemistry
DSSTox RID
DSSTox FilelD
FIGURE 3 DSSTox Standard Chemical Fields included in all published DSSTox data files, with the STRUCTURE-related fields
automatically generated from the STRUCTURE field, and the STRUCTURE.Shown field indicating the relationship to TestSubstance
fields.
coverage for models. However, this high-level endpoint summa-
rization is not generally endorsed by toxicologists, particularly
for complex endpoints such as developmental toxicity (Julien
et al. 2004), and can effectively obfuscate meaningful patterns or
associations of chemicals aligned with biological mechanisms.
As toxicity data models for study areas such as these become
increasingly populated and available, modelers and toxicologists
alike will be presented with a large and varied array of new
possibilities for defining groups or "profiles" of effects that can
constitute intermediate endpoints and anchors for structure-
based modeling and exploration (Yang et al. 2006b). The
ease and flexibility conferred by toxicity data models in how
aggregated toxicity endpoints are defined, and structure-indexed
data are extracted, will add an increasingly valuable degree of
freedom to predictive modeling.
DSSTox DATABASE NETWORK:
ENDPOINTS AND DATA LINKAGES
A primary goal of EPA's DSSTox project (Richard and
Williams 2002a; Richard 2004; EPA DSSTox 2007) is to publish
structure-annotated toxicity data files for use in structure-
activity modeling. Whereas toxicity data models focus on
the standardization and "read-across" of experimental toxicity
data fields, DSSTox is contributing the standardized chemical
description layer to these efforts and providing "read-across"
in chemical structure space. A major focus of this effort has
been on the quality review of the chemical annotation and
representation with respect to the toxicity information (EPA
DSSTox 2007). In order to accommodate and represent the
diverse chemical content of public toxicity databases and inven-
tories, DSSTox Standard Chemical Fields (Fig. 3) also make a
clear distinction between the assigned chemical structure and its
relationship to the actual substance tested, which can be a single
compound, or a defined mixture, polymer, macromolecule,
or active ingredient in a formulation. In addition, the entire
DSSTox chemical inventory is now centrally indexed by unique
record ID, generic substance ID, and chemical (i.e., structure)
ID to better serve data management needs, as well as to be
100% compatible with indexing of the large and growing
PubChem inventory (NCBI PubChem 2007), which currently
exceeds 7 million compounds, as well as with ACToR, which
incorporates the entire PubChem inventory into its even larger
aggregated inventory. Currently, there are 11 published DSSTox
structure-data (SD) files available for download, spanning nearly
7,000 unique substances (EPA DSSTox 2007).
107
Toxicity Data Informatics
Previous
TOG
Next
-------
Chemical Structures
Toxiclty
Model Schema
Toxicity Data
DSSTox Summary
Toxicity Data Files
Compound
Cheml
Chem2
ChernS
Chem4
Chem5
Chein6
Chein?
ChemB
Tox1
rat
Tox2
male
Tox3
+
Tox4
lung
FIGURE 4 Representation of structure-based toxicity data mining as exploring relationships between corresponding layers in the
chemistry and toxicity data model domains, fed by experimental toxicity data, and with DSSTox data files represented as flat tabular
slices of summary toxicity data in corresponding chemical space.
A number of published DSSTox data files include the sorts
of highly summarized "toxicity endpoints" that have been the
traditional focus of SAR prediction modeling efforts. Since
its inception, the DSSTox project additionally has sought to
provide an enhanced level of toxicity endpoint description
and annotation to encourage alternative modeling strategies
and use of the data files in structure-searching and chemical
relational database applications (Richard 2004; Benigni et al.
2007b). Although not approaching the level of experimental
detail captured in the toxicity data models discussed above,
each summarized toxicity endpoint can be considered as a
high-level "slice" of a hierarchical or layered toxicity data
model, with less summarized endpoints extracted from lower
levels of the data model. An example, with correspondingly
smaller chemical coverage of actives, is a "carcinogenicity call"
defined by tumor incidence in a particular target organ (e.g.,
liver) and rodent species (e.g., rat). The analogy of a slice is
apt, since DSSTox structure-data files (SD files) are "flat" files,
representing summary data in a single-layer, tabular spreadsheet
form (Fig. 4), whereas toxicity data models are hierarchical in or-
ganization, with multiple nested layers (Fig. 2) accommodating
increasingly detailed description (Yang et al. 2006b). In addition
to data aggregation, this data transformation, or "flattening,"
is a required step to effectively interface toxicity data with
SAR modelers (Richard and Williams 2002b), as well as with
other types of prediction approaches and structure-searchable
bioassay data resources, such as PubChem (NCBI PubChem
2007).
Issues pertaining to varied representations of summa-
rized toxicity data in relation to rodent carcinogenicity
will be illustrated by consideration of the recently updated
CPDB Summary Tables (CPDB Summary Tables 2007;
Gold et al. 1997, 1999) and the corresponding DSSTox
CPDBAS (Carcinogenic Potency Database—All Species) data
file, CPDBAS_v5a_1547_25Oct2007 (Gold et al. 2007) contain-
ing data for 1547 chemical substances, or records. In contrast
to a detailed toxicity data model that captures data at the study
level (Fig. 2), each column of the published CPDB Summary
Tables represents many layers of aggregation and consensus,
with chronic bioassay study results pooled from the results of
published literature studies and government research institute
publications (NTP and NCI) that meet broadly defined criteria
for CPDB inclusion (CPDB 2007; Gold et al. 1997). Target site
results are pooled to the level of species/sex, and TD50 potency
values are computed for pooled species results. There remains,
however, potentially rich descriptive "read-across" information
in the CPDB Summary Tables with respect to compound,
genotoxicity, species, sex, target sites, and consensus potency.
The DSSTox CPDBAS file is an enhanced version of the CPDB
Summary Tables that includes DSSTox Standard Chemical
Fields and 32 DSSTox Source-specific toxicity content fields
that are designed to facilitate SAR modeling, data mining,
structure searching, and relational text searching (Table 1). Half
of the 32 DSSTox Source-specific toxicity content fields listed
in Table 1 closely correspond to entries in the published CPDB
Summary Tables (CPDB Summary Tables 2007), whereas the
remaining fields represent added or modified content to the
DSSTox file. The latter include molar unit fields for TD50
potency values, text note fields containing expanded footnotes
from the original CPDB tables, DSSTox notes pertaining to
A. M. Richard et al.
108
Previous
TOC
-------
TABLE 1 List of DSSTox Source-specific toxicity data fields for Carcinogenic Potency Database—All Species (CPDBAS)
CPDBAS_v5a_1547_25Oct2007a
Values
Modifications to CPDB entry Nonblank entries*
Mutagenicity.SAL.CPDB
TD50.Rat.mg
TD50_Rat_mmol
TD50.Rat.Note
TargetSites.Rat_Malec
TargetSites.Rat.Female
TargetSites_Rat_BothSexes
TD50.Mouse.mg
TD50.Mouse.mmol
TD50.Mouse.Note
TargetSites_Mouse_Male
TargetSites_Mouse_Female
TargetSites.Mouse.BothSexes
TD50.Hamster.mg
TD50.Hamster.mmol
TD50_Hamster_Note
TargetSites.Hamster.Male
TargetSites.Hamster.Female
TargetSites.Hamster.BothSexes
TD50.Dog.mg
TargetSites_Dog
TD50_Rhesus_mg
TargetSites.Rhesus
TD50.Cynomolgus.mg
TargetSites.Cynomolgus
TD50_Dog_Rhesus_Cynomolgus_Note
ActivityCategory.SingleCellCall
ActivityCategory.MultiCellCall
ActivityCategory.MultiCellCalLDetails
Note.CPDBAS
NTP.TechnicalReport
ChemicalPage_URL
"Positive," "negative"
Numeric
Numeric
Defined text
Defined text
Defined text
Defined text
Numeric
Numeric
Defined text
Defined text
Defined text
Defined text
Numeric
Numeric
Defined text
Defined text
Defined text
Defined text
Numeric
Defined text
Numeric
Defined text
Numeric
Defined text
Defined text
"1" or"0"
"1" or"0"
"Multisite active,"
"multisex active,"
"multispecies active"
"Multisex inactive,"
"multispecies inactive"
Memo—version history
notes
Text
URL to CPDB chemical data
page
CPDB
CPDB—eliminated footnotes
Added to DSSTox file
CPDB—expanded text
of footnotes
CPDB—no abbreviations
CPDB—no abbreviations
CPDB—no abbreviations
CPDB—eliminated footnotes
Added to DSSTox file
CPDB—expanded text
of footnotes
CPDB—no abbreviations
CPDB—no abbreviations
CPDB—no abbreviations
CPDB—eliminated footnotes
Added to DSSTox file
CPDB—expanded text
of footnotes
CPDB—no abbreviations
CPDB—no abbreviations
CPDB—no abbreviations
CPDB—eliminated footnotes
CPDB—no abbreviations
CPDB—eliminated footnotes
CPDB—no abbreviations
CPDB—eliminated footnotes
CPDB—no abbreviations
CPDB—expanded text
of footnotes
Added to DSSTox file
Added to DSSTox file
Added to DSSTox file
Added to DSSTox file
Added to DSSTox file
Added to DSSTox file
860
586
565
1,086
975
59
445
434
929
942
22
45
45
73
67
2
3
5
9
24
10
19
1,544
1,151
1,151
415
1547
aField definitions can be found at: http://www.epa.gov/dsstox/sdf.cpdbas.html #SDFFields.
^Nonblank entries for activity fields include activity measures, as well as indications of "no positive results" out of a total of 1 547 chemical records in
CPDBAS.
cTargetSites field entries are expanded from the three-letter abbreviations used in the CPDB Summary Tables (e.g., bladder, kidney, lung, liver, etc.).
CPDB version updates and affected records, the NTP Technical
Report number where applicable, and a URL (Uniform Re-
source Locator, i.e., website address) field linking to the CPDB
chemical data page summary for each chemical record to enable
structure locating of these web pages. The numbers of nonblank
entries in each data field are listed in Table 1 to convey chemical
coverage.
In addition to the above content fields, a set of three
"ActivityCategory" fields have been added to the DSSTox
CPDBAS data file to offer a variety of summarized carcinogenic
activity representations for possible use in SAR modeling efforts.
The first of these fields, ActivityCategory_SingleCellCall, rep-
resents a low-evidence, conservative carcinogenic call and is the
activity measure most typically employed in past SAR modeling
109
Toxicity Data Informatics
Previous
TOC
-------
TABLE 2 Combinations of "SingleCell" and "MultiCell" activity calls in CPDBAS.vSa, broken down by MultiCellCalLDetails, with each
row representing a candidate set of Model compounds
Call
ActivityCategory. ActivityCategory .MultiCellCalLDetails
SingleCellCall MultiCellCall "Multisite" "Multisex" "Multispecies"
Total Incidences Compound
CPDBAS_v5a Setc
1
1
1
1
Active (1) 1
1
1
1
1
0
0
Inactive (0) 0
0
0
xa
1
1
1
1
1
1
1
1
X
0
0
0
0
X X
V x
x v
X X
V V
V x
x v
V V
Total MultiCellCall
V x
V V
V x
V V
Total MultiCellCall
X
X
X
V
V
V
V
V
Incidences (1/1)
X
X
V
V
Incidences (0/0)
224*
81
113
8
123
27
37
193
582
169
266
15
288
569
A1
A2
A3
A4
A5
A6
A7
A8
A9
11
12
13
14
15
a"X" denotes a negative condition (i.e., condition not met for compartment).
^Number represents total incidence compounds satisfying listed conditions (e.g., in Model A1, 224 compounds have a SingleCellCall = 1, but do
NOT also have a MultiCellCall = 1; i.e., only 1 tumor target site in one species/sex cell is listed for all 224 compounds in this set).
cEach row represents a set of conditions for defining "Active" and "Inactive" compound sets for use in modeling, such that a model can be specified
by pairing of rows as in [A1: 11] for only SingleCellCall results, or [A4-A8:I3,I4] for pooled "multispecies" results.
studies of chemical carcinogenicity (Benigni 1997; Benigni
et al. 2007a). This field has a value of "1" if at least one tumor
site (e.g., bladder, liver, lung, etc.) is listed for any experimental
animal "cell" (e.g., rat male, rat female, mouse male, etc.),
and a value of "0" otherwise. Although attractive to modelers
for its large coverage of chemical space, this SingleCellCall
representation of carcinogenic activity most likely overestimates
carcinogenic risk and obscures a number of other potentially
important carcinogenic activity measures contained within the
data file. With few exceptions, known human carcinogens
produce tumors in multiple test species, usually at multiple
target sites (Ashby and Patton 1993; Tennant 1993). To facilitate
consideration of more broadly weighted measures of carcino-
genic activity in SAR modeling and data mining, two summary
activity fields, ActivityCategory_MultiCellCall and Activity-
Category_MultiCellCall_Details, were added to the DSSTox
CPDBAS data file. The ActivityCategoryJVIultiCellCall field
is assigned a value of "1" if a compound meets one or more
conditions of "multisite," "multisex," or "multispecies" across
the 12 CPBDAS Tai^etSites fields listed in Table 1, and is
assigned a value of "0" only if no tumor sites are reported for
any test cell AND if more than one species/sex cell experiment
was carried out reporting "no positive results." Hence, this field
conveys weighted evidence in support of either an inactive or
active call (1 or 0), as well as weighted evidence of carcinogenic
activity in relation to potential human risk (multisite, multisex,
multispecies).
Table 2 lists a variety of conditions for defining active
(9) and inactive (5) sets of compounds with respect to
carcinogenicity, derived from different combinations of Sin-
gleCellCall and MultiCellCall conditions. These compound
sets provide various views of the carcinogenic activity landscape
within CPDBAS. A particularly interesting observation is that
Compound Set A9, with presumably the most restrictive set of
conditions ("multisite" AND "multisex" AND "multispecies")
has the largest numbers of compounds for any MultiCellCall
active compound set, which supports a nondiscriminating
biological generality to the carcinogenic response for a sig-
nificant fraction of compounds. A number of questions can
be easily posed and explored within CPDBAS by virtue of
the structure-activity-annotation and data organization of this
file, such as: How do the chemical and biological landscapes
of these compound sets differ? Do particular target sites,
or combinations of target sites, cluster preferably in any of
these active sets? Does genotoxic activity cluster differently
in these active sets? Are there distinguishing characteristics
(biological or chemical) of those compounds that are exclusively
SingleCellCall active (e.g., mouse liver carcinogens)? Which
chemical sets have greater biological relevance, or produce
more internally consistent models? In addition, with increasing
standardization across the toxicity data landscape, including
genotoxicity and other data domains, the possibility to bring
added layers of information to bear on these sorts of questions
becomes possible.
In addition to elaborating toxicity data fields within in-
dividual DSSTox data files, the DSSTox Project has moved
in the past few years to publish important toxicity-related
chemical inventories within the EPA and the NTP. These in-
clude "structure-index" files, containing only DSSTox Standard
Chemical Fields, of chemical inventories for new predictive
toxicology initiatives, such as the EPA's ToxCast Program
and the NTP High-Throughput Screening Program (DSSTox
File Names: TOXCST, NTPHTS) (Houck et al. 2007; Smith
et al. 2007). These files are serving to centrally structure-index
bioassay data being generated in the ToxCast and NTP HTS
testing programs, enabling these programs to interface with
PubChem and ACToR, which will house these bioassay data.
In addition, "structure-index locator" files are published for
A. M. Richard et al.
110
Previous
TOC
-------
x*EPA DSETox
SUucture-Browsef VI.03
Search
File Incidences 1 f Search Details
?Help
Output Options
Oueiy
Results Type
Hits
Display
Stiiictuie:
^x
J^^'
r
Exact matches
Substructures
Similarity > 80%
1
2
6
Details
Details
Details
7
I Choose Format _^J Save
Print
7
DSSTox
Substance Similarity Structure
ID Score'. M.itch
Substance
Name
Substance
CASRN Description
Details
pi
N-4,
20112 100 CH3 N=-NH CH3
.,„ .n -| single chemical
compound
CPDBAS
IRtSTR
HPVCS!
TOXCST
27603
96.3
CH3CH3
CHS
1,3.5-Triazine-2.4-diamine. 6-chloro-N-
(1,1-dimetfiyletfiyl)-N'-etfiyl-
5915-41-3
single chemical
compound
HPVCSI
FIGURE 5 View of the DSSTox Structure-Browser resulting from a search initiated from the drawn Atrazine structure, showing Search
Details across DSSTox data files.
on-line toxicity data inventories, including the NTP On-line
Database (NTP On-line Database 2007; Burch et al. 2007) and
the EPA High Production Volume Information System (EPA
HPV 2007; Wolf et al. 2007) (DSSTox File Names: NTPBSI,
HPVISD). These files include URLs to chemical-specific source
data pages to enable a user to be linked directly to on-line
data web pages through a structure search. DSSTox files for the
on-line CPDB Summary Tables (Gold et al. 2007) and the EPA
IRIS (Integrated Risk Information System) Database (Backus
et al. 2007) (DSSTox File Names: CPDBAS, IRISTR) include
both a rich complement of data (over 30 toxicity data fields in
both cases) and URLs for chemical-specific data pages linking
to extensive on-line data resources on the EPA IRIS and the
CPDB source websites (EPA IRIS 2007; CPDB 2007).
With the recent launch of the public DSSTox Structure-
Browser (Transue and Richard 2007), users can now perform
on-line structure searches (exact match/substructure/similarity)
of the entire published DSSTox chemical inventory, with the
option to search or view results by DSSTox data file. The
DSSTox Structure-Browser is built from publicly available tools
and open access source code, and is designed to be easy
to understand and use by regulators and toxicologists, with
hundreds of hot links provided to file names and download
pages, field definitions, help pages, and URL links to database
documentation and Source websites. A specially designed
feature of this browser is that it can be easily called up
(http://www.epa.gov/dsstox_structurebrowser/) from a source-
collaborator website (e.g., CPDB, NTP, IRIS, HPV-IS), and
can be directed by a simple extension to the URL (e.g.,
(http://www.epa.gov/dsstox_structurebrowser/?dbs = NPTBSI)
to confine the structure search to the corresponding published
DSSTox structure-locator file (Transue and Richard 2007).
This effectively delivers local structure-search capability directly
to previously isolated on-line toxicity databases. The NTP
On-line Database (NTP On-line Database 2007) has already
taken advantage of this new capability and for the first
time is offering structure searching through the DSSTox
Structure-Browser directly from its website. In addition, the
EPA's IRIS and HPV-IS websites (EPA IRIS 2007; EPA HPV
2007) are expected to incorporate this feature in the near
future.
Accessing the DSSTox Structure-Browser directly, a user can
perform a name or CASRN text search, or by typing a SMILES
or drawn structure can perform exact/substructure/similarity
searches across the entire published DSSTox chemical inven-
tory, with search results displayed by generic substance ID.
Sample screen shots of structure-search results for "Atrazine"
are shown in Figures 5 and 6. Figure 5 illustrates the structure
"read-across" capability to identify this substance and analog
substances in multiple DSSTox data files. The Substance Results
page shown for IRISTR (Fig. 6) provides a URL link for "EPA
IRIS Chemical Substance Data Page" that takes the user to
the on-line IRIS Quick View Document for "Atrazine" (EPA
IRIS 2007). The current published DSSTox chemical data
inventory will soon be deposited into PubChem (estimated by
March 2007), with data files such as CPDBAS and IRISTR,
111
Toxicity Data Informatics
Previous
TOC
-------
SfOwssjr VT.03
Search
File incidences
Search Details
Substance Results ?Help
CH3—(
IRISTR:
EPA Irtteijinted Risk Information System (IRIS) Striietuie-
Index Locator File (544 retoids)
IRlSTR_v1 a_544_28Jul2007
IRISTR Soitice Website
View IRIS Chemical Substance Data Page
Output Options
[Choose Format^] Save
DSSTox_RID
DSSTox_Gei»eiic_SID
TeslSiilist.iiice Clietiiic.ilN.iiiie
TestSnbstance_CASRN
TestS u bsta it c e_D ese t i pti on
STRUCTURE_Sri»wii
StudyType
Species
Oral_RfD_Assessed
0 1 al_Rf D_Ci iti ca I Eff e €te
0 r al_RfD_m 0_|> e i _kg_d ,iy
Qral_RfD_inni<»l_j>e(_kg_day
Oral_RfD_Notes
Oral_RfD_CoiifIdenee
lnliala»ion_RfC_Asses$ed
liihalatie>n_RfC_CiiticaIEffecls
WtOfEvidence_Caficer_Assessed
WtGfEvideiice_1986GiiideltneCa1egijfies
DrinkiiigWater_P(eeiiisorEffect_TumorType
Inli.il.itioii^UnitRisk Assessfiil
li)lialation_P(eaiisoiEffect_TumotType
Tata I Assessi 1 1 e i its
23892
20112
Atrazine
1912-24-9
single chemical compound
tested chemical
Human Health Exposure Toxicity Review for Risk Assessment
cancer, acute; short-term; sub-chronic; chronic, developmental
rodent; human; dog; rabbit
1
decreased body weight gain; cardiac toxieity; moderate-to-severe dilation right
atrium
0.035 mg/kg-bw/day
1.62 mmol/kg-bw/day
NOAEL (No observed adverse effect level). 3.5 mg/kg-day
High
0
Not assessed under the IRIS program.
0
Not assessed under the IRIS program.
0
Not assessed under the IRIS program.
0
Not assessed under the IRIS program.
1
FIGURE 6 View of the DSSTox Structure-Browser resulting from a search initiated from the drawn Atrazine structure, showing
Substance Results within a particular DSSTox data file.
which contain many columns of summary toxicity information,
deposited under several DSSTox-defined PubChem "assay IDs,"
or AIDs, to take full advantage of new and evolving PubChem
assay search and SAR clustering features. Once deposited,
PubChem users will be provided direct access to the DSSTox
data inventory and chemical data page URLs. In addition,
PubChem CIDs (Compound IDs) will maplil onto DSSTox
CIDs and will enable auto-generation of URLs for PubChem
CID data pages to be directly incorporated into the DSSTox
Structure-Browser. This will allow users to link from the DSSTox
Substance Results page directly to the corresponding PubChem
Compound (CID) results page by the click of a button, provid-
ing easy access to PubChem bioassay results, structure-analog
pages, and PubChem links to data on external sites, such as
TOXNET (NLM TOXNET 2007). A similar CID compatibility
will enable direct linkage from DSSTox Substance Results pages
to the EPA ACToR system, discussed in the next section. Hence,
new structure-searching and locating capabilities are effectively
opening a bidirectional information highway and new data
mining opportunities between previously isolated toxicity data
A. M. Richard et al.
inventories and the larger world of structure-indexed bioassay
information (Richard et al. 2006).
ACToR: WHERE IS THE DATA
AND WHAT TYPES OF DATA ARE
OUT THERE?
The ACToR program (Aggregated Computational Toxicol-
ogy Resource) is tackling the larger objective of surveying
all publicly available chemical toxicity resources of potential
interest, and building tools to allow the construction of toxicity
data sets for chemical structure "read-across" and modeling.
An effort of this kind faces three significant problems. First,
it is often useful to merge different types of data to build
a predictive model (e.g., chemical structure, biochemical and
other in vitro data and in vivo toxicology data); second, these
data are likely to be distributed over a wide range of sources; and,
third, the data in each of those sources is typically organized
in a unique scheme. The ACToR program is taking a broad
112
Previous
TOC
Next
-------
TABLE 3 Assay categories being incorporated into the EPA Aggregated Computational Toxicology Resource (ACToR) database to
allow a user to broadly survey data availability for chemicals of interest
No.
1
2
3
Assay category
Physicochemical
Biochemical
Genomics
Description
Physical and chemical properties (in
vitro and/or in silico)
Biochemical (non-cell-based) (in vitro
and/or in silico)
Gene expression values or signatures
Examples
Molecular weight, logP, boiling point
Enzyme inhibition or receptor binding
constants
Result of in vitro or in vivo microarray
4 Cellular
5 Tissue
6 In vivo toxicology (tabular
primary)
7 In vivo toxicology (study
listing primary)
8 In vivo toxicology (tabular
secondary)
9 In vivo toxicology (summary
calls)
10 In vivo toxicology (summary
report via URL)
Cell-based assay
Tissue slice assays
Tabulated results from primary
animal-based studies of chemical
effect
Primary studies are available but have
not been tabulated
Tabulated data from secondary sources
of in vivo toxicology studies
Derived summary determinations of risk
Links to text reports on the web for
which specific data values are not
directly accessible in tabular form
analysis
Cell culture cytotoxicity
Tissue slice cytotoxicity
Clinical chemistry, histopathology,
developmental, and reproductive
assays
Clinical chemistry, histopathology,
developmental, and reproductive
assays
Clinical chemistry, histopathology,
developmental, and reproductive
assays
Chemicals determined to pose a defined
risk of human cancer
Reports from EPA Integrated Risk
Information System (IRIS), National
Toxicology Program (NTP)
approach to managing these three issues by bringing together
as much publicly available data on chemicals of environmental
interest as possible, into a limited number of databases linked
together by unique chemicals identifiers (Judson et al. 2007).
ACToR is being built with open source MySQL relational
database technology (http://www.mysql.com/). The web-based
system is presently available in a prototype version on the
EPA intranet, and is scheduled for public release in 2008.
Currently, the largest components of ACToR are the chemical
and assay databases, which contain chemical structure, in vitro
bioactivity data, and summary toxicology data for over 500000
compounds derived from more than 200 sources, including
the EPA; FDA; Centers for Disease Control and Prevention
(CDC); National Institutes of Health (NIH); state agencies;
corresponding government agencies in Canada, Europe, and
Japan; universities; the World Health Organization (WHO);
and nongovernmental organizations (NGOs). In the past year,
the DSSTox internal data structure has been redesigned to align
with both PubChem and ACToR, and all of the structures and
data from the DSSTox inventory now are incorporated into
ACToR. In addition, ACToR is serving as the primary data
management system for the ToxCast program (Dix et al. 2007)
and will include the ToxRefDB reference data, as well as all
newly generated HTS data. ACToR aggregates data from a large
number of sources and presents the data in a unified format on a
chemical-by-chemical basis. The system is designed to facilitate
the construction of modeling data sets, which can consist of
many chemicals and data from many assays.
Data that are potentially of interest for chemical toxicology
modeling exist in many places and forms and, as discussed
previously, can contain multiple levels of detail. For ACToR
construction, it is accepted that there is no single available
schema that is ideal for holding all types of existing chemical-
activity data; hence, several subschema are incorporated, often
by copying other databases in their entirety. For instance,
ACToR-affiliated databases include ToxRefDB, already de-
scribed, and EPA's High Production Volume Information
System (EPA HPVIS 2007), which contains detailed data from
Organisation for Economic Cooperation and Development
(OECD) guideline SIDS (Screening Information Data Set)
in vivo studies on high-production volume chemicals, along
with other industry-submitted environmental and toxicological
data.
The diverse data that exist on environmental chemicals cur-
rently are incorporated into ACToR and organized according
the basic categories listed in Table 3. Some of these types of data
are directly usable in modeling efforts (Nos. 1-6, 8, 9), whereas
others simply provide pointers to more detailed data (often still
in text form) (Nos. 7, 10) that could be mined manually to
build relevant tabular data sets for specific sets of chemicals.
The benefit of having all of these data available through the
ACToR web interface is that it is relatively straightforward to
build a list of target chemicals and then to systematically extract
data sets for further analyses.
ACToR is specifically designed to aggregate all available
information for each of a large number of chemicals. Currently,
CASRNs are employed as a unique identifier to link data from
multiple sources. Using CASRNs for this purpose has several
known drawbacks: they are not always available or unique for
a given substance (e.g., CASRNs can be retired and replaced),
they do not typically distinguish to the level of compound
purity grade (e.g., analytical vs. technical grade), and they
113
Toxicity Data Informatics
Previous
TOC
-------
are tied to a nonpublic registry system (Chemical Abstracts
Service, http://www.cas.org/). Nonetheless, these are the most
consistent and widely used chemical identifiers for indexing
public and environmental regulatory data resources, and they
index many existing public databases for which chemical
structures are currently unavailable. Hence, for the larger "read-
across" objectives of ACToR, CASRNs are sufficiently general
and widely available to serve as the initial basis for compound
indexing and aggregation.
Data are being systematically imported into ACToR from
a large number of public sources, which are referred to as
"data collections." A data collection will usually include a set
of substances (unique chemicals) and may have corresponding
compounds (chemical structures) and one or more assays.
The largest source of data currently in ACToR in terms
of substances and assay data points is PubChem (NCBI
PubChem 2007), which is itself a compilation of multiple
data sources (57 of which currently have data included
in ACToR). Most assay data in PubChem originate from
HTS assays run by the Molecular Libraries Screening Cen-
ters Network (MLSCN) (Austin et al. 2004) on compounds
from the Molecular Libraries Small Molecule Repository
(http://mlsmr.glpg.com/MLSMR_HomePage/). However, the
vast majority of chemicals in PubChem have no assay
data and come from collections of molecular structures
from chemical manufacturer catalogs (e.g., SIGMA-Aldrich,
http://www.sigmaaldrich.com/) or virtual screening libraries
(e.g., ZINC, http://zinc.docking.org/). The balance of the data
collections within ACToR pertain more specifically to environ-
mental chemicals and were extracted from a wide range of public
governmental and nongovernmental resources listed previously.
To be included in ACToR, a data collection must meet
several criteria. First, it has to be publicly available with no
restrictions on redistribution. An important goal of the ACToR
project is to create a widely usable, freely distributable, open
source system. Any conclusions drawn from these data should
be subject to independent confirmation, which is made possible
by this open source data model. Second, if a source consists of
a web-accessible database, an index of the chemicals in the
database is required in order to link that web resource back into
ACToR.
The ACToR system is currently being employed to support
the ToxCast environmental screening and prioritization project,
seeking to identify among large lists of chemicals of potential
environmental interest those for which some reference toxicity
data exist and those that are good candidates for toxicity
screening. The goal of the ToxCast program is to screen all
widely used chemicals using in vitro and in silico techniques
and to prioritize for detailed testing those chemicals with
"signatures" indicating possible toxicity. The derivation of these
predictive signatures is the object of phase I of the ToxCast
project, which is currently under way (Dix et al. 2007; EPA
ToxCast 2007).
It is estimated that there are about 30,000 unique chemicals
in widespread use in many countries, and for a high percentage
of these there is little available toxicology information. A
first challenge is to identify these chemicals and important
subsets that should be subject to a screening effort. ACToR has
incorporated ~500,000 unique chemicals from many sources
(where a unique chemical corresponds to one CASRN in
ACToR). The initial step in this data construction was to
select several data collections that comprise the most widely
used chemicals in the United States and build an overlap set.
These are comprised of High Production Volume chemicals
(HPVs) (EPA HPV 2007), pesticide and antimicrobial active
and inert ingredients (EPA Pesticides 2007), the EPA Toxic
Release Inventory (EPA TSCA 2007), the EPA drinking water
contaminant lists (EPA Drinking Water 2007), chemicals that
are part of the EPA's endocrine disrupter screening program
(EPA EDSP 2007), and the so-called medium-production
volume chemicals listed as part of the EPA Inventory Update
Rule list (EPA IUR 2007), which are chemicals manufactured
in excess of 10,0001b/year. In total, there are 11139 unique
chemicals in this aggregated collection.
One can easily browse individual chemicals within ACToR
to determine what data have been derived from the data
collections. For comparison to earlier example results shown
for Leadscope and DSSTox (Figs. 1, 5, and 6), the chemical
"Atrazine," CASRN [1912-24-9] is chosen to illustrate the
types of information indexed and easily retrieved in ACToR.
ACToR screenshots for a portion of the Atrazine-retrieved
results are presented in Figure 7. Information for Atrazine is
found in 54 different data collections. It has 31 physicochemical
parameters estimated using the EPA EPI Suite package (EPA
EPI Suite 2007) and over 1,000 HTS assay values (these include
multiple dose measures in the same assay) for a wide variety
of biochemical and cell-based assays contained in PubChem.
There is tabulated in vivo toxicology data from IRIS, CPDB,
World Health Organization (WHO) Classification of Pesticides
(WHO IPCS 2007), the EPA Drinking Water Standards and
Health Advisories (EPA Water Quality 2007), and the EPA
Risk-based Concentrations database (EPA RBC 2007). The
Pesticide Action Network (PAN 2007), Scorecard (Scorecard
2007), and the EPA Office of Pesticide Programs (EPA Pesticides
2007) each provides summary toxicity determinations, and PAN
and Scorecard indicated that Atrazine is a suspected carcinogen,
whereas the EPA has determined that it is not likely to be
a human carcinogen. However, all three sources indicate that
the chemical is a reproductive toxin, and the EPA further lists
the mode of action as endocrine disruption. The chemical is
manufactured or used in amounts between 10 M Ib/year and
50 M Ib/year in the United States, is on the EPA Superfund
Amendments and Reauthorization Act 110 Superfund Site
Priority Contaminant List (EPA SARA 2007), and is subject to
the National Primary Drinking Water Regulations (EPA Water
Quality 2007) and the New Jersey Right-to-Know Hazardous
Substances Regulation (NJ Hazardous Substances 2007). A
Material Data Safety Sheet is available from the CDC (CDC
ISCS 2007). Using the MESH (Medical Subject Headings)
link to PubMed (PubMed MESH 2007), one can find 1969
references on Atrazine, including 41 reviews. Using the link to
TOXNET DART (Developmental and Reproductive Toxicol-
ogy Database; NLM TOXNET 2007), one finds 63 references
relevant to reproductive toxicology. Genetic toxicology and
immunology studies are available from the NTP (NTP On-line
Database 2007).
A first approximation of the total world of data available
for understanding toxicity can be gleaned from some summary
statistics within ACToR for the set of 11,139 environmental
chemicals selected for screening and prioritization. Chemical
structures are readily available for 7,990 (72%) of the target list.
This relatively low rate is a consequence of the fact that many
of these high-use chemicals are complex mixtures, including
petroleum streams and substances such as plant oils, tars, ashes,
A. M. Richard et al.
114
Previous
TOC
-------
Chemical Summary: Atrazine
,f^,
GCID: 110 JA_/J^
CASRN: 1912-24-9 t-i'^'^^H'^'^^
CID: 11 978 I
CCID: 11 976 ^-l.
Formula C8H14CIN5
MW 215.61333
SMILES CIC1=NC(=NC(=N1)NC(C)C)NCC
InChI— 1/C8H14CIN5/c1-4-1 0-7-1 2-8(9)1 3-8(
3/h5H,4H2,1-3H3,(H2,10,1 1,12,13,14)
°C, QUEUED
Status
SuhstancesHide
SCID Name
112 Atrazine
. „, 1 ,3 ,5-Tri a z i n e-2 ,4- d i a rn i n e , 6- c h I o ro-
~ rnethylethyl)-
7119 Atrazine
7664 atrazine
8690 Atrazine
10641 Atrazine
11569 Atrazine
18479 Atrazine
24089 Atrazine
24159 Atrazine
24338 ATRAZINE
35738 Atrazine
35739 Atrazine
39956 Atrazine
49204 1,35-Tnazine-2,4-diamine,6-chloro-
rnethylethyl)-
Assay Data by Assay Category
* Show: PhysicoChemical (311
* Show: Biochemical (10791
* Show: Cellular tDl
N * Show: Tissue (01
CI • Show: Organ ID)
* Show: Organism (01
* Show: In vivo toxicology (tabular primary) (D)
* Show: In vivo loxicoloav (studv list in a primarvl (81
* Show: In vivo toxicology (tabular secondary) (631
* Show: In vivo toxicology (summary calls) (181
ii-7iii-w7i * Show: In vivo loxicoloav (sumrnarv report via URL1 (101
* Show: Regulation (31
* Show: Category (01
* Show: Chemical Summary URL (131
* Show: Chemical Use Level (51
• Show: Description (41
Data Collection Source SID
CPDBAS DSSTOX DSSTOX 20112
N-ethyl-N\"-(1- HPVCSI DSSTOX DSSTOX 27446
IRISTRJ3SSTOX DSSTOX_23892
NCTRER DSSTOX DSSTOX 22356
NTPBSI DSSTOX DSSTOX 30756
NTPHTS DSSTOX DSSTOX 33202
ToxCasl 320 DSSTOX 40346
ATSDR ToxFaq ATSDR ToxFac
EDC73 EDC73_8
EPA DWC EPA DWC 5
EXTOXNET EXTOXNET 18
INCHEMJARC INCHEMJARC,
INCHEM IARC INCHEM IARC
ITERJTERA ITERJTERAJ56
\l- pt h v I- NV- (1 -
x c L j i jf i j •< * i] HID QC m II ID QC m "3OQ
HJh. OD Uj- IUK OD UZ J^.y
Assay Data by Phenotype . Study Type
• Show: Hazard (341
* Show: AcuteTox (221
* Show: SubchronicTox (1)
* Show: ChronicTox (211
* Show: Carcinogenicity (E51
* Show: GeneTos (381
• Show: DevTox 122)
* Show: ReproTox (231
• Show: NeuroTox (201
• Show: DevNeuroTox (Dl
* Show: ImmunoTox 1211
* Show: DermalTox (11
* Show: RespiratoryTox (Dl
• Show: NephroTox (01
* Show: HepatoTox (11
* Show: Endocrine (21
• Show: CardioTox (D)
* Show: EcoTox (19)
• Show: FoodSafe 031
• Show: ToxOther (01
Deep Data Tables
* Show: TOXNET Toxicology (101
• Show: EPA HPVIS ID)
* Show: ToxRefDB (01
FIGURE 7 Sample views of the EPA Aggregated Computational Toxicology Resource (ACToR) results for Atrazine showing general
substance characteristics, incidence across ACToR data collections, and types of data available.
and plant products. About half of these chemicals have some
publicly available toxicology data within the sets of information
currently compiled. Primary in vivo toxicology data (taken
into ACToR from the original testing source) is available for
1,447 chemicals (13%), and secondary in vivo toxicology data
(taken into ACToR from a secondary source that compile and
summarize data from primary sources) is available for a total of
1,405 chemicals (13%). A total of 5205 chemicals (47%) have
one or more summary in vivo toxicity calls or determinations,
which are derived by experts who have curated data from the
primary scientific literature. Finally 5,244 chemicals (47%) have
one or more summary text reports on chemical toxicity available
on the Internet. However, many of these, especially from
the European Substances Information System Low Production
Volume list (ESIS HPV-LPV 2007), simply state that no hazard
or toxicology information is available for that chemical. These
are conservative numbers as there are still large collections
of data yet to be compiled and loaded into ACToR. The
bottom line, however, is that there is relatively little detailed
in vivo toxicology information available for the majority of
these environmental chemicals. The toxicology "data gap" for
commonly used chemicals is well known (Applegate and Baer
2006).
ACToR is a rapidly evolving system. Future developments
will include incorporation of additional data collections, ex-
traction of tabular data from on-line text documents linked
115
Toxicity Data Informatics
-------
to chemicals, addition of more curated chemical structures,
and the construction of a more flexible query and data export
interface. ACToR will be used for constructing training and
validation data sets for the ToxCast chemical screening and
prioritization effort, and for building computational models
linking chemical structure with in vitro and in vivo assays. It
is anticipated that this large structure-searchable database will
also be a valuable resource for reviewers within the EPA and
other regulatory agencies who are examining new chemicals
submitted for marketing approval. In relation to the toxicity
data models and the DSSTox project discussed previously,
ACToR can be considered a super aggregator and data mining
facilitator. Since it is a fully relational database system, ACToR
is capable of incorporating databases such as ToxRefDB in
their entirety, retaining the full internal data model structure.
Efforts to encourage toxicity data model standards, such as
ToxML, will equally enhance "read-across" data capability both
within the aggregated ACToR system and across federated,
or independently maintained, databases on the Internet. The
strengths of the DSSTox project, in terms of quality structure-
toxicity annotation of a growing data inventory, will also be fully
incorporated into ACToR. In return, the ACToR system will
add a full relational searchability across all DSSTox data fields
and files, wherever fields are standardized and "read-across" is
possible.
CONCLUSIONS
A confluence of new public data initiatives focused on
the world of chemical toxicity information, as represented
here by ToxML, DSSTox, and ACToR, is creating a more
unified, linked landscape of public toxicity data and propelling
the field of toxicity data informatics forward. The essential
elements for advancing such capabilities are shared standards
for chemical and toxicity data representation, involvement and
engagement of toxicity domain experts, transparent rules for
data aggregation, and chemical structure and text "read-across"
in relation to all levels of toxicity data and chemical-indexed
information.
A prominent concern of toxicologists encountering these
tools for the first time is to question how data quality
and the sufficiency of an experimental toxicology study are
judged in the process of data capture. Secondly, what is the
role of toxicology domain experts in defining appropriate
means for aggregating data into meaningful "endpoints" for
modeling? Any predictive toxicology approach, including SAR
modeling, must rely heavily on the peer review process and
author conclusions to self-regulate the quality of published
toxicology studies and reports. Having said that, however,
when a data model succeeds in faithfully capturing the key
elements of a toxicological experiment, sufficient to allow a
toxicology domain expert to judge the overall merits of the
study and its conclusions, it has delivered a useful and unbiased
representation of that experiment. Furthermore, a data model
that lends itself to flexible means for aggregating toxicity data
and defining summarized endpoints for modeling can easily
support alternative viewpoints and changing attitudes from
within the toxicology community as to what endpoints are more
meaningful and relevant to hazard identification and human
risk assessment. Hence, the engagement of toxicology experts
in this process is essential to this enterprise of building useful
predictive models.
A. M. Richard et al.
There are three general types of currently available databases
and technology in the hazard identification and risk assessment
fields: (1) databases storing the results of toxicity experiments,
(2) databases for use by regulatory agencies, and (3) aggregated
databases to support SARs and predictive modeling. A vision
of the future is that these three levels of databases will be
stored in the same database with a well-designed database
schema and will communicate seamlessly. Initiatives such as
ACToR, DSSTox, and ToxML are significant steps in this
direction. Entry of new toxicity data in the public domain occurs
largely from government and academic research institutions
and government regulatory agencies, through past, present and
future data submissions and public disclosure laws. With the
broad adoption of shared data standards and data models,
and quality review of chemical information, data submissions
from the private sector to a regulatory agency can take place
electronically and seamlessly between respective databases. All
stakeholders in this process stand to benefit from the growth
of quality curated and standardized public toxicity databases,
and the field of predictive toxicology will be one of the largest
beneficiaries.
REFERENCES
Ames, B. N. 1971. The detection of chemical mutagens with enteric
bacteria. In Chemical Mutagens: Principles and Methods for Their
Detection, Vol. 1, ed. A. Hollaender, Plenum Press, New York-
London 267-282
Applegate, J. S., and Baer, K. 2006. Strategies for closing the chemical
data gap. Center for Progressive Reform (White Paper) 602:1-19.
Ashby, J., and Paton, D. 1993. The influence of chemical structure on the
extent and sites of carcinogenesis for 522 rodent carcinogens and
55 different human carcinogen exposures. Mutat. Res. 286:3-74.
Auletta.A. E., Brown, M., Wassom, J. S.,and Cimino, M. C. 1991. Current
status of the Gene-Tox Program. Environ. Health Perspec 96:33-36.
Austin, C. P., Brady, L S., Insel, T. R., and Collins, F. S. 2004. NIH Molecular
Libraries Initiative. Science 306:1138-1139.
Backus, G. S., Wolf, M. A., Burch, J., and Richard, A. M. 2007. DSSTox
EPA Integrated Risk Information System (IRIS) Toxicity Review Data:
SDF File and Documentation: IRISTR.v1a.544.28Jul2007.www.
epa.gov/ncct/dsstox/sdf.iristr.html. Accessed January 24, 2008.
Benigni, R. 1997. The first US National Toxicology Program exercise on
the prediction of rodent carcinogenicity: definitive results. Mutat.
Res. 387:35-45.
Benigni, R., Bossa, C., Richard, A. M., and Yang, C. 2007b. A novel
approach: Chemical relational databases, and the role of the
ISSCAN database on assessing chemical carcinogenicity. Annali dell'
Institute Super/ore di Sanita. In press.
Benigni, R., Netzeva, T. I., Benfenati, E., Bossa, C., Franke, R., Helma, C.,
Hulzebos, E., Marchant, C., Richard, A., Woo, Y. T, and Yang, C.
2007a. The expanding role of predictive toxicology: an update on
the (Q)SAR models for mutagens and carcinogens. J. Environ. Sci.
Health C 25:53-97.
Burch, J., Eastin, W. C., Bowden, B., Wolf, M. A., and Richard,
A. M. 2007. DSSTox National Toxicology Program Bioas-
say On-line Database Structure-Index Locator File: SDF File
and Documentation, NTPBSI_v2a_2293_24Aug2007.www.epa.gov/
ncct/dsstox/sdf_ntpbsi.html. Accessed January 24, 2008.
CDC ICSC. 2007. Centers for Disease Control and Prevention,
National Institute for Occupational Safety and Health, Inter-
national Chemical Safety Cards, http://www.cdc.gov/niosh/ipcs/
nicstart.html. Accessed January 24, 2008.
CPDB. 2007. Berkeley Carcinogenic Potency Database. http://potency.
berkeley.edu/. Accessed January 24, 2008.
116
Previous
TOC
-------
CPDB Summary Tables. 2007. Summary Table of Chemicals in the Carcino-
genic Potency Database: Results for Positivity, Potency (TD50), and
Target Sites, http://potency.berkeley.edu/chemicalsummary.html.
Accessed January 24, 2008.
Contrera, J. R, Matthews, E. J., Kruhlak, N. L, and Benz, R. D. 2005.
In silica screening of chemicals for bacterial mutagenicity using
electrotopological E-state indices and MDL QSAR software. Regul.
Tax. Pharmacol. 43:313-323.
Dix, D. J., Houck, K. A., Martin, M. T, Richard, A. M., Setzer, W., and
Kavlock, R. J. 2007. The ToxCast program for prioritizing toxicity
testing of environmental chemicals. Tox. Sci. 95:5-12.
EPA Drinking Water. 2007. U.S. Environmental Protection Agency's Office
of Ground Water and Drinking Water—Contaminants. http://www.
epa.gov/ogwdw/contaminants/index.html. Accessed January 24,
2008.
EPA DSSTox. 2007. U.S. Environmental Protection Agency's Dis-
tributed Structure-Searchable Toxicity (DSSTox) Database Network.
http://www.epa.gov/ncct/dsstox/index.html. Accessed January 24,
2008.
EPA EDSP. 2007. U.S. Environmental Protection Agency's En-
docrine Disruption Screening Program, http://www.epa.gov/
scipoly/oscpendo/. Accessed January 24, 2008.
EPA EPI Suite. 2007. U.S Environmental Protection Agency's Estimation
Programs Interface Suite(tm) for Microsoft® Windows, XR http://
www.epa.gov/opptintr/exposure/pubs/episuite.htm. Accessed Jan-
uary 24, 2008.
EPA HPV. 2007. U.S. Environmental Protection Agency's High Pro-
duction Volume Challenge Program and Information Sys-
tem. http://www.epa.gov/hpv/ and http://www.epa.gov/hpvis/.
Accessed January 24, 2008.
EPA IRIS. 2007. U.S. Environmental Protection Agency's Integrated Risk
Information System Database, http://www.epa.gov/iris/. Accessed
January 24, 2008.
EPA IUR. 2007. U.S. Environmental Protection Agency's Inventory Update
Rule list, http://www.epa.gov/oppt/iur/. Accessed January 24, 2008.
EPA Pesticides. 2007. US Environmental Protection Agency's Office
of Pesticide Programs, http://www.epa.gov/pesticides/. Accessed
January 24, 2008.
EPA RBC. 2007. U.S. Environmental Protection Agency's Hu-
man Health Risk Assessment, Risk-based Concentration Ta-
ble. http://www.epa.gov/reg3hwmd/risk/human/index.htm. Ac-
cessed January 24, 2008.
EPA Regulating Pesticides. 2007. U.S. Environmental Protection
Agency's Office of Pesticide Programs, Regulating Pesticides.
http://www.epa.gov/pesticides/regulating/. Accessed January 24,
2008.
EPA SARA. 2007. U.S. Environmental Protection Agency's Su-
perfund Amendments and Reauthorization Act, 110 Su-
perfund Site Priority Contaminant List, http://www.epa.gov/
lawregs/laws/cercla/html. Accessed January 24, 2008.
EPA ToxCast 2007. U.S. Environmental Protection Agency's Na-
tional Center for Computational Toxicology ToxCastTM Program.
http://www.epa.gov/ncct/toxcast/. Accessed January 24, 2008.
EPA TSCA. 2007. U.S. Environmental Protection Agency's Toxic
Substances Control Act Release Inventory, http://www.epa.gov/
opptintr/newchems/pubs/invntory.htm. Accessed January 24, 2008.
EPA Water Quality. 2007. U.S. Environmental Protection Agency's Water
Quality Criteria, Drinking Water Standards and Health Advisories.
http://www.epa.gov/waterscience/criteria/drinking/. Accessed Jan-
uary 24, 2008.
ESIS HPV-LPV. 2007. European Substances Information System Low
Production Volume Chemicals, http://ecb.jrc.it/esis/index.php?
PGM=hpv. Accessed January 24, 2008.
Gold, L. S., Manley, N. B., Slone, T. H., and Rohrbach, L. 1999.
Supplement to the Carcinogenic Potency Database (CPDB): Results
of animal bioassays published in the general literature in 1993 to
1994 and by the National Toxicology Program in 1995 to 1996.
Environ. Health Perspect. 107(Suppl. 4):527-600.
Gold, L. S., Slone, T. H., Ames, B. N., Manley, N. B., Garfinkel, G. B.,
Rohrbach, L. 1997. Chapter 1: carcinogenic potency database. In
Handbook of Carcinogenic Potency and Genotoxicity Databases,
eds. Gold L. S., and Zeiger E., CRC Press, Boca Raton, FL,
1-605.
Gold, L. S., Slone, T. H., Williams, C. R., Burch, J. M., Stewart, T.
W., Swank, A. E., Beidler, J., and Richard, A. M. 2007. DSSTox
Carcinogenic Potency Database Summary Tables—All Species,
SDF Files and Documentation, CPDBAS.v5a.1527.250ct2007.
http://www.epa.gov/ncct/dsstox/sdf_cpdbas.html. Accessed Jan-
uary 24, 2008.
Gombar, V. K., Enslein, K., and Blake, B. W. 1995. Assessment
of developmental toxicity potential of chemicals by quantitative
structure-toxicity relationship models. Chemosphere 31:2499-
2510.
Houck, K., Dix, D., Judson, R., Martin, M., Wolf, M., Kavlock,
R., and Richard, A. M. 2007. DSSTox EPA ToxCast High
Throughput Screening Testing Chemicals Structure-Index File:
SDF File and Documentation: TOXCST.v2a.320.25Sep2007.
http://www.epa.gov/ncct/dsstox/sdf.toxcst.html. Accessed January
24, 2008.
Judson, R., Richard, A., Dix, D., Elloumi, F., Martin, M., Cathey, T.,
Transue, T., Spencer, R., and Wolf, M. 2007. ACToR—Aggregated
Computational Toxicology Resource. Toxicol. Appl. Pharmacol.
Submitted
Julien, E., Willhite, C. C., Richard, A. M., and DeSesso, J. M. 2004.
Challenges in constructing statistically-based SAR models for
developmental toxicity. Birth Defects Res. Part A 70:902-911.
Kazius, J., McGuire, R., and Bursi, R. 2005. Derivation and validation of
toxicophores for mutagenicity prediction. J. Med. Chem. 48:312-
320.
Leadscope, Inc. 2007. FDA databases, http://www.leadscope.com/
fda.databases/. Accessed January 24, 2008.
Martin, M. T, Houck, K. A., McLaurin, K., Richard, A. M., and
Dix, D. J. 2007. Linking regulatory toxicological information on
environmental chemicals with high-throughput screening (HTS) and
genomic data. The Toxicologist CD J. Soc. Toxicol. 96:219-220.
NCBI PubChem. 2007. National Institutes of Health, National Library
of Medicine, National Center for Biotechnology Information, Pub-
Chem Project, http://pubchem.ncbi.nlm.nih.gov/. Accessed January
24, 2008.
NJ Hazardous Substances. 2007. New Jersey Right-to-Know
Hazardous Substances Regulation, http://web.doh.state.nj.us/
rtkhsfs/indexfs.aspx. Accessed January 24, 2008.
NLM TOXNET. 2007. National Institutes of Health, National Li-
brary of Medicine (NLM), Toxicology Data Network (TOXNET).
http://toxnet.nlm.nih.gov. Accessed January 24, 2008.
NTP On-line Database. 2007. National Institute of Health & Environmental
Sciences, National Toxicology Program (NTP) On-line Database.
http://ntp.mehs. nih.gov/ntpweb/ index. cfm?objectid = 72016020-
BDB7-CEBA-F3E5A7965617C1C1. Accessed January 24, 2008.
PAN. 2007. Pesticide Action Network, http://www.pesticideinfo.org/
lndex.org. Accessed January 24, 2008.
PubMed MESH. 2007. National Institutes of Health, National Library of
Medicine, National Center for Biotechnology Information, Medical
Subject Headings for indexing articles for MEDLINE/PubMed.
http://www.ncbi.nlm.nih.gov/sites/entrez?db=mesh. Accessed Jan-
uary 24, 2008.
Richard, A. M. 2004. DSSTox Website launch: Improving public access
to databases for building structure-toxicity prediction models.
Prec//n/ca2:103-108.
Richard, A. M. 2006. Future of predictive toxicology: an expanded view of
"chemical toxicity"—future of toxicology perspective. Chem. Res.
Toxicol. 19:1257-1262.
Richard, A. M., Gold, L. S., and Nicklaus, M. C. 2006. Chemical
structure indexing of toxicity data on the internet: Moving to-
wards a flat world. Curr. Opin. Drug Discovery Develop. 9:314-
325.
117
Toxicity Data Informatics
Previous
TOC
-------
Richard, A. M., and Williams, C. R. 2002a. Distributed structure-
searchable toxicity (DSSTox) database network: a proposal. Mutat.
Res. 499:27-52.
Richard, A. M., and Williams, C. R. 2002b. Public sources of mutagenicity
and carcinogenicity data: Use in structure-activity relationship
models. In QSARs of Mutagens and Carcinogens, eds. Benigni R.,
CRC Press, New York, pp. 145-173.
Scorecard. 2007. Scorecard Pollution Information Site, Health Effects.
http://www.scorecard.org/health-effects/. Accessed January 24,
2008.
Singh, A. V, Knudsen, K. B., and Knudsen, T. B. 2005. Computational
systems analysis of developmental toxicity: design, development
and implementation of a Birth Defects Systems Manager (BDSM).
Reproductive Toxicol. 19:421-439.
Smith, C., Collins, B., Tice, R., Wolf, M. A., and Richard,
A. M. 2007. DSSTox National Toxicology Program High
Throughput Screening Structure-Index File: SDF File and Doc-
umentation: NTPHTS.v2a.1408.25Jul2007. http://www.epa.gov/
ncct/dsstox/sdf_ntphts.html. Accessed January 24, 2008.
Swartz, C. D., Parks, N., Umbach, D. M., Ward, W. 0., Schaaper, R. M.,
and DeMarini, D. M. 2007. Enhanced mutagenesis of Salmonella
tester strains due to deletion of genes other than uvrB. Environ.
Mol. Mutagen. 48:694-705.
Tennant, R. W. 1993. Stratification of rodent carcinogenicity bioassay
results to reflect relative human hazard. Mutat. Res. 286:111-
118.
ToxML. 2007. A publicly available Toxicity XML standard and controlled vo-
cabulary for representing toxicity data. Database schema available
at Leadscope Internet site, http://www.leadscope.com/toxml.php.
Accessed January 24, 2008.
Transue, T, and Richard, A. M. 2007. U.S. Environmental Protection
Agency DSSTox Structure-Browser v1.03. http://www.epa.gov/
dsstox.structurebrowser/. Supporting documentation available
on the Internet at: http://www.epa.gov/ncct/dsstox/
StructureBrowserlnfo.html. Accessed January 24, 2008.
WHO IPCS. 2007. World Health Organization's International Programme
on Chemical Safety, Classification of Pesticides by Hazard, http://
www.who.int/ipcs/publications/pestiddes_hazard/en/. Accessed
January 24, 2008.
Wolf, M. A., Martin, M., and Richard, A. M. 2007. DSSTox EPA
High Production Volume Information System Data—Structure-
Index Locator File: SDF File and Documentation, Launch
File version: HPVISD.v1a.1006.250ct2007. www.epa.gov/ncct/
dsstox/sdf_hpvisd.html. Accessed January 24, 2008.
Yang, C., Arvidson, K., Aveston, S., Benigni, R., Benz, R. D., Contrera, J.,
Dierkes, A. N., Xing, Hasselgren-Arnby, C., Jaworksa, J., Matthews,
E., Kruhlak, N., Kemper, R., Rathman, J. F, Richard, A. M.
2007. Understanding genetic toxicity from data mining: Process
of building knowledge by integrating multiple genetic toxicity
databases. Toxicol. Mech. Meth. Present volume.
Yang, C., Benz, R. D., and Cheeseman, M. A. 2006a. Landscape of
current toxicity databases and database standards. Curr. Opin. Drug
Discovery Develop. 9:124-133.
Yang, C., Richard, A. M., and Cross, K. P. 2006b. The art of data mining
the minefields of toxicity databases to link chemistry to biology.
Curr. Comput. Aided Drug Design 2(2): 135-1 50.
A. M. Richard et a/.
118
Previous
TOC
-------
Risk Analysis, Vol. 29, No. 4, 2009
DOI:10.1111/j.l539-6924.2008.01168.x
Commentary
Toxicity Testing in the 21st Century: Implications for Human
Health Risk Assessment
Robert J. Kavlock,1* Christopher P. Austin,2 and Raymond R. Tice3
The risk analysis perspective by Daniel Krewski and
colleagues lays out the long-term vision and strate-
gic plan developed by a National Research Coun-
cil committee/1^ sponsored by the U.S. Environmen-
tal Protection Agency (EPA) with support from the
U.S. National Toxicology Program (NTP), to "ad-
vance the practices of toxicity testing and human
health assessment of environmental agents." Com-
ponents of the vision include chemical characteriza-
tion; the use of human-cell-based, high-throughput
assays that cover the diversity of toxicity pathways;
targeted testing using animals to fill in data gaps;
dose-response and extrapolation modeling; and the
generation and use of population-based and human
exposure data for interpreting the results of toxicity
tests. The strategic plan recognizes that meeting this
vision will require a major research effort conducted
over a period of a decade or more to identify all of the
important toxicity pathways, and that a clear distinc-
tion must be made between which pathway perturba-
tions are truly adverse (i.e., would likely lead to ad-
verse health outcomes in humans) and those that are
not. Krewski et al note that achieving this vision in
a reasonable timeframe (i.e., decades) would require
the involvement of an interdisciplinary research
1National Center for Computational Toxicology, Office of Re-
search and Development, U.S. Environmental Protection Agency,
Research Triangle Park, NC, USA.
2NIH Chemical Genomics Center, National Human Genome
Research Institute, National Institutes of Health, MSC 3370,
Bethesda, MD, USA.
Biomolecular Screening Branch, National Toxicology Program,
National Institute of Environmental Health Sciences, Research
Triangle Park, NC, USA.
* Address correspondence to Robert J. Kavlock, Director, Na-
tional Center for Computational Toxicology, Office of Research
and Development, U.S. Environmental Protection Agency, Re-
search Triangle Park, NC 27711, USA; tel: 919-541-2326; fax: 919-
541-1194; Kavlock.Robert@epamail.epa.gov.
institute that would be coordinated and funded pri-
marily by the U.S. federal government and that
would foster appropriate intramural and extramu-
ral research. It is expected that this approach would
greatly increase the number of compounds that can
be tested, while providing data more directly rele-
vant for conducting human health risk assessment.
The NTP though its Roadmap,4 the National Insti-
tutes of Health (NIH) Chemical Genomics Center
(NCGC) through the Molecular Libraries Initiative,5
and the EPA through its ToxCast™ program6 and its
draft Strategic Plan for the Future of Toxicity Testing
have individually recognized the need to bring inno-
vation into the assessment of the toxicological activ-
ity of chemicals, and each has made progress in do-
ing so. However, the grand challenge put forth by
Krewski et al. requires an effort unparalleled in the
field of toxicology and risk assessment.
In recognition of the importance of the NRC re-
and to accelerate progress in this area, two
NIH institutes and EPA have entered into a for-
mal collaboration known as Tox21 to identify mech-
anisms of chemically induced biological activity,
prioritize chemicals for more extensive toxicological
evaluation, and develop more predictive models of
in vivo biological response.(2) Consistent with the vi-
sion outlined by Krewski et al., success in achieving
these goals is expected to result in methods for toxi-
city testing that are more scientific and cost effective
as well as models for risk assessment that are more
mechanistically based. As a consequence, a reduction
or replacement of animals in regulatory testing is an-
ticipated to occur in parallel with an increased abil-
ity to evaluate the large numbers of chemicals that
4 Available at: http://ntp.niehs.nih.gov/go/vision.
5 Available at: http://www.ncgc.nih.gov/.
6 Available at: epa.gov/ncct/toxcat.
485 0272-4332/09/0100-0485$22.00/l © 2008 Society for Risk Analysis
Previous
TOC
-------
486
Kavlock, Austin, and Tice
currently lack adequate toxicological evaluation. Ul-
timately, Tox21 is expected to deliver biological ac-
tivity profiles that are predictive of in vivo toxicities
for the thousands of understudied substances of con-
cern to regulatory authorities in the United States, as
well as in many other countries.
The Tox21 collaboration is being coordinated
through a five-year Memorandum of Understand-
ing (MoU),7 which leverages the strengths of each
organization. The MoU builds on the experimental
toxicology expertise at the NTP, headquartered at
the NIH National Institute of Environmental Health
Sciences (NIEHS); the high-throughput screening
(HTS) technology of the NIH Chemical Genomics
Center (NCGC), managed by the National Hu-
man Genome Research Institute (NHGRI); and
the computational toxicology capabilities of the
EPA's National Center for Computational Toxicol-
ogy (NCCT). Each party brings complementary ex-
pertise to bear on the application of novel method-
ologies to evaluate large numbers of chemicals for
their potential to interact with the myriad of biologi-
cal processes relevant to toxicity. A central aspect of
Tox21 is the unique capabilities of the NCGCs high-
speed, automated screening robots to simultaneously
test thousands of potentially toxic compounds in bio-
chemical and cell-based HTS assays, and an ability
to target this resource toward environmental health
issues. As mentioned by Krewski et al, EPA's Tox-
Cast™ Program is an integral and critical compo-
nent for achieving the Tox21 goals laid out in the
MoU.
To support the goals of Tox21, four fo-
cus groups—Chemical Selection, Biological Path-
ways/Assays, Informatics, and Targeted Testing—
have been established; these focus groups represent
the different components of the NRC vision de-
scribed by Krewski et al. The Chemical Selection
group is coordinating the selection of chemicals for
the Tox21 compound library to test at the NCGC.
A chemical library of nearly 2,400 chemicals selected
by NTP and the EPA is already under study at the
NCGC and results from several dozen HTS assays
are already available. In the near term, this library
will be expanded to approximately 8,400 compounds,
with an additional ~1,400 compounds selected by
the NTP, ~1,400 compounds selected by the EPA,
and ~2,800 clinically approved drugs selected by
the NCGC. Compound selection is currently based
largely on the compound having a defined chemical
7 Available at: http://ntp.niehs.nih.gov/go/28213.
structure and known purity; on the extent of its sol-
ubility and stability in dimethyl sulfoxide (DMSO),
the preferred solvent for HTS assays conducted at
the NCGC; and on the compound having low volatil-
ity. Implementing quality control procedures for en-
suring identify, purity, and stability of all compounds
in the library is an important responsibility of this
group. A subset of the Tox21 chemical library will be
included in Phase II of the ToxCast program, which
will examine a broader suite of assays in order to
evaluate the predictive power of bioactivity signa-
tures derived in Phase I. Phase II of ToxCast will be
launched by the summer of 2009.
The Biological Pathways and Assays group is
identifying critical cellular toxicity pathways for in-
terrogation using biochemical- and cell-based high-
throughput screens and prioritizing HTS assays for
use at the NCGC. Assays already performed at the
NCGC include those to assess (1) cytotoxicity and
activation of caspases in a number of human and
rodent cell types, (2) up-regulation of p53, (3) ago-
nist/antagonist activity for a number of nuclear re-
ceptors, and (4) differential cytotoxicity in several
cell lines associated with an inability to repair various
classes of DNA damage. Other assays under consid-
eration include those for a variety of physiologically
important molecular pathways (e.g., cellular stress
responses) as well as methods for integrating human
and rodent hepatic metabolic activation into reporter
gene assays. Based on the results obtained, this group
will construct test batteries useful for identifying haz-
ard for humans and for prioritizing chemicals for fur-
ther, more in-depth evaluation.
The Informatics group is developing databases
to store all Tox21-related data and evaluating the re-
sults obtained from testing conducted at the NCGC
and via ToxCast™ for predictive toxicity patterns.
To encourage independent evaluations and/or anal-
yses of the Tox21 test results, all data as well as
the comparative animal and human data, where
available, will be made publicly accessible via var-
ious databases, including EPAs Aggregated Com-
putational Toxicology Resource (ACToR), NIEHS'
Chemical Effects in Biological Systems (CEBS), and
the National Center for Biotechnology Information's
PubChem.
As HTS data on compounds with inadequate
testing for toxicity becomes available via Tox21,
there will be a need to test selected compounds in
more comprehensive assays. The Targeted Testing
group is developing strategies and capabilities for this
purpose using assays that involve higher order testing
Previous
TOC
-------
Commentary
487
systems (e.g., roundworms (Caenorhabditis elegans),
zebrafish embryos, rodents).
In addition to the testing activities, the MoU pro-
motes coordination and sponsorship of workshops,
symposia, and seminars to educate the various stake-
holder groups, including regulatory scientists and the
public, with regard to Tox21-related activities. Per-
sons interested in following the progress of Tox21
are invited to join the EPA's Chemical Prioritization
Community of Practice,8 which meets monthly via
teleconference.
Given the scope of the challenge presented by
Krewski et al., success will require a long-term con-
certed effort by a large number of investigators,
working in a coordinated manner. The Tox21 consor-
tium welcomes participation in our effort by individ-
ual scientists and by organizations. The implications
for success of this effort are considerable. If success-
ful, we will be able to address regulatory demands
such as those placed by the Food Quality Protection
Act for the endocrine screening program9 and the
new European Community Regulation on chemicals
and their safe use, known as REACH (Registration,
Evaluation, Authorization and Restriction of Chem-
ical Substances),10 identify key modes of action on a
scale not imaginable even a few years ago, direct a
much more efficient and effective use of animals in
toxicity testing, identify potentially susceptible sub-
populations based on the presence of polymorphisms
8 Available at: http://www.epa.gov/ncct/practice_community/
category_priority.html.
9 Available at: http://www.epa.gov/opp00001/regulating/laws/fqpa/.
l° Available at: http://ec.europa.eu/environment/chemicals/reach/
reach_intro.htm.
in toxicity pathways, screen the effects of mixtures,
and study emerging issues like the safety of nano-
materials. The acquisition of data from broad-scale
HTS programs also creates demands to integrate this
knowledge and understand the implications for sys-
tems biology, and to have risk assessors trained and
conversant in the new technologies and their utilities.
While the ultimate goal of eliminating the use of an-
imals in toxicology testing might seem unattainable,
it is only by carefully evaluating the relevance and
reliability of strategies based on in vitro test meth-
ods that the utility and limitations of such an ap-
proach can be determined and decisions made on
how best to conduct toxicology testing in the future.
To do otherwise will result in increasing demands
being placed on systems never designed to handle
the large numbers of chemicals in need of evalua-
tion, and continued reliance on test methods based
on empirical observation rather than on mechanistic
understanding.
DISCLAIMER
The research described in this report has been
funded by one or more of the participating federal
agencies. The report does not necessarily reflect the
views of the respective organizations.
REFERENCES
1. National Research Council. Toxicity Testing in the 21st Cen-
tury: A Vision and A Strategy. Washington, DC: National
Academy Press, 2007.
2. Collins FS, Gray GM, Bucher JR. Transforming environmental
health protection. Science, 2008: 319:906-907.
Previous
-------
Toxicology and Applied Pharmacology 238 (2009) 80-89
Contents lists available at ScienceDirect
Toxicology and Applied Pharmacology
journal homepage: www.elsevier.com/locate/ytaap
Toxicogenomic effects common to triazole antifungals and conserved between rats
and humans
Amber K. Goetz a'b, David J. Dix a'*
a National Center for Computational Toxicology, Office of Research and Development, U.S. Environmental Protection Agency, Research Triangle Park, North Carolina 27711, USA
b Department of Environmental and Molecular Toxicology, North Carolina State University, Raleigh, North Carolina 27695, USA
ARTICLE INFO
Article history:
Received 16 January 2009
Revised 13 April 2009
Accepted 22 April 2009
Available online 3 May 2009
Keywords:
Myclobutanil
Propiconazole
Triadimefon
Toxicogenomics
CAR
PXR
ABSTRACT
The triazole antifungals myclobutanil, propiconazole and triadimefon cause varying degrees of hepatic
toxicity and disrupt steroid hormone homeostasis in rodent in vivo models. To identify biological pathways
consistently modulated across multiple timepoints and various study designs, gene expression profiling was
conducted on rat livers from three separate studies with triazole treatment groups ranging from 6 h after a
single oral gavage exposure, to prenatal to adult exposures via feed. To explore conservation of responses
across species, gene expression from the rat liver studies were compared to in vitro data from rat and human
primary hepatocytes exposed to the triazoles. Toxicogenomic data on triazoles from 33 different treatment
groups and 135 samples (microarrays) identified thousands of probe sets and dozens of pathways
differentially expressed across time, dose, and species — many of these were common to all three triazoles, or
conserved between rodents and humans. Common and conserved pathways included androgen and estrogen
metabolism, xenobiotic metabolism signaling through CAR and PXR, and CYP mediated metabolism.
Differentially expressed genes included the Phase I xenobiotic, fatty acid, sterol and steroid metabolism genes
Cyp2b2 and CYP2B6, Cyp3al and CYP3A4, and Cyp4a22 and CYP4A11; Phase II conjugation enzyme genes
Ugtlal and UGT1A1; and Phase III ABC transporter genes Abcbl and ABCB1. Gene expression changes caused
by all three triazoles in liver and hepatocytes were concentrated in biological pathways regulating lipid,
sterol and steroid homeostasis, identifying a potential common mode of action conserved between rodents
and humans. Modulation of hepatic sterol and steroid metabolism is a plausible mode of action for changes in
serum testosterone and adverse reproductive outcomes observed in rat studies, and may be relevant to
human risk assessment.
Published by Elsevier Inc.
Introduction
Myclobutanil, propiconazole and triadimefon are agrichemical
fungicides with a 1,2,4-N-substituted triazole moiety that binds the
heme portion of fungal cypSl, inhibiting fungal lanosterol-14a-
demethylase activity, blocking ergosterol biosynthesis and thus
controlling several species and strains of fungi (Ghannoum and Rice,
1999; Vanden Bossche et al., 1990). All three of the triazoles in this
study are used in the control of brown patch, dollar spot, rust, and
several other fungal and plant diseases to protect turf, fruit, and
vegetable and seed commodities.
Triazole fungicides exhibit a range of toxicological properties in
mammalian species. Effects reported from triazole exposures in rodents
* Disclaimer: The United States Environmental Protection Agency through its Office
of Research and Development funded and managed the research described here. It has
been subjected to Agency administrative review and approved for publication.
* Corresponding author. National Center for Computational Toxicology, Mail Drop
D343-03, U.S. Environmental Protection Agency, Research Triangle Park, NC 27711, USA.
Fax: +1 919 5411194.
E-mail address: dix.david@epa.gov (D.J. Dix).
0041-008X/S - see front matter. Published by Elsevier Inc.
doi:10.1016/j.taap.2009.04.016
include male infertility for myclobutanil and triadimefon in rats, liver
tumorigenicity for propiconazole and triadimefon in mice, thyroid
tumors for triadimefon in rats, and some measure of reproductive and/
or hepatic toxicity for all three of these triazoles (Goetz et al., 2007; U.S.
EPA, 1995,2005a, 2005b, 2006). Studies examining developmental and
reproductive effects of myclobutanil, propiconazole, and triadimefon
exposures beginning in gestation and continuing to adulthood in rats
have demonstrated that all three triazoles caused increased serum
testosterone levels (Goetz et al., 2007). Several studies in mice and rats
on the hepatic and thyroid toxicity of triazole fungicides have identified
modes of action, in some cases conserved across rodent species, and
possibly relevant to the assessment of risk to human health (Allen et al.,
2006; Wolf etal., 2006).
Genomic data from these and other studies have linked triazole
specific toxicological endpoints to demonstrate the ability of high-
content biology to delineate potential pathways of toxicity (Goetz
et al., 2006; Hester et al., 2006; Hester and Nesnow, 2008; Martin
et al., 2007; Tully et al., 2006; Ward et al., 2006). To date, in vivo gene
expression profiles have demonstrated that triazoles appear to modu-
late CAR and PXR, and subsequently perturb hepatic lipid, sterol, and
steroid and xenobiotic metabolism pathways. The concordance of
Previous
-------
A.K. Goetz, DJ. Dix / Toxicology and Applied Pharmacology 238 (2009) 80-89
81
these in vivo observations and gene expression findings demonstrated
the ability of genomics to identify potential modes of action and
toxicity pathways.
In the present study, genomic data from a series of in vivo and in
vitro studies on three triazoles (myclobutanil, propiconazole and tria-
dimefon) were analyzed to test the hypothesis that there are con-
served hepatic biological pathways active in rat liver and human
hepatocytes that are commonly perturbed by this chemical class.
While changes in individual genes did vary between the in vivo and in
vitro studies, comparison of homologous genes across species and
mapping these to biological pathways facilitated meaningful evalua-
tions of functional significance. These differentially expressed genes
and affected pathways were then related to the toxicological end-
points from these same in vivo and in vitro studies, creating a frame-
work for understanding the toxicity pathways common to triazoles
and conserved in rats and humans.
Materials and methods
Prenatal to adult rat study. Details of the animal husbandry and
study design for the prenatal to adult rat study have been previously
published (Goetz et al., 2007). For this and all other studies; animal
care, handling, and treatment at EPA or by its contractors was con-
ducted in American Association for Accreditation of Laboratory Animal
Care — International accredited facilities, and all procedures were
approved by an Institutional Animal Care and Use Committee. For the
prenatal to adult rat study conducted at EPA, pregnant Wistar Han rats
were exposed to triazoles in feed from gestation day 6 (GD6) until
weaning of the pups. Offspring were housed with their respective
dams until weaning on postnatal day 23 (PND23). Males were then
removed from the dams and exposed until PND92. Feed was mixed
with triazoles by Bayer CropScience (Kansas City, MO) as part of a
Materials Cooperative Research and Development Agreement
between the U.S. EPA and the U.S. Triazole Task Force. Control animals
were fed 5002 Certified Rodent Diet with acetone vehicle added. The
six treatment groups carried forward to genomic analysis were myclo-
butanil (MYC) at 500 or 2000 ppm (equivalent to 32.9 and 133.9 mg/
kg/day); propiconazole (PPZ) at 500 or 2500 ppm (equivalent to 31.9
and 169.7 mg/kg/day); or triadimefon (TDF) at 500 or 1800 ppm
(equivalent to 33.1 and 139.1 mg/kg/day). One male from each litter
was euthanized by asphyxiation with carbon dioxide and necropsied at
PND92 and tissues collected for transcriptional profiling.
Repeated dose oral gavage studies. In collaboration with EPA, Gene
Logic Laboratories Inc. (Gaithersburg, MD) conducted one of the
toxicogenomic studies presented herein. Male Sprague-Dawley rats
10 weeks of age were administered vehicle control (15% Alkamuls EL-
620 in deionized water), myclobutanil (300 mg/kg/day) or tria-
dimefon (175 mg/kg/day) by oral gavage. Control and treated groups
were necropsied and tissues were collected at 6 h and 24 h after the
first dose, and 24 h after the 14th dose (336 h); a total of six treatment
groups. Histopathology and gene expression profiling in the liver, and
serum hormone levels, were conducted on samples from this study.
A second oral gavage study was conducted for EPA by Iconix
Pharmaceuticals Inc. (Mountain View, CA) under contract EP-D-05-
006. This study has been previously described in Martin et al. (2007).
However, the genomic data presented in Martin et al. (2007) was
derived from the CodeLink array platform. The genomic data in the
present study were derived from Affymetrix microarrays, using RNA
from the same rat liver samples assessed in Martin et al. (2007). Male
10 week old Sprague-Dawley rats were obtained from Charles River
Laboratories (Hollister, CA) and acclimated for 5-9 days prior to
dosing. Animals received vehicle control (15% Alkamuls EL-620),
myclobutanil (300 mg/kg/day), propiconazole (300 mg/kg/day), or
triadimefon (175 mg/kg/day) every 24 h by oral gavage at a volume of
10 ml/ kg for 72 h total; a total of three treatment groups. Animals were
euthanized by asphyxiation with carbon dioxide and necropsied with-
out fasting 24 h after final dose to assess select endpoints including
serum testosterone levels and transcriptional profiling analysis of
the liver.
Liver RNA isolation. Total RNA was extracted from individual rat
liver samples using TRI Reagent (Molecular Research Center Inc.,
Cincinnati, OH) according to the manufacturer's protocol. For quality
control, RNA A26o^A28o ratios were assessed via NanoDrop Fluoro-
spectrometer (NanoDrop Technologies, Inc., Wilmington, DE). RNA
absorbance readings with a range 1.8-2.1 were followed with DNase
treatment, Total RNA Cleanup (Qjagen RNeasy), and then checked for
RNA quality using a model 2100 Bioanalyzer (Agilent Technologies
Inc., Palo Alto, CA). Samples with a ratio of 28 S:18 S rRNA > 1.6 were
accepted for subsequent use in DNA microarrays. RNA was stored at
— 80 °C until labeling for microarray hybridization or PCR analysis. For
samples from the Gene Logic study, RNA exaction was done by Gene
Logic according to that company's standard operating procedures
(Gene Logic, Gaithersburg, MD).
Rat and human primary hepatocytes. Rat and human hepatocytes
were cultured, and RNA from these cells was prepared under EPA
contract 4D-6182-NTSA with CellzDirect Inc. (Durham, NC). Rat
primary hepatocytes were derived from PND60 male Sprague-Dawley
rats. For each treatment group, two 60 mm plates at a density of
3.5xl06 hepatocytes/plate were generated, allowing for duplicate
measures of each treatment. Hepatocytes were cultured for 24-36 h
prior to exposure of test chemicals. Primary cultures of human
hepatocytes were prepared from human liver tissue from three anony-
mous donors. Tissue samples were derived from normal remnants of
resected liver tissue, which had been removed due to the presence of
metastatic tumors or from non-transplantable donor organs. Primary
hepatocytes were exposed to either vehicle control (DMSO 0.1%);
positive controls: phenobarbital (rat, 200 uM; human, 1000 uM),
pregnenolone 16-alpha carbonitrile (PCN, 30 uM in rat), rifampicin
(Rif, 30 uM in human); myclobutanil, propiconazole, or triadimefon
(10,30, or 100 uM); a total of nine triazole treatment groups from each
specie. The doses used were estimated from an initial dose range-
finding study to represent a maximum tolerated dose without
significant cell lysis. Rat and human hepatocytes were isolated by a
collagenase perfusion method described by LeCluyse et al. (1996,
2000). Final cell viability prior to plating was determined by the
Trypan Blue exclusion test and was >75% in all cases. Following
isolation, hepatocytes were resuspended in DMEM containing 5% fetal
calf serum, insulin and dexamethasone (1 uM) and added to 60 mm
NUNC Permanox® dishes (~4xl06/dish) coated with a simple
collagen, type I, substratum and allowed to attach for 3-6 h at 37 °C
in a humidified chamber with 95%/5% air/C02. After attachment,
dishes were swirled and medium containing debris and unattached
cells were aspirated. Fresh ice-cold serum-free culture medium
containing 50 nM dexamethasone, 6.25 ug/ml insulin, 6.25 ug/ml
transferrin, 6.25 ng/ml selenium (1TS+) and 0.25 mg/ml Matrigel®
was added to culture vessels, which were immediately returned to the
humidified chamber. Medium was changed on a daily basis. Cultures of
hepatocytes were maintained for 36-48 h prior to initiating expe-
riments with chemicals, vehicle, and positive controls. Hepatocytes
were dosed for 3 consecutive days refreshing media every 24 h. Twenty
four hours following the final treatment period, media was removed
from hepatocytes for cytotoxicity analysis through determination of
lactate dehydrogenase leakage from the cells (see below). Cells were
washed with Hanks' Balanced Salt Solution and lysed with RLT Buffer
(Qjagen) and frozen for subsequent analysis.
Hepatocyte cytotoxicity. Cytotoxicity was assessed by measuring
levels of lactate dehydrogenase (LDH) using the CytoTox-ONE™
Homogenous Membrane Integrity Assay (Promega Corp., San Luis
Previous
-------
82
A.K. Goetz, DJ. Dix / Toxicology and Applied Pharmacology 238 (2009) 80-89
Obispo, CA). LDH released into the culture medium was measured
with a 10 min coupled enzymatic assay that results in the conversion
of resazurin into resorufin. One hundred microliters of supernatant
was mixed with 100 ul of CytoTox-One reagent (lactate, NAD+, and
resazurin as substrates in the presence of diaphorase). The mixture
was incubated at room temperature for 10 min, followed with an
addition of 50 ul/well of stop solution. Plates were read using the
FLUOstar Optima (BMG LABTECH, Offenburg, Germany) and
fluorescence was measured with an excitation wavelength of
560 nm and an emission wavelength of 590 nm. Generation of the
fluorescent resorufin product is proportional to the amount of LDH.
The average fluorescence values from sample, maximum LDH release
(total lysis) and culture medium blank were used to calculate the
percent cytotoxicity for a given treatment group.
Hepatocyte RNA isolation. RNA was isolated from lysed hepato-
cytes by initial homogenization with a Turrax Ultra T8 homogenizer.
Samples were individually loaded onto Qjagen AllPrep columns for
isolation of both genomic DNA and RNA from each treatment group.
This method included two steps to remove genomic DNA contami-
nation. The first step was passage of lysates through a gDNA affinity
column that binds strongly to gDNA while allowing RNA to pass. The
second step was removal of gDNA involved on-column DNAse
digestion of gDNA, then subsequent washing steps to remove the
DNAse from the columns. RNA was eluted in RNAse-free water,
analyzed by NanoDrop spectrophotometry, and frozen at — 70 °C. RNA
was confirmed free of gDNA by lack of PCR amplicon. RNA was highly
pure (A26o/A28o>1.7) and not degraded (28 S/18 S rRNA ratio > 1.5).
Microarray hybridization and scanning. The entire microarray pro-
cessing, except for the Gene Logic liver samples, was conducted for
EPA by Expression Analysis Inc. (Durham, NC) under contract 68-D-
04-002. Five micrograms of purified total RNA from 3-6 biological
replicates per treatment group was hybridized to Affymetrix Rat
Genome 230 2.0 or Human Genome U133 Plus 2.0 GeneChip® arrays
according to the Affymetrix GeneChip Expression Analysis Technical
Manual (www.affymetrix.com). In brief, after purified RNA passed
quantity and quality assessment (A26o/A28o ratio with range 1.8 to 2.1
acceptable), double stranded cDNA was synthesized from RNA using
reverse transcriptase and an oligo-dT primer. The cDNA served as the
template in an in vitro transcription (IVT) reaction that produced
amplified amounts of biotin-labeled antisense mRNA. Biotinylated
RNA (labeled cRNA) served as the microarray target. The cRNA was
fragmented using heat and magnesium (Mg + 2), reducing the cRNA to
25-200 base fragments to facilitate efficient and reproducible
hybridization. cRNA was combined with the hybridization cocktail,
containing 3 nM B2 control oligo, 10 mg/ml Herring Sperm DNA,
50 mg/ml BSA, 100% DMSO, and 2x hybridization buffer (NaCl 5 M,
MES hydrate Sigma Ultra, MES Sodium Salt, EDTA Disodium Salt 0.5 M,
10% Tween-20). GeneChips were hybridized at 45 °C for 16 h. After
hybridization the chip was washed and stained with fluorescent
streptavidin-phycoerythrin, binding to biotin for detection. Signal
amplification using anti-Streptavidin antibody and biotinylated goat
IgG antibody was used to bind biotin and provide an amplified fluor
that emits light when the chip is scanned with the GeneChip® Scanner
3000. Gene Logic also used the Affymetrix Rat Genome 230 2.0
GeneChip for gene expression profiling, according to that company's
standard operating procedures (Gene Logic, Gaithersburg, MD) and
the standard Affymetrix protocols.
Microarray data analysis. To minimize non-biological factors such
as the total amount of target hybridized to each array, the signal values
for each hybridization were multiplied by a scaling factor to achieve a
mean intensity equal to 500. The converted .eel files were loaded into
the JMP Genomics program (v6.0.02; SAS Inc., Cary, NC), Log2 trans-
formed, normalized using interquartile normalization, and analyzed
for significant changes in transcript levels through row-by-row
modeling using one-way analysis of variance (ANOVA). For initial
exploratory analysis, principle component analysis (PCA) was applied
using JMP Genomics. Comparisons were made between controls and
each treatment group with statistical cut-offs applied at a p-value
adjusted false discovery rate (FDR) of 5-10%, and an absolute
difference of |1.2| or greater. Probe sets representing transcribed loci,
unknown genes, and image clones were removed from the final list of
each analysis. The Affymetrix .eel files can be accessed through the
Gene Expression Omnibus (GEO; www.ncbi.nlm.nih.gov/geo); series
accession numbers GSE10408, GSE10409, GSE10410 and GSE10411.
Pathway analysis. Ingenuity Pathways Analysis (IPA; Ingenuity®
Systems, www.ingenuity.com) was used for pathway level analyses.
Canonical pathway analysis identified the pathways from the IPA
library that were most significant to the dataset. For each dataset, all
the probe sets on the microarrays were uploaded into Ingenuity™
(rat: 31,099; human: 54,000 probe sets). Focus genes from each
dataset were defined as those with a fold change of |1.2| or greater and
the adjusted p-value for an FDR of 5 or 10%. Focus genes were overlaid
onto a molecular network defined within the Ingenuity Pathway
Knowledge Base (IPKB). Genes that met the fold change and p-value
cut-off, and associated with a canonical pathway in the IPKB were
considered in the analysis. The significance of the association between
the dataset and the canonical pathway was measured using the ratio
of the genes from the dataset that mapped to the pathways divided by
the total number of genes that mapped to the canonical pathway.
Significance was calculated using the right-tailed Fisher's Exact Test by
comparing the number of focus genes that participated in a given
pathway, relative to the total number of occurrences of these genes in
all pathway annotations in the IPKB. Using this methodology, over-
represented pathways containing more focus genes than expected by
chance were identified.
Quantitative PCR. To further compare in vivo and in vitro liver models, a
subset of homologous genes were identified from the microarray datasets,
and these were selected for confirmatory testing. TaqMan^-based quan-
titative reverse transcription polymerase chain reaction (qRT-PCR) was
used to determine the relative levels of Abcbl, Cyplal, Cyp2bl/2, Cyp3al,
Cyp3a2, Cyp4al and Ugtlal mRNAs in rat liver and primary hepatocyte
samples; and ABCB1, CYP1A1, CYP1A2, CYP2B6, CYP3A4, CYP4A11,
UGT1A1 and SLC01B1 mRNAs in human hepatocyte samples. Primer/
probe sets specific for each enzyme were utilized from Applied Biosystems
(Foster City, CA) for Abcbl (Rn00561753_ml), Cyplal (Custom assay, Log
# 1045-28, Lot#001), Cyp2bl/2 (Custom assay, Lot#001), Cyp3al
(Rn01640761_gl), Cyp3a2 (Rn00756461_ml), Cyp4al
(Rn00598510_ml), Ugtlal (Rn00754947_ml), ABCB1
(Hs00184500_ml), CYP1A1 (Hs00153120_ml), CYP1A2
(Hs01070369_ml), CYP2B6 (Hs00167937_gl), CYP3A4
(Hs00430021_ml), CYP4A11 (Hs00167961_ml), UGT1A1
(Hs02511055_sl) and SLC01B1 (Hs00272374_ml). The exception to
this was for Cyp2bl and Cyp2b2, for which the primer/probe set could
detect either gene transcript — that is why these results are hereafter
referred to as Cyp2bl/2. A two-step RT-PCR process was performed by
initial reverse transcription of approximately 200 ng of total RNA in a
60 ul reaction using the High Capacity cDNA archive Kit (Applied Bio-
systems, Foster City, CA), followed by quantitative PCR amplification with
isoform-specific primer/probe sets on 2 ul of each reverse transcribed
cDNA. Reactions were characterized by the PCR cycle threshold (CT)
automatically determined by the PE Applied Biosystems ABI 7900HT
Sequencer software. CT values were within the linear phase (log scale) of
exponential growth for all targets. CT values were determined for target
genes and an endogenous control gene ((3-actin); each sample was
normalized to both (3-actin control and vehicle control. A difference of
one CT was considered equivalent to a two-fold difference in gene
expression (exponential relationship, i.e. RQ=2~DDQ). Sample means
Previous
-------
A.K. Goetz, DJ. Dix / Toxicology and Applied Pharmacology 238 (2009) 80-89
83
for each replicate were determined along with the standard error of the
mean if appropriate and percent of adjusted positive control. Relative
fold changes in mRNA content were analyzed using the Kruskal-Wallis
nonparametric ANOVA with Dunn's multiple comparisons post test,
measures with p< 0.05 were considered significant.
Results
Hepatocyte cytotoxicity
Overall, there was only significant cell lysis for one triazole
treatment group, and for the majority of rat and human samples the
percent cytotoxicity was 1% or less (Fig. 1). There was an increase in
cytotoxicity for one rat sample following 100 uM propiconazole
treatment, but not the other two rat samples in that treatment group.
An increase in cytotoxicity was observed for the human 100 uM
propiconazole treatment group, to approximately a 27% level.
Serum testosterone levels in vivo
In the Gene Logic repeated dose oral gavage study, myclobutanil
significantly increased serum testosterone levels 24 h after a single
exposure of rats by oral gavage (Fig. 2). Elevated testosterone levels for
triadimefon or for myclobutanil at other timepoints did not achieve
statistical significance.
CeneChip quality control analysis
Affymetrix microarrays with a scaling factor greater than 15.0,
indicative of poor data quality, were removed from the analysis
in order to reduce technical variation due to assay noise. Within
the prenatal to adult exposure study, three GeneChips from the
liver dataset were removed prior to normalization and statistical
analysis based on scaling factor. After this removal, 3-7 GeneChips
(i.e. biological replicates) were still available from each treat-
ment group. In the four additional datasets analyzed, no micro-
arrays had a scaling factor of 15.0 or greater, and all arrays were
analyzed.
Probe set level analysis
Statistical significance was determined by a combination of a p-
value threshold adjusted for a false discovery rate (FDR) of 5-10%, and
an absolute difference of |1.2| or greater. The p-value for an FDR of 5%
V
/
3
D Rat 304
D Rat 293
P Rat 256
V V j? s .y
Fig. 1. (A) Rat hepatocyte cytotoxicity: LDH leakage as a function of treatment (B) Human hepatocyte cytotoxicity: LDH leakage as a function of treatment. RFU: relative fluorescence
units. *p<0.01.
Previous
-------
A.K. Goetz, DJ. Dix / Toxicology and Applied Pharmacology 238 (2009) 80-89
300%
- -O - Myclobutanil 225 mg/kg/day
Triadimefon 175 mg/kg/day
6 hr 24 hr 336 hr
Exposure Period
Fig. 2. Serum testosterone levels from the Gene Logic repeated dose study using adult male Sprague-Dawley rats. Triazoles were administered by oral gavage; dose is in mg/kg/day.
*p<0.05
for the Gene Logic 6 h rat liver samples was p<0.00029, for 24 h
p<0.00252, and for 336 h p< 0.00162; for the Iconix 72 h rat liver
samples with an FDR of 5%: p< 0.00036; and for the prenatal to adult
PND92 rat liver samples with an FDR of 10%: p< 0.00072. For the
primary hepatocyte samples the p-value for an FDR of 5% was
p< 0.00004 for rat, and p< 0.00025 for human. Probe sets that
interrogated unknown genes or transcribed loci were removed from
the probe set list. The number of probe sets meeting p-value and fold
change criteria across the 33 treatment groups ranged from zero to
1439, and are presented in Supplemental Data Table 1.
Gene level analysis
Gene level analysis focused on differentially expressed genes
identified by microarray, whether or not they mapped to significantly
affected biological pathways. The goal was to identify genes common
to triazole toxicity, consistent across multiple studies and timepoints,
and conserved between rat and human. Table 1 presents rat genes
from the microarray analysis common to two or more triazoles, and
consistent across multiple studies and timepoints. Only statistically
significant gene expression changes are presented in Table 1 (p-value
Table 1
Common and consistent triazole gene expression changes in rat liver and hepatocytes detected by microarray.
Accession
number
AY082609
AF286167
AF257746
AF072816
NM_053754
NM_024484
NIVL022407
M23995
NIVL133586
L46791
X00469
AI454613
NIVL013105
AI639276
U46118
D38381
NIVL016999
M33936
AA893326
NIVL020540
NMJB1576
NMJM9283
NIVL131906
U95011
NMJM7136
M13506
J02612
AF461738
NM_022228
AA945082
AF228917
Gene
symbol
Abcbl
Abcbl
Abcbl
Abcc3
AbcgS
Alasl
Aldhlal
Aldhla4
Ces2
Ces3
Cyplal
Cyp2b2
Cyp3al
Cyp3a3
Cyp3al3
Cyp3al8
Cyp4alO
Cyp4al4
Cyp4al4
Gstm4
For
Slc3a2
Slcola4
Slcola4
Sqle
Udpgtr2
Ugtlal
Ugtlal
Ugt2al
Yc2
Zdhhc2
Myclobutanil
6 h
1.5
1.5
1.7
1.9
1.3
5.4
1.3
-1.4
1.6
1.4
1.5
24 h
1.7
1.6
1.6
1.4
1.4
1.9
2.7
3.7
2.4
6.0
1.6
1.3
1.4
1.7
1.2
1.8
2.1
1.4
2.1
1.4
1.5
1.3
1.7
1.3
72 h 336 h P92
mid
1.6 1.8
1.9 2.9
5.2
2.2 2.0
4.2 6.2 2.7
1.8
1.6
1.6
1.6
2.0
3.2
1.4
1.5
3.2 2.6
1.5 1.2
Propiconazole
P92 RPH 72 h P92 P92
high 100 mid high
1.5
1.4
2.0 2.2 2.9
4.7
1.7 2.8 2.3
-2.8
3.0
3.3 3.2 2.8 4.7
-2.2
-1.8
1.4
2.1
2.0 2.1
2.0 2.4
1.8
1.4 1.4
2.5 1.9
1.3 1.3
Triadimefon
RPH RPH 6 h
30 100
1.2
1.3
1.4
1.7
-2.0 -3.2
5.2
-3.6 -1.4
-3.8 -1.6
-3.0
1.6
1.2
1.3
24 h
1.6
1.7
1.7
1.6
2.2
2.3
4.2
2.1
5.9
1.6
1.4
2.3
1.4
1.5
1.3
1.9
1.8
2.0
1.8
1.8
1.5
1.5
1.4
72 h 336 h
1.4
1.4
1.5
2.2
1.8
1.9 3.4
3.5
2.4 2.8
2.8 5.7
1.7
1.6
2.0
2.4
1.3
1.7
1.9
2.8
1.6
1.6
1.3
2.6 2.9
1.2 1.3
P92 P92
mid high
1.4
1.4
1.8
2.1
4.8
3.4
5.6
2.8 7.6
1.4
2.2
1.4
2.2
2.0 2.0
1.3 1.7
3.0
1.8
3.1
1.4
RPH RPH RPH
10 30 100
-2.0 -2.8
Note Average fold change is derived from all biological replicates per treatment group. All results are statistically significant, common to two or more triazoles, and consistent across
multiple timepoints or studies. RPH (rat primary hepatocytes).
Previous
-------
A.K. Goetz, DJ. Dix / Toxicology and Applied Pharmacology 238 (2009) 80-89
85
threshold adjusted for an FDR of 5-10%, and an absolute difference of |
1.21 or greater). The common and consistent genes from Table 1 are
readily identifiable as components from a xenobiotic metabolism
pathway modulated by the triazoles, and also potentially significant to
triazole metabolism. These rat genes encode the transporters, and
Phase 1 and 11 metabolic enzymes for hepatocyte uptake, oxidation-
reduction, conjugation and excretion into the canalicular space of the
liver (Fig. 3A). Other genes identified in Table 1 include additional
Phase 1 enzymes, mostly P450 s (Cyplal, Cyp2b2, Cyp3al3, Cyp3al8,
Cyp4alO, Cyp4al4) but also P450 oxidoreductase (For), aldehyde
dehydrogenase (Aldhla4) and carboxylesterase (Ces2); and the Phase
11 enzyme Yc2. The increased expression of Ces2 seen in triazole
exposed livers was not observed in rat primary hepatocytes, in fact,
another carboxylesterase isoform (Ces3) was significantly decreased
in rat hepatocytes by all three triazoles (Table 1; Supplemental Data
Table 7). Extended lists of all the gene expression changes common to
two or more triazoles for each individual rat liver study or timepoint
are presented in the Supplemental Data Tables 2-6, and similar results
for individual genes from rat primary hepatocytes are in Supplemental
Data Table 7. There were many differences between the rat in vivo and
in vitro model systems at the individual gene level. One of many
examples of this was the decreased transcript levels for the gluco-
corticoid receptor (Nr3cl) by myclobutanil and triadimefon after
24 hour exposure in vivo, but not in vitro.
Based on the microarray data alone, changes in gene expression
common to multiple triazoles, and consistent across rat studies were
not conserved in human hepatocytes. In order to identify conserved
triazole effects on the expression of human gene homologs for the rat
Phase 1 and Phase 11 metabolic genes presented in Fig. 3A, results from
qPCR have to be considered. There were, however, numerous genes
commonly affected by multiple triazoles in either the rat or human
primary hepatocytes individually, and these are presented in Supple-
mental Data Tables 7 and 8, respectively.
Only propiconazole and triadimefon had significant effects on gene
expression in human hepatocytes, and commonalities between the
two triazoles were limited; myclobutanil had no overall significant
effect on gene expression. Genes involved in endogenous and xeno-
biotic metabolism (alcohol dehydrogenases ADH1A and ADH1B), and
the regulation of cell cycle and cell fate (IGF binding protein subunit
1GFALS, MAS-related G protein receptor MRGPRF) were all down-
regulated following exposure to the higher concentrations of 30 or
100 uM of propiconazole or triadimefon.
A RAT
Aldhlal
r
fk v^ *"
Triazole c—->j
Q)Slc3a2 ^Cyp3a3
r*Ssico1a4
Triazole-OH
3Gsim4
Udpgtr2
Ugtlal
Ugt2a1
Triazole-CONJ
bl
caX
'-v)Abcg5 I—p/
C—C—CH3'
6—CH CH
O
B HUMAN QCYPIAZ Q)UGT1A1 Q)ABCB1\
r(~ Triazole ^^ Triazole-OH ^^ Triazole-CONJ
Q)SLC01B QCYP3A4
O CH3
C—C—CH3«»
Fig. 3. Proposed CAR/PXR regulated xenobiotic metabolism pathway modulated by triazole antifungals. Indicated genes encode transporters and Phases I and II metabolic enzymes
for hepatocyte uptake of triazoles, and subsequent oxidation-reduction, conjugation and excretion into the canalicular space of the liver. (A) Rat solute carrier family (Slcla4, Slc3a2)
uptake; oxidation-reduction by P450 s (Cyplal, CypSal, Cyp3a3) or aldehyde dehydrogenase (Aldhlal); glutathione S-transferase (Gstm4) or UDP-glucuronosyltransferase
(Udpgtr2, Ugtl al, Ugt2al) conjugation; ATP-binding cassette transporter (Abcbl, AbccS, AbcgS) excretion. (B) Human solute carrier family transporter (SLC01B) uptake; oxidation-
reduction by P450 s (CYP1A1, CYP1A2, CYP3A4); UDP-glucuronosyltransferases (UGT1 Al) conjugation; ATP-binding cassette transporter (ABCB1) excretion. Genes regulated by CAR
(Q) PXR (Q), or both (Q)) are indicated.
Previous
-------
Table 2
Comparison of microarray and quantitative PCR assessment of triazole induced changes in gene expression.
Treatment
72 h rat liver
Myclobutanil
Propiconazole
Triadimefon
GD6-PND92 rat liver
Myclobutanil
Propiconazole
Triadimefon
Treatment
mg/kg/day
300
300
175
134
170
139
MM
Abcbl
Array
1.44
1.16
1.21
1.24
1.48
1.41
qPCR
4.53
1.25
-1.08
-2.37
1.64
-4.99
Cyplal
Array
1.71
1.96
1.35
1.76
3.02
5.60
qPCR
15.07
34.93
5.64
5.82
2131
79.99
Cyp2bl/2
Array3
4.17
3.16
2.80
3.29
4.67
7.63
qPCR
170.21
37.53
26.63
64.57
132.12
63.13
Cyp3al
Array
1.10
1.07
1.16
-1.12
1.08
1.98
qPCR
9.88
7.72
4.16
2.05
2.04
18.20
Cyp3a2
Array15 qPCR
-1.19
-1.10
-1.47
3.22
4.01
1.57
Cyp4al
Arrayc
-1.16
-1.35
-1.33
-1.73
-1.59
-1.30
qPCR
-3.41
-3.68
-4.59
-1.91
-1.53
-2.70
Ugtlal
Array
1.22
1.06
1.16
1.50
1.82
1.78
qPCR
1.30
-1.54
-1.20
1.44
1.32
7.51
Rat primary hepatocytes
PB
PCN
Myclobutanil
Propiconazole
Triadimefon
Treatment
200
30
30
100
30
100
30
100
MM
1.05
1.03
1.08
1.23
1.08
1.45
1.18
1.25
ABCB1
Array
1.37
1.50
1.52
1.30
1.29
1.46
1.81
2.83
qPCR
1.15
1.09
1.08
1.20
1.34
1.09
1.06
1.10
CYP1A1
Array
1.28
-2.29
1.18
1.50
-1.11
1.48
1.10
1.33
qPCR
2.39
2.03
3.22
1.54
3.08
-1.33
1.59
-1.02
CYP1A2
Array
23.42
-2.09
14.66
2.83
10.93
-1.93
3.94
1.66
qPCR
1.80
1.61
1.80
2.03
2.93
1.63
1.84
2.02
CYP2B6
Array
4.30
9.68
14.34
20.83
22.93
9.29
20.32
27.61
qPCR
3.51
3.05
5.98
8.00
7.69
4.56
4.65
6.36
CYP3A4
Array qPCR
-1.56
-1.36
-1.66
-2.65
-1.81
-3.60
-1.75
-2.24
CYP4A11
Array
-4.65
-27.63
-6.80
-3238
- 14.03
- 287.36
-11.24
-53.65
qPCR
1.17
1.14
1.08
1.16
1.65
-1.06
1.06
1.16
UGT1A1
Array
-1.85
-3.54
-1.65
-1.82
-1.80
-3.59
-2.84
-2.51
SLC01B1
qPCR Arrayd qPCR
Human primary hepatocytes
PB
RIF
Myclobutanil
Propiconazole
Triadimefon
1000
30
30
100
30
100
30
100
1.97
1.13
1.19
1.54
1.56
1.49
1.68
2.13
4.97
2.82
2.87
2.59
4.28
1.72
4.18
2.58
1.92
2.70
1.30
1.94
2.27
4.96
1.30
1.57
6.13
11.52
3.50
12.63
18.68
64.28
3.93
5.83
1.06
1.16
1.26
1.56
1.52
1.83
-1.35
1.03
1.82
2.12
4.20
4.43
5.14
3.77
1.48
2.44
2.60
2.11
1.87
1.75
1.83
2.36
1.52
1.59
26.36
27.12
25.22
1331
14.08
8.33
9.67
10.54
2.50 16.05
2.46 35.73
1.60 8.36
1.62 6.68
1.64 6.58
-1.55 1.19
1.84 10.46
1.99 7.66
-1.90
-1.70
-1.20
-1.65
-1.72
-1.35
-1.84
-2.23
-2.86
-2.85
1.11
-2.66
-1.58
-2.34
-2.07
-4.49
1.33
1.35
1.04
1.02
1.05
1.08
-1.03
1.27
7.93 1.01 2.32
9.61 1.02 2.09
8.94 - 1.02 2.70
5.69 -1.02 1.37
9.74 1.01 2.24
2.36 -1.01 -1.24
8.12 -1.01 1.50
5.79 1.02 1.18
Note. Significant fold change relative to control in bold. Reference chemicals: PB (phenobarbital) for CAR; PCN (pregnane 16ct carbonitrile) and RIF (rifampicin) for PXR.
a Probe set representing Cyp2b2 on Affymetrix microarray.
b No representative probe set on Affymetrix microarray.
c Probe set representing Cyp4alO on Affymetrix microarray.
d Probe set representing SLC01A2 on Affymetrix microarray.
Previous
TOC
-------
Table 3
Triazole induced changes in gene expression mapped to biological pathways from both rat in vivo, and rat and human in vitro studies.
Pathways
Androgen-estrogen metabolism
Arachidonic acid metabolism
Arginine-proline metabolism
Ascorbate-aldarate metabolism
Bile acid biosynthesis
Butanoate metabolism
Fatty acid metabolism
Galactose metabolism
Glutamate metabolism
Glutathione metabolism
Glycerolipid metabolism
Glicine-serine-threonine metabolism
Glycolysis-gluconeogenesis
Histidine metabolism
Linoleic acid metabolism
Lysine degradation
Metabolism xenobiotics CYP450 s
Nitrogen metabolism
Pentose-glucuronate metabolism
Propanoate metabolism
Pyruvate metabolism
Retinol metabolism
Starch-sucrose metabolism
Sterol biosynthesis
Tryptophan metabolism
Xenobiotic metabolism signaling
Liver Liver
6 hour 225 24 hour
mg/kg 225 mg/kg
MYC TDF MYC
X
XXX
X
X
X X
XXX
X
X X
X X
X
X
X
X
XXX
X
X
X
X
X
X
X
X X
XXX
TDF
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
Liver
72 hour
300 mg/kg/ day
MYC
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
PPZ TDF
X
X
X X
X X
X X
X
X
X
X X
X
X
X
X X
Liver Liver GD6-PND92 Liver GD6-PND92
336 hour mid-dose high-dose
225 mg/kg/
day
MYC
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
TDF MYC
X
X X
X X
X
X
X
X
X
X
X X
X X
X
X
X X
X X
X X
PPZ TDF MYC
X
XXX
X
XXX
X
X
X
XXX
X
X
X
X
XXX
X
PPZ
X
X
X
X
X
X
X
X
X
X
TDF
X
X
X
X
X
X
X
X
X
X
X
X
X
RPH
30 MM
PPZ TDF
X
X
X
X
X
X
X
X X
X
X
X
X
X
RPH
100 MM
MYC PPZ
X
X X
X
X X
X X
X
X
X
X X
X
X X
X X
X
X
HPH HPH
30 MM 100 MM
TDF PPZ TDF PPZ TDF
X
X X X X
X
X
X XXX
X X X X
X X X X
X
X X X X
X X X X
X
X
Note Only pathways consistently affected across three or more exposure periods are indicated. Complete lists of all pathways for each triazole are in Supplemental Data Table 10A-C
Previous
TOC
-------
A.K. Goetz, DJ. Dix / Toxicology and Applied Pharmacology 238 (2009) 80-89
Genes identified by microarray that were differentially expressed
in response to only individual triazoles, but were consistent across
multiple timepoints or studies, are tabulated in Supplemental Data
Tables 9A-C.
Quantitative PCR
To compare in vivo and in vitro liver effects, and confirm and
expand on the microarray results, seven rat and eight human
homologous genes were analyzed by qPCR (Table 2). PCR analysis
was restricted to the 72 h timepoint of the shorter term repeated dose
gavage studies since all three triazoles had this timepoint in common.
In general, qPCR indicated more robust results than the microarrays;
in several cases the qPCR results demonstrated greater magnitude of
change relative to control, or gained statistical significance where
array results were equivocal. This was the case for rat Abcbl, Cyplal,
Cyp2bl/2, Cyp3al, and Cyp4al; and human ABCB1, CYP1A1, CYP1A2,
CYP2B6, CYP3A4 and UGT1A1, the combination of microarray and
qPCR data confirmed and magnified the effects of triazoles on these
rat and human genes. It is worth noting that for a number of genes and
in vitro chemical treatment groups, there appeared to be a greater or
more statistically significant effect at 30 uM, than at the very high top
concentration of 100 uM.
With the addition of qPCR data, it was clear that the Cyplal
induction common to triazoles in rat liver, was not consistent in rat
hepatocytes. However, qPCR did confirm that induction of human
CYP1A1 and CYP1A2 was conserved in human hepatocytes. The
induction of rat Cyp2bl/2 and Cyp3al, and human CYP2B6 and
CYP3A4 indicated by microarrays as common to the triazoles, was
now clearly a consistent, conserved and robust response in both
species. Consistency across rat in vivo and in vitro results was
especially strong for the inductions of Cyp2bl/2 and Cyp3al, and the
suppression of Cyp4al in both model systems. The common
induction of Ugtlal by all three triazoles in the rat liver following
gestational to adult exposure was detected by a mix of qPCR and
array results, but little to no effect was observed after 72 h exposure
in vivo, or in rat hepatocytes. However, the induction of UGT1A1 in
human hepatocytes by at least two, if not all three of the triazoles
was fairly robust. Response of the rat and human hepatocytes to the
reference chemicals phenobarbital (PB), pregnane 16a carbonitrile
(PCN), and rifampicin (R1F) in regard to expression of ABCB1,
CYP1A1, Cyp2bl/2, CYP2B6, Cyp3al, CYP3A4, and UGT1A1 was
consistent with CAR and PXR activation in the two species. A similar
pattern of expression was observed for all three triazoles, in both
species.
The overall conserved response across rat and human hepatic
systems supported a common, consistent and conserved response
involving a xenobiotic metabolism pathway not only modulated by
the triazoles, but also metabolizing the triazoles in both rat liver and
hepatocytes (Fig. 3A), and human hepatocytes (Fig. 3B). Human genes
encoding the transporters SLC01B and ABCB1, the Phase I enzymes
CYP1A1, CYP1A2 and CYP3A4, and the Phase 11 enzyme UGT1A1 all fit
into this pathway. In contrast to the rat models, there was no
suppression of CYP4A11 in human hepatocytes, indicating some
possible differences in how human PPARa and fatty acid metabolism
responds to triazoles.
Pathway level analysis
A pathway based approach comparing gene expression across
triazoles, time, dose, in vivo and in vitro models, and species placed
effects on individual gene expression into meaningful biological and
toxicological context. Pathways were considered commonly affected
when all three triazoles caused significant effects at a given timepoint.
Consistent pathways were those affected across multiple timepoints,
or between rat liver and rat hepatocytes for a triazole. Conserved
pathways were those affected in both rat and human hepatocytes, by
the same or all of the triazoles. Table 3 shows the 26 biological
pathways significantly affected by triazoles consistently in three or
more exposure periods. Several consistently perturbed metabolic
pathways were common to all three triazoles and also conserved
between rat and human models. These included androgen-estrogen
metabolism, bile acid biosynthesis, P450 xenobiotic metabolism, and
xenobiotic metabolism signaling. Three energy pathways affected in a
consistent, common and conserved manner were glycerolipid meta-
bolism, glycolysis-gluconeogenesis, and starch-sucrose metabolism.
Several pathways consistent and common in both rat models, but not
conserved in humans, were arachidonic acid metabolism and fatty
acid metabolism. Nine of the 33 rat and human treatment groups had
no affected pathways, including all six of the lowest concentration
(10 uM) treatments in vitro (rat and human). Lists of all the affected
pathways from the 33 different treatment groups analyzed are
presented in Supplemental Data Tables 10A-C.
Discussion
The goal of this study was to compare rat in vivo and in vitro
models, to human in vitro models and to identify common, consistent,
and conserved gene expression changes that better characterized the
modes of action for triazole toxicities. The potential of short term in
vivo or in vitro assays for predicting longer-term effects was also
explored in this comparative toxicogenomic analysis. Measuring
changes in gene expression across different exposure periods and
doses, triazole chemicals, and biological models revealed differential
gene expression patterns common to the triazole antifungals and
conserved across the species. These differentially expressed genes
mapped to the fatty acid catabolism, lipid homeostasis, and steroid
and xenobiotic metabolism pathways; representing common biologi-
cal processes affected by triazole treatments in rat liver, and rat and
human primary hepatocytes.
The consistent modulation of target pathways over several in vivo
studies and timepoints indicated that triazoles were affecting fatty
acid catabolism, lipid homeostasis and xenobiotic metabolism in the
rat liver, similar to what has been reported previously in Tully et al.
(2006) and Martin et al. (2007). Several CAR and PXR regulated genes
were differentially expressed by all three triazoles. Short-term
exposure studies showed an increase in Cyp2B and Cyp3A genes,
supporting the hypothesis that all three triazoles caused an early
adaptive response through the CAR and PXR receptors. However,
continuing increased expression of Cyp2B and 3A after longer-term
exposures indicated a maladaptive, toxic response to triazoles; a
viable mechanism of action for the observed hepatomegaly and
hepatocyte hypertrophy by all three triazoles. Increased expression of
specific rat genes in the steroid metabolism pathway following
triazole exposures included Cyp2b2, Cyp3al3, and Cyp3al8. Cyp2b2
and Cyp3al8 are involved in the 16(3- and 6a-hydroxylation of
testosterone, while Cyp3al3 is involved in the metabolism of pro-
gesterone. These changes in gene expression are likely in response to,
and closely related to the effects on steroid homeostasis previously
reported in these studies (Goetz et al., 2007; Martin et al., 2007).
Similar increases in CYP2B6 and CYP3A4 were observed in human
hepatocytes exposed to triazoles.
A few differences stood out between the rat in vivo and in vitro
model systems. The transcript levels for the glucocorticoid receptor
(Nr3cl) were decreased by myclobutanil and triadimefon in vivo, but
not in vitro. Nr3cl, in addition to CAR, is induced by androgen and
xenobiotics and upregulates expression of Cyp2b genes (Honkakoski
and Negishi, 2000). Down regulation of Nrc31 by the two triazoles that
are clearly reproductive toxicants may be significant to understanding
more fully hormone-dependent effects from these exposures.
A select number of SLCs and ABC transporter genes were con-
sistently modulated by triazole exposure. Triazole exposure increased
Previous
-------
A.K. Goetz, DJ. Dix / Toxicology and Applied Pharmacology 238 (2009) 80-89
89
transcript levels of Slcola4 and Abcbl in vivo. Slcola4 is an organic
anion transporter, transporting negatively charged substrates includ-
ing bile acids and estrogen conjugates whereas Abcbl is a multidrug
transporter P-glycoprotein which is active during liver regeneration
and hepatocarcinogenesis (Meng et al., 2002; Ortiz et al., 2004).
Additional transporters involved in steroid metabolism, cholesterol
absorption, amino acid and bile acid transport were also upregulated
by triazoles. The transcript levels of Slcola4, Abcbl, and Abcc3,
mediated by CAR and PXR (Guo et al., 2002; Jigorel et al., 2005, 2006;
Staudinger et al., 2003) were upregulated in the rat liver and
hepatocytes. A similar increase in expression of ABCB1 was observed
in human hepatocytes exposed to triazoles. It is likely that this
induction of transporter genes increased the potential for increased
uptake of triazoles into hepatocytes, and subsequently increased
excretion of hydrophilic metabolites of triazoles. The Phase 111
transporter gene expression profiles suggested increased fatty acid,
bile acid, and triazole metabolite transport in the rat liver and primary
hepatocytes.
This study demonstrated the value of gene expression profiling in
confirming a mode of action and underlying mechanisms of triazole
induced toxicity. Pathway-based and gene-level analyses were able to
define specific patterns of gene expression, delineating common
mechanisms in triazole toxicity. Comparison between liver and rat
primary hepatocytes showed a targeted effect on genes involved in early
fatty acid catabolism, and to a lesser degree on Phase 111 trans porters and
sterol metabolism. Several consistently perturbed metabolic pathways
were common to all three triazoles and conserved between rat and
human models. These conserved pathways included androgen-estrogen
metabolism, bile acid biosynthesis, P450 xenobiotic metabolism, and
xenobiotic metabolism signaling. The subsets of genes involved in fatty
acid catabolism/lipid homeostasis, Phase 111 transporters, and metabo-
lism of testosterone are robust candidates for biomarkers defining
potential mechanisms of action for disruption of testosterone home-
ostasis. Furthermore, these gene expression changes indicate shifts in
lipogenesis to fatty acid oxidation for increased energy production, and
increased expression of Phase 111 transporters for the uptake and
excretion of triazoles in the hepatocytes — both mechanisms of triazole
toxicity which were common across the triazoles. Overall, the results
from these genomic analyses provide new leads for understanding the
modes of action responsible for triazole toxicities. These analyses
revealed functional categories of chemical response genes that indicate
mechanisms and provide direction for further research on triazole
mechanisms of action.
Acknowledgments
The authors thank Drs Wenjun Bao and Russ Wolfinger (SAS Inc.,
Cary, NC) for expert advice on data analysis; and Drs Hongzu Ren
(EPA) and Stephen Ferguson (CellzDirect Inc., Durham, NC) for
excellent technical support. We also thank Dr Douglas Wolf (EPA/
ORD) for technical review of this manuscript; and Ms. Jennifer Hill
for excellent management of the EPA contracts with Expression
Analysis Inc. (Durham, NC), and CellzDirect. Microarrays and
reagents for a portion of this study were provided by Affymetrix
Inc. (Santa Clara, CA) as part of a Materials Cooperative Research and
Development Agreement with EPA. AKG was supported by EPA/
North Carolina State University Cooperative Training Agreement
#CT826512010.
Appendix A. Supplementary data
Supplementary data associated with this article can be found, in
the online version, at doi:10.1016/j.taap.2009.04.016.
References
Allen, J.W., Wolf, D.C., George, M.H., Hester, S.D., Sun, G., Thai, S-E, Delker, DA, Moore,
T., Jones, C, Nelson, G., Roop, B.C., Leavitt, S., Winkfleld, E., Ward, W.O., Nesnow, S.,
2006. Toxicity profiles in mice treated with hepatotumorigenic and non-
hepatotumorigenic triazole conazole fungicides: propiconazole, triadimefon, and
myclobutanil. Toxicol. Pathol. 34, 853-862.
Ghannoum, MA, Rice, L.B., 1999. Antifungal agents: mode of action, mechanisms of
resistance, and correlation of these mechanisms with bacterial resistance. Clin.
Microbiol. Rev. 12, 501-517.
Goetz, A.K., Bao, W, Ren, H., Schmid, J.E., Tully, D.B., Wood, C.R., Rockett, J.C., Narotsky,
M.G., Sun, G., Lambert, G.R., Thai, S-E, Wolf, D.C., Nesnow, S., Dix, D.J., 2006. Gene
expression profiling in the liver of CD-I mice to characterize the hepatotoxicity of
triazole fungicides. Toxicol. Sci. 215, 274-284.
Goetz, A.K., Ren, H., Schmid, J.E., Blystone, C.R., Thillainadarajah, I., Best, D.S., Nichols,
H.P., Strader, LE, Wolf, D.C., Narotsky, M.G., Rockett, J.C, Dix, D.J., 2007. Disruption
of testosterone homeostasis as a mode of action for the reproductive toxicity of
triazole fungicides in the male rat. Toxicol. Sci. 95, 227-239.
Guo, G.L, Staudinger, J., Ogura, K., Klaassen, C.D., 2002. Induction of rat organic anion
transporting polypeptide 2 by pregnenolone-16ct-carbonitrile is via interaction
with pregnane x receptor. Mol. Pharmacol. 61,832-839.
Hester, S.D., Wolf, D.C, Nesnow, S., Thai, S-E, 2006. Transcriptional profiles in liver from
rats treated with tumorigenic and non-tumorigenic triazoles conazole fungicides:
propiconazole, triadimefon, and myclobutanil. Toxicol. Pathol. 34, 879-894.
Hester, S.D., Nesnow, S., 2008. Transcriptional responses in thyroid tissues from rats
treated with a tumorigenic and a non-tumorigenic triazole conazole fungicide.
Toxicol. Appl. Pharmacol. 227, 357-369.
Honkakoski, P., Negishi, M., 2000. Review article: regulation of cytochrome P450 (CYP)
genes by nuclear receptors.]. Biochem. 347, 321-337.
Jigorel, E., LeVee, M., Boursier-Neyret, C, Bertrand, M., Fardel, 0., 2005. Functional
expression of sinusoidal drug transporters in primary human and rat hepatocytes.
Drug Metab. Dispos. 33,1418-1422.
Jigorel, E., LeVee, M., Boursier-Neyret, C., Parmentier, Y., Fardel, 0., 2006. Differential
regulation f sinusoidal and canalicular hepatic drug transporter expression by
xenobiotics activating drug-sensing receptors in primary human hepatocytes. Drug
Metab. Dispo. 34,1756-1763.
LeCluyse, E.L, Bullock, PL, Parkinson, A., Hochman, J.H., 1996. Cultured rat hepatocytes.
Pharm. Biotechnol. 8,121-159.
LeCluyse, E, Madan, A., Hamilton, G., Carroll, K., Dean, R., Parkinson, A., 2000.
Expression and regulation of cytochrome P450 enzymes in primary cultures of
human hepatocytes. J Biochem. Mol. Toxicol. 14,177-188.
Martin, M.T., Brennan, R., Hu, W, Ayanoglu, E., Lau, C., Ren, H., Wood, C.R., Corton, J.C.,
Kavlock, R.J., Dix, D.J., 2007. Toxicogenomic study of triazole fungicides and
perfluoroalkyl acids in rat livers predicts toxicity and categorizes chemicals based
on mechanisms of toxicity. Toxicol. Sci. 97, 595-613.
Meng, L.J., Wang, P., Wolkoff, A.W., Kim, R.B., Tirana, R.G., Hofmann, A.F., Pang, K.S.,
2002. Transport of the sulfated, amidated bile acid, sulfolithocholyltaurine, into rat
hepatocytes is mediated by oatpl and oatp2. Hepatology. 35,1031-1040.
Ortiz, D.E, Moseley, J., Calderon, G., Swift, A.L., Li, S., Arias, I.M., 2004. Identification of
HAX-1 as a protein that binds bile salt export protein and regulated its abundance in
the apical membrane of Madin-Darby canine kidney cells. J. Biol. Chem. 279,
32761-32770.
Staudinger, J.L, Madan, A., Carol, K.M., Parkinson, A., 2003. Regulation of drug
transporter gene expression by nuclear receptors. Drug Metab. Dispos. 31,523-527.
Tully, D.B., Bao, W, Goetz, A.K., Blystone, C.R., Ren, H., Schmid, J.E., Strader, LE, Wood,
C.R., Best, D.S., Narotsky, M.G., Wolf, D.C., Rockett, J.C., Dix, D.J., 2006. Gene
expression profiling in liver and testis of rats to characterize the toxicity of triazole
fungicides. Toxicol. Appl. Pharmacol. 215, 260-273.
U.S. Environmental Protection Agency, 1995. Myclobutanil; pesticide tolerances. Office
of Prevention, Pesticides and Toxic Substances, Washington DC. Fed. Regist. 60,
40500-40503.
U.S. Environmental Protection Agency, 2005a. Myclobutanil; pesticide tolerances for
emergency exemptions. Office of Prevention, Pesticides and Toxic Substances,
Washington DC. Fed. Regist. 70,49499-49507.
U.S. Environmental Protection Agency, 2005b. Propiconazole; pesticide tolerances for
emergency exemptions. Office of Prevention, Pesticides, and Toxic Substances,
Washington DC. Fed. Regist. 70,43284-43292.
U.S. Environmental Protection Agency, 2006. Triadimefon. Preliminary Human Health
Risk Assessment Revised. Office of Prevention, Pesticides, and Toxic Substances.
Washington DC.
Vanden Bossche, H., Marichal, P., Gorrens, J., Coene, M-C, 1990. Biochemical basis for
the activity and selectivity of oral antifungal drugs. Br. J. Clin. Pract. Suppl. 71,
41-46.
Ward, W.O., Delker, DA, Hester, S.D., Thai, S-E, Wolf, D.C., Allen, J.W, Nesnow, S., 2006.
Transcriptional profiles in liver from mice treated with hepatotumorigenic and
nonhepatotumorigenic triazole conazole fungicides: propiconazole, triadimefon,
and myclobutanil. Toxicol. Pathol. 34, 863-878.
Wolf, D.C, Allen, J.W., George, M.H., Hester, S.D., Sun, G., Moore, T, Thai, S-E, Delker, D.,
Winkfleld, E., Leavitt, S., Nelson, G., Roop, B.C., Jones, C., Thibodeaux, J., Nesnow, S.,
2006. Toxicity profiles in rats treated with tumorigenic and nontumorigenic
triazole conazole fungicides: propiconazole, triadimefon, and myclobutanil.
Toxicol. Pathol. 34, 895-902.
Previous
-------
This article was downloaded by: [US EPA Environmental Protection Agency]
On: 2 September2009
Access details: /Access Details: [subscription number 910755252]
Publisher Informa Healthcare
Informa Ltd Registered in England and Wales Registered Number: 1072954 Registered office: Mortimer House,
37-41 Mortimer Street, London W1T 3JH, UK
Toxicology Mechanisms and Methods
Publication details, including instructions for authors and subscription information:
http://www. info rmaworld. co m/smpp/title~content=t713396575
Understanding Genetic Toxicity Through Data Mining: The Process of Building
Knowledge by Integrating Multiple Genetic Toxicity Databases
C. Yang *; C. H. Hasselgren b; S. Boyerb; K. Arvidson c; S. Aveston d; P. Dierkes d; R. Benignie; R. D. Benz f;
J. Contrera f; N. L. Kruhlakf; E. J. Matthews f; X. Han a; J. Jaworska h; R. A. Kemper <; J. F. Rathman '; A. M.
Richard k
a Leadscope, Inc., Columbus, OH b Computational Toxicology, Safety Assessment, AstraZeneca R&D,
Molndal, Sweden c U.S. Food and Drug Administration, Center for Food Safety and Applied Nutrition, Office
of Food Additive Safety, College Park, MD d Unilever, Safety and Environmental Assurance Centre, Bedford,
Bedfordshire, England e Environment and Health Department, Istituto Superiore di Sanita', Rome, Italy ' U.S.
Food and Drug Administration, Center for Drug Evaluation and Research, Office of Pharmaceutical Science,
Informatics and Computational Safety Analysis Staff 10903 New Hampshire Avenue, Silver Spring, MD 9
DuPont Haskell Global Centers for Health & Environmental Sciences, Newark, DE h Procter & Gamble,
Central Product Safety, Strombeek, Sever, Belgium '' Boehringer Ingelheim Pharmaceuticals Inc., Ridgefield,
CT J Department of Chemical and Biomolecular Engineering, The Ohio State University, Columbus, OH k
National Center for Computational Toxicology, U.S. Environmental Protection Agency, Research Triangle
Park, NC
Online Publication Date: 01 February 2008
To cite this Article Yang, C., Hasselgren, C. H., Boyer, S., Arvidson, K., Aveston, S., Dierkes, P., Benigni, R., Benz, R. D., Contrera, J.,
Kruhlak, N. L., Matthews, E. J., Han, X., Jaworska, J., Kemper, R. A., Rathman, J. F. and Richard, A. M.(2008)'Understanding
Genetic Toxicity Through Data Mining: The Process of Building Knowledge by Integrating Multiple Genetic Toxicity
Databases',Toxicology Mechanisms and Methods,18:2,277 — 295
To link to this Article: DOI: 10.1080/15376510701857502
URL: http://dx.doi.org/10.1080/15376510701857502
PLEASE SCROLL DOWN FOR ARTICLE
Full terms and conditions of use: http://www.informaworld.com/terms-and-conditions-of-access.pdf
This article may be used for research, teaching and private study purposes. Any substantial or
systematic reproduction, re-distribution, re-selling, loan or sub-licensing, systematic supply or
distribution in any form to anyone is expressly forbidden.
The publisher does not give any warranty express or implied or make any representation that the contents
will be complete or accurate or up to date. The accuracy of any instructions, formulae and drug doses
should be independently verified with primary sources. The publisher shall not be liable for any loss,
actions, claims, proceedings, demand or costs or damages whatsoever or howsoever caused arising directly
or indirectly in connection with or arising out of the use of this material.
Previous
-------
Toxicology Mechanisms and Methods, 18:277-295, 2008
ISSN: 1537-6516 print; 1537-6524 online
DOI: 10.1080/15376510701857502
informa
healthcare
Understanding Genetic Toxicity Through Data Mining:
The Process of Building Knowledge by Integrating
Multiple Genetic Toxicity Databases
C. Yang
Leadscope, Inc., 1393 Dublin Road,
Columbus, OH, 43215
C. H. Hasselgren and S. Boyer
Computational Toxicology, Safety
Assessment, AstraZeneca R&D, 431 83
Molndal, Sweden
K. Arvidson
U.S. Food and Drug Administration, Center
for Food Safety and Applied Nutrition, Office
of Food Additive Safety, 5100 Paint Branch
Parkway, College Park, MD, 20740
S. Aveston and P. Dierkes
Unilever, Safety and Environmental
Assurance Centre, Colworth House,
Sharnbrook, Bedford, Bedfordshire, MK44
1LQ, England
R. Benigni
Istituto Superiore di Sanita', Environment
and Health Department, Viale Regina Elena
299,00161, Rome, Italy
R. D. Benz, J. Contrera, N. L. Kruhlak,
and E. J. Matthews
U.S. Food and Drug Administration, Center
for Drug Evaluation and Research, Office of
Pharmaceutical Science, Informatics and
Computational Safety Analysis Staff 10903
New Hampshire Avenue, Silver Spring, MD
20993-0002
X. Han
DuPont Haskell Global Centers for Health &
Environmental Sciences, Newark, DE 19714
J. Jaworska
Procter & Gamble, Central Product Safety,
lOOTemselaan, 1853 Strombeek—Bever,
Belgium
R. A. Kemper
Boehringer Ingelheim Pharmaceuticals Inc.,
175 Briar Ridge Rd, Ridgefield, CT,
06877-0368
J. F. Rathman
Department of Chemical and Biomolecular
Engineering, The Ohio State University, 140
W, 19th Ave., Columbus, OH 43210
A. M. Richard
National Center for Computational
Toxicology, U.S. Environmental Protection
Agency, Research Triangle Park, NC 27711
Received 20 November 2007;
accepted 5 December 2007.
This article is not subject to United States
Copyright laws.
Address correspondence to C. Yang, 1393
Dublin Road, Columbus, OH 43215,
Leadscope, Inc. E-mail:
cyang@leadscope.com
ABSTRACT Genetic toxicity data from various sources were integrated into
a rigorously designed database using the ToxML schema. The public database
sources include the U.S. Food and Drug Administration (FDA) submission
data from approved new drug applications, food contact notifications, generally
recognized as safe food ingredients, and chemicals from the NTP and CCRIS
databases. The data from public sources were then combined with data
from private industry according to ToxML criteria. The resulting "integrated"
database, enriched in pharmaceuticals, was used for data mining analysis.
Structural features describing the database were used to differentiate the
chemical spaces of drugs/candidates, food ingredients, and industrial chemicals.
In general, structures for drugs/candidates and food ingredients are associated
with lower frequencies of mutagenicity and clastogenicity, whereas industrial
chemicals as a group contain a much higher proportion of positives. Structural
features were selected to analyze endpoint outcomes of the genetic toxicity
studies. Although most of the well-known genotoxic carcinogenic alerts were
identified, some discrepancies from the classic Ashby-Tennant alerts were
observed. Using these influential features as the independent variables, the
results of four types of genotoxicity studies were correlated. High Pearson
correlations were found between the results of Salmonella mutagenicity and
mouse lymphoma assay testing as well as those from in vitro chromosome
aberration studies. This paper demonstrates the usefulness of representing a
chemical by its structural features and the use of these features to profile
a battery of tests rather than relying on a single toxicity test of a given
chemical. This paper presents data mining/profiling methods applied in a
weight-of-evidence approach to assess potential for genetic toxicity, and to
guide the development of intelligent testing strategies.
Keywords Genetic Toxicity; Databases; ToxML; SAR; QSAR; Structural Alerts
INTRODUCTION
Computational toxicology is the application of computational methods to iden-
tify, characterize, and assess the hazards and risks that chemicals pose to human
health and the environment (Environmental Protection Agency [EPA] NCCT 2007).
One aspect of computational toxicology involves implementing reliable predictive
methods similar to those traditionally relied upon in structure-activity relationships
277
Previous
TOC
-------
(SARs) employed for small molecules. Current research
trends include computational screening of chemicals and
extension to molecular-level understanding via bioassays and
genomics/proteomics methodologies (Richard 2006). The Na-
tional Toxicology Program's (NTP) high-throughput screening
project is a good example of this trend (NTP NICEATM 2007).
Profiling compounds based on both chemical and biological
domains as a first step in understanding complex toxicity can
be applied using predictive toxicology to screen and prioritize
chemicals in both the development and safety assessment stages.
Whether a method involves profiling biological endpoints
or building quantitative structure-activity relationship (QSAR)
models, these computational approaches require sufficiently
large databases with qualified data. Issues related to data
availability, size, and quality are obstacles for both private
and public sector entities. To develop biologically meaningful
methods that accurately predict toxicity requires good coverage
of the relevant chemical space; however, this may not be feasible
due to the labor-intensive nature of experimental toxicology
research. Data sources in private sector entities are often limited
in scope by their targets of interest. In addition, most of the
information accumulated during the lengthy period of product
discovery is typically not kept in an easily accessible format nor
is well designed for data exchange/sharing and integration with
other resources. Thus, for this paper, an integrated database was
constructed from various private sector databases to augment
public and commercial sources with an emphasis on genetic
toxicity as an example of the benefits that can be derived from
having such an approach.
Genetic toxicity is one area in which chemical reactivity
mechanisms and modes of action (MOA) are at least somewhat
understood. The results of a battery of genetic toxicity tests
remain of interest to many industries and regulatory agencies
(Contrera et al. 2005; Matthews et al. 2006a, 2006b; Benigni
et al. 2007). Structural alerts (SAs) developed from genetic
toxicity studies provide a basis for screening for genotoxic
carcinogens and have been discussed extensively in the literature
(Ashby 1985; Ashby and Tennant 1991; Kazius et al. 2005;
Benigni 2004; Kirkland et al. 2005). A comprehensive review
based on a workshop held in support of the European Union's
REACH legislation has also been published (Benigni et al.
2007). Recently, many review articles and communications
have been reported in response to REACH as well as the
Seventh Amendment for Cosmetics (Kirkland et al. 2007;
Jacobson-Kram and Contrera 2007; Tweats et al. 2007; O'Brien
et al. 2006; Cimino 2006; Yang et al. 2006). In this paper, the
benefit of constructing a toxicity database with a rigorous set
of data models for data mining is demonstrated. Further, the
database was profiled for structures and various biological effects
in an attempt to understand the relationships between chemical
and biological domains for various genetic toxicity endpoints.
Genetic toxicity profiles in this paper are based on the
four common study types: bacterial mutagenesis, mammalian
mutagenesis, in vitro chromosome aberration (in vitro CA),
and in vivo micronucleus (MN). An improved understanding
of the chemical-biological domains opens doors for identifying
structural features relevant for each toxicity endpoint. The
ability to determine structural feature-level correlations to
biology allows us to derive MOA knowledge from the data.
This process is demonstrated in this paper using the database
integrated from sources including the U.S. Food and Drug
Administration (FDA), the NTP, and private industries. The
TABLE 1 Composition of the 2006 integrated database based
on data source1
Database
source
CDER
CFSAN-
Indirect
CFSAN-PAFA
NTP + CCRIS
+ Other2
Private
CFSAN-
CDER Indirect
231 0
131
CFSAN-
PAFA
0
0
356
NTP +
CCRIS +
Other2
17
14
184
1,930
Private
0
1
3
43
919
10nly the structurable compound records are included in the count
statistics.
2Tokyo-Eiken 2007.
ultimate goal of the methodology is to implement a systematic
weight-of-evidence (WOE) approach based on clearly defined
and transparent rules to assess toxicity and to guide the
development of intelligent testing strategies.
MATERIALS AND METHODS
Database Coverage
Data sources and ToxML database
The integrated database used in this investigation consists
of publicly available data assembled as the Leadscope SAR-
ready database (2006 version) and a collection of information
from private industries. Table 1 summarizes the sources of
the data in this integrated database. The FDA databases in
Table 1 were developed using a ToxML database standard
through a cooperative research and development agreement
(CRADA) between Leadscope and the U.S. FDA.1 ToxML is
an XML standard developed to represent toxicity experiments
with an ontology to provide an endpoint-specific set of
controlled vocabulary. ToxML has been described in other
articles, including another paper in this issue (Richard et al.
2008). The term "SAR-ready" refers to a database in which the
ToxML experimental data, possibly compiled from multiple
source databases into an integrated database, has undergone
further processing and aggregation to produce summary study-
type endpoint assessments (i.e., assigned "calls" for individual
chemicals). The data content included approved NDAs (new
drug applications), FCNs (food contact notifications) for
packaging, sanitizers, etc., and the FDA's PAFA (priority-based
assessment of food additives) database. The PAFA database
includes toxicity data on a variety of food ingredients, including
direct food additives (e.g., aspartame, sucralose), food contact
substances/indirect food additives (e.g., packaging materials,
sanitizers, plastics additives), generally recognized as safe
(GRAS) ingredients (e.g., sodium benzoate, phosphoric acid),
and flavors. The chemicals in the PAFA database, however, are
not all FDA-regulated food ingredients (FDA CFSAN EAFUS
1 The data compiled for the FDA/Leadscope CRADA are available through
the U.S. Freedom of Information Act (FOIA); no proprietary data were used
to build the FDA CRADA databases.
C Yang et al.
278
Previous
TOC
-------
TABLE 2 Study type and genetic toxicity recorded in the SAR-ready database
Study types
Compound level calls
Bacterial mutagenesis
Mammalian mutagenesis
In vitro chromosome
aberrations
In vivo chromosome aberrations
In vivo micronucleus
In vitro micronucleus
Bacterial mutation. Salmonella mutation, strain-level mutations forTA100, TA98,
TA1535, andTA1537
Mammalian mutation, MLA, CHL V79, CHO HPRT mutations
In vitro chromosome aberration, CHO chromosome aberration, CHL (v79) chromosome
aberration, human blood (leukocytes, peripheral blood, etc.) chromosome aberration
Insufficient data for analysis
In vivo MN, mouse MN (peripheral blood, bone marrow), rat MN (peripheral blood, bone
marrow)
Insufficient data for analysis
2007). Approximately 70% of the compounds in the PAFA
database are flavoring agents. Other public sources include
the NTP (NTP 2007), the Chemical Carcinogenesis Research
Information System (CCRIS 2007), Tokyo-Eiken (Tokyo-Eiken
2007), and several primary publications (Crebelli et al. 1999).
These public databases are enriched with chemicals termed in
this paper as "industrial" covering a wide variety of compounds,
including agricultural chemicals and consumer products. In this
paper, chemicals that are not classified as drugs/candidates or
food ingredients are, in general, grouped as "industrial." The
Carcinogenic Potency Database (CPDB), a popular source for
Salmonella mutagenicity data, was not included in the SAR-ready
or integrated databases because no study protocols or references
are available from the source. Approximately 73% of the CPDB
compounds with Salmonella data, however, are found in the
SAR-ready database with study details, references, and calls.
A total of four private industry databases constructed
according to the same ToxML standard were used as examples
to demonstrate the benefits of integration. These private
sources were categorized as drugs/candidates, food additives,
and industrial chemicals. The therapeutic indications associated
with the drugs/candidates included inflammation, immunol-
ogy, cardiovascular, gastrointestinal, respiratory, antipsychotics,
analgesics, anticancer, anti-infection, and antiviral categories.
The actual chemical structures of the proprietary compounds
were not required and not used for any of the analyses in this
paper; the sharing of knowledge derived from these structures
was accomplished solely through general structural feature
statistics. Definition and methods of obtaining structure fea-
tures and statistics are described in this Materials and Methods
section.
ToxML evaluation of genetic toxicity
studies
The ToxML standard has been used to transform public
databases as well as to construct databases from FDA and private
sector data. In addition, the process of building an SAR-ready
genetic toxicity database was developed by defining a set of
fields to meet the inclusion criteria to be used in endpoint
assessment. The criteria were defined by consultation with
scientists from the FDA and industry. Table 2 lists the genetic
toxicity study types in the SAR-ready database; it contains a
total of 2,428 compounds in six genetic toxicity study types.
Table 3 lists specific fields and subfields that were deemed
critical for study validation. These guidelines were also used
to resolve call conflicts that arose when combining studies from
multiple sources: calls were assigned for the studies only if the
information defined in Table 3 was available. A compound
was not used if there were no acceptable studies meeting these
criteria reported in the database. The ToxML approach used
here for genetic toxicity endpoint assessment applies transparent
and systematic rules across the diverse databases to aggregate
data from individual experiments across test systems to define
a "call" for each chemical compound. A single study was
designated "positive" if any of the test strains, cell lines, or
target cells resulted in a positive response. A single compound
was designated positive if it was reported positive in more than
50% of the studies. In contrast, a compound was considered to
be negative if reported negative in more than two-thirds (67%)
of the studies. The outcome for a compound reported positive
in 33% to 50% of the studies is assigned as "intermediate."
In the binary classification analysis presented in this paper,
TABLE 3 ToxML fields essential for determining compound-level calls in the genetic toxicity database
Top levels of ToxML
Compound level
Study level
Test level
Fields
ID, Chemical name, InChl
Background
System
Control
Dosage regimen
Conditions
Results
Fields
ID (Registry numbers. . .), chemical name (IUPAC name
active ingredient. . .)
Study type, study source, study start date, reference
Species, strain, metabolic activation, sex, cell, cell line.
of exposure
Negative control, positive control
, Orange Book
target cell, route
Concentration/dose, solvent/vehicle, frequency, treatment time
Solvent vehicle, scoring technique, stain
Test call, cytotoxicity, precipitation
279
Data Mining Integrated Multiple Genetic Toxicity Databases
Previous
TOC
-------
compounds in the intermediate group were included with
the positives. If a compound was reported only once in the
database, but met the requirements shown in Table 3, then
the compound "call," or endpoint assessment, was given as
reported in the source database. Calls were made for the test
endpoints that are more commonly accepted by regulatory
agencies (Table 2). In vivo chromosome aberration and in vitro
MN study results were not used because of insufficient data
to apply systematic rules for construction of endpoint calls.
Merging this SAR-ready database with the private industry data
using the same assessment criteria resulted in the integrated
genetic toxicity database of 3,220 chemicals with four genetic
toxicity study types, which were then used for the data mining
and analysis.
Structural Analysis
Structural feature-based representation
The Leadscope genetic toxicity databases are represented
structurally by chemical classes. Structural features defined in
the Leadscope chemical hierarchy (Roberts et al. 2000) were
used to characterize and analyze compound classes of the
various genetic toxicity databases. Leadscope features include
classes of benzenes, functional groups, heterocycles, fused
rings, and pharmacophores. Leadscope software fragments the
chemical structure of a given compound into corresponding
structural features (i.e., specific functional groups, substitution
patterns, etc.) for automatic grouping. The features describing
the chemical structures in the database were exported as a table
of chemical fingerprints from Leadscope v2.4 for each of the
data sources in the integrated database. Examples are shown
in Table 4. The fingerprint table is a matrix of [^(compound
structures) x /'(chemical features)], where each row corresponds
to one particular structure (compound) and each column
corresponds to a chemical feature; a value of 1 indicates that
the compound contains that particular feature, whereas a value
of 0 indicates that it does not.
Principal component analysis
Principal component analysis (PGA) is a multivariate data
analysis technique with numerous diverse and important
applications. One important feature of PGA is that it is ideal
for handling data sets in which the number of observations,
>S(compound structures), is less than the number of descriptor
variables, F (chemical features), a situation for which conven-
tional modeling approaches are inadequate. In the computation
of principal components (PCs), the original data can be thought
of as S independent observations of F random variables. The
maximum possible number of principal components equals
the rank (r) of the data matrix, and therefore r < min(S,
F). Generally, a large proportion of the overall variability is
accounted for by a small number,/?, of the PCs, where p <^ r.
Thus, the original S x F matrix, [SF], can be replaced by a new
matrix of S observations x p latent variables with very little loss
of information, where p <$; F. PCA as such is an unsupervised
technique (i.e., its aim is simply to project the original high
dimensional set into a lower dimensional one spanned by
orthogonal components). Nevertheless, it can be used for
pattern recognition purposes as well. Pattern recognition on
the principal component scores is performed as with any other
type of descriptors; however, pattern recognition on PCs has
C Yang et al.
several noteworthy advantages. In many instances the first
component, even though it is the one explaining the highest
proportion of variance, only reflects the scaling properties of
the studied system, and is a pure "size" indicator, whereas the
next components describe the "shape," or quality properties
of the system (Darroch and Mosimann 1985). In addition, the
principal components are orthogonal to each other and, hence,
are uncorrelated and do not carry any redundant information;
this allows for the generation of well-conditioned models by
avoiding mutual correlation of regressors that creates ambiguity
in the model selection procedure. Since the PC scores are linear
combinations of all of the original variables, each of the original
variables contributes to each PC in the presence of all other
variables, making this a truly multivariate approach (Jolliffe
2002; Benigni and Giuliani 1994).
Structural domains of the databases can thus be visualized
by plotting various PCs. Although the full /^-dimensional PC
space cannot be visualized if/1 > 3, projection plots in 2-
or 3-dimensions are often very useful. Leadscope structural
features were used as independent variables; structural features
appearing in 20 or more compounds were included in this
unsupervised analysis. As noted, it is generally possible to select
a small number, p, of PCs that explain most of the variation
in the data set. In order to discard PCs not relevant for the
problem under study, the eigenvector coefficients of the/1 PCs
are analyzed. Removed from consideration were PCs associated
with eigenvectors in which all coefficients are the same sign
(either all positive or all negative). As stated above, the PCs
with all coefficients of the same sign are "size" components, and
describe the number of "active" Leadscope features, whereas the
PCs with mixed signs describe the different types of chemicals in
the databases: the latter is the use of PCs that we were interested
in. The principal component analyses in Figures 2 and 4 were
performed using MatLab 7.4.
Statistics for characterizing the
chemical-biological domain
Z-SCOre StdtistiC. The compounds in the database were
grouped by pre-defined substructure searching. The resulting
chemical classes can be correlated with the biological responses,
after which the mean and the standard deviation of each
compound class are calculated. The z-score statistic provides a
metric to estimate whether the compound class is more highly
associated with the biological response than the mean of the
whole database. The z-score compares the mean activity of a
subset to the expected value according to:
z=(xi- x0)
(1)
where x\ and x-i are the mean responses of the subset and
full set, respectively, and n\ and ng are the set sizes and so2
is the sample variance of the full set. Equation 1 represents a
z-score adjusted for set sizes to correct for the fact that small
sets with extreme values would have too large an impact if
conventional z-scores were used. For example, the adjusted z-
score will balance the following two situations: (1) a particular
class has only a few compounds but all have much higher values
than the average of the whole data set; (2) a particular class has a
large number of compounds but also a broad distribution with
280
Previous
TOC
-------
TABLE 4 A selection of influential differentiating features of chemicals in the integrated database acrn*;*; the nenetic tnxicitv cmdnnint<;
Chemical features
Alcohol alkenyl, cyc-
Alkyl
Aryl
Aldehyde Alkyl
Aryl
Amlne amirte(NH2). alKyt-
am>ne(NH2). aryl-
sec-ammef NH) aryl-
lart-amine, aryl-
Azo
Bases, nucleositfes
Benzene benzene. l-alkylammo-
benzene. 1-carbonylarnino-
benzeoe 1 -heleroamino-
benzerte, 1 -methyl-
benzene. t.2.3.4-fused
Epoxide
Ether ether, alkenyl. cyc-
Furan
Malide halkte. alkenyl-
haltde. alkyl-
halide. p-alkyl-
nalkte. s-alkyl-
tialide aryi-
Hydrazine hydrazine. alky), acyc-
hydrazine. aryl-
Kelone ketone. af kenyl , cyc-
ketone. aryl-
Iminomethyl. alkenyt, eye-
Nilro nil/o
nitro, aryi-
Nilroso and Nitrosamine
Phosphorous groups
Heterocyde benzimtaazole
benzopyran
imidazole
pyrazine1. Cells with z-soores >2 are in bold fonts lo indicate a strong association with positives. The full feature table of the integrated database will be
available online: http:Wivww.lead5CQpe.com. http «epa.gov'ncct'dsstox and httpiWambit.acad.bg
Previous
-------
its own average being greater than the that of the whole data set.
In this case, the adjusted z-score will be greater for the second
case although the class mean is smaller than in the first case. In
this paper, all the z-scores were calculated using equation 1. The
features and statistics (means, z-scores, and frequencies) of each
of the database sources were exported from Leadscope v2.4.
The database sources include the SAR-ready genetic toxicity
database and the private collections. The exported data were
combined and the statistics (means, z-scores, and frequencies)
were recalculated for the entire integrated database outside
of Leadscope software. Examples of the resulting integrated
database are listed in Table 4.
Multivariate Correlation. Finding structural features
that correlate with toxicity can be mathematically represented
by:
[SFf[ST] = [FT]
(2)
where [SF] represents a fingerprint table. [ST] is the response
matrix, with each row corresponding to a structure and each
column a toxicity endpoint of interest. The [FT] matrix
obtained by the above operation is thus an array in which
each row corresponds to a particular chemical feature and
each column a toxicity endpoint. Thus, [FT] provides a direct
connection between structural features and endpoints, which
subsequently is used to discover structural alerts through the
analysis techniques presented in this paper.
For profiling chemicals with structural features and toxicity
endpoints, relevant structural features must be selected. These
structural features are correlated with toxicity endpoints, and
selected features become the independent variables in the
subsequent multivariate analysis. The feature selection criteria
to differentiate with toxicity endpoints are summarized below.
For a given endpoint, features that give the highest absolute
z-scores are selected. Across the endpoints, features resulting in
the largest variance across the different endpoints are selected,
which is equivalent to finding the largest difference between
minimum and maximum values. It is also desirable to find
features with the highest or lowest mean values across the
endpoints with small variance. In addition, the frequency
of a feature (i.e., the number of compounds that a feature
describes in the data set) is also considered. The correlation
is quantitatively expressed by the Pearson correlation:
(3)
where Sj and Sj are the variances of the i andj variable and Sy the
covariance. A pair-wise Pearson correlation coefficient for assays
i and j is calculated using z-scores of selected features against
the particular test endpoints using features as observations.
RESULTS AND DISCUSSION
Chemical Domain of Integrated
Databases
Structures in the database
Approximately 16% of the structures in the SAR-ready
database are drugs, 16% are food ingredients (direct and
C Yang et al.
indirect), and 68% are industrial chemicals. When the four
private sources are combined with the SAR-ready database
to create the integrated genetic toxicity database, the counts
for drugs/candidates increased to 30% and the general indus-
trial chemical counts decreased to 48%. Table 1 shows the
compound counts by data sources included in the integrated
database. There are very few overlaps between the various
sources with the exception of the PAFA data from CFSAN.
About 52% of the PAFA food ingredients were also found in
NTP/CCRIS, as shown in Table 1. The chemicals present in
the intersection of PAFA and NTP/CCRIS are mainly flavoring
(78%) and coloring agents (15%).
The chemical structures in the databases can be further
characterized by structural classes and substructural fragments.
The NTP and CCRIS databases contain mostly industrial
chemicals with increased presence of pyran(H), benzopyran,
and alkyl and aryl halides features. In contrast, drug compounds
from the FDA CDER enrich the heterocyclic classes but do not
generally contain reactive groups associated with mutagenicity
such as unsaturated aldehydes or N-nitroso features. Food
ingredients do not contain carbamates or common heterocycles
such as 1,4-dihydro pyridines, thiophenes, and thiazoles.
Figure 1 depicts the Leadscope compound class distribution of
the integrated database. Structural features differentiating the
three broad substance use types, without regard to endpoint
classification, are selected as examples.
Structural domain comparisons
Two of the most important reasons for building the
integrated database from multiple sources were (1) to diversify
and augment the chemical spaces of the public databases, which
are heavily biased toward industrial chemicals; and (2) to address
the biological responses of the various chemical domains and
determine if SAR patterns relating the chemical and biological
domains can be discovered. In addition, the integrated database
permits an analysis of the biological activity associated with
specific chemical structural features when these features are
present in different chemical environments. It is the context
of where and how the features are positioned and with the
combinations of other features in the whole molecule that is
critical. The makeup of a whole molecule by various structural
features is manifested in the chemical substance use types, for
example, drugs vs. surfactants. For this reason, adding biological
and chemical structure information of drugs/candidates and
food ingredients to the set of general industrial chemicals
was extremely important. It should be stressed again that the
chemical structure information was in the form of features
statistics as shown in Table 4.
Approximately 50% of the SAR-ready database structures
were randomly selected and their Leadscope structural features
exported as chemical fingerprints from Leadscope v.2.4. This
random sampling preserved the ratio between the chemical
substance use types (i.e., drugs, food ingredients) and general
industrial chemicals. The structures were then clustered using
principal component analysis (PCA) using the structural feature
fingerprints as descriptors. In this analysis, only those features
present in 20 or more compounds were included. As noted
previously, a large proportion of the variation in the data
is typically captured by a relatively small number of latent
variables (principal components). If one chooses p PCs, then
282
Previous
TOC
-------
Drugs/candidates
Food ingredients
Industrial chemicals
Aldehyde
Carbamate
Epoxide
Halide
alkyl halide
aryl halide
Heterocycle
Imidazole
Naphthalene, 1-subst
Nitro
Oxolane(H)
Pyran(H)
Piperidine
Pyridine
Quinoline
Thiazole
10 100
0.1 1 10 100 0.1 1 10 100 0.1
% Frequency of features
FIGURE 1 Structural classes differentiating the broad substance use categories within the integrated database (3,220 compounds).
benzene polyhalides
azo dyes and
benzene sulfonates
PC,
FIGURE 2 Chemical domain of the databases represented in the latent variable space defined by the three most differentiating principal
components. Orange denotes drugs/candidates, green for industrial, and blue for food ingredients chemicals. A: SAR-ready databases
only. B: Integrated databases.
283
Data Mining Integrated Multiple Genetic Toxicity Databases
Previous
Next
-------
similar compounds will tend to cluster together in the p-
dimensional space defined by the PCs.
The scores of the three most strongly discriminating princi-
pal components (PC2, PCS, and PC4) are plotted in Figure 2A
for the SAR-ready database. The PC plots effectively differenti-
ate the various database sources; most of the defined organics in
the group of industrial compounds from NTP/CCRIS clustered
together except for two local islands. A portion of the industrial
chemicals, namely the benzene polyhalide group, clustered
separately with most of the loading on PCS. The other island
included benzene sulfonates and azo benzenesulfonyl dyes.
Some of the food ingredients such as flavoring and coloring
agents were clustered closely with the industrial chemicals.
Many of the marketed drug compounds with heterocycles were
mostly differentiated as outliers in the SAR-ready database.
When compounds from the four proprietary collections
were added to the SAR-ready database and PCA repeated,
the structures from chemical and consumer product industries
overlap significantly with the chemical space of NTP/CCRIS
and food ingredients (Fig. 2B). However, the region of chemical
space populated by drug candidates expanded considerably and
is clearly separated from the structures in NTP/CCRIS. The drug
space can be further differentiated by the therapeutic indications
using the same structural features. This domain analysis
validates a notion of drug-like compounds and biology-driven
chemical analogs. Structural domains give further insights
beyond those obtained from analysis of the experimental or
calculated physical properties of various structures.
Biological Domain of the Integrated
Database
Endpoints of toxicity database
Six genetic toxicity study types were included in the SAR-
ready database (Table 2) and four types within the integrated
database were analyzed in depth. Bacterial mutagenesis data
were grouped by four individual Salmonella strains. Mammalian
mutagenesis data were evaluated from mouse lymphoma (MLA)
studies. In vitro chromosome aberration data were grouped by
Chinese hamster lung (CHL) or Chinese hamster ovary (CHO)
cell tests. In vivo micronucleus data include both mouse and rat
studies. These further aggregations of data represent "endpoint
assessment" to define activity categories for use in SAR analysis.
The databases are also profiled for their distributions of
genetic toxicity study calls (i.e., positive and negative com-
pounds). These study call counts reflect the result of the
aforementioned endpoint assessment rule, instead of that given
in the original databases. For example, eugenol (CAS 97-53-0,
2-Methoxy-4-(2-propenyl)phenol) was considered positive in
the CHO chromosome aberration test with and without S9 in
the study reports from the PAFA database. However, the NTP
study report considers this chemical to be a negative without S9,
and weakly positive with S9 under similar test conditions. In the
SAR-ready and integrated databases, the compound was thus
counted as positive for CHO chromosome aberrations, both
with and without metabolic activation. However, in all cases,
the original test and study calls are preserved in the database
along with the treatment-level results.
Distributions of endpoint outcomes
The distributions shown in Figures 3 to 6 represent the
compound statistics when the ToxML endpoint criteria were
applied to determine an aggregated call for each compound
for a given endpoint. In general, both FDA submissions
and private collections are quite low in positive compounds.
The number of positive compounds increased significantly by
adding data from the public sources. Taking the Salmonella
endpoint as an example, drugs from the FDA CDER database
contain fewer than 9% positives, whereas the food ingredients
and the industrial chemicals from CCRIS/NTP give 14% and
39%, respectively. Thus, in the integrated database, the overall
frequency of positives is 33%.
3000
2500
2000
1500
1000
500
0
I positive
j negative
FIGURE 3 Distribution of negative/positive outcomes for Salmonella reverse mutation.
C Yang et a/.
284
Previous
TOC
-------
1000
800
600
400
200
J positive
negative
#
FIGURE 4 Distribution of negative/positive outcomes for in
vitro mammalian mutation.
The results in Figures 3 to 6 also demonstrate that many
more positive studies have historically been reported in the
in vitro chromosome aberrations and mammalian mutagenesis
than in bacterial mutagenesis or in in vivo MN. In the MLA2
test, the proportion of positives is high for the industrial
chemicals from CCRIS and NTP (71%); the same source also
gives 52% positives for in vitro chromosome aberration studies.
Drugs/candidates that have been selected for development seem
to result in much lower positives in the MLA tests (14% positive)
as well as in the in vitro chromosome aberrations (26%) tests
than industrial chemicals. One of the likely explanations is that
drug candidates that were found to be positive in these tests were
dropped from development and their data were not reported.
The positive frequency trend of in vivo MN follows the pattern
of Salmonella mutagenesis: 7% for CDER drugs, 33% for the
food ingredients from CFSAN, 43% for the chemicals in CCRIS
and NTP, and 33% for the integrated database. Of the CFSAN
compounds, more than 90% of the positive compounds in
Figures 3 to 6 are from CFSAN PAFA data; 70% to 80% of
these PAFA positive compounds are flavoring agents.
Exploring correlations between observations across the
genetic toxicity experimental results using a read-across strategy
requires data for many compounds and many different toxico-
logical endpoints. However, at a compound level, there were
only 65 total compounds out of 3,220 in common that had all
possible test results across the Salmonella, MLA, CHL, or CHO
in vitro chromosome aberration, and rodent MN endpoints. A
partial explanation for the low number of compounds with
a complete test profile is that the integrated database does
not capture all available genetic toxicity data; our stringent
2The MLA data included in the combined database were not evaluated
using the new global equivalency factor recommended by the International
Workshop for Genotoxicity Testing. Re-evaluation of the MLA data using
this factor would undoubtedly decrease the percentage of positive MLA
findings in this data set (Moore et al. 2006).
285
inclusion criteria forced us to exclude many data points. Based
on this sparse data resource, it is impossible to obtain a
meaningful correlation across the genetic toxicity endpoints at
the compound level. However, a strategy for resolving issues
related to data scarcity in the chemical-biological domain is
presented in the next section. By expanding the chemical space
from compounds to structural features, biologically meaningful
and statistically valid correlations can be made.
Chemical and Biological Activity
Domain Analysis
In this section, the biological activity domain of the
integrated database is analyzed in greater detail at the compound
level across the various endpoints, similarities and dissimilarities
of findings between the tests are compared, and then the
analysis is expanded to the feature level. In compound-level
analysis, a specific endpoint outcome of a single compound
is used as an observation. In feature-level analysis, the average
outcome for the particular endpoint is calculated so that the
observation is linked to specific chemical structural features
that may be present in many compounds. Finally, structure
domain profiling across the four genetic toxicity endpoints
is explored at the feature level. Correlations of selected
structural features with various endpoints are used to link
the observations in the two domains, chemical structures and
biology.
Reverse bacterial mutagenesis profile
Since the Salmonella test is one of the most common and
important genetic toxicity tests, detailed discussion at the strain
level is presented. Common questions asked when testing is
performed using different strains include whether there are
correlations between test results of different Salmonella strains
and whether there is subsequent structural differentiation.
Table 5 shows contingency tables for the compound counts
for the responses across four Salmonella strains. The numbers
in parentheses represent the expected values of the appropriate
null distribution; that is, these counts are what one would expect
if the two strains being compared were completely independent
(uncorrelated). Large differences between actual and expected
counts suggest a significant association between two strains.
For example, only 55 compounds were found to be TA100
negative and TA1535 positive; under the null hypothesis, 235
compounds were expected in this category. This table indicates
that the concordance (overall agreement on both positives
and negatives) of any two strains ranges from 85% to 90%.
However, when examined separately, sensitivity (agreement on
positives) and specificity (agreement on negatives) values reveal
some interesting patterns. For positive TA1535 or TA1537,
there is a 0.85 probability that TA100 is positive. If TA1537
is positive, then the probability that TA98 is also positive is
~0.85, whereas the proportion of positive TA1535 compounds
that are also TA98 positive is only 0.55. This may be understood
from the standpoint of mutation mechanism differences (i.e.,
base substitution vs. frameshift mutation). It is interesting
to note that the reverse observation is not always true. For
example, only 50% of TAlOO-positive compounds are TA1535
positive; similarly, only 50% of TA98-positive compounds
are positive in TA1537. This means that TA100 and TA98
strains are much more sensitive and/or not differentiating than
Data Mining Integrated Multiple Genetic Toxicity Databases
Previous
TOC
-------
1000
800
600
400
FIGURE 5 Distribution of negative/positive outcomes for in vitro chromosome aberration.
their counterparts, TA1535 and TA1537, respectively. This
observation can be explained by the fact that the pKMlOl
plasmid was introduced into the former two strains specifically
to increase sensitivity (McCann et al. 1975).
If TA98 is positive, then the probability of TA100 also
being positive is ~0.85; however, a positive TA100 outcome
corresponds to a positive TA98 outcome for only 68% of the
compounds. This suggests that the TA98 test may be a more
conservative test than TA100. This analysis may be interpreted
as an indication of more frequent base substitution mutations
than frameshift mutations for the chemicals in this database.
Examining the negatives, if either TA100 or TA98 is negative,
then probabilities that the other three strains will also be
negative are >0.90 in both cases. These observations may justify
the use of TA98 and TA100 strains to make a conservative
decision to avoid false negatives. Based on historical data,
if the compound is negative in both TA98 and TA100, the
compound has a >0.9 likelihood of being negative in all other
strains. A similar recommendation based on analysis of test data
across these Salmonella strains for more limited chemical space
coverage was previously published by NTP (Mortelmans and
Zeiger 2000).
This investigation also examined the correlations between
the biological activity response and chemical structural fea-
ture domains. Within the chemical domain, a compound is
described by specific chemical features, which in turn are
used for data mining of SARs. For simple visualization, the
Salmonella data were interpreted according to the ToxML
criteria for endpoint assessment (i.e., if any one of the four
test strains is positive, then the Salmonella assay is considered
positive). In Figure 7A, Salmonella outcome was then projected
onto the chemical domain of the integrated database for
three of the different substance use types identified by PGA
(Fig. 2B). Not surprisingly, most of the hot spots are found
within the industrial chemical clusters located in the center.
Several local islands within this dense space are identified
with high frequency of Salmonella mutagenicity, including
nitroso groups, aromatic polyhalides, aromatic amines, and
epoxides. In addition, a cluster along the diagonal of the
PC2 and PCS axis with positive PC2 scores clearly separates
a group of drugs/candidates with positive Salmonella mutagenic
potential (Fig. 7A). It's worth noting that these drugs are mostly
antineoplastics and antivirals, which often have exemptions for
positive genetic toxicity studies. About 30% of the cluster of
azo dyes and sulfonyl benzenes was associated with Salmonella
mutagenicity. This analysis gives a rationale for pursuing local
group-based QSARs for Salmonella mutagenicity. Not only can
structural domains be differentiated, but the biological profiles
can also be differentiated using structural features.
SAR analyses of Salmonella mutagenicity data of chemical
structural features have been widely published (Benigni 2004;
Vogel et al. 1998). Well-known chemical classes identified in
TABLE 5 Contingency table for Salmonella mutagenicity in the integrated database
TA100 +
TA100-
TA1535 +
248 (68)
55 (235)
TA1535-
250 (430)
1680(1500)
TA100 +
TA100-
TA1535 +
TA1535-
TA1 537 +
147(36)
28(139)
100(21)
72(151)
TA1 537 -
219(330)
1399(1288)
122(201)
1531 (1452)
TA100 +
TA100-
TA1535 +
TA1535-
TA1537 +
TA1537-
TA98 +
488(161)
124(451)
1 53 (44)
196(305)
144(27)
135(252)
98-
230 (557)
1885(1588)
130(239)
1743(1634)
29(146)
1493(1376)
Values in parentheses are the expected compound counts based on the null hypothesis.
C Yang et al.
286
Previous
TOC
-------
500
400
300
200
100
positive
negative
Oil
FIGURE 6 Distribution of negative/positive outcomes for in vivo micronucleus.
those studies, capable of direct alkylation or cross-linking of
DNA molecules, are also found in the present databases. Table 4
summarizes the comparisons across the four Salmonella strains
for prominent structural features in the database. The major
structural features associated with Salmonella mutagenicity are p-
alkyl halides (including nitrogen mustard), nitroso compounds,
epoxides, aziridines, and O-alkyl sulfites. Structural features
differentiating individual Salmonella strains can be also found.
For example, epoxides, nitroso compounds, hydrazines, and
alkyl halides seem to preferentially induce point mutations of
TA100 and TA1535. Of the two, TA100 seems to be affected
by much wider types of compound classes and, therefore, it is
much less discriminating.
TA98 and TA1537 strains reflect frameshift mutations
and are associated with intercalating agents or large reactive
groups. Polycyclic aromatic hydrocarbons (PAHs) provide a
good example; both TA1537 and TA98 give positive re-
sponses to the PAH chemical class. Many of these PAHs are
also associated with TAlOO-positive responses, but not with
TA1535-positive responses. Features such as azo, 2-hydroxy
naphthalene, secondary aryl amine(NH), and 1-alkylamino
benzene are positively associated with TA98 and TA1537. The
presence of furan, quinoline, and 2-oxynaphthalene features in
a compound are associated with TA98 mutagenesis. However,
most of these heterocycles are nonmutagenic to Salmonella,
with the exceptions of quinoline, pyrazole, and a small
number of well-known compounds containing aziridine or
benzopyrans.
There are a number of structural features that correlate
positively with all Salmonella strains, especially with TA100,
TA1537, and TA98. These include primary aryl amine, 2-
amino naphthalene, nitrosamine, quinone, p-alkyl halide, 4-
oxo benzopyran, quinoline, and 1,3-benzothiazole. Aromatic
amines and halides are worth mentioning in more detail.
Aromatic amines belong to a class that requires transformations
to metabolites (e.g., by N-hydroxylation). As shown in Table 4,
primary aromatic amines affect all strains, whereas secondary
amines are associated with only frameshift mutations. Tertiary
aromatic amines are much less likely to induce positive effects.
This observation supports the notion that more complex modes
of action are involved in the aromatic amines. In the case of alkyl
halides, both primary and secondary alkyl halides are highly
positively associated with the three strains. The formation of
small DNA adducts via carbonium ion formation may explain
this observation. Aryl halides, on the other hand, do not easily
form carbonium ions and, hence, are not found to be highly
associated with positive results in TA1535 and TA98. Aryl nitro
groups, especially 1-nitro benzenes, are much more potent than
alkyl nitro groups across all four strains, which can be explained
by Sf^Ar reaction through the resonance effects of nitro groups
of the aromatic rings.
A combined analysis of the biological activity profiles in
Figures 3 to 6 and chemical structural features in Table 4
allows us to next consider whether there are preferential
nonconcordant features across the Salmonella strains. For
example, are there structural features that are consistently
positive for TA100 and negative for TA1535? This answer
can be found by comparing the two groups of compounds
for TA100+/TA1535- vs. TA100-/TA1535-. Aryl nitro and
l-alkyl-4-amino(NH2)-benzene features belong to this group.
Features contributing preferentially to TA98 over TA1537 are
primary aromatic amines, alkenyl and aryl halides, and aryl nitro
groups. On the other hand, phenols, l-R-3-hydroxybenzenes
or l-carbonyl-2-hydroxybenzenes, are associated positively with
TA1537 mutation but negatively with TA98. The compounds
resulting in preferential mutation of TA1535 over TA100
contain alkyl oximes and, in general, have more heterocycles
such as l,3-diazine(H).
One systematic way of identifying features that are preferen-
tial in this way is to employ multivariate analysis of selected
features. In this process, instead of finding commonalities
of individual compounds within a cell in the contingency
287
Data Mining Integrated Multiple Genetic Toxicity Databases
Previous
TOC
-------
PC,
(a)
PC,
PC,
(d)
FIGURE 7 Mapping of genetic toxicity outcomes onto structural domains of the integrated database represented in the latent variable
space defined by the three most differentiating principal components. Red and green symbols denote positive and negative outcomes,
respectively. Gray points mark regions in the chemical space in which there are no toxicity data, (a): Salmonella reverse mutation, (b): In
vitro mammalian mutation, (c): In vitro chromosome aberration, (d): In vivo micronucleus.
table, selected structural features are directly correlated with the
biological profiles. Mathematically, two matrices are prepared:
ST [structure x strain calls] and SF [structure x features]. Matrix
multiplication (eq 2) yields matrix FT [features x strain calls].
Analysis of the resulting matrix of structural features against
the mutagenicity across various strains allows more in-depth
questions on the structure-toxicity relationships to be addressed.
From the Leadscope structural hierarchy, 83 chemical fea-
tures were selected (as described in the Materials and Methods
section) as descriptor variables to correlate with the four strain
outcomes. A partial but representative list of statistical results is
given in Table 4 by way of example. Figure 8 provides a multiple
scatterplot of the outcomes of the four strains based on the
83 feature observations, each point representing a feature. The
C. Yang et a/.
288
Previous
Next
-------
halide, alkyl-
halide, p-alkyl-
TA100
TA1535
Pearson cor relations
TA100-TA1535:
TA100-TA1537:
TA98 - TA1537:
TA93-TA1535:
0,93
0.73
080
0.84
060
•
V ••:
'• '• '
TA1537
v--
TA98
FIGURE 8 Scatter plots of feature z-scores computed from the
outcomes of each Salmonella strain and compared pair-wise with
the other three strains. This analysis used all data in the integrated
database.
values plotted here are the z-scores of the mutagenic outcome
between 0 and 1. The outcomes of TA100 and TA98 test strains
are 93% correlated, whereas the correlations forTA100/TA1535
and TA98/TA1537 are 73% and 84%, respectively. However, it
is important to emphasize that the purpose of this scatterplot
matrix is not to simply look for high Pearson correlation
coefficients; rather, it is to easily identify concordant and
nonconcordant features.
Between TA100 and TA1535, 4-oxo benzopyran is one of
the nonconcordant feature groups for positive TAlOO/negative
TA1535. Between TA98 and TA1537, l,3,5-triazine(H), furan,
and s-alkyl halide features are associated with negative TA1537
regardless of TA98. Features such as 1-amino (primary) naph-
thalene, 1-oxy naphthalene, and O-alkyl sulfonates are TA1537
positive regardless of the outcomes of other strains. Between
TA100 and TA98, aziridine, nitrosamine, nitroso, and epoxide
features are nonconcordant. Nonconcordant features can enrich
the structural rules that differentiate between strains. This
knowledge can also be used to establish testing strategies. For
example, if a compound contains features that the combination
of TA100 and TA98 strains can detect for mutagenic potential,
then testing the strain combination should be sufficient. If
a compound contains features such as 4-oxo benzopyran,
l,3,5-triazine(H), furan, oximes, hydrazine, or s-alkyl halide,
then it will be desirable to test TA1535/TA1537 as well.
These observations are consistent with previous findings in the
literature (Prival and Zeiger 1998).
In vitro mammalian mutagenesis
Although the mammalian mutagenesis data in the integrated
database include V79 and CHO HPRT as well as MLA assays,
due to the limited number of available studies for the former
two, only the MLA studies are discussed in detail. Sixty percent
of the 639 compounds are found to be mutagenic when
mouse lymphoma cells are used (Fig. 4). Figure 7B is a result
of projecting the MLA outcomes onto the chemical domain
depicted in Figure 2B. The positive cluster of MLA is more
densely populated in the industrial chemical space, but smaller
than the region occupied by the Salmonella-positive cluster. In
addition, the positive frequency of MLA varies depending on
the substance use type. Of the 639 compounds with MLA data,
there are 163 drugs/candidates and 61 food ingredients; the
percent positives is 34% for the drugs/candidates, 49% for
food ingredients, and 72% for industrial chemicals, resulting
in 60% total positive. In general, the positive frequency in the
integrated database increases in the order of drugs/candidates,
food ingredients, and industrial chemicals as shown in Table 6.
This observation raises an important point: when dealing with
correlations across genetic toxicity endpoints, it is important
to profile the relationships by both compound classes and
substance use types. For example, aryl alcohol groups in
drugs/candidates seem to give higher positive rates in MLA tests
than in food ingredients and industrial chemicals. Quinolines
have a stronger influence on the industrial chemicals than on
the drugs/candidates and food ingredients. Table 7 illustrates
TABLE 6 Reported genetic toxicity outcome by substance use types in the integrated database
% Positives
Substance use type
Drugs/candidates1
Food ingredients2
Industrial chemicals3
Total positive (%)
Salmonella
19.1
17.0
41.6
33.1
MLA
30.4
49.2
71.6
56.2
In Vitro
CA
38.6
29.5
50.4
44.1
In Vitro CA
(CHL)
52.2
21.6
76.4
46.1
In Vitro CA
(CHO)
46.2
40.4
47.3
46.6
In Vivo
MN
25.4
37.3
50.0
33.3
Total
882
391
1,945
3,218
1The drugs/candidates in the integrated database include drugs from the FDA CRADA CDER 2006 genetic toxicity database and drugs/candidates
from private sources. Drugs from FDA CDER do not include all historical marketed drugs information.
2The food ingredients data in the integrated database combined the FDA CRADA CFSAN 2006 database with a small amount of additional data
from private sources. An examination of the CFSAN data indicates that compounds that tested positive in one or more of the genetic toxicity assays are
disproportionately associated with chemicals used as flavors. Risk assessors account for genotoxic test results in conjunction with other information as
part of their overall chemical risk assessment of these substances.
3The industrial chemicals in the integrated database include a wide variety of substance use types (agricultural chemicals, surface active agents,
solvents, organic catalysts, etc.), which are not further differentiated in this paper. The public sources of the industrial chemicals are mostly from the
public databases, not from the regulatory agencies. Only a small number of chemicals were added from the private sources.
289
Data Mining Integrated Multiple Genetic Toxicity Databases
Previous
TOC
-------
TABLE 7 Interdependency of some features and substance use types across the genetic toxicity endpoints
Chemical features
Aromatic p-amines
Alcohol, aryl
Halides, alkyl
Ketone, alkenyl, cyc-
Imidazole
Quinoline
Substance use
type
Drug
Food
Industry
Drug
Food
Industry
Drug
Food
Industry
Drug
Food
Industry
Drug
Food
Industry
Drug
Food
Industry
Total
frequency
58
3
353
55
54
237
52
6
157
30
19
39
45
8
25
32
5
23
Salmonella
2.22
-0.64
11,14
0.99
137
1.86
-0.53
f3.11
5.98
-0.11
2.87^H
-1.46
-1.32
-1.12
2.29
-0.35
0.43
4.72
Average
MLA
1.55
none
2.74
3.47
0.80
1,34
1.56
1.45
0.79
1.22
none
-0.11
1.76
1.45
-0.19
0.44
-0.98
1-42
z-score
In vitro CA
0.67
-,,3
4.39
0.33
1,65
!-0.20
0.84
-0.47
2.07
143
100
125
1.85
2.02
1 12
1.79
None
2.64
In vivo MN
1.11
-0.77
-0.96
0.80
| 0.89
-2.53
0.64
None
-1.47
2.17
1.56
1.75
-0.77
1.85
-0.58
-0.23
-0.77
-1.00
Table cells are shaded gray if absolute z-scores > I. Cells with z-scores >2 are in bold fonts to indicate a strong association with positives.
the effects of these two factors, structural features and substance
use types, on the genetic toxicity outcomes.
When comparing both mammalian and bacterial mutations,
the MLA gives more positive findings than the Salmonella
mutagenicity assay. Endpoints for the Salmonella assay and
MLA are concordant (either both positive or both negative)
for 30% of the compounds in the database, whereas only ~5%
(32 compounds) are Salmonella positive and MLA negative.
A feature that is Salmonella positive is very likely to also be
MLA positive. There are features preferentially associated with
positive MLA results regardless of Salmonella outcomes. They
include s-alkyl halide, aryl aldehyde, acyclic alkyl hydrazine,
alkylamino benzene, thioxomethyl, 1,3-benzodioxole, and
furan. The azo group was associated with positivity in Salmonella
but not in MLA, whereas epoxides are positive in both tests.
In vitro chromosome aberration
Two study types accepted by regulatory agencies to estimate
the clastogenic potential of a compound are the in vitro
chromosome aberration and in vivo MN tests. In this integrated
database, historical results from the CHL and CHO cell lines,
as well as human blood cells, were available for comparisons.
Of the 793 compounds, 228 have CHL data and 638 have
CHO; however, only 73 compounds have been tested in both
CHL and CHO cells; hence, the comparison of the two at the
compound level is not reliable.
The in vitro chromosome aberration endpoint has many
more positive outcomes compared to the in vivo MN assay, as
illustrated in Figures 5, 6, and 7c-d. Of the 986 compounds
with reported values at this endpoint, 44% were positive.
Further breaking down to the cell lines showed that 46% of
C Yang et al.
the compounds were positive in CHL (total 228) or CHO
(total 638) cell lines. Also, many more positives are found in
the industrial chemicals group than in the drugs/candidates
group, and a much higher positive frequency was found for
industrial chemicals when tested with CHL cell lines (76%)
than with CHO (47%) or human blood cell lines. A lower
positive frequency was found for the drugs/candidates group
when human peripheral blood cell lines were used (27%). In
contrast, the positive frequency of the CHO cell lines was
consistently 40% to 50% across the substance use types. These
results are summarized in Table 7.
It is worth noting that there were no drugs/candidates from
private sources added for in vitro chromosome aberrations to
the integrated database. In general, in this database a much
higher frequency of positives are found in the in vitro chromo-
some aberration studies than in the in vivo micronucleus.
As shown in Table 4, a portion of the same features triggering
positives for Salmonella and MLA results are also associated
with positive outcomes for in vitro chromosome aberrations.
Important features include aromatic amines, alkyl halides, nitro
groups, nitroso groups, epoxides, and furans. Primary aromatic
amines and alkyl halides preferentially are associated with
positives in industrial chemicals (Table 7). Features such as
alkenyl ketones, imidazoles, purines, nucleoside bases, and
quinolines are also associated with positive results in the in
vitro chromosome aberration test.
In vivo micronucleus
Most of the in vivo MN studies in the integrated database
were performed using mouse models. For this analysis, rat
and mouse data with bone marrow and peripheral blood as
290
Previous
TOC
-------
target cells were combined under a general classification as
"rodent" to give a more statistically sound base. For the rodent
MN, a smaller number of positive studies (34%) was found
compared to the in vitro chromosome aberrations. Of the total
381 compounds, 211 were drugs/candidates and 26% of these
were positive. Food ingredients and industrial chemicals make
up 18% and 24% of the MN data, respectively; 37% of the
food ingredients and 50% of the industrial chemicals were
positive.
When correlating structural features to a biological outcome,
many of the prominent features that were associated with
other genetic toxicity endpoints, including in vitro chromo-
some aberrations, are not highly correlated with MN results.
The features associated positively with MN are nucleoside
bases, cyclic alkenyl alcohols, cyclic alkenyl ethers, alkenyl
ketones, nitrosamines, nitroso groups, 1,3-dioxanes, 2,4-dioxo
pyrimidines, and cyclic alkenyl iminomethyl groups. A set
of features that are not positively correlated with the other
Genetox endpoints, but only with both clastogenicity tests, in-
clude alkenyl ketones (Michael acceptors), etoposides, purines,
oxolanes, and alkyl chlorides. Within the integrated database,
there are several compound clusters that were found to be
positive in either the in vitro chromosome aberration or
in vivo MN assay. They show similar structural features to
etoposides, ellipticines, or flavonoids, containing one or more
combinations of anthraquinones, purines, oxolanes, imidazoles,
indoles, and benzopyran groups. These structures are well-
known topoisomerase I and II inhibitors, which may be related
to the topoisomerase-induced clastogenicity (Snyder and Gillies
2002; Lynch et al. 2003). The detailed feature analysis in Tables 5
and 7 may give insights to improve the understanding of the
clastogenicity knowledge base.
Profile across genetic toxicity endpoints
The final process of learning from these data involves
profiling the structural domains with genetic toxicity based
on the four individual study endpoints and their structural
relationships. The same process of expanding the compound-
level observations to multidimensional profiles of structural
features is adopted. Table 8 provides two-way contingency tables
across the Salmonella, MLA, in vitro chromosome aberration,
and in vivo MN assays. Concordances (overall agreement) for
these two-way comparisons of the endpoints range from 60% to
75%. However, the sensitivity (agreement on positives) of the
contingency tables ranges widely, from 36% to 83%, whereas the
specificity (agreement on negatives) is approximately 55% for
all pairs. Salmonella and MLA outcomes are 83% concordant
for positives, but only 57% concordant for negatives based
on a total of 612 compounds. Although only based on 180
compounds, findings on in vitro chromosome aberrations and
in vivo MN results are interesting. Only about 37% of the
compounds were positive in both in vitro and in vivo testing,
which is one source of the concern for the validity of using
in vitro chromosome aberration for clastogenicity potential
assessment. However, if a compound is negative in the in vitro
test, then there is a 94% chance that it is also negative in in
vivo testing. Considering that in vitro chromosome aberrations
showed only ~55% specificity (agreement on negatives) to
Salmonella and MLA assays, the high specificity to the MN
assay is worth noting.
In general, Salmonella and MLA results exhibit an increas-
ing proportion of positives from drugs/candidates, to food
ingredients, to industrial chemicals as listed in Table 6. In
contrast, the chromosome aberration test using CHL cells had
higher positive rates for drugs/candidates than food ingredients.
Industrial chemicals had the highest positive frequency in
all tests. Sixty percent of the MN results were based on
compounds in the drugs/candidates group, while only 19% of
the in vitro chromosome aberration data were from this same
group. Industrial chemicals make up ~60% of the chromosome
aberration and 24% of the MN studies. There may be two
reasons for the higher positive frequency for the in vitro
chromosome aberration results in the integrated database. One
is that the biology of the in vivo MN is more differentiating
than that of in vitro chromosome aberration. The other is due
to an artifact of the integrated database, where there were more
data from industrial chemicals for the in vitro chromosome
aberration assay and more drugs/candidates for the in vivo
MN assay. Overall, the lower frequency of positives in the
drug/candidates and food ingredients groups compared to the
industrial chemical group is a reflection of the greater safety
margins applied to compounds intended for consumption by
humans.
As explained in previous discussions under specific end-
points, Table 7 summarizes chemical features having large
variations across substance use types (drugs/candidates, food in-
gredients, industrial chemicals) and genetic toxicities. Aromatic
primary amines correlate strongly with high positive z-scores
for industrial chemicals and drugs/candidates for Salmonella,
in vitro chromosome aberration, and MLA tests, whereas the
effects on the food ingredients were somewhat negative. For in
vivo MN, drugs/candidates were still positively correlated, but
the industrial chemicals were not. For the cyclic alkenyl ketone
feature, z-scores are very similar and positive for all substance
use types for both in vitro chromosome aberrations and in vivo
TABLE 8 Contingency table for genetic toxicity outcomes in the integrated database
Salmonella +
Salmonella —
MLA + MLA-
161(110) 34(85) Salmonella +
185(236) 232(319) Salmonella -
MLA +
MLA-
In Vitro
CA +
219(146)
191 (264)
156(124)
34 (66)
In Vitro
CA-
98(171)
381 (342)
92(124)
97 (75)
Salmonella +
Salmonella —
MLA +
MLA-
In Vitro CA +
In Vivo
MN +
18(10)
32 (40)
156(124)
34 (66)
30(16)
In Vivo
MN-
31 (39)
171 (48)
92(124)
97 (75)
50 (64)
Values in parentheses are the expected compound counts based on the null hypothesis.
291
Data Mining Integrated Multiple Genetic Toxicity Databases
Previous
TOC
-------
nitre, heteraamino benzene
Salmonella
MLA
Pearson correlations:
Salmonella - MLA: 0.72
Salmonella - MCA: 0.68
MLA - ivtCA: 0.67
MLA - MN: 0.42
MCA - MN: 0.33
ivtCA
ketone, alkenyl
ketone. cyclic altenyl
MN
FIGURE 9 Scatter plots of feature z-scores computed from the
outcomes of each genetic toxicity test and compared pair-wise
with the other three tests. Test abbreviations: Salmonella (all
four strains), MLA (mouse lymphoma mutation), ivtCA (in vitro
chromosome aberration), MN (in vivo micronucleus).
MN. On the other hand, the same feature shows very different
patterns for other endpoints (e.g., Salmonella mutagenicity).
Such observations explain why it is difficult to draw meaningful
conclusions based solely on compound classes and a single
endpoint. A better approach is to describe compounds at the
feature level for diverse substance use types profiled across
multiple endpoints.
For building structural trends, the four study types were
compared using the multivariate analysis method described
in the Materials and Methods section. A total of 203 features
were extracted from the Leadscope chemical hierarchy using the
systematic feature selection method. Table 4 lists examples of
these features and Figure 9 displays the pair-wise correlations
between the genetic toxicity assays for the selected features
as observations (points on the plots). As expected from
the contingency table analysis, Salmonella reverse mutation
correlates quite well with in vitro chromosome aberration as well
as with MLA. As mentioned before, the in vitro chromosome
aberration tests were correlated more closely with Salmonella
and MLA mutations than with clastogenicity of the in vivo MN
studies.
The correlating features for each pair are identified in Figure 9
and Table 4. The well-known genotoxic compound classes such
as nitroso/nitrosamines, nitro groups, phosphorus-containing
groups, and quinones are associated positively with all four
study types in genetic toxicity endpoints. Aldehydes are highly
associated only with the MLA. Aromatic amines significantly
affect all of the in vitro genetic toxicity tests, but do not
have a large impact on the in vivo MN test. The azo group
is positively associated with Salmonella mutagenicity, but not
with other endpoints. PAHs with the 1,2,3,4-fused benzene
feature show high positive correlation with mutagenicity but
not with clastogenicity endpoints. Alkyl halides and epoxides
have positive effects in all of the genetic toxicity tests except in
the MN test. Nucleoside bases do not impact Salmonella, but
C Yang et al.
influence other genetic toxicity endpoints. On average, phenols
and thioxomethyl groups contribute positively only to MLA.
Compounds with imidazole groups and quinoline features are
preferentially contributing to positive in vitro chromosome
aberrations.
Validation of Structural Rules
The ultimate goal of data mining is to transform all of
the observations discussed above into knowledge from which
structural feature rules can then be built. The contingency
table and multivariate analyses provide systematic methods to
explore the data. The objective of this paper is to describe the
analysis process; the next step-building structural alerts from
the extracted knowledge-will be demonstrated in a later paper.
However, the structural rules available in the current literature
can be applied to the integrated database to evaluate their
validity. As an example, Table 9 shows how Ashby-Tennant
alerts (Ashby 1985) are correlated with the data in the integrated
database across the four endpoints using the z-score statistic
as a quantitative metric. From Table 9, roughly two-thirds of
the rules are appropriate to Salmonella mutagenicity. Within
the integrated database, these published rules (Kazius et al.
2005) show correct classification rates of 60% to 70% for
Salmonella mutagenicity in vitro chromosome aberration and
MLA assays, but only 39% for in vivo MN results. Most of
the published rules were not statistically significant based on
this integrated database. Again, this is not surprising since
60% of the compounds with MN results in this database are
from the drugs/candidates group, a group whose chemical
space the Ashby-Tennant rules or other published rules may
not fully represent. In this study, due to the addition of a
private collection with more MN data, new features such as
cyclic alkenyl ketones, cyclic alkenyl ether, and nucleoside
bases are found. The cyclic alkenyl ketone ether group was
preferential to in vivo MN, whereas alkenyl ketone (general
Michael acceptor) and nucleoside base features represent both
in vitro chromosome aberration and in vivo MN outcomes.
Not many compounds in the drugs/candidates group in the
integrated database contained Michael acceptor-related features.
These features are not yet considered rules since the statistical
base of the data in the integrated database was not large enough
and the chemical structures were too diverse. It is important
to point out again that a substructural fragment by itself is not
necessarily sufficient to define an alert, but that combinations
of such features in a substance are necessary to explain the
significance and mode of action of each.
Weight-of-Evidence Approach
by Profiling
Genetic toxicity studies are often viewed as a battery of
surrogate screening tests for carcinogenicity. In some cases,
single-endpoint approaches have been shown to work fairly
well; for example, Salmonella mutagenesis is known to be a good
predictor of genotoxic carcinogenicity (Benigni et al. 2000).
However, many previous publications have convincingly argued
that using individual genetic toxicity endpoints to estimate
carcinogenicity will generally not give a complete understanding
and will often give unsatisfactory predictions; these researchers
conclude that what is needed are methods of using information
292
Previous
TOC
-------
TABLE 9 Some structural rules across the genetic toxicity endpoints
z-sco re
Chemical features
Total frequency Salmonella
MLA
In vitro CA
In vivo MN
Ashby-Tennant alerts
N-chloramine
N-methylol
alkyl aldehyde
Alkyl hydrazine (R2NNH2)
Alkyl hydrazine (R2NNR2)
Alkyl hydrazine (RNHNH2)
Alkyl hydrazine (RNHNHR)
Alkyl phosphonate
Aromatic N-oxide
Aromatic amine (NH2)
Aromatic amine, N-hydroxy
Aromatic amine, dialkyl
Aromatic azo
Aromatic nitro
Aryl methyl halide
Aziridine
Epoxide
Monohaloalkyene
Nitrogen mustard
IMitrosamine, dialkyl
Primary alkyl halide
Propriolactone
Urethane derivatives
Features for clastogenicity
from the integrated
database
Cyclic alkenyl ketone ether
Alkenyl ketone
Nucleoside bases
930
5
3
24
2
1
4
4
8
5
414
2
76
79
199
8
6
59
88
8
24
88
4
51
30
34
28
0.80
-0.60
-1,0
0.50
0,10
0.83
2.7
2.4
2.9
2.8
I
across multiple endpoints to give more reliable predictions.
Models based on a WOE approach provide exactly this type of
advantage (Matthews et al. 2006b). WOE methods are diverse,
varying from heuristic rules to Bayesian optimization. The
multivariate analysis is one way to quantify WOE. In this paper,
we are taking a step toward multiple domain correlations using
structural features and the endpoint outcomes represented by
the features.
The notion that MLA and in vitro chromosome aberration
experiments give too many false positives originates from the
correlations of these endpoints individually to carcinogenicity
(Matthews et al. 2006a). From the perspective of feature profiling
based on the battery of genetic toxicity tests, it would be
better to view MLA and chromosome aberration outcomes
not as individual predictors, but rather as components of a
multivariate approach in which the aggregate profile over all
endpoints provides a biological fingerprint for any compound
of interest. As long as the assays are biologically meaningful
and are not used in isolation to correlate a particular genetic
toxicity endpoint to carcinogenicity, the high percentage of
negatives in the Salmonella data tests and high percentage of
positives in the MLA data tests are in fact giving insights about
the biological activities of the compound classes and chemical
structural features. If a certain combination of positive features
shows up more frequently in the industrial chemicals, then it is
reasonable to expect to find more positives when the compound
makeup of the data source is mostly industrial chemicals. It may
be possible to use all of the in vitro genetic toxicity endpoints
as assays (in a preliminary screen) and to include the aggregate
profile of these results to correlate to in vivo toxicity endpoints
or to carcinogenicity. The multivariate method described in
this paper can be applied to assess the WOE quantitatively. The
selected features for these multivariate correlations in fact are
candidates for the predictors in a QSAR model.
Lastly, one of the most important potential applications of
this profiling method is to guide development of intelligent
testing strategies. These methods can help researchers decide
what chemicals should be tested and what assays should be
employed in order to obtain reliable toxicity data as efficiently
as possible and to avoid unnecessary animal testing. Data
mining to establish profiles of chemicals both in chemical and
biological domains will assist such decisions. For example, if
293
Data Mining Integrated Multiple Genetic Toxicity Databases
Previous
TOC
-------
a drug/candidate contains a primary aromatic amine and if
this feature is a major reason for concern, then a positive
observation in Salmonella mutagenicity may signal a concern
for other genetic toxicity endpoints because positive z-scores
are observed for all tests for this particular feature-substance
type combination (Table 7) and also because of the relatively
strong correlation of Salmonella results with the other three tests
(Table 8 and Figure 9), and testing of in vivo MN may be
considered. A negative observation in the Salmonella assay for
this group indicates that both chromosome aberration and MN
testing may not be necessary because of the strong correlation
of Salmonella mutagenicity and in vitro chromosome aberration
for this primary aromatic amine group (Table 7) and because of
the observation that in vivo MN is most likely negative when in
vitro chromosome aberration is negative (Table 8). On the other
hand, if an industrial chemical contains a primary aromatic
amine or alkyl halide group, the chemical is likely to be positive
in three of the in vitro tests if these features are responsible for
the observed toxicity. Industrial chemicals with these features
alone do not usually trigger the in vivo MN outcome, as shown
in Table 7.
Although genetic toxicity tests, in general, are usually per-
formed simultaneously as a full battery for the sake of efficiency,
the principle of intelligent testing strategies demonstrated here
can be applied to other types of biological safety tests. Overall,
intelligent testing strategies can be developed by understanding
the interaction between structural features and the molecular
makeup defined in the substance use types and their effects
across the various genetic toxicity and clinical endpoints. Thus,
this paper presents data mining and profiling methods for a
WOE approach to assess toxicity and for guiding the general
development of intelligent testing strategies.
ACKNOWLEDGMENTS
The authors dedicate this paper to the memory of Dr. Gary
Hollingshaus at DuPont whose vision of applying predictive
data mining within the industrial workflow instigated the
formation of the Leadscope In Silico Toxicology (LIST) focus
group. The idea of ToxML and construction of aggregating
databases was borne from many discussions with Drs. Gary
Hollingshaus, Philip Lee, and Dan Kleierin Haskell Laboratory.
The authors also thank Leadscope focus group members
Mitchell Cheeseman, Yan Gu, Dale Johnson, Julie Mayer,
Donna Morrall, Richard Mueller (deceased), Chad Nelson,
Grace Patlewicz, Gregory Pearl, Rene Sotomayor, Michelle
Twaroski, Anita White, and Alan Wilson. Funding for the
ToxML database and LIST focus group was provided in
part by the U.S. NIST Advanced Technology Program (70
NANB4H3003).
Disclaimer. This paper does not reflect the policies of
the U.S. FDA CDER, U.S. FDA CFSAN, or U.S. EPA. The
content of this paper is only a reflection of scientific research
conducted in one of many collaborations with the agencies.
Proprietary structures, toxicity data, and other information from
private sources were not disclosed outside of Leadscope Inc.
The information presented in this paper is derived solely from
analysis of structural feature statistics. The status of the testing of
these chemicals during lead discovery and product development
has never been disclosed.
REFERENCES
Ashby, J. 1985. Fundamental structural alerts to potential carcinogenicity
or non-carcinogenicity. Environ. Mutagen. 7:919-921.
Ashby, J., and Tennant, R. W. 1991. Definitive relationships among
chemical structure, carcinogenicity and mutagenicity for 301
chemicals tested by the U.S. NTP. Mutat. Res. 257:229-306.
Benigni, R. 2004. Computational prediction of drug toxicity: The case
of mutagenicity and carcinogenicity. Drug Discov. Today Techno!.
1(4):457-463.
Benigni, R., and Giuliani, A. 1994. Quantitative modeling and biology:
the multivariate approach. Am. J. Physiol. 266:R1697-R1704.
Benigni, R., Giuliani, A., Franke, R., and Gruska, A. 2000. Quantitative
structure-activity relationships of mutagenic and carcinogenic aro-
matic amines. Chem. Revs. 100:3697-3714.
Benigni, R., Netzeva, E., Bossa, C, Franke, R., Helma, C, Hulzebos, E.,
Merchant, C., Richard, A. M., Woo, Y. I, and Yang, C. 2007. The
expanding role of predictive toxicology: an update on the (Q)SAR
models for mutagens and carcinogens. J. Environ. Sd. Health C.
Environ. Carcinog. Ecotoxicol. Rev. 25:1-43.
Chemical Carcinogenesis Research Information System. 2007.
Available on the Internet at http://toxnet.nlm.nih.gov/cgi-
bin/sis/htmlgen?CCRIS. Accessed November 28, 2007.
Cimino, M. C. 2006. Comparative overview of current international
strategies and guidelines for genetic toxicology testing for regu-
latory purposes. Environ. Molec. Mutagen. 47(5):362-390.
Contrera, J. F, Matthews, E. J., Kruhlak, N. L, and Benz, R. D. 2005.
In silico screening of chemicals for bacterial mutagenicity using
electrotopological E-state indices and MDL QSAR software. Regul.
Toxicol. Pharmacol. 43:313-323.
Crebelli, R., Carere, A., Leopardi, P., Conti, L., Fassio, F., Raiteri, F., Barone,
D., Ciliutti, P., Cinelli, S., and Vericat, J. A. 1999. Evaluation of 10
aliphatic halogenated hydrocarbons in the mouse bone marrow
micronucleus test. Mutagenesis 14(2):207-21 5.
Darroch, J., and Mosimann, J. E. 1985. Canonical and principal compo-
nents of shape. Biometrika 72:241-252.
Environmental Protection Agency National Centerfor Computational Tox-
icology. 2007. http://www.epa.gov/comptox/index.html. Accessed
November 2, 2007.
Food and Drug Administration CFSAN EAFUS. 2007. U.S. FDA Center
for Food Safety and Applied Nutrition, Everything Added to Food
in the United States, http://www.cfsan.fda.gov/~dms/eafus.html.
Accessed November 28, 2007.
Jacobson-Kram, D.,and Contrera, J. F. 2007. Genetic toxicity assessment:
employing the best science for human safety evaluation part I: early
screening for potential human mutagens. Toxicol. Sci. 96(1):16-20.
Jolliffe, I. T. 2002. Principal Component Analysis, 2nd ed. Springer Series
in Statistics, Springer-Verlag, New York.
Kazius, J., McGuire, R., and Bursi, R. 2005. Derivation and validation of
toxicophores for mutagenicity prediction. J. Med. Chem. 48:312-
320.
Kirkland, D. J., Aardema, M., Banduhn, N., Carmichael, P., Fautz, R.,
Meunier, J. R., and Pfuhler, S. 2007. In vitro approaches to develop
weight of evidence (WoE) and mode of action (MoA) discussions
with positive in vitro genotoxicity results. Mutagenesis 22(3): 161-
175.
Kirkland, D., Aardema, M., Henderson, L, and Miiller, L. 2005. Evaluation
of the ability of a battery of three in vitro genotoxicity tests to
discriminate rodent carcinogens and non-carcinogens. Mutat. Res.
584(1-2): 1-2 56.
Lynch, A., Harvey, J., Aylott, M., Nicholas, E., Burman, M., Siddiqui, A.,
Walker, S., and Rees, R. 2003. Investigations into the concept of
inhibitor-induced clastogenidty. Mutagenesis 18(4):345-353.
Matthews, E. J., Kruhlak, N. L., Cimino, M. C., Benz, R. D., and Contrera,
J. F. 2006a. An analysis of genetic toxicity, reproductive and
developmental toxicity, and carcinogenicity data: I. Identification of
carcinogens using surrogate endpoints. Regul. Toxicol. Pharmacol.
44:83-96.
C. Yang et a/.
294
Previous
TOC
-------
Matthews, E. J., Kruhlak, N. L, Cimino, M. C., Benz, R. D., and Contrera,
J. F. 2006b. An analysis of genetic toxicity, reproductive and
developmental toxicity, and carcinogenicity data: II. Identification
of genotoxicants, reprotoxicants, and carcinogens using in silico
methods. Regul. Toxicol. Pharmacol. 44:97-110.
McCann, J., Spingarn, N. E., Kobori, J., and Ames, B. N. 1975. Detection
of carcinogens as mutagens: bacterial tester strains with R factor
plasmids. Proc. Nat. Acad. Sd. USA 72(3):979-983.
Moore, M. M., Honma, M., Clements, J., Bolcsfoldi, G., Burlinson, B.,
Cifone, M., Clarke, J., Delongchamp, R., Durward, R., Fellows, M.,
Gollapudi, B., Hou, S., Jenkinson, P., Lloyd, M., Majeska, J., Myhr,
B., O'Donovan, M., Omori, T., Riach, C., San, R., Stankowski, L.
F. Jr., Thakur, A., Van Goethem, F., Wakuri, S., and Yoshimura, I.
2006. Mouse lymphoma thymidine kinase gene mutation assay:
follow-up meeting of the International Workshop on Genotoxicity
Tests-Aberdeen, Scotland, 2003-assay acceptance criteria, positive
controls, and data evaluation. Environ. Mol. Mutagen. 47:1-5.
Mortelmans, K., and Zeiger, E. 2000. The Ames Salmonella microsome
mutagenicity assay. Mutat. Res. 455:29-60.
National Toxicology Program (NTP) On-line Database. 2007. National
Institute of Health & Environmental Sciences NTP On-line Database.
http://ntp-apps.niehs.nih.gov/ntp_tox/index.cfm. Accessed Novem-
ber 2, 2007.
National Toxicology Program NICEATM. 2007. The NTP Interagency
Center for the Evaluation of Alternative Toxicological Methods
(NICEATM); Interagency Coordinating Committee on the Val-
idation of Alternative Methods (ICCVAM). http://iccvam.niehs.
nih.gov/methods/methods.htm. Accessed November 2, 2007.
O'Brien, J., Renwick, A. G., Constable, A., Dybing, E., Muller, D. J. G.,
Schlatter, J., Slob, W., Tueting, W., van Benthem, J., Williams, G.
M., and Wolfreys, A. 2006. Approaches to the risk assessment of
genotoxic carcinogens in food: A critical appraisal. Food Chem.
Toxicol. 44(10):!613-1635.
Prival, M. J., and Zeiger, E. 1998. Chemicals mutagenic in Salmonella
typhimurium strain TA1535 but not in TA100. Mutat. Res. Genetic
Toxicol. Environ. Mutagen. 412(3):251-260.
Richard, A. M. 2006. Future of toxicology—predictive toxicology: an ex-
panded view of chemical toxicity. Chem. Res. Toxicol. 19(10):1257-
1262.
Richard, A., Yang, C., and Judson, R. 2007. Toxicity data informatics:
supporting a new paradigm for toxicity prediction. Toxicol. Mech.
A/7ef/i.Present volume
Roberts, G., Myatt, G. J., Johnson, W. P., Cross, K. P., and
Blower, P. E. 2000. LeadScope: software for exploring large
sets of screening data. J. Chem. Inf. Comput. Sci. 40:1302-
1314.
Snyder, R. D. Gillies, P. J. 2002. Evaluation of the clastogenic,
DNA intercalative, and topoisomerase ll-interactive properties of
bioflavonoids in Chinese hamster V79 cells. Environ. Molec.
Mutagen. 40:266-276
Tokyo-Eiken. 2007. Tokyo Metropolitan Institute of Public
Health, Mutagenicity of food additives. http://www.tokyo-
eiken.go.jp/henigen/index.htm. Accessed November 2, 2007.
Tweats, D. J., Scott, A. D., Westmoreland, C., and Carmichael,
P. 2007. Determination of genetic toxicity and potential car-
cinogenicity in vitro—challenges post the Seventh Amendment
to the European Cosmetics Directive. Mutagenesis 22(1 ):5-
13.
Vogel, E. W., Barbin, A., Nivard, A. J. M., Stack, H. F, Waters, M. D.,
Paul, H. M., and Lohman, P. H. M. 1998. Heritable and cancer
risks of exposures to anticancer drugs: inter-species comparisons
of covalent deoxyribonucleic acid-binding agents. Mutat. Res.
400:509-540.
Yang, C., Benz, D. R., and Cheeseman, M. A. 2006. Landscape of current
toxicity databases and database standards. Curr. Opin. Drug Discov.
Develop. 9(1): 124-133.
295
Data Mining Integrated Multiple Genetic Toxicity Databases
Previous
TOC
-------
Available online atvvww.sciencedirect.com
ELSEVIER Toxicology and Applied Pharmacology 277 (2008) 163-178
Toxicology
ScienceDirect ^ Applied
Pharmacology
www.elsevier.com/locate/ytaap
Contemporary Issues in Toxicology
Understanding mechanisms of toxicity: Insights from
drug discovery research
Keith A. Houck *, Robert J. Kavlock
National Center for Computational Toxicology, Office Research and Development, United Stated Environmental Protection Agency,
Research Triangle Park, NC 27711, USA
Received 4 April 2007; revised 28 September 2007; accepted 11 October 2007
Available online 4 November 2007
Abstract
Toxicology continues to rely heavily on use of animal testing for prediction of potential for toxicity in humans. Where mechanisms of toxicity have
been elucidated, for example endocrine disruption by xenoestrogens binding to the estrogen receptor, in vitro assays have been developed as surrogate
assays for toxicity prediction. This mechanistic information can be combined with other data such as exposure levels to inform a risk assessment for the
chemical. However, there remains a paucity of such mechanistic assays due at least in part to lack of methods to determine specific mechanisms of toxicity
for many toxicants. A means to address this deficiency lies in utilization of a vast repertoire of tools developed by the drug discovery industry for
interrogating the bioactivity of chemicals. This review describes the application of high-throughput screening assays as experimental tools for profiling
chemicals for potential for toxicity and understanding underlying mechanisms. The accessibility of broad panels of assays covering an array of protein
families permits evaluation of chemicals for their ability to directly modulate many potential targets of toxicity. In addition, advances in cell-based
screening have yielded tools capable of reporting the effects of chemicals on numerous critical cell signaling pathways and cell health parameters. Novel,
more complex cellular systems are being used to model mammalian tissues and the consequences of compound treatment. Finally, high-throughput
technology is being applied to model organism screens to understand mechanisms of toxicity. However, a number of formidable challenges to these
methods remain to be overcome before they are widely applicable. Integration of successful approaches will contribute towards building a systems
approach to toxicology that will provide mechanistic understanding of the effects of chemicals on biological systems and aid in rationale risk assessments.
Published by Elsevier Inc.
Keywords: High-throughput screening; High-content screening; In vitro toxicology; Mechanism of toxicity; Systems toxicology; Predictive toxicology
Contents
Introduction 164
High-throughput screening 164
HTS for modulators of drug-metabolizing enzymes 165
High-throughput genotoxicity assays 166
Ion channel targets for toxicity screening 167
Receptor targets for toxicity screening 167
Broad pharmacological profiling 167
Complex cellular toxicity assays 168
Model organism toxicity assays 170
Data analysis 171
Experimental considerations 172
Challenges of HTS application to toxicology 175
* Corresponding author. NCCT/ORD (D343-03), US EPA, 109 TW Alexander Dr, Research Triangle Park, NC 27711, USA. Fax: +1 919 685 3371.
E-mail address: houck.keith@epa.gov (K.A. Houck).
0041-008X/$ - see front matter. Published by Elsevier Inc.
doi:10.1016/j.taap.2007.10.022
Previous
-------
164
K.A. Houck, R.J. Kavlock / Toxicology and Applied Pharmacology 277 (2008) 163-178
Conflict of interest statement 176
References 176
Introduction
Toxicology has traditionally focused on the effects of
exogenous chemicals on living organisms through intensive
studies done one chemical at a time. Such approaches have
served to illuminate the modes of action of many classes of
chemicals and provided detailed mechanistic understanding of
the molecular targets of toxicity for some. However the costs of
this approach have been high and mechanistic understanding
remains limited. Toxicology studies rely heavily on use of
vertebrate animals, an expensive undertaking in both time and
money with debatable predictive power for human safety. For
example, carcinogenicity studies are conducted using a 40-year-
old model requiring 400 or more study animals and two years of
exposure at a cost of millions of dollars (Bucher, 2002). Despite
the investment of resources, debate continues on the utility of
these data in predicting carcinogenicity potential in humans
(Ennever and Lave, 2003). The problem lies chiefly with an
inability to discern mechanisms of toxicity for the majority of
toxicants tested using the "black box" whole animal assays;
hence, cross-species extrapolation and low-dose, real-life expo-
sure effects become very difficult to appropriately assess. With
increasing public concern over the minimal toxicity information
available for thousands of large volume-production chemicals
produced and used in commerce, the inadequacy of existing
methods presents a sizeable quandary (De Rosa et al., 2003;
National Research Council, 2006). REACH legislation in the
European Union covers approximately 30,000 chemicals and
would require millions of animals and billions of Euros to
conduct safety assessment on all of these using traditional
methods (Van der Jagt et al., 2004). Such an enormously
expensive undertaking would likely provide useful data in
defining the toxicity (or lack thereof) for many chemicals;
however, without mechanistic understanding, debate over risk of
a subset of these chemicals would most assuredly ensue. What is
needed are cost-effective screening assays that would not only
identify chemicals of safety concern, but provide quantitative
and mechanistic information to inform rationale risk assessment.
In the pharmaceutical and biotechnology industries over the
past 15 years, enormous resources have been invested in devel-
oping efficient means to screen compounds against large numbers
of potential therapeutic targets. The dramatic advances made in
high-throughput screening (HTS) technologies now permit ready
profiling of biological activity of large chemical libraries using
multiwell plates and automated liquid handling equipment.
Although developed to support drug discovery, toxicologists
have begun applying these batch-testing methodologies to large
numbers of chemicals using in vitro bioassays and model organ-
ism screens to characterize potential for toxicity and understand
mechanisms of action. This review will look at applications of
HTS techniques to toxicology and their potential impact on
shifting testing paradigms for evaluating chemicals for risk.
High-throughput screening
HTS techniques are used primarily in the pharmaceutical
industry in support of lead generation projects whose goal is to
efficiently sort through enormous numbers of compounds for
leads, the starting chemical structures for the drug development
process. Large libraries of organic small molecules or natural
products are batch-tested against biological targets in industry-
standard 96- and 384-well plates, occasionally using even
higher density 1536- or 3456-well plates. Relying heavily on
automation and robotics, throughput ranges from thousands to
1,000,000 samples tested per day depending on the specific
assay format (Table 1). Rate-limiting steps often lie in the type
of signal detection required for the assay; simple fluorescence
signals on the high end of the throughput range and cellular
imaging assays at the lower end. Costs range from well under
$ 1 per well for reagents and consumables to $ 10-75 per well for
work performed at contract research laboratories.
Although the numbers of chemicals requiring toxicity
screening is not in the range of the numbers of chemicals
typically screened for lead generation for drug discovery, the
opportunity to broadly profile compounds for toxicity with a
variety of assays efficiently and at low cost makes HTS an
attractive approach. This strategy, i.e. few chemicals tested
against a large number of assays, is the converse of the drug
discovery paradigm where many compounds are tested against
one biological target, but makes use of the same efficiency
infrastructure used for HTS assays. Initial work on adopting
HTS techniques for toxicity testing occurred in the pharmaceu-
tical industry and centered largely on the cytochrome P450
monooxygenase (CYPs) drug metabolizing enzymes since
interference with these enzymes was a commonly encountered
problem during development of new drug candidates (Crespi
and Stresser, 2000). This area continues to receive much focus
as new technologies are brought to bear on inadequacies of
existing approaches. Beyond drug metabolizing enzymes, in
vitro assays for specific targets associated with toxicity, e.g. ion
channel assays, have been developed and are now routinely run.
In addition, pharmacology profiling panels targeting represen-
tative members of many protein families are used to understand
Table 1
Definitions of screening modes
Screening mode Abbreviation Samples tested Example
per day
Low-throughput
Medium-
throughput
High-
throughput
Ultra-high-
throughput
LTS
MTS
HTS
uHTS
1-500
500-10,000
10,000-100,000
> 100,000
Animal models
Fluorescent cellular
microscopic imaging assay
Fluorescent enzymatic
inhibition assay
Beta-lactamase cell
reporter assay
Previous
TOC
-------
K.A. Houck, R.J. Kavlock / Toxicology and Applied Pharmacology 277 (2008) 163-178
165
unexpected "off-target" activities of compounds in develop-
ment. Use of in vitro cellular assays for toxicity testing has been
rapidly growing using a number of different methods including
automated, fluorescence imaging technology to profile cell
health parameters; genotoxicity screening in mammalian cells;
and cell systems biology approaches. Finally, adaptation of
several model organism assays to HTS-compatible formats has
also begun to impact toxicity testing paradigms. The application
of these approaches to toxicity prediction and mechanistic
understanding will be discussed in this review.
HTS for modulators of drug-metabolizing enzymes
CYPs are the most important class of enzymes metabolizing
exogenous chemicals as well as many endogenous substrates.
Metabolism of exogenous chemicals or interference with the
activities of these enzymes can result in toxicities manifested in
a number of ways. First, the chemical itself can be directly acted
on by the enzymes, primarily in the liver and intestines,
resulting in production of a toxic metabolite or, conversely, the
detoxification of a toxic chemical (Shimada, 2006). For
Pharmaceuticals, this is of obvious importance in determining
whether the active agent survives first-pass metabolism in the
liver to reach its target tissue in sufficient dose. For environ-
mental chemicals, understanding whether CYP metabolism
activates chemicals to toxic metabolites that would be missed in
in vitro systems lacking CYP activity or, equally important,
whether CYP metabolism results in efficient inactivation of
toxic chemicals are critical requirements needed for rationale
risk assessment.
Secondly, inhibition or induction of CYPs by drugs or
xenobiotics can disrupt the normal kinetics of other drugs and
chemicals resulting in increased or decreased levels and
potential toxicities. Such interactions, called drug-drug inter-
actions in the pharmaceutical field, have many clinically
important examples. In one common case, exogenous ligands
activate nuclear receptors, in particular pregnane X receptor
(PXR), leading to induction of mRNA for a variety of CYPs,
subsequent increased metabolic activity of the expressed
enzymes and, hence, reduced bioavailability of therapeutic
drugs such as oral contraceptives (Handschin and Meyer, 2003).
Such activities can also seriously effect the clinical development
of novel therapeutic drugs by confounding efficacy trials. As a
result, the pharmaceutical industry has developed HTS
approaches to permit efficient screening of chemicals for their
activities in modulating the drug metabolizing enzymes. These
assays, described below, allow the quantitative measurement of
effect of the chemicals directly on their molecular targets, e.g.
specific CYPs or nuclear receptors. This can then be used in
conjunction with human exposure predictions or measurements,
to estimate the potential for these interactions to be of a sig-
nificant toxicity concern.
HTS methods to determine the effects of CYP enzymatic
activity on exogenous chemicals has been approached in a
number of ways. Standard, low-throughput, methods allow
exposure of chemicals to a source of CYP activity in 96-well
plates followed by extraction of the chemical and determination
of amount of material remaining using liquid chromatography/
mass spectrometric methods (LC/MS/MS) (Lau et al., 2002;
Shibata et al., 2002). Since the vast majority of cell lines have
insignificant levels of most of the CYP enzymes, sources of
metabolizing capacity are recombinant CYP enzymes, hepato-
cyte subcellular fractions such as S9 or microsomes, or primary
hepatocytes themselves, freshly isolated or cyropreserved. Use
of hepatocytes also provides the inclusion of Phase II conju-
gation systems to more closely represent the in vivo environ-
ment. These "loss-of-parent" assays are useful in understanding
pharmacokinetic parameters in the drug development pro-
cess but are of less utility in toxicology. Reasons for this are
predominately the lack of a priori knowledge of what the
metabolite(s) will be and whether they are inherently more or
less toxic than the parent. Determination of these effects
generally has been beyond the screening level of information
provided by such assays. Interest in coupling such biotransfor-
mation systems to toxicity testing has led to several systems of
potential value. In one approach, cytotoxicity in cell lines stably
expressing individual, recombinant CYP enzymes can be com-
pared to the parental cell line lacking the CYP (Yoshitomi et al.,
2001; Bernauer et al., 2003). While such approaches can be
high-throughput and useful for mechanistic understanding
for specific chemicals, concern over lack of a complete spec-
trum of metabolic responses, including enzyme induction, has
prevented the widespread adoption of these lines for high-
throughput toxicity screening.
Alternatively, co-culture systems consisting of metabolically
competent cells, i.e. primary hepatocytes, coupled with one or
more additional target cell types provided a means to incorporate
biotransformation into toxicity screening. Although notoriously
variable from batch-to-batch, improvements in methods for use
of cryopreserved human hepatocytes have demonstrated greater
consistency and preservation of metabolic function including
enzyme induction (Roymans et al., 2005; Kafert-Kasting et al.,
2006). Innovative platforms to carry out toxicity screening with
primary hepatocytes have been described although these new
microfluidic-based technologies have yet to be widely accepted
(Viravaidya et al., 2004; Lee et al., 2006a). In another variation
on this theme, the MetaChip system couples CYP catalysis of
chemicals using immobilized CYPs with cytotoxicity endpoints
determined by cell viability staining (Lee et al., 2005). Finally,
the IdMOC system permits co-culture of five cell types along
with primary hepatocytes in an interconnected system allowing
exposure of metabolites of test compound to a variety of cells
representing different tissues (Li et al., 2004). Although these
methods have yet to make a significant impact in toxicology,
they represent progress in dealing with a critical problem
otherwise requiring whole animal testing. These or similar
methods will likely receive growing attention in the future.
HTS methods to assess the potential for compounds to elicit
pharmacokinetic drug interactions via inhibition of cytochrome
CYP activities are well established in the drug development
industry. Most of the major species of CYP enzymes are readily
available from commercial sources as recombinant proteins
produced in Escherichia coli or baculovirus-insect cell expres-
sion systems. Semi-automated assays validated under Good
Previous
TOC
-------
166
K.A. Houck, R.J. Kavlock / Toxicology and Applied Pharmacology 277 (2008) 163-178
Laboratory Practices (GLP) standards that measure inhibition of
specific substrate conversion using LC/MS/MS for detection
have been described (Walsky and Obach, 2004). Assays with
much higher throughput also exist and are reliant on quenched
fluorescent substrates that fluoresce following CYP-mediated
metabolism (Crespi et al., 1997). Because this is a homogenous
assay with a fluorescent readout, the assay can be readily
miniaturized using very low reaction volumes suitable for
screening in 1536-well plates (Trubetskoy et al., 2005). Phase II
conjugative enzymes have been less well developed for HTS
approaches. Methods possibly useful in HTS format for UDP-
glycosyl transferase enzymes and phenol sulfotransferase
(SULT1A1) have been published but do not appear to be in
routine use (Trubetskoy and Shaw, 1999; Frame et al., 2000).
Induction of drug metabolizing enzyme activity, another
potential mode of drug interaction toxicity, has traditionally
been determined through measurement of enzymatic activity in
primary hepatocyte cultures following drug exposure, a rela-
tively low-throughput method (Luo et al., 2004; Lin, 2006).
CYP3A4 monooxygenase activity has been the focus of much
attention for drug interactions as it is responsible for metabolism
of more than 50% of pharmaceuticals (Guengerich, 1999).
Recent work that identified the nuclear receptor PXR as a key
regulator of CYP3A monooxygenase expression provided the
insight to develop HTS assays for inducers of this important
drug interaction pathway (Lehmann et al., 1998; Moore and
Kliewer, 2000). HTS luciferase reporter gene assays, typically
performed in stably or transiently transfected liver cell lines
such as 4HIIE or HuH7, are now routinely run to identify
activators of PXR. Strong correlations between inducers of the
PXR reporter gene activity and ability to induce CYP3A
activity have been demonstrated. Lower throughput enzymatic
activity assays in hepatocytes are reserved for confirmation
assays (Luo et al., 2002; Cui et al., 2005). Reporter gene assays
compatible with HTS for other nuclear receptors controlling
drug metabolism enzyme induction; e.g. constitutive androstane
receptor, liver X receptor, vitamin D receptor, and farnesoid X
receptor; have also been described (Faucette et al., 2006; Houck
et al., 2004; Bettoun et al., 2003; Lee et al., 2006b). The more
distantly related aryl hydrocarbon receptor (AhR), activated by
ligands including dioxin, regulates expression of the drug-
metabolizing enzymes of the CYP1A family. Reporter gene
assays have been established in liver cell lines, e.g. HepG2, and
can be used in HTS format (Yueh et al., 2005). Good correlation
between reporter gene activity and induction of mRNA for
CYP1A1 as well as additional genes regulated by AhR was
shown.
High-throughput genotoxicity assays
A battery of in vitro assays including the Ames bacterial
reverse mutation test, the mouse lymphoma tk gene mutation
assay, and the micronucleus clastogenicity assay is traditionally
used to measure the potential for genotoxic activity of chemicals.
These are not high-throughput methods, are relatively costly,
and are not ideal with regard to either sensitivity (Ames) or
specificity (lymphoma tk and micronucleus) (Kirkland et al.,
2005). In addition, for drug development, a positive finding
often results in termination of work on a chemical series, perhaps
prematurely, since additional medicinal chemistry modifications
to the chemical series may eliminate a positive test finding. Tests
are generally reserved for relatively late-stage drug development
candidates. However, recent advances provide high-throughput
adaptations and alternatives to these genotoxicity assays.
Although not yet validated to the extent they can be used for
regulatory purposes, they expand the possibilities for providing
useful prioritization information. For example, in drug dis-
covery, the HTS methods permit an earlier screening of a much
larger pool of potential drug candidates, thus drawing attention
to possible liabilities earlier in the development process. For
environmental health concerns, a wider net can be cast to more
efficiently prioritize chemicals for more detailed evaluation of
carcinogenicity concerns, in particular if sensitivity and
specificity of the test or combinations of tests are superior.
The Ames reverse mutagenesis test requires relatively large
quantities of test chemical, approximately 250 mg, and is not
easily automated (Miller et al., 2005). An adaptation, known at the
Ames II assay, uses bacteria grown in liquid suspension in
microtiter plates rather than on agar plates (Gee et al., 1994). This
method was shown to have a very good concordance with the
standard Ames testing procedure (Fluckiger-Isler et al., 2004). It
reduces the amount of chemical required for testing and is
compatible with limited automation providing a means of signifi-
cantly increasing the throughput of this important genotoxicity
test.
A mammalian reporter cell line for DNA damage has been
developed from the human, p53-proficient, lymphoblastoid cell
line TK6 (Hastwell et al., 2006). The resulting cell line, GenM-
T01, contains a stably integrated, enhanced green fluorescent
protein (EGFP) reporter gene under the control of the upstream
promoter region of the human GADD45a (growth arrest and
DNA damage) gene. GADD45a is induced in a p53-dependent
manner in response to a wide variety of DNA damaging agents.
Initial validation efforts have shown very high sensitivity and
specificity for distinguishing genotoxic carcinogens from non-
genotoxic carcinogens and non-genotoxic non-carcinogens
(Hastwell et al., 2006). However, only chemicals not requiring
metabolic activation were used for the validation. A single
chemical, cyclophosphamide, was used to demonstrate that
addition of an S9 mix could be used to metabolically activate a
pro-genotoxin; however, the method used was not conducive to
HTS format. Further validation of this reporter cell line is
warranted given the potential to broadly and accurately screen
for genotoxic chemicals. Unfortunately, in lieu of a method to
provide biotransformation capacity, this assay cannot provide a
complete picture of genotoxicity potential, an issue of relevance
in general for in vitro toxicity assays.
Finally, automated cellular imaging technology has been
recently been applied to the micronucleus clastogenicity assay
resulting in much higher throughput and lower compound
requirement than the assay relying on traditional microscopy. In
this assay, cells are treated with test chemicals in microtiter
plates, stained, read on automated, fluorescence microscope
imaging systems, and scored using an image analysis algorithm.
Previous
TOC
-------
K.A. Houck, R.J. Kavlock / Toxicology and Applied Pharmacology 277 (2008) 163-178
167
Compound requirement is reduced from tens of milligrams to
1-3 mg and throughput is greatly increased (Lang et al., 2006).
Ion channel targets for toxicity screening
The hERG potassium ion channel is essential for normal
electrical activity in the heart by mediating ventricle repolariza-
tion (Sanguinetti and Tristani-Firouzi, 2006). Linkage of hERG to
pathology was demonstrated through mutations in the receptor
that caused cardiac arrhythmias. This promiscuous receptor is
blocked by a relatively wide variety of pharmaceutical
compounds resulting in prolongation of the QT interval of
cardiac rhythm. Examples include Seldane, Hismanal, Propulsid
and Raxar, all withdrawn from the market at least in part as a result
of these findings (Fermini and Fossa, 2003). Therefore, major
effort was put into developing HTS methods to screen drug
candidates early in the discovery process to identify poten-
tial problems. Several types of assays in HTS format were
develope —a receptor-binding assay, an ion efflux assay, and a
fluorescence assay based on membrane potential dyes. The
receptor-binding assay, while sensitive and reliable, detects
chemicals that bind to the receptor but may or may not induce
QT prolongation (Finlayson et al., 2001). Thus compounds active
in the receptor-binding assay must be followed up with assays
such as voltage-clamping electrophysiology measurements. The
rubidium efflux assay can provide modest throughput and a
functional endpoint, but is not as readily automated as the binding
assay due to requirement for a flame atomic absorption spectro-
meter as a reader (Cheng et al., 2002). Alternatively, use of the
radioactive rubidium isotope 86Rb and scintillation counting can
be used but generation of 86Rb waste in an HTS format is not
desirable (Weir and Weston, 1986). This method may also result
in right-shifted concentration-response curves indicating reduced
potencies relative to patch-clamp-derived values (Murphy et al.,
2005). Use of fluorescent, voltage-sensitive dyes in combination
with a fluorometric imaging reader is an additional HTS method
used for hERG although again sensitivity is reported as
significantly right-shifted relative to patch-clamp values (Murphy
et al., 2005). Development of patch-clamp instruments capable of
parallel measurements providing moderate-throughput offered
alternatives to the binding and rubidium efflux assay. However
published results suggest the instrument performs no better than
the rubidium assay with regard to sensitivity (Sorota et al., 2005).
Pharmaceutical compounds in development are now routinely
screened against hERG, but it does not appear that environmental
chemicals have been evaluated for activity against this channel.
Interpretation of the significance of activity of environmental
chemicals against hERG would need to consider the relevant
exposure scenarios for active chemicals together with the potency
of the chemicals for channel inhibition.
Receptor targets for toxicity screening
Beyond the specific molecular targets described above, the
drug discovery process has generated HTS assays against a large
number of candidate therapeutic targets. Some of these also
represent likely potential toxicity screening targets. For exam-
ple, there has been much interest in the potential for environ-
mental chemicals to act as endocrine disrupters in humans and
wildlife (Kavlock et al., 1996; Waring and Harris, 2005;
Hutchinson et al., 2000). One mechanism of action for endocrine
disrupters is to act directly as ligands to steroid hormone nuclear
receptors, in particular estrogen, androgen and thyroid hormone
receptors. High-throughput, gene reporter assays using lucifer-
ase or beta-lactamase have been developed that can easily be
used to screen for both agonist and antagonist activity for these
as well as other nuclear receptors (Giddings et al., 1997; Chin
etal., 2003). In addition, functional biochemical assays exist that
are capable of measuring receptor activation through ligand-
induced recruitment of coactivator proteins. A number of assay
formats based on protein:protein interactions have been
developed using the interacting domains of the nuclear receptor
and the coactivator protein. Ligand-induced interactions are
measured by fluorescence polarization with a fluorescently
labeled small peptide (Ozers et al., 2005), fluorescence reso-
nance energy transfer using fluorescently labeled receptor and
coactivator (Liu et al., 2003), or Amplified Luminescence
Proximity Homogenous Assay (AlphaScreen), a bead-based
approach with very high sensitivity (Rouleau et al., 2003).These
assays, too, can be run in agonist or antagonist mode. Each of
these assay formats is subj ect to a variety of types of interference,
e.g. fluorescence quenching, resulting in potential false positive
and false negative results. However, combining two formats, e.g.
cell-based gene reporter assay with biochemical coactivator
recruitment, would greatly reduce the number of inaccurate
findings. Traditional receptor binding assays using radiolabeled
ligands are also available as means to confirm results from the
other assays. Indeed, these binding assays can themselves be
configured to be conducted in high-throughput format using
scintillation proximity assay beads (Nichols et al., 1998).
Application of traditional binding study analysis methods, e.g.
Lineweaver-Burk plots, gives insight into mechanism of
binding that permits appropriate assessment of the results of in
vitro screens (Laws et al., 2006). Potency, efficacy and
mechanism of binding provide data that can be used to determine
the effects of the chemicals under given human exposure
scenarios, thus aiding in understanding the likelihood of hazard.
Recent work has linked cardiac valvulopathy with therapeutic
drugs having activity at the 5HT-2b G-protein-coupled receptor
(GPCR) including norfenfluramine and ergot-based Parkinson
drugs (Fitzgerald et al., 2000; Schade et al., 2007). HTS binding
assays for 5HT-2b are readily available and the receptor has been
reported to be one of the more promiscuous ones screened in
pharmacology safety profiling panels (Whitebread et al., 2005).
Given the growing evidence of the serious side effects associated
with 5HT-2b ligands, screening against this receptor is likely to
become a routine part of safety assessment for new drug develop-
ment. Screening of environmental chemicals for activity at this
receptor does not appear to have been reported.
Broad pharmacological profiling
Other molecular targets are less well characterized with
respect to demonstrated direct associations with clinical or
Previous
TOC
-------
168
K.A. Houck, R.J. Kavlock / Toxicology and Applied Pharmacology 277 (2008) 163-178
environmental toxicities. However, understanding of the
importance of many of these targets in normal physiology as
well as a variety of pathologies suggest that chemicals with
adequate potency against a target may represent a potential for
toxicity given sufficient exposure. As a result, a standard
method of assessing the possible liabilities of chemicals during
their development as new drugs is to broadly screen the
candidates against a panel of molecular targets in HTS assays
(Whitebread et al., 2005). Such an approach has also been
suggested for evaluating environmental chemicals for potential
for hazard (Dix et al., 2007). These panels vary in size as well as
the degree of known associations with clinical side effects and
toxicities. A number of research services companies provide
access to hundreds of these assays. GPCRs, kinases, proteases,
nuclear receptors, phosphatases, phosphodiesterases and other
protein families are included in these panels. Unfortunately,
details on use of the data derived from screening with these
panels have not been made readily available since most
screening is done in conjunction with proprietary drug discov-
ery research in the private sector. However, several publications
demonstrated use of a panel of 92 ligand-binding assays for
identifying relationships between molecular structure and broad
biological activity profiles (Fliri et al., 2005a) and relationship
to drug adverse side effects (Fliri et al., 2005b). Further
investigation of this type of approach is warranted given the
modest success despite using relatively crude datasets, e.g.
screening at a single concentration, the limited number and
breadth of assays employed, and toxicity endpoints derived only
from clinical side-effect data on commercial drug labels.
Several commercial databases have been developed containing
HTS data to support development of computational models to
predict toxicity including Bioprint (Mao et al., 2006) and
RSMDB (http://www.novascreen.com/rsmdb.asp). Such mod-
els are empirically derived based on profiles of known toxic
chemicals. Unfortunately, their general utility is difficult to
judge as they are not in the public domain. A currently growing
public database of informative HTS data is PubChem (http://
pubchem.ncbi.nlm.nih.gov/), the repository for screening
results from the National Institute of Health's Molecular
Libraries Roadmap (Austin et al., 2004). Screening conducted
at 10 Molecular Libraries Screening Centers includes a wide
range of targets and assay types that eventually will include
many of interest to the toxicology community. Currently, a
collection of ~ 100,000 small molecules constitutes the screen-
ing library and includes both compounds with known biological
activities as well as novel structures.
Complex cellular toxicity assays
The ability to link activity against specific molecular targets
with toxicity endpoints has significant limitations as evidenced
by the relatively few examples of clearly defined biochemical
assays for safety assessment. Future work likely will lead to
increased mechanistic understanding of various toxicities and
may identify additional targets for which surrogate assays can
be used. However these biochemical systems inherently lack the
complexity that may be required for mechanisms of toxicity to
be manifested. Such complexity does exist in whole animal
assays although important species differences often pose
problems in extrapolation to human safety assessment. With
the great interest in reduction of the number of animals used for
toxicity testing due to costs and ethical concerns, much attention
has been focused on use of cell-based assays for prediction of
toxicity. Use of cellular models provides a much higher level of
complexity than simple biochemical assays. The emergent
properties of cell system models may be more suitable for
detecting toxicity not necessarily linked to single molecular
targets. In addition, these properties may serve to put activity
against molecular targets in the context of normal in vivo
biology where robust homeostatic networks provide a buffering
capacity against chemical stresses to specific molecular entities.
Finally, the relatively easy accessibility of cells and cell lines
from a number of different species including humans facilitates
comparison of chemical activities across species. Research of
this nature may greatly help in understanding the reliability of
extrapolation of results from test animal to human.
Early work on use of cell-based assays for toxicity prediction
used primarily cytotoxicity endpoints such as ATP content,
membrane leakage, or cell number to screen for activity against
a variety of cell lines. Such data is useful for identifying acutely
toxic compounds as demonstrated by the Multicenter Evalua-
tion of In Vitro Cytotoxicity program carried out by the
Scandinavian Society for Cell Toxicology (Walum et al., 2005).
Cell lines, in particular human ones, showed good predictive
ability for 50 acute human toxicants. The coefficient of deter-
minations of linear regressions of ICsoS for the human cell line,
with LD50 values in man obtained from clinical and forensic
medicine handbooks, generated r2=0.69 vs 0.65 and 0.61 in
mouse and rat in vivo studies, respectively. This was increased
to r2=0.85 when toxicokinetic information on blood-brain
barrier penetration was added for 32 of the chemicals. Such an
approach, however, has significant drawbacks including lack of
mechanistic information and inability to predict non-acutely
toxic compounds. However recent efforts in several areas have
provided new opportunities to utilize cellular models to predict
toxicity and provide more insight into mechanisms of action.
Through evaluation of a wide variety of cell signaling and
stress-response pathways, activities of chronic toxicants are
more likely to be detected since these complex networks include
large numbers of potential toxicity targets. The impact of
automated, epifluorescence microscopy on development of
toxicity assays in support of this will be discussed below. In
addition, technological advances permitting novel high-
throughput approaches will also be reviewed.
Microscopy in cell biology has long been a descriptive
technology limiting its utility in quantitative studies, particu-
larly in large-scale ones. However, development of automated,
epifluorescence imaging platforms and robust image analysis
algorithms, together termed high-content imaging (HCI), has
provided the means to do high-throughput, quantitative analysis
of cellular phenotypic assays (Rausch, 2006). These assays
gather data at the single cell level and report information on
many parameters, an approach often referred to as high-content
screening (Giuliano et al., 1997). Among the many parameters
Previous
TOC
-------
K.A. Houck, R.J. Kavlock / Toxicology and Applied Pharmacology 277 (2008) 163-178
169
that can be quantified are cell signaling pathways, protein ex-
pression levels, cell cycle status, receptor internalization,
cytoskeletal integrity, energy metabolism status, nuclear mor-
phology, post-translational protein modifications, cell move-
ment, and cell differentiation. These parameters are measured
using specific fluorescent stains, fluorescently labeled anti-
bodies, and/or expression of GFP-tagged proteins. The imaging
platforms allow multiplexing of fluorescent endpoints with up to
four colors capable of being detected in a single well (Bertelsen,
2006). Although typically done as fixed endpoint assays, kinetic
analysis is also possible using instruments containing integrated
environmental chambers. Application of this technology to
toxicology has paralleled its development for drug discovery
uses in recent years.
A commonly employed application of HCI is to determine
the effect of chemicals on a set of parameters measuring cell
homeostasis. A variety of endpoints can be measured and
multiplexing permits efficient screening of chemicals and
increases sensitivity towards detection of toxic compounds
(O'brien et al., 2006). By combining measurement of cell
number and nuclear morphology (Hoechst 33342 dye), intra-
cellular calcium content (fluo-4 dye), mitochondrial membrane
potential (TMRM dye) and membrane permeability (TOTO-3
dye), sensitivity in detecting 243 hepatotoxic chemicals out of a
collection of 611 compounds was 93%, a significant improve-
ment over the <25% obtained using conventional cytoxicity
endpoints. In this study, human HepG2 hepatoblastoma cells
were used with exposure to compounds optimized to three days.
Of the endpoints measured, cell proliferation was the most
sensitive endpoint, perhaps reflecting the complexity of the
physiological processes regulating it, followed by mitochon-
drial membrane potential and nuclear morphology, with
membrane permeability and calcium levels being least sensitive
(O'brien et al., 2006). The complimentary nature of these mul-
tiple endpoints reflects different mechanisms of toxicity for the
chemicals tested and resulted in the high sensitivity obtained.
This study did not directly address mechanistic interpretation of
the results obtained; rather it presented an in vitro surrogate
system for detecting compounds likely to cause hepatotoxicity
in vivo.
Additional HCI endpoints would likely provide insight into
specific mechanisms of toxicity. For example, reactive oxygen
species generation can be determined with the use of the dye
dichlorofluorescein (Phillips et al., 2005). Nuclear morphology
is easily measured with Hoechst dye. Interference with the cell
cycle can be measured by monitoring the expression of cell
cycle-specific proteins, e.g. cyclins (Gasparri et al., 2006).
Mitotic index assays can be performed by labeling phosphory-
lated histone proteins (Gasparri et al., 2004). Apoptosis can be
detected with multiplexed endpoints including nuclear con-
densation and caspase activation (Fennell et al., 2006). Out-
growth of neurites from neuronal precursor cells (Richards
et al., 2006) or cell lines can be monitored and used to determine
potential neurotoxicity demonstrated by selective activity
against neurites relative to other cytotoxic effects. As previously
mentioned in the genotoxicity assay section, micronucleus
detection as a measure of clastogenicity is feasible in a relatively
high-throughput format (Lang et al., 2006). With all of these
phenotypic assays, multiple endpoints can be combined in a
single assay provided they are within the spectral limitations
of the fluorescent dyes used and instrument capabilities.
Since chemicals with similar mechanisms of toxicity would
be expected to have similar cellular phenotypic effects, mea-
surement of HCI endpoints can be used as a classification
system. Quantifying and statistical clustering of a large number
of endpoints resulted in successful classification of 45 of 51
chemicals as to mechanism of action for a diverse set of
chemicals representing at least 12 different mechanism classes,
e.g. actin poisons and topoisomerase II inhibition (Adams et al.,
2006).
Combining the ability to quantify phenotypic changes
relating to toxicity with the emerging opportunity to interrogate
primary human cells may yield assays of high value as in vitro
surrogate toxicity screens. A number of features of HCI make
this an ideal technology for use with primary cells. Relatively
few cells are required as typically only 100-500 cells are
counted per well, accommodating the often very limited
availability of primary cells as well as batch-to-batch variability
issues (Borchert et al., 2006; Mayer et al., 2006). Many
endpoints measured by HCI rely on antibody staining or
fluorescent dyes of endogenous proteins, thus eliminating the
need for protein over-expression or other cell engineering
tactics. In recent years, a large variety of primary human (and
other species) cells have been made available commercially.
Advances in use of pluripotent stem cells differentiated into
desirable target cell types such as cardiomyocytes present
opportunities for providing ready supplies of appropriate cell
types of interest for testing (Sartipy et al., 2006). In addition,
more complex cell culture systems may be possible containing
two or more different cell types to better represent in vivo
physiology. Morphological or specific antigenic staining can be
used to distinguish between the populations. Such a system, for
example, might represent the interaction of liver Kupffer cells
and hepatocytes, perhaps critical for understanding hepatocar-
cinogenesis and providing a means to measure effects of
chemicals on cell-cell interactions not seen in single cell
cultures (Roberts et al., 2007).
Another strategy that has generated mechanistic under-
standing, using a HTS panel of cell proliferation assays, has
been employed for more than a decade at the National Cancer
Institute (NCI). Profiling of tens of thousands of chemicals for
growth inhibitory activity across the National Cancer Institute's
~ 80 tumor cell line screen created an information-rich database
useful for understanding mechanism of action (Shoemaker
et al., 2002). Chemicals were clustered based on structural
information and biological data using a self organizing maps
(SOM's) statistical approach resulting in groups of compounds
with similar cytotoxicity profiles. As an example of the use of
this technique for understanding mechanism, a cluster contain-
ing a number of known inhibitors of mitochondrial complex I of
the electron transport chain was studied (Glover et al., 2007).
Ten chemicals with unknown mechanism of action were select-
ed from the cluster and subsequently five of these shown to be
potent inhibitors of complex I activity.
Previous
TOC
-------
170
K.A. Houck, R.J. Kavlock / Toxicology and Applied Pharmacology 277 (2008) 163-178
Berg et al. (2006) followed a similar strategy through use of
an HTS panel of four human cell-based assays consisting of
primary endothelial and peripheral blood mononuclear cells,
alone or in co-culture, and measured multiple inflammation-
related endpoints, primarily by ELISA. In testing of a set of 44
compounds consisting of a wide variety of pharmacological
tools and drugs, they successfully grouped compounds by
mechanism of action, e.g. modulators of NFkB signaling or the
phosphatidylinositol 3-kinase/Akt signal transduction pathway,
through hierarchical clustering and function similarity mapping
of the activity profiles in the assays. This included successful
classification of compounds with mechanisms unrelated to
inflammatory system modulation. They attributed their success
to use of primary cells that maintain appropriate regulatory
processes and signaling pathway interactions. This approach
seems well-suited towards mechanistic understanding of
toxicity. Development of additional complex cellular models
and profiling of larger libraries of pharmacological tools would
provide a resource for understanding and predicting potential
toxicities for new chemicals by highlighting specific mechan-
isms affected by the tested compounds.
Similarly, MacDonald et al. (2006) profiled the activity of
107 drugs and pharmacological tools in 49 different human cell-
based assays at several time-points generating a matrix of 127
different measurements for each compound. The assays
consisted of cellular reporter assays using protein complemen-
tation technology for diverse protein interactions within cells
including those involved in cell-cycle, apoptosis, mitogenesis,
proteolysis, GPCR signaling, cytoskeletal function, DNA
damage, nuclear receptor signaling, stress and inflammation.
Data were analyzed by clustering and revealed known
mechanisms of action as well as novel activities for some of
the drugs. A subset of the assays was shown to be predictive for
antiproliferative activity.
An additional approach for cell-based toxicity screening for
characterizing mechanism of toxicity used dynamic monitoring of
cytotoxicity by microelectronic sensors in 96- or 384-well arrays
(Kirstein et al., 2006). The assay determines changes in cell status
including viability, morphology and adherence by measuring
electrical impedance in real-time. This approach has several
attractive features including a label-free detection system and the
ability to use any attached cell type. Testing of 11 reference
toxicants on the mouse fibroblast cell line BALB/c 3T3 yielded a
correlation coefficient of r2=0.98 comparing in vitro ICso with
mouse in vivo LD50 values (Xing et al., 2006). Prediction of acute,
in vivo toxicity by this method is likely limited to classes of
chemicals with direct effects on target cells. In addition to
accurately predicting acutely toxic effects of chemicals, specific
kinetic patterns of response to different toxicants over a 72 hr
period provides insight into possible mechanisms of action by
comparison to patterns induced by compounds of known
mechanism of action (Xing et al., 2005; Solly et al., 2004).
Model organism toxicity assays
While complex cell culture systems of primary cells can
provide unique insights into in vivo pathophysiology, they will
never completely model the higher level interactions present in
an intact organism. For that reason, efforts have been made to
adapt HTS technologies to assays using whole animals in the
form of model organisms that may yield important mechanistic
information. Model organisms used include both invertebrates,
e.g. the yeast, Saccharomyces cerevisiae (Fairn et al., 2007),
and nematode, Caenorhabditis elegans (Anderson et al., 2004);
and vertebrates, e.g. the zebrafish, Danio rerio (Zhang et al.,
2003).
Ease of growth, a completely sequenced genome and
availability of a wide range of genetic mutants has resulted in
extensive use of S. cerevisiae for understanding physiological
and pathophysiological processes in higher level organisms.
Several recent approaches demonstrated the utility of yeast in
examining mechanisms of toxicity. Chemogenomic profiling in
S. cerevisiae, a method that determines chemical sensitivity
using a whole genome screen, was used to identify the ability of
an apoptosis-inducing chemical, the isoprenoid farnesol, to kill
cells through generation of reactive oxygen species by the
electron transport chain. The Pkc 1 signaling pathway was shown
to mediate farnesol-induced cell death through modulation of the
generation of reactive oxygen species (Fairn et al., 2007).
Parsons et al. (2006) used a complete library of yeast haploid
deletion mutants (~ 5000 strains) to test for hypersensitivity to
82 chemicals and natural product extracts which included many
drugs and other pharmacological agents. Screening of 5000
strains was done efficiently through parallel treatment using
large pools of deletion strains with identification of specific
strains determined by molecular barcodes, unique 20-mer
oligonucleotide tags associated with the specific gene disruption
and quantitated by hybridization to oligonucleotide arrays com-
plimentary to the barcode sequences (Giaever et al., 2002).
Using both hierarchical clustering and a factorgram method that
allows chemicals to be included in multiple sets, chemicals with
similar modes-of-action were grouped and correlated with
effects on specific cellular pathways and protein targets. Crude
natural product extracts were also demonstrated to display the
activity profile of their constituent active component suggesting
utility in screening mixtures and poorly characterized samples.
The same method was applied to examining the activity of 12
chemicals with DNA-damaging activity (Lee et al., 2005). Both
known and novel functional interactions were discovered by this
screening method and distinctions between types of DNA
damage and response shown.
C. elegans is a well-studied organism easily grown in the lab
on agar or in liquid medium and fed with bacteria. All 959
somatic cells have been characterized with respect to lineage
and, importantly for screening, are visible by microscopy.
Extensive genetic manipulation of the worm provided avail-
ability of a large number of mutant strains (C. elegans WWW
Server: http://elegans.swmed.edu/) and C. elegans is conducive
to RNAi studies simply by soaking the worm in dsDNA or
feeding with bacteria containing specific plasmids (Kamath and
Ahringer, 2003). Thus hypersensitivity and hyposensitivity to
specific toxicants can be linked to over- or under-expression of
genes providing a link to toxic mechanism (Lindblom et al.,
2001; Liao and Yu, 2005). For example, arsenic-induced
Previous
TOC
-------
K.A. Houck, R.J. Kavlock / Toxicology and Applied Pharmacology 277 (2008) 163-178
171
oxidative stress is more toxic to "/-glutamylcysteine synthetase
(GCS-1) mutant C. elegans than to wild-type demonstrating the
role of glutathione in protecting from arsenate-induced
oxidative stress. Quantitative handling of the worms as well
as detection of assay endpoints are key factors limiting
screening to modest throughput. Development of the COPAS
Biosorter, a flow-cytometer modified for larger objects, allowed
sorting of up to 100 worms per second as well as fluorescent
analysis of signals along the worm's axis providing a means to
automate the assays (Pulak, 2006). Adaptation of HCI instru-
mentation to measuring endpoints of C. elegans assays is a
likely future development in this field.
D. rerio, the zebrafish, as well as other fish species such as
the fathead minnow, have been used for environmental toxicant
testing for many years in the role of a sentinel species (Mizell
and Romig, 1997). Although informative as to presence of
toxicants, little mechanistic information resulted. More recently,
the zebrafish has been used as a model organism for a wide
variety of research including drug discovery and toxicology. In
particular, the transparent embryo of the zebrafish allow
changes to organ morphology to be detected with a dissecting
microscope with more detailed information obtained by
visualization of serial sections stained with specific antibodies
or by in situ hybridization. Using this approach, Haendel et al.
(2004) demonstrated malformations, in particular a twisted
notochord, induced by the dithiocarbamate pesticide Metam
Sodium and correlated this with altered expression of collagen
2al in the notochord sheath. Further work that allowed
identification of a target in a pathway regulating collagen 2al
synthesis could then provide an opportunity to compare across
species the target's existence and function as well as the potency
of the chemical for the target relative to real-world exposure
levels. Parng et al. (2007) applied immunostaining and
morphometric analysis to studying neurotoxicity in zebrafish
and distinguished effects of a variety of human neurotoxicants
on different neuron populations. Relative to larger vertebrates,
these studies are inexpensive and capable of testing moderate
numbers of chemicals, although not in the traditional HTS
range, but are still laborious and are generally not applied to
large chemical libraries. In order to improve throughput of
scoring phenotypic endpoints in toxicity assays, transgenic
embryos expressing GFP in specific locations utilizing tissue-
specific promoters have been generated. Transgenic fish are
relatively easily made by injection of DNA vectors at the one- or
two-cell stage. This permits screening for effects on the labeled
tissue using automated fluorescent microscopy of fish embryos
in microtiter plates (Burns et al., 2005). Effects on the devel-
oping heart can be determined through use of a transgenic
zebrafish expressing GFP in the myocardium and measuring
heart rate by an HCI system. In addition to genetic manipulation
of the zebrafish, a large collection of mutant strains of the fish
have been collected (http://zfin.org/zf_info/dbase/db.html) and,
through comparison to effects induced by toxicants, may
provide clues to mechanisms of toxicity. Zebrafish can also be
injected with morpholino oligonucleotides permitting a reverse
genetics approach potentially useful in elucidating mechanisms
of action (Nasevicius and Ekker, 2001). As with C. elegans,
high-throughput approaches have just begun to be adapted for
screening of zebrafish. The ability to interrogate intact,
vertebrate organisms, particularly during developmental stages,
gives access to a wide variety of toxicities perhaps detectable
only in whole animals. Elucidating the mechanisms involved,
evaluating that mechanism across species, and testing across
species against any molecular targets identified will greatly
enhance risk assessment.
Data analysis
As discussed above, many methods exist for efficiently and
relatively inexpensively generating data by high-throughput
means on large collections of chemicals. Such data in isolation
have limited utility. Only through integration and analysis of
these data is meaningful understanding of mechanisms of
toxicity likely to result. Towards this goal, two general analysis
methods have been applied. The first, classification based on
similarity of bioactivity profiles to those of chemicals of
known toxic mechanism, relies on a relatively large number of
screening assays to ensure comprehensive coverage of possible
mechanisms. The idea is based on the assumption that
mechanistically related toxic chemicals will display similar
patterns of biological activity recognized through characteristic
signatures in in vitro tests. On the other hand the converse of this
idea, i.e. compounds with similar in vitro signatures will have
similar in vivo toxicities, does not necessarily follow due to a
lack of higher level biological interactions in these in vitro
assays. Exhaustive screening of all protein targets is not
practical. However, because polypharmacology, i.e. the ability
of chemicals to bind to more than one protein target, is common
and occurs most often between members of the same protein
family, sufficient representation of protein families in the
screening panel may provide an adequate sampling of specific
targets or their related family members to develop signatures
(Paolini et al., 2006). Chemicals with known mechanism of
toxicity that are used as reference standards will likely play a key
role in interpreting these data. Profiles of biological activity for
these reference chemicals anchored to in vivo toxicity endpoints
can serve as classifiers to suggest potential hazard for chemicals
of unknown toxicity. Discovery of novel patterns of activity
could suggest new mechanisms of action outside the list of
reference compounds and trigger new research hypotheses.
Alternatively, a quantitative determination of the effect of
individual chemicals on molecular targets and cellular signaling
pathways, coupled with information on phenotypic endpoints,
can be applied in a cell systems toxicology approach. These
hypothesis-driven datasets would be integrated based on
knowledge of cellular and higher level systems responses
using computational and informatics tools from systems biology
research (Fig. 1). Such work would help define toxicity
pathways and could lead to development of cellular response,
pathway-based assays that would serve as key targeted-
information screening assays of the type described by the
National Research Council (NRC, 2007). These pathway assays
would report perturbation of components of the pathways by
tested chemicals suggesting potential for hazard and increasing
Previous
TOC
-------
172
£yl flbudt, «../ Kavlock / Toxicology and Applied Pharmacology 277 (2008) 163-178
Cell Responses
Proliferation
Apoptosis
Mitochondrial Fx
ERK Pathway
Wnt Pathway
NFkB Translocation
Cytokine 1 Release
Cytokine 2 Release
Ca2+ Release
SMAD translocation
FKHLR nuclear translocation
Etc.
Biochemical
Targets
Kinase 1
Kinase 2
Kinase 3
GPCR1
GPCR2
GPCR3
Ion channel 1
Ion channel 2
PDE1
PDE2
Nuclear Receptor 1
Nuclear Receptor 2
Etc.
Mechanism
Fig. 1. Cell systems toxicology. Chemicals are screened through panels of biochemical targets and cellular phenotypic screens. Data are in
biology tools to identify likely mechanisms of action. (Illustration used with permission from BioCarta.)
using systems
priority for selective in vivo toxicity testing. Quantitative
modeling of affected systems, assuming availability of sufficient
information, would then provide key, mechanistically relevant
input into rationale risk assessment for the toxicant. Such
modeling would take into account the pathway being perturbed
not in isolation, but as part of a dynamic regulatory system
composed of multiple interacting pathways. This systems ap-
proach may reveal unexpected environmental effects on human
disease as well as the presence of compensatory responses that
maintain cellular homeostasis despite environmental stress.
Experimental considerations
As with all high-throughput assays, there are a number of
important caveats that must be considered in designing
screening strategies and interpreting results. Data quality is of
utmost importance. The "garbage in, garbage out" lesson holds
very true for HTS data and has serious implications for at-
tempting to screen for potential for hazard or illuminate mech-
anistic understanding of results. For this reason, all HTS testing
needs to be conducted under stringent, well-validated condi-
tions that include determination of Z'-scores, signal-to-back-
ground, reproducibility, plate positional effects and appropriate,
biologically relevant controls (Zhang et al., 1999). The Z'
statistic is widely used in HTS assay validation and relates
variability around the minimum and maximum signals of the
assay to the assay window size (Mean Maximum Signal - Mean
Minimum Signal) (Fig. 2). Values of > 0.5 indicate suitability of
the assay for HTS applications. Assays with lower values can
still be used but with likely significantly higher false positive
and negative results (Zhang et al., 1999). Signal-to-background
should generally be as large as possible using biologically
relevant controls to determine the window. Day-to-day
reproducibility is best determined by measuring ICso or ECso
values from independent runs and calculating variability. Plate
positional effects are best detected with analysis of variance of
rows and columns combined with good data visualization tools.
In addition, data quality of HTS assays is subject to a variety of
artifactual interferences potentially resulting in false positives
5 3000
0) 2500 '•
J =
Relative
_i _i r
•* • • • • .•*•••
• % % „ • •••••# #•*•• ••••./•••• ,
r. "..*•" .*.•••..*••*•* :'•••>
n Min • Max
Z' = 1 - [(3*STDmax + 3*STDmin)/(AVGmax-AVGmm)] = 0.76
^^jPWItefl} OIQ4I%lj3*Il^^ DC ttoflftyftyjJDi CtcfTHTtD O3
0 10 20 30 40 50 60 70 80 90
Well Number
Fig. 2. Example of calculation of Z'-statistic for validation of an HTS screening
assay. Data are derived from two 96-well plates; one with solvent control (Min)
and one with positive control (Max). The Z' of 0.76 indicates an assay highly
suitable for HTS. No plate positional effects are observed.
Previous
-------
K.A. Houck, R.J. Kavlock / Toxicology and Applied Pharmacology 277 (2008) 163-178
173
Table 2
Common modes of interference with HTS assays
Assay type
Interference
Cause
Accommodation
Fluorescence, luminescence
Fluorescence
Fluorescence, luminescence,
colorimetric
AlphaScreen
Enzymatic inhibition
Luminescent reporter
Fluorescent imaging
Cell-based assay
Innerfilter effect Compound or substrate absorption of
(quench of signal) excitation or emission light from tracer
Autofluorescence Compound or other assay component
emission of light in range of assay signal
Light scattering Compound or other assay
(quench of signal) component insolubility
Signal quench Singlet oxygen quench
Non-specific loss of signal Aggregation of enzyme by compound
Non-specific loss of signal
Loss of signal
General cytotoxicity
Direct inhibition of luciferase
enzymatic activity
Photobleaching
Varied
Read of compound alone; use of red-shifted dyes; secondary
assay in alternative format; reduce screening concentration
Read of compound alone; use of red-shifted dyes; secondary
assay in alternative format
Reduce screening concentration; increase organic
solvent concentration in assay buffer
Test in control AlphaScreen assay not requiring target;
secondary assay in alternative format
Inclusion of detergent in assay buffer; reduce
screening concentration
Run control assay for inhibition with added recombinant
luciferase; secondary assay in alternative format
Use more stable dye
Run control assay for cell viability, e.g. ATP
content assay; reduce screening concentration
and false negatives. Table 2 lists some of the most common of
these problems. Strategies to deal with specific issues must be
incorporated into screening programs. Beyond assay validation
and interference concerns, there are several other issues leading
to false positives and false negatives which must be kept in
perspective (Table 3). As previously mentioned, pro-toxicants,
i.e. those requiring biotransformation to a toxic metabolite, will
require a metabolic activation system to detect activity.
Technological advances are urgently needed in this area. Poor
aqueous solubility, particularly for many environmental chemi-
cals which are often very lipophilic, is also of great concern
since screening assays are usually conducted in an aqueous
buffer.
The potential problems described above can be of very high
concern in the field of toxicology where false negative
screening results may cause significant toxicities and mechan-
istic understanding to be overlooked. Development of assays
with both high sensitivity and high specificity is the obvious
goal for such research, although this is usually difficult to
achieve. In drug discovery, false negative rates are treated as
less critical since the goal is to identify a viable starting point for
drug development, not find every compound with activity at the
target. False positive rates in drug discovery present a problem
only to the extent that actives become too numerous to
thoroughly follow-up on and result in missed opportunities
for finding the rare, true positives. For toxicity screening, false
negative rates need to be minimized, often at the cost of lowered
specificity. This can be accomplished with highly robust assays,
testing samples in replicate, appropriate chemical handling, and
testing at multiple concentrations. Within reasonable limits, the
lowered specificity usually accompanying increased sensitivity
can be handled by two methods: 1) confirmation assays that
eliminate false positives resulting from statistical variation or
procedural errors, and 2) testing against the same target in an
alternative format to eliminate artifacts due to interference with
the assay technology. As further insurance against false
negatives, redundancy can be built in to an overall screening
strategy. For example, screening against a wide panel of
molecular targets in combination with cellular pathway screens
that encompass many of these same targets in a cellular context
gives multiple, independent opportunities to detect activity.
Current limitations to HTS approaches requiring technical
solutions include lack of biotransformation capacity capable of
metabolism of a chemical to a toxic form or, conversely, its
inactivation, and difficulty dealing with volatile or aqueously
insoluble chemicals.
Table 3
Critical issues with HTS assays for toxicity screening
Issue
Cause
Accommodation
False negative Lack of biotransformation to active metabolite
False negative Poor aqueous solubility of chemical
False negative Poor solubility of chemical stock in standard solvent
False negative Testing concentration too low
False negative Lack of appropriate assay
False negative Statistical issue
False positive Testing concentration too high
False positive Lack of appropriate biological complexity for
adaptation to toxicant stress
False positive Lack of biotransformation to inactive metabolite
Include biotransformation system in assay
Determine solubility with nephelometer; increase organic solvent
in assay (if tolerated); reduce screening concentration
Change solvent and ensure compatibility with assay
Increase testing concentration to limits of solubility
Expand potential target coverage with broader panels or complex cellular assays
Use appropriately validated assays; replicate testing; use
concentration-response testing format
Use concentration relevant to exposure scenarios; interpret
quantitative result in the context of relevant exposure scenarios
Develop/use more complex cellular assays
Include biotransformation system in assay
Previous
TOC
-------
174
K.A. Houck, R.J. Kavlock / Toxicology and Applied Pharmacology 277 (2008) 163-178
With a myriad of assay possibilities, how can toxicity
screening be efficiently utilized to recognize toxicants and shed
light on mechanism of action? One possibility would be as
illustrated in Fig. 3. Initial screening would occur with a
comprehensive panel of cellular assays covering cell response/
signaling pathways and cell health parameters. Testing concen-
trations and interpretation of concentration-response results
Chemicals
would ideally be informed by realistic exposure scenarios. Use
of negative control compounds such as Generally Regarded As
Safe (GRAS) substances would help provide confidence in
positive findings. Model organism assays that capture complex
biology not capable of being detected in simpler formats may be
included here as well. These pathway assays would likely have
varying levels of linkage to known toxicity outcomes based on
aqueous solubility | aqueous insolubility
volatile ^ non-volatile
Special handling required
Special handling required
Testing at exposure-
relevant concentrations
Cellular Response
Pathways
Cell Health Panels
Model
v^^ organism ^>
active
inactive
Predict
biotransformation
no metabolites I metabolites
r
Low priority
Consider testing
metabolites
Protein superfamily panel biochemical screens
kinases
phosphatases
GPCRs
transcription
factors
NRs
proteases
CYPs
DMA modifying
enzymes
other
enzymes
Integrate results;
mechanism hypothesis
Exposure
determination
PB/PK
Modeling
Selective
Animal testing
Risk Assessment
Fig. 3. Chemical toxicity testing flow scheme.
Previous
-------
K.A. Houck, R.J. Kavlock / Toxicology and Applied Pharmacology 277 (2008) 163-178
175
testing of reference toxicants. Experience and knowledge
gained over time would demonstrate which of the pathways
can be confidently called toxicity pathways. Chemicals with
activity in one or more of these assays would then pass to a large
screening panel of in vitro, biochemical targets. Activity pat-
terns in these assays would lead to hypotheses about mechanism
of action. The strength of the evidence linking the in vitro
results with in vivo toxicity would drive decisions about the
necessity of additional selective biochemical, cellular and/or
animal testing. Animal testing may also be required to under-
stand pharmacokinetic parameters that may influence toxicity.
Together, these data would provide important qualitative and
quantitative information to be used for a risk assessment
analysis of the chemical. The capability to carry our most of this
effort currently exists although means to deal with biotransfor-
mation, highly volatile chemicals, and poorly aqueously soluble
chemicals are still needed.
Challenges of HTS application to toxicology
As recently noted by the National Research Council (2007),
toxicology is reaching a pivotal movement. The emergence of a
plethora of advances in molecular biology, cell culture tech-
niques, computer science and bioinformatics is providing an
opportunity to acquire information on the cellular and molecular
pathways affected by chemicals in ways unimagined only a
decade ago. Large numbers of biological targets as identified in
this review can be probed for interactions with hundreds to
thousands of chemicals in short periods of time. The NRC
envisions a future of toxicology that includes multi-faceted
exploration of chemical characterization (e.g. physical and
chemical properties, environmental fate and transport, metabo-
lism, and interaction with cellular components), toxicity
pathways (defined as a 'cellular response pathway that, when
sufficiently perturbed in an intact animal, are expected to result
in adverse health effects') and targeted in vivo testing (to further
explore and quantify information obtained by exploring toxicity
pathways). Combined, the NRC foresees these elements
providing the basis for a more informed assessment of human
health risks based on a deeper understanding of the mode of
action by which toxic effects are induced, including the key
molecular and other biological targets in the pathways.
Many of the new tools now available to toxicology origi-
nated in the pharmaceutical industry as they strove to streamline
the process of drug discovery and development. While overall
assessments of the success of that industry are cloaked in
confidential business information, this review has identified a
number of efforts that have at least provided preliminary
glimpses of their potential to detect toxicity pathways. Never-
theless, there are numerous differences and challenges in
translating the experience of drug development to environ-
mental toxicity. Notable of these are: (1) drugs are developed as
biologically active compounds, and hence may be more
amenable to such an approach than environmental chemicals,
many of which have no intended biological activity; (2) the
chemical space covered by drug development is considerably
narrower than that of environmental chemicals, which do not
have to display particular ADME characteristics that make drugs
bioavailable and efficacious; (3) the metabolism of drugs is
generally well characterized, as is the activity of any metabolites;
(4) selection of chemical libraries used for drug discovery
include consideration of solubility, something not necessarily of
importance for environmental chemicals; (5) while there may
be only 400-500 drugable targets, the number of potential
biological off-targets for drugs and environment chemicals is
likely to be quite large, necessitating a very broad search for
toxicity pathways; (6) for assessment of environmental chemi-
cals, confidence in the value of a negative results in any assay
will have to be higher than for drugs, which will certainly
undergo additional scrutiny as preclinical studies continue on
lead compounds; (7) the probability that chemicals are going to
interact with more than a single toxicity pathway, with exposure
intensity and duration playing important determinants as to
which one(s) are key to inducing adverse health effects; (8) the
likelihood that the outcomes of perturbing toxicity pathways will
be cell-type dependent, reflecting the variability in expression of
coactivators, corepressors or other mitigating factors; and (9) the
aspect that at least some forms of toxicity are dependent on
higher order interactions of cells in tissues or organs what may
not be apparent by a reductionist evaluation of isolated cells and/
or pathways. Even this limited list of obstacles exposes the
daunting challenge in creating a new paradigm for toxicological
evaluation, and it also points to the need for a strategic approach
to design a research program that can begin to chip away at the
obstacles.
The application of HTS in toxicology could be twofold, one
essentially bottom up, the second top down. In the bottom up
approach, a single or small number of chemicals could be
analyzed against a vast array of targets to isolate the key toxicity
pathways. In the top down approach, a relatively large number
of chemicals could be assayed against a small number of key
targets (this is essentially the approach being used for detection
of endocrine disrupting chemicals that act via interaction with
estrogen, androgen or thyroid hormone function). A middle
ground would try to maximize both the numbers of chemicals
assayed and the breadth of the assays so as to achieve a
biologically based prioritization process. This process could
then be applied to chemicals that have large potential for human
exposure, but for which have avoided extensive traditional
toxicological evaluations due to uses not covered by the data-
intensive regulatory statutes such as Federal Insecticide,
Fungicide and Rodenticide Act. It is this latter application that
should serve as fertile test of the utility of HTS technologies,
and one that is being explored by the EPA (Dix et al., 2007).
Under this ToxCast program, several hundred chemicals with
well characterized toxicological effects are being evaluated
against more than 400 biological targets. The biological targets
include activities against a variety of enzymes, kinases,
phosphatases, ion channels, nuclear receptor binding and
response modulation, exocrine function of stimulated cells,
transcript profiles of exposed primary cell types, and responses
of model organisms. The chemicals currently being looked at
are primarily pesticidal agents that have subchronic, chronic,
developmental and reproductive assays available. Since they are
Previous
TOC
-------
176
K.A. Houck, R.J. Kavlock / Toxicology and Applied Pharmacology 277 (2008) 163-178
designed to have some biological activity, the probability of
detecting interactions with some cellular pathways should
be relatively high, and the resulting bioactivity profiles across
all targets should provide a rich dataset to evaluate whether
the particular profiles can be associated with certain specific
health outcomes. The current status of this program is available
at http://epa.gov/ncct/toxcast.
Conflict of interest statement
The authors report no conflicts of interest with respect to this
manuscript.
This work was reviewed by EPA and approved for publi-
cation but does not necessarily reflect official Agency policy.
Mention of trade names or commercial products does not
constitute endorsement or recommendation by EPA for use.
References
Adams, C.L., Kutsyy, V., Coleman, D.A., Cong, G., Crompton, A.M., Elias,
K.A., Oestreicher, D.R., Trautman, J.K., Vaisberg, E., 2006. Compound
classification using image-based cellular phenotypes. Methods Enzymol.
414, 440-468.
Anderson, G.L., Cole, R.D., Williams, P.L., 2004. Assessing behavioral toxicity
with Caenorhabditis elegans. Environ. Toxicol. Chem. 23, 1235-1240.
Austin, C.P., Brady, L.S., Inse, 1T.R., Collins, F.S., 2004. NIH molecular
libraries initiative. Science 306, 1138-1139.
Berg, E.L., Kunkel, E.J., Hytopoulos, E., Plavec, I., 2006. Characterization of
compound mechanisms and secondary activities by BioMAP analysis.
J. Pharmacol. Toxicol. Methods 53, 67-74.
Bernauer, U., Glatt, H., Heinrich-Hirsh, B., Liu, Y, Muckel, E., Vieth, B.,
Gundert-Remy, U., 2003. Heterologous expression of mouse cytochrome
P450 2el in V79 cells: construction and characterization of the cell line and
comparison with V79 cell lines stably expressing rat P450 2E1 and human
P450 2E1. ATLA 31, 21-30.
Bertelsen, M., 2006. Multiplex analysis of inflammatory signaling pathways
using a high-content imaging system. Methods Enzymol. 414, 348-363.
Bettoun, DJ., Burris, T.P., Houck, K.A., Buck II, D.W., Stayrook, K.R., Khalifa,
B., Lu, J., Chin, W.W., Nagpal, S., 2003. Retinoid X receptor is a nonsilent
major contributor to vitamin D receptor-mediated transcriptional activation.
Mol. Endocrinol. 17, 2320-2328.
Borchert, K.M., Sells Galvin, R.J., Hale, L.V., Trask, O.J., Nickischer, D.R.,
Houck, K.A., 2006. Screening for activators of the wingless type/Frizzled
pathway by automated fluorescent microscopy. Methods Enzymol. 414,
140-150.
Bucher, J.R., 2002. The National Toxicology Program rodent bioassay: designs,
interpretations, and scientific contributions. Ann. N.Y. Acad. Sci. 982,
198-207.
Burns, C.G., Milan, D.J., Grande, E.J., Rottbauer, W., MacRae, C.A., Fishman,
M.C., 2005. High-throughput assay for small molecules that modulate
zebrafish embryonic heart rate. Nat. Chem. Biol. 1, 263-264.
Cheng, C.S., Alderman, D., Kwash, J., Dessaint, J., Patel, R., Lescoe, M.K., Kinrade,
M.B., Yu, W., 2002. A high-throughput HERG potassium channel function
assay: an old assay with a new look. Drug Dev. Ind. Pharm. 28, 177-191.
Chin, J., Adams, A.D., Bouffard, A., Green, A., Lacson, R.G., Smith, T.,
Fischer, PA., Menke, J.G., Sparrow, C.P., Mitnaul, L.J., 2003. Miniaturiza-
tion of cell-based beta-lactamase-dependent FRET assays to ultra-high
throughput formats to identify agonists of human liver X receptors. Assay
Drug Dev. Technol. 1, 777-787.
Crespi, C.L., Stresser, D.M., 2000. Fluorometric screening for metabolism-
based drug-drug interactions. J. Pharmacol. Toxicol. Methods 44,325-331.
Crespi, C.L., Miller, V.P., Penman, B.W., 1997. Microtiter plate assays for
inhibition of human, drug-metabolizing cytochromes P450. Anal. Biochem.
248, 188-190.
Cui, X., Thomas, A., Han, Y, Palamanda, J., Montgomery, D., White, R.E.,
Morrison, R.A., Cheng, K.C., 2005. Quantitative PCR assay for
cytochromes P450 2B and 3A induction in rat precision-cut liver slices:
correlation study with induction in vivo. J. Pharmacol. Toxicol. Methods 53,
215-218.
De Rosa, C.T., Nickle, R., Faroon, O., Jones, D.E., 2003. The impact of
toxicology on public health policy and service: an update. Toxicol. Ind.
Health 19, 115-124.
Dix, D J., Houck, K.A., Martin, M.T., Richard, A.M., Setzer, R.W, Kavlock, R J.,
2007. The ToxCast Program for prioritizing toxicity testing of environmental
chemicals. Toxicol. Sci. 95, 5-12.
Ennever, F.K., Lave, L.B., 2003. Implications of the lack of accuracy of the
lifetime rodent bioassay for predicting human carcinogenicity. Regul.
Toxicol. Pharmacol. 38, 52-57 (Aug).
Fairn, G.D., Macdonald, K, McMaster, C.R., 2007. A chemogenomic screen in
Saccharomyces cerevisiae uncovers a primary role for the mitochondria in
farnesol toxicity and its regulation by the Pkcl pathway. J. Biol. Chem. 282,
4868-4874.
Faucette, S.R., Sueyoshi, T, Smith, C.M., Negishi, M., Lecluyse, E.L., Wang,
H., 2006. Differential regulation of hepatic CYP2B6 and CYP3A4 genes by
constitutive androstane receptor but not pregnane X receptor. J. Pharmacol.
Exp. Ther. 317, 1200-1209.
Fennell, M., Chan, H., Wood, A., 2006. Multiparameter measurement of caspase
3 activation and apoptotic cell death in NT2 neuronal precursor cells using
high-content analysis. J. Biomol. Screen. 11, 296-302.
Fermini, B., Fossa, A.A., 2003. The impact of drug-induced QT interval
prolongation on drug discovery and development. Nat. Rev. Drug Discov. 2,
439-447.
Finlayson, K, Turnbull, L., January, C.T., Sharkey, J., Kelly, J.S., 2001. [3H]
dofetilide binding to HERG transfected membranes: a potential high
throughput preclinical screen. Eur. J. Pharmacol. 430, 147-148.
Fitzgerald, L.W, Burn, T.C., Brown, B.S., Patterson, J.P., Corjay, M.H.,
Valentine, P.A., Sun, J.H., Link, J.R., Abbaszade, I., Hollis, J.M., Largent,
B.L., Hartig, P.R., Hollis, G.F., Meunier, P.C., Robichaud, A.J., Robertson,
D.W., 2000. Possible role of valvular serotonin 5-HT(2B) receptors in the
cardiopathy associated with fenfluramine. Mol. Pharmacol. 57, 75-81.
Fliri, A.F., Loging, W.T., Thadeio, P.P., Volkmann, R.A., 2005a. Biological
spectra analysis: linking biological activity profiles to molecular structure.
Proc. Natl. Acad. Sci. U. S. A. 102, 261-266.
Fliri, A.F., Loging, W.T., Thadeio, P.P., Volkmann, R.A., 2005b. Analysis of
drug-induced effect patterns to link structure and side effects of medicines.
Nat. Chem. Biol. 1, 389-397.
Fluckiger-Isler, S., Baumeister, M., Braun, K., Gervais, V, Hasler-Nguyen, N.,
Reimann, R., Van Gompel, J., Wunderlich, H.G., Engelhardt, G., 2004.
Assessment of the performance of the Ames II assay: a collaborative study
with 19 coded compounds. Mutat. Res. 558, 181-197.
Frame, L.T., Ozawa, S., Nowell, S.A., Chou, H.C., DeLongchamp, R.R.,
Doerge, D.R., Lang, N.P., Kadlubar, F.F., 2000. A simple colorimetric assay
for phenotyping the major human thermostable phenol sulfotransferase
(SULT1A1) using platelet cytosols. Drug Metab. Dispos. 28, 1063-1068.
Gasparri, F., Mariani, M., Sola, F., Galvani, A., 2004. Quantification of the
proliferation index of human dermal fibroblast cultures with the Array Scan
high-content screening reader. J. Biomol. Screen. 9, 232-2343.
Gasparri, F., Cappella, P., Galvani, A., 2006. Multiparametric cell cycle analysis
by automated microscopy. J. Biomol. Screen. 11, 586-598.
Gee, P., Maron, D.M., Ames, B.N., 1994. Detection and classification of
mutagens: a set of base-specific Salmonella tester strains. Proc. Natl. Acad.
Sci. U. S. A. 91, 11606-11610.
Giaever, G., Chu, A.M., Ni, L., Connelly, C., Riles, L., Veronneau, S., Dow, S.,
Lucau-Danila, A., Anderson, K., Andre, B., Arkin, A.P., Astromoff, A., El-
Bakkoury, M., Bangham, R., Benito, R., Brachat, S., Campanaro, S.,
Curtiss, M., Davis, K., Deutschbauer, A., Entian, K.D., Flaherty, P., Foury,
F., Garfmkel, D.J., Gerstein, M., Gotte, D., Guldener, U., Hegemann, J.H.,
Hempel, S., Herman, Z., Jaramillo, D.F., Kelly, D.E., Kelly, S.L., Kotter, P.,
LaBonte, D., Lamb, D.C., Lan, N., Liang, H., Liao, H., Liu, L., Luo, C.,
Lussier, M., Mao, R., Menard, P., Ooi, S.L., Revuelta, J.L., Roberts, C.J.,
Rose, M., Ross-Macdonald, P., Scherens, B., Schimmack, G., Shafer, B.,
Shoemaker, D.D., Sookhai-Mahadeo, S., Storms, R.K., Strathern, J.N.,
Previous
TOC
-------
K.A. Houck, R.J. Kavlock / Toxicology and Applied Pharmacology 277 (2008) 163-178
111
Valle, G., Voet, M., Volckaert, G., Wang, C.Y., Ward, T.R., WiLhelmy, J.,
Winzeler, E.A., Yang, Y., Yen, G., Youngman, E., Yu, K., Bussey, H., Boeke,
J.D., Snyder, M., Philippsen, P., Davis, R.W., Johnston, M., 2002.
Functional profiling of the Sacchammyces cerevisiae genome. Nature 418,
387-391.
Giddings, S.J., Clarke, S.E., Gibson, G.G., 1997. CYP4A1 gene transfection
studies and the peroxisome proliferator-activated receptor: development of a
high-throughput assay to detect peroxisome proliferators. Eur. J. Drug
Metab. Pharmacokinet. 22, 315-319.
Giuliano, K.A., DeBiasio, R.L., Dunlay, R.T., Gough, A., Volosky, J.M., Zock,
J., Pavlakis, G.N., Taylor, D.L., 1997. High-content screening: a new
approach to easing key bottlenecks in the drug discovery process. J. Biomol.
Screen. 2, 249-259.
Glover, C.J., Rabow, A.A., Isgor, Y.G., Shoemaker, R.H., Covell, D.G., 2007.
Data mining of NCI's anticancer screening database reveals mitochondrial
complex I inhibitors cytotoxic to leukemia cell lines. Biochem. Pharmacol.
73,331-340.
Guengerich, P.P., 1999. Cytochrome P-450 3A4: regulation and role in drug
metabolism. Annu. Rev. Pharmacol. Toxicol. 39, 1-17.
Handschin, C., Meyer, U.A., 2003. Induction of drug metabolism: the role of
nuclear receptors. Pharmacol. Rev. 55, 649-673.
Haendel, M.A., Tilton, R, Bailey, G.S., Tanguay, R.L., 2004. Developmental
toxicity of the dithiocarbamate pesticide sodium metam in zebrafish.
Toxicol. Sci. 81, 390-400.
Hastwell, P.W, Chai, L.L., Roberts, K.J., Webster, T.W, Harvey, J.S., Rees, R.W,
Walmsley, R.M., 2006. High-specificity and high-sensitivity genotoxicity
assessment in a human cell line: validation of the GreenScreen HC GADD45a-
GFP genotoxicity assay. Mutat. Res. 607, 160-175.
Houck, K.A., Borchert, K.M., Hepler, C.D., Thomas, J.S., Bramlett, K.S.,
Michael, L.F., Burris, T.P., 2004. T0901317 is a dual LXR/FXR agonist. Mol.
Genet. Metab. 83, 184-187.
Hutchinson, T.H., Brown, R., Brugger, K.E., Campbell, P.M., Holt, M., Lange, R.,
McCahon, P., Tattersfield, L J., van Egmond, R., 2000. Ecological risk assess-
ment of endocrine disrupters. Environ. Health Perspect. 108, 1007-1014.
Kafert-Kasting, S., Alexandrova, K., Barthold, M., Laube, B., Friedrich, G.,
Arseniev, L., Hengstler, J.G., 2006. Enzyme induction in cryopreserved
human hepatocyte cultures. Toxicology 220, 117-125.
Kamath, R.S., Ahringer, J., 2003. Genome-wide RNAi screening in Caenor-
habditis elegans. Methods 30, 313-321.
Kavlock, R.J., Daston, G.P., DeRosa, C., Fenner-Crisp, P., Gray, L.E., Kaattari,
S., Lucier, G., Luster, M., Mac, M.J., Maczka, C., Miller, R., Moore, J.,
Rolland, R., Scott, G., Sheehan, D.M., Sinks, T, Tilson, H.A., 1996.
Research needs for the risk assessment of health and environmental effects
of endocrine disrupters: a report of the U.S. EPA-sponsored workshop.
Environ. Health Perspect. 104 (Suppl4), 715-740.
Kirkland, D., Aardema, M., Henderson, L., Muller, L., 2005. Evaluation of the
ability of a battery of three in vitro genotoxicity tests to discriminate rodent
carcinogens and non-carcinogens I. Sensitivity, specificity and relative
predictivity. Mutat. Res. 584, 1-256.
Kirstein, S.L., Atienza, J.M., Xi, B., Zhu, J., Yu, N., Wang, X., Xu, X., Abassi,
Y.A., 2006. Live cell quality control and utility of real-time cell electronic
sensing for assay development. Assay Drug Dev. Technol. 4, 545-553.
Lang, P., Yeow, K., Nichols, A., Scheer, A., 2006. Cellular imaging in drug
discovery. Nat. Rev. Drug Discov. 5, 343-356.
Lau, YY, Sapidou, E., Cui, X., White, R.E., Cheng, K.C., 2002. Development of
a novel in vitro model to predict hepatic clearance using fresh, cryopreserved,
and sandwich-cultured hepatocytes. Drug Metab. Dispos. 30, 1446-1454.
Laws, S.C., Yavanhxay, S., Cooper, R.L., Eldridge, J.C., 2006. Nature of the
binding interaction for 50 structurally diverse chemicals with rat estrogen
receptors. Toxicol. Sci. 94, 46-56.
Lee, M.Y, Park, C.B., Dordick, IS., Clark, D.S., 2005. Metabolizing enzyme
toxicology assay chip (MetaChip) for high-throughput microscale toxicity
analyses. Proc. Natl. Acad. Sci. U. S. A. 102, 983-987.
Lee, P.J., Hung, P.J., Rao, V.M., Lee, L.P., 2006a. Nanoliter scale
microbioreactor array for quantitative cell biology. Biotechnol. Bioeng.
94, 5-14.
Lee, F.Y, Lee, H., Hubbert, M.L., Edwards, P.A., Zhang, Y, 2006b. FXR, a
multipurpose nuclear receptor. Trends Biochem. Sci. 31, 572-580.
Lehmann, J.M., McKee, D.D., Watson, M.A., Willson, T.M., Moore, J.T.,
Kliewer, S.A., 1998. The human orphan nuclear receptor PXR is activated
by compounds that regulate CYP3A4 gene expression and cause drug
interactions. J. Clin. Invest. 102, 1016-1023.
Li, A.P., Bode, C., Sakai, Y, 2004. A novel in vitro system, the integrated
discrete multiple organ cell culture (IdMOC) system, for the evaluation of
human drug toxicity: comparative cytotoxicity of tamoxifen towards normal
human cells from five major organs and MCF-7 adenocarcinoma breast
cancer cells. Chem. Biol. Interact. 150, 129-136.
Liao, V.H., Yu, C.W, 2005. Caenorhabditis elegans gcs-1 confers resistance to
arsenic-induced oxidative stress. BioMetals 18, 519-528.
Lin, J.H., 2006. CYP induction-mediated drug interactions: in vitro assessment
and clinical implications. Pharm. Res. 23, 1089-1116.
Lindblom, T, Pierce, G., Sluder, A., 200 I.AC, elegans orphan nuclear receptor
contributes to xenobiotic resistance. Curr. Biol. 11, 864-868.
Liu, J., Knappenberger, K.S., Kack, H., Andersson, G., Nilsson, E., Dartsch, C.,
Scott, C.W., 2003. A homogeneous in vitro functional assay for estrogen
receptors: coactivator recruitment. Mol. Endocrinol. 17, 346-355.
Luo, G., Cunningham, M., Kim, S., Burn, T, Lin, J., Sinz, M., Hamilton, G.,
Rizzo, C., Jolley, S., Gilbert, D., Downey, A., Mudra, D., Graham, R.,
Carroll, K., Xie, J., Madan, A., Parkinson, A., Christ, D., Selling, B.,
LeCluyse, E., Gan, L.S., 2002. CYP3A4 induction by drugs: correlation
between a pregnane X receptor reporter gene assay and CYP3A4 expression
in human hepatocytes. Drug Metab. Dispos. 30, 795-804.
Luo, G., Guenthner, T, Gan, L.S., Humphreys, W.G., 2004. CYP3A4 induction
by xenobiotics: biochemistry, experimental methods and impact on drug
discovery and development. Curr. Drug Metab. 5, 483-505.
MacDonald, M.L., Lamerdin, I, Owens, S., Keon, B.H., Bilter, O.K., Shang, Z.,
Huang, Z., Yu, H., Dias, J., Minami, T., Michnick, S.W, Westwick, J.K.,
2006. Identifying off-target effects and hidden phenotypes of drugs in
human cells. Nat. Chem. Biol. 2, 329-337.
Mao, B., Gozalbes, R., Barbosa, F., Migeon, J., Merrick, S., Kamm, K., Wong, E.,
Costales, C., Shi, W, Wu, C., Froloff, N., 2006. QSAR modeling of in vitro
inhibition of cytochrome P450 3A4. J. Chem. Inf. Model 46, 2125-2134.
Mayer, T., Jagla, B., Wyler, M.R., Kelly, P.O., Aulner, N., Beard, M., Barger, G.,
Tobben, U., Smith, D.H., Branden, L., Rothman, J.E., 2006. Cell-based
assays using primary endothelial cells to study multiple steps in
inflammation. Methods Enzymol. 414, 266-283.
Miller, J.E., Vlasakova, K., Glaab, WE., Skopek, T.R., 2005. A low volume,
high-throughput forward mutation assay in Salmonella typhimurium based
on fiuorouracil resistance. Mutat. Res. 578, 210-224.
Mizell, M., Romig, E.S., 1997. The aquatic vertebrate embryo as a sentinel for
toxins: zebrafish embryo dechorionation and perivitelline space microinjec-
tion. Int. J. Dev. Biol. 41, 411-423.
Moore, J.T., Kliewer, S.A., 2000. Use of the nuclear receptor PXR to predict
drug interactions. Toxicology 153, 1-10.
Murphy, S.M., Palmer, M., Poole, M.F., Padegimas, L., Hunady, K., Danzig, J.,
Gill, S., Gill, R., Ting, A., Sherf, B., Brunden, K., Stricker-Krongrad, A.,
2005. Evaluation of functional and binding assays in cells expressing either
recombinant or endogenous hERG channel. J. Pharmacol. Toxicol. Methods
54, 42-55.
Nasevicius, A., Ekker, S.C., 2001. The zebrafish as a novel system for functional
genomics and therapeutic development applications. Curr. Opin. Mol. Ther.
3, 224-228.
National Research Council, Committee on Toxicity Testing and Assessment of
Environmental Agents, 2006. Toxicity Testing for Assessment of Environ-
mental Agents: Interim Report. The National Academies Press.
National Research Council, 2007. Toxicity Testing in the Twenty-first Century:
A Vision and Strategy. Prepublication edition available at: http://dels.nas.
edu/dels/reportDetail.php?Flink_id=4286&session_id=.
Nichols, J.S., Parks, D.J., Consler, T.G., Blanchard, S.G., 1998. Development of
a scintillation proximity assay for peroxisome proliferator-activated receptor
gamma ligand binding domain. Anal. Biochem. 257, 112-119.
O'brien, P.J., Irwin, W, Diaz, D., Howard-Cofield, E., Krejsa, C.M., Slaughter,
M.R., Gao, B., Kaludercic, N., Angeline, A., Bernardi, P., Brain, P.,
Hougham, C., 2006. High concordance of drug-induced human hepatotoxi-
city with in vitro cytotoxicity measured in a novel cell-based model using
high content screening. Arch. Toxicol. 80 (9), 580-604 (Sep).
Previous
TOC
-------
178
K.A. Houck, R.J. Kavlock / Toxicology and Applied Pharmacology 277 (2008) 163-178
Ozers, M.S., Ervin, K.M., Steffen, C.L., Fronczak, J.A., Lebakken, C.S.,
Carnahan, K.A., Lowery, R.G., Burke, T.J., 2005. Analysis of ligand-
dependent recruitment of coactivator peptides to estrogen receptor using
fluorescence polarization. Mol. Endocrinol. 19, 25-34.
Paolini, G.V., Shapland, R.H., van Hoorn, W.P., Mason, J.S., Hopkins, A.L.,
2006. Global mapping of pharmacological space. Nat. Biotechnol. 24,
805-815.
Parng, C., Roy, N.M., Ton, C., Lin, Y., McGrath, P., 2007. Neurotoxicity
assessment using zebrafish. J. Pharmacol. Toxicol. Methods 55, 103-112.
Parsons, A.B., Lopez, A., Givoni, I.E., Williams, D.E., Gray, C.A., Porter, J.,
Chua, G., Sopko, R., Brost, R.L., Ho, C.H., Wang, J., Ketela, T., Brenner, C.,
Brill, J.A., Fernandez, G.E., Lorenz, T.C., Payne, G.S., Ishihara, S., Ohya,
Y., Andrews, B., Hughes, T.R., Frey, B J., Graham, T.R., Andersen, R.J.,
Boone, C., 2006. Exploring the mode-of-action of bioactive compounds by
chemical-genetic profiling in yeast. Cell 126, 611-625.
Pulak, R., 2006. Techniques for analysis, sorting, and dispensing of C. elegans
on the COPAS flow-sorting system. Methods Mol. Biol. 351, 275-286.
Phillips, G.W., Irwin, W.A., Howard-Cofield, E.J., Randle, L.E., Abraham, V.C.,
Haskins, J.R., O'Brien, P.J., 2005. Incorporation of an oxidative stress
biomarker into high content screening (HCS) for human toxicity potential.
llth Annual Meeting of the Society for Biomol. Screen. (SBS), P05020.
Rausch, O., 2006. High content cellular screening. Curr. Opin. Chem. Biol. 10,
316-320.
Richards, G.R., Smith, A.J., Parry, F., Platts, A., Chan, O.K., Leveridge, M.,
Kerby, J.E., Simpson, P.B., 2006. A morphology- and kinetics-based
cascade for human neural cell high content screening. Assay Drug Dev.
Technol. 4, 143-152.
Roberts, R.A., Ganey, P.E., Ju, C., Kamendulis, L.M., Rusyn, I, Klaunig, J.E.,
2007. Role of the kupffer cell in mediating hepatic toxicity and carcinogenesis.
Toxicol. Sci. 96,2-15.
Rouleau, N., Turcotte, S., Mondou, M.H., Roby, P., Bosse, R., 2003.
Development of a versatile platform for nuclear receptor screening using
AlphaScreen. J. Biomol. Screen. 8, 191-197.
Roymans, D., Annaert, P., Van Houdt, J., Weygers, A., Noukens, J., Sensenhauser,
C., Silva, J., Van Looveren, C., Hendrickx, J., Mannens, G., Meuldermans, W,
2005. Expression and induction potential of cytochromes P450 in human
cryopreserved hepatocytes. Drug Metab. Dispos. 33, 1004-1016.
Sanguinetti, M.C., Tristani-Firouzi, M., 2006. hERG potassium channels and
cardiac arrhythmia. Nature 440, 463-469.
Sartipy, P., Bjorquist, P., Strehl, R., Hyllner, J., 2006. Pluripotent human stem
cells as novel tools in drug discovery and toxicity testing. IDrugs 9,702-705.
Schade, R., Andersohn, F., Suissa, S., Haverkamp, W, Garbe, E., 2007. Dopamine
agonists and the risk of cardiac-valve regurgitation. N. Engl. J. Med. 356,
29-38.
Shibata, Y, Takahashi, H., Chiba, M., Ishii, Y, 2002. Prediction of hepatic
clearance and availability by cryopreserved human hepatocytes: an
application of serum incubation method. Drug Metab. Dispos. 30, 892-896.
Shimada, T., 2006. Xenobiotic-metabolizing enzymes involved in activation and
detoxification of carcinogenic polycyclic aromatic hydrocarbons. Drug
Metab. Pharmacokinet. 21, 257-276.
Shoemaker, R.H., Scudiero, D.A., Melillo, G., Currens, M.J., Monks, A.P.,
Rabow, A.A., Covell, D.G., Sausville, E.A., 2002. Application of high-
throughput, molecular-targeted screening to anticancer drug discovery. Curr.
Top. Med. Chem. 2, 229-246.
Solly, K., Wang, X., Xu, X., Strulovici, B., Zheng, W, 2004. Application of real-
time cell electronic sensing (RT-CES) technology to cell-based assays.
Assay Drug Dev. Technol. 2, 363-372.
Sorota, S., Zhang, X.S., Margulis, M., Tucker, K., Priestley, T., 2005.
Characterization of a hERG screen using the Ion Works HT: comparison to
a hERG rubidium efflux screen. Assay Drug. Dev. Technol. 3, 47-57.
Trubetskoy, O.V, Shaw, P.M., 1999. A fluorescent assay amenable to measuring
production of D-glucuronides produced from recombinant UDP-glycosyl
transferase enzymes. Drug Metab. Dispos. 27, 555-557.
Trubetskoy, O.V, Gibson, J.R., Marks, B.D., 2005. Highly miniaturized formats
for in vitro drug metabolism assays using vivid fluorescent substrates and
recombinant human cytochrome P450 enzymes. J. Biomol. Screen. 10,
56-66.
Van der Jagt, K., Munn, S., Torslov, J., de Bruijn, J., 2004. Alternative
approaches can reduce the use of test animals under REACH. Report to
the Directorate General JRC, Institute for Health and Consumer
Protection, European Commission, (http://ihcp.jrc.ec.europa.eu/docs/ecb/
Reducing_the_use_of_test_animals_under_REACH_IHCP_report.pdf).
Viravaidya, K., Sin, A., Shuler, M.L., 2004. Development of a microscale cell
culture analog to probe naphthalene toxicity. Biotechnol. Prog. 20,316-323.
Walsky, R.L., Obach, R.S., 2004. Validated assays for human cytochrome P450
activities. Drug Metab. Dispos. 32, 647-660.
Walum, E., Hedander, J., Garberg, P., 2005. Research perspectives for pre-
screening alternatives to animal experimentation On the relevance of
cytotoxicity measurements, barrier passage determinations and high
throughput screening in vitro to select potentially hazardous compounds
in large sets of chemicals. Toxicol. Appl. Pharmacol. 207 (2 Suppl),
393-397.
Waring, R.H., Harris, R.M., 2005. Endocrine disrupters: a human risk? Mol.
Cell. Endocrinol. 244, 2-9.
Weir, S.W, Weston, A.H., 1986. The effects of BRL 34915 and nicorandil on
electrical and mechanical activity and on 86Rb efflux in rat blood vessels.
Br. J. Pharmacol. 88, 121-128.
Whitebread, S., Hamon, J., Bojanic, D., Urban, L., 2005. Keynote review: in
vitro safety pharmacology profiling: an essential tool for successful drug
development. Drug Discov. Today 10, 1421-1433.
Xing, J.Z., Zhu, L., Jackson, J.A., Gabos, S., Sun, X.J., Wang, X.B., Xu, X.,
2005. Dynamic monitoring of cytotoxicity on microelectronic sensors.
Chem. Res. Toxicol. 18, 154-161.
Xing, J.Z., Zhu, L., Gabos, S., Xie, L., 2006. Microelectronic cell sensor assay
for detection of cytotoxicity and prediction of acute toxicity. Toxicol.
In Vitro 20, 995-1004.
Yoshitomi, S., Ikemoto, K., Takahashi, J., Miki, H., Namba, M., Asahi, S., 2001.
Establishment of the transformants expressing human cytochrome P450
subtypes in HepG2, and their applications on drug metabolism and
toxicology. Toxicol. In Vitro 15, 245-256.
Yueh, M.F., Kawahara, M., Raucy, J., 2005. Cell-based high-throughput
bioassays to assess induction and inhibition of CYP1A enzymes. Toxicol.
In Vitro 19, 275-287.
Zhang, J.H., Chung, T.D., Oldenburg, K.R., 1999. A simple statistical parameter
for use in evaluation and validation of high throughput screening assays.
J. Biomol. Screen. 4, 67-73.
Zhang, C., Willett, C., Fremgen, T., 2003. Zebrafish: an animal model for
toxicological studies. Curr. Protoc. Toxicol., Suppl. 17, 1-18.
Previous
TOC
-------
Research
Use of Cell Viability Assay Data Improves the Prediction Accuracy
of Conventional Quantitative Structure-Activity Relationship Models
of Animal Carcinogenicity
Hao Zhu,1-2 Ivan Rusyn,1-3 Ann Richard,4 and Alexander Tropsha1-2
1Carolina Environmental Bioinformatics Research Center, laboratory for Molecular Modeling, Division of Medicinal Chemistry and
Natural Products, School of Pharmacy, and 3Department of Environmental Sciences and Engineering, School of Public Health,
University of North Carolina at Chapel Hill, Chapel Hill, North Carolina USA; 4National Center for Computational Toxicology, Office of
Research and Development, U.S. Environmental Protection Agency, Research Triangle Park, North Carolina, USA
BACKGROUND: To develop efficient approaches for rapid evaluation of chemical toxicity and human
health risk of environmental compounds, the National Toxicology Program (NTP) in collaboration
with the National Center for Chemical Genomics has initiated a project on high-throughput screen-
ing (HTS) of environmental chemicals. The first HTS results for a set of 1,408 compounds tested for
their effects on cell viability in six different cell lines have recently become available via PubChem.
OBJECTIVES: We have explored these data in terms of their utility for predicting adverse health
effects of the environmental agents.
METHODS AND RESULTS: Initially, the classification k nearest neighbor (£NN) quantitative struc-
ture—activity relationship (QSAR) modeling method was applied to the HTS data only, for a
curated data set of 384 compounds. The resulting models had prediction accuracies for training,
test (containing 275 compounds together), and external validation (109 compounds) sets as high as
89%, 71%, and 74%, respectively. We then asked if HTS results could be of value in predicting
rodent Carcinogenicity. We identified 383 compounds for which data were available from both the
Berkeley Carcinogenic Potency Database and NTP—HTS studies. \C^e found that compounds clas-
sified by HTS as "actives" in at least one cell line were likely to be rodent carcinogens (sensitivity
77%); however, HTS "inactives" were far less informative (specificity 46%). Using chemical
descriptors only, £NN QSAR modeling resulted in 62.3% prediction accuracy for rodent Carcino-
genicity applied to this data set. Importantly, the prediction accuracy of the model was significantly
improved (72.7%) when chemical descriptors were augmented by HTS data, which were regarded
as biological descriptors.
CONCLUSIONS: Our studies suggest that combining NTP—HTS profiles with conventional chemical
descriptors could considerably improve the predictive power of computational approaches in
toxicology.
KEYWORDS: carcinogenesis, computational toxicology, high-throughput screening, QSAR. Environ
Health Persfect 116:506-513 (2008). doi:10.1289/ehp.!0573 available via http://dx.doi.org/
[Online 4 January 2008]
The traditional approaches for in vivo animal
chemical safety testing are costly, time consum-
ing, and have a low throughput (Bucher and
Portier 2004). To improve the efficiency of
assessing potential human health hazards of
environmental chemicals, the National
Toxicology Program (NTP) at the National
Institute of Environmental Health Sciences
(NIEHS) recently initiated the High Through-
put Screening (HTS) project (NTP 2007;
Inglese et al. 2006; Xia et al. 2007). The NTP-
HTS effort aims to develop high-throughput
biological assays that aid in predicting a
chemical's potential for in vivo toxicity in a
manner that is both informative of mecha-
nisms and pathways and relevant to human
health risk assessment. These assays are
expected to help in prioritizing compounds
for targeted animal testing. Recently, a set of
1,408 chemical agents, many with known
in vivo toxicity profiles, was screened in six
human cell lines for cytotoxicity and other phe-
notypic end points. The HTS results, including
complete dose-response data for all tested com-
pounds, were made publicly available through
PubChem [National Center for Biotechnology
Information (NCBI) 2007]. These data can be
explored in terms of assessing the relevance of
HTS screening to predictive toxicology.
Accurate prediction of the adverse effects of
chemical substances on living systems, identifi-
cation of possible toxic alerts, and compound
prioritization for animal testing are the primary
goals of computational toxicology. Rapid
expansion of experimental data sets that com-
bine data on chemical structure and various
toxicity end points for numerous environmen-
tal agents {e.g., NTP [NTP 2007]; Berkeley
Carcinogenic Potency Database [CPDB 2007];
and Distributed Structure-Searchable Toxicity
database [DSSTox; U.S. Environmental
Protection Agency (U.S. EPA) 2007]} provides
novel opportunities to explore the relationships
between chemical structure and toxicity using
cheminformatics approaches. Application of
advanced cheminformatics tools, such as quan-
titative structure—activity relationship (QSAR)
methods, to the analysis of these data may pro-
vide means for accurate prediction of chemical
toxicity of untested compounds, allowing for
prioritization of compounds for subsequent
animal testing.
QSAR modeling aims to establish rigorous
correlations between the chemical descriptors
of a set of compounds and their experimentally
studied biological activities. Many different
QSAR approaches have been developed over
nearly 50 years of research (Beresford et al.
2004; Dearden 2003; Johnson et al. 2004;
Schultz et al. 2003a). Recent trends in the field
have focused on model validation as the key
part of model development to ensure signifi-
cant external predictive power of QSAR mod-
els. Traditional QSAR models are developed
based on chemical descriptors alone (Klopman
et al. 2004; Richard 2006). In some cases, addi-
tional physicochemical properties, such as water
partition coefficient (logP) (Klopman et al.
2003), water solubility (Stoner et al. 2004), and
melting point (Mayer and Reichenberg 2006)
were used successfully to augment computed
chemical descriptors and improve the predic-
tive power of QSAR models. These studies sug-
gest that using hybrid descriptor sets in QSAR
modeling could prove beneficial.
The availability of HTS data on large sets
of chemical agents offers an attractive avenue
for exploring its utility in hybrid descriptor-
based QSAR modeling. In this respect, the
NTP-HTS data represent attractive and poten-
tially mechanistically relevant in vitro "biologi-
cal" descriptors for modeling the adverse health
effects in vivo. Our study tested a hypothesis
that improved QSAR predictions can be
Address correspondence to A. Tropsha, Campus Box
7360, 327 Beard Hall, University of North
Carolina, Chapel Hill, NC 27599-7360 USA.
Telephone: (919) 966-2955. Fax: (919) 966-0204.
E-mail: alex_tropsha@unc.edu
Supplemental Material is available online at http://
www.ehponline.org/members/2008/10573/suppl.pdf
We thank R. Tice (National Institute of Environ-
mental Health Sciences) for valuable comments.
This work was supported, in part, by grants from
the National Institutes of Health (GM076059 and
ES005948) and the U.S. EPA (RD832720).
This manuscript was approved for publication by the
U.S. EPA National Center for Computational
Toxicology. However, the content does not necessarily
reflect the views and policies of the U.S. EPA and men-
tion of trade names or commercial products does not
constitute endorsement or recommendation for use.
The authors declare they have no competing
financial interests.
Received 19 June 2007; accepted 3 January 2008.
506
VOLUME 1161 NUMBER 41 April 2008 • Environmental Health Perspectives
Previous
TOC
-------
Biological descriptors in QSAR modeling of carcinogenicity
developed using a combination of chemical and
biological descriptors of environmental chemi-
cals. To this end, we have developed QSAR
models based on NTP-HTS data using the k
nearest neighbor (knn) approach. Initially, we
modeled the NTP—HTS results separately to
explore the inherent relationship between
chemical structure and its effect on cell viability.
Next, we evaluated if a correlation exists
between the NTP-HTS assay results and their
in vivo rodent carcinogenic potency, as reported
in the CPDB. Subsequently, the HTS results
were used as biological descriptors that were
combined with chemical descriptors to develop
£NN QSAR models for predicting rodent car-
cinogenicity summary calls of the compounds.
Finally, we attempted to examine the relative
significance of the HTS "descriptors" in the
resulting models and their interplay with chem-
ical descriptors. Our studies demonstrate that
adding NTP—HTS data to chemical descriptors
employed in conventional QSAR modeling
affords improved models that may advance the
use of computational approaches in toxicology.
Our current studies were limited to explor-
ing the value of cell viability assays in predict-
ing rodent carcinogenicity as one example of
in vivo toxicity end point. This limitation was
because in vivo rodent carcinogenicity is the
only end point reported in the CPDB for a sig-
nificant fraction of compounds also tested for
their effect on cell viability. Certainly, as addi-
tional chemicals with known in vivo responses
are tested in cell-based assays, we will continue
to explore similar approaches in correlating the
in vitro and in vivo data.
Methods
Data sources. NTP-HTS data set. The
NTP-HTS assay results were obtained from
PubChem (NCBI 2007), and chemical struc-
tures associated with these results were pro-
vided by the DSSTox (U.S. EPA 2007)
database. The complete data set included
1,408 compounds that were tested in six cell
lines at the National Institutes of Health
(NIH) Chemical Center Genomics (NCGC)
(Inglese et al. 2006; Xia et al. 2007). The cell
lines used for screening of the effect of chemi-
cal agents on cell viability included BJ [human
foreskin fibroblast; PubChem bioassay identi-
fier (AID) no. 421], HEK293 (transformed
human embryonic kidney cell line; AID no.
427), HepG2 (human hepatoma; AID no.
433), Jurkat (clone E6-1, human acute T-cell
leukemia; AID no. 426), MRC-5 (human
lung fibroblast; AID no. 434), and SK-N-SH
(human neuroblastoma; AID no. 435).
Details on the assays and the testing protocols
can be found in PubChem. For the purposes
of this work, the data set was curated as fol-
lows. First, we removed duplicate data entries
for 55 chemical records with identical chemi-
cal structures (i.e., keeping one of the two
identical records) and 14 records for which
molecular structure could not be obtained.
Second, inorganic and organometallic com-
pounds as well as compound mixtures were
excluded since these do not have conventional
chemical descriptors used in QSAR studies.
The curated subset of the original NTP-HTS
data set used in this work included 1,289
unique organic compounds [Supplemental
Material, Table 1 (online at http://www.
ehponline.org/members/2008/10573/suppl.
pelf)]. The "activity" classification for each
compound, for each HTS assay, was assigned
by NCGC as reported in PubChem. HTS
studies included the 55 duplicate compounds.
The analysis of assay results for these duplicate
compounds demonstrated that the HTS data
were highly reproducible (Xia et al. 2007).
The CPDB database. We obtained the
rodent carcinogenicity data from the CPDB
(CPDB 2007; Gold et al. 1991). The CPDB
provides a systematic and unifying source of
the outcomes from in vivo animal chemical
carcinogenicity studies. The most recent release
of the CPDB includes experimental data on
testing of 1,481 diverse chemicals in one or
both sexes of rats and mice, reporting out-
comes on 35 possible target organ/tissue sites.
A chemical structure—annotated version of the
CPDB summary tables consolidating all
species was published on the U.S. EPA
DSSTox website (U.S. EPA 2007) with addi-
tional summary activity categorizations and
was used for the present study. For modeling
purposes, chemical agents in the CPDB were
categorized as follows: active (multisite, multi-
sex, or multispecies carcinogens), marginally
active (single-site carcinogens), inactive (non-
carcinogenic in more than two test cells and no
active results), or no conclusion (insufficient
results). Of the 1,466 compounds classified as
"active" or "inactive" in the CPDB, 314 were
represented in the NTP-HTS data set (178
active and 136 inactive) and used in this study.
A complete list of these agents is provided in
the Supplemental Material, Table 2 (online at
http://www.ehponline.org/members/2008/
10573/suppl.pdf).
MolConnZ chemical descriptors. The
MolConnZ software (eduSoft LC, Ashland,
VA, USA) affords computation of a wide range
of topologic indices of molecular structure.
These indices include but are not limited to
the following descriptors: simple and valence
path, cluster, path/cluster and chain molecular
connectivity indices, kappa molecular shape
indices, topologic and electrotopologic state
indices, differential connectivity indices, graph
radius and diameter, Wiener and Platt indices,
Shannon and Bonchev—Trinajstic information
indices, counts of different vertices, counts of
paths, and edges between different kinds of
vertices (Hall et al. 1991; Kier 1986, 1987;
Kier and Hall 1991). Overall, MolConnZ
produces over 400 different descriptors. Those
with zero value or zero variance were removed.
The remaining descriptors were range scaled,
as the absolute scales for MolConnZ descrip-
tors can differ by orders of magnitude.
Accordingly, our use of range scaling avoided
giving descriptors with significantly higher
ranges a disproportional weight on distance
calculations in multidimensional MolConnZ
descriptor space.
QSAR modeling. Selection of test and
training sets. The curated NTP-HTS data set
(consisting of 1,289 unique organic com-
pounds) was subdivided into multiple training/
test set pairs using the sphere exclusion pro-
gram developed in our laboratory (Golbraikh
et al. 2003). The number of compounds
included in the test set was gradually increased
to obtain the largest possible test set for which
accurate predictions could be obtained from
models developed for the corresponding small-
est possible training set.
The procedure implemented in the present
study begins with the calculation of the distance
matrix D between points that represent com-
pounds in the descriptor space. Let -Dm;n and
-Dmax be the minimum and maximum elements
of D, respectively. TV probe sphere radii, R, are
defined by the following formulas: R^m = R\ =
Anim -#max = RN = Anax/4, Rf = RI +
(*-l)*CfyrJ?i)/CW-l), where i= 2, ..., N-l.
Each probe sphere radius corresponds to one
division in the training and the test set. A
sphere-exclusion algorithm used in the present
study consisted of the following steps: (i) ran-
domly select a compound; (ii) include it in the
training set; (iii) construct a probe sphere
around this compound; (iv) select compounds
from this sphere and include them alternately
into the test and the training sets; (v) exclude all
compounds from within this sphere from fur-
ther consideration; and (vi) if no more com-
pounds are left, stop. Otherwise let m be the
number of probe spheres constructed and n be
the number of remaining compounds. Let d^
(i=l,...,m;j=l,...,n) be the distances between the
remaining compounds and the probe sphere
centers. Select a compound corresponding to
the lowest Rvalue and go to step (ii). This
algorithm guarantees that at least in the entire
descriptor space (i) representative points of the
test set are close to representative points of the
training set (test set compounds are within the
applicability domain defined by the training
set); (ii) most of the representative points of the
training set are close to representative points of
the test set; and (iii) the training set represents
the entire modeling set (i.e., there is no subset
in the modeling set that is not represented by a
similar compound in the training set)
(Golbraikh et al. 2003). Consequently, the
sphere exclusion algorithm could maximize the
diversity of the training/test sets in the descrip-
tor space used for modeling. Because of the
Environmental Health Perspectives • VOLUME 116 I NUMBER 4 I April 2008
507
Previous
TOC
-------
Zhu et al.
stochastic nature of the algorithm, the composi-
tion of training and test sets is different for dif-
ferent original data set divisions. For example,
we tested the results of more than 40 data set
divisions generated by the sphere exclusion and
found that any two training sets had no more
than 85% identical compounds.
The statistical significance of models was
characterized with the standard leave-one-out
cross-validated R2 (if) for the training sets and
the conventional R2 for the test sets. Models
were considered acceptable if both q and R
were larger than the arbitrary cutoff values
(0.65 was used as a cutoff in this study).
Models that did not meet these cutoff criteria
were discarded. Additional details of this
approach are described elsewhere (Golbraikh
et al. 2003; Golbraikh and Tropsha 2002b).
£NN QSAR method. The £NN QSAR
method employs the £NN pattern recognition
principle and a variable selection procedure.
Initially, a subset of nvar (number of selected
variables) descriptors is selected randomly. The
model developed with this set of descriptors is
validated by leave-one-out cross-validation,
where each compound is eliminated from the
training set, and its biological activity is pre-
dicted as the average activity of k most similar
molecules (k = 1 to 5). The weighted molecu-
lar similarity was characterized by the modi-
fied Euclidean distance between compounds
in the nvar subspace of the multidimensional
descriptor space. Generally, the Euclidean dis-
tances in the descriptor space between a com-
pound and each of its k nearest neighbors are
not the same. Thus, the activity of each of the
k neighbors of a compound was given a weight
that was higher for close neighbors and lower
for distant neighbors as follows (Equations 1
and 2):
[1]
exp(-<)'
y = 2 w,y,
where dj is the Euclidean distance between
the compound and its k nearest neighbors; Wj
is the weight for every individual nearest
neighbor; j/; is the actual activity value for
nearest neighbor i; and y is the predicted
activity value. A method of simulated anneal-
ing with the Metropolis-like acceptance crite-
ria is used to optimize the variable selection.
In summary, the £NN QSAR algorithm
generates both an optimum k value and an
optimal nvar subset of descriptors, that afford a
QSAR model with the highest training set
model accuracy as estimated by the q2 value.
Further details of the £NN method implemen-
tation, including the description of the simu-
lated annealing procedure used for stochastic
sampling of the descriptor space, are given in
our previous publications (Ng et al. 2004;
Shen et al. 2003; Zheng and Tropsha 2000).
Applicability domain of &NN QSAR
models. Formally, a QSAR model can predict
the target property for any compound for
which chemical descriptors can be calculated.
However, because all the models are developed
in £NN QSAR modeling by interpolating
activities of the nearest neighbor compounds
only in the relevant training sets, a special
applicability domain (i.e., similarity threshold)
should be introduced to avoid making predic-
tions for compounds that differ substantially
from the training set molecules. This proce-
dure resembles that for identifying chemical
outliers prior to the onset of modeling.
To measure similarity, each compound is
represented by a point in the Af-dimensional
descriptor space (where M is the total number
of descriptors in the descriptor pharma-
cophore) with the coordinates X^, Xi2, ..., XiM,
where Xj s are the values of individual descrip-
tors. The molecular similarity between any two
molecules is characterized by the Euclidean dis-
tance between their representative points. The
Euclidean distance d^ between two points i
and j (which correspond to compounds i and
j) in Af-dimensional space can be calculated as
follows (Equation 3):
[3]
Compounds with the smallest distance
between one another are considered to have
Table 1. Summary of the biological activity of chemical agents screened in NTP-HTS assays.
Classification
Actives
Inconclusives
Inactives
BJ
42
44
1,203
HEK293
63
79
1,147
HepG2
41
47
1,201
Jurkat
121
89
1,079
MRC-5
37
44
1,208
SK-N-SH
74
54
1,161
All tests
140
90
1,059
Table 2. Rodent carcinogenicity classification (CPDB database) for 314 NTP-HTS compounds.
Rats
Classification
Active
Inactive
Total
Male
121
150
271
Female
111
154
265
Mice
Male
123
153
276
Female
134
140
274
the highest similarity. The similarities of com-
pounds in our training set are compiled to
produce an applicability domain threshold,
Dj^ calculated as follows (Equation 4):
DT = V -
[4]
Here, y is the mean Euclidean distance to the
nearest neighbor of each compound within
the modeling set, O is the standard deviation
of these Euclidean distances, and Zls an arbi-
trary parameter to control the significance
level. On the basis of previous studies (Shen
et al. 2002), we set the default value of this
parameter to 0.5, which formally places the
boundary for which compounds will be pre-
dicted at one-half of the SD (assuming a
Boltzmann distance distribution between
£NN compounds in the training set). Thus, if
the distance of the external compound from
at least one of its k nearest neighbors in the
training set exceeds this threshold, the predic-
tion is considered unreliable.
Robustness of QSAR models. y-Randomi-
zation (randomization of response) is a widely
used approach to establish the model robust-
ness. It consists of rebuilding the models using
randomized activities of the modeling set and
subsequent assessment of the model statistics. It
is expected that models obtained for the model-
ing set with randomized activities should have
significantly lower predictivity for the external
validation set than the models built using the
modeling set with real activities, or the total
number of acceptable models based on the ran-
domized modeling set satisfying the same cutoff
criterion (q2 and R2> 0.65) is much less than
that based on the real modeling set. If this con-
dition is not satisfied, real models built for this
modeling set are not reliable and should be dis-
carded. This test was applied to all data divi-
sions considered in this study.
Results
Table 1 provides a summary of the classifica-
tion of the chemical agents used for these stud-
ies with respect to their "biological activity"
(i.e., the effect on cell viability) in each of the
six cell lines used for screening. In the entire
NTP-HTS data set of unique 1,289 com-
pounds, 140 were defined as "active" and 90 as
"inconclusive" based on one or more active or
inconclusive calls recorded in PubChem across
the six cell lines, respectively. The majority of
compounds—1,059—were recorded in
PubChem as "inactive" in all experiments.
Overall, the NTP-HTS data set contains
314 compounds that can be mapped to the
CPDB database and classified as carcinogenic
according to DSSTox "multisite, multisex, or
multispecies" summary designations (Table 2).
QSAR modeling of NTP-HTS data using
chemical descriptors, QSAR modeling of
the NTP-HTS data was desired to establish
508
VOLUME 116 I NUMBER 41 April 2008 • Environmental Health Perspectives
Previous
TOC
-------
Biological descriptors in QSAR modeling of carcinogenicity
predictive models of HTS assays that can be
used to impute such data for future compound
libraries that may be tested. In addition, our use
of the y-randomization test as part of modeling
procedures could be viewed as an independent
statistical test of the "nonrandomness" of the
HTS data. The curated NTP-HTS data set has
a biased distribution of active and inactive com-
pounds (16% actives and inconclusives vs. 84%
inactives). This is characteristic of most of the
available biological data sets (such as those
deposited in PubChem), which are dominated
by inactive compounds. To address this bias,
we used a (dis)similarity search to exclude a
considerable fraction of inactive compounds
from the data set to balance the active/inactive
ratio for modeling purposes. To this end, we
calculated the Molecular ACCess System
(MACCS) structural keys (Renner and
Schneider 2006) for all 1,289 compounds in
the data set, using the MOE software
(Chemical Computing Group, Montreal,
Canada). All the active compounds were used
as a probe subset, and the Tanimoto coeffi-
cients (Schultz et al. 2003b; Willett and
Winterman 1986) between each inactive com-
pound and the probe subset were calculated
based on their MACCS keys. The inactive
compound was selected into the modeling set
only if it had a relatively high Tanimoto simi-
larity (> 0.7) with one or more active/inconclu-
sive compounds. Using this approach, 244 of
the original 1,079 inactive compounds were
selected because of their relatively high similar-
ity to the active compounds. Thus, the final
data set for the classification QSAR modeling
included a total of 384 compounds (140 actives
and 244 inactives). The rationale for this
approach to selecting (a subset of) inactive com-
pounds for the classification modeling is that it
is more challenging to establish robust models
when the two classes of active and inactive
compounds include relatively similar molecules.
It is quite obvious that if the two classes of
compounds (i.e., active or inactive) are chemi-
cally dissimilar as judged by a simple similarity
metric such as Tanimoto coefficients, then no
additional statistical modeling using sophisti-
cated data mining techniques is necessary. We
did not include any compounds with incon-
clusive results in our modeling studies.
Because it is critical to demonstrate that
QSAR models have high prediction accuracy
for external validation data sets (Golbraikh and
Tropsha 2002a; Zhang et al. 2006), 109 com-
pounds (37 actives and 72 inactives) were ran-
domly selected for external model validation.
The remaining 275 compounds (103 actives
and 172 inactives) were used for modeling, and
multiple training and test sets were generated.
The variable selection £NN QSAR models were
developed for each training set, and the predic-
tive power of each model was assessed against
the corresponding test set. The acceptability
cutoff values of the leave-one-out cross-valida-
tion accuracy and the prediction accuracy for
the test set were set to 0.65 (Kovatcheva et al.
2004). Because the data set was unbalanced,
we used the average of sensitivity and speci-
ficity to represent the overall predictive power
of a model in this study. Therefore, the overall
predictive accuracy of each model was defined
as the average of the correctly predicted active
ratio (sensitivity) and the correctly predicted
inactive ratio (specificity) (de Lima et al.
2006). The total number of models that satis-
fied the accuracy threshold criteria was 599,
and the statistical characteristics of 15 most sig-
nificant £NN models are shown in Table 3.
Our previous studies have demonstrated
that the highest external prediction accuracy of
QSAR models is achieved using the consensus
approach, that is, by averaging the predictions
from individual models (Tropsha et al. 2003).
The consensus prediction results for 109 com-
pounds in the external validation set are
provided in Table 4. The sensitivity and speci-
ficity of the consensus prediction were 56.8%
and 90.2%, respectively. Thus, the overall pre-
dictive power was 73.5%, that is, similar to
that for the training/test sets (Table 3).
To ensure high external validation accuracy
of the training set models, we also considered
their applicability domains. This restriction
decreases the number of compounds consid-
ered for the prediction but increases the relia-
bility so that higher accuracy is typically
expected. Indeed, after removing compounds
outside the applicability domain of our train-
ing set models, the coverage of the external set
was reduced to 88%. However, the accuracy of
prediction for actives and inactives improved
to 65.4% and 92.9%, respectively (i.e., total
accuracy increased to - 80%).
It is interesting to see whether the £NN
HTS models could make reliable predictions of
the remaining 835 inactive compounds, which
were excluded because they were relatively dis-
similar to the compounds used in the model-
ing procedure. The consensus prediction gave
64.1% predictive accuracy for all 835 com-
pounds. After excluding 138 compounds out
of applicability domain, the coverage was
reduced to 83.5%, but the predictive accuracy
increased to 80.1%
The y-randomization test was performed as
well. For the modeling set with real HTS
results, there were 599 models that satisfied the
criterion ofif/K2 > 0.65 (Table 3), whereas for
the data set with randomized HTS results, only
5 models that had if/IP > 0.65 were generated.
These results indicate that our models are sta-
tistically robust.
The utility of the NTP-HTS data for
QSAR modeling of rodent carcinogenicity, A
total of 314 NTP-HTS compounds are repre-
sented in the CPDB. A summary of HTS
activity and rodent carcinogenicity of these
agents is shown in Table 5. Seventy-seven
percent of the compounds classified by
NTP—HTS as "active" are also categorized as
rodent carcinogens. On the contrary, only
46% of NTP—HTS "inactive" agents are classi-
fied by the CPDB as noncarcinogenic in
rodents. At the same time, the large fraction of
compounds found inactive in HTS assays
effectively renders the current assays insuffi-
cient in terms of predicting the in vivo toxicity.
To further examine whether in vitro
NTP-HTS data could improve the pre-
diction accuracy for in vivo rodent carcino-
genicity testing, we applied the hybrid
descriptor-based QSAR modeling that uti-
lized both biological (NTP-HTS output) and
chemical [MolConnZ (eduSoft LC)] descrip-
tors. First, all 314 compounds were randomly
divided into two sets. The modeling set com-
prised 264 compounds, whereas 50 randomly
selected compounds were designated as the
external validation set. After calculating
Table 3. Statistical information of the 15 most statistically significant /d\IN QSAR models based on the
275-compound modeling set.
Model ID
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
Average
N-training
141
140
141
140
141
190
228
140
140
149
140
190
140
149
141
154
Pred.-training
0.90
0.91
0.92
0.88
0.88
0.90
0.86
0.92
0.89
0.88
0.87
0.85
0.88
0.87
0.92
0.89
N-test
119
121
120
123
120
85
47
121
116
122
124
85
125
118
123
111
Pred.-test
0.73
0.71
0.69
0.73
0.73
0.71
0.74
0.67
0.70
0.70
0.72
0.73
0.70
0.71
0.66
0.71
NNN
1
1
1
1
1
1
1
1
4
1
1
1
1
1
1
1
Abbreviations: N-training, number of compounds in the training set; Pred.-training, the overall predictivity of the training
set; N-test, number of compounds in the test set; Pred.-test, the overall predictivity of the test set; NNN, number of the
nearest neighbors used for prediction.
Environmental Health Perspectives • VOLUME 116 I NUMBER 4 I April 2008
509
Previous
TOC
-------
Zhu et al.
chemical descriptors using the MolConnZ
software, we combined the NTP-HTS data
(a total of seven binary biological descriptors
including the active/inactive call for each cell
line separately and one for the entire experi-
ment, i.e., a compound was considered active
if it was active in at least one cell line) with
the MolConnZ chemical descriptors to create
a hybrid chemico-biological descriptor set.
Although we appreciate that the six cell lines
originate from different organs, it is notewor-
thy that great similarity was observed in cyto-
toxicity profiles across the entire panel of cell
lines (R. Tice, personal communication).
Furthermore, the number of active com-
pounds for each individual cell line is rela-
tively small, thus we combined the data. After
using the sphere exclusion method to generate
training/test set pairs from the same modeling
set of compounds, two types of £NN QSAR
models were developed. One was built using
only the MolConnZ chemical descriptor set
(340 variables), and the other was built using
the combined chemico-biological descriptor
set (347 variables).
kNN QSAR models were selected based
on the cf/R2 cutoff of 0.65/0.65 in this model-
ing development process. One hundred three
£NN models developed using chemical
descriptors alone that passed these criteria,
whereas this number nearly doubled to 198
when a combined chemico-biological descrip-
tor set was used. Although data from each of
the six cell lines or their combination were
given equal weight in defining the global
NTP-HTS activity of each compound, the
prognostic value of each cell line varied with
regard to its usefulness for predicting the
rodent carcinogenicity of a chemical. Figure 1
shows the frequency of use of each biological
descriptor in the 198 successful £NN QSAR
models. The predictive power of the QSAR
models was verified using the external valida-
tion set of 50 compounds not used in training
set modeling (Table 6). QSAR modeling
using MolConnZ descriptors only [referred to
as £NN-MolConnZ (£NN-MZ) models]
achieved 69.2% sensitivity and 55.5% speci-
ficity (Table 7). In contrast, 78.6% sensitivity
and 66.7% specificity were achieved when the
combined chemico-biological descriptor set
(referred to as £NN-MZHTS models) was
used for modeling. The overall prediction
accuracy rate increased significantly from
62.3% to 72.7% and the coverage of the
external set increased from 88% to 92%, that
is, more external compounds were found
within (numerically) the same applicability
domain when using the hybrid descriptor set.
The y-randomization test was also per-
formed for the carcinogenicity modeling using
MZ descriptors only and using the MZ and
HTS descriptors. Using randomized carcino-
genicity results, no models could be found to
satisfy the criterion ofif/R2 > 0.65, indicating
that our models were statistically robust.
Discussion
This study evaluated the potential of HTS cell
assays as novel biological predictors of adverse
health effects caused by chemicals in vivo in ani-
mal studies. To this end, we have evaluated the
HTS data for hundreds of chemical agents
tested in six cell lines and focused on com-
pounds that were also studied for their carcino-
genicity in chronic two-year cancer bioassays by
the NTP. Although HTS results provided com-
plete dose—response data, we used only binary
activity summary data (i.e., actives or inactives)
because of the binary nature of the CPDB data
(i.e., carcinogenic or not carcinogenic). Our ini-
tial analysis has established a strong correlation
between the chemical structures of the com-
pounds and their effects in cell-based assays.
However, we have demonstrated, not surpris-
ingly, that the results of testing compounds in
cell viability assays do not serve as unequivocal
predictors of their carcinogenicity in vivo.
Specifically, the data indicated a fairly strong
predictivity of cell growth inhibition toward
animal carcinogenicity (i.e., a positive cell
viability assay response has a strong probability
Table 4. Consensus prediction for 109 compounds in the external validation set.
Consensus prediction
After applicability domain applied
Model characteristics
Pred. actives (n)
Pred. inactives (n)
Sensitivity (%)
Specificity(%)
Overall predictive power (%)a
Exp. actives
21
16
56.8
90.2
73.5
Exp. inactives
7
65
Exp. actives
17
9
65.4
92.9
79.2
Exp. inactives
5
65
Abbreviations: Exp., experimental; Pred., predicted.
"The overall predictive power is the average value of sensitivity (predictive rate of actives) and specificity (predictive rate
of inactives).
Table 5. The relationship between HTS activity and rodent carcinogenicity of 314 compounds.
Content of CPDB
HTS actives
HTS inconclusives
HTS inactives
of predicting carcinogenicity in vivo) but low, if
any, predictivity of the in vivo carcinogenicity
on the compound effects in cell viability assays
(i.e., there are many carcinogens that do not
elicit responses in the cell viability assays). Thus,
to maximize the utility of in vitro assays results
for predicting the in vivo data, we considered
building QSAR models of the in vivo chemical
carcinogenicity using HTS results as additional
biological descriptors of underlying chemical
structures.
There are several major potential applica-
tions of biological descriptors in QSAR model-
ing that may advance the science and practice
of computational toxicology. In our computa-
tional experiments, the binary contributions of
all six HTS cell line test results were treated
equally a priori. The variable selection £NN
QSAR approach yielded 198 externally predic-
tive models. Because of the nature of the
method, these models differ in the choice of
descriptors resulting from the variable selection
procedure for the final model. Thus, the mod-
els could be analyzed for the frequency of
occurrence of different descriptors that could
reveal chemical determinants of a compound's
carcinogenicity, as well as possible utility of the
individual HTS assays. Figure 1 shows the fre-
quency of occurrence of seven HTS descriptors
in the 198 £NN QSAR models described
above. The analysis of this distribution, espe-
cially in the context of chemical structure of
tested compounds, may provide clues concern-
ing the usefulness of different cell lines for
screening purposes.
For example, the HTS-Jurkat and HTS-
HepG2 biological descriptors were found in the
majority of the successful models. Jurkat and
HepG2 are human tumor cell lines derived
from a T-cell leukemia and hepatocellular carci-
noma, respectively. Jurkat cells grow in suspen-
sion with a relatively fast doubling time of about
22 hr. In contrast, HepG2 cells grow as attached
cultures with a doubling time of about 37 hr.
Both cell lines retain some metabolic capacity
toward xenobiotics and are used frequently for
in vitro testing (Mersch-Sundermann et al.
2004; Nagai et al. 2002). Compared with
HTS-Jurkat and HTS-HepG2 cells, the
HTS-HEK293 descriptor (a human embryonic
120
8 o 100
if 8°
= -S 60
CPDB actives (n)
CPDB inactives (n}
Correlation (%)
30
77
12
13
136
114
46
Figure 1. Seven HTS descriptors with their fre-
quency of use in the 198 kNN QSAR model.
510
VOLUME 116 I NUMBER 41 April 2008 • Environmental Health Perspectives
Previous
TOC
-------
Biological descriptors in QSAR modeling of carcinogenicity
kidney cell line) was found in much smaller
numbers of successful models, and all but
two compounds active in this cell line were
also found to be active in other cell lines.
Therefore, assay results for the tested com-
pounds in HEK293 cells may be redundant
with respect to rodent carcinogenicity model-
ing conducted here.
Interestingly, the predictions for 8 of the 50
compounds in the external test set were differ-
ent using the £NN-MZ versus £NN-MZHTS
models. The apparent reason for this disparity
(in the context of £NN QSAR approach used
in this study) is due to the change of nearest
neighbors in the training set of these 8 com-
pounds using the MolConnZ (eduSoft LC)
descriptors only versus using the hybrid chemi-
cal—HTS descriptors. For example, the com-
pound 2,4-dichlorophenol (CAS no. 120-83-2)
has 1,2-benzenediol (CAS no. 120-80-9),
1,4-benzenediol (CAS no. 123-31-9), and
4-chlorobenzene-l,2-diamine (CAS no.
95-83-0) as its nearest neighbors in the training
set as defined by £NN-MZ modeling
(Table 6). After including HTS descriptors, its
nearest neighbors in the training set change to
2-chloro-/>-phenylenediamine (CAS no.
61702-44-1), l-amino-4-methoxybenzene
(CAS no. 20265-97-8), and/-nitroaniline
(CAS no. 100-01-6) instead. Thus, the addi-
tion of HTS descriptors affects the similarity
relationships between compounds based purely
on their chemical descriptors. As shown in this
study, the addition of HTS descriptors, on
average, improves the prediction accuracy of
in vivo carcinogenicity.
We further analyzed the interplay between
the significance of the bioassay and that of spe-
cific chemical descriptors in the context of
in vivo carcinogenicity by comparing the
occurrence of top chemical descriptors in
QSAR models with and without HTS descrip-
tors. Table 8 shows chemical descriptors that
occur most frequently in successful (i.e., exter-
nally predictive) QSAR models using chemical
descriptors only. This table also reports the
change in occurrence of these descriptors after
HTS descriptors are included. Because the
number of successful £NN QSAR models
increased significantly from 103 to 198 after
HTS descriptors were used, we also include in
Table 8 the ratio of occurrence to the total
number of models, which may better indicate
the significance of the descriptors.
The descriptors for each final kNN QSAR
model are chosen as a result of the stochastic
variable selection procedure that maximizes the
correlation between descriptors and carcino-
genicity. We reasoned that the analysis of
occurrence of various chemical descriptors
before and after inclusion of HTS descriptors
in modeling may be interpreted in terms of
their relative information content with respect
to the in vivo toxicity. Thus, those chemical
descriptors that have a similar ratio of occur-
rence in models with or without HTS descrip-
tors (exemplified by descriptors 1, 2, and 7)
contribute to successful models independently
of the biological descriptors. For compounds
whose predicted activity is primarily determined
by the presence of these particular chemical
descriptors and unaffected by the addition of
Table 6. Consensus prediction of 50 compounds in the external validation set using the /d\IN QSAR models
based on two different descriptor sets.
CAS no.
79005
106934
90120
86577
634935
120832
99558
67630
96695
619170
298817
75058
50782
50760
86500
92875
57578
80057
75274
115286
91645
4342034
103231
333415
62737
828002
98011
87683
67721
122667
58935
121755
298000
150685
1212299
759739
98953
67209
59870
55185
636215
106478
122601
103855
1918021
57681
79196
108054
1330207
17924924
Name CPDB actives
1,1,2-Trichloroethane +
1,2-Dibromoethane +
1-Methylnaphthalene
1-Nitronaphthalene
2,4,6-Trichloroaniline +
2,4-Dichlorophenol
5-Nitro-o-toluidine +
Isopropanol
4,4-Thiobis(6-ferf-butyl-m-cresol)
4-Nitroanthranilicacicl
8-Methoxypsoralen +
Acetonitrile
Acetylsalicylicacid
Actinomycin D +
Azinphosmethyl
Benzidine +
Propiolactone +
BisphenolA
Bromodichloromethane +
Chlorendicacid +
Coumarin +
Dacarbazine +
Di(2-ethylhexyl)adipate +
Diazinon
Dichlorvos +
Dimethoxane +
Furfural +
Hexachloro-1,3-butadiene +
Hexachloroethane +
Hydrazobenzene +
Hydrochlorothiazide
Malathion
Methyl parathion
Monuron +
/V./V'-Dicyclohexylthiourea
/V-Ethyl-n-nitrosourea +
Nitrobenzene +
Nitrofurantoin +
Nitrofurazone +
/V-Nitrosodiethylamine +
o-Toluidine hydrochloride +
p-Chloroaniline
Phenyl glycidyl ether +
Phenylthiourea
Picloram
Sulfamethazine +
Thiosemicarbazide
Vinyl acetate +
Xylenes (mixed)
Zearalenone +
MZ MZHTS
+ +
+ +
-
+ +
+ +
+
+ +
+
+ +
- -
+ +
+ +
-
I +
-
+ +
+ +
+
+ +
I I
+ +
_ _
+ +
- -
+ +
-
+ +
+
+ +
-
I I
- -
-
_ _
I |
+ +
+ +
I +
I I
+ +
-
-
+ +
+
- -
-
+ +
+ +
+ +
+
Abbreviations: +, carcinogenic; -, noncarcinogenic; I, inconclusive because out of the applicability domain; MZ, models
based on MolConnZ descriptors only; MZHTS, models based on the combination of MolConnZ and HTS descriptors.
Table?. Summary of the statistical parameters of the prediction results of 50 external compounds.
Chemical descriptors only
Combined descriptors
Model characteristics
Pred. actives
Pred. inactives
Sensitivity (%)
Specificity(%)
Overall predictivity(%)
Coverage (%)
Exp. actives
18
8
69.2
55.5
62.3
88
Exp. inactives
8
10
Exp. actives
22
6
Exp. inactives
6
12
78.6
66.7
72.7
92
Abbreviations: Exp., experimental; Pred., predicted.
Environmental Health Perspectives • VOLUME 116 I NUMBER 4 I April 2008
511
Previous
TOC
-------
Zhu et al.
Table 8. Summary of the top 10 atom and bond type MozConnZ chemical descriptors used in successful
kNN QSAR models before and after using HTS descriptors.
No. Descr Name Illustration
Freq_MZ Ratio_MZ Freq_MZHTS Ratio_MZHTS
Snitroso
nHBintS
naasN
SHBintS
nHssNH
SdsN
SdsssP
9 SsBr
10 SHssNH
Sum of E-states of nitroso group
nnitroso Number of nitroso group
Number of hydrogen bond acceptor
and donor pairs separated by
3 skeletal bonds
I
38
34
27
36.9%
33.0%
26.2%
73
36.9%
34.8%
15.7%
H bond
donor
H bond
acceptor
Number of aromatic nitrogen
with substitute
Sum of E-state of strength for
potential hydrogen bonds if
separated by 3 skeletal bonds
l
H bond | | 3
donor H A
H bond
acceptor
Number of amine groups
,H
Sum of E-states for nitrogens
with one single bond and one
double bond
Sum of E-states for phosphors
with three single bonds and
one double bond
Sum of E-states for bromines
R Br
Sum of H E-states for hydrogens
in amine groups.
25
24.3%
42
21.2%
24 23.3% 41
20.7%
24 23.3% 23
24 23.3%
19
18.4%
24.2%
10.6%
19
18
18.4%
17.5%
45 22.7%
25 12.6%
N
\
Abbreviations: Descr_Name, name of descriptor; Freq_MZ, frequency of occurrence in successful kNN models only
using only MolConnZ descriptors; Ratio_MZ, ratio of occurrence in successful QSAR models using only MolConnZ
descriptors; Freq_MZHTS, frequency of occurrence in successful kNN models using MolConnZ and HTS descriptors;
Ratio_MZHTS, ratio of occurrence in successful QSAR models using MolConnZ and HTS descriptors.
HTS descriptors, this implies that the HTS
adds no new information to the prediction of
in vivo carcinogenicity. Conversely, if the fre-
quency of a chemical descriptor decreases sig-
nificantly after the HTS descriptors are
included, it is less important than, and likely
redundant with, the biological descriptors. In
these cases, the biological descriptor is clearly
adding new, biologically significant information
that is not as effectively captured by the chemi-
cal descriptor.
Interestingly, descriptors 1 and 2
(7Y-nitroso compounds) were selected as the
most important chemical descriptors in our
models, and their importance is relatively unaf-
fected by inclusion of HTS descriptors. The
large majority of 7V-nitroso compounds have
been found to produce genotoxic effects and to
cause tumor development in laboratory ani-
mals, as they are metabolized to reactive elec-
trophilic species causing damage to various
cellular constituents such as DNA, constituting
a key event in the carcinogenic mechanism
(Brambilla and Martelli 2007). Because such
metabolic transformations do not generally
occur in cellular systems, the significance of all
but one of the HTS assays (i.e., HepG2) in
predicting events relevant to the carcinogenic-
ity for these compounds is likely to be mini-
mal. To the contrary, the NTP-HTS data
show that cells are highly sensitive to the effects
of amine-type compounds (descriptors 6 and
10) and biological descriptors are better predic-
tors of the carcinogenic potential of these
agents than structure alone. Among all 30 car-
cinogens that are also active in HTS tests, 15
are amines. A similar observation can be made
for organic compounds containing phosphorus
(descriptor 8). Most of the remaining chemical
descriptors, which approximately delineate
neighborhoods of chemical space, have similar
distribution among the models with or with-
out the HTS descriptors. Hence, HTS descrip-
tors offer no additional value as predictors of
carcinogenicity for these chemical subsets. As
more HTS data are generated, the above analy-
sis suggests a strategy that can be used to eluci-
date possible mechanistic relevance of HTS
assays to carcinogenicity prediction within
areas of chemical space approximately defined
by chemical descriptors.
Conclusions
We have examined the utility of in vitro
NTP—HTS data for predicting in vivo adverse
health effects (i.e., carcinogenicity) of environ-
mental agents. Our analysis suggests that
NTP—HTS results have limited predictive
power by themselves for rodent carcinogenicity.
This result is not surprising, given the relatively
low frequency of positives across the HTS
assays (16%) and that cell viability (i.e., cell
death) may not be directly related to the car-
cinogenic potential of a compound. However,
512
VOLUME 116 I NUMBER 41 April 2008 • Environmental Health Perspectives
Previous
TOC
-------
Biological descriptors in QSAR modeling of carcinogenicity
our data suggest that using the NTP-HTS
results as biological fingerprint descriptors of
generalized xenobiotic-induced pathophysiol-
ogical processes helps improve the overall
QSAR-based prediction accuracy of rodent car-
cinogenicity compared with those based on
chemical descriptors alone. While the mecha-
nistic relevance of the HTS assays in predicting
rodent carcinogenicity is unclear at present, the
empirical evidence of the significance of the
biological descriptors for the computational
modeling purposes is compelling and should
motivate continued investigation. Furthermore,
as additional sets of compounds with known
in vivo toxicity responses are investigated in
cell-based viability assays, we shall continue to
develop models similar to those reported in this
article for additional toxicity end points. The
present analysis suggests that as more mechanis-
tically relevant HTS data are generated and a
greater number of compounds are screened,
computational toxicology tools could be used
to select most relevant HTS assays (cell lines
and/or measurements) and prioritize chemical
agents for screening. With sufficient improve-
ments in resulting model predictive perfor-
mance, in vitro HTS bioassays, coupled with
traditional chemical structure-based descriptors,
may be ultimately helpful in prioritizing or even
partially replacing in vivo toxicity testing.
CORRECTION
The following corrections have been made
from the original manuscript published
online. In the Abstract under "Methods and
Results," the phrase "curated data set of
557 compounds" has been changed to
"curated data set of 384 compounds." The
sentence "The resulting models had predic-
tion accuracies for training, test (containing
400 compounds together), and external vali-
dation (157 compounds) sets as high as
79%, 79%, and 84%, respectively" has been
changed to "The resulting models had pre-
diction accuracies for training, test (contain-
ing 275 compounds together), and external
validation (109 compounds) sets as high as
89%, 71%, and 74%, respectively."
REFERENCES
Beresford AP, Segall M, Tarbit MH. 2004. In silico prediction of
ADME properties: are we making progress? Curr Opin
Drug Discov Develop 7:36-42.
Brambilla G, Martelli A. 2007. Genotoxic and carcinogenic risk
to humans of drug-nitrite interaction products. Mutat Res
635:17-52.
Bucher JR, Portier C. 2004. Human carcinogenic risk evalua-
tion. Part V: The National Toxicology Program vision for
assessing the human carcinogenic hazard of chemicals.
ToxicolSci 82:363-366.
CPDB (The Carcinogenic Potency Database). 2007. The
Carcinogenic Potency Project. Available: http://potency.
berkeley.edu/cpdb.html [accessed 14 December 2007].
Dearden JC. 2003. In silico prediction of drug toxicity. J Comput
Aided MolDes 17:119-127.
de Lima P, Golbraikh A, Oloff S, Xiao Y, Tropsha A. 2006.
Combinatorial QSAR modeling of P-glycoprotein sub-
strates. J Chem Inf Mod 46:1245-1254.
Golbraikh A, Shen M, Xiao Z, Xiao YD, Lee KH, Tropsha A. 2003.
Rational selection of training and test sets for the develop-
ment of validated QSAR models. J Comput Aided Mol Des
17:241-253.
Golbraikh A, Tropsha A. 2002a. Beware of q2! J Mol Graph
Model 20:269-276.
Golbraikh A, Tropsha A. 2002b. Predictive QSAR modeling
based on diversity sampling of experimental datasets for
the training and test set selection. J Comput Aided Mol
Des 16:357-369.
Gold LS, Slone TH, Manley NB, Garfinkel GB, Hudes ES,
Rohrbach L, et al. 1991. The Carcinogenic Potency
Database: analyses of 4000 chronic animal cancer experi-
ments published in the general literature and by the U.S.
National Cancer Institute/National Toxicology Program.
Environ Health Perspect96:11-15.
Hall LH, Mohney B, Kier LB. 1991. The electrotopological
state—an atom index for QSAR. Quan Struct Act Relat
10:43-51.
Inglese J, Auld DS, Jadhav A, Johnson RL, Simeonov A, Yasgar
A, et al. 2006. Quantitative high-throughput screening: a
titration-based approach that efficiently identifies biologi-
cal activities in large chemical libraries. Proc Natl Acad
Sci USA 103:11473-11478.
Johnson DE, Smith DA, Park BK. 2004. Linking toxicity and
chemistry: think globally, but act locally? Curr Opin Drug
Discov Devel 7:33-35.
Kier LB. 1986. Indexes of molecular shape from chemical
graphs. Acta Pharmaceutica Jugoslavia 36:171-188.
Kier LB. 1987. Inclusion of symmetry as a shape attribute in
kappa-index analysis. Quant Struct Act Relat 6:8-12.
Kier LB, Hall LH. 1991. A differential molecular connectivity
Index. Quant Struct Act Relat 10:134-140.
Klopman G, Zhu H, Ecker G, Chiba P. 2003. MCASE study of the
multidrug resistance reversal activity of propafenone
analogs. J Comput Aided Mol Des 17:291-297.
Klopman G, Zhu H, Fuller MA, Saiakhov RD. 2004. Searching for
an enhanced predictive tool for mutagenicity. SAR QSAR
Environ Res 15:251-263.
Kovatcheva A, Golbraikh A, Oloff S, Xiao YD, Zheng W,
Wolschann P, et al. 2004. Combinatorial QSAR of ambergris
fragrance compounds. J Chem Inf Comput Sci 44:582-595.
Mayer P, Reichenberg F. 2006. Can highly hydrophobic organic
substances cause aquatic baseline toxicity and can they
contribute to mixture toxicity? Environ Toxicol Chem
25:2639-2644.
Mersch-Sundermann V, Knasmuller S, Wu XJ, Darroudi F,
Kassie F. 2004. Use of a human-derived liver cell line for
the detection of cytoprotective, antigenotoxic and
cogenotoxic agents. Toxicology 198:329-340.
Nagai F, Hiyoshi Y, Sugimachi K, Tamura HO. 2002. Cytochrome
P450 (CYP) expression in human myeloblastic and lym-
phoid cell lines. Biol Pharm Bull 25:383-385.
NCBI (National Center for Biotechnology Information). 2007.
The PubChem Project. Available: http://pubchem.
ncbi.nlm.nih.gov [accessed 3 September 2007].
Ng C, Xiao Y, Putnam W, Lum B, Tropsha A. 2004. Quantitative
structure-pharmacokinetic parameters relationships
(QSPKR) analysis of antimicrobial agents in humans using
simulated annealing k-nearest-neighbor and partial least-
square analysis methods. J Pharm Sci 93:2535-2544.
NTP (National Toxicology Program). 2007. NTP High Throughput
Screening Initiative. Available: http://ntp.niehs.nih.gov/go/
28213 [accessed 14 December 2007].
Renner S, Schneider G. 2006. Scaffold-hopping potential of lig-
and-based similarity concepts. ChemMedChem 1:181-185.
Richard AM. 2006. Future of toxicology—predictive toxicology:
an expanded view of "chemical toxicity." Chem Res Toxicol
19:1257-1262.
Schultz TW, Cronin MTD, Netzeva Tl. 2003a. The present status
of QSAR in toxicology. J Mol Struct Theochem 622:23-38.
Schultz TW, Netzeva Tl, Cronin MT. 2003b. Selection of data
sets for QSARs: analyses of Tetrahymena toxicity from
aromatic compounds. SAR QSAR Environ Res 14:59-81.
Shen M, LeTiran A, Xiao Y, Golbraikh A, Kohn H, Tropsha A. 2002.
Quantitative structure-activity relationship analysis of func-
tionalized amino acid anticonvulsant agents using k nearest
neighbor and simulated annealing PLS methods. J Med
Chem 45:2811-2823.
Shen M, Xiao Y, Golbraikh A, Gombar VK, Tropsha A. 2003.
Development and validation of k-nearest-neighbor QSPR
models of metabolic stability of drug candidates. J Med
Chem 46:3013-3020.
Stoner CL, Gifford E, Stankovic C, Lepsy CS, Brodfuehrer J,
Prasad JVNV, et al. 2004. Implementation of an ADME
enabling selection and visualization tool for drug discov-
ery. J Pharm Sci 93:1131-1141.
Tropsha A, Gramatica P, Gombar VK. 2003. The importance of
being earnest: validation is the absolute essential for suc-
cessful application and interpretation of QSPR models.
QSAR Combi Sci 22:69-77.
U.S. EPA (U.S. Environmental Protection Agency). 2007.
Distributed Structure-Searchable Toxicity (DSSTox) Public
Database Network. Available: http://www.epa.gov/ncct/
dsstox/index.html [accessed 3 September 2007].
Willett P, Winterman V. 1986. A comparison of some measures
for the determination of intermolecular structural similarity
measures of intermolecular structural similarity. Quan
Struct Activ Relat 5:18-25.
Xia M, Huang R, Witt KL, Southall N, Fostel J, Cho MH, et al. .
2007. Compound cytotoxicity profiling using quantitative
high-throughput screening. Environ Health Perspect
doi:10.1289/ehp.10727 [Online 22 November 2007].
Zhang S, Golbraikh A, Oloff S, Kohn H, Tropsha A. 2006. A novel
automated lazy learning QSAR (ALL-QSAR) approach:
method development, applications, and virtual screening
of chemical databases using validated ALL-QSAR models.
J Chem Inf Model 46:1984-1995.
Zheng W, Tropsha A. 2000. Novel variable selection quantita-
tive structure—property relationship approach based on
the k-nearest-neighbor principle. J Chem Inf Comput Sci
40:185-194.
Environmental Health Perspectives • VOLUME 116 I NUMBER 4 I April 2008
513
Previous
TOC
-------
Genetic Epidemiology 32: 767-778 (2008)
A Comparison of Analytical Methods for Genetic Association Studies
Alison A. Motsinger-Reif,1 David M. Reif,2 Theresa J. Fanelli3 and Marylyn D. Ritchie3*
1Bioinformatics Research Center, Department of Statistics, North Carolina State University, Raleigh, North Carolina
2National Center for Computational Toxicology, US Environmental Protection Agency, Research Triangle Park, North Carolina
3Center for Human Genetics Research and Department of Molecular Physiology and Biophysics, Vanderbilt University Medical School,
Nashville, Tennessee
The explosion of genetic information over the last decade presents an analytical challenge for genetic association studies. As
the number of genetic variables examined per individual increases, both variable selection and statistical modeling tasks
must be performed during analysis. While these tasks could be performed separately, coupling them is necessary to select
meaningful variables that effectively model the data. This challenge is heightened due to the complex nature of the
phenotypes under study and the complex underlying genetic etiologies. To address this problem, a number of novel
methods have been developed. In the current study, we compare the performance of six analytical approaches to detect both
main effects and gene-gene interactions in a range of genetic models. Multifactor dimensionality reduction, grammatical
evolution neural networks, random forests, focused interaction testing framework, step-wise logistic regression, and explicit
logistic regression were compared. As one might expect, the relative success of each method is context dependent. This
study demonstrates the strengths and weaknesses of each method and illustrates the importance of continued methods
development. Genet. Epidemiol. 32:767-778, 2008. © 2008 Wiley-Liss, Inc.
Key words: genetic association study; epistasis; multifactor dimensionality reduction; grammatical evolution neural
networks; focused interaction testing framework
Contract grant sponsor: National Institutes of Health; Contract grant numbers: HL65962; GM62758; AG20135.
*Correspondence to: Marylyn D Ritchie, Center for Human Genetics Research, Department of Molecular Physiology and Biophysics, 519
Light Hall, Vanderbilt University Medical School, Nashville, TN 37232-0700. E-mail: ritchie@chgr.mc.vanderbilt.edu
Received 4 lanuary 2008; Accepted 3 April 2008
Published online 16 lune 2008 in Wiley InterScience (www.interscience.wiley.com).
DOI: 10.1002/gepi.20345
INTRODUCTION
The identification and characterization of genetic
variants that predict common, complex disease is an
important priority in the field of genetic epidemiol-
ogy. It is hypothesized that such diseases are the
result of the complex interplay between a myriad of
genetic and environmental factors [Moore, 2003;
Moore and Williams, 2005; Ritchie, 2005; Templeton,
2000; Thornton-Wells et al., 2004]. Additionally, as
genotyping technology has advanced, the volume of
genetic information available for analysis has ex-
ploded. This explosion of information, combined
with the complex genetic architecture, has resulted
in a difficult analytical challenge [Moore and
Ritchie, 2004; Ritchie et al., 2005]. The dimensionality
involved in the evaluation of combinations of many
such variables quickly diminishes the usefulness of
traditional, parametric statistical methods. Referred
to as the curse of dimensionality [Bellman, 1961], as
the number of genetic or environmental factors
increases and the number of possible interactions
increases exponentially, many contingency table
cells will be left with very few, if any, data points.
This results in a crucial need for analytical methods
that can simultaneously perform variable selection
tasks along with statistical modeling in such high-
dimensional data.
To address this challenge, a number of analytical
and computational methods have been developed to
detect and model genetic associations [Hahn et al.,
2003; Kooperberg et al., 2001; Moore, 2003; Nelson
et al., 2001; Ritchie et al., 2001,2003a; Tahri-Daizadeh
et al., 2003; Zhu and Hastie, 2004]. Since it is unlikely
that any one analytical method will be ideal in all
situations, it is important to empirically evaluate the
strengths and weaknesses of a variety of computa-
tional approaches. In the current study, we evaluate
the performance of multifactor dimensionality re-
duction (MDR) [Ritchie et al., 2001], grammatical
evolution neural networks (GENN) [Motsinger et al.,
2006a; Motsinger-Reif et al., 2008a], focused interac-
tion testing framework (FITF) [Millstein et al., 2006],
random forests (RF) [Breiman, 2001], and logistic
regression (LR) [Hosmer and Lemeshow, 2000] on
main effect, two-locus, and three-locus genetic
models. We approach this comparison from an
end-user's perspective, using commonly available
© 2008 Wiley-Liss, Inc.
-------
768
Motsinger-Reif et al.
software packages and author recommended con-
figuration parameters.
As expected the relative performance of each
method is context dependent, as each method has
its individual strengths and weaknesses. This study
highlights the utility of each method in their
respective context and stresses the importance of
continued methods development. No single method
will be optimal for all data scenarios, and this study
illustrates this fact. The goal of this study is to aid a
common user in deciding which computational
method is most appropriate for their particular data.
METHODS
MULTIFACTOR DIMENSIONALITY
REDUCTION (MDR)
MDR was designed to detect gene-gene interac-
tions in the presence or absence of main effects in
case-control studies in human genetics [Hahn et al.,
2003; Ritchie et al., 2001]. MDR has been shown to
have high power to detect interactions in a wide
range of simulated data [Motsinger and Ritchie,
2006b; Ritchie et al., 2003a; Velez et al., 2007] and has
identified interactions in common complex diseases
such as multiple sclerosis [Brassat et al., 2006;
Motsinger et al., 2007], coronary artery disease
[Agirbasli et al., 2006], and diabetic nephropathy
[Hsieh et al., 2006].
MDR has previously been described in detail
[Hahn et al., 2003; Motsinger and Ritchie, 2006a;
Ritchie et al., 2001]. Figure 1 illustrates the MDR
algorithm. The data set is divided into multiple
partitions for cross-validation before analysis begins.
Cross-validation [Hastie et al., 2001] is an important
part of the algorithm in order to find a model that
not only fits the given data but can also predict on
unseen data. In this study, five-fold cross-validation
is used, so that 4/5 of the data comprise the training
set and the remaining 1 /5 of the data comprises the
testing set [Motsinger and Ritchie, 2006b].
In step one, an exhaustive list of n combinations of
genetic loci to evaluate from the list of all variables is
created. Next, each of the n combinations is arranged
in contingency tables in fc-dimensional space with all
possible combinations as individual cells in the
table. Then, the number of cases and controls for
each locus combination is counted. In step three, the
ratio of cases to controls within each cell is
calculated. Each genotype combination is then
labeled as "high risk" or "low risk" of the phenotype
of interest based on comparison of the ratio to a
threshold. The threshold used is dependent on the
ratio of cases and controls within the data set. If the
ratio within a multifactor combination is above that
seen in the data, it is labeled as "high risk" and if it
is below, it is labeled as "low risk." This step
compresses multidimensional genotype data into
one dimension with two classes.
The high-risk/low-risk profile for each of the
multifactorial combinations represents the MDR
model for a particular combination of multi-locus
genotypes. Balanced accuracy, the arithmetic mean
of sensitivity and specificity, is calculated for each
model. Balanced accuracy is used as the fitness
metric in the MDR algorithm to solve the challenges
Stec 1
List of multi-locus
combinations to
evaluate
1,2
1,3
1,4
Factor 1
Factor 1
s
)
Step 2°
1 .
d
M
I
J
L
•
• *
i. i
L
Step3
Step 6
Training Accuracy
1.2 73.77
1.3 69.22
Testing Set step 7 r^ ^
Factor 1 * I—
HR | HR
S.
LR
LR
LR
LR
i
Testing Accuracy
1.4 72.30
CV Consistency
1.4 9
3.
2.80
0.33
0.13
0.23
3,33
0.58
1.33
0.40
2.71
Step4|T=1
Training Set
Factor 1
Step 5
72,30 p=0.002
Fig. 1. Multifactor dimensionality reduction. The steps correspond to those described in the Methods section.
Genet, Epidemiol.
Previous
TOC
-------
Analytical Methods for Genetic Epidemiology
769
presented by an imbalanced number of cases and
controls in a data set. In combination with adjusting
the threshold used in assigning high- or low-risk
status, it has been shown that balanced accuracy
makes MDR robust to class imbalance [Velez et al.,
2007]. In the case of balanced data, balanced
accuracy is mathematically equivalent to classifica-
tion accuracy. The best fc-locus model is selected and
the model is evaluated against the testing group and
testing accuracy is calculated. Prediction accuracy is
then calculated for the testing set. This is repeated
for each cross-validation interval (i.e. training set
and testing set) and the average training accuracy
and testing accuracy are calculated. Among all of the
fc-locus models created, the single model with the
highest cross-validation consistency is chosen as the
best fc-locus model. This process is completed for
each k = 1 to N loci combinations that are computa-
tionally feasible. An optimal fc-locus model is chosen
for each level of k considered, so a one-locus model,
two-locus model, three-locus model, etc. collectively
comprise a set.
Once this set of models is computed, a final model
is chosen. The final model is selected based on
maximization of both testing accuracy and cross-
validation consistency. Testing accuracy is how well
the model predicts risk/disease status in indepen-
dent testing sets generated through cross-validation
and is calculated as described above. Cross-valida-
tion consistency is the number of times a model is
identified across the cross-validation sets. For five-
fold cross-validation, the consistency can range from
one to five. A higher value of cross-validation
consistency represents stronger support for the
model. When testing accuracy and cross-validation
indicate different models, the rule of parsimony is
used to choose the simpler model.
GRAMMATICAL EVOLUTION NEURAL
NETWORKS (GENN)
GENN is a machine-learning approach designed
to detect gene-gene interactions in the presence or
absence of marginal main effects [Motsinger et al.,
2006a; Motsinger-Reif et al., 2008a]. GENN was
designed to perform variable selection without the
computational burden of exhaustively searching all
possible variable combinations. GENN has success-
fully identified interactions in a wide range of data
simulations [Motsinger et al., 2006a,b,c; Motsinger-
Reif et al., 2008a], and a real-data application in HIV
immunogenomics [Motsinger-Reif et al., 2008a].
Methodology and software have been previously
described for GENN [Motsinger et al., 2006a; Mot-
singer-Reif et al., 2008a]. GENN utilizes grammatical
evolution (GE) [O'Neill and Ryan, 2001] to optimize
the inputs, architecture, and weights of an NN.
Details of GE can be found in O'Neill and Ryan
[2003]. Briefly, a binary string genome is mapped
into a functional NN according to the rules specified
in pre-defined Backus-Naur Form grammar. The
grammar used in GENN is available in Motsinger-
Reif et al. [2008a] or from the authors on request.
Evolutionary operators operate at the level of the
binary string genome, automatically evolving the
optimal NN for a given data set.
The GENN algorithm is depicted in Figure 2. In
step one the user initializes a set of parameters in the
configuration file. These parameters specify details
of the evolutionary process implemented in GENN.
These items include: crossover rate, mutation rate,
STEP I
pupUUimw
IHI eninncoi
pvm.itrfcuief.gtiifrMitii.
,u,Jnm_,,,i
r, '.!•
™tan.«»
^T^^omd
»ii.*™._B»
^,.rt.0,_-.
iu
»
ii
7
1 I
DOI
:
.:
»
ICCC
STEP 2
10 1
• o
L
• •
STEP 5
GKNN Model
LITUI Pmliclliiii Emir
19.25
21.55
19.26%
22.12%
GKiNN Models
Classification Error
19.25
22,12
24,3?
28,14
Tournament ;
Fig. 2. Grammatical evolution neural networks. The steps correspond to those described in the Methods section.
Previous
TOC
Genet. Epidemiol.
-------
770
Motsinger-Reif et al.
population size, maximum number of generations,
type of selection, and type of crossover. Second, the
data are divided into 10 equal parts for 10-fold cross-
validation. Cross-validation is implemented to de-
velop a model that not only fits the data at hand but
that can also generalize to future, unseen data. In 10-
fold cross-validation, 9/10 of the data are used for
training, to develop an NN model, and the other 1 /
10 of the data are used to evaluate the predictive
ability of the model. This is repeated for each
possible 9/10:1/10 split of the data. Third, an initial
population of random solutions is generated using
sensible initialization [O'Neill and Ryan, 2003].
Sensible initialization guarantees functional NNs in
the initial population. Details of this process can be
found in [Ritchie et al., 2003b]. Fourth, each newly
generated NN is evaluated on the data in the
training set and its fitness recorded. The fitness
function used is balanced error (or 1—balanced
accuracy), where balanced accuracy is (sensitivity+
specificity)/2. Higher accuracy represents higher
reproductive fitness in the GENN population. This
fitness function was implemented to ensure GENN
is robust to class imbalance in a data set [Hardison
et al., 2008]. Fifth, a user-specified selection techni-
que selects the best solutions for crossover and
reproduction. A proportion of the best solutions will
also be directly copied (reproduced) into the new
generation. Another proportion of solutions will be
used for crossover with other best solutions. The
cycle begins with this newly created generation,
which is equal in size to the original population. This
cycle continues until either a classification error of
zero is found or a user-specified limit on the number
of generations is reached. After each generation, an
optimal solution is identified. At the end of GENN
evolution, the overall best solution is selected as the
optimal NN. Sixth, this best GENN model is tested
on the testing portion of the data to estimate the
predictive balanced error of the model. Steps two
through six are performed 10 times with the same
parameter settings, each time using a different 9/10
of the data for training and 1/10 of the data for
testing. At the end of a GENN analysis, 10 models
are generated—one best model from each cross-
validation interval. A final model is chosen based on
maximization of the cross-validation consistency of
variables/loci across the 10 models. A higher value
of cross-validation consistency represents stronger
support for the model.
RANDOM FORESTS (RF)
RF is a machine-learning technique that builds a
forest of classification trees wherein each component
tree is grown from a bootstrap sample of the data,
and the variable at each tree node is selected from a
random subset of all variables in the data [Breiman,
2001]. The final classification of an individual is
determined by voting over all trees in the forest.
There are several advantages of the RF method that
make it a promising technique for genetic associa-
tion studies. First, it can handle a large number of
input variables. Second, it estimates the relative
importance of variables in determining classifica-
tion, thus providing a metric for feature selection.
Third, RF produces a highly accurate classifier with
an internal unbiased estimate of generalizability
during the forest building process. Fourth, RF is
fairly robust in the presence of etiological hetero-
geneity and relatively high amounts of missing data
[Lunetta et al., 2004]. Finally, and of increasing
importance as the number of input variables
increases, learning is fast and computation time is
modest even for very large data sets [Robnik-
Sikonja, 2004]. RF has successfully identified disease
susceptibility variants in a variety of real-data
applications in genetic epidemiology such as drug
response [Sabbagh and Darlu, 2006] and dermato-
myositis [Mamyrova et al., 2006].
Each tree in the forest is constructed as follows
from data having N individuals and M explanatory
variables:
(1) Choose a training sample by selecting N indivi-
duals, with replacement, from the entire data set.
(2) At each node in the tree, randomly select m
variables from the entire set of M variables in the
data. The absolute magnitude of m is a function
of the number of variables in the data set and
remains constant throughout the forest building
process.
(3) Choose the best split at the current node from
among the subset of m variables selected above.
(4) Iterate the second and third steps until the tree is
fully grown (no pruning).
Repetition of this algorithm yields a forest of trees,
each of which has been trained on bootstrap samples
of individuals (see Fig. 3). Thus, for a given tree,
certain individuals will have been left out during
training. Prediction error and variable importance is
estimated from these "out-of-bag" individuals.
The out-of-bag (unseen) individuals are used to
estimate the importance of particular variables by
randomly permuting the values of that variable and
testing whether these permutations adversely affect
the predictive ability of trees in classifying out-of-
bag samples. If randomly permuting values of a
particular variable does not affect the predictive
ability of trees on out-of-bag samples, that variable is
assigned a low importance score. If randomly
permuting the values of a particular variable
drastically impairs the ability of trees to correctly
predict the class of out-of-bag samples, then the
importance score of that variable will be high. By
running out-of-bag samples down entire trees
during the permutation procedure, interactions are
taken into account when calculating importance
Genet. Epidemiol.
-------
Analytical Methods for Genetic Epidemiology
771
M variables
M variables
m variables
Fig. 3. Random forests. The steps correspond to those described
in the Methods section.
scores, since class is assigned in the context of other
variable nodes in the tree.
The recursive partitioning trees comprising an
RF provide an explicit representation of variable
interaction [Breiman et al., 1984; Province et al.,
2001]. Thus, these models may uncover interactions
among factors that do not exhibit strong marginal
effects, without demanding a pre-specified model
[McKinney et al., 2006]. Additionally, tree methods
are suited to dealing with certain types of genetic
heterogeneity, since splits near the root node define
separate model subsets in the data. RFs capitalize on
the solid benefits of decision trees and have
demonstrated excellent predictive performance
when the forest is diverse (i.e. trees are not highly
correlated with each other) and composed of
individually strong classifier trees [Breiman, 2001;
Bureau et al., 2005].
FOCUSED INTERACTION TESTING FRAME-
WORK (FITF)
The FITF was recently developed to detect
epistatic interactions that predict disease risk. De-
tails of the FITF algorithm and software can be
found in Millstein et al. [2006]. FITF is a modification
of the interaction testing framework (ITF) method,
which pre-screens all possible gene sets to focus on
those that potentially are the most informative and
reduce the multiple testing problem by reducing the
number of statistical tests performed. FITF has been
shown to outperform MDR when interactions
involved additive, recessive, or dominant genes
[Millstein et al., 2006]. Additionally, FITF has
successfully identified a multi-locus model that
predicts childhood asthma [Millstein et al., 2006].
The ITF strategy performs a series of LR analyses
in incremental stages, where the highest-order
interaction parameter considered increases at each
subsequent stage. In stage one, the main effect of
each genetic variant is considered, in stage two, all
pair-wise combinations are tested, in stage three all
three-way interactions are tested, etc. In order to
avoid re-testing the same effects, if a variant or
multi-locus combination is declared significant in an
earlier stage, those variants are not re-tested in
subsequent stages. The overall type I error is
controlled by dividing the overall desired a level
by the number of stages and allocating this adjusted
a level to each stage. Within each stage, the
significance threshold is adjusted by controlling the
false discovery rate [Benjamini and Hochberg, 1995].
This approach tests for interactions in the presence
of no main effects.
The FITF algorithm modifies the ITF approach to
reduce the overall number of variants tested with an
initial filter process. A j^ goodness-of-fit statistic
that compares the observed with the expected
Bayesian distribution of multi-locus genotype com-
binations in a combined case-control population is
used in a prescreening initial stage. This statistic,
referred to as the chi-square subset (CSS), has the
form:
CSS =
Ej
E(n{)
where M, is the observed number of subjects
(regardless of case/control status) in the z'th geno-
type group and r is the total number of genotype
groups. The expected M,-, noted as £(«,-), is estimated
based on the sample marginal genotype frequencies
of each gene.
LOGISTIC REGRESSION (LR)
LR is a derivative of linear regression that fits a
function to continuous or discrete independent vari-
ables based on a dichotomous dependent variable
[Hosmer and Lemeshow, 2000]. LR uses a transforma-
tion of the logistic distribution to develop a function
based on the independent variables. The logistic
distribution provides a flexible platform with a
clinically meaningful interpretation. The distribution
is transformed by the logit function, allowing tradi-
tional regression techniques to be applied. Using this
formulation, the predicted values from regression are
dichotomous with binomially distributed errors. Once
a regression function has been derived using iterative
fitting techniques, least squares is used to minimize
errors, based on the binomial distribution and is given
as a logarithmic transformation of the maximum
likelihood, called a log-likelihood. Like linear regres-
sion, partial derivatives of the likelihood function are
evaluated to minimize error.
Previous
TOC
Genet. Epidemiol.
-------
772
Motsinger-Reif et al.
One of the most common procedures for variable
selection in a LR analysis is step-wise logistic
regression (step LR) [Hosmer and Lemeshow,
2000]. In the step-wise procedure, each variable is
tested for independent effects, and those variables
with significant effects are included in the model. In
a second step, interaction terms of those variables
with significant main effects are included, and
significant effects are included in the model.
LR is a de facto standard for traditional association
studies. Using independent variables to predict a
dichotomous dependent variable, LR by definition
lacks the ability to characterize purely interactive
effects. Only variables that contain an independent
main effect will be included in the final model. To
properly evaluate non-linear purely interactive
effects, combinations of variables must be encoded
as a single variable for inclusion in the analysis. Such
an encoding scheme can be computationally expen-
sive, depending on the number of variables used.
In the current study, we use stepLR to assess its
performance to perform both variable selection and
statistical modeling tasks. Additionally, since LR
modeling is the standard in genetic epidemiology,
we use explicit LR (eLR) as a "positive control" to
assess the strength of the genetic signal in the
simulated data. In eLR, the known simulated effect
is explicitly modeled. So for example, for a two-locus
interactive model, an LR equation would be built
with parameters for the independent effects of both
loci and the interaction term.
DATA SIMULATION
Simulated data sets that exhibit both main effects
and gene-gene interactions were generated in order
to compare the performance of the above methods in
a variety of situations. Multiple disease models were
generating with varying allele frequencies, herit-
ability, odds ratios, and number of functional
polymorphisms.
We simulated case-control data with both main
effects and purely epistatic models. We generated
data sets demonstrating main effects to test the
performance of all the methods studied to perform
simpler variable selection tasks. We also simulated
models that exhibit epistatic effects in the absence of
main effects. Epistasis occurs when the combined
effect of two or more genes on a phenotype could
not be predicted from their individual genotypes.
Epistasis is increasingly accepted as a common
feature of the genetic architecture of common,
complex disease [Moore, 2003] and presents a
difficult analytical challenge. For this study, epistatic
disease models with no main effects were simulated
to challenge the analytical methods. While an
effectively infinite range of genetic models could
be simulated, simple main effect and purely epistatic
models were chosen to represent extremes from an
analytical point of view.
Two different minor allele frequency scenarios
were chosen: 0.2 and 0.4. Two different values of
heritability were used in our simulations: 1 and 5%.
Roughly, heritability in the broad sense describes the
proportion of the total phenotype/disease that is
due to genetic effects. More specifically, the exact
heritability calculations used can be found in
Culverhouse et al. [2002]. A range of odds ratio
values was selected for each heritability value. For
the lower heritability models, odds ratios of 1.2, 1.4,
1.6, 1.8, and 2.0 were simulated. For the higher
heritability models, odds ratios of 2.0, 2.5, 3.0, 3.5,
and 4.0 were simulated. This range of heritability
and odd ratio values were chosen to represent a
"worst-case scenario" of common diseases to test the
lower limits of the methods. By testing these
extremely low-signal models, the study is able to
distinguish differences in the lower performance
limits of the analytical methods. It is assumed that
any method that can find such minimal effects
should have greater posterior probabilities to find
more substantial effects. Additionally, null data
(with no disease model) were simulated to estimate
the type I error rates of each method.
A range of functional interacting loci was simu-
lated using penetrance functions. Penetrance func-
tions define the probability of disease given a
particular genotype combination to model the
relationship between genetic variations and disease
risk. The range of functional loci selected included
one-, two-, and three -locus interaction models. All
penetrance functions used in the current study are
available from the authors on request.
Data sets were simulated using software described
by Moore et al. [2004]. All possible combinations of
allele frequency, heritability/odds ratio, and inter-
acting loci were modeled, resulting in 30 models.
One hundred data sets were generated for each
model, resulting in 3,000 data sets. Each data set
included 1,000 total individuals—500 cases and 500
controls. While the number of functional loci present
varied between models, the total number of single
nucleotide polymorphisms (SNPs) were constant in
each data set (100 SNPs per individual).
DATA ANALYSIS
All replicates of all disease models were analyzed
by each of the following methods: MDR, GENN, RF,
FITF, stepLR, and eLR. The number of times each
method identified the correct simulated model
(including all correct loci and only correct loci) as
the final or best model across the 100 replicates was
used to estimate the posterior probability of identi-
fying the simulated model. We have chosen to use
posterior probability, rather than empirical power to
evaluate performance in these simulations, because
Genet. Epidemiol.
-------
Analytical Methods for Genetic Epidemiology
773
each of the methods used for this comparison has
different prior probabilities for detecting interaction
effects. For example, the prior probability for step
LR to detect a purely epistatic two-locus model is
extremely small (close to zero) because there are no
main effects to condition on. However, the prior
probability for MDR is quite high because MDR
performs an exhaustive search. For all methods
except RF, the final model selected must be
statistically significant to contribute to the posterior
probability estimate. For MDR and GENN, permuta-
tion testing was used to empirically estimate
statistical significance, with a P-value of 0.05
considered significant. For the eLR, step LR, and
FITF analyses, a P-value of 0.05 was used in the
analysis to determine significance. Because RF is
most commonly used as a filter, and not for strict
association testing, no estimate of statistical signifi-
cance was used in the RF posterior probability
estimations.
While we understand this performance assess-
ment is strict, we feel it is the fairest comparison
across many methods. Also, since the end-goal of
many analyses is to find the best model that predicts
disease status, we feel this is a practical assessment.
Null data were used to estimate the type I error rate
of each method. Alpha was set to 0.05 for each
method, and the type I error rate was estimated as
the number of times a significant result was found
across 100 replicates. Again, because RF is most
commonly used as a filter, no false-positive rate was
estimated.
All analyses were performed with a common
user's perspective in mind. Configuration para-
meters for each software program were set according
to the author default settings or recommendations.
MDR analysis was performed using a Linux
version of the MDR software (compiled and bench-
marked on a PC with a 600 MHz Pentium-Ill
running Red Hat 2.2.5-15, written in C and compiled
with the GNU C compiler). MDR software is
currently distributed in a JAVA version with a
graphical user interface or in a C library. The most
current open-source versions are available at
www.epistasis.org/mdr.html. MDR has also been
added to Weka-CG, which is available from the same
Web site. MDR evaluated all possible single-locus
through four-locus models, and a single final model
was chosen using the heuristic described above. No
level of interaction was pre-selected for final model
selection. Final models containing too few or too
many loci based on the simulated model did not
contribute to the probability estimate. This strict
method of final model selection should be consid-
ered in interpreting the results presented below.
GENN analysis was performed using a Linux
software package that is available from the authors
on request. The GA used to evolve the binary string
that is transcribed into an NN has the following
parameters in the current implementation: crossover
rate = 0.9, mutation = 0.01, population = 200 per
deme, demes = 10, max generations = 200, codon
size = 8, GE wrapping count = 2, min chromosome
size (in terms of codons) = 50, max chromosome
size = 1,000, selection = tournament, and sensible
initialization depth = 10. The island model of paral-
lelization is used, where the best individual is
passed to each of the other processes after every 25
generations [Cantu-Paz, 2000], to prevent stalling in
local minima. The genome was derived from GAlib
(version 2.4.5), and a typical GA one-point crossover
of linear chromosomes is used [Motsinger et al.,
2006b]. For the results presented below, final model
selection was performed based on cross-validation
consistency as described above and only models
with the correct number of loci, and only correct loci
included in the model counted toward the prob-
ability estimate. Incomplete models (models that did
not include all the simulated disease loci) or models
containing the simulated disease model plus addi-
tional noise loci were not included in the probability
estimated. Again, this strict performance assessment
should be kept in mind when interpreting the
results.
RF analysis was performed using the freely
available R package randomForest [Ihaka and
Gentleman, 1996; R Development Core Team,
2006]. This package is based on the original Fortran
code available in Breiman and Cutler [2004]. For
each of the data sets, forests comprising 10,000 trees
were grown. Variable importance was calculated
using the out-of-bag permutation test. The relative
importance (rank) of variables was determined from
the mean decrease in Gini index using the out-of-bag
permutation testing procedure [Breiman et al., 1984].
Again, a very strict performance assessment is used
in the results presented below. The simulated locus/
loci must be the top ranking locus/loci in the RF
analysis to contribute to the probability estimate. RF
is often used as a filter approach, where a pre-
specified number of the top loci will be used in a
second stage of association analysis. In this study, RF
is being used for association testing and not as a
filter. This use should be kept in mind in interpreting
the results.
FITF software is freely available at http://
hydra.usc.edu/fitf and was originally written for
Windows. Because of the computational burden of
running large-scale simulation studies in Windows,
source code was requested and kindly shared by Dr.
W. James Gauderman. The configuration parameters
used in the current study are as recommended in the
software instructions, and include CSS cutoff 2 = 3,
CSS Cutoff 3 = 6, 0^ = 0.016667, a2 = 0.016667,
a3 = 0.016667.
Both stepLR and eLR analyses were performed
using SAS v9.1 commercial software. For the eLR
analysis, because the known simulated model was
Genet. Epidemiol.
-------
774
Motsinger-Reif et al.
explicitly modeled, the probability results presented
are estimated as the number of times the eLR model
was statistically significant at P<0.05 across the 100
replicates.
RESULTS
The type I error results demonstrate that each
method had nominal type I error rates. The type I
error for each method was as follows: MDR—3%,
FITF—3%, GENN—3%, eLR-^%, and sLR—2%.
Table I summarizes the posterior probability
results for all six methods for the lower heritability
models (1%). As these results show, these simula-
tions really do test the lower limits of all six
methods, as the genetic effects in this data are
extremely small. The low posterior probability of
even eLR for the two- and three-locus interaction
models demonstrates the difficulty of statistically
modeling such small effects. The other methods
must also perform variable selection along with
TABLE I. Posterior probability results from 1%
heritability models
Posterior probability to detect
simulated model (%)
Number of
disease loci
1
2
3
OR
1.2
1.4
1.6
1.8
2
1.2
1.4
1.6
1.8
2
1.2
1.4
1.6
1.8
2
MAP
0.2
0.4
0.2
0.4
0.2
0.4
0.2
0.4
0.2
0.4
0.2
0.4
0.2
0.4
0.2
0.4
0.2
0.4
0.2
0.4
0.2
0.4
0.2
0.4
0.2
0.4
0.2
0.4
0.2
0.4
GPNN
6
6
36
40
69
84
80
90
99
99
0
0
0
4
10
4
5
24
30
14
0
0
0
0
0
0
0
0
0
0
MDR
1
4
30
29
67
74
85
86
94
93
0
0
2
3
8
8
33
20
46
22
0
0
0
0
0
0
4
2
0
0
FITF
0
0
0
0
21
4
56
34
91
49
0
0
0
0
12
1
35
1
37
0
0
0
0
0
0
0
2
0
0
0
RF
4
7
37
34
73
86
88
96
98
99
0
0
0
0
0
0
0
1
0
0
0
0
0
0
0
0
0
0
0
0
stepLR
42
52
68
69
95
93
88
97
85
88
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
eLR
48
52
84
76
100
96
100
100
100
100
24
44
10
33
96
98
100
81
98
96
2
6
8
6
0
3
15
17
0
3
statistical modeling, so lower performance as
compared to eLR is not surprising. For the lowest
effect single-locus models (OR 1.2-1.6), stepLR
outperforms the other four computational ap-
proaches and has comparable performance to eLR.
For the higher effect single-locus models (OR
1.8-2.0), all the methods perform very well. In
particular, GENN and RF have the highest perfor-
mance. The performance of those two methods is
comparable to that of eLR.
For the lowest effect size two-locus models, all five
computational approaches have extremely low
posterior probabilities. For the higher effect size
models, GENN, MDR, and FITF show an improve-
ment as compared to stepLR and RF. The low
performance of all methods is not surprising given
the very small effect sizes and the added challenge
of identifying interacting variables as opposed to
main effects. The posterior probabilities to detect the
three-locus models with such small effect sizes are
near or at zero for all of the methods. Even the
performance of eLR is extremely low.
Table II gives the results of the analyses of the
higher heritability genetic models. For the single-
locus models, all methods have reasonably high
performance. Notably, stepLR has the lowest poster-
ior probability of the five computational methods.
RF has the highest probability (100%) of the
computational methods (tied with eLR) to detect
all single-locus models. GENN has the next highest
performance, and the performance of MDR is pretty
comparable. FITF has a much lower probability than
GENN, MDR, or RF for the very weakest models
(OR 1.2).
For the two-locus models, the posterior probability
of correctly detecting the interaction both stepLR
and RF is minimal (approaching or at zero). This is
not surprising, since both are dependent on at least a
small marginal main effect to detect interactions,
and these models are purely epistatic. GENN, MDR,
and FITF all have a reasonably high probability of
detecting the two-locus interactions. MDR has the
highest performance across most models, and FITF
has the lowest of these three methods. For the three-
locus models, the only method with a reasonable
posterior probability to detect the interaction is
MDR. Notably, the performance of MDR is much
higher than even that of eLR.
In summary, as expected, the relative performance
of each method is context dependent. Figures 4-6
graphically summarize these results. These figures
present box plots of the distributions of posterior
probability results (presented in Tables I and II) over
a range of MAP and OR. Overall, MDR had
consistently high performance in main effect, two-
locus, and three-locus models. GENN had high
performance in main effect and two-locus models,
but had poor posterior probabilities to detect three-
locus interactions. FITF had relatively high perfor-
Genet. Epidemiol.
-------
Analytical Methods for Genetic Epidemiology
775
TABLE II. Posterior probability results from 5%
Posterior Probabilities to Detect Two-Locus Models
heritability
models
Posterior
probability
simulated
Number of
disease loci
1
2
3
OR
2
2.5
3
3.5
4
2
2.5
3
3.5
4
2
2.5
3
3.5
4
MAP
0.2
0.4
0.2
0.4
0.2
n A.
U.'i
0.2
0.4
0.2
0.4
0.2
0.4
0.2
0.4
0.2
0.4
0.2
0.4
0.2
0.4
0.2
0.4
0.2
0.4
0.2
0.4
0.2
0.4
0.2
0.4
GPNN
95
99
99
99
99
QQ
"o
99
100
100
100
10
62
71
82
92
95
95
97
95
73
0
0
2
0
6
1
0
o
0
0
MDR
94
93
98
98
96
01
"1
93
99
95
98
60
82
93
93
95
100
93
96
98
83
14
15
71
45
51
80
62
95
14
93
modei
FITF RF
84
48
98
94
98
QQ
""
97
98
97
99
0
10
68
72
82
96
90
85
96
48
7
1
38
22
1
0
1
o
0
3
100
100
100
100
100
1 nn
1UU
100
100
100
100
52
2
26
1
6
4
9
1
5
1
0
0
0
0
0
0
0
o
0
0
to detect
(%)
stepLR
88
88
93
82
95
QQ
OO
92
91
92
87
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
o
0
0
eLR
100
100
100
100
100
1 nn
1UU
100
100
100
100
99
99
100
100
100
100
100
100
100
100
18
21
19
21
16
3
30
13
17
2
O
° '
0
1 3-
m
if-
* 8-
o -
A A ^ i
n P t
m
li.J
T
•
•
• _
GENN MDR FITF RF stepLR eLR GENN MDR FITF RF stepLR eLR
1 % Heritability 5% Heritability
Fig. 5. Posterior probabilities to detect two-locus epistatic
models.
Posterior Probabilities to Detect Three-Locus Models
1
_S-
$
Is-
£
°
r
i
*a-
o -
'
'
r*^ t
_ A J!_ _ _ 9 ±. I
1
T
; t
1 -L
GENN MDR FITF RF stepLR eLR GENN MOR FITF RF stepLR eLR
l« Heritability 5% Heritability
Fig. 6. Posterior probabilities to detect three-locus epistatic
models.
Posterior Probabilities to Detect Single-Locus Models
T T
GENN MDR FITF RF stepLR eLR GENN MDR FITF RF stepLR eLR
1 % Heritability 5% Heritability
Fig. 4. Posterior probabilities to detect single-locus models.
mance in detecting main effects and two-locus
interactions, but consistently had lower performance
than MDR and GENN. RF had extremely high
posterior probabilities to detect main effects, but
was limited in the case of purely epistatic models.
StepLR had high performance in detecting main
effects, but was also unable to detect interactions in
the absence of main effects.
DISCUSSION
The results of this study demonstrate the utility of
the methods investigated and emphasize the im-
portance of continued methods development. For
the smallest effect sizes simulated, none of the
methods had decent posterior probabilities to detect
interactive effects. Also, these simulations did not
try to model more complex scenarios, like genetic
heterogeneity or phenocopy. The performance of
each of these methods should be evaluated in
increasingly complex models.
The success of MDR in all the models simulated is
not surprising, given its many previous successes in
both real and simulated data [Agirbasli et al., 2006;
Cho et al., 2004; Hsieh et al., 2006; Motsinger and
Previous
TOC
Genet. Epidemiol.
-------
776
Motsinger-Reif et al.
Ritchie, 2006b; Ritchie et al., 2001, 2003a]. The
success of GENN for main effect and two-locus
models was also expected, given its previous
successes in real and simulated data [Motsinger
et al., 2006a,b,c; Motsinger-Reif et al., 2008a], but its
poor performance with the three-locus models was
surprising. This highlights a disadvantage of an
evolutionary computation approach in exploring
purely epistatic models—it is much less likely that
three loci will be stochastically assembled into a
model to evaluate than two loci. The exhaustive
search approach of MDR avoids this risk, but is
highly computationally intensive. In relatively small
data sets like those simulated here, the computa-
tional burden is not a limitation, but as genotyping
technology pushes the field toward genome-wide
association as the norm, this approach will become
infeasible without decreasing the number of vari-
ables using a filtering technique. The evolutionary
computation approach will be much more realistic
for such data sets; however, detecting purely
epistatic models will be a daunting task. It was
encouraging that the GENN results in the single and
two-locus models were comparable to MDR, because
of its computational advantage. Additionally, there
is evidence that the non-exhaustive search strategy
of GENN may have other advantages over MDR,
especially in the case of genetic heterogeneity [Mot-
singer-Reif et al., 2008a].
Both GENN and MDR outperformed the FITF
approach in epistatic models. The increase in
probabilities for these two methods may be due to
the many corrections for multiple testing used in
FITF. Because GENN and MDR both utilize permu-
tation distributions for significance testing, correc-
tion for multiple testing is unnecessary. While the
filter stage of FITF does reduce the number of tests
performed with the ITF strategy, there are still a very
large number of tests that are corrected for. This
correction may limit the performance of FITF in
comparison to GENN and MDR.
Both RF and stepLR had very high posterior
probabilities to detect main effects, but were both
unable to detect purely epistatic models. There is a
solid theoretical explanation for these results, since
both require marginal main effects to perform
variable selection tasks. Future extensions/modifi-
cations of these approaches should consider this
limitation and modify the variable selection process
to capture pure interactions. Some groups have in
fact begun to make modifications in this way
[Bureau et al., 2005].
The results of the current study were presented as
posterior probabilities to detect the simulated model,
as an empirical estimate of the performance of each
method. While these results allow a comparison of
the posterior probabilities between methods, they do
not reflect the differences in the prior probabilities of
each method. The prior probabilities of each method
differ based on the model building process of each.
For example, in the case of stepwise LR, the prior
probability of selecting any multi-locus interaction
model under the null is very close to zero. This is
due to the hierarchical model building process,
where only variables with significant main effects
are even considered in the evaluation of potential
interactions. MDR, on the other hand, has a
theoretically higher prior probability of detecting
multi-locus interactions because of the exhaustive
search process used ensures higher-order models
will be evaluated.
Bayes factors comparisons [Goodman, 1999] are a
commonly used tool to combine prior and posterior
information in a ratio that provides evidence in favor
of one model specification versus another. Such
comparisons would be useful in interpreting the
results of the current study given the important
differences in the prior probabilities of each method,
but unfortunately a formal Bayes Factor comparison
is not possible in the current study. Calculations
result in effectively undefined Bayes Factors (data
not shown) due to the extremely small prior
probabilities under the null of each of the methods.
The very general low false-positive rates (discussed
previously) of each of the methods reflect the very
low prior probability of detecting specific genetic
models in null data.
While formalization of these differences is not
possible, it is important to remember these differ-
ences when choosing an analysis strategy. If it is
hypothesized that the underlying genetic etiology
includes purely epistatic models, step LR would be a
theoretically inappropriate choice. MDR or GENN
may be better selections. If it is hypothesized that the
underlying genetic etiology includes interactions
with strong main effects, a number of analytical
methods may be appropriate. If it is hypothesized
that main effects result in the phenotype of interest,
the list of appropriate methods extends further.
While the list of computational methods compared
in the current study is not exhaustive, this study
compares and contrasts several important novel
analytical approaches in genetic epidemiology. By
thoroughly understanding the strengths and weak-
nesses of each method, researchers can choose
appropriate analytical tools for their particular data
and questions.
ACKNOWLEDGMENTS
We to thank Dr. W. James Gauderman for provid-
ing code for the FITF algorithm. These results do not
reflect the official policy of the US Environmental
Protection Agency.
Genet. Epidemiol.
-------
Analytical Methods for Genetic Epidemiology
777
REFERENCES
Agirbasli D, Agirbasli M, Williams SM, Phillips III JA. 2006.
Interaction among 5,10 methylenetetrahydrofolate reductase,
plasminogen activator inhibitor and endothelial nitric oxide
synthase gene polymorphisms predicts the severity of cor-
onary artery disease in Turkish patients. Coron Artery Dis
17:413-117.
Bellman R. 1961. Adaptive Control Processes. Princeton: Princeton
University Press.
Benjamini Y, Hochberg Y. 1995. Controlling the false discovery
rate: a practical and powerful approach to multiple testing. J R
Stat Soc Ser B (Methodological) 57:289-300.
Brassat D, Motsinger AA, Caillier SJ, Erlich HA, Walker K, Steiner
LL, Cree BA, Barcellos LF, Pericak-Vance MA, Schmidt S,
Gregory S, Hauser SL, Haines JL, Oksenberg JR, Ritchie MD.
2006. Multifactor dimensionality reduction reveals gene-gene
interactions associated with multiple sclerosis susceptibility in
African Americans. Genes Immun 7:310-315.
Breiman L. 2001. Random forests. Mach Learn 45:5-32.
Breiman L, Cutler A. 2004. Random forests. http://www.Stat.-
Berkeley.Edu/~Breiman/RandomForests/Cc_Home Htm
Breiman L, Friedman JH, Olshen RA, Stone CJ. 1984. Classification
and Regression Trees. New York: Chapman & Hall.
Bureau A, Dupuis J, Falls K, Lunetta KL, Hayward B, Keith TP,
Van EP. 2005. Identifying SNPs predictive of phenotype using
random forests. Genet Epidemiol 28:171-182.
Cantu-Paz E. 2000. Efficient and Accurate Parallel Genetic
Algorithms. Boston: Kluwer Academic Publishers.
Cho YM, Ritchie MD, Moore JH, Park JY, Lee KU, Shin HD, Lee
HK, Park KS. 2004. Multifactor-dimensionality reduction
shows a two-locus interaction associated with Type 2 diabetes
mellitus. Diabetologia 47:549-554.
Culverhouse R, Suarez BK, Lin J, Reich T. 2002. A perspective on
epistasis: limits of models displaying no main effect. Am J
Hum Genet 70:461-171.
Goodman S. 1999. Toward evidence-based medical statistics. 2:
The Bayes factor. Ann Intern Med 130:1005-1013.
Hahn LW, Ritchie MD, Moore JH. 2003. Multifactor dimension-
ality reduction software for detecting gene-gene and gene-
environment interactions. Bioinformatics 19:376-382.
Hardison NE, Fanelli TJ, Reif DM, Ritchie MD, Motsinger-Reif
AA. 2008. Balanced accuracy makes grammatical evolution
neural networks robust to class imbalance. Proceedings of the
Genetic and Evolutionary Algorithm Conference. 2008, in
press.
Hastie T, Tibshirani R, Friedman JH. 2001. The Elements Of
Statistical Learning. Springer Series in Statistics. Basel: Springer.
Hosmer DW, Lemeshow S. 2000. Applied Logistic Regression.
New York: John Wiley & Sons Inc.
Hsieh CH, Liang KH, Hung YJ, Huang LC, Pei D, Liao YT, Kuo
SW, Bey MS, Chen JL, Chen EY 2006. Analysis of epistasis for
diabetic nephropathy among type 2 diabetic patients. Hum
Mol Genet 15:2701-2708.
Ihaka R, Gentleman R. 1996. R: a language for data analysis and
graphics. J Comput Graphical Stat 5:299-314.
Kooperberg C, Ruczinski I, LeBlanc ML, Hsu L. 2001. Sequence
analysis using logic regression. Genet Epidemiol 21 :S626-S631.
Lunetta KL, Hayward LB, Segal J, Van EP. 2004. Screening large-
scale association study data: exploiting interactions using
random forests. BMC Genet 5:32.
Mamyrova G, O'hanlon TP, Monroe JB, Carrick DM, Malley JD,
Adams S, Reed AM, Shamim EA, James-Newton L, Miller FW,
Rider LG. 2006. Immunogenetic risk and protective factors for
juvenile dermatomyositis in Caucasians. Arthritis Rheum
54:3979-3987.
McKinney BA, Reif DM, Ritchie MD, Moore JH. 2006. Machine
learning for detecting gene-gene interactions: a review. Appl
Bioinformatics 5:77-88.
Millstein J, Conti DV, Gilliland FD, Gauderman WJ. 2006. A testing
framework for identifying susceptibility genes in the presence
of epistasis. Am J Hum Genet 78:15-27.
Moore JH. 2003. The ubiquitous nature of epistasis in determining
susceptibility to common human diseases. Hum Hered 56:73-82.
Moore JH, Ritchie MD. 2004. STUDENTJAMA. The challenges of
whole-genome approaches to common diseases. J Am Med
Assoc 291:1642-1643.
Moore JH, Williams SM. 2005. Traversing the conceptual divide
between biological and statistical epistasis: systems biology
and a more modern synthesis. Bioessays 27:637-646.
Moore J, Hahn L, Ritchie M, Thornton T, White BC. 2004. Routine
discovery of high-order epistasis models for computational
studies in human genetics. Appl Soft Comput 4:79-86.
Motsinger AA, Ritchie MD. 2006a. Multifactor dimensionality
reduction: an analysis strategy for modeling and detecting
gene-gene interactions in human genetics and pharmacoge-
nomics studies. Hum Genomics 2:318-328.
Motsinger AA, Ritchie MD. 2006b. The effect of reduction in cross-
validation intervals on the performance of multifactor dimen-
sionality reduction. Genet Epidemiol 30:546-555.
Motsinger AA, Dudek SM, Hahn LW, Ritchie MD. 2006a.
Comparison of Neural Network Optimization Approaches
for Studies of Human Genetics. Lecture Notes in Computer
Science, vol. 3907. Berlin: Springer, p 103-114.
Motsinger AA, Hahn LW, Dudek SM, Ryckman KK, Ritchie MD.
2006b. Alternative cross-over strategies and selection techni-
ques for grammatical evolution optimized neural networks.
Proceedings of the Genetic and Evolutionary Algorithm
Conference, vol. 1, p. 947-949.
Motsinger AA, Reif DM, Dudek SM, Ritchie MD. 2006c. Under-
standing the evolutionary process of grammatical evolution
neural networks for feature selection in genetic epidemiology.
IEEE Trans Comput Intell Bioinformatics Comput Biol 1-8.
Motsinger AA, Brassat D, Caillier SJ, Erlich HA, Walker K, Steiner
LL, Barcellos LF, Pericak-Vance MA, Schmidt S, Gregory S,
Hauser SL, Haines JL, Oksenberg JR, Ritchie MD. 2007.
Complex gene-gene interactions in multiple sclerosis: a multi-
factorial approach reveals associations with inflammatory
genes. Neurogenetics 8:11-20.
Motsinger-Reif AA, Dudek SM, Hahn LW, Ritchie MD. 2008a.
Comparison of approaches for machine-learning optimization
of neural networks for detecting gene-gene interactions in
genetic epidemiology. Genet Epidemiol Feb 8 [Epub ahead of
print].
Motsinger-Reif AA, Fanelli TJ, Davis AC, Ritchie MD. 2008b.
Grammatical evolution neural networks is robust to common
types of noise common to genetic epidemiology studies. BMC
Res Notes, in press.
Nelson MR, Kardia SL, Ferrell RE, Sing CF. 2001. A combinatorial
partitioning method to identify multilocus genotypic partitions
that predict quantitative trait variation. Genome Res
11:458-470.
O'Neill M, Ryan C. 2001. Grammatical evolution. IEEE Trans Evol
Comput 5:349-357.
O'Neill M, Ryan C. 2003. Grammatical Evolution: Evolutionary
Automatic Programming in an Arbitrary Language. Boston:
Kluwer Academic Publishers.
Genet. Epidemiol.
-------
778
Motsinger-Reif et al.
Province MA, Shannon WD, Rao DC. 2001. Classification methods
for confronting heterogeneity. Adv Genet 42:273-286.
R Development Core Team. 2006. R: a language and environment
for statistical computing. http://www R-Project Org Vienna,
Austria.
Ritchie MD. 2005. Bioinformatics approaches for detecting gene-
gene and gene-environment interactions in studies of human
disease. Neurosurg Focus 19:E2.
Ritchie MD, Hahn LW, Roodi N, Bailey LR, Dupont WD, Parl FF,
Moore JH. 2001. Multifactor-dimensionality reduction reveals
high-order interactions among estrogen-metabolism genes in
sporadic breast cancer. Am J Hum Genet 69:138-147.
Ritchie MD, Hahn LW, Moore JH. 2003a. Power of multi-
factor dimensionality reduction for detecting gene-gene inter-
actions in the presence of genotyping error, missing data,
phenocopy, and genetic heterogeneity. Genet Epidemiol
24:150-157.
Ritchie MD, White BC, Parker JS, Hahn LW, Moore JH. 2003b.
Optimization of neural network architecture using genetic
programming improves detection and modeling of gene-gene
interactions in studies of human diseases. BMC Bioinformatics
4:28.
Ritchie MD, Carillo MW, Wilke RA. 2005. Computational
approaches for pharmacogenomics. Pac Symp Biocomput
245-247.
Robnik-Sikonja M. 2004. Improving random forests. Machine
learning: Ecml 2004. Proceedings 3201:359-370.
Sabbagh A, Darlu P. 2006. Data-mining methods as useful tools for
predicting individual drug response: application to CYP2D6
data. Hum Hered 62:119-134.
Tahri-Daizadeh N, Tregouet DA, Nicaud V, Manuel N, Cambien F,
Tiret L. 2003. Automated detection of informative combined
effects in genetic association studies of complex traits. Genome
Res 13:1952-1960.
Templeton A. 2000. Epistasis and Complex Traits. Oxford: Oxford
University Press, p 41-57.
Thornton-Wells TA, Moore JH, Haines JL. 2004. Genetics, statistics
and human disease: analytical retooling for complexity. Trends
Genet 20:640-647.
Velez D, White BC, Motsinger AA, Bush WS, Ritchie MD, Moore
JH. 2007. A balanced accuracy function for epistasis modeling
in imbalanced datasets using multifactor dimensionality
reduction. Genetic Epidemiology 31:306-315.
Zhu J, Hastie T. 2004. Classification of gene microarrays by
penalized logistic regression. Biostatistics 5:427-443.
Genet. Epidemiol.
-------
© 2008 Wiley-Liss, Inc.
Birth Defects Research (Part B) 83:522-529 (2008)
Review Article
A Lifestage Approach to Assessing Children's Exposure
Elaine A. Cohen Hubal,1* Jacqueline Moya,2 and Sherry G. Selevan3
1U.S. Environmental Protection Agency, Office of Research and Development, National Center for Computational Toxicology,
Research Triangle Park, NC
U.S. Environmental Protection Agency, Office of Research and Development, National Center for Environmental Assessment,
Washington, DC
3Consultant, Silver Spring MD, formerly at the U.S. Environmental Protection Agency, Office of Research and Development,
National Center for Environmental Assessment, Washington, DC
Understanding and characterizing risks to children has been the focus of considerable research efforts at the U.S.
Environmental Protection Agency (EPA). Potential health risks resulting from environmental exposures before
conception and during pre- and postnatal development are often difficult to recognize and assess because of a
potential time lag between the relevant periods of exposure during development and associated outcomes that may be
expressed at later lifestages. Recognizing this challenge, a lifestage approach for assessing exposure and risk is presented
in the recent EPA report titled A Framework for Assessing Health Risks of Environmental Exposures to Children (U.S. EPA,
2006). This EPA report emphasizes the need to account for the potential exposures to environmental agents during all
stages of development, and consideration of the relevant adverse health outcomes that may occur as a result of such
exposures. It identifies lifestage-specific issues associated with exposure characterization for regulatory risk assessment,
summarizes the lifestage-specific approach to exposure characterization presented in the Framework, and discusses
emerging research needs for exposure characterization in the larger public-health context. This lifestage approach for
characterizing children's exposures to environmental contaminants ensures a more complete evaluation of the potential
for vulnerability and exposure of sensitive populations throughout the life cycle. Birth Defects Res (Part B) 83:522-529,
2008. © 2008 Wiley-Liss, Inc.
INTRODUCTION
Potential health risks resulting from environmental
exposures before conception and during pre- and
postnatal development are often difficult to recognize
and assess due to a potential time lag between the
relevant timing of exposure and of outcomes that may be
expressed at any lifestage including those far removed
from that of exposure. Lifestages are defined here as
temporal stages (or intervals) of life that have distinct
anatomical, physiological, behavioral and/or functional
characteristics that contribute to potential differences in
environmental exposures (U.S. EPA, 2006). The consid-
eration of lifestage-specific periods of unique suscept-
ibility in relation to childhood activities, behaviors, and
intakes was recognized in the International Life Sciences
Institute (ILSI) 2001 workshop (Olin and Sonawane,
2002) and has been the focus of significant efforts in the
EPA Office of Research and Development.
The need for a lifestage approach to exposure and risk
assessment is highlighted in the recent EPA report
entitled A Framework for Assessing Health Risks of
Environmental Exposures to Children (hereinafter referred
to as the "Framework")(U.S. EPA, 2006). This report
followed a series of policy statements (U.S. EPA, 1995;
Executive Order, 1997) and regulatory statutes such as
the Food Quality Protection Act and the Safe Drinking
Water Act (U.S. 104th Congress, 1996aand 1996b)
explicitly requiring consideration of risks to children in
risk assessments conducted by the Agency. The Frame-
work emphasizes the need to account for potential
exposures to environmental agents during all stages of
development, and consideration of the relevant adverse
health outcomes that may occur as a result of such
exposures (Brown et al., this issue). A report developed
in parallel by the World Health Organization, Principles
for Evaluating Health Risks in Children Associated with
Exposure to Chemicals (WHO, 2006), also focuses on the
potential vulnerability of children to chemical exposures
and the potential for increased risk of adverse affects
from early-life exposure.
Further, as understanding improves for how to assess
and mitigate health risks resulting from exposures to
individual environmental pollutants, environmental
health scientists are turning their attention toward
characterizing relationships between multiple environ-
mental factors and complex diseases such as asthma and
obesity (Schwartz, 2006; U.S. EPA 2003a). Because
*Correspondence to: Elaine Cohen Hubal, U.S. Environmental Protection
Agency, National Center for Computational Toxicology, Mail Drop B205-
01, Research Triangle Park, NC 27711.
E-mail: hubal.elaine@epa.gov
Received 8 July 2008; Accepted 15 September 2008
Published online in Wiley InterScience (www.interscience.wiley.com)
DOI: 10.1002/bdrb.20173
-------
LIFESTAGE APPROACH TO ASSESSING CHILDRENS EXPOSURE
523
preconception, prenatal, infant, and childhood exposures
may impact both early lifestage health outcomes as well
as development of disease later in life, a lifestage
approach is required to identify and assess these
exposures. Carefully designed exposure studies and
multifactorial exposure analyses of one form or another
are required in order to conduct national-scale regula-
tory-based risk assessments, conduct community-based
risk screening and remediation, support epidemiology
studies investigating gene-environment interactions, and
characterize exposure and risk for public health tracking.
In this paper, lifestage-specific issues associated with
exposure characterization for regulatory risk assessment
are identified; the lifestage-specific approach to exposure
characterization that is presented in the Framework is
summarized; and emerging research needs for exposure
characterization in the larger public-health context are
discussed.
ROLE OF LIFESTAGE-SPECIFIC EXPOSURE
ASSESSMENT
Exposure characterization is the risk analysis step in
which human interaction with an environmental agent of
concern is evaluated (WHO, 2004). Exposure (sometimes
referred to as potential dose) is defined as the contact of
an individual or population with an agent of concern
(WHO, 2004). Exposure assessment is defined as the
process of estimating the magnitude, frequency, and
duration of an exposure, along with characteristics of the
exposed individual or population. Ideally, this process
will provide descriptions of the sources, pathways,
routes, and uncertainties associated with exposures of
interest (WHO, 2004). Sometimes exposure can be
measured directly, but more often exposure must be
estimated. Armstrong et al., (2000) note that the EPA
Guidelines for Exposure Assessment (U.S. EPA, 1992)
provide the approaches needed to assess exposures,
including those of children. But that due to potentially
unique exposure patterns that may result during devel-
opment, EPA guidance needs to be supplemented by
further commentary on application of exposure assess-
ment approaches to children. In addition, the Framework
(U.S. EPA, 2006) notes that children may be more or less
vulnerable than adults, but without data on exposure
and response and without systematic evaluation of these
data, determining which lifestage may be most vulner-
able is challenging. The Framework provides information
on application of traditional exposure characterization
approaches over the course of development as well as
references to current lifestage-specific data for exposure
factors and exposure scenarios.
An important objective of the Framework is to provide
a holistic approach for assessing lifestage-specific risks.
As such, significant emphasis is placed on the need for
iteration between exposure, dose-response, and hazard
analysis steps of the risk assessment process, as depicted
in Figure 1. The amount of agent that enters an
individual after crossing an exposure surface (e.g.,
surface of respiratory tract, gastrointestinal tract, skin,
placental barrier) is referred to as the dose (WHO, 2004).
Clearly, not all exposures will result in a significant dose
(e.g., contaminated hands may be washed before dermal
absorption or oral transfer can occur). Yet, it is the dose of
the toxic moiety at the target tissue that will ultimately
Lifestage-Specific Problem Formulation
Lifestage-Specific Analysis
Lifestage-Specific
Exposure Characterization
Evaluation of Available Exposure Data
j
Lifestage-Specific Exposure Analysis
Variability, Sensitivity, and Uncertainty
Iteration with Hazard and Dose-!
Characterization
x
Lifestage-Specific Exposure Characterization
Lifestage- Lifestage-
Specific Specific
Lifestage-Specific Risk Characterization
Risk Communication/Management
Fig. 1. Flow diagram for lifestage-specific exposure character-
ization. [CAP] Using the lifestage-specific exposure information
identified in problem formulation, exposure is estimated using a
tiered approach. The lifestage-specific exposure is characterized
by discussing the variability and uncertainty in the results. Key
sources of variability and uncertainty can be assessed using
sensitivity analysis. Iteration with hazard characterization and
dose-response characterization (illustrated by dashed arrows)
occurs throughout this process to ensure that critical windows of
exposure are considered.
cause health effects. The approach outlined in the
Framework encourages evaluation of the potential for
adverse health outcomes during all developmental life-
stages, based on knowledge of exposure, critical win-
dows of development for organ systems, MOAs,
anatomy, physiology, and behavior that can affect
external exposure and internal dose metrics. The primary
purpose of a lifestage-specific exposure characterization
is a detailed description of the potential for exposure
during preconception or developmental lifestages that is
relevant for assessing potential health risks at these and
subsequent lifestages.
EXPOSURE CHARACTERIZATION IN
FRAMEWORK
Overview of Process
Exposure characterization (Fig. 1) begins in the
problem formulation phase with identification of poten-
tial sources, pathways, and scenarios. Pathways and
media that may be relevant for the various lifestages are
depicted in Figure 2. For example, depending on the life
stage, exposure media can include amniotic fluid, human
milk, air, water, soil, and food. Problem formulation sets
the stage to guide collection of available exposure data
and other required information for exposure
Birth Defects Research (Part B) 83:522-529, 2008
Previous
TOC
-------
524
COHEN HUBAL ET AL.
Prenatal:
All exposures to the fetus occur
transplacentally or via physical
factors. The mother's exposure to
environmental media can be a
significant source of exposure for
environmental media for the fetus.
Infant/Young Child:
Exposures for the infant and
young adult can occur through all
environmental media. When
breastfed, the mother's exposure
to environmental media can be an
additional source of exposure to
the infant.
Older Child/Adolescent:
i Air ! |Water!
Exposures to the child and j ^j^1 i--K Mother ^
adolescent can occur through all " if ijl
environmental media. The ; other! | Diet
mother's exposure is no longer a
factor for the child.
Fig. 2. Exposure routes during developmental lifestages. [CAP] These three illustrations show the different routes of exposure by
lifestage for children. The solid lines in the figure represent relevant exposure, whereas dotted lines represent exposures that are not
relevant to the specific lifestage. During gestation, the majority of exposures (except for physical factors) occur transplacentally through
exposure to the mother. After birth, exposures may either be directly to the child, with an additional route from the mother for those
agents that may be present in human milk.
characterization. Potentially significant exposure scenar-
ios are identified and evaluated to conduct a lifestage-
specific exposure analysis. Variability, sensitivity, and
uncertainty analyses are conducted to determine impact
of the available exposure data on the resulting analysis.
The results of the exposure characterization are iterated
with the hazard and dose-response analyses (Makris
et al., 2008). The purpose of this interaction is to identify
any critical developmental windows of susceptibility
(i.e., development periods with potential for increased
likelihood of an adverse effect) that were not considered
in the initial exposure characterization, or to identify
important exposure periods (i.e., periods of potentially
higher exposure) that were not considered in the hazard/
dose-response assessment. Finally, the exposure charac-
terization is summarized in a narrative that includes
discussion of the confidence in the analysis results based
on available data. This information feeds into the
comprehensive lifestage-specific risk characterization
(see Brown et al., 2008).
Conceptual Model
Problem formulation results in development of a
conceptual model. Here, the assessor identifies poten-
tially relevant lifestage-specific exposure scenarios by
considering the "windows" of early lifestage exposure
that could lead to toxicological outcomes in children or at
later lifestages. Several exposure considerations for
development of the conceptual model are discussed in
the Framework (U.S. EPA, 2006). These include perform-
ing a preliminary examination of the data to determine
the lifestages likely to be exposed given the chemical
properties, fate and transport, uses of the environmental
agent(s), and the defined scope of the assessment. For
example, lipophilic chemicals such as dioxins and
polychlorinated biphenyls (PCBs) can accumulate in
human milk (Solomon and Weiss, 2002); lead, which is
strongly absorbed to soil, can be ingested by children
during hand-to-mouth behavior (Landrigan et al.,
2002). The conceptual model involves a qualitative
Birth Defects Research (Part B) 83:522-529, 2008
Previous
TOC
-------
LIFESTAGE APPROACH TO ASSESSING CHILDRENS EXPOSURE
525
characterization of the potential sources, pathways of
exposures (including exposure media and routes),
exposure scenarios (lifestages, time frames, locations,
and activities), and pattern of exposures (magnitude and
duration) to parents or children, as appropriate.
An important issue to consider during development of
the conceptual model is the potential for lifestage-specific
vulnerabilities resulting from intrinsic susceptibility as
well as difference in exposure patterns. This includes a
qualitative understanding of lifestage-specific activity
patterns to identify potentially highly exposed lifestages
(Tulve et al., 2002; Cohen Hubal et al, 2000). For example,
a small pilot of children in a daycare setting found
difference in potential dermal exposure between infants
(6-12 months) and preschool age children (2-3 years)
(Cohen Hubal, 2006). Pesticide loadings on bodysuits and
levels measured using handwipes were higher for infants
than for the preschool age children. In addition, exposure
assessment case studies presented by Firestone et al.
(2007) demonstrate the need to consider lifestage-specific
differences when estimating exposures and identifying
data gaps. For example, in the case study evaluating
dietary intake of pesticides, significant statistical differ-
ences were found, based on 50th percentiles for children
aged 1 to < 3 years compared to each of children 1 to < 2
years and children 2 to <3 years. Although the magnitude
of these differences for these age bins is small, (<0.02 |j,g/
kg/day), these may be significant when aggregate
exposure over all pathways is considered and when these
exposures are considered in the context of intrinsic
susceptibility at a particular developmental window. In
addition, Firestone et al. (2007) discuss how limited data
for particular age groups may impact exposure estimates.
As such, consideration of potential differences prior to
beginning the assessment will facilitate interpretation and
transparent communication of results as well as design of
studies required to develop additional exposure data.
The Framework calls for application of EPA's Guidance
on Selecting Age Groups for Monitoring and Assessing
Childhood Exposures to Environmental Contaminants
(U.S. EPA, 2005) as a starting point for identifying and
selecting potentially important age bins for
exposure analysis (Firestone et al., 2007). This guidance
includes expert analysis of existing generic
exposure data and provides a detailed discussion of
how these age groups were developed and how to
implement them in an assessment. In brief, the recom-
mended age groups are based on the current under-
standing of differences in behavior and physiology that
may impact exposures to children. (Windows of intrinsic
susceptibility were also considered in a general sense.
For more detailed discussion of lifestage-specific sus-
ceptibility, see Makris et al., 2008.) The standard age
groups presented in the EPA Guidance include: birth to
<1 month; 1 to <3 months; 3 to <6 months; 6 to <12
months; 1 to < 2 years; 2 to < 3 years; 3 to < 6 years; 6 to
<11 years; 11 to <16 years; and 16 to <21 years
(Firestone et al., 2007). Importantly, the Framework
extends the scope of required exposure age bins to
facilitate characterization of exposure during preconcep-
tion and prenatal lifestages. The EPA's Child-Specific
Exposure Factors Handbook is another tool referenced in
the Framework that can be used to identify age-specific
behaviors that may result in higher exposure levels
(U.S. EPA, 2002).
Birth Defects Research (Part B) 83:522-529, 2008
Traditionally, the conceptual model has considered
human exposure in the context of the source-to-effects
paradigm (U.S. EPA, 2003a); however, the Framework
advocates for a person-oriented approach. In this
approach, the individual or population is the center of
the exposure model and the assessor searches from the
point of contact to identify a source(s) (Price et al., 2003).
In this way, potentially important time periods of
exposure, exposure pathways, and vulnerable indivi-
duals or populations can be identified. A person-
oriented exposure approach is necessary to address the
potential time lag between the relevant periods of
exposure during development and associated outcomes
that may be expressed at later lifestages.
Exposure Measurement and Estimation Approach
Four major types of information are used to character-
ize exposure: questionnaire-based metrics, surrogate
exposure metrics, personal exposure measures, and
biomonitoring data. Questionnaire-based metrics are
often the basis for exposure classification in epidemiolo-
gical studies. These data include information on activity
patterns, diet, and product use, and they are useful for
indicating specific behaviors or consumption patterns
that may put the individual in contact with the
environmental chemical. Questionnaires used to collect
children's exposure data are usually administered by a
trained technician using a proxy (e.g., parent or
caregiver) or filled out by the caregiver (e.g., in diaries).
Observational studies using techniques such as video-
taping have also proven useful for collecting activity data
(Zartarian et al., 1998). Surrogate exposure metrics are
typically ambient measurements of air, water, or food
(i.e., market basket) collected at community or regional
locations. Locations where these data are collected may
vary depending on the life stage. For example, infants
and toddlers are more likely to be in contact with rugs,
floors, lawns, and chemicals that tend to sink to low
levels (WHO, 2006). The environment of adolescents is
more varied and may include the home, school, social,
and occupational settings (WHO, 2006). These data are
often useful for characterizing background exposure
levels. Personal exposure data are collected at the
individual level. Personal air monitors and duplicate
diet samples are examples of this type of direct or point-
of-contact information. In the past, personal exposure
monitoring has been difficult to implement with young
children due to the size of monitors and the level of input
required from participants. Emerging technologies in
small-scale sensors, and wireless transmission provide
the promise for collection of needed personal exposure
data across all developmental lifestages. Finally, biomo-
nitoring data include measurement of chemical, metabo-
lite, or molecular markers in biological fluids/tissues
(e.g., blood, urine, human milk, cord blood, amniotic
fluid). Exposure biomonitoring data are used extensively
in epidemiology and in public health surveillance and
have the advantage of measuring total exposure from all
routes and sources. In general, observational studies
designed specifically to characterize important sources
and pathways of exposure will collect several, or even all
four, types of exposure information.
In the Framework, three approaches for calculating
exposures are discussed: point-of-contact, scenario
-------
526
COHEN HUBAL ET AL.
evaluation, and dose reconstruction (U.S. EPA, 1992). To
conduct the lifestage-specific exposure characterization,
a calculation approach is selected on the basis of
available data and the risk assessment questions that
were defined during the problem formulation phase.
The point-of-contact, or direct, approach requires
measurements of chemical concentrations at the point
where exposure occurs (at the interface between the
person and the environment) and records on length of
contact with each chemical. This approach, often used in
occupational settings, does not require additional in-
formation on an individual's characteristics or behaviors
because exposure is measured directly.
Using the dose-reconstruction approach, estimates of
exposure are developed from population-level biomoni-
toring data. Urine, blood, nail, saliva, hair, and feces are
commonly used to measure biomarkers of exposure
(WHO, 2006). Maternal biomarkers include amniotic
fluid and human milk (WHO, 2006). Other media used
to assess prenatal and postnatal exposure include first
teeth, meconium, and cord blood (WHO, 2006). At the
present time, most biomarker data are difficult to
interpret, either because the presence of a biomarker
may not be unique (e.g., many stressors may result in a
change in the same biomarker) or there may not be
adequate exposure pathway information to link the
biomarker to the exposure. As such, careful study design
is required to collect biomonitoring information for use
in characterizing exposure for risk assessment (Albertini
et al., 2006). Currently, this assessment approach is most
successful for characterizing exposure to persistent
compounds. Using biomonitoring data to characterize
intermittent exposures to rapidly metabolized and
cleared compounds requires a more complicated study
design incorporating repeated measures and information
on potential sources. Additional lifestage-specific meth-
odological considerations arise and need to be addressed
in design of biomonitoring studies (Barr et al., 2005).
The scenario evaluation approach, sometimes referred
to as the indirect approach, utilizes data on chemical
concentration, frequency and duration of exposure, as
well as information on the exposed lifestage. In this
approach, models are required to link the different data
to estimate individual or population-level exposures or
distributions of exposures. Currently, modeled estimates
(i.e., using the scenario evaluation approach) are often
used to conduct risk assessments necessary to make
regulatory decisions. As such, the scenario evaluation
approach is the focus of the Framework and is discussed
in more detail below.
Exposure Data and Information
The objectives and scope of the risk assessment,
defined in the problem formulation phase, provide the
focus for identifying all of the relevant human exposure
data and other necessary information. To focus on risk
from exposure to children, data are required on sources
and exposure media concentrations that have been
identified in the locations where children spend time.
These may change by developmental stage. For example,
sources may be identified in: (1) residence and workplace
for pregnant and lactating women; (2) residence, daycare
and outdoor play areas for infants and toddlers; (3)
residence, school, and locations of after-school activities
for school-age children; and (4) residence, school, and
locations of after-school activities and workplace for
adolescents.
For a given source, exposure media (e.g., water, soil/
dust/sediments, food, and objects/surfaces) and expo-
sure routes (i.e., inhalation, ingestion, dermal absorption,
and indirect ingestion) define the pathway of exposure
(Cohen Hubal et al., 2000b).
Exposure media and routes may change with lifestage.
Figure 2 highlights the stages of development and their
relevant exposure routes. For example, the fetus will be
exposed to cord blood and amniotic fluid, the infant to
breast milk, the teething child to surfaces of toys and
other objects (both intended and unintended) through
mouthing, and the school-age child to contaminants on
classroom floors.
For any given pathway, a set of associated exposure
scenarios describes how an exposure takes place, and is
used to estimate distribution of exposure. Available data
resources are outlined in the Framework. Though many
of the large exposure studies have focused on the adult
lifestage, these data are significant and are supplemented
by smaller and more recent studies specific to early
lifestages (Fenske et al., 2005; Gilliland et al., 2005;
Morgan et al., 2006).
Analysis Level or Tiered Assessment
The Framework proposes a tiered approach for
exposure characterization with extensive feedback at
each tier to the hazard identification/dose-response
analyses. The tiered approach is useful for the efficient
allocation of resources. Typically, an exposure character-
ization will begin with a screening-level assessment and
then, if there appears to be significant exposure or an
unacceptable level of uncertainty, a second, more refined
level of analysis will be conducted. The major difference
among the levels of assessment reflects the different
assumptions that are used (i.e., conservative assumptions
versus more refined assumptions). An important point
raised in the Framework is that probabilistic techniques
may be required at either level of analysis depending on
the types of scenarios that are being evaluated. This is a
departure from tiered approaches that have been
proposed previously (e.g., Reiss et al., 2003). Because a
lifestage assessment requires screening and potential
evaluation of the full range of age bins as well as multiple
sources and pathways, a probabilistic approach may be
the most efficient and most conservative, even at the
screening tier.
Screening Assessment. The purpose of a screen-
ing tier as described in the Framework is to identify
potentially important pathways, scenarios, and vulner-
able lifestages as well as to rule out insignificant ones.
Bounding values for exposure factors and conservative
simplifying assumptions are used at this level of analysis.
As a result, the output may have a high level of
uncertainty. Typically, screening assessments are used
when a quick exposure estimate is needed and to
prioritize additional work. Thus, screening assessments
make use of readily available measurement data, models,
and conservative assumptions to fill in data gaps.
Historically, deterministic calculations were used in most
screening-level exposure analyses. However, exposure
assessments have become increasingly complex, and
Birth Defects Research (Part B) 83:522-529, 2008
-------
LIFESTAGE APPROACH TO ASSESSING CHILDRENS EXPOSURE
527
probabilistic techniques may be useful when, for exam-
ple, exposure parameters have large variability or when
multiple sources exist (U.S. EPA, 2001).
Should a potential risk be identified based on the
bounding assumptions used in this level of analysis, a
more refined tier of analysis would be necessary to
address uncertainties in this screening-level exposure
assessment.
Refined Assessment. This tier provides more
detail for potentially relevant scenarios and potentially
sensitive age groups that may have been identified in the
screening assessment. The goal is often to estimate the
distribution of exposure for the relevant lifestages. Based
on results of the sensitivity analysis conducted for the
screening-level assessment, significant exposure factors
and important assumptions are revisited to develop
more realistic estimates of exposure.
This more advanced analysis may include the applica-
tion of sophisticated modeling tools to develop exposure
estimates for use in regulatory decisions. Few of these
models are designed currently to specifically address
lifestage exposures. As a result, data on the age bins used
in the models and outputs produced by the models may
not address the specific age groups of interest for a
complete lifestage-specific assessment (U.S. EPA, 2005;
Firestone et al., 2007). The Framework provides sugges-
tions for approaches to address data limitations and
associated uncertainties remaining following this refined
tier of analysis.
Supplemental Data Collection. Based on results
of the refined assessment and the associated sensitivity
and uncertainty analyses, specific data needs may be
identified. The Framework suggests that if the objectives
of the risk assessment indicate that any specific un-
certainties in the exposure characterization should be
addressed, then collection of new data to address these
may be needed.
Variability and Uncertainty in LifeStage Exposure
Assessment
Variability is defined as the heterogeneity of values
over time, space, or different members of a population.
Variability implies real differences among members of
that population (WHO, 2004). The Framework recognizes
the significant inter-individual variability in early life-
stages due to rapid physiologic, anatomic, and behavior-
al changes. Even within a relatively narrow age group,
variability may be significant. This variability affects the
determination of upper percentiles of exposure and its
associated risk. That is, given a high-quality/high-
quantity set of data for each age group, there may still
be significant variability for a particular exposure factor,
set of factors, or exposure pathway. The better the data
and the characterization of this variability, the better the
basis for final selection of age groups for a specific
assessment.
Uncertainty can be defined as imperfect knowledge
concerning the present or future state of an organism,
system, or (sub)population under consideration (WHO,
2004). Uncertainties include considerations related to (1)
missing, incomplete and/or incorrect knowledge; and (2)
ignorance and/or lack of awareness (WHO, 2004).
Currently, many assessments of early life exposures will
have significant uncertainty due to limited exposure
factor data for the relevant lifestages. Significant un-
certainty may also result from a lack of information and
understanding of relationships between early life ex-
posures and potential effects later in life.
Probabilistic approaches can be used to identify and
quantitatively characterize the important factors and
associated uncertainty and variability to better assess
early lifestage exposures. As mentioned previously, the
Framework recognizes that although probabilistic tech-
niques have traditionally been applied in advanced
levels of exposure assessment, these approaches may be
required at screening levels of analysis for complex
exposures and for consideration of multiple lifestages.
CONCLUSIONS
Contributions of the Framework
The lifestage approach presented in the Framework for
characterizing children's exposures to environmental
contaminants ensures a more complete evaluation of
the potential for vulnerability and exposure of sensitive
populations during development. Several important
issues highlighted in the Framework will serve to
improve exposure assessments for all lifestages. The
need to coordinate and iterate between exposure char-
acterization, hazard identification, and dose-response
assessment is emphasized throughout the Framework.
Increased coordination between the three analysis steps
of a risk assessment will ensure that critical windows of
exposure and hazard are considered. In addition, the
Framework promotes incorporating lifestage considera-
tions into Agency risk assessments despite database
limitations and scientific uncertainties. Recognizing that
there will always be information and knowledge gaps,
the Framework advocates for transparency in identifying
these gaps while providing suggested approaches for
moving forward to characterize exposures with available
data. Finally, the Framework emphasizes the importance
of characterizing variability and identifying significant
uncertainties as an integral part of conducting lifestage-
specific exposure assessment. Availability of the Frame-
work is sure to stimulate research that will facilitate
better understanding of variability in exposures based on
lifestage, and that will address some of the most
important uncertainties.
Research Needs
As risk assessments are conducted to address complex
questions related to potential outcomes across lifestages
from early life exposures, tools and approaches from
other fields will have to be applied. New methodologies
may have to be developed to incorporate into risk
assessment the knowledge about physiological and
behavioral differences across life stages. This may
require identifying analogies in other disciplines and
then conducting research to adapt and incorporate
relevant expertise, tools, and advances in these other
fields. Relevant fields may include statistics, systems
analysis, ecological risk assessment, social sciences,
genomics, and systems biology. Moving exposure analy-
sis forward for lifestage-specific risk assessment high-
lights the importance of developing predictive exposure
metrics, models, and lifestage-specific exposure factors
data. An increased emphasis on characterizing the links
Birth Defects Research (Part B) 83:522-529, 2008
-------
528
COHEN HUBAL ET AL.
between exposure, dose, and health outcome is essential
to ensure that critical windows of susceptibility are
addressed. The Framework also highlights the need for
identifying and characterizing vulnerable individuals
and populations. Finally, a broad definition of environ-
ment, environmental stressors, and exposure may be
needed to address these emerging issues in lifestage-
specific risk assessment.
The Framework document and this manuscript focus
on lifestage-specific issues associated with exposure
characterization for regulatory risk assessment. How-
ever, as regulators are called on to address the implica-
tions of environmental policy decisions on public health,
data developed in large-scale epidemiology studies such
as the National Children's study (Needham et al., 2005;
Ozkaynak et al., 2005) will be important for informing
these analyses. Improved approaches and guidance for
designing and implementing a high-quality exposure
component will be critical for success of studies
investigating environmental contributions to complex
disease and gene-environment interactions.
A major challenge in assessing early life exposures is
the limited availability of efficient and affordable
methods for comprehensively monitoring exposures
and internal dose. Improved exposure monitoring tools
are critical for identifying vulnerable populations, study-
ing gene-environment interactions, and testing impacts
of intervention and regulation. Emerging tools in
molecular biology provide the potential to develop
cellular and molecular indicators of exposure that can
be used to assess the vulnerability of humans across
lifestage to environmental stressors. Development of
molecular indicators of exposure combined with devel-
opment of nanotechnology-based sensors will present
the opportunity for the simultaneous, near real-time
measurement of biologically relevant exposures to multi-
ple real-world stressors. Limited research is underway in
this area (Weis et al., 2005). However, a significant
research effort is required to apply these technologies to
provide data required to characterize lifestage-specific
exposures for improved regulatory risk assessment and
public health policy.
ACKNOWLEDGMENTS
The authors wish to acknowledge the contributions of
the following current and former EPA scientists authors
of the EPA's Framework document: Rebecca C. Brown,
Stanley Barone Jr., Hisham El-Masri, Susan Y. Euling,
Carole Kimmel, Susan L. Makris, Babasaheb Sonawane,
Tracey Thomas, and Chad M. Thompson.
Disclaimer: This work was conducted by the U.S. EPA.
It has been subjected to the Agency's administrative
review and has been approved for publication.
REFERENCES
Albertini R, Bird M, Doerrer N, Needham L, Robison S, Sheldon L, Zenick
H. 2006. The use of biomonitoring data in exposure and human
health risk assessments. Environ Health Perspect 114(11):1755-1762.
Armstrong TW, Hushka LJ, Tell JG, Zaleski RT. 2000. A tiered approach
for assessing children's exposure. Environ Health Perspect 108:
469^174.
Barr DB, Wang RY, Needham LL. 2005. Biologic monitoring of exposure
to environmental chemicals throughout the life stages: requirements
and issues for consideration for the national children's study.
Environ Health Perspect 113:1083-1091.
Brown RC, Barone Jr. S, Kimmel CA. 2008. Children's health risk
assessment: incorporating a lifestage approach into the risk assess-
ment process. Birth Defects Res (Part B) 83:511-521.
Cohen Hubal EA, Egeghy P, Leovic K, Akland G. 2006. Measuring
potential dermal transfer of a pesticide to children in a daycare
center. Environ Health Perspect 114(2):264-269.
Cohen Hubal EA, Sheldon LS, Burke JM, McCurdy TR, Berry MR, Rigas
ML, Zartaria VG, Freeman NCG. 2000a. Exposure assessment for
children: a review of the factors influencing exposure of children,
and the data available to characterize and assess that exposure.
Environ Health Perspect 108(6):475^86.
Cohen Hubal EA, Sheldon LS, Zufall MJ, Burke JM, Thomas KW. 2000b.
The challenge of assessing children's residential exposure to
pesticides. Journal of Exposure Analysis and Environmental Epide-
miology 10:638-649.
Executive Order 13045: Protection of children from environmental health
risks and safety risks. 2000. Federal Register April 21, 1997.
62(78):19883-19888. Accessed online September 25, 2007, at http://
www.epa.gov/fedrgstr/eo/eol3045.htm
Fenske RA, Bradman A, Whyatt RM, Wolff MS, Barr DB. 2005. Mini-
monograph: lessons learned for the assessment of children's
pesticide exposure: critical sampling and analytical issues for future
studies. Environ Health Perspect 113:1455-1462.
Firestone M, Moya J, Cohen-Hubal E, Zartarian V, Xue J. 2007. Identifying
age groups for exposure assessments and monitoring. Risk Analysis
27(3):701-714.
Gilliland F, Avol E, Kinney P, Jerrett M, Dvonch T, Lurmann F,
Buckley T, Breysse P, Keeler G, de Villiers T, McConnell R. 2005.
Mini-monograph: air pollution exposure assessment for epidemio-
logic studies of pregnant women and children: lessons learned from
the Centers for Children's Environmental Health and Disease
Prevention Research. Environ Health Perspect: 113:1447-1454.
Landrigan PJ, Schechter CB, Lipton JM, Fahs MC, Schwartz J. 2002.
Environmental pollutants and disease in American children: esti-
mates of morbidity, mortality, and costs for lead poisoning, asthma,
cancer, and developmental disabilities. Environ Health Perspect
110(7):721-728.
Makris SL, Thompson CM, Euling SY, Selevan SG, Barone Jr. S, Sonaware
BR. 2008. A lifestage-specific approach to hazard and dose-response
characterization for children's health risk assessment. Birth Defects
Res (Part B) 83:530-546.
Morgan MK, Sheldon LS, Croghan CW, Chuang JC, Lordo RA, Wilson
NK, Lyu C, Brinkman M, Morse N, Chou YL, Hamilton C, Finegold
JK, Hand K, and Gordon SM . 2006. A pilot study of children's total
exposure to persistent pesticides and other persistent organic
pollutants (CTEPP). Volume I: Final Report. Contract Number 68-
D-99-011, U.S. Environmental Protection Agency, Office of Research
and Development, Research Triangle Park, NC. Accessed online
November 8, 2007, at http://www.epa.gov/heasd/ctepp/
ctepp_report.pdf
Needham LL, Ozkaynak H, Whyatt RM, Barr DB, Wang RY, Naeher L,
Akland G, Bahadori T, Bradman A, Fortmann R, Liu L-J, Morandi M,
O'Rourke MK, Thomas K, Quackenboss J, Ryan PB, Zartarian V.
2005. Exposure assessment in the national children's study:
introduction. Environ Health Perspect 113:1076-1082.
Olin SS, Sonawane BR. 2003. Workshop to develop a framework for
assessing risks to children from exposure to environmental agents.
Environ Health Perspect 111(12):1524-1526. Accessed online Novem-
ber 11, 2007, at http://ehp.niehs.nih.gov/members/2003/6183/
6183.html
Ozkaynak H, Whyatt RM, Needham LL, Akland G, Quackenboss J. 2005.
Exposure assessment implications for the design and implementa-
tion of the National Children's Study. Environ Health Perspect
113:1108-1115.
Price PS, Chaisson CF, Koontz M, Wilkes C, Ryan B, Macintosh D,
Georgopoulos P. 2003. Construction of a comprehensive chemical
exposure framework using person oriented modeling. Developed for
The Exposure Technical Implementation Panel, American Chemistry
Council, Contract 1388. Accessed online September 24, 2007, at
http://www.theUfeHnegroup.org/lifeline/documents/comprehensive_
chemical_exposure_framework. pdf
Reiss R, Anderson EL, Lape J. 2003, A framework and case study for
exposure assessment in the voluntary children's chemical evaluation
program. Risk Analysis 23(5):1069-1084.
Schwartz DA. 2006. The importance of gene-environment interactions
and exposure assessment in understanding human diseases. J Expo
Anal Environ Epidem 16(6):474-476.
Solomon, Weiss. 2002. Mini-monograph: chemical contaminants in breast
milk chemical contaminants in breast milk: time trends and regional
variability. Environ Health Perspect 110(6):339-347.
Tulve NS, Suggs JC, McCurdy T, Cohen Hubal EA. 2002. Frequency of
mouthing behavior in young children. J Expo Anal Environ Epidem
12:259-264.
Birth Defects Research (Part B) 83:522-529, 2008
-------
LIFESTAGE APPROACH TO ASSESSING CHILDRENS EXPOSURE
529
U.S. 104th Congress. 1996a. Food Quality Protection Act (FQPA). P.L. 104-
170. Accessed online September 25, 2007, at http://~www.epa.gov/
pesticides /regulating/laws/fqpa/gpogate.pdf
U.S. 104th Congress. 1996b. Safe Drinking Water Act (SDWA) amend-
ments. P.L. 104-182. Accessed online September 25, 2007, at http://
www.epa.gov/safewater/sd~wa/text.html
U.S. EPA. 1992. Guidelines for exposure assessment. U.S. Environmental
Protection Agency, Risk Assessment Forum, Washington, DC, 600Z-
92/001, 1992. Accessed online September 24, 2007, at http://
cfpub.epa.gov/ncea/raf/recordisplay.cfm?deid — 15263
U.S. EPA. 1995. Policy on Evaluating Health Risks to Children. Office of
the Administrator, Washington, DC. Accessed online September 25,
2007, at http://~www.epa.gov/OSA/spc/2poleval.htm
U.S. EPA. 2001. Risk Assessment Guidance for Superfund (RAGS),
volume Ill-part A: process for conducting probabilistic risk assess-
ment. Office of Solid Waste and Emergency Response. EPA 540-R-02-
002. Accessed online September 24, 2007, at http://www.epa.gov/
oswer/riskassessment/ragsSa/index.htm
U.S. EPA. 2002. Child-specific exposure factors handbook. Office of
Research and Development, Washington, DC. EPA/600/P-00/002B.
Accessed online September 24, 2007, at http://cfpub.epa.gov/ncea/
cfm/recordisplay.cfm?deid = 55145
U.S. EPA. 2003a. Human health research strategy. Office of Research and
Development, Washington, DC. EPA/600/R-02/050. Accessed on-
line September 24, 2007, at http://www.epa.gov/nheerl/human
health/HHRSJinal_web.pdf
U.S. EPA. 2003b. Framework for Cumulative Risk Assessment. (Fig. 1-3).
U.S. Environmental Protection Agency, Office of Research and
Development, National Center for Environmental Assessment,
Washington Office, Washington, DC. EPA/600/P-02/001F, 2003.
Accessed online September 24, 2007, at http://cfpub.epa.gov/ncea/
raf/recordisplay.cfm?deid — 54944
U.S. EPA. 2004. Risk assessment principles and practices. Office of
Research and Development, Washington, DC. EPA/100/B-04/001.
Accessed online October 25, 2007, at http://www.epa.gov/osa/
pdfs /ratf-f inal.pdf
U.S. EPA. 2005. Guidance on selecting age groups for monitoring and
assessing childhood exposures to environmental contaminants. Risk
Assessment Forum, Washington, DC. EPA/630/P-03/003F. Accessed
online on September 24, 2007, at http://cfpub2.epa.gov/ncea/cfm/
recordisplay.cfm?deid = 146583
U.S. EPA. 2006. A framework for assessing health risks of environmental
exposures to children. National Center for Environmental Assess-
ment, Office of Research and Development, Washington, DC. EPA/
600/R-05/093F. Accessed September 24, 2007, at http://cfpub.epa.
gov/ncea/cfm/recordisplay.cfm?deid = 158363
Weis BK, Balshaw D, Barr JR, Brown D, Ellisman M, Lioy P, et al. 2005.
Personalized exposure assessment: promising approaches for human
environmental health research. Environ Health Perspect 113(7):
840-848.
WHO. 2004. IPCS risk assessment terminology. Harmonization Project
document no. 1, World Health Organization, Geneva. Accessed
online October 25, 2007, at http://www.who.int/ipcs/methods/
harmonization/areas /terminology /en/index, html
WHO. 2006. Principles for evaluating health risks in children
associated with exposure to chemicals. International Programme
on Chemical Safety (IPCS) Environmental Health Criteria
document (EHC-237). World Health Organization. Accessed online
September 25, 2007, at http://whqlibdoc.who.int/publications/
2006/92415723 7X_eng.pdf
Birth Defects Research (Part B) 83:522-529, 2008
-------
© 2008 Wiley-Liss, Inc.
Birth Defects Research (Part A) 82:177-186 (2008)
Fetal Alcohol Syndrome (FAS) in C57BL/6 Mice Detected
through Proteomics Screening of the Amniotic Fluid
Susmita Datta,1* Delano Turner,2 Reetu Singh,3* L. Bruno Ruest,3
William M. Pierce Jr.,2 and Thomas B. Knudsen3
Department of Bioinformatics and Biostatistics, School of Public Health and Information Science,
University of Louisville, Louisville, Kentucky 40292
Biomolecular Mass Spectrometry Laboratory, Department of Pharmacology and Toxicology,
School of Medicine, University of Louisville, Louisville, Kentucky 40292
3 Department of Molecular, Cellular and Craniofacial Biology, Birth Defects Center, School of Dentistry,
University of Louisville, Louisville, Kentucky 40292
Received 22 August 2007; Revised 11 December 2007; Accepted 12 December 2007
BACKGROUND: Fetal Alcohol Syndrome (FAS), a severe consequence of the Fetal Alcohol Spectrum Disor-
ders, is associated with craniofacial defects, mental retardation, and stunted growth. Previous studies in
C57BL/6J and C57BL/6N mice provide evidence that alcohol-induced pathogenesis follows early changes in
gene expression within specific molecular pathways in the embryonic headfold. Whereas the former (B6J)
pregnancies carry a high-risk for dysmorphogenesis following maternal exposure to 2.9 g/kg alcohol (two
injections spaced 4.0 h apart on gestation day 8), the latter (B6N) pregnancies carry a low-risk for malforma-
tions. The present study used this murine model to screen amniotic fluid for biomarkers that could poten-
tially discriminate between FAS-positive and FAS-negative pregnancies. METHODS: B6J and B6N litters
were treated with alcohol (exposed) or saline (control) on day 8 of gestation. Amniotic fluid aspirated on
day 17 (n = 6 replicate litters per group) was subjected to trypsin digestion for analysis by matrix-assisted
laser desorption-time of flight mass spectrometry with the aid of denoising algorithms, statistical testing,
and classification methods. RESULTS: We identified several peaks in the proteomics screen that were
reduced consistently and specifically in exposed B6J litters. Preliminary characterization by liquid chroma-
tography tandem mass spectrometry and multidimensional protein identification mapped the reduced peaks
to alpha fetoprotein (AFP). The predictive strength of AFP deficiency as a biomarker for FAS-positive litters
was confirmed by area under the receiver operating characteristic curve. CONCLUSIONS: These findings in
genetically susceptible mice support clinical observations in maternal serum that implicate a decrease in
AFP levels following prenatal alcohol damage. Birth Defects Research (Part A) 82:177-186, 2008. © 2008
Wiley-Liss, Inc.
Key words: alcohol; pregnancy; FAS; FASD; mouse; C57BL/6J; C57BL/6NCrl; amniotic fluid; proteomics;
Random Forest; alpha fetoprotein
Presented at the 47th annual meeting of the Teratology Society, June 23-28,
2007, Pittsburgh PA.
Grant sponsor: NIH P20-RR/DE17702 (S. D.).
Grant sponsor: NIH/NIEHS P30ES014443-01A1 (S. D.)
Grant sponsor: NIH RO1-AA13205 (T. B. K.).
Grant sponsor: Biomolecular Mass Spectrometry Laboratory; Grant number:
S10-RR11368.
Grant sponsor: State of Kentucky Physical Facilities Trust Fund (W. M. R).
Grant sponsor: University of Louisville School of Medicine (W. M. R).
Grant sponsor: University of Louisville Research Foundation (W. M. R).
Grant sponsor: Heart and Stroke Foundation of Canada (L. B. R.).
L. Bruno Ruest's current address: Dept. of Biomedical Sciences, Baylor Col-
lege of Dentistry/TAMHSC, 3302 Gaston Ave., Dallas TX 75246.
Thomas B. Knudsen's current address: National Center for Computational
Toxicology, US Environmental Protection Agency, 109 T.W. Alexander Dr.
(B205-01), Research Triangle Park NC 27711.
The authors declare they have no competing financial interests. Although Dr.
Knudsen's current address is the U.S. Environmental Protection Agency this
work was conducted and analyzed while he was on the faculty at University
of Louisville. The content does not reflect the views of the Agency, nor does
the mention of trade names or commercial products constitute endorsement
or recommendations for use.
Correspondence to: Susmita Datta, Dept. of Bioinformatics and Biostatistics,
School of Public Health and Information Science, University of Louisville,
Louisville, KY 40292. E-mail: susmita.datta@louisville.edu
*Reetu Singh's current address: The Hamner Institutes for Health Sciences, 6
Davis Drive, Research Triangle Park, NC 27709-2137.
Published online 31 January 2008 in Wiley InterScience (www.interscience.
wiley.com).
DOI: 10.1002/bdra.20440
Birth Defects Research (Part A): Clinical and Molecular Teratology 82:177-186 (2008)
Previous
TOC
-------
178
DATTA ET AL.
INTRODUCTION
The Fetal Alcohol Syndrome (FAS) comprises a subset
of Fetal Alcohol Spectrum Disorders (FASD) and includes
a consistent pattern of physical abnormalities in the face
and eye, with intrauterine growth retardation and neuro-
developmental deficits (Jones et al., 1973). Although the
consequences of prenatal alcohol damage have been
well-characterized in children, the full range of effects in
FAS/FASD remains difficult to diagnose (Bertrand et al.,
2004). Developmental stage at the time of prenatal alco-
hol exposure, and maternal genotype or maternal-fetal
interactive effects, coupled with differing patterns and
amounts of alcohol consumption by the mother account
for some of the known variability and uncertainty in
alcohol-induced end points (Streissguth et al., 1996;
Viljoen et al., 2001; Sulik, 2005).
Early detection of FAS/FASD is highly desired for the
purposes of early intervention, both prenatally in terms
of reducing alcohol consumption through the remainder
of pregnancy, and postnatally, in terms of initiating
measures that may improve the child's performance
(Streissguth et al., 1996). As such, research leading to dis-
covery and validation of biomarkers that can better
inform healthcare decisions and interventions in alcoholic
pregnancies is needed (Bearer et al., 2005; Goodlett et al.,
2005). Studies have shown the utility of alcohol-derived
fatty acid ethyl esters in the meconium as indicative of
maternal alcohol consumption during pregnancy (Bearer
et al., 2005). Although this screening may ultimately
prove useful to quantify a risky drinking pattern, the
extent to which fatty acid ethyl esters address sensitive
stages of exposure or the risk for alcohol-related birth
defects is not yet clear. Identification of biomarker(s) of
effect could strengthen a prenatal alcohol screening pro-
gram by linking exposure with fetal changes (Goodlett
et al., 2005; Chen et al., 2005; Green et al., 2007). The dis-
covery and validation of general and specific biomarkers
for FAS/FASD in well-defined animal models can
advance this effort.
In the research described here, we tested the hypothe-
sis that protein complexity of the amniotic fluid (AF) can
change in association with the risk for alcohol-related
birth defects. AF is a colorless composite of water, pro-
tein, minerals electrolytes, hormones, environmental pol-
lutants, and exfoliated cells. During the normal course of
pregnancy the AF is conditioned by amniocytes that pro-
duce cytokines, lipids, prostaglandins, and growth factors
in response to local and endocrine signals, and by fetal
urination, lung secretion, swallowing, and intestinal
absorption (Cheung and Brace, 2005). In amniocentesis
some AF is aspirated for diagnostic purposes during the
second trimester, usually at 16-18 weeks when the AF
peaks in volume. Biochemical analysis of AF can reveal
specific developmental disorders; for example, alpha-feto-
protein (AFP) is elevated in the AF of fetuses with open
NTDs including anencephaly and spina bifida (Jones
et al., 2001). In contrast, AFP levels are sometimes abnor-
mally low in AF of fetuses with Down's syndrome
(trisomy 21) (Yamamoto et al., 2001). Some fetal testis
proteins in the AF were decreased in males born to alco-
hol users (Westney et al., 1991). Because the human AF
proteome database comprises more than 400 different
gene products (Tsangaris et al., 2005), a large scale analy-
sis of proteins (proteomics) may reflect multiple changes
that can be systematically linked with alcohol-induced
birth defects.
Proteomics-based methods have been used to study
brains from healthy and chronic alcoholic individuals
(Lewohl et al., 2004; Alexander-Kaufman et al., 2006). To
our knowledge, no such studies have been performed on
AF of alcoholic pregnancies. The present study used a
well-defined murine model to screen AF for biomarkers
that could potentially discriminate between FAS-positive
and FAS-negative pregnancies. This model comprises
closely related C57BL/6 mouse strains that respond dif-
ferently to maternal alcohol exposure on gestation day
(GD)8. Whereas C57BL/6J litters (B6J) carry a high-risk
for dysmorphogenesis following maternal exposure to
two 2.9 g/kg injections of ethanol alcohol spaced 4.0 h
apart on GD8, C57BL/6NCrl litters (B6N) carry a low-
risk for malformations (Green et al., 2007). AF was har-
vested on day 17 of gestation to capture the optimal yield
of AF (Cheung and Brace, 2005) at a stage in gestation
that can identify FAS-positive and FAS-negative litters
(Green et al., 2007). We applied statistical methods to
proteomics to identify unique signatures in the MALDI-
TOF mass spectrum of AF that could be anchored to the
risk for FAS and to differentiate between alcohol-exposed
litters in the sensitive strain relative to other groups.
MATERIALS AND METHODS
Animals and Exposure
C57BL/6J and C57BL/6NCrl mice 18-19 g in weight
were purchased from The Jackson Laboratory (Bar
Harbor, ME) and from Charles River Laboratories
(Wilmington, MA), respectively. These lines derived from
C57BL/6 by strict inbreeding at The Jackson Laboratory
(B6J) and the NIH followed by Charles River (B6N) and
differ by 1.6% polymorphism in microsatellite markers
(Hovland et al., 2000). Mice were housed in static mi-
croisolater cages with bedding that was absorbent, non-
nutritive, and nontoxic. The colonies cohabited the same
animal room and were maintained on a 12 h photoperiod
(06.00-18.00 h light). Diet was Purina mouse chow and
tap water ad libitum. Mice were acclimated to the room
for at least 10 days before breeding. The animal protocol
for this study was reviewed and approved by the Institu-
tional Animal Care and Use Committee at the University
of Louisville. Carbon dioxide asphyxiation followed by
cervical dislocation was the method of euthanasia.
For timed pregnancies we bred experienced males to
nulliparous females (20-22 g body weight on average) for
4—6 h starting at 07.30-8.00 h. Detection of a vaginal plug
at 13.30-14.00 h was regarded as evidence of coitus and
this day designated GDO. Dams showing 2-3 g weight
gain at 09.30 h on GD8 were assumed pregnant (typically
four to eight somite pair stage). The experimental design
had four treatment groups, with six to seven litters per
group as follows: (I) control B6J; (II) alcohol B6J; (III) con-
trol B6N; and (IV) alcohol B6N. Treatment used a stand-
ard model of i.p. injection of 22% absolute ethanol (v/v)
in isotonic saline given on GD8 by two i.p. injections
spaced 4 h apart (Webster et al., 1980; Sulik and John-
ston, 1983; Sulik et al., 1986; Kotch and Sulik, 1992; Green
et al., 2007). Each injection doses the dam with 2.9 g/kg
ethanol. Control litters received vehicle (saline) alone in
the same manner. All injections were at 0.5 mL per 30 g
maternal body weight.
Birth Defects Research (Part A) 82:177- 186 (2008)
Previous
TOC
-------
BIOMARKERS OF FAS/FASD
179
AF Collection
Pregnant dams were euthanized on GD17. After hys-
terectomy, the uterus was examined for resorptions. Indi-
vidual intact amniotic sacs containing the fetus were
carefully dissected from the myometrium and decidua
using fine tweezers and iridectomy scissors. AF was aspi-
rated from each individual AF cavity using a sterile
microsyringe and placed in a sterile microcentrifuge tube
on wet ice. Fetuses were inspected for evidence of gross
malformations, weighed, and fixed (after hypothermia) in
neutral-buffered formalin. Phenotype data from this
study were combined with similar data from our previ-
ous study (Green et al., 2007). Because the "mother" was
the unit of exposure in this model, the "litter" comprised
the unit of sampling and also was the basis for group
comparison. Two-way ANOVA (substrain, treatment,
interaction) was fitted to explain the variability in the ter-
atological outcomes using GraphPad Prism version 4.02
for Windows (GraphPad Software, San Diego, CA;
www.graphpad.com.) When ANOVA revealed a signifi-
cant group-wise effect (p < .05), postanalysis was per-
formed using Bonferroni-corrected multiple comparison
tests, as indicated.
AF Sample Preparation
The AF was inspected for clarity. Samples that were
cloudy or pink were rejected, as were samples from dead
fetuses. AF samples were then pooled for fetuses within
each litter, except any fetuses with NTDs were processed
individually so as not to contaminate the pooled AF with
uninformative samples. The AF yield for proteomics
analysis was at least 100-150 |iL pooled AF for each litter
replicate (n = 6 litter replicates). In total we analyzed
24 pooled AF samples from the four groups that met the
criteria for analysis, with six samples coming from each
group plus a few smaller samples from the grossly
abnormal fetuses. AF samples were centrifuged at 800 Xg
for 5 min at 4°C to remove amniocytes. Samples were
stored at —20°C until processing. A 50 \iL aliquot of AF
was placed in a clean microtube with 10 \iL 6M urea,
vortexed, and left at room temperature for 20 min. The
denatured samples were reduced with 10 \iL of 20 mM
dithiothreitol at 56°C for 45 min and alkylated with
10 \iL of 55 mM iodoacetamide at room temperature in
the dark. Ultrapure water (10 |iL) was added to dilute
the urea followed by addition of 10 \iL methylated tryp-
sin (Promega, Madison, WI; catalog #V5113, 100 ng/|iL)
in 50 mM ammonium bicarbonate buffer. Samples were
incubated overnight at 37°C. Trypsinized digests were
desalted via C-18 ZipTip (Millipore, Billerica, MA; cata-
log #ZTC18SO96) by aspirating three to five times.
Digests were washed with 0.1% formic acid and eluted
from the ZipTip with 10 \iL 60% acetonitrile in 0.1% for-
mic acid.
MALDI-TOF and Tandem MS
Aliquots of AF trypsin hydrolysate were spotted to a
MALDI target plate using 10 mg/mL alpha-cyanohy-
droxyl cinnamic acid solution as the matrix. Tryptic pep-
tide fragments were resolved on a Micromass ToFSpec2E
(Micromass/Waters, Milford, MA) mass spectrometer.
The instrument was set to reflectron mode using a
337 nm nitrogen laser and the instrument was operated
in positive ion mode for the m/z range of 500 to 4,000 Da.
Twelve spectra were collected automatically using set
locations on each sample well. Each spectrum consisted
of 40 laser firings (480 total laser firings) averaged to
improve signal-to-noise ratios. Internal tryptic hydrolysis
peaks were used to calibrate the instrument to a mass ac-
curacy of 75 ppm or less. Spectral data patterns were
compared to a battery of databases that were in-house
and on world wide web-based data searching resources
for the pattern matching discussed later.
Preprocessing of Mass Spectra
We applied three preprocessing steps: standardization,
denoising, and alignment. The first preserved higher mo-
lecular mass protein fragments at low abundance by
accounting for the nonuniform baseline and variability in
maximum intensity across the mass spectra by the
method of Satten et al. (2004). In this method, each spec-
trum was standardized using only information from that
spectrum. Let x denote mass-to-charge ratio (m/z) and let
y(x) be the corresponding spectral intensity. The spectra
are standardized by replacing the intensity y(x) with
y*(x) = (y(x) - Q0.5(x)}/{Qo.75(x) - Q0.25(x)}, where Qa(x)
is a local estimate of the ath quantile of spectral inten-
sities at m/z ratio x given by
with
Q+(x0)=mm{Wh(x0,y)>a},
E
(2)
where
(3)
and, for a set C , I{C} = 1 if C is true and = 0 otherwise.
Note that h is a user selectable width defining a neigh-
borhood of x0. In other words, we centered the spectra
using a (local) estimate of the median spectral intensity
and divided by a local estimate of the interquartile range.
We used interquartile range as a measure of scale
because it is insensitive to outlier peak intensities. The
function Wi,(x0, y) is the proportion of weights {1 — (x —
Xo)2/h2 that correspond to intensities y(x) that are less
than or equal to y. This choice for Qa is a variant of that
proposed by Ducharme et al. (1995).
The standardized spectra (step 1) were next subjected
to a denoising algorithm to separate noise from potential
peaks and for additional smoothing and alignment.
Although standardized spectra have a common scale and
are fairly homoscedastic, they still contain a mixture of
noise and signal. Denoising ensures that the features
used for classification correspond to real m/z peaks and
increases confidence in the scientific validity of the classi-
fication procedure (Sorace and Zhan, 2003). Again, we
selected a method that uses only the information in a sin-
gle spectrum. Note that the standardized spectral inten-
Birth Defects Research (Part A) 82:177- 186 (2008)
Previous
TOC
-------
180
DATTA ET AL.
sity y*(x) can be negative; in fact the median of the y*(x)
values is typically zero. While it may be difficult to sepa-
rate noise from signal using those standardized inten-
sities that are positive, the negative standardized inten-
sities presumably represent pure noise; therefore, we esti-
mated standard error based on the negative intensities
for a spectrum (Satten et al., 2004). We used those m/z's
for which the standardized intensities were at least three
standard deviations apart from the average standardized
intensity.
All spectra following standardization (step 1) and
denoising (step 2) were then aligned by a simple binning
technique. In step 3 the "features" across spectra were
binned into common intervals of bandwidth = 0.1 Da.
Maximum intensity within an interval was assigned to
the midpoint m/z of the interval. Following alignment the
features are hereto forward defined as "peaks" that were
subjected to statistical analysis.
Statistical Analysis for Proteomics
A univariate analysis was used to find any peaks that
would differentiate between control and exposed samples
in the sensitive (B6J) substrain, between control and
exposed samples in the insensitive (B6N) substrain, and
between control groups in both substrains (B6N, B6J). For
each comparison we used ANOVA, taking the mean
intensities in the comparison groups and forming t-statis-
tics for each and every peak. Multiple hypothesis cor-
rection (Benjamin! and Hochberg, 1995) was applied to
control the false discovery rate at a 10% level. We also
pursued groupwise classification (e.g., FAS-positive vs.
FAS-negative) using a Random Forest (RF) method
(Breiman, 2001). Considered one of the best off-the-shelf
classifiers currently available (Satten et al., 2004), RF
returns a list of variables (m/z) that are deemed to be
most useful for classifying the group of samples. For
each variable (peak) an importance measure is provided
that represents the variable's ability to distinguish control
versus alcohol-treated samples, and therefore could be a
candidate biomarker. Due to the randomness inherent to
the algorithm we ran RF multiple times to obtain the best
classification rate.
Provisional Peptide Fragment Identification
Uses of MALDI-TOF in proteomics studies prior to
trypsin digestion for each protein yields a limited num-
ber of peaks spread across a large range of m/z ratios.
Provisional identification of some proteins is possible
based on searchable properties of the peptides, such as
molecular mass. This peptide mass fingerprinting used
the Aldente search engine (Tuloup et al., 2002; http://
www.expasy.org/tools/aldente/) with the following
search parameters: molecular mass range taken to be 0 to
150 kDa; fixed modification of cysteine residues by car-
boxyamidomethylation; variable oxidation modification
of methionine; no restriction placed on isoelectric point;
and species selected as Mus musculus and genus Roden-
tia. Because each peptide may produce a range of frag-
ments after trypsinization the complexity of the mass
spectrum increases dramatically, resulting in more pep-
tide fragments that are displayed over a tighter range of
m/z ratios (e.g., mostly below 3,000 Da). Thus, peptide
mass fingerprinting was coupled with liquid chromatog-
raphy with tandem MS (LC-MS/MS) or multidimen-
sional protein identification technology (MuDPIT) to
derive sequence information on some of the peptides. For
LC-MS/MS, AF samples were denatured and digested as
above, desalted on a Cig spin column, and fractionated
with strong cation exchange resin. Strong cation
exchange fractions were concentrated to ~1 \iL with a
SpeedVac and diluted with 5% acetonitrile/0.1% formic
acid to ~7.5 \iL. Some fractions were subjected to analy-
sis on a Waters CapLC coupled to a Waters Q-TOF API-
US mass spectrometer. The LC eluate was coupled to a
nano-LC sprayer and MS/MS spectra were acquired with
data-dependent scanning. Only ions with 2+, 3+, or 4+
charges were selected for MS/MS analysis. MuDPIT anal-
ysis (Washburn et al., 2001; Welters et al., 2001) included
salt pulses given to free peptides from cation-exchange
resin, reversed phase resin separation, and MS/MS of
eluted peptide fragments. For both LC-MS/MS and
MuDPIT the MS/MS spectra were searched against
Swiss-Prot database with Protein-Lynx 4.0. The mass
error allowed was 25 ppm and a minimum three consec-
utive residues were required for a positive match.
Following preliminary characterization an in silico
trypsin digestion was performed to generate the peptide
fragment patterns of candidate proteins. This method
used ProteinProspector v4.0.8 (http://prospector.ucsf.
edu/), with "trypsin" selected as the enzyme. Other pa-
rameters selected included: peptide fragment mass range
was 800^1,000 Da, minimum fragment length of five,
maximum of two missed cleavages allowed, and cysteine
modification using carbamidomethylation.
RESULTS
Fetal Characteristics
Results from teratological evaluation (this study) were
combined with similar data from the previous study
(Green et al., 2007) to gauge the net response between
B6J and B6N pregnancies. Results on GD17 are shown
for mean incidence rate of resorptions, mean fetal weight,
and mean percentage of viable fetuses with overt malfor-
mations (Table 1). Mean resorption rates were 9.2% for
the control and exposed B6N litters. Although resorption
rates trended higher in exposed B6J litters (29.2% resorp-
tions) versus control B6J litters (13.1% resorptions), due
to highly variable litter effects the differences in resorp-
tion rates were not statistically significant. Mean fetal
weight was reduced in both substrains following alcohol
exposure (9-11% reduction vs. controls). Two-way
ANOVA (treatment, substrain, interaction) identified sig-
nificant treatment-related effects. Malformations mostly
involved the eye (coloboma, microphthalmia) and were
significant for treatment (p < .001) and treatment X sub-
strain interaction (p = .006). Postanalysis localized the
significant effects to the exposed B6J group. Malforma-
tion rates were 26.0% in exposed B6J pregnancies versus
6.0% in exposed B6N pregnancies. Consistent with the
previous study (Green et al., 2007), these findings suggest
that we may define sensitivity to prenatal alcohol based
on increased risk for malformations (and resorptions),
because fetal weight reduction was evident in either sub-
strain under the treatment conditions employed here. B6J
pregnancies carry a high-risk for dysmorphogenesis fol-
lowing maternal exposure to 2.9 g/kg alcohol whereas
B6N pregnancies carry a low-risk.
Birth Defects Research (Part A) 82:177- 186 (2008)
Previous
TOC
-------
BIOMARKERS OF FAS/FASD
181
Table 1
Fetal Effects of GD8 Alcohol Exposure in B6J and B6N Pregnancies Evaluated on GDI 7
Substrain
B6J
B6J
B6N
B6N
Two-way ANOVA
Group
Control
Alcohol
Control
Alcohol
(p-value)
Litters
(«)
22
16
20
19
Treatment
Substrain
Interaction
Resorptions
(% per litter)
13.1 ± 3.6
29.2 ± 6.4
9.2 ± 2.9
9.2 ± 5.5
0.057
0.064
0.239
Fetal weight
(mean g per fetus)
0.725 ± 0.025
0.647 ± 0.022*
0.735 ± 0.021
0.667 ± 0.027**
<0.001
0.509
0.643
Malformations
(% per litter)
0.0 ± 0.0
26.0 ± 7.8***
1.9 ± 1.4
6.0 ± 2.8
<0.001
0.052
0.006
Mean ± standard error, data compiled from previous (Green et al., 2007) and present studies.
Bonferroni-corrected t test for control (2 X saline, GD8) versus alcohol (2 X 2.9 g/kg ethanol, GD8) groups within each substrain;
*P < .05, **p < .01, ***p < .001.
Comparative Analysis of Aligned Mass Spectra
AF was aspirated on GDI 7 and pooled for proteomics
screen. Samples were pooled within a litter avoiding any
dead fetuses, bloody or cloudy AF aspirates, or fetuses
with open NTDs. AF passing acceptance criteria (n = 6
per group) was trypsinized for direct analysis by
MALDI-TOF mass spectrometry. Typical mass spectrum
profiles are shown for control and alcohol-exposed B6J
samples for the 500^,000 m/z region (Fig. 1) and to dem-
onstrate the signal amplification on raw profiles prepro-
cessed through the standardization and denoising algo-
rithms (Fig. 2). Although peak sizes in the MALDI-TOF
profiles are not considered to reflect an accurate quantita-
tive measure of peptide fragments, the alignment of six
replicated litters per treatment group returned a robust
reduction in several peaks across the different biological
conditions of the experiment. Three biologically relevant
groupwise comparisons were considered: control versus
alcohol-exposed samples in B6J (high-risk) pregnancies;
control versus alcohol-exposed samples in B6N (low-risk)
pregnancies; and control groups in B6J versus B6N preg-
nancies. The most important classifying m/z peaks are
shown in Table 2.
Analysis of the control versus alcohol-exposed B6J
samples returned three affected m/z peaks with inten-
sities that were reduced in a highly-significant manner,
with p values well below 0.00001: m/z = 3,163.5, 1,495.8,
and 1,369.6 (Table 2). These changes were evident irre-
spective of the number of malformed fetuses per litter in
the overall AF sample prior to, or after, exclusions of
NTDs or bloody/cloudy samples. As such, there was no
bias in those AF samples that met criteria in terms of the
number of malformed fetuses in the sample by group,
100-j .
.
%-
100-
~
0-
1337.90 1479.95
913.56
897.58
v»
, J
B
705
,1
J
91361
50
11
1009.66
1010.66
M,ft.
,
1337.91
1250.74
%>
86.65 1
•4
f
\
147
1609.96
ILL
1663.96
/
i
IBB;
I,. ,
3.17
II,
9.99
ll
236
2365.37,,
i
l.l,r.
L39 3026.75
2367.37 3025.74ho28.73
.. . j M - r3029'74
1883.17
1481,98
*"
,1,
1882.18
AlJ
1963.20 2366.43
„
/
MM
2367.41
2555.50 302f'80
Jlj r2682J1 I
500
1000
1500
2000
2500
3000
3500
nix
Figure 1. Sample MALDI-TOF spectrum of murine amniotic fluid samples collected on GD17. Samples from control (A) and alcoholic
(B) B6I mouse fetuses: 50 |iL aliquots were digested with methylated trypsin and 1 |iL aliquots were prepared in an alpha-cyano matrix
for linear and reflected MALDI-TOF.
Birth Defects Research (Part A) 82:177- 186 (2008)
Previous
TOC
Next
-------
182
Raw
DATTA ET AL.
Standardized and Denoised
CD
o
CD -
00
0
+
CD -
0
o
j[
J I, . ,l,
1 1 1 1 1 1
o
1000 1200 1400
to
CD
o
o -
CNJ
O
LO -
O -
O _
LO
L...
1000
1200
1400
m/z m/z
Figure 2. Preprocessing effects. Example of raw (left panel) and preprocessed (right panel) mass spectra (1,000-1,500 m/z region).
relative to the number of malformed fetuses per group in
the overall sample prior to exclusions. In contrast, a com-
parison of B6N samples revealed no peaks that were dif-
ferentially affected by alcohol in comparison to the con-
trols. Thus, preliminary screening of AF peptides re-
vealed a consistent display of peptide fragments across
samples and a significant differential display of several
peaks in the high-risk (B6J) pregnancies, but not in the
low-risk (B6N) pregnancies. One of these peaks (3163.5)
was significantly different in comparing control pregnan-
cies between the two substrains.
We next analyzed the AF peptide fragment profiles to
determine which specific peaks would discriminate the
FAS-positive group. Two of the top three altered peaks
in alcohol-exposed B6J pregnancies, namely m/z values
3,163.5 and 1,495.8, were also amongst the most impor-
Table 2
Significantly Altered (Reduced) Mass Spectrum
Peaks Identified in the Preliminary Murine Amniotic
Fluid Proteomics Screen
Group comparison Significantly
(n = 6) altered peaks (m/z)
B6J control vs.
B6N control vs
B6J control vs.
exposed
. exposed
B6N control
3,163.5
1,495.8
1,369.6
None
3,163.5
Top-significant
classifiers (m/z)
3,163.5
1,495.8
None
3,163.5
2,802.3
2,228.2
Spectra of mass/charge (m/z) ratios from trypsinized sam-
ples by MALDI-TOF in reflectron mode were preprocessed by
the three-step schema (standardized, denoised, aligned) and
subjected to statistical analysis (n = 6 independent litters per
analysis).
tant variables to classify the samples (Table 2). Repeated
use of the RF algorithm yielded classification accuracy as
high as 75 versus 75% predicted for a random classifier
that ignores the data. Therefore, in a limited sample size
of n = 6 we achieved reasonable success in distinguish-
ing the FAS-positive group using straight MALDI-TOF
analysis of the trypsinized AF sample. Peak 1,495.8,
which was the second most significant peak in terms of
the univariate t test, emerged as a strong diagnostic bio-
marker based on importance measures for classifying the
alcohol-exposed B6J proteomic profiles across multiple
RF runs (Table 2). In contrast, repeated use of the RF
algorithm failed to classify alcohol-exposed and control
B6N proteomic profiles. This method yielded 21-28%
classification accuracy, which is even worse than purely
random assignment (50%) and well-below the 75% accu-
racy with the B6J comparison. Again, this is consistent
with the FAS phenotype anchor.
Running the RF procedure to compare B6J and B6N
control groups yielded classification accuracies up to
65%, with three peaks of m/z = 3,163.5, 2,802.3, and
2,228.2 emerging as the top-significant classifiers (Table
2). Although marginal in accuracy by recursive RF, one
of these peaks (3,163.5) was significantly different at a p-
value well-below .00001. Another peak (2,228.2) trended
toward the effect but was not statistically significant, per-
haps as a limitation of the small sample size (n = 6).
Specificity-Sensitivity Analysis
Area under the receiver operating characteristic (ROC)
curve (AUROC) was computed to determine the predic-
tive classification power (sensitivity/specificity) of diag-
nostic peaks identified in the MALDI-TOF screen. To this
end, we selected the top five peaks (peaks 1-5) with the
largest absolute t statistic with regards to the capacity to
differentiate AF proteomic profiles between control and
Birth Defects Research (Part A) 82:177- 186 (2008)
Previous
TOC
-------
BIOMARKERS OF FAS/FASD
183
0,0
0.2
0,4 0.6
1-Specificity
0,8
1.0
0.0
0.2
0,4 0.6
1 -Speeifcily
0,8
1.0
Figure 3. Specificity-sensitivity analysis of the five top significant classifier peaks in the FAS-positive diagnostic profile. ROC curves
were drawn for ROC characteristics of the top five peaks from MALDI-TOF based on linear discriminant classifier of alcohol-exposed
and control B6J samples (named peaks 1-5, based on p values). Bootstrapped samples were randomly divided into training and test sets
for linear discriminant analysis with varying classification cut-offs. ROC curves plotted 1-specificity versus sensitivity for peaks 1-4 (left
panel) and peaks 4-5 (right panel). Maximum area under the ROC curve (1.0) was achieved using peaks 1-4, and near-unity (0.92) with
peaks 4-5.
alcohol-exposed B6J pregnancies. The first three of these
(peaks 1-3) were statistically significant after multiple
hypotheses correction using Benjamini and Hochberg
false discovery rate control at 10%. Due to the limitation
in sample sizes (n = 6), we generated 50 bootstrap sam-
ples from the original samples to perform a cross-valida-
tory calculation (Efron and Tibshirani, 1993) and deter-
mine ROC of a linear discriminant classifier using these
peaks (Datta and de Padilla, 2006). Bootstrapped samples
were randomly divided into training set (35 samples)
and test set (15 samples) from each group for linear dis-
criminant analysis (Fisher, 1936) with varying classifica-
tion cut-offs. Based on results for the test samples, we
plotted 1-specificity values against the sensitivity values
in order to construct the ROC curves (Fig. 3). The maxi-
mum AUROC (1.0) was achieved using intensity values
of peaks 1-3 or peaks 1-4, indicating ideal classification
performance. We also obtained unity AUROCs con-
structed with any of the top three peaks together with
peaks 4-5; however, AUROC dropped slightly (0.92)
when the ROC curve was drawn using only two chan-
nels (peaks 4 and 5).
Provisional Peptide Identification
Several more formal approaches were used for further
characterization of discriminating peaks in the MADLI-
TOF screen. First, peptide mass fingerprinting was
applied to the MALDI-TOF mass spectra. Because we
were dealing with clean AF samples we assumed the
complexity of proteins was low enough for at least some
useful information to be drawn regarding the most abun-
dant peptide species present. Among the top-scoring can-
didates the most frequent occurrences were: mouse
alpha-fetoprotein precursor (P02772), mouse serum albu-
min precursor (P07724), and myosin regulatory light
chain 2 skeletal muscle isoform (P97457). These peptides
were detected in essentially all B6J and B6N control sam-
ples. Many peptides showed weaker redundancy across
samples; a typical example is shown for control and
alcohol-exposed B6J and B6N samples with regards to
the five top-scoring candidate peptides (Table 3). In all
four cases the top two scoring candidates were, respec-
tively, alpha-fetoprotein and serum albumin. Whereas
both proteins belong to the albuminoid gene family the
current AF screen did not indicate differences in the
MALDI-TOF peptides derived from serum albumin.
An effort was made using in silica trypsin digestion to
map the top peaks identified by statistical analysis (Table
2) onto the most abundant proteins identified by peptide
mass fingerprinting. Although we could not identify the
most significantly reduced peak in the alcohol-exposed
B6J pregnancies (m/z = 3,163.5), the second-most signifi-
cantly reduced peak, which was also the second-most im-
portant classifying peak (m/z = 1,495.8), was successfully
mapped to mouse alpha-fetoprotein precursor. The third-
most significantly reduced peak (m/z = 1,369.6) was also
identified as a potential peptide of mouse alpha-fetopro-
tein precursor, assuming one missed in silica cleavage.
Analysis of some B6J samples by LC-MS/MS and MuD-
PIT further implicated both peaks 1,495.8 and 1,369.6
(assuming one missed cleavage) as trypsin-induced pep-
tide fragments cleaved from mouse alpha-fetoprotein
precursor protein. The related amino acid sequences cor-
responded to residues 514-526 (DETYAPPPFSEDK) and
191-201 (ADNKEECFQTK), respectively, of this 605
amino acid protein. Furthermore, among the top 20-
altered peaks observed in alcohol-exposed B6J samples in
terms of absolute t statistic, nine of them (m/z = 1,495.8,
Birth Defects Research (Part A) 82:177- 186 (2008)
Previous
TOC
-------
184
DATTA ET AL.
Table 3
Top Five Scoring Proteins (in Terms of Lowest
pValue from MALDI-TOF Comparison) Shown
for Representative Control and Alcohol-Exposed
AF Samples in B6J and B6N Pregnancies
Name of candidate protein
(UniProt)
Primary
accession Amino
number acids Coverage
B6J (saline)
Mouse alpha-fetoprotein
precursor P02772 605 43%
Mouse serum albumin precursor P07724 608 18%
N-terminal acetyltransferase
complex ARD1 subunit
homologA Q9QY36 235 44%
Short/branched chain specific
acyl-CoA dehydrogenase,
mitochondrial precursor Q9DBL1 432 30%
Myosin regulatory light chain 2,
skeletal muscle isoform P97457 169 38%
B6J (ethanol)
Mouse alpha-fetoprotein
precursor P02772 605 30%
Mouse serum albumin precursor P07724 608 23%
Phospholipid hydroperoxide
glutathione peroxidase,
mitochondrial precursor O70325 197 42%
Serine/threonine-protein kinase
TBK1 Q9WUN2 729 22%
Potassium channel
tetrameriesation Q9D7X1 259 35%
B6N (saline)
Mouse alpha-fetoprotein
precursor
Mouse serum albumin precursor
Receptor-interacting serine/
threonine-protein kinase 2
Vinculin
Caspase-4 subunit plO
B6N (ethanol)
Mouse alpha-fetoprotein
precursor
Mouse serum albumin precursor
Putative SplOO-related protein
Spermatid-specific linker histone
Hl-like protein
Succinyl-CoA ligase
P02772
P07724
P58801
Q64727
P70343
P02772
P07724
Q99388
Q9QYLO
Q9Z218
605
608
539
1,066
373
605
608
208
170
433
45%
34%
34%
19%
58%
36%
21%
49%
51%
33%
Based on the Aldante search engine (http://www.expasy.org/
tools/aldente/).
1,369.6, 1,774.9, 1,556.8, 1,638.8, 1,337.7, 1,685.9, 897.5,
897.6) were predicted from an in silica trypsin cleavage
of mouse alpha-fetoprotein precursor protein. These
matches cover different regions of the protein and
include bins of 100, 200, 300, and 500 amino acid resi-
dues. Therefore, we conclude from these findings that
reduced detection of mouse alpha-fetoprotein precursor
protein accounted for two of three major classifier peaks
that can distinguish the alcohol-exposed B6J litters. The
third classifier (peak 3,163.5) differentiated between the
sensitive strain and the insensitive strain in the unex-
posed, but also significantly differentiated between the
exposed and unexposed in the sensitive strain; however,
the identity of this peptide was not determined in the
present study. By these criteria mouse alpha-fetoprotein
precursor protein levels were not altered in the alcohol-
exposed B6N pregnancies. A few B6J fetuses were
excluded from AF pooling that may have been dead or
severely malformed for days prior to AF procurement.
Although it is feasible to assay AF from individual grav-
ida, such abnormal fetuses must ultimately be procured
at a much earlier gestational stage to be useful. In fact,
the individual AF from a few severely malformed fetuses
from the exposed B6J group that we did examine by
MALDI-TOF profiles were not informative because there
were no abnormal fetuses to compare from the control
B6J or exposed B6N groups (not shown).
DISCUSSION
The optic primordium is a critical target of alcohol in
experimental teratogenesis and in FASD (Green et al.,
2007; Higashiyama et al., 2007). Early gestational expo-
sure to alcohol reprograms genetic networks during ini-
tiation of the FAS in mice (Green et al., 2007). That effect
was demonstrated in the GD8 mouse embryonic headfold
at 3 h following a single maternal injection of ethanol
(2.9 g/kg). In the aftermath of global genetic responses
that clearly differentiated high-risk (B6J) from low-risk
(B6N) inbred lines of C57BL/6 mice (Green et al., 2007),
results from the present study show that dysmorphogen-
esis was associated with eventual changes in the AF-com-
partment of the fetus that could be detected by MALDI-
TOF mass spectrometry and specialized data analysis
methods. Because the current FAS animal model is not
highly penetrant the pooled AF samples would have
included grossly unaffected as well as malformed fetuses.
Both pedigrees (B6J, B6N) were exposed to alcohol and
for both test substrains we measured an effect of the
alcohol exposure in terms of fetal weight reduction, but
only one substrain showed a response in terms of
increased malformation rates. In contrast to malforma-
tions, the fetal weight reduction was more evenly distrib-
uted across a litter. Thus, the general and specific bio-
markers for FAS/FASD that might emerge from such an
AF analysis on GD17 following acute maternal alcohol
intoxication on GD8 can only be anchored to the
increased risk for malformations on a litter basis. These
changes may be summarized as follows: (a) the AF pro-
teome in alcohol-exposed B6J pregnancies showed a
highly significant drop in the abundance of three peaks
(m/z = 3,163.5, 1,495.8, 1,369.6); (b) ROC analysis found
these peaks to be highly sensitive and specific for classi-
fying the susceptible group by exposure; (c) two of these
peaks (1,495.8, 1,369.6) mapped to mouse alpha-fetopro-
tein precursor protein, as did 9 of the 20 most altered
peaks based on in silica digestion; and (d) none of these
peaks were found to be altered by alcohol in the B6N
substrain. Taken together, these findings suggest that dis-
crete changes to the AF proteome can be anchored to the
observed risk for alcohol-related birth defects in a mouse
model for FAS. We interpret these changes to represent
an ability to better identify fetuses more likely to be
affected with FAS dysmorphology in association with the
incidence rate of detectable malformations.
Recently, alpha-fetoprotein has been considered as a
biomarker for perinatal distress (Mizejewski, 2007). Dis-
cordant levels of AFP in AF have been indicative of
Birth Defects Research (Part A) 82:177- 186 (2008)
Previous
TOC
-------
BIOMARKERS OF FAS/FASD
185
structural defects in the brain and spinal cord (elevated)
or low birth weight-fetal growth restriction (reduced). In
general, developmental regulation of AFP may be con-
nected with growth and differentiation or perinatal stres-
sors as reflected in the functional role attributed to AFP
in small molecule binding and transport (e.g., fatty acids,
retinoids, hormones, heavy metals, drugs, and toxicants).
Although interesting for the pathogenesis of developmen-
tal defects, this is dispensable for major organogenesis as
shown by the lack of malformations in AFP knockout
mice (Gabant et al., 2002).
Realization of alpha-fetoprotein as a general biomarker
for FAS has practical implications for understanding
alcohol's mode of action on the fetus as well as potential
translation to clinical diagnostics. On one hand, alpha-
fetoprotein is released from various cell types and gains
access to the extracellular fluids such as the AF compart-
ment (AF-AFP) and maternal serum (MS-AFP). There-
fore, lower amounts of AF-AFP may secondarily yield
lower MS-AFP levels that could, ultimately, reflect an
increased risk of alcohol-related malformations in babies
whose mothers drink heavily during pregnancy. In fact,
low MS-AFP was found to predict FAS correctly in 59%
of alcoholic pregnancies (Halmesmaki et al., 1987). That
study followed several standard diagnostic proteins
(human placental lactogen, pregnancy specific beta-1-gly-
coprotein, and alpha-fetoprotein) in 35 pregnant problem
drinkers and 14 abstinent control women, and concluded
that low alpha-fetoprotein and pregnancy specific beta-1-
glycoprotein in maternal serum were useful indicators in
predicting FAS. We arrived at the same link between
alpha-fetoprotein and FAS through a completely inde-
pendent, non a priori discovery-based screen of the AF
proteome and support the scientific justification for moni-
toring MS-AFP as part of prenatal care of drinking
women and early screening for alcohol-damaged fetuses.
On the other hand, the reduction in AF-AFP raises
questions regarding alcohol's mode of action on the AF
proteome. The presence of alpha-fetoprotein has been
detected almost universally in postimplantation embryos,
yolk sac, amnion, embryonic disc, and early primitive
streak stages for all mammalian species studied so far
(Mizejewski, 2004). Apart from the long-running debate
on alpha-fetoprotein's role in brain development, ele-
vated MS-AFP is a clinical biomarker for NTDs such as
spina bifida or anencephaly (Brock and Sutcliffe, 1972).
The failure of neural tube closure results in leakage of
this serum protein into the AF and MS at higher levels
than normal. In contrast, MS-AFP is abnormally low in
some pregnancies that carry trisomy 21 (Davis et al.,
1985). The clinical "triple test" performed at 14—22 weeks
of pregnancy is used to screen fetal and placental prod-
ucts in serum samples of expectant mothers >35 years in
age to detect trisomy 21 (Spencer et al., 1997; Caserta
et al., 1998; Wald et al., 2006a,b; Mizejewski, 2007). This
test measures alpha-fetoprotein levels along with uncon-
jugated estradiol and human chorionic gonadotropin that
are reduced in trisomy 21 pregnancies.
The alpha-fetoprotein precursor is synthesized at high
levels by fetal liver cells and visceral yolk sac endoder-
mal cells. Thus, acute gestational exposure to alcohol
likely alters the AF proteome as a secondary consequence
of fetal development. Because fetal growth retardation
was observed in both strains (B6J, B6N) but reduced AFP
was detected only in one strain (B6J), the data do not
suggest that growth retardation may have affected the
AFP level produced in the fetal liver. Furthermore, we
are not aware of studies that implicate liver dysfunction
in FAS children; however, the drop in AF-AFP could
reflect the aftermath of acute gestational alcohol exposure
on liver and/or yolk sac development. This might have
implications on multiple tissues because alpha-fetopro-
tein functions as a binding protein for small molecules
such as vitamin D, estrogens, fatty acids, and metals
(Mizejewski, 2004; Gitlin and Boesman, 1966; Attardi and
Ruoslahti, 1976). In addition to its role as a molecular
troubleshooter alpha-fetoprotein contains sequence motifs
that render it a druggable target in diagnostics or thera-
peutics (Mizejewski, 2004; Uriel, 1989). Perhaps the motif
sequence DETYAPPPFSEDK (m/z = 1,495.8) could pro-
vide a molecular target for early FAS diagnosis or thera-
peutic intervention, through an understanding of the
small molecules that might bind to this peptide domain.
Additional studies will be needed to establish the link
between dysregulation of the embryonic transcriptome
(Green et al., 2007) and disruption of alpha-fetoprotein in
the AF proteome (current study) following gestational
alcohol exposure.
ACKNOWLEDGMENTS
The authors are grateful to Kenneth Lyons Jones, M.D.
of the University of California San Diego for thoughtful
input.
REFERENCES
Alexander-Kaufman K, James G, Sheedy D, et al. 2006. Differential pro-
tein expression in the prefrontal white matter of human alcoholics: a
proteomics study. Mol Psychiatry 11:56-65.
Attardi B, Ruoslahti E. 1976. Foetoneonatal oestradiol-binding protein in
mouse brain cytosol is alpha-fetoprotein. Nature 263:685-687.
Bearer CF, Stoler JM, Cook ID, et al. 2005. Biomarkers of Alcohol Use in
Pregnancy. Alcohol Research and Health, NIAAA publications 28:38-
43 (pubs.niaaa.nih.gov/publications/arh28-l /38-43.pdf).
Benjamini Y, Hochberg Y. 1995. Controlling the false discovery rate: a
practical and powerful approach to multiple testing. J Royal Statist
Soc Series B Methodol 57:289-300.
Bertrand }, Floyd RL, Weber MK, et al. 2004. National Task Force on
FAS/FAE. Fetal alcohol syndrome: guidelines for referral and
diagnosis. Atlanta, GA: Centers for Disease Control and Preven-
tion (http://www.cdc.gov/ncbddd/fas/documents/FAS_guidelines_
accessible.pdf).
Breiman L. 2001. Random Forests. Machine Learning 45:5-32.
Brock DJH, Sutcliffe RG. 1972. Alpha-fetoprotein in the antenatal diagno-
sis of anencephaly and spina bifida. Lancet 2:197-199.
Caserta D, Baldi M, Carta G, et al. 1998. Tri-test: clinical considerations
on 1784 cases. Minerva Ginecol 50:73-75.
Chen SY, Charness ME, Wilkemeyer MF, et al. 2005. Peptide-mediated
protection from ethanol-induced neural tube defects. Dev Neurosci
27:13-19.
Cheung CY, Brace RA. 2005. Amniotic fluid volume and composition in
mouse pregnancy. J Soc Gynecol Investig 12:558-562.
Datta S, de Padilla LM. 2006. Feature selection and machine learning with
mass spectrometry data for distinguishing cancer and non-cancer
samples. Statistical Methodology: Special Issue on Bioinformatics
3:79-92.
Davis RO, Casper P, Huddleston JF, et al. 1985. Decreased levels of amni-
otic fluid AFP associated with Down syndrome. Am J Obstet Gynecol
153:541-544.
Ducharme GR, Gannoun A, Guertin M-C, et al. 1995. Reference values
obtained by kernel-based estimation of quantile regressions. Biomet-
rics 51:1105-1116.
Efron B, Tibshirani RJ. 1993. An Introduction to the Bootstrap. New York:
Chapman & Hall.
Fisher RA. 1936. The Use of Multiple Measurements in Taxonomic Prob-
lems. Ann Eugenics 7:179-188.
Birth Defects Research (Part A) 82:177- 186 (2008)
Previous
TOC
-------
186
DATTA ET AL.
Gabant P, Forrester L, Nichols J, et al. 2002. Alpha-fetoprotein, the major
fetal serum protein, is not essential for embryonic development but
is required for female fertility. Proc Natl Acad Sci USA 99:12865-
12870.
Gitlin D, Boesman M. 1966. Serum AFP, albumin, and -G-globulin in the
human conceptus. J Clin Invest 45:1826-1830.
Goodlett CR, Horn KH, Zhou FC. 2005. Alcohol teratogenesis: mecha-
nisms of damage and strategies for intervention. Exp Biol Med
230:394-406.
Green ML, Singh AV, Zhang Y, et al. 2007. Reprogramming of genetic
networks during initiation of fetal alcohol syndrome. Dev Dynam
236:613-631.
Halmesmaki E, Autti I, Granstrom ML, et al. 1987. Prediction of fetal alco-
hol syndrome by maternal alpha fetoprotein, human placental lactogen
and pregnancy specific beta 1-glycoprotein. Alcohol Suppl 1:473^76.
Higashiyama D, Saitsu H, Komada M, et al. 2007. Sequential developmen-
tal changes in holoprosencephalic mouse embryos exposed to ethanol
during the gastrulation period. Birth Defects Res A Clin Mol Teratol
79:513-523.
Hovland DN Jr, Cantor RM, Lee GS, et al. 2000. Identification of a murine
locus conveying susceptibility to cadmium-induced forelimb malfor-
mations. Genomics 63:193-201.
Jones EA, Clement-Jones M, James OFW, et al. 2001. Differences between
human and mouse alpha-fetoprotein expression during early devel-
opment. Ann Hum Genet 198:555-559.
Jones KL, Smith DW, Ulleland CH, et al. 1973. Pattern of malformation in
offspring of chronic alcoholic mothers. Lancet 1:1267-1271.
Kotch LE, Sulik KK. 1992. Experimental fetal alcohol syndrome: proposed
pathogenic basis for a variety of associated facial and brain anoma-
lies. Am J Med Genet 44:168-176.
Lewohl JM, Van Dyk DD, Craft GE, et al. 2004. The application of pro-
teomics to the human alcoholic brain. Ann NY Acad Sci 1025:14-26.
Mizejewski GJ. 2004. Biological roles of alpha-fetoprotein during preg-
nancy and perinatal development. Exp Biol Med 779:439^63.
Mizejewski GJ. 2007. Physiology of alpha-fetoprotein as a biomarker for
perinatal distress: relevance to adverse pregnancy outcome. Exp Biol
Med (Maywood) 232:993-1004.
Satten GA, Datta S, Moura H, et al. 2004. Standardization and denoising
algorithms for mass spectra to classify whole-organism bacterial
specimens. Bioinformatics 20:3128-3136.
Sorace JM, Zhan M. 2003. A data review and re-assessment of ovarian
cancer serum proteomic profiling. BMC Bioinformatics 4:24.
Spencer K, Muller F, Aitken DA. 1997. Biochemical markers of trisomy 21
in amniotic fluid. Prenat Diagn 17:31-37.
Streissguth AP, Barr HM, Kogan J, et al. 1996. Understanding the occur-
rence of secondary disabilities in clients with fetal alcohol syndrome
(FAS) and fetal alcohol effects (FAE): Final report to the Centers for
Disease Control and Prevention (CDC). Final Report to the Centers
for Disease Control and Prevention (CDC), August, 1996, Seattle:
University of Washington, Fetal Alcohol & Drug Unit, Tech. Rep. No:
96-06.
Sulik KK. 2005. Genesis of alcohol-induced craniofacial dysmorphism.
Exp Biol Med 230:366-375.
Sulik KK, Johnston MC. 1983. Sequence of developmental alterations fol-
lowing acute ethanol exposure in mice: craniofacial features of the fe-
tal alcohol syndrome. Am J Anat 166:257-269.
Sulik KK, Johnston MC, Daft PA, et al. 1986. Fetal alcohol syndrome and
DiGeorge anomaly: critical ethanol exposure periods for craniofacial
malformations as illustrated in an animal model. Am J Med Genet
(suppl) 2:97-112.
Tsangaris G, Weitzdorfer R, Pollak D, et al. 2005. The amniotic proteome.
Electrophoresis 26:1168-1173.
Tuloup M, Hoogland C, Binz PA, et al. 2002. A new Peptide Mass Finger-
printing tool on ExPASy: ALDentE. Swiss Proteomics Society 2002
congress. Applied Proteomics. Lausanne (CH), Dec:3-5.
Uriel J. 1989. The physiological role of alpha-fetoprotein in cell growth
and differentiation. J Nucl Med Allied Sci 33:12-17.
Viljoen DL, Carr LG, Foroud TM, et al. 2001. Alcohol dehydrogenase-2»2
allele is associated with decreased prevalence of fetal alcohol syn-
drome in the mixed-ancestry population of the Western Cape Prov-
ince, South Africa. Alcohol Clin Exp Res 25:1719-1722.
Wald NJ, Barnes IM, Birger R, et al. 2006a. Effect on Down syndrome
screening performance of adjusting for marker levels in a previous
pregnancy. Prenat Diagn 26:539-544.
Wald NJ, Morris JK, Ibison J, et al. 2006b. Screening in early pregnancy
for pre-eclampsia using Down syndrome quadruple test markers.
Prenat Diagn 26:559-564.
Washburn MP, Wolters D, Yates JR. 2001. Large-scale analysis of the yeast
proteome by multidimensional protein identification technology. Nat
Biotechnol 19:242-247.
Webster WS, Walsh DA, Lipson AH, McEwen SE. 1980. Teratogenesis af-
ter acute alcohol exposure in inbred and outred mice. Neurobehav
Toxicol 2:227-234..
Westney L, Bruney R, Ross B, et al. 1991. Evidence that gonadal hormone
levels in amniotic fluid are decreased in males born to alcohol users
in humans. Alcohol Alcoholism 26:403^07.
Wolters DA, Washburn MP, Yates JR. 2001. An automated multidimen-
sional protein identification technology for shotgun proteomics. Anal
Chem 73:5683-5690.
Yamamoto R, Azuma M, Wakui Y, et al. 2001. Alpha-fetoprotein microhe-
terogeneity: a potential biochemical marker for Down's syndrome.
Clin Chim Acta 304:137-141.
Birth Defects Research (Part A) 82:177- 186 (2008)
Previous
TOC
-------
© 2009 Wiley-Liss, Inc.
Birth Defects Research (Part A) 85:732-740 (2009)
Inducible 70 kDa Heat Shock Proteins Protect Embryos
from Teratogen-Induced Exencephaly: Analysis Using
Hspala/alb Knockout Mice
Marianne Barrier,1'2* David J. Dix,3 and Philip E. Mirkes1'2
1Birth Defects Research Laboratory, Division of Genetics and Development, Department of Pediatrics, University of Washington,
Seattle, Washington
Department of Veterinary Physiology and Pharmacology, Texas A&M University, College Station, Texas
3Office of Research and Development, U.S. Environmental Protection Agency, Research Triangle Park, North Carolina
Received 20 February 2009; Revised 28 May 2009; Accepted 28 May 2009
BACKGROUND: It is well known that a variety of teratogens induce neural tube defects in animals; however,
less is known about proteins that play a role in protecting embryos from teratogen-induced neural tube
defects. Previously, our laboratory has shown that embryos overexpressing the 70-Da heat shock proteins
(HSPs) Hspala and Hspalb were partially protected from the deleterious effects of exposure to hyperther-
mia in vitro. METHODS: In the present studies, we have used a transgenic mouse in which both of the
stress-inducible HSPs Hspala and Hspalb were deleted by homologous recombination. Time-mated
Hspala/alb (KO) and wildtype (WT) mice were exposed to hyperthermia in vivo on gestational day
8.5. RESULTS: Results show that 52% of the gestational day 15 fetuses from KO litters were exencephalic,
whereas only 20% of WT fetuses were affected. In addition, 6% of treated KO fetuses also exhibited eye
defects (microphthalmia and anopthalmia), defects not observed in WT fetuses exposed to hyperthermia.
Lysotracker red staining and caspase-3 enzyme activity were examined within 10 hours after exposure to
hyperthermia, and significantly greater levels of apoptosis and enzyme activity were observed in the KO
embryos compared with WT embryos. CONCLUSIONS: These results show that embryos lacking the Hspala
and Hspalb genes are significantly more sensitive to hyperthermia-induced neural tube and eye defects, and
this increased sensitivity is correlated with increased amounts of apoptosis. Thus, these results also suggest
that Hspala and Hspalb play an important role in protecting embryos from hyperthermia-induced congenital
defects, possibly by reducing hyperthermia-induced apoptosis. Birth Defects Research (Part A) 85:732-740,
2009. © 2009 Wiley-Liss, Inc.
Key words: Hsp70; Hspala; Hspalb; Hyperthermia; neural tube defect
INTRODUCTION
Neural tube defects (NTDs) are some of the most com-
mon congenital defects, with approximately 4000 preg-
nancies per year, or 12 per day, affected by an NTD in
the United States (Finnell et al., 2000). Neural tube
defects are related to failure of the embryonic neural
folds to fuse properly along the neuroaxis. Two common
forms of NTDs are anencephaly (incomplete closure of
the anterior neural folds) and spina bifida (incomplete
closure along the posterior neural folds). In humans,
potential causitive agents for NTDs include retinoids,
pesticides, organic solvents, ionizing radiation, vinyl
chloride, water nitrates, and disinfection by-products
(Finnell et al., 2000; Padmanabhan, 2006). In addition, the
anticonvulsant medications carbamazepine and valproic
Additional Supporting Information may be found in the online version of this
article.
Presented in part at the 46th annual meeting of the Teratology Society
June 24 to 29, 2006 at the Loews Ventana Canyon Resort in Tucson,
Arizona.
The U.S. Environmental Protection Agency through its Office of Research
and Development partially funded and collaborated in the research
described here. It has been subjected to agency review and approved for
publication.
Supported by National Institutes of Health grants R01ES07026, R01ES08744,
P30ES07033.
Philip E. Mirkes's current address: Department of Veterinary Physiology and
Pharmacology, Texas A&M University, College Station, Texas.
"Correspondence to: Marianne Barrier, US EPA, ORD, Research Triangle Park,
NC 27711. E-mail: barrier.marianne@epa.gov
Published online 28 July 2009 in Wiley InterScience (www.interscience.
wiley.com).
DOI: 10.1002/bdra.206lO
Birth Defects Research (Part A): Clinical and Molecular Teratology 85:732- 740 (2009)
Previous
TOC
-------
HSPA1A AND HSPA1B SUPPRESS TERATA
733
acid have also been implicated in NTDs in humans
(Lammer et al., 1987; Rosa, 1991). Several studies impli-
cate maternal hyperthermia as a risk factor for NTDs in
humans (Edwards, 1986; Graham et al., 1998; Moretti
et al., 2005).
Previous work from our laboratory has shown that
hyperthermia primarily targets cells in the developing
rodent central nervous system, inducing excessive levels
of cell death (Mirkes, 1985; Mirkes and Little, 1998, 2000)
In addition to inducing heat shock protein synthesis
(Lindquist and Craig, 1988; Li and Nussenzweig, 1996;
Nagata, 1996; Welch, 1987), hyperthermia also rapidly
activates the extracellular signal-regulated protein kinases
(ERKs), c-JUN N-terminal kinases (JNK) and stress-
activated protein kinase (p38) signal transduction path-
ways in postimplantation mouse embryos (Kyriakis and
Avruch, 1996; Verheij et al., 1996; Mirkes et al., 2000).
Although a variety of physical and chemical agents are
known that can disrupt embryonic development in ani-
mals and humans, little is known about factors that can
be activated to protect early postimplantation mamma-
lian embryos from these teratogens. Nonetheless, an
extensive literature documents that heat shock proteins
(HSPs), particularly 70 kDa Hspala and Hspalb, can pro-
tect cells exposed to a variety of toxic exposures such as
heat, radiation, oxidative stress, and chemical toxins
(Jaattela et al., 1992; Moseley, 1996; Nollen et al., 1999;
Hunt et al., 2004; Mayer and Bukau, 2005; Niu et al.,
2006).
In mice, the 70 kDa Hspa (formerly known as Hsp70)
family consists of 13 known members in NCBI Entrez
Gene (http://www.ncbi.nlm.nih.gov/). The majority of
these 13 Hspa genes and proteins are constitutively
expressed in the absence of stress. Constitutively
expressed Hspa are known to act as chaperones that
assist in folding, transport, assembly, and function of
proteins in the cytoplasm, mitochondria, endoplasmic
reticulum, and nucleus (Beckmann et al., 1990; Shi and
Thomas, 1992; Georgopoulos and Welch, 1993). In addi-
tion to these constitutively expressed members of the
family, two other members, Hspala and Hspalb, are rap-
idly induced in response to various stresses. In this arti-
cle, we will use Hspala/alb to indicate Hspala and
Hspalb, unless specifically indicated otherwise. These in-
ducible Hspa presumably function in a manner similar to
the constitutive Hspa; however, they do so in the context
of stress-induced alterations in cellular metabolism
(Gabai et al., 1995; Kampinga et al., 1995; Stege et al.,
1995). In addition, more recent evidence suggests that in-
ducible Hspala/alb play a direct role in protecting cells
from a variety of stresses by inhibiting stress-induced ap-
optosis (Mosser et al., 1997; Mosser et al., 2000; Didelot
et al., 2006). For example, Hspala/alb affect caspase-de-
pendent apoptosis by inhibiting translocation of the Bcl-2
family member Bax (Stankiewicz et al., 2005), inhibiting
cytochrome c release from the mitochondria and prevent-
ing the activation of caspase-3 through modulation of the
apoptosome (Li et al., 2000; Matsumori et al., 2006; Steel
et al., 2004; Tsuchiya et al., 2003). Hspala/alb are also
thought to affect caspase-independent apoptosis, when a
protective effect was seen after the transfection of
Hspala/alb in cells treated with exogenous caspase
inhibitors (Creagh et al., 2000). Other mechanisms by
which Hspala/alb is believed to affect stress-induced
apoptosis are being an effector of the antiapoptotic
Akt/PKB, prosurvival kinase (Barati et al., 2006; Rafiee
et al., 2006), inhibiting the activation of a key mitogen-
activated protein kinase, JNK1 (c-Jun N-terminal Kinase;
Meriin et al., 1999; Yaglom et al., 1999; Gabai et al., 2000;
Park et al., 2001; Lee et al., 2005), and inhibiting the
release of proapoptotic protein Smac/DIABLO from
myocyte mitochondria (Jiang et al., 2005).
Rat and mouse postimplantation embryos are capable
of responding to heat stress with the induction of several
HSPs (Mirkes, 1987; Walsh et al., 1987; Higo et al., 1989;
Bennett et al., 1990; Honda et al., 1991), the most promi-
nent being Hspala/alb. Hspala/alb can be induced in
day 10 rat embryos (day 8.5 in mice) by temperatures
above 40°C. At these temperatures, synthesis of Hspala/
alb can be detected within 30 to 60 minutes after expo-
sure (Mirkes, 1987), and accumulation of Hspala/alb
protein can be detected within 2.5 hours (Mirkes and
Doggett, 1992). Once synthesized, Hspala/alb protein
can be detected in the embryo for up to 24 hours. Thus,
temperatures that exceed the normal growth tempera-
tures (37-38°C) by more than 3°C rapidly induce the syn-
thesis and accumulation of specific HSPs.
We hypothesize that the induction of Hspala/alb after
heat shock helps protect embryos from hyperthermia-
induced NTDs by reducing the induction of apoptosis.
The goal of the present study was to observe Hspala/
alb protection of early postimplantation mouse embryos
and to determine the molecular mechanisms underlying
these protective effects.
MATERIALS AND METHODS
Hspala/alb Knockout and Wildtype Mice
In the present studies, we have used a transgenic
mouse in which both of the stress-inducible members of
the 70-kDa HSP family (Hspala/alb) were deleted by
homologous recombination from C57B1/6-J mice. The
University of Washington received heterozygous
Hspala/alb knockout (KO) mice from the USEPA. The
heterozygotes were then crossed to produce Hspala/alb
KO and wildtype (WT) lines which were then maintained
separately. Knockout and WT genotypes were verified
with PCR analysis of tail clips using F191:5'GTVl CAC
TTT AAA CTC CCT CC 3' and R644 5'CTG CTT CTC TTG
GCT TCG3' primers. Knockout and WT embryos were
evaluated for somite number at several gestational time
points from 8 days, 22 hours to 9 days, 12 hours (4 litters
per time point) to determine whether there is a difference
in developmental timing between the two strains.
In Vivo Hyperthermia Exposure
Females were mated overnight with males of the same
strain and checked for a vaginal plug at 8:00 AM the next
morning. For females with a plug, gestational day 0 was
determined to start at midnight the night before. To
determine pregnancy, the weight of each plugged female
was recorded at the time of plug check and again on the
morning of gestational day 8.5. Mice that demonstrated a
weight gain of 2 gm or more were assumed to be preg-
nant, and those that gained less were assumed not to be
pregnant. For females with plugs, the beginning of gesta-
tion was set at midnight of the previous night. At noon
on gestational day 8.5, the pregnant dams were exposed
for 10 minutes to a water bath set at 38°C for controls
Birth Defects Research (Part A) 85:732-740 (2009)
Previous
TOC
-------
734
BARRIER ET AL.
and 43°C for hyperthermia treatment. For all treatments,
the pregnant dam was placed in a 50-ml conical tube
with the tip removed for air flow and holes drilled into
the sides to allow water to circulate around the animal.
The restrained mouse was then partially submerged in
the water bath set at the appropriate temperature. Mater-
nal core body temperature was monitored and recorded
via a rectal probe (Skin Temp Probe from MiniMitter,
Bend, OR). A Minilogger (MiniMitter, Bend, OR) was
used to display real-time temperature on a computer
using in-house visualization software. The depth of the
mouse in the water bath was adjusted as needed to
achieve consistent temperature exposure levels. For the
control treatment, maternal temperature was kept
between 37° and 38°C. For the hyperthermia treatment,
maternal temperature was monitored that so it reached
43°C at the 7-minute mark and held at 43°C for the
remaining 3 minutes of the exposure. Following
the 10-minute exposure, the animal was removed from
the water bath, pat-dried briefly with paper towels, and
quickly placed in a 38°C incubator and monitored until
core temperature returned to normal. The treated mice
were then returned to their cage until the embryo collec-
tion time point. At the appropriate time point, the preg-
nant dams were euthanized by cervical dislocation, and
the gravid uteri were removed for embryo extraction.
Gestational Day 15.5 Evaluations
At least 10 hyperthermia- and 5 control-treated litters
from each Hspala/alb KO and WT strains were eval-
uated for NTDs. Time-mated females were treated at ges-
tational day 8.5, and the litters were extracted at gesta-
tional day 15.5. For each litter, the numbers of live and
resorbed fetuses were recorded, and the fetuses were
evaluated for developmental abnormalities. Differences
in percent resorbed and abnormal fetuses among treat-
ment groups were analyzed using a one-way ANOVA
with Bonferroni's Multiple Comparison test performed
using GraphPad Prism version 5.01 for Windows
(GraphPad Software, San Diego, CA).
Tissue Collection for Western Blot Analysis
Time-mated females were treated at gestational day
8.5. At 1, 2.5, and 5 hours after the 10-minute exposure
(either 38°C or 43°C), embryos were removed from the
uterus for analysis. Embryos from three litters were col-
lected for each of the 12 strain (Hspala/alb KO, WT) X
treatment (Hyperthermia, Control) X time point (1-, 2.5-,
5-hour) combinations. Embryos dissected from the uterus
and surrounding membranes were flash frozen or stored
in RNAlater (Am7020, Applied Biosystems, Foster City,
CA). Protein was extracted with the miRvana Kit
(AM1561, Applied Biosystems, Foster City, CA) and
quantified using BCA protein assay (23227; Pierce, Rock-
ford, IL). Samples were run on 12.5% PAGE gels (Protean
4 system, BioRad) and transferred to PolyScreen PVDF
hybridization transfer membrane (PerkinElmer, Boston,
MA) using a semidry transfer apparatus (Ellard Instru-
mentation, Ltd., Monroe, WA). Immunoblot analysis was
performed using 3% nonfat dry milk in Tris-buffered
saline/0.5% Tweens for blocking and antibody dilutions.
The primary antibodies used were for the inducible
forms of Hspala/alb (SPA-810; Stressgen, Ann Arbor,
MI) and beta-actin (A3854; Sigma-Aldrich, Inc., St. Louis,
MO) as a loading control. Membranes were incubated
overnight (Hspala/alb) or 2 hours (actin). The secondary
antibody used was HRP-linked antimouse secondary
(NA931; GEHealthcare Bio-Sciences Corp. Piscataway,
NJ). ECL Plus Western Blotting Detection System
(RPN2132; GEHealthcare Bio-Sciences Corp., Piscataway,
NJ) was used for detection of the antibodies and mem-
branes were imaged with the Kodak Image Station 440CF
(Eastman Kodak Co., Rochester, NY).
Lysotracker Red Staining of Lysosomes in
Whole Embryos
Time-mated females were treated at gestational day
8.5. At 10 hours after the 10-minute exposure (38°C or
43°C), embryos were removed from the uterus for analy-
sis. At least three litters were collected for each of the
four strain (Hspala/alb KO, WT) X treatment (hyper-
thermia, control) combinations. Embryos dissected from
the uterus and surrounding membranes were stained
with 5 \iM Lysotracker Red (Molecular Probes, Inc.,
Eugene, OR) in IX Hank's balanced salt solution (Molec-
ular Probes, Inc., Eugene, OR) for 30 minutes at 38°C in
the dark. Embryos were then fixed in 4% paraformalde-
hyde overnight at 4°C. The fixed embryos were mounted
on microscope slides using ProLong Gold antifade
mounting media and Coverwell imaging chamber gas-
kets (Invitrogen, Carlsbad, CA). The head regions of the
embryos were imaged using the BioRad Radiance
2000MP microscope (Bio-Rad Laboratories, Inc., Rich-
mond, CA) with 20X objective, 0.75NA (or 10X, 0.45NA)
at the Texas A&M Image Analysis Laboratory. Lyso-
tracker fluorescence was detected using 568-nm excitation
wavelength and 590-nm emission wavelength. Images
were analyzed using Metamorph (Universal Imaging,
West Chester, PA). In brief, each image was corrected by
subtracting the background and transformed to a binary
image. The pixels present were then counted to deter-
mine the signal per unit area. Differences in Lysotracker
staining among treatment groups were analyzed using a
one-way ANOVA with Bonferroni's Multiple Compari-
son test performed using GraphPad Prism version 5.01
for Windows (GraphPad Software).
Caspase-3 Enzyme Assay
Time-mated females were treated at gestational day
8.5. At 5 hours after the 10-minute exposure (either 38°C
or 43°C), embryos were removed from the uterus for
analysis of caspase-3 enzyme activity. Three litters were
collected for each of the four strain (Hspala/alb KO,
WT) X treatment (hyperthermia, control) combinations.
The embryos were dissected from the uterus and sur-
rounding membranes, and then the heads were removed
and stored in phosphate buffered saline at —20°C for
approximately 3 weeks until use. Each embryo head was
processed individually by adding cell lysis buffer (Cas-
pase-3 Cellular Activity Assay Kit PLUS, AK-703; BIO-
MOL Research Laboratories, Plymouth Meeting, PA) and
disrupting the tissue by pipetting with a standard plO
pipette tip. Caspase-3 activity was measured for each
head using the BIOMOL Caspase-3 Cellular Activity
Assay Kit according to the manufacturer's instructions
and using the Power Wave XC Universal Microplate
Spectrophotometer plate reader (BIO-TEK Instruments,
Birth Defects Research (Part A) 85:732-740 (2009)
Previous
TOC
-------
HSPA1A AND HSPA1B SUPPRESS TERATA
735
Inc., Winooski, VT). Activity readings were taken every
10 minutes for 4 hours. Enzyme activity was determined
by the rate of cleavage of a caspase-3 colorimetric sub-
strate (DEVD-pNA) normalized to embryo protein con-
tent (fmols/min/ug protein) as described in the Quanti-
Zyme Assay Kit. A one-way ANOVA with Bonferroni's
multiple comparison test was performed using GraphPad
Prism version 5.01 for Windows (GraphPad Software) to
determine significance in pairwise comparisons of
enzyme activity.
Hspa la/alb
Wild Type
Untreated
3B°C Control
43"C Heat Shock
Hspala/alb
Knockout
Untreated
3B°C Control
4 y C Heat Shock
Hipala/alb Protein
RESULTS
Hspala/alb Expression in Hspala/alb WT and
KO Embryos
In the present study we used a transgenic mouse in
which both of the stress-inducible members of the
70-kDa HSP family (Hspala/alb) were deleted by
homologous recombination. In preliminary experiments,
we performed western blot analyses to confirm the lack
of inducible Hspala/alb protein. As shown in Figure 1,
we detected little if any expression of HSPala/alb in KO
embryos compared to a robust induction of HSPala/alb
in WT embryos. Untreated litters of KO and WT embryos
were evaluated for neural fold closure and somite num-
ber at multiple gestational time points (8 days, 22 hours;
9 days, 6 hours; 9 days, 12 hours) to determine whether
there is a difference in developmental timing between
the two strains. Based upon a statistical analysis (Mann
Whitney, unequal variance, two-tailed, unpaired) of dif-
ferences in somite number between KO and WT
embryos, only the difference at 9 days, 6 hours (WT-20,
KO-17 somites) was statistically significant at the 95%
confidence interval (See Supplemental Table SI).
Hyperthermia-Induced NTDs in Hspala/alb WT
and KO Embryos
Hyperthermia and control-treated fetuses of both
Hspala/alb KO and WT litters were evaluated at gesta-
tional day 15.5 to observe morphologic abnormalities
resulting from the treatments (Fig. 2). Among WT
fetuses, hyperthermia exposure at gestational day 8.5
induced exencephaly in 20% of hyperthermia-exposed
embryos (Fig. 2A; Table 1). Among KO fetuses, hyper-
thermia exposure at gestational day 8.5 induced exence-
phaly in 52% of hyperthermia-exposed embryos, a signif-
icant 2.6-fold increase in the incidence of exencephaly
compared to WT fetuses (p < 0.001; Table 1). In addition,
6% of hyperthermia-treated KO embryos exhibited eye
defects (microphthalmia and anopthalmia), defects not
observed in WT embryos exposed to hyperthermia
(Fig. 2B, Table 1). Even in the absence of hyperthermia
exposure, KO embryos were more sensitive to maldevel-
opment (2.5% exencephaly among KO fetuses vs. 0%
among WT fetuses) and in utero death (18% resorption
among KO embryos/fetuses vs. 6% among WT embryos/
fetuses; Table 1). There are significantly more affected
embryos (resorptions + defects) in hyperthermia-treated
KO litters, than in control KO or hyperthermia-treated
WT litters (Table 1). These results show that embryos
lacking the inducible Hspala/alb genes are more sensi-
tive to hyperthermia-induced neural tube and eye defects
compared to their counterparts that contain these genes.
In addition, embryos lacking the inducible Hspala/alb
WT KO
1 hour
WT KO
2.5 hours
Time After Exposure (hours)
Figure 1. Western blot analysis of inducible Hspala/alb protein
expression in untreated, control, and heat shock-treated Hspala/
alb KO and WT embryos show induction of Hspala/alb in WT,
but not KO embryos. (A) Representative Western blot image of
Hspala/alb protein expression in WT and KO samples as com-
pared to a protein standard of inducible Hspala/alb. (B) Chart
summarizing the relative expression levels of Hspala/alb protein
at 3 time points after treatment. [Color figure can be viewed in the
online issue, which is available at www.interscience.wiley.com.]
genes are more prone to maldevelopment (exencephaly,
resorption) even in the absence of a known teratogenic
exposure.
Apoptosis in Hspala/alb WT and KO Embryos
To test the hypothesis that the increase in exencephaly
among KO fetuses is related to an increase in hyperther-
mia-induced apoptosis during neural tube closure, we
used the lysosome stain, Lysotracker Red, to assess the
location and abundance of lysosomes as indicators of cell
death. We focused our analysis on the head regions of
the embryos, particularly along the anterior neural folds.
There appeared to be a moderate increase in Lysotracker
Red staining in the prosencephalon of the Hspala/alb
WT embryo in response to heat shock, whereas the
increase is observed to be more widespread through
much of the head region of the Hspala/alb KO embryo
(Fig. 3). Analysis of the signal per unit area showed sig-
nificant increases in Lysotracker Red staining in response
to hyperthermia in both Hspala/alb WT (fourfold, p <
0.01) and KO (sixfold, p < 0.001) embryos (Fig. 4). The
signal levels after hyperthermia treatment were signifi-
Birth Defects Research (Part A) 85:732-740 (2009)
Previous
TOC
-------
736
BARRIER ET AL.
A)
Unaffected
Exencephaly
Gestational day 15 C57BI/6 Hspa1a/a1b WT mice after heat shock
B)
unaffected exencephaly exencephaly & anopthalmia
unaffected microptnalmia anopthalmia unaffected
Gestational day 15 C57BI/6 Hspa1a/a1b KO mice after heat shock
Figure 2. Evaluation of C57B1/6 Hspala/alb-WT and KO mice on gestational day 15. Representative images of unaffected fetuses compared to
those with exencephaly and/or eye defects. [Color figure can be viewed in the online issue, which is available at www.interscience.wiley.com.]
cantly higher (1.7-fold, p < 0.01) in the Hspala/alb KO
embryos than in the Hspala/alb WT embryos (Fig. 4).
We also used a caspase-3 enzyme activity assay to
evaluate levels of apoptosis in embryo heads. Caspase-3
is an effector caspase, which is activated only during the
apoptotic process and is therefore a marker of apoptotic
activity. We observed a significant sevenfold (p < 0.05)
increase in caspase-3 activity in response to heat shock in
the Hspala/alb-KO litters, but a nonsignificant threefold
increase in the Hspala/alb-WT litters (Fig. 5). The cas-
pase-3 activity levels after heat shock treatment were sig-
nificantly higher (4.6-fold, p < 0.05) in the Hspala/
alb-KO embryos compared with the Hspala/alb-WT
embryos (Fig. 5).
DISCUSSION
It is well known that a variety of teratogens induce
NTDs in animals; however, less is known about proteins
that play a role in protecting embryos from teratogen-
induced NTDs. Thus, the first goal of the present study
was to determine whether Hspala/alb, a protein that is
rapidly induced in response to various stresses including
hyperthermia, protects early postimplantation mouse
embryos from teratogen-induced malformations. To do
this, we compared the levels of hyperthermia-induced
defects in WT and HSPala/alb null fetuses. Our results
clearly show that embryos/fetuses lacking HSPala/alb
are significantly more sensitive to hyperthermia-induced
Birth Defects Research (Part A) 85:732-740 (2009)
Previous
TOC
-------
HSPA1A AND HSPA1B SUPPRESS TERATA
737
Table 1
Summary of evaluations of C57B1/6 Hspala/alb-WT and KO mice on gestational day 15. A one-way ANOVA
analysis with Bonferroni's multiple comparison posttest was used to determine significance in pairwise
comparisons of treatment/strain groups for percent malformations and resorptions (GraphPad Prism 5.01)
Hspala/alb WT
Hspala/alb KO
Control
Heat Shock
Control
Significant Comparisons with 1-way Anova + Bonferroni's MCT.
WT, wildtype; KO, knockout; MCT, multiple comparison posttest; Ctrl, control; HS, heat shock.
aKO Ctrl vs. KO HS p < 0.001, WT HS vs. KO HS p < 0.001.
bKO Ctrl vs. KO HS p < 0.01, WT HS vs. KO HS p < 0.001.
Heat
Total no. of litters
Total no. of pups
No. (%) with exencephaly
No. (%) with eye defects
No. (%) resorbed
No. (%) affected
(All defects + Resorptions)
8
67
0 (0%)
0 (0%)
4 (6%)
4 (6%)
11
90
17 (20%)
0 (0%)
9 (9%)
26 (27.5%)
5
36
1 (2.5%)
0 (0%)
9 (18%)
10 (20.5%)
11
81
43 (52%)a
5 (6%)
11 (12%)
56 (60%)b
NTDs (exencephaly) and eye defects. Thus, one of the
functions of HSPala/alb during neural tube closure is to
act as a suppressor of teratogenic effects, at least as it
applies to hyperthermia. Additional research is required
to determine whether HSPala/alb can suppress defects
induced by other teratogens, particularly those teratogens
that do not induce the expression of HSPala/alb.
Despite the fact that embryos/fetuses lacking HSPala/
alb are significantly more sensitive to hyperthermia-
induced NTDs (exencephaly) and eye defects, approxi-
mately half of the embryos exposed to hyperthermia
Hspala/alb Wildtype Hspala/alb Knockout
Figure 3. Representative images of Lysotracker Red-stained
Hspala/alb-WT and KO embryo head regions demonstrate lev-
els of apoptosis found along the neural tube following control
and heat shock-treatment. For orientation, the general forebrain
(F), midbrain (M), and hindbrain (H) regions are indicated.
exhibit apparently normal neural tube closure and eye
development. Similarly, even in the presence of inducible
HSPala/alb, approximately 20% of the embryos exposed
to hyperthermia fail to complete neural tube closure,
resulting in exencephaly. Thus, although HSPala/alb
plays an important role in protecting embryos from
hyperthermia-induced defects, this protection is not com-
plete. This suggests that there must be other protective
factors in early postimplantation embryos. One such fac-
tor may be Hsp25, which has been shown to be protec-
tive in cells (Landry et al., 1989; Lavoie et al., 1993); we
have also shown that it is constitutively expressed in
postimplantation rat embryos and is also significantly
induced by exposure to hyperthermia (Mirkes et al.,
1996). Given the potential to modulate the expression of
heat shock proteins therapeutically (Herbst and Wanker,
2007; Putics et al., 2008; Roesslein et al., 2008), it will be
important to determine whether Hsp25 is also a suppres-
Lysotracker Red
(Overall Utter Averages)
0.20
I
f
Significant Comparisons
WT Ctrl vs WT HS - • p < 0-01
KO Ctrl VJ KO HS - ** p < O.OOJ
WT HS vs KO HS • • p •: 0.01
C±D
Hspala/alb
Wildtype
Cont
Hspala/alb
Krockout
rol
*
*
Hspala/alb
Wildtype
Heat
Hspala/alb
Knockout
Shock
0,00
Figure 4. Litter averages for Lysotracker Red signal in the head
region of Control- and Heat Shock-treated Hspala/alb KO and
WT embryos. The increase in Lysotracker Red staining in
response to heat shock was significant in both Hspala/alb-KO
and WT embryos. The level of staining after heat shock treat-
ment was significantly higher in the Hspala/alb-KO embryos
than in the Hspala/alb-WT embryos.
Birth Defects Research (Part A) 85:732-740 (2009)
Previous
TOC
-------
738
BARRIER ET AL.
Caspase-3 Enzyme Activity Assay
|Gverali Litter Averages)
_ 06
IO.M
£ 0.4 -
2-
C o<
Significant Comparisons
KO Ctrl vi KO MS - * p < 0,05
WT HS vs KO HS - * p i 0.05
Hspals/alb
Wildtyp*
Hsp»la/aib
Knockout
Hspsla/alb
Wildtype
Hspala/alb
Knockoy?
Heat shot*
Figure 5. Litter average for caspase-3 enzyme specific activity in
control- and hyperthermia-treated Hspala/alb KO and WT
embryo heads. The increase in caspase-3 enzyme activity in
response to heat shock was significantly greater in the Hspala/
alb-KO embryos (sevenfold) than in the Hspala/alb-WT
embryos (3.3-fold). The caspase-3 enzyme activity levels after
heat shock treatment were significantly higher in the Hspala/
alb-KO embryos than in the Hspala/alb-WT embryos.
sor of teratogenic effects. In addition, it is important to
identify other suppressors of teratogenic effects. Recent
work from our laboratory has shown that p53, a key cel-
lular regulator that determines whether a cell will arrest
or die in response to a variety of stresses, also suppresses
hyperthermia-induced exencephaly in mouse fetuses
(Hosako et al., 2009). Although far from conclusive, our
finding that HSPala/alb and p53, both of which are
known to regulate the apoptotic pathway, function as
suppressors of teratogenic effects suggests that apoptosis
is causally linked to the failure of neural tube closure
and the development of exencephaly in hyperthermia-
treated embryos.
Although our evaluations of somite numbers in
untreated KO and WT embryos showed a trend of fewer
somites in the KO embryos at the 8 day, 22 hour and
9 day, 6 hour time points, only the latter difference is
statistically significant. This difference disappears by the
9 day, 12 hour time point. A trend line fit to the somite
data for the KO (R-square = 0.9932) and WT (R-square =
0.9998) litters estimate an increase of one somite every
1.6 hours for KO and 1.8 hours for WT. Based on the for-
mulas for these trend lines, there would be an approxi-
mately four-somite difference at the time of treatment on
gestational day 8.5 (~6.5 somites in KO and 10.5 somites
in WT). Although we cannot rule out the possibility that
the nonsignificant trend of fewer somites in KO
compared with WT embryos around the time of exposure
to hyperthermia plays a role in the observed increase in
affected embryos in KO litters, it is unlikely that this is
the only contributing factor considering the increase in
maldevelopment observed in control-treated KO litters.
In addition to showing that embryos/fetuses lacking
HSPala/alb are significantly more sensitive to hyper-
thermia-induced NTDs (exencephaly) and eye defects,
our results show that embryos/fetuses lacking HSPala/
alb are significantly more sensitive to the induction of
exencephaly and intrauterine demise in the absence of a
hyperthermia exposure. These results raise two interest-
ing questions: what is the cause of exencephaly and
resorptions in KO embryos/fetuses in the absence of
hyperthermia, and why are KO embryos more sensitive
than WT embryos in the absence of hyperthermia-
induced Hspala/alb? The cause of exencephaly and
resorptions in KO embryos/fetuses could be either devel-
opmental errors or an exposure that is teratogenic to KO
but not WT embryos/fetuses. Although we cannot rule
either of these causes in or out, the fact that our control
pregnant mice are restrained and immersed in a water
bath at 38°C suggests that this treatment might constitute
a teratogenic exposure in KO but not WT embryos. Exen-
cephaly was not observed in untreated litters from either
WT or KO breeding colonies. Perhaps a more interesting
question is why some KO embryos develop exencephaly
or die in utero without exposure to hyperthermia. It
might be that although Hspala/alb are "inducible,"
these genes are also expressed constitutively at low levels
in the absence of any inducer. Although we do not have
definitive data suggesting that this is the case, the West-
ern blot data presented in Figure 1 do show faint bands
at the position expected for Hspala/alb. If Hspala/alb
is indeed constitutively expressed in the absence of
Hspala/alb inducers, this Hspala/alb presumably plays
some roles, one of which might be to inhibit abnormal
levels of programmed cell death. The absence of this pro-
tective function in KO embryos then might sensitize cer-
tain embryos to maldevelopment (exencephaly) or death
when confronted with developmental errors or a low-
level teratogenic exposure. Clearly, more research will be
needed to answer these intriguing questions.
Given our initial finding that HSPala/alb is a suppres-
sor of teratogenic effects; our second goal was to begin to
elucidate the molecular mechanisms underlying the pro-
tective effects of HSPala/alb. In this effort, we were
guided by our finding that hyperthermia-induced exence-
phaly is accompanied by hyperthermia-induced apopto-
sis in the neural folds and surrounding tissues (Fig. 3).
Although we do not know how the increase in apoptosis
is directly related to the observed exencephaly, the data
show an increase in cell death in tissues (neural folds)
known to play a key role in neural tube closure. The fact
that apoptosis is induced in any other tissue as well does
not detract from our conclusion that apoptosis plays a
role in hyperthermia-induced exencephaly. This increase
in cell death might interfere with the normal closure of
the neural tube in this region and is consistent with the
morphology of the observed defect of exencephaly in
which a section of the cranial region is absent. In addi-
tion, we were guided by an extensive literature showing
that one of the important functions of HSPala/alb is to
inhibit apoptosis at various points in the extrinsic and
intrinsic apoptotic pathways (Arya et al., 2007). Thus, we
hypothesized that the induction of Hspala/alb after heat
shock helps to protect embryos from hyperthermia-
induced neural tube defects by reducing the level
of hyperthermia-induced apoptosis. Conversely, we
hypothesized that in the absence of the antiapoptotic
effects of HSPala/alb, the increased exencephaly in null
embryos should be accompanied by an increase in apo-
ptosis. Results of the present studies show that embryos
lacking the inducible Hspala/alb genes are significantly
Birth Defects Research (Part A) 85:732-740 (2009)
Previous
TOC
-------
HSPA1A AND HSPA1B SUPPRESS TERATA
739
more sensitive to hyperthermia-induced NTDs and that
this increased sensitivity is associated with increased lev-
els of apoptosis. Although not formally proven, our
results from HSPala/alb KO embryos support the hy-
pothesis that HSPala/alb normally protects embryos
from hyperthermia-induced congenital defects, possibly
by reducing hyperthermia-induced apoptosis.
Although our data clearly show that hyperthermia-
induced apoptosis is significantly elevated in neurulating
mouse embryos in the absence of HSPala/alb, we do
not know the specific anti-apoptotic mechanisms of
HSPala/alb that are compromised and responsible for
the increased apoptosis. HSPala/alb is known to block
apoptosis at a number of points in the apoptotic path-
way. For example, HSPala/alb can inhibit apoptosis by
blocking translocation of Bax from the cytoplasm to the
mitochondria (Gotoh et al., 2004; Stankiewicz et al., 2005),
by inhibiting the formation of a functional apoptosome
complex through interaction with Apaf-1 (Beere et al.,
2000; Saleh et al., 2000), by blocking the activation of Bid
(Gabai et al., 2002), and by blocking the migration of AIF
from mitochondria to nuclei (Ravagnan et al., 2001). We
also know that hyperthermia both induces the rapid
expression of HSPala/alb (Mirkes, 1987; Mirkes and
Doggett, 1992; Mirkes et al., 1994; Thayer and Mirkes,
1997; Mirkes et al., 1999) and activates the mitochondrial
apoptotic pathway in postimplantation rodent embryos
(Mirkes and Little, 2000; Mirkes et al., 2001; Little and
Mirkes, 2002; Little et al., 2003) by inducing the release of
cytochrome c from mitochondria and the subsequent acti-
vation of both initiator (caspase-9) and effector caspases
(caspase-3,—6, and—7). Additional research is required to
determine the mechanisms by which HSPala/alb inter-
acts with apoptotic pathways in rodent embryos and
thereby modulates the levels of programmed and/or
teratogen-induced apoptosis.
ACKNOWLEDGMENTS
The authors wish to dedicate this manuscript to
Dr. Tom Shepard in honor of his numerous contributions
to the field of teratology. We also thank Sucheol Gil, Jen-
nifer Faske, Roula Mounemne, and Elizabeth M. Watson
for outstanding technical assistance.
REFERENCES
Arya R, Mallik M, Lakhotia SC. 2007. Heat shock genes - integrating cell
survival and death. J Biosci 32:595-610.
Barati MT, Rane MJ, Klein JB, McLeish KR. 2006. A proteomic screen
identified stress-induced chaperone proteins as targets of Akt phos-
phorylation in mesangial cells. J Proteome Res 5:1636-1646.
Beckmann RP, Mizzen LE, Welch WJ. 1990. Interaction of Hsp 70 with
newly synthesized proteins: implications for protein folding and as-
sembly. Science 248:850-854.
Beere HM, Wolf BB, Cain K, et al. 2000. Heat-shock protein 70 inhibits ap-
optosis by preventing recruitment of procaspase-9 to the Apaf-1
apoptosome. Nat Cell Biol 2:469^75.
Bennett GD, Mohl VK, Finnell RH. 1990. Embryonic and maternal heat
shock responses to a teratogenic hyperthermic insult. Reprod Toxicol
4:113-119.
Creagh EM, Carmody RJ, Cotter TG. 2000. Heat shock protein 70 inhibits
caspase-dependent and -independent apoptosis in Jurkat T cells. Exp
Cell Res 257:58-66.
Didelot C, Schmitt E, Brunet M, et al. 2006. Heat shock proteins: endoge-
nous modulators of apoptotic cell death. Handb Exp Pharmacol
(172):171-198.
Edwards MJ. 1986. Hyperthermia as a teratogen: a review of experimental
studies and their clinical significance. Teratog Carcinog Mutagen 6:
563-582.
Finnell RH, Gelineau-van Waes J, Bennett GD, et al. 2000. Genetic basis of
susceptibility to environmentally induced neural tube defects. Ann N
Y Acad Sci 919:261-277.
Gabai VL, Mabuchi K, Mosser DD, Sherman MY. 2002. Hsp72 and stress
kinase c-jun N-terminal kinase regulate the bid-dependent pathway in
tumor necrosis factor-induced apoptosis. Mol Cell Biol 77:3415-3474.
Gabai VL, Yaglom JA, Volloch V, et al. 2000. Hsp72-mediated suppression
of c-Jun N-terminal kinase is implicated in development of tolerance
to caspase-independent cell death. Mol Cell Biol 20:6826-6836.
Gabai VL, Zamulaeva IV, Mosin AF, et al. 1995. Resistance of Ehrlich
tumor cells to apoptosis can be due to accumulation of heat shock
proteins. FEES Lett 375:21-26.
Georgopoulos C, Welch WJ. 1993. Role of the major heat shock proteins
as molecular chaperones. Annu Rev Cell Biol 9:601-634.
Gotoh T, Terada K, Oyadomari S, Mori M. 2004. hsp70-DnaJ chaperone
pair prevents nitric oxide- and CHOP-induced apoptosis by inhibi-
ting translocation of Bax to mitochondria. Cell Death Differ 11:
390-402.
Graham JM Jr, Edwards MJ, Edwards MJ. 1998. Teratogen update: gesta-
tional effects of maternal hyperthermia due to febrile illnesses and re-
sultant patterns of defects in humans. Teratology 58:209-221.
Herbst M, Wanker EE. 2007. Small molecule inducers of heat-shock
response reduce polyQ-mediated huntingtin aggregation. A possible
therapeutic strategy. Neurodegener Dis 4:254-260.
Higo H, Lee JY, Satow Y, Higo K. 1989. Elevated expression of proto-
oncogenes accompany enhanced induction of heat-shock genes after
exposure of rat embryos in utero to ionizing irradiation. Teratog
Carcinog Mutagen 9:191-198.
Honda K, Hatayama T, Takahashi K, Yukioka M. 1991. Heat shock pro-
teins in human and mouse embryonic cells after exposure to heat
shock or teratogenic agents. Teratog Carcinog Mutagen 11:235-244.
Hosako H, Francisco LE, Martin GS, Mirkes PE. 2009. The roles of p53
and p21 in normal development and hyperthermia-induced malfor-
mations. Birth Defects Res B Dev Reprod Toxicol 86:40-47.
Hunt CR, Dix DJ, Sharma GG, et al. 2004. Genomic instability and
enhanced radiosensitivity in Hsp70.1- and Hsp70.3-deficient mice.
Mol Cell Biol 24:899-911.
Jaattela M, Wissing D, Bauer PA, Li GC. 1992. Major heat shock protein
hsp70 protects tumor cells from tumor necrosis factor cytotoxicity.
Embo J 11:3507-3512.
Jiang B, Xiao W, Shi Y, et al. 2005. Heat shock pretreatment inhibited the
release of Smac/DIABLO from mitochondria and apoptosis induced
by hydrogen peroxide in cardiomyocytes and C2C12 myogenic cells.
Cell Stress Chaperones 10:252-262.
Kampinga HH, Brunsting JF, Stege GJ, et al. 1995. Thermal protein dena-
turation and protein aggregation in cells made thermotolerant by var-
ious chemicals: role of heat shock proteins. Exp Cell Res 219:536-546.
Kyriakis JM, Avruch J. 1996. Protein kinase cascades activated by stress
and inflammatory cytokines. Bioessays 18:567-577.
Lammer EJ, Sever LE, Oakley GP, Jr. 1987. Teratogen update: valproic
acid. Teratology 35:465^73.
Landry J, Chretien P, Lambert H, et al. 1989. Heat shock resistance con-
ferred by expression of the human HSP27 gene in rodent cells. J Cell
Biol 109:7-15.
Lavoie JN, Hickey E, Weber LA, Landry J. 1993. Modulation of actin
microfilament dynamics and fluid phase pinocytosis by phosphoryla-
tion of heat shock protein 27. J Biol Chem 268:24210-24214.
Lee JS, Lee JJ, Seo JS. 2005. HSP70 deficiency results in activation of c-Jun
N-terminal Kinase, extracellular signal-regulated kinase, and cas-
pase-3 in hyperosmolarity-induced apoptosis. J Biol Chem 280:
6634-6641.
Li CY, Lee JS, Ko YG, et al. 2000. Heat shock protein 70 inhibits apoptosis
downstream of cytochrome c release and upstream of caspase-3 acti-
vation. J Biol Chem 275:25665-25671.
Li GC, Nussenzweig A. 1996. Thermotolerance and heat shock proteins:
possible involvement of Ku autoantigen in regulating Hsp70 expres-
sion. EXS 77:425^49.
Lindquist S, Craig EA. 1988. The heat-shock proteins. Annu Rev Genet
22:631-677.
Little SA, Kim WK, Mirkes PE. 2003. Teratogen-induced activation of
caspase-6 and caspase-7 in early postimplantation mouse embryos.
Cell Biol Toxicol 19:215-226.
Little SA, Mirkes PE. 2002. Teratogen-induced activation of caspase-9 and
the mitochondrial apoptotic pathway in early postimplantation
mouse embryos. Toxicol Appl Pharmacol 181:142-151.
Matsumori Y, Northington FJ, Hong SM, et al. 2006. Reduction of cas-
pase-8 and -9 cleavage is associated with increased c-FLIP and
increased binding of Apaf-1 and Hsp70 after neonatal hypoxic/ische-
mic injury in mice overexpressing Hsp70. Stroke 37:507-512.
Mayer MP, Bukau B. 2005. Hsp70 chaperones: cellular functions and mo-
lecular mechanism. Cell Mol Life Sci 62:670-684.
Birth Defects Research (Part A) 85:732-740 (2009)
Previous
TOC
-------
740
BARRIER ET AL.
Merlin AB, Yaglom JA, Gabai VL, et al. 1999. Protein-damaging stresses
activate c-Jun N-terminal kinase via inhibition of its dephosphorylation:
a novel pathway controlled by HSP72. Mol Cell Biol 19:2547-2555.
Mirkes PE. 1985. Effects of acute exposures to elevated temperatures on
rat embryo growth and development in vitro. Teratology 32:259-266.
Mirkes PE. 1987. Hyperthermia-induced heat shock response and thermo-
tolerance in postimplantation rat embryos. Dev Biol 119:115-122.
Mirkes PE, Cornel LM, Wilson KL, Dilmann WH. 1999. Heat shock pro-
tein 70 (Hsp70) protects postimplantation murine embryos from the
embryolethal effects of hyperthermia. Dev Dyn 214:159-170.
Mirkes PE, Doggett B. 1992. Accumulation of heat shock protein 72 (hsp
72) in postimplantation rat embryos after exposure to various periods
of hyperthermia (40 degrees -43 degrees C) in vitro: evidence that
heat shock protein 72 is a biomarker of heat-induced embryotoxicity.
Teratology 46:301-309.
Mirkes PE, Doggett B, Cornel L. 1994. Induction of a heat shock response
(HSP 72) in rat embryos exposed to selected chemical teratogens. Ter-
atology 49:135-142.
Mirkes PE, Little SA. 1998. Teratogen-induced cell death in postimplanta-
tion mouse embryos: differential tissue sensitivity and hallmarks of
apoptosis. Cell Death Differ 5:592-600.
Mirkes PE, Little SA. 2000. Cytochrome c release from mitochondria of
early postimplantation murine embryos exposed to 4-hydroperoxycy-
clophosphamide, heat shock, and staurosporine. Toxicol Appl Phar-
macol 162:197-206.
Mirkes PE, Little SA, Cornel L, et al. 1996. Induction of heat shock protein 27
in rat embryos exposed to hyperthermia. Mol Reprod Dev 45:276-284.
Mirkes PE, Little SA, Umpierre CC. 2001. Co-localization of active cas-
pase-3 and DNA fragmentation (TUNEL) in normal and hyperther-
mia-induced abnormal mouse development. Teratology 63:134—143.
Mirkes PE, Wilson KL, Cornel LM. 2000. Teratogen-induced activation of
ERK, JNK, and p38 MAP kinases in early postimplantation murine
embryos. Teratology 62:14-25.
Moretti ME, Bar-Oz B, Fried S, Koren G. 2005. Maternal hyperthermia
and the risk for neural tube defects in offspring: systematic review
and meta-analysis. Epidemiology 16:216-219.
Moseley PL. 1996. Heat shock proteins: a broader perspective. J Lab Clin
Med 128:233-234.
Mosser DD, Caron AW, Bourget L, et al. 1997. Role of the human heat
shock protein hsp70 in protection against stress-induced apoptosis.
Mol Cell Biol 17:5317-5327.
Mosser DD, Caron AW, Bourget L, et al. 2000. The chaperone function of
hsp70 is required for protection against stress-induced apoptosis.
Mol Cell Biol 20:7146-7159.
Nagata K. 1996. Regulation of thermotolerance and ischemic tolerance.
EXS 77:467^81.
Niu P, Liu L, Gong Z, et al. 2006. Overexpressed heat shock protein 70
protects cells against DNA damage caused by ultraviolet C in a dos-
e-dependent manner. Cell Stress Chaperones 11:162-169.
Nollen EA, Brunsting JF, Roelofsen H, et al. 1999. In vivo chaperone activ-
ity of heat shock protein 70 and thermotolerance. Mol Cell Biol 19:
2069-2079.
Padmanabhan R. 2006. Etiology, pathogenesis and prevention of neural
tube defects. Congenit Anom (Kyoto) 46:55-67.
Park HS, Lee JS, Huh SH, et al. 2001. Hsp72 functions as a natural inhibi-
tory protein of c-Jun N-terminal kinase. Embo J 20:446^56.
Putics A, Vodros D, Malavolta M, et al. 2008. Zinc supplementation
boosts the stress response in the elderly: Hsp70 status is linked to
zinc availability in peripheral lymphocytes. Exp Gerontol 43:
452-461.
Rafiee P, Theriot ME, Nelson VM, et al. 2006. Human esophageal micro-
vascular endothelial cells respond to acidic pH stress by PI3K/AKT
and p38 MAPK-regulated induction of Hsp70 and Hsp27. Am J
Physiol Cell Physiol 291:C931-945.
Ravagnan L, Gurbuxani S, Susin SA, et al. 2001. Heat-shock protein 70
antagonizes apoptosis-inducing factor. Nat Cell Biol 3:839-843.
Roesslein M, Schibilsky D, Muller L, et al. 2008. Thiopental protects
human T lymphocytes from apoptosis in vitro via the expression of
heat shock protein 70. J Pharmacol Exp Ther 325:217-225.
Rosa FW. 1991. Spina bifida in infants of women treated with carbamaze-
pine during pregnancy. N Engl J Med 324:674-677.
Saleh A, Srinivasula SM, Balkir L, et al. 2000. Negative regulation of the
Apaf-1 apoptosome by Hsp70. Nat Cell Biol 2:476^83.
Shi Y, Thomas JO. 1992. The transport of proteins into the nucleus
requires the 70-kilodalton heat shock protein or its cytosolic cognate.
Mol Cell Biol 12:2186-2192.
Stankiewicz AR, Lachapelle G, Foo CP, et al. 2005. Hsp70 inhibits heat-in-
duced apoptosis upstream of mitochondria by preventing Bax trans-
location. J Biol Chem 280:38729-38739.
Steel R, Doherty JP, Buzzard K, et al. 2004. Hsp72 inhibits apoptosis
upstream of the mitochondria and not through interactions with
Apaf-1. J Biol Chem 279:51490-51499.
Stege GJ, Kampinga HH, Konings AW. 1995. Heat-induced intranuclear
protein aggregation and thermal radiosensitization. Int J Radiat Biol
67:203-209.
Thayer JM, Mirkes PE. 1997. Induction of Hsp72 and transient nuclear
localization of Hsp73 and Hsp72 correlate with the acquisition and
loss of thermotolerance in postimplantation rat embryos. Dev Dyn
208:227-243.
Tsuchiya D, Hong S, Matsumori Y, et al. 2003. Overexpression of rat heat
shock protein 70 is associated with reduction of early mitochondrial
cytochrome C release and subsequent DNA fragmentation after per-
manent focal ischemia. J Cereb Blood Flow Metab 23:718-727.
Verheij M, Bose R, Lin XH, et al. 1996. Requirement for ceramide-initiated
SAPK/JNK signalling in stress-induced apoptosis. Nature 380:75-79.
Walsh DA, Klein NW, Hightower LE, Edwards MJ. 1987. Heat shock and
thermotolerance during early rat embryo development. Teratology
36:181-191.
Welch WJ. 1987. The mammalian heat shock (or stress) response: a cellu-
lar defense mechanism. Adv Exp Med Biol 225:287-304.
Yaglom JA, Gabai VL, Meriin AB, et al. 1999. The function of HSP72 in
suppression of c-Jun N-terminal kinase activation can be dissociated
from its role in prevention of protein damage. J Biol Chem 274:
20223-20228.
Birth Defects Research (Part A) 85:732-740 (2009)
Previous
TOC
-------
Genes and Immunity (2008), 1-8
© 2008 Macmillan Publishers Limited All rights reserved 1466-4879/08 $32.00
www.nature.com/gene
ORIGINAL ARTICLE
Integrated analysis of genetic and proteomic data
identifies biomarkers associated with adverse events
following smallpox vaccination
DM Reif1, AA Motsinger-Reif2, BA McKinney3, MT Rock4, JE Crowe Jr4-5-6 and JH Moore7-8
^National Center for Computational Toxicology, US Environmental Protection Agency, Research Triangle Park, NC, USA; ^Department
of Statistics, Bioinformatics Research Center, North Carolina State University, Raleigh, NC, USA; department of Genetics, University
of Alabama School of Medicine, Birmingham, AL, USA; ^Department of Pediatrics, Vanderbilt University Medical Center, Nashville,
TN, USA; 5Department of Microbiology and Immunology, Vanderbilt University Medical Center, Nashville, TN, USA; 6Program in
Vaccine Sciences, Vanderbilt University Medical Center, Nashville, TN, USA; 7Department of Genetics, Dartmouth Medical School,
Lebanon, NH, USA and 8Computational Genetics Laboratory, Dartmouth Medical School, Lebanon, NH, USA
Complex clinical outcomes, such as adverse reaction to vaccination, arise from the concerted interactions among the myriad
components of a biological system. Therefore, comprehensive etiological models can be developed only through the integrated
study of multiple types of experimental data. In this study, we apply this paradigm to high-dimensional genetic and proteomic
data collected to elucidate the mechanisms underlying the development of adverse events (AEs) in patients after smallpox
vaccination. As vaccination was successful in all of the patients under study, the AE outcomes reported likely represent the
result of interactions among immune system components that result in excessive or prolonged immune stimulation. In this
study, we examined 1442 genetic variables (single nucleotide polymorphisms) and 108 proteomic variables (serum cytokine
concentrations) to model AE risk. To accomplish this daunting analytical task, we employed the Random Forests (RF) method
to filter the most important attributes, then we used the selected attributes to build a final decision tree model. This strategy is
well suited to integrated analysis, as relevant attributes may be selected from categorical or continuous data. Importantly, RF is
a natural approach for studying the type of gene-gene, gene-protein and protein-protein interactions we hypothesize to be
involved in the development of clinical AEs. RF importance scores for particular attributes take interactions into account, and
there may be interactions across data types. Combining information from previous studies on AEs related to smallpox
vaccination with the genetic and proteomic attributes identified by RF, we built a comprehensive model of AE development that
includes the cytokines intercellular adhesion molecule-1 (ICAM-1 or CD54), interleukin-10 (IL-10), and colony stimulating
factor-3 (CSF-3 or G-CSF) and a genetic polymorphism in the cyokine gene interleukin-4 (IL4). The biological factors included
in the model support our hypothesized mechanism for the development of AEs involving prolonged stimulation of inflammatory
pathways and an imbalance of normal tissue damage repair pathways. This study shows the utility of RF for such analytical
tasks, while both enhancing and reinforcing our working model of AE development after smallpox vaccination.
Genes and Immunity advance online publication, 16 October 2008; doi:10.1038/gene.2008.80
Keywords: smallpox; Random Forests; integrated analysis; genetic; proteomic; interactions
Introduction
Live attenuated vaccinia virus, delivered intradermally,
is the vaccine given to immunize individuals against
smallpox. Although vaccination of healthy adults with
vaccinia virus induces a protective response in the
majority of individuals immunized, vaccinia virus is
reactogenic in a significant number of vaccinees.1 The
most common adverse events (AEs) after vaccination
include fever, lymphadenopathy (swelling and tender-
ness of lymph nodes) and a generalized acneiform rash.
Correspondence: Dr DM Reif, National Center for Computational
Toxicology, US Environmental Protection Agency, D343-03, 109 TW
Alexander Drive, Research Triangle Park, NC 27711, USA.
E-mail: reif.david@epa.gov
Received 11 June 2008; revised and accepted 27 August 2008
Collectively, these clinical reactions suggest that indivi-
duals suffering AEs have immune responses beyond the
necessary magnitude, or sustain the immune response
longer than necessary.
To elucidate the complex pathophysiology underlying
unwanted responses to vaccination, we gathered high-
dimensional genetic and proteomic data in a cohort of
subjects in which a portion experienced an AE after
primary immunization with Aventis Pasteur smallpox
vaccine. Through a comprehensive examination of
systemic (serum) cytokine/chemokine changes com-
bined with the characterization of polymorphisms in a
large panel of candidate genes, we sought to provide a
thorough portrayal of the complex genetic and proteomic
interplay behind the development of AEs. Knowledge
of how risk factors in a subject's genetic back-
ground interact with dynamically changing levels of
Previous
TOC
Next
-------
Integrated analysis of smallpox vaccination data
DMReifefa/
immunological proteins could shed light on important
therapeutic targets or pathways to direct vaccine
modification and pre-vaccination screening procedures.
It is increasingly gaining acceptance that complex
clinical outcomes, such as adverse reaction to vaccina-
tion, arise from the concerted interactions among the
myriad components of a biological system.2 Complica-
ting genetic factors, such as multiple contributing loci
and/or susceptibility alleles, incomplete penetrance and
epistasis, are further convoluted by proteomic, metabo-
lomic and environmental effects.3 If such a multiscale
system is to be understood, then interactions among its
many attributes must be considered.4 Although there is
considerable intuitive appeal to the incorporation of
multiple types of biological data, simultaneous analysis
of information on different scales of measurement (that
is, continuous proteomic data and categorical genetic
data) creates additional analytical challenges. Therefore,
appropriate computational analysis methods must
traverse large numbers of input variables and handle
diverse data types. For this study, we employed a two-
stage analytical strategy. The first step was to filter a list
of over 1500 genetic and proteomic attributes, taking
interactions within and across data types into account,
down to an analytically tractable subset of candidates.
The second step involved careful statistical and biolo-
gical exploration of the filtered subset of candidate
attributes, resulting in a final model of AE development.
For the first (filter) step, we implemented a random
forest™ (RF) approach.5 RF is a machine learning
technique that builds a forest of classification trees by
sampling, with replacement, from the data and selecting
the attribute at each tree node from a random subset of
all attributes. The RF method offers many advantages for
the analysis of diverse biological data. First, it can handle
a large number of input attributes, both discrete (for
example, single nucleotide polymorphisms, or SNPs)
and continuous (for example, microarray expression
levels or data from high-throughput proteomic techno-
logies). Second, RF estimates the relative importance of
attributes in discriminating between classes (in this case,
AE status), thus providing a metric for feature selection.
Third, RF produces a highly accurate classifier with an
internal unbiased estimate of generalizability during the
forest-building process. Fourth, RF is robust in the
presence of etiological heterogeneity and missing data.6
Finally, learning is fast and computation time is modest
even for very large data sets.7
In the second (modeling) step, we took advantage of
the tractable number of attributes identified by the RF
filter to explore thoroughly the statistical and biological
relationships among the attributes and AE outcomes.
Decision trees were used to derive a descriptive,
biologically interpretable model of the functional inter-
actions among the attributes associated with systemic
AEs. Our final model justified our multiscale analysis
strategy, in that it included the cytokines intercellular
adhesion molecule-1 (ICAM-1 or CD54), interleukin-10
(IL-10) and colony stimulating factor-3 (CSF-3 or G-CSF),
as well as an SNP in interleukin-4 (IL4). Evaluating our
final model from an immunological perspective, we
conclude that AEs in response to smallpox vaccination
result from the hyperactivation of inflammatory path-
ways, leading to excess recruitment and stimulation of
monocytes in peripheral tissues. This model is consistent
with work demonstrating overstimulation of inflamma-
tory and tissue damage repair pathways developed in
earlier studies of AEs after smallpox vaccination.8"11
Materials and methods
Study subjects
Vaccines, study subjects and clinical vaccine study
design have been described in detail.9 Briefly, 148 (116
with recorded AE information) healthy adults were
enrolled at the Vanderbilt University Medical Center as
part of a multicenter study of primary immunization
against smallpox using the Aventis Pasteur smallpox
vaccine at National Institutes of Health (NIH) Vaccine
and Treatment Evaluation Units. NIH-DMID Protocol 02-
054 was implemented. Volunteers were eligible if they
had no smallpox vaccination scar, no history of vaccinia
virus immunization, normal renal and hepatic serum
chemistry values, no contraindications against immuni-
zation (pregnancy, immunosuppression or eczema) and
negative serum test results for hepatitis B surface
antigen, hepatitis C virus antibody, rapid plasma reagin
and HIV-1 ELISA. There were a total of 61 subjects for
whom both genetic and proteomic data were gathered.
Individuals were asked to self-identify race; white (60)
and Asian (1) were the only categories identified in this
cohort. To facilitate comparison with earlier studies, and
because there was no statistical difference in age, gender
or race according to AE status (data not shown), the data
were not adjusted for these covariates.
Clinical assessments
Details of the clinical assessments have been described
earlier.9 For all study subjects, a team of trained
physicians and nurse providers examined the medical
history and clinical symptoms to ensure consistent
clinical assessment. Subjects were examined on five
visits within the first month after vaccination and were
assessed for occurrence of an AE. Collection of serum for
cytokine measurements occurred at the evaluation just
before vaccination (baseline) and at the evaluation
between days 5 and 7 post-vaccination (acute phase).
Although all AEs were noted, only systemic AEs were
considered in this study, as we expected these to be
associated more strongly with serum cytokine expression
than would an AE displayed only at the site of
inoculation. Systemic AEs included fever, generalized
rash and lymphadenopathy. Specifically, fever was
defined as an oral temperature of > 38.3 °C. Generalized
rash was defined as skin eruptions on non-contiguous
areas in reference to the site of vaccination. Detailed
descriptions of the acneiform rashes considered in this
study have been described.12 Lymphadenopathy was
defined as enlargement or tenderness of regional lymph
nodes attributed to vaccination. For subjects on which
both genetic and proteomic data were gathered, 16
subjects experienced a systemic AE and 45 subjects did
not experience an AE.
Identification of genetic polymorphisms
The custom SNP panel used in this study was based on
the NCI SNP500 Cancer project13 and has been described
earlier.14 The majority of SNPs included on the panel
target soluble factor mediators and signaling pathways,
Genes and Immunity
Previous
TOC
Next
-------
Integrated analysis of smallpox vaccination data
DMReifefa/
many of which have immunological significance.
Genotyping for SNPs was performed using DNA
amplified directly from Epstein-Barr virus-transformed
B cells generated from peripheral blood samples col-
lected from each subject. Genotyping was performed at
the Core Genotyping Facility of the National Cancer
Institute (NCI, Gaithersburg, MD, USA). Genotypes were
generated using the Illumina GoldenGate assay
technology. Of the 1536 SNPs assayed, a total of
1442 genotypes passed standard quality control filters.
In Reif et al.,15 the complete list of SNPs analyzed is
available.
Quantification of serum cytokine levels
Serum samples were obtained just prior to vaccination
(baseline) and 6-9 days after vaccination (acute), as
described earlier in detail.9 Serum samples were col-
lected in 5 ml Vacutainer serum separator tubes (Becton
Dickinson, San Jose, CA, USA) and were centrifuged at
700 x g for lOmin. The serum then was collected,
aliquoted into cryovials (Sarstedt Inc., Numbrecht,
Germany) and stored at -80 °C until assayed. Cytokine
concentrations were determined using rolling circle
amplification technology-enhanced custom dual anti-
body sandwich immunoassay arrays, as described.16"19
The expression levels of 108 protein analytes were
measured in 100 ul serum aliquots from the patient
samples. Glass slides held 12 replicate spots of mono-
clonal capture antibodies specific for each analyte.
Duplicate samples of sera were incubated for 2 h, washed
and then incubated with secondary biotinylated poly-
clonal antibodies. The 'rolling circle' method was then
used to amplify signals.17 Quality control measures were
used to optimize antibody pairs, minimize array-to-array
variation and standardize procedures of chip manufac-
turing.17 ATecan LS200 unit was used to scan arrays and
customized software was used to determine mean
fluorescence intensities. In addition, 15 serial dilutions
of recombinant analytes at known concentrations (stu-
died in parallel on each slide) were used to develop best-
fit equations for each analyte, and the upper and lower
limits of quantitation were defined. Changes in serum
cytokine concentrations were calculated as percent
change from the subject's baseline value because of the
broad individual range of systemic cytokine expression
before and after immunization.
Random Forests
An RF is a collection of decision tree classifiers,
where each tree in the forest has been trained using a
bootstrap sample of individuals from the data, and
each split attribute in the tree is chosen from among
a random subset of attributes. Classification of
individuals is on the basis of aggregate voting over all
trees in the forest.
Each tree in the forest was constructed as follows from
data having N=61 individuals and M = 1552 explana-
tory (genetic plus proteomic) attributes:
(1) The method chose a training sample by selecting
N individuals, with replacement, from the entire data
set.
(2) At each node in the tree, m attributes were selected
randomly from the entire set of M attributes in the
data. The absolute magnitude of m was a function of
the number of attributes in the data set and remained
constant throughout the forest-building process.
The method chose the best split at the current node
from among the subset of m attributes selected above.
We iterated the second and third steps until the tree
(3)
(4)
was fully grown (no pruning).
Repetition of this algorithm yielded a forest of trees, each
of which had been trained on bootstrap samples of
individuals (see Figure 1). Thus, for a given tree, certain
individuals were left out during training. Prediction
error and attribute importance were estimated from
these 'out-of-bag' individuals.
The out-of-bag (unseen) individuals were used to
estimate the importance of particular attributes accord-
ing to the following logic: If randomly permuting values
of a particular attribute did not affect the predictive
ability of trees on out-of-bag samples, then that attribute
was assigned a low importance score. If, however,
randomly permuting the values of a particular attribute
drastically impaired the ability of trees to correctly
predict the class of out-of-bag samples, then the
importance score of that attribute was high. By running
out-of-bag samples down entire trees during the permu-
tation procedure, attribute interactions were taken into
account when calculating importance scores, as class was
assigned in the context of other attribute nodes in the
tree.
The recursive partitioning trees comprising an RF
provide an explicit representation of attribute interaction
that is readily applicable to the study of interactions
among multiple data types.20-21 These models may
uncover interactions among genes, proteins and/or
environmental factors that do not exhibit strong marginal
effects. In addition, tree methods are suited to dealing
with certain types of genetic heterogeneity, as splits near
the root node define separate model subsets in the data.
RFs capitalize on the solid benefits of decision trees and
M attributes
M attributes
Entire
datasct
Bootstrap
sample
Out-of-bag
individuals
m attributes
Stepl
Step 2
Step 3
Figure 1 Construction of individual trees using the Random Forest
method from a full data set of N individuals and M attributes.
Proceeding from the root node, individual subjects were classified
into terminal AE status leaves according to the value of that
individual's genetic or proteomic attribute at each node. The steps
correspond to those described in the text.
Previous
Genes and Immunity
-------
Integrated analysis of smallpox vaccination data
DMReifefa/
have demonstrated excellent predictive performance
when the forest is diverse (that is, trees are not highly
correlated with each other) and composed of individu-
ally strong classifier trees.5'22 The RF method is a natural
approach for studying gene-gene, gene-protein or
protein-protein interactions because importance scores
for particular attributes take interactions into account
without demanding a pre-specified model.23
Decision trees
To represent the interactions among genetic and/or
proteomic attributes associated with AEs, decision trees
were chosen to build the final model because of their
ready interpretability and explicit modeling of attribute
interactions. The tree classified individual subjects into
AE groups by proceeding down a dichotomous tree,
where the genetic or proteomic attribute at each node (or
split) was selected for the gain in information it
provided. Gain in information was attributed when
knowledge about the variation in this attribute separated
subjects into appropriate AE classes. When interpreting
the tree, attributes at each node were taken in the context
of attributes at nodes closer to the root—thus allowing an
explicit representation of attribute interactions. To aug-
ment the generalizability of our final model, we
stipulated that at least five subjects must appear in each
terminal (status) leaf and used 10-fold cross-validation
(CV) to estimate the predictive ability of the final model.
Although CV accuracy was reduced by allowing trees
with less than five subjects in terminal nodes, CV
accuracy proved to be insensitive to changes in other
tree parameters for these data. We used the implementa-
tion of the C4.5 decision-tree algorithm provided in the
Weka machine learning software package to obtain our
final model.24
Data analysis strategy
RF analysis was performed using the freely available R
package randomForest.25-26 This package is based on the
original Fortran code available at the website cited in
Breiman and Cutler.27 RF was used to analyze data sets
containing each biological data type separately and in
parallel, resulting in two stratified data sets (genetic only;
proteomic only) and a combined data set (both genetic
and proteomic attributes). Genetic attributes were treated
as categorical, whereas proteomic attributes were treated
as continuous values. For each genetic, proteomic, or
combined data set, forests comprised of 10 000 trees were
grown. Attribute importance was calculated using the
out-of-bag permutation test described above. The relative
importance (rank) of functional genetic attributes and
related proteomic attributes was determined from the
mean decrease in the Gini index using the out-of-bag
permutation testing procedure. The relative importance
determined from the mean decrease in classification
accuracy produced nearly identical results both here and
in extensive simulation studies.28
The simulation studies28 used the current data as the
basis for a range of simulated models, providing
guidance for the parameters in the analysis discussed
here. The relative rank of simulated genetic and
proteomic predictors was evaluated for a range of filter
cutoffs and on both stratified and combined data sets.
Results from these data-based simulation studies demon-
strated high confidence that AE-associated attributes
Genes and Immunity
Previous
having relatively meager effects would be ranked in the
top 10% of attributes in RF analysis, and that analysis of
the combined (genetic and proteomic) data was generally
advantageous. Therefore, we chose the top 10% of
attributes as ranked by RF as candidates for inclusion
in our final model. To represent the interactions among
genetic and/or proteomic attributes associated with AEs,
we built a decision tree model.
Biological interpretation of our final model was aided
by Chilibot (chip literature robot) knowledge mining
software.29 Chilibot inferred relationship networks
among the attributes in the final model based on
linguistic analysis of relevant records from public
biomedical literature databases. The natural language
processing approach used by Chilibot is superior to
standard co-occurrence text mining approaches because
parsing text into sentences can characterize the type of
relationship (for example, inhibition or stimulation)
between input terms. The terms given explicitly to
Chilibot as input were 'ICAM-1,' 'IL-10,' 'IL-4' and
'CSF-3/ as well as the alternate gene names 'CD54' and
'G-CSF (for ICAM-1 and CSF-3, respectively). The
software automatically adds syntactic synonyms (for
example, "IL 10,' 'IL-10,' 'IL10,' and so on) to the search
criteria. Because the goal of this study is hypothesis
generation, as opposed to strict hypothesis testing,
Chilibot was used to aid in discovery rather than using
any pre-defined network relationships.
Results
Filtering of important attributes using RFs
Supplementary Table 1 lists all attributes having an
importance rank in the top 10% relative to all attributes
in the combined data set. Figure 2 depicts the attribute
importance score landscape over the entire data set. This
landscape proved robust to changes in RF parameters,
provided that a sufficiently large forest (10 000 trees) was
grown. RF identified both genetic and proteomic
attributes as important discriminators of AE status.
Approximately one-third of the attributes identified as
important were genetic, with the remaining two-thirds
being proteomic. Although this distribution among data
Importance scores for all attributes
High
Attribute importance
Figure 2 Attribute importance 'landscape' showing the shape of
the importance curve ranking all attributes in the combined (genetic
plus proteomic) data set. Attributes above the horizontal line
indicate a relative importance rank in the top 10% (90th percentile)
of all attributes in the data set.
-------
Integrated analysis of smallpox vaccination data
DMReifefa/
types may reflect systematic patterns concerning the
etiology of AE outcomes, the bias toward proteomic
attributes probably arose out of the fact that the cytokine
array was specifically designed to capture variation in
important systemic mediators. In contrast, the genetic
data include candidate SNPs in and around genes having
a variety of immunological functions. In addition, with
multiple SNPs per gene, correlation existing among
polymorphisms (that is, haplotypes) could drive down
RF importance scores for particular SNPs, as RF might
select any SNP from within a haplotype at a particular
node. Indeed, the IL4 SNP in our final model was part of
a group of four SNPs in IL4 having nearly identical
importance scores, and Haploview analysis showed
them to be in high linkage disequilibrium, providing
evidence that these genetic polymorphisms are inherited
as a haplotype.30 In this context, linkage disequilibrium
has an impact akin to etiological heterogeneity, which is a
concern in any association study. The heterogeneity
concern is part of the rationale for using RF as a first-
stage filter that identifies a handful (the top 10%) of
attributes for further consideration. The effect of
repeated samplings over many thousands of trees gives
all attributes an unbiased opportunity to demonstrate AE
association, even if importance scores for groups of SNPs
in linkage disequilibrium are slightly tamped down.
Thus, attributes whose importance scores may be
tamped down by phenomena such as linkage disequili-
brium still have a chance to surpass our 10% importance
threshold over a sufficiently large forest of resampled
trees, whereas slightly down-weighted importance
scores may push interesting attributes below an overly
strict first-stage threshold in a smaller forest. Consider-
ing the RF importance rank of attributes included in our
final model relative to all attributes in the combined data
set, all three proteomic attributes were ranked in the top
1%, and the IIA SNP (rs 2243290) was ranked in the top
5%. Relative to their respective data types, the IIA SNP
was ranked in the top 1% among all attributes in the
genetic data set, and ICAM-1, CSF-3 and IL-10 were
ranked in the top 1% among all proteomic attributes.
Modeling the association of genetic and proteomic biomarkers
with AEs
Having filtered out the noise using RFs, we used a
decision tree representation to explore interactions
among the attributes in our filtered list related to AE
status. The final decision tree model is shown in Figure 3.
Our final model included four variables—three proteo-
mic attributes and one genetic attribute. Change in
ICAM-1 concentration comprises the root node of the
tree, with subsequent nodes composed of change in IL-10
concentration, a SNP in IL4, and change in CSF-3
concentration. Imposing our minimum of five indivi-
duals per terminal (AE status) leaf, this tree correctly
classified 89% of individuals (with seven misclassifica-
tions) in the full data set and achieved a 10-fold CV
(prediction) accuracy of 75%.
Figure 4 characterizes the biological relationships
among the attributes in the tree using Chilibot. Inter-
active relationships were characterized into one of three
types based on the verbs connecting pairs of attributes in
the biomedical literature as follows: (1) Stimulatory
relationships were connected by verbs such as 'activate,'
'stimulate' or 'enhance.' (2) Inhibitory relationships were
Previous
(7/2)
Figure 3 Final model of genetic and proteomic factors contributing
to AE development. Each node (oval) constitutes a decision point
based on the genotype of genetic attributes (IL4 SNP) or whether the
concentration change from baseline in proteomic attributes (ICAM-
1, IL-10 and CSF-3) was above (upward-pointing arrows) or below
(downward-facing arrows) a calculated threshold. Starting at the
root node (ICAM-1), subjects were classified into AE status leaves
(rectangles) by proceeding along the decision points at each
attribute node. Given below each terminal leaf is the total number
of subjects classified into that AE status group/the number of
subjects incorrectly assigned to that AE status group.
Figure 4 Biological relationships among the attributes in our final
model characterized using Chilibot. Connections between each
attribute node (oval) are denoted according to the type of interactive
relationship they represent: stimulatory (solid), both stimulatory
and inhibitory (dotted) or neutral (dashed). Arrowheads indicate
that interactions between particular biological attributes are bi-
directional.
connected by verbs such as 'decrease,' 'attenuate' or
'inhibit.' (3) Neutral relationships were assigned when
the nature of the relationship could not be determined
contextually. Mining the biomedical literature suggested
interactive relationships connecting all of the attribute
nodes in our final model. Stimulatory, inhibitory or
neutral pair-wise interactive relationships were identi-
fied between each of ICAM-1, IL-10, IIA and CSF-3.
Genes and Immunity
-------
Integrated analysis of smallpox vaccination data
DMReifefa/
Thorough examination of the networks inferred facili-
tated the biological interpretation of the final model
discussed below.
Discussion
Our final model provides an immunologically plausible
and testable biological mechanism of AE occurrence after
smallpox vaccination that includes both genetic and
proteomic factors. The analytical strategy used is appro-
priate for the study of complex phenotypes, as outcomes
such as AE development likely result from the interplay
of multiple genetic, proteomic and environmental fac-
tors.31'32 The decision tree trained on the attributes
passing our RF filter proposes a solid biological model
of AE development.
The attributes included in this tree point to an
important role of one particular immune cell type: is
monocytes. Monocytes are bone marrow-derived circu-
lating blood cells that are precursors of tissue macro-
phages. Monocytes are recruited actively to the sites of
inflammation, where they differentiate into macrophages
in tissues. These macrophages play important roles in
coordinating both innate and adaptive immune re-
sponses. Macrophages are activated by microbial pro-
ducts such as endotoxin and by T-cell cytokines such as
interferon-y. Activated macrophages phagocytose and
kill microorganisms, secrete pro-inflammatory cytokines
and present antigens to helper T cells.
The root node of the tree we developed is ICAM-1
(CD54), where small changes from baseline concentra-
tion (<11%) of ICAM-1 predict a non-AE response to
vaccination and high changes from baseline concentra-
tion (>11%) point toward AE risk, depending on factors
in subsequent nodes. ICAM-1 is mainly expressed on
endothelial cells, T cells, B cells and monocytes. It
functions in cell-cell adhesion, which plays a crucial
role in monocyte differentiation into macrophages, as
entry into tissues is necessary. In addition, ICAM-1
expression is upregulated in mature monocytes,33 aiding
in cell adhesion and the eventual differentiation into
macrophages. Circulating monocytes are in random
contact with endothelial cells, and the adhesion molecule
E-selectin slows the monocyte by inducing rolling of the
monocyte along the endothelial surface before firm
attachment to vascular cell adhesion molecule 1 or
ICAM-1, which interact with integrins on the monocyte
surface. Once the monocyte is tightly bound, it then
migrates between endothelial cells.34'35 Excessive levels
of ICAM-1 might cause an 'over-recruitment' of mono-
cytes into tissue, triggering an unnecessarily active
innate inflammatory response.
For individuals with large changes in ICAM-1, the next
node in the tree is IL-10, where changes from baseline
>85% are associated with AEs. IL-10 is produced by
activated macrophages and some helper T cells for which
a major function is to inhibit activated macrophages and,
therefore, to maintain homeostatic control of innate and
cell-mediated immune reactions. Changes in IL-10 levels
may indicate an imbalance in this delicate homeostasis,
leading to AEs.
For individuals with mild changes in IL-10 concentra-
tion, the next node is an SNP in the gene encoding IL-4.
In an earlier genetic study of two vaccination cohorts
(including a subset of individuals in the present data),
this same IL-4 polymorphism was associated with AEs
(P = 0.05 and P = 0.06 in the first and second cohort,
respectively).15 Interestingly, by including proteomic
factors in this study, our model indicates that the AE
risk conferred by this SNP is dependent on proteomic
context. IL-4 is a cytokine produced mainly by the TH2
subset of CD4+ helper T cells, whose functions include
the induction of the differentiation of TH2 cells from
naive CD4 + precursors, stimulation of IgE production by
B cells and suppression of interferon-y-dependent
macrophage functions.36"38 Although direct functional
significance of the SNP is unknown, it is reasonable that
the different genotypes could result in functionally
different versions of the IL-4 protein or in different
bioavailability levels of IL-4. The fact that multiple SNPs
in IL4 achieved nearly identical importance scores
indicates that variation within the IL4 gene region may
be related functionally to the development of AEs.
Because of the intricate cross-talk between macrophages
and the TH2 response in maintaining homeostasis, it is
plausible that the major IL4 genotype (CC) is associated
with calming the activated macrophage response and
directing the acquired immune system to progress in
response to vaccine presentation, whereas the variant
genotypes (AC or AA) fail to calm the innate response,
presenting increased AE risk.
For individuals having one of the variant genotypes at
IL4, the lowest node of the tree is CSF-3 (G-CSF). G-CSF
is a cytokine produced by activated T cells, macrophages
and endothelial cells at the sites of infection, which acts
on the bone marrow to mobilize and increase the
production of neutrophils to replace those consumed in
inflammatory reactions. In our model, increased levels of
CSF-3 after vaccination (change >78%) indicated in-
creased risk of suffering an AE. This finding implies
another possible over-recruitment event in the develop-
ment of AEs, as neutrophils have been associated with
host tissue damage and failure to terminate acute
inflammatory responses.39 This reaction is consistent
with the types of AE symptoms observed in this study
and with the overall proposed biological mechanisms of
AE development.
The results of this study provide a viable biological
hypothesis of AE occurrence after smallpox vaccination
that is experimentally testable. Our model includes both
genetic and proteomic biomarkers. Allowing for such an
integrative model is an important strength of our
analytical strategy. It is increasingly recognized that the
pathophysiology of complex clinical outcomes hinges on
biological factors acting on multiple levels.40 Therefore,
the formulation of robust etiological models must take
this inherent complexity into account and capitalize on
the power of modern experimental data-generating
techniques.
We conclude that AEs after smallpox vaccination result
from hyperactivation of inflammatory signals, leading to
excess recruitment and stimulation of monocytes in
peripheral tissues. Our analysis identifies a set of
interacting genetic and proteomic candidates associated
with AEs, such as ICAM-1, IL-10, IIA and CSF-3. As the
proteomic measurements occurred early in the period
after vaccination, before most AEs presented themselves
clinically, our model could be used as a diagnostic tool in
the prediction of AEs. Of course, the ultimate goal of
Genes and Immunity
Previous
TOC
Next
-------
Integrated analysis of smallpox vaccination data
DM Reif eta/
such a study is the identification and characterization of
biological risk factors contributing to the inappropriate
immune response to vaccination. We present a hypo-
thesized mechanism of AE development that targets
specific elements of systemic inflammatory pathways for
further study.
Future studies should further evaluate the reproduci-
bility of the current model, given that the number of
vaccinated subjects meeting the criteria for inclusion and
having both genetic and proteomic data was relatively
small. Ideally, our model would be evaluated for
replication in an entirely independent sample. However,
the validity of our current model can be assessed through
the statistical process of internal CV (where our model
achieved a 75% prediction accuracy) and through
comparison of these results with our earlier studies of
genetic15 or proteomic8 data alone. In this study, our RF
approach with the combined data identified all attributes
highlighted in the earlier proteomic study8 (ICAM-1,
CSF-3, Eotaxin and TIMP-2) and two of the three genes
highlighted in the earlier genetic study15 (MTHFR and
IIA but not IRF1). Although the IRF1 polymorphisms
were not ranked in the top 10% of all attribute
importance scores in the combined data set, these
attributes would have passed the top 10% filter criteria
relative to only genetic attributes. Given that the subset
of subjects used in this study (that is, those having both
genetic and proteomic data) has only partial overlap
with subjects in either of the earlier studies, we feel that
the current results are remarkably stable.
Finally, our hypothesized model must be tested at the
bench. The functional consequences of genetic variability
in IL4 should be characterized fully. Time series studies
with dense measurement points are needed to shed light
on the dynamic interplay between the signaling of
ICAM-1, IL-10 and CSF-3. Additional data are needed
on the effects of these cytokines in other physiological
compartments. Careful assessment of external factors
(such as nutrition, fitness and relevant environmental
exposures) influencing protein expression should be
considered in future studies. The results from this study
suggest that analysis of the molecular and cellular basis
of complex clinical phenomena will require an experi-
mental approach that takes into account the broader
spatial and temporal physiological context of complex
biological systems.
Acknowledgements
This work was supported by the National Institutes of
Health (NIH)/National Institute of Allergy and Infec-
tious Diseases (NIAID) Vaccine Trials and Evaluation
Unit (contract N01-AI-25462, study DMID 02-054); NIH/
NIAID (Grants R21-AI-59365, K25-AI-064625 and R01-
AI-59694) and NIH/National Institute of General Med-
ical Sciences (NIGMS) (Grant R01-GM-2758). Cytokine
analysis was a kind gift of Stephen Kingsmore, PhD, and
Molecular Staging Incorporated. Genotype analysis was
a kind gift of Stephen Chanock, MD, and the NCI Center
for Cancer Research. Kathryn Edwards, MD, coordinated
the original acquisition of the data analyzed for this
study. The United States Environmental Protection
Agency (EPA), through its Office of Research and
Previous
Development, collaborated in the research described
here. It has been subjected to Agency review and
approved for publication, although it does not necessa-
rily represent the views or polices of the US EPA.
References
1 Kemper AR, Davis MM, Freed GL. Expected adverse events in a
mass smallpox vaccination campaign. Eff Clin Pract 2002; 5: 84-90.
2 Reif DM, White BC, Moore JH. Integrated analysis of genetic,
genomic and proteomic data. Expert Rev Proteomics 2004; 1:
67-75.
3 Maniolo TA, Collins FS. Genes, environment, health, and
disease: facing up to complexity. Hum Hered 2007; 63: 63-66.
4 Nicholson JK. Global systems biology, personalized medicine
and molecular epidemiology. Mol Syst Biol 2006; 3: 1-6.
5 Breiman L. Random forests. Much Learn 2001; 45: 5-32.
6 Lunetta KL, Hayward LB, Segal J, Van EP. Screening large-
scale association study data: exploiting interactions using
random forests. BMC Genet 2004; 5: 32.
7 Robnik-Sikonja M. Improving random forests. Proc Eur Conf
Much Learn 2004; 3201: 359-370.
8 McKinney BA, Reif DM, Rock MT, Edwards KM, Kingsmore
SF, Moore JH et al. Cytokine expression patterns associated
with systemic adverse events following smallpox immuniza-
tion. / Infect Dis 2006; 194: 444-453.
9 Rock MT, Yoder SM, Talbot TR, Edwards KM, Crowe Jr JE.
Adverse events after smallpox immunizations are associated
with alterations in systemic cytokine levels. / Infect Dis 2004;
189: 1401-1410.
10 Rock MT, Yoder SM, Talbot TR, Edwards KM, Crowe Jr JE.
Cellular immune responses to diluted and undiluted Aventis
Pasteur smallpox vaccine. / Infect Dis 2006; 194: 435-443.
11 Talbot TR, Stapleton JT, Brady RC, Winokur PL, Bernstein DI,
Germanson T et al. Vaccination success rate and reaction
profile with diluted and undiluted smallpox vaccine: a
randomized controlled trial. JAMA 2004; 292: 1205-1212.
12 Talbot TR, Bredenberg HK, Smith M, LaFleur BJ, Boyd A,
Edwards KM. Focal and generalized folliculitis following
smallpox vaccination among vaccinia-naive recipients. JAMA
2003; 289: 3290-3294.
13 Garcia-Closas M, Malats N, Real FX, Yeager M, Welch R,
Silverman D et al. Large-scale evaluation of candidate genes
identifies associations between VEGF polymorphisms and
bladder cancer risk. PLoS Genet 2007; 3: e29.
14 Packer BR, Yeager M, Burdett L, Welch R, Beerman M, Qi L et al.
SNPSOOCancer: a public resource for sequence validation, assay
development, and frequency analysis for genetic variation in
candidate genes. Nucleic Acids Res 2006; 34: D617-D621.
15 Reif DM, McKinney BA, Motsinger-Reif AA, Chanock SJ,
Edwards KM, Rock MT et al. Genetic basis for adverse events
after smallpox vaccination. / Infect Dis 2008; 198: 16-22.
16 Kader HA, Tchernev VT, Satyaraj E, Lejnine S, Kotler G,
Kingsmore SF et al. Protein microarray analysis of disease
activity in pediatric inflammatory bowel disease demonstrates
elevated serum PLGF, IL-7, TGF-betal, and IL-12p40 levels in
Crohn's disease and ulcerative colitis patients in remission
versus active disease. Am J Gastroenterol 2005; 100: 414^23.
17 Perlee L, Christiansen J, Dondero R, Grim wade B, Lejnine S,
Mullenix M et al. Development and standardization of
multiplexed antibody microarrays for use in quantitative
proteomics. Proteome Sci 2004; 2: 9.
18 Schweitzer B, Wiltshire S, Lambert J, O'Malley S, Kukanskis K,
Zhu Z et al. Inaugural article: immunoassays with rolling circle
DNA amplification: a versatile platform for ultrasensitive
antigen detection. Proc Natl Acad Sci USA 2000; 97: 10113-10119.
19 Schweitzer B, Roberts S, Grimwade B, Shao W, Wang M, Fu Q
et al. Multiplexed protein profiling on microarrays by rolling-
circle amplification. Nat Biotechnol 2002; 20: 359-365.
Genes and Immunity
TOC
Next
-------
Integrated analysis of smallpox vaccination data
DM Reif eta/
20 Breiman L, Friedman JH, Olshen RA, Stone CJ. Classification
and Regression Trees. Chapman & Hall: New York, 1984.
21 Province MA, Shannon WD, Rao DC. Classification methods
for confronting heterogeneity. Adv Genet 2001; 42: 273-286.
22 Bureau A, Dupuis J, Falls K, Lunetta KL, Hayward B, Keith TP
et al. Identifying SNPs predictive of phenotype using random
forests. Genet Epidemiol 2005; 28: 171-182.
23 McKinney BA, Reif DM, Ritchie MD, Moore JH. Machine
learning for detecting gene-gene interactions: a review. Appl
Bioinformatics 2006; 5: 77-88.
24 Witten IH, Frank E. Data Mining: Practical Machine Learning
Tools and Techniques, 2nd edn. Morgan Kaufmann: San
Francisco, 2005.
25 Ihaka R, Gentleman R. R: a language for data analysis and
graphics. / Comput Graph Stat 1996; 5: 299-314.
26 R Development Core Team. R: a language and environment
for statistical computing. R foundation for statistical comput-
ing. Available at http://www.R-project.org, 2006.
27 Breiman L, Cutler A. Random forests. Available at http://
www.stat.berkeley.edu/ ~ breiman/RandomForests/cc_home.
htm, 2004.
28 Reif DM, Motsinger AA, McKinney BA, Crowe Jr JE, Moore JH.
Feature selection using a random forests classifier for the
integrated analysis of multiple data types. In: Proceedings of the
IEEE Symposium on Computational Intelligence in Bioinformatics
and Computational Biology, 2006, pp 171-178.
29 Chen H, Sharp BM. Content-rich biological network con-
structed by mining PubMed abstracts. BMC Bioinformatics
2004; 8: 5-147.
30 Barrett JC, Fry B, Mailer J, Daly MJ. Haploview: analysis and
visualization of LD and haplotype maps. Bioinformatics 2005;
21: 263-265.
31 Moore JH. The ubiquitous nature of epistasis in determining
susceptibility to common human diseases. Hum Hered 2003;
56: 73-82.
32 Wilke RA, Reif DM, Moore JH. Combinatorial pharmacoge-
netics. Nat Rev Drug Discov 2005; 4: 911-918.
33 Most J, Schwaeble W, Drach J, Sommerauer A, Dierich MP
Regulation of the expression of ICAM-1 on human mono-
cytes and monocytic tumor cell lines. / Immunol 1992; 148:
1635-1642.
34 Peters W, Charo IF. Involvement of chemokine receptor 2 and
its ligand, monocyte chemoattractant protein-1, in the devel-
opment of atherosclerosis: lessons from knockout mice. Curr
Opin Lipidol 2001; 12: 175-180.
35 Zittermann SI, Issekutz AC. Basic fibroblast growth factor
(bFGF, FGF-2) potentiates leukocyte recruitment to inflamma-
tion by enhancing endothelial adhesion molecule expression.
Am J Pathol 2006; 168: 835-846.
36 Eslick J, Scatizzi JC, Albee L, Bickel E, Bradley K, Perlman H.
IL-4 and IL-10 inhibition of spontaneous monocyte apoptosis
is associated with Flip upregulation. Inflammation 2004; 28:
139-145.
37 Mangan DF, Robertson B, Wahl SM. IL-4 enhances
programmed cell death (apoptosis) in stimulated human
monocytes. / Immunol 1992; 148: 1812-1816.
38 Soruri A, Kiafard Z, Dettmer C, Riggert J, Kohl J, Zwirner J. IL-
4 down-regulates anaphylatoxin receptors in monocytes and
dendritic cells and impairs anaphylatoxin-induced migration
in vivo. J Immunol 2003; 170: 3306-3314.
39 Serhan CN, Savill J. Resolution of inflammation: the beginning
programs the end. Nat Immunol 2005; 6: 1191-1197.
40 Hood L. Systems biology: integrating technology, biology, and
computation. Mech Ageing Dev 2003; 124: 9-16.
Supplementary Information accompanies the paper on Genes and Immunity website (http://www.nature.com/gene)
Genes and Immunity
Previous
TOC
Next
-------
ELSEVIER
Available online at www.sciencedirect.com
-"• ScienceDirect
Toxicology in Vitro 22 (2008) 296-300
Toxicology
in Vitro
www. elsevier. com/locate/toxinvit
9,10-Phenanthrenequinone induces DNA deletions and forward
mutations via oxidative mechanisms in the
yeast Saccharomyces cerevisiae
Chester E. Rodriguez a>1, Zhanna Sobol b>1, Robert H. Schiestl b'*
a Department of Pharmacology, Geffen School of Medicine, Center for the Health Sciences, Los Angeles, CA 90095-1735, USA
Departments of Pathology and Environmental Health Sciences, Geffen School of Medicine and School of Public Health, UCLA, Los Angeles, CA, USA
Received 15 June 2007; accepted 4 September 2007
Available online 12 September 2007
Abstract
The estimated cancer risk from diesel exhaust particles (DEP) in the air is approximately 70% of the cancer risk from all air pollutants.
DEP is comprised of a complex mixture of chemicals whose carcinogenic potential has not been adequately assessed. The polycyclic aro-
matic hydrocarbon quinone 9,10-phenanthrenequinone (9,10 PQ) is a major component of DEP and a suspect genotoxic agent for DEP
induced DNA damage. 9,10 PQ undergoes redox cycling to produce reactive oxygen species that can lead to oxidative DNA damage.
We used two systems in the yeast Saccharomyces cerevisiae to examine possible differential genotoxicity of 9,10 PQ. The DEL assay
measures intra-chromosomal homologous recombination leading to DNA deletions and the CAN assay measures forward mutations
leading to canavanine resistance. Cells were exposed to 9,10 PQ aerobically and anaerobically followed by DNA damage assessment.
The results indicate that 9,10 PQ induces DNA deletions and point mutations in the presence of oxygen while exhibiting negligible effects
anaerobically. In contrast to the cytotoxicity observed aerobically, the anaerobic effects of 9,10 PQ seem to be cytostatic in nature, reduc-
ing growth without affecting cell viability. Thus, 9,10 PQ requires oxygen for genotoxicity while different toxicities exhibited aerobically
and anaerobically suggest multiple mechanisms of action.
© 2007 Elsevier Ltd. All rights reserved.
Keywords: 9,10-Phenanthraquinone; 1,4-Benzoquinone; DNA deletions; DEL assay; Canavanine assay; Forward mutations; Saccharomyces cerevisiae;
Yeast
1. Introduction
Exposure to diesel exhaust particles (DEP) in urban air
represents an important cancer risk factor. According to
the Multiple Air Toxics Exposure Study for the South
Coast Air Basin (MATES II study) in Southern Califor-
nia, DEP emissions represent 70% of all carcinogenic risk
from ambient measurements in the South Coast Air Basin
(2000). The levels of airborne DEP have increased dra-
Abbreviations: 9,10 PQ, 9,10-Phenanthrenequinone; 1,4 BQ, 1 4-Benzo-
quinone; DEP, Diesel Exhaust Particles.
* Corresponding author. Tel.: +310 267 2087; fax: +310 267 2578.
E-mail address: rschiestl@mednet.ucla.edu (R.H. Schiestl).
1 Contributed Equally.
0887-2333/S - see front matter © 2007 Elsevier Ltd. All rights reserved.
doi:10.1016/j.tiv.2007.09.001
matically over the last few decades due to the increased
use of diesel-based engines, which provide higher fuel effi-
ciency and lower carbon dioxide emissions than gasoline-
based engines, but emit between 30-100 times more par-
ticulate matter into the atmosphere (Ma and Ma, 2002;
Peterson and Saxon, 1996). It has been estimated that
DEP constitute as much as 40% of the respirable particu-
late matter in a city such as Los Angeles where the daily
human intake has been estimated to be as much as 300 (ig
(Ma and Ma, 2002). Considering that the lung clearance
of DEP has been estimated to be about 18 days, signifi-
cant accumulation can be expected to take place (Sun
et al., 1984).
The composition of DEP consists of an inert carbona-
ceous core onto which a complex mixture of chemical
Previous
-------
C.K Rodriguez et al. I Toxicology in Vitro 22 (2008) 296-300
297
entities is adsorbed. Over 450 different chemical species
have been identified in the organic layer of DEP including
transition metals, a variety of polycyclic aromatic hydro-
carbons (PAHs), nitroaromatic hydrocarbons, quinones,
aldehydes, ketones, aliphatic hydrocarbons, and heterocy-
clic compounds (Li et al., 2004). It is not clear, however,
which components from the complex mixture of chemicals
mediate carcinogenicity induced by DEP, but quinones
represent suspect agents in this process since DEP-induced
toxicity is consistent with quinone-like chemistry (Arimoto
et al., 1999; Bai et al., 2001; Sagai et al., 1993). Quinones
such as 9,10 PQ are substantial constituents of DEP and
can mediate the production of reactive oxygen species,
potentially leading to oxidative DNA damage (Bolton
et al., 2000). Moreover, quinones can also exert DNA dam-
age through their actions as electrophiles, resulting in cova-
lent adducts at nucleophilic centers of DNA. In particular,
9,10 PQ is one of the most toxic quinones found in DEP
with measured levels of about 24 (ig per gram of DEP
(Cho et al., 2004). In this study, the genotoxicity profile
of 9,10 PQ was examined by two different assays estab-
lished in the yeast S. cerevisae. The yeast DEL assay is a
measure of chromosomal rearrangements (Schiestl, 1989;
Schiestl et al., 1989) while the CAN assay represents a mea-
sure for point mutations (Whelan et al., 1979). The DEL
assay measures a DNA deletion event that occurs as a
result of intrachromosomal homologous recombination
between 400 base pair repeats separated by 6 k basepairs
of DNA. An elevated frequency of homologous recombi-
nation is associated with genomic instability and an
increased risk of cancer (Bishop and Schiestl, 2003). Thus,
the DEL assay measures a genotoxic endpoint that is rele-
vant in carcinogenesis. The CAN assay measures resistance
to the toxic compound canavanine. Canavanine is an ana-
log of arginine and enters the cell via the arginine-specific
permease CAN1. A forward point mutation event disrupts
the function of CAN1 thereby inhibiting the uptake of can-
avanine and rendering the cell canavanine resistant. The
goal of this study is to gain insight about how 9,10 PQ
leads to DNA damage. Incubations were carried out under
aerobic and anaerobic conditions to distinguish redox
cycling from the direct actions of the quinone. Since 9,10
PQ is not considered an electrophile, observable DNA
damage was expected to be strictly aerobic. For compari-
son purposes, the genotoxicity of 1,4-benzoquinone (1,4
BQ) was also examined as a quinone whose toxicity has
been exclusively attributed to its actions as an electrophile
and does not undergo redox cycling (Kondrova et al.,
2007).
2. Materials and methods
2.1. Chemicals
9,10-Phenanthrenequinone (CAS 84-11-7), 1,4-benzo-
quinone (CAS 106-51-4) and Canavanine (CAS 543-38-
4) were purchased from Aldrich (Milwaukee, WI).
2.2. Yeast strains and growth conditions
The S. cerevisiae strain used for the DEL assay is the
diploid strain RSI 12 with the genotype: MATa/MATa,
ura3-52, leu2-3,112/leu2-A98, trp5-27/TRP5, arg4-3/
ARG4, ade2-40/ade2-101, ilvl-92/ILVl, HIS3::pRS6/
his3 A 300, LYS2/lys2-801. This strain was created by
R.H. Schiestl (Schiestl and Prakash, 1988).
The S. cerevisiae strain used for the canavanine resis-
tance assay was wildtype Y433 (provided by P. Hieter) with
the following genotype: MATa ura3-52 leu2-A98 ade 2-101
ilvl-92 his3-A200 Iys2-801.
RSI 12 and Y433 cells were grown on YPAD (1% yeast
extract, 2% peptone, 2% dextrose, and 80.0 (ig/ml adenine)
plates solidified with 2% agar.
RSI 12 cells were grown in suspension using synthetic
complete liquid growth media lacking leucine (SC-leu med-
ium). Y433 cells were grown in YPAD (1% yeast extract,
2% peptone, 2% dextrose, and 80.0 (ig/ml adenine) syn-
thetic liquid medium.
2.3. DEL Assay
The yeast DEL assay was performed as previously
described with minor modifications (Brennan et al.,
1994). In brief, part of an RSI 12 colony was used to inoc-
ulate 5.0 ml of SC-leu liquid medium. After overnight incu-
bation at 30 °C and 275 RPM, the concentration of cells
was determined by hemocytometer-counting, and aliquots
containing 1.0 x 107 cells were adjusted to 5.0ml with
SC-leu medium before exposure to 9,10 PQ or 1,4 BQ for
17 hours at 30 °C. The final concentrations of 9,10 PQ in
the media were: 2.5 (iM, 5 (iM, 10 (iM, 15 (iM and 20 (iM
for aerobic conditions and 20 (iM, 30 (iM, 40 (iM and
50 (iM under anaerobic conditions. The final concentra-
tions for 1,4 BQ under both conditions were 60 (iM and
100 (iM. Quinones were dissolved in acetone which consti-
tuted 0.4% of the final incubation mixture. Incubations
were carried out in 25 ml Erlenmeyer flasks equipped with
gas tight rubber septa. Anaerobic conditions were achieved
by purging the cultures with nitrogen gas, using a syringe
needle through the septa, for one hour at 30 °C with shak-
ing prior to addition of test compound. Aerobic experi-
ments were performed similarly, with the exception that
the flasks were not swept with nitrogen and the septa con-
tained a small opening to introduce air into the headspace.
Following incubation, cells were pelleted (3200 rpm,
10 minutes, 4 °C), washed, and re-suspended in 1 ml of
sterilized distilled water. Cells were then counted by hemo-
cytometer, diluted, and a volume corresponding to 50-100
cells was plated on synthetic complete (SC) medium and
synthetic complete medium lacking histidine (SC-his).
Plates were incubated at 30 °C for 2-3 days before being
analyzed for viable and histidine revertant colonies, respec-
tively. A dose-dependent doubling of the control recombi-
nation frequency and statistically significant increase in
recombination frequency over the control is considered a
Previous
-------
298
C.K Rodriguez et al. I Toxicology in Vitro 22 (2008) 296-300
positive response in the yeast DEL assay (Kirpnick et al., doubling of the untreated frequency was considered a posi-
2005). tive effect.
2.4. Canavanine-resistance assay
3. Results
The canavanine-resistance forward mutation assay was
carried out similarly as the DEL assay and as previously
described with minor modifications (Whelan et al., 1979).
In brief, a volume of 5.0 ml of YPAD liquid medium was
inoculated with a single Y433 colony. After overnight incu-
bation at 30 °C and 275 rpm, cells were counted with a
hemocytometer, and adjusted to 1.0 xlO7 cells/ml in a vol-
ume of 5.0 ml of YPAD liquid medium before incubation
with 9,10 PQ or 1,4 BQ for seven hours under aerobic
and anaerobic conditions, as similarly described for the
DEL assay. The final concentrations in the media were
20 uM 9,10 PQ (aerobic), 60 (iM 9,10 PQ (anaerobic) and
100 (iM 1,4 BQ. The resulting cultures were centrifuged
(3200 rpm, 10 minutes, 4 °C), washed, and re-suspended
in 1.0 ml of sterilized distilled water. Following counting
by hemocytometer, samples were diluted and a volume cor-
responding to 100-200 cells was plated on SC medium
lacking arginine in the presence and absence of canavanine
at a concentration of 219 (iM. Plates were then incubated
at 30 °C for 2-3 days and subsequently analyzed for viable
and canavanine-resistance mutants, respectively. Similarly
to the DEL assay, a statistically significant increase and
3.1. DNA Deletions in diploid RSI 12 cells
The DEL assay measures DNA deletions, a subgroup of
chromosomal rearrangement events. These events most
likely occur in response to DNA double strand breaks
(Galli and Schiestl, 1998) and a wide range of clastogenic
compounds induce DEL recombination (Kirpnick et al.,
2005). Thus, DNA deletion induction is a gauge of a com-
pound's ability to cause DNA double strand breaks. Table
1 depicts the effect of 9,10 PQ and 1,4 BQ on the S. cerevi-
siae diploid strain RSI 12 under aerobic and anaerobic con-
ditions. 9,10 PQ leads to significant induction of DNA
deletions starting at the 5 (iM dose. Cell viability and gen-
eration number begin to significantly decrease at the 15 (iM
exposure concentration. In the absence of oxygen, 9,10 PQ
does not induce DNA deletions or diminish cell viability
even though the exposure concentrations range from
20 uM to 50 uM. At the 40 (iM dose 9,10 PQ induces a sig-
nificant decrease in generation number. The 50 (iM dose
also decreases the generation number but the decrease is
not statistically significant. The control quinone 1,4 BQ
does not lead to an increase in DNA deletions and does
Table 1
DNA deletions, survival and generation number in diploid S. cerevisiae cells under aerobic and anaerobic conditions
9,10 PQ (Aerobic)
Dose (uM)
% Plating efficiency
generations
DEL Events/104 viable cells
0.0
57.9 ±13. 19
4.78 ±0.14
0.93 ±0.39
2.5
49.06 ± 12.66
4.74 ±0.23
1.46 ±0.42
5.0
50.41 ± 13.87
4.60 ± 0.24
1.84 ±0.44*
10.0
52.15 ±6.93
4.58 ± 0.22
2.08 ± 0.52*
Induction of DEL
1.00
1.57
1.97
2 23
15.0
23.42 ± 13.30*
0.96 ±0.25™
7.98 ±1.90™
8.56
20.0
1.79 ± 1.03**
0.77 ± 0.58**
22.40 ± 15.03*
24.03
9,10 PQ (Anaerobic)
Dose (uM)
% Plating efficiency
generations
DEL Events/104 viable cells
Induction of DEL
0.0
56.12 ± 10.17
4.89 ±0.18
1.16 ±0.60
1.00
20.0
60.14 ± 16.6
4.66 ±0.24
0.70 ±0.24
0.61
30.0
56.81 ±8. 12
4.49 ± 0.30
0.98 ± 0.25
40.0
70.14 ±7.28
3.77 ±0.03**
1.54 ±0.17
50.0
40.07 ± 16.1
2.43 ±1.56
1.86 ±0.30
0.84
1.33
1.60
1,4 BQ (Aerobic)
Dose (uM)
% Plating efficiency
generations
DEL Events/104 viable cells
Induction of DEL
0.0
47.64 ± 20.25
5.14 ±0.06
1.34 ±0.58
1.00
60.0
82.83 ± 8.86
4.36 ±0.40*
1.58±0.13
1.18
100.0
69.56 ± 20.29
4.43 ± 0.47
1.97 ±0.18
1.47
1,4 BQ (Anaerobic)
Dose (uM)
% Plating efficiency
generations
DEL Events/104 viable cells
Induction of DEL
0.0
72.66 ±49.38
5.11 ±0.17
1.3 ±0.56
1.00
60.0
122.44 ± 38.44
4.05 ±0.38*
1.11 ±0.03
0.85
100.0
75.25 ±5.28
4.17 ±0.47*
1.46 ±0.19
1.12
(*) P value < 0.05, (**) P value < 0.01. Statistical significance determined by student's t-tssi with comparison to the untreated control (zero dose). Data is
presented as the average ± standard deviation. DEL event frequency = number of deletion events per 104 viable cells.
Previous
-------
C.K Rodriguez et al. I Toxicology in Vitro 22 (2008) 296-300
299
not decrease cell viability up to 100 (iM. This was true for
both aerobic and anaerobic conditions. There was a slight
but statistically significant decrease in generation number
for this quinone under both exposure conditions.
3.2. CAN Mutations in haploid Y433 cells
Mutations that confer canavanine resistance in yeast are
largely (90%) single base-pair alterations such as base sub-
stitutions or frameshift mutations (Tishkoff et al., 1997).
The CAN assay in the current study provides information
about the base damaging activity induced by 9,10 PQ
(Table 2). The point mutation frequency was determined
at a single exposure dose of 20 (iM for 9,10 PQ and
100 uM 1,4 BQ. At this concentration 9,10 PQ induced a
significant increase in point mutations as well as a signifi-
cant decrease in both cell viability and generation number.
In the absence of oxygen, there was no significant increase
in point mutations or decrease in cell viability for 9,10 PQ.
However, there was a significant reduction in generation
number. The control quinone 1,4 BQ led to an average
Table 2
Forward mutations, survival and generation number in haploid S.
cerevisiae cells under aerobic and anaerobic conditions
9,10 PQ (Aerobic)
Dose (uM)
% Plating efficiency
generations
CANR Events/106 viable cells
Induction of mutations
9,10 PQ (Anaerobic)
Dose (uM)
% Plating efficiency
generations
CANR Events/106 viable cells
Induction of Mutations
1,4 BQ (Aerobic)
Dose (uM)
% Plating efficiency
generations
CANR Events/106 viable cells
Induction of mutations
1,4 BQ (Anaerobic)
Dose (uM)
% Plating efficiency
generations
CANR Events/106 viable cells
Induction of mutations
0.0
141.42 ±9.63
4.99 ± 0.39
0.16 ±0.19
1.00
20.0
29.57 ± 4.06*
0.64 ± 0.66™
2.16 ±0.49™
13.50
0.0
173.00 ±94.40
5.26 ± 0.46
0.13 ±0.05
1.00
0.0
141.42 ±9.63
4.99 ± 0.39
0.16 ±0.19
1.00
0.0
173.00 ±94.40
5.26 ± 0.46
0.13 ±0.05
1.00
60.0
214.67 ±30.89
2.63 ±0.51™
0.61 ±0.76
4.69
100.0
19.96 ± 10.04*
0.13 ±0.22**
5.78 ±5.21
36.13a
100.0
19.96 ± 10.04*
0.96 ± 0.66™
0.98 ±0.34*
7.54
(*) P value < 0.05, (**) P value < 0.01. Statistical significance determined
by student's t-test with comparison to the untreated control (zero dose).
Data is presented as the average ± standard deviation. CANR Frequency
= number of CANR events per 106 viable cells.
a Although 1,4 BQ leads to a 36-fold increase in CAN mutations, the
induction of point mutations is not statistically significant because the
actual frequency of CANR events per 106 viable cells is highly variable
(large standard deviation).
36 fold increase in point mutations in the presence of oxy-
gen but the increase was not statistically significant due to
large standard deviation. In the absence of oxygen, 1,4 BQ
led to a statistically significant 7.5 fold increase in point
mutations. Under both aerobic and anaerobic conditions,
1,4 BQ caused a significant decrease in cell viability and
generation number.
4. Discussion
The goal of this study was to examine the DNA damag-
ing activity of 9,10 PQ under aerobic and anaerobic condi-
tions. The DEL assay for DNA deletions and the CAN
assay for point mutations in the yeast S. cerevisiae pro-
vided a measure of genotoxicity. The findings of this study
reveal that 9,10 PQ leads to DNA deletions and point
mutations only in the presence of oxygen. In the absence
of oxygen, 9,10 PQ caused a significant decrease in genera-
tion number in both the diploid and haploid strains but at
approximately three-fold higher concentrations than in the
presence of oxygen. This finding is consistent with previ-
ously observed results where the IC50 for growth inhibition
was 13.81 (iM under aerobic conditions and 36.00 (iM
anaerobically (Rodriguez et al., 2004). The quinone 1,4
BQ does not induce DNA deletions and does not cause a
loss of cell viability in the diploid strain at concentrations
as high as 100 (iM. However, it does cause a decrease in
generation number. In the haploid strain, 1,4 BQ leads to
a decrease in cell viability and generation number in the
presence of oxygen and leads only to a decrease in genera-
tion number in the absence of oxygen. This non-redox
cycling quinone also leads to forward mutations in the
absence of oxygen. The genotoxicity probably occurs
through direct interaction with DNA.
The growth inhibition mediated by 9,10 PQ and 1,4 BQ
in the absence of oxygen is likely a result of GAPDH inhi-
bition. Both quinones have been previously shown to inhi-
bit GAPDH by an oxygen-independent mechanism
(Rodriguez et al., 2005). Inhibition of GAPDH activity
may have an effect on cell cycle progression because active
GAPDH has been shown to stimulate cell proliferation and
reverse cyclin B inhibition by other enzymes (Carujo et al.,
2006). The current study concludes that the activity of 9,10
PQ leads to growth inhibition under both aerobic and
anaerobic conditions but leads to DNA deletions and point
mutations only in the presence of oxygen. Furthermore, the
electrophilic, non-redox cycling 1,4 BQ leads to point
mutations independent of oxygen but does not cause
DNA deletions.
Acknowledgements
This study was supported by a UC Toxic Substances Re-
search and Teaching Lead Campus Program fellowship to
CR and ZS, and an EPA STAR fellowship to ZS.
Previous
-------
300
C.K Rodriguez et al. I Toxicology in Vitro 22 (2008) 296-300
References
Arimoto, T., Yoshikawa, T., Takano, H., Kohno, M., 1999. Generation of
reactive oxygen species and 8-hydroxy-2'-deoxyguanosine formation
from diesel exhaust particle components in L1210 cells. The Japanese
Journal of Pharmacology 80, 49-54.
Bai, Y., Suzuki, A.K., Sagai, M., 2001. The cytotoxic effects of diesel exhaust
particles on human pulmonary artery endothelial cells in vitro: role of
active oxygen species. Free Radical Biology & Medicine 30, 555-562.
Bishop, A.J., Schiestl, R.H., 2003. Role of homologous recombination in
carcinogenesis. Experimental and Molecular Pathology 74, 94-105.
Bolton, J.L., Trush, M.A., Penning, T.M., Dryhurst, G., Monks, T.J.,
2000. Role of quinones in toxicology. Chemical Research in Toxicol-
ogy 13, 135-160.
Brennan, R.J., Swoboda, B.E., Schiestl, R.H., 1994. Oxidative mutagens
induce intrachromosomal recombination in yeast. Mutation Research
308, 159-167.
Carujo, S., Estanyol, J.M., Ejarque, A., Agell, N., Bachs, O., Pujol, M.J.,
2006. Glyceraldehyde 3-phosphate dehydrogenase is a SET-binding
protein and regulates cyclin B-cdkl activity. Oncogene 25, 4033-4042.
Cho, A.K., Di Stefano, E., You, Y., Rodriguez, C.E., Schmitz, D.A.,
Kumagai, Y., Miguel, A.H., Eiguren-Fernandez, A., Kobayashi, T.,
Avol, E., Froines, J.R., 2004. Determination of Four Quinones in
Diesel Exhaust Particles, SRM 1649a, and Atmospheric PM2.5.
Aerosol Science and Technology 38, 68-81.
Galli, A., Schiestl, R.H., 1998. Effects of DNA double-strand and single-
strand breaks on intrachromosomal recombination events in cell-cycle-
arrested yeast cells. Genetics 149, 1235-1250.
Kirpnick, Z., Homiski, M., Rubitski, E., Repnevskaya, M., Hewlett, N.,
Aubrecht, J., Schiestl, R.H., 2005. Yeast DEL assay detects clastogens.
Mutation Research 582, 116-134.
Kondrova, E., Stopka, P., Soucek, P., 2007. Cytochrome P450 destruction
by benzene metabolites 1,4-benzoquinone and 1,4-hydroquinone and
the formation of hydroxyl radicals in minipig liver microsomes.
Toxicology In Vitro 21, 566-575.
Li, N., Alam, J., Venkatesan, M.I., Eiguren-Fernandez, A., Schmitz, D.,
Di Stefano, E., Slaughter, N., Killeen, E., Wang, X., Huang, A.,
Wang, M., Miguel, A.H., Cho, A., Sioutas, C, Nel, A.E., 2004. Nrf2 is
a key transcription factor that regulates antioxidant defense in
macrophages and epithelial cells: protecting against the proinnamma-
tory and oxidizing effects of diesel exhaust chemicals. Journal of
Immunology 173, 3467-3481.
Ma, J.Y., Ma, J.K., 2002. The dual effect of the particulate and organic
components of diesel exhaust particles on the alteration of pulmonary
immune/inflammatory responses and metabolic enzymes. Journal of
Environmental Science and Health, Part C: Environmental Carcino-
genesis and Ecotoxicology Reviews 20, 117-147.
MATES-II, 2000. Multiple Air Toxics Exposure Study in the South Coast
Air Basin (MATES-II). A report of the South Coast Air Quality
Management District. Available: http://www.aqmd.gov/matesiidf/
matestoc.htm.
Peterson, B., Saxon, A., 1996. Global increases in allergic respiratory
disease: the possible role of diesel exhaust particles. Annals of Allergy,
Asthma and Immunology 77, 263-268, quiz 269-270.
Rodriguez, C.E., Shinyashiki, M., Froines, J., Yu, R.C., Fukuto, J.M.,
Cho, A.K., 2004. An examination of quinone toxicity using the
yeast Saccharomyces cerevisiae model system. Toxicology 201,
185-196.
Rodriguez, C.E., Fukuto, J.M., Taguchi, K., Froines, J., Cho, A.K., 2005.
The interactions of 9,10-phenanthrenequinone with glyceraldehyde-3-
phosphate dehydrogenase (GAPDH), a potential site for toxic actions.
Chemico-Biological Interactions 155, 97-110.
Sagai, M., Saito, H., Ichinose, T., Kodama, M., Mori, Y., 1993. Biological
effects of diesel exhaust particles. I. In vitro production of superoxide
and in vivo toxicity in mouse. Free Radical Biology & Medicine 14,
37-47.
Schiestl, R.H., Prakash, S., 1988. RAD1, an excision repair gene of
Saccharomyces cerevisiae, is also involved in recombination. Molec-
ular and Cellular Biology 8, 3619-3626.
Schiestl, R.H., 1989. Nonmutagenic carcinogens induce intrachromoso-
mal recombination in yeast. Nature 337, 285-288.
Schiestl, R.H., Gietz, R.D., Mehta, R.D., Hastings, P.J., 1989. Carcin-
ogens induce intrachromosomal recombination in yeast. Carcinogen-
esis 10, 1445-1455.
Sun, J.D., Wolff, R.K., Kanapilly, G.M., McClellan, R.O., 1984. Lung
retention and metabolic fate of inhaled benzo(a)pyrene associated with
diesel exhaust particles. Toxicology and Applied Pharmacology 73,
48-59.
Tishkoff, D.X., Filosi, N., Gaida, G.M., Kolodner, R.D., 1997. A novel
mutation avoidance mechanism dependent on S. cerevisiae RAD27 is
distinct from DNA mismatch repair. Cell 88, 253-263.
Whelan, W.L., Gocke, E., Manney, T.R., 1979. The CAN1 locus of
Saccharomyces cerevisiae: fine-structure analysis and forward muta-
tion rates. Genetics 91, 35-51.
Previous
-------
Toxicology Letters 181 (2008) 148-156
Contents lists available at ScienceDirect
Toxicology Letters
journal homepage: www.elsevier.com/locate/toxlet
Comparing single and repeated dosimetry data for perfluorooctane
sulfonate in rats1^
Leona A. Harris3'*, Hugh. A. Barton5'1
' Department of Mathematics and Statistics, The College of New Jersey, Swing, NJ 08628, USA
b National Center for Computational Toxicology, Environmental Protection Agency, Research Triangle Park, JVC 27711, USA
ARTICLE INFO
Article history:
Received 27 May 2008
Received in revised form 10 July 2008
Accepted 10 July 2008
Available online 29 July 2008
Keywords:
Perfluorooctane sulfonate
Pharmacokinetic modeling
PBPK
ABSTRACT
Perfluorooctane sulfonate (PFOS) is a member of a class of perfluorinated chemicals used in a variety
of consumer and industrial applications because of their oleophobic and hydrophobic properties. It has
been shown to cause toxicity in adult and developing laboratory animals. Because PFOS has also been
shown to be widely distributed throughout the environment, there have been concerns about its poten-
tial health risk to humans. Limited pharmacokinetic data for PFOS are available in rodents and humans,
while epidemiological studies of workers and extensive toxicity studies in rodents have been performed.
The existing pharmacokinetic and toxicity database in rodents can be useful in the cross-species extrap-
olations needed to evaluate and interpret internal dosimetry in humans. A mathematical model that
describes the disposition of PFOS in adult rats following intravenous, oral, and chronic dietary exposures
was developed to gain a better understanding of the pharmacokinetics of PFOS and to determine whether
single-dose kinetics are predictive of repeated-dose kinetics. In order to characterize existing time-course
data, time-dependent and concentration-dependent changes in the pharmacokinetic parameters for uri-
nary and biliary clearance and liver distribution were needed. Whether these time-dependent changes
represent inconsistencies across experiments, effects of aging in the rats, or chemically induced changes
in pharmacokinetics remains to be determined.
© 2008 Elsevier Ireland Ltd. All rights reserved.
1. Introduction
Perfluorooctane sulfonate (PFOS) and related perfluoroalkyl
acids are used for a variety of consumer and industrial appli-
cations because of their oleophobic and hydrophobic properties.
These chemicals have been used in fabric protection products, the
coatings of paper plates and microwave popcorn bags, firefighting
foams, herbicides, denture cleaners, shampoos, and floor polishes
(Lau et al., 2007; OECD, 2002; Trudel et al, 2008). Manufactur-
ing of PFOS and substances known or suspected to generate PFOS
through degradation pathways was ceased by the largest manu-
facturer (3M) between 2000 and 2002. PFOS has been shown to
be persistent and widely distributed in the environment. It has
been detected in the liver and blood of wildlife across the globe
* This work was reviewed by US EPA and approved for publication but does not
necessarily reflect official Agency policy. Mention of trade names or commercial
products does not constitute endorsement or recommendation by EPA for use.
* Corresponding author.
E-mail addresses: harrisl@tcnj.edu (L.A. Harris), habarton@alum.mit.edu
(HughA Barton).
1 Current address: Pharmacokinetics, Dynamics and Metabolism Department,
Pfizer Global Research & Development, Groton, CT 06340, USA.
0378-4274/S - see front matter © 2008 Elsevier Ireland Ltd. All rights reserved.
doi:10.1016/j.toxlet.2008.07.014
in North America, Europe, Asia, and Antarctica, in blood samples
of 3M fluorochemical-production workers, as well as in blood sam-
ples of non-occupationally exposed humans, at levels which appear
to be declining since the phase-out (Calafat et al., 2007; Giesy and
Kannan, 2001; Olsen et al., 2005, 2008).
Epidemiological studies focused on workers and more recently
on the general population, notably pregnant women and their off-
spring have observed limited effects that may not be reproduced in
other studies (Alexander and Olsen, 2007; Fei et al., 2008). Toxic-
ity studies have evaluated effects in adult and developing animals.
Rats and nonhuman primates have been used in repeated exposure
studies of adults. Liver enlargement has been observed in rats and
monkeys (Seacat et al., 2002, 2003a), while hepatocellular adeno-
mas were observed in rats in a chronic toxicity study (3M Company,
2002). That study also observed evidence of thyroid effects. Limited
malformations arose from developmental exposures, which may
be due to maternal toxicity rather than direct effects of PFOS (Lau
et al., 2004, 2007). However, offspring of mice and rats exposed
during pregnancy demonstrated dose-dependent toxicity, notably
mortality of newborn pups shortly after birth at high doses. Direct
correlations of some effects with blood or liver concentrations of
PFOS have been reported, but limited data exist for describing PFOS
pharmacokinetics (Johnson et al., 1979a,b).
Previous
TOC
-------
LA. Harris, HughA. Barton / Toxicology Letters 181 (2008) 148-156
149
Johnson et al. (1979b) found that PFOS was well absorbed when
a single oral dose of potassium PFOS (mean dose, 4.2mg/kg) was
administered to three male Charles River CD rats. In this study, 95%
of the dose was absorbed within 24 h. In a similar study, Johnson
et al. (1979a) found that PFOS distributes primarily to the liver and
the blood when a single intravenous dose of potassium PFOS (mean
dose, 4.2 mg/kg) was administered to six male Charles River CD rats.
By 89 days, 25% of the dose was found in the liver, 3% of the dose
was found in the plasma, and 42.8% of the dose had been excreted
in the urine and feces. Therefore, since there is no evidence to sug-
gest that PFOS is further metabolized, the half-life for elimination
from the body, tj/2, for male rats appears to be greater than 89
days. Johnson et al. (1984) showed that cholestyramine adminis-
tered in feed to rats for several days after an intravenous dose of
potassium PFOS increases the fecal elimination of PFOS substan-
tially and, hence, PFOS appears to undergo marked enterohepatic
recirculation.
PFOS and related chemicals have become a major focus in envi-
ronmental toxicology (Lau et al., 2007; Trudel et al., 2008). The
persistence and wide distribution of PFOS in the environment has
caused concerns about its potential health risk to humans. Lim-
ited pharmacokinetic data for PFOS in humans have been used to
estimate serum elimination half-lives between 5 and 9 years, con-
siderably higher than the serum elimination half-life of 7.5 days
for adult rats, though the half-life for elimination from the body
was greater than 89 days in rats (Johnson et al., 1979b; OECD,
2002; Olsen et al., 2007). While there is also a limited pharma-
cokinetic database in rodents, extensive toxicity studies in rodents
have been performed. Therefore a mathematical model describing
the pharmacokinetics of PFOS in rats would be useful in performing
cross-species extrapolations needed to evaluate internal dosimetry
in humans.
The objective of this research was to develop a physiologi-
cally based pharmacokinetic (PBPK) model of PFOS that could be
used to gain a better understanding of the pharmacokinetics of
PFOS in rats following a variety of exposure scenarios. The PBPK
model described in this paper was used to evaluate the existing
time course data for PFOS concentrations in the liver, plasma, red
blood cells, urine, and feces following intravenous, oral, and chronic
dietary exposures to determine whether single-dose kinetics are
predictive of repeated-dose kinetics. In this evaluation, inconsis-
tencies among the different exposure scenarios were observed and
led to the use of time-dependent and concentration-dependent
changes in the pharmacokinetic parameters for PFOS to obtain rea-
sonable predictions of the time course data. The time-dependent
and concentration-dependent changes presented in this paper
are hypotheses that require additional experimental investigation
to better understand the biological processes underlying these
changes.
2. Methods
2.1. Model structure
The pharmacokinetics of PFOS in adult male and nonpregnant female rats can
be described using standard PBPK compartmental modeling (Gerlowski and Jain,
1983; Nestorov, 2003). The PBPK model presented in this paper describes the body
as a set of tissue compartments representing the organs and tissues in the body
interconnected by blood flow through the compartments. See Fig. 1 for a graphi-
cal representation of the model. A system of differential equations based on mass
balance is used to track the uptake and absorption of PFOS into the blood stream,
transport of PFOS from one tissue compartment to the next via the plasma, and
elimination of PFOS from the body via the urine and feces. For purposes of this
model, plasma and serum are assumed interchangeable. The system variables, rep-
resenting the amounts of PFOS in the various tissue compartments, are defined in
Table 1 and the model parameters are defined in Tables 2 and 3. Each tissue com-
partment in the model is described as a diffusion-limited compartment (including
sub-compartments for the capillary blood in the tissue and the tissue cells) cons is-
Upper
0,.
/
urine
a,
QL
I
*.l
Lo\
P*A^*P:A Plasma
— H —
Red Blood Cells
Rest of Body Plasma
— M—
Rest of Body Tissue
Liver Plasma
— H—
Liver Tissue
u
ver 1
fact ( — > feces
oral •
Fig. 1. Conceptual representation of a physiologically based pharmacokinetic model
for PFOS exposure in rats. The boxes represent tissue compartments and the arrows
represent plasma flow to/from the tissue compartments and diffusional transfer
across the cell membrane.
tent with an assumption that the rate of diffusion across the cell membrane is slow
compared to the blood flow rate into the tissue (Gerlowski and Jain, 1983). Since
PFOS has been shown to distribute primarily to the liver and blood (Johnson et al.,
1979a), compartments representing the blood, liver, and rest of the body have been
included in the model. In addition, a two-compartment gastrointestinal tract has
been included as a route of entry.
Table 1
Model variables and concentrations
Variable Description
Calculation
ApiaSF Amount of PFOS in the plasma (mg)
Ap:A Amount of PFOS-bound albumin in
plasma (mol)
ARBC Amount of PFOS in the red blood
cells (mg)
AU Amount of PFOS in the urine (mg)
Ag Amount of PFOS in the rest of body
tissue (mg)
ARP Amount of PFOS in the rest of body
plasma (mg)
AUGI Amount of PFOS in the upper
gastrointestinal tract (mg)
ALGI Amount of PFOS in the lower
gastrointestinal tract (mg)
AF Amount of PFOS in the feces (mg)
AL Amount of PFOS in the liver tissue
(mg)
ALP Amount of PFOS in the liver plasma
(mg)
CPlas Total concentration of PFOS in
plasma (mg/L)
CpiaSF Free concentration of PFOS in
plasma (mg/L)
CAP Free concentration of albumin in
plasma (M)
CP:A Concentration of PFOS-bound
albumin in plasma (M)
CRBC Concentration of PFOS in red blood
cells (mg/L)
CR Concentration of PFOS in rest of
body tissue (mg/L)
CRP Concentration of PFOS in rest of
body plasma (mg/L)
CL Concentration of PFOS in liver
tissue (mg/L)
CLP Concentration of PFOS in liver
plasma (mg/L)
State variable
State variable
State variable
State variable
State variable
State variable
State variable
State variable
State variable
State variable
State variable
CplasF = Cplas - CP:A MWp 1000
CAP = CA - Cp:A
CR=/
CRP =ARp/\
CL=AL/VL
Previous
TOC
-------
150
LA. Harris, HughA. Barton / Toxicology Letters 181 (2008) 148-156
Table 2
Physiological parameters defined for the PFOS PBPK model
Parameter
Description
Value
Source
BW
/Plas
VL
VIP
VLT
WoB
WoBP
WoBT
Qc
Qplas
OK
QL
OR
Rat body weight
Volume of plasma
Volume of red blood cells
Volume of blood
Fraction of plasma in blood
Volume of liver
Volume of liver plasma
Volume of liver tissue
Volume of rest of body
Volume of rest of body plasma
Volume of rest of body tissue
Blood flow rate by the heart (cardiac output)
Plasma flow rate by the heart
Plasma flow rate to the kidneys
Plasma flow rate to the liver
Plasma flow rate to the rest of body
0.288 kg
0.0090 L
0.0076 L
0.0166 L
0.5434
0.0105 L
0.0012 L
0.0093 L
0.2090 L
0.0045 L
0.2045 L
5.5432 L/h
3.0122L/h
0.4247 L/h
0.5512 L/h
2.5875 L/h
Johnson etal. (1979a)
Johnson etal. (1979a)
Johnson etal. (1979a)
Calculated as VPiis + VRBc
Calculated as Vp,as/(Vpias + VRBC)
Brown et al. (1997)
Brown et al. (1997)
Calculated as VL - VLP
Calculated as 0.82(BW) - VBld - VL
Brown etal. (1997)
Calculated as VRoB - VROBP
Brown et al. (1997)
Calculated as/piasQ£
Brown et al. (1997)
Brown etal. (1997)
Calculated as Qpias - QL
2.2. Blood compartment
The blood has sub-compartments for the plasma and red blood cells. The equa-
tions representing the blood compartment describe the transport of PFOS through
the systemic plasma to/from the tissue compartments, the binding of PFOS to the
plasma protein albumin, the diffusion of PFOS into the red blood cells, and the elim-
ination of PFOS into the urine. The differential equation for the amount of PFOS in
the plasma is given by
at
- = QL(CLP - Cp,aS
- Cplas) + PARBC ( ^£ _ Cp,asF ) - ClurCp,as,
J RBC
The first two terms describe the transport of PFOS through the plasma to/from the
liver and the rest of the body, respectively; the third term represents the diffusion
of PFOS across the red blood cell membrane; and the fourth term describes uri-
nary clearance as a first-order process removing PFOS from the arterial plasma.
This simplified description of kidney filtration is commonly used in PBPK mod-
eling when the kidney is lumped together with the other tissues in the rest of
body compartment (Krishnan and Andersen, 1994; Nestorov, 2003). In the model
the urinary clearance rate is described as a fraction of the kidney blood flow rate
(Clur=/urQk).
It has been shown that PFOS binds to the albumin in the blood (Jones et al., 2003).
This affects the availability of PFOS to distribute to the other tissue compartments
(Mendel, 1992). Here we assume that the total plasma concentrations of PFOS, CPias,
are available to distribute to the liver and the rest of body compartments, while
only the free concentrations of PFOS in the plasma (CplasF = Cplas-Cp:AMWp 1000)
are available to distribute to the red blood cells. The differential equation for the
amount of PFOS bound to albumin in the plasma is given by
,, . CplasF r ,/ j. r
-
where CAF = CA - CP:A is the free concentration of albumin available to bind to PFOS
in the plasma.
The differential equation for the amount of PFOS in the red blood cells is given
by
d^RBC D. \r CRBC \
— - 7— = PARBC CPlasF - - -
dt \_ PRBC )
where PARBc is the permeability cross-product for the red blood cells, PRBc is the
RBC-to-plasma partition coefficient, and CRBC/PRBC is the concentration of PFOS free
to leave the red blood cells.
23. Liver compartment
The diffusion-limited compartment representing the liver has sub-
compartments for the liver plasma and the liver tissue. The differential equation
for the amount of PFOS in the liver plasma is given by
s - CLP) + PALiv ± -
The first two terms describe the transport of PFOS to/from the systemic plasma
and diffusional transfer between the plasma and tissue, while the last two terms
describe the distribution of PFOS to the liver from the upper and lower sections of
Table 3
Biochemical and kinetic parameters defined for the PFOS PBPK model
Parameter
kau
kal
fed
kb
a
/ur
kf
PARBC
PARoB
PALiv
Kd
Kon
koff
MWA
MWp
MWP:A
N
CA
BMax
PL
PRBC
PROB
Description
Rate of absorption from the upper GI tract
Rate of absorption from the lower GI tract
Rate of transfer from upper-lower GI tract
Maximum rate of biliary elimination
Biliary elimination rate of decrease
Fraction of urinary clearance
Rate of fecal elimination
Red blood cells permeability cross-product
Rest of body permeability cross-product
Liver permeability cross-product
Equilibrium disassociation constant
Association rate for PFOS-albumin
Disassociation rate for PFOS-albumin
Molecular weight of albumin
Molecular weight of PFOS-
Molecular weight of P:A complex
Number of binding sites
Total concentration of albumin in plasma
Maximum binding capacity
Liver-to-plasma partition coefficient
RBC-to-plasma partition coefficient
Rest of body-to-plasma partition coefficient
Value
0.114 h-1
1.0 h-1
0.01 h-1
8.0 h-1
5.0 h-1
0.28
5.0 h-1
0.002(BW)°-75 L/h
O.OOl(BW)0-75 L/h
0.00025(BW)OJ5L/h
10-7M
IQSM-'s-1
10-2 s-1
66,000g/mol
499.12g/mol
66,499.1 2 g/mol
1
0.00041 M
0.00041 M
8.66
801.3
0.47
Source
Estimated
Estimated
Estimated
Estimated
Estimated
Estimated
Estimated
Estimated
Estimated
Estimated
Estimated
Estimated
Calculated as /Cdkon
Seshagiri and Adiga (1989)
3M (2001), PubChem (2008)
Calculated as MWA + MWP
Assumption
Teeguarden and Barton (2004)
Calculated as N x CA
Estimated
Estimated
Estimated
Previous
TOC
-------
LA. Harris, HughA. Barton / Toxicology Letters 181 (2008) 148-156
151
the gastrointestinal tract due to oral absorption and enterohepatic recirculation. The
differential equation for the amount of PFOS in the liver tissue is given by
In the first term, PALiv represents the liver permeability cross-product, PL represents
the liver-to-plasma partition coefficient, and Q./PL represents the concentration of
PFOS free to leave the liver tissue. The second term represents the biliary elimina-
tion of PFOS from the hepatocytes, the cells of the liver tissue that produce bile,
to the lower gastrointestinal tract. A time-dependent biliary elimination rate that
decreases with time, kb(t) = k^l(at + 1 ), was used to capture the elimination dynamics
of PFOS in feces. The parameter a is a measure of the rate of decrease (see Table 3).
2.4. Rest of body compartment
The rest of body compartment represents the rest of the organs and tissues in
the body not included in the blood and liver compartments. It is a diffusion-limited
compartment that has sub-compartments representing the plasma and tissue cells
in the remaining organs and tissues. The differential equations for the amounts of
PFOS in the rest of body plasma and the rest of body tissue cells are given by
,,S - CRp) + PARoB - CRp
As in the blood and the liver compartments, these equations describe the transport of
PFOS to/from the systemic plasma and diffusion transfer across the cell membrane.
2.5. Absorption from the gastrointestinal (GI) tract
A two-compartment model is used to describe the uptake of PFOS into gastroin-
testinal tract following oral administration, absorption of PFOS into the liver via the
portal venous system, and elimination of PFOS in the feces. The differential equation
for the amount of PFOS in the upper gastrointestinal tract is given by
The first term represents the transfer of PFOS to the liver due to absorption via the
portal venous system and the second term represents the transfer of PFOS to the
lower gastrointestinal tract. The differential equation for the amount of PFOS in the
lower gastrointestinal tract is given by
(Mia
dt
+ kb(t)AL - k,,ALGl -
The first two terms represent input from the upper gastrointestinal tract and from
the liver tissue via the bile; the third term describes the transfer of PFOS to the liver;
and the fourth term describes fecal elimination.
2.6. Routes of elimination
The model describes two primary routes for the elimination of PFOS from the
body: urine and feces. The differential equations for the amounts of PFOS in the
urine and feces are given by
dAu
~ = clurCplasF
daily body weights and food consumption levels needed to estimate daily oral doses
for each of the dose groups. Although rats eat episodically during the 12-h dark cycle
(Yuan, 1993), the relatively slow clearance of PFOS makes it reasonable to treat the
dietary exposure as a daily bolus oral dose. The PBPK model treats a single, oral dose
as an initial amount in the upper gastrointestinal tract, so an iterative process of
starting and stopping the model each day, adding the new daily dose to the current
amount in the uppergastrointestinal tract was used to simulate these chronic dietary
exposures.
2.8. Model implementation and parameterization
The model was implemented and parameterized using the software MATLAB
(The MathWorks, Inc., Natick, MA). Organ-specific and species-specific parameters
(e.g. cardiac output, plasma flow rates, and organ volumes) obtained from the lit-
erature are given in Table 2. An intravenous data set (Johnson et al., 1979a) was
used to calculate the tissue-to-plasma partition coefficients, PRBc, PL, and PR. This
study reported that at 89 days following a single intravenous dose of 4.2 mg/kg of
potassium PFOS, 42.8% of the dose had been excreted in urine and feces and 25.21%,
2.81%, and 0.47% of the dose remained in the liver, plasma, and red blood cells,
respectively. Assuming that the distributional pseudo-equilibrium between the tis-
sues and plasma was attained at 89 days (Lam et al., 1982; Khor and Mayersohn,
1991), this data was used to estimate each partition coefficient as the ratio of tissue
concentration to plasma concentration at 89 days (see Table 3). In this model, the
total plasma concentration of PFOS at 89 days estimated by Johnson et al. (1979a)
was used to estimate PL and PR; however, it is assumed that only the free plasma con-
centration of PFOS is available to distribute to the red blood cells. Therefore PRBc was
estimated as the ratio of the red blood cell concentration of PFOS to the free plasma
concentration of PFOS at 89 days. Here the free plasma concentration of PFOS at 89
days is calculated by solving the binding equation T=(BMaxF)/(/(d+F) + F for Fusing
the total plasma concentration at 89 days, T, as estimated by Johnson et al. (1979a).
The value of PUBC estimated here may be an upper bound in light of recent measure-
ments with human blood that did not detect PFOS in red blood cells (Ehresman et
al., 2007).
The remaining chemical-specific parameters were estimated by fitting the
model to experimental data using a two-step process. In the first step, a sub-model
consisting of all of the model compartments except the uppergastrointestinal tract
was fit to the data in the intravenous study (Johnson et al., 1979a). The upper
gastrointestinal tract was excluded from the model in this step because this com-
partment is only needed to simulate the pharmacokinetics of PFOS following oral
uptake. Therefore the parameters associated with this compartment, kiu and ktt, do
not contribution to the distribution of PFOS following intravenous administration.
In the second step of the parameter estimation process, the parameters obtained
from the first step were fixed and used to estimate k,a and kti by fitting the entire
model to an oral gavage data set (Johnson et al., 1979b). Refer to Table 3 for a list
of the resulting parameter values obtained from fitting the model to experimental
data.
3. Results
The PBPK model described in this paper was used to predict
the pharmacokinetics of PFOS in rats following intravenous, oral,
and chronic dietary exposures. Model simulations tracking the time
courses of PFOS concentrations in the liver, plasma, red blood cells,
urine, and feces are compared to experimental data to determine
whether single-dose PFOS kinetics are predictive of repeated-dose
kinetics.
dAr
~dT
= kfALG
2.7. Dietary modeling
The model was used to simulate a 104-week dietary study reported by the 3M
Company (2001, 2002). In the study, male and female Sprague-Dawley rats were
assigned to six dose groups (40-70 rats/sex/dose level) and were fed diets containing
potassium PFOS concentrations of 0 ppm, 0.5 ppm, 2 ppm, 5 ppm, or 20 ppm. This
paper will not address Dose Group 1 (the control group given 0 ppm) and Dose
Group 6 (a recovery group given 20 ppm). The intent of the 104-week study was
to analyze the sub-chronic and chronic effects of PFOS exposure in rats. To study
the sub-chronic effects of PFOS exposure at weeks 4 and 14, a small group of rats
(5 rats/sex/dose level/sacrifice) were added to each dose group. These results were
reported by Seacat et al. (2003a,b). Over the 104-week period of the study, body
weight changes, food consumption levels, and mean daily intake levels (mg/kg/day))
were measured consistently. These measurements were obtained wee klyforthe first
16 weeks and then monthly for the remainder of the study. In order to use the model
to simulate these daily dietary exposures, linear interpolation was used to obtain
3.1. Single-dose kinetics
Two single-dose pharmacokinetic studies (Johnson et al.,
1979a,b) were used to estimate model parameters and determine
whetherthe model is capable of accurately predicting PFOS kinetics
following acute exposures. Results from simulating a single intra-
venous dose of 4.2 mg/kg of potassium PFOS in six male rats are
shown in Fig. 2. The graphs compare model simulations to the data
to illustrate the ability of the model to replicate the distribution of
PFOS to the liver, plasma, red blood cells, urine, and feces for 90 days
following exposure. The results from the intravenous data set sug-
gest that the elimination of PFOS in feces decreases overtime while
liver amounts remain high. In order to simulate this behavior, the
biliary elimination rate was treated as a decreasing function of time.
In the model, a mean body weight of 288 g, the initial body weight of
the rats in the Johnson et al. (1979a) study, was used as an estimate
Previous
TOC
-------
152
LA. Harris, HughA. Barton / Toxicology Letters 181 (2008) 148-156
30 40 50 60
Time (days)
Plasma
Liver
Fig. 2. Intravenous exposure in male rats. Model simulations are compared to experimental data for a single intravenous dose of 4.2 mg/kg of potassium PFOS in male rats
(Johnson et al., 1979a). (a) Model simulations (solid and dashed curves) track the cumulative percent of dose in urine and feces overtime, (b) Model simulations predict the
percent of dose in tissues on day 89.
of the body weight for the 90-day period following exposure. It is
possible that the body weights may have increased to more than
500 g during this time (Charles River Laboratories, 2008); however,
body weight changes were not reported in thejohnson et al. (1979a)
study. Using body weight changes for an equivalent 90-day period
as reported in a 2-year study (3M Company, 2002) and using the
parameters listed in Tables 2-3, the model is able to reproduce the
results of the IV study shown in Fig. 2. Since similar results were
obtained when using a fixed body weight and increasing the body
weight over time, while the extent of the body weight changes in
thejohnson et al. (1979a) study are unknown, the parameter values
using a fixed body weight are reported here. The model was also
used to simulate a single oral dose of 4.2 mg/kg of potassium PFOS
in three male rats. These results are shown in Fig. 3. The graphs
show that while the model is able to characterize the elimination
of PFOS in urine over time, it was difficult to fit the model to the
plasma and feces time courses using the limited number of data
values.
D
O
. . , ...
Data: Urine
— Model: Feces
Data: Feces
Time (hrs)
0
_
Data: Plasma
Fig. 3. Oral administration in male rats. Model simulations (solid and dashed curves) are compared to experimental data for a single oral gavage dose of 4.2 mg/kg of potassium
PFOS in male rats (Johnson et al., 1979b).
Previous
TOC
-------
LA. Harris, HughA. Barton / Toxicology Letters 181 (2008) 148-156
153
Table 4
Model parameter values used to
changes in urinary elimination and
sures in male rats
reflect time-dependent and dose-dependent
liver partitioning during chronic dietary expo-
Parameter
Value
/ur
b (high-dose)
b (mid-dose and mid-high dose)
b (low-dose)
r (all doses) (h)
c (all doses)
0.01
0.3
0.6
0.9
3360
7
3.2. Repeated-dose kinetics
To capture the biological changes observed in the plasma and
liver PFOS concentration data, the urinary elimination rate con-
stant, /ur, and the liver-to-plasma partition coefficient, PL, were
treated as functions of time. To account for the decreasing plasma
levels, the following Hill function is used to describe the urinary
elimination rate constant as an increasing function of time:
/ur(t)=/ur
bt4
•4 if4
where/ur(t) has an initial value of/ur and increases to a maximum
value of/ur + b, and r is a measure of the rate of increase. Similarly,
to account for decreasing liver concentrations, the liver-to-plasma
partition coefficient is described as a decreasing function of time
in the following way:
Using the parameters in Table 3 that were estimated from the
intravenous and oral data sets and a smaller fraction of urinary
clearance, /ur (see Table 4), the model was used to predict PFOS
liver and plasma levels in male rats for 104 weeks of daily dietary
exposures. Results from simulating Dose Groups 2-5 of the study
are shown in Fig. 4. The graphs suggest that the model predicts
liver and plasma concentrations reasonable well in earlier weeks.
However, decreasing plasma and liver concentrations in the data
between weeks 14 and 53 suggest that there are unknown biologi-
cal changes that cause increased elimination of PFOS from the body
and changes in the liver partitioning (see Fig. 4d). These biological
changes make it impossible for the model to predict PFOS kinetics
in later weeks using the same set of parameter values.
The parameter r in the equations for/ur(t) and PL(£) was cho-
sen in such a way that the changes in the urinary elimination and
liver partitioning would occur between weeks 14 and 53. Refer to
Tables 4-5 for a list of the parameter values used to reflect these
changes during chronic dietary exposures in male and female rats.
Results from simulating these time-dependent changes in males
are shown in Figs. 5-7. These figures illustrate that the changes
needed to accurately predict PFOS plasma levels were dependent
on the dose level. The high-dose group (20 ppm) required less of an
increase in/ur than the low (0.5 ppm), mid- (2 ppm), and mid-high
(5 ppm) dose groups. For the high-dose group (see Fig. 5), the frac-
104 Week Dietary Study - Male
100 120
20 40 60 80
Weeks
100 120
if
\ 3; 20
104 Week Dietary Study - Male
20 40
100 120
00 20 40
100 120
Weeks
(C)
104 Week Dietary Study - Male
40
80
100 120
60
Weeks
2000
13 1500
r
iiooo
soo
0
104 Week Dietary Study - Male
60
Weeks
Weeks
Fig. 4. Chronic dietary exposures in male rats. Model simulations (solid curves), without time-dependent changes in the urinary elimination and liver partitioning, are
compared to data from a 104-week dietary study reported by 3M Company (2002) in which male rats were fed diets containing (a) 0.5 ppm, (b) 2 ppm, (c) 5 ppm, and (d)
20 ppm of potassium PFOS.
Previous
TOC
-------
154
LA. Harris, HughA. Barton / Toxicology Letters 181 (2008) 148-156
104 Week Dietary Study - Male
104 Week Dietary Study - Male
I
|_150
"E "^
| 50
o
I o
40 60 80
Weeks
100 120
CO
SS 4
Q_
.S 3
I?
e o* 2
§ 1
c
o
0 Q
*
xx"^ ^"1""\
y' ^"-^^
/ ^^—- — _ „___ ~__— - -^
/
__________U i____J L__________X_^,
"
_
0 20 40 60 80 100 1;
40 60 80
Weeks
100 120
Fig. 5. Time-dependent changes in urinary elimination and liver partitioning during
chronic (high-dose) dietary exposures in male rats. Model simulations (solid curves)
are compared to data for the high-dose group (20 ppm).
Table 5
Model parameter values used to reflect time-dependent changes in urinary elimi-
nation and liver partitioning during chronic dietary exposures in female rats
Parameter
Value
b (low-dose, mid-dose, mid-high-dose)
r (all doses) (h)
c (all doses)
0.01
0.3
3360
7
tion of urinary clearance increases over time to a maximum value
of/ur + b = 0.31. Fig. 6 show that the mid and mid-high dose levels
required a similar level of increase in /ur. For these dose groups,
the fraction of urinary clearance increases over time to a maxi-
mum value of/ur + b = 0.61. In order to predict the plasma levels in
the low-dose group, an even larger fraction of urinary clearance is
required (see Fig. 7). In the low dose group, the fraction of urinary
elimination increases overtime to a maximum value of/ur + b = 0.91.
Time-dependent changes in the urinary elimination and liver
partitioning were also required to simulate chronic dietary expo-
sures in female rats for the low-, mid-, and mid-high dose groups.
However, the changes in/ur were not dose-dependent as in the male
rats. For each of these dose groups, the fraction of urinary clearance
was increased over time to a maximum value of/ur+ b = 0.31. For
the high-dose group, plasma levels reached pseudo-steady state (as
40 60 80 100 120
Weeks
Fig. 7. Time-dependent changes in urinary elimination and liver partitioning during
chronic (low-dose) dietary exposures in male rats. Model simulations (solid curves)
are compared to data for the low-dose group (0.5 ppm).
in the data) and, hence, only time-dependent changes in the liver
partitioning were required. Results from simulating these time-
dependent changes for all of the dose groups are shown in Fig. 8.
The results in Figs. 4-8 suggest that the urinary elimination rate
may be a function of the age of the animals, i.e. as the animals age
the urinary elimination rate increases. This provides some support
for the need to decrease the original estimate of/ur from 0.28 to 0.01.
The original estimate of/ur was calculated using an intravenous
study in which male rats were 8 weeks old at the onset of exposure,
approximately the age of the animals at the start of the dietary
study, so it might have been expected that the same value of /ur
could be used.
4. Discussion
The pharmacokinetics and resulting tissue dosimetry for per-
fluoroalkyl acids has been under increasing investigation due to
the availability of measured blood concentrations in human pop-
ulations (Calafat et al., 2007; Olsen et al., 2005, 2008) and the
substantial differences in half lives reported for different species
(summarized in Lau et al., 2007 and Trudel et al., 2008). In addition,
plasma concentrations of perfluoroalkyl acids have been measured
fairly frequently in the animal toxicity studies facilitating toxi-
E 15
104 Week Dietary Study - Male
60 SO
Weeks
O).
I
a. 40
ii.
| 400
c —J
•S -i 200
2 o)
c 3
o 100
104 Week Dietary Study - Male
Weeks
I
Weeks
Weeks
Fig. 6. Time-dependent changes in urinary elimination and liver partitioning during chronic (mid-dose and mid-high dose) dietary exposures in male rats. Model simulations
(solid curves) are compared to data for the (a) mid-dose (2 ppm), and (b) mid-high dose group (5 ppm).
Previous
-------
LA. Harris, HughA. Barton / Toxicology Letters 181 (2008) 148-156
155
(a).
IE
2 2>
104 Week Dietary Study - Female
40 60 80
Weeks
100 120
40 60 80 100 120
Weeks
(b) (B 40
p O!
E 310
5 0
O 0
•S 100
11
104 Week Dietary Study - Female
60
Weeks
60
Weeks
(C)5
400
300
200
>
' 100
0
104 Week Dietary Study - Female
60
Weeks
-j-r-
60
Weeks
(d)«
SE
Si
2 3 500
'
104 Week Dietary Study - Female
300
200
100
0
C
' ^ j
'M
20 40 60 80 100 12
Weeks
60
Weeks
Fig. 8. Time-dependent changes in urinary elimination and liver partitioning during chronic dietary exposures of female rats. Model simulations (solid curves) are compared
to data from the 104-week dietary study in which female rats were fed diets containing (a) 0.5 ppm, (b) 2ppm, (c)5ppm, and(d) 20ppm of potassium PFOS. The high-dose
group only required changes in the liver partitioning.
cological evaluations based upon plasma concentrations, rather
than the typical analyses for environmental chemicals based upon
exposure doses. Measurements of PFOS concentrations in a 2-year
dietary study were a major focus of the analysis presented here (3M
Company, 2001, 2002). These analyses offer the potential to better
interpret human biomonitoring data if the pharmacokinetics are
reasonably well understood.
In contrast to perfluorooctanoate acid (PFOA), for which fairly
robust pharmacokinetic data sets exist in several species (Kudo and
Kawashima, 2003), PFOS has been the subject of fewer investiga-
tions. Data for PFOA and PFOS in monkeys were previously analyzed
using a model proposing saturable resorption in the kidneys result-
ing in greater clearance at high plasma concentrations and slower
clearance at lower concentrations resulting in a longer appar-
ent half life (Andersen et al., 2006). However, pharmacokinetic
analyses for PFOA in rats have not demonstrated concentration-
dependent clearance (Wambaugh et al., submitted for publication).
Concentration-dependent clearance can result in apparent discrep-
ancies between single-dose pharmacokinetic studies, such as blood
time course data following intravenous or oral dosing, and repeated
dosing studies, particularly when the single-dose studies do not
involve an adequate range of concentrations. Such discrepancies
have led to proposals for altering pharmacokinetic parameters such
as the volume of distribution (Washburn et al., 2005).
The modeling reported here indicates discrepancies among dif-
ferent studies as well as an apparent need for time-dependent
changes in pharmacokinetic parameters for PFOS over the course of
a 2-year study in rats. Based upon the measured data in the study,
changes in body weight and food consumption levels over time
were modeled. To account for decreasing plasma and liver levels
in the data, time-dependent changes in urinary elimination and
liver partitioning were made. Other possible changes that were not
modeled in this effort include decreasing absorption over time and
increasing fecal elimination over time. However, the latter would
involve increasing biliary elimination which essentially has been
shut down due to the decreased fecal elimination in the single-
dose intravenous study. Whether the necessary time-dependent
changes are a consequence of the PFOS exposure itself, the aging
of the animals, or some combination of the two is unclear. Dur-
ing the writing of this manuscript, Tan et al. (2008) published a
related model for the disposition of perfluoroalkylacids in rats and
monkeys. The authors similarly determined that time-dependent
changes were needed to accurately characterize the data. They had
similar difficulties fitting the oral PFOS data set and chose to rep-
resent the data from the intravenous study and the oral study as
a single oral dose; their analyses did not include the dietary study
data analyzed here.
Using only plasma concentrations it is not possible to completely
distinguish between changes in tissue distribution and clearance,
so the time-dependent changes described here are hypotheses
requiring additional experimental investigation. Measurements of
daily PFOS urinary and fecal excretion, as well as plasma and tis-
sue concentrations during a repeated-dose study, would be useful
in testing these hypotheses. While the major gender differences
observed for PFOA are not observed with PFOS, there were mod-
est differences in the pharmacokinetic parameters required for
Previous
TOC
-------
156
LA. Harris, HughA. Barton / Toxicology Letters 181 (2008) 148-156
describing the two sexes in the chronic study. No concentration-
dependent effects were observed for the females, while time and
concentration-dependent changes in clearance were utilized for
describing the chronic male rat plasma data. These results are
inconsistent with the saturable resorption hypothesis in that the
highest clearance is predicted for the lowest concentration, rather
than the highest. Some additional data on PFOS in blood, urine,
and feces has recently been presented (Chang S, Hart JA, Ehres-
man DJ, Butenhoff JL, personal communication) that may provide
additional perspectives on the analyses presented here and by Tan
et al. (2008). Developing a clearer understanding of the pharma-
cokinetics of PFOS will be beneficial for better interpreting the
relationship between blood concentrations measured in the animal
toxicity studies and those measured in humans.
Conflict of interest statement
There are no conflicts of interest.
Acknowledgement
The authors would like to thank Jennifer Seed for useful com-
ments on this article.
References
3M Company, 2001.104-Week dietary chronic toxicity study with perfluorooctane
sulfonic acid potassium salt (PFOS T-6295) in rats. 3M Study Report, Covance
Study No.: 6329-183. Analytical laboratory report on determination of the pres-
ence and concentration of perfluorooctanesulfonate (PFOS) in the serum of
Sprague-Dawley rats exposed to potassium perfluorooctanesulfonate via gav-
age. 3M Report No. Fact Tbx-002, LRN-U2121.
3M Company, 2002. 104-Week dietary chronic toxicity and carcinogenicity study
with perfluorooctane sulfonic acid potassium salt (PFOS; T-6295) in rats. Final
Report. 3M Company, St. Paul, MN. January 2, 2002. U.S. EPA Docket AR-226-
1051 a.
Alexander, B.H., Olsen, G.W., 2007. Bladder cancer in perfluorooctanesulfonyl fluo-
ride manufacturing workers. Ann. Epidemiol. 17,471-478.
Andersen, M.E., Clewell III, H.J., Tan, Y.M., Butenhoff, J.L., Olsen, G.W., 2006. Phar-
macokinetic modeling of saturable, renal resorption of perfluoroalkylacids in
monkeys—probing the determinants of long plasma half-lives. Toxicology 227,
156-164.
Brown, R.P., Delp, M.D., Lindstedt, S.L., Rhomberg, L.R., Beliles, R.P., 1997. Physiological
parameter values for physiologically based pharmacokinetic models. Toxicol.
Ind. Health 13, 407-484.
Calafat, A.M., Wong, L.Y., Kuklenyik, Z., Reidy, J.A., Needham, L.L, 2007. Polyfluo-
roalkyl chemicals in the U.S. population: data from the National Health and
Nutrition Examination Survey (NHANES) 2003-2004 and comparisons with
NHANES 1999-2000. Environ. Health Perspect. 115,1596-1602.
Charles River Laboratories, 2008. Weight Chart for the Crl:CD(SD) Rat. http://
www.criver.com/research_models_and_services/research_models/CD_IGS.html.
Ehresman, D.J., Froehlich, J.W., Olsen, G.W., Chang, S.C., Butenhoff, J.L., 2007. Compar-
ison of human whole blood, plasma, and serum matrices for the determination
of perfluorooctanesulfonate (PFOS), perfluorooctanoate (PFOA), and other fluo-
rochemicals. Environ. Res. 103,176-184.
Fei, C, Mclaughlin, J.K., Tarone, R.E., Olsen, J., 2008. Fetal growth indicators and
perfluorinated chemicals: a study in the Danish National Birth Cohort. Am. J.
Epidemiol. 168, 66-72.
Gerlowski, L.E., Jain, R.K., 1983. Physiologically based pharmacokinetic modeling:
principles and applications. J. Pharm. Sci. 72,1103-1127.
Giesy, J.P., Kannan, It, 2001. Global distribution of perfluorooctane sulfonate in
wildlife. Environ. Sci. Technol. 35,1339-1342.
Johnson, J.D., Gibson, S.J., Ober, R.F., 1979a. Extent and route of excretion and tis-
sue distribution of total carbon-14 in rats after a single intravenous dose of
FC-95-14C. Project No. 8900310200, Riker Laboratories, Inc., St. Paul, MN. U.S.
EPA Docket AR-226-0006.
Johnson, J.D., Gibson, S.J., Ober, R.F., 1979b. Absorption of FC-95-14C in rats after a
single oral dose. Project No. 8900310200, Riker Laboratories, Inc., St. Paul, MN.
U.S. EPA Docket AR-226-0007.
Johnson, J.D., Gibson, S.J., Ober, R.F., 1984. Cholestyramine-enhanced fecal elim-
ination of carbon-14 in rats after the administration of ammonium! 14C]
perfluorooctanoate or potassium [14C] perfluorooctanesulfonate. Fund. Appl.
Toxicol. 4, 972-976.
Jones, P.O., Hu, W., De Coen, W., Newsted, J.L., Giesy, J.P., 2003. Binding of perfluori-
nated fatty acids to serum proteins. Environ. Toxicol. Chem. 22, 2639-2649.
Khor, S.P., Mayersohn, M., 1991. Potential error in the measurement of tissue to blood
distribution coefficients in physiological pharmacokinetic modeling: residual
tissue blood. I. Theoretical considerations. Drug Metab. Dispos. 19,478-485.
Krishnan, It, Andersen, M.E., 1994. Physiologically based pharmacokinetic modeling
in toxicology. In: Hayes, A.W. (Ed.), Principles and Methods in Toxicology, 3rd ed.
Raven Press, Ltd., New York, pp. 149-188.
Kudo, N., Kawashima, Y, 2003. Toxicity and toxicokinetics of perfluorooctanoic acid
in humans and animals. J. Toxicol. Sci. 28, 49-57.
Lam, G., Chen, M., Chiou, W.L., 1982. Determination of tissue to blood partition
coefficients in physiologically-based pharmacokinetic studies. J. Pharm. Sci. 71,
454-456.
Lau, C, Butenhoff, J.L., Rogers, J.M., 2004. The developmental toxicity of perfluo-
roalkyl acids and their derivatives. Toxicol. Appl. Pharmacol. 198, 231-241.
Lau, C, Anitole, It, Hodes, C., Lai, D., Pfahles-Hutchens, A., Seed, J., 2007. Perfluo-
roalkyl acids: a review of monitoring and toxicological findings. Toxicol. Sci. 99,
366-394.
Mendel, C.M., 1992. The free hormone hypothesis: distinction from the free hormone
transport hypothesis. J.Androl. 13,107-116.
Nestorov, I., 2003. Whole body pharmacokinetic models. Clin. Pharmacokinet. 42,
883-908.
OECD, 2002. Co-Operation on Existing Chemicals: Hazard Assessment of Per-
fluorooctane Sulfonate (PFOS) and Its Salts. Joint Meeting of the Chemicals
Committee and the Working Party of Chemicals, Pesticides and Biotechnology,
Organisation for Economic Cooperation and Development.
Olsen, G.W., Huang, H.Y, Helzlsouer, It]., Hansen, K.J., Butenhoff, J.L., Mandel, J.H.,
2005. Historical comparison of perfluorooctanesulfonate, perfluorooctanoate,
and other fluorochemicals in human blood. Environ. Health Perspect. 113,
539-545.
Olsen, G.W., Burris, J.M., Ehresman, D.J., Froehlich, J.W., Seacat, A.M., Butenhoff, J.L.,
Zobel, L.R., 2007. Half-life of serum elimination of perfluorooctanesulfonate,
perfluorohexanesulfonate, and perfluorooctanoate in retired fluorochemical
production workers. Environ. Health Perspect. 115,1298-1305.
Olsen, G.W., Mair, D.C., Church, T.R., Ellefson, M.E., Reagen, W.It, Boyd, T.M., Herron,
R.M., Medhdizadehkashi, Z., Nobiletti, J.B., Rios, J.A., Butenhoff, J.L., Zobel, L.R.,
2008. Decline in perfluorooctanesulfonate and other polyfluoroalkyl chemicals
in American Red Cross Adult Blood Donors, 2000-2006. Environ. Sci. Technol.
42, 4989-4995.
PubChem, 2008. http://pubchem.ncbi.nlm. nih.gov/summary/summary.cgi?cid=
3736298.
Seacat, A.M., Thomford, P.J., Hansen, K.J., Olsen, G.W., Case, M.T., Butenhoff, J.L.,
2002. Subchronic toxicity studies on perfluorooctanesulfonate potassium salt
in cynomolgus monkeys. Toxicol. Sci. 68, 249-264.
Seacat, A.M., Thomford, P.J., Hansen, K.J., Clemen, L.A., Eldridge, S.R., Elcombe, C.R.,
Butenhoff, J.L, 2003a. Sub-chronic dietary toxicity of potassium perfluorooc-
tanesulfonate in rats. Toxicology 183,117-131.
Seacat, A.M., Thomford, P.J., Hansen, K.J., Clemen, L.A., Eldridge, S.R., Elcombe, C.R.,
Butenhoff, J.L., 2003b. Erratum to "sub-chronic dietary toxicity of potassium
perfluorooctanesulfonate in rats". Toxicology 192, 263-264.
Seshagiri, P.B., Adiga, PR., 1989. Comparison of biotin binding protein of pregnant
rat serum with rat serum albumin. J. Biosci. 14, 221-231.
Tan, Y.M., Clewell III, H.J., Andersen, M.E., 2008. Time dependencies in perfluoroocty-
lacids disposition in rat and monkeys: a kinetic analysis. Toxicol. Lett. 177,38-47.
Teeguarden, J.G., Barton, H.A., 2004. Computational modeling of serum-binding pro-
teins and clearance in extrapolations across life stages and species for endocrine
active compounds. Risk Anal. 24, 751-770.
Trudel, D., Horowitz, L., Wormuth, M., Scheringer, M., Cousins, I.T., Hungerbuhler, It,
2008. Estimating consumer exposure to PFOS and PFOA. Risk Anal. 28,251-269.
Wambaugh, J.F., Barton, H.A., Setzer, R.W., submitted for publication. Comparing
models for PFOA pharmacokinetics Using Bayesian analysis. J. Pharmacokinet.
Pharmacodyn.
Washburn, S.T., Bingman, T.S., Braithwaite, S.It, Buck, R.C., Buxton, L.W., Clewell, H.J.,
Haroun, LA., Kester, J.E., Rickard, R.W., Shipp, A.M., 2005. Exposure assessment
and risk characterization for perfluorooctanoate in selected consumer articles.
Environ. Sci. Technol. 39,3904-3910.
Yuan, J., 1993. Modeling blood/plasmaconcentrations in dosed feed and dosed drink-
ing water toxicology studies. Toxicol. Appl. Pharmacol. 119,131-141.
Previous
TOC
-------
Journal of Environmental Science and Health Part C, 27:57—90, 2009 f*\ Tavlof &. FfanClS
Copyright © Taylor & Francis Group, LLC
ISSN: 1059-0501 (Print); 1532-4095 (Online)
DOI: 10.1080/10590500902885593
Predictive Models for
Corcinogenicity and
Mutagenicity: Frameworks,
State-of-the-Art, and
Perspectives
E. Benfenati,1 R. Benigni,2 D. M. DeMarini,3 C. Helma4
D. Kirkland,5 T. M. Martin,6 P Mazzatorta7
G. Ouedraogo-Arras,8 A. M. Richard,9 B. Schilter,7
W. G. E. J. Schoonen,10 R. D. Snyder,11 and C. Yang12
1Istituto di Ricerche Farmacologiche "Mario Negri," Milano, Italy
2Istituto Superiore di Sanita, Environment and Health Department, Rome, Italy
Environmental Carcinogenesis Division, US EPA, Research Triangle Park, North
Carolina, USA
4In Silica Toxicology, Basel, Switzerland
5Covance Laboratories Ltd, Harrogate, United Kingdom
6Sustainable Technology Division, National Risk Management Research Laboratory,
US EPA, Cincinnati, Ohio, USA
7Nestle Research Center, Quality and Safety Department, Lausanne, Switzerland
8L'Oreal, Safety Research Department, Aulnay-sous-Bois, France
9National Center for Computational Toxicology, US EPA, Research Triangle Park,
North Carolina, USA
10Schering-Plough Research Institute, Oss, The Netherlands
11 Schering-Plough Research Institute, Summit, New Jersey, USA
12 Center for Food Safety and Applied Nutrition, Food and Drug Administration, College
Park, Maryland, USA
Mutagenicity and carcinogenicity are endpoints of major environmental and regula-
tory concern. These endpoints are also important targets for development of alternative
methods for screening and prediction due to the large number of chemicals of potential
concern and the tremendous cost (in time, money, animals) of rodent carcinogenicity
bioassays. Both mutagenicity and carcinogenicity involve complex, cellular processes
that are only partially understood. Advances in technologies and generation of new data
Received January 29, 2009; accepted March 9, 2009.
Address correspondence to E. Benfenati, Head, Laboratory of Environmental Chem-
istry and Toxicology, Istituto di Richerce Farmacologiche "Mario Negri," Via La Masa
19, Milan 20156, Italy. E-mail benfenati@marionegri.it
57
Previous I TOC
-------
58 E. Benfenati et al.
will permit a much deeper understanding. In silica methods for predicting mutagenicity
and rodent carcinogenicity based on chemical structural features, along with current
mutagenicity and carcinogenicity data sets, have performed well for local prediction
(i.e., within specific chemical classes), but are less successful for global prediction (i.e.,
for a broad range of chemicals). The predictivity of in silica methods can be improved by
improving the quality of the data base and endpoints used for modelling. In particular,
in vitro assays for clastogenicity need to be improved to reduce false positives (relative
to rodent carcinogenicity) and to detect compounds that do not interact directly with
DNA or have epigenetic activities. New assays emerging to complement or replace some
of the standard assays include Vitotox™, GreenScreenGC, and RadarScreen. The needs
of industry and regulators to assess thousands of compounds necessitate the develop-
ment of high-throughput assays combined with innovative data-mining and in silica
methods. Various initiatives in this regard have begun, including CAESAR, OSIRIS,
CHEMOMENTUM, CHEMPREDICT, OpenTox, EPAA, and ToxCast™. In silica meth-
ods can be used for priority setting, mechanistic studies, and to estimate potency. Ulti-
mately, such efforts should lead to improvements in application of in silica methods for
predicting carcinogenicity to assist industry and regulators and to enhance protection
of public health.
Key Words: Carcinogenicity; mutagenicity; QSAR; in silica; predictive methods
INTRODUCTION
The EC-funded SCARLET (Structure-activity relationships leading experts
in mutagenicity and carcinogenicity) project was designed to investigate the
current status and future application of predictive models for carcinogenicity
and mutagenicity. The project organized a Workshop in Milan, Italy, 2-4 April
2008. Participants discussed the potentials and the issues of the methods but
also considered societal and industrial needs and the possibilities of interac-
tions with and incorporation of a wider body of scientific research dealing with
carcinogenicity and toxicity in general.
Studies on carcinogenicity and mutagenicity span diverse fields such as bi-
ology, toxicology, biochemistry, and chemistry. These studies are important not
only from a scientific point of view but also for the societal consequences re-
lated to the toxicity of carcinogenic and mutagenic compounds. To explore the
different perspectives, different tools may be preferable. In silico tools are those
based on computer programs and include so-called (quantitative) structure-
activity relationships; i.e., (Q)SAR methods. In silico tools offer advantages for
several scenarios, particularly where test data are unavailable or prohibitively
expensive and time consuming to generate. For this reason, it is useful to pro-
vide a general overview of the evolving field of carcinogenicity and mutagenic-
ity studies toward the goal of enhancing in silico approaches. More details on
the SCARLET project and the workshop can be found on the Internet (1).
It is not feasible or practical to report all the contributions and positions
related to the use of in silico methods for predicting mutagenicity and carcino-
genicity. Here we will organize the discussion in five areas to assess the utility
Previous I TOC
-------
Predictive Models for Carcinogenicity and Mutagenicity 59
of ire silico tools for different potential scenarios:
1. The scientific framework of carcinogenicity and mutagenicity studies;
2. The needs of industry and regulators;
3. State-of-the art of the methods of ire silico prediction for carcinogenicity and
mutagenicity;
4. Connecting new and traditional data into a new scenario;
5. Conclusions.
THE SCIENTIFIC FRAMEWORK OF CARCINOGENICITY AND
MUTAGENICITY STUDIES
The meeting opened with a brief overview of the emerging knowledge of mecha-
nisms underlying mutagenesis and carcinogenesis. The importance of the "mu-
tagenesis paradigm" was emphasized because this is the general step-wise pro-
cess by which a cell deals with DNA damage. Most chemical mutagens do not
directly induce mutations (i.e., a change in nucleotide sequence in the DNA).
Instead, mutagens induce DNA damage, which can be, for example, a DNA
adduct (a molecule bound covalently to a nucleotide) or a single- or double-
strand break. In either case, the primary nucleotide sequence is not changed.
A complex array of signalling pathways detects the DNA damage and directs
the cell to do one of three general things: repair the damage, convert the dam-
age to a mutation, or signal the cell to die (apoptosis). Thus, mutagenesis is
a cellular process involving enzymatic activities and usually DNA replication.
Consequently, mutagens make DNA damage and cells make mutations. Mod-
elling this complex process directly may be possible in the future as a clearer
understanding of the underlying signalling pathways emerges.
Carcinogenesis is now clearly understood to proceed in a Darwinian
process in which cells with a growth advantage are selected in a step-wise
fashion, resulting in a tumor. Both mutational and epigenetic (non-mutational)
events are involved and are necessary for this process. Epigenetic events are
changes in gene expression and do not involve a change in nucleotide sequence
(mutation). Gene expression can be modulated by methylation of DNA or
methylation or acetylation of histones, which are proteins surrounding the
DNA. Carcinogenesis can be initiated by either a genetic (mutational) or
epigenetic (non-mutational) event. However, both are ultimately necessary for
the formation of a tumor.
More and more studies are directed toward development of alternative
tests to animal studies. However, given the high frequency of irrelevant
positives from ire vitro mammalian cell tests, unless there is improved accu-
racy of these tests to predict ire vivo genotoxic or carcinogenic hazard, many
Previous
-------
60 E. Benfenati et al.
substances will be classified inappropriately; i.e., falsely labelled hazardous
or even banned. Improvements to basic cell culture are needed to avoid
reactions between test substance and culture medium that can result in
production of clastogenic levels of hydrogen peroxide. A review of the top
concentration (10 mM) for testing non-toxic and freely soluble substances
is urgently needed to see if this concentration is justified. Many different
measures of cytotoxicity can be used to determine an appropriately cytotoxic
top concentration, but there is now evidence that these do not always select
the same concentration. If a measure is used that underestimates the toxicity,
then higher-than-warranted concentrations may be used and positive artefacts
may occur. Also, the levels of cytotoxicity required in chromosomal aberration
and mouse lymphoma assays are largely based on old published data, and
it is not certain that these can be substantiated if more modern protocols
are used (2). These and other factors that may contribute to a high rate of
irrelevant positive results were highlighted by Kirkland et al. (3). A 3-year
research program is underway that is funded by the EU cosmetic industry
association COLIPA and supported by the UK National Institute for the 3Rs,
together with ECVAM, to evaluate changes in study design that would reduce
the frequency of irrelevant positive results.
Also discussed was the concept of non-covalent DNA interactions. For ex-
ample, DNA groove-binding or intercalation, which has not been modelled ad-
equately, might also explain many "false positive" findings. Nearly one half of
the marketed drugs that are not structurally alerting but are still positive in
in vitro cytogenetics assays may operate through DNA intercalation and topoi-
somerase poisoning (4, 5). This conclusion has been drawn both from cell-based
testing and 3-D DNA docking evaluations.
Some Examples of Recent In Vitro Techniques
Recently, three different assays were examined for implementation as mu-
tagenicity, genotoxicity, and/or clastogenicity assays; i.e., Vitotox, GreenScreen
GC, and RadarScreen. For clastogenicity also, in vivo experiments have to be
performed with rodents.
Vitotox
The Vitotox assay (Thermo Technologies, Finland) is performed in
Salmonella typhimurium strain TA104 (6-8). The assay is based on the ac-
tivation of an SOS repair system by genotoxic compounds. In these bacteria, a
luciferase gene of the beetle Vibrio frescio is introduced by molecular design,
and this gene is under the transcriptional control of the recN promoter. This
recN promoter, in turn, is strongly repressed, which prevents the luciferase
gene expression under control conditions. In the presence of a DNA-damaging
genotoxic compound, the RecA regulator protein recognizes the resulting
Previous I TOC
-------
Predictive Models for Carcinogenicity and Mutagenicity 61
Table 1: Predictive values of Vitotox, GreenScreen GC and RadarScreen assays in
comparison with scores for full Ames tests and a set of 156,81, and 154
compounds, respectively. The compounds are references, a specific set of
pre-clinical and clinical developmental compounds of Organon, and a specific
clastogenic set of steroidal compounds (20)
Vitotox
Sensitivity
Specificity
Concordance
90
90
90
n
48
108
156
GreenScreen GC
39
98
74
n
33
48
81
RadarScreen
55
52
53
n
47
107
154
free ends or mismatches in DNA. This results in a cascade of biochemical
interactions and reactions leading to the de-repression of the strong recN pro-
moter and subsequent transcription of the luciferase gene. This will enhance
the production of the luciferase protein. After addition of luciferin and ATP,
the synthesis of light can then be visualized. Light enhancement is an indi-
cation for genotoxicity. A second S. typhimurium TA104 strain, constitutively
expressing the luciferase gene, is used as a control strain for measuring cyto-
toxicity. A ratio score of 1.5 between the genotoxicity and cytotoxicity strains
is used as a cut-off for real genotoxicity.
The assay can be performed in the presence or absence of S9 fraction of rat
liver homogenates induced with Aroclor 1254. With the reference compounds,
4-nitroquinolin-l-oxide and benzo[a]pyrene, reproducible results were found
in independent assays. Table 1 shows the results of these tests in comparison
with the Ames test. The concordance of the Vitotox test and the full Ames test
was 90%. The sensitivity was 90% for the true positives, and the specificity
was 90% for the true negatives. The potential advantage of this assay relative
to the standard Ames assay is that smaller volumes/amounts of sample are
needed, cytotoxicity is determined, and it is faster (results are obtained in 3 h
versus 3 days).
GreenScreen GC and RadarScreen
The GreenScreen GC (Centronix, UK) and RadarScreen (reMYND, Bel-
gium) assays are carried out in yeast (Saccharomyces cerevisiae). The advan-
tage of these yeast assays is that chromosomal segregation in the meiotic phase
in yeast resembles that of vertebrates. In these assays, the focus is on the acti-
vation of the RAD54 promoter, which is the analogue of the RAD51 promoter in
vertebrates (9). Both RAD54 and RAD51 are recombinational repair genes that
belong to the class of RAD52 genes, which are involved in the repair of chro-
mosomal double-strand break damage (10-13). In the GreenScreen GC strain,
the RAD54 gene has been replaced by the gene encoding Green Fluorescence
Protein (GFP) (14-17), whereas in the RadarScreen strain the /3-galactosidase
Previous I TOC
-------
62 E. Benfenati et al.
(LacZ) gene was introduced (18, 19). The end-point measurement with GFP is
assessed after 24 h and with /3-galactosidase after 6 h. The control analysis for
cytotoxicity can be carried out in the same sample by adsorption measurement
with a spectrophotometer.
A drawback of the GFP system is that some compounds gave autofluores-
cence at the wavelength of GFP. Another issue is that the addition of the S9
fraction caused quenching of the GFP signal. With a specific set of 100 candi-
date pharmaceuticals, autofluorescence occurred with 15 compounds, whereas
for 15 other compounds activation with the rat-liver S9 fraction was essen-
tial. These problems could be overcome with the use of the /3-galactosidase
enzyme. This enzyme can cleave 6-0-/3-galactopyranosyl-luciferin (Promega,
Madison, WI) into free luciferin, which in turn can be quantified with luciferase
of the firefly. At the wavelength of 650 nm, no quenching was found with rat-
liver. Due to luminometry, this assay became very sensitive, leading to a better
identification of clastogenic compounds. With the reference compounds, methyl
methanesulphonate and benzo[a]pyrene, reproducible results were found in in-
dependent assays.
With both the GreenScreen GC and RadarScreen assays, the predictivity
with the full Ames assay was relatively low compared with the Vitotox eval-
uation, leading to sensitivity of 39% and 55% and specificity of 98% and 52%,
respectively. Table 2 shows the results of these tests compared with the clasto-
genicity test. This new RadarScreen assay should be a good predictor for the
identification of real clastogenic and/or carcinogenic compounds. For example,
among 40 steroidal compounds (20) a predictivity score concordance of 82.5%
was calculated. With respect to the compounds tested in the full Ames test,
12 Ames negative compounds were identified with clastogenic human lym-
phoblast or CHO assays. For these 12 compounds, a concordance of 66% was
found between the RadarScreen and Ames assay. Overall analysis of 132 com-
pounds with this RadarScreen assay, compared with the available in vitro clas-
togenicity/aneuploidy assays, found that the sensitivity for the positive com-
pounds was 80%, the specificity for the negative compounds was 77%, and the
Table 2: Predictive values for Vitotox, GreenScreen GC and RadarScreen assays in
comparison with scores for in vitro clastogenicity/aneuploidy for a set of 132,44,
and 130 compounds, respectively. The compounds are references, a specific set of
pre-clinical and clinical developmental compounds of Organon and a specific
clastogenic set of steroidal compounds (20)
Vitotox
Sensitivity
Specificity
Concordance
29
89
51
n
85
47
132
GreenScreen GC
22
95
57
n
23
21
44
RadarScreen
80
77
78
n
83
47
130
Previous I TOC
-------
Predictive Models for Carcinogenicity and Mutagenicity 63
Table 3: Sensitivity values of Vitotox and RadarScreen assays separately and
combined according to scores for in vitro mutagenicity in the Ames test for a set of
156 compounds, for in vitro clastogenicity assays for a set of 85 compounds, and
for in vitro carcinogenicity assays using tests as in (21) for a set of 50 compounds,
these being references, a specific set of pre-clinical and clinical developmental
compounds of Organon, and a specific clastogenic set of steroidal compounds
(20)
Sensitivity (+ and -)
Mutagenicity
Vitotox
RadarScreen
Vitotox + RadarScreen
43/48
26/47
23/24
90
55
96
Clastogenicity
5/17
66/83
66/85
29
80
78
Carcinogenicity
3/10
40/49
41/50
30
82
82
concordance was 78% (Table 2). This same analysis for the GreenScreen GC
assay with 44 compounds showed a sensitivity of 22%, a specificity of 95%, and
a concordance of 57% (Table 2).
A positive score in a mutagenicity, genotoxicity, clastogenicity, or carcino-
genicity assay is seen as a strong negative factor for a compound that leads
most times to removal of the compound. Introduction and the combined usage
of the Vitotox and RadarScreen tests may lead to a better deselection and/or
ranking of the lead compounds in the early phase of the selection process with
respect to genotoxic and clastogenic compounds. The sensitivity of the Vitotox
and RadarScreen assays together was 96% compared with the Ames assay, 78%
with clastogenicity tests, and 82% with in vitro carcinogenicity tests described
in (21) (Table 3).
Thus, the combination of the Vitotox and RadarScreen in vitro assays leads
to a very high predictivity for in vitro mutagenicity, clastogenicity, and in vitro
carcinogenicity tests. Moreover, with the introduction of the AhR assays with
rat H4IIE and human HepG2 cells, a possible risk factor for non-genotoxic
carcinogens might be identified early in the selection process.
THE NEEDS OF INDUSTRY AND REGULATORS
The chemical industry is facing increasing challenges regarding the safety as-
sessment of chemicals by regulatory agencies throughout the world. In par-
ticular, in Europe REACH (Registration, Evaluation, Authorization, and re-
striction of Chemicals) took effect on 1 June 2007. Due to the large number
of chemicals to evaluate, REACH promotes the use of alternative methods to
animal testing to achieve these goals within a reasonable timeframe. REACH
requires that a large number of substances be tested for genotoxicity and, be-
cause positive results in vitro usually trigger additional in vivo testing, a large
role for predicting/confirming these results is envisaged.
Previous I TOC
-------
64 E. Benfenati et al.
For the cosmetic industry, the seventh amendment to the European Cos-
metic Directive will also phase out the use of animals for testing ingredients
starting in March 2009. Therefore, the use of non-animal methods is manda-
tory for the seventh amendment, whereas it is a recommendation for REACH.
Both regulations affect not only companies located in Europe but also those
exporting chemicals to Europe. There are incentives for seeking alternative
methods (in vitro and in silico methods) in hazard/risk assessment. It should
be emphasized that structure-activity concepts are already used, for example,
by the U.S. Environmental Protection Agency (EPA).
The EC Research Projects
To deal with this new situation and these new regulatory requirements, the
European Union (EU) has promoted a wide range of initiatives to adequately
prepare for REACH. Several EC-funded projects relate exclusively or partly to
the REACH regulation and in silico models, and others will start soon.
CAESAR (Computer Assisted Evaluation of industrial chemical Sub-
stances According to Regulations) is an EC-funded project, coordinated by E.
Benfenati (Milan), that is devoted to developing in silico models for REACH for
five endpoints: carcinogenicity, mutagenicity, reproductive toxicity, bioconcen-
tration factor, and skin sensitization. All models will be made freely available
at the Web site: http://www.caesar-project.eu.
Because the purpose is related to regulation, careful attention has been
given to the source of the data to be used for QSAR models, including checking
the suitability of the experimental procedure according to REACH. Further-
more, great care has been given to the data quality from a chemical and toxic-
ity point of view. Many mistakes have been found in the data, even when data
have been taken from recent papers. Double checking of all chemical struc-
tures found that more than 10% of the chemical structures were incorrect (22).
Conversely, some sources, such as the EPA DSSTox database (23), were found
to be of high quality.
OSIRIS (Optimized Strategies for Risk Assessment of Industrial Chem-
icals based on Intelligent Combinations of Non-Test and Test Information) is
an EC-funded project, coordinated by G. Schuiirmann (Leipzig), that is devoted
to developing intelligent testing strategies for REACH. In silico, in vitro, and
in vivo data will be integrated with exposure scenario. Software will be devel-
oped to assist users in determining if the data are sufficient for modelling or
if further experiments are needed. Models from CAESAR also will be incorpo-
rated within OSIRIS.
Other EC-funded projects are highly relevant to supporting REACH
legislation, even if the project descriptions were not dedicated specifically
to REACH. We already introduced SCARLET, coordinated by E. Benfenati
(Milan). The EC-funded project CHEMOMENTUM, coordinated by P. Bala
Previous I TOC
-------
Predictive Models for Carcinogenicity and Mutagenicity 65
(Warsaw), will produce software for automatic QSAR modelling. The user will
have the capability to build up workflows and extract data from databases and
chemical structures from repositories. Chemical descriptors will be calculated
automatically, including 3D descriptors, and a battery of algorithms will be
available for modelling. A grid-based approach will allow different parts of the
centralized workflow to be physically present in different locations without af-
fecting software performance. This user-friendly scheme will allow simplified
and seamless modelling, whereas today the user has to switch from the dif-
ferent components. CHEMOMENTUM also will incorporate docking tools. The
output of the model will also be available in QSAR Model Reporting Format as
specified by requirements of the Organization for Economic Cooperation and
Development (OECD). Models for REACH will be available that utilize the
data produced within CAESAR.
CHEMPREDICT is an EC-funded project coordinated by E. Benfenati (Mi-
lan). This project is devoted to developing models focused on new simplified
chemical descriptors that are based on the simplified molecular input line en-
try system (SMILES) format. Endpoints for modelling include carcinogenicity
and genotoxicity.
OpenTox is a new EC project coordinated by B. Hardy (DouglasConnect)
that started in 2008. It is focused on the development of an open-source frame-
work providing unified access to toxicity data, (Q)SAR models, and validation
procedures.
The European Chemicals Bureau (ECB), among others, has promoted stud-
ies aimed at clarifying the state of the art in the field of (Q)SARs for muta-
gens and carcinogens. These studies have been described in official EU reports
(24, 25) as well as in scientific publications (26, 27).
Several initiatives have been implemented to help companies fulfill the
requirements of legislation. The EPAA (European Platform for Alternative
Approaches to Animal testing) is one such initiative (28) that aims to facili-
tate efforts to share knowledge and research to accelerate implementation and
acceptance of the 3Rs (Replacement, Reduction, Refinement). The EPAA was
founded in 2005 from a joint initiative of the European Commission companies
and trade associations. This partnership consists of a steering committee, a
mirror group that provides critical input to the steering committee, and five
working groups. The EPAA will map currently used and ongoing projects in
the area of the 3Rs, implement projects where there are gaps, and promote the
acceptance and the use of the 3Rs. The EPAA is seeking to partner not only
with European stakeholders but with international initiatives as well.
The Situation in the Pharmaceutical Industry
Within the pharmaceutical industry, the focus of structure-activity mod-
elling is typically on the development of new drugs with a beneficial effect
Previous I TOC
-------
66 E. Benfenati et al.
on a particular disease at an acceptable low dose level. Treatment should im-
prove the well-being of the patient, whereas unwanted side-effects should be
absent or very low. However, if the drug has to be administered at relatively
high dose levels, side effects may be induced due to supra-pharmacology or off-
target pharmacology. Moreover, at high dosages, the compound itself and/or its
metabolites may induce adverse side effects, such as genotoxicity, carcinogenic-
ity, reprotoxicity, hepatotoxicity, nephrotoxicity, cardiotoxicity, neurotoxicity, or
blood or skin toxicity. Within the portfolio of Organon, the toxicity failure rate
from 1960 to 2000 was due mainly to genotoxicity/carcinogenicity (20%) and re-
protoxicity (20%), followed by hepatotoxicity (12%), nephrotoxicity (12%), and
cardiotoxicity (12%). This portfolio is completely different from that of Roche
during the same period of time in which genotoxicity/carcinogenicity reached
only a level of 6% and reprotoxicity a level of 2%. On the other hand, hepato-
toxicity, nephrotoxicity, and cardiotoxicity reached levels of 20%, 4%, and 16%,
respectively. These figures can easily be explained by virtue of the focus on
reproductive medicine within Organon. One of the main topics in this field is
male and female contraception as well as hormone replacement therapy. The
effects of estrogens, progestagens, and androgens with respect to these toxicity
areas are well known. An overall reduction in the attrition rate of compounds
with 20% to 50% failure rates by means of introducing early toxicity screening
might lead to a sharp cost reduction in drug development (29-31).
The strategy within a pharmaceutical industry should be to start with
early toxicity screens as early as possible in the discovery process. Possible
toxicity screens could be divided into mutagenicity, genotoxicity, clastogenic-
ity, cytotoxicity, cytochrome P450 induction, and nuclear receptor activation.
These predictive toxicity assays should be an integral part of the "ranking and
selection" process of candidate drugs, should be on a medium to high through-
put level, should only use a small amount of compound (1 to 20 mg), and should
be carried out with a limited amount of man power. Implementation of in sil-
ico procedures with structure-activity relationship (SAR) programs, such as
DEREK, TOPKAT, MultiCASE, and Mutalert, can already be initiated at the
stage of hit selection, in vitro assays for measuring mutagenicity, genotoxic-
ity, clastogenicity, non-genotoxic carcinogenicity, cytotoxicity, nuclear receptor
activation, and cytochrome P450 enzyme activation, as well as competition as-
says, can be implemented from the point of lead selection to the final choice of
the development candidate in the preclinical selection phase.
Genotoxicity Tests within the Pharmaceutical Industry
According to the guidelines of the U.S. Food and Drug Administration
(FDA), four different endpoints testing mutagenicity and clastogenicity are
considered for a new drug approval or food ingredients notifications. These
tests are:
Previous I TOC
-------
Predictive Models for Carcinogenicity and Mutagenicity 67
1. An in vitro test for gene mutation in bacteria, i.e., Salmonella reverse mu-
tation;
2. An in vitro test with cytogenetic evaluation of chromosomal damage, e.g.,
in vitro micronucleus or in vitro chromosome aberration;
3. In vitro mammalian mutation, i.e., in vitro mouse lymphoma Tk+/~ assay;
and
4. An in vivo test for chromosomal damage using rodent hematopoietic cells,
i.e., micronucleus assay.
These in vitro tests are still relatively time-consuming, have a relatively
low throughput, and require a relatively large amount of compound.
Non-genotoxic Carcinogens within the Pharmaceutical Industry
Activation of the AhR with dioxin (TCDD) leads in both rats and humans
to an increased incidence of liver tumors (32-34). TCDD is, therefore, classi-
fied as a non-genotoxic carcinogen. The difficulty in using AhR activation as
a marker for non-genotoxic carcinogenesis lies in the fact that not all AhR in-
ducers are necessarily non-genotoxic carcinogens. For instance, grapes, fruits,
and vegetables, which are generally considered very healthy, contain chemicals
that enhance AhR activity and, thus, affect AhR-driven pathways. The main
question is whether these protective mechanisms increase or decrease tumor
incidence. Thus, a compound that activates the AhR can either be beneficial
or carcinogenic. Because it is difficult to predict in which class an AhR acti-
vator will fall, it is advisable to steer away from AhR activation during drug
development, even though good compounds may be thrown out with the bad.
For the identification of at least one specific group of non-genotoxic carcino-
gens, it is relevant to measure the activation of the rat and/or human AhR. For
this, simple cellular assays are available in both rat H4IIE and human HepG2
cells, both of which make use of the metabolism of 3-cyano-7-ethoxycoumarin
(CEC) by cytochrome P450 enzyme 1A1 and/or 1A2—enzymes that are induced
following AhR activation. For TCDD and 3-methylcholanthrene, the activity is
similar in both cell lines. However, species-specific differences have also been
observed; e.g., indigo activity is dominant in the rat cell line, whereas indirubin
is a much stronger AhR agonist in the human cell line. On the other hand, com-
pounds such as menadione activated only the human AhR receptor, whereas
flutamide, Org D, PCB156, and PCB157 were specific for the rat receptor (35).
This species difference has also been observed for a number of Organon pro-
prietary compounds.
As mentioned above, existing in silico models may "miss" prediction of
many genotoxicity tests, which in turn may be related at least in part to the
fact that the existing computational programs cannot detect non-covalent DNA
Previous
-------
68 E. Benfenati et al.
interactions. Very few DNA intercalators were employed in the training sets of
these models; even if they were, most of them were classical fused planar multi-
ring compounds such as acridines rather than the atypically structured inter-
calators reported recently (4, 5). Because of the complexity of the SARs and the
probable need to account for electrostatic and other (i.e., van der Waals, hydro-
gen bonding) effects, it will be difficult to improve programs such as DEREK or
MCASE to identify such molecules. As the genotoxicity database grows, how-
ever, it may be possible to establish a predictive tool for non-covalent chemi-
cal/DNA interactions.
The Food Industry and the Need of In Silico Models for
Carcinogenicity and Mutagenicity
In recent years there has been mounting concern about food as a source of
exposure to potentially toxic chemicals. It has been estimated that there are
over five million man-made chemicals known, of which 70,000 are in use today.
The application of continuously improving analytical methods has revealed
that many of these chemicals can enter the food chain and result in human
exposure.
Food chemical risk assessment is the scientific process used to characterize
the health significance of potentially harmful chemicals in food. Classically, it
comprises four steps: (1) hazard identification; (2) hazard characterization; (3)
exposure assessment; and (4) risk characterization. In general, hazard identifi-
cation and characterization rely on toxicological data obtained in experimental
animals, mainly rodents. Because toxicological information is limited or absent
for the majority of the inadvertent food-borne chemicals, the assessment of the
health significance of such chemicals is difficult or impossible. Nevertheless,
the detection of such chemicals in food products may trigger not only heavy
management action (e.g., public recall) but also public concern resulting in loss
of consumer confidence for the food supply. In such situations, the availability
of reliable tools to establish levels of safety concern without hard toxicological
data appears of particular importance to ensure adequate consumer protection
without undue overconservatism. This should ultimately allow optimal use of
the limited resources available.
Solutions to this general issue are not straightforward. Obviously, experi-
mental toxicology is not a practical tool to deal with situations requiring fast
decision making. Furthermore, even if sufficient facilities to perform toxicolog-
ical testing within a relevant time frame were available, it still can be ques-
tioned whether testing a large number of substances would be a rational and
practical approach. In this context, in silico predictive models have obvious
advantages in terms of time, cost, and animal protection.
In silico strategies are already proactively and successfully used for pre-
clinical screening in pharmaceutical discovery pipelines in which an early
Previous I TOC
-------
Predictive Models for Carcinogenicity and Mutagenicity 69
identification of toxicological hazard offers a clear competitive advantage. Such
efforts allow the exclusion of chemicals that could potentially produce unac-
ceptable adverse effects in further regulatory toxicology tests. The situation
of the food industry is different and requires the development of alternative
models with the following specific characteristics:
• Risk rather than hazard-based. In the food context, the most likely appli-
cation of computational toxicology models would be in the establishment
of the level of safety concern associated with the inadvertent/accidental
presence of a chemical in products. This requires not only qualitative in-
formation on the potential hazardous properties of the chemical (e.g., car-
cinogenicity) but also quantitative information (e.g., carcinogenic potency),
allowing the derivation of a margin of exposure (MoE) with the estimated
intake. The interpretation of the size of the MoE (e.g., allowing for various
uncertainties such as inter- and intra-species differences) would likely help
to make decisions at the management level.
• Reliable, high sensitivity. Most (Q)SAR predictive models suffer from in-
herent poor sensitivity; i.e., the ability to correctly identify true positives
(36). Modellers, partially because they are often confronted with non-
representative datasets, have focused their attention on identification of
toxicophores that are overly general and, as a result, models tend to have
many false positives. This has made computational toxicology a useful tool
for high-throughput screening (HTS) but different strategies should be op-
timized if the target is to have a low number of false negatives or a high
concordance.
• Global. Compounds found in foods and food ingredients present a high
structural diversity and complexity that may be greater than synthetic
Pharmaceuticals (37) and, therefore, require the development of global in
silico models.
Ideally, in silico toxicology strategies should predict adverse effects in
the human population. Because the toxicological training databases currently
available consist mainly of in vitro and animal data with high limitations to
predict human situations (38), the development of such models will always
constitute a significant challenge. Their practical application in the food sector
will depend on their potential to accurately predict endpoints that are cur-
rently used to make food safety management decisions. This includes the need
to establish confidence limits. The acceptance of these models will be possible
only if the analysis is fully transparent. Therefore, the promotion of validated,
freely available tools based on open-source codes, such as those developed by
ECB and EPA, is recommended.
Previous I TOC
-------
70 E. Benfenati et al.
STATE-OF-THE-ART OF IN SILICO METHODS FOR PREDICTION OF
CARCINOGENICITYAND MUTAGENICITY
To properly evaluate the utility of current in silico methods, it has to be clari-
fied that different purposes are envisioned and, thus, different evaluations are
possible. Also, for this reason, practical applications of in silico programs for
prediction of mutagenicity or carcinogenicity have ranged in utility from indis-
pensable to useless. Next we list some of the different ways that in silico tools
for prediction of mutagenicity and carcinogenicity can be employed.
Priority Setting
Models are usually used to set priorities among chemicals for further test-
ing. For this use, several (Q)SAR models and databases are commercially (or
freely) available, typically for alert-identification and read-across. Due to the
differences in the systems (knowledge-based versus artificial intelligence, SAR
versus QSAR, applicability domains, and extent to which mode-of-action is
considered in the model development), combining several models appears to
be more sound than relying on a single one. The so-called global and local
models may be used for the purpose of priority setting. Because regular up-
dates of the models are released, care should be taken to re-assess their per-
formances on a regular basis. Therefore, active participation of industry is
needed to evaluate and improve existing in silico models so they can meet
their needs. For REACH, although the chemicals to be characterized have
been chosen on the basis of the amounts produced annually, it may be valu-
able to further prioritize them using global SAR. Thus, those chemicals that
pose major concern may be identified and given a higher priority for further
evaluation.
Mechanistic Investigation
When using human knowledge-based systems such as Derek (Lhasa Ltd.)
or OncoLogic (U.S. EPA) or any model based on a mechanistic understanding
(as opposed to models based purely on statistics), it is possible to gain insight
into the mechanism underlying the mutagenicity/carcinogenicity.
Quantitative Evaluation of the Potency
In REACH this evaluation is requested in the case of genotoxic compounds
to assess if the expected exposure level for the scenario of use of the chemical
compound will produce an unacceptable risk.
Although a particular model may provide results unacceptable for a cer-
tain use, for a different purpose the same model can be useful. For instance,
a global model with overall prediction accuracy of 65-70%, as can be the case
Previous I TOC
-------
Predictive Models for Carcinogenicity and Mutagenicity J
for carcinogenicity models, might be considered unsuitable as a substitute for
traditional testing methods; however, the same model might be useful when
combined with other considerations as a means for prioritization. Further-
more, even within the same regulation, such as REACH, in some cases a clas-
sifier model can be useful, for instance to identify the presence for a certain
mutagenic fragment, whereas for other purposes, such as in support of risk
assessment, a quantitative model with less uncertainty is necessary because
the toxic effect has to be considered along with the exposure level.
There are many publications on the databases, structural alerts, and mod-
els, and a number of commercial products are built on these. Publicly available
genetic toxicity and carcinogenicity data sets include CCRIS (39), EPA Gene-
tox (40), NTP (41), IARC (21, 42), EPA IRIS (43), U.S. FDA CRADA database
(44), Tokyo-Eiken (45), Mutants (46), CPDB (47, 48), ISSCAN (49), and pri-
mary publications. Commercial databases derived from these sources also are
available from Leadscope (50), Lhasa (51), and MDL (52). These databases pro-
vide test results form both regulatory-accepted test protocols as well as other
screening methods.
Based on these data, commercial prediction models are currently mar-
keted in the form of global models, including MultiCASE (53), TOPKAT (54),
MDL QSAR (55), and Leadscope FDA Model Applier (56), whereas OncoLogic
(57) and LAZAR (58) are freely available for use. Derek from Lhasa offers a
knowledge-base classification. Toxtree is a software tool developed by ECB
(through IdeaConsult Ltd.) that is able to estimate different types of toxic haz-
ards by applying structural rules; it is an open-source, freely available appli-
cation that can be downloaded from the ECB Web site (59). The new module
predicts mutagenicity and carcinogenicity by applying a revised, updated list
of Structural Alerts (SA), and, when applicable, three QSARs for congeneric
classes (see details in the Toxtree scientific manual) (24, 25). The Leadscope
system provides data-mining and prediction methods based on both biologi-
cal and chemical databases. Currently, researchers at the U.S. FDA and U.S.
EPA use Leadscope for chemical and biological read-across based on databases
and to build predictive models. Many of the above prediction methods tend to
rely heavily on chemical structures and summarized biological endpoint data.
Quite often these summarized endpoints are too far removed from the original
experimental measures and, hence, have lost at least some of their biological
context.
In mutagenicity and carcinogenicity, it is possible to distinguish between
(a) coarse-grain methods, relying on the recognition of SAs; (b) fine-tuned ap-
proaches, which include Quantitative Structure-Activity Relationships (QSAR)
methods for congeneric classes of chemicals (same chemical scaffold, same pre-
sumed mechanism of action); and (c) global QSAR models that attempt to
combine elements of the previous two approaches and address more chemical
classes (60).
Previous I TOC
-------
72 E.Benfenatietal.
Furthermore, some studies are based on human expertise, which identifies
a series of structural alerts, whereas other tools are based on techniques used
to discover the presence of relationships not yet known; data-mining tools are
often used in the latter case. Examples of models that codify human knowledge
include HazardExpert, OncoLogic, Toxtree, and DEREK. MultiCASE, Lead-
scope, and LAZAR use software based on the data-driven discovery of geno-
toxic or chemical fragments identified by specific automated algorithms. There
are also mechanistic-based models that rely on prediction of likely chemical
reactions (61).
A comparative analysis of existing lists of SAs derived for the prediction
of rodent carcinogenicity has indicated that they have a prediction accuracy
of 65% for the rodent carcinogenicity bioassay and an even better prediction
accuracy of 75% for Salmonella mutagenicity results; i.e., surprisingly, rodent
carcinogenicity SAs predict mutagenicity better than they predict rodent car-
cinogenicity (26). In addition, these SAs and the Salmonella assay have been
shown to be equally predictive of the rodent carcinogenicity data. Overall, the
SAs are a powerful tool for coarse-grain characterization of the chemicals; i.e.,
for description of sets of chemicals, preliminary hazard characterization, cate-
gory formation for regulatory purposes, or for selecting subsets of chemicals to
submit to fine-tuned QSAR analyses for priority setting. A previous analysis on
the priority setting criteria adopted by the U.S. National Toxicology Program
in selecting chemicals to be bioassayed has shown that the structural criteria
adopted to short-list suspect chemicals were able to enrich the target up to
ten times. In fact, 70% of the chemicals bioassayed as suspect carcinogens (i.e.,
SAs or positive Salmonella data) were carcinogens, whereas only 7% of the
chemicals bioassayed based on production/exposure considerations were car-
cinogenic (62). This result points to the high reliability of SAs for priority set-
ting. On the other hand, the SAs are overly general and are not well suited as
a tool for discriminating between positives and negatives of congeners within
a chemical class; this is the role of the local, fine-tuned QSARs for congeneric
classes (27).
A survey of local QSARs for congeneric classes of chemicals has shown that
these are classified into (a) models for the gradation of the potency of the pos-
itives (mutagens or carcinogens) and (b) those that discriminate between pos-
itives and negatives. This is a crucial difference with respect to models such
as those for aquatic toxicity, where it is assumed that all the chemicals can be
scaled along one axis of potency, ranging from highly potent to weakly potent
chemicals. In mutagenicity and carcinogenicity this does not generally hold
true; i.e., models for potency most often fail to separate positives from nega-
tives. Thus, the models can be applied in two phases: first to separate positives
from negatives, and second to assess the potency of the chemicals predicted as
positive in the first phase (63).
The survey on QSARs included (a) a short list of promising models; (b) re-
calculation of the statistics; and (c) most importantly, the performance of real
Previous I TOC
-------
Predictive Models for Carcinogenicity and Mutagenicity 73
external predictivity tests. The latter consisted in selecting from the literature
test chemicals falling in the same applicability domains of the training sets,
but that had never been considered by the authors of the models. The QSARs
selected were all scientifically interpretable, had good internal statistics and
cross-validation, but varied widely in their external predictivity. The QSARs
for potency had an external prediction ability in the range 30-70% correct
(percentage of chemicals whose potency was correctly predicted within 1 log
unit). On the other hand, the QSARs for activity (yes/no) had an external pre-
diction ability in the range 70-100% correct. This indicates that classification
estimates (e.g., yes/no) generally are much more reliable than estimating data
points or relative potency rankings. This also confirms the dichotomy between
QSARs for potency and QSARs for positivity/negativity (64).
Another important result of the survey was that internal validation mea-
sures (e.g., cross-validation, other statistics) are not good predictors of external
predictivity (26, 64). Hence, they should be considered only as a means for bet-
ter describing the performances of training sets. It should be emphasized that
the external predictivity of high quality, local QSARs (70-100%) is in the same
range as the intra-assay agreement of the generally reliable and reproducible
Salmonella mutagenicity assay (80-85%) (65). Hence, the uncertainty inher-
ent to the two methods is comparable. In addition, as indicated above, rodent
carcinogenicity SAs correlate with rodent carcinogenicity bioassay results on a
large database to the same extent as the Salmonella assay results.
The OECD guidelines point out that to facilitate the consideration of a
(Q)SAR model for regulatory purposes, the model should be associated with
a mechanistic interpretation if possible (66). Furthermore, mechanistically
based (Q)SARs provide (a) a common ground of discussion for modelers, tox-
icologists, and regulators; (b) additional tools for minimizing the possibility
of chance correlation; (c) intelligible information to guide synthetic chemists
in preparing safer chemicals; and (d) a rational foundation for developing a
QSAR science.
On the other hand, the large amount of legacy data and the anticipated ex-
plosion of toxicity-related information expected to be generated over the next
year (see, for instance, the ToxCast™ program below), call for the application
of flexible data-mining tools. In the past, for instance, Bursi and coworkers (67)
showed results of a global SAR model for mutagenicity with accuracy similar
to that of the intra-assay experiments for the Ames test mentioned above. Re-
cently, Gini and Ferrari showed similar results obtained within the CAESAR
project (68). It should be mentioned that for mutagenicity data, modelling data
sets consisting of several thousands of compounds are available and, in this
case, the role of modern computer techniques are suitable to screen a wide se-
ries of possibilities. In this way, it is possible to explore relationships between
the presence of a certain fragment and the toxicity and, thus, to mimic the
process done manually by the human experts. Thus, data-mining tools can be
used to explore data in new ways and to identify novel toxicity mechanisms.
Previous I TOC
-------
74 E. Benfenati et al.
A further contribution of global QSAR models (69) was presented by Toropov
and coworkers that showed some models to predict potency for carcinogenicity
and mutagenicity using simple descriptors based on SMILES format.
Examples of QSAR Methods Based on Data-mining
for Carcinogenicity
We will now present some data-mining studies in more detail. Martin
and coworkers developed several different QSAR methodologies for acute
aquatic toxicity in order to model large, noncongeneric data sets (70). The
methodologies include the Hierarchical clustering method, the FDA MDL
QSAR method, the single-model method, and the nearest-neighbor method.
These methods were shown to yield excellent prediction results (70). The
hierarchical clustering approach uses Ward's method (71) to divide an ex-
perimental toxicity training set into a series of structurally similar clus-
ters where each cluster is assumed to represent a common mode-of-action.
The structural similarity is defined in terms of 2-D and 3-D descrip-
tors. A genetic algorithm-based technique is used to generate statistically
valid QSAR models for each cluster. The toxicity for a given query com-
pound is estimated using the average of the predictions from the clus-
ter models whose chemicals are structurally the most similar to the query
compound.
The FDA MDL QSAR method is a variation of the clustering methodol-
ogy of Contrera and coworkers (72). In this method, the prediction for each
test chemical is made using a unique model that is fit to the chemicals from
the entire training set that are the most similar to the test compound. In the
single-model method, a multilinear regression (MLR) model is fit to the entire
data set using molecular descriptors as independent variables. In the nearest-
neighbor method, the predicted toxicity is simply the average of the chemicals
in the training set that are most similar to the test chemical.
The predictive ability of the QSAR methods developed by Martin and
coworkers was evaluated using several different carcinogenicity datasets.
First, the methods were evaluated using a small congeneric set of aromatic
amines. Franke and coworkers reported that they were able to develop excel-
lent correlations for this data set using multilinear regression models (73). It
was shown that cross validation might overestimate the predictive ability of
regression models if it consists of only refitting the model coefficients to the
training sets for the different cross-validation folds. It is suggested that one
could obtain a conservative estimate of the potential prediction accuracy of
multilinear regression methods by using a genetic algorithm to fit a new mul-
tilinear model to the training set for each cross validation fold. The different
QSAR methods achieved prediction concordances of 60-66% (averaged over the
different sex-species sets) for the aromatic amines data sets.
Previous I TOC
-------
Predictive Models for Carcinogenicity and Mutagenicity 75
Next, the predictive ability of the QSAR methods was evaluated using the
larger data sets contained in the Carcinogenic Potency Database (CPDB) (47).
Each sex-species dataset contained ~600-750 noncongeneric chemicals. The
fraction of carcinogenic compounds for each data set were 47%, 45%, 43%, and
44% for the male rat, female rat, male mouse, and female mouse sex-species
datasets, respectively. The QSAR methods of Martin and coworkers achieved
prediction concordances of about 61-63%, sensitivities of 48-60%, and speci-
ficities of 65-73% (averaged over the different sex-species sets) from 5-fold
cross-validations. The results achieved for the CPDB were similar to those
achieved by QSAR-based approaches in the two NTP training exercises (74).
It has been suggested that in order to successfully model large, noncon-
generic carcinogenicity data sets, one should develop a series of more focussed
QSAR models (75). To test this strategy the predictive ability of class-specific
models was compared with the predictive ability of the noncongeneric QSAR
methods described above. The training set for each of the class-specific models
consisted of only the chemicals in that particular class while for the noncon-
generic QSAR methods the training data for the different classes were pooled
together. The comparison was performed using a data set of 280 chemicals
taken from the NTP database (75). Benigni and Richard assigned the chemi-
cals in the NTP data set to ten different structural classes (e.g., electrophilic
alkylating agents and halogenated aliphatic compounds). The data set was
separated randomly into training (80%) and prediction sets (20%) 10 times.
Sampling was done so that there was an equal number of cancer and non-
cancer scores for each chemical class (for both the training and prediction sets).
The results for the 10 different prediction sets were pooled together. The class-
based models achieved an average prediction concordance of about 58%. The
hierarchical and nearest-neighbor methodologies achieved slightly lower pre-
diction concordances of 55% and 57%, respectively. These results indicate that
it may be possible to correlate noncongeneric datasets without manually di-
viding the datasets into classes (although one could argue that the results are
inconclusive due to the low prediction concordances). The prediction concor-
dances were lower for the NTP data set compared with the CPDB and aromatic
data sets because composite cancer scores were modelled.
INCORPORATING NEW AND TRADITIONAL DATA INTO A NEW
SCENARIO
Carcinogenicity and Mutagenicity Data: New Initiatives to
Improve Access and Utility for Modelling
A number of new initiatives are underway to improve access to exist-
ing public carcinogenicity and mutagenicity data for use in modelling, to
Previous I TOC
-------
76 E. Benfenati et al.
encourage use of less summarized activity classifications, to create linkages
and structure-searchable access to publicly available sources of toxicity data,
and to infuse new types of high-throughput biological test data (i.e., biochemi-
cal, cell-based, etc.) along with chemical structure considerations into the pre-
diction modelling paradigm (76, 77). These various initiatives offer the promise
of moving the current paradigm for toxicity prediction toward one that can be
applied more broadly and confidently to larger swaths of chemical space and
to a greater diversity of toxicity endpoints (78).
Current structure-based SAR models for prediction of chemical carcino-
genicity and mutagenicity rely on a relatively small number of publicly avail-
able data resources in which the data being modelled are typically highly
summarized and aggregated representations of the actual experimental re-
sults (i.e., positive and negative calls) (79). The Berkeley Carcinogenic Potency
Database (CPDB) (47), which includes bioassay results for more than 1500 sub-
stances curated from literature reports, has been commonly employed in this
way for past SAR modelling studies. EPAs DSSTox Database Network (23)
offers elaborated and quality reviewed structure-data file (SD file) represen-
tations of public toxicity data sets, such as the CPDB-A11 Species (CPDBAS)
Summary Tables (48) as well as expanded data linkages and coverage of chem-
ical space for carcinogenicity and mutagenicity. In particular, the most recently
published DSSTox CPDBAS SD file includes a number of new species-specific
summary activity fields, along with a species-specific normalized score for car-
cinogenic potency (TD50) and sex/species-specific tumor incidences (80). To fur-
ther facilitate use of these summary activity fields and associated data within
the CPDBAS data file, these chemical structure-associated data have been de-
posited in the large, publicly available PubChem database (81) as seven "bioas-
says" or PubChem AIDs (PubChem Assay Identifiers); i.e., AIDs for CPDBAS
Mutagenicity, Rat, Mouse, Hamster, Dog-Primates, SingleCellCall, and Mul-
tiCellCall. Separate indexing of component activity classifications within the
PubChem system allows a user to take full advantage of the tools and capa-
bilities within PubChem for "read-across" and SAR clustering in bioassay and
structure space; i.e., allowing for comparisons of CPDB compounds across the
entire PubChem inventory (millions of compounds, hundreds of assays).
The entire DSSTox published data file inventory (> 16,000 records, >8,000
unique compounds) has been deposited/updated within PubChem, enabling a
user to cross-reference between the two systems. A user can now link directly
from PubChem substance and bioassay listings to DSSTox data files, docu-
mentation, and Source chemical data pages where available (e.g., to the CPDB
or National Toxicology Program online chemical data pages). Having estab-
lished this direct correspondence between PubChem and DSSTox CIDs (Com-
pound/structure IDs), the DSSTox Structure-Browser (82) now incorporates a
direct link from the DSSTox structure-search results page to the corresponding
PubChem Compound (CID) summary results page, allowing less experienced
Previous I TOC
-------
Predictive Models for Carcinogenicity and Mutagenicity J J
users (e.g., toxicologists, risk assessors) to directly access relevant PubChem
bioassay information, links, and data for similar substances.
The concept of chemical "toxicity profiling" has gained new prominence
and importance in the field of toxicity prediction and recently has been rec-
ommended as a long-term goal for toxicity screening and assessment by a
prominent advisory committee and U.S. government agencies (83, 84). Tox-
icity profiling can occur at two levels: (1) profiles of in vivo responses are
made possible by increasing availability of detailed observational data and ex-
perimental measures associated with chronic toxicity studies, and (2) newer
in vitro high-throughput screening (HTS) data offer a means for broadly char-
acterizing a chemical's biological profile in terms of target interactions, path-
way perturbations, and cellular responses. Several initiatives are currently un-
derway to harness legacy toxicity data from diverse domains of study (cancer
bioassays, genetic toxicity, developmental and reproductive toxicity, skin sen-
sitization, etc.) into hierarchical toxicity data models suitable for building re-
lational databases (85, 86). These historical reference data are necessary to
anchor and validate new predictive toxicology approaches based on alterna-
tive in vitro test methods as well as to eventually move away from current in
vivo rodent test systems to mechanistic pathway-based test data more directly
relevant to humans.
The ToxCast™ project within the U.S. EPA's National Center for Compu-
tational Toxicology is a prominent example of a new predictive toxicology ini-
tiative that is aggressively harnessing legacy in vivo data as well as employing
new HTS in vitro technologies and toxicity profiling concepts (76, 87). As part
of this effort, the ToxRef database has been built to house in vivo toxicity data
in standardized data model representations, and it is being populated with ro-
dent bioassay study data (chronic, developmental, reproductive) for hundreds
of pesticidal active ingredients registered for use in the United States across
a range of toxicity investigation areas. Public release of various representa-
tions of these data, in conjunction with other ToxCast™ data, occurred in late
2008, through venues such as DSSTox and PubChem, and the new EPA AC-
ToR (Aggregated Computational Toxicology Resource) system (88, 89), which
will house all ToxCast™ data as well as supporting publicly available data.
Phase I of the ToxCast™ effort is generating data in hundreds of HTS bio-
chemical and cell-based assays for 320 selected compounds, mostly pesticidal
actives (90), for which a rich profile of toxicity data exists within ToxRef or
other public sources. The goal is to develop candidate predictive signatures for
various toxicity endpoints to undergo further testing and validation in Phase
II.
Microarray data (i.e., information on the changes in gene expression of
many genes) are becoming more available each year, and structure-annotated
data bases containing such information will soon be available for general use.
The complex array of genes that are mutated or whose expression is modulated
Previous I TOC
-------
78 E. Benfenati et al.
is only incompletely understood at this point. However, microarray data, along
with epigenetic and genetic data, will permit a more precise modelling of the
cancer process and improve predictive toxicology in silico.
The generation of large amounts of new HTS and in vitro data, coupled
with enrichment and elaboration of reference in vivo toxicity data in the public
domain, offer new challenges and opportunities to SAR modellers. Although
SAR modelling has had many successes and has been an extremely valuable
tool for toxicity screening and prioritization in the absence of biological data,
the limitations of a structure-only approach to prediction are well known and
enumerated in the literature. In general, chemical class-based approaches that
offer a greater chance of mechanistic coherence can be applied more confidently
to prediction, but in a narrow range of chemical space. In contrast, global SAR
prediction approaches can be applied more broadly, albeit usually with less
confidence. The concepts of chemical similarity and chemical class can be use-
fully employed in the new paradigm to focus investigation on regions of HTS
activity space, exploring differences within the space. By the same token, clus-
ters within HTS activity profile space (such as in a heat map representation)
can be projected onto chemical space, potentially implicating members of mul-
tiple chemical classes and offering new SAR hypotheses for further investiga-
tion. HTS, or "fast biology" results can also potentially be employed as "biolog-
ical descriptors" in a traditional QSAR paradigm; i.e., coupled with traditional
chemical descriptors for SAR/QSAR model construction (91).
Finally, adding layers of richness to data model representations of legacy
in vivo toxicity results, such as in the ToxRef database, yields a great variety of
new activity profile representations, or "endpoints," for use in guiding and an-
choring SAR and HTS predictive toxicology investigations. A smaller region of
chemical space associated with a more focused activity profile, in turn, is likely
to be offset by the greater potential mechanistic coherence of the data and,
therefore, greater potential for modelling and prediction success. The effective
incorporation of SAR concepts into ToxCast™ and similar toxicity modelling
efforts will be crucial for their ultimate success.
Two Case Studies of Data-mining in a New, Broader Scenario
These case studies describe the approaches investigated within regulatory
agencies to go beyond the traditional QSAR paradigm for predictive toxicology
and to include biology more explicitly into the QSAR process.
Case Study 1: Genetic Toxicity Data-mining with Integrated
Database
The predictive data-mining methodology was applied to data from vari-
ous regulatory agencies and industry partners. Some findings from this case
Previous I TOC
-------
Predictive Models for Carcinogenicity and Mutagenicity 79
study were recently published (92) in which the FDA CRADA SAR (Cooper-
ative Research and Development Agreement Structure Activity Relationship)
genetic toxicity database (2006 version) was integrated with proprietary indus-
try data. The proprietary data were shared by Leadscope structural features
statistics without the actual connection tables. The 3220 chemicals in this in-
tegrated database were 30% drugs, 22% food ingredients, and 48% industrial
chemicals, the latter including agricultural chemicals. Various data sources
were integrated according to the ToxML criteria (80, 86). For compounds to be
scored for assessment (e.g., test calls), ToxML requires the data to be accom-
panied by information on test system, including conditions such as controls,
dosage regimen, and cytotoxicity (92). The ToxML data model and data entry
tool (ToxML Editor) are freely and publicly available (93). The genetic toxic-
ity profile of these chemicals was analyzed by structural features across the
various strains of Salmonella typhimurium reverse mutagenesis (Salmonella
mutation), mouse lymphoma mutation, in vitro chromosome aberration (ivt
CA), and in vivo micronucleus (micronucleus).
Structural features associated with point mutations—in particular, base
substitutions or frame shifts—were found. For example, alkyl halides are
highly correlated with mutagenicity in all Salmonella strains, whereas aryl
halides are not. Well-known groups such as epoxides, nitro, nitroso, and
quinines are highly associated with all four genetic toxicity endpoints. On
the other hand, structural features such as azo, benzimidazole, and quinolines
are correlated with mutagenicity but not with clastogenicity. Features such as
alkenyl ketones, aryl aldehyde, pyrazine (H), and base nucleosides are associ-
ated only with clastogenicity. When the four genetic toxicity outcomes are cor-
related using the structural features, two mutagenicity tests using Salmonella
and mouse lymphoma correlated well. Salmonella mutagenesis outcome also
correlated well with that of in vitro chromosome aberrations. However, in vivo
micronucleus did not correlate well with in vitro chromosome aberrations. If a
chemical is ivt CA negative, it will probably be micronucleus negative. It is im-
portant to note that these genetic toxicity screening tests should be used more
as a profile rather than as an individual predictor for carcinogenicity. These
structural features can be further refined to form structural alerts grouped by
chemical reactivity.
Structural alerts representing positive carcinogenicity/negative genotox-
icity were extracted from the data set. Many structural alerts for genotoxic
carcinogens for general industrial chemicals were consistent with literature
results (94). The landscape of these alerts changed significantly when the
structure space changed from industrial chemicals toward drugs. Several new
structural alerts representing non-genotoxic carcinogens were presented. One
of the structural alerts included the statin analogs.
To further understand the biology of the statins, various target or-
gan lesions from the chronic studies were compared with the SAR-ready
Previous I TOC
-------
80 E. Benfenati et al.
carcinogenicity database (45). In chronic studies, liver lesions included cen-
trilobular necrosis, hypertrophy, and vacuolar degeneration of perilobular hep-
atocytes, cellular atypia, fatty change, and bile duct hyperplasia. Liver organ
weight increases and ALT/AST enzyme level increases were also noticed dur-
ing the chronic studies. These statin analogs also showed thyroid lesions, thy-
roid organ weight increase, and CPK enzyme increase. In the carcinogenicity
database, adenoma, carcinoma, and fibrosarcoma of liver were observed as well
as thyroid follicular cell adenoma. The increased incidence of liver and thyroid
tumors is connected by a well-known mechanism of thyroid-stimulating hor-
mone instigating the liver microsomal enzymes (95). The same connection was
also reported previously from data-mining the CPDB database (96). As pre-
sented in this case, understanding non-genotoxic carcinogens requires under-
standing of biological mechanisms involved in target organ effects. Therefore,
a quality database providing in-depth target organ findings in chronic studies
along with the carcinogenicity and genetic toxicity data can be vitally impor-
tant to further our knowledge of carcinogens.
Case Study 2: NTP High throughout Screening and
Understanding Genotoxicity and Carcinogenicity
One of the questions that can be posed of a database of genetic toxic-
ity screening tests is whether the current genetic toxicity tests sufficiently
reflect the different mechanisms involved in carcinogenesis. In this regard,
the HTS campaign initiated by the U.S. National Toxicology Program (NTP)
is worth discussing. The objectives of that project are to develop methods to
screen and prioritize the nomination of chemicals for rodent bioassays and
to look for approaches to gain insights on mode-of-actions for various toxi-
city endpoints. Potentially, some of these data and methods can lead to im-
provements in predictive toxicology. NTP initially selected 1408 chemicals and
tested against 24 bioassays. The chemicals included food ingredients (13%),
agricultural chemicals (17%), drugs and hormones (20%), and general indus-
trial chemicals (50%). The 24 bioassays included caspase and kinase activi-
ties of NCGC cell lines (3T3, BJ, SHSY5, H4IIe, Hek293, HepG2, HUVEC,
Jurkat, N2A,,IkB signalling protein, JNK Alpha), and SKNSH, MRC5, Renal,
and Mesenchymal assays. The bioassay panel also includes 7 FRED and 13
NCGC strains for cell viability (97).
When compounds were clustered against the bioassays and cell viabilities,
most of the compounds were not differentiated by activities; however, some
blocks of chemicals did separate based on cell viabilities. The differentiation
of cell viabilities and activities against compounds increased markedly when
the observations were based on smaller units of structural features (i.e., frag-
ments of molecules) rather than individual compounds. Using cell viability
to probe for rodent acute toxicity, biological and chemical fingerprints were
Previous I TOC
-------
Predictive Models for Carcinogenicity and Mutagenicity
investigated. The bioassay profile also has been compared with the genotox-
icity and carcinogenicity endpoints. Of the 1408 chemicals, 543 had rodent
carcinogenicity data, 1112 had Salmonella data, 344 had mammalian muta-
genesis data, 428 had ivt CA data, and 223 had micronucleus data within
the database sources mentioned above. At the compound level, there were too
many missing values to permit meaningful correlations between the bioassays
and toxicity data (i.e., too few cases where the same compound had all types of
assay data across the data sources).
When the observations were based on structural features and statistics
were recalculated, the trends reported previously among the four genetic tox-
icity tests in the first data-mining case study were again observed. Struc-
tural features correlating either positively or negatively across the different
endpoints were selected to recalculate the statistics. Based on the 1408 data
set, Salmonella mutagenesis (R = 0.52) and in vitro chromosome aberra-
tions (R = 0.39) showed some correlations to rodent carcinogenicity, in which
rodent carcinogenicity was defined as induction of tumors in both rat and
mouse.
From this method, structural features positive for both rodent carcino-
genicity and genotoxicity included aromatic amines, azo, epoxides, halides,
and nitroso/nitrosamine. Pyridine (H) was found to correlate more with non-
genotoxic carcinogens. Genotoxic but not carcinogenic features included alkyl
nitro, benzimidazole, quinoline, and 1,4-diamino benzene features. These re-
sults are quite consistent with earlier reports on the classes of rodent carcino-
genicity (98, 99). Using the feature analyses, a compound-class-driven cate-
gorization of the correlations between bioassays and toxicity endpoints was
conducted by 2-D clustering of correlations between the various toxicity and
bioassays. A heat-map generated from these data was used to tease out par-
ticular biological assays correlating highly with rodent carcinogenicity for a
specific set of compound classes.
To summarize, a battery of genotoxicity screening tests can be used for
profiling compounds to understand carcinogenicity potential. Structural alerts
with genotoxic-carcinogenic outcome probabilities stratified by potency can be
developed based on the feature-based methodology. The current NCGC bioas-
says from the first NTP HTS campaign may not correlate well with genotoxic-
ity and carcinogenicity at a global compound level. However, toxicity of a par-
ticular class of chemicals correlates relatively well with some of the biological
assays. In addition, NTP has expanded their screen to include other cell-based
and activity assays.
The importance of the two case studies presented above is not to emphasize
an efficacy of one particular test or assay, but rather to approach the genetic
toxicity and carcinogenicity problem in a new way by linking chemical struc-
tures to biological effects with the introduction of new molecular-level assays
to yield potential mechanistic insights. In the long run, predictive data-mining
Previous I TOC
-------
82 E. Benfenati et al.
methods can help in revisiting data, testing strategies, and past presumptions
involved in risk assessment.
CONCLUSIONS
In QSAR models for carcinogenicity (mainly) and mutagenicity, there are still
a number of open problems and unresolved issues. Most of these involve the
point of application (e.g., screening vs. late-stage) and interpretation of soft-
ware output. "False positives" generated by programs such as DEREK or
MC4PC are greatly reduced by applying expert knowledge that takes into ac-
count the chemical context of the alert or biophore, and whether hydrolysis or
metabolism are likely to convert the molecule to a true alert as defined origi-
nally and confirmed in the literature. The "false negative" designation, on the
other hand, suggests that the learning sets on which the prediction systems
were based initially may lack some essential knowledge. Studies using both
a cell-based system and 3-D DNA docking/electrostatic modelling have shown
that many "false negatives" can be explained by non-covalent binding (e.g.,
DNA intercalation or groove binding), and that the genotoxicity of such inter-
actions is largely the result of topoisomerase II inhibition. This type of chemi-
cal/DNA interaction was and still is poorly understood and, consequently, may
be trained inappropriately in the learning sets of the most commonly used in
silico programs. Moreover, because most of these putative intercalating agents
do not possess classical, planar intercalating structures, simple visual inspec-
tion does not allow prediction of non-covalent binding to DNA.
In order to significantly improve in silico models for carcinogenicity and
mutagenicity, it is crucial to understand and accept that there are still prob-
lems with the experimental methods dedicated to study these endpoints. Thus,
several of the problems that appear as (Q)SAR problems are actually typical
of the general limitations of the current experimental techniques and state of
the knowledge. (Q)SAR models, for their part, are more suitable to statistical
treatment of the data, which highlight their accuracy, sensitivity, specificity,
reliability, and predictivity. Different tools are useful, depending on whether
the model is a classifier (SAR) or a regression model.
In the case of the in vitro models designed to replace in vivo methods, it is
more and more common to have a statistical evaluation for the false positives
and negatives of the method. However, similar objective appraisal is missing
in animal models when compared with human toxicity. The extrapolation from
animal models to humans is not an easy task given the paucity of data on the
latter. In addition, the variability of the in vivo data is poorly described.
These intrinsic scientific problems are complicated by different purposes
and intended uses of the models. Models for regulations are strictly linked
to legal specifications that depend on the specific regulation. Even within the
Previous I TOC
-------
Predictive Models for Carcinogenicity and Mutagenicity 83
same regulation, different possible uses of the modelling tools are possible. For
example, within the REACH legislation, carcinogenicity has to be described as
a category for prioritization and classification and labelling. However, contin-
uous quantitative estimates of potency are also needed to estimate the risk of
the carcinogen within a given scenario of exposure.
There are several possible uses of the in silico models, such as prioritisa-
tion, screening, mechanistic studies, and support for risk assessment. Today
the acceptance of in silico tools and predictive models, some based on incorpo-
ration of newer HTS data, for carcinogenicity as alternative methods to in vitro
or in vivo testing, is still highly debated, and the varied discussions at the
workshop demonstrated this. One reason for this is the lack of knowledge and
experience produced on the proposed alternative methods. It was mentioned
that even the Ames test, when originally proposed, was severely criticized. De-
spite this, after decades, mutagenicity is now a requested endpoint in several
regulations.
It may happen that a scientist strongly supports and advocates for use of
a particular method with which he or she is comfortable. However, the techni-
cal, theoretical differences between the different models should have a lower
emphasis compared with the advantages for the user and practical utility of
models for a certain application. As Galileo said, hypotheses have to be exper-
imentally evaluated. In the case of in silico models, they have to be proven
to work for their intended use. There are at least two possible applications:
in silico models can classify potential carcinogens or mutagens or they can
predict a potency from a continuous model, with an estimated confidence in-
terval, namely the error range. Depending on the errors in the quantitative
predictions or classifications, the method can be used as a screening tool or as
a substitute to in vitro or in vivo testing if the error is acceptable. In the case
of models for regulatory purposes, where conservative measures for ensuring
public safety is of primary concern, the errors that give rise to false negatives
are much more relevant and high specificity is thus critical. The measure of
sensitivity and specificity is much more important than that of concordance.
This holds true both for classification and continuous QSAR models. The pre-
dicted residuals are more appropriate than just R2.
To cope with the scientific problems of a better understanding and predic-
tion of carcinogenicity and mutagenicity, new efforts have to be planned and
organized, integrating different tools, not only in silico. We discussed the new
possibilities of a large screening of chemical substances and some ongoing ini-
tiatives. The challenge is to reinforce and expand knowledge on toxicity phe-
nomena by introducing and using new experimental data that are more easily
generated than with the classical in vivo methods. The use of these data is
changing the perspective of toxicity evaluation. The huge amount of data will
offer new ways to explore relationships between data of different origin as a
contribution to the understanding of toxicity. In this evolving scenario there
Previous I TOC
-------
84 E. Benfenati et al.
will be a need for powerful data-mining tools that are capable of extracting
knowledge from a complex multidimensional space. New initiatives requiring
a paradigm shift are increasing, and the hope is that a better understanding of
toxicity phenomena will be achieved for ensuring safer chemicals and stream-
lined, yet protective regulatory procedures.
ACKNOWLEDGEMENTS
We acknowledge the financial support of the European Commission, project
SCARLET, contract no. SP5A-CT-2007-044166. This manuscript has been re-
viewed by the National Health and Environmental Effects Research Labora-
tory and National Center for Computational Toxicology, U.S. Environmental
Protection Agency, and is approved for publication. Approval does not signify
that the contents reflect the views of the Agency, nor does mention of trade
names or commercial products constitute endorsement or recommendation for
use.
REFERENCES
1. StruCture-Activity Relationships Leading ExperTs in mutagenicity and carcino-
genicity (SCARLET), European Commission Project SPSA-CT-2007-014166. Available
at: http://scarlet-project.eu
2. Moore, MM, Honma, M, Clements, J, Bolcsfoldi, G, Burlinson, B, Cifone, M,
Clarke, J, Delongchamp, R, Durward, R, Fellows, M, Gollapudi, B, Hou, S, Jenkinson, P,
Lloyd, M, Majeska, J, Myhr, B, O'Donovan, M, Omori, T, Riach, C, San, R, Stankowski,
LF Jr, Thakur, A, Van Goethem, F, Wakuri, S, Yoshimura, I. Mouse Lymphoma Thymi-
dine Kinase Gene Mutation Assay: Follow-up Meeting of the International Workshop
on Genotoxicity Tests-Aberdeen, Scotland, 2003—Assay acceptance criteria, positive
controls, and data evaluation. Environ Mol Mutagen 2006;47:1—5.
3. Kirkland, D, Pfuhler, S, Tweats, D, Aardema, M, Corvi, R, Darroudi, F,
Elhajouji, A, Glatt, H, Hastwell, P, Hayashi, M, Kasper, P, Kirchner, S, Lynch, A,
Marzin, D, Maurici, D, Meunier, J-R, Miiller, L, Nohynek, G, Parry, J, Parry, E, Thybaud,
V, Tice, R, van Benthem, J, Vanparys, P, White, P. How to reduce false positive results
when undertaking in vitro genotoxicity testing and thus avoid unnecessary follow-up
animal tests: Report of an ECVAM Workshop. Mutation Research 2007;628:31-55.
4. Snyder, RD, Ewing, DE, Hendry, LB. Evaluation of DNA intercalation potential
of pharmaceuticals and other chemicals by cell-based and three-dimensional computa-
tional approaches. Environ Mol Mutagen 2004;44:163-173.
5. Snyder, RD, Ewing, D, Hendry, L. DNA intercalative potential of marketed drugs
testing positive in in vitro cytogenetics assays. Mutation Res 2006;609:47-59.
6. van der Lelie, D, Regniers, L, Borremans, B, Provoost, A, Verschaeve, L. The
VITOTOX test, an SOS bioluminescence Salmonella typhimurium test to measure
genotoxicity kinetics. Mutagenesis 1997;20:449—454.
7. Verschaeve, L, Van Compel, J, Thilemans, L, Regniers, L, Vanparys, P, van der
Lelie, D. VITOTOX® bacterial genotoxicity and toxicity test for the rapid screening of
chemicals. Environ Mol Mutagen 1999;33:240-248.
Previous I TOC
-------
Predictive Models for Carcinogenicity and Mutagenicity 85
8. Van Compel, J, Woestenborghs, F, Beerens, D, Mackie, C, Cahill, PA, Knight,
AW, Billinton, N, Tweats, DJ, Walmsley, RM. An assessment of the utility of the yeast
GreenScreen assay in pharmaceutical screening. Mutagenesis 2005;20:449^154.
9. Pastink, A, Eeken, JCJ, Lohman, PHM. Genomic integrity and the repair of
double-strand DNA breaks. Mut Res 2001;480-481:37-50.
10. Clever, B, Interthal, H, Schmuckli-Maurer, J, King, J, Sigrist, M, Heyer, W-F. Re-
combinational repair in yeast: functional interactions between RadSl and Rad54 pro-
teins. EMBO J 1997;16:2535-2544.
11. Sonoda, E, Sasaki, MS, Buerstedde, J-M, Bezzubova, O, Shinohara, A, Ogawa, H,
Takata, M, Yamaguchi-Iwai, Y, Takeda, S. RadSl-deficient vertebrate cells accumulate
chromosomal breaks prior to cell death. EMBO J 1998;17:598-608.
12. Arbel, A, Zenvirth, D, Simchen, G. Sister chromatid-based DNA repair is medi-
ated by RAD54, not by DMC1 or TID1. EMBO J 1999;18:2648-2658.
13. Dronkert, MLG, Beverloo, HB, Johnson, RD, Hoeijmakers, JHJ, Jasin, M,
Kanaar, R. Mouse RAD54 affects DNA Double-strand break repair and sister chromatid
exchange. Mol Cell Biol 2000;20:3147-3156.
14. Van Compel, J, Woestenborghs, F, Beerens, D, Mackie, C, Cahill, PA, Knight,
AW, Billinton, N, Tweats, DJ, Walmsley, RM. An assessment of the utility of the
yeast GreenScreen assay in pharmaceutical screening. Mutagenesis 2005;20:449—
454.
15. Billinton, N, Barker, MG, Michel, CE, Knight, AW, Heyer, W-D, Goddard, NJ,
Fielden, PR, Walmsley, RM. Development of a green fluorescent protein reporter for a
yeast genotoxicity biosensor. Biosensores Bioelectronics 1998;13:831-838.
16. Cahill, PA, Knight, AW, Billington, N, Barker, MG, Walsh, L, Keenan, PO,
Williams, CV, Tweats, DJ, Walmsley, RM. The GreenScreen genotoxicity assays: a
screening validation programme. Mutagenesis 2004;19:105—119.
17. Knight, AW, Billinton, N, Cahill, PA, Scott, A, Harvey, JS, Roberts, KJ, Tweats,
DJ, Keenan, PO, Walmsley, RM. An analysis of results from 305 compounds tested with
the yeast RAD54-GFP genotoxicity assay (GreenScreen GC)—including relative predic-
tivity of regulatory tests and rodent carcinogenesis and performance with autofluores-
cent and coloured compounds. Mutagenesis 2007;22:409^il6.
18. Cole, GM, Schild, D, Lovett, ST, Mortimer, RK. Regulation of RAD54- and
RAD52-lacZ gene fusions in Saccharomyces cerevisiae in response to DNA damage.
Mol Cell Biol 1987;7:1078-1084.
19. Averbeck, D, Averbeck, S. Induction of the genes RAD54 and RNR2 by various
damaging agents in Saccharomyces cerevisiae. Mut Res 1994;315:123-138.
20. Joosten, HFP, Acker, FAA, Dobbelsteen van den DJ, Horbach GJMJ, Krajnc, EL
Genotoxicity of hormonal steroids. Toxicol Lett 2004;151:113-134.
21. International Agency for Research on Cancer (IARC). Available at: http://
www.iarc.fr/ENG/Databases/index.php.
22. Zhao, C, Boriani, E, Ghana, A, Roncaglkioni, A, Benfenati, E. A new hybrid
system of QSAR models for predicting bioconcentration factor (BCF). Chemosphere
2008:73:1701-1707.
23. EPA DSSTox. U.S. Environmental Protection Agency's Distributed Structure-
Searchable Toxicity (DSSTox) Database Network. 2009. Available at: http://www.
epa.gov/ncct/dsstox/
24. Benigni, R, Bossa, C, Jeliazkova, NG, Netzeva, TI, Worth, AP. The Benigni/Bossa
rulebase for mutagenicity and carcinogenicity—a module of Toxtree. Report EUR 23241
Previous I TOC
-------
86 E. Benfenati et al.
EN. 2008. Luxembourg, Office for the Official Publications of the European Communi-
ties. EUR—Scientific and Technical Report Series.
25. Benigni, R, Bossa, C, Netzeva, TI, Worth, AP. Collection and evaluation of
(Q)SAR models for mutagenicity and carcinogenicity. EUR 22772 EN. Luxembourg, Of-
fice for the Official Publications of the European Communities. EUR—Scientific and
Technical Research Series; 2007.
26. Benigni, R, Bossa, C. Predictivity and reliability of QSAR models: the case of
mutagens and carcinogens. Toxicol Mechanisms Meth 2008;18:137-47.
27. Benigni, R, Netzeva, TI, Benfenati, E, Bossa, C, Rainer, R, Helma, C, Hulzebos, E,
Marchant, C, Richard, A, Woo, Y-T, Yang, C. The expanding role of predictive toxicology:
an update on the (Q)SAR models for mutagens and carcinogens. J Environ Sci Health
C Environ Carcinog Ecotoxicol Revs 2007;25:53-97.
28. European Partnership for Alternative Approaches to animal testing (EPAA).
Available at: http://ec.europa.eu/enterprise/epaa/brochure.htm
29. Brown, D, Superti-Furga, G. Rediscovering the sweet spot in drug discovery. Drug
Discovery Today 2003;8:1067-1077.
30. DiMasi, JA. Risks in new drug development: Approval success rates for investi-
gational drugs. Clin Pharmacol Ther 2001;69:297-307.
31. Kola, I, Landis, J. Can the pharmaceutical industry reduce attrition rates? Na-
ture Reviews Drug Discovery 2004;3:711-715.
32. Fingerhut, MA, Halperin, WE, Marlow, DA, Piacitelli, LA, Honchar, PA, Sweeney,
MH, Greife, AL, Dill, PA, Steenland, K, Suruda, AJ. Cancer mortality in work-
ers exposed to 2,3,7,8-tetrachlorodibenzo-p-dioxin. N Engl J Med 1991;324:212-
218.
33. Manz, A, Berger, J, Dwyer, JH, Flesch-Janys, D, Nagel, S, Waltsgott, H. Can-
cer mortality among workers in chemical plant contaminated with dioxin. Lancet
1991;338:959-964.
34. Zober, A, Messerer, P, Huber, P. Thirty-four-year mortality follow-up of BASF
employees exposed to 2,3,7,8-TCDD after the 1953 accident. Int Arch Occup Environ
Health 1990; 62:139-157.
35. Westerink, WMA, Stevenson, JCR, Schoonen, WGEJ. Pharmacologic profiling of
human and rat cytochrome P450 1A1 and 1A2 induction and competition. Arch Toxicol
2008;82:909-921. Doi 10.1007/S00204-008-0317-7.
36. Snyder, RD, Smith, MD. Computational prediction of genotoxicity: room for im-
provement. Drug Discovery Today 2005;10:1119-1124.
37. Ertl, P, Roggo, S, Schuffenhauer, A. Natural Product-likeness score and its appli-
cation for prioritization of compound libraries. J Chem Inf Model 2008;48:68-74.
38. Contrera, JF, Jacobs, AC, DeGeorge, JJ. Carcinogenicity testing and the eval-
uation of regulatory requirements for pharmaceuticals. Regul Toxicol Pharmacol
1997;25:130-145.
39. CCRIS (Chemical Carcinogenesis Research Information System), developed and
maintained by National Cancer Institute. Available at: http://toxnet.nlm.nih.gov/cgi-
bin/sis/htmlgen?CCRIS
40. Gene-Tox: Created by US, EPA. Available at: http://toxnet.nlm.nih.gov/cgi-
bin/sis/htmlgen?GENETOX
41. National Toxicology Program (NTP). Available at: http://ntp.niehs.nih.gov
Previous I TOC
-------
Predictive Models for Carcinogenicity and Mutagenicity 87
42. International Programme on Chemical Safety (IPCS). INCHEM, a product co-
operation between the International Programme on Chemical Safety (IPCS) and
the Canadian Centre for Occupational Health and Safety (CCOHS). Available at:
http ://www.inchem .org/
43. IRIS EPA. Integrated risk information system for carcinogen classification, orig-
inally prepared in US EPA. Available at: http://www.epa.gov/iris/index.htm
44. FDA Cooperative Research and Development Agreemen) database. Leadscope.
Available at: http://www.leadscope.com/fda_databases/
45. Tokyo-Eiken, Tokyo Metropolitan Institute of Public Health providing primary
mutagenicity of food additives for about 300 chemicals. Available at: http://www.tokyo-
eiken.go.jp/henigen/
46. The Mutants, a database of mutagenicity/genetic toxicity database spon-
sored by Dr. Motoi Ishidate. Available at: http://members.jcom.home.ne.jp/mo-
ishidate/index.html
47. CPDB Summary Tables. Summary Table of Chemicals in the Carcinogenic Po-
tency Database: Results for Positivity, Potency (TD50), and Target Sites. 2007. Avail-
able at: http://potency.berkeley.edu/chemicalsummary.html
48. Gold, LS, Slone, TH, Williams, CR, Burch, JM, Stewart, TW, Swank, AE, Bei-
dler, J, Richard, AM. DSSTox Carcinogenic Potency Database Summary Tables—All
Species, SDF Files and Documentation, CPDBAS_v5c_1547_20NOV2008. 2008. Avail-
able at: http://www.epa.gov/ncct/dsstox/sdLcpdbas.html
49. Benigni, R, Bossa, C, Richard, AM, Yang, C. A novel approach: chemical rela-
tional databases, and the role of the ISSCAN database on assessing chemical carcino-
genicity. Ann 1st Super Sanita 2008; 44:48-56.
50. Leadscope Inc., Ohio, USA. http://www.leadscope.com
51. Lhasa Limited, Leeds, UK. http://www.lhasalimited.org/
52. RTECS, Registry of toxic effects of chemical substances, created by NIOSH,
maintained and marketed by MDL Symix. Available at: http://mdl.com/products/
predictive/rtecs/index.jsp
53. MultiCASE MC4PC. Available at: http://www.multicase.com/products/prod01.htm
54. TOPKAT. Available at: http://accelrys.com/products/discovery-studio/toxicology/
55. MDL QSAR. Available at: http://www.mdl.com/products/predictive/qsar/index.jsp
56. Leadscope FDA Model Applier. Available at: http://www.leadscope.com/product.
info.php?productsJd=66
57. OncoLogic. A computer system to evaluate the carcinogenic potential of chemi-
cals. Available at: http://www.epa.gov/oppt/newchems/tools/oncologic.htm
58. LAZAR (Lazy Structure Activity Relationship). Available at: http://www.
predictive-toxicology.org/lazar/
59. Toxtree. Available at: http://ecb.jrc.it/qsar/
60. Benigni, R. Structure-activity relationship studies of chemical mutagens and
carcinogens: mechanistic investigations and prediction approaches. Chem Revs
2005;105:1767-1800.
61. Sello, G, Sala, L, Benfenati, E. Predicting toxicity: A mechanism of action
model of chemical mutagenicity. Mutat Res Fundam Mol Mech Mutag 2001;479:141-
171.
Previous I TOC
-------
88 E. Benfenati et al.
62. Fung, VA, Huff, J, Weisburger, EK, Hoel, DG. Predictive strategies for select-
ing 379 NCI/NTP chemicals evaluated for carcinogenic potential: scientific and public
health impact. Fund Appl Toxicol 1993;20:413-36.
63. Benigni, R. Structure-activity relationship studies of chemical mutagens and
carcinogens: mechanistic investigations and prediction approaches. Chem Revs
2005;105:1767-800.
64. Benigni, R, Bossa, C. Predictivity of QSAR. J Chem Inf Model. 2008;48:971-980.
65. Piegorsch, WW, Zeiger, E. Measuring intra-assay agreement for the Ames
Salmonella assay. In: Rienhoff, O, Lindberg DAB (Eds.), Lecture Notes in Medical In-
formation (Statistical Methods in Toxicology), Springer-Verlag, Berlin; 1991: 35-41.
66. OECD. The Report from the Expert Group on (Quantitative) Structure Activity
Relationship ([Q]SARs) on the Principles for the Validation of (Q)SARs. 49Paris, OECD.
OECD Series on Testing and Assessment; 2004.
67. Katzius, J, McGuire, R, Bursi, R. Derivation and validation of toxicophores for
mutagenicity prediction. J Med Chem 2005;48:312-320.
68. Ferrari, T, Gini, G. A new Predictive Model of Mutagenicity, with statisti-
cal analysis and validation using data-mining tools in WEKA. Poster presented at
SCARLET workshop, April 2-4, 2008, Milan, Italy. Available at: http://www.scarlet-
project.eu/posters/FerrarLT-scarlet.pdf
69. Toropov, AA, Toropova, AP, Benfenati, E. QSAR modelling of carcinogenicity
and mutagenic potentials by optimal SMILES-based descriptors. Poster presented at
SCARLET workshop, April 2-4, 2008, Milan, Italy. Available at: http://www.scarlet-
project.eu/posters/Toropov _A-scarlet.pdf
70. Martin, TM, Harten, P, Venkatapathy, R, Das, S, Young, DM. A hierarchical clus-
tering methodology for the estimation of toxicity. Toxicol Mech Method 2008;18:251-
266.
71. Romesburg, HC. Cluster Analysis for Researchers. Belmont, CA: Lifetime Learn-
ing Publications; 1984.
72. Contrera, JF, Matthews, EJ, Benz, RD. Predicting the carcinogenic potential of
Pharmaceuticals in rodents using molecular structural similarity and E-state indices.
Regulat Toxicol Pharmacol 2003;38:243-259.
73. Franke, R, Gruska, A, Giuliani, A, Benigni, R. Prediction of rodent carcinogenic-
ity of aromatic amines: a quantitative structure-activity relationships model. Carcino-
genesis 2001;22 (9):1561-1571.
74. Bristol, DW, Wachsman, JT, Greenwell, A. The NIEHS Predictive-Toxicology
Evaluation Project. Environ Health Persp 1996;104(Supplement 5):1001-1010.
75. Benigni, R, Richard, AM. QSARS of mutagens and carcinogens: Two case studies
illustrating problems in the construction of models for noncongeneric chemicals. Mutat
Res 1996;371:29-46.
76. Dix, DJ, Houck, KA, Martin, TM, Richard, AM, Setzer, W, Kavlock, RJ. The
ToxCast™ program for prioritizing toxicity testing of environmental chemicals. Tox
Sci 2007;95:5-12.
77. Richard, AM, Yang, C, Judson, R. Toxicity Data Informatics: Supporting a New
Paradigm for Toxicity Prediction. Tox Mech Meth 2008;18:103-118.
78. Richard, AM. Future of Predictive Toxicology: An Expanded View of "Chem-
ical Toxicity"—Future of Toxicology Perspective. Chem Res Toxicol 2006;19:1257-
1262.
Previous I TOC
-------
Predictive Models for Carcinogenicity and Mutagenicity 89
79. Richard, AM, Williams, CR. Public sources of mutagenicity and carcinogenicity
data: Use in structure-activity relationship models. In: Benigni R (Ed.), QSARs of Mu-
tagens and Carcinogens. New York: CRC Press; 2002:145-173.
80. Richard, AM, Yang, C, Judson, R. Toxicity Data Informatics: Supporting a New
Paradigm for Toxicity Prediction. Tox Mech Method 2008;18:103-118.
81. NCBI PubChem. National Institutes of Health, National Library of Medicine,
National Center for Biotechnology Information, PubChem Project. 2007. Available at:
http://pubchem.ncbi.nlm.nih.gov/
82. Transue, T, Richard, AM. U.S. Environmental Protection Agency
DSSTox Structure-Browser v2.0. 2008. Available at: http://www.epa.gov/dsstox-
structurebrowser/
83. National Research Council (NRC). Toxicity Testing in the 21st Century: A Vision
and a Strategy. Washington, DC: National Academies Press; 2007.
84. Collins, FS, Gray, GM, Bucher, JR. Transforming Environmental Health Protec-
tion. Science 2008;319: 906-907.
85. Martin, TM, Houck, KA, McLaurin, K, Richard, AM, Dix, DJ. Linking Regu-
latory Toxicological Information on Environmental Chemicals with High-Throughput
Screening (HTS) and Genomic Data. The Toxicologist CD. J Soc Toxicol 2007;96:219-
220.
86. Yang, C, Benz, RD, Cheeseman, MA. Landscape of current toxicity databases and
database standards. Curr Opin Drug Discovery Develop 2006;9:124—133.
87. Houck, K, Dix, D, Judson, R, Martin, M, Wolf, M, Kavlock, R, Richard, AM.
DSSTox EPA ToxCast High Throughput Screening Testing Chemicals Structure-Index
File: SDF File and Documentation: TOXCST.v3a.320.12FEB2009. 2008. Available at :
http://www.epa.gov/ncct/dsstox/sdLtoxcst.html.
88. Judson, R, Richard, AM, Dix, D, Elloumi, F, Martin, M, Cathey, T, Transue, T,
Spencer, R, Wolf, M. ACToR—Aggregated Computational Toxicology Resource. Toxicol
Appl Pharmacol 2008;233:7-13.
89. ACToR. Aggregated Computational Toxicology Resource. Available at: http://
actor.epa.gov/actor/
90. EPA ToxCast. U.S. Environmental Protection Agency's National Center for Com-
putational Toxicology ToxCastTM Program. 2009. Available at: http://www.epa.gov/
comptox/toxcast/
91. Zhu, H, Rusyn, I, Richard, AM, Tropsha, A. The Use of Cell Viability Assay
Data Improves the Prediction Accuracy of Conventional Quantitative Structure Activ-
ity Relationship Models of Animal Carcinogenicity. Enviro Health Persp 2008;116:506-
513.
92. Yang, C, Arnby, CH, Arvidson, K, Aveston, S, Benigni, R, Benz, RD, Boyer, S,
Contrera, J, Dierkes, P, Han, X, Jaworska, J, Kemper, RA, Kruhlak, NL, Matthews, EJ,
Rathman, JF, Richard, AM. Understanding Genetic Toxicity through Data-mining: The
Process of Building Knowledge by Integrating Multiple Genetic Toxicity Databases. Tox
Mech Method 2008;18:277-295.
93. Leadscope Inc. ToxML editor. Ohio, USA. http://www.leadscope.com/toxmLeditor.
94. Ashby, J. Fundamental structural alerts to potential carcinogenicity or non-
carcinogenicity. Environ Mutagen 1985;7:919-921.
95. Klassen, CD. Casarett & DoulPs Toxicology: The basic science of poisons. New
York: McGraw-Hill; 1996.
Previous I TOC
-------
90 E. Benfenati et al.
96. Yang, C, Richard, AM, Cross, KP. The Art of Data-mining the Minefields of Toxic-
ity Databases to Link Chemistry to Biology. Curr Comp-Aided Drug Design 2006;2,135—
150.
97. Xia, M, Huang, R, Witt, KL, Southall, N, Fostel, J, Choi, MH, Jadhav, A, Smith,
CS, Inglese, J, Portier, CJ, Tice, RR, Austin, CP. Compound Cytotoxicity Profiling Us-
ing Quantitative High-Throughput Screening, Environ Health Persp 2008;116-284-291.
Available at: http://www.ehponline.org.
98. Gold, LS, Bernstein, L, MaGaw, R, Slone, TH. Interspecies extrapolation in car-
cinogenesis: Prediction between rats and mice. Environ Health Persp 1989;81:211—219.
99. Benigni, R, Giuliani, A, Franke, R, Gruska, A. Quantitative structure-
activity relationships of mutagenic and carcinogenic aromatic amines. Chem Revs
2000;100:3697-3714.
Previous
TOC
-------
TOXKOLOGICAL SCIENCES 102(1), 15-32 (2008)
doi: 10.1093/toxsci/kfm286
Advance Access publication November 17, 2007
Predicting Maternal Rat and Pup Exposures: How Different are They?
Miyoung Yoon*'tjl and Hugh A. Bartonf2
*National Research Council Research Associateship Program at U.S. Environmental Protection Agency, Research Triangle Park, North Carolina;
^US EPA Human Studies Facility, 104 Mason Farm Road, Chapel Hill, North Carolina 27599; and ^.National Center for Computational Toxicology,
U.S. Environmental Protection Agency, Research Triangle Park, North Carolina 27711
Received September 17, 2007; accepted November 14, 2007
Risk and safety assessments for early life exposures to
environmental chemicals or Pharmaceuticals based on cross-
species extrapolation would greatly benefit from information on
chemical dosimetry in the young. Although relevant toxicity
studies involve exposures during multiple life stages, the mother's
exposure dose is frequently used for extrapolation of rodent
toxicity findings to humans and represents a substantial source of
uncertainty. A compartmental pharmacokinetic model augmented
with biological information on factors changing during lactation
and early postweaning was developed. The model uses adult
pharmacokinetics, milk distribution, and relevant postnatal
biology to predict dosimetry in the young for chemicals. The
model addressed three dosing strategies employed in toxicity
studies (gavage, constant ppm diet, and adjusted ppm diet) and
the impact of different pharmacokinetic properties such as rates
of clearance, milk distribution, and volume of distribution on the
pup exposure doses and internal dosimetry. Developmental delays
in clearance and recirculation of chemical in excreta from the pup
to mother were evaluated. Following comparison with data for
two chemicals, predictions were made for theoretical chemicals
with a range of characteristics. Pup exposure was generally lower
than the mother's with a shorter half-life, lower milk transfer,
larger volume of distribution, and gavage dosing, while higher
with longer half-life, higher milk transfer, smaller volume of
distribution, and dietary exposures. The present model demon-
strated pup exposures do not always parallel the mother's. The
model predictions can be used to help design early life toxicity and
pharmacokinetic studies and better interpret study findings.
Key Words: early life dosimetry; biological modeling; lactational
exposure.
Disclaimer: This work was reviewed by EPA and approved for publication,
but does not necessarily reflect official Agency policy. Mention of trade names
or commercial products does not constitute endorsement or recommendation by
EPA for use.
1 Present address: The Hamner Institutes of Health Sciences, 6 Davis Drive,
Research Triangle Park, NC 27709.
2 To whom correspondence should be addressed at National Center for
Computational Toxicology, B205-1, Office of Research and Development, US
Environmental Protection Agency, 109 TW Alexander Dr., Research Triangle
Park, NC 27711. Fax: (919)-541-1994. E-mail: habarton@alum.mit.edu.
Published by Oxford University Press 2007.
Evaluating potential risks from early life exposures is more
challenging than evaluating risks in adults, in part, because the
relevant toxicity studies including one- or two-generation
reproductive, developmental, and developmental neurotoxicity
studies involve multiple life stages (e.g., gestation, lactation,
and postnatal growth of offspring). Currently, the average daily
dose given to the mother is used for extrapolation to humans,
even when the effects are observed in the offspring. To
improve extrapolation of animal toxicity data to humans,
information on chemical dosimetry in the young during critical
developmental windows would be needed (Barton, 2005).
However, dosimetry data from early life exposures are scarce
for environmental chemicals, and even for pharmaceuticals, in
multigeneration studies. Poorly characterized pup dosimetry
during lactational and early postweaning periods is a substantial
source of uncertainty in the extrapolation of rodent toxicity
findings to humans along with uncertainty in the identification
of critical developmental windows. In recent years, predictions
of perinatal internal exposures have been made using
computational pharmacokinetic modeling for a number of
environmental and pharmaceutical chemicals (reviewed in
Corley et al. (2003)). However, it is generally difficult to
develop a full physiologically based pharmacokinetic model
because of limitations on pharmacokinetic information during
the relevant periods as well as limited information regarding
physiological parameters for early life stages (e.g., gestation,
lactation, and early postweaning).
Knowledge of pup dosimetry can contribute not only to
applying study results in evaluating risks but also for
improving toxicity study designs. A critical factor determining
chemical concentrations in pups would be the extent and
pattern of maternal chemical exposure because it determines
chemical concentrations in the adult animal in repeated dosing
scenarios (Saghir et al., 2006; Yuan, 1993). Several different
dosing approaches are used in toxicity studies including diet,
drinking water, gavage, and, if appropriate, dermal and
inhalation. Although dietary exposure often represents a rele-
vant exposure method for chemicals, in some cases it is
difficult to use due to technical problems preparing chemical-
fortified diet or a need to accurately determine maternal dose
levels leading to the use of gavage dosing. Only limited
Previous
TOC
-------
16
YOON AND BARTON
TABLE 1
List of the Theoretical Test Compound Categories
1
2
3
4
5
6
Chemical category
Base case (Vd° = 0.7, Pm = 1)
Small volume of distribution (Vd = 0.2)
Large volume of distribution (Vd = 2.5)
High milk transfer (Pm = 10)
Moderate milk transfer (Pm = 3)
Low milk transfer (Pm = 0.1)
Abbreviation
Base
SmVd
LgVd
HIghPm
MidPm
LowPm
Chemical properties
Uniformly distributed throughout the body /approximately
distributed to total body water
Milk concentration equals the maternal blood concentration
Limited distribution to tissues
Highly bound to plasma protein
Distribution to storage depot
Milk concentration greatly exceeds maternal blood level
Milk concentration is moderately higher than maternal blood level
Milk concentration is lower than maternal blood level
Note. We evaluated 16 different theoretical test compounds in the present study. Two test compounds were defined for each of the six chemical categories, one
with a short half-life and the other with a long half-life. Categories 2-6 include 10 different theoretical compounds, each of which possesses the same chemical
properties as the base case except for the one factor varied to the value noted in parentheses. For base case chemicals, the impacts of delayed elimination capacity
or excreta recirculation were evaluated for each half-life. Otherwise, prenatal development of elimination and no recirculation were modeled for all the test
compounds in categories 2-6. The details of pharmacokinetic parameter values used for each test compounds are listed in Table 3.
"Vd represents the body weight-normalized volume of distribution (I/kg) here and in other tables.
consideration has been given to potential differences in the
amount of chemical transferred to the suckling neonates
resulting from gavage versus dietary administration of a com-
pound (Arnold et al, 2000). There has been a concern about the
potential overexposure during lactation due to highly increased
maternal food consumption during this period, which has been
discussed as a potential cause of misinterpretation of increased
neonatal toxicity during lactation (Hanley and Watanabe, 1985).
To that end, a modified dietary administration regimen is
sometimes employed in reproductive toxicity study, which
adjusts the chemical concentration in diet based on historical
food intake data during lactation to maintain relatively constant
exposures during this period (Hanley et al., 2002).
The present study was intended to provide a tool to predict
pup dosimetry using limited biological and pharmacokinetic
properties of test compounds and, thus, to help design toxicity or
pharmacokinetic studies in early postnatal periods as well as to
help understand findings from such studies. The goals of this
research were to evaluate whether one could use data on adult
pharmacokinetics and milk transfer in conjunction with a bi-
ologically based model to predict pup dosimetry to a reasonable
approximation and then evaluate how different pharmacokinetic
properties (e.g., rates of clearance and milk distribution) would
affect the pup exposure doses (e.g., from milk) and circulating
concentrations. A classical compartmental pharmacokinetic
modeling approach was employed, which was supported by
biological information on changing factors (e.g., increasing pup
body weight) during lactation and early postweaning period.
Three dosing approaches employed in toxicity studies
(i.e., unadjusted ppm diet, adjusted ppm diet, and gavage) were
simulated to compare resulting maternal and neonatal dosimetry.
Model performance was evaluated by comparing the model to
previously reported lactational exposure data for two chemicals.
Subsequently, exposures were simulated for 16 theoretical
compounds with a variety of different characteristics bench-
marked from environmental chemicals and/or pharmaceuticals.
The properties of these theoretical test compounds are listed in
Table 1, categorized in six cases varying elimination rate,
volume of distribution, and milk transfer. The present model
simulates pharmacokinetics in the dam and pups for the parent
compound only. This model enables simultaneous consideration
of several factors with potential effects on pup exposures and
consequently provides a means to predict the overall impact of
these factors on dosimetry in the young. From the results of this
modeling exercise, we have begun to derive general insights
about pup exposures and different study designs for chemicals
with different properties.
METHODS
Model Structure
The model was coded and all the simulations were performed using
acslXtreme (version 2.0.1.7, Aegis, Inc., Huntsville, AL). The structure of the
biologically based pharmacokinetic model for chemical exposures of the dam
and pups is illustrated in Figure 1. Simulation of chemical exposures was
performed for 28 days after birth, of which the first 21 days were the lactational
period followed by 1 week postweaning. The model describes changes in body
weight, milk production and consumption, and food consumption during those
4 weeks as well as exposure by three methods—gavage, unadjusted feeding, or
adjusted feeding. All abbreviations and symbols used in describing the model
structure are listed in the legend of Figure 1. Equations for parameter values
that change over the duration of simulation are presented in Table 2. Values for
two other parameters, BWd and Vm, which change during the simulation period
were incorporated in the model using TABLE functions in acslXtreme as
explained later in this section. All other chemical parameters are listed in
Table 3.
Model Structure for the Dam
The present model uses a one-compartment pharmacokinetic model
structure, in which the dam and pups were each represented as one central
Previous
TOC
-------
BIOLOGICAL MODELING OF RAT EARLY LIFE DOSIMETRY
17
Maternal Exposure
RFO,
Food
Unadjusted or Adjusted
vs.
Gavage
Feeding
Dose
Gavage
Dose
Kari
Karf
Dam
Neonatal Exposure: Lactational
Weaning after PND21
Neonatal Exposure: Post weaning
Milk
Kan
Pups
N*Vn
Recirculation
Birth to PND14
FIG. 1. Schematic representation of BBPK model for chemical exposures during lactation and early postweaning. Abbreviations used in the present model are
as follows: Dam (subscript d), the compartment for the mother; Kad, first-order absorption constant for the dam (per hour); yd, volume of distribution of the dam (1);
Ked, first-order rate constant for chemical elimination from the dam excluding milk secretion (per hour); /fL, rate constant for chemical secretion via milk from
the dam (per hour); Milk (subscript, m), the conceptual compartment for milk; Vm, volume of milk secreted from the dam/ingested by the N pups (I/day); Pups
(subscript p), the compartment for the pups, as a litter; Kap, first-order absorption constant for the pups (per hour); N, the number of pups per litter; Vp, volume of
distribution of an individual pup (1); Kep, first-order rate constant for chemical elimination from the pups (per hour); RFDd, rate of feed dosing in the dam (g/h);
RFDp, rate of feed dosing in the pups (g/h). Chemical concentration in each compartment is expressed as C with a subscript for corresponding compartment;
Cd, concentration in the dam (mg/1); Cm, concentration in the milk (mg/1); Cp, chemical concentration in the pups (mg/1).
compartment. Elimination of a test chemical (e.g., metabolic or urinary) was
described as a first-order process with a rate constant, Ked (per hour); saturable
metabolism was not modeled. Elimination of chemical from the dam through
milk was modeled as a separate process determined by another elimination rate
constant, /fL (Per hour) (Fig. 1). The chemical absorption process of the dam
was a first-order process described by Kad (per hour). The model structure for
dosing is detailed in a following section. The volume of distribution of the dam
compartment (Vd) was defined as a product of the body weight-normalized
volume of distribution (Vdd, I/kg) with the body weight of the dam (BWd, kg).
Changes in the amount of the test chemical in the dam (mg/h) are described
in Equations 1-3, where Ad represents the amount of chemical in the dam (mg).
The overall chemical change in the dam was a function of the chemical
absorbed from the absorption site (RABd), elimination (Ked X Ad), and
secretion via milk (/fL * ^d) (Equation 1).
dAd/dt = RABd - Ked X Ad - KL X Ad.
RABd = dABd/dt = Ka
-------
18
YOON AND BARTON
TABLE 2
Equations Incorporated in the Model to Describe Changing Parameters"
Simulation periods
Weekl
Week2
WeekS
Week4
References
BWP
FOODd
FOOD/
AJ
Unadf
Adf
R
Prenatal
Delay*
0.0022 + 0.0028 X PND
(12 + 82 X PND)/(8.6 + PND)
0
1 1
FID/121e FID/202
1 1
(1
- 0.00014 X PND2 + 0.0000052 X PND3
(65000000 X e{-°-67 x PND)) +
(0.569 + (9.74 - 0.569))/((1 + e(22 ~
1
FID/269
1
X PND)/(PND + 7)
Doerflinger and Swithers (2004)
17
PND))/0.9)
1
1
1
Shirley (1984)
Redman and Sweney
Shirley (1984)
(1976)
Notes. Bwp, body weight of the pup (kg, individual pup); FOODd, daily food consumption by the dam (g); FOODp, daily food consumption by the pup
(g, individual pup); AJ feeding dose adjustment factor; R, ratio of Kep/Ked, defining developmental pattern of elimination capacity in the pup.
These equations were written in DISCRETE block in acslXtreme CSL file.
kpOODp equation was in effect from PND17 and onward.
c"Unadj" represents the unadjusted feeding dosing simulation.
rf"Adj" represents the adjusted feeding dosing simulation.
TID refers to the reference intake and the numbers represent the food intake by the dam during the indicated simulation week.
^"Prenatal" refers to development of adult elimination capability occurring before birth so it modeled as constant after birth.
*"Delay" refers to delayed development of elimination capacity simulated in the model. The elimination capacity was modeled to reach the half maximum on
PND7.
TABLE 3
Chemical Parameters and Simulation Conditions for the 16 Theoretical Test Compounds
Vd (I/kg)
Ka (per hour)
Ke (per hour)
Category
Half-life Dam (Vdd) Pup (Vd ) Pm Dam Pup
Dam
Pup
Recirculation
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
Base Short
Long
w/Developmental delay Short
Long
w/Recirculation Short
SmVd
LgVd
HighPm
MidPm
LowPm
Long
Short
Long
Short
Long
Short
Long
Short
Long
Short
Long
0.7
0.7
0.7
0.2
2.5
0.7
0.7
0.7
0.7
0.7
0.7
0.2
2.5
0.7
0.7
0.7
1 2
1 2
1 2
1 2
1 2
10 2
3 2
0.1 2
2
2
2
2
2
2
2
2
0.7°
0.036
0.7
0.03
0.7
0.03
0.7
0.03
0.7
0.03
0.7
0.03
0.7
0.03
0.7
0.03
0.7°
0.036
0.7 X R
0.03 X R
0.7
0.03
0.7
0.03
0.7
0.03
0.7
0.03
0.7
0.03
0.7
0.03
No
No
Yes
No
No
No
No
No
Note. Abbreviations used for describing category are from Tables 1 and 2. Values in bold highlight the base case for comparison purposes and the changes in
individual parameters defining specific cases.
"ti/2 = 1 h.
btl/2 = 24 h.
Previous
TOC
-------
BIOLOGICAL MODELING OF RAT EARLY LIFE DOSIMETRY
19
When modeling developmental delays in elimination, Kep was modeled to
be proportionally related with the mother's value, i.e., following Kep = R X
Ked, where R indicates the developmental pattern of pups' elimination capacity.
Delayed development of elimination was modeled using a Michaelis-Menten
type curve (Table 2).
Structure for Lactational Transfer
The dam and pup compartments were connected with a conceptual milk
compartment, the volume of which (Vm, 1) does not refer to actual existing
volume as a separate compartment, but rather represents the postnatal day
(PND)-dependent volume of milk produced per day. It was assumed that the
pups consume all the milk produced without any delay between production and
ingestion. The rate of milk production and suckling was assumed to be constant
without any circadian variation. A simplified description of milk intake was
used that did not describe separate suckling episodes throughout the day, but
rather used a continuous input to the pups at a constant rate.
The rate constant for lactational transfer/secretion of chemical (/fL) was
derived from two predetermined factors, the Vm and the milk partition
coefficient or ratio of the chemical concentration in milk to the dam's central
compartment (Pm). Pm was employed in the model as an index of the extent of
chemical transfer into milk relative to the levels of chemical in plasma (i.e.,
mother's central compartment). It was assumed that the chemical concentration
in milk was in instant equilibration with maternal blood and consequently
parallels her concentration. Pm was assumed to be constant during the whole
lactational period. By definition,
Cm = Pm X Cd. (7)
The amount of chemical secreted in milk (Am, mg) can be expressed as
a function of the clearance to milk (Clm, 1/h) and the milk concentration
(Cm, mg/1) expressed as the product of the concentration in the dam and the
milk partition coefficient:
dAm/di = Clm X Cm = Clm X CA X Pm.
(8)
The milk clearance is the volume of milk produced per day (ym) divided by 24
h. However, the model was specified in terms of rate constants, so we need to
derive the milk elimination rate constant /fL (per hour) on the z'th PND:
dAm/di = .KLXCdXVd. (9)
Setting Equations 8 and 9 equal and rearranging terms obtains:
£L = PmXVrm/(24XVd). (10)
Now the rate of lactational transfer of chemical (RML) is expressed as:
RML = KL XAd = dAML/dt,
(H)
where AML represents the amount of chemical secreted through milk obtained
by the integration of Equation 11 and available for absorption to the pups in
Equation 6.
Model Structure for Dosing
The model included exposure of the dam by three dosing approaches—
gavage, unadjusted feeding (constant ppm in diet), or adjusted feeding (weekly
changes in ppm in diet). All these were intended to provide the same target
dose, 15 mg/kg/day either throughout the study (gavage, adjusted feeding) or at
the appropriate baseline period (unadjusted feeding). After weaning, direct
dosing to the pups by gavage or unadjusted feeding was included in the model
for 1 week. The dosing to the pup was modeled for an individual pup, rather
than for the combined litter of eight pups. In the case of dietary exposure, the
pup started eating the same diet as the dam on PND 17 as described later. After
weaning, the pups were assumed to consume unadjusted diet.
Gavage. Gavage dosing was modeled as a bolus dose scheduled once
a day (coded in a DISCRETE block in acslXtreme) at the target dose of
15 mg/kg/day (ODOSEO). For gavage dosing, the daily dose to the dam or pups
(Dosed or Dosep in Equations 3 and 6) was defined as:
Dose = ODOSEO X BW.
(12)
Dietary exposure. To simulate feeding exposure, two factors that
determine the amount and rate of chemical input via diet were incorporated
in the model, the amount of food intake per day (FOOD, g/day) and the diurnal
pattern of food consumption using the mean percentage of total food intake
during 1-h intervals (FOODPC, %/h). Feeding exposure for each consecutive
simulation day was modeled as a continuous addition of the chemical to the
absorption site of the dam or pups at a specific rate (RFC, g/h) using the
TABLE function in acslXtreme.
RFC = FOODPC/100 X FOOD.
(13)
Integrating Equation 13 gives the amount of food consumed (AFC, g). The
amount of feed consumed per day (FOOD) was introduced using the DISCRETE
block for the dam and when applicable, for the pups, to accommodate daily
changes. The values for FOOD were determined by the equations in Table 2.
The chemical concentration in diet (FEEDO, mg/g diet) to achieve the target
dose (TARGET, 15 mg/kg/day) was derived using the mean food consumption
(FID, g/kg/day) by the dam during gestation or during the first week of lactation
as a reference point, depending on the simulation scenarios.
FEEDO = TARGET/FID.
(14)
For unadjusted feeding, FEEDO was used for the whole duration of simulation
without any modification. In order to simulate adjusted feeding exposure, a feeding
dose-adjustment factor (AJ) was incorporated in the model to appropriately reduce
chemical concentration in food during lactation based on the extent of increase in
food intake during lactation compared to the reference intake (FID) (Table 2).
= FID/mtake;.
FEED = FEEDO X AJ,
(15)
(16)
where Intake, indicates the mean food intake (g/kg/day) during the zth week of
lactation and FEED represents the adjusted chemical concentration in food
(mg/g food). Intake, was adapted from historical intake data (Shirley, 1984) for
which the intake by the dam and pups was not discriminated, as is typical in
toxicity studies. Adjustment of chemical concentration in food was modeled on
a weekly basis and only for lactation period, so AJ = 1 was used for the
postweaning period returning the concentration in food to the initial
concentration. Consequently, adjusted and unadjusted feed concentrations
were the same after weaning (Table 2). Although the direct dosing of the pups
through food during the first week after weaning was expressed as unadjusted
feeding, some toxicity studies also adjust the diet during this period.
The rate of chemical dosing via feeding (RED, mg/h) is:
RFD = FEED X RFC.
(17)
Hence, the dose from dietary exposure (FDOSE, mg) (i.e., the amount of
chemical consumed via diet) was obtained by integrating RFD and then utilized
as Dose in Equations 3 and 6 when dietary administration was simulated.
Recirculation of excreta. Neonatal rats are known to be unable to
eliminate wastes without maternal stimulation for several days after birth
(Henning, 1981). Hence, it was expected that much of the chemical eliminated
from the pups returned to the dam through this process. In order to simulate the
recirculation of excreta between the dam and pups, the amount of chemical
eliminated from the pups (AEP, g) was modeled as an additional chemical input
to the dam without loss for the first 2 weeks of postnatal period (Fig. 1). AEP
was added to the Dose in Equation 3 during those 2 weeks, where AEP was
. X AD).
Model Parameterization
The present model incorporated known changes in biological parameters
during lactation and the early postweaning period. Modeled kinetic properties
Previous
TOC
-------
20
YOON AND BARTON
for the test compounds were benchmarked using data from real chemicals. The
rationale for biological and chemical parameterization of the current model is
detailed in the Supplementary section. Parameter values were obtained from
literature when possible, but several assumptions were made due to limited data
availability for these early life stages in rats. Efforts were made to obtain values
within the same study and/or for the same species of rats whenever possible.
Biological/pharmacokinetic parameters changing during lactation and early
postweaning were modeled with either linear interpolations between reported
time points of measurements using the TABLE function in acslXtreme or curve
fitting to reported data points using nonlinear regression tools in Prism 4
(GraphPad Software, Inc., San Diego, CA). Equations from fitted curves are
listed in Table 2. The values used in TABLE functions for BWd and ym are
reported in the Supplementary section. The equation for consumption of food
and the equation for utilizing the TABLE value for daily milk volume were
written in a DISCRETE block, so that they varied on a daily basis, but stayed
constant during each of the 24 h. When converting data points from previously
published figures in the literature into numbers in order to incorporate them in
the model, Digitizelt software was used (version 1.5, www.digitizeit.de).
Biological Parameters
The biological data incorporated in the present model are shown in Figures 2
and 3.
Body weights. Maternal body weight (BWd) changes during lactation and
postweaning periods were incorporated into a TABLE function based on
published values for Sprague-Dawley rats (Shirley, 1984). The growth rate of
the neonates was derived from pup body weight data for Sprague-Dawley rats
(Doerflinger and Swithers, 2004) as shown in Figure 2 and Table 2.
Food consumption. Food intake for the dam was adapted from the same
study from which the body weight data were obtained (Shirley, 1984). Shirley
reported the total observed intake as maternal intake, although food
consumption by the pups in later part of lactation was observed. In order to
calculate food intake solely by the dam, an estimated amount of food eaten by
a whole litter (average size reported as 9.5 pups) was subtracted from the total
reported intake. For this purpose, the amount of food consumed per day by an
individual pup (g/pup/day) reported in another study with Sprague-Dawley rats
(Redman and Sweney, 1976) was multiplied by 9.5. The onset of diet
consumption by the pups was introduced as PND17 in the present model. The
estimated mother-only intake values (g/dam/day) incorporated in the model are
plotted in Figure 2 along with the observed total intake values during the last
5 days of lactation for comparison purposes. The pattern of feeding by the pups
was modeled with intake starting on PND17 and rapidly increasing until
PND21, followed by a continuous increase over the postweaning period
(Redman and Sweney, 1976)(Fig. 2).
Dietary dose adjustment. To simulate the adjusted diet dosing regimen by
reducing the chemical concentration on a weekly basis, the feeding dose
adjustment factor AJ described in Equations 15 and 16 was included in the model
as shown in Table 2. Values for FID and Intake, were derived from gestational
and lactational food intake data using Sprague-Dawley rats (Shirley, 1984).
Diurnal variation in feeding behavior. The diurnal feeding behavior of
the dam and its changing pattern during the lactation and postweaning periods
were included in the model based on data for Wistar rats (Strubbe and Gorrisen,
1980). It was incorporated in the model as % total daily intake per hour
(FOODPCd, %/h) as described earlier, for which four different patterns were
defined for each week of lactation and postweaning using TABLE functions
applied sequentially for the corresponding simulation week.
The diurnal fluctuation of feeding rates was also applied in modeling food
intake by the pups. From PND17 to weaning, it was assumed that the pups follow
the same feeding pattern as the dam during this period, i.e., circadian variations in
feeding were not yet obvious (Doerflinger and Swithers, 2004; Redman and
Sweney, 1976). The diet consumption pattern in Sprague-Dawley rat pups on
PND25 was adapted to represent the feeding pattern during the postweaning
periods in the current model (Redman and Sweney, 1976). As in the case of the
28
100-1
_c
T3
O Measured (Dam+Pups)
x Estimated (Dam only)
14
PND
21
28
0.08-1
0.06-
O)
i 0.04-
ffi
0.02'
0.00
BW, pup
14
PND
21
28
12-1
-------
BIOLOGICAL MODELING OF RAT EARLY LIFE DOSIMETRY
21
0.100 -I
0.075 -
•D 0.050 H
0.025
0.000
X Experimental
-B- Modeled
A Theoretical
14
21
PND
FIG. 3. Theoretical milk yield volume over lactation. "Experimental"
represents data points from Knight et al. (1984). "Modeled" represents
recreated Vm values used for creating the TABLE function. The connecting
solid line shows the simulated Vm values over time. "Theoretical" represents
Vm values calculated to match the suggested caloric requirement from milk for
the growing pups (Stole et al., 1966).
dam, the feeding pattern was incorporated in the model as hourly rates of food
intake (FOODPCp%, total daily intake/h) using a TABLE function.
Milk consumption. In this model, the maternal milk yield equals the milk
intake by the pups with no losses. Daily milk intake by the pups was
incorporated in the model as a combination of experimentally measured values
for early lactation and theoretical values based on the energy requirement for
growing rat pups for the later part of lactation. Milk consumption was
simplified using continuous suckling, rather than attempting to capture its
episodic occurrence. Since pup intake of milk and food were not available from
a single study, multiple sources were utilized to create estimates that were also
evaluated to insure that the caloric intake was consistent with the growth of the
pups.
In order to construct a milk intake curve for the early period of lactation,
milk yield data determined from Wistar rats with a litter size of 10 were utilized
(Knight et al., 1984). We adopted the values for PNDs2 and 6 determined by
the tritiated water dilution technique for the TABLE function for milk intake
during the first week of lactation. For the later part of lactation, theoretically
derived milk intake values were incorporated in the model, based on the
suggested caloric requirement of pups (Stole et al., 1966), calories provided by
independent feeding from PND17 and onwards (Redman and Sweney, 1976),
and the observed ability of neonatal rats to respond to caloric deficit and
consume either milk or diet at appropriate levels to match their energy needs
(Henning, 1981). The milk intake in the second and third weeks of lactation
was set to meet the suggested caloric requirement of 45 kcal/100 g body weight
(Stole et al., 1966), either provided solely by milk from PND7 to PND16 or
provided both by milk and diet from PND17 onwards, i.e., milk volume
suckled by the pups during these last 5 days was estimated to fulfill the energy
requirement not already provided by the diet. The overall milk intake pattern in
the current model consists of two experimental data points for PND2 and PND6
and 15 estimated values for PNDs7-21 that were utilized in the TABLE
function. A smooth transition from the experimentally measured milk intake to
the calculated values was possible because the milk intake values calculated
using the two approaches were very similar for PND6 (Fig. 3). Since the Knight
et al. (1984) data were for milk intake by 10 pups, daily milk intake per
kilogram pup body weight was calculated using the reported body weight in the
same paper, and then the daily milk yield for 8 pups was calculated using pup
body weight simulated in the model (Doerflinger and Swithers, 2004). Caloric
values of rat milk were derived from the milk composition data and the energy
value of its components (Bornschein et al., 1977; Luckey et al., 1954).
Physiological fuel energy value of 3.41 kcal/g for Certified Rodent Diet5002,
which is often used in multigeneration toxicity studies (Hanley et al., 2002;
Hinderliter et al., 2005), was utilized to calculate caloric value of rat chow
(www.labdiet.com). Sometimes a different diet (e.g., DietSOOS), which has
higher calories (i.e., 3.50 kcal/g) than Diet5002, is used for lactating dams and
the diet switched after weaning (Howdeshell et al., 2007; Rayner et al., 2007).
However, these small differences in energy value of diets were not expected to
make a substantial difference either in milk intake estimation or food intake by
the pups during late lactation and postweaning period. For instance, only 2.5%
less Vmilk value was estimated at most using DietSOOS in the simulation. The
constructed curve for changing milk intake was incorporated in the model to
simulate the daily milk intake for the eight pups (I/day, Vm). Circadian
variation in milk intake was not modeled, i.e., constant suckling throughout
a day was assumed as suggested from a few studies (Godbole et al., 1981;
Redman and Sweney, 1976).
Chemical Properties of Theoretical Test Compounds
The present model was run for a series of hypothetical chemicals with
different characteristics denoted as six categories (Table 1). A total of 16 dif-
ferent chemicals were simulated incorporating different chemical parameters
and conditions in the model (Table 3). These chemical parameter values were
either benchmarked from actual chemicals or derived from a few assumptions
detailed here.
Oral bioavailability. Oral bioavailability of the test chemical administered
via gavage, feeding, and milk transfer was assumed to be 100% by setting the
value of F (used in Equations 3 and 6) as 1.
Absorption. The absorption of the chemical was simulated as a rapid
process both in the dam and pups. The absorption constants for the dam and
pups were assumed to be the same (Kaj = Kap) and to be constant over the
duration of simulation.
Distribution. The volumes of the distribution for the dam (Vd, 1) or pups
(Vp, 1) were calculated by multiplying the body weight-normalized volume of
distribution sealer (Vd, I/kg) for the dam or pups with its body weight (kg). The
same Vd values were used both for the dam and the individual pup (Vdd =
Vdp) and kept constant during the whole simulation. Three values of Vd were
used to simulate different distribution scenarios: limited distribution to tissues
(Vd = 0.2), distribution to total body water (Vd = 0.7), and distribution to
a storage depot (Vd = 2.5).
Extent of milk transfer. A ratio of the concentration in dam's plasma to
her milk (Pm) was used to describe the extent of chemical transfer to milk
(Equation 7) since milk was assumed to be in instant equilibrium with the
mother's concentration in her central compartment. The ratio was constant
during the whole lactational period. Four Pm values were adopted in the present
model: Pm = 0.1 (milk < plasma), Pm = 1 (milk = plasma), Pm = 3, and
Pm = 10 (milk > plasma). Milk concentrations compared to the dam's plasma
or blood for several environmental chemicals and Pharmaceuticals in rats fall
into the Pm ranges used, including perfluorooctanoate (ratio fa 0.1),
2,4-dichlorophenoxyacetic acid (2,4-D, ratio fa 1), tetrachloroethylene (milk
to blood partition coefficient ~ 10), zidovudine (milk to serum ratio ~ 1), and
ranitidine (milk to serum ratio fa 10) (Alcorn and McNamara, 2002;
Byczkowski et al, 1994; Hinderliter et al, 2005; McNamara et al, 1996;
Sturtz et al, 2006). Biologically, milk concentration can reflect distribution
dependent on physical chemical properties (e.g., partition coefficient) and also
biochemical properties (e.g., active transport and protein binding). In the
present model, Pm was varied with a fixed Vd, i.e., 0.7 I/kg, so high Pm cases
would reflect chemicals actively transferred into milk rather than highly
lipophilic chemicals for which a higher Vd would be expected as well as
high Pm.
Elimination. The elimination rate constants in the dam (Ked) and pup
(Kep) were set to the same values and treated as constant, except when
investigating the impact of developmental delays. The elimination rate
Previous
TOC
-------
22
YOON AND BARTON
TABLE 4
Chemical Parameters and Simulation Conditions for the Compounds for Model Benchmarking of 2,4-D and OTA
Vd (I/kg)
Ka (per hour)
Ke (per hour)
Dam
Pup
Pm
Dam
Pup
Dam
Pup
Recirculation
OTA
Prenatal development
Developmental delay
2,4-D
Adult female Vd
Smaller Vd
0.43
0.43
0.18
0.077
0.43
0.43
0.18
0.077
0.6
0.6
1.1
1.1
2
2
3.9
3.9
2
2
3.9
3.9
0.0067
0.0067
0.33
0.33
0.0067
0.0067 X R
0.33
0.33
No
No
No
No
Note. The parameter values for OTA and 2,4-D were adapted from previously published adult female pharmacokinetic parameters (Li et al., 1997; Timchalk,
2004). To simulate the developmental delay in elimination capacity for OTA, R implemented the function in Table 2. For 2,4-D lactational exposure simulations,
the volume of distribution was reduced to the smaller Vd reported for this chemical (Timchalk, 2004).
constants were chosen to give half-lives (f1/2 = 0.693/Ke) of 1 and 24 h,
representing rapid and slower elimination, respectively, in nonlactating rats.
During lactation, the half-life in the dam can differ because the total elimination
includes excretion via milk which can shorten the overall half-life. A half-life of
24 h was considered the longest reasonable to consider in the current model
structure because it does not directly include gestational exposures, which
would be expected to result in a substantial body burden carrying over into
lactation.
Allometric scaling was not employed when incorporating the elimination
constant in the model, so the half-life does not change with age. Hence, Ked and
Kep were constant, independent of body weight or corresponding age over the
period of simulation, except when modeling developmental delays in the pups
(Table 2). Kep was expressed as proportion of Ked using R = Kep/Ked. When
R = 1, the pups' overall elimination capability was at adult levels at birth.
Alternatively, a delayed pattern of elimination was modeled, with half
maximum activity reached on PND7. This development pattern was based on
critical changes in neonatal kidney morphology and function observed within
the first postnatal week (Kavlock and Gray, 1982). However, it should be noted
that it only represents one possible scenario of changes in elimination capacity
during rat development, and this formula does not refer to any specific
metabolic or renal process (see Supplementary section for other possible
developmental patterns).
Defining base cases. It was necessary to set a point of reference to which
other simulation results could be compared. This was indicated as a "base case".
The pharmacokinetic parameters for the base chemicals are shown in
Table 3. Two base chemicals were defined, one for short half-life and the other
for longer half-life compounds. For the base case simulations, elimination
capacity in the pups was assumed to have already reached an adult level at birth
(R = 1), milk concentrations were equal to maternal plasma concentrations, and
distribution was to total body water. Recirculation of excreta between the dam
and pups was not included in the base case simulation. To evaluate the impact
of selected biological or chemical factors on the extent of neonatal exposure,
these factors were varied one factor at a time using alternative values listed in
Table 3, and the simulation results were compared to those from the base case.
Each chemical property was varied with all other properties fixed to base case
values in the current study to explore the range of model predictions.
Alternatively, one can run the model with different sets of chemical parameters
based on the properties of actual chemicals. This approach is required for the
model evaluation by comparison with measured data for specific compounds.
Model Evaluation
The model performance was evaluated using previously published
lactational transfer studies for 2,4-D and ochratoxin A (OTA) in rats. Model-
predicted concentrations of these chemicals in the dam, pups, and milk were
compared to published values. All the model assumptions and parameters
described earlier were applied in the two benchmarking simulations (e.g., milk
consumption and pup growth) except 2,4-D or OTA-specific pharmacokinetic
parameters and the reported experimental conditions were applied instead of
theoretical values and study designs. (Table 4)
Simulation of OTA Exposure via Milk in Rats
Placental and lactational transfer of OTA in Sprague-Dawley rats were
measured in a cross-fostering study (Hallen et al., 1998). The model was used
to simulate exposure of the dam to OTA during lactation and the first week after
weaning; the predicted concentrations in the dam, pups, and milk were
compared to the published values for PNDsl4 and 21. We benchmarked the
model to the data from one of the cross-fostering groups in which the pups were
born from an unexposed dam and nursed by a foster mother exposed to the
toxin throughout premating, gestation, and lactation. Therefore, it was
necessary to consider the chemical in the dam resulting from the prelactational
exposure. It was reported that OTA was given to the dam by gastric intubation
at 50 ug/kg/day five times per week during five consecutive weeks including
2 weeks of premating and 3 weeks of gestation followed by 7 days/week for the
3 weeks of lactation (Hallen et al., 1998). Since the present model does not
have either premating or gestational periods, the amount of OTA still in the
dam resulting from the 5 weeks of exposure before birth was assumed to be
equal to that from 5 weeks of repeated exposure for an adult female. Hence, the
prelactational exposure was simply modeled as repeated oral gavage of OTA to
the adult female following the dosing regimen described above. To simulate
this repeated OTA exposure in the adult female, selected parameter values were
changed, i.e., the milk yield was set to zero, and the body weight of the dam
was held constant at early gestational weight, 0.25 kg (Shirley, 1984). OTA
pharmacokinetic parameters were adapted from the two-compartment model
structure of Li et al. (1997) using adult female Sprague-Dawley rats. The beta
phase elimination constant and steady-state volume of distribution were
adopted as Ke and Vd for the current one-compartment model. The rapid
absorption constant, Ka = 2, was utilized for OTA simulation based on
observation of efficient absorption of OTA from the gastrointestinal tract after
oral administration in F344 rats (Zepnik et al., 2003). A developmental delay in
elimination was also modeled (Table 4).
Simulation of 2,4-D Exposure via Milk in Rats
Sturtz et al. (2006) administered 2,4-D to Wistar rat dams with eight pups per
litter during lactation using an adjusted feeding method. 2,4-D concentrations in
the serum of the dam and pups and in the milk were determined on PND16. In
this study, the 2,4-D concentration was adjusted by comparing the extent of food
intake to the most recent intake for the two preceding days during lactation.
Lacking details, we approximated this study design and modeled the diet
Previous
TOC
-------
BIOLOGICAL MODELING OF RAT EARLY LIFE DOSIMETRY
23
1.0-
0,8-
0.6-
0,4-
0.2-
0.0
A)
Prenatal
7 14 21
PND
simCd simCp
1.0
0.8-
0.6
0,4-
0.2
0.0
B)
Delayed
14
21
sim Cm A Cp v Cm
PND
Cd
FIG. 4. Simulation of OTA concentration in the dam, pups, and milk. Predicted concentrations in the dam, pup, and milk are indicated as sim Cd, sim Cp, and
sim Cm, respectively. The published data points on PNDsl4 and 21 are represented with mean ± SD (Hallen et al., 1998). They are plotted in the middle of the
corresponding day of measurement. The predictions using prenatal and delayed developmental patterns of elimination are illustrated in A and B, respectively.
concentration adjustment as a weekly event using the food intake in the first week
of lactation. The model was evaluated for predicting the dosimetry for the lowest
exposure dose only, i.e., 15 mg/kg/day, because elimination of 2,4-D in rats has
been shown to saturate at higher exposure levels, e.g., 50 mg/kg and above
(Gorzinski et al., 1987). The value for Pm was 1.1 according to the findings by
Sturtz et al. (2006) at 15 mg/kg/day. As listed in Table 4, two different
simulations were performed to compare the model with the experimental data.
The first simulation (adult female Vd) used the 2,4-D pharmacokinetic
parameters derived from the nonlactating adult female rats (Griffin et al.,
1997; Timchalk, 2004). In the second simulation (smaller Vd), the volume of
distribution (Vd) was shifted to a value, which was the lower end reported for
2,4-D (Timchalk, 2004), while keeping the elimination rate constant unchanged.
Dose Metric Calculation
Daily dose metrics simulated include the maximum concentration of the
chemical (Cmax, mg/1), 24-h cumulative area under the curve (AUC, mg X h/1),
and daily dose (mg/kg/day). The percentage of AUCp to AUCd was also
reported as a dose metric of relative risk for the pups, as suggested in Corley
et al. (2003). All the dose metrics were calculated for three selected PNDs—4,
16, and 28—representative for early lactation, (moderately) late lactation, and
early postweaning.
The differences between the values on PND, and PND,_i were referred to as
the daily dose metrics on z'th PND for the 24-h AUC, the relative risk, and Cmax.
The daily dose for gavage dosing was a fixed value of 15 mg/kg/day. In the
case of dietary exposure, the daily dose on PND, is given by the Equation 18.
(FDOSE; -FDOSE;_!)/BW;.
(18)
During lactation, the daily dose to the pups was the sum of chemical input
from milk and diet (if applicable), where the milk dose on PND, is given by
Equation 19.
(AML; -
(19)
This gives the daily dose to the pups as the sum of milk dose (Equation 19) and
dietary dose (Equation 18), which is applicable only for the last 5 days of
lactation (i.e., PNDsl7-21).
RESULTS
Evaluation of Model Performance
The model performance was assessed using the published data
for 2,4-D and OTA. Figure 4 compares model simulations with
previously reported OTA data (Hallen et al., 1998). The reported
Cd and Cm values fell in the predicted concentration ranges
either simulated with adult female pharmacokinetic parameters
(Fig. 4A) or with delayed development of pup elimination
incorporated (Fig. 4B). The Cp was reasonably predicted in both
cases showing 13% underestimation and 20% overestimation
compared to the measured value on PND14, respectively.
Table 5 reports the simulated 2,4-D concentrations in the
dam, pups, and milk on PND 16 from two different simulations
with varying Vd. The predicted 2,4-D levels in the dam, pups,
and milk were consistently about one-third of reported values
with the adult female Vd (Table 5). With a smaller Vd, higher
values were achieved that were closer to the measured values
(Table 5). Incorporating recirculation and delayed development
of elimination in the model made little differences compared to
the predicted 2,4-D levels with adult Vd (data not shown).
These two limited tests indicate that the model can
reasonably approximate pup exposures when data on milk
transfer and kinetics in the nonlactating female are available.
Predicted Dose Metrics for Theoretical Chemicals
Exposure levels in the dam and pups were predicted and
several internal or external dose metrics were calculated as
indicated in "Methods" section. Since it was not practical to
report all the predictions from the simulations, selected results
TABLE 5
Benchmarking the Model to Published 2,4-D Data
Dam (Cd) (mg/1) Pup (Cp) (mg/1) Milk (Cm) (mg/1)
Sturtz et al"
Adult female Vd6
SmVd6'c
26.09 ± 4.24
4.6-9.1
8.0-18
6.34 ± 1.68
1.5-2.1
6.6-9.5
28.92 ± 3.04
5.1-10
8.9-19
"Mean ± SE as reported in (Sturtz et al., 2006).
^Predicted Cmin-Cmax values were reported for each dose metric.
The volume of distribution for 2,4-D was varied to the smaller value
reported for this chemical (Timchalk, 2004).
Previous
TOC
-------
24
YOON AND BARTON
A)
12
6 •
12-
6 -
3 -
Short half life compound
Dam
Gavage Unadj Feed Adj Feed
IP f r\
7 14 21
PND
Pup
Gavage Unadj Feed Adj Feed
28
14
PND
21
28
B)
75 -i
60
^,45-
£30-
15 •
"01
75 n
60
45
30 -
15
Long half life compound
Dam
Gavage Unadj Feed Adj Feed
7 14 21
PND
Pup
Gavage Unadj Feed Adj Feed
28
14
PND
21
28
AUCd
AUCp
AUCd
AUCp
FIG. 5. Predicted concentrations and AUC values for base case compounds. The base case represents a rapidly absorbed chemical distributed to total body
water. It partitions to milk with a concentration equal to maternal plasma. Simulations are for the 3-week lactational period followed by 1-week postweaning. The
24-h AUC for dams and pups were calculated for each indicated PND.
are highlighted below. Complete listings of predicted dose
metrics (Cmax, AUC, and daily exposure dose) as well as
figures showing all the simulated concentration profiles over
time are in the Supplementary section.
Base Case Comparison of Alternative Exposure Methods,
Short Half-life
The concentrations in the dam and her pups predicted during
the 4 weeks simulation and the resulting 24-h AUC values on
selected PNDs for the short half-life base case compound are
shown in Figure 5A. hi the dam, gavage dosing resulted in
higher peak concentrations compared to the two dietary
exposures. For dietary exposures, the peak levels reflected
the changing pattern of food intake during lactation and after
weaning. When the dam was dosed via unadjusted feeding
(i.e., constant ppm in diet), the peak levels were observed to
increase during the first week of lactation as expected from
a rapid increase in dam's food consumption in this period, while
the peaks in the second and third week of lactation increased only
slightly. This behavior differed from the expectation based on
food intake by the dam which increased during these latter two
weeks, although not as extensively as in the first week (Fig. 2).
The similarity of the peak concentrations is attributable to the
Previous
TOC
-------
BIOLOGICAL MODELING OF RAT EARLY LIFE DOSIMETRY
25
altered feeding pattern of the dam during lactation which
results in less fluctuation in feeding rates during the day.
Although the peak concentrations were approximately constant
during the second and third weeks, internal exposure (AUCd)
on PND16 was higher than on PND4 indicating that the dam's
increased food intake indeed resulted in greater exposure
(Fig. 5A). If the chemical concentration in diet was adjusted
during lactation, peak values were lower than those from the
unadjusted dosing regimen. As intended for the adjusted diet
protocol, the internal exposure was maintained at approximately
constant levels during lactation as implied by the similar
predicted AUCd for PNDs4 and 16 (Fig. 5A). Since the
adjustment was simulated only for the lactational period and
thus the chemical concentration was the same for adjusted and
unadjusted exposures, the concentration profiles and AUCd of
the dam for postweaning period were identical for the two die-
tary dosing methods (Fig. 5A).
The concentration in the pups (Cp) during lactation was
substantially lower than in the mother (Cd) for all three dosing
approaches (Fig. 5A). Gavage dosing was predicted to have the
greatest difference, with the pup's peak concentrations about
50 fold lower than the mother's (see Supplementary tables). The
time profile for the chemical concentration in the pups during
lactation did not parallel the mother's. When the dam was
gavaged, the pup concentration was predicted to continuously
decrease as they grew, while the mother's concentration stayed
constant. The rate of chemical input through milk appeared not
to be fast enough to prevent chemical dilution by growth of the
pup, i.e., increasing pup volume of distribution, hi simulations of
feeding exposure, the profiles of peak concentrations in mother
and pup were very different from PND17 and onwards, which
was attributable to the initiation of consumption of dosed-feed
by the pups (Fig. 5A). During the 4th week of simulation, the
pups and dam were either dosed directly by gavage or ate the
unadjusted diet. The pups' internal exposure during this
postweaning period was the same as the mother's exposure
when gavaged. Pups fed chemical via diet had higher peak
concentrations and a larger AUC than their mother's (Fig. 5A),
which was attributable to the higher food consumption per kg
body weight of the pups compared to adults. Compared to
lactational exposure, the postweaning exposure of pups was
much higher for both gavage and feeding. The differences in pup
peak concentrations between lactation and postweaning expo-
sures were greater for gavage dosing, but in terms of AUC, the
extent of the lactation versus postweaning differences were
similar in gavage and feeding.
Base Case Comparison of Alternative Exposure Methods,
Long Half-life
Figure 5B shows the predicted concentration profiles in the
dam and pups for the long half-life compound. The daily peak
concentrations were the highest with unadjusted feeding
exposure, and the AUCd from this dosing method was also
greater than gavage or adjusted feeding approaches in the dam.
Similar to the short half-life case, the predicted AUCd levels
indicated that the exposures to the dam via gavage and adjusted
feeding were to be very similar. The slight increase in
concentration in the gavaged dam after weaning was due to
cessation of chemical excretion in milk. The concentration
profiles in the pups roughly parallel the mother's for the three
dosing methods though lower and showing smaller daily
fluctuations (although this partially may be due to modeling
continuous suckling, rather than episodic behavior), hi the case
of long half-life chemical exposure, the concentrations in the
pups, Cp, approached the mother's levels, especially for the first
week, but were still lower. Adjusted feeding and gavage
produced very similar lactational exposure in the pups in terms
of both peak concentrations and AUCP during the later part of
lactation until the pups initiated their own feeding (Fig. 5B). The
impact on Cp of the pups' independent feeding on treated diet
was not as great as that observed for the short half-life compound
exposure. Postweaning exposure of the pups was the same as the
dam for gavage, while higher than mother's for feeding exposure.
When compared to the lactational exposure, the postweaning
exposures were observed to be higher for both gavage and
feeding, but the difference between the two periods was not as
great as was the case for the short half-life compound in terms of
peak concentration and the predicted AUCp (Figs. 5A and 5B).
When comparing the lactation and postweaning exposures, the
greatest difference was caused by adjusted feeding followed by
gavage and then unadjusted feeding. It should be noted that the
exposure to the dam determines the pup's exposure during
lactation, while the pups' own exposure determines their
exposure level after weaning. It was also noticeable that dietary
exposure produced higher peak concentrations in the post-
weaning pups compared to gavage, in the long half-life chemical
exposure simulations, while it resulted in lower peak concen-
trations than gavage for the short half-life chemicals (Fig. 5B).
Changes in Volume of Distribution
Simulations were performed to determine the impact of
changing chemical properties on the extent of pup exposure.
By changing Vd and Pm, the exposures of the dam and pups
varied to different extents from the base case predictions, but
some of the patterns remained similar to those for the base
cases (see Figures in Supplementary section). Only distinctive
features resulting from these variations are described here. The
AUC values were utilized to compare the pups' exposure to the
dam's and among different periods of postnatal life. The impact
of varying chemical properties on the dose metrics other than
AUC including the concentration profiles over time and Cmax
can be found in the Supplementary section. The predicted
AUCp and AUCd values were compared with the volume of
distribution (Vd) varying from the base case scenarios. Results
from gavage dosing are shown for a short half-life compound
as an example and unadjusted feeding for long half-life
Previous
TOC
-------
26
YOON AND BARTON
A)
Short half life, Gavage
1
.c
*
01
E
Vd=0.2 Vd=0.7
100 -
80 -
60 •
40 -
20-
I
I
1
(Base)
Oil
4 16 23 4 16 28
PND
Vd=2.5
ELJUl
4 16 28
B)
Long half life, Unadj Feed
Vd=0.2 Vd=0.7 Vd=2.5
5000 •
4000-
1 3000 •
*o>
E 2000 •
1000-
0-
r [
J
(Base)
flfil nfL-
4 16 28 4 16 28 4 16 28
PND
FIG. 6. Impact of varying Vd on predicted AUC values of the dam (open bars) and pups (closed bars) for two selected exposure scenarios. Other
pharmacokinetic parameters were the same as for the base case (e.g., Pm = 1).
compound (Fig. 6). The smaller the Vd, the higher the relative
level of AUCp compared to AUCa. The impact of varying Vd
on AUCp was observed both for lactation and postweaning
period, in terms of absolute values. However, it should be
noted that the relative levels of AUCp and AUCa during
postweaning were the same regardless of Vd values, although
the absolute values were different (Fig. 6). The impact of
changing Vd on the relative AUCp to AUCa during lactation
was more pronounced for the case of the long half-life
compound. With the smallest Vd used in the model, AUCp
values were greater than AUCa during lactation by more than
twofold (Fig. 6B). Similar trends were observed for the other
two dosing methods (data in Supplementary section). Com-
paring dosing methods, the relative level of AUCp to AUCa
was greater in gavage dosing simulations, though to only
a small extent, compared to the two dietary methods for all Vd
cases. Unadjusted and adjusted feeding methods resulted in
similar relative exposure levels for the pups compared to the
mother for all Vd used.
Changes in Milk Partitioning
Three Pm values in addition to the base case with milk
concentrations equal to maternal concentrations (Pm =1) were
assessed—milk concentrations 10-fold lower (Pm = 0.1),
threefold higher (Pm = 3), and 10-fold higher (Pm = 10) than
mother's plasma concentrations. Predicted AUC values using
gavage dosing for the short half-life compounds and adjusted
feeding for the long half compounds are illustrated in Figure 7,
while predictions for the other two dosing regimen are listed in
the Supplementary section. The impact of varying Pm on the
pup exposure was observed mostly during lactation as shown
by increased (or decreased) AUCp values observed on PNDs4
and 16 with higher (or lower) Pm, while PND28 values were
unaffected as this is 7 days after exposure to milk ceased. For
short half-life compounds, increasing Pm resulted in less
difference in internal exposures for dams (AUCa) and pups
(AUCp) (Fig. 7A). The AUCa values decreased with increasing
Pm, which also contributed to reducing the difference between
mother and pups though the extent of this decrease was smaller
than the corresponding increase in AUCp. Long half-life
compounds exhibited pup exposures ranging from much lower
than maternal to much greater than maternal with increasing
Pm, which substantially contributed to the decline in the
maternal exposure with increasing milk transfer (Fig. 7B). The
other two dosing methods produced similar trends (see
Supplementary section). When comparing dosing strategies,
A)
Short half life, Gavage
B)
Long half life, Adj Feed
35 Pm=0.1 Pm=1 Pm=3 Pm=10 All Pm=0.1 Pm=1 Pm=3 Pm=10 All
28 -
^ 21-
^
•K
Ul ...
£ u'
7 •
(Base)
i
1400-
1050-
_j
£ 700-
O)
£
350-
(Base)
1
1 1
nil
J:
;
i
4 16 4 16 4 16 4 16 28 4 16 4 16 4 16 4 16 28
PND PND
FIG. 7. Impact of varying Pm on predicted AUC values of the dam (open bars) and pups (closed bars) for two selected exposure scenarios. Other
pharmacokinetic parameters were the same as the base case (e.g., Vd = Vp = 0.7).
Previous
TOC
-------
BIOLOGICAL MODELING OF RAT EARLY LIFE DOSIMETRY
27
A) Short half life compound
6 t -- Prenatal
— Delayed
5-
4 -
B)
fl)
g 280 •
0)
§ 210-
£t
6? 140-
0.
0 -
4 16 28
PND
70-
60-
50-
40 -
30-
20-
10-
0
Long half life compound
• Prenatal
- Delayed
350
0)
I/)
g 280
1 210-1
-Q
g? 140 -
70-
o
14
PND
21
28
16
PND
28
FIG. 8. Predicted concentration of the pups with delayed development of elimination. These plots present the results using unadjusted feeding. AUCp is shown
as % of AUCp from the corresponding half-life base case simulation on each selected PND. The base case assumes prenatal development of elimination to adult
capacity.
the relative level of AUCp to AUCa was greater in gavage
dosing simulations, although to a small extent, compared to the
two dietary methods, for all varied Pm cases. The relative level
of AUCp to AUCd was almost the same for unadjusted feeding
and adjusted feeding methods for each simulation case (see
Supplementary section).
Developmental Delay in Elimination
In order to compare the base case with a different pattern for
postnatal development of elimination capacity, one possible
scenario of a developmental delay was simulated (DevKe in
Supplementary section). The delay was described with
a Michaelis-Menten profile (i.e., a rectangular hyperbola) with
half maximal capacity on PND7. Figure 8 shows the
concentration profiles over time in pups with delayed de-
velopment of elimination compared with those born with adult
capacity (base case) for unadjusted feeding exposures. The
delay in elimination capacity in the pups resulted in higher
concentrations and AUCp compared to the base case. The
greatest impact was predicted to occur during the first week
with a two- to threefold increase in AUCp for the short half-life
chemical (Fig. 8A) and a smaller increase for the long half-life
chemical (Fig. 8B). The impact was reduced as the elimination
capacity approached adult levels for both half-life cases. The
impact on peak concentrations appeared to be more persistent.
Other dosing approaches showed similar trends, in terms of the
timing of the maximum effect on AUC, the profiles of the
impact over time, and greater impacts in the short half-life case,
though the extent of maximum effect varies according to the
different dosing methods and half-life of a chemical.
Role of Excreta Recirculation
Concentrations in the pups with or without recirculation
were compared using predictions from adjusted feeding
simulations in Figure 9. Excreta recirculation between the
dam and pups had a small impact, with a maximum effect of
a 10-30% increase in AUCp for the long half-life case only
(Fig. 9B). Cp was higher than in the base case during the
second week for the long half-life chemical simulation. The
impact of recirculation on concentrations and thus AUCp for
the short half-life chemical was negligible (Fig. 9A). Results
for other exposure methods were similar (see Supplementary
section).
Predicted Pups' Daily Dose
Daily doses to the pup (mg/kg/day) were calculated using
the model and compared to the daily dose administered to the
dam for three selected PNDs (Fig. 10). Daily doses to the pup
(mg/kg/day) from six different cases were graphed and
compared to the predicted dam's daily doses, with gavage
Previous
TOC
Next
-------
28
YOON AND BARTON
Short half life compound
B)
01
in
a
jjj 100-
15
ft
a so -
O
3
0 -
4 16 28
PND
60-
45-
o»
£ 30-
15-
Long half life compound
- Base
- Recirculation
14
PND
21
28
01
in
ra
o
« 100-
as
J3
l-o-
O
< 0-
4 16 28
PND
FIG. 9. Predicted concentration in pups with recirculation of excreta simulated. Simulation results using adjusted feeding are presented. AUCp is shown as %
of AUCp from the corresponding half-life base case simulation on each selected PND.
providing the targeted 15 mg/kg/day. Overall, the amount of
chemical delivered per kilogram body weight of pups per day
was smaller than that of the dam during lactation for short half-
life compounds for all simulation cases (Fig. 10A). Differences
between the dam's and the pup's doses during lactation were
smaller with a smaller Vd or a higher Pm for short half-life
chemicals (Fig. 10A). For long half-life compounds, pups
received comparable doses as the dam. The daily doses to the
pups were predicted to be higher than the dam's on PND4
when Vd was smaller or Pm was greater than 1 during the first
week of lactation, while in the later period of lactation these
cases were predicted to have similar levels as the dam's
(Fig. 10B). After weaning, gavage dosing was predicted to
provide the same dose to the dam and pups, while feeding was
predicted to result in a higher chemical dose to the pups
(Figs. 10A and 10B).
DISCUSSION
The purpose of this study was to answer the question, to how
much chemical are pups exposed during lactation and early
postweaning? Although pharmacokinetic information on lac-
tational transfer and dosimetry in developing rat pups has been
noted as a critical data need in extrapolating early life toxicity
studies to humans, only limited information is available
(Barton et al., 2006). Pharmacokinetic studies involving early
life stages are technically challenging and approaches are less
standardized than for adults, even for pharmaceuticals. The
current model demonstrated that it could provide insights into
what dosimetry would be during lactation and early post-
weaning periods in dams and pups. These insights can be
applied to pharmacokinetic and toxicity study design, in-
terpretation of toxicity study results, and risk assessment
applications.
The present model for rats was based on known changes in
biology and exposure characteristics during early postnatal
periods together with chemical kinetics adapted from adults.
Model evaluation against limited published lactational transfer
data for OTA and 2,4-D indicated that reasonable predictions
could be made using this information whether the chemical was
given via diet (2,4-D) or by gavage (OTA). While two- or
threefold discrepancies were observed by comparison with
some data, this was considered acceptable for obtaining
plausible initial estimates for a range of chemicals.
Model predictions were obtained to derive insights for
a range of considerations important in one-generation toxicity
settings, such as characterizing exposures during lactation,
comparison of pup exposures to the mother's, comparison of
exposures via milk to direct exposure after weaning, and
evaluation of alternative dosing approaches. The present study
clearly showed that exposure of the pups frequently does not
Previous
TOC
Next
-------
BIOLOGICAL MODELING OF RAT EARLY LIFE DOSIMETRY
29
A)
25 i
20
>v
a
5 15
O)
O) 10 -
5 -
0
Short half life compound
Gavage
B)
50
40
m
•C 30
20 -
Long half life compound
Gavage
1
40-
30-
20 -
10-
0
PND4
Unadj Feed
PND16
JT
PND28
I
PND4
PND16
75
60
ns
•a 45-
r°
PND28
Unadj Feed
PND4
PND16
PIMD28
PND4
PIMD16 PND2B
25-
20-
•5 15-
"§)
"a 10 •
E
5 •
Adj Feed
J
50-
40-
t 30-
0!
0> 20-
10-
o
1
?
/
/
/
/
/
/
/
/
/
/
/
Adj Feed
'
>
*
$
'
>
>
•
I
PN04 PND16 PND28 PND4 PND16 PN028
1 1 Dam BSH Pup-LgVd EE3 Pup-HighPrn
Pup-Base K^a Pup-LowPm Pup-RCC
nrm Pup-SmVd W7^ Pup-MidPm •• Pup-AN cases
FIG. 10. Predicted daily exposure dose (mg/kg/day) of the test chemical to the pups from milk or direct dosing. For each dosing method, seven different
exposure scenarios were simulated, and resulting pup daily dose were compared to the mother's dose for three selected days. For PND28, the daily doses to the
pups are the same for all cases, so only one column was plotted for each exposure scenario.
parallel maternal exposure. However, it was possible to
delineate several general features of the neonatal exposure
pattern. The extent of lactational exposure of the pups to short
half-life compounds was generally lower than that of the dam,
while it was often comparable or even higher than maternal
levels for longer half-life compounds. Factors such as a lower
concentration in milk than in maternal blood and a volume of
distribution larger than total body water tended to result in
lower pup exposures compared to dam and consequently
greater differences between maternal and neonatal exposures
for both half-life compounds. Less difference was predicted
between the maternal and neonatal exposures when the milk
concentrations were greater than maternal blood levels, and the
volume of distribution was smaller than total body water level.
With these characteristics, the pup exposure level could exceed
that of the dam for long half-life compounds. But, it also needs
to be noted that even when the test compound's half-life is
short, the pup exposure level could be comparable to the
mother, if there are factors that tend to elevate exposure, e.g.,
smaller volume of distribution, delay in development of
elimination, and recirculation. Pup exposures comparable to
the mother's for a short half-life compound would be more
likely with a dietary exposure regimen rather than gavage
because gavage tended to result in more pronounced differ-
ences between maternal and neonatal exposure. Recirculation
of excreta between the dam and the pups can contribute to pup
exposure for longer half-life compounds when elimination is
largely renal clearance, rather than by metabolism, by raising
maternal exposure. Recirculation would be a particularly
important factor to consider in cross-fostering studies where
the pups would expose the "nonexposed" foster mother.
Predictions from the present model can be used as initial
estimates for better designing and interpreting toxicity studies.
Current findings suggest that there are some cases for which
Previous
TOC
-------
30
YOON AND BARTON
one should think about alternative exposure methods other than
milk to achieve exposure to the pups. Particular attention
would be needed when the pup exposure was predicted to be
extremely low, e.g., less than 1% of maternal exposure, which
was observed when the milk concentration was much less than
the mother's blood concentration or when the volume of
distribution was large for a short half-life compound. In these
cases, one would need to be careful determining the windows
of susceptibility or interpreting the potency of the chemical. If
pup exposure was estimated from maternal levels for such
a chemical, then the period of lactation could be interpreted
incorrectly as not being a susceptible period when instead there
was simply a lack of adequate exposure in the pups and the
lactational period had effectively not been evaluated. If this
were the case, an alternative study design such as direct dosing
of the pups would need to be considered (Moser et al., 2005).
Conversely, if the chemical exhibited toxicity with this very
low level of milk transfer, then it can be suspected as a very
potent compound. Model predictions also could help in
amending the study design by providing information as to
whether toxicities seen at certain doses may be related to
excessively high pup exposure. During lactation, a notable
example of this case is high milk transfer of a chemical with
a long half-life.
The model predictions could be informative about whether
abrupt changes in exposure levels after weaning, due to the
initiation of direct dosing of pups, result in toxicity. For
example, predictions from the current modeling showed that
postweaning exposure was much higher than lactational
exposure when gavaging short half-life chemicals. This abrupt
increase in concentration in the pup could produce premature
death or other toxicities in the neonates right after weaning,
which could be misinterpreted as a critical window for the test
chemical. Long half-life chemicals tended to show a greater
similarity between lactational and postweaning exposures,
though not in every case (e.g., low Pm). In the postweaning
period, the pup exposure is predicted to be even higher than
adult exposure under dietary exposure conditions due to the
higher feeding rate of pups on a body weight basis compared to
adults, while the same when gavage dosing is used.
Collectively, initial estimates of pup exposure from the current
model could provide valuable information to understand
a chemical's effect based upon limited information on adult
pharmacokinetics and milk transfer to facilitate design of
follow-up pharmacokinetic or toxicity studies.
Although different dosing methods can lead to very different
pharmacokinetic outcomes (Saghir et al., 2006), there have
been few direct studies of the potential differences when the
dam is given a chemical via gavage versus dietary adminis-
tration. Arnold et al. (2000) reported that similar doses of
hexaclorobenzene or Aroclor 1254 administered by two
different dosing methods, gavage or feeding, did not result in
similar exposures in suckling neonates. The marked increase in
maternal food consumption during lactation was suggested to
account for the difference, which was also supported by the
current model predictions. The administration of chemicals at
different dose rates, as exemplified by gavage versus dietary
administration, could result in differences in the developmental
outcomes, particularly those dependent upon peak concentration.
The amount and pattern of lactational exposure from
adjusted dietary dosing for longer half-life compounds were
predicted to be more similar to gavage than unadjusted diet as
shown by similar internal exposure levels and peak concen-
trations before the onset of pup's independent feeding on solid
diet. Unadjusted feeding generally produced higher pup
exposure than these two approaches. These predictions will
help not only to choose an appropriate dosing approach to
achieve a pup exposure level similar to its mother's but also to
provide insight into what to expect when different dosing
methods are applied. Dietary exposure is often considered to be
the most relevant dosing approach for life-stage testing, unless
there is a technical problem, so we sought to determine for
what chemicals this dosing method would be most relevant.
When the chemical is rather slowly eliminated and the milk
transfer is moderate (i.e., the Pm being between 1 and 3) and/or
the volume of distribution is slightly smaller than the total body
water, then one can expect that the dam and pups would be
exposed at similar levels via feeding. Again, it should be taken
into consideration that the unadjusted diet exposure in a number
of cases for long half-life compounds will lead to higher
exposures than expected from the predetermined target dose
due to the increased maternal food intake during lactation.
Current modeling highlighted some critical data needs in
predicting postnatal dosimetry; time-dependent changes in milk
partitioning are one example. The milk concentration was
assumed to parallel maternal plasma levels by a constant
proportionality, the milk to plasma ratio, throughout lactation.
Known changes in milk during lactation may call into question
this assumption for at least some classes of chemical (Luckey
et al., 1954). This would matter especially for chemicals whose
partition to milk is highly affected by milk composition. Such
chemicals include highly fat-soluble compounds due to
significantly higher fat content in milk the first few days after
birth and its rapid decline thereafter, rngestion of highly
lipophilic compounds such as hexachlorobenzene and poly-
chlorinated biphenyls by suckling pups was observed to be
elevated shortly after birth correlated with high milk fat content
(Arnold et al., 2000). Secretion of chemicals that bind to milk
proteins may also be affected by such time-dependent changes
in milk composition. Thus, more accurate estimation of milk to
blood ratios for test compounds would be obtained by
measuring them experimentally throughout lactation. It also
would be important to determine milk composition throughout
lactation because currently available information is very
limited. With additional data, methods could be developed to
predict passive distribution of chemicals based upon milk
composition and the physical chemical properties of the
chemical.
Previous
TOC
-------
BIOLOGICAL MODELING OF RAT EARLY LIFE DOSIMETRY
31
While the current model has value as a tool to make initial
predictions, it is also plausible that adult pharmacokinetics
during lactation would differ from those in nonlactating adults
or that there were major age-dependent changes in the pups that
could impact a specific chemical. There are changes in fluid
handling during lactation in the dam, so distribution character-
istics or renal clearance of a chemical could be affected
although rats do not show decreased circulating albumin levels
until the final days of pregnancy, in contrast to humans (Stock
et al., 1980). Developmental pharmacokinetic changes in pups
can be extensive during the first weeks of life as exemplified by
changes in enzymes and transporters (Li et al., 2002; Yoon
et al., 2006). The developmental pattern of elimination capacity
was one possible example of developmental delay in
elimination capacity based on critical changes during neonatal
kidney maturation (Kavlock and Gray, 1982). If one knew
important pharmacokinetic determinants in the adult, e.g.,
metabolism by a specific enzyme, binding to a specific protein,
or movement via a transporter, this could potentially be
incorporated into the current model (See Supplementary
section for examples). As the complexity and detail increase,
it would be necessary to shift to using a physiologically based
pharmacokinetic model in place of the one-compartment model
used here. Collecting data to evaluate the model predictions for
any specific compound would be essential.
It will be necessary to modify the model to expand its
general uses as well. Examples would include accounting for
metabolism or transport and related saturation kinetics as well
as lactational transfer of metabolites, modeling very long half-
life compounds, and modeling premating and pregnancy
periods. The activity of saturable active transport systems can
be important for the pharmacokinetics of pharmaceutical and
environmental chemicals, including for distribution to milk.
For example, high lipophilicity is not the only factor that can
lead to high milk concentrations. The milk concentration of the
lipophilic carboxylic acid herbicide 2,4-D was shown to be
similar to maternal blood (Sturtz et al., 2006), while less
lipophilic (or more water soluble) compounds like antiviral
drugs have shown several fold higher milk concentrations
compared to mother's blood due to active transport in rats
(Alcorn and McNamara, 2002). Early life toxicity testing
usually involves the dam's exposure to the test chemical prior
to impregnation and continued exposure during gestation and
lactation. As a result, the neonate (and the fetus) is exposed to
the test chemical and its metabolites prenatally and during
lactation. Hence, simulation of the gestational carry over would
be critical in predicting lactational exposure accurately for very
long half-life compounds.
We have demonstrated a modeling approach to predict
maternal and pup external and internal exposures by combining
biological information in the literature on body weight
changes, milk production, and food consumption with adult
nonpregnant rat pharmacokinetics and information on milk
distribution. The resulting predictions can be used to (1) design
pharmacokinetic studies to evaluate the predictions, (2) evaluate
early life toxicity study design choices (e.g., exposure method),
and (3) develop hypotheses to interpret toxicity study findings.
Extending this approach by incorporating more pharmacoki-
netic determinants in the model for distribution to milk and
early postnatal pharmacokinetics would improve predictions
for specific chemicals as well as facilitate development of
further generalizations concerning the extent of pup exposures
for chemicals with different pharmacokinetic characteristics.
The current modeling analyses predicted substantial differences
between maternal and pup external and internal exposures
indicating that risk assessment approaches based upon maternal
exposure doses are of limited utility when considering early
childhood exposures.
SUPPLEMENTARY DATA
Supplementary section are available online at http://toxsci.
oxfordjournals.org/.
FUNDING
National Research Council Research Associateship Award at
the US Environmental Protection Agency to M.Y. (EPA-NRC
# CR82879001).
ACKNOWLEDGMENTS
The authors acknowledge Dr. Suzanne Fenton for her review and comments.
REFERENCES
Alcorn, J., and McNamara, P. J. (2002). Acyclovir, ganciclovir, and zidovudine
transfer into rat milk. Antimicrob. Agents Chemother. 46, 1831-1836.
Arnold, D. L., Bryce, F. R., Clegg, D. J., Cherry, W., Tanner, J. R., and
Hayward, S. (2000). Dosing via gavage or diet for reproduction studies:
A pilot study using two fat-soluble compounds-hexachlorobenzene and
aroclor 1254. Food Chem. Toxicol. 38, 697-706.
Barton, H. A. (2005). Computational pharmacokinetics during developmental
windows of susceptibility. /. Toxicol. Environ. Health A 68, 889-900.
Barton, H. A., Pastoor, T. P., Baetcke, K., Chambers, J. E., Diliberto, J.,
Doerrer, N. G., Driver, J. H., Hastings, C. E., lyengar, S., Krieger, R., et al.
(2006). The acquisition and application of absorption, distribution,
metabolism, and excretion (ADME) data in agricultural chemical safety
assessments. Crit. Rev. Toxicol. 36, 9-35.
Bornschein, R. L., Fox, D. A., and Michaelson, I. A. (1977). Estimation of
daily exposure in neonatal rats receiving lead via dam's milk. Toxicol. Appl.
Pharmacol. 40, 577-587.
Byczkowski, J. Z., Kinkead, E. R., Leahy, H. F., Randall, G. M., and
Fisher, J. W. (1994). Computer simulation of the lactational transfer of
tetrachloroethylene in rats using a physiologically based model. Toxicol.
Appl. Pharmacol. 125, 228-236.
Corley, R. A., Mast, T. J., Carney, E. W., Rogers, J. M., and Daston, G. P.
(2003). Evaluation of physiologically based models of pregnancy and
Previous
TOC
-------
32
YOON AND BARTON
lactation for their application in children's health risk assessments. Crit. Rev.
Toxicol. 33, 137-211.
Doerflinger, A., and Swithers, S. E. (2004). Effects of diet and handling on
initiation of independent ingestion in rats. Dev. Psychobiol. 45, 72-82.
Godbole, V. Y., Grundleger, M. L., Pasquine, T. A., and Thenen, S. W. (1981).
Composition of rat milk from day 5 to 20 of lactation and milk intake of lean
and preobese Zucker pups. /. Nutr. Ill, 480-487.
Gorzinski, S. J., Kociba, R. J., Campbell, R. A., Smith, F. A., Nolan, R. J., and
Eisenbrandt, D. L. (1987). Acute, pharmacokinetic, and subchronic lexicological
studies of 2,4-dichlorophenoxyacetic acid. Fundam. Appl. Toxicol. 9,423-435.
Griffin, R. J., Godfrey, V. B., Kim, Y. C., and Burka, L. T. (1997). Sex-
dependent differences in the disposition of 2,4-dichlorophenoxyacetic acid in
Sprague-Dawley rats, B6C3F1 mice, and Syrian hamsters. Drug Metab.
Dispos. 25, 1065-1071.
Hallen, I. P., Breitholtz-Emanuelsson, A., Hull, K., Olsen, M., and
Oskarsson, A. (1998). Placental and lactational transfer of ochratoxin A in
rats. Nat. Toxins 6, 43-49.
Hanley, T. R., Jr, Breslin, W. J., Quasi, J. F., and Carney, E. W. (2002).
Evaluation of spinosad in a Iwo-generalion dielary reproduction sludy using
Sprague-Dawley rals. Toxicol. Sci. 67, 144-152.
Hanley, T. R., Jr, and Walanabe, P. G. (1985). Measuremenl of solid feed
consumption patterns in neonalal rals by 141Ce-radiolabeled microspheres.
Toxicol. Appl. Pharmacol. 77, 496-500.
Henning, S. J. (1981). Poslnalal development Coordination of feeding,
digestion, and metabolism. Am. J. Physiol. 241, G199-G214.
Hinderliler, P. M., Mylchreesl, E., Gannon, S. A., Bulenhoff, J. L., and
Kennedy, G. L., Jr (2005). Perfluorooclanoale: Placenlal and laclalional
Iransport pharmacokinelics in rals. Toxicology 211, 139-148.
Howdeshell, K. L., Furr, J., Lambrighl, C. R., Rider, C. V., Wilson, V. S., and
Gray, L. E., Jr (2007). Cumulative effecls of dibulyl phlhalale and
dielhylhexyl phlhalale on male ral reproductive Iracl development Altered
felal steroid hormones and genes. Toxicol. Sci. 99, 190-202.
Kavlock, R. J., and Gray, J. A. (1982). Evaluation of renal function in neonalal
rals. Biol. Neonate 41, 279-288.
Knighl, C. H., Docherty, A. H., and Peaker, M. (1984). Milk yield in rals in
relation to activity and size of Ihe mammary secretory cell population.
/. Dairy Res. 51, 29-35.
Li, N., Hartley, D. P., Cherringlon, N. J., and Klaassen, C. D. (2002). Tissue
expression, ontogeny, and inducibilily of ral organic anion Iransporting
polypeplide 4. /. Pharmacol. Exp. Ther. 301, 551-560.
Li, S., Marquardl, R. R., Frohlich, A. A., Villi, T. G., and Crow, G. (1997).
Pharmacokinelics of ochratoxin A and ils metabolites in rals. Toxicol. Appl.
Pharmacol. 145, 82-90.
Luckey, T. D., Mende, T. J., and Pleasanls, J. (1954). The physical and
chemical characterization of ral's milk. /. Nutr. 54, 345-359.
McNamara, P. J., Meece, J. A., and Paxlon, E. (1996). Active Iransport of
cimelidine and ranilidine into Ihe milk of Sprague Dawley rals.
/. Pharmacol. Exp. Ther. 277, 1615-1621.
Moser, V. C., Walls, I., and Zoelis, T. (2005). Direcl dosing of preweaning
rodenls in loxicily testing and research: Deliberations of an ILSI RSI Expert
Working Group. Int. J. Toxicol. 24, 87-94.
Rayner, J. L., Enoch, R. R., Wolf, D. C., and Fenlon, S. E. (2007). Alrazine-
induced reproductive Iracl alterations after Iransplacenlal and/or laclalional
exposure in male Long-Evans rals. Toxicol. Appl. Pharmacol. 218, 238-248.
Redman, R. S., and Sweney, L. R. (1976). Changes in diel and patterns of
feeding activity of developing rals. /. Nutr. 106, 615-626.
Saghir, S. A., Mendrala, A. L., Bartels, M. J., Day, S. J., Hansen, S. C.,
Sushynski, J. M., and Bus, J. S. (2006). Slralegies to assess systemic
exposure of chemicals in subchronic/chronic diel and drinking water sludies.
Toxicol. Appl. Pharmacol. 211, 245-260.
Shirley, B. (1984). The food intake of rals during pregnancy and lactation. Lab.
Anim. Sci. 34, 169-172.
Slock, B., Dean, M., and Levy, G. (1980). Serum protein binding of drugs
during and after pregnancy in rals. /. Pharmacol. Exp. Ther. 212, 264-268.
Stole, V., Knopp, J., and Slolcova, E. (1966). Iodine, solid diel, water and milk
intake by laclaling rals and Iheir offsprings. Physiol. Bohemoslov. 15,
219-225.
Slrubbe, J. H., and Gorissen, J. (1980). Meal patterning in Ihe laclaling ral.
Physiol. Behav. 25, 775-777.
Slurtz, N., Bongiovanni, B., Rassello, M., Ferri, A., de Duffard, A. M., and
Duffard, R. (2006). Detection of 2,4-dichlorophenoxyacelic acid in ral milk
of dams exposed during lactation and milk analysis of Iheir major
componenls. Food Chem. Toxicol. 44, 8-16.
Timchalk, C. (2004). Comparative inler-species pharmacokinelics of
phenoxyacelic acid herbicides and related organic acids: Evidence lhal
Ihe dog is nol a relevanl species for evaluation of human heallh risk.
Toxicology 200, 1-19.
Yoon, M., Madden, M. C., and Barton, H. A. (2006). Developmental
expression of aldehyde dehydrogenase in ral: A comparison of liver and lung
development Toxicol. Sci. 89, 386-398.
Yuan, J. (1993). Modeling blood/plasma concenlralions in dosed feed and
dosed drinking water toxicology sludies. Toxicol. Appl. Pharmacol. 119,
131-141.
Zepnik, H., Volkel, W., and Dekanl, W. (2003). Toxicokinelics of Ihe
mycoloxin ochratoxin A in F 344 rals after oral adminislralion. Toxicol.
Appl. Pharmacol. 192, 36^4.
Previous
TOC
-------
Journal of Exposure Science and Environmental Epidemiology (2009), 1-10 ijfife
© 2009 Nature Publishing Group All rights reserved 1559-0631/09/S32.00 ^UP
www.nature.com/jes
Research needs for community-based risk assessment: findings from a
multi-disciplinary workshop
YOLANDA ANITA SANCHEZa, KACEE DEENERb, ELAINE COHEN HUBALC, CARRIE KNOWLTONa,
DAVID REIFC AND DEBORAH SEGALd
3Association of Schools of Public Health, Washington, DC, USA
bU.S. Environmental Protection Agency, Office of Research and Development, National Center for Environmental Assessment, Washington, DC, USA
CU.S. Environmental Protection Agency, Office of Research and Development, National Center for Computational Toxicology, Washington, DC, USA
dU.S. Environmental Protection Agency, Office of Research and Development, National Center for Environmental Research, Washington, DC, USA
Communities face exposures to multiple environmental toxicants and other non-chemical stressors. In addition, communities have unique activities and
norms that influence exposure and vulnerability. Yet, few studies quantitatively consider the role of cumulative exposure and additive impacts.
Community-based risk assessment (CBRA) is a new approach for risk assessment that aims to address the cumulative stressors faced by a particular
community, while incorporating a community-based participatory research framework. This paper summarizes an Environmental Protection Agency
(EPA) sponsored workshop, "Research Needs for Community-Based Risk Assessment." This workshop brought together environmental and public
health scientists and practitioners for fostering an innovative discussion about tools, methods, models, and approaches for CBRA. This workshop was
organized around three topics: (1) Data and Measurement Methods; (2) The Biological Impact of Non-Chemical Stressors and Interaction with
Environmental Exposures; and (3) Statistical and Mathematical Modeling. This report summarizes the workshop discussions, presents identified research
needs, and explores future research opportunities in this emerging field.
Journal of Exposure Science and Environmental Epidemiology advance online publication, 25 February 2009; doi:10.1038/jes.2009.8
Keywords: population-based studies, epidemiology, exposure modeling, analytical methods, empiricalfstatistical modeling, PBPK modeling.
Introduction
Communities face myriad exposures to environmental
toxicants and other non-chemical stressors. Although
environmental epidemiology studies do consider multiple
risk factors, few studies quantitatively considered the full
range of complex interactions between the multiple environ-
mental agents (chemical, biological, and social stressors)
within a targeted population or within a geographic area in
influencing health outcomes. The handful of quantitative
cumulative risk assessments, conducted to date, consider only
the additive impacts of chemical agents that share a common
mode of action (Castorina et al., 2003; Payne-Sturges et al.,
2004a; Payne-Sturges et al., 2004b; Caldas et al., 2006; U.S.
Environmental Protection Agency, 2006a) or a common
exposure media (Fox et al., 2004; Teuschler et al., 2004). In
addition, some studies have investigated the interaction of
environmental stressors that lead to negative health outcomes
1. Address all correspondence to: Deborah Segal, US EPA, Office of
Research and Development, National Center for Environmental Research,
Room 3108, 1025 F. Street, NW, Washington, DC 20004, USA. Tel.:
+ 202 343 9797. Fax: + 202 233 0677. E-mail: Segal.deborah@epa.gov
Received 8 September 2008; accepted 29 December 2008
(Cary et al., 1997; Morrison et al., 1998; Erren et al., 1999).
These studies, however, provide little information on
susceptibility factors, interactive effects of biological re-
sponses, or social stressors that may modify toxic response.
The 1996 Food Quality and Protection Act expanded risk
assessment for evaluating chemical mixtures and environ-
mental contaminants that target similar body mechanisms
(U.S. Congress, 1996). Cumulative risk assessment was
discussed extensively in the International Life Sciences
Institute (ILSI, 1999) publication, titled A Framework for
Cumulative Risk Assessment, which focused on the cumula-
tive toxicity, exposure, and risk characterization of multiple
environmental contaminants. However, the methods for
cumulative risk assessment are still evolving (U.S. Environ-
mental Protection Agency, 2003; U.S. Environmental
Protection Agency, 2007a). The 2003 Environmental
Protection Agency (EPA) publication, Framework for
Cumulative Risk Assessment, widely expanded the definition
of cumulative risk assessment to include non-chemical
stressors and the concept of population vulnerability. The
EPA defined cumulative risk assessment as "an analysis,
characterization and possible quantification of the combined
risks to health and or the environment from multiple agents
or stressors (U.S. Environmental Protection Agency, 2003)."
Previous
TOC
Next
-------
Sanchez et al.
Research needs for community-based risk assessment
The key aspects of the EPA Framework for Cumulative
Risk Assessment are: (1) understanding the combined effects
of more than one agent/stressor; (2) considering non-
chemical stressors; (3) focusing on identifying and character-
izing vulnerable human and ecological populations; and
(4) using a place-based or population-based analysis for risk
assessments, which elicits community expertise. This EPA
framework generated a paradigm shift in risk assessment by
greatly expanding the concept of an environmental "stressor"
to include chemical, biological, physical, and psychosocial
agents (U.S. Environmental Protection Agency, 2003). The
framework also includes the concept of population vulner-
ability — certain disadvantaged, underserved, and over-
burdened communities face conditions that can exacerbate
environmental burdens. More specifically, mechanisms of
vulnerability identified in the framework include differences
in individual or population susceptibility, exposure, prepa-
redness, and ability to recover (U.S. Environmental Protec-
tion Agency, 2003; National Environmental Justice
Advisory Committee, 2004).
Communities generally share a common geographic
location and/or common experience (traditions, diet, beha-
vioral norms). Therefore, cumulative risk assessment for
communities is inherently place based (for a discussion on
place-based public health, see (Patychuk, 2007; Yeboah,
2005). Community norms influence diet and activities that
determine how individuals come into contact with environ-
mental contaminants (U.S. Environmental Protection
Agency, 2003, 2006b; National Environmental Justice
Advisory Committee, 2004). In addition, non-chemical
stressors may affect the health outcome of exposure to
environmental contaminants (White et al., 2007). For
example, stress has been shown to exacerbate lead toxicity
(Bellinger et al., 1988; Tong et al., 2000; Cory-Slechta et al.,
2004). The EPA has reflected this understanding of place-
based cumulative risk assessment in the 2006 Human Health
Multi-Year Plan, by expanding the long-term goal of
"Research on Cumulative Risk" to include research on
community-based cumulative risk assessment. This includes
the development and application of tools and approaches for
assessing community risk, and the application of community-
based tools, as well as approaches for assessing exposure to
environmental contaminants and non-chemical stressors
(U.S. Environmental Protection Agency, 2006b; U.S.
Environmental Protection Agency, 2007a).
Therefore, community-based risk assessment (CBRA) is
defined here as a model of risk assessment that addresses
multiple chemical and non-chemical stressors faced by a
particular community, while incorporating a community-
based participatory research framework and a transparent
process to instill confidence and trust among the community
members (U.S. Environmental Protection Agency, 2003,
2007b; National Environmental Justice Advisory Committee,
2004). CBRA may include characteristics of a community that
cannot be identified and assessed through traditional risk
assessment paradigms, such as social and cultural dynamics
of the community or resources, strengths, and relationships
within the community (Israel et al., 1998). Community and
stakeholder involvement is critical in harnessing community
knowledge and to better understand complex cumulative
exposures (U.S. Environmental Protection Agency, 2003;
National Environmental Justice Advisory Committee, 2004;
Menzie et al., 2007). Traditionally, this community knowl-
edge has been difficult to obtain using conventional research
and risk assessment methods (Israel et al., 1998; Israel et al.,
2001; O'Fallon and Dearry, 2001; Corburn, 2005). Com-
munity-based participatory research frameworks can aid
efforts to involve the community, and can integrate insightful
community information that can advance environmental
health research (O'Fallon and Dearry, 2002; Corburn, 2005;
Israel et al., 2005) and risk assessment.
CBRA reflects the recommendations put forth in the
National Environmental Justice Advisory Council's
(NEJAC) report, titled "Ensuring Risk Reduction in
Communities with Multiple Stressors: Environmental Justice
and Cumulative Risks/Impacts," and echoes the interests of
the EPA's Office of Environmental Justice and Office of
Children's Health Protection (National Environmental
Justice Advisory Committee, 2004; U.S. Environmental
Protection Agency, 2006b). CBRA evolved from the EPA
and the NEJAC publications regarding frameworks for
cumulative risk assessment, which indicated the need for
dealing with risk on a community-by-community basis (U.S.
Environmental Protection Agency, 2003).
Including CBRA in the regulatory decision-making
process poses some significant challenges. These challenges
include the need to assess toxicity of mixtures, measure
vulnerability of populations, and to evaluate interactions
among multiple stressors, chemical or non-chemical (U.S.
Environmental Protection Agency, 2003). Additional chal-
lenges are associated with the need to partner with
stakeholders for obtaining community knowledge. The
traditional EPA risk assessments evaluate the hazardous
properties of substances, assess the extent of human
exposure, and characterize the risk of adverse health effects
(National Research Council, 1983). Many of these risk
assessments aim to protect the most sensitive individuals and/
or groups in the general population. In contrast, CBRA
would characterize additional community-level stressors and
measures of vulnerability to help inform risk evaluation and
decision-making at the local level. This may include how
community-level stressors and vulnerability factors interact
with environmental contaminant exposures to impact the
overall risk to the individuals within the defined community.
To address the scientific challenges inherent in CBRA, the
Office of Research and Development (ORD) at EPA
sponsored a workshop, titled "Research Needs for Commu-
nity-Based Risk Assessment." This workshop focused on
Journal of Exposure Science and Environmental Epidemiology (2009), 1-10
Previous
TOC
Next
-------
Research needs for community-based risk assessment
Sanchez et al.
three topics: (1) Data and Measurement Methods; (2) The
Biological Impact of Non-Chemical Stressors and Interac-
tion with Environmental Exposures; and (3) Statistical and
Mathematical Modeling. (U.S. Environmental Protection
Agency, 2007b). This paper provides an overview of the
workshop, presents research needs identified based on results
of the workshop, and highlights themes to further advance
the science behind CBRA.
Workshop overview
From 18 October 2007 to 19 October 2007, in the Research
Triangle Park, NC, the EPA's ORD sponsored the
"Research Needs for Community-Based Risk Assessment"
workshop. This multi-disciplinary workshop was coordi-
nated by a small organizing committee that developed four
basic questions regarding CBRA, which became the frame-
work of the workshop:
• What research has been conducted?
• What is the current state of the science?
• What are the research needs?
• How can community-based information be quantified in a
way that is useful for EPA risk assessments?
Approximately 85 people attended the workshop. Partici-
pants included the EPA employees, contractors, or fellows,
spanning eight offices and five regions. Other participants
were affiliated with the National Institute of Environmental
Health Sciences (NIEHS), academia, other research insti-
tutes, local government, or community advocacy groups.
Before the workshop, participants were encouraged to read
the 2007 Environmental Health Perspectives Mini-Mono-
graph on cumulative risk assessment (Callahan and Sexton,
2007; DeFur et al., 2007; Menzie et al., 2007; Ryan et al.,
2007; Sexton and Hattis, 2007). The workshop was divided
into three topics, as described earlier. Day one of the
workshop included presentations on the state of the science
for each topic. Small breakout sessions on each topic
occurred on day two, at which 10-20 participants discussed
the research needs for their respective topics. Session chairs
compiled and summarized the research needs identified by the
breakout group participants, and these were presented back
to the full group for further discussion. A summary report
that included details of the workshop presentations and
discussion was prepared (U.S. Environmental Protection
Agency, 2007b).
Session I: Data and Measurement Methods
Measuring chemical and non-chemical stressors, susceptibil-
ity factors, and health outcomes at the community-level will
play an important role in CBRA. Some examples include,
measuring personal and community exposures to multiple
chemical stressors, monitoring the time-activity behavior,
Journal of Exposure Science and Environmental Epidemiology (2009), 1-10
measuring markers of susceptibility, and tracking early health
outcomes. This session explored currently available tools and
methods.
Session I included three presentations: "Development of
Nanoscaled Sensor Systems for Detecting and Monitoring
Environmental Chemical Agents" by Desmond Stubbs of
Oak Ridge Center for Advanced Studies (ORCAS); "Data
Collection Platforms for Integrated Longitudinal Surveys of
Human Exposure-Related Behavior" by Paul Kizakevich of
RTI International; and "Assessment Methods for Commu-
nity-Based Risk Assessment" by Elaine Faustman of the
University of Washington.
The first speaker, Dr. Stubbs, focused on the application of
emerging technologies for measuring exposures to chemical
stressors. He summarized results of an earlier workshop co-
sponsored by the EPA and the ORCAS as the background
and context for his presentation (U.S. Environmental
Protection Agency and Oak Ridge Center for Advanced
Studies, 2006c). Both the EPA and the NIEHS identified the
need for a rugged, lightweight, low-cost, wearable, real-time
sensor capable of multi-analyte detection with minimal
burden to the individual. The "gold standard" was defined
as the ability to simultaneously detect multiple chemical
agents in the field with the same sensing system and to link
this data to a specific biological event. Such a device would be
capable of remote data acquisition, location recording, and
measurement of both the concentration and frequency of
environmental exposure. Dr. Stubbs identified the ongoing
research on several devices for use in exposure assessment,
including passive radio-frequency identification tags, an
electronic nose (i.e., "dog-on-a-chip"), microelectromagnetic
sensors, and interferometric optical sensors. He then discussed
microfabricated cantilever array platforms and the potential
for these to provide lightweight, wearable multi-analyte
sensors. Dr. Stubbs also described the possibility of linking
the pea-size sensing and telemetry unit to a receiver unit the
size of a small personal digital assistant, designed to be carried
in a pocket. The personal digital assistant unit could have
analysis and display capability, and support global position-
ing and bio-monitoring device interfaces. Preliminary results
suggest that these devices are capable of real-time detection
(sub-second scale) of low vapor pressure chemical compounds
in the subparts per billion range. The potential power of these
new small-scale technologies for measuring personal and
community-level exposures to a wide range of chemical
constituents and stressors was recognized by many CBRA
workshop participants.
The second speaker, Dr. Kizakevich, focused on ap-
proaches for measuring exposure-related behaviors for
assessing risk. He presented details on the development of a
system that integrates multiple real-time data collection
streams and survey modes on a handheld Pocket PC
platform (Whitmore and Kizakevich, 2004). The objectives
of this research are to develop, validate, and evaluate
Previous
TOC
Next
-------
Sanchez et al.
Research needs for community-based risk assessment
innovative methods for the TALE (time/activity/location/
exertion-level) data, dietary consumption data, and data on
use of consumer products, including pesticide products,
household cleaning products, and personal care products.
The system integrates diaries and questionnaires with a
collection of wireless peripheral devices for monitoring
physical and physiological data.
The RTI researchers are also exploring different methods
for collecting data and evaluating these methods using
feedback from the study population. Three Pocket PC diary
modes were studied: interactive menus, voice questionnaires,
and passive periodic photos. Innovations, such as passive
microenvironment identification (i.e., beacons), passive exer-
tion assessment, wireless product use event markers, wireless
interfaces, intelligent prompting, GPS tracking, and auto-
mated daily review for collecting the data both accurately and
with a low participant burden are also being investigated. The
system design emphasizes easy reconfiguration for supporting
varied study requirements, investigator needs, and partici-
pant preferences. Dr. Kizakevich noted that data collected
during piloting of these approaches will be made available
after the next round of monitoring. Originally, the goal was
to determine the best method for collecting the exposure-
related behavior data. However, based on the first round of
evaluation, it is clear that the best technology for collecting
behavior data will be determined by the objectives of a given
study or risk assessment. Results of the RTI research will
provide information that will allow investigators and
communities to determine which method is best for their
needs.
In the third presentation, Dr. Faustman considered study
approaches and data requirements for characterizing the
exposure and risk factors to assess individual- and commu-
nity-level risks. She presented three types of studies
conducted by the University of Washington investigators to
understand pesticide exposures in children (Thompson et al.,
2003, 2008; Vigoren et al., 2007). The three studies presented
were a community-based participatory research project, a
longitudinal multiple sampling project aimed at understand-
ing between- and within-family variability, and a longitudinal
cohort study. Dr. Faustman also identified the importance of
collaboration between researchers and community members,
by presenting these three study examples. Throughout her
presentation, Dr. Faustman emphasized the need for study
designs to integrate the wide range of data required to
conduct CBRAs. Although the researchers typically had
access to general statistics on pesticide usage, an important
insight into the potential sources and pathways was obtained
from community participants that proved integral for
understanding exposures. Her final message focused on the
need to develop and incorporate biomarkers of exposure,
susceptibility, and effect into studies for identifying vulner-
able groups and to understand risks. Genomic and gene
expression analysis technologies are being applied in some of
the studies by the University of Washington, and have the
potential to improve prediction of exposure-response and at-
risk individuals in communities.
These presentations provided insight into the tremendous
challenges and wide range of data needs associated with
characterizing stressors for CBRA. All the speakers identified
the need for efficient tools for monitoring personal exposures
to better identify vulnerable groups, understand significant
exposure pathways, and develop targeted interventions.
Novel measurement methods for monitoring environmental
stressors (small-scale sensors), collecting exposure-related
behavior data (wireless, real-time survey methods), and
developing biomarkers of exposure, susceptibility, and effect
(genomic and gene expression analyses) were highlighted in
the context of CBRA.
Session II: The Biological Impact of Non-Chemical
Stressors and Interaction with other Environmental
Exposures
There is a recognized need to incorporate non-chemical
stressors into cumulative risk assessment (U.S. Environmen-
tal Protection Agency, 2003; National Environmental Justice
Advisory Committee, 2004). Most public health research on
non-chemical stressors have focused on the health effects
from exposure to chronic stress (Negro-Vilar, 1993; Bjorn-
torp, 2001; Kramer et al., 2001; Maccari et al., 2003; Strine
et al., 2004; Wright, 2005; Tamashiro et al., 2007; Suglia
et al., 2008). However, CBRA provides an opportunity to
investigate the non-chemical stressors that might interact with
environmental contaminants. Therefore, this session focused
on understanding the health impacts of non-chemical
stressors, specifically chronic stress, and their ability to
interact with exposure to environmental toxicants affecting
the risk of adverse health outcomes.
Session II included three presentations: "Social Stress,
Stress Hormones, and Neurotoxins", by James Herman of
the University of Cincinnati; "Intersections of Social
Ecology, Neurobehavioral Development, and Environmental
Contamination" by Bernard Weiss of the University of
Rochester; and "Social Environment as a Modifier of
Chemical Exposures" by Robert Wright of the Harvard
University School of Public Health.
Dr. Herman described the biological systems that mediate
stress responses. Herman and Seroogy (2006) had broadly
defined stress as a "real or perceived threat to homeostasis."
The secretion of glucocorticoid hormones, particularly
cortisol, function to return the body to homeostasis after
stress. However, a prolonged secretion of cortisol and other
glucocorticoids due to chronic stress inhibits neurogenesis.
This can contribute to deleterious effects on the body and
brain, including immune system dysfunction, depression, and
cognitive decline (Herman and Seroogy, 2006). Dr. Herman
also highlighted that this process can exacerbate other
effective disease states, such as schizophrenia and bipolar
Journal of Exposure Science and Environmental Epidemiology (2009), 1-10
Previous
TOC
Next
-------
Research needs for community-based risk assessment
Sanchez et al.
disease. He also emphasized the potential for the interaction
of specific environmental neurotoxicants and chronic stress,
because they both represent "hits" on a target system in the
multi-hit hypothesis of toxicity (White et al., 2007). More-
over, both environmental neurotoxicants and chronic stress
can modulate glucocorticoid secretion, which can work
together to potentiate the effects on nerve cells and neurons.
This phenomenon has been shown by exposure to lead and
chronic stress (Bellinger et al., 1988; Tong et al., 2000; Cory-
Slechta et al., 2004; Bellinger, 2008).
Dr. Weiss, the second speaker, focused on the interaction
between exposure to neurotoxicants and social disadvantage,
referring to his review of children's vulnerability to environ-
mental contaminants (Weiss and Bellinger, 2006). He used
lead exposure as a case study and discussed the sometimes-
difficult-to-quantify effects on intellectual quotient and
behavior associated with lead exposure. Dr. Weiss also
emphasized the disparate exposure of people of lower
socioeconomic status to both lead and chronic stress,
explaining that lead is only one example of differential
exposure of neurotoxins to this population. Other examples
of differential exposure to neurotoxins include environmental
tobacco smoke (Barbeau et al., 2004a; Barbeau et al.,
2004b), pesticides (Sexton et al., 2006), and mercury (Payne-
Sturges and Gee, 2006).
The third speaker, Dr. Wright, integrated the information
presented by the other speakers with his detailed discussion of
the effects of environmental contaminants on neuronal
function. He also used lead exposure as a case study. Lead
exposure stimulates neurotransmitter release causing inap-
propriate firing of neurons and the blockage of calcium
channels required for proper neuron function. Citing
numerous animal studies, Dr. Wright illustrated that positive
social environments can mitigate the effects of environmental
toxicants, such as lead (Guilarte et al., 2003; Weaver et al.,
2004). Similar studies in humans have shown that social
determinants can alter susceptibility to environmental con-
taminants (Tong et al., 2000; Clougherty et al., 2007). A new
birth cohort, the Early Life Exposure in Mexico to
Environmental Toxicants Project (ELEMENT), has been
established for investigating these interactions. This project
will examine stress, lead exposure, iron deficiency, and
neurodevelopment with a holistic perspective. The long-term
goals of ELEMENT are to: (1) identify factors that increase
and/or decrease metal toxicity; (2) understand the biology
of metal neurotoxicity; (3) prevent toxicity; and (4) treat
toxicity after it has occurred, by finding the appropriate
intervention(s).
Together, these three presentations provided a compre-
hensive picture of current knowledge about how the brain
responds to chronic stress and how this response can interact
with exposure to environmental contaminants. This over-
view helped set the stage for the breakout sessions, which
charged the session participants for identifying gaps in our
Journal of Exposure Science and Environmental Epidemiology (2009), 1-10
understanding of the connections between the chronic stress
and environmental contaminants toxicity.
Session III: Statistical and Mathematical Modeling
There are statistical and modeling challenges involved in
viewing organisms and the environment as they really are an
integrated whole. Traditional biostatistical approaches, such
as linear regression, data stratification or transformation, and
others are useful, yet have important limitations when
handling high-dimensional data of disparate types. The
Session III discussions included integrating data that vary
across space and time, pooling datasets drawn from multiple
sources, and creating accessible and user-friendly methods for
public participation.
This session included three presentations: "Community
Based Risk Assessment—A Statistician's Perspective" by
Louise Ryan of the Harvard School of Public Health; "A
Multi-Site Time Series Study of Hospital Admissions and Fine
Particles: A Case-Study for National Public Health Surveil-
lance" by Francesca Dominici of the Bloomberg School of
Public Health at Johns Hopkins University; and "Risk
Assessment!Risk Communication: Understanding the Commu-
nity" by Thomas Schlenker of the Public Health Depart-
ment of Madison-Dane County, WI.
The first speaker, Dr. Ryan, discussed examples of
community-focused research studies that were similar in
terms of having sparse data, a clever combination of data
from multiple sources, and the inclusion of spatiotemporal
modeling in the study designs. The most successful studies
integrated both personal and community-level data to
overcome issues of sparse data and unknown confounding
factors (Ryan, 2008). As uncertainty tends to be large when
dealing with data collected in real-world communities, it is
important to measure characteristics of the community in
addition to individuals. Appropriate statistical techniques,
such as spatiotemporal and hierarchical models, are of great
practical use in such studies that require synthesis of
information from multiple sources. However, researchers
must be cautioned against overinterpreting model results and
placing too much emphasis on f-values disconnected from
other relevant information. For complex problems, the
results must undergo rigorous sensitivity analyses in order
to fine-tune the models. Dr. Ryan called for continued work
for developing tools capable of combining information
measured on multiple scales and degrees of uncertainty, so
that the community-based models are robust with respect to
time, space, and other perturbations.
Dr. Dominici, the second speaker, discussed the utility of a
national system for tracking population health. She stated
that population health research could be advanced rapidly by
integrating the existing databases (each containing separate
information on environmental, social, and economic factors
that impact health) and by designing new statistical models to
describe the associated risk factors. Dr. Dominici highlighted
Previous
TOC
Next
-------
Sanchez et al.
Research needs for community-based risk assessment
how multi-site studies comparing day-to-day variations in
hospital admission rates with day-to-day variations in
pollution levels within the same community are used to
estimate city-specific pollution effects relative to confounding
effects, such as trend, season, and weather (Dominici et al.,
2006). Results have indicated that effects are consistent
across location, and that there is a lag between air pollution
exposure and respiratory effect. These preliminary results
indicate that flux in the levels of air pollution affect health.
Such studies provide an impetus for linking national
databases and developing appropriate analysis methods to
investigate risk at the local level. Owing to the small
attributable risk for air pollution and the large number of
potential confounders, single-site studies generally display
increased statistical error. Therefore, a national system for
analyzing data from multiple locations in a systematic
fashion is necessary to reliably assess population health.
Dr. Schlenker, the final session speaker, emphasized that
accurate and valid risk assessments cannot be carried out
unless there is an understanding of the community and
communication between the community and researchers. He
illustrated this point with examples of community-based
studies involving lead and manganese, in which commu-
nicating a story about the "life" of these metals in the body
instead of merely providing data and scientific jargon about
internal disposition was crucial to success (Schlenker, 1989).
Dr. Schlenker proposed that providing examples of how a
model has been (or can be) used at the community level is the
best way to take complicated models and move them into a
context, in which they can be trusted and understood by the
community. The research can benefit fully from community
guidance and case-specific advice, by including communities
in the analysis process.
Collectively, the speakers for Session III identified several
key requirements for successful modeling for CBRA—
including data collected across spatiotemporal scales, in-
formation on multiple communities for elucidating commu-
nity-specific risk factors, comprehensive community
involvement, and appropriate statistical analysis methods.
The speakers identified the need for continued development
of methods for analyzing disparate data types, integrating
existing (and nascent) databases, and working to mean-
ingfully include communities in all stages of research.
Workshop results
Emerging Themes
The major outcome of the workshop is the resulting list of
research needs (see Table 1) and a list of suggestions to
enhance CBRA (see Table 2) elicited from the summary
document. Many broad ideas were mentioned in more
than one workshop session topic, suggesting emerging and
crosscutting themes. The three overarching themes, which
are inclusive of the individual research needs, were identified:
(1) scientific tools and methods to better measure and
evaluate exposures and health outcomes at a community
level; (2) environmental health infrastructure; and (3) com-
munity involvement processes.
The need for scientific tools (methods and models) to
better measure and evaluate exposures and health outcomes
at a community level was identified throughout the work-
shop. On the basis of workshop presentations of the state of
the art for monitoring and modeling, it was clear that some
very sophisticated tools are available and that much of the
research effort could be focused on adapting and applying
these tools to the specific objectives of a CBRA. In Session I,
the potential for emerging monitoring technologies to provide
low-burden, real-time data on the full range of community
environmental stressors was identified. Furthermore, Session
III participants suggested that statistical techniques are
needed to better evaluate health outcomes at the community
level, including techniques for synthesizing information from
multiple datasets, reduce limitations of small population size,
and characterize group-level effects. Workshop participants
also identified the need for adjustments to the traditional risk
framework. Session II participants recommended amend-
ments in the risk paradigm to incorporate vulnerability and
non-chemical stressors. Session I participants identified a
need for methodology and modeling changes to include
qualitative data and incorporate social (e.g., poverty, access
to medical care, chronic stress) variables as modeling
parameters. A major outcome of the workshop was the
recognition that a new conceptual model for risk assessment
may be needed.
A second theme that emerged throughout the workshop
was the need for an environmental health infrastructure to
address the current gaps in data and data accessibility to
foster multidisciplinary research required for CBRA. Session
III participants suggested a better infrastructure is needed to
create an enhanced access to existing databases and develop
transparent modeling methods for diverse disciplines, entities
of government, research groups, and community organiza-
tions. All the three sessions advocated for an enhanced
infrastructure, which could ensure multiple levels of local,
state, tribal, and federal entities working together on CBRA.
In addition, all the sessions resulted in the acknowledgment
of the need to facilitate cross-disciplinary teams within public
health practice and research, social science, and environ-
mental health science.
A final reoccurring theme was the need for community
involvement. This will require the establishment and fostering
of effective working relationships between the community
and researchers in addition to the community and govern-
ment agencies. This may involve training on the options
available for community involvement (such as the use of a
community-based participatory research framework) within
government agencies and among research institutes. In
Journal of Exposure Science and Environmental Epidemiology (2009), 1-10
Previous
TOC
Next
-------
Research needs for community-based risk assessment
Sanchez et al.
Table 1. Research needs for community-based risk assessment (CBRA) by workshop session and emerging theme
Emerging themes
#1 #2 #3
Session I: Data Needs and Measurement Methods for CBRA X
Develop metrics, indicators, and biomarkers for exposure and health tracking surveillance
Develop simple and low-cost monitoring methods for pollutants and pathogens at the individual and community level over space and X
through time (including real time)
Develop simple and low-cost monitoring methods for non-chemical stressors at the individual and community level over space and X
through time (including real time)
Develop enhanced sensor technologies for providing real-time data on individual and community level measures of exposure to X
environmental stressors
Create accessible and well-documented databases with links to the full range of exposure information, to include an infrastructure for X
facilitating addition of data by investigators and the sharing of data and tools used to characterize environmental stressors
Identify and adapt indices used currently in social sciences for measuring community-level psychosocial health X
Translate more qualitative social indices into a form that is useful for quantitative risk assessments X
Session II: The Biological Impact of Non-Chemical Stressors and Interaction with Other Environmental Exposures
Review social variables of importance for health in the context of the EPA risk assessment X
Develop approaches for incorporating vulnerability into risk assessment models X
Develop techniques to incorporate important social variables as modeling parameters X
Develop techniques to use community characteristics as proxies of psychosocial exposure X
Understand the interaction (chemical dose-response relationships) of chemical and non-chemical stressors, specifically psychosocial X
stress
Obtain data on baseline variability of psychosocial stress hormones among the population in order to understand inter- and intra- X
individual variability
Develop tools to monitor psychosocial stress levels in real time (develop biomarkers) at individual and community levels X
Incorporate psychosocial stress into physiologically based pharmacokinetic (PBPK) and physiologically based pharmacodynamic X
(PBPD) models
Examine differential activity patterns between social groups X
Session III: Statistical and Mathematical Modeling for CBRA
Compare various monitoring and modeling techniques to assess value and ease of use X
Develop techniques to integrate existing datasets on population health for future predictions/modeling X
Develop and apply advanced statistical techniques to: characterize group-level effects, synthesize information from multiple datasets, X
extrapolate data across communities, reduce limitations of small population studies, account for possible underestimation of exposure,
etc.
Increase the ability of Hierarchical Bayesian Model to add data from multiple sources and scales X
Develop spatiotemporal models that can adjust for information at multiple scales and levels of accuracy (temporal, spatial, or data from X
multiple sources)
Develop better geospatial techniques to characterize communities X
Explore emerging geospatial tools (e.g., Google Earth) X
Develop hierarchical datasets gathered at multiple levels that can be mapped collected, organized, and accessed by community members X
Improve methods for interpreting biomonitoring data X
Develop transparent modeling methods that can be used collaboratively with the community X X
Better communicate methods and results of complex models X
Emerging themes: #1, Scientific tools and methods to better evaluate health outcomes at a community level (including a new framework for risk assessment);
#2, Environmental health infrastructure; #3, Community involvement processes.
addition, paradigm shifts within agencies and research
institutes may be necessary to initiate CBRA (U.S.
Environmental Protection Agency, 1999; National Environ-
mental Justice Advisory Committee, 2004). Although not
discussed explicitly, the community-based participatory
research or community involvement in decision-making is
not without challenges. For example, truly involving the
community as active participants in research or decision-
making is expensive and requires a great deal of resources.
Additionally, there may be a lack of trust, as well as
differences in goals, values, and perspectives between the
community members, scientists (Israel et al., 2005), and risk
assessors.
Review of Research Needs
Some research needs and other suggestions identified at the
workshop (Tables 1 and 2) include using tools, methods, or
approaches that exist, but are not currently being applied
to risk assessment. Session I indicated the need to refine
sensor technologies for providing real-time data on commu-
Journal of Exposure Science and Environmental Epidemiology (2009), 1-10
Previous
TOC
Next
-------
Sanchez et al.
Research needs for community-based risk assessment
Table 2. Crosscutting suggestions to enhance community-based risk assessment (CBRA) by emerging theme
Emerging themes
#1 #2 #3
Cross-Cutting Suggestions
Develop a new framework to integrate all chemical, non-chemical, and vulnerability issues into risk assessment X
Establish attributes of successful and unsuccessful case studies (deliberative processes where communities partner with the EPA). X
Integrate community knowledge for risk assessment X
Develop tools/methods to elicit community knowledge for risk assessment X
Establish models, tools, and frameworks from other disciplines (specifically the ecological sciences) that would be useful for human X
health risk assessment
Create access to databases that give information at the local level X
Integrate multidisciplinary teams to undertake CBRA research X
Integrate multi-agency (federal, state, local) partnerships to address CBRA X
Utilize community training modules on basic environmental health and risk assessment X
Focus on research that is directly usable by community or its local health or environmental department (community-driven research) X
Establish training modules in academia/agencies on how to conduct community-based participatory research X
X
X
X
X
X
X
X
X
X
X
X
X
X
Emerging themes: #1, Scientific tools and methods to better evaluate health outcomes at a community level (including a new framework for risk assessment);
#2, Environmental health infrastructure; #3, Community involvement processes.
nity environmental stressors. Further examples of the
existing tools and methods are techniques in use by social
scientists or ecologists. Session II suggested the need to
identify and apply indices for measuring community-level
psychosocial health (e.g., community cohesion) and use
community characteristics as proxies of psychosocial expo-
sure. All the session groups identified the need for methods to
work effectively with community members, such as how to
elicit community knowledge and address research problems
that are applicable to the community. Such approaches may
already be in use among social science and public health
disciplines, but risk assessors may be able to further refine
these methods for their work. Altogether, it is important to
understand models, tools, and frameworks from other
disciplines that would be useful for human health risk
assessment.
Other identified research needs included the development
of new tools and methods that could be useful for CBRA.
This may require multidisciplinary collaborations to create
novel techniques and modeling approaches. For example,
techniques used for translating more qualitative social indices
into quantitative risk assessments were addressed in Session I.
Session III identified the need to develop transparent
modeling methods that can be used collaboratively within
the community. Additional needs addressed in Session II
include approaches for incorporating vulnerability into risk
assessment models and techniques for including important
social variables (e.g., poverty, access to medical care, chronic
stress, etc.) as modeling parameters.
Next steps
There have been multiple actions taken, such as the
workshop to advance CBRA research. First, EPA's
National Center for Environmental Research (NCER)
compiled the workshop proceedings, which are now available
online. This document includes a copy of most presentations,
a final agenda, and a summary report capturing all the
presentations and discussions of the workshop (U.S.
Environmental Protection Agency, 2007b). Second, the
NCER created a Listserv to disseminate information and
resources relevant to CBRA (to enlist, see: http://www.epa.
gov/ncer/CBRA web site *forthcoming*). Most of the
information is digested in a monthly bulletin. Third, NCER
is establishing a CBRA Science Page on its Web site, which
will provide information regarding CBRA to the general
public and to the research community. In addition to these
NCER activities, other parts of the EPA are supporting
CBRA. For example, the EPA's Risk Assessment Forum
has been a proponent of CBRA. The EPA's CARE
(Community Action for a Renewed Environment) program
is a community-based cooperative agreement program,
which helps to build broad-based partnerships for reducing
environmental risks at the local level. Also, the EPA's
Region 6 is partnering with the Ponca Tribe of Northern
Oklahoma, EPA's Office of Research and Development, the
University of North Texas, and the Oklahoma State
University to conduct a cumulative risk assessment of the
Tribe, examining holistically the effects of numerous
environmental stressors on tribal lands.
The next step to support CBRA within the EPA is for
NCER to incorporate CBRA into its extramural research
program. NCER's mission is to support high-quality
research by the nation's leading scientists, which will improve
the scientific basis for decisions on national environmental
issues to help the EPA achieve its goals. In 2009, EPA/
NCER plans to issue a Request for Applications (RFA)
soliciting research to further the field of CBRA. This RFA
Journal of Exposure Science and Environmental Epidemiology (2009), 1-10
Previous
TOC
Next
-------
Research needs for community-based risk assessment
Sanchez et al.
will aim to address some of the major research needs
identified in this workshop.
Acknowledgements
The authors are grateful to the many workshop presenters for
their thoughtful contributions, and the workshop partici-
pants for their valuable insights and discussions. The authors
would like to acknowledge Nigel Fields for his breakthrough
concepts on community-based risk assessment, and for his
energy and inspiration.
Disclaimer
This publication was developed under Cooperative Agree-
ment No. #X3-83085001 awarded by the US Environmental
Protection Agency (EPA) to the Association of Schools of
Public Health (ASPH). It has not been formally reviewed by
the EPA. The views expressed in this paper are solely those of
the authors and do not necessarily reflect those of the
Agency. The EPA does not endorse any products or
commercial services mentioned in this publication.
References
Barbeau E.M., Krieger N., and Soobader MJ. Working class matters: socio-
economic disadvantage, race/ethnicity, gender, and smoking in NHIS 2000.
Am J Public Health 2004a: 94(2): 269-278.
Barbeau E.M., McLellan D., Levenstein C, DeLaurier G.F., Kelder G., and
Sorensen G. Reducing occupation-based disparities related to tobacco: roles for
occupational health and organized labor. AmJInd Med 2004b: 46(2): 170-179.
Bellinger D., Leviton A., Waternaux C., Needleman H., and Rabinowitz M. Low-
level lead exposure, social class, and infant development. Neurotoxicol Teratol
1988: 10(6): 497-503.
Bellinger D.C. Lead neurotoxicity and socioeconomic status: conceptual and
analytical issues. Neurotoxicology 2008: 29(5): 828-832.
Bjorntorp P. Do stress reactions cause abdominal obesity and comorbidities? Obes
Rev 2001: 2(2): 73-86.
Caldas E.D., Boon P.E., and Tressou J. Probabilistic assessment of the cumulative
acute exposure to organophosphorus and carbamate insecticides in the
Brazilian diet. Toxicology 2006: 222(1-2): 132-142.
Cary R., Clarke S., and Delic J. Effects of combined exposure to noise and
toxic substances-critical review of the literature. Ann Occup Hyg 1997: 41(4):
455-465.
Callahan M.A., and Sexton K. If cumulative risk assessment is the answer, what is
the question? Environ Health Perspect 2007: 115(5): 799-806.
Castorina R., Bradman A., McKone T.E., Barr D.B., Harnly M.E., and
Eskenazi B. Cumulative organophosphate pesticide exposure and risk
assessment among pregnant women living in an agricultural community: a
case study from the CHAMACOS cohort. Environ Health Perspect 2003:
111(13): 1640-1648.
Clougherty J.E., Levy J.I., Kubzansky L.D., Ryan P.B., Suglia S.F., Canner
M.J., and Wright R.J. Synergistic effects of traffic-related air pollution and
exposure to violence on urban asthma etiology. Environ Health Perspect 2007:
115(8): 1140-1146.
Corburn J. Street Science: Community Knowledge and Environmental Health
Justice. MIT Press, Cambridge, Massachusetts, 2005.
Cory-Slechta D.A., Virgolini M.B., Thiruchelvam M., Weston D.D., and Bauter
M.R. Maternal stress modulates the effects of developmental lead exposure.
Environ Health Perspect 2004: 112(6): 717-730.
DeFur P.L., Evans G.W., Cohen Hubal E.A., Kyle A.D., Morello-Frosch R.A.,
and Williams D.R. Vulnerability as a function of individual and group
resources in cumulative risk assessment. Environ Health Perspect 2007: 115(5):
817-824.
Dominici R, Peng R.D., Bell M.L., Pham L., McDennott A., Zeger S.L., and
Samet J.M. Fine particulate air pollution and hospital admission for
cardiovascular and respiratory diseases. JAMA 2006: 295(10): 1127-1134.
Erren T.C., Jacobsen M., and Piekarski C. Synergy between asbestos and smoking
on lung cancer risks. Epidemiology 1999: 10(4): 405-411.
Fox M.A., Tran N.L., Groopman J.D., and Burke T.A. Toxicological resources
for cumulative risk: an example with hazardous air pollutants. Regul Toxicol
Phamacol 2004: 40(3): 305-311.
Guilarte T.R., Toscano C.D., McGlothan J.L., and Weaver S.A. Environmental
enrichment reverses cognitive and molecular deficits induced by developmental
lead exposure. Ann Neural 2003: 53(1): 50-56.
Herman J.P., and Seroogy K. Hypothalamic-pituitary-adrenal axis, glucocorti-
coids, and neurologic disease. Neural Clin 2006: 24(3): 461-481, vi.
International Life Sciences Institute. A Framework for Cumulative Risk
Assessment. In: Mileson B., Faustman E., Olin S., Ryan P.B., Ferenc S.,
and Burke T. (eds.). Washington, DC, 1999.
Israel B.A., Schulz A.J., Parker E.A., and Becker A.B. Review of community-
based research: assessing partnership approaches to improve public health.
Amu Rev Public Health 1998: 19: 173-202.
Israel B.A., Schulz A.J., Parker E.A., and Becker A.B. Community-
based participatory research: policy recommendations for promoting a
partnership approach in health research. Educ Health (Abingdon) 2001:
14(2): 182-197.
Israel B.A., Parker E.A., Rowe Z., Salvatore A., Minkler M., Lopez J., Butz A.,
Mosley A., Coates L., Lambert G., Potito P.A., Brenner B., Rivera M.,
Romero H., Thompson B., Coronado G., and Halstead S. Community-based
participatory research: lessons learned from the Centers for Children's
Environmental Health and Disease Prevention Research. Environ Health
Perspect 2005: 113(10): 1463-1471.
Kramer M.S., Goulet L., Lydon J., Seguin L., McNamara H., Dassa C., Platt
R.W., Chen M.F., Gauthier H., Genest J., Kahn S., Libman M., Rozen R.,
Masse A., Miner L., Asselin G., Benjamin A., Klein J., and Koren G. Socio-
economic disparities in preterm birth: causal pathways and mechanisms.
Paediatr Perinat Epidemiol 200\: 15(Suppl 2): 104-123.
Maccari S., Darnaudery M., Morley-Fletcher S., Zuena A.R., Cinque C.,
and Van Reeth O. Prenatal stress and long-term consequences: implications
of glucocorticoid hormones. Neurosci Biobehav Rev 2003: 27(1-2):
119-127.
Menzie C.A., MacDonell M.M., and Mumtaz M. A phased approach for
assessing combined effects from multiple stressors. Environ Health Perspect
2007: 115(5): 807-816.
Morrison H.I., Villeneuve P.J., Lubin J.H., and Schaubel D.E. Radon-progeny
exposure and lung cancer risk in a cohort of Newfoundland fluorspar miners.
Radiat Res 1998: 150(1): 58-65.
National Environmental Justice Advisory Committee. Ensuring Risk Reduction in
Communities with Multiple Stressors: Environmental Justice and Cumulative
Risks I Impacts, Vol. EPA 300/R-04/903. U.S. Environmental Protection
Agency, Washington, DC, 2004.
National Research Council. Risk Assessment in the Federal Government: Managing
the Process. National Academy Press, Washington, DC, 1983.
Negro-Vilar A. Stress and other environmental factors affecting fertility
in men and women: overview. Environ Health Perspect 1993: 101(Suppl 2):
59-64.
O'Fallon L.R., and Dearry A. Commitment of the National Institute
of Environmental Health Sciences to community-based participatory
research for rural health. Environ Health Perspect 2001: 109(Suppl 3):
469-473.
O'Fallon L.R., and Dearry A. Community-based participatory research as a tool
to advance environmental health sciences. Environ Health Perspect 2002:
110(Suppl 2): 155-159.
Patychuk D.L. Bridging place-based research and action for health. Can J Public
Health 2007: 98(Suppl 1): S70-S73.
Payne-Sturges D., and Gee G.C. National environmental health measures for
minority and low-income populations: tracking social disparities in environ-
mental health. Environ Res 2006: 102(2): 154-171.
Payne-Sturges D.C., Burke T.A., Breysse P., Diener-West M., and Buckley T.J.
Personal exposure meets risk assessment: a comparison of measured and
Journal of Exposure Science and Environmental Epidemiology (2009), 1-10
Previous
TOC
Next
-------
Sanchez et al.
Research needs for community-based risk assessment
modeled exposures and risks in an urban community. Environ Health Perspect
2004a: 112(5): 589-598.
Payne-Sturges D.C., Schwab M., and Buckley T.J. Closing the research loop: a
risk-based approach for communicating results of air pollution exposure
studies. Environ Health Perspect 2004b: 112(1): 28-34.
Ryan L. Combining data from multiple sources, with applications to environ-
mental risk assessment. Stat Med 2008: 27(5): 698-710.
Ryan P.B., Burke T.A., Cohen Hubal E.A., Cura J.J., and McKone T.E. Using
biomarkers to inform cumulative risk assessment. Environ Health Perspect
2007: 115(5): 833-840.
Schlenker T. The effects of lead in Milwaukee's water. Wis Med J 1989: 88(10):
13-15.
Sexton K., Adgate J.L., Fredrickson A.L., Ryan A.D., Needham L.L., and
Ashley D.L. Using biologic markers in blood to assess exposure to multiple
environmental chemicals for inner-city children 3-6 years of age. Environ
Health Perspect 2006: 114(3): 453-459.
Sexton K., and Hattis D. Assessing cumulative health risks from exposure to
environmental mixtures - three fundamental questions. Environ Health
Perspect 2007: 115(5): 825-832.
Strine T.W., Ford E.S., Balluz L., Chapman D.P., and Mokdad A.M. Risk
behaviors and health-related quality of life among adults with asthma: the role
of mental health status. Chest 2004: 126(6): 1849-1854.
Suglia S.F., Ryan L., Laden F., Dockery D.W., and Wright RJ. Violence
exposure, a chronic psychosocial stressor, and childhood lung function.
Psychosom Med 2008: 70(2): 160-169.
Tamashiro K.L., Nguyen M.M., Ostrander M.M., Gardner S.R., Ma L.Y.,
Woods S.C., and Sakai R.R. Social stress and recovery: implications for body
weight and body composition. Am J Physiol Regul Integr Comp Physiol 2007:
293(5): R1864-R1874.
Teuschler L.K., Rice G.E., Wilkes C.R., Lipscomb J.C., and Power F.W. A
feasibility study of cumulative risk assessment methods for drinking water
disinfection by-product mixtures. / Toxicol Environ Health A 2004: 67(8-10):
755-777.
Thompson B., Coronado G.D., Grossman J.E., Puschel K., Solomon C.C., Islas
L, Curl C.L., Shirai J.H., Kissel J.C., and Fenske R.A. Pesticide take-home
pathway among children of agricultural workers: study design, methods, and
baseline findings. / Occup Environ Med 2003: 45(1): 42-53.
Thompson B., Coronado G.D., Vigoren E.M., Griffith W.C., Fenske R.A.,
Kissel J.C., Shirai J.H., and Faustman E.M. Para ninos saludables: a
community intervention trial to reduce organophosphate pesticide exposure in
children of farmworkers. Environ Health Perspect 2008: 116(5): 687-694.
Tong S., McMichael A.J., and Baghurst P.A. Interactions between environmental
lead exposure and sociodemographic factors on cognitive development. Arch
Environ Health 2000: 55(5): 330-335.
U.S. Congress. Food Quality Protection Act of 1996. Public Law 104-170, United
States of America; 1996.
U.S. Environmental Protection Agency. Concepts, Methods, and Data Sources for
Cumulative Health Risk Assessment of Multiple Chemicals, Exposures and
Effects: A Resource Document. EPA Office of Research and Development,
Washington, DC, 2007a. EPA/600/R-06/013F.
U.S. Environmental Protection Agency. EPA's Framework for Community-Based
Environmental Protection. EPA Office of Policy and Office of Reinvention,
Washington, DC, 1999. EPA 237/K-99/001.
U.S. Environmental Protection Agency. EPA's Framework for Cumulative Risk
Assessment. EPA Office of Research and Development, Washington, DC,
2003. EPA 630/P-02/001F.
U.S. Environmental Protection Agency Organophosphorus Cumulative Risk
Assessment—2006 Update. EPA Office of Pesticide Programs, Washington,
DC, 2006a. EPA HQ-OPP-2006-0618.
U.S. Environmental Protection Agency. Human Health Research Program Multi-
Year Plan (FY 2006-2013). EPA Office of Research and Development,
Washington, DC, 2006b.
U.S. Environmental Protection Agency Proceedings of the U.S. EPA
Workshop on Research Needs for Community-Based Risk Assessment.
EPA Office of Research and Development, Research Triangle Park, NC,
2007b. Available at: http://es.epa.gov/ncer/cbra/presentations/ll_18_07/
1 l_18_07_workshop.html.
U.S. Environmental Protection Agency and Oak Ridge Center for Advanced
Studies. Nanotechnology Applications in Environmental Health: Big Plans for
Little Particles. EPA Office of Research and Development, Research Triangle
Park, NC, 2006c. Available at: http://www.epa.gov/ncct/communications.
html.
Vigoren E.M., Griffith W.C., Krogstad F.T.O., Coronado G.D., Thompson B.,
and Faustman E. Formal Uncertainty Analysis in the Interpretation of
Organophosphate Pesticide Metabolite Concentrations. In: Society for Risk
Analysis Annual Meeting; Dec 9-12, 2007; San Antonio, TX, 2007.
Weaver I.C., Cervoni N., Champagne F.A., D'Alessio A.C., Sharma S., Seckl
J.R., Dymov S., Szyf M., and Meaney M.J. Epigenetic programming by
maternal behavior. Nat Neurosci 2004: 7(8): 847-854.
Weiss B., and Bellinger D.C. Social ecology of children's vulnerability to
environmental pollutants. Environ Health Perspect 2006: 114(10): 1479-1485.
White L.D., Cory-Slechta D.A., Gilbert M.E., Tiffany-Castiglioni E., Zawia
N.H., Virgolini M., Rossi-George A., Lasley S.M., Qian Y.C., and Basha
M.R. New and evolving concepts in the neurotoxicology of lead. Toxicol Appl
Pharmacol 2007: 225(1): 1-27.
Whitmore R., and Kizakevich P. Progress Report: Data Collection Platforms for
Integrated Longitudinal Surveys of Human Exposure-Related Behavior. EPA
Grant Number R831541; 2004.
Wright RJ. Stress and atopic disorders. J Allergy Clin Immunol 2005: 116(6):
1301-1306.
Yeboah D.A. A framework for place based health planning. Aust Health Rev 2005:
29(1): 30-36.
10
Journal of Exposure Science and Environmental Epidemiology (2009), 1-10
Previous
TOC
Next
-------
Journal of Exposure Science and Environmental Epidemiology (2008), 1-6 ijfife
© 2008 Nature Publishing Group All rights reserved 1559-0631/08/S30.00 ^UP
www.nature.com/jes
Review
Exposure science and the U.S. EPA National Center for Computational
Toxicology
ELAINE A. COHEN HUBALa, ANN M. RICHARDa, IMRAN SHAHa, JANE GALLAGHER13, ROBERT KAVLOCKa,
JERRY BLANCATOa AND STEPHEN W. EDWARDSb
^National Center for Computational Toxicology, US Environmental Protection Agency, Research Triangle Park, North Carolina, USA
bNational Health and Environmental Effects Laboratory, US Environmental Protection Agency, Research Triangle Park, North Carolina, USA
The emerging field of computational toxicology applies mathematical and computer models and molecular biological and chemical approaches to explore
both qualitative and quantitative relationships between sources of environmental pollutant exposure and adverse health outcomes. The integration of
modern computing with molecular biology and chemistry will allow scientists to better prioritize data, inform decision makers on chemical risk
assessments and understand a chemical's progression from the environment to the target tissue within an organism and ultimately to the key steps that
trigger an adverse health effect. In this paper, several of the major research activities being sponsored by Environmental Protection Agency's National
Center for Computational Toxicology are highlighted. Potential links between research in computational toxicology and human exposure science are
identified. As with the traditional approaches for toxicity testing and hazard assessment, exposure science is required to inform design and interpretation
of high-throughput assays. In addition, common themes inherent throughout National Center for Computational Toxicology research activities are
highlighted for emphasis as exposure science advances into the 21st century.
Journal of Exposure Science and Environmental Epidemiology advance online publication, 5 November 2008; doi:10.1038/jes.2008.70
Keywords: exposure modeling, toxicology, bioinformatics, toxicogenomics.
Introduction
Computational toxicology is a new and high-priority
research area in US Environmental Protection Agency (US
EPA, 2003; Kavlock et al, 2007). Defined as the application
of mathematical and computer models to predict adverse
effects and to better understand the mechanism(s) through
which a given chemical induces harm, computational
toxicology provides approaches to explore both qualitative
and quantitative relationships between sources of environ-
mental pollutant exposure and adverse health outcomes. This
integration of modern computing with molecular biology and
chemistry will allow scientists to better prioritize data, inform
decision makers on chemical risk assessments and understand
a chemical's progression from the environment to the target
tissue within an organism and ultimately to the key steps that
trigger an adverse health effect.
In February 2005, US EPA established the National
Center for Computational Toxicology (NCCT) to conduct
Address all correspondence to: Dr. Elaine A. Cohen Hubal, National
Center for Computational Toxicology, US Environmental Protection
Agency, Mail Drop B205-01, Research Triangle Park, NC 27711, USA.
Tel.: + 1 919 541 4077, Fax: +919 685 3334.
E-mail: hubal.elaine@epa.gov
Received 18 September 2008; accepted 23 September 2008
and sponsor research in this area. The overall goal of ORD's
research program on Computational Toxicology is to use
emerging technologies to improve quantitative risk assess-
ment and reduce uncertainties in the source-to-adverse
outcome continuum by providing ultimately systems level
understanding of biological processes and their perturbation.
The importance and relevance of this mission and NCCT-
initiated research has received strong support with the recent
release of the National Academy of Sciences report calling
for a transformative shift in toxicity testing and risk
assessment (NRC, 2007). Toxicity Testing in the 21st
Century: A Vision and a Strategy, calls for a collaborative
effort across the toxicology community to rely less on animal
studies and more on in vitro tests using human cells and
cellular components to identify chemicals with toxic effects.
A framework for implementing this long-range vision is
provided by the recently formalized collaboration between
two NIH institutes (NIEHS and NHGRI) and the EPA to
use high-speed, automated screening methods to efficiently
test compounds for potential toxicity (Collins et al., 2008).
These high visibility efforts in toxicity testing and
computational toxicology raise important research questions
and opportunities for exposure scientists. The National
Academies report authors (NRC, 2007) emphasize that,
population-based data and human exposure information are
Previous
TOC
Next
-------
Cohen Hubal et al.
Exposure and complex
required at each step of their vision for toxicity testing; and
that these data will continue to play a critical role in both
guiding development and use of the toxicity information.
The NCCT Computational Toxicology program has
identified the need to include exposure information for
chemical prioritization, modeling system response to chemi-
cal exposures across multiple levels of biological organization
and linking information on potential toxicity of environ-
mental contaminants to real-world health outcomes (e.g.,
complex disease). As a starting point, several common
themes have emerged among the NCCT research projects.
These themes are of particular interest to exposure scientist as
we consider how to best incorporate the tools of computa-
tional toxicology into exposure research as well as how to
best contribute to research in computational toxicology. The
research conducted in the NCCT is designed to address the
need for: (1) characterization of the target system across
levels of biological organization; (2) improved linkages across
the source-to-outcome continuum; and (3) a shift from linear
source-to-dose paradigm to a systems-based approach. In
addition, the complexity of the systems under study and the
multidimensional nature of data produced using emerging
technologies requires extensive collaboration and advanced
environmental informatic capabilities. In this paper, with
these common themes in mind: potential links between
research conducted in the US EPA's National Center for
Computational Toxicology and human exposure science are
discussed; the need for exposure science to address chemical
screening, prioritizing and toxicity testing in the 21st century
is identified; and priority research areas for exposure
scientists are proposed.
NCCT research activities
Toxcast: Prioritizing the Toxicity Testing of Environmental
Chemicals
Globally there is a need to characterize potential risk to
human health and the environment that arises from the
manufacture and use of tens of thousands of chemicals. In
2007, US EPA's NCCT launched ToxCast™ to develop a
cost-effective in vitro approach for prioritizing the toxicity
testing of large numbers of chemicals in a short period of time
(Dix et al., 2007). Using data from state-of-the-art high-
throughput screening (HTS) bioassays developed in the
pharmaceutical industry, ToxCast™ is building computa-
tional models to forecast the potential human toxicity of
chemicals. The premise underlying ToxCast™ is that
toxicological response is driven by interactions between
chemicals and biomolecular targets. For most environmental
chemicals the protein targets and biological effects underlying
potential adverse effects have yet to be identified or
characterized. The strategy of ToxCast™ is to focus on a
diverse range of assays and data types to identify potential
targets. The ToxCast™ program will apply a multiple target
matrix approach to address this goal. The matrix contains an
expanded number of potential targets whose chemical
interactions may be characterized by in silica models,
biochemical assays, cell-based in vitro assays (based on both
human and animal tissues) and nonmammalian models. The
resulting data span levels of biological organization: mole-
cular, cellular, tissue and whole organism. The overall
pattern across many assays and data types will be used to
develop a fingerprint or bioactivity profile that can be used as
a predictor of toxicity. These hazard predictions will provide
EPA regulatory programs with science-based information
helpful in prioritizing chemicals for more detailed toxicolo-
gical evaluations and lead to more efficient use of animal
testing. The resulting data will also provide insights into
modes of action of chemical toxicity in an unprecedented and
unbiased manner. This, in turn, has implications for
identifying potentially susceptible populations, both from a
life-stage viewpoint, but also from a genetic (polymorphic)
standpoint as toxicity pathways intersect with disease path-
ways.
The ToxCast™ program is being implemented using a
tiered multiphase approach (see www.epa.gov/ncct/toxcast).
In phase I, over 300 well-characterized chemicals have been
profiled in over 400 HTS end points. These end points
include biochemical assays of protein function, cell-based
transcriptional reporter assays, multi-cell interaction assays,
transcriptomics on primary cell cultures and developmental
assays in zebrafish embryos. Almost all of the phase 1
compounds have been tested in traditional toxicology tests,
including developmental toxicity, multigeneration studies and
subchronic and chronic rodent bioassays. Phase 1 ToxCast™
signatures will be defined and evaluated by the ability to
predict outcomes from existing mammalian toxicity testing
and identify toxicity pathways that are relevant to human
health effects.
ToxCast phase II, scheduled to launch in FY09 will bring
the total number of chemicals screened to nearly 1000. These
additional compounds will represent broader chemical
structure and use classes and some pharmaceutical agents
with known adverse side effects, to evaluate the predictive
bioactivity signatures developed in phase I. As a result of the
memorandum of understanding (MOU) with National
Toxicology Program/NIEHS and NIH Chemical Economics
Center/NHGRI, additional chemical screening capability is
being made accessible and it is now projected that more than
5000 chemicals will be entering the high-throughput screen-
ing program of the NCGC within the next year. It is
anticipated that successful conclusion of ToxCast™ phases I
and II will provide EPA regulatory programs with a tool for
rapidly and efficiently screening compounds and prioritizing
further toxicity testing.
As computational analyses of ToxCast™ phase I data begin,
the need to consider exposure potential for selecting phase II
Journal of Exposure Science and Environmental Epidemiology (2008), 1-6
Previous
TOC
Next
-------
Exposure and complex
Cohen Hubal et al.
Stressor
Environmental
Source
| Ambient
* Exposure
Environmental
Source
Personal
\ ^ Exposure
Perturbation
Perturbation
Internal Exposure
(Tissue Dose)
Dose to Cell
Dose of Stressor
Molecules
Population
Individual
ft
Tissue
Cell
Biological
Molecules
Disease
Incidence/Prevalence
Disease State
(Changes to Health Status)
Dynamic Tissue Changes
(Tissue Injury)
Dynamic Cell Changes
(Alteration in Cell Division,
Cell Death)
Dynamic Changes
in Intracellular Processes
Figure 1. Cascade of exposure-response processes for integrating exposure science and toxicogenomic mode-of-action information.
chemicals as well as for providing real-world relevance for
interpretation of toxicity screening has been identified by
NCCT. As ToxCast™ and related research activities provide
information on key events required to incorporate mode-of-
action information along the continuum, similar key exposure
metrics at comparable resolution will need to be identified
(Figure 1). This would build upon the great strides made in the
last 20 years on PBPK modeling and expand on that success
as emerging technologies blur the boundaries between
exposure and effects sciences. The ultimate goal would be an
integrated program in which biomarkers of exposure and
bioindicators of effects are jointly determined and can be used
to enhance biologically based dose-response models by
providing measured parameters linking relevant exposures to
the probability of an adverse outcome (NRC, 2007).
Distributed Structure-Searchable Toxicity Database
Network: Informatics for Environmental Health Risk
Assessment
Specific activities at the NCCT include research to define
chemical properties that can be used as indicators of potential
toxicity for use in prioritization of toxicity testing as well as to
construct computational models of chemical interactions with
biological systems for human health risk assessment. This
research requires creation of flexible databases covering a
broad range of chemical space so that the wide range of
multidimensional data spanning levels of biological organiza-
tions across the source-to-outcome continuum can be
accessed, combined and interpreted using novel approaches.
The Distributed Structure-Searchable Toxicity (DSSTox)
Database Network is creating a chemical data foundation for
improved structure activity and predictive toxicology capabil-
Journal of Exposure Science and Environmental Epidemiology (2008), 1-6
ities, and broad linkages to chemical data resources across and
outside of EPA (Richard et al., 2006). The DSSTox website
(US EPA, 2008) publishes downloadable, chemical structure
files associated with toxicity data in a variety of formats, along
with documentation and links to source information, quality
review procedures and guidance for users. Standardized
chemical structure annotation of a diverse array of toxicol-
ogy-related data and resources, coupled with the online
DSSTox Structure Browser, are providing structure-search-
ability and direct access to these data (including EPA's
Integrated Risk Information System, Fat-head minnow acute
toxicity database and High-Production Volume Chemical lists,
the National Toxicology Program's — NTP Bioassay
database, as well as estrogen-receptor binding data, rodent
carcinogenicity data and most recently, gene expression data).
The DSSTox project is also providing primary structure-
annotation and cheminformatics support to both the NTP HTS
and EPA ToxCast™ programs in conjunction with NCCT's
Aggregated Computational Toxicity Resource (ACToR) pro-
ject, slated for public release in late 2008 (Richard et al., 2008).
The latter is providing a relational database platform for
surveying vast Internet data resources pertaining to environ-
mental toxicology (hundreds of thousands of chemicals),
including high- and medium-production chemical lists and
exposure-related data (Judson et al., in press). ACToR will also
serve as the primary storage and analysis resource for the
ToxCast HTS data, linking these data to standardized historical
lexicological test results and broader chemical resources.
Similarly, it is imperative that exposure data be accessible
and linked to the rapidly growing base of toxicity data.
Development of consolidated data and knowledge bases for
exposure is a high priority. Existing tools and platforms that
Previous
-------
Cohen Hubal et al.
Exposure and complex
are currently being implemented with environmental toxicity
information should be considered to provide the most useful
links to existing toxicity and environmental health data.
Relevance and value of exposure information for toxicology
and risk assessment will increase dramatically if links to these
data are immediately apparent to an investigator searching the
universe of toxicity and health data for a given compound.
Chemical structure-annotation of exposure-related data, such
as could be provided by DSSTox, and incorporation of such
data into the new ACToR resource, will greatly enhance
linkages between these exposure data and toxicity-related
human health end points. In a preliminary demonstration, a
DSSTox file of 60 chemical structures was created to index
chemical-related content within the EPA Children's Total
Exposure to Persistant Pesticides and Other Persistant
Pollutants (CTEPP) database (US EPA, 2006). Ideally,
conversion of text tables within CTEPP pdf documents would
be tagged and indexed in web-accessible files such that
chemical structure-searches on the Internet could locate
relevant exposure content. These sorts of linkages have the
potential to bring the toxicology and exposure science research
communities into closer alignment and foster more productive
interaction.
v-Liver™: Characterizing Toxicity Pathways and
Extrapolating Dose-Response
ToxCast™ has generated an unprecedented amount of rodent
data for discovering in vitro biomarkers of adverse outcomes in
vivo, which will be vital for prioritizing chemicals for further
testing. Rodent liver toxicity is currently the most frequent
cause for the regulation of orally consumed environmental
chemicals. The Virtual Liver Project v-Liver™ will utilize
ToxCast™ and other public and agency data to aid in
extrapolating in vitro assays to clinical outcomes across
chemicals, doses, genders, life stages and populations (Kav-
lock et al., 2007). Virtual Tissues offer a novel translational
paradigm for predicting target organ toxicity by fusing
molecular and cellular systems modeling for physiologically
relevant simulation (Knudsen and Kavlock, 2008). The goal
of v-Liver™ is to quantitatively simulate liver injury due to
chronic chemical exposure by modeling the linkage of
perturbed molecular pathways with adaptive or adverse
processes leading to changes of cell state, and the integration
of this response through a dynamic cellular network giving rise
to macroscopic tissue alterations. Histopathology is currently
the clinical gold standard for estimating adverse liver out-
comes. In the long-term, the Virtual Liver's ability to
quantitatively predict tissue lesions from molecular and cellular
networks dynamics will help in accurately assessing human
risks from exposure to environmental stressors.
The first phase of v-Liver™ is a proof of concept focused
on a subset of ToxCast™ chemicals and apical toxicity end
points. These initially include a subset of pathways from
nuclear receptor activation to proliferative sublobular lesions in
rodents through a combination of cellular mitogenic, muta-
genic and regenerative proliferation processes. Currently, data
are being gathered on relevant molecular, cellular and tissue-
level quantitative data on chemicals with known toxicological
profiles to cross-validate the in silica modeling approach. In
addition, qualitative and quantitative information on normal
and pathologic processes across levels of biological organiza-
tion is being curated in a knowledge base for virtual tissue
construction. Finally, the multiscale molecular, cellular and
tissue responses are simulated via an agent-based modeling
(ABM) approach. Here, the liver tissue is being conceptualized
as an ecosystem of heterogeneous cells. The ABM approach
attempts to faithfully model the microanatomic heterogeneity
of the complex hepatic acinus as a network of parenchymal
and nonparenchymal "agents" in a nutrient and xenobiotic
gradient. Agents model the dynamic behavior of liver cells to
their microenvironment by processing endogenous and xeno-
biotic inputs through molecular circuits. ABM approaches
encapsulate molecular, cellular and tissue complexity effectively
enabling in silica simulation of functional liver unit(s) across
species, chemicals, doses and times. The modularity of the
approach also simplifies integration with physiologic models at
the organism scale.
The liver's response to environmental chemicals spans
multiple levels of organization — from molecular interactions
to alterations in tissue structure. Novel computational
approaches are required to ensure that information on
biological effects is developed at environmentally relevant
exposures. Integrating exposure, organism-level ADME and
Virtual Tissues is vital for assessing the risk of adverse
outcomes in human populations. Significant work in exposure
modeling has focused on application of mass-balance
approaches to model chemical fate and transport from source
to individual to internal dose. This research should be
continued with specific emphasis on developing inputs
required to predict response of biological systems to
environmental perturbations such as those required for v-
Liver™. In addition, there continues to be significant
challenges associated with modeling and predicting individual
and population interaction with the environment. Similar to
tissue-level biological systems, imbedded complexity (e.g.,
feedback, multiple scales, multiple stressors, etc.) of the
higher-level systems requires consideration of a novel
approach. Conceptualizing a population as an ecosystem of
heterogeneous individuals and the individual as an ecosystem
of heterogeneous behavior will facilitate holistic modeling of
human-environment interaction. Use cases for data rich
compounds should be developed to test utility of this
approach for identifying key exposure events that can support
tissue-level predictions. In the long-term, the source-to-
outcome modeling paradigm may become more integrative
by capitalizing on emerging multiscale systems modeling
approaches such as those being applied in the v-Liver™
project.
Journal of Exposure Science and Environmental Epidemiology (2008), 1-6
Previous
TOC
Next
-------
Exposure and complex
Cohen Hubal et al.
Mechanistic Indicators of Childhood Asthma Study:
Understanding Environmental Factors of Complex Disease
Ultimately, the systems of primary interest to human-health
risk assessors are those at the individual and population level.
Emerging tools in molecular biology provide the potential to
develop cellular and molecular indicators of exposure that
can be used to assess the vulnerability of humans to
environmental stressors. The Mechanistic Indicators of
Childhood Asthma (MICA) Study has been designed to
incorporate state-of-the-art technologies to examine the
physiological and environmental factors that interact to
increase the risk of asthmatic responses (http://www.epa.gov/
dears/studies.htm). Collected markers of susceptibility,
exposure and effects are being used to analyze and
characterize combined risk factors that relate to asthma
severity in a cohort of children. The MICA study provides an
opportunity to advance a system-based approach for
evaluating complex relationships among environmental
factors, physiological biomarkers and health outcomes. This
study also provides a platform for applying and testing
computational approaches to evaluate multifactorial-multi-
dimensional data that are becoming standard output of
environmental health and ecogenetic studies (NRC, 2008)
and to use these data for hypothesis development.
The MICA study is primarily a clinically based observa-
tional children's study. Multiple measures of health status,
asthma severity, environmental exposure and gene expression
have been collected in a case/control cohort of 200 children
(aged 9-12 years). Environmental samples have been
collected with a focus on three broad classes of particulate-
associated chemicals: volatile organic compounds, metals and
polycyclic aromatic hydrocarbons. In the NCCT component
of the MICA study, advanced statistical and machine
learning methods are being applied in combination with
mechanistic information to evaluate multiple types of
biomarker data collected in MICA (similar approaches for
combining data presented by Reif et al., in press). Methods
and tools are being applied to evaluate and visualize gene
expression data in novel ways. These approaches are being
used to characterize the relationship between rat and human
response, to characterize the relationship between gene
expression and biological response and to evaluate the utility
of gene expression data collected in a human cohort study for
understanding relationships between exposure, susceptibility
and early effects (Reif et al., in preparation; Heidenfelder
et al., submitted).
The MICA study provides a case example for how
exposure information and computational toxicology has the
potential to provide a mechanistic interpretation of biomo-
nitoring data, whether these data are "classical" concentra-
tion measurements or "toxicogenomic" markers in relation
to exposure patterns, routes and pathways. As methods for
assessing health risks resulting from exposures to individual
environmental pollutants improves, environmental health
Journal of Exposure Science and Environmental Epidemiology (2008), 1-6
scientists are turning attention toward characterizing relation-
ships between multiple environmental factors and complex
disease. The MICA study is an example of this shift.
Computational tools and approaches for efficiently char-
acterizing exposure potential of environmental compounds
are required for screening and prioritizing as well as for
environmental health studies. One possibility is to formulate
an exposure classification index based on a limited set of
metrics designed to efficiently cover exposure space. Applica-
tion of environmental informatic approaches may help to
identify the critical metrics for representing personal exposure
over time, place, life stage and lifestyle or behavior. Such an
approach could also inform exposure data collection at the
personal, residential, community and ambient level. The
MICA study also serves as an example of the type of study
outlined in Figure 1 where biomarkers of exposure and
bioindicators of effects are jointly determined and linked
computationally to support modeling for risk assessment.
Exposure science for computational toxicology
Clearly the new field of computational toxicology provides
significant opportunities for exposure scientists. The chal-
lenge is to move forward and consider new approaches for
measuring, modeling and assessing exposure to address 21st
century research needs for environmental health risk assess-
ment. Consideration of analogies in hazard assessment may
help to inform our path forward.
The NRC Vision (2007) of a shift to characterizing toxicity
pathways requires a commensurate shift to characterizing
exposure across all levels of biological organization (Figure 1).
Interpretation of toxicogenomic hazard data requires con-
textual relevance. Pathways identified using HTS approaches
such as those being developed in the ToxCast program are
being anchored to apical end points using conventional
toxicity data. Similarly, understanding relevant perturbations
leading to these toxicogenomic end points require anchoring
stressors to real-world human exposure (e.g., biomonitoring
data and other conventional exposure metrics). As illustrated
in the examples below, new approaches to risk assessment
require exposure science to extend beyond traditional
boundaries and predict exposures down to the molecular
level. This requires consideration of the interactions between
exposure and effect and highlights the need for interdisciplin-
ary teams to define these interactions.
Suter (1999) notes that conventionally risk assessment
considered the process by which a release of a contaminant
results in exposure of a target or receptor. Induction of effects
is assumed to occur after the exposure process facilitating
separate analysis of exposure and hazard. However, as risk
assessment has moved to address risks resulting from
exposures to multiple stressors, this assumption is no longer
appropriate. Effects at one organizational level affect others
Previous
TOC
Next
-------
Cohen Hubal et al.
Exposure and complex
resulting in complex health outcomes. So, rather than
considering flow of contaminants along the source-to-
outcome continuum, there is a need to characterize cascades
of alternating processes and states in an overall network
(Suter, 1999).
As toxicity testing relies more on evaluating the mode of
action for compounds, systems approaches describing the
molecular basis of disease (Loscalzo et al., 2007) are being
considered for risk assessment purposes (Edwards and Preston,
2008). With this approach, molecular networks for disease can
be generated (Schadt and Lum, 2006) and used to derive key
event networks for use in mode of action determination. The
resulting networks describe the overall connectivity of the
system along with the perturbations of that system resulting in
certain disease states. Mode of action can now be defined as the
perturbations of this "normal" state by a specific stressor or
mixture. Such a holistic systems approach demands exposure
metrics and models to characterize key stressors at a level of
resolution commensurate with that of the response or effects.
An example of this type of approach has previously been
demonstrated for ecological risk assessment (Ankley et al.,
submitted; Ekman et al., 2007, 2008). Development and
application of toxicogenomic molecular indicators of exposure
(e.g., Sen et al., 2007) and nanotechnology-based sensors
(Weis, 2005) provides the potential to mechanistically link
traditional exposure metrics and end points measured in HTS
assays. Together, a focus on mode of action and characteriza-
tion of stressors at all levels of biological organization enables
the vision for toxicity testing in the 21st century set forth by the
NRC (NRC, 2007) by providing a framework in which to
interpret toxicity pathway perturbations.
The NCCT research program and the NRC vision present
tremendous challenges and opportunities for exposure
science. In May 2008, US EPA established a Community
of Practice in Exposure Science for Toxicity Testing, Screen-
ing and Prioritization (ExpoCop) to provide a forum for
promoting the advancement of exposure science to begin to
address some of the challenges alluded to in this paper (US
EPA, 2008). We look forward to a broad participation from
the exposure science community as we continue this
important dialog.
Disclaimer
The US Environmental Protection Agency, through its
Office of Research and Development funded and managed
the research described here. It has been subjected to Agency's
administrative review and approved for publication.
References
Ankley, et al. Endocrine disrupting chemicals in fish: developing exposure
indicators and predictive models of effects based on mechanism of action.
submitted.
Collins F.S., Gray G.M., and Bucher J.R. Transforming environmental health
protection. Science 2008: 319: 906-907.
Dix D.J., Houck K.A., Martin M.T., Richard A.M., Setzer R.W., and Kavlock
R.J. The ToxCast program for prioritizing toxicity testing of environmental
chemicals. Toxicol Sci 2007: 95(1): 5-12.
Edwards S.W., and Preston R.J. Systems biology and mode of action based risk
assessment. Toxicol Sci 2008 doi:10.1093/toxsci/kfn!90.
Ekman D.R., Teng Q., et al. NMR analysis of male fathead minnow urinary
metabolites: a potential approach for studying impacts of chemical exposures.
Aquat Toxicol 2007: 85(2): 104-112.
Ekman D.R., Teng Q., et al. Investigating compensation and recovery of fathead
minnow (Pimephales promelas) exposed to 17alpha-ethynylestradiol with
metabolite profiling. Environ Sci Technol 2008: 42(11): 4188-4194.
Heidenfelder B.L., Reif D.M., Cohen Hubal E.A., Hudgens E.E., Bramble L.A.,
Wagner J.G., Harkema J.R., Morishita M., Keeler G.J., Edwards S.W., and
Gallagher J.E. Comparative microarray analysis and pulmonary morpho-
metric changes in brown Norway rats exposed to ovalbumin and/or
concentrated air particulates. Submitted.
Judson R., Richard A., Dix D., Houck K., Elloumi F., Martin M., Cathey T.,
TransueT.R., Spencer R., and Wolf M. ACToR — aggregated computational
toxicology resource. Toxicol Appl Pharmacol, Available online 18 July 2008.
Kavlock R.J., Ankley G., Blancato J., Breen M., Conolly R., Dix D., Houck K.,
Hubal E., Judson R., Rabinowitz J., Richard A., Setzer R.W., Shah I.,
Villeneuve D., and Weber E. Computational toxicology a state of the science
mini review. Toxicol Sci 2007: 103(1): 14-27.
Knudsen T.B., and Kavlock R.J. Comparative Biomformatics and Computational
Toxicology, Abbot B., and Hansen D. (Eds.). 3rd edn. Taylor & Francis 2009.
Loscalzo J., Kohane I., and Barabasi A.L. Human disease classification in the
postgenomic era: a complex systems approach to human pathobiology. Mol
Syst Biol 2007: 3: 124.
National Research Council of the National Academies (NRC). Toxicity Testing in
the 21st Century: A Vision and A Strategy. The National Academies Press,
Washington, DC, 2007.
National Research Council of the National Academies (NRC). The National
Children's Study Research Plan: A Review. The National Academies Press,
Washington, DC, 2008.
Richard A., Yang C, and Judson R. Toxicity data informatics: supporting a new
paradigm for toxicity prediction. Toxicol Mech Meth 2008: 18: 103-118.
Richard A.M., Gold L.S., and Nicklaus M.C. Chemical structure indexing of
toxicity data on the Internet: Moving toward a flat world. Curr Opin Drug
Discov Devel 2006: 9(3): 314-325.
Reif D.M., et al. Integrating demographic, clinical, and environmental exposure
information to identify genomic biomarkers associated with subtypes of
childhood asthma. 2008 Joint Annual Conference ISEE/ISEA. Pasadena, CA.
Reif D.M., Motsinger A.A., McKinney B.A., Edwards K.M., Chanock S.J.,
Rock M.T., Crowe Jr J.E., and Moore J.H. Integrated analysis of genetic and
proteomic data identifies biomarkers associated with systemic adverse events
following smallpox vaccination. Genes Immun, published online 16 October
2008. doi:10.1038/gene.2008.80.
Schadt E.E., and Lum P.Y. Thematic review series: systems biology approaches to
metabolic and cardiovascular disorders. Reverse engineering gene networks to
identify key drivers of complex disease phenotypes. / Lipid Res 2006: 47:
2601-2613.
Sen B., Mahadevan B., and DeMarini D.M. Transcriptional responses to complex
mixtures — a review. Mutat Res 2007: 636(2007): 144-177.
Suter G.W. Developing conceptual models for complex ecological risk assessments.
Him Ecol Risk Assess 1999: 5(2): 375-396.
U.S. EPA. A Framework for a Computational Toxicology Research Program. Office
of Research and Development, Washington, DC, 2003 EPA 600/R-03/065
http://www.ep a. gov/comp to x/publications/comptoxframework06_02_04.pdf.
U.S. EPA. A Pilot Study of Children's Total Exposure to Persistent Pesticides and
Other Persistent Organic Pollutants (CTEPP). Volume I: Final Report.
Contract Number 68-D-99-011, U.S. Environmental Protection Agency,
Office of Research and Development, Research Triangle Park, NC, 2006
Available online at http://www.epa.gov/heasd/ctepp/ctepp_report.pdf.
U.S. EPA. EPA Community of Practice: Exposure Science for Toxicity Testing,
Screening, and Prioritization 2008: http://www.epa.gov/ncct/practice_community/
exposure_science.htmlAccessed September 16, 2008.
Weis B.K., Balshaw D., Barr J.R., Brown D., Ellisman M., Lioy P., et al.
Personalized exposure assessment: promising approaches for human environ-
mental health research. Environ Health Perspect 2005: 113(7): 840-848.
Journal of Exposure Science and Environmental Epidemiology (2008), 1-6
Previous
TOC
Next
-------
Reproductive Toxicology, In Press
Fetal malformations and early embryonic gene expression response in
cynomolgus monkeys maternally exposed to thalidomide
Makoto Emaa*, Ryota Iseb, Hirohito Katoc, Satoru Onedad, Akihiko Hirosea, Mutsuko
Hirata-Koizumi3, Amar V. Singh6, Thomas B. Knudsen, and Toshio Iharac
a Division of Risk Assessment, Biological Safety Research Center, National Institute
of Health Sciences, Tokyo, Japan
b Shin Nippon Biomedical Laboratories (SNBL), Ltd., Tokyo, Japan
c Shin Nippon Biomedical Laboratories (SNBL), Ltd., Kagoshima, Japan
d SNBL USA, Ltd., Everett, WA, USA.
e Contractor to NCCT, Lockheed-Martin, Research Triangle Park NC, USA 27711
f National Center for Computational Toxicology (NCCT), U.S. Environmental
Protection Agency, Research Triangle Park NC, USA 27711
* Corresponding author:
Makoto Ema, DVM, PhD.
Division of Risk Assessment, Biological Safety Research Center, National Institute
of Health Sciences, 1-18-1, Kamiyoga, Setagaya-ku, Tokyo 158-8501, Japan
Tel: +81-3-3700-9878
Fax: +81-3-3700-1408
E-mail: ema@nihs.go.jp
Previous I TOC
-------
ABSTRACT
The present study was performed to determine experimental conditions for thalidomide
induction of fetal malformations and to understand the molecular mechanisms underlying
thalidomide teratogenicity in cynomolgus monkeys. Cynomolgus monkeys were orally
administered thalidomide at 15 or 20 mg/kg/day on days 26-28 of gestation, and fetuses were
examined on day 100-102 of gestation. Limb defects such as micromelia/amelia, paw/foot
hyperflexion, polydactyly, syndactyly, and brachydactyly were observed in seven of eight
fetuses. Cynomolgus monkeys were orally administered thalidomide at 20 mg/kg on day 26 of
gestation, and whole embryos were removed from the dams 6 h after administration. Three
embryos each were obtained from the thalidomide-treated and control groups. Total RNA was
isolated from individual embryos, amplified to biotinylated cRNA and hybridized to a custom
Non-Human Primate (NHP) GeneChip® Array. Altered genes were clustered into genes that
were up-regulated (1,281 genes) and down-regulated (1,081 genes) in thalidomide-exposed
embryos. Functional annotation by Gene Ontology (GO) categories revealed up-regulation of
actin cytoskeletal remodeling and insulin signaling, and down-regulation of pathways for
vasculature development and the inflammatory response. These findings show that thalidomide
exposure perturbs a general program of morphoregulatory processes in the monkey embryo.
Bioinformatics analysis of the embryonic transcriptome following maternal thalidomide
exposure has now identified many key pathways implicated in thalidomide embryopathy, and
has also revealed some novel processes that can help unravel the mechanism of this important
developmental phenotype.
Key words: Thalidomide; Teratogenicity; Fetal malformation; Gene expression profile;
Embryo; Cynomolgus monkey
Previous I TOC
-------
1. INTRODUCTION
Thalidomide (a-phthalimidoglutarimide) was synthesized in West Germany in 1953 by
the Chemie Griinenthal pharmaceutical firm, and was marketed from October 1957 into the
early 1960s. It was used for treating nausea and vomiting late during pregnancy and was also
said to be effective against influenza. The first case of the phocomelia defect, although not
recognized at the time as drug-related, was presented by a German scientist in 1959;
subsequently, malformed children were reported in 31 countries [1]. A pattern of defects of
limbs as well as the ocular, respiratory, gastrointestinal, urogenital, cardiovascular and nervous
systems caused by maternal thalidomide exposure during early pregnancy was observed. Limb
defects such as phocomelia, amelia, micromelia, oligodactyly, and syndactyly were the most
common malformations [2]. After removal from the global market in 1962, thalidomide was
reintroduced in 1998 by the biotechnology firm Celgene as an immunomodulator for treatment
of erythema nodosum leprosum, a serious inflammatory condition of Hansen's disease, and in
orphan status for treating Crohn's disease and several other diseases [1].
Animal species are not equally susceptible or sensitive to the teratogenicity of chemical
agents, and some species respond more readily than others [3]. For thalidomide, a variety of
developmental toxic effects were reported in 18 animal species, but the responses have been
highly variable across species. Limb defects that mimic human thalidomide embryopathy have
only been observed and replicated in a few strains of rabbits and in primates [1,3,4]. Eight of 9
subhuman primates treated with thalidomide showed characteristic limb reduction
malformations ranging from amelia to varying degrees of phocomelia at a dosage and timing
comparable to those observed in human thalidomide embryopathy [3,5]. Since the first report of
Previous I TOC
-------
thalidomide appeared 50 years ago, considerable information regarding the therapeutic
applications of this drug has accumulated, but the mechanisms by which thalidomide produce
congenital malformations are still not well understood [2,3,5].
The nonhuman primate Macaca fascicularis (cynomolgus monkey) is widely used in
prenatal developmental studies because of year-round rather than seasonal breeding behavior
[6]. Kalter [5] noted that nonhuman primates, especially macaques and baboons, are favorable
for mechanistic studies; however, only two full reports of the teratogenicity of thalidomide in
cynomolgus monkeys are available [7,8]. In those studies, cynomolgus monkeys were given
thalidomide by gavage at doses of 5 to 30 mg/kg/day during gestation days 20 to 30, and fetuses
were examined morphologically. The findings of these studies determined the critical period
and doses of thalidomide required for production of fetal malformations in this macaque species.
Although amounts taken were not always accurately recorded in humans, available documents
show that typical malformations resulted from the ingestion of as little as 25 mg three times a
day or 100 mg/day for 3 days during the sensitive period, equivalent to an astonishingly small
dosage of about 1 mg/kg/day [5]. In teratology studies using cynomolgus monkeys, the timing
of dosing was comparable to the human one and the doses were estimated to be 5 to 30 times
higher those which produced typical malformations in humans [5,7,8].
Knowledge of the patterns of altered gene expression in embryonic target organs on a
global scale is an important consideration for understanding the mechanisms of teratogenesis
[9-13]. The application of cDNA microarray technology, a genome-wide analysis technique, to
cynomolgus monkeys facilitates the rapid monitoring of a large number of gene alterations in
Previous I TOC
-------
this species [14]. In order to obtain information about the molecular mechanisms underlying the
detrimental effects of thalidomide teratogenicity, the present study has determined the
experimental conditions required to produce thalidomide-induced fetal defects that mimicked
human abnormalities in cynomolgus monkeys and then profiled altered patterns of gene
expression in these embryos during the critical period. The dosing used in the present study was
15 or 20 mg/kg/day thalidomide given by gavage to pregnant dams at days 26-28 of gestation
for teratological evaluation, and 20 mg/kg given on day 26 for gene expression profiling 6 h
post-treatment.
2. MATERIALS and METHODS
2.1. Teratological evaluation
The teratology study was performed at SNBL USA, Ltd. (Everett, WA, USA) in
compliance with the Animal Welfare Act and recommendations set forth in The Guide for the
Care and Use of Laboratory Animals [15]. Only females showing 25-32 day menstrual cycles
were used in these experiments. Each female monkey was paired with a male of proven fertility
for three days between days 11-15 of the menstrual cycle. When copulation was confirmed, the
median day of the mating period was regarded as day 0 of gestation. Pregnancy was confirmed
on day 20 or day 25 by ultrasound (SSD-4000, Aloka Co., Mitaka, Japan) under sedation
induced by intramuscular injection of 5% ketamine hydrochloride (Sigma Chemical Co., St
Louis, MO, USA). The monkeys were given (±)-thalidomide (Lot no. SEH7050, Wako Pure
Chemical Industries, Ltd., Osaka, Japan) at 15 or 20 mg/kg/day by oral administration using
gelatin capsules (Japanese Pharmacopiae grade) on days 26 to 28 of gestation. The dosage was
adjusted to the body weight on day 25 of gestation. Cesarean section was performed on day
Previous I TOC
-------
100-102 of gestation under deep anesthesia induced by intramuscular injection of 5% ketamine
hydrochloride (0.1-0.2 ml/kg) and inhalation of isoflurane (0.5-2.0%, Baxter, Liberty Corner,
NJ, USA). Salivation was inhibited by atropine (0.01 mg/kg, Phenix Pharmaceutical, St. Joseph,
MO, USA). Fetal viability was recorded, and the fetuses were euthanized by intraperitoneal
injection of pentobarbital and phenytoin solution (Euthasol®, Virbac Corp., Fort Worth, TX,
USA). Fetuses were sexed and examined for external anomalies after confirmation of the
arrested heart-beat. After the completion of external examinations, fetuses were examined for
internal abnormalities.
2. 2. Microarray experiments
The animal experiments were performed at Shin Nippon Biomedical Laboratories
(SNBL), Ltd. (Kagoshima, Japan) in compliance with the Guideline for Animal
Experimentation (1987), and in accordance with the Law Concerning the Protection and Control
of Animals (1973) and the Standards Relating to the Care and Management of Experimental
Animals (1980). This study was approved by the Institutional Animal Care and Use Committee
of SNBL and performed in accordance with the ethics criteria contained in the bylaws of the
SNBL committee.
Each female monkey was paired with a male of proven fertility for one day between
day 11 and day 15 of the menstrual cycle. Pregnant females, aged 5-8 years and weighing
2.84-3.76 kg on day 22 of gestation, were allocated randomly to two groups, each with three
monkeys, and housed individually. The monkeys were orally dosed with (±)-thalidomide (Lot
no. SDH7273/SDJ3347, Wako Pure Chemical Industries, Ltd., Osaka, Japan) at 0 or 20 mg/kg
Previous I TOC
-------
by oral administration of a gelatin capsule on day 26 of gestation, which was during the critical
period for thalidomide-induced teratogenesis [7,8]. Dosage was adjusted to the body weight on
day 22 of gestation. Control monkeys received the capsule only.
2.3. RNA sample collection
Hysterectomy was performed under terminal anesthesia at 6 h after the administration
of thalidomide on day 26 of gestation. Whole embryos were rapidly removed from the uterus
using a stereomicroscope and immersed in sterilized physiological saline. Three embryos each
in the thalidomide-treated and control groups were obtained for RNA analysis and stored at
-70 °C until further processing. General factors of maternal age, weight and date of processing
these samples are shown in Table 1. Embryos were processed simultaneously, and aside from
the blocking factors in Table 1, all 6 samples were handled concurrently through RNA isolation
and hybridization.
Table 1: Procurement of cynomolgus embryos at SNBL for microarray study
Group
control
thalidomide
Embryo
001
002
003
101
102
103
Maternal age
in years
6
1
8
5
6
8
Maternal bw
in kg (day 22)
3.76
2.84
3.68
2.97
3.01
3.14
Date of embryo
collection (day 26)
Nov. 2, 2006
Dec. 2, 2006
Dec. 2, 2006
Oct. 30, 2006
Nov. 6, 2006
Nov. 24, 2006
*. eel filename
(NIHS)
137255bpcynall.cel
137256bpcynall.cel
137257bpcynall.cel
137258bpcynall.cel
137259bpcynall.cel
137260bpcynall.cel
Previous
TOC
-------
2.4. RNA preparation and labeling
Total RNA was isolated from each day-26 embryo, amplified to cRNA, and biotin-labeled
for analysis on the Affymetrix NHP GeneChip® Array at Gene Logic Inc. (Gaithersburg, MD,
USA) using the TRIzol method and RNeasy columns according to protocols from Affymetrix
(Santa Clara, CA, USA). The 28S/18S rRNA ratio of isolated RNA was assessed using a
Bioanalyzer (Agilent Technologies, Palo Alto, CA, USA) and found to be of sufficiently high
quality. Biotinylated cRNA was finally cleaned up and fragmented by limited hydrolysis to a
distribution of cRNA fragment sizes below 200 bases.
2.5. Affymetrix NHP GeneChip® Array and hybridization
Biotinylated cRNA samples from control and exposed embryos (n=3 each) were
hybridized using Biogen Idec's (NASDAQ: BIIB) proprietary Affymetrix NHP GeneChip®
Array platform. This microarray chip contains a comprehensive representation of the
Cynomolgus genome derived from Biogen Idec's proprietary sequencing efforts, from which
Gene Logic (www.genelogic.com/) subsequently obtained the exclusive rights to provide as a
service (personal communication, Jun Mano, Gene Logic). The steps for hybridization followed
a protocol described in the Gene Logic GeneChip® Analysis Manual (Gaithersburg, MD, USA).
Probe-sets for this analysis consisted of cynomolgus expressed sequence tags (ESTs), published
rhesus monkey ESTs, predictive coding sequences from the rhesus genome, and human genes
not represented by monkey sequences. Because of the incomplete state of annotation for the
cynomolgus genome at the time this study was undertaken, we used human, mouse and rat gene
annotations to characterize monkey genes on the NHP GeneChip® Array. This reasonably
assumes that most cynomolgus sequences are well-annotated by human ortholog information.
Previous I TOC
-------
After hybridization the GeneChip® arrays were scanned and raw signal values were subjected
to subsequent normalization and processing.
2.6. Microarray data processing and analysis
Probe-level data normalization from the 6 *.cel files used the robust multichip average
(RMA) method with perfect-match (PM) but not mismatch (MM) data from the microarrays.
RMA returns a single file containing the 51,886 probes in 6 columns of normalized data,
representing the Iog2-intensity of each probe. To query differential transcript abundance
between sample groups, the Iog2 ratio of treated (Q) to reference (R) was computed for all 6
samples, with R being the average of the 3 controls. The 6 columns were centered to MEDIAN
= 0.00 and scaled to STDEV = 0.50 [10,12]. These data were loaded to GeneSpring GX7.3
software (Agilent Technologies, Redwood City CA, USA) for one-way analysis of variance
(ANOVA) by treatment group. Due to the small sample size (n=3) and limited annotation of the
cynomolgus genome for this preliminary analysis we relaxed the selection criterion by not
applying a false-discovery rate filter. Genes or probes passing the statistical (ANOVA) filter at a
P-value of 0.05 were subjected to K-means clustering, with cluster Set 1 and Set 2 that were
up-regulated and down-regulated, respectively, in the thalidomide-exposed versus control
embryos. Entrez gene identifiers were used for bioinformatics evaluation
(http://www.ncbi.nlm.nih.gov/).
3. RESULTS
3.1. Teratological evaluation
Previous I TOC
-------
To confirm thalidomide embryopathy in the cynomolgus colony under the conditions
used for this study, pregnant dams were given thalidomide at 15 and 20 mg/kg on days 26-28 of
gestation. Four fetuses were obtained at each dose for teratological evaluation (Table 2).
Although we did not observe a clear dose-response in this limited number of fetuses, we did
observe a number of cases with limb defects consistent with human thalidomide embryopathy.
Figure 1 shows external appearance of fetuses of dams exposed to thalidomide on GD 26-28.
Bilateral amelia in the fore-/hindlimbs was noted in one female fetus at 20 mg/kg, and bilateral
micromelia in the hindlimbs was observed in four fetuses at 15 mg/kg. Deformities of the paw
and/or foot including hyperflexion, ectrodactyly, polydactyly, syndactyly, brachydactyly, and/or
malpositioned digits, were observed in all fetuses at 15 mg/kg and in two fetuses at 20 mg/kg.
Tail anomalies were found in one fetus at 15 mg/kg and three fetuses at 20 mg/kg. Small penis
was noted in one fetus each in both thalidomide-treated groups. No internal abnormalities were
noted in any of the thalidomide-treated fetuses examined here. This confirmed the relevant
sensitivity of cynomolgus embryos to thalidomide, based on a maternally administered dose of
15-20 mg/kg during days 26-28 of gestation.
3.2. Genes altered by thalidomide
The embryonic transcriptome was evaluated at 6 h after 20 mg/kg maternal thalidomide
exposure on day 26. For this analysis, we used a proprietary Non-Human Primate (NHP)
microarray having representation of the cynomolgus genome (see Methods for details). The
NHP array includes 18,293 cynomolgus genes and 8,411 Rhesus genes as well as genes from
several other species. The 6-array dataset conforming to MIAME standards resides in the Gene
Expression Omnibus repository (www.ncbi.nlm.nih.gov/geo/) under platform accession number
10
Previous I TOC
-------
GPL8393 (series GSM389350-389355). A thalidomide-sensitive subset of genes in the
embryonic transcriptome was reflected in the high-percentage of present calls for genes whose
expression levels showed >1.5-fold difference between thalidomide-treated and control
embryos.
Statistical (ANOVA) analysis identified 2,362 genes that differed significantly between
control and thalidomide groups (P < 0.05). The heat map for these genes showed a clear pattern
(Figure 1). K-means clustering partitioned them into primary sets of up-regulated (1,281) genes
and down-regulated (1,081) genes for thalidomide relative to control embryos. Corresponding
files for the up-regulated (Set 1) and down-regulated (Set 2) genes are provided as a
supplement.
Previous
11
TOC
-------
Table 2. Morphological findings in fetuses of cynomolgus monkeys given thalidomide on days 26-28
Target
Findings
Forelimb
Amelia
Dose
Fetus no.
Gender
1 5 mg/kg
1234
Female Male Female Female
_
20 mg/kg
5678
Male Male Male Female
- - - B
Paw
Hyperflexion B -
Ectrodactyly L -
Accessory digit(s) * L —
Polydactyly * - R
Brachydactyly - -
Hindlimb
Micromelia B
Amelia - -
Foot
Hyperflexion - B
Ectrodactyly - B
Polydactyly - -
Syndactyly R -
Brachydactyly - -
Malpositioned digit(s) - -
B
B B B
B
R
B
L
B
R
B
B
B
Craniofacial
Trunk
Tail
Bent or curled tail
Short tail
External Genital Organs
Small penis
- : No anomaly was observed.
+: Anomaly was observed.
B: Bilateral anomaly was observed.
R: Unilateral (right side) anomaly was observed.
L: Unilateral (left side) anomaly was observed.
* Polydactyly means (almost) complete extra digits existed, and accessory digit incomplete "digit like tissue"
attached to a normal digit.
12
Previous
TOC
-------
Figure 1. Malformed fetuses of cynomolgus maternal monkeys exposed to thalidomide on CDs 26-28.
A) The fetus of maternal monkey given thalidomide at 15 mg/kg/day exhibiting brachydactyly in the
paw, micromelia in the hindlimb, hyperflexion, ectrodactyly and brachydactyly in the foot and curled
tail. B) The fetus of maternal monkey given thalidomide at 20 mg/kg/day exhibiting amelia in the fore-
and hindlimb and bent tail.
Previous
13
TOC
-------
Figure 2. Molecular abundance profiles of the thalidomide-sensitive genes in the cynomolgus
embryonic transcriptome on day 26 of gestation. RNA was isolated from day 26 embryos 6 h after
maternal exposure to 20 mg/kg thalidomide or vehicle control. Values represent Log2 ratios of
treated/reference, where the reference is an average of all three controls for each gene. ANOVA returned
2362 genes that were significantly different between the groups (n=3, P < 0.05). The heat map visualizes
the genes in rows and the embryos in columns, and the histogram shows the distribution of genes in each
cluster. Columns left to right: 1-3 from control embryos (#001, #002, #003) and 4-6 from thalidomide
embryos (#101, #102, #103). Genes were partitioned by K-means clustering into two primary expression
clusters with 1,281 up-regulated genes (red) and 1,081 down-regulated genes (green).
14
Previous
TOC
-------
3.3. Annotation systems
Ranking functional categories of genes in an expression cluster is an important step to
unravel the cellular functions and pathways represented in the differentially expressed gene list.
To derive the highest-ranking biological themes across the up-/down-regulated gene lists,
Entrez gene IDs were annotated by Gene Ontology (GO) category using the Database for
Annotation, Visualization, and Integrated Discovery (http://appsl.niaid.nih.gov/david/). Table 3
lists the significantly over-represented themes when the 1,281 up-regulated genes (Table 3 A)
and 1,081 down-regulated genes (Table 3B) were mapped by GO category. We used lev el-4
annotation for Biological Processes, Cellular component and Molecular Function as well as
curated pathways from the KEGG (Kyoto Encyclopedia of Genes and Genomes) open source
pathway resource to obtain categories passing by Fisher exact test (P < 0.05). For clarity and
greater specificity we limited the categories in Table 3 to those having at least 10 hits for
sensitivity and no more than 50 hits to improve specificity. Supplemental Table 3 provided
electronically includes the gene identifiers for each category.
Integrated biological processes evident across the up-regulated categories addressed the
regulation of cellular growth, including cell cycle progression, DNA repair and nucleic acid
transport. Other up-regulated biological processes addressed the regulation of metabolism, the
cytoskeletal cycle, heart development and vesicle transport. Many of these processes were
logically reflected in the ontologies for cellular components addressing the nucleo-ribosomal
system, the microtubule network, and molecular functions for GTPase activity and actin binding.
Up-regulated signaling pathways (KEGG) included several oncogenic growth pathways as well
as the TGF-beta, GnRH and insulin signaling pathways.
15
Previous I TOC
-------
Table 3A. GO-annotated biological categories for genes up-regulated in the embryo following maternal thalidomide
exposure
Category
GOTERM_BP_4
GO:0015931
GO:0050658
GO:0050657
GO:0051236
GO:0051028
GO:0045941
GO:0007507
GO:0051276
GO:0006281
GO:0022618
GO:0031325
GO:0009893
GO:0051169
GO:0016481
GO:0006461
GO:0045786
GO:0009892
GO:0031324
GO:0000074
GO:0051726
GO:0007010
GO:0016192
GOTERM_CC_4
GO:0005830
GO:0005681
GO:0000785
G 0:0031965
Term
Biological Process (level 4)
nucleobase, nucleoside, nucleotide and nucleic acid transport
RNA transport
nucleic acid transport
establishment of RNA localization
mRNA transport
positive regulation of transcription
heart development
chromosome organization and biogenesis
DNA repair
protein-RNA complex assembly
positive regulation of cellular metabolic process
positive regulation of metabolic process
nuclear transport
negative regulation of transcription
protein complex assembly
negative regulation of progression through cell cycle
negative regulation of metabolic process
negative regulation of cellular metabolic process
regulation of progression through cell cycle
regulation of cell cycle
cytoskeleton organization and biogenesis
vesicle-mediated transport
Cellular component (level 4)
cytosolic ribosome (sensu Eukaryota)
spliceosome
chromatin
nuclear membrane
'ount
15
13
13
13
11
40
15
45
28
12
42
44
14
28
27
19
38
32
42
42
41
39
PValue
0.001
0.002
0.002
0.002
0.007
0.000
0.006
0.000
0.001
0.035
0.000
0.000
0.035
0.003
0.005
0.022
0.002
0.009
0.005
0.005
0.008
0.013
List
Total
694
694
694
694
694
694
694
694
694
694
694
694
694
694
694
694
694
694
694
694
694
694
Pop
Hits
100
87
87
87
79
326
128
394
267
116
416
445
145
300
295
209
436
387
526
529
526
509
Pop
Total
13532
13532
13532
13532
13532
13532
13532
13532
13532
13532
13532
13532
13532
13532
13532
13532
13532
13532
13532
13532
13532
13532
Log 2 Fold
Change
+2.92
+2.91
+2.91
+2.91
+2.71
+2.39
+2.28
+2.23
+2.04
+2.02
+ 1.97
+ 1.93
+ 1.88
+ 1.82
+ 1.78
+ 1.77
+ 1.70
+ 1.61
+ 1.56
+ 1.55
+ 1.52
+ 1.49
10
16
22
15
0.017
0.004
0.001
0.012
743
743
743
743
76
134
194
136
14201
14201
14201
14201
+2.51
+2.28
+2.17
+2.11
16
Previous
-------
GO:0012506
GO:0005874
GO:0005635
GO:0005768
GO:0005694
GO:0030529
vesicle membrane
microtubule
nuclear envelope
endosome
chromosome
ribonucleoprotein complex
GO TERM MF 4 Molecular Function (level 4)
GO:0051427 hormone receptor binding
GO:0051020 GTPase binding
GO:0003712 transcription cofactor activity
GO:0003779 actin binding
GO:0008234 cysteine-type peptidase activity
KEGG_PATHWAY
hsa05220 Chronic myeloid leukemia
hsa05222 Small cell lung cancer
hsa05215 Prostate cancer
hsa04350 TGF-beta signaling pathway
hsa04912 GnRH signaling pathway
hsa04910 Insulin signaling pathway
Table 3B. GO-annotated biological categories for genes down-regulated in the embryo following maternal thalidomide
exposure
Category
GOTERM_BP_4
GO:0008284
GO:0007517
GO:0009889
GO:0006417
GO:0032940
GO:0001944
Term
Biological Process (level 4)
positive regulation of cell proliferation
muscle development
regulation of biosynthetic process
regulation of translation
secretion by cell
vasculature development
13
23
18
18
32
41
10
11
41
27
15
10
11
11
11
11
14
i the el*
Count
24
16
18
14
23
15
0.030
0.005
0.015
0.028
0.011
0.047
0.001
0.003
0.000
0.002
0.027
0.016
0.016
0.016
0.020
0.026
0.025
nbryofoll
PValue
0.000
0.006
0.005
0.027
0.004
0.026
743
743
743
743
743
743
578
578
578
578
578
225
225
225
225
225
225
'owing
List
Total
556
556
556
556
556
556
125
233
182
196
385
584
57
78
311
302
172
74
87
87
90
94
134
'.rnal ti
Pop
Hits
240
177
207
174
287
191
14201
14201
14201
14201
14201
14201
12599
12599
12599
12599
12599
4214
4214
4214
4214
4214
4214
halidomid
Pop
Total
13532
13532
13532
13532
13532
13532
+ 1.99
+ 1.89
+ 1.89
+ 1.76
+ 1.59
+ 1.34
+3.82
+3.07
+2.87
+ 1.95
+ 1.90
+2.53
+2.37
+2.37
+2.29
+2.19
+ 1.96
e
Log2Fold
Change
-2.43
-2.20
-2.12
-1.96
-1.95
-1.91
17
Previous
-------
GO:0045045
GO:0051246
GO:0006873
GO:0006954
GO:0016192
GO:0042127
GO:0019752
GO:0046907
GOTERM_CC_4
GO:0005625
GO:0005768
GO:0005789
GO:0044432
GO:0005624
GO:0005783
GOTERM_MF_4
GO:0030594
GO:0051020
GO:0016747
GO:0004175
secretory pathway
regulation of protein metabolic process
cellular ion homeostasis
inflammatory response
vesicle-mediated transport
regulation of cell proliferation
carboxylic acid metabolic process
intracellular transport
Cellular component (level 4)
soluble fraction
endosome
endoplasmic reticulum membrane
endoplasmic reticulum part
membrane fraction
endoplasmic reticulum
Molecular Function (level 4)
neurotransmitter receptor activity
GTPase binding
transferase activity, transferring other than amino-acyl groups
endopeptidase activity
KEGG_PATHWAY
hsa04640 Hematopoietic cell lineage
hsa04612 Antigen processing and presentation
18
23
16
22
35
34
36
40
0.020
0.008
0.031
0.012
0.004
0.005
0.012
0.043
556
556
556
556
556
556
556
556
239
307
214
301
509
499
572
714
13532
13532
13532
13532
13532
13532
13532
13532
12
10
0.005
0.024
223
223
85
80
4214
4214
-1.83
-1.82
-1.82
-1.78
-1.67
-1.66
-1.53
-1.36
21
15
28
30
44
46
0.004
0.039
0.031
0.047
0.026
0.049
602
602
602
602
602
602
244
196
435
494
749
827
14201 -:
14201
14201
14201
14201
14201
>.03
.81
.52
.43
.39
.31
14
11
15
31
0.000
0.002
0.028
0.012
531
531
531
531
99
78
188
463
12599
12599
12599
12599
-3.36
-3.35
-1.89
-1.59
-2.67
-2.36
Results for the embryo 6 h after a teratogenic dose of thalidomide (20 mg/kg) on day-26 of gestation for 1,281 significantly up-regulated genes
(Table 3A) and 1,081 significantly down-regulated genes (Table 3B) based on the population of arrayed genes. The annotated system used the
NIH/NIAID Database for Annotation, Visualization, and Integrated Discovery (DAVID) at level 4. Count refers to the number of altered genes in
the ontology (min = 10 and max = 50). P Value refers to results from Fisher exact test (P < 0.05); List Total refers to the number of annotated
genes on the array; Pop Hits and Pop Total refers to the number of annotated genes in the database for the category and overall; Log2 Fold
Change is computed as the mean Log2 (treated / control) for genes in the category . Note: see electronic supplement Table 2 for gene identifiers
in each listing.
18
Previous
-------
Integrated biological processes evident across the down-regulated categories addressed
ion homeostasis and cellular secretion. These processes were logically reflected in the
ontologies for cellular components addressing the endoplasmic reticulum, GTPase activity
and transferases. Other down-regulated biological processes addressed cell growth, muscle
and vasculature development, and the inflammatory response - consistent with KEGG
pathways for hematopoietic cells and antigen processing.
4. DISCUSSION
The results from this study show that a teratogenic dose of thalidomide (20 mg/kg)
significantly alters global gene expression profiles in the cynomolgus monkey embryo within
6 h of exposure on day 26 of gestation. Bioinformatics analysis of the embryonic
transcriptome following maternal thalidomide exposure revealed up-regulation in several
signaling pathways with roles in morphogenesis and oncogenesis (e.g., TGF-beta, insulin
signaling), and down-regulation of the endoplasmic reticulum and inflammatory response. As
might be anticipated, this implies a broad reaction of the embryo to the mechanism of
thalidomide and a generalized reprogramming of pathways known to be important in
development and teratogenesis.
The dosing scenario used in the present study was 15 or 20 mg/kg/day thalidomide given
by gavage to pregnant dams at days 26-28 of gestation for teratological evaluation, and 20
mg/kg given on day 26 for gene expression profiling 6 h post-treatment. The teratological
exposure induced limb malformations consistent with earlier studies with thalidomide in
19
Previous I TOC
-------
pregnant macaques. For example, it was previously reported that two fetuses with amelia
were obtained from two of four cynomolgus monkeys given thalidomide by gavage at 10
mg/kg/day on days 32 to 42 after commencement of menses (approximately equivalent to
days 20 to 30 of gestation) and that the fetal malformations were similar to malformations
reported in children whose mothers had taken thalidomide during pregnancy [7]. Forelimb
malformations in the cynomolgus fetus were noted following a single oral administration of
thalidomide on days 25, 26 or 27 of gestation at 10 and 30 mg/kg and daily administration on
days 25 to 27 of gestation at 5 mg/kg, and both fore- and hindlimb malformations were
observed following a single oral administration on day 25 or 28 of gestation at 30 mg/kg [8].
The present study, taken together with the previous studies [7,8], indicate that orally
administered thalidomide induces fetal malformations in cynomolgus monkeys similar to
human pregnancies and furthermore localizes the vulnerable period to days 25 to 28 of
gestation and the effective doses to 5 to 30 mg/kg/day.
Given the limitations of working with this species the preliminary application of a
custom NHP microarray, the analysis at one dose and time point, and the incomplete state of
annotation of the macaque genome, the current study design focused on RNA collected from
individual embryos rather than the specific target organ system (forelimb, hindlimb). Ideally a
follow-up study on focused gene expression analysis should be performed for specific
embryonic limbs in which malformations have been induced with thalidomide; however, the
present study is among the first to provide genomic information on the initial changes in gene
expression occurring in macaque embryos during the critical events following a teratogenic
dose of thalidomide. A total of 43 and 26 functional categories of redundant genes were up-
20
Previous I TOC
-------
and down-regulated, respectively, based on the GO annotation system for human Locus Link
identifiers.
Statistically, the top-ranked 20 up-regulated genes included 4 hits to cell shape and
polarity genes: KIAA0992 (twice), FNML2, FMNL3. Palladin, encoded by the KIAA0992
gene, plays a role in cytoskeletal organization, embryonic development, cell motility, and
neurogenesis [16]. Formin-related proteins play a role in Rho GTPase-dependent regulation
of the actin cytoskeletal cycle and have been implicated in morphogenesis, cell movement
and cell polarity [17]. Several genes in the focal adhesion/actin cytoskeleton pathway were
up-regulated. Guanine nucleotide exchange factors (GEFs) DOCK1, which forms a complex
with RhoG, and VAV2 and ARHGEF7 that act on Rho family GTPases, play a fundamental
role in small G-protein signaling pathways that regulate numerous cellular processes
including actin-cytoskeletal organization [18-22]. To further understand the mechanisms of
thalidomide induced-teratogenicity the regional and developmental stage of expression for
these genes and corresponding proteins should be determined; however, these preliminary
findings suggest that thalidomide perturbs a general program involving the up-regulation of
Rho family GTPases and their GEFs.
One candidate pathway for the control of cytoskeletal remodeling evident in studies
of early induction of the Fetal Alcohol Syndrome (FAS) in mouse embryos is the receptor
tyrosine kinase (RTK) signaling pathway, mediating insulin-like growth factors [12]. Genes
in the RTK insulin signaling pathway were significantly up-regulated by thalidomide
treatment as in FAS. AKT1 and GSK3(5, which were up-regulated by thalidomide, are key
21
Previous I TOC
-------
genes in this pathway. AKT1, a serine-threonine protein kinase, is regulated by PDGF and
insulin through PI-3 kinase signaling [23-25]. GSK3(5, a substrate of AKT, is a
proline-directed serine-threonine kinase that was initially identified as a phosphorylating and
inactivating glycogen synthase [26]. IGF-I and IGF-II are expressed in the anterior and
posterior mesodermal cells of the developing limbs [27-29]. IGF-I can influence chick limb
outgrowth [29-31] and regulate muscle mass during early limb myogenesis [32]. Although
these facts may implicate IGF signals as a potential mediator of thalidomide embryopathy, the
present study did not find significant expression or thalidomide-induced alteration in the
global pattern of several key transcripts in this signaling pathway, including IGFBPs 13, 5, 6
and 7, IGF1, IGF1R, and IRS 14 (data not shown). It is certainly plausible that thalidomide
exposure may locally alter upstream events in IGF-1 signaling without necessarily altering the
molecular abundance profiles of the pathway in the developing limb of monkey embryos. On
the other hand, our preliminary microarray analysis does find evidence for up-regulation of
GSK3(5 and AKT1 transcripts that are downstream in the insulin signaling pathway. Effects
on TGF-beta and WNT signaling may be critical here. Thalidomide-induced oxidative stress
in chick embryos can enhance signaling through BMPs (bone morphogenetic proteins),
leading to up-regulation of the WNT antagonist Dickkopfl (Dkkl) and subsequent cell death
[33]. We note here a significant up-regulation of genes in the TGF-beta pathway and
similarities with genes in the cytoskeletal cycle and WNT pathways for the murine FAS [12].
Some of the responsive genes found in this study are known to play roles in vascular
development pathways. For example, vascular endothelial growth factor (VEGF) was
down-regulated and platelet-derived growth factor receptor (5 (PDGFR(5) was up-regulated
22
Previous I TOC
-------
during early stages in thalidomide embryopathy. VEGF is a key stimulator of vascular cell
migration and proliferation and acts directly on endothelial cells, whereas PDGF attracts
connective tissue cells that can also stimulate angiogenesis. The reciprocal effect on these
transcript profiles, potentially leading to an overall decrease in VEGF/ PDGFRP activities,
might be predicted to interfere with vascular cell recruitment and proliferation in the
developing embryo or limb. It is well known that thalidomide reduces the activity or
production of VEGF and TNF-a, leading to inhibition of angiogenesis [34]. The present
microarray data are consistent with this effect. Furthermore, VEGF stimulates PDGFRP and
induces tyrosine phosphorylation [35]. The reciprocal effect that maternal thalidomide
exposure had on these transcripts may suggest a key event in the programming or induction of
vascular cells or their progenitors has been disrupted within 6 h after exposure. This notion is
supported by the study of D'Amato et al. [36] that suggested limb defects caused by
thalidomide were secondary to inhibition of blood vessel growth in the developing limb bud.
Down-regulation of the vascular development program is consistent with this notion and with
the supposition that correct limb bud formation requires a complex interaction of both
vasculogenesis and angiogenesis during development [37]. Perhaps these genes might be
considered as potential biomarkers of thalidomide-induced teratogenesis in cynomolgus
monkeys. A recent study with the teratogenic thalidomide metabolite, CPS49, has shown
direct evidence for suppression of endothelial angiogenetic sprouting and failure to establish a
normal vascular network as a key event in thalidomide embryopathy [38]. CPS49 mimics the
antiangiogenic properties, but not anti-inflammatory properties, of thalidomide.
23
Previous I TOC
-------
Finally, the inflammatory response pathway was found to be significantly
down-regulated in the early thalidomide embryome. Although down-regulation of the
inflammatory response might be anticipated to protect the embryo, studies in laboratory
animals have implicated a role for reactive oxygen species (ROS) in thalidomide
embryopathy [39]. In that study, thalidomide was found to preferentially increase ROS in
embryonic limb cells from a sensitive species (rabbit) but not the insensitive species (rat).
Down-regulation of the inflammatory pathways in thalidomide-exposed monkey embryos
reinforces this notion.
In conclusion, these findings show that thalidomide exposure perturbs a general
program of morphoregulatory processes in the cynomolgus monkey embryo. Bioinformatics
analysis has now identified many key pathways implicated in thalidomide embryopathy in
cynomologus monkeys, and has also revealed some novel processes that can help unravel the
mechanism of this important developmental phenotype. Several pathways, including actin
cytoskeleton remodeling and downstream insulin signaling-related genes, in addition to
vascular development pathways may provide candidate biomarkers for key events underlying
the teratogenicity of thalidomide in primates. To clarify the molecular mechanisms further
studies must examine protein expression, phosphorylation, and other modifications in the
precursor target organ system.
24
Previous I TOC
-------
ACKNOWLEDGEMENTS
This work was partially supported by Health and Labour Sciences Research Grants
(Research on Regulatory Science of Pharmaceuticals and Medical Devices:
H16-Kenkou-066; Research on Risk of Chemical Substances: H17-Kagaku-001;) from the
Ministry of Health, Labour and Welfare of Japan. The bioinformatics analysis was performed
at the National Center for Computational Toxicology, US EPA. Authors are grateful to Dr.
Robert MacPhail of EPA's National Health and Environmental Effects Research Laboratory
for helpful comments on the manuscript.
Disclaimer: The U.S. EPA, through its Office of Research and Development
collaborated in the research described here. It has been subjected to agency review and
approved for publication. The authors declare they have no competing financial interests.
25
Previous I TOC
-------
REFERENCES
[1] Schardein JL, Macina OT. Thalidomide: In: Human Developmental Toxicants-Aspects of
Toxicology and Chemistry, Boca Raton: CRC press, Taylor & Francis Group; 2007:
127-41.
[2] Hansen JM, Carney EW, Harris C. Differential alteration by thalidomide of the
glutathione content of rat vs. rabbit conceptuses in vitro. Reprod Toxicol
1999;13:547-54.
[3] Schardein JL. Thalidomide: the prototype teratogen: In: Chemically Induced Birth Defects,
3rd edition, revised and expanded, New York: Marcel Dekker Inc; 2000: 89-120.
[4] Teo SK, Denny KH, Stirling DI, Thomas SD, Morseth S, Hoberman AM. Effects of
thalidomide on developmental, peri- and postnatal function in female New Zealand white
rabbits and offspring. Toxicol Sci 2004;81:379-89
[5] Kalter H. Thalidomide: In: Teratology in the Twentieth Century-Congenital
Malformations in Humans and how their Environmental Causes were Established,
Amsterdam: Elsevier Science; 2003:167-75.
[6] Yoshida T. Introduction: In: The TPRC Handbook on the Care and Management of the
Laboratory Cymonolgus Monkey, Yoshida T, Fujimoto K eds, Tokyo: Springer Japan;
2006:1-3.
[7] Delahunt CS, Lassen LJ. Thalidomide syndrome in monkeys. Science 1964;146:1300-5.
[8] Hendrickx AG. The sensitive period and malformation syndrome produced by
thalidomide in the crab-eating monkey (Macacafascicularis). J Med Prim
1973;2:267-76.
26
Previous I TOC
-------
[9] Finnell RH, Gelineau-van Waes J, Eudy JD, Rosenquist TH. Molecular basis of
environmentally induced birth defects. Ann Rev Pharmacol Toxicol 2002;42:181-208.
[10] Singh AV, Knudsen KB, Knudsen TB. Computational systems analysis of
developmental toxicity: design, development and implementation of a birth defects
systems manager (BDSM). Reprod Toxicol 2005; 19,421-39.
[11] Daston, G.P. Genomics and developmental risk assessment. Birth Defects Res (Part A)
2007;79:l-7.
[12] Green ML, Singh AV, Zhang Y, Nemeth KA, Sulik KK, Knudsen TB. Reprogramming
of genetic networks during initiation of the fetal alcohol syndrome. Dev Dyn
2007;236:613-31.
[13] Knudsen TB and Kavlock RJ. Comparative bioinformatics and computational toxicology.
In: Developmental Toxicology volume 3, Target Organ Toxicology Series. (B Abbott and
D Hansen, editors) New York: Taylor and Francis, 2008;pp 311-360.
[14] Gene Logic. NHP GeneChip® Array Service [accessed September 14, 2007]
http://www.genelogic.com/docs/pdfs/NHP W.pdf
[15] Institute of Laboratory Animal Research, Commission of Life Sciences, National
Research Council. Guide for the Care and Use of Laboratory Animals, Washington DC:
The National Academies Press; 1996.
[16] Otey CA, Rachlin A, Moza M, Arneman D, Carpen O. The
palladin/myotilin/myopalladin family of actin-associated scaffolds. Int Rev Cytol
2005;246:31-58.
27
Previous I TOC
-------
[17] Yayoshi-Yamamoto S, Taniuchi I, Watanabe T. FRL, a novel formin-related protein,
binds to Rac and regulates cell motility and survival of macrophages. Mol Cell Biol
2000;20:6872-81.
[18] Marignani PA, Carpenter CL. Vav2 is required for cell spreading. J Cell Biol
2001;154:177-86.
[19] Brugnera E, Haney L, Grimsley C, Lu M et al. Unconventional Rac-GEF activity is
mediated through the Dockl80-ELMO complex. Nature Cell Biol 2002;4:574-582.
[20] Katoh H, Negishi M. RhoG activates Racl by direct interaction with the
DocklSO-binding protein Elmo. Nature 2003;424:461-4.
[21] Rosenberger G, Jantke I, Gal A, Kutsche K. Interaction of alphaPIX (ARHGEF6) with
beta-parvin (PARVB) suggests an involvement of alphaPIX in integrin-mediated
signaling. Hum Mol Genet 2003;12:155-67.
[22] Shin EY, Woo KN, Lee CS, Koo SH, Kim YG, Kim WJ, Bae CD, Chang SI, Kim EG.
Basic Fibroblast Growth Factor Stimulates Activation of Racl through a p85 PIX
Phosphorylation-dependent Pathway. J Biol Chem 2004;279:1994-2004.
[23] Burgering BM, Coffer PJ. Protein kinase B (c-Akt) in phosphatidylinositol-3-OH kinase
signal transduction. Nature 1995;376:599-602.
[24] Franke TF, Yang SI, Chan TO, Datta K, Kazlauskas A, Morrison DK, Kaplan DR,
Tsichlis PN. The protein kinase encoded by the Akt proto-oncogene is a target of the
PDGF-activatedphosphatidylinositol 3-kinase. Cell 1995;81:727-36.
[25] Kohn AD, Kovacina KS, Roth RA. Insulin stimulates the kinase activity of RAC-PK, a
pleckstrin homology domain containing ser/thr kinase. EMBO J 1995;14:4288-95.
28
Previous I TOC
-------
[26] Cross DA, Alessi DR, Cohen P, Andjelkovich M, Hemmings BA. Inhibition of glycogen
synthase kinase-3 by insulin mediated by protein kinase B. Nature 1995;378:785-9.
[27] Streck RD, Wood TL, Hsu MS, Pintar JE. Insulin-like growth factor I and II and
insulin-like growth factor binding protein-2 RNAs are expressed in adjacent tissues
within rat embryonic and fetal limbs. Dev Biol 1992;151:586-96.
[28] van Kleffens M, Groffen C, Rosato RR, van den Eijnde SM, van Neck JW,
Lindenbergh-Kortleve DJ, Zwarthoff EC, Drop SL. mRNA expression patterns of the
IGF system during mouse limb bud development, determined by whole mount in situ
hybridization. Mol Cell Endocrinol 1998;138:151-61.
[29] Stephens TD, Bunde CJ, Fillmore BJ. Mechanism of action in thalidomide teratogenesis.
Biochem Pharmacol 2000;59:1489-99.
[30] Dealy CN, Kosher RA. Studies on insulin-like growth factor-I and insulin in chick limb
morphogenesis. Dev Dyn 1995;202:67-79.
[31] Dealy CN, Kosher RA. IGF-I, insulin and FGFs induce outgrowth of the limb buds of
amelic mutant chick embryos.Development 1996;122:1323-30.
[32] Mitchell PJ, Johnson SE, Harmon K. Insulin-like growth factor I stimulates myoblast
expansion and myofiber development in the limb. Dev Dyn 2002;;223:12-23.
[33] Knobloch J, Shaughnessy JD Jr, and Riither U. Thalidomide induces limb deformities by
perturbing the Bmp/Dkkl/Wnt signaling pathway. FASEB J 2007;21:1410-21.
[34] Eisen T, Boshoff C, Mak I, Sapunar F, Vaughan MM, Pyle L, Johnston SR, Ahern R,
Smith IE, Gore ME. Continuous low dose Thalidomide: a phase II study in advanced
melanoma, renal cell, ovarian and breast cancer. Br J Cancer 2000;82:812-7.
29
Previous I TOC
-------
[35] Ball SG, Shuttleworth CA, and Kielty CM. Vascular endothelial growth factor can signal
through platelet-derived growth factor receptors. J Cell Biol 2007;177:489-00.
[36] D'Amato RJ, Loughnan MS, Flynn E, Folkman J. Thalidomide is an inhibitor of
angiogenesis. ProcNatl Acad Sci USA 1994;91:4082-5.
[37] Seifert R, Zhao B, Christ B. Cytokinetic studies on the aortic endothelium and limb bud
vascularization in avian embryos. Anat Embryol (Berl) 1992; 186:601-10.
[38] Therapontos C, Erskine L, Gardner ER, Figg WD, Vargesson N. Thalidomide induces
limb defects by preventing angiogenic outgrowth during early limb formation. PNAS
Early Edition 2009; www.pnas.org/cgi/doi/10.1073/pnas.0901505106.
[39] Hansen JM, Harris KK, Philbert MA, Harris C. Thalidomide modulates nuclear redox
status and preferentially depletes glutathione in rabbit limb versus rat limb. J Pharmacol
Exp Therap 2002;300:768-76.
30
Previous I TOC
-------
Systems Biology in Reproductive Medicine
Inhibition of Rat and Human Steroidogenesis by Triazole
Antifungals
Journal:
i
Manuscript ID:
i
Manuscript Type:
1
Date Submitted by the
Author:
Complete List of Authors:
i
Keywords:
Systems Biology in Reproductive Medicine
draft
Research Article
Dix, David; U.S. EPA
myclobutanil, CYP17, testosterone, triadimefon, propiconazole
i
Manuscript Central
URL: http:/mc.manuscriptcentral.com/sbrm Email: AARS@comcast.net
Previous
TOC
-------
Page 2 of 30 Systems Biology in Reproductive Medicine
1
2
3
4
5
6
„ Inhibition of Rat and Human Steroidogenesis by Triazole Antifungals
9
10
11
12 Amber K. Goetz1'2, John C. Rockett1, Hongzu Ren1, Inthirany Thillainadarajah1,
13
\\ David J.Dix1
16
17
18
19 Office of Research and Development, U.S. Environmental Protection Agency, Research Triangle
p-i
22 Park, NC 27711, USA; Department of Environmental and Molecular Toxicology, North Carolina
23
24 State University, Raleigh, NC 27695, USA
25
26
27
2g Correspondence and reprint requests:
29
30 Dr. David Dix
31
32 National Center for Computational Toxicology (D343-03)
33
_. Office of Research & Development
35 U.S. Environmental Protection Agency
36
37 Research Triangle Park, NC 27711
38
39 E-mail: dix.david@epa.gov.
40
J" Tel: 919 541 2701
42
43
44
45 Running Title: Triazole effects on Steroidogenesis
46
47
48
49
Disclaimer: The United States Environmental Protection Agency through its Office of Research and
ou
51
52 Development funded and managed the research described here. It has undergone Agency review and
53
54 been approved for publication.
oo
56
57
58
59
60
URL: http:/mc.manuscriptcentral.com/sbrm Email: AARS@comcast.net
Previous I TOC
-------
Systems Biology in Reproductive Medicine Page 3 of 30
1
2
3 Abstract
4
5 Environmental chemicals that alter steroid production could interfere with male reproductive
6
development and function. Three agricultural antifungal triazoles (myclobutanil, propiconazole and
9
10 triadimefon) that are known to modulate expression of cytochrome P450 (CYP) genes and
11
12 enzymatic activities were tested for effects on steroidogenesis in rat in vivo and in vitro, and human
I o
14
,.,- in vitro model systems. Hormone production was measured in testis organ cultures from untreated
16
17 adult and neonatal rats, following in vitro exposure to 1, 10, or 100 uM of myclobutanil or
18
19 triadimefon. Myclobutanil and triadimefon reduced media levels of testosterone by 40-68% in the
21
22 adult and neonatal testis culture, and altered steroid production in a manner that indicated CYP 17-
23
24 hydroxylase/17,20 lyase (CYP17A1) inhibition at the highest concentration tested. Rat to human
25
Oft
2° comparison was explored using the H295R (human adrenal adenocarcinoma) cell line. Following 48
28
29 hour exposure to myclobutanil, propiconazole or triadimefon at 1, 3, 10, 30, or 100 jiM, there was an
30
31 overall decrease in estradiol, progesterone and testosterone by all three triazoles. These data indicate
32
33
_. that myclobutanil, propiconazole and triadimefon are weak inhibitors of testosterone production in
35
36 vitro. However, in vivo exposure of rats to triazoles resulted in increased serum and intra-testicular
37
38 testosterone levels. This discordance could be due to higher concentrations of triazoles tested in
39
40
>,. vitro, and differences within an in vitro model system lacking neuroendocrine control.
42
43
44
45 Key Words: myclobutanil, triadimefon, propiconazole, testosterone, CYP 17
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
2
URL: http:/mc.manuscriptcentral.com/sbrm Email: AARS@comcast.net
Previous I TOC
-------
Page 4 of 30 Systems Biology in Reproductive Medicine
1
2
3 Introduction
4
5 Conazoles are triazole fungicides used for crop protection and pharmaceutical treatment of
6
fungal infections. They inhibit cytochrome P450 (CYP) 51 by competitively binding to the heme
9
1 o component of the CYP enzyme (Ghannoum and Rice, 1999). In fungal cells this binding depletes
11
12 ergosterol levels, which leads to a build up of precursor sterols in the cellular membrane, disrupting
13
14
^ c turgor pressure and triggering cytotoxicity. Conazoles disrupt other CYPs, including steroidogenic
16
17 CYPs (CYP17A1, CYP19A1) and consequently there is concern that inadvertent exposure to
18
19 agricultural conazoles may inhibit steroidogenesis and adversely affect reproduction in humans and
21
22 other mammalian species (Zarn et al., 2003).
23
24 It has been shown that some triazole conazoles inhibit aromatase (CYP 19) conversion of
25
Oft
2° testosterone to estrogen (Andersen et al., 2002; Trosken et al., 2004; Vinggaard et al., 2000). In
28
29 addition, there are reports that have investigated the effects of agricultural triazoles on testosterone
30
31 synthesis. The triazoles hexaconazole and flusilazole inhibited testosterone synthesis in Leydig cell
32
33
_. culture and increased the incidence of Leydig cell tumors in rats (Inchem IPCS, 1990, 1995). In
35
36 contrast, the triazoles myclobutanil and triadimefon, did not cause Leydig cell tumors in rats, but did
37
38 increase serum testosterone levels in adult male rats after 14 days. This occurred without affecting
39
40
41 serum luteinizing hormone (LH), estradiol, or stimulating Leydig cell hyperplasia (Goetz et al., 2007;
42
43 Tully et al., 2006).
44
45 The current study was designed to investigate the effect of myclobutanil, propiconazole and
46
47
43 triadimefon on testis testosterone synthesis in vivo or in vitro, and to determine whether and how
49
50 exposure to these triazoles alters testis testosterone levels in conjunction with increased serum
51
cp
^ testosterone levels. In the first experiment, the effects of triadimefon following exposure in vivo on
Oo
54
55 serum and testis testosterone levels were measured in rats to test the hypothesis that increased testis
56
57 testosterone production contributes to the reported elevated serum testosterone levels. We selected
58
59
60
O
URL: http:/mc.manuscriptcentral.com/sbrm Email: AARS@comcast.net
Previous I TOC
-------
Systems Biology in Reproductive Medicine Page 5 of 30
1
2
3 triadimefon because it had caused the most robust increases in serum testosterone in previous rat
4
5 experiments (Goetz et al., 2007; Tully et al., 2006). Serum LH was measured to determine if it
6
contributed to altered testosterone production, and serum estradiol to test if inhibition of aromatase
9
1 o occurred in vivo. In the second experiment, rat in vivo to in vitro comparisons were explored by
11
12 measuring intra-testicular testosterone production in organ cultures of neonatal and adult rat testes
13
14
^ c exposed to varying concentrations of either myclobutanil or triadimefon to assess in vitro testis
16
17 testosterone production. In the third experiment, rat to human comparisons were explored by
18
19 measuring hormone production (testosterone, estradiol, progesterone) in the H295R cell line
21
22 following exposure to myclobutanil, propiconazole or triadimefon.
23
24
25
2® Materials and Methods
28
29 Animal Husbandry. All animal procedures were approved by the U.S. Environmental Protection
30
31 Agency's National Health and Environmental Effects Research Laboratory Institutional Animal Care
32
33
_. and Use Committee. All animals were purchased from Charles River Laboratories (Raleigh, NC) and
35
36 housed in an Association for Assessment and Accreditation for Laboratory Animal Care-
37
38 International accredited facility. Animals were individually housed in polypropylene boxes
39
40
41 containing Alpha-Dri® bedding (Shepherd Specialty Papers, Watertown, TN), and subjected to a 12
42
43 hour: 12 hour lightdark cycle under controlled temperature (22 ± 2°C) and humidity (40-60%).
44
45 Animals were provided unlimited access to LabDiet 5002 Rodent Diet (PMI LabDiet, Richmond, IN)
46
47
43 and water. Feed was prepared by Bayer CropScience (Kansas City, MO) as part of a Materials
49
50 Cooperative Research and Development Agreement between the USEPA and the US Triazole Task
51
cp
^ Force. Control animals were fed 5002 Certified Rodent Diet. The triadimefon treated group received
Oo
54
55 feed containing 1800 ppm triadimefon.
56
57
58
59
60
4
URL: http:/mc.manuscriptcentral.com/sbrm Email: AARS@comcast.net
Previous I TOC
-------
Page 6 of 30 Systems Biology in Reproductive Medicine
1
2
3 In vivo dosing and sample collection. The treated animals started dietary exposure on postnatal day
4
5 (PND) 60. Male Wistar Han rats (n = 15) were fed rat chow 5002 containing 1800 ppm triadimefon.
6
Treatment lasted 30 days, body weight and feed were weighed on a weekly basis to determine dose.
9
10 Animals were tail bled 14 days into dosing (PND74) for testosterone measurements. On day 30
11
12 (PND 90) animals were decapitated between 08:30 and 10:30 and then necropsied. Trunk blood was
13
14
^ c collected for serum measurements of testosterone, estradiol, luteinizing hormone, and prolactin.
16
17 Liver, epididymis, ventral prostate, seminal vesicle, and pituitary were collected and weighed. Testes
18
19 were weighed and snap frozen in liquid nitrogen and stored at -80°C until analysis. The frozen right
21
22 testis was homogenized in cold Dulbecco's PBS (Gibco, Grand Island, NY) with an Ultra-Turrax T25
23
24 homogenizer (Janke-Kunkel IKA, Boutersem, Belgium), and centrifuged at 4°C for 10 min at 4,000
25
Oft
2° refusing a Beckman J2-21M centrifuge. Supernatant was then centrifuged with a 5417R centrifuge
28
29 (Eppendorf, Westbury, NY) at 20,000 rcf for 10 min at 4°C. Supernatant was collected and stored at
30
31 -80°C until analysis of intratesticular testosterone measurements.
32
34 Rat in vitro testis culture. Testes from adult (PND 90-100, n = 5-8) and neonatal (PND 1, n = 5
35
36 litters) Sprague Dawley rats were used in the in vitro study. Testis parenchyma was sliced into -100
37
38 mg pieces for each adult testis and neonatal testes were left intact. Testis tissue was incubated in 1.5
39
40
4,, ml of M199 media (Gibco, Grand Island, NY) supplemented with 0.2% bovine serum albumin
42
43 (Sigma, St. Louis, MO) and 10% charcoal/dextran treated fetal bovine serum (Hyclone, Logan, UT).
44
45 Human chorionic gonadotropin (hCG), an LH receptor agonist, was used to stimulate testosterone
46
47
43 production. Human chorionic gonadotropin (Sigma, St. Louis, MO) was added at 100 mU/mL to all
49
50 treatment groups except the negative controls. Technical grade (>95% purity) myclobutanil (LKT
51
rp
^ Laboratories Inc., St. Paul, MN) or triadimefon (Bayer CropScience, Kansas City, KS) was added to
Oo
54
55 a final concentration of 1, 10, or 100 jiM. Each test chemical was premixed with ethanol to aid
56
57 dilution; the final ethanol volume was 0.05% of the total culture volume. Positive control medium
58
59
60
URL: http:/mc.manuscriptcentral.com/sbrm Email: AARS@comcast.net
Previous I TOC
-------
Systems Biology in Reproductive Medicine Page 7 of 30
1
2
3 (plus hCG, minus test chemical) and negative control medium (minus hCG and test chemical) both
4
5 contained 0.05% ethanol. Tissues were incubated at 34°C in 2.0 ml siliconized tubes rotated at -10
6
rpm. At three time points, 0.5, 1.5, and 2.5 hr, the media was removed and replenished with fresh
9
1 o media containing the appropriate triazole and dose concentration. All treatments were replicated in
11
12 triplicate for each adult rat.
13
14
. c H295R cells. H295R human adrenocortical carcinoma cell lines were obtained from the American
16
17 Type Culture Collection (ATCC CRL-2128; ATCC, Manassas, VA) and grown in 75 cm2 flasks with
18
19 12.5 ml of supplemented medium at 37°C with a 5% CC>2 atmosphere. H295R cultures were
21
22 performed by Dr. Xiaowei Zhang in the Department of Zoology at Michigan State University, East
23
24 Lansing, MI. Supplemented medium was a 1:1 mixture of Dulbecco's modified Eagle's medium
25
26 with Ham's F-12 Nutrient mixture (Sigma, St. Louis, MO) with 15 mM HEPES buffer. The medium
28
29 was supplemented with 1.2 g/L Na2CC>3, ITS + Premix (1 ml Premix/ 100 ml medium), and 12.5 ml/
30
31 500 ml NuSerum (BD Biosciences, San Jose, CA). Final component concentrations in the medium
32
33
_. were as follows: 15 mM HEPES, 6.25 ug/ml insulin, 6.25 ug/ml transferring, 6.25 ng/ml selenium,
35
36 1.25 mg/ml bovine serum albumin, 5.35 ug/ml linoleic acid, and 2.5% NuSerum. The medium was
37
38 changed two to three times per week and cells were detached from flasks for subculturing by use of
39
40
41 trypsin/ EDTA (Sterile IX Trypsin-EDTA; Life Technologies Inc., Grand Island, NY). Cells were
42
43 exposed to triazoles in 6-well Tissue Culture Plates (Nalgene Nunc Inc., Rochester, NY). To
44
45 minimize potential effects of hormones in the serum during exposures, Nu-Serum was replaced with
46
47
43 2.5% charcoal dextran treated FBS (HyClone Laboratories, Inc. Logan, UT) immediately before cells
49
50 were exposed to triazoles at 1, 3, 10, 30, or 100 uM concentration in dimethyl sulfoxide (DMSO,
51
rp
^ Sigma-Aldrich, St. Louis, MO) for 48 h. At the end of this culture period media was collected and
Oo
54
55 shipped frozen to EPA for hormone measurements.
56
57
58
59
60
6
URL: http:/mc.manuscriptcentral.com/sbrm Email: AARS@comcast.net
Previous I TOC
-------
Page 8 of 30 Systems Biology in Reproductive Medicine
1
2
3 H295R cell viability. Protocol is adapted from Hilscherova et al., 2004. Cells were visually
4
5 inspected under a microscope to evaluate viability and cell numbers. Cell viability was determined
6
using the live/dead cell viability kit (Molecular Probes, Eugene, OR).
9
10 Hormone Measurements. Rat intratesticular levels of testosterone, androstenedione, 17alpha-
11
12 hydroprogesterone, and progesterone were measured from the culture media using coat-a-count 125I
13
14
^ c radioimmunoassay kits (Diagnostic Products Corporation, Los Angeles, CA). Serum LH was
16
17 measured using the rat disassociation enhanced lanthanide flourometric immunoassay (DELPHIA)
18
19 (Haavisto, 1993). Serum prolactin (PRL) was measured by radioimmunoassay according to
21
22 manufacturer's instructions, using materials supplied by the National Hormone and Pituitary Agency
23
24 for LH and PRL: iodination preparation 1-6, reference preparation RP-3, and antisera S-9. lodination
25
^ material was radiolabelled with 125I (DuPont/New England Nuclear) by a modification of the
28
29 chloramine-T method (Greenwood, 1963). Con6 control standards (Diagnostic Products Corporation,
30
31 Los Angeles, CA) were used to verify assay quality. Lactate dehydrogenase (LDH) levels in the
32
33
_. media, an indicator of cytotoxicity, were measured using an LDH detection kit (Roche Diagnostics,
35
36 Indianapolis, IN). Estradiol, progesterone, and testosterone levels in the H295R culture media were
37
38 assayed in duplicate using appropriate Coat-A-Count radioimmunoassay (RIA) kits (Diagnostic
39
40
41 Products Co., Los Angeles, CA) according to manufacturer's instructions.
42
43 Statistical analysis. The average of the three replicates for each adult rat testis was used for
44
45 statistical analysis. The litter was the unit of analysis for the neonatal culture. Rat in vitro hormone
46
47
43 data and LDH data, both loglO transformed, were analyzed using the SAS GLM procedure (SAS
49
50 Institute Inc., Gary, NC). Since each adult or litter was exposed to all eight treatments within the
51
rp
^ testis culture assay, animal or litter was used as a blocking factor within analyses of variance.
Oo
54
55 Pairwise t-tests were used to test for any differences between treatment groups and the control. Rat
56
57 in vivo hormone data (loglO transformed), body and tissue weight data were analyzed using a two-tail
58
59
60
7
URL: http:/mc.manuscriptcentral.com/sbrm Email: AARS@comcast.net
Previous I TOC
-------
Systems Biology in Reproductive Medicine Page 9 of 30
1
2
3 t-test. Statistical significance between control and treatment was set at p<0.05. H295R hormone data
4
5 was analyzed using ANOVA, measures with p<0.05 were considered significant. Students t-test was
6
used for further comparisons between control and treatment groups.
9
10
11
12 Results
13
14
^ c In vivo triadimefon effects. The mean dose received was 126 mg triadimefon/kg body weight/day.
16
17 Feed intake of treated males was on average 10% less than the control males over the course of
18
19 dosing. During the first week of treatment, body weights were significantly decreased 8-10% at 1800
21
22 ppm triadimefon compared to the controls, however, the rate of body weight gain was similar
23
24 between treatment groups throughout the remainder of the study (Figure 1). Four individuals (one
25
Oft
2° control and three treated) were removed from the study due to factors not related to treatment.
28
29 Liver weights were increased (27% absolute, 37% adjusted for body weight) after 30 days
30
31 exposure (Table 1). Absolute pituitary and paired epididymal weights were decreased (9.8% and
32
33
_. 5.2%, respectively) at necropsy. There were no treatment effects on the androgen-dependent tissues;
35
36 ventral prostate or seminal vesicle. Serum testosterone levels were unaffected following 2 weeks
37
38 exposure, however they were increased after 4 weeks exposure along with intra-testicular testosterone
39
40
41 levels (Table 1). Serum levels of estradiol, LH, and prolactin were elevated, but not statistically
42
43 significant.
44
45 Rat in vitro testis cultures. Testosterone production remained fairly constant (<11% change) in the
46
47
43 positive control adult testis tissue after each successive time point (0.5, 1.5, and 2.5 hr; hCG
49
50 stimulated) (Figure 2A). Testosterone levels increased ~ 88% between the 0.5 and 1.5 hr time point
51
rp
^ in the positive control neonatal testis, and remained constant during the 1.5 and 2.5 hr time point
Oo
54
55 (Figure 2B). Accounting for incubation time, testosterone production decreased slowly after each
56
57 successive time point in the negative control adult and neonatal testis cultures that were not
58
59
60
8
URL: http:/mc.manuscriptcentral.com/sbrm Email: AARS@comcast.net
Previous I TOC
-------
Page 10 of 30 Systems Biology in Reproductive Medicine
1
2
3 administered hCG and triazole treatment. The control data suggests hCG continued to stimulate
4
5 testosterone production in both adult and neonatal testis cultures; and testosterone production
6
0 decreased over time without hCG in the testis cultures.
o
9
1 o Following administration of each triazole in the adult testis cultures, testosterone levels were
11
12 reduced by a statistically significant amount only at the highest concentration tested (Figure 2A).
13
14
^ c Testosterone levels were reduced by myclobutanil or triadimefon by less than 50% following 100
16
17 |iM treatment compared to the respective control at 0.5 h with hCG. At 1.5 and 2.5 h, testosterone
18
19 levels decreased 63-68% following 100 jiM triadimefon treatment and ~ 50% following 100 jiM
21
22 myclobutanil treatment (Figure 2A). In the neonatal testis culture, testosterone levels were reduced
23
24 following myclobutanil (100 jiM ) and triadimefon (10 and 100 jiM) treatment, which suggests that
25
Oft
2° triadimefon might be the stronger inhibitor of the two (Figure 2B). Neonatal testosterone production
28
29 was more pronounced at the 0.5 h time point by myclobutanil and triadimefon compared to the 1.5
30
31 and 2.5 h time points. This may be due to testosterone levels increasing over time, as was seen in the
32
33
_. control groups containing hCG. Neonatal testosterone production was reduced -57% by triadimefon
35
36 (10 and 100 jiM) at 1.5 and 2.5 h. The 100 jiM myclobutanil treatment reduced neonatal testosterone
37
38 production by 65% and 40% at 1.5 and 2.5 h, respectively. The variability among the time points
39
40
41 was higher in the neonatal testis culture and likely due to the use of the litter vs. individual adults for
42
43 statistical analysis.
44
45 The hormone pattern after chemical treatment suggests CYP17A1 was inhibited. Media
46
47
43 levels of androstenedione were reduced by both chemicals in the adult (Figure 3 A) and neonatal
49
50 (Figure 3B) testis cultures, but not at all concentrations and time points that reduced testosterone
51
rp
^ levels. In the adult testis cultures, the levels of 17alpha-hydroxyprogesterone (Figure 4A) and
Oo
54
55 progesterone (Figure 5A) were increased by myclobutanil and triadimefon at lower concentration
56
57 levels than those that affected the androgens. In the neonatal testis cultures, the levels of 17alpha-
58
59
60
9
URL: http:/mc.manuscriptcentral.com/sbrm Email: AARS@comcast.net
Previous I TOC
-------
Systems Biology in Reproductive Medicine Page 11 of 30
1
2
3 hydroxyprogesterone (Figure 4B) and progesterone (Figure 5B) were also increased. The progestins
4
5 were not detected in the media of the neonatal testis cultures at all time points. LDH levels were
6
variable in the adult testis culture, which made cytotoxicity evaluation difficult. Analysis of the log
9
1 o transformed LDH data found no significant differences among the treatment groups, with the
11
12 exception of 100 uM triadimefon at the 2.5 hour time point. At this concentration and time point,
I o
14
. c LDH levels were significantly lower than the control (p < 0.011, data not shown). Cytotoxicity in the
16
17 neonatal testis cultures could not be evaluated since LDH was not detected in these cultures.
18
19 H295R cell viability. No adverse effects on cell growth or viability were observed following triazole
21
22 treatments (data not shown).
23
24 H295R hormone assays. Relative change in estradiol, progesterone and testosterone levels by
25
Oft
2° myclobutanil, propiconazole, or triadimefon in H295R cells are shown in Figure 6A/B. Myclobutanil
28
29 reduced the levels of estradiol in a concentration-dependent manner. Propiconazole and triadimefon
30
31 produced an increase in estradiol at the lower doses of 1 and 3 uM but a decrease as the dose
32
33
_. increased. All three triazoles decreased levels of progesterone at all doses assessed, however
35
36 progesterone levels returned to control levels at the highest dose of triadimefon. Testosterone levels
37
38 were decreased by all three triazoles in a consistent manner. In addition to the overall decrease in
39
40
41 hormone levels, estradiol levels were consistently greater than progesterone and testosterone levels
42
43 following propiconazole and triadimefon treatment indicating a possible disruption in the conversion
44
45 of estradiol, i.e. CYP17A1 activity. Established serum standards, a low, medium, and high control,
46
47
43 were used to quantitate variation among hormone assays. The intra-assay coefficient of variation
49
50 (CoV) range was 1-13 %, and the inter-assay CoV was 14-19 % among the low, medium, and high
51
rp
^ controls for testosterone assays. The androstenedione intra-assay CoV range was 0-9% and inter-
Oo
54
55 assay CoV was 7-17%. The 17alpha-hydroxyprogesterone intra-assay CoV range was 0-12% and
56
57
58
59
60
10
URL: http:/mc.manuscriptcentral.com/sbrm Email: AARS@comcast.net
Previous I TOC
-------
Page 12 of 30 Systems Biology in Reproductive Medicine
1
2
3 inter-assay CoV was 2-11%. The progesterone intra-assay CoV range was 2-12% and inter-assay
4
5 CoV was 19-21%.
6
7
8
9
10 Discussion
11
12 Toxicology studies using animals and in vitro cellular or tissue preparations have been used to
13
14
^ c study the toxic effects and mechanism of action of chemicals, to determine the effective and safe dose
16
17 of drugs in humans, and the risk of toxicity from chemical exposures. In vitro testing allows for a
18
19 specific evaluation of a chemical's ability to alter the synthesis of steroids, however the absorption,
21
22 distribution, metabolism, and elimination (ADME) of the chemical is not accounted for in such
23
24 testing (Gray et al., 1997). It has been demonstrated that several conazoles disrupt CYP enzymes,
25
Oft
2° including steroidogenic CYPs, and consequently there is concern that inadvertent exposure to
28
29 agricultural conazoles may inhibit steroidogenesis and adversely affect reproduction in humans and
30
31 other mammalian species (Zarn et al., 2003). In the present study we examined each chemical's
32
33
_. ability to affect testosterone production, which may be helpful in understanding the mode of action
35
36 for the reproductive effects observed in toxicology studies of conazoles. This series of experiments
37
38 was designed to address the hypothesis that the triazoles myclobutanil and triadimefon elicit their
39
40
41 reproductive effects through inhibition of testosterone production in the testis.
42
43 Results from this set of experiments demonstrate that all three triazoles were weak inhibitors
44
45 of testosterone production in vitro and suggest that, at least for myclobutanil and triadimefon,
46
47
4g inhibition of CYP17A1 occurs in vitro. Inhibition of CYP17A1 has been reported with several
49
50 imidazole compounds (Ayub and Levell, 1987; Engelhardt et al., 1991), and triazole compounds such
51
rp
^ as hexaconazole (Lloyd, 1991) and flusilazole (Inchem IPCS, 1995). Inhibition of the LH signal
Oo
54
55 transduction pathway, as stimulated by hCG, may have contributed to lowered testosterone
56
57 production. Although this hypothesis was not tested in our study design, the increased progestin
58
59
60
URL: http:/mc.manuscriptcentral.com/sbrm Email: AARS@comcast.net
Previous I TOC
-------
Systems Biology in Reproductive Medicine Page 13 of 30
1
2
3 levels in the cultures treated with myclobutanil or triadimefon suggest that hCG stimulation of the LH
4
5 pathway was not altered.
6
The data suggest that changes in hormone levels were not due to direct cytotoxicity. Indirect
9
1 o evidence from the progestin hormone data suggests an absence of cytotoxicity; levels of these
11
12 hormones continued to rise with increasing concentrations of the triazoles. This effect was also
I o
14
^ c observed in the adult testis culture, where no significant cytotoxicity was observed
16
17 Inhibition of testosterone synthesis was the hypothesis used to explain the increased incidence
18
19 of Ley dig cell tumors observed by the triazoles hexaconazole and flusilazole, and a similar
21
22 hypothesis was proposed that ketoconazole would induce similar tumors by the same mechanism if
23
24 tested under the USEPA guideline criteria (i.e., maximum tolerated dose and length of exposure)
25
Oft
27 (Cook et al., 1999). Triadimefon and myclobutanil appear to inhibit CYP17A1 in vitro, decrease
28
29 testosterone levels in vitro, but increase serum testosterone levels in vivo and do not induce Leydig
30
31 cell tumors. The difference across the triazole and imidazoles may be a result of the strength with
32
33
_. which different conazoles bind and inhibit metabolizing and/or steroidogenic CYPs. This marked
35
36 reduction in testosterone production by individual conazoles is likely due to a threshold dose
37
38 response as part of the mode of action for the production of Leydig cell tumors in rats.
39
40
41 In the dietary exposure experiment, triadimefon increased serum testosterone levels similar to
42
43 previous reports on triadimefon, myclobutanil and propiconazole (Goetz et al., 2007; Inchem IPCS,
44
45 1985; Tully et al., 2006). Serum testosterone levels were not increased during the first two weeks of
46
47
43 dosing suggesting this effect developed over time. The increased intra-testicular level of testosterone
49
50 by triadimefon suggests increased testis testosterone production contributes to the increase in serum
51
rp
^ testosterone. In addition, treatment related hepatic adaptive response (presumably causing a decrease
Oo
54
55 in liver testosterone metabolism and clearance) also likely contributed to the increased circulating
56
57 levels of testosterone.
58
59
60
12
URL: http:/mc.manuscriptcentral.com/sbrm Email: AARS@comcast.net
Previous I TOC
-------
Page 14 of 30 Systems Biology in Reproductive Medicine
1
2
3 The in vitro data demonstrated the opposite effect on testosterone production, producing a
4
5 discrepancy between the in vitro and in vivo results. Several triazole compounds have been examined
6
in both in vivo and in vitro studies, such as fluconazole (Hanger et al., 1988) and R76713 and its
9
1 o enantiomers (Wouters et al., 1990). In both cases, the inhibitory effects observed in vitro on
11
12 androgen synthesis and concomitant increase in the precursor progestins (Wouters et al., 1990),
13
14
^ c which is indicative for some effect on the CYP17A1 enzyme, were not repeated in the in vivo
16
17 experiments. As noted in these other studies, it is likely that the in vitro triadimefon concentrations
18
19 which significantly inhibit testosterone production (100 jiM) are not achieved through dietary
21
22 exposure in the adult rat, and another mechanism of action is stimulating the increased serum
23
24 testosterone levels at lower concentrations of triadimefon.
25
Oft
2° A similar situation occurs with the triazoles tebuconazole and epoxiconazole (Taxvig et al.,
28
29 2007, 2008). Reproductive toxicity studies with tebuconazole and epoxiconazole demonstrated
30
31 female virilization (increased anogenital distance) by both triazoles; fetal male feminization
32
33
_. following exposure to tebuconazole with concomitant decrease in fetal testosterone levels; and
35
36 increased testosterone levels in the dams following epoxiconazole which was observed following
37
38 myclobutanil exposure to pregnant dams as well (Goetz et al., 2007). Although the overall results
39
40
41 from these reproductive toxicity studies show that many of the azole fungicides behave similarly
42
43 following gestational exposure, the profile of action in vivo varies. The route of exposure (oral);
44
45 method of exposure, whether dietary or by gavage with a vehicle such as corn oil; and duration of
46
47
43 exposure will all certainly have a significant impact on the outcome of the studies. It is important to
49
50 define the effects observed at the high dose levels in these reproductive toxicity studies and
51
cp
^ extrapolate them to low dose effects.
Oo
54
55 The mechanism of action responsible for the increased serum and testis testosterone levels is
56
57 not clear. The elevated testosterone levels would be expected to reduce LH levels through a negative
58
59
60
1 ^
URL: http:/mc.manuscriptcentral.com/sbrm Email: AARS@comcast.net
Previous I TOC
-------
Systems Biology in Reproductive Medicine Page 15 of 30
1
2
3 feedback loop. However, triadimefon exposure did not significantly alter serum LH suggesting that
4
5 the hypothalamic-pituitary-gonadal (HPG) axis may be altered. Triadimefon could be acting as an
6
androgen receptor (AR) antagonist to stimulate increased testosterone production, but it is unlikely
9
1 o since high levels of triadimefon/triadimenol are needed to inhibit AR function (Okubo et al., 2004).
11
12 In addition, the androgen-dependent ventral prostate and seminal vesicle weights were unaffected in
13
14
^ c this study, and other studies do not show much evidence of AR antagonism in androgen-dependent
16
17 tissues (Goetz et al., 2007; Tully et al., 2006). Triadimefon has been reported to be an aromatase
18
19 inhibitor in vitro, but this inhibition was not evident in serum estradiol within our study and others
20 ' y
21
22 (Goetz et al., 2007; Tully et al., 2006) which suggests that an altered HPG regulation from reduced
23
24 estradiol levels did not occur.
25
Oft
2° One explanation for a disrupted HPG axis that has not been explored is by means of altered
28
29 neurotransmitters within the hypothalamus. Triadimefon exposure has been shown to affect behavior
30
31 presumably by altering neurotransmitters within the brain (Crofton et al., 1988; Reeves et al., 2004a,
32
33
,. 2004b; Walker and Mailman, 1996). If triadimefon is affecting neurotransmitters within the
35
36 hypothalamus, this mechanism could disrupt the HPG axis. This hypothesis was investigated by
37
38 measuring serum PRL levels as a proxy of altered dopamine levels within the hypothalamus (Waeber
39
40
41 et al., 1983), but there was no effect. The mechanism by which triadimefon increases testis
42
43 testosterone production requires further investigation and should include examination of
44
45 hypothalamic-pituitary axis regulation.
46
47
43 In summary, triadimefon, myclobutanil and propiconazole show weak inhibition of
49
50 testosterone production in vitro in rat testis cultures, and the hormone data suggests inhibition of
51
rp
^ CYP17A1. However, in vivo triadimefon increased rat testis testosterone production and serum
Oo
54
55 testosterone levels. The mechanism of action for increased testis testosterone levels in vivo is
56
57 unresolved, but may possibly involve a disruption of the HPG axis. Further studies into the effects of
58
59
60
14
URL: http:/mc.manuscriptcentral.com/sbrm Email: AARS@comcast.net
Previous I TOC
-------
Page 16 of 30
Systems Biology in Reproductive Medicine
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
triadimefon and other triazoles on the pituitary, hypothalamus, testis, and steroidogenesis will be
needed to fully elucidate the dose-response of mechanisms relevant to human health risk assessments.
Acknowledgements:
AKG was supported by U.S. EPA and N.C. State University Cooperative Training Agreement No.
CT826512010.
URL: http:/mc.manuscriptcentral.com/sbrm Email: AARS@comcast.net
15
Previous
TOC
-------
Systems Biology in Reproductive Medicine Page 17 of 30
1
2
3 References
4
5 Andersen, H.R., Vinggaard, A.M., Rasmussen, T.H., Gjermandsen, I.M., Bonefeld-J0rgensen, E.G.
6
(2002). Effects of Currently Used Pesticides in Assays for Estrogenicity, Androgenicity, and
9
10 Aromatase Activity In Vitro. Toxicology and Applied Pharmacology. 179, 1-12.
11
12
13
14
,.,- Ayub M. and Levell M. J. (1987). Inhibition of Testicular 17alpha-hydroxylase and 17,20 lyase but
16
17 not 3beta-hydroxysteroid dehydrogenase or 17beta-hydroxysteroid oxidoreductase by Ketoconazole
18
19 and Other Imidazole Drugs. Journal of Steroid Biochemistry. 28, 521-531.
21
22
23
24 Cook J. C., Klinefelter G. R., Hardisty J. H., Sharpe R. M., Foster P. M. D. (1999). Rodent Leydig
25
Oft
2° Cell Tumorigenesis: A Review of the Physiology, Pathology, Mechanisms, and Relevance to
28
29 Humans. Critical Reviews in Toxicology. 29, 169-261.
30
31
32
34 Crofton K. M., Boncek V. M., Reiter L. W. (1988). Hyperactivity Induced by Triadimefon, A
35
36 Triazole Fungicide. Fundamental and Applied Toxicology. 10, 459-465.
37
38
39
40
41 Engelhardt D., Weber M. M., Miksch T., Abedinpour F., Jaspers C. (1991). The Influence of
42
43 Ketoconazole on Human Adrenal Steroidogenesis: Incubation Studies with Tissue Slices. Clinical
44
45 Endocrinology. 35, 163-168.
46
47
48
49
50 Ghannoum, M.A. and Rice, L.B. (1999). Antifungal Agents: Modes of Action, Mechanisms of
51
rp
^ Resistance, and Correlations of the Mechanisms with Bacterial Resistance. Clinical Microbiology
Oo
54
55 Reviews. 12, 501-517.
56
57
58
59
60
16
URL: http:/mc.manuscriptcentral.com/sbrm Email: AARS@comcast.net
Previous I TOC
-------
Page 18 of 30 Systems Biology in Reproductive Medicine
1
2
3 Goetz, A.K., Ren, H., Schmid, I.E., Blystone, C.R., Thillainadarajah, I, Best, D.S., Nichols, H.,
4
5 Strader, L.F., Narotsky, M.G., Wolf, D.C., Rockett, J.C., Dix, D. J. (2007). Disruption of Testosterone
6
Homeostasis as a Mode of Action for the Reproductive Toxicity of Triazole Fungicides in the Male
9
10 Rat. lexicological Sciences. 95,227-239.
11
12
13
14
^ c Gray L. E., Kelce W. R., Wiese T., Tyl R., Gaido K., Cook J., Klinefelter G., Desaulniers D., Wilson
16
17 E., Zacharewski T., Waller C., Foster P., Laskey J., Peel J., Giesy J., Laws S., McLachlan J., Breslin
18
19 W., Cooper R., Giulio R. Di, Johnson R., Purdy R., Mihaich E., Safe S., Sonnenschein C., Welshons
21
22 W., Miller R., McMaster S., Colborn T. (1997) Endocrine Screening Methods Workshop: Detection
23
24 of Estrogenic and Androgenic Hormonal and Antihormonal Activity for Chemicals that Act Via
25
Oft
Receptor or Steroidogenic Enzyme Mechanisms. Reproductive Toxicology. 11, 719-750.
28
29
30
31 Greenwood F.C., Hunter W. M., Glover J. S. (1963). The Preparation of 1-131-Labelled Human
32
33
_. Growth Hormone of High Specific Radioactivity. The BiochemicalJournal. 89, 114-123.
35
36
37
38 Hanger, D.P., Jevons, S., and Shaw, J.T.B. (1988). Fluconazole and Testosterone: In Vivo and In
39
40
4,, Vitro Studies. Antimicrobial Agents and Chemotherapy. 32, 646-648.
42
43
44
45 Haavisto AM, Pettersson K, Bergendahl M, Perheentupa A, Roser JF, Huhtaniemi I. (1993) A
46
47
43 Supersensitive Immunofluorometric Assay for Rat Luteinizing Hormone. Endocrinology. 132, 1687-
49
50 1691.
51
52
53
54
55
56
57
58
59
60
17
URL: http:/mc.manuscriptcentral.com/sbrm Email: AARS@comcast.net
Previous I TOC
-------
Systems Biology in Reproductive Medicine Page 19 of 30
1
2
3 Hilscherova, K., Jones, P.D., Gracia, T., Newsted, J.L., Zhang, X., Sanderson, J.T., Yu, R.M.K., Wu,
4
5 R.S.S., Giesy, J.P. (2004). Assessment of the effects of chemicals on the expression often
6
steroidogenic genes in the H295R cell line using real-time PCR, Toxicol. Sci. 81, 78-89.
9
10
11
12 Inchem IPCS, FAO/WHO (1985). Joint Meeting on Pesticide Residues: 733. Triadimefon, Pesticide
13
^ c residues in food evaluations Part II Toxicology.
16
17
18
19 Inchem IPCS, FAO/WHO (1990). Joint Meeting on Pesticide Residues: 810. Hexaconazole,
21
22 Pesticide Residues in Food Evaluations in Toxicology.
23
24
25
26 Inchem IPC S, F AO/WHO (1995). Joint Meeting on Pesticide Residues: 896. Flusilazole, Pesticide
28
29 Residues in Food Evaluations Part II Toxicological & Environmental.
30
31
32
33
_. Lloyd S. C. (1991). Effects of hexaconazole (ICIA523) on the steroidogenic function of isolated rat
35
36 and human Leydig cells. ICI Central Toxicology Laboratory.
37
38
39
40
41 Okubo T., Yokoyama Y., Kano K., Soya Y., Kano I. (2004). Estimation of the Estrogenic and
42
43 Antiestrogenic Activities of Selected Pesticides by MCF-7 Cell Proliferation Assay. Archives of
44
45 Environmental Contamination and Toxicology. 46, 445-453.
46
47
48
49
50 Reeves R., Thiruchelvam M., Cory-Slechta D. A. (2004a) Development of Behavioral Sensitization
51
^ to the Cocaine-Like Fungicide Triadimefon is Prevented by AMP A, NMD A, DA Dl but not DA D2
Oo
54
55 Receptor Antagonists. Toxicological Sciences. 79, 123-136.
56
57
58
59
60
18
URL: http:/mc.manuscriptcentral.com/sbrm Email: AARS@comcast.net
Previous I TOC
-------
Page 20 of 30 Systems Biology in Reproductive Medicine
1
2
3 Reeves R., Thiruchelvam M., Cory-Slechta D. A. (2004b). Expression of Behavioral Sensitization to
4
5 the Cocaine-Like Fungicide Triadimefon is Blocked by Pretreatment with AMP A, NMDA and DA
6
Dl Receptor Antagonists. Brain Research. 1008, 155-167.
9
10
11
12 Taxvig, C., Hass, U., Axelstad, M., Dalgaard, M., Boberg, J., Andersen, H.R., Vinggaard, A.M.
13
14
. c (2007). Endocrine-Disrupting Activities In Vivo of the Fungicides Tebuconazole and Epoxiconazole.
16
17 Toxicological Sciences. 100, 464-473.
18
19
20
22 Taxvig, C., Vinggaard, A.M., Hass, U., Axelstad, M., Metzdorff, S., Nellemann, C. (2008).
23
24 Endocrine-Disrupting Properties In Vivo of Widely Used Azole Fungicides. InternationalJournal of
25
26 Andrology.3l, 170-177.
28
29
30
31 Trosken, E.R., Schloz, K., Lutz, R.W., Volkel, W., Zarn, J.A., Lutz, W.K. (2004). Comparative
32
33
_. Assessments of the Inhibition of Recombinant Human CYP19 (aromatase) by Azoles Used in
35
36 Agriculture and as Drugs for Humans. Endocrine Research. 30, 387-394.
37
38
39
40
41 Tully, D.B., Bao, W., Goetz, A.K., Blystone, C.R., Ren, H., Schmid, I.E., Strader, L.F., Wood, C.R.,
42
43 Best, D.R., Narotsky, M.G., Wolf, D.C., Rockett, J.C., Dix, D.J. (2006). Gene Expression Profiling
44
45 in Liver and Testis of Rats to Characterize the Toxicity of Triazole Fungicides. Toxicology and
46
47
43 Applied Pharmacology. 215,260-273.
49
50
51
^ Vinggaard, A.M., Hnida, C., Breinholt, V., Larson, J.C. (2000). Screening of Selected Pesticides for
Oo
54
55 Inhibition of CYP19 Aromatase Activity In Vitro. Toxicology In Vitro. 14, 277-234.
56
57
58
59
60
19
URL: http:/mc.manuscriptcentral.com/sbrm Email: AARS@comcast.net
Previous I TOC
-------
Systems Biology in Reproductive Medicine Page 21 of 30
1
2
3 Waeber C., Reymond O., Reymond M., Lemarchand-Beraud T. (1983). Effects of Hyper- and
4
5 Hypoprolactinemia on Gonadtrophin Secretion, Rat Testicular Luteinizing Hormone/Human
6
Chorionic Gonadotropin Receptors and Testosterone Production by Isolated Leydig Cells. Biology of
Q
1 o Reproduction. 28,167-177.
11
12
13
14
,.,- Walker Q. D. and Mailman R. B. (1996). Triadimefon and Triadimenol: Effects on Monoamine
16
17 Uptake and Release. Toxicology and Applied Pharmacology. 139, 227-233.
18
19
20
21
22 Wouters, W., De Coster, R., van Dun, J., Krekels, M.D.W.G., Dillen, A., Raeymaekers, A., Freyne,
23
24 E., Van Gelder, J., Sanz, G., Venet, M., Janssen, M. (1990). Comparative Effects of the Aromatase
25
26 Inhibitor R76713 and of its Enantiomers R83839 and R83842 on Steroid Biosynthesis In Vitro and In
28
29 Vivo. Journal of Steroid Biochemistry and Molecular Biology.37, 1049-1054.
30
31
32
33
_. Zarn, J.A., Bruschweiler, B.J., Schlatter, J.R. (2003). Azole Fungicides affect Mammalian
35
36 Steroidogenesis by Inhibiting Sterol 14alpha-demthylase and Aromatase. Environmental Health
37
38 Perspectives. Ill, 255-261.
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
20
URL: http:/mc.manuscriptcentral.com/sbrm Email: AARS@comcast.net
Previous I TOC
-------
Page 22 of 30
Systems Biology in Reproductive Medicine
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
Table 1: Weight and hormone measurements (± SEM) from control and treated animals following 30
days dietary exposure to triadimefon .
Parameter
Body Weight (g)
Liver Weight (g)
Live weight adjusted for body weight
Testis weight (g)
Epididymides (g)
Ventral Prostate (g)
Seminal Vesicle (g)
Pituitary (mg)
Serum T (14 day) (ng/mL)
Serum T (30 day) (ng/mL)
Intra-testicular T (ng/mL)
LH (ng/mL)
Estradiol (pg/mL)
Prolactin (ng/mL)
Control"
0 ppm
366.08 ±5. 63
12.977 ±0.282
3.54
3.474 ±0.055
1.162 ±0.016
0.379 ±0.020
1.148 ±0.044
10.2 ±0.02
2.63 ±0.60
2.15 ±0.39
33.70 ±5. 71
0.433 ±0.069
15.04 ±1.30
4.81 ±0.82
Triadimefon
ISOOppm
338.25 ±5. 84**
16.499 ±0.447***
4.88
3.485 ±0.066
1.102 ±0.020*
0.371 ±0.017
1.133 ±0.052
9.2 ±0.02*
2.72 ±0.46
6.20 ±1.26**
70.76 ±12.05 **
0.680 ±0.128
18.08±1.10
8. 00 ±1.99
Percent change
92.4
127.1
137.8
-
94.8
-
-
90.2
-
288.3
210.0
-
-
-
control animals (n = 14) and treated animals (n=12); * p < 0.05, ** p < 0.01, *** p < 0.0001.
URL: http:/mc.manuscriptcentral.com/sbrm Email: AARS@comcast.net
21
Previous
TOC
-------
Systems Biology in Reproductive Medicine Page 23 of 30
1
3 FIGURE LEGENDS
4
5 Figure 1: Body weight of control and triadimefon treated rats over the course of dosing. Solid line =
6
control, dashed line = triadimefon ISOOppm treatment. ** = p<0.01,*** = p< 0.001.
9
1 o Figure 2: In Vitro testosterone production by the adult (A) and neonatal (B) testis after myclobutanil
11
12 (Myc) and triadimefon (Tri) exposure. Asterisks indicate a significant difference (* p < 0.05, ** p <
13
14
^ c 0.01, and *** p < 0.001) between the treatment group and control group ((+), hCG with no chemical)
16
1 7 at each time point. (-) refers to tissue not stimulated by hCG and without chemical treatment.
18
1 9 Figure 3: In Vitro androstenedione production by the adult (A) and neonatal (B) testis after
21
22 myclobutanil (Myc) and triadimefon (Tri) exposure. Asterisks indicate a significant difference (* p
23
24 < 0.05, ** p < 0.01, and *** p < 0.001) between the treatment group and control group ((+), hCG
25
Oft
2° with no chemical) at each time point. (-) refers to tissue not stimulated by hCG and without chemical
28
29 treatment.
30
31 Figure 4: In Vitro 17alpha-hydroxyprogesterone production by the adult (A) and neonatal (B) testis
32
33
_. after myclobutanil (Myc) and triadimefon (Tri) exposure. Asterisks indicate a significant difference
35
36 (* p < 0.05, ** p < 0.01, and *** p < 0.001) between the treatment group and control group ((+),
37
38 hCG with no chemical) at each time point. (-) refers to tissue not stimulated by hCG and without
39
40
41 chemical treatment.
42
43
44
45 myclobutanil (Myc) and triadimefon (Tri) exposure. Asterisks indicate a significant difference (* p
46
47
43
49
50
51
52
treatment.
Figure 5: In Vitro progesterone production by the adult (A) and neonatal (B) testis after
myclobutanil (Myc) and triadimefon (Tri) exposure. Asterisks indicate a significant difference
< 0.05, ** p < 0.01, and *** p < 0.001) between the treatment group and control group ((+), hCG
with no chemical) at each time point. (-) refers to tissue not stimulated by hCG and without chemical
54
55 Figure 6: Hormone measures from H295R cell media. (A) Each panel presents progesterone,
56
57 estradiol and testosterone changes, for each of the three triazoles (myclobutanil, propiconazole,
58
59
60
22
URL: http:/mc. manuscriptcentral.com/sbrm Email: AARS@comcast.net
Previous I TOC
-------
Page 24 of 30
Systems Biology in Reproductive Medicine
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
triadimefon). (B) Each panel presents results with all three triazoles, for each of the three hormones.
All data presented as percent change relative to controls. * p<0.05, ** p<0.01, *** p<0.001.
URL: http:/mc.manuscriptcentral.com/sbrm Email: AARS@comcast.net
23
Previous
TOC
-------
Systems Biology in Reproductive Medicine
Page 25 of 30
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
Figure 1:
400-
200
• Control
•Triadimefon 1800ppm
60 65 70 75 80 85 90
Post-natal Day
URL: http:/mc.manuscriptcentral.com/sbrm Email: AARS@comcast.net
24
Previous
TOC
-------
Page 26 of 30
Systems Biology in Reproductive Medicine
Figure 2A:
0.7 -i
§" 0.6 -
LLI
CO
±, 0.5
O)
E
c
o>
s
0)
ts
o
ts
0)
0.4 -
0.3 -
0.2 -
0.1 -
0.0
JL***
to
Myc 1 Myc 10 Myc 100 Tri 1 Tri 10 Tri 100
Treatments
Figure 2B:
30 n
Myc 1 Myc 10 Myc 100 Tri 1
Treatments
Tri 10 Tri 100
URL: http:/mc.manuscriptcentral.com/sbrm Email: AARS@comcast.net
25
Previous
TOC
-------
Systems Biology in Reproductive Medicine
Page 27 of 30
Figure 3A:
0.16 -i
Myc 1 Myc 10 Myc 100 Tri 1 Tri 10 Tri 100
Treatments
Figure 3B:
i
Myc 1 Myc 10 Myc 100 Tri 1
Treatments
Tri 10 Tri 100
URL: http:/mc.manuscriptcentral.com/sbrm Email: AARS@comcast.net
26
Previous
TOC
-------
Page 28 of 30
Systems Biology in Reproductive Medicine
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
Figure 4A:
0.25 n
0.00
Myc1 MydO MyclOO Tri 1 Tri 10 Tri 100
Treatments
Figure 4B:
-. 3.5 -,
LU
CO
3.0 -
to 2.5
S
1*2.0
0)
l"<
I/)
0)
O) 1.0
Q
Q.
I 0.5 H
0.0
Myc 1 Myc 10 Myc 100
Treatments
Tri1
Tri 10 Tri 100
URL: http:/mc.manuscriptcentral.com/sbrm Email: AARS@comcast.net
27
Previous
TOC
-------
Systems Biology in Reproductive Medicine
Page 29 of 30
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
Figure 5A:
I
0.00
Myc1 MydO MydOO Tri 1 Tri 10 Tri 100
Treatments
Figure 5B:
3.0 -i
Myc1 MydO MydOO Tri 1 Tri 10 Tri 100
Treatments
URL: http:/mc.manuscriptcentral.com/sbrm Email: AARS@comcast.net
28
Previous
TOC
-------
Page 30 of 30
Systems Biology in Reproductive Medicine
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
Figure 6:
A 120%
100%
80%
3
-------
Regulatory Toxicology and Pharmacology xxx (2009) xxx-xxx
Contents lists available at ScienceDirect
Regulatory Toxicology and Pharmacology
journal homepage: www.elsevier.com/locate/yrtph
Evaluation of high-throughput genotoxicity assays used in profiling the US EPA
ToxCast™ chemicals
Andrew W. Knighta'*, Stephen Littleb, Keith Houckb, David Dixb, Richard Judsonb, Ann Richard5,
Nancy McCarrollc, Gregory Akermanc, Chihae Yangd, Louise Birrell3, Richard M. Walmsley3
" Centronix Ltd., CTF Building, 46 Crafton Street, Manchester, M13 9NT, UK
b National Center for Computational Toxicology (D343-03), Office of Research and Development, US Environmental Protection Agency, Research Triangle Park, NC 27711, USA
c Health Effects Division, Office of Pesticide Programs, US Environmental Protection Agency, 1200 Pennsylvania Ave., NW (MC 7509P), Washington, DC 20460, USA
d Office of Food Additive Safety (HFS-275), Center for Food Safety and Applied Nutrition, US Food and Drug Administration, College Park, MD 20740, USA
ARTICLE INFO
Article history:
Received 27 April 2009
Available online xxxx
Keywords:
Genotoxicity
In vitro
High-throughput screening
ToxCast
Pesticides
Hazard assessment
GreenScreen HC
CellCiphr
CellSensor
p53
CADD45 alpha
ABSTRACT
Three high-throughput screening (HTS) genotoxicity assays—GreenScreen HC GADD45a-GFP (Centronix
Ltd.), CellCiphr p53 (Cellumen Inc.) and CellSensor p53RE-bla (Invitrogen Corp.)—were used to analyze
the collection of 320 predominantly pesticide active compounds being tested in Phase I of US. Environ-
mental Protection Agency's ToxCast™ research project. Between 9% and 12% of compounds were positive
for genotoxicity in the assays. However, results of the varied tests only partially overlapped, suggesting a
strategy of combining data from a battery of assays. The HTS results were compared to mutagenicity
(Ames) and animal tumorigenicity data. Overall, the HTS assays demonstrated low sensitivity for rodent
tumorigens, likely due to: screening at a low concentration, coverage of selected genotoxic mechanisms,
lack of metabolic activation and difficulty detecting non-genotoxic carcinogens. Conversely, HTS results
demonstrated high specificity, >88%. Overall concordance of the HTS assays with tumorigenicity data was
low, around 50% for all tumorigens, but increased to 74-78% (vs. 60% for Ames) for those compounds pro-
ducing tumors in rodents at multiple sites and, thus, more likely genotoxic carcinogens. The aim of the
present study was to evaluate the utility of HTS assays to identify potential genotoxicity hazard in the
larger context of the ToxCast project, to aid prioritization of environmentally relevant chemicals for fur-
ther testing and assessment of carcinogenicity risk to humans.
© 2009 Elsevier Inc. All rights reserved.
1. Introduction
The ToxCast™ project of the National Center for Computational
Toxicology (NCCT) at the US. Environmental Protection Agency
(EPA) aims to develop a cost-effective approach for rapid prioritiza-
tion of the assessment of toxicity of large numbers of environmen-
tally relevant chemicals (EPA ToxCast, 2009). Using data from
state-of-the-art high-throughput screening (HTS) and high-content
imaging bioassays, ToxCast is building computational models to
forecast the potential toxicity of these compounds to humans.
These hazard predictions will provide EPA regulatory programs
with science-based information helpful in prioritizing compounds
for more detailed toxicological evaluations and, ultimately, lead to
more efficient use of animal testing. Investigating modern HTS as-
says for genotoxicity is a key element of the ToxCast project and an
important component of toxicity risk assessment. The hypothesis
is that useful and predictive data can be obtained by integrating
the results from large batteries of individual HTS tests. Such
* Corresponding author. Fax: +44 161 606 7337.
E-mail address: andrew.knight@gentronix.co.uk (A.W. Knight).
0273-2300/S - see front matter © 2009 Elsevier Inc. All rights reserved.
doi: 10.1016/j.yrtph.2009.07.004
integrated data will supplement traditional assays for evaluating
chemical genotoxicity in support of carcinogenicity assessments,
and provide useful insights into mechanisms of toxicity, which in
turn will impact on risk assessment (Dix et al., 2007; Houck and
Kavlock, 2008).
To test this hypothesis, an initial set of 320 compounds was
chosen based on a significant amount of accessible historical toxi-
cological data and was designated as the ToxCast Phase I Chemical
Dataset. Most of the ToxCast Phase I compounds are pesticide ac-
tive compounds with hazard assessment toxicological data from
the US EPA Office of Pesticide Programs (EPA OPP, 2009a).
The regulatory requirement for mutagenicity testing to support
a pesticide registration is found in the EPA's Code of Federal Regu-
lations (CFR) Title 40 Part 158 (EPA CFR, 2009). In 1991, the Agency
revised these guidelines to identify specific mutagenicity testing to
be performed; these appear in the OPP Pesticide Assessment
Guidelines, Subdivision F, Hazard Evaluation: Human and Domes-
tic Animals. The Subdivision F guideline was revisited in 1996 by
the Office of Prevention, Pesticides and Toxic Substances (OPPTS)
and is presented as the OPPTS Harmonized Test Guidelines, Series
870 Health Effects Volume I-III (EPA OPPTS, 2009.) The existing
Please cite this article in press as: Knight, A.|
Regul. Toxicol. Pharmacol. (2009), doi:10.1C
Previous
i profiling the US EPA ToxCast™ chemicals.
-------
A.W. Knight et al./Regulatory Toxicology and Pharmacology xxx (2009) xxx-xxx
EPA test battery is a three-tiered system that includes tests for
gene mutation in bacteria (Ames test); mammalian cells (e.g.,
mouse lymphoma, Chinese hamster ovary (CHO) cells, CHO strain
AS52 and/or Chinese hamster V79 lung fibroblasts), and; chromo-
somal aberrations in vitro (mammalian cells) or in vivo (mouse
micronucleus assay). If positive results are obtained in one or more
of the first tier assays, subsequent tests are performed to confirm
these results; pesticides may also be assessed for potential germi-
nal cell effects (Dearfield et al., 1991; Waters et al., 1993). This ap-
proach is similar to the strategy developed by the United Kingdom
and is in general agreement with strategies devised by other Euro-
pean Economic Community nations (Cimino, 2006). Within the
statutory limitations of the Federal Insecticide, Fungicide, and
Rodenticide Act (FIFRA), such mutagenicity testing is required for
all general use pesticides, including terrestrial, aquatic, green-
house, forestry, domestic outdoor and indoor for both food crop
and non-food uses (EPA FIFRA, 2009).
Environmental regulators and researchers have looked to the
pharmaceutical industry for modern techniques to assess genotox-
icity (Muller et al., 1999; US Federal Register, 2008). However, the
established genetic toxicity tests currently recognized by regula-
tory authorities, such as the Ames test, chromosome aberration
test, in vitro micronucleus test or in vitro mouse lymphoma thymi-
dine kinase mutation assay, are not suited to higher throughput
compound screening as they require too much compound, have
too low a throughput, and require too much time to conduct (Cus-
ter and Sweder, 2008). In vitro assays can be overly sensitive when
used for broad chemical screening due to a lack of accounting for
lower in vivo bioavailability and lack of deactivation mechanisms
present in in vitro systems. In some cases, prokaryotes give nega-
tive results for compounds known to specifically interact with
eukaryotic targets such as chromatin DNA and the different pro-
teins involved in DNA metabolism or chromosome segregation
(Walmsley, 2005), leading to low assay sensitivity and falsely neg-
ative predictions of in vivo hazard. Nevertheless, the Ames test is
often used as a predictor of rodent carcinogenesis (Ames et al.,
1973a,b; Matthews et al., 2006). Conversely, it has recently been
recognized that in vitro mammalian genotoxicity assays, such as
the micronucleus test, demonstrate low specificity, i.e., the ability
to correctly identify non-rodent-carcinogens, leading to falsely po-
sitive predictions of in vivo hazard (Kirkland et al., 2005, 2007).
This may in part be explained by the highly toxic and physiologi-
cally unattainable doses applied in these assays in order to detect
bonafide carcinogens (Kirkland, 1992). A prerequisite for the Tox-
Cast project is that such assays should be in an HTS format so that
the assessment of large numbers of compounds can be conducted
efficiently and cost effectively, while also attempting to maintain
accuracy in their prediction of in vivo hazard.
The pharmaceutical industry has employed scaled-down ver-
sions of regulatory assays and other screening approaches. High-
throughput, microplate-based bacterial mutagenicity assays such
as Ames II (Fliickiger-Isler et al., 2004), SOS umuC Chromotest
(Reifferscheid et al., 1991; Reifferscheid and Heil, 1996) and Vito-
tox (Muto et al., 2003), have been shown to be effective screening
alternatives to the standard Ames test and have demonstrated high
concordance with this test using selected compounds. The devel-
opment of higher throughput, miniaturized versions of in vitro
mammalian tests presents a greater challenge due to the fragility
of the cell lines, automating labor-intensive multi-step processes
and the requirement for manual, and often more subjective, scor-
ing. This said, a protocol for a higher throughput micronucleus test
using mouse lymphoma cells and automated scoring by flow
cytometry has been developed (Bryce et al., 2008). However, the
number of validation compounds and the diversity of their chem-
istry are currently limited, and scoring by flow cytometry will ulti-
mately place a limit on the testing throughput attainable. There is
also a microplate format thymidine kinase mutation assay,
although the assay still requires many weeks for dose setting and
the selection of revertants for counting (Chen and Moore, 2004).
The purpose of this paper is to evaluate recently developed HTS
methodologies from Centronix Ltd., Cellumen Inc., and Invitrogen
Corp., the latter performed by the National Institutes of Health
Chemical Genomics Center (NCGC), in the context of the particular
chemical space and larger aims of the ToxCast project. These assays
each reflect different aspects of the cellular response to a genotoxic
challenge, which can lead to DNA damage, mis-repair and muta-
tions and, ultimately, to tumorigenesis and carcinogenesis.
The Centronix 'GreenScreen HC assay uses a human lymphoblas-
toid TK6 cell line (Watanabe et al., 1995), which has been genetically
modified by incorporating a green fluorescent protein (GFP) reporter
based on the regulation of the human GADD45a (growth arrest and
DNA damage) gene (Hastwell et al., 2006). GADD45a mediates the
cell's responses to genotoxic stress and the GFP fluorescence repor-
ter includes p53 regulatory elements that ensure specific and dose-
dependent response from the gene reporter. Validation studies have
shown that the assay responds positively to all classes of genotoxic
damage, but unlike other genetic toxicity assays, appears particu-
larly suited to achieving both high sensitivity and specificity for dis-
criminating genotoxic rodent carcinogens from non-carcinogens
and carcinogens acting by epigenetic mechanisms (Hastwell et al.,
2006). The Cellumen 'CellCiphr' cytotoxicity profiling panel consists
of fluorescent probes for 10 cell features of cytotoxicity of which one
is a DNA damage response in human HepG2 cells, measured by p53
activation via a fluorescent anti-p53 antibody (Vernetti et al., 2009).
The Invitrogen 'CellSensor' assay uses a beta-lactamase reporter
gene under the control of p53 response elements stably integrated
into HCT-116 cells (Brattain et al., 1981). This test system employs
proprietary 'GeneBLAzer' technology based on fluorescence reso-
nance energy transfer (FRET) (Zlokarnik et al., 1998). p53 is known
to act in a 'gate keeper' role during the cell cycle, ensuring genetic
and cellular integrity (Lane, 1992). Through sequence specific and
also non-specific binding, p53 can act in a variety of ways in the cell
in response to genotoxic stress. When DNA damage is sensed it can
activate DNA repair proteins, it can hold the cell cycle at the Gl/S
regulation checkpoint until repair is effective, or it can initiate apop-
tosis if the DNA damage cannot be repaired. Thus p53 is central to
many of the cell's DNA damage response pathways and anti-cancer
mechanisms (Liu and Kulesz-Martin, 2001). In this way the p53 as-
says serve as a broader screen and will also respond to the effects of
cytotoxicity-induced cellular regeneration and neoplasia that can
lead to tumor formation.
Fundamentally carcinogenesis, and to a lesser extent mutagene-
sis, are complex multistage and multi-pathway processes. There are
more than 100 distinct types of cancer, and corresponding subtypes
of organ specific tumors. In addition, cells need to pass through sev-
eral separate physiological changes to develop the novel capabili-
ties acquired during tumor development, and breach the cell's
inherent defense mechanisms (Hanahan and Weinberg, 2000).
Genotoxicity, i.e., direct or indirect DNA damage which may result
in DNA mutation or changes in chromosome structure or number, is
only one dimension in the process and indeed some carcinogens act
through non-genotoxic modes of action (MOAs). For example, in a
wide survey of marketed Pharmaceuticals, Brambilla and Martelli
(2009) reported that out of 315 drugs with both genotoxicity and
carcinogenicity data, 24% were carcinogenic in at least one sex of
mice or rats but test negative in genotoxicity assays, while the same
percentage were positive in both carcinogenicity and genotoxicity
assays. It is not sufficient to say that a compound which produces
positive results in in vitro mutation assays will be a genotoxic car-
cinogen in vivo (EPA, 2007). There are many forms of DNA damage
that genotoxic compounds can exert including; single or double
strand breaks, alteration of bases, formation of covalent adducts,
Please cite this article in press as: Knight, A.W.I
Regul. Toxicol. Pharmacol. (2009), doiilO.lOiel
Previous
1-ofiling the US EPA ToxCast™ chemicals.
-------
A.W. Knight et al./Regulatory Toxicology and Pharmacology xxx (2009) xxx-xxx
oxidative damage, and a myriad of alterations that disrupt the
process of DNA replication or repair. In turn, the cell has a multitude
of sensors, transducers and effectors that orchestrate the response
of the cell to the genotoxic assault (Harper and Elledge, 2007;
Norbury and Hickson, 2001). Thus no single HTS assay, whether
based on specific mutation events or particular cellular responses
to DNA damage, will do a comprehensive job of accurately predict-
ing animal carcinogenicity. Rather the hope is that each HTS assay
will be a surrogate for one of the dimensions in the process and
that, by combining the results of a battery of diverse assays in a
statistical manner, they will provide a satisfactory screen for poten-
tial carcinogenic compounds (Waters et al., 1988).
The results from screening the 320 ToxCast compounds in the
three HTS assays have been compared to one standard genotoxicity
assay in the form of historical mutagenicity (Ames) data, available
for a subset of the test compounds, and animal tumorigenicity stud-
ies conducted in support of regulatory requirements. The main
objective was to assess the performance of the HTS data alongside
results obtained from a conventional genotoxicity assay for which
the HTS assays provide supportive complementary information,
considering both for informing prediction of rodent carcinogenicity
endpoints. In vivo tumorigenicity data for this set of compounds
have been extracted from the recently published chronic toxicity re-
sults from the US EPA ToxRefDB database, which includes potency
results for multiple rodent species and various sites of tumor devel-
opment (Martin et al., 2009; EPA ToxRefDB, 2009). Finally, the assays
have been assessed for their practical utility for deployment in HTS
campaigns applied to screening hundreds or thousands of diverse
chemicals.
2. Materials and methods
2.1. Selection of the test compound set
The compounds chosen for Phase I of the ToxCast program were
selected to include a wide diversity of chemical classes and modes
of action. Of the many groups of environmental compounds that re-
quire prioritization by regulatory authorities, pesticide compounds
are amongst the most characterized and prevalent in toxicity dat-
abases, thus providing a comprehensive set of toxicity endpoints
against which to judge the performance of the HTS assays. From
the approximately 1000 conventional pesticide actives registered
by EPA, the selection was narrowed down based on: (i) the degree
of overlap with other databases such as those of the National Tox-
icology Program (NTP, 2009) and EPA DSSTox (EPA DSSTox, 2009)
and (ii) suitability for HTS assays, i.e., giving lower priority to inor-
ganics, organometallics, highly lipophilic (high AlogP as a measure
of octanoI/water partitioning) and smaller volatile compounds with
molecular weights <150 (Dix et al., 2007).
The final compound set is comprised of 309 unique structures;
291 pesticides actives, 8 metabolites and 10 industrial chemicals
with relevance to other toxicological programs. Of the 291 pesticide
actives, 273 are registered and 18 are unregistered. Three com-
pounds (bensulide, diclofop methyl and prosulfuron) were included
in triplicate, randomly distributed in the compound set to allow for
the assessment of assay reproducibility. Five compounds (3-iodo-2-
propynylbutylcarbamate, chlorsulfuron, dibutyl phthalate, EPTC
and fenoxaprop-ethyl) were included in duplicate but sourced from
different suppliers in order to check for potential differences in test
results with source and small differences in purity. Compound pro-
curement and preparation was handled by Compound Focus Inc., a
subsidiary of BioFocus DPI (South San Francisco, CA, USA). Regard-
ing chemical sources, including replicates: 193 of chemicals were
purchased from Sigma-Aldrich (St. Louis, MO, USA) principally
Pestanal standards from Riedel-de Haen; 101 from Crescent
Chemicals Co. (Islandia, NY, USA); 5 from ChemService Inc. (West
Chester, PA, USA); 4 from Wako Chemicals USA Inc. (Richmond,
VA, USA); 2 from Alfa Aesar (Ward Hill, MA, USA); 1 from Acros
(Geel, Belgium); 1 from TCI America (Portland, OR, USA); 2 from
Battelle (in turn obtained from Sigma-Aldrich and Matrix Scientific
(Columbia, SC, USA)); and 11 chemicals were procured from the
EPA National Pesticide Standard Repository (which came from var-
ious chemical companies). All compounds were coded and blind-
tested by the assay operators and only identified at the end of the
study for the purposes of comparative data analysis. The com-
pounds were supplied to the assay operators as frozen aliquots dis-
solved in dimethylsulfoxide (DMSO) at a concentration of 20 mM.
2.2. GreenScreen HC assay (Centronix Ltd.)
The GreenScreen HC assay uses two genetically modified TK6
cell lines: the GADD45a-GFP reporter strain (GenM-TOl) and a con-
trol strain (GenM-COl) containing an out-of-frame EGFP gene, such
that a functional and fluorescent GFP protein is not produced. The
control strain was used to allow effective correction for any test
compound's autofluorescence, or non-specific induced cellular
fluorescence, that may otherwise give a false indication of GFP
induction in the reporter strain. A suspension of 2 x 106 cells per
ml in a proprietary assay medium are added to a dilution series
of the test compound, and separately to a standard genotoxicant
(methylmethane sulfonate), in 96-well microplates. At 24 and
48 h time-points during incubation of the microplate at 37 °C, 5%
C02, the measurement of the induction in cellular fluorescence
was indicative of genotoxicity, while the measurement of the
reduction in optical absorbance, proportional to the inhibition of
cell proliferation, was used to quantify general cytotoxicity, both
with reference to statistically defined thresholds (Hastwell et al.,
2006). Measurements were made in an Ultra 384 (Tecan, Theale,
UK) microplate reader. Recently the microplate layout has been
modified for HTS applications, testing 12 compounds per plate.
Each compound was tested over 3 serial dilutions (200, 100 and
50 (J.M), with the top test concentration limited by the 1% v/v tol-
erance of DMSO in the GreenScreen HC assay. Using this strategy,
only very potent genotoxins active at lower concentrations would
cause significant cytotoxicity, triggering re-testing from a lower
concentration. A full description of the protocol employed in this
exercise has been published elsewhere (Knight et al., 2009).
Cytotoxicity was assessed by the percentage reduction in cell
proliferation (relative cell density) compared to that achieved in
the vehicle-treated controls. It is important to note that this is
not a measure of cell viability or death. If the cell density relative
to a vehicle-treated control fell below 80% at 1 test concentration
the compound was deemed cytotoxic and if extended over 2 or 3
concentrations, strongly cytotoxic. Otherwise the compound was
considered negative for cytotoxicity. Fluorescence induction in
the test strain was corrected for both cell density and autofluores-
cence with reference to the control strain. If induction of GFP fluo-
rescence relative to a vehicle-treated control exceeded 50% at 1
test concentration the compound was deemed genotoxic and if ex-
tended over 2 or 3 concentrations, strongly genotoxic. Otherwise
the compound was considered negative for genotoxicity.
2.3. CellCiphr Cytotox Profiling Panel—p53 endpoint. (Cellumen Inc.)
Assessment of the DNA damage response in human-derived
HepG2 cells was determined by measurement of p53 activation
via a fluorescent anti-p53 antibody, one of 10 phenotypic end-
points in the Cellumen CellCiphr Cytotox Profiling Panel (Vernetti
et al., 2009). The 20 mM stock of test compound was diluted with
100% DMSO in 9 twofold dilution steps to create 10 concentrations
from 200 to 0.39 u,M. HepG2 cells in log-phase growth were seeded
Please cite this article in press as: Knight, A.|
Regul. Toxicol. Pharmacol. (2009), doi:10.1C
Previous
i profiling the US EPA ToxCast™ chemicals.
-------
A.W. Knight et al./Regulatory Toxicology and Pharmacology xxx (2009) xxx-xxx
in collagen-coated 384 well microplates at 3 cell concentrations
corresponding to different exposure times, [4.3 x 103 (acute),
2.4 x 103 (early), 1.2 x 103 (chronic) cells/well], allowed to settle
for 30 min at room temperature and incubated at 37 °C, 5% C02
for 16 h prior to treatment. Following exposure, cells were fixed
in 4% buffered formalin, the nucleus stained with Hoechst 33342,
and for phospho-p53 using Alexa Fluor 488 antibody. Plates were
read on an Arrayscan HCS Reader (Thermo-Fisher) using channel
1 to quantitate valid objects (cells) defined by Hoescht staining
and p53 activity in channel 2 by quantifying the amount of Alexa
Fluor 488 antibody fluorescence in the area defined by the nuclear
dye. Data were normalized to vehicle-treated control cells. Re-
sponses are measured at 3 different time-points (30 min (acute),
24 h (early) and 72 h (chronic)) for each of 10 concentrations of
the test compounds (serial dilutions from 200 u,M) tested in dupli-
cate along with positive and negative controls, to collect the data to
determine the half-maximal activity (AC50). AC50 values were
determined by fitting the data to the Hill equation using the Con-
doseo module of Genedata Screener (Genedata AG, Basal, Switzer-
land). A positive result was concluded if the p53 AC50 was
calculated to be below 200 u,M, provided the AC50 was lower than
the IC50 for cell loss/cytotoxicity at for that time point.
2.4. Invitrogen CellSensor p53RE-bla HCT-116 assay (NCGC)
HCT-116 cells with a stably integrated beta-lactamase reporter
gene under control of p53 response elements were suspended in
OPTI medium including 0.5% dialyzed FBS, and plated onto 1536
well assay plates at a density of 4000 cells/well (5 u.1 of assay med-
ium with cells suspended at 8 x 105 cell/ml) using a Flying Reagent
Dispenser (Aurora Discovery, Carlsbad, CA). After the plates were
incubated at 37 °C, 5% C02 for 6 h, 23 nL of compounds dissolved
in DMSO, positive controls or DMSO alone were transferred to
the assay plate by a pin tool (Kalypsys, San Diego, CA) resulting
in a 217-fold dilution. The final compound concentration in the
5 u.1 assay volume ranged from 1.2 nM to 92 u,M in 15 concentra-
tions. Nultilin-3 was used as a positive control. The plates were
then incubated for a further 16 h at 37 °C, 5% C02. Subsequently,
1 ul of GeneBLAzer™ B/G FRET substrate (Invitrogen, CA) was
added, the plates were incubated at room temperature for 2 h,
and fluorescence intensity at 460 and 530 nm emission was mea-
sured with excitation at 405 nm using an EnVision microplate
reader (Perkin Elmer, Shelton, CT). Data were expressed as the ratio
of emissions at 460 nm/530 nm. For primary data analysis, read-
ings for each titration point were first normalized relative to the
Nultilin-3 control (12 u,M, 100%) and wells containing the vehicle
only (basal, 0%), and then corrected by applying a pattern correc-
tion algorithm using control plates containing the DMSO diluent
alone. Concentration-response titration points for each compound
were fitted to the Hill equation and concentrations of half-maximal
activity (AC50) and maximal response (efficacy) values were calcu-
lated. (Invitrogen, 2009). A positive result was concluded if the p53
AC50 was calculated to be below 92 uM.
3. Reference data
Two comparative data sets were compiled: collated data from
the Salmonella typhimurium reverse mutagenesis assay (Ames test)
extracted from multiple public sources; and chronic toxicity re-
sults from the EPA ToxRefDB.
3.1. Ames test data
In order to provide a link and a fair comparison between the
new in vitro HTS genotoxicity data generated in this project and
the chronic toxicity data from rodent bioassays, results from a reg-
ulatory in vitro assay, the bacterial Salmonella mutagenesis (Ames)
test, were compiled from the Leadscope toxicity database (2008)
that includes data from a variety of sources: the US National Insti-
tute of Environmental Health Sciences' National Toxicology Pro-
gram (NTP, 2009); the US National Library of Medicine's
Chemical Carcinogenesis Research Information System (CCRIS,
2009); the US National Institute for Occupational Safety and
Health's Registry of Toxic Effects of Chemical Substances (RTECS,
2009); the Carcinogenic Potency Database (CPDB, 2009); and the
Tokyo Metropolitan Institute of Public Health Mutagenicity of Food
Additives Database (TMIPH, 2009). Where conflicting data were
observed between different studies, a straightforward algorithm
was applied to arrive at a positive, intermediate or negative score.
The procedure was based on the number of reported positive and
negative studies, as well as rules based on content, i.e., the mini-
mum assay information a data source should provide before it is
graded and the completeness and quality of the data record (Yang
et al., 2008). Using this approach Ames data were publicly available
for 108 of the 309 unique test compound structures. The Ames data
and the details of this analysis are provided in the supplementary
information Appendix A and on the ToxCast home page (EPA
ToxCast, 2009).
3.2. Chronic toxicity data
Rodent tumorigenesis data from chronic, two-year bioassays
were collated from the ToxRefDB database, as of October 2008
(Martin et al., 2009; EPA ToxRefDB, 2009). These data allowed dis-
crimination of tumorigenicity based on occurrence in rats or mice,
males or females, and single or multiple sites within each animal
category. For the purposes of statistical comparison, positive and
negative scores were given in three categories: (i) rodent tumori-
gen (rat and/or mouse); (ii) multiple species tumorigens, and;
(iii) multiple site tumorigen. In these three categories data were
available for 273, 212 and 273 test compounds, respectively. In
addition, positive results were ranked by potency. Hence further
comparisons were made to a subset of the most potent tumorigens
by selecting those compounds with a potency <15 mg/kg/day.
4. Results
Full results for all 320 test chemicals are provided in the supple-
mentary information Appendix B and on the ToxCast home page
(EPA ToxCast, 2009). This comprises a comparative genetic toxicity
database comparing results for the 309 unique compounds, provid-
ing chemical name, CAS number, DSSTox chemical IDs, color in
solution, qualitative (positive/negative) and quantitative (lowest
effective concentration or AC50) genotoxicity results, a positive
or negative call for Ames mutagenicity data and ToxRefDB tumor-
igenicity data with a delineation between rodent, multiple species,
multiple site and most potent rodent tumorigens.
4.1. Physicochemical properties
For assays based on optical endpoints, there is a potential for
interference to occur in the measurement of optical absorbance
and fluorescence from compounds that are autofluorescent, col-
ored or that precipitate from aqueous solution upon incubation
with cells and assay medium. A preliminary assessment of the de-
gree of interference for the ToxCast compound set was determined
using the GreenScreen HC assay where measurements of conven-
tional optical absorbance and fluorescence are made. In this assay
each compound was initially tested from 200 uM, and 58 com-
pounds (18.1%) were found to form significant precipitates after
Please cite this article in press as: Knight, A.W.I
Regul. Toxicol. Pharmacol. (2009), doi:10.1016l
Previous
1-ofiling the US EPA ToxCast™ chemicals.
-------
A.W. Knight et al./Regulatory Toxicology and Pharmacology xxx (2009) xxx-xxx
incubation with test cells and assay medium. Since higher optical
absorbance readings can artificially increase the estimation of cell
density, and reduce sensitivity for assessment of cytotoxicity, in
this project, these compounds were re-tested from a 2- or 4-fold
lower starting concentration to confirm the accuracy of the initial
test result. After optimizing the test concentration, only 9 com-
pounds (2.9%) produced optical absorbance values greater than
20% that of the cell density achieved by the vehicle-treated con-
trols. Fourteen compounds (4.5%) were visibly colored in solution;
however, the color was not sufficiently high as to limit the analysis.
Ninety-three compounds (30.1%) were greater than 20% more
fluorescent than the background autofluorescence of the vehicle-
treated control strain and assay medium. However, the use of the
non-fluorescent control strain to correct the measured fluores-
cence induction coupled with a simple corrective method, based
on fluorescent polarization measurements for the most autofluo-
rescent compounds to enhance the discrimination of GFP,
adequately compensated for the test compound's autofluorescence
in all cases (Knight et al., 2002). Thus for testing compounds from a
concentration of 200 u,M, these results suggest that the physico-
chemical properties of the test compounds would present minimal
challenge for an assay with optical endpoints.
4.2. Reproducibility
Table 1 shows the genotoxicity and cytotoxicity results for the
GreenScreen HC and CellCiphr p53 assays for the 8 compounds
with replicate samples randomly distributed in the 320 ToxCast
compound set. The data show that both assays were highly
consistent within each set of replicates, giving identical genotoxi-
city results and closely matching cytotoxicity results. Upon inspec-
tion of the raw data for GreenScreen HC, dibutyl phthalate and
prosulfuron showed very similar dose-dependent trends in cyto-
toxicity responses between replicates. Only one test of prosulfuron
produced a reduction in relative cell density just over the signifi-
cance threshold at 48 h and only at the highest test concentration
(200 u,M). There appears to be no difference in the reproducibility
results based on chemical supplier. The CellSensor p53 assay gave
negative results for all replicates and thus data for this assay is not
included in Table 1.
4.3. Effective concentration ranges
The supply of pre-solubilized library compounds in DMSO solu-
tion automatically restricts the highest concentrations that can be
tested because of the inherent cytotoxicity associated with DMSO.
Compounds were tested up to 200 u,M in the GreenScreen HC and
CellCiphr p53 assays and to 92 uM in the CellSensor p53 assay.
Comparing the concentrations at which a significant positive geno-
toxicity result was recorded allows an assessment of the sensitivity
of each assay in terms of chemical concentration, bearing in mind
the caveats of the restricted concentration range and defined dilu-
tion strategy. The average AC50 for a positive result in each assay
(LEC in the case of GreenScreen HC) were as follows: GreenScreen
HC (65 uM) > CellCiphr p53 (58 uM) > CellSensor p53 (31 uM).
These findings indicate all assays show similar chemical sensitivity
for a genotoxicity response.
4.4. Screening hit rates for genotoxicity and cytotoxicity
The following data analysis is based on the 309 unique
compounds in the set omitting replicates. Where data for the rep-
licate compounds were not consistent for the same assay, the
strongest positive response was taken as the 'definitive' result for
Table 1
Genotoxicity and cytotoxicity results for the 8 compounds which were represented in duplicate or triplicate in the ToxCast chemical set. Groups of 2 replicates are from different
chemical suppliers. Groups of 3 replicates are true replicates from the same chemical stock Key: +, positive; + +, strong positive (GreenScreen HC assay only); —, negative; LEC,
Lowest effective concentration; AC50, Active concentration for 50% effect; CFLID, project's chemical ID code.
Chemical name
CAS
number
Replicate CFIJD
#
Source Purity GreenScreen HC CellCiphr p53
GreenScreen HC CellCiphr p53
Genotoxicity LEC Genotoxicity AC50
(um) Min
(um)
Cytotoxicity LEC Cytotoxicity AC50
(um) Min
(um)
3-Iodo-2-
propynylbutylcarbamate
Bensulide
Chlorsulfuron
Dibutyl phthalate
Diclofop-methyl
EPTC
Fenoxaprop-ethyl
Prosulfuron
Agreement in all replicates
55406-53-
6
741-58-2
64902-72-
3
84-74-2
51338-27-
3
759-94-4
66441-23-
4
94125-34-
5
1
2
1
2
3
1
2
1
2
1
2
3
1
2
1
2
1
2
3
TV000274 Crescent
TV000023 Sigma
TV000097 Sigma
TV000334 Sigma
TV000318 Sigma
TV000060 Crescent
TV000264 Sigma
TV000026 Sigma
TV000290 Alfa_Aesar
TV000369 Sigma
TV000333 Sigma
TV000103 Sigma
TV000178 Sigma
TV000259 Crescent
TV000266 Crescent
TV000374 Sigma
TV000138 Sigma
TV000311 Sigma
TV000303 Sigma
97 + +
97.1 + +
99.5 -
99.5 -
99.5 -
99
99.9 -
99.63 -
98.6 -
99.2 -
99.2 -
99.2 -
98.7 -
97
98
98.4 -
98.4 -
98.4 -
98.4 -
8/8
12.5 + 8.5 ++ 12.5 +
12.5 + 81.0 + + 12.5 +
+ + 25 +
+ + 25 +
+ + 25 +
_ - -
-
+ + 50 +
+ 200 +
+ + 50 -
+ + 50 +
+ + 50 +
-
-
+ + 25 +
+ + 50 +
-
+ 200 -
-
8/8 7/8 7/8
8.6
3.2
30.5
34.4
40.4
199
186
191
198
120
169
Please cite this article in press as: Knight
Regul. Toxicol. Pharmacol. (2009), doi:10
!&
Previous I
^^n
B'^^VPQIf^^B PmlilinS the US EPA ToxCast™
chemicals.
-------
AW. Knight et al./Regulatory Toxicology and Pharmacology xxx (2009) xxx-xxx
Table 2
Summary of the genotoxicity screening hit rates for positive results in the
genotoxicity HTS assays.
Assay
GreenScreen HC
CellCiphr p53
CellSensor p53
Genotoxicity
Number
32
27
36
%
10.4
8.7
11.7
Cytotoxicity
Number
231
171
%
74.8
55.3
comparative analysis, representing evidence of a risk for a positive
result in that particular assay. Table 2 summarizes the screening
hit rates for the 3 HTS assays for the compound collection. The
number of positive results for genotoxicity was similar between
the 3 assays, averaging just over 10%. The hit rate of 10.4% for
GreenScreen HC was only marginally higher than the 7.3% demon-
strated previously in a larger collection of 1266 compounds in a li-
brary of pharmacologically active compounds, LOPAC (Knight et al.,
2009). This may in part reflect testing from 200 u,M in this study
rather than 100 u,M in the larger collection. Nine percent of the
GreenScreen HC positive results were only positive at 200 u,M,
while the most common lowest effective concentration (LEC) was
100 (J.M (34% of positive results), comfortably within the test con-
centration range.
While the GreenScreen HC, CellCiphr p53 and CellSensor p53
assays produced similar numbers of positive results (32, 27 and
36, respectively) the overlap between data sets was relatively
small. The number of positive results which were common to
CellCiphr p53 and GreenScreen HC was 6. The number of positive
results which were common to GreenScreen HC and CellSensor
p53 was 9. Eleven positive compounds were detected in both the
CellCiphr p53 and CellSensor p53 sets. Seventy one compounds
(23.0%) were positive in at least one of the three HTS assays. The
overlap in positive results between assays is shown in Fig. 1.
The number of positive results for cytotoxicity was higher than
for genotoxicity in the CellCiphr p53 and the GreenScreen HC as-
says. The highest frequency of positive cytotoxicity results was
demonstrated in the GreenScreen HC assay (74.8%). This is signifi-
cantly higher than the 33% rate seen previously in the LOPAC col-
lection, for example (Knight et al., 2009). This is likely due to the
difference in the nature of the chemical space, i.e., pesticides are
designed to reduce the viability of microorganisms, plants and ani-
mals, rather than providing therapeutic benefit. It should be noted
that this discrepancy is not as a result of higher dose used in this
study compared to an evaluation of the LOPAC compounds, as
the most frequent LEC reported for the positive cytotoxicity results
(53.1% of compounds) was 50 u,M. The CellCiphr p53 assay re-
ported a similarly high number of positive cytotoxicity results
(55.3%).
GreenScreen HC
CellCiphr p53
X
CellSensor p53
Fig. 1. Overlap of positive genotoxicity results predicted by the HTS assays.
4.5. Prediction of mutagenicity and tumorigenicity data
The principle aim of conducting in vitro genotoxicity assays is to
predict the potential for genotoxicity or carcinogenicity in vivo. The
predictive ability and accuracy of the HTS assays were determined
by comparison to published Ames test data and the chronic toxic-
ity results (tumorigenicity) from the US EPA ToxRefDB (Martin
et al., 2009). Commonly used terms for comparative analysis and
their definitions are taken from Cooper and co-workers and are
paraphrased here (Cooper et al., 1979). Also included is the term
'relative predictivity', introduced by Kirkland, which is the ratio
of real to false results (Kirkland et al., 2005).
A compound tested in any particular assay (e.g., Test A) can
have a positive outcome for which there may be either a positive
result (a), or a negative result (b), from a second, comparative
in vivo test for carcinogenicity. Thus the total number of positives
for Test A is (a + b). Similarly, Test A can have a negative outcome
for which there might be either a positive (c), or a negative (d), re-
sult from the comparative test. Thus, the total number of negatives
for Test A is (c + d). It follows that the total number of positive re-
sults from the second comparative test is (a + c) and the total num-
ber of negative results from the second comparative test is (b + d).
Similarly, the total number of compounds for which there are data
for both tests, represented by N, is (a + b + c + d). The following
terms were calculated from these basic figures:
• Sensitivity, the percentage of correctly identified positives = [a/
(a + c)] x 100.
• Specificity, the percentage of correctly identified negatives = [d/
(b + d)] x 100.
• Concordance, the percentage of correctly identified results, both
positive and negative = [(a + d)/N] x 100.
• Balanced accuracy = mean of the sensitivity and specificity.
• Prevalence, percentage of positive calls in the second, compara-
tive test results = [(a + c)/N] x 100.
• Positive predictivity = (a/(a + b)).
• Negative predictivity = (d/(c + d)).
• Relative predictivity of a positive result for carcinogenicity is the
fraction of carcinogens giving a positive genotoxicity result,
divided by the fraction of non-carcinogens giving a positive
genotoxicity result = [(a/(a + c))/(b/(b + d))].
• Relative predictivity of a negative result for non-carcinogenicity
is the fraction of non-carcinogens giving a negative genotoxicity
result, divided by the fraction of carcinogens giving a negative
genotoxicity result = [(d/(b + d))/(c/(a + c))].
The higher the relative predictivity value obtained, the more
predictive the testing strategy will be and values above 2 are con-
sidered significant (Kirkland et al., 2005). The concordance gives a
measure of overall accuracy since it combines the percentage of
correctly identified results both positive and negative. Although
the balanced accuracy is a simple mean of the sensitivity and spec-
ificity results, it allows a better measure of performance if it is the
case that an assay has mainly positive or mainly negative results,
since sensitivity may be very high with correspondingly and dis-
proportionately low specificity, or vice versa.
Table 3A-E details the comparison of the HTS assay results to
the corresponding Ames mutagenicity test results (available for
108 compounds) and rodent tumorigenicity results (available from
ToxRefDB for 273 of the compounds in the ToxCast set).
Table 3A compares the results of the HTS assays to the pub-
lished Ames test results. While the terms 'sensitivity' and 'specific-
ity' are generally considered only for comparison with carcinogenic
endpoints, in this table the authors have used the same terms for
comparison to Ames results in order to provide the reader with a
measure that is more readily comparable to subsequent analyses
Please cite this article in press as: Knight, A.W.I
Regul. Toxicol. Pharmacol. (2009), doi:10.1016l
Previous
1-ofiling the US EPA ToxCast™ chemicals.
-------
A.W. Knight et al./Regulatory Toxicology and Pharmacology xxx (2009) xxx-xxx
Table 3
Comparison of the performance of the HTS assays in the prediction of the Salmonella mutagenicity (Ames) test and rodent tumorigenicity.
GreenScreen GreenScreen
A. Prediction of Ames data
Ames + 6
Ames - 4
Number of comparisons 108
Sensitivity (% correct positives) 13.0
Specificity (% correct negatives) 93.5
Concordance 59.3
Balanced accuracy 53.3
Positive predictive value 0.60
Negative predictive value 0.59
Relative predictivity (positives) 2.02
Relative predictivity (negatives) 1.08
GreenScreen
B. Prediction of rodent tumorigenicity
Rodent + 22
Rodent - 7
Number of comparisons 273
Sensitivity (% correct positives) 14.9
Specificity (% correct negatives) 94.4
Concordance 51.3
Balanced accuracy 54.6
Positive predictive value 0.76
Negative predictive value 0.48
Relative predictivity (positives) 2.65
Relative predictivity (negatives) 1.11
GreenScreen
+
CellSensor p53
40 5
58 6
108
95% conf. lim. p-Value
(0, 27)
(81, 100)
(0.5, 0.7)
(0.5, 0.7)
(1.6, 2.6)
(0.8, 1.5)
GreenScreen
126
118
95% conf.
lim.
(7, 23)
(85, 100)
(0.7, 0.8)
(0.4, 0.6)
(2.5, 2.9)
(0.9, 1.4)
GreenScreen
-
<0.001 10.9
<0.001 90.3
56.5
50.6
0.004 0.45
0.673 0.58
0.002 1.12
0.795 1.01
CellSensor CellSensor
p53 + p53 -
p-Value
<0.001
<0.001
<0.001
0.409
<0.001
0.459
21
12
273
14.2
90.4
49.1
52.3
0.64
0.47
1.48
1.05
CellSensor
p53 +
127
113
95% conf.
lim.
(6, 22)
(82, 99)
(0.6, 0.7)
(0.4, 0.5)
(1.3, 1.7)
(0.8, 1.4)
CellSensor
p53-
CellSensor p53
41
56
95% conf. lim.
(0, 25)
(78, 100)
(0.3, 0.6)
(0.5, 0.7)
(0.7, 1.6)
(0.7, 1.4)
CellCiphr
p53 +
p-Value
<0.001
<0.001
0.002
0.702
0.001
0.762
17
9
273
11.5
92.8
48.7
52.1
0.65
0.47
1.60
1.05
CellCiphr
p53 +
p-Value
<0.001
<0.001
0.603
0.959
0.622
0.988
CellCiphr
p53-
131
116
95% conf.
lim.
(3, 19)
(85, 100)
(0.6, 0.7)
(0.4, 0.5)
(1.4, 1.8)
(0.8, 1.3)
CellCiphr
p53-
CellCiphr p53 CellCiphr p53
4
10
108
8.7
83.9
51.9
46.3
0.29
0.55
0.54
0.92
p-Value
<0.001
<0.001
<0.001
0.742
<0.001
0.757
42
52
95% conf. lim.
(0, 23)
(72, 96)
(0.2, 0.4)
(0.5, 0.6)
(0.1, 1.1)
(0.6, 1.3)
Ames Ames
23
13
90
44.2
65.8
53.3
55.0
0.64
0.46
1.29
1.18
Ames
+
29
25
95% conf.
lim.
(31,58)
(50, 82)
(0.5, 0.7)
(0.3, 0.6)
(1.0, 1.7)
(0.7, 1.8)
Ames
-
p-Value
<0.001
<0.001
0.013
0.583
0.034
0.578
p-Value
0.032
0.001
0.164
0.497
0.136
0.532
C. Prediction of most potent tumorigens <15 mg/kg/day
Rodent + 7
Rodent - 22
Number of comparisons 273
Sensitivity (% correct positives) 15.2
Specificity (% correct negatives) 90.3
Concordance 77.7
Balanced accuracy 52.8
Positive predictive value 0.24
Negative predictive value 0.84
Relative predictivity (positives) 1.57
Relative predictivity (negatives) 1.07
GreenScreen
+
39
205
95% conf.
lim.
(5,27)
(85, 95)
(0.1, 0.3)
(0.82, 0.86)
(1.0, 2.4)
(0.9, 1.2)
GreenScreen -
p-Value
0.687
0.003
0.145
0.387
0.140
0.427
7
26
273
15.2
88.5
76.2
51.9
0.21
0.84
1.33
1.04
CellSensor
p53 +
39
201
95% conf.
lim.
(5, 27)
(83, 94)
(0.1, 0.3)
(0.82, 0.86)
(0.7, 2.2)
(0.9, 1.2)
CellSensor
p53-
p-Value
0.705
0.025
0.390
0.562
0.392
0.577
4
22
273
8.7
90.3
76.6
49.5
0.15
0.83
0.90
0.99
CellCiphr
p53 +
42
205
95% conf.
lim.
(0, 20)
(85, 95)
(0.1, 0.3)
(0.81, 0.85)
(0.3, 1.8)
(0.9, 1.2)
CellCiphr
p53-
p-Value
0.123
0.003
0.755
0.880
0.727
0.853
14
22
90
56.0
66.2
63.3
61.1
0.39
0.80
1.65
1.50
Ames +
11
43
95% conf.
lim.
(40, 72)
(56, 77)
(0.2, 0.6)
(0.74, 0.9)
(1.0, 2.6)
(1.2, 1.9)
Ames -
p-Value
0.001
0.271
0.141
0.009
0.109
0.008
D. Prediction of multiple species rodent tumorigens
Rodent + 5
Rodent - 17
Number of comparisons 212
Sensitivity (% correct positives) 11.6
Specificity (% correct negatives) 89.9
Concordance 74.1
Balanced accuracy 50.8
Positive predictive value 0.23
Negative predictive value 0.80
Relative predictivity (positives) 1.16
Relative predictivity (negatives) 1.02
38
152
95% conf.
lim.
(0, 24)
(84, 96)
(0.1, 0.3)
(0.77, 0.83)
(0.6, 2.0)
(0.9, 1.2)
p-Value
0.130
0.001
0.635
0.800
0.700
0.861
5
18
212
11.6
89.3
73.6
50.5
0.22
0.80
1.09
1.01
38
151
95% conf.
lim.
(1, 24)
(83, 95)
(0.1, 0.3)
(0.77, 0.83)
(0.5, 1.9)
(0.9, 1.2)
p-Value
0.116
<0.001
0.776
0.873
0.837
0.928
3
13
212
7.0
92.3
75.0
49.6
0.19
0.80
0.91
0.99
40
156
95% conf.
lim.
(0, 19)
(86, 98)
(0.1, 0.3)
(0.77, 0.82)
(0.3, 1.7)
(0.8, 1.2)
p-Value
0.026
<0.001
0.782
0.940
0.726
0.906
9
16
62
95% conf.
lim.
50.0
63.6
59.7
56.8
0.36
0.76
1.38
1.27
9
28
p-Value
(33, 71)
(50, 77)
(0.2, 0.5)
(0.7, 0.8)
(0.6, 2.6)
(1.0, 1.8)
0.034
0.218
0.440
0.192
0.471
0.163
(continued on next page)
Please cite this article in press as: Knight,
Regul. Toxicol. Pharmacol. (2009), doi:10
ffi Previous I TOC 1 Nex
I^^^Hi profiling the
US EPA ToxCast™ chemicals.
-------
A.W. Knight et al./Regulatory Toxicology and Pharmacology xxx (2009) xxx-xxx
Table 3 (continued)
GreenScreen GreenScreen
+ —
E. Prediction of multiple site rodent twnorigens
Rodent + 13 45
Rodent - 16 199
Number of comparisons 273
95% conf.
Sensitivity (% correct positives)
Specificity (% correct negatives)
Concordance
Balanced accuracy
Positive predictive value
Negative predictive value
Relative predictivity (positives)
Relative predictivity (negatives)
22.4
92.6
77.7
57.5
0.45
0.82
3.01
1.19
lim.
(12, 34)
(87, 98)
(0.4, 0.6)
(0.79, 0.84)
(2.5, 3.7)
(1.1, 1.4)
p-Value
0.750
<0.001
<0.001
0.033
<0.001
0.019
CellSensor CellSensor
p53 + p53 -
10
23
273
17.2
89.3
74.0
53.3
0.30
0.80
1.61
1.08
48
192
95% conf.
lim.
(6, 29)
(84, 95)
(0.2, 0.4)
(0.77, 0.83)
(1.1,2.3)
(0.9, 1.3)
p-Value
0.409
<0.001
0.059
0.327
0.050
0.332
CellCiphr CellCiphr
p53 + p53 -
8
18
273
13.8
91.6
75.1
52.7
0.31
0.80
1.65
1.06
50
197>
95% conf.
lim.
(4, 25)
(86, 97)
(0.2, 0.4)
(0.77, 0.82)
(1.2, 2.3)
(0.9, 1.2)
p-Value
0.147
<0.001
0.047
0.416
0.026
0.453
Ames
+
14
22
90
50.0
64.5
60.0
57.3
0.39
0.74
1.41
1.29
Ames
14
40
95% conf.
lim.
(33, 65)
(51, 76)
(0.2, 0.5)
(0.67, 0.81)
(0.8, 2.3)
(1.0, 1.7)
p-Value
0.019
0.425
0.305
0.127
0.302
0.086
in Table 3B-E. Due to the limit on the number of available Ames
data, this assessment has the least number of chemical compari-
sons (N = 108). The prevalence of positive results in the Ames assay
(42.6%) was much higher than that observed for the HTS assays
(see Table 2) and this bias must be borne in mind when consider-
ing these results. The predominant reason for the higher positive
prevalence in the Ames data is likely due to testing at considerably
higher concentrations, as defined in ICH guidelines (US Federal
Register, 2008). Ames testing is often carried out with concentra-
tions up to 5 mg/plate or 10 mM if solubility allows; this is approx-
imately 50-fold greater than the highest concentration used in the
HTS assays. Secondly, none of the HTS assays incorporated meta-
bolic activation which is routinely included in Ames testing via
rat liver homogenates (S9 fraction). All three HTS assays showed
similar correlation with the Ames data. Prediction of positive Ames
test results (sensitivity) was low for all assays, between 8.7% and
13.0%. Prediction of negative Ames test results (specificity) was
high for all HTS assays examined and highest for GreenScreen HC
(93.5%). We caution, however, that given the high percentage of
negative HTS results the assay data are biased towards negative
predictions and hence we expect better performance in predicting
negative results in the regulatory assays. The overall concordance
was similar for all assays, averaging 55.9%. Using the relative pre-
dictivity for positives as a comparison metric for Ames (Table 3A),
GreenScreen HC was significantly better than the CellSensor p53
and CellCiphr p53 assays (p-values 0.002, 0.622 and 0.034, respec-
tively). Confidence intervals and p-values were calculated using a
parametric bootstrapping technique that compares predictions
generated by any given model (GreenScreen HC, CellCiphr p53
and Cell Sensor p53) with predictions obtained from a null hypoth-
esis model. The null model draws random samples from a binomial
distribution in which the probability of an active result equals the
proportion of actives observed in the actual experimental results.
Statistics were obtained by comparing the results for the model
of interest with two thousand samplings of the null model. Ap-va-
lue of 0.002, for example, therefore indicates that the probability of
observing a value greater than the reported statistic by random
chance is only 0.2%. Low p-values (<0.05) are obtained for models
that perform significantly better than the null model for the asso-
ciated statistic. High p-values denote results for which the model
prediction is no better than random chance for that particular sta-
tistical measure.
Table 3B is derived from comparing assay results to rodent
tumorigenicity from rodent chronic bioassays in ToxRefDB, con-
sidering any significant tumor formation, irrespective of species
or site. The number of chemical comparisons was 273, reflecting
a larger amount of available data and comprehensive overlap of
the ToxRefDB and ToxCast databases. Sensitivity of the HTS assays
was again low in this set, averaging 13.5% potentially due to the
aforementioned lack of metabolic activation in the HTS assays.
Consequently the Ames test results demonstrated a higher sensi-
tivity of 44.2%. However, sensitivity will also be low in this com-
parison as in vitro genotoxicity assays are not expected to readily
detect non-genotoxic carcinogens. Overall concordance was very
similar for all in vitro assays including the regulatory Ames data,
averaging 49.7%. It seems equally apparent that, whereas the
Ames test had a much higher sensitivity than the 3 HTS assays
(44.2% vs. 11.5-14.9%), the HTS assays had much better specific-
ity, again due in part to the imbalanced nature of the HTS data-
sets, i.e., a high prevalence of negative results. Using the relative
predictivity for positives as a metric to compare the rodent
tumorigenicity, GreenScreen demonstrated the highest value
(2.65) although all the HTS assays showed significant predictivity
with low p-values ^0.001 (cf. p-value of 0.136 for Ames). These
latter data are mirrored in the positive predictive values for the
HTS and Ames assays.
Table 3C compares assay results with only the most potent
tumorigens, based on oral exposure values, where positive results
were given only for compounds with a potency ^15 mg/kg/day.
Compared to the prediction of all tumorigens (Table 3B) a signifi-
cant improvement in concordance figures for all assays was noted
when potency was factored into activity, averaging 76.8%. Once
again the sensitivity of the HTS assays was lower than that of the
Ames test, whereas specificity was higher. Relative predictivity
for a positive result was, on the whole, lower than the previous
comparison reflecting the lower number of potent tumorigens—
only 46 of the 148 rodent tumorigens had a potency ^15 mg/kg/
day. Broadening the analysis to potencies ^50 mg/kg/day in-
creased the number of potent tumorigens to 76, however, the per-
formance figures from the different assays changed only
marginally for all parameters examined (data not shown).
Table 3D compares assay results to multiple species rodent car-
cinogens, i.e., compounds that have been shown to produce tumors
in both rats and mice. In this comparison the accuracy of prediction
(concordance) was improved for each of the cellular HTS assays,
compared with the analysis of all tumorigens, averaging 74.2%
and was significantly higher than Ames (59.7%). The number of
available comparisons was lower (212 compounds) due to the
requirement for available data in both rats and mice. In compari-
sons using the small set of most the most potential tumorigens
(Table 3C) and multiple species rodent tumorigens (Table 3D) the
positive predictive value and relative predictivity of all the assays
considered was low and associate with high p-values, demonstrat-
ing lower confidence in the results.
Table 3E compares assay results to multiple site rodent carcin-
ogens, i.e., compounds which have been shown to produce tumors
Please cite this article in press as: Knight, A.W.I
Regul. Toxicol. Pharmacol. (2009), doiilO.lOiel
Previous
1-ofiling the US EPA ToxCast™ chemicals.
-------
A.W. Knight et al./Regulatory Toxicology and Pharmacology xxx (2009) xxx-xxx
in multiple sites in rats and/or mice, such as liver, kidney, thyroid,
testes, spleen or lung. In this comparison the data set is more
complete with 273 comparisons possible. In this case, for all the
HTS assays, concordance was further improved, from an average
of 50% for all rodent tumorigens to 74-78% for multiple site
tumorigens. Relative predictivity for a positive result was signifi-
cantly high for GreenScreen HC (3.01, p-value <0.001) compared
to the p53 assays and this assay also demonstrated a correspond-
ingly high specificity of 92.6%. Again these data are reflected in
the positive and negative predicted values. Most noteworthy,
however, was the increase in sensitivity of the GreenScreen HC
assay (nearly doubling to 22.4%) over Table 3D where compari-
sons were made on the basis of multiple species carcinogens. In
this case, a compound that produces tumors at multiple sites irre-
spective of species or gender can be viewed as a particularly clear
carcinogenic response, most likely due to a genotoxic mechanism
of action, a factor that could lead to the improved sensitivity in
this assay.
4.6. Nature of the compounds and mechanisms of action
The ToxCast compound set was highly diverse in terms of both
MOAs and chemical structure (i.e., a large number of chemical clas-
ses). The majority of classes of pesticide had fewer than 3 chemical
representatives in the set (<1%) and therefore although this enables
a more accurate assessment of assay performance in a wide chem-
ical space, the response of individual assays to particular chemical
classes and features is more difficult to model and will require
more detailed investigation. This is a long term objective of future
studies, incorporating data from structure-activity relationship
(SAR), other regulatory genetic toxicology and MOA studies as well
as other relevant endpoints.
5. Discussion
The HTS assays studied represent two gene targets in their end-
points, p53 and GADD45a in a p53 competent cell line. The tumor
suppressor p53 protein appears to sense multiple types of DNA
damage and regulates many parts of the cell's DNA damage re-
sponse, activating DNA repair, cell cycle arrest and induction of
apoptosis. p53 activity is maintained at low levels in healthy cells
but is rapidly induced by ionizing radiation and several genotoxic
and chemotherapeutic compounds. Hence p53 is an obvious candi-
date for a marker of genotoxic effects. However, p53 is a multifunc-
tional protein with potential for diverse modifications and
biochemical properties. Its functionality is only partially under-
stood in terms of how p53 coordinates the DNA damage response
in specific cell types, the nature of the interactions between p53
and DNA repair proteins and the mechanism of p53-dependent
apoptosis (Liu and Kulesz-Martin, 2001; Meek, 2004). Induction
of p53 can also occur in response to a range of non-genotoxic stres-
ses leading to growth arrest or apoptosis. Ohno et al. (2008) re-
cently conducted an 80 compound validation study of a
genotoxicity test system based on p53R2 gene expression with a
luciferase reporter. Although reporting that this system could be
used for the rapid screening of genotoxic potential, the ability of
the assay to detect genotoxic effects was unclear. A subsequent fo-
cused study of 27 compounds with diverse genotoxic mechanisms
revealed that the test compound's potency in the assay was related
to MOA and that well-known genotoxic antimetabolites (e.g., pur-
ine synthesis and metabolism inhibitors such as 6-mercaptopur-
ine) and HDAC (histone deacetylase) inhibitors (e.g., trichostatin
A) were not detected.
The transcriptional regulation of GADD45a is also complex and
includes p53-dependent induction and regulation by BRCA1,
c-MYC and NF-kB. GADD45a induction has been observed in re-
sponse to a wide range of genotoxic and growth arrest stresses
and upon exposure of cells to UV and a wide range of genotoxic
compounds with diverse MOAs. The protein is thought to play a
role in DNA repair, cell cycle checkpoints, apoptosis in response
to genotoxic stress and antitumorigenesis as well as in the mainte-
nance of genomic stability (Hastwell et al., 2006).
In the context of the complexity of biological pathways and pro-
cesses of mutagenesis and carcinogenesis and with the differing
nature of the cell lines and endpoints used in the three HTS assays,
it is not surprising that there is only limited overlap between the
positive results of these HTS assays (see Fig. 1) and Ames positive
results. This gives some weight to the argument for the use of a
battery of related screens probing different aspects of the cellular
response from which data can be pooled. Furthermore, the non-
overlapping features of the different HTS assays considered in the
present study also may reflect variable sensitivities of each assay to
different chemical classes and, hence, provide support for the
broader ToxCast approach in which HTS assays are to be chosen,
combined and applied mindful of the chemical space being pre-
dicted. Clearly, however, these are longer term objectives beyond
the scope of the present study.
Two compounds, both fungicides, were positive in all three HTS
assays; benomyl and fluoxastrobin. In many studies benomyl has
produced negative results for gene mutations, structural chromo-
some aberrations and DNA damage and does not react directly
with DNA. Benomyl inhibits fungal growth by binding to tubulin
and disrupting microtubule assembly. A similar genotoxic MOA
in mammalian cells results in aneuploidy in vitro and in vivo
(Bentley et al., 2000). Based on a MOA analysis, there is plausible
evidence that benomyl acts through an aneuploidy mechanism
(McCarroll et al., 2002). Benomyl has also previously been reported
as a positive in the GreenScreen HC assay with an LEC concentra-
tion of 12 (J.M (Hastwell et al., 2006) compared to an LEC of
50 u,M for this study. Fluoxastrobin is a strobilurin fungicide that
acts through inhibition of mitochondrial respiration, which can re-
sult in the production of free electrons that react with oxygen to
form superoxide, known to cause oxidative stress and DNA damage
(Bartlett et al., 2002). Of 5 other strobilurin fungicides in the data
set, all 5 were positive in the CellSensor p53 assay and 4 of these
were positive in the GreenScreen HC assay, yet none were positive
in the CellCiphr p53 assay. A strobilurin present in the ToxCast
compound collection, azoxystrobin, has been shown to produce a
weak clastogenic response in mammalian cells in vitro at cytotoxic
doses. However, in animals azoxystrobin was negative in assays for
chromosomal damage and general DNA damage at high dose lev-
els. Hence, strobilurins are considered to have low toxicity and
consequently low risk to non-target organisms (EPA, 1997). This
class of compounds highlights the complications of extrapolating
in vitro test results to human health risk assessments.
It is unusual in a comparative study such as this to have a
wealth of tumorigenicity data to reference for the majority of com-
pounds. In pharmaceutical R&D, a positive Ames result will often
remove a compound from the development process and follow-
up in vivo studies will only be performed for candidates that are
developed for life threatening-conditions or have other highly
favorable therapeutic indications, although exceptions include
compounds from oncology and anti-viral campaigns. Overall the
HTS assays have demonstrated low sensitivity for compounds that
have been shown to be active tumorigens, and similarly low posi-
tive predictivity for Ames results. Screening hit rates for positive
results were consistent between assays averaging 10%. This is
likely explained by the limitation of concentrations achievable in
screening, the lack of exogenous metabolic activation, and the
probability that many tumorigens may be acting via non-muta-
genic MOAs. In this small set of three assays, screening only picked
Please cite this article in press as: Knight, A.|
Regul. Toxicol. Pharmacol. (2009), doi:10.1C
Previous
i profiling the US EPA ToxCast™ chemicals.
-------
10
A.W. Knight et al./Regulatory Toxicology and Pharmacology xxx (2009) xxx-xxx
up the most potent genotoxic tumorigens, and thus low sensitivity
was a trade-off for the advantage of screening of large numbers of
compounds with low sample concentration requirements. Con-
versely, the HTS assays demonstrated high specificity, consistently
over 88%, and hence produce a low number of what may be inter-
preted as false positive results when comparing to rodent tumori-
genicity data. In the context of the predicted dose, high specificity
in screening for hazard assessment is desired, otherwise concerns
over product safety are falsely raised, triggering more detailed,
lengthy and expensive mechanistic investigations, often in vivo,
and delaying or halting development or deployment of the com-
pound. However, low assay sensitivity is also of concern and less
desirable for hazard identification and screening in the context of
environmental risk assessment, since public safety is the primary
objective. Hence, limitations in the sensitivity of individual assays
to predict the in vivo endpoint may need to be compensated by
additional tests or assays with higher sensitivity and different per-
formance characteristics. The measure of concordance is biased by
the low number of positive results and hence low sensitivity and
high specificity of the HTS assays and is therefore a less useful met-
ric in this study. Overall concordance with tumorigenicity data was
low, around 50% for all tumorigens, but increased to 74-78% for the
HTS assays (vs. 60% for Ames) for compounds producing tumors in
rodents at multiple sites.
Through numerous validation studies it has become clear that
carcinogens partition into two main groups; genotoxic and non-
genotoxic MOAs. Many mechanisms are a part of the latter group,
including inhibition of apoptosis, down-regulation of gap junction
intercellular communication, enhancement of cellular prolifera-
tion, and peroxisome proliferation (Yamasaki et al., 1996; Combes,
2000; Vinken et al., 2008). The concordance with tumorigenicity
data was highest when only the most potent tumorigens and those
compounds that produced tumors at multiple sites and in multiple
species were considered, and were therefore more likely to have a
genotoxic mode of action (EPA, 2007). This is in agreement with
Ashby and Tennant's observations that the use of genotoxicity
screening assays will enable the detection of trans-species and
multiple site rodent carcinogens, while the detection of tissue,
sex or species-specific carcinogens will only be achieved by con-
ducting lifetime carcinogenicity bioassays (Ashby and Tennant,
1988). Compounds that produce tumors in different rodent species
or at more than one site are more likely to be acting by a genotoxic
mechanism, i.e., directly reacting with DNA or interfering with its
production, maintenance or translocation, and thus are more likely
to be detected by in vitro genotoxicity tests (EPA, 2007). This dis-
tinction and approach has been taken since no other attempt has
been made to classify the carcinogens in the ToxCast compound
set as either genotoxic or non-genotoxic carcinogens. Indeed this
is not a trivial exercise and even where data exist there is often
an incomplete picture and differences in 'expert opinion' as to
the exact mechanism of carcinogenicity. Furthermore for some
compounds, both genotoxic and non-genotoxic mechanisms may
play a role (Yamasaki et al., 1996).
Many environmental compounds are tested at maximum toler-
ated doses in vivo in rodent chronic bioassays, and at up to 10 mM
in regulatory genotoxicity tests, often at doses or concentrations
that could not be attained in actual human exposure (Kirkland
et al., 2007). Whereas many compounds may produce quantifiable
genotoxic effects in this high-dose testing, the concentration at
which this is achieved is an important consideration in risk assess-
ment. In the HTS screening assays utilized in the present study,
testing was carried out at lower concentrations than those at
which some regulatory tests are performed; hence it might be ex-
pected to detect only a subset of the more potent genotoxic spe-
cies, i.e., show low comparative sensitivity. ToxRefDB includes
potency figures in the form of a lowest effect level in mg/kg body
weight/day over the two years of exposure at which the chronic tu-
mor endpoint was observed.
The authors stress that none of the HTS assays being considered
here have been developed with the aim of replacing or accurately
predicting the results of the Ames test. Miniaturized tests such as
Ames II (Fliickiger-Isler et al., 2004) and other bacterial assays do
this adequately. Rather, HTS tests for genotoxicity were developed
as early screening tools with the aim of increasing testing effi-
ciency and concordance with carcinogenicity endpoints as com-
pared to the low-throughput, standard regulatory tests (Custer
and Sweder, 2008). Furthermore, these HTS genotoxicity assays
probe cellular pathways that likely play a role in mutagenic and
carcinogenic processes for many chemicals.
The lack of capacity for metabolic activation is a significant lim-
itation of the cell lines currently used in the HTS assays examined
here. It has long been known that cytochrome P450 enzymes facil-
itate principally phase I metabolism of xenobiotics and are highly
conserved in virtually all eukaryotic and many prokaryotic cells.
Also, it is well-known that the metabolic activation of progenotoxic
or procarcinogenic compounds often leads to the formation of
reactive electrophilic intermediates (Nebert and Dalton, 2006).
P450 enzymes are predominantly localized in the liver. Hence liver
homogenates (S9 fraction) from rats dosed with P450 inducing
compounds are commonly added to the assay medium of in vitro
assays to provide a level of metabolic competency and enable
detection of progenotoxic compounds. This lack of metabolic com-
petency is being addressed in recent modifications of the Green-
Screen HC assay that incorporate co-exposure of the test
compound and S9 fraction (Jagger and Tate et al., 2009). Likewise,
the CellCiphr and CellSensor assays are also being modified to be
run in primary cell systems with more complete metabolic
capacity.
The 2-year rodent bioassay is generally considered the gold
standard for assessing potential risk of carcinogenicity to humans.
However, such assays are expensive, time-consuming and low-
throughput. In the context of assessing risk it should be noted that,
while rodent bioassays are currently the best approach we have to
assess the potential for human carcinogenicity, they may overesti-
mate the risk of mutagenesis and carcinogenesis. The European
Medicines Agency quotes in their guidance for carcinogenicity test-
ing for Pharmaceuticals that "Since the early 1970s, many investi-
gations have shown that it is possible to provoke a carcinogenic
response in rodents by a diversity of experimental procedures,
some of which are now considered to have little or no relevance
for human risk assessment" (EMA, 2008). Rodent bioassays with
environmental chemicals are conducted up to high doses
approaching what is maximally tolerated by the animal, and the
subsequent chronic cytotoxicity can lead to increased cell division
to replace damaged cells, and increased DNA replication with a
greater potential to produce mutations. This likely contributes to
the high positive rate in rodent tumorogenicity studies, as was
the case in ToxRefDB with just under half the compounds (148)
identified as rodent tumorigens. The positive predictivity of the
HTS data was lowest when attempting to correlate with multiple
species tumorigens. Although both species considered here are ro-
dents, there are frequently significant differences in response be-
tween rats and mice even of the same sex. For example, it has
been reported that carcinogenicity in mice predicts carcinogenicity
in rats with an accuracy of only 70% (Lave et al., 1988) and there
are obvious cautionary implications in extrapolating the results
further for human risk assessment (Knight et al., 2006). However,
hazard information from rodent bioassays alone should not be con-
strued as complete indicators of human risk; EPA considers many
factors in the process of interpreting such information relative to
human exposure and risk. The vast majority of these pesticidal
compounds have been evaluated by EPA as having limited carcin-
Please cite this article in press as: Knight, A.W.I
Regul. Toxicol. Pharmacol. (2009), doiilO.lOiel
Previous
1-ofiling the US EPA ToxCast™ chemicals.
-------
A.W. Knight et al./Regulatory Toxicology and Pharmacology xxx (2009) xxx-xxx
11
ogenic potential in humans (EPA OPP, 2009b), and to be safe when
used according to EPA registration(s). Data from the Ames assay,
for example, are only a part of the regulatory genetic toxicology
battery results that contributes to the weight of the evidence ap-
proach taken by EPA in consideration of carcinogenesis.
Several physical characteristics are required for an effective and
useful HTS assay. These are high sample throughput, high repro-
ducibility, low compound requirement, and low cost of consum-
ables and equipment. The CellSensor p53 assay demonstrated the
shortest protocol taking 24 h to complete; GreenScreen HC takes
48 h, while CellCiphr p53 takes 24-72 h depending on the expo-
sure duration. The CellSensor p53 assay can achieve the highest
throughput of 30,000 to 100,000 compounds per week in a fully
automated laboratory. GreenScreen HC and CellCiphr demonstrate
a lower but still significantly high-throughput of up 500-700 com-
pounds per week, yet can be readily set up manually using auto-
dispensing handheld pipettes. All three HTS assays used proprie-
tary assay reagents and cell lines available from Centronix Ltd.,
Millipore or Invitrogen. Reproducibility was specifically tested by
the random distribution of replicates of test compounds within
the ToxCast compound set. The GreenScreen HC and CellCiphr
p53 assays demonstrated a very high degree of reproducibility
both in qualitative and quantitative results and for both genotoxi-
city and cytotoxicity (see Section 4.2). Genotoxicity results for Cell-
Sensor were negative in all cases, so a fair analysis of
reproducibility could not be made for this assay.
The next stage of this work will entail combining the HTS assay
results presented here with a much larger set of HTS assay results
from the ToxCast project, along with data from the full EPA regula-
tory test battery genetic toxicology, MOA analysis and structure-
activity relationship (SAR) approaches to data analysis, to derive
predictive signatures for in vivo endpoints. We fully expect differ-
ent types of HTS assays, conjoined with chemical class characteris-
tics, will contribute to prediction of each in vivo toxicity endpoint.
These initial models built from the ToxCast Phase 1 chemicals and
data will be validated on larger sets of compounds, and ultimately
the most predictive models could be employed for screening and
prioritization of large lists of environmental chemicals, in align-
ment with the needs of risk assessment. Thus, the HTS genotoxicity
indicator assays presented here are likely to find greatest value
when deployed in combination with other types of pathway indi-
cators or less related endpoints, such as cytotoxic or SAR measures.
In conclusion, whereas comparisons to Ames and tumorigenic-
ity data have been useful in this particular project in order to as-
sess the relevance and quality of the results produced, positive
data from these high-throughput genotoxicity assays alone are
not suitable surrogates for standard regulatory genotoxicity as-
says; neither can these HTS results be used in isolation as direct
predictors of animal tumorigenicity. However, data from these
HTS assays can reasonably be applied in four areas: (i) screening
for compounds capable of causing genetic damage in vitro and thus
used for 'hazard identification'; i.e., the rapid and accurate identi-
fication of compounds with the potential to induce carcinogenic-
ity; (ii) as one part of a weight of evidence assessment to inform
on the likelihood of a compound's adverse effect for humans; (iii)
to aid the determination the mode of action for carcinogenicity;
and (iv) to aid prioritization of a compound for further follow-up
testing in in vitro assays and rodent bioassays (EPA, 2007).
Conflict of interest statement
Andrew Knight, Louise Birrell and Richard Walmsley are
employees of Centronix Ltd., who developed and manufacture
the GreenScreen HC assay. All other authors have no conflicts of
interest to declare.
Disclaimer
The United States Environmental Protection Agency through its
Office of Research and Development partially funded and collabo-
rated in the research described here. It has been subjected to
Agency review and approved for submission and peer review. Ref-
erence to any specific commercial products, process, or service by
trade name, trademark, manufacturer, or otherwise, does not nec-
essarily constitute or imply its endorsement, recommendation, or
favoring by the United States Government.
Acknowledgements
Chihae Yang wishes to thank Leadscope and US FDA Center for
Food Safety and Applied Nutrition for the use of their databases to
enable the statistical analysis.
Appendices A and B Supplementary data
Supplementary data associated with this article can be found, in
the online version, at doi:10.1016/j.yrtph.2009.07.004.
References
Ames, B.N., Lee, F.D., Durston, W.E., 1973a. An improved bacterial test system for the
detection and classification of mutagens and carcinogens. Proc. Nat. Acad. Sci.
USA 70, 782-786.
Ames, B.N., Durston, W.E., Yamasaki, E., Lee, F.D., 1973b. Carcinogens are mutagens:
a simple test system combining liver homogenates for activation and bacteria
for detection. Proc. Nat. Acad. Sci. USA 70, 2281-2285.
Ashby, J., Tennant, R.W., 1988. Chemical structure, Salmonella mutagenicity and
extent of carcinogenicity as indicators of genotoxic carcinogenesis among 222
chemicals tested in rodents by the US NCI/NTP. Mutat. Res. 204,17-115.
Bartlett, D.W., Clough, J.R., Godwin, J.R., Hall, A.A., Hamer, M., Parr-Dobrzanski, B.,
2002. The strobilurin fungicides. Pest. Man. Sci. 58, 649-662.
Bentley, K.S., Kirkland, D., Murphy, M., Marshall, R., 2000. Evaluation of thresholds
for benomyl- and carbendazim-induced aneuploidy in cultured human
lymphocytes using fluorescence in situ hybridization. Mutat. Res. 464, 41-51.
Brambilla, G., Martelli, A., 2009. Update on genotoxicity and carcinogenicity testing
of 472 marketed Pharmaceuticals. Mutat. Res. 681, 209-229.
Brattain, M.G., Fine, W.D., Khaled, F.M., Thompson, J., Brattain, D.E., 1981.
Heterogeneity of malignant cells from a human colonic carcinoma. Cancer
Res. 41, 1751-1756.
Bryce, S.M., Avlasevich, S.L, Bemis, J.C., Lukamowicz, M., Elhajouji, A., Van
Goethemc, F., De Boeck, M., Beerens, D., Aerts, H., Van Compel, J., Collins, J.E.,
Ellis, P.C., White, A.T., Lynch, A.M., Dertinger, S.D., 2008. Intel-laboratory
evaluation of a flow cytometric, high content in vitro micronucleus assay.
Mutat. Res. 650, 181-195.
CCRIS, 2009. US National Library of Medicine's Toxicology Data Network (TOXNET)
Chemical Carcinogenesis Research Information System. Available from: . (accessed January 2009).
Chen, T., Moore, M.M., 2004. Screening for chemical mutagens using the mouse
lymphoma assay. In: Van, Z., Caldwell, G.W. (Eds.), Methods in Pharmacology
and Toxicology: Optimization in Drug Discovery: In Vitro Methods. Humana
Press, pp. 337-352.
Cimino, M.C., 2006. Comparative overview of current international strategies and
guidelines for genetic toxicology testing for regulatory purposes. Environ. Mol.
Mutagen. 47, 362-390.
Combes, R.D., 2000. The use of structure-activity relationships and markers of cell
toxicity to detect non-genotoxic carcinogens. Toxicol. in Vitro 14, 387-399.
Cooper, J.A., Saracci, R., Cole, P., 1979. Describing the validity of carcinogen
screening tests. Br. J. Cancer 39, 87-89.
CPDB, 2009. Berkeley Carcinogenic Potency Database. Available from: . (accessed January 2009).
Custer, L.L., Sweder, K.S., 2008. The role of genetic toxicology in drug discovery and
optimization. Curr. Drug. Met. 9, 978-985.
Dearfield, K.L, Auletta, A.E., Cimino, M.C., Moore, M.M., 1991. Considerations in the
US Environmental Protection Agency's testing approach for mutagenicity.
Mutat. Res. 258, 259-283.
Dix, D.J., Houck, KA, Martin, M.T., Richard, A.M., Setzer, R.W., Kavlock, R.J., 2007.
The ToxCast program for prioritizing toxicity testing of environmental
chemicals. Toxicol. Sci. 95, 5-12.
EMA, 2008. European Medicines Agency, ICH Topic SIB carcinogenicity: testing for
carcinogenicity of Pharmaceuticals, Step 5, Note for guidance on
carcinogenicity: testing for carcinogenicity of Pharmaceuticals (CPMP/ICH/
299/95). March 2008. Available from: (accessed January 2009).
Please cite this article in press as: Knight, A.|
Regul. Toxicol. Pharmacol. (2009), doi:10.1C
Previous
i profiling the US EPA ToxCast™ chemicals.
-------
12
A.W. Knight et al.f Regulatory Toxicology and Pharmacology xxx (2009) xxx-xxx
EPA, 1997. Pesticide tolerance petition filing for azoxystrobin. Fed. Reg. Doc. 97-
5683. Tuesday, March 11,1997.
EPA, 2007. Framework for determining mutagenic mode of action for
carcinogenicity. Using EPA's 2005 Cancer Guidelines and Supplemental
Guidance for Assessing Susceptibility from Early-Life Exposure to
Carcinogens. EPA 120/R-07/002-A. External Peer Review Draft September
2007. Available from: .
(accessed January 2009).
EPA CFR, 2009. US Environmental Protection Agency's Code of Federal Regulations
Title 40 Part 158. Available from: . (accessed January 2009).
EPA DSSTox, 2009. US Environmental Protection Agency's Distributed Structure-
Searchable Toxicity Data Network. Available from: (accessed January 2009).
EPA FIFRA, 2009. US Environmental Protection Agency's Federal Insecticide,
Fungicide, and Rodenticide Act. Available from: . (accessed January 2009).
EPA OPP, 2009a. US Environmental Protection Agency's Office of Pesticide Programs.
Available from: . (accessed January
2009).
EPA OPP, 2009b. US Environmental Protection Agency's Office of Pesticide
Programs. Available from: . (accessed
January 2009).
EPA OPPTS, 2009. US Environmental Protection Agency's Office of Prevention,
Pesticides and Toxic Substances Harmonized Test Guidelines, Series 870 Health
Effects Volume I-III. Available from: .
(accessed January 2009).
EPA ToxCast, 2009. US Environmental Protection Agency's National Center for
Computational Toxicology ToxCast Project. Available from: . (accessed January 2009).
EPA ToxRefDB, 2009. US Environmental Protection Agency's National Center for
Computational Toxicology ToxRefDB (Toxicology Reference Database).
Available from: . (accessed January 2009).
Fliickiger-Isler, S., Baumeister, M., Braun, K., Gervais, V., Hasler-Nguyen, N.,
Reimann, R., Van Compel, J., Wunderlich, H.-G., Engelhard, G., 2004.
Assessment of the performance of the Ames II™ assay: a collaborative study
with 19 coded compounds. Mutat. Res. 558,181-197.
Hanahan, D., Weinberg, R.A., 2000. The hallmarks of cancer. Cell 100, 57-70.
Harper, J.W., Elledge, S.J., 2007. The DNA damage response: ten years after. Mol. Cell
28, 739-745.
Hastwell, P.W., Chai, L-L, Roberts, K.J., Webster, T.W., Harvey, J.S., Rees, R.W.,
Walmsley, R.M., 2006. High-specificity and high sensitivity genotoxicity
assessment in a human cell line: validation of the GreenScreen GADD45a-GFP
genotoxicity assay. Mutat. Res. 607,160-175.
Houck, K.A., Kavlock, R.J., 2008. Understanding mechanisms of toxicity: Insights
from drug discovery research. Toxicol. App. Pharmacol. 277,163-178.
Invitrogen, 2009. CellSensor™ p53RE-bla HCT-116 Cell-based Assay Protocol.
Available from: . (accessed January 2009).
Jagger, C, Tate, M., Cahill, P.A., Hughes, C, Knight, A.W., Billinton, N., Walmsley,
R.M., 2009. Assessment of the genotoxicity of S9-generated metabolites using
the GreenScreen HC GADD45a-GFP assay. Mutagenesis 24, 35-50.
Kirkland, D., Aardema, M., Henderson, L, Miiller, L, 2005. Evaluation of the ability of
a battery of three in vitro genotoxicity tests to discriminate rodent carcinogens
and non-carcinogens I. Sensitivity specificity and relative predictivity. Mutat.
Res. 584, 1-256.
Kirkland, D., Pfuhler, S., Tweats, D., Aardema, M., Corvi, R., Darroudi, F., Elhajouji, A.,
Glatt, H., Hastwell, P., Hayashi, M., Kasper, P., Kirchner, S., Lynch, A., Marzin, D.,
Maurici, D., Meunier, J.R., Miiller, L, Nohynek, G., Parry, J., Parry, E., Thybaud, V.,
Tice, R., van Benthem, J., Vanparys, P., White, P., 2007. How to reduce false
positive results when undertaking in vitro genotoxicity testing and thus avoid
unnecessary follow-up animal tests: report of an ECVAM workshop. Mutat. Res.
628, 31-55.
Kirkland, D.J., 1992. Chromosomal aberration tests in vitro: problems with protocol
design and interpretation of results. Mutagenesis 7, 95-106.
Knight, A., Bailey, J., Balcombe, J., 2006. Animal carcinogenicity studies: 1. Poor
human predictivity. ATLA 34,19-27.
Knight, A.W., Birrell, L, Walmsley, R.M., 2009. Development and validation of a
higher throughput screening approach to genotoxicity testing using the
GADD45a-GFP GreenScreen HC assay. J. Biomol. Scr. 14, 16-30.
Knight, A.W., Goddard, N.J., Billinton, N., Cahill, P.A., Walmsley, R.M., 2002.
Fluorescence polarisation discriminates green fluorescent protein from
interfering autofluorescence in a microplate assay for genotoxicity. J.
Biochem. Biophys. Meth. 51,165-177.
Lane, D.P., 1992. p53, guardian of the genome. Nature 358,15-16.
Lave, L.B., Ennever, F.K., Rosenkranz, H.S., Omenn, G.S., 1988. Information value of
the rodent bioassay. Nature 336, 631-633.
Liu, Y., Kulesz-Martin, M., 2001. p53 protein at the hub of cellular DNA damage
response pathways through sequence-specific and non-sequence specific DNA
binding. Carcinogenesis 22, 851-860.
Martin, M.T., Judson, R.S., Reif, D.M., Kavlock, R.J., Dix, D.J., 2009. Profiling chemicals
based on chronic toxicity results from the US EPA ToxRef database. Environ.
Health Perspect. 117, 392-399.
Matthews, E.J., Kruhlak, N.L, Cimino, M.C., Benz, R.D., Contrera, J.F., 2006. An
analysis of genetic toxicity, reproductive and developmental toxicity, and
carcinogenicity data: I. Identification of carcinogens using surrogate endpoints.
Regul. Toxicol. Pharm. 44, 83-96.
McCarroll, N.E., Protzel, A., loannou, Y., Frank Stack, H.F., Jackson, M.A., Waters, M.D.,
Dearfield, K.L., 2002. A survey of EPA/OPP and open literature on selected
pesticide chemicals III. Mutagenicity and carcinogenicity of benomyl and
carbendazim. Mutat. Res. 512,1-35.
Meek, D.W., 2004. The p53 response to DNA damage. DNA Repair 3,1049-1056.
Miiller, L, Kikuchi, Y., Probst, G., Schechtman, L, Shimada, H., Sofuni, T., Tweats, D.,
1999. ICH-Harmonised guidances on genotoxicity testing of Pharmaceuticals:
evolution, reasoning and impact. Mutat. Res. 436, 195-225.
Muto, S., Baba, H., Uno, Y., 2003. Evaluation of the Vitotox™ test as a high-
throughput genotoxicity assay. Environ. Mut. Res. Commun. 25, 69-75.
Nebert, D.W., Dalton, T.P., 2006. The role of cytochrome P450 enzymes in
endogenous signaling pathways and environmental carcinogenesis. Nat. Rev.
Cancer 6, 947-960.
Norbury, C.J., Hickson, I.D., 2001. Cellular response to DNA damage. Annu. Rev.
Pharmacol. Toxicol. 41, 367-401.
NTP, 2009. National Institutes of Environmental Health Sciences National
Toxicology Program. Available from: . (accessed
January 2009).
Ohno, K., Ishihata, K., Tanaka-Azuma, Y., Yamada, T., 2008. A genotoxicity test
system based on p53R2 gene expression in human cells: Assessment of its
reactivity to various classes of genotoxic chemicals. Mutat. Res. 656, 27-35.
Reifferscheid, G., Heil, J., 1996. Validation of the SOS/umu test using test results of
486 chemicals and comparison with the Ames test and carcinogenicity data.
Mutat. Res. 369, 129-145.
Reifferscheid, G., Heil, J., Oda, Y., Zahn, R.K., 1991. A microplate version of the SOS/
umu-test for rapid detection of genotoxins and genotoxic potentials of
environmental samples. Mutat. Res. 253, 215-222.
RTECS, 2009. US National Institute for Occupational Safety and Health's Registry of
Toxic Effects of Chemical Substances. Available from: . (accessed January 2009).
TMIPH, 2009. Tokyo Metropolitan Institute of Public Health Mutagenicity of Food
Additives. Available from: . (accessed
January 2009).
US Federal Register, 2008. International Conference on Harmonization: draft
guidance on S2(R1) genotoxicity testing and data interpretation for
Pharmaceuticals intended for human use; availability. Available from:
.
(accessed June 2009).
Vernetti, L, Irwin, W., Giuliano, K.A., Cough, A., Johnston, P., Taylor, D.L, 2009.
Cellular systems biology applied to pre-clinical safety testing: a case study of
CellCiphr cytotoxicity profiling. In: Xu, J.J., Ekins, S. (Eds.), Drug Efficacy, Safety
and Biologies Discovery: Emerging Technologies and Tools. Wiley and Sons.
Vinken, M., Doktorova, T., Ellinger-Ziegelbauer, H., Ahr, H.-J., Lock, E., Carmichael, P.,
Roggen, E., van Delft, J., Kleinjans, J., Castell, J., Bort, R., Donato, T., Ryan, M.,
Corvi, R., Keun, H., Ebbels, T., Athersuch, T., Sansone, S.-A., Rocca-Serra, P.,
Stierum, R., Jennings, P., Pfaller, W., Gmuender, H., Vanhaecke, T., Rogiers, V.,
2008. The carcinoGENOMICS project: Critical selection of model compounds for
the development of omics-based in vitro carcinogenicity screening assays.
Mutat. Res. 659, 202-210.
Walmsley, R.M., 2005. Genotoxicity screening: the slow march to the future. Exp.
Opin. Drug Metabol. Toxicol. 1, 261-268.
Watanabe, T., Kataoka, T., Mizuta, S., Kobayashi, M., Uchida, T., Imai, K., Wada, H.,
Kinoshita, T., Murate, T., Mizutani, S., Saito, H., Hotta, T., 1995. Establishment
and characterization of a novel cell line, TK-6, derived from T cell blast crisis of
chronic myelogenous leukemia, with the secretion of parathyroid hormone-
related protein. Leukemia 9,1926-1934.
Waters, M.D., Stack, H.F., Jackson, M.A., Bridges, B.A., 1993. Hazard identification:
efficiency of short-term tests in identifying germ cell mutagens and putative
nongenotoxic carcinogens. Environ. Health Perspect. 101, 61-72.
Waters, M.D., Stack, H.F., Rabinowitz, J.R., Garrett, N.E., 1988. Genetic activity
profiles and pattern recognition in test battery selection. Mutat. Res. 205,119-
138.
Yamasaki, H., Ashby, J., Bignami, M., Jongen, W., Linnainmaa, K., Newbold, R.F.,
Nguyen-Ba, G., Parodi, S., Rivedal, E., Schiffmann, D., Simons, J.W.I.M., Vasseur,
P., 1996. Nongenotoxic carcinogens: development of detection methods based
on mechanisms: a European project. Mutat. Res. 353, 47-63.
Yang, C, Hasselgren, C.H., Boyer, S., Arvidson, K., Aveston, S., Dierkes, P., Benigni, R.,
Benz, R.D., Contrera, J., Kruhlak, N.L, Matthews, E.J., Han, X., Jaworska, J.,
Kemper, R.A., Rathman, J.F., Richard, A.M., 2008. Understanding genetic toxicity
through data mining: the process of building knowledge by integrating multiple
genetic toxicity databases. Toxicol. Mech. Meth. 18, 277-295.
Zlokarnik, G., Negulescu, P.A., Knapp, T.E., Mere, L, Burres, N., Feng, L, Whitney, M.,
Roemer, K., Tsien, R.Y., 1998. Quantitation of transcription and clonal selection
of single living cells with beta-lactamase as reporter. Science 279, 84-88.
Please cite this article in press as: Knight, A.W.I
Regul. Toxicol. Pharmacol. (2009), doiilO.lOiel
Previous
1-ofiling the US EPA ToxCast™ chemicals.
-------
Submitted to Chemical Research in Toxicology
Molecular Modeling for Screening Environmental Chemicals
for Estrogenicity: Use of the Toxicant-Target Approach
Journal:
i
Manuscript ID:
i
Manuscript Type:
P_________________^^
Date Submitted by the
Author:
j~_____________________
Complete List of Authors:
Chemical Research in Toxicology
tx-2009-00135x.R2
Article
Rabinowitz, James; US EPA, NCCT
Little, Stephen; US EPA, NCCT
Laws, Susan; US EPA, NHEERL
Goldsmith, Michael-Rock; US EPA, NERL
<$>scholaroNi
Manuscript Central
ACS Plus Environment
Previous
TOC
-------
Page 2 of 35 Submitted to Chemical Research in Toxicology
1
2
3
4 Molecular Modeling for Screening Environmental
o
6
7
s Chemicals for Estrogenicity: Use of the Toxicant-Target
10
11
12
13
14
15
16
17
18
19
20
21 National Center for Computational Toxicology, U.S. Environmental Protection Agency, Research
22
23 Triangle Park, NC
24 &
25
26
27
28
29 RECEIVED DATE (to be automatically inserted after your manuscript is accepted if required
30
31 according to the journal that you are submitting your paper to)
32
33
Approach
James R. Rabinowitz*, Stephen B. Little, Susan C. Laws1 and Michael-Rock Goldsmith2
TITLE RUNNING HEAD Molecular Modeling - Screening for Estrogenicity
AUTHOR FOOTNOTE *rabinowitz.james(g),epa.gov, corresponding author
36
37
38
39
40 ( 1 ) S . Laws is a member of the Reproductive Toxicology Division, NHEERL, U. S .EPA
41
42
4 o (2) During this study M-R. Goldsmith was a cross Office of Research and Development postdoctoral fellow working in the NCCT. He is
presently in the National Exposure Research Laboratory, U.S. EPA, Research Triangle Park, NC
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
ACS Paragon Plus Environment
Previous I TOC
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
Submitted to Chemical Research in Toxicology
Table of Contents Graphic
Page 3 of 35
ACS Paragon Plus Environment
2
Previous
TOC
-------
Page 4 of 35 Submitted to Chemical Research in Toxicology
ABSTRACT
2
3
4
5
6 There is a paucity of relevant experimental information available for the evaluation of the
7
8 potential health and environmental effects of many man made chemicals. Knowledge of the potential
9
10
.. pathways for activity provides a rational basis for the extrapolations inherent in the preliminary
12
13 evaluation of risk and the establishment of priorities for obtaining missing data for environmental
14
15 chemicals. The differential step in many mechanisms of toxicity may be generalized as the interaction
16
1 g between a small molecule (a potential toxicant) and one or more macromolecular targets. An approach
19
20 based on computation of the interaction between a potential molecular toxicant and a library of
21
22 macromolecular targets of toxicity has been proposed for preliminary chemical screening. In the current
23
24
25 study, the interaction between a series of environmentally relevant chemicals and models of the rat
26
27 estrogen receptors (ER) was computed and the results compared to an experimental data set of their
28
29 relative binding affinities. The experimental data set consists of 281 chemicals, selected from the U.S.
30
31
22 EPA's Toxic Substances Control Act (TSCA) inventory, that were initially screened using the rat uterine
33
34 cytosolic ER-competitive binding assay. Secondary analysis, using Lineweaver-Burk plots and slope
35
3® replots, was applied to confirm that only fifteen of these test chemicals were true competitive inhibitors
O /
38
39 of ER binding with experimental inhibition constants (Ki) less than 100 |jM. Two different rapid
40
41 computational "docking" methods have been applied. Each provides a score that is a surrogate for the
42
44 strength of the interaction between each ligand-receptor pair. Using the score that indicates the strongest
45
46 interaction for each pair, without consideration of the geometry of binding between the toxicant and the
47
48 target, all of the active molecules were discovered in the first 16% of the chemicals. When a filter is
49
51 applied based on the geometry of a simplified pharmacophore for binding to the ER, the results are
52
53 improved and all of the active molecules were discovered in the first 8% of the chemicals. In order to
54
55 obtain no false negatives in the model that includes the pharmacophore filter only 8 molecules are false
56
58 positives. These results indicate that molecular "docking" algorithms that were designed to find the
59
60 chemicals that act most strongly at a receptor (and therefore are potential pharmaceuticals) can
ACS Paragon Plus Environment
Previous I TOC
-------
Submitted to Chemical Research in Toxicology Page 5 of 35
efficiently separate weakly active chemicals from a library of primarily inactive chemicals. The
2
3 advantage of using a pharmacophore filter suggests that the development of filters of this type for other
4
5 receptors will prove valuable.
6
7
8
9 KEYWORDS Computational molecular docking, Toxicant-Target interactions, Estrogen receptor,
10
11 Endocrine disruption
12 F
13
. c Abbreviations: (KBA) Relative binding affinity, (ER) estrogen receptor, (KIERBL) The label in DSSTox
16
17 (15) for the primary experimental data set used in this study (14), (1C50) The concentration of a test
18
1 ^ chemical that inhibits the maximal specific binding of 0.33 nM radiolabeled (H) 17-/3-estradiol (E2) to
21
22 rat uterine cytosolic ER by 50%, (Ki) Inhibition constant for the test chemical, i.e., micromolar
23
24 concentration that will bind to half the ER at equilibrium in the absence of radioligand or other
25
Oft
competitors, (TPR) True positive ratio, the ratio of the positives discovered divided by the total number
28
29 of positives, (FPR) False positive ratio, the ratio of the number of false positives divided by the number
30
31 of negatives, (HPTE) 2,2-Bis-(p-Hydroxyphenyl)-l,l,l-Trichloroethane).
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
ACS Paragon Plus Environment
Previous I TOC
-------
Page 6 of 35 Submitted to Chemical Research in Toxicology
1
2
3
4 INTRODUCTION
o
6
7
o
The potential risks due to exposure to man-made chemicals in the environment must be
y
10
11 evaluated in order to protect human health and the environment. Little or no experimental information
12
13 is available about the potential biological effects of as many as 75% of these chemicals (1). In addition
14
16 the data that are available are often insufficient to adequately evaluate the potential hazards of each of
17
18 these chemicals. Therefore, there is a compelling need to develop information that would enable the
19
20 screening of the potential health and environmental effects of large numbers of man-made chemicals (2).
21
22
23 While animal studies have historically been preferred for risk assessment, such data sets are often
24
25 difficult and expensive to obtain and yet still require significant extrapolation to be applied to this task.
26
27
28
29 The National Academy of Science report, "Toxicity Testing in the Twenty-First Century"
30
31 presents a vision of a new paradigm for toxicity testing (3). This vision is based on the identification
o^
33
34 and characterization of toxicity pathways. The capacity of specific chemicals to participate in these
35
36 pathways would then be interrogated using rapid (m vitro or perhaps computational) test methods.
37
38 Integration of the data using informatics approaches that consider the causal nature of these pathways
39
40
41 could then be applied for the evaluation of the risks posed by a specific chemical. In support of this
42
43 vision, efforts are underway to devise approaches that combine more rapidly obtained experimental data
44
^ and data computed from chemical structure with informatics approaches to perform toxicity evaluations
46
47
43 (4). In the initial stage, these approaches may be seen as extrapolations from more readily obtainable
49
50 data through the more traditionally applied animal data to the evaluation of potential health and
51
C f}
^r environmental effects. Until more experience is obtained and success is demonstrated, these
Oo
54
55 approaches should be considered as methods to screen chemicals and develop testing priorities. After
56
57 sufficient experience, the support of this approach through the more traditionally accepted animal data
58
^ may become unnecessary and this type of evidence used directly to evaluate chemical toxicity.
ACS Paragon Plus Environment
Previous I TOC
-------
Submitted to Chemical Research in Toxicology Page 7 of 35
Knowledge of the potential mechanisms of toxicity provides a rational basis for extrapolation
2
3 from chemicals where there is a great deal of data on toxicity to chemicals for which little data exists.
4
5 The differential step in many mechanisms of chemical toxicity may be generalized as the interaction
6
between a small molecule (a potential toxicant) and one or more macromolecular targets. The small
8
9
1 o molecule may be the chemical itself or one of its descendents. Modeling the potential of a molecule for
11
12 specific interactions of this type is a source of insight into its potential toxicity. In a previous paper the
13
application of an approach based on computation of the interaction between a potential molecular
I O
16
17 toxicant and a library of macromolecular targets of toxicity was proposed for chemical screening (5). In
18
19 order to use a library of this type to assess the potential for untested chemicals to be toxic and to
20
21
determine the most likely pathways for toxicity, a rapid method to evaluate interactions between the
23
24 molecule and the target is needed. Molecular "docking" (6-8) has been developed to screen large
25
26 libraries of chemicals for molecules that interact with specific sites on proteins and therefore are
27
po
2Q potential pharmaceutical agents (9-11). It also has been used to identify xenoestrogens (12) but has
30
31 infrequently applied to investigate the potential toxicity of weaker agents (5, 13). It is a rapid method
32
33 that depends on a surrogate for the interaction energy (a scoring function) that is determined
34
35
36 heuristically from data determined from many interactions between small molecules and proteins. In the
37
38 current study this computational approach is applied to a set of environmentally relevant chemicals
39
40 interacting with the estrogen receptor (ER) and the computational results compared to results from a
41
42
43 recent experimental study that determined the relative binding affinities (KB A) of these molecules to the
44
45 rat (ER) in a complex preparation (14-16).
46
47
48
49 In the experimental study, the KB As were determined by the ability of the natural ligand, to bind
50
51 to the estrogen receptors in a rat tissue preparation in the presence of increasing amounts of the test
o^
53
54 chemical. This assay is currently being evaluated for inclusion in a test battery to be used for
55
56 determining the potential of environmental chemicals to disrupt endocrine function (17, 18). It
57
^ simulates the initiating step of a process that results in estrogen receptor mediated physiological
oy
60
ACS Paragon Plus Environment
Previous I TOC
-------
Page 8 of 35 Submitted to Chemical Research in Toxicology
responses. Chemicals that are active in this assay may have tissue dependent agonist and/or antagonist
2
3 properties. Measuring the response as a function of the concentration of the test chemical allows true
4
5 binders to be separated from chemicals that interfere with the binding of the natural ligand through other
6
mechanisms. These results are analogous to what is modeled in molecular "docking" computer
8
9
1 o experiments and as such form an excellent system for exploring the use of computational "docking" as a
11
12 tool in a toxicity screen. The data set has the additional advantage that most of chemicals that bind are
13
found to bind weakly and "docking" methods can be tested for their capacity to find chemicals that bind
I O
16
17 weakly to a receptor and are not drug like.
18
19
20
21 METHODS
22
23
24
25 Molecular modeling software tools, which have been developed for the discovery of novel
26
27 pharmaceutical agents, were applied to identify chemicals that bind weakly to the rat estrogen receptor.
28
29
2Q These tools attempt to order a set of chemicals by their capacity to bind to a specific site in a protein (7).
31
32 This is accomplished by computationally "docking" each chemical into the ligand binding pocket of the
33
34 receptor. In molecular "docking" the most favorable interaction between the chemical - binding pocket
OO
oc
37 pair (toxicant-target in this case) is identified and quantified.
38
39
40
41 The protein targets used in this study were derived from crystallographic structures of the
42
43 estrogen receptors each with a ligand bound. The atoms of the ligands were computationally removed
44
45 from crystal structures to create computational targets for "docking". In order to compare the results
46
47
43 from this computational study to the experimental results from the cytosolic preparation (14, 16) four
49
50 targets were created. Multiple targets were required because the rat uterine cytosol used in the
51
52
co experimental study contained both ERa and ERp and the experimental method do not distinguish
54
55 between agonist-like and antagonists-like binding. Crystal structures were not available from the
56
^J Protein Data Bank (PDB) (19) for the rat ERa. Therefore the rat ERa targets were derived from human
58
59
60 ERa (20, 21) by homology models. There is 96.6% homology between the ligand binding domain of the
ACS Paragon Plus Environment
Previous I TOC
-------
Submitted to Chemical Research in Toxicology Page 9 of 35
rat ERa and the ligand binding domain of the human ERa. The homology models were developed in
2
3 Molecular Operating Environment substituting the ligand binding domain of the rat receptor sequence
4
5 for the similar human receptor sequence (22-24). The model targets derived in this way were
6
8 subsequently optimized by molecular mechanics using the AMBER94 force-field (25). Crystal
9
10 structures of the rat ERp are available directly from the PDB for both agonist (26) and antagonist (27)
11
12
co-crystallized conformations. The AMBER94 force-field was also used to optimize the structures of
I O
14
15 these targets. In this manner four targets were created, ERa (agonist), ERa (antagonist), ERp (agonist)
16
and ERp (antagonist).
19
20
21
22 A structural data file containing each of the 281 chemicals included in the experimental study
23
24 (14) was obtained from DSSTox (16). The label for this data set in DSSTox is KIERBL. The
25
26 structures were imported into the Molecular Operating Environment software (22) and the molecular
OQ
29 geometries were optimized using the MMFFx force field (28). For the purposes of this study a chemical
30
31 was considered active (e.g, a true competitive binder for the ER) if it had an experimentally determined
32
00
^ Ki and an ICso less than 100 \iM in the experimental study (14). Thus, of the 281 chemicals, 15 were
o^
35
36 classified as active ER binders. 266 were classified as inactive for this study. Of those classified as
37
38 inactive for this study, the experimental study left some uncertainty about 11 of those chemicals. Of
39
40
41 those 11 chemicals, 9 had limited competitive binding curves because of poor solubility and had ICso
42
43 that were uncertain in the experimental study but were greater than 100 (jM, while the other 2 chemicals
44
^ had partial competitive binding curves that were deemed sufficient to obtain an ICso values but those
46
47
48 values were greater than 100 |jM, more than 5 orders of magnitude greater than the natural ligand. All
49
50 chemicals used in this study are reported in the U.S. EPA's Distributed Structure-Searchable Toxicity
52
53 (DSSTox) Public Database Network where all experimental results and a Structure Data File with
54
55 chemical identifiers are available for public access (15, 16).
56
57
58
59
60
Q
ACS Paragon Plus Environment
Previous I TOC
-------
Page 10 of 35 Submitted to Chemical Research in Toxicology
Two different molecular "docking" approaches, using two different software packages were
2
3 employed. The first package, eHiTS (29, 30) relies on a decomposition of the optimized structure of
4
5 potential ligands into fragments. Then individually docking each fragment into the binding site,
6
obtaining multiple fragment poses and using a graph matching scheme to reconstruct the structure of the
8
9
10 potential ligand based on connectivity tables. The reconstructed molecular poses are subsequently
11
12 locally optimized and a score (a surrogate for the binding energy) for each pose is obtained from a
13
heuristic scoring function (29). The second algorithm, FRED (31, 32) applies a systematic exploration
I O
16
17 of the rotational and translational space of each putative ligand in the target space to discover the
18
19 configurations for the putative ligand-target interaction with the best score. It has been shown to be
20
21
sufficiently accurate and rapid for virtual screening for leads for potential pharmaceuticals (33, 34).
23
24 Both of the "docking" methods used in this study have been shown to be successful at finding drug-like
25
26 (molecules that bind strongly to the receptor) estrogenic molecules (29, 33). Their capacity to separate
27
po
2Q weak binders from non-binders is a primary goal of this computational study.
30
31
33 RESULTS
34
35
oc
3-7 The experimental results of Laws and co-workers (KIERBL) (14) provide an excellent data set
38
39 for investigating the application of the toxicant-target approach for the prioritization of chemicals for
40
41 potential toxicity. The advantages of comparing computational molecular "docking" results to these
43
44 experimental results are: first, there are a number of excellent crystal structures of both a and |3 estrogen
45
46 receptors available in the Protein Data Bank (20, 21, 26, 27 and others) that can be used to synthesize
47
4Q macromolecular targets for computer "docking", second, a library of 281 chemicals was tested in the
50
51 same laboratory with the same protocol for their capacity to compete with radiolabeled 17-|3-estradiol
52
CO
^~ for their binding to the rat ER and third, the experiments yield a relatively direct measurement of what is
O^
55
56 modeled in computational "docking", the energy of interaction between the test chemicals and the
57
58 receptor compared to the energy of interaction of the receptor with 17-|3-estradiol.
59
60
Q
ACS Paragon Plus Environment
Previous I TOC
-------
Submitted to Chemical Research in Toxicology Page 11 of 35
In the current study each of the 281 chemicals was computationally "docked" into 4 rat ER
2
3 targets. The large structural differences between targets obtained from co-crystallization with an agonist
4
5 or an antagonist chemical were verified by observation (results not shown). The targets created from the
6
agonist crystal structures are smaller and enclosed while the targets created from the antagonist crystal
8
9
1 o structures are larger, more open and will accommodate larger molecules.
11
12
13
14 The best scores for each molecule interacting with each target are reported for each
15
16 computational approach and used for further analysis in this study. The chemical names and the
1 8
1Q experimental determination of binding from the criteria described in the methods section and the
20
21 measured values along with the "docking" scores for the interaction of that chemical with each of the
22
23 four targets using both methods are in Table 1 of the supporting material. For this data set each
24
25
2g chemical that binds (with the exception of 17-|3-estradiol) has an experimental Ki that is 3 to more than
27
28 5 orders of magnitude larger than 17-|3-estradiol. The rat uterine cytosol used as the source of receptors
29
30
31 in the binding assays contained both the alpha and beta subtypes, with the predominant type being ERa
32
33 (35). Models that attempted to realistically combined ERa and ERp scores were used but they did not
34
35
35 significantly improve the results when compared to those from consideration of ERa alone (data not
37
38 shown). Thus, in the remainder of this analysis only the scores for ERa with both molecular docking
39
40
41 methods are used. Since the experimental results do not distinguish between molecules that bind like an
42
43 agonist and molecules that bind like an antagonist, scores for both targets have been combined to give a
44
^ composite score for each chemical. The composite score for each chemical is chosen as the score that
46
47
43 indicates the most stable interaction for that chemical. Comparison of the results for agonist and
49
50 antagonist targets might indicate whether the chemical was likely to be an agonist or an antagonist. The
51
rp
^ composite score for each chemical was considered along with the separate scores for the agonist and
Oo
54
55 antagonist targets.
56
57
58
59
60
ACS Paragon Plus Environment
Previous I TOC
-------
Page 12 of 35 Submitted to Chemical Research in Toxicology
The results from computational "docking" are scores (a surrogate for the interaction energy) for
2
3 each potential ligand-protein pair. The scores are determined by the internal scoring functions of the
4
5 computational method used. The parameters in the scoring functions have been determined from
6
multiple experimental measurements of protein-small molecule structures (30, 32). They are not
8
9
10 adjusted to optimize the results for comparison to these specific receptors or experimental results. For
11
12 the purpose of predicting potential activity a demarcation between likely active and likely inactive
13
chemicals must be chosen. An example of this demarcation is shown in Figure 1. The only adjustable
I O
16
17 parameter in this approach is the choice of demarcation between the predicted likely active chemicals
18
19 and the predicted likely inactive chemicals (line Q-R in Figure 1). In Figure 2 the predicted true positive
20
21
ratio (TPR) is plotted as a function of the predicted false positive ratio (FPR) as the position of the
23
24 demarcation line between predicted positive and negative (Q-R) is varied. Table 1 shows a synopsis of
25
26 these results for ER* targets for both "docking" methods. Using the eHiTS method the results for the
28
29 agonist target are superior to the results for the antagonist target and the composite score. Using the
30
31 FRED method the results for the agonist target are superior to the results for the antagonist target and
32
o o
^~ composite score for identifying 14 of the 15 active chemicals but finding one of the active chemicals
O^
35
36 (4,4'-sulfonyldiphenol) is more difficult, that chemical is best found by the antagonist target. All the
37
38 active chemicals appear in the first 16% of the data set when the eHiTS, fragment based approach is
39
® used with the agonist target and 39% (27% when one positive chemical is excluded) when FRED is
42 A
43 used.
44
45
46
47 For the preceding results the best score for each chemical was chosen without consideration of
48
49 the geometry of binding between the toxicant and the target. Many experimental studies of estrogen
50
52 receptor binding have shown that specific interactions between the atoms in the ligand and atoms in the
53
54 receptor play a key role in molecular recognition. As a result, a pharmacophore for binding to the
55
56 estrogen receptor has been developed (36). In theory, "docking" algorithms should insure the proper
o /
co
eg orientation of the ligand and the target so that the requirements of this pharmacophore are satisfied
60
ACS Paragon Plus Environment
Previous I TOC
-------
Submitted to Chemical Research in Toxicology Page 13 of 35
because the scoring functions should include the interactions that favor the pharmacophore recognition
2
3 process. However, the development of scoring functions is a continuing process (37, 38) and it has been
4
5 found that including pharmacophores, in the form of constraints on the allowed docking poses, improves
6
results in some cases (39, 12 and 40). In the docking program FRED, constraints of this type may be
8
9
10 applied to the geometry of the interacting molecules. The set of chemicals was docked again under the
11
12 constrained condition that only geometries that produced two appropriate hydrogen bonds between the
13
toxicant and the target (equivalent of the hydrogen bonds made by the hydroxyl group of the A-ring of
I O
16
17 17-p-estradiol to an arginine and glutamate in the binding pocket of the receptor) were permitted. The
18
19 resulting scores using this approach in FRED are shown in Table 1 of the supporting material. The plots
21
22 of TPR as a function of FPR for these results are shown in Figure 3. Table 2 shows these results for
23
24 various choices of demarcation. All positive chemicals are now found in the first 14% (8% if the same
25
2® difficult chemical is omitted) of the molecules and while the results for finding all 15 chemicals is better
28
29 for the antagonist target than the agonist target, they are quite similar and the agonist target finds the first
30
31 14 more rapidly. The eHiTS program does not yet contain the option of applying externally determined
32
o o
^ pharmacophore driven constraints directly but it is possible to use the insight from the results generated
O^
35
36 by FRED to enhance the results obtained from eHiTS. In the library of possible ligand-receptor poses
37
38 generated by eHiTS, all poses that do not satisfy the constraints have been eliminated. Those results are
39
® also shown in Table 1 of the supporting material and the plots of TPR as a function of FPR for these
42
43 results are shown in Figure 4. Table 2 shows the summary of these results for various choices of
44
45 demarcation (Q-R). All positive chemicals appear in the first 8% of the molecules. For the eHiTs
46
method with the constraints the agonist target consistently yields the best results. The addition of a
49
50 simplified pharmacophore constraint significantly decreases the number of false positives that would
51
52 result from a demarcation scheme that minimizes the false negative rate for all approaches and
53
eliminates no positive chemicals in this data set. For all the "docking" results in this study, molecules
56
57 that could not be "docked" were inactive in the experimental study.
58
59
60
17
ACS Paragon Plus Environment
Previous I TOC
-------
Page 14 of 35 Submitted to Chemical Research in Toxicology
With the exception of 17-|3-estradiol all of the chemicals in this data set that displace the natural
2
3 ligand, bind only weakly to the estrogen receptor. The computational methods used in this study were
4
5 designed to increase the discovery rate of molecules that bind efficaciously to specific protein targets
6
8 and have been shown to enrich data bases for potential pharmaceutical agents by identifying chemicals
9
10 that bind strongly to receptor targets, including the estrogen receptor (33). Applying the same set of
11
12 computational tools, an approach has been described that is capable of separating chemicals that bind
I O
14
15 weakly to the estrogen receptor from those chemicals that do not bind to the receptor at all, at least for
16
17 the example of this data set. In order to explore the activity domain of the approach used in this study, a
18
19 set of chemicals that are primarily potent estrogen receptor binders was docked into targets constructed
21
22 from the human estrogen receptor (20, 21). The activities of these chemicals at the estrogen receptors
23
24 have been measured in a similar manner to the measurements in the KIERBL data set in a single study
25
2® (41). Chemical information and the experimental determination of KB A (from (41)) along with the
28
29 "docking" scores for the ERa receptor are shown in Table 2 of the supporting material. The
30
31 demarcation lines (Q-R), established previously for the KIERBL data set with each docking method,
32
33
34 were then used to separate the active chemicals from the inactive chemicals in this additional data set of
35
36 primarily strong binders. With the eHiTS approach all chemicals that compete with E2 for the estrogen
37
38 receptor are identified by the agonist target with the exception of coumestrol, methoxychlor (a very
39
40
4
-------
Submitted to Chemical Research in Toxicology Page 15 of 35
In a previous paper (5) the long term goal of developing a library of molecular targets for
2
3 chemical toxicity was introduced. In order to use a library of this type to assess the potential for
4
5 untested chemicals to be toxic and to determine the most likely pathways for toxicity, a rapid method to
6
evaluate interactions between the molecule and the (macromolecular) targets is needed. Molecular
8
9
10 "docking" has been used to screen large libraries of chemicals for molecules that interact strongly with
11
12 specific sites on proteins and therefore are potential pharmaceutical agents. In this study we evaluate the
13
capacity of two "docking" methods to discover chemicals that interact weakly (3-5 orders of magnitude
I O
16
17 less than the natural ligand) with the estrogen receptor.
18
19
20
21 In the KIERBL data of weak binders used in this study only 5% of the chemicals have any
22
23 measured capacity to compete with 17-|3-estradiol for the ER in the tissue preparation. It is difficult to
24
25
26 estimate how many of the untested industrial chemicals are likely to interact with particular biological
27
28 receptors. However, it is likely that the amount for each receptor is small and their interactions with the
29
o r\
^ receptors weak when compared to the natural ligand. (Chemicals that interact strongly should be more
O I
32
33 easily discovered.) For this reason, the experimental data set used in this study is more relevant to
34
35 environmental circumstances than a data set composed of pharmaceuticals or other chemicals that have
36
oy
been designed to exhibit pharmaceutical like biological activity.
08
39
40
._ The choice of the ERs as a model target for this exploration of the application of "docking" as a
43
44 tool for exploring potential toxicity has added advantages. First, the target is of environmental interest
45
46 and has specific relevance for endocrine disrupter screening. Second there are a relatively large number
47
4Q of crystal structures of the ERs co-crystallized with various agonist and antagonist ligands, in the
50
51 literature. The availability of many similar crystal structures supplies leverage to counteract a short
52
53 coming of the docking methods used in this study, that is, the protein targets are not considered flexible.
54
56 The computational targets do not respond to the potential ligand and do not relax to provide a better fit
57
58 for each ligand. While there are methods that include protein flexibility (42, 43), they are currently too
59
60
ACS Paragon Plus Environment
Previous I TOC
-------
Page 16 of 35 Submitted to Chemical Research in Toxicology
computationally intensive to be considered for this type of application. In examples like the estrogen
2
3 receptor where there are a number of crystal structures available, flexibility may be included by
4
5 constructing multiple targets for the same receptor and accepting the "best" score for each toxicant-
6
target pair. In this study we have utilized targets for both the agonist and antagonist configuration of the
8
9
1 o receptor but only a single target for each of these major receptor configurations.
11
12
13
14 Two different "docking" packages with two different approaches and different scoring functions
15
16 have been used. The most useful results were obtained when a simplified pharmacophore filter was
1 8
1Q applied to constrain the allowed "docking" results. By eliminating toxicant-target poses that do not
20
21 satisfy a predetermined pharmacophore, the filter significantly reduces the number of chemicals that
22
23 would be classified as false positives for each approach used, yet its application had no effect on the
24
25
2g discovery of true positives in this data set, that are up to 5 orders of magnitude weaker than 17-0-
27
28 estradiol. The constraints of this pharmacophore based filter could be satisfied by poses generated with
29
o r\
^ 102 of the molecules considered in this study.
O I
32
33
The pharmacophore filter was derived from decades of experience with the estrogen receptors
oO
36
37 (36) and is used in this study in a simplified form. It is composed of only two of the three hydrogen
38
39 bonds found in the interaction between strong binders and the ER. More complex pharmacophores have
40
._ been proposed (44, 45). There is a balance between minimizing the false discovery rate and discovering
43
44 all potential positive chemicals that must be carefully considered when increasing the complexity of the
45
46 filter. The same pharmacophore filter used to discover chemicals that bind strongly to a receptor may
47
4Q not be appropriate in a chemical screen designed to also find chemicals that bind weakly. The
50
51 pharmacophore filter used in this study contains the geometry for a single hydrogen bond donor
52
53 interacting with a glutamate and a single hydrogen bond acceptor interacting with an arginine. For other
54
56 potential receptor targets the available literature may not be as extensive and a similar approach for the
57
58 development of a pharmacophore filter may not be feasible. An alternative approach for the
59
60
ACS Paragon Plus Environment
Previous I TOC
-------
Submitted to Chemical Research in Toxicology Page 17 of 35
development of pharmacophore filters is to use computational methods to identify a pharmacophore
2
3 from crystal structures of the target protein co-crystallized with an array of ligands (46, 47).
4
5
6
7 The pesticide methoxychlor could not fully satisfy even this simple pharmacophore. It contains
8
9 the necessary hydrogen bond acceptor but does not contain a potential hydrogen bond donor.
10
Experimental evidence suggests it interacts weakly and in a complex manner with the estrogen receptors
13
14 (48-50). HPTE, (2,2-Bis-(p-Hydroxyphenyl)-l,l,l-Trichloroethane), a metabolite of methoxychlor,
15
16 does satisfy the filter and is a more active estrogen receptor binder than its parent (48, 49) (See table 2 of
1 8
1Q the supplemental material for the docking results for HTPE). An even simpler pharmacophore may
20
21 provide the best approach for including all potential weakly estrogen molecules.
22
23
24 The degree of improvement in the results of this study through application of the pharmacophore
25
Oft
filter was surprising because the specific interactions characterized by the filter should already be in the
28
29 scoring functions used in "docking". This result suggests that docking methods could be improved (at
30
31 least relative to their capacity to discover weak binders) by improving scoring functions. Molecular
32
00
^ docking with a pharmacophore filter might be incorporated into a tiered or other structured approach for
o^
35
36 chemical screening (see for instance 51). The application of a pharmacophore filter could be
37
38 incorporated into one of the initial tiers and all molecules that could not satisfy the filter in any docking
39
pose would be eliminated before docking. Each individual pose would still be required to satisfy the
42
43 filter for scoring during the "docking" phase which would be part of a later tier.
44
45
46
47 The "docking" results obtained from the ERa agonist target are consistently better than the
48
docking results for all other individual targets and the composite scores for KIERBL data set of weak
ou
51
52 binders. Part of the explanation is that the tissue preparations in the experimental study contain much
53
54 more ERa than ERp (35). All of the active chemicals are discovered most rapidly by using only the
oo
cc
57 agonist target and the addition of results for the antagonist target by creating composite scores increases
58
59 the false discovery rate. This may result from something specific about the KIERBL data set (that
60
ACS Paragon Plus Environment
Previous I TOC
-------
Page 18 of 35 Submitted to Chemical Research in Toxicology
perhaps it does not contain antagonists). The decrease in the discovery rate when composite scores
2
3 (constructed from agonist and antagonist results) are used may result from the crystal structures used to
4
5 create the targets. As a result of a more favorable crystal structure interactions in the antagonist target,
6
there might be a constant favorable offset. This would introduce false positives through the antagonist
8
9
10 target when compared to the agonist target. The current results suggest that this is the case and an
11
12 adjustment of the scores when comparing results from different targets is needed. Clearly, the
13
discovery rates of this study (with the KJERBL data set) would not be changed by including docking at
I O
16
17 an antagonist target and combining it with the results from the agonist target with an offset but for data
18
19 sets that contained (bulky) antagonists the results would be improved. Most of the molecular
20
21
identification features are the same for the agonist and antagonist targets but the antagonist target can
23
24 accommodate larger molecules. The crystal structure of ERa co-crystallized with hydroxy-tamoxifen
25
26 was used to create the antagonist target in this study. Hydroxy-tamoxifen is too bulky to dock well into
28
29 the agonist target. Similar bulky known antagonist molecules were not discovered with only an agonist
30
31 target when "docking" the series of pharmaceutical like molecules. It is likely that a general screen
32
o o
^~ would need both targets. As an element in a tiered or another type of ordered approach for determining
O^
35
36 chemicals that bind to the estrogen receptor, agonist and antagonist targets might be used separately
37
38 after prefiltering steps based on various molecular descriptors for each target.
39
40
41
42 The result of the application of a "docking" tool to the toxicant-target paradigm is a list of
43
chemicals ordered by their score. The score for a molecule is a surrogate for the interaction energy
4O
46
47 between the potential ligand and the target. These scores are derived from scoring functions that are
48
49 obtained from consideration of the general form of the interaction between a protein and a small
50
52 molecule and experimental data describing many protein small molecule interactions. The scoring
53
54 functions do not depend on the specific chemicals or the specific protein being considered. The score
55
56 for a specific protein-ligand interaction depends on the geometry of the interaction and the description of
o /
co
eg the general scoring function. The development of the ordered list is independent of and does not require
60
17
ACS Paragon Plus Environment
Previous I TOC
-------
Submitted to Chemical Research in Toxicology Page 19 of 35
any data on other chemicals interacting with this specific target. This is different than many other
2
3 computational methods for evaluating potential toxicity where data for a training set of chemicals
4
5 interacting with the specific receptor is required. The use of "docking" tools requires only a crystal
6
structure of the target or a similar protein if homology models are used.
8
9
10
The variable line Q-R (see Figurel) used to dichotomize the set of molecules is the only
13
14 adjustable parameter in this approach. Optimizing the placement of Q-R allows the determination of the
15
16 best separation between active and inactive chemicals in the data set and facilitates identification of all
17
1 8
1Q the positive chemicals in the smallest subset of molecules. This is a reasonable approach where the
20
21 necessary relevant data exists but in cases where that data is not available or in cases where there are
22
23 other rationales (perhaps resource driven) for choosing the number of chemicals to be tested the ordered
24
25
26 list provides a justification for the order of chemicals to be tested.
27
28
29 When chemicals in the initial data set are compared to a large set of known estrogen receptor
30
31 binders (52) and three physicochemical properties that are relevant for ER binding (but probably not a
32
00
^ complete set of relevant properties relevant to ER binding) are computed (53. 54) and employed for
o^
35
36 comparison, the KIERBL data set is found to be more diverse than the set of known estrogen binders
37
38 (see Figure 6). To a major extent this is the case because most of the chemicals are inactive. When only
39
the active chemicals are considered 4 of the 15 active chemicals are found to be within the
42
43 physicochemical space of the external strong binding chemicals, 8 are adjacent to that space and 3 of the
44
45 active chemicals are outside that space. About 10% of the nonbinding chemicals are contained in the
46
space of external strong binding chemicals.
49
50
52 CONCLUSIONS
53
54
55
56 In the primary experimental library considered in this study, only 14 of the 280 (excluding 170-
57
58 estradiol) chemicals have measured Ki values and an IC50 less than lOOuM. Computational molecular
59
60
1 8
ACS Paragon Plus Environment
Previous I TOC
-------
Page 20 of 35 Submitted to Chemical Research in Toxicology
"docking" methods are able to identify all of these chemicals using targets constructed from ER crystal
2
3 structures. For the best approach applied in this study, the 14 positive chemicals are identified with only
4
5 8 false positive assignments. All of the computational molecular docking approaches and comparisons
6
used in this study give reasonable results. All chemicals from the experimental library that could not be
8
9
1 o "docked" in the biomolecular targets were inactive.
11
12
13
14 The molecular docking algorithms applied in this study were developed to aid in pharmaceutical
15
16 discovery. As such they were designed to find the molecules with the greatest activity. However, the
1 8
^g active chemicals in this study (with the exception of 17|3-estradiol itself) all have activities 3-5 orders of
20
21 magnitude smaller than 17-|3-estradiol and are discovered by these "docking" algorithms. 95% of the
22
23
24 chemicals in the library considered in this study are not active. This is not typical of computational
25
26 studies of environmental chemicals where the chemical library often contains more positive chemicals
27
28 and attempts are made to balance the data base. This data set is most likely more similar to the problem
30
3^ seen in environmental circumstances where for any biomolecular target the majority of the chemicals
32
33 will be negative. The relevant problem is to identify the few chemicals that are positive.
34
35
36
37 This approach is based on modeling the forces that determine the interaction between a small
38
39 molecule and a biomolecular target. A score (a surrogate for that interaction energy) is obtained for each
41
42 molecule. The actual determination of likely positive and negative chemicals depends on the choice of a
43
44 line of demarcation (called Q-R in Figure 1) imposed on the continuous spectrum of scores for the
45
model interaction. This value may be adjusted to provide a more or less conservative evaluation. In
48
49 order to use this approach, there must be high quality structural descriptions of the biomolecular targets
50
51 or similar macromolecules. Previous data on the activity of other chemicals acting at this target are not
52
CO
^~ required.
O^
55
56
In this approach a single step in a complex path for toxicity is modeled. It is an indicator of
oo
59
60 suitability of the potential toxicant to take part in that single facet of more general mechanisms for
1Q
ACS Paragon Plus Environment
Previous I TOC
-------
Submitted to Chemical Research in Toxicology Page 21 of 35
toxicity. There are undoubtedly other pathways and perhaps other steps in a single pathway for toxicity
2
3 that may be influenced by a potential chemical toxicant. As a library of targets is developed,
4
5 interactions with many targets may be combined to provide a more complete picture of potential
6
chemical toxicity. The results generated by the toxicant-target approach may be integrated with other
8
9
1 o types of data, chemical information, pharmacophores and bioassay data. These different types of data
11
12 may be best combined in a tiered approach that recognizes the strengths of each type of data and the
13
requirements for obtaining it.
I O
16
17
18
19
on
2" ACKNOWLEDGMENT: MRG was supported during a portion of this work by NHEERL-DESE
22
23 Training Agreement, EPA CT829471. This work was reviewed by EPA and approved for publication
24
25 but does not necessarily reflect official Agency policy. We thank Drs. Chang, Dix and Kavlock for
26
27
28 reading and commenting on this manuscript and Dr. Richard for helpful advice.
29
30
32 SUPPORTING INFORMATION PARAGRAPH Table 1 of the supporting material contains the
33
34 name, CAS number, smiles code, measured ICso and the activity calls based on the criteria described in
35
3® this study for each chemical. It also contains all the docking data (scores) used in this study, four targets
O /
38
39 ERa, ER|3, from both agonist and antagonist crystal structures and four methods, FRED without
40
41 constraints, Fred with 2 constraints, eHiTS without constraints and eHiTS results with the constraints
42
.. from FRED superimposed on the scores.
45
46
Jg SUPPORTING INFORMATION PARAGRAPH Table 2 of the supporting material contains the
49
50 name, CAS number, smiles code, RBAs for ERa from reference (41). It also contains the docking data
51
52
53 (scores) for both ERa the agonist and antagonist targets using both Fred and eHiTS with out constraints.
54
55
56
57
58 REFERENCES
59
60
70
ACS Paragon Plus Environment
Previous I TOC
-------
Page 22 of 35 Submitted to Chemical Research in Toxicology
(1) Judson, R., Richard, A., Dix D., Houck, K., Martin, M., Kavlock, R., Dellarco, V., Henry, T.,
2 Holderman, T., Sayre, P., Tan, S., Carpenter, T. and Smith, E., (2009) The Toxicity Data Landscape for
3 Environmental Chemicals, Environmental Health Perspectives, 117(5), 685-695.
4
5 (2) Richard, A.M., (2006) Future of Toxicology— Predictive Toxicology: An Expanded View of
6 "Chemical Toxicity". Chemical Research in Toxicology, 19(10), 1257-1262.
7
8 (3) National Research Council (NRC). (2007). Toxicity Testing in the 21st Century: A Vision and a
9 Strategy. National Academy Press, Washington, DC. Available at: http://www.nas.edu.
10
11 (4) Dix, D. I, Houck, K. A., Martin, M. T., Richard, A, M, Setzer, R. W., and Kavlock, R. 1(2007) The
12 ToxCast Program for Prioritizing Toxicity Testing of Environmental Chemicals. Tax. Sci. 95, 9-12.
I O
14 (5) Rabinowitz, J. R., Goldsmith, M-R., Little, S. B., and Pasquinelli, M. A. (2008) Computational
16 Molecular Modeling for Evaluating the Toxicity of Environmental Chemicals: Prioritizing Bioassay
17 Requirements. Environmental Health Perspectives, 116,573-577.
18
19 (6) Kuntz, ID., (1992) Structure-based strategies for drug design and discovery. Science
20 257(5073): 1078-1082.
21
22 (7) Halperin, I, Ma, B., Wolfson, H., Nussinov, R., (2002). Principles of docking: an overview of
23 search algorithms and a guide to scoring functions. Proteins 47(4):409-443.
24
25 (8) Sousa, S.F., Fernandes, P.A., Ramos, M.J., (2006) Protein-ligand docking: current status and future
26 challenges. Proteins Struct Funct Bioinform 65:15-26.
28 (9) Abagyan, R., Totrov, M., (2001) High-throughput docking for lead generation. Curr Opin Chem
y J8/o/5(4):375-382.
O\J
31
3_ (10) Bissantz C., Folkers G.,, and Rognan D., (2000) Protein-Based Virtual Screening of Chemical
33 Databases. 1. Evaluation of Different Docking/Scoring Combinations. J. Med. Chem. 43:4759-4767.
34
35 (11) Cozzini P. and Dottorini T., (2004) Is it possible docking and scoring new ligands with few
36 experimental data? Preliminary results on estrogen receptor as a case study. European J. Med. Chem.
37 31:601-609.
38
39 (12) Amadasi, A., Mozzarelli, A., Meda, C., Maggi, A., and Cozzini, P., (2009) Identification of
40 Xenoestrogens in Food Additives by an Integrated In Silico and In Vitro Approach. Chem.Res. Toxicol.,
JJ 22(l):52-63
(13) Ekins, S., (2004) Predicting undesirable drug interactions with promiscuous proteins in silico. Drug
45 Discov Today 9(6):276-285.
46
47 (14) Laws, S. C., Yavanhxay, S., Cooper, R. L., and Eldridge, J. C. (2006) Nature of the Binding
48 Interaction for 50 Structurally Diverse Chemicals with Rat Estrogen Receptors. Tox. Sci. 94, 46-56.
49
50 (15) Richard, A. M., and Williams, C. R. (2002) Distributed Structure-Searchable Toxicity (DSSTox)
51 Public Database Network: A Proposal. Mutation Research, 499,27-52.
52
53 (16) S. Laws, J. Kariya, M. Wolf, and A.M. Richard (2009) DSSTox EPA Estrogen Receptor Ki
54 Binding Study (Laws et al.) Database - (KIERBL): SDF file and documentation, Launch version:
55 KIERBL_vla_278_17Feb2009, www.epa.gov/ncct/dsstox/sdf_kierbl.html
56
cl 07) http://epa.gov/endo/pubs/edspoverview/fmalrpt.htm (accessed 3/24/09)
58
59
60
01
ACS Paragon Plus Environment
Previous I TOC
-------
Submitted to Chemical Research in Toxicology Page 23 of 35
(18) Nffl NTP/ICCVAM Report (2002) Current Status of Test Methods for Detecting Endocrine
3
2 Disrupters: In Vitro Estrogen Receptor Binding Assays, NIEHS, RTF, NC
4 http://iccvam.ni ehs.nih.gov/docs/endo_docs/fmall002/erbndbrd/ERBd034504.pdf
Q (19) Berman, H.M., Westbrook, I, Feng, Z., Gilliland, G., Bhat, T.N., Weissig, G.H., Shindyalov, I.N.,
7 Bourne, P.E. (2000) The Protein Data Bank , Nucleic Acids Research, 28, 235-242
8
9 (20) PDB ID: 1GWR, Warnmark, A., Treuter, E., Gustafsson, J.A., Hubbard R.E., Brzozowski A.M.,
10 Pike A.C. (2002) Interaction of transcriptional intermediary factor 2 nuclear receptor box peptides with
11 the coactivator binding site of estrogen receptor alpha. JBiol Chem. 277(24):21862-8.
12
13 (21) PDB ID: 3ERT, Shiau A.K., Barstad D., Loria P.M., Cheng L., Kushner P.J., Agard D.A., Greene
14 G.L., (1998) The structural basis of estrogen receptor/coactivator recognition and the antagonism of this
16 interaction by tamoxifen. Cell. 95(7):927-37.
1g (22) MOE; Chemical Computing Group, Inc.: Montreal, Quebec, Canada, 2008.
19
20 (23) Levitt, M. Accurate Modelling of Protein Conformation by Automatic Segment Matching. J. Mol.
21 Biol. 226, 507-533 (1992).
22
23 (24) T. Fechteler; U.: Schomburg, D. Dengler: Prediction of Protein Three-Dimensional Structures in
24 Insertion and Deletion Regions: A Procedure for Searching Data Bases of Representative Protein
25 Fragments Using Geometric Criteria. J. Mol. Biol. 253 (1995), 114-131
26
27 (25) Cornell, W.D., Cieplak, P., Bayly, C.I., Gould I.R., Merz, Jr., K.M., Ferguson, D.M., Spellmeyer,
28 D.C., Fox, T., Caldwell, J.W., and Kollman, P.A., A Second Generation Force Field for the Simulation
of Proteins, Nucleic Acids, and Organic Molecules, J. Am. Chem. Soc. 1995, 117, 5179-5197.
32 (26) PDB ID: 1HJ1, Pike A.C., Brzozowski A.M., Walton J., Hubbard R.E., Thorsell A.G, Li Y.L.,
33 Gustafsson J.A., Carlquist M., (2001) Structural insights into the mode of action of a pure antiestrogen.
34 Structure. 9(2): 145-53.
35
36 (27) PDB ID: 2J7X, Pike, A.C.W., Brzozowski, A.M., Hubbard, R.E., Walton, J., Bonn, T.,
37 Thorsell, A.-G., Engstrom, O., Ljunggren, J., Gustaffson, J.-A., Carlquist, M. , Structure of Agonist-
38 Bound Estrogen Receptor Beta LED in Complex with Lxxll Motif from NcoaS , to be published
39
40 (28) Halgren T., (1996) Merck molecular force field. I. Basis, form, scope, parameterization and
^2 performance of MMFF94; J. Comput. Chem. 17 490-519
(29) http://www.simbiosys.ca/ehits/index.html (Accessed 2/22/09)
*g (30) Zsoldos, Z., Reid, D., Simon, A, Sadjad, B.S., Johnson, A.P., (2006) eHiTS: An Innovative
47 Approach to the Docking and Scoring Function Problems, Current Protein and Peptide Science, 2006,
48 7,421-435.
49
50 (31) http://www.eyesopen.com/products/applications/FRED.html (Accessed 2/22/09)
51
52 (32) McGann, M.R., Almond, H.R., Nicholls, A., Grant, J.A., and Brown, F.K., (2003)
53
54 Gaussian Docking Functions, Biopolymers, 68, 76-90.
55
56 (33) Schultz-Gasch, and T., Stahl, M., (2003) Binding Characteristics in structure-based virtual
57 screening:evaluation of current docking tools. J Mol Model, 9, 47-57.
58
59
60
00
ACS Paragon Plus Environment
Previous I TOC
-------
Page 24 of 35 Submitted to Chemical Research in Toxicology
(34) Kellenberger, E., Rodrigo, J., Muller, P., and Rognan, D., (2004) Comparative evaluation of eight
2 docking tools for docking and virtual screening accuracy. Proteins: Structure, Function, and
3 Bioinformatics 57, 225-242.
4
5 (35) Shughrue, P.J., Lane, M.V, Scrimo, P.J., and Merchenthaler, I. (1998) Comparative distribution of
6 estrogen receptor-a (ER-a) and P (ER-P) mRNA in the rat pituitary, gonad, and reproductive tract.
7 Steroids. 63(10):498-504
8
9 (36) Anstead, G.M., Carlson, K.E. and Katzenellenbogen, J.A.(1997) The estradiol pharmacophore:
10 Ligand structure-estrogen receptor binding affinity relationships and a model for the receptor binding
] 2 site Steroids, 62(3), 268-303.
13
(37) Stahl, M., Rarey M.,(2001). "Detailed analysis of scoring functions for virtual screening." JMed
15 Chem 44(7): 1035-42.
16
17 (38) Jain, A. N., (2006). "Scoring functions for protein-ligand docking." Curr Protein Pept Sci 7(5):
18 407-20.
19
20 (39) Muthasa, D., Sabnis, Y.A., Lundborga, M., Karlen, A., (2008) Is it possible to increase hit rates in
21 structure-based virtual screening by pharmacophore filtering? An investigation of the advantages and
22 pitfalls of post-filtering, Journal of Molecular Graphics an d Modelling 26(8), 1237-1251.
23
24 (40) Peach, M.L., Nicklaus, M.C., (2009) "Combining docking with pharmacophore filtering for
improved virtual screening" J. Cheminformatics http://www.j cheminf.com/content/1/1/6
26
27 (41) Kuiper, G.G.J.M., Carlsson, B., Grandien, K., Enmark, E., Hagblad, J., Nilsson S., Gustafsson, J.-
28 A. (1997) "Comparison of Ligand Binding Specificity and Transcript Tissue Distribution of Estrogen
^ Receptors a and P" Endocrinology 138(3): 863-879.
oU
31 (42) Carlson H.A., (2002). Protein flexibility and drug design: how to hit a moving target. Curr Opin
32 Chem Biol 6(4):447-452.
33 v '
34 (43) Sherman W., Day T., Jacobson M.P., Friesner R.A., Farid R., (2006). Novel procedure for
^ modeling ligand/receptor induced fit effects. JMedChem 49:534 -553.
oD
3g (44) Mukherjee S., Nagar S., Mullick S., Mukherjee A., Saha A., (2008). Pharmacophore
39 mapping of arylbenzothiophene derivatives for MCF cell inhibition using classical and 3D space
40 modeling approaches. Journal of Molecular Graphics and Modelling, 26(6), 884-992.
41
42 (45) Yang, J.-M., Shen T.-W., (2005) "A Pharmacophore-Based Evolutionary Approach for Screening
43 Selective Estrogen Receptor Modulators" Proteins: Structure, Function and Bioinformatics, 59, 205-
44 220.
45
46 (46) Ekins, S., Erickson, J.A., (2002) "A Pharmacophore for Human Pregnane X Receptor Ligands
Jg "Drug Metabolism and Distribution, 30(1), 96-99.
49
5Q (47) Podolyan, Y., Karypis,G., (2009) "Common Pharmacophore Identification Using Frequent Clique
51 Detection Algorithm" J. Chem. Inf. Model, 49 (1), 13-21.
52
53 (48) Gaido, K.W., Leonard, L.S., Maness, S.C., Hall, J.M., McDonnell, D.P., Saville, B., Safe, S.,
54 (1999) "Differential Interaction of the Methoxychlor Metabolite 2,2-Bis-(p-Hydroxyphenyl)-1,1,1-
55 Trichloroethane with Estrogen Receptors a and P" Endocrinology, 140(12), 5746-5753.
56
57 (49) Blair, R.M., Fang, H., Branham, W.S., Hass, B.S., Dial, S.L., Moland, C.L., Tong, W., Shi, L.,
58 Perkins, R., Sheehan, D.M., (2000) "The Estrogen Receptor Relative Binding Affinities of 188 Natural
60 and Xenochemicals: Structural Diversity of Ligands" Toxicological Sciences 54, 138-153.
73
ACS Paragon Plus Environment
Previous I TOC
-------
Submitted to Chemical Research in Toxicology Page 25 of 35
(50) Laws, S.C., Carey, S.A., Ferrell, J.M., Bodman, G.J., Cooper R.L., (2000) "Estrogenic Activity of
2 Octylphenol, Nonylphenol, Bisphenol A and Methoxychlor in Rats" lexicological Sciences, 54 154-
3 167.
5 (51) Fang, H., long, W., Welsh, W.J., Sheehan, D.M., (2003) "QSAR Models in Receptor-Mediated
6 Effects: the Nuclear Receptor Superfamily" Theo Chem 622, 113-125.
7
8 (52) Liu,T., Lin,Y., Wen,X., Jorrisen, R.N., Gilson, M.K., (2007) "BindingDB: "A Web-Accessible
9 Database of Experimentally Determined Protein-Ligand Binding Affinities" Nucleic Acids Research 35,
10 D198-D201.
11
12 (53) Ertl, P., Rohde, B., Selzer, P., (2000) Fast Calculation of Molecular Polar Surface Area as a Sum
. . of Fragment-Based Contributions and Its Application to the Prediction of Drug Transport Properties; J.
15 Med. Chem. 43 3714-3717. (as implemented in 22)
16
17 (54) Wildman, S.A., Crippen, G.M., (1999) Prediction of Physiochemical Parameters by Atomic
18 Contributions; J. Chem. Inf. Comput. Sci. 39 (5) 868-873. (as implemented in 22)
19
20
21
22
23
24
25
26
27 FIGURE CAPTIONS
28
29
30 Figure 1. An Idealized Example of the Number of Chemicals as a Function of Docking Score. In this
31
o o
~Z example the scoring function imperfectly separates the positive chemicals from the negative chemicals.
OO
34
35 A line Q-R is drawn and used to make a prediction of the activity of each chemical. The predictions
36
37 made based on the docking scores are a function of the position of Q-R as it moved horizontally.
38
39
40 Figure 2a. Docking results using the method FRED with no constraints. The True Positive Rate (TPR)
41
42
43 is shown as a function of the False Positive Rate (FPR) for agonist, antagonist targets and a composite
44
45 score constructed from selecting the best score of the two for each chemical. They are compared to
46
47 random selection.
48
49
1-1 Figure 2b. Docking results using the method eHiTS with no constraints. The TPR is shown as a
52
53 function of the FPR for agonist, antagonist targets and a composite score constructed from selecting the
54
55 best score of the two for each chemical. They are compared to random selection.
56
57
58
59
60
ACS Paragon Plus Environment
Previous I TOC
-------
Page 26 of 35 Submitted to Chemical Research in Toxicology
Figure 3. Docking results using the method FRED with 2 constraints. The TPR is shown as a function
2
3 of the FPR for agonist, antagonist targets and a composite score constructed from selecting the best
4
5 score of the two for each chemical. They are compared to random selection.
6
7
8 Figure 4. Docking results using the method eHiTS with 2 constraints. The TPR is shown as a function
9
10
1 1 of the FPR for agonist, antagonist targets and a composite score constructed from selecting the best
12
1 3 score of the two for each chemical. They are compared to random selection.
14
15
16 Figure 5. Comparison of docking scores for the KJERBL data set and a set of known strong estrogen
1 8
1Q agonists (see reference 41). See tables 1 and 2 of the supplemental material for scores for all chemicals
20
21 considered. Only a single active chemical has a score that is above the demarcation determined for the
22
23 weak binders.
24
25
Oft
Figure 6. Comparison of the KIERBL data set of weak binding chemicals to a larger set of known ER
28
29 binding chemicals. It is shown in the three dimensions of relevant (but probably not complete)
30
31 computed (22) physical properties, log P (54), total polar surface area (TPSA) (53), and molecular
32
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
weight.
75
ACS Paragon Plus Environment
Previous I TOC
-------
Submitted to Chemical Research in Toxicology
Page 27 of 35
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
Table 1.
A Summary
of the results for each approach
constraints
Number of True
Positives
Identified
Fred agonist
Fred-
antagonist
Fred-
composite
eHiTS agonist
eHiTS
jmtagonist
eHiTS
5
16
20
23
8
18
18
10
51
57
59
18
50
51
applied.
14
76 (27)
105
104
36
95
68
without
75
153 (54)
109 (39)
119 (42)
46 (16)
97 (35)
74 (26)
Table 1. A Summary of the results for each approach without constraints applied. Each
entry is the number of chemicals that would be called positive when Q-R is adjusted to
yield the number of true positives heading that column. For example using the eHiTS
method and the rat ERa agonist target to obtain the first 5 true positives 8 chemicals are
called positive. In that case there are 3 false positives 10 false negatives and 263 true
negatives. The numbers in parenthesis are the percent of the data base. For the last
column there are no false negatives.
ACS Paragon Plus Environment
26
Previous
TOC
-------
Page 28 of 35
Submitted to Chemical Research in Toxicology
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
Table 2. A Summary of the results for each approach when 2
constraints from the pharmacophore are applied.
Number of True
Positives
Identified
10
14
15
Fred agonist
Fred-
antagonist
Fred-
composite
eHiTS agonist
eHiTS
antagonist
10
14
18
15
14
19
39 (14)
37 (13)
48 (17)
23(8)
42 (15)
eHiTS
composite
10
20
32 (11)
Table 2. A Summary of the results for each approach when 2 constraints from the
pharmacophore are applied. Each entry is the number of chemicals that would be called
positive when Q-R is adjusted to yield the number of true positives heading that column.
For example using the eHiTS method and the rat ERa agonist target to obtain the first 5
true positives 7 chemicals are called positive. In that case there are 2 false positives 10
false negatives and 264 true negatives. The numbers in parenthesis are the percent of the
database that number represents. For the last column there are no false negatives.
ACS Paragon Plus Environment
27
Previous
TOC
-------
Submitted to Chemical Research in Toxicology
Page 29 of 35
Q
05
O
E
CD
6
CD
.0
E
Score
R
Predicted
.2
-------
Page 30 of 35
Submitted to Chemical Research in Toxicology
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
AQ
Results with FRED no Constraints
1.0
0.8
CD
-i— <
CO
K.
CD
O
Q_
CD
0.6
0.4
0.2
0.0
Agonist
Antagonist
Composite
Random
0.0
0.2
0.4 0.6
False Positive Rate
ACS Paragon Plus Environment
0.8
1.0
Previous
-------
Submitted to Chemical Research in Toxicology
Page 31 of 35
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
AQ
Results with eHiTS no Constraints
co
CD
o
Q_
CD
Agonist
Antagonist
Composite
Random
0.0
0.2
0.4 0.6
False Positive Rate
ACS Paragon Plus Environment
0.8
1.0
Previous
-------
Page 32 of 35
Submitted to Chemical Research in Toxicology
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
AQ
Results with FRED 2 Constraints
1.0
0.8
co
CD
O
Q_
0.6
0.4
0.2
0.0
Agonist
Antagonist
Composite
Random
0.0
0.2
0.4 0.6
False Positive Rate
ACS Paragon Plus Environment
0.8
1.0
Previous
-------
Submitted to Chemical Research in Toxicology
Page 33 of 35
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
AQ
Results with eHiTS 2 Constraints
1.0
0.8
CD
CD
o
Q_
(D
0.6
0.4
0.2
0.0
Agonist
Antagonist
Composite
Random
0.0
0.2
0.4 0.6
False Positive Rate
ACS Paragon Plus Environment
0.8
1.0
Previous
-------
Page 34 of 35
Submitted to Chemical Research in Toxicology
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
AQ
eHiTS (Log KO
-10 -9
-, c -5 -A -
'7
FRED (score/ 10")
-1 10
Strong binder
D Weak binder (KIERBL)
ACS Paragon Plus Environment
Previous
-------
Submitted to Chemical Research in Toxicology
Page 35 of 35
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
AQ
TPSA/20(A*)
Q KIERBL non-binding
• KIERBL weak-binding
Q ER strong binding (bindingdb.org)
ACS Paragon Plus Environment
Previous
-------
Bioactivity Profiling Using BioMAP Cell Systems
Houck et al
In Press
Profiling Bioactivity of the ToxCast Chemical Library Using BioMAP
Primary Human Cell Systems
Keith A. Houck,a David J. Dix,a Richard S. Judson,a Robert J. Kavlock,a Jian
Yang,b Ellen L. Bergb
"National Center for Computational Toxicology
Office of Research and Development
United States Environmental Protection Agency
Research Triangle Park, NC 27711
bBioSeek, Inc.
310 Utah, Suite 100
South San Francisco, CA 94080
Other contact information:
dix.david@epa.gov
iudson.richard@epa.gov
kavlock. robert@epa.gov
eberg@bioseekinc.com
919-541-2701
919-541-3085
919-541-2326
650-552-0721
Address correspondence to:
Keith Houck, Ph.D.
US EPA
109 T.W. Alexander Dr.
D343-03
Research Triangle Park, NC 27711
Tel: 919-541-5519
Fax: 919-685-3371
Email: houck.keith@epa.gov
Disclaimer: The United States Environmental Protection Agency through its Office of
Research and Development funded and managed the research described here. It has
been subjected to Agency administrative review and approved for submission and peer
review.
Word Count: 5992
Short Title: Bioactivity Profiling Using BioMAP Cell Systems
03JUNE2009 Version
Previous
1/27
TOC
Revised/Submitted
-------
Bioactivity Profiling Using BioMAP Cell Systems Houck et al
ABSTRACT
The complexity of human biology has made prediction of health effects as a
consequence of exposure to environmental chemicals especially challenging. Complex
cell systems, such as the Biologically Multiplexed Activity Profiling (BioMAP) primary,
human, cell-based disease models, leverage cellular regulatory networks to detect and
distinguish chemicals with a broad range of target mechanisms and biological processes
relevant to human toxicity. Here we utilize the BioMAP human cell systems to
characterize effects relevant to human tissue and inflammatory disease biology following
exposure to the 320 environmental chemicals in the Environmental Protection Agency's
(EPA) ToxCast phase I library. The ToxCast chemicals were assayed at four
concentrations in eight BioMap cell systems, with a total of 87 assay endpoints resulting
in over 100,000 data points. Within the context of the BioMap database, ToxCast
compounds could be classified based on their ability to cause overt cytotoxicity in
primary human cell types, or according to toxicity mechanism class derived from
comparisons to activity profiles of BioMap reference compounds. ToxCast chemicals
with similarity to inducers of mitochondrial dysfunction, cAMP elevators, inhibitors of
tubulin function, inducers of endoplasmic reticulum stress or NFicB pathway inhibitors
were identified based on this BioMap analysis. This dataset is being combined with
additional ToxCast datasets for development of predictive toxicity models at the EPA.
Key Words: Toxicology, primary human cells, bioactivity profiling, chemical genetics
03JUNE2009 Version 2/27 Revised/Submitted
Previous I TOC
-------
Bioactivity Profiling Using BioMAP Cell Systems Houck et al
INTRODUCTION
Alternatives to whole animal testing of environmental and industrial chemicals
are needed for understanding the toxicity potential of the many thousands of chemicals
and materials in commercial use. Cost, animal welfare concerns, and relevance to
human risk are the major issues driving this need. The U.S. EPA has initiated a large-
scale effort, ToxCast, which is investigating high-throughput, in vitro assays as a means
to develop predictive toxicology models. The goal of the project is to compose broad
bioactivity profiles characterizing the in vitro biological activity of a reference set of
chemicals1. These profiles will be correlated with in vivo toxicity endpoints culled from
extensive in vivo animal studies, including those used for pesticide registration
submissions to the EPA (e.g., chronic toxicity endpoints)2. Generation of bioactivity
profiles for chemicals lacking toxicity information and analysis of similarity to reference
chemicals would then be applied to predict potential for toxicity. The ultimate goal is to
provide an efficient means to prioritize for detailed study the approximately 10,000
industrial and environmental chemicals of potential concern for which minimal toxicity
data currently exists.3
One important focus of the ToxCast project is measuring the chemical
perturbation of critical cellular signaling pathways that may represent potential modes of
chemical toxicity. This is being accomplished using in vitro bioactivity profiles derived
from screening the ToxCast chemical library against both specific molecular targets
using high-throughput, biochemical screening assays, as well as testing in a large suite
of cellular assays. One such cellular system that characterizes pharmaceutical drug
function is based on statistical analysis of protein expression in a panel of assays using
primary human cells stimulated in complex environments. This approach, termed
Biologically Multiplexed Activity Profiling (BioMAP), provides characterization of drug
function across a broad range of tissue and disease biology. The pattern of activity in
these systems allows classification of molecules according to mechanism of action,
possibly providing insights into clinical phenomena4"6. The cell systems contain primary
human cells in different environments relevant to vascular inflammation and immune
activation. BioMAP profiling has been shown to detect and discriminate multiple
functional drug classes including glucocorticoids; TNF-a antagonists; and inhibitors of
03JUNE2009 Version 3/27 Revised/Submitted
Previous I TOC
-------
Bioactivity Profiling Using BioMAP Cell Systems Houck et al
calcineurin, HMG-CoA reductase, heat shock protein 90, inosine monophosphate
dehydrogenase, phosphodiesterase 4, phosphoinositide kinase-3, and p38 mitogen
activated kinase, among others.4"6
BioMAP systems are designed to model complex human disease and tissue
biology by stimulating human primary cells, single cell type or defined mixtures of cell
types, such that multiple disease- and tissue-relevant signaling pathways are
simultaneously active. The choice of cell types and stimulations is guided by knowledge
of relevant disease biology and mechanisms. Chemical effects are then recorded by
measuring biologically meaningful protein readouts relevant to significant biological
responses (e.g., inflammation, tissue remodeling).4"7 Activity profiling in BioMAP
systems, in conjunction with BioSeek's database that contains profiles of hundreds of
experimental pharmaceutical compounds and approved therapeutics, provided one
interpretation of potential modes of toxicity. The primary goal of the present study was to
generate BioMAP profiles that could then be used as part of the entire ToxCast in vitro
dataset for classifying environmental chemicals in the ToxCast database based on
predictive signatures and putative toxicity pathways.
MATERIALS AND METHODS
Cell culture
BioMAP systems employed are shown in Table 1. Human umbilical vein
endothelial cells (HUVEC) were pooled from multiple donors, cultured according to
standard methods, and plated into microtiter plates at passage four. Human neonatal
foreskin fibroblasts (HDFn) from three donors were pooled and cultured according to
standard methods. HDFn were plated in low serum conditions 24 hr before stimulation
with cytokines. Primary human bronchial epithelial cells, arterial smooth muscle cells
and keratinocytes were cultured according to standard methods. Peripheral blood
mononuclear cells (PBMC) were prepared from buffy coats from normal human donors
according to standard methods. Monocyte-derived macrophages were differentiated in
the presence of M-CSF according to standard procedures. Concentrations/amounts of
agents added to confluent microtiter plates to build each system: cytokines (IL-1(3, 1
03JUNE2009 Version 4/27 Revised/Submitted
Previous I TOC
-------
Bioactivity Profiling Using BioMAP Cell Systems Houck et al
ng/ml; TNF-a, 5 ng/ml; IFN-y, 20 ng/ml; IL-4, 5 ng/ml), activators (histamine, 10 |jM;
SAg, 20 ng/ml or LPS, 2 ng/ml), growth factors (TGF-(3, 5 ng/ml; EGF, bFGF, and
PDGF-BB, 10 ng/ml), PBMC (7.5x104 cells/well) or macrophages (3.5x104 cells/well). All
primary human cells utilized in this work were obtained via commercially available
sources.
Compounds
Compounds were tested at 40, 13.3, 4.4 and 1.48 uM for the study, in a single
well per readout parameter. Compounds were prepared in DMSO from 20 mM stock
solutions, added 1 hr before stimulation of the cells, and were present during the whole
24 hr stimulation period. Final DMSO concentration was 0.2%. Colchicine, 3.3 uM, was
included as a positive control. Compounds were tested in a blinded fashion and included
three sets of triplicate samples and five sets of duplicates for quality control purposes.
Plate formats. Templates were prepared with seven compounds (four
concentrations) per 96-well plate. One positive control (colchicine) and eight negative
control wells (0.2% DMSO) were employed on each plate. Left and rightmost rows (A1-
H1, A12-H12) were not employed for EPA compounds.
ELISA
The levels of readout parameters were measured by ELISA as described.9
Briefly, microtiter plates are treated, blocked, and then incubated with primary antibodies
or isotype control antibodies (0.01-0.5 ug/ml) for 1 hr. After washing, plates were
incubated with a peroxidase-conjugated anti-mouse IgG secondary antibody or a biotin-
conjugated anti-mouse IgG antibody for 1 hr followed by streptavidin-HRP for 30 min.
Plates were washed and developed with TMB substrate and the absorbance (OD) was
read at 450 nm (subtracting the background absorbance at 650 nm). Quantitation of
TNF-a and PGE2 in the LPS system was done using commercially available kits
according to the manufacturer's directions. Proliferation of PBMC (T cells) was
quantified by Alamar blue reduction and proliferation of adherent cell types was
quantified by SRB staining.10
03JUNE2009 Version 5/27 Revised/Submitted
Previous I TOC
-------
Bioactivity Profiling Using BioMAP Cell Systems Houck et al
Other Assessments
Overtly adverse effects of compounds on cells were determined by 1) measuring
alterations in total protein (SRB assay), 2) measuring the viability of peripheral blood
mononuclear cells; and 3) microscopic visualization. SRB was performed by staining
cells with 0.1% sulforhodamine B after fixation with 10% TCA, and reading wells at 560
nm. PBMC viability was assessed by adding Alamar blue to PBMC that had been
cultured for 24 hours in the presence of activators and compounds and measuring its
reduction after 8 hr. Samples were assessed visually according to the following scheme:
2.0=cobblestone (unactivated phenotype); 1.0=activated (normal phenotype); 0.5=lacy
or sparse; 0.375=rounded; 0.25=sparse and granular; 0.1=no cells in well. During this
procedure, cells are also assessed for the presence of compound precipitates, and
samples were flagged if precipitates are observed.
Data analysis
Measurement values for each parameter in a treated sample were divided by the
mean value from six DMSO control samples (from the same plate) to generate a ratio.
All ratios were then log 10 transformed. Visual categorical scores (see above) were
similarly converted (Iog10 ratios of 0.3, 0.0, -0.3, -0.4, -0.6, and -1.0). Significance
prediction envelopes were calculated for historical controls (99% and 95%). Hit
prediction envelopes (99% and 95%) were also calculated from historical controls. Hit
prediction envelopes differ from significance envelopes in that they are calculated for the
entire profile, not for individual readout parameters. Thus a 95% hit envelope will contain
95% of control profiles, and therefore will depend on the specific systems and readouts
selected for analysis.8'31 Overtly cytotoxic compounds are identified as generating
profiles with one or more of the following readouts below the indicated thresholds: SRB
< -0.3, PI or PBMC cytotoxicity <-0.1 or Visual score <-0.6 in one or more systems. The
complete set of results for the 320 chemicals for each of the 87 endpoints can be
accessed at: http://www.epa.gov/ncct/toxcast/.
For analysis of profile similarities, overtly cytotoxic compound profiles were
removed. The correlation metric was a combination of similarity metrics in addition to
Pearson's correlation (J. Yang, E. Berg, personal communication). This approach was
03JUNE2009 Version 6/27 Revised/Submitted
Previous I TOC
-------
Bioactivity Profiling Using BioMAP Cell Systems Houck et al
found to improve the accuracy of mechanism classification with test data sets (not
shown), due to the diversity of BioMAP profile characteristics (wide variation in the
number of active readouts, the number of active systems, and in the amplitude of
biomarker readout changes). Thus, the similarity metrics used for the analyses of
profiles included Pearson's correlation, a real value Tanimoto metric (=A-B/(||A||+||B|| -
A-B, where A and B are the two profile vectors), and a system weighted-averaged real
value Tanimoto metric ( = £ (i=system) Wi*Ti/£Wi, where Ti is the real value Tanimoto
score for the ith system, Wi is the weight for the ith system, Wi = number of markers in
the system/(1 + exp(-(max.ratio of the two profiles in this system-0.09)*100) ). The real
value Tanimoto metric was employed as a scaled version for filtering profile similarities.
The scaled version is calculated by normalizing each profile to the unit vector (eg. A =
A/||A||) first, then applying the formula given above. Similar profiles were identified as
those having Pearson correlations > 0.7, and tanimoto scores > 0.5 (or as otherwise
indicated). Thresholds for these metrics were set using specific sets of training and test
sets of profile data (not shown). The function similarity map uses the results of pairwise
correlation analysis to project the "proximity" of related profiles from multi-dimensional
space to two dimensions. The two dimensional projection coordinates were generated
by applying a modified nonlinear mapping technique, using a modified stress function by
Clark. A gradient descent minimization method was used to minimize the modified stress
function, starting from a set of initial positions (e.g. from principal components analysis).
Distances between compounds are representative of their similarities and lines are
drawn between compounds whose profiles are sufficiently similar, with metrics that are
above all selected thresholds (passing all filters).
RESULTS
Overview of ToxCast compound activities in BioMAP systems.
320 ToxCast proof-of-concept chemicals (list available at:
http://www.epa.gov/ncct/toxcast/) were tested in eight BioMAP model systems (Table I)
at four concentrations, from 1.3 - 40 |o,M. In each assay system, 7-14 biomarker
readouts were measured for a total of 87 readout measurements per compound per
concentration tested. Active compounds were identified as those compounds for which
03JUNE2009 Version 7/27 Revised/Submitted
Previous I TOC
-------
Bioactivity Profiling Using BioMAP Cell Systems Houck et al
biomarker readout changes at one or more concentrations resulted in the overall
compound profile appearing outside of a 95% hit prediction envelope (developed from
replicates of negative control sample replicates, see Materials and Methods). Of the 320
compounds tested, 219 (68%) were active, based on these criteria. The complete
BioMAP data set for the ToxCast chemicals can be accessed through the internet at:
http://www.epa.gov/ncct/toxcast/.
Cytotoxicity of ToxCast compounds in BioMAP Systems.
Each BioMAP system (Table I) contains one or more assay endpoints that
correlate with overt cell cytotoxicity. Such cytotoxicity may confound interpretation of
mechanistic information by causing changes in biomarker levels through effects
secondary to cell death. Thus concentrations of chemicals inducing significant
cytotoxicity were excluded from further analysis. These cytotoxicity endpoints include
sulforhodamine B (SRB) staining for total protein, alamar blue assessment of peripheral
blood mononuclear cell metabolic activity (PBMC Cytotox.), and a morphologic score
(Vis). The Vis score classifies cell shape as defined in Materials and Methods.
Cytotoxicity of compounds can depend on cell type as well as the cellular environment.
Compound exposures that resulted in a >50% reduction in total protein levels (Log
ratiolO of SRB < 0.3) were considered overtly cytotoxic, and by this criteria, 70 ToxCast
compounds were cytotoxic to at least one cell type at one or more concentrations.
Figure 1 shows the distribution of cytotoxicities (indicated by red) by cell type (system)
and concentration. Cell-type selective cytotoxicities were observed with a number of
compounds; for example metiram-zinc was cytotoxic to PBMC and fibroblast-containing
systems, but not endothelial cells, and perfluorooctane sulfonic acid was selectively
cytotoxic to epithelial cells. Of the compounds exhibiting overt cytotoxicity, these were
most frequently cytotoxic to endothelial cells (3C system) and least frequently cytotoxic
to smooth muscle cells (SM3C). Although these two systems contain different primary
cell types (endothelial cells versus smooth muscle cells), they are stimulated with the
same combination of stimuli (IL-1p, TNFa and IFNy), suggesting a cell type-specific
difference in toxicity mechanisms. As shown in Fig. 1, many of the cytotoxicities
observed show a sharp concentration-response relationship. In addition, a few
compounds show a reverse concentration effect; they are less cytotoxic at higher
concentrations. Compounds showing this effect include fentin, (2-
03JUNE2009 Version 8/27 Revised/Submitted
Previous I TOC
-------
Bioactivity Profiling Using BioMAP Cell Systems Houck et al
benzothiazolylthio)methyl thiocyanate (TCMTB), captafol, and captan. This reverse
concentration effect can be attributed to poor solubility at higher concentrations, or
possibly biological effects such as induction of a protective response like the phase II
enzyme glutathione S-transferase. Solubility was assessed visually on all compounds
and noted precipitations included captafol on initial cell treatment and captan following
24-hr incubation.
Reproducibility of BioMAP profiles.
In order to assess the reproducibility of compound activities in these assays, a
positive control, colchicine (a tubulin-binding mitotic poison), was included on every
assay plate. Figure 2 shows the overlay of the replicate control data in a profile plot.
While the amplitude of biomarker changes for each readout shows variability between
replicates (i.e. from plate to plate), the overall profile shape is very consistent and can be
used as a measure of reproducibility. Among replicate samples of colchicine, the
Pearson correlation coefficients between any two replicates was high, and ranged from
0.82 - 0.97. Included in the ToxCast chemical library were three chemicals each present
as three independent samples. Correlation coefficients for the 40 |o,M testing
concentration between any two replicates had ranges of 0.77-0.89 for bensulide, 0.61-
0.80 for diclofop-methyl, and 0.20-0.43 for prosulfuron. Note that prosulfuron had very
little significant activity and was thus based on data that largely did not fall outside the
95% significance envelope defined by the solvent controls. Across all assays, average
coefficient of variation of the fold-change for the triplicates was 7.2%.
Correlation analysis of ToxCast compound profiles.
The BioMAP profiles of ToxCast compounds were compared to one another for
similarity by pairwise correlation analysis (see Materials and Methods). Figure 3 shows
a visualization of these relationships in a function homology map generated by a
comparison of the resulting correlations and display of the correlations by non-linear
projection in two dimensions (Sammon mapping, a method of multidimensional scaling
that preserves distance and topology of data).8 Compound similarities that are significant
(above a selected threshold) are shown as connected lines. Compound profiles
displaying overt cytotoxicity (Fig. 1) were excluded from the map shown.
03JUNE2009 Version 9/27 Revised/Submitted
Previous I TOC
-------
Bioactivity Profiling Using BioMAP Cell Systems Houck et al
Analysis of selected ToxCast compounds.
The pair-wise correlation analysis suggested that ToxCast compounds could be
classified into functional groups (clusters in Fig. 3) by BioMAP profiling. Compounds
were then further analyzed by comparison to reference BioMAP profiles to evaluate
similarity of ToxCast compounds to agents with known mechanisms of action. Examples
of similarities and mechanism classes that were discovered by this analysis are listed in
Table II.
Mitochondria! dysfunction.
Compounds with potential to induce mitochondrial dysfunction were identified by
their similarity to reference compounds. Reference compounds included oligomycin A,
an inhibitor of mitochondrial ATPase; shaoguamycin (complex I inhibitor); myxathiazol
(complex II inhibitor); and antimycin A (complex III inhibitor). Oligomycin A was tested in
the study along with ToxCast compounds, while the profiles for other reference
compounds were from a reference database, as described.11 Figure 4 shows an overlay
of the BioMAP profiles for the readouts measured in common for two ToxCast
compounds representative of this class, pyraclostrobin and trifloxystrobin, along with the
profile of the reference, myxathiazol. Pyraclostrobin and trifloxystrobin are strobilurin
fungicides known to inhibit mitochondrial function as a mode of their fungicidal action as
do a number of other compounds listed in this class (Table II).12 Additional compounds
with profiles that significantly match this class include the conazole fungicides and others
that inhibit sterol biosynthesis as a mode of action suggesting a connection between
sterol biosynthesis and mitochondrial function. Sterol biosynthesis inhibitors in this
mechanism class include prochloraz, triadimenol, difenoconazole, fenarimol, and
flusilazole. A possible mechanism for this activity may be seen in Saccharomyces
cervisiae where many of the genes that are required for efficient uptake and/or transport
of sterols are also required for mitochondrial functions.13 Reduced mitochondrial
membrane sterol content affects the adenine nucleotide transporter in the inner
mitochondrial membrane, leading to decreased intra-mitochondrial ATP levels.14
Vertebrate mitochondria contain P450 enzymes involved in sterol metabolism and may
suffer the same fate induced by interference with sterol biosynthesis.
03JUNE2009 Version 10/27 Revised/Submitted
Previous I TOC
-------
Bioactivity Profiling Using BioMAP Cell Systems Houck et al
Compounds with potential to induce endoplasmic reticulum stress.
Compounds with the potential to induce endoplasmic reticulum stress were
identified by their similarity to a group of compounds (A23187, thapsigargin,
chlorambucil, sodium azide, cycloheximide and rotenone), previously identified as
having this potential.9 These compounds are known to interact with diverse targets, for
example, chlorambucil is a nitrogen mustard alkylating agent that is used in
chemotherapy, and covalently modifies many cellular targets in addition to DMA; A23187
is a calcium ionophore; thapsigargan inhibits the SERC ATPase; cycloheximide inhibits
acyl transferase II and protein synthesis; sodium azide inhibits mitochondrial Complex IV
and rotenone inhibits Complex I . The ability of these compounds to induce endoplasmic
stress may be through shared secondary targets or common downstream mechanisms
involving the generation of reactive species.11 Several of the chemicals in this group are
inhibitors of mitochondrial dysfunction, such as propargite and rotenone, or sodium
channel modulators, such as the pyrethroids resmethrin and tefluthrin. Interestingly, the
pyrethroids, like rotenone, have been associated with effects on dopaminergic nerve
pathways and could play a role in neurological disorders, including Parkinson's
Disease.15"16 As shown in Fig. 5A, the compounds in this class are somewhat diverse
and many show strong concentration-dependent differences in their profiles, consistent
with compounds affecting multiple targets. Many of the compounds in this class become
overtly cytotoxic to cells at higher concentrations.
A/F/cB inhibitors.
Compounds with the ability to inhibit the NFicB pathway were identified by
similarity to reference NFicB inhibitors including Ro106-9920 (kBa ubiquitination
inhibitor), BAY 11-7085 (kBa phosphorylation inhibitor), SC-514 and IKK-2 inhibitor IV
(IKK-2 inhibitors), or dimethyl fumarate.11 As previously described, many chemicals in
this group act through covalent modification of NFicB pathway components resulting in
inactivation of the NFkB transcription factor.8'17"18 The dithiocarbamate pesticides such
as dazomet, maneb and metam-sodium appear to have such activity. Dithiocarbamates
have also been described as potent inhibitors of NFKB activation in cellular assays.19
Fig. 5B shows a BioMAP overlay of dazomet with dimethyl fumarate.
03JUNE2009 Version 11/27 Revised/Submitted
Previous I TOC
-------
Bioactivity Profiling Using BioMAP Cell Systems Houck et al
Elevators ofcAMP.
ToxCast compounds that up-regulate cAMP levels were identified by similarity to cAMP-
elevating reference compounds, including the phosphodiesterase IV inhibitors ICI-63197
and rolipram. Fig. 5C shows an overlay of the BioMAP profiles of azatrine, cyanazine
and simazine, herbicides of the chlorotriazine class with highly related structures. One
notable feature of these profiles is the strong inhibition of TNFa in the BioMAP LPS
system, and inhibition of PAI-1 levels in the HDF3CGF system, hallmark activities of
phosphodiesterase inhibitors that differentiate them from other mechanisms. Atrazine
has been demonstrated to inhibit phosphodiesterases and to activate a cyclic-AMP
response element reporter gene.20"21 Elevated cAMP can induce aromatase expression
causing increased androgen conversion to estrogens in a wide range of tissues.22 In
addition, atrazine was shown to increase activity for an aromatase reporter gene.21
These activities could explain, in part, the reported endocrine disrupter activity of
atrazine resulting from amplification of cAMP; possibly an increase in the cAMP-
responsive cellular kinase SGK-1; and nuclear receptor NR5A (SF-1) phosphorylation, a
regulator of aromatase gene expression.21
Microtubule function and estrogen receptor signaling.
We have previously described similarity between BioMAP profiles of the estrogen
receptor agonist, 17beta-estradiol, its metabolite, 2-methoxyestradiol and microtubule
destabilizers.8 Compounds with similarity to this class were identified by their similarity to
colchicine, vinblastine (microtubule destabilizers) 17beta-estradiol, and/or 2-
methoxyestradiol. Fig. 5D shows an overlay of benomyl and fludioxonil with paclitaxel.
Benomyl, an antifungal compound, inhibits fungal cell mitosis by binding to microtubules
and deforming their structure. It also has been demonstrated to disrupt mammalian cell
microtubules and inhibit proliferation and mitosis in HeLa cells with IC50 of around 5
micromolar, less potent than its activity in fungal cells.23 This may be responsible for the
teratogenic activity of benomyl.24 The phenylpyrrole fungicide fludioxonil is a natural
product analog with an unknown mode of action.25 However, fludioxonil has been shown
to be a clastogen and inducer of polyploidy in CHO cells, activities that are consistent
with effects on microtubule function.26 Teratogenic activity of fludioxonil has not been
demonstrated, although whether the parent molecule reaches the developing fetus
during developmental toxicity testing is not known.
03JUNE2009 Version 12/27 Revised/Submitted
Previous I TOC
-------
Bioactivity Profiling Using BioMAP Cell Systems Houck et al
DISCUSSION
The goal of the ToxCast program is to develop algorithms that predict potential
for chemical toxicity using data generated by in vitro bioactivity profiling assays
combined with physicochemical properties. Given this ambitious goal of developing a
comprehensive approach that will cover a large number of toxicity mechanisms, the
profiling assays should encompass a wide range of potential toxicity targets. In part, this
requirement is satisfied by high-throughput, biochemical screening assays against a
large number of specific molecular targets.1 However, having assays for all potential
individual toxicity targets is not practical. First, knowledge of specific molecular targets of
toxicity is incomplete and the ability to assay all proteins for effects of chemical
perturbation is not presently possible. Second, targets of chemical toxicity may be a
function of emergent properties of more complex systems such as the cell or intact
organism that are not present at the biochemical level.27 For these reasons, we have
included complex cellular assay systems as part of the ToxCast program. The BioSeek
cell systems models not only provide such cellular assays, but were conducted in
primary human cells and thus may provide a more direct link to human health effects
than would transformed cell lines typically used in cell-based screening.28
Eight BioMAP systems were used in the present study yielding 7-14 readouts per
system for a total of 87 readouts per compound. The different readouts were not
selected for known relevance to toxicity. The selection of readouts and design of the
assay systems was directed towards optimal detection and discrimination of diverse
target and pathway mechanisms.4"6 We did not assume a priori that individual readouts
were related to toxicity, but rather our hypothesis was that if the patterns of readouts in
these assays can be correlated with diverse mechanisms, some of these may be
relevant to toxicity. Despite the large amount of data generated, only a relatively narrow
breadth of specific pathophysiology was covered, chiefly vascular and inflammatory
biology. However, the ubiquitous nature of critical cell signaling pathways makes it
possible that a relatively higher percentage of important toxicological targets are present
in these systems. Indeed, past experience with the BioMAP systems demonstrated the
ability to discern signatures of activity for disease modulators not directly related to the
03JUNE2009 Version 13/27 Revised/Submitted
Previous I TOC
-------
Bioactivity Profiling Using BioMAP Cell Systems Houck et al
biology of the specific BioMAP system, for example detection of oncology and
cardiovascular drug activities in systems designed for inflammatory mechanisms.4 The
key to success under this assumption is a reference library of bioactivity profiles
covering important mechanisms of toxicology both to validate the assays and to
distinguish mechanisms of action of unknown chemicals. Correlation of bioactivity
profiles of a library of pharmacological probes and drugs with signatures generated by
screening the ToxCast chemical library provided evidence for mechanisms of potential
toxicity for a subset of the library.
Using statistical correlation methods to classify ToxCast chemicals with respect
to potential mechanisms of bioactivity revealed at least five reasonable matches with
profiles in the BioSeek database of known compounds. These mechanisms: NFicB
inhibitors, cAMP elevation, inducers of mitochondrial dysfunction, endoplasmic reticulum
stress and microtubule inhibition, were associated with 76 of the 309 unique chemical
structures in the library. The types of correlated activities were sometimes related to the
intended mechanism of action of the chemical, e.g. pesticides that act through inhibition
of mitochondrial respiratory chain function in insects or that destabilize microtubule
function in fungus. In other cases, these appear not to be related to intended mechanism
of action, e.g. inhibition of NFicB activation, cAMP elevation or TNF-a secretion. Such
effects may or may not reflect toxicity mechanisms. However, they do shed light on
mechanisms of bioactivity and potential molecular targets which may prove useful in
evaluating the potential toxicity of new chemicals. As we gain greater understanding of
the importance of the many different signaling pathways, we will better be able to
describe what constitutes a "toxicity pathway" and assess the significance of a
chemical's perturbation of it. An important component of this effort is the profiling of
compounds of known pharmacological and toxicological effects to measure their ability
to modulate these pathways. This will provide validation of the toxicity link of the
pathway as well as reference profiles to use for assessing the effects of new chemicals.
It was somewhat surprising that for the most part, chemicals in the ToxCast
library generated relatively weak signatures relative to the profiles of the reference
pharmacological probes and drugs. All of the compounds in the ToxCast library,
consisting largely of pesticide active ingredients, are bioactive and include many with
significant toxicities when tested at high doses in animals.2 Since bioactivity is a function
03JUNE2009 Version 14/27 Revised/Submitted
Previous I TOC
-------
Bioactivity Profiling Using BioMAP Cell Systems Houck et al
of dose, it is important to consider the in vitro screening concentration relative to the
high-dose, in vivo animal toxicology. It is not clear whether the concentrations used in
this profiling effort, i.e. 40 |o,M top concentration, is a reasonable representation of the
effective tissue concentrations associated with lowest effect levels in the in vivo toxicity
testing.2 Toxicity testing in rodents is generally done up to the maximally tolerated dose,
and without tissue dosimetry for comparison to concentration response in cell-based, in
vitro assays, interpretation remains a challenge. As part of their development, many of
these pesticides were designed to be selectively toxic towards pest species and the
weak signatures detected in the BioMAP profiles may reflect this built in safety margin.
Not all compounds produced statistically significant profiles that closely matched
reference compounds in the BioSeek database. There are several possible explanations
for this. Compound signatures may reflect poly-pharmacology, i.e. activity against
multiple pathways, resulting in confounding signatures of activity that change with testing
concentration. Many of the probes or drugs used to generate the reference bioactivity
profiles are very selective for specific pathways as an inherent function of their utility as
a pharmacological probe or drug. Activities against targets other than the intended may
only occur at much higher concentrations due to medicinal chemistry efforts and
selective screening. The environmental chemicals, on the other hand, may never have
been subject to such engineering and thus could display multiple overlapping profiles at
the concentrations tested without one predominant behavior. Alternatively, there may be
no reference compounds in the database for accurate comparison. For these cases, use
of knowledge about the ToxCast chemicals may provide useful information for future
screening using the BioMAP systems. Extensive in vivo toxicity data for the compounds
are available in a relational database2 and a variety of research data exists in the
scientific literature describing modes of action and toxicities for many of the compounds.
Further analysis and predictive modeling of the BioMAP profiles of the Toxcast
compounds may yield classes of chemicals that can be correlated with toxicity endpoints
or mechanisms. This would serve to guide interpretation of future testing of chemicals
with little or no toxicity information available.
In addition to the methods of analysis described here, individual endpoints will be
included in a more comprehensive collection of data that make up the EPA's ToxCast
project. Data from 8-10 in vitro assays sources (both cell-free and cell-based), plus
03JUNE2009 Version 15/27 Revised/Submitted
Previous I TOC
-------
Bioactivity Profiling Using BioMAP Cell Systems Houck et al
physicochemical properties and chemical structure information is being use to develop
predictive "signatures" of in vivo toxicity as captured in ToxRefDB. A signature in this
context is a rule or algorithm against which the in vitro data for a chemical is tested. If
the chemical matches the signature, then it is predicted to be toxic for the endpoint being
evaluated. We are using a variety of statistical and machine learning methods to mine
the ToxCast data.29 The primary input from the BioMAP data described here are Lowest
Effective Concentrations (LEG) which are the lowest tested concentrations at which a
given chemical shows a BioMAP response that is statistically significantly different from
background. For each BioMAP system, we exclude components where the LEG is
greater than or equal to the LEG at which cytotoxicity is seen for the chemical in the
particular BioMAP system.
The future of toxicity testing requires the ability to determine the effects of
chemicals on important cellular toxicity pathways in human cells and interpret the
significance of any perturbation.30 Bioactivity profiling of compounds in the primary,
human BioMAP cell systems facilitates this for many pathways involved in vascular and
inflammatory biology. Coupled with a reference database of many compounds with
known toxicities, this approach provides a means to interpret bioactivity in the context of
effects on human or rodent pathophysiology. Much work remains to be done in terms of
expanding the diversity of toxicological probes for the reference database, expanding the
pathophysiology covered by the complex cellular assays, and relating in vitro screening
concentrations to actual human exposure levels. In addition, environmental toxicology
typically involves effects resulting from exposure to complex mixtures of chemicals.
While not addressed in this study, the pathway-based approach may allow the ability to
discern unexpected effects of exposure to mixtures that would only be apparent in
complex cell biology models. Overall, this approach represents an important first step in
the ability to use primary human cells in vitro as model systems for evaluating
environmental chemicals for risks to human health.
03JUNE2009 Version 16/27 Revised/Submitted
Previous I TOC
-------
Bioactivity Profiling Using BioMAP Cell Systems
Houck et al
ACKNOWLEDGEMENTS
The data presented in this paper was generated under EPA contract EP-W-07-
039 with BioSeek Inc., as part of the ToxCast research program. The authors thank
John Nanartowicz and John Southerland for the procurement and management of this
contract.
03JUNE2009 Version
Previous
17/27
TOC
Revised/Submitted
-------
Bioactivity Profiling Using BioMAP Cell Systems Houck et al
REFERENCES
1. Dix, DJ, Houck, KA, Martin, MT, Richard, AM, Setzer, W, and Kavlock, RJ: The
ToxCast program for prioritizing toxicity testing of environmental chemicals.
Toxicol Sci 2007;95:5-12.
2. Martin M, Judson RS, Reif DM, Kavlock RJ and Dix DJ: Profiling Chemicals
Based on Chronic Toxicity Results from the U.S. EPA ToxRef Database. Environ
Health Perspect, in press.
3. Judson R, Richard A, Dix DJ, Houck K, Martin M, Kavlock R, Dellarco V, Henry
T, Holderman T, Sayre P, Tan S, Carpenter T, Smith E: The Toxicity Data
Landscape for Environmental Chemicals. Environ Health Perspect, in press.
4. Kunkel, EJ, Plavec, I, Nguyen, D, Melrose, J, Rosier, ES, Kao, LT, Wang, Y,
Hytopoulos, E., Bishop, AC, Bateman, R, Shokat, KM, Butcher, EC, & Berg, EL:
Rapid Structure-Activity and Selectivity Analysis of Kinase Inhibitors by BioMAP
analysis in Complex Human Primary Cell-Based Models. /Assay Drug Dev
7ec/?no/2004a;2:431-41.
5. Kunkel, EJ, Dea, M, Ebens, A, Hytopoulos, E, Melrose, J, Nguyen, D, Ota, KS,
Plavec, I, Wang, Y, Watson, SR, Butcher, EC, Berg, EL: An Integrative Biology
Approach for Analysis of Drug Action in Models of Human Vascular Inflammation.
FASEB Journal 2004b; 18:1279-81.
6. Berg EL, Kunkel EJ, Hytopoulos E, Plavec I: Characterization of compound
mechanisms and secondary activities by BioMAP analysis. J Pharmacol Toxicol
Methods 2006; 53:67-74.
7. Berg EL, Kunkel EJ, Hytopoulos E: Biological complexity and drug discovery: a
practical systems biology approach. Syst Biol (Stevenage) 2005;152:201-6.
8. Plavec, I, Sirenko, O, Privat, S, Wang, Y, Dajee, M, Melrose, J, Nakao, B,
Hytopoulos, E, Berg, EL, & Butcher, EC: Method for Analyzing Signaling
Networks in Complex Cellular Systems. Proc Natl Acad Sci USA
2004;101:1223-28.
9. Melrose J, Tsurushita N, Liu G, Berg EL: IFN-gamma inhibits activation-induced
expression of E- and P-selectin on endothelial cells. J Immunol 1998;161;2457-
2464.
03JUNE2009 Version 18/27 Revised/Submitted
Previous I TOC
-------
Bioactivity Profiling Using BioMAP Cell Systems Houck et al
10. Ahmed SA, Gogal, RM Jr, Walsh, JE: A new rapid and simple non-radioactive
assay to monitor and determine the proliferation of lymphocytes: an alternative to
[3H]thymidine incorporation assay. J Immunol Methods 1994;170:211-24.
11. Berg EL et al., in preparation.
12. Bartlett DW, Clough JM, Godwin JR, Hall AA, Hamer M, Parr-Dobrzanski B: The
strobilurin fungicides. Pest Manag Sci 2002;58:649-62.
13. Reiner S, Micolod D, Zellnig G, Schneiter R: A genomewide screen reveals a
role of mitochondria in anaerobic uptake of sterols in yeast. Mol Biol Cell
2006;17:90-103.
14. Astin AM and Haslam JM: The effects of altered membrane sterol composition on
oxidative phosphorylation in a haem mutant of Saccharomyces cerevisiae.
Biochem J 1977; 166:287-98.
15. Nasuti C, Gabbianelli R, Falcioni ML, Di Stefano A, Sozio P, Cantalamessa F:
Dopaminergic system modulation, behavioral changes, and oxidative stress after
neonatal administration of pyrethroids. Toxicology 2007; 229:194-205.
16. Ryu EJ, Harding HP, Angelastro JM, Vitolo OV, Ron D, Greene LA:
Endoplasmic reticulum stress and the unfolded protein response in cellular
models of Parkinson's disease. J Neurosci. 2002;22:10690-8.
17. Toledano MB and Leonard WJ: Modulation of transcription factor NF-kappa B
binding activity by oxidation-reduction in vitro. Proc Natl Acad Sci USA
1991;88: 4328-32.
18. Chen, L, Fischle, W, Verdin, E, Greene, WC: Duration of nuclear NF-kappaB
action regulated by reversible acetylation. Science 2001 ;293:1653-7.
19. Schreck R, Meier B, Ma'nnel DN, Droge W, Baeuerle PA: Dithiocarbamates as
potent inhibitors of nuclear factor kappa B activation in intact cells. J Exp Med
1992; 175:1181-94.
20. Roberge M, Hakk H, Larsen G: Atrazine is a competitive inhibitor of
phosphodiesterase but does not affect the estrogen receptor. Toxicol Lett
2004;154:61-68.
21.Suzawa M, Ingraham HA: The herbicide atrazine activates endocrine gene
networks via non-steroidal NR5A nuclear receptors in fish and mammalian cells.
PLoSOA/E2008;3:e2117.
22. Fan W, Yanase T, Morinaga H, Gondo S, Okabe T, Nomura M, Komatsu T,
Morohashi K, Hayes TB, Takayanagi R, Nawata H: Atrazine-induced aromatase
03JUNE2009 Version 19/27 Revised/Submitted
Previous I TOC
-------
Bioactivity Profiling Using BioMAP Cell Systems Houck et al
expression is SF-1 dependent: implications for endocrine disruption in wildlife
and reproductive cancers in humans. Environ Health Perspect 2007; 115:720-7.
23. Gupta K, Bishop J, Peck A, Brown J, Wilson L, and Panda D: Antimitotic
Antifungal Compound Benomyl Inhibits Brain Microtubule Polymerization and
Dynamics and Cancer Cell Proliferation at Mitosis, by Binding to a Novel Site in
Tubulin. Biochemistry 2004; 43: 6645 -6655.
24. Kavlock, RJ, Chernoff, N, Gray, Jr, LE , Gray, JA, Whitehouse, D: Teratogenic
effects of benomyl in the Wistar rat and CD-1 mouse, with emphasis on the route
of administration. Toxicol. Appl. Pharmacol 1982;62:44-54.
25. Rosslenbroich H-J, Stuebler D: Botrytis cinerea—history of chemical control and
novel fungicides for its management. Crop Protection 2000; 19:557-561.
26. U.S. Environmental Protection Agency: Fludioxonil; Pesticide Tolerance. Fed
Regist 1998;63:53820-53826.
27. Bhalla US,, lyengar R: Emergent Properties of Networks of Biological Signaling
Pathways. Science 1999;283: 381 -387.
28. Horrocks C, Halse R, Suzuki R, Shepherd PR: Human cell systems for drug
discovery. Curr Opin Drug Discov Devel 2003;6:570-5.
29. Judson, R, Elloumi, F, Setzer, RW, Li, Z, Shah, I: A Comparison of Machine
Learning Algorithms for Chemical Toxicity Classification Using a Simulated Multi-
Scale Data Mode. BMC Bioinformatics 2008 ;9:241.
30. National Research Council: Toxicity Testing in the 21st Century: A Vision and a
Strategy Washington D.C.: National Academies Press 2007.
31. Storey, JD, & Tibshirani, R: Statistical significance for genomewide studies. Proc
NatlAcad Sci U SA 2003; 100:9440-5.
03JUNE2009 Version 20/27 Revised/Submitted
Previous I TOC
-------
Bioactivity Profiling Using BioMAP Cell Systems
Houck et al
FIGURES AND TABLES
Table I. Eight BioMAP Systems utilized in this study. BioMAP Systems listed according to their
short names are comprised of the cell types shown cultured and stimulated with the
environmental factors (added along with test compounds) for 24 hours. For each System, the
biomarker readouts listed (number of readouts is shown in parentheses) are measured at 24 or
72 hours as described in Materials and Methods.
System
3C
~. *
LPS ^
SAO.
BE3C
HDF3CGF
KF3CT
SM3C
Cell Types
Endothelial cells
Endothelial cells
Peripheral Blood
Mononuclear Cells +
Endothelial cells
Peripheral Blood
Mononuclear Cells +
Endothelial cells
Bronchial epithelial
cells
Fibroblasts
Keratinocytes +
Fibroblasts
\fescularsmooth
muscle cells
Environment
IL-1|!+TNF-a+IFN-v
IL-4+histamine
TLR4
TCR
IL-lp+TNF-a+IFN-v
IL-lfS+TNF-a-HFN-v
+bFGF+EGF+PDGF-BB
IL-1(}+TNF-a+IFN-v
•tTGF-il
IL-l|J+TNF-a+IFN-v
Readouts
MCP-1, VCAM-'i, ICAM-i.Throribomodulin, Tissue
Factor, E-selectin, uPAR. IL-6, MIG. HLA-DR, Prolif.
Vis., SRB (J3)
VEGFRII, P-selectin, VCAM-' , uPAR, Eotaxin-3, MCP-
1,SRB(7)
CD40. VCAM-1, Tissue Factor, MCP-', E-selectin, IL-
1a, IL-6. M-CSF, TNF-a, PGE2. SRB (11)
MCP--.. CD3B. CD40, CD69, E-selectin, IL-G. MIG,
PBMC Cytotox, SRB, Proliferation (10)
uPAR, IP-10, MIG, HLA-DR. IL-J a, MMPA PAH,
SRB. TGF-bl,tPAuPA(11)
VCAM-1, IP-10, IL-8, MIG, Collagen III, M-CSF, MMP-1,
PAI-1, Proliferation, TIMP-1, EGFR, SRB ('2>
MCP-1, ICAM-', IP--0, IL-'a, MMP-9, TGF-bl, TIMP-2,
uPA SRB (9)
MCP-1, VCAM-1, Thronbomodulin, Tissue Factor, IL-
6, LDLR, SAA uPAR, IL-6. MIG, HLA-DR, M-CSF.
Prolif., SRB (14}
03JUNE2009 Version
21/27
Revised/Submitted
Previous
TOC
-------
Bioactivity Profiling Using BioMAP Cell Systems
Houck et al
Table II. Table of ToxCast compounds with assigned mechanisms identified by similarity of
BioMAP profiles to reference compounds (see Results). Reference compounds that were tested
in the study along with ToxCast compounds are shown in italics.
(Z, E )-F en py rox i mate
Azoxystrobin
Bromoxynil
Butralin
Cyazofamid
D-cis,trans-Allethrin
Difenoconazole
Difenzoquat Metilsulfate
Dimethomorph
Ethofumesate
Famoxadone
Fenamidone
Mitochondria! Dysfunction
Fenarimol
Fenitrothion
Fenoxycarb
Fipronil
Fluoxastrobin
Flusilazole
tndoxacarb
Methoxychlor
Methyl (sothiocyanate
MGK
Norflurazon
Novaluron
Oligomycin A
Oxadiazon
Oxyfluorfen
Paclobutrazol
Prochloraz
Propargite
Pyraclostrobin
Pyridaben
Tebufenpyrad
Thiazopyr
Triadimenol
Trifloxystrobin
17p-Estradiot
3-tedo-2-P ropy ny I bu ty Icarbamate
Abamectin
Benomyl
Chlorpyrifes Oxon
Microtubule Inhibitors
Cyprodinil
Dinicortazole
Emamectln Benzoate
Endosulfan
Fludioxonil
Hexythiazox
Niclosamide
Parathion-Methyl
Prodiamine
Pyridaben (high dose)
ER Stress Inducers
A2318?
Chlorambucil
Cycloheximicle
Dithiopyr
Flumiciorac-Pentyl
Propargite
Resmethrin
Rotenone
Tefluthrin
Zoxamide
3-lodo-2-P ropy ny I buty Icarbamate
Acetochlor
Bendiocarb
Benfluralin
Captan
Dazomet
Dimethyl fuma/afe
Formetanate Hydrochloride
IKK-2 Inhibitor IV
Maneb
Metam-Sodium Hydrate
cAMP Elevation
Atrazine
Cyanazine
Propazine
Simazine
03JUNE2009 Version
22/27
Revised/Submitted
Previous
TOC
-------
Bioactivity Profiling Using BioMAP Cell Systems
Houck et al
I
a
o
Fig. 1. Distribution of cytotoxicities observed for 70 ToxCast compounds. Cytotoxicity
is indicated by SRB measurements or Alamar blue (SAg PBMC). The threshold for
cytotoxicity (High) is indicated by measurements of <-0.3 (Red). Orange and yellow
indicate measurements of-0.2 to -0.3 (Mod) and -0.1 to -0.2 (Low), respectively. No
significant effects on SRB measurements is indicated by white (None). Table columns
are organized by system and concentration (highest concentration on the left in each
system).
System
Concentration
,,
BE3C HDF3CGF KF3CT
IPS
(PBMC) 3C
4H
SM3C
'EL
»:.•• H :•- :
• . . f • . • .. i
DryzBln
. . .,....,. ., .,
Crl- '- 1"
~ .-;. p
-tfl
H=TE
•MHo^
03JUNE2009 Version
23/27
Revised/Submitted
Previous
TOC
-------
Bioactivity Profiling Using BioMAP Cell Systems
Houck et al
Fig. 2. Overlay of BioMAP profiles of positive control compound (colchicine) replicates.
Colchicine was tested in the 8 BioMAP Systems (Figure 1) at 3.3 uM and included as a positive
control on every plate (template) used in the study. All replicates (T1-T16) are shown. The
biomarker readouts measured (see Methods) are indicated along the x-axis. The y-axis shows
the Iog10 expression ratios of the readout level measurements relative to solvent (DMSO buffer)
controls. Each data point represents a single well. The grey area above and below the dashed
line indicates the 95% significance envelope of DMSO negative controls.
BioMAP Systems
Cytotoxicity Readouts
Readout Parameters (Biomarkers)
03JUNE2009 Version
Previous
24/27
TOC
Revised/Submitted
-------
Bioactivity Profiling Using BioMAP Cell Systems
Houck et al
Fig. 3. Function Similarity Map for EPA ToxCast Compounds. Compound profiles in 8 BioMAP
systems were compared by pain/vise correlation and correlations analyzed for significance and
subjected to non-linear projection (see Materials and Methods). Compounds at concentrations
resulting in overt toxicity to cells are not included. Compound clusters discussed in text include:
(1) mitochondrial inhibitors; (2) NFidB inhibitors; (3) cAMP elevators; and (4) inducers of
endoplasmic reticulum stress.
A.OWLI-
03JUNE2009 Version
Previous
25/27
TOC
Revised/Submitted
-------
Bioactivity Profiling Using BioMAP Cell Systems
Houck et al
Fig. 4. BioMAP profiles of pyraclostrobin and trifloxystrobin compared to reference compound
myxathiazol, an inhibitor of mitochondrial electron transport chain complex II. Compounds
were tested at the indicated concentrations as described in Materials and
Methods
•0.9
-l.D
-1.1
-1.2
•U
i -:
• Myxothiazol, 1.111 uM
• Pyractostrobin. 13.333 uM
• Trifloxystrabin, 13.333 uM
III
TFF
!«i
03JUNE2009 Version
26/27
Revised/Submitted
Previous
TOC
-------
Bioactivity Profiling Using BioMAP Cell Systems
Houck et al
Fig. 5. BioMAP profiles of selected test chemicals and reference compounds showing profiles
characteristic of A) endoplasmic reticulum stress, B) NFkB inhibition, C) cAMP elevation, and D)
inhibition of microtubule function. Compounds were tested at the indicated concentrations as
described in Materials and Methods.
A.
BE.3C HDF3CGF Kl '•< I
JI]17r'TjTljTf ^^pTi
rrrnrijn mrn
p{,||WTT^|pT'iTT|»
03JUNE2009 Version
Previous
27/27
TOC
Revised/Submitted
-------
Pa9e 1 of 14 ToxSci Advan^tffl&iigaiSeiKraa July 14, 2009
1
2
3
4
5 Biologically-Relevant Exposure Science for 21s Century Toxicity
® Testing
8
9 Elaine A. Cohen Hubal*
10
11 National Center for Computational Toxicology, U.S. EPA, Research Triangle Park, NC
12 27711
13
14
15
1 e * Author to whom correspondence should be addressed
17 U.S. Environmental Protection Agency
18 National Center for Computational Toxicology
™ Mail Drop B205-01
21 Research Triangle Park, NC 27711
22 Telephone: (919)541-4077
23 Internet: hubal.elaine@epa.gov
24
25 Send express mail to:
?7 U.S. Environmental Protection Agency
28 Mail Drop E205-01
29 4930 Old Page Road
30 Research Triangle Park,NC 27709
31
^2 Running Title: EXPOSURE AND THE NRC VISION
34
35 Keywords: Exposure, biomarkers, exposure ontology, knowledge bases, and network
36 models
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60 Published by Oxford University Press 2009.
Previous I TOC
-------
Toxicological Sciences Page 2 of 14
1
2
3
4
5
6 ABSTRACT
7
8 High visibility efforts in toxicity testing and computational toxicology including the
9 recent NRC report, Toxicity Testing in the 21st Century: a Vision and Strategy (NRC,
11 2007), raise important research questions and opportunities for the field of exposure
12 science. The authors of the National Academies report (NRC, 2007) emphasize that
13 population-based data and human exposure information are required at each step of their
14 vision for toxicity testing, and that these data will continue to play a critical role in both
15 guiding development and use of the toxicity information. In fact, state-of-the-art
exposure science is essential for translation of toxicity data to assess potential for risk to
1 g individuals and populations and to inform public health decisions. As we move forward
19 to implement the NRC vision, a transformational change in exposure science is required.
20 Application of a fresh perspective and novel techniques to capture critical determinants at
21 biologically-motivated resolution for translation from controlled in vitro systems to the
~: open, multifactorial system of real-world human-environment interaction will be critical.
24 Development of an exposure ontology and knowledgebase will facilitate extension of
25 network analysis to the individual and population for translating toxicity information and
26 assessing health risk. Such a sea change in exposure science is required to incorporate
27 consideration of lifestage, genetic susceptibility, and interaction of non-chemical
28 stressors for holistic assessment of risk factors associated with complex environmental
disease. A new generation of scientific tools has emerged to rapidly measure signals
31 from cells, tissues, and organisms following exposure to chemicals. Investment in 21st
32 century exposure science is now required to fully realize the potential of the NRC vision
33 for toxicity testing.
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
-------
Page 3 of 14 Toxicological Sciences
1
2
3. INTRODUCTION
4
5
6 The NRC report, Toxicity Testing in the 21st Century: a Vision and Strategy (NRC, 2007)
7 articulates a long-range vision and promotes a transformation in toxicity testing based on
8 a rapidly evolving understanding of molecular pathways—the role of these in normal cell
^ function and in toxicity. The key aspect of the NRC vision and proposed paradigm shift
11 in toxicity testing is that new tools are available to examine toxicity pathways in a depth
12 and breadth that has not been possible before. In response to the NRC report, efforts
13 underway to apply high-throughput-screening (HTS) approaches for chemical
14 prioritization and toxicity testing have been accelerated (Collins et al, 2008, Dix et al,
15 2007). As a result, an explosion of HTS data for in vitro toxicity assays will become
available over the next few years. How will this new toxicity information be translated
1 g to assess potential for risks to individuals and populations from environmental exposures
19 and to improve public health?
20
21 The authors of the National Academies report (NRC, 2007) emphasize that population-
~: based data and human exposure information are required at each step of their vision for
24 toxicity testing. Exposure needs highlighted in the NRC report include (1) human
25 exposure data to select doses for toxicity testing and facilitate development of
26 environmentally-relevant hazard information; (2) biomonitoring data relating real-world
27 human exposures with concentrations that perturb toxicity pathways to identify
28 biologically-relevant exposures; and (3) information on host susceptibility and
background exposures to interpret and extrapolate (i.e., translate) in Protest results for
31 risk assessment. While the importance of exposure information for design and
32 interpretation of toxicity testing under the NRC vision was clearly identified, it was
33 beyond the charge of the committee to address the required science and resources to meet
34 this need. As a result, current discrepancies in the scientific foundation for hazard and
^ exposure characterization are rapidly increasing.
37
38 Others, however, have recognized that just as interpretation of toxicogenomic hazard data
39 requires anchoring to apical endpoints for contextual relevance, understanding relevant
40 perturbations leading to these toxicogenomic endpoints requires anchoring stressors to
4^ real-world human exposure (biologically-relevant exposure metrics). New approaches
43 for toxicity testing and risk assessment require systems-based consideration of
44 interactions between exposure and effect as well as the science to predict exposures down
45 to the molecular level (Edwards and Preston, 2008; Cohen Hubal et al., 2008; Sheldon
46 and Cohen Hubal, 2009). Wild (2005) has proposed the need for a "step change" in
47 exposure assessment and has articulated a vision for exposure measurement
2® commensurate with that of the NRC vision for toxicity testing. Wild has called for an
CQ "exposome," or measurement of the life-course of environmental exposures to provide
51 the evidence base for public health decisions to address environmental health. Wild and
52 others (e.g., Weis et al, 2005) discuss the potential of emerging technologies to provide
53 this new generation of exposure information. Finally, Lange et al. (2007) have discussed
^4 the need to integrate heterogeneous ontologies into interdisciplinary knowledge systems
56 to unify scientific fields and harness the full potential of exposure and health outcome
57
58
59
60
-------
Toxicological Sciences Page 4 of 14
1
2
3
data. Lange et al. illustrate a framework and call for development of a knowledge system
5 to seamlessly compute relationships across the source-to-outcome continuum.
6
7 In this paper, the case is presented for a transformation in exposure science
8 commensurate with the transformation in toxicology presented by the NRC. A new
^ generation of tools to rapidly characterize biologically-relevant exposures and link to
11 environmentally-relevant hazard is required to employ toxicity data for holistic risk
12 assessment and to inform public health decisions. Research initiatives required to
13 develop this exposure science include: (1) application of systems biology network
14 modeling to identify exposure metrics and models for characterizing key stressors at
15 biologically-motivated resolution; (2) development and application of advanced
technologies to measure key exposure metrics (e.g., biomarkers to measure internal
1g exposure, sensors to measure personal exposure); and (3) development of an exposure-
19 hazard knowledge system to facilitate risk assessment. The imperative for this
20 transformation and an outline of the required initiatives follow.
21
22
24 THE CASE FOR EXPOSURE
25
26 Exposure characterization is the risk analysis step in which human interaction with an
27 environmental agent of concern is evaluated. Exposure is defined as the contact between
28 an agent and a target (WHO, 2004). Although the primary application of this definition
for risk assessment has been to the individual or human population as a target of exposure
31 and a chemical as an agent of exposure, the target of exposure can be an organ, tissue or
32 cell, and the agent of exposure can be a biological, physical, or psychosocial stressor or
33 the byproduct of given exposure agent (Figure 1). Exposure assessment is defined as
34 evaluation of exposure of a system, organism, or (sub)population to an agent (and its
^ derivatives). The process may include estimating the magnitude, frequency, and duration
37 of an exposure, along with characteristics of the exposed individual or population (WHO,
38 2004). In limited cases, exposure can be measured directly, but more often due to current
39 scientific limitations exposure must be estimated (Cohen Hubal et al., 2008; Paustenbach
40 2000).
41
43 Low-level and prevalent environmental exposures may contribute substantially to the
44 burden of common complex disease (Hemminki et al, 2006, Gibson, 2008).
45 Understanding the relationships between environmental exposures and health outcomes
46 requires integration of a wide range of factors—extrinsic (e.g. environmental), intrinsic
47 (e.g. genotypic), and mechanistic (e.g. lexicological)—to support health studies and
4® characterize risk. Assessing complex human-health risks associated with exposures to
CQ chemicals requires that hazard, susceptibility, and exposure are all reliably characterized.
51
52 Characterization of susceptibility is rapidly advancing through application of microarray
53 technology for genotyping and investment in large genome-wide association (GWA)
^4 studies (McCarthy et al, 2008). Epidemiologists are now facing the challenges associated
56 with interpreting this massive amount of genomic variation data for understanding
57 etiology of complex environmental disease. Calls for tools to characterize and unravel
58
59
60
Previous I TOC
-------
Page 5 of 14 Toxicological Sciences
1
2
3
interacting genetic and environmental factors have begun (Collins, 2006; Manolio and
5 Collins, 2007). Similarly, the NRC has presented a vision for advancing characterization
Q of hazard through application of high-throughput screening (HTS) methods and system
7 biology approaches to elucidate toxicity pathways. As a result, toxicologists are facing
8 the challenges of translating the high content toxicity data that is now being generated
^ (Dix et al, 2007) to inform risk assessment. The high-priority need for research to
11 interpret these hazard data in the context of real-world exposure has been identified by
12 risk assessors. At the same time, however, characterization of exposure remains
13 primitive by comparison (e.g., scenario-based assessment of exposures of sentinel
14 products in non-standardized scenarios versus measurement of real-time personal
15 exposure) and resources to improve the scientific basis of exposure assessment are
limited or nonexistent. This lack of balance in efforts to improve measuring hazard and
1 g exposure is less than ideal for providing advancement in risk assessment.
19
20 Just as Wild (2005) questions whether or not fundamental knowledge about genetics will
21 improve understanding of disease etiology at the population level, we should question
~: whether or not fundamental knowledge of toxicity pathways will improve understanding
24 of real-world human-health risk. Accurate assessment of chemical exposures remains an
25 outstanding and largely unmet challenge in toxicology and risk assessment. One side of
26 the hazard-exposure equation continues to be refined while the other remains subject to
27 crude characterization based largely on indirect estimates and default assumptions. Due
28 to the complex nature of the human system, predictions of potential health risks
associated with chemical exposures will be limited by the least resolved or least
31 understood component of the system. By focusing resources exclusively on improving
32 hazard characterization we compromise the ability to fully realize benefits of the NRC
33 vision. Just as a new generation of scientific tools is being applied to rapidly assess toxic
34 response resulting from chemical exposures, there is a critical need to develop methods
^ for characterizing environmental exposures at biologically-relevant resolution to translate
37 HTS toxicity results for human health risk assessment.
38
39
40 REQUIRED INITIATIVES
41
43 What does the real world look like and how can we capture a picture (i.e., model) of the
44 real world that will facilitate risk assessment and allow us to make important
45 environmental health decisions? Understanding relationships among multiple
46 environmental factors and complex disease, as well as characterizing environmental
47 health risk factors requires collection and analysis of a wide range of data. Information
2® on the characteristics of multiple stressors (chemical, physical, biological and
CQ psychosocial), the characteristics of the human receptor (genetics, health status, life stage,
51 behaviors, social factors, etc.) at multiple levels of organization (individual, community,
52 population), and the temporal and spatial patterns of exposures and outcomes must be
53 considered. Strategic research is required to identify key determinants of exposure to
^4 capture the essence of this multifactorial reality. What are the critical elements of
56 exposure in a given context? What are the key metrics for characterizing these exposure
57 elements in that context? What is the required resolution for measuring key metrics and
58
59
60
Previous I TOC
-------
Toxicological Sciences Page 6 of 14
1
2
o
modeling exposures so that these are relevant for developing and interpreting hazard
5 information to assess health risks? Finally, can new scientific understanding and tools in
6 biological, computational, and information sciences be leveraged to develop rapid,
7 inexpensive approaches for characterizing biologically-relevant exposure?
8
9 Under the NAS vision, tools developed by the pharmaceutical industry are being applied
11 to transform toxicity testing. Similarly, exposure scientists must leverage advanced
12 measurement and computational tools from disparate, but related fields to transform
13 exposure science. New technologies must be applied to move from our current crude
14 indirect estimates of exposure to the biologically-based metrics required to interpret
15 emerging toxicity data and advance human-health risk assessment. In addition, the
complexity of the multifactorial systems under study and the resulting multidimensional
1 g data produced using emerging technologies require application of environmental
19 informatic capabilities and advanced computational tools to model and link exposures to
20 health outcomes. A combination of discovery and engineering (mechanistic)-based
21 modeling approaches for hypothesis development and testing are required. Statistical
~: data-mining and machine-learning approaches are required to extract information from
24 extant data on critical exposure determinants, link exposure information with toxicity
25 data, and identify limitations and gaps in exposure data. Engineering or mechanistic
26 approaches are required to model the human-environment system and to test our
27 understanding of this system.
28
29
Systems biology: exposure at all levels of biological organization
31
32 In the NRC vision, the authors propose systems-biology evaluation of signaling networks
33 to characterize perturbations of toxicity pathways and as the basis of a new toxicity-
34 testing paradigm. Environmental stressors (i.e., exposure) leading to perturbations of
^ toxicity pathways are simplified and treated as unidirectional and one dimensional.
37 Fortunately, systems theory also provides the required conceptual framework for linking
38 exposure science and toxicology to study, characterize, and make predictions about the
39 complex interactions between humans and environmental chemicals and associated
40 feedback across levels of biological organization (Figure 1). A systems-biology approach
41 for holistic study of environmental disease and risk assessment considers coupled
43 networks that span multiple levels of biological organization. These networks describe
44 the overall connectivity of the system. Mechanistic understanding is derived by
45 characterization of these networks and impacts of perturbations due to behavioral and
46 environmental influences. Edwards and Preston (2008) present the conceptual basis for
47 extending network analysis to inform risk assessment. Networks at different levels of the
4® system can be used to merge molecular-level changes with measured events at the
CQ individual or population level. Molecular networks are developed based on data from
51 'omic measurements. Key event networks, where each node ideally represents a toxicity
52 pathway, are abstracted from the molecular network based on biological interpretation
53 and targeted experimentation (both in vitro and in vivo). Adverse outcomes are driven by
^4 the impact of an individual's genetics, epigenetics and exposure profile. Connectivity at
56 the population level is driven by common genetics, lifestyle, and environment. An
57
58
59
60
Previous I TOC
-------
Page 7 of 14 Toxicological Sciences
1
2
3
example of the type of approach described by Edwards and Preston has been partially
5 demonstrated for an ecological model of endocrine disruption (Ankley, et al., 2009).
6
7 Gohlke et al (2009) present an example of how this approach can be applied using gene-
8 centered databases to develop linked networks to explore interplay between genetic and
^ environmental factors for metabolic syndrome and neuropsychiatric disorders. The
11 analysis presented by Gohlke and coauthors highlights significant gaps in exposure
12 information required to extend this approach to assess and mitigate human health risks.
13 The Comparative Toxicogenomic Database (Davis et al., 2008) used to compile
14 environmental factor-gene/protein relationships does indeed provide an important model
15 for how exposure information can and should be made accessible to facilitate
investigation of gene-environment-disease relationships. However, because the CTD is
1 g limited to curated information on direct chemical-gene interactions and direct gene-
19 disease associations, chemical-disease relationships must be inferred. Here again, real-
20 world exposure information is required to translate molecular insights to assess risks to
21 individuals and populations.
22
23
24 Just as key cellular processes may be associated with multiple complex outcomes, it is
25 likely that exposures to multiple xenobiotic compounds may elicit perturbation of the
26 same key toxicity pathways. Understanding the critical determinants of multifactorial
27 perturbations and feedback in the human-environment system is required to interpret
28 toxicity data for risk assessment. A systems approach for assessing risk provides a
holistic view of interactions between a chemical stressor and biological entity from the
31 molecular level through to the level of the organism and/or population (Figure 1). Such a
32 holistic systems approach demands exposure metrics and models to characterize key
33 stressors at a level of resolution commensurate with that of the response or effects
34 (biologically-motivated resolution).
35
oc
37 Biologically-relevant exposure metrics
38
39 The major challenge to realizing the full potential of the NRC vision is the limited
40 availability of efficient and affordable methods for measuring biologically-relevant
4^ exposures. Biologically-relevant exposure metrics are those that can be directly
43 associated with key events in a disease process and with an individual's exposure profile.
44 Based on this need to characterize biologically-relevant environmental exposures, Wild
45 (2005) has proposed investment in development of exposure biomarkers to improve our
46 ability to understand and mitigate environmental impacts on human health. In fact, we
47 are faced with the same critical need for advanced exposure science if we are to realize
2® the NRC vision. Though Wild's vision for an "exposome" has been articulated in the
CQ context of characterizing the environmental contribution to etiology of common complex
51 disease, the basic principles behind this call are germane for human health risk
52 assessment. As early as 1999, Groopman and Kensler highlighted the challenges
53 associated with developing biomarkers and interpreting biomarker data to sort out the
^4 interactions of multiple chemicals, multiple exposures and relation of these to health
56 outcomes. With appropriate investment, a new generation of technologies may provide
57 the tools to address some of these challenges. Limited examples follow, but
58
59
60
Previous I TOC
-------
Toxicological Sciences Page 8 of 14
1
2
o
opportunities to adapt a wide range of sensors and biomarkers to measure chemical
5 stressors and/or derivatives of these at all levels of biological organization should be
6 considered.
7
8 Currently, as advocated by the NAS vision, investment in 'omic technologies is focused
^ on understanding and characterizing toxicity or hazard. Yet, these tools may also yield a
11 new generation of exposure metrics. In a second report (NRC, 2007a), the NAS has
12 called for further development of toxicogenomic technologies to increase capabilities in
13 exposure assessment. There are early indications that with appropriate investment, this
14 area of research could provide important approaches for assessing real-world exposures.
15 Wild (2009) considers application of transcriptomics for development of exposure
biomarkers to improve exposure assessment in epidemiology. Fry et al (2007) present
1 g an exciting example of the potential to link environmental exposure and altered gene
19 expression. In a study conducted in Thailand, Fry and coworkers identified expression
20 signatures from babies born to arsenic-unexposed and -exposed mothers that were highly
21 predictive of prenatal arsenic exposure in a subsequent test population. Resulting
~: signatures, based on a very small number of genes, show promise as biomarkers of
24 arsenic exposure. Other studies have investigated altered global gene expression
25 associated with exposure to cigarette smoke, benzene, metal fumes and air pollution.
26 While the limited research conducted to date suggests that environmental exposures elicit
27 changes in gene expression specific to the type of exposure, significant scientific hurdles
28 remain. However, careful targeted investment should be directed to develop 'omic tools
to link real-world exposure and in vitro hazard information.
31
32 Biomarkers alone will not provide the full range of information required to characterize
33 exposure and make critical links to hazard for risk assessment. Direct, noninvasive, and
34 sensitive detection of chemical stressors in relevant environmental and biological media
o c
^ could prove to be the most effective means of assessing exposure. Recently, Schwartz
37 and Collins (2007) identified the need for better environmental biosensors to study gene-
38 environment interactions associated with complex disease. Research in this area is also
39 required to apply toxicity testing results to assess human health risk. Advances in
40 nanotechnology and related development of small-scale sensors promise to facilitate
41 comprehensive monitoring of exposure, dose, and associated indicators of early effect.
43 Nanotechnologies offer the potential to improve exposure and risk assessment by
44 facilitating collection of large numbers of measurements on very small numbers of
45 molecules at a low cost. It is currently possible to develop micro- and nano-scale sensor
46 arrays that can detect specific sets of harmful agents in the environment (Andreescu et al,
47 2009). Provided adequate informatics support, these sensors can be used to monitor
4® multiple agents in real time and the resulting data can be accessed remotely. The
CQ potential also exists to extend these small-scale monitoring systems to the individual level
51 to detect personal and in vivo distributions of toxicants (Barry et al, 2009; Weis et al.
52 2005).
53
^4 Together, application and development of exposure assessment tools such as advanced
56 molecular indicators of exposure (Sen et al, 2007, Wild 2009) and nanotechnology-based
57 sensors (Barry et al, 2009; Andreescu, 2009) present the opportunity for simultaneous
58
59
60
-------
Page 9 of 14 Toxicological Sciences
1
2
3
measurement of biologically-relevant exposures to multiple real-world stressors as well
5 as the potential to mechanistically link traditional exposure metrics and endpoints
6 measured in HTP in vitro assays.
7
8 Exposure-hazard knowledge system
9
10
11 The NRC vision emphasized the importance of extrapolation modeling to: (1) provide
12 quantitative, mechanistic understanding of dose-response relationship for perturbations
13 by environmental agents; (2) predict human exposures leading to tissue concentrations
14 comparable to concentrations causing perturbations in vitro, and (3) provide a basis for
15 addressing background chemical exposures, background disease processes, and host
susceptibility (NRC, 2007). Development and use of models that can efficiently address
1 g these critical components for translation and risk assessment requires capabilities to
19 collect, organize, retrieve, and link large amounts of disparate, multidimensional
20 exposure and hazard information.
21
OO
~: Significant energy and resources have been committed to collate and improve access to
24 genomic, toxicology, and health data (Richard et al 2008; Judson et al 2008; Davis et al
25 2008). Lacking from these information resources is the real-world exposure data
26 required to translate molecular insights for assessing risks at the individual and
27 population level. Knowledge-discovery based tools are new to the exposure science
28 community. Yet, these tools are absolutely critical as these provide the opportunity to
efficiently leverage exposure information for extrapolation modeling and translation of in
31 vitro HTS toxicity data for risk assessment.
32
33 Translation of the hazard information developed under the NRC vision will require a
34 holistic risk assessment knowledge system that includes ontologies and databases to
^ facilitate computerized collection, organization, and retrieval of exposure, hazard, and
37 susceptibility information. In addition, this system must be compatible and linkable to
33 the larger environmental health universe of information to facilitate risk assessment for
39 improved public health decisions (Richard 2006, 2008; Judson et al. 2008; Davis 2008).
40 An exposure ontology consistent with those being used in toxicology and other health
4^ sciences is required to formally represent exposure concepts, the relationships between
43 these concepts and most important the relationships between exposure, susceptibility, and
44 toxicology domains.
45
46 Lange et al. (2007) illustrate a framework for an interdisciplinary knowledge system to
47 link agriculture, food science, nutrition, and health. Although this vision is presented in a
2® slightly different context, the concept as outlined is directly relevant for 21st century
CQ human health risk assessment. Just as we are experiencing a transformation in
51 toxicology, the food science community has seen a shift from discovery of essential
52 nutrients for human survival to characterization of complex, multifactorial interactions
53 among food, diet, and health. The authors argue for standardized ontologies to define
^4 relationships, allow for automated reasoning, and facilitate meta-analyses. This same
56 capability is clearly required to develop biologically-relevant exposure metrics, design in
57 vitro toxicity tests to measure environmentally-relevant hazard, and to incorporate
58
59
60
Previous I TOC
-------
Toxicological Sciences Page 10 of 14
1
2
3
information on susceptibility and background exposures for interpretation of these data to
5 assess real-world risks to individuals and populations.
6
7
8 EXPOSURE SCIENCE FOR THE 21ST CENTURY
9
10
11 The need for radical improvement in exposure science is not academic. Characterization
12 of biologically-relevant exposure is required to translate advances and findings in
13 computational toxicology to information that can be directly used to support risk
14 assessment for decision making and improved public health. New technologies must be
15 applied to both toxicology and exposure science if the ultimate goal of evaluating risk to
humans is to be achieved. Just as authors of the NRC report (2007) recognize the need
1 g for broad-based support to achieve their vision for toxicity testing, realization of
19 objectives for 21st century risk assessment will require significant investment in exposure
20 science and development of capacity across both the public and private sector.
21 Ultimately, this additional investment will maximize contributions of emerging toxicity
~: testing approaches toward improved understanding of relationships between
24 environmental factors and human health outcomes.
25
26 Recognition that improvement in exposure science is required to characterize and manage
27 risks associated with environmental stressors is broad based. A National Academies
28 workshop conducted in June 2009, Exposure Science in the 21st Century, focused on the
role of exposure science in health studies, risk assessment, and risk prevention (NRC,
31 2009). At this workshop, the director of US EPA's National Exposure Research
32 Laboratory announced plans for a new National Academies committee on exposure
33 science in the 21st century (Reiter, 2009). Formation of this committee would be a
34 critical step toward building the scientific basis for exposure characterization to protect
^ environmental and public health. At the same time, the risk assessment community
3-, cannot wait to initiate important research in exposure science to meet rapid advances in
38 toxicity testing and critical needs for translating emerging HTP hazard data if the NRC
39 vision (2007) is to be realized.
40
41
42
43 Acknowledgements
44 I would like to thank Drs. Tina Bahadori and Steve Edwards for helpful discussions. I
45 would also like to thank Dr. Christopher Wild for so vividly highlighting the critical need
46 for reliable exposure assessment tools.
47
2® Disclaimer
CQ This manuscript is has being subjected to Agency review through the Office of Research
51 and Development and been cleared for publication by the US Environmental Protection
52 Agency.
53
54 Figure Captions
55 5 F
Figure 1. Cascade of exposure-response processes for integrating exposure science and
co toxicogenomic mode-of-action information. (Cohen Hubal et al., 2008, JESEE)
59
60
Previous I TOC
-------
Page 11 of 14 Toxicological Sciences
1
2
3
5 REFERENCES
6
7 Andreescu, S, J Njagi, C Ispas, MT Ravalli. (2009) JEM Spotlight: Applications of
8 advanced nanomaterials for environmental monitoring. JEnviron Monit. 11(1):27-40.
9
10
11 Ankley GT, Bencic DC, Breen MS, Collette TW, Conolly RB, Denslow ND, Edwards
12 SW, Ekman DR, Garcia-Reyero N, Jensen KM, Lazorchak JM, Martinovic D, Miller DH,
13 Perkins EJ, Orlando EF, Villeneuve DL, Wang RL, Watanabe KH. (2009) Endocrine
14 disrupting chemicals in fish: Developing exposure indicators and predictive models of
15 effects based on mechanism of action. Aquatic Toxicol92(3): 168-178.
\ g Barry RC, Y Lin, J Wang, G Liu, CA Timchalk. (2009) Nanotechnology-based
19 electrochemical sensors for biomonitoring chemical exposures. J Expo Sci Environ
20 Epidemiol. 19(1): 1-18.
21
OO
~: Cohen Hubal, EA, J Moya, SG Selevan. (2008) A lifestage approach to assessing
24 children's exposure. Birth Defects Res (Part B) 83:522-529.
25
26 Cohen Hubal EA, Richard AM, Imran S, Gallagher J, KavlockR, Blancato J, Edwards S.
27 (2008) Exposure science and the US EPA National Center for Computational
28 Toxicology. J Expo Sci Environ Epidemiol. doi: 10.1038/jes.2008.70 [Online: Nov 5
30 2°°8]
31
32 Collins FS, Gray GM, Bucher JR. (2008) Transforming environmental health protection.
33 5cfence319(5865):906-7.
34
35 Collins FS. (2006) The case for a US prospective cohort study of genes and
36 environment. Nature 429':47'5-77'.
37
O Q
39 Davis, AP, CG Murphy, MC Rosenstein, TC Wiegers, CJ Mattingly. (2008) The
40 Comparative Toxicogenomics Database facilitates identification and understanding of
41 chemical-gene-disease associations: arsenic as a case study. BMC Medical Genomics
42 1:48.
43
A A
45 Dix DJ, Houck KA, Martin MT, Richard AM, Setzer RW, Kavlock RJ. (2007) The
45 ToxCast program for prioritizing toxicity testing of environmental chemicals. Toxicol Sci.
47 95(1):5-12.
48
49 Edwards, SW, Preston, RJ. (2008) Systems biology and mode of action based risk
y assessment. Toxicol Sci. 106(2):312-318.
O I
52
53 Fry RC, Navasumrit P, Valiathan C, Svensson JP, Hogan BJ, Luo M, Bhattacharya S,
54 Kandjanapa K, Soontararuks S, Nookabkaew S, Mahidol C, Ruchirawat M, Samson LD.
55 (2007) Activation of inflammation/NF-kappaB signaling in infants born to arsenic-
56 exposed mothers. PLoS Genet. 3(ll):e207.
O /
58
59
60
Previous I TOC
-------
Toxicological Sciences Page 12 of 14
1
2
o
Gibson, G. (2008) The environmental contribution to gene expression profiles. Nat Rev
5 Genet 9:57 5-81.
6
7 Gohlke, JM, R Thomas, Y Zhang, MC Rosenstein, AP Davis, C Murphy, KG Becker, CJ
8 Mattingly and CJ Portier. (2009) Genetic and environmental pathways to complex
^ diseases. SMC Systems Biology3:46.
12 Groopman, JD, and TW Kensler (1999) The light at the end of the tunnel for chemical-
13 specific biomarkers: daylight or headlight? Carcinogenesis 20( 1)1-11.
14
15 Hemminki, K, L Bermejo, JL Bermejo, A Forsti. (2006) The balance between heritable
17 and environmental aetiology of human disease. Nat Rev Genet 7(12) 958-65.
18
19 Judson R, A Richard, D Dix, K Houck, F Elloumi, M Martin, T Cathey, TR Transue, R
20 Spencer, M Wolf. (2008) ACToR - Aggregated Computational Toxicology
21 Resource. ToxicolAppliedPharmacol. 233(1):7-13.
23
24 Lange, MC, DG Lemay, JB German. (2007) A multi-ontology framework to guide
25 agriculture and food towards diet and health. JSciFoodAgric 87:1427-1434.
26
27 McCarthy MI, GR Abecass, LR Cardon, DB Goldstein, J Little, JPA loannidis, JN
28 Hirschhorn. (2008) Genome-wide association studies for complex traits: consensus,
__ uncertainty and challenges. Nat Rev Genet 91:356-59.
31
32 Manolio, TA, FS Collin. (2007) Genes, environment, health, and disease: facing up to
33 complexity. Hum Hered63(2):63-66.
34
35 National Research Council of the National Academies (NRC). (2007) Toxicity Testing in
the 21st Century: A Vision and a Strategy. The National Academies Press. Washington,
38 DC
39
40 National Research Council of the National Academies (NRC). (2007a) Applications of
41 Toxicogenomic Technologies to Predictive Toxicology and Risk Assessment. The
42 National Academies Press. Washington, DC.
43
/L National Research Council of the National Academies (NRC). (2009) Eighth Workshop
4g of the Standing Committee on Risk Analysis Issues and Reviews. Exposure Science In
47 The 21st Century. June 18-19, 2009. National Academy of Sciences, Washington, DC.
48 http://dels.nas.edu/best/risk_analysis/ExposureBackground.shtml (accessed June 29,
49 2009).
50
52 National Research Council of the National Academies (NRC). (2009b) Standing
53 Committee on Use of Emerging Science for Environmental Health.
54 http://dels.nas.edu/envirohealth/ (accessed July 1, 2009)
55
56
57
58
59
60
Previous I TOC
-------
Page 13 of 14 Toxicological Sciences
1
2
3
Paustenbach DJ. 2000. The practice of exposure assessment: A state-of-the-art review
5 (Reprinted from Principles and Methods of Toxicology, 4th edition, 2001). J Toxicol
6 Environ Health B Crit Rev 3(3): 179-291.
7
8 Richard AM, Gold LS, and Nicklaus MC (2006) Chemical structure indexing of toxicity
^ data on the Internet: Moving toward a flat world. Current Opinion in Drug Discovery &
]° Development(CODDD), 9(3):314-325.
12
13 Richard, A, C Yang, R Judson (2008) Toxicity data informatics: Supporting a new
14 paradigm for toxicity prediction. Tox. Mech. Meth. 18:103-118.
15
16 Reiter, L. (2009) EPA's Perspective on Exposure Science and Goals for the Workshop.
17 National Research Council, Eighth Workshop of the Standing Committee on Risk
Analysis Issues and Reviews. Exposure Science In The 21st Century. June 18-19, 2009.
20 National Academy of Sciences, Washington, DC.
21
22 Schwartz D and Collins FS. (2007) Environmental biology and human disease. Science
23 316(5825):695-696.
24
^c
Sen, B, Mahadevan B and DeMarini DM ( 2007) Transcriptional responses to complex
2i mixtures—A review. Mutation Research 636 (2007) 144-177.
28
29 Sheldon, LS, and EA Cohen Hubal. (2009) Exposure as part of a systems approach for
30 assessing risk. Environ Health Perspectdoi:W. 1289/ehp.0800407 [Online 8 April 2009]
31
32 Weis, BK, D Balshawl, et al. (2005). Personalized exposure assessment: Promising
approaches for human environmental health research. Environ Health Perspect 113(7):
3* 840-848.
36
37 WHO. (2004). IPCS Risk Assessment Terminology. Harmonization Project Document
38 No. 1, World Health Organization, Geneva. Accessed online on 06/29/09 at
39 http://www.who.int/ipcs/methods/harmonization/areas/terminology/en/index.html
40
41
42 Wild, CP. (2005) Complementing the Genome with an "Exposome": The Outstanding
43 challenge of environmental exposure measurement in molecular epidemiology. Cancer
44 Epidemiol Biomarkers Prev 14(8) 1847-18 5 0.
45
46 Wild, CP. (2009) Environmental exposure measurement in cancer epidemiology.
^ Mutagenesis. 24(2) 117-125.
49
50
51
52
53
54
55
56
57
58
59
60
Previous I TOC
-------
Toxicological Sciences
Page 14 of 14
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
AQ
Stressor
Perturbation
Biological
Receptor
Perturbation
Outcome
Environmental
Source
| Ambient
Exposure
Environmental
Source
U
Population
Personal
Exposure
J
Individual
Internal Exposure'
(Tissue Dose)
Dose to Cell
Dose of Stressor
Molecules
J
Tissue
J
Cell
J
Biological
Molecules
Disease
Incidence/Prevalence
Disease State
(Changes to Health Status)
Dynamic Tissue Changes
(Tissue Injury)
Dynamic Cell Changes
(Alteration in Cell Division,
Cell Death)
Dynamic Changes in
Intracellular Processes
-------
TOXICOLOGICAL SCIENCES 95(1), 5-12 (2007)
doi: 10.1093/toxsci/kfll03
Advance Access publication September 8, 2006
FORUM
The ToxCast Program for Prioritizing Toxicity Testing of
Environmental Chemicals
David J. Dix,1 Keith A. Houck, Matthew T. Martin, Ann M. Richard, R. Woodrow Setzer, and Robert J. Kavlock
National Center for Computational Toxicology (D343-03), Office of Research and Development, U.S. Environmental Protection Agency,
Research Triangle Park, North Carolina 27711
Received May 24, 2006; accepted August 30, 2006
The U.S. Environmental Protection Agency (EPA) is developing
methods for utilizing computational chemistry, high-throughput
screening (HTS), and various toxicogenomic technologies to predict
potential for toxicity and prioritize limited testing resources toward
chemicals that likely represent the greatest hazard to human health
and the environment. This chemical prioritization research pro-
gram, entitled "ToxCast," is being initiated with the purpose of
developing the ability to forecast toxicity based on bioactivity
profiling. The proof-of-concept phase of ToxCast will focus upon
chemicals with an existing, rich toxicological database in order to
provide an interpretive context for the ToxCast data. This set of
several hundred reference chemicals will represent numerous
structural classes and phenotypic outcomes, including tumorigens,
developmental and reproductive toxicants, neurotoxicants, and
immunotoxicants. The ToxCast program will evaluate chemical
properties and bioactivity profiles across a broad spectrum of data
domains: physical-chemical, predicted biological activities based on
existing structure-activity models, biochemical properties based on
HTS assays, cell-based phenotypic assays, and genomic and
metabolomic analyses of cells. These data will be generated through
a series of external contracts, along with collaborations across EPA,
with the National Toxicology Program, and with the National
Institutes of Health Chemical Genomics Center. The resulting
multidimensional data set provides an informatics challenge re-
quiring appropriate computational methods for integrating various
chemical, biological, and toxicological data into profiles and models
predicting toxicity.
Key Words: high-throughput screening; toxicogenomics; chemo-
informatics; bioinformatics.
Across several U.S. Environmental Protection Agency (EPA)
programs, there is a clear need to develop methods for
evaluating large numbers of environmental chemicals for
potential toxicity and to use the resulting information to
Disclaimer: This work was reviewed by EPA and approved for publication
but does not necessarily reflect official Agency policy.
'To whom correspondence should be addressed. Fax: (919) 541-1194.
E-mail: dix.david@epa.gov.
Published by Oxford University Press 2006.
prioritize the use of testing resources toward those chemicals
and endpoints that present the greatest likelihood of risk to
human health and the environment. This need can be addressed
through the experience of the pharmaceutical industry in the
use of state of the art, high-throughput screening (HTS),
toxicogenomics, and computational chemistry tools for the
discovery of new drugs (Table 1), with appropriate adjustments
to the needs of environmental toxicology. Thus, a research pro-
gram entitled "ToxCast" has been initiated within EPA to
develop an ability to forecast toxicity based on bioactivity
profiling. Ultimately, ToxCast's purpose is to develop methods
of prioritizing chemicals for further screening and testing to
assist EPA programs in the management and regulation of
environmental contaminants.
Over the past decade, HTS has developed into a primary tool
for drug discovery based upon bioactivity screening of the
drugable proteome (Fliri et al., 2005b; Janzen and Hodge,
2006). On a more limited scale, HTS has also been adapted to
agrochemical discovery for the analysis of target species and
model organisms (Smith et al., 2005; Tietjen et al., 2005).
Recently, HTS applications to toxicology have been expanding
as a useful complement to traditional toxicology (Bhogal et al.,
2005; Fliri et al., 2005a; Kikkawa et al., 2006). In the federal
sector, the National Institutes of Health Chemical Genomics
Center (NCGC) has been established (http://www.ncgc.nih.
gov/). The NCGC is using industrial-scale HTS technologies to
collect data that is useful for developing small-molecule
chemical probes for basic biological research (Austin et al.,
2004).
Traditional toxicology testing involves screening com-
pounds through in vivo and in vitro tests focused on defined
endpoints (e.g., neurotoxicity, developmental toxicity) or mech-
anisms of action (e.g., mutagenicity, cytotoxicity, regenerative
hyperplasia). However, EPA is confronted with a large number
of compounds to evaluate and faced with the difficulty of
prioritizing scarce resources. Thus, environmental toxicology
is challenged by (1) too many compounds to evaluate through
endpoint-based in vivo testing and (2) inadequate models or
Previous
TOC
-------
DIX ET AL.
TABLE 1
Selected Examples of Biological Screening and Chemoinformatics Relevant to the Development of ToxCast
Approach
Citation
Description
HTS
Chemoinformatic
surveys
Structure-activity
studies
Austin et al. (2004)
Berg et al. (2006)
Fliri et al. (2005b)
Fliri et al. (2005a)
Melnick et al. (2006)
O'Brien et al. (2006)
Scherf et al. (2000)
Smith et al. (2005)
Tietjen et al. (2005)
Walum et al. (2005)
Richard et al. (2006)
Yang et al. (2006b)
Ekins et al. (2003)
O'Brien and DeGroot (2005)s
Poroikov et al. (2003)
NIH Molecular Libraries Initiative expanding use of small-molecule chemical probes for
biological research
BioMAP profiling based on activity of -100 drugs in cell-based assays designed to incorporate
biological complexity
Biological activity spectra for 1567 compounds (primarily drugs) based on interactions with 92
ligand-binding assays
Utility of biological activity spectra for predicting drug-induced adverse effects
Effects of 1400 kinase inhibitors on panel of 35 tyrosine kinase-dependent cellular assays in
dose-response format
High-content screening of > 600 compounds in HepG2 cells demonstrated human toxicity
potential with 80% sensitivity and 90% specificity
Correlated gene expression changes with drug activity patterns in 60 human cell lines—one of
first to integrate large amounts of genomic and pharmacology data
Application of HTS to agrochemical discovery
HTS assays for development of new herbicides, insecticides, and fungicides included use of
technologies capable of evaluating > 200,000 chemicals per year
Combination of HTS in combination with basic biokinetic information to improve
identification of toxic compounds
Public initiatives accelerating integration of diverse biological information with standardized
chemical structure annotation
Strategy for mining structure-integrated toxicity databases to link chemical structure to
biological endpoint
Public data on 1750 molecules to train computer models that predict inhibition of CYP2D6 and
CYP3A4
Data for 58,963 compounds on human ether-a-go-go related gene channel and 2410
compounds on inhibition of CYP2D6 combined to create predictive model of toxicty
Prediction of activity spectra for substances for total of 565 different outcomes, resulted in
64 million predictions on -25,000 chemicals in National Cancer Institute Open Database,
used to select compounds for further testing
knowledge of mechanism for many types of toxicity to design
suitable in vitro testing. These challenges are also faced by
other organizations including the U.S. Food and Drug Admin-
istration, European Union member countries, the Organization
for Economic Cooperation and Development, and the regulated
community (i.e., the pharmaceutical, agrochemical, and con-
sumer products industries). There is an important need to dis-
tinguish between compounds that present little or no concern
from those with the greatest likelihood of causing an adverse
effect in the target species. High-throughput, high-content, and
toxicogenomic screening methods applied to predictive toxi-
cology provide opportunities for addressing these challenges.
The underlying hypothesis for ToxCast is that toxicological
response is driven by interactions between chemicals and
biomolecular targets. In most cases, these targets are part of the
cellular proteome (e.g., receptors, ion channels, kinases). How-
ever, for most environmental chemicals the protein targets and
biological effects underlying potential adverse effects have yet
to be defined or characterized. Because suitable assays to query
these have remained elusive, a more global approach of bio-
activity profiling is a critical goal in environmental toxicology.
This goal is embodied in the ToxCast program, which will
focus on a multiple target matrix approach rather than a single
target, directed vector approach. The matrix contains an ex-
panded number of potential targets whose chemical interac-
tions may be characterized by in silica models, biochemical
assays, cell-based in vitro assays, and nonmammalian animal
models.
ENABLING HTS AND TOXICOGENOMICS
TECHNOLOGIES
Modern computational chemistry and molecular and cellular
biology tools allow researchers to characterize abroad spectrum
of physical and biological properties for large numbers of
chemicals (Bredel and Jacoby, 2004; Table 1). Genomics,
transcriptomics, proteomics, and metabolomics technologies
are components of this modern molecular biology toolkit.
However, though omics technologies produce large amounts
of data per sample, they are not truly high throughput, and the
per chemical cost can be significant. Thus, the primary driver
transforming drug discovery has been HTS technologies
(Macarron, 2006). HTS is comprised of assays in miniaturized
format that can be either target or phenotype based. Target-based
assays usually measure either binding or function of proteins
Previous
TOC
-------
TOXCAST CHEMICAL PRIORITIZATION
TABLE 2
Differences in the Application of HTS to Pharmaceutical
Research versus Environmental Toxicology
Antimicrobials
750
Pesticide Actives
Chemical space
Chemical numbers
Intended mechanism
of action (MOA)
Target potency
Off-target effects
Error rate
Parent activity
Pharmaceutical
research
Narrow
104-106
Generally known
and specific
High
Often understood
False positives
not acceptable
Chemical design factor
Environmental
toxicology
Broad
102-104
May not exist
Generally low
May or may not be
due to intended MOA
False negatives
not acceptable
Usually unknown
cell free or in engineered cells. Phenotype-based assays monitor
more complex endpoints in cells or whole organisms. These
assays utilize small quantities of reagents and test chemicals and
can be quite cost and time effective for analyzing larger numbers
of chemicals.
The ability to generate broad-based bioactivity profiles for
large libraries of compounds in coordinated portfolios of
biochemical and cellular assays has become the norm in the
pharmaceutical sciences for drug discovery. As bioactivity
profiles for compound libraries have grown, the potential of
these profiles for identifying off-target mechanisms and
potential liabilities has begun to emerge (Bhogal et al., 2005;
Fliri et al, 2005b,c; Klekota et al, 2006; Melnick et al, 2006).
HTS technology optimized for drug discovery is now being
refocused to applications in toxicological screening. It is im-
portant to appreciate, however, the significant and substantial
differences between the application of HTS to pharmaceutical
research versus environmental toxicology (Table 2). The
chemical space and numbers, the targeting and potency, and
most importantly the intolerance for false negatives are all key
differences that will impact assay selection and study design
for ToxCast. The aim of drug discovery HTS is to find a small
number of active compounds amenable to subsequent optimi-
zation for drug development, and in this pursuit, false negatives
are generally not a major concern. HTS for toxicology must
determine the activity of all compounds tested, and false neg-
atives are of greater concern from a public safety standpoint.
ENVIRONMENTAL CHEMICALS RELEVANT
TO TOXCAST
There are potentially 10,000 or more environmental chem-
icals from several EPA programs in need of prioritization for
further testing. Antimicrobials, pesticidal inerts, high produc-
tion volume (HPV: > 1 million Ibs/year) chemicals, inventory
update rule (> 10,000 Ibs/year, < 1 million Ibs/year) chemicals,
H2OCCL
7324
HPV Chemicals
3300
Pesticidal Inerts
3310
IUR Chemicals
5400
FIG. 1. Environmental chemicals present challenges to regulatory Agen-
cies in prioritizing for further screening and testing. Chemical pesticide actives
are relatively modest in number (826) and have robust toxicological test data to
inform hazard characterization. However, toxicity data on antimicrobials and
pesticidal inerts (http://www.epa.gov/pesticides/regulating/index.htm), HPV
(http://www.epa.gov/opptintr/chemrtk/hpvchmlt.htm), inventory update rule
(http://www.epa.gov/oppt/iur/index.htm), and drinking water contaminate
candidate list (http://www.epa.gov/safewater/mcl.htmlttmcls) chemicals are
less complete, making prioritization more difficult.
and drinking water contaminant candidate list chemicals (Fig. 1)
generally have limited toxicological data available for hazard
and risk assessments. As ToxCast moves beyond initial proof-
of-concept, thousands of environmental chemicals from various
EPA domains can be considered for the ToxCast program.
Looking beyond U.S. borders, there may be utility for a program
like ToxCast in Europe's Registration, Evaluation, and Autho-
rization of Chemicals (REACH) program. In 2003, the Euro-
pean Commission adopted the REACH proposal as a new
regulatory framework for chemicals manufactured or imported
at more than 1 ton per year. After final adoption of the REACH
legislation, which is expected by the end of 2006, REACH
legislation is likely to be in force by mid 2007 (http://
ec.europa.eu/environment/chemicals/reach/reach_intro.htm).
For the ToxCast proof-of-concept, conventional chemical
pesticide actives are an ideal set of compounds for a number of
reasons. Currently, registered pesticide actives are relatively
modest in number (about 800), yet these actives represent
a fairly diverse set of structural classes (Table 3). Furthermore,
these chemicals were all designed to have biological activity
targeted against a pest species. This biological activity
promises to provide a diverse range of positive results in the
biochemical and cellular assays of ToxCast. Most importantly
for the purposes of ToxCast, the pesticide actives have a wealth
of uniform toxicological test data to inform hazard character-
ization. This existing information, and EPA's evaluation and
interpretation of these data in current risk assessments, will
Previous
TOC
-------
DIX ET AL.
TABLE 3
Chemical Pesticide Actives Use Categories and
Chemical Classes"
Pesticide category
Total
Pesticide classification
Classified
Fungicide
Herbicide
Insecticide
Unclassified
Pesticide-use categories
Food use
Antimicrobial
Number of unique
chemicals
826
559
134
173
192
267
270
219
Number of chemical
classes
163
112
58
66
62
102
100
76
"Categories and classes derived from EPA's OPP Information Network.
provide the critical and necessary context for interpreting
ToxCast data.
TABLE 4
Examples of Key Targets, Mechanisms, and Toxicities for the
ToxCast Chemical Prioritization Program
Molecular targets
Biological mechanisms
or pathways
Kinases, phosphatases,
proteases
G-protein-coupled
receptors, steroid, and
nonsteroidal nuclear
receptors
Gamma-amino-butyric
acid receptors,
ion channels
DNA, topoisomerases,
ligases, and helicases
Caspases, cyclins
(dependent kinases)
Signal transduction
pathways
Receptor-mediated
pathways
Apoptosis
Oxidative
stress
DNA recombination
and repair
Cell cycle
Toxicities
Teratogenicity
Reproductive
(Developmental)
neurotoxicity
(Developmental)
immunotoxicity
Genotoxicity
Carcinogenicity
DESIGN OF THE TOXCAST RESEARCH PROGRAM
ToxCast is designed to populate multiple data domains of
increasing biological relevance and experimental cost, from
in silico to in vitro, and perhaps even in vivo with nonmam-
malian model organisms (Fig. 2). Associations between data
domains and across chemicals can be made in order to generate
bioactivity fingerprints and to group or bin chemicals. It is these
larger patterns gleaned from bioactivity profiling across a broad
range of assays that can be associated with either chemical
structure (Fliri et al, 2005b,c; Melnick et al, 2006), or with
known toxicity of reference chemicals in a proof-of-concept
Physical-Chemical Indicators'?^.
Bio-Computational lndicators\%^
I Biochemical Based Indicators
Cellular Based Indicators
FIG. 2. The multiple data domains that will comprise the ToxCast research
program increase in both biological relevance and cost, along the continuum
from in silico to in vitro to in vivo models. Associations between data domains,
and across chemicals, will be used to bin or group chemicals with similar
bioactivity profiles.
study. It is from these associations or correlations between chem-
ical structure, bioactivity profile, and toxicity outcome that the
predictive power of ToxCast will be derived. The chemical and
biological diversity in ToxCast will afford an opportunity to
establish qualitative connections and quantifiable linkages
between chemical structure, biological activity, and known or
predicted toxicity.
In the course of identifying screening targets (Table 4) or
assays suitable for ToxCast (Table 5), two key considerations
are the technical and economic feasibility of pursuing that
target or assay for thousands of chemicals. Rather than just the
TABLE 5
Assay and Chemical Selection Criteria for ToxCast
Assays
Chemicals (for proof-of-concept)
Capacity for thousands of chemicals
Broad spectrum of genes, proteins,
and metabolites
Optimized for HTS or 'omics
Linkage to known lexicological
MOA
If in vivo, nonmammalian models
Ability to test concentration-response
Metabolic capability
Minimal false negatives
Currently available
Completeness, quality, and relevancy
of toxicological data
Use in other toxicity testing
programs (e.g., NTP HTS, HPV)
MW, solubility, ALogP—suitability
for HTS
Range of toxicities, modes of action
Structural chemical classes, and
enough member per class for
grouped analysis
Chemical use classes
Role of metabolism in toxicity
Suitable purity, enantiomeric forms
available
Note. MW, molecular weight.
Previous
TOC
-------
TOXCAST CHEMICAL PRIORITIZATION
drugable proteome, ToxCast sets out to survey a broad spec-
trum of genes, proteins, and metabolites that comprise the cel-
lular "interactome." Pathway-based analyses may also identify
effects on higher level signaling, in addition to discrete targets
within the cellular interactome. These pathways could serve as
a good middle ground between biochemical or other target-
focused assays and more phenomenological, phenotypic, or high-
content assays. Thus, the range of potential targets and assays
is very broad, and increasing biological relevance will have to
be balanced against increasing cost for various data domains
(Fig. 2). Two abiding requirements for ToxCast assays will be
the ability to minimize false negatives relating to hazards to
human health and the current availability of these assays from
reliable sources.
The majority of ToxCast data will come from a diverse series
of assay types that collectively evaluate a broad spectrum of
bioactivities (Fig. 3). Like prior examples in the literature (Fliri
et al, 2005c; Janzen and Hodge, 2006; Melnick et al, 2006),
ToxCast data will include numerous HTS assays delineating
biological effects. Eventually, ToxCast is designed to research
thousands of chemicals, requiring a managed library of envi-
ronmental chemicals and sophisticated chemical and biological
informatics to identify meaningful data associations. HTS
biochemical assays will be supplemented by cellular assays
for more complex biological effects and toxicities (Table 4,
Fig. 3). The potential of high-content screening (Borchert et al.,
2005; O'Brien et al., 2006) and microelectronic monitoring
(Xing et al., 2005) of cells for detection of specific toxicities
will be researched. In addition, the use of the nematode
Caenorhabditis elegans (in collaboration with the National
Institute of Environmental Health Sciences [NIEHS]; Schwartz
et al., 2004), and the zebrafish Danio rerio will be explored as
models of mammalian toxicity. Biological samples from these
various in vitro and in vivo assays will also be utilized for
supplemental genomics and metabolomics.
FIG. 3. Data generation for the ToxCast program will begin with a man-
aged chemical library, then flow from seven integrated types of analyses
evaluating a broad spectrum of bioactivities. These data will be interpretively
linked within the ToxCast database and a structured strategy developed to
predict toxicity.
While much of the ToxCast data are likely to come from
HTS enzyme and receptor assays, an important complement to
these data will be derived from assays using complex formats
of human, nonhuman primate, or rodent cells for detecting bio-
transformation and complex toxicities. These are capable of
detecting secondary effects (e.g., altered membrane perme-
ability) resulting from chemically induced perturbations of the
interactome. For example, in vitro primary hepatocyte models
of the liver are commonly used to screen for metabolism and
toxicity of xenobiotics, but primary hepatocytes rapidly lose
liver-specific functions under standard cell culture conditions.
Advances in tissue engineering and in silica modeling are
enabling development of novel engineered approaches (Allen
et al., 2005; Sivaraman et al., 2005) that could improve
chemical hazard testing by recreating the three-dimensional
microscale of the liver. Such tissue engineering raises new
possibilities for the study of complex toxicological processes
in vitro (Griffith and Swartz, 2006), and the convergence of
HTS and toxicogenomic data with systems biology is creating
opportunities for developing bioengineered and computational
models that more realistically replicate hepatic architecture and
function. Toward this end, ToxCast could provide critical data
for defining processes such as nuclear receptor-mediated reg-
ulation of xenobiotic-metabolizing enzymes, useful to systems
biology models of toxicity. It is through systems biology that
the issue of metabolism and biotransformation may be best
addressed within ToxCast. A recent workshop organized by the
European Center for the Validation of Alternative Methods
(Coecke et al, 2006) emphasized the need to account for
biotransformation with appropriate methods and to consider
how such information can be incorporated into computer
models for hazard identification.
Toxicogenomic assays, specifically the highly parallel pro-
filing of gene expression and cellular metabolites in ToxCast
biological samples can be an important adjunct or alternative to
biochemical HTS profiling. For example, nuclear receptor bind-
ing and activity could be assessed by monitoring expression of
suites of genes that are the transcriptional targets for specific
nuclear receptors of interest. The appropriate target genes can
be identified by a complementary suite of positive internal
control ligands (e.g., testosterone for the androgen receptor, ri-
fampicin for the human pregnane X receptor) utilized in
ToxCast cellular assays. Receptor activities could then be as-
sessed based on expression of receptor-modulated genes and
utilized as an efficient toxicogenomics in vitro assay for
characterization of environmental chemicals (Yang et al.,
2006a).
SELECTION OF PROOF-OF-CONCEPT CHEMICALS
The essential first step for the ToxCast program is to conduct
a demonstration phase using reference chemicals that have
an existing, rich toxicological database (i.e., registered chemical
Previous
TOC
-------
10
DIXETAL.
pesticide actives). Several hundred reference chemicals repre-
sentative of differing structural classes and phenotypic
outcomes (e.g., carcinogens, reproductive toxicants, neuro-
toxicants) will need to be evaluated in ToxCast's wide net of
assays and endpoints for this proof-of-concept. As the program
matures, the assays and endpoints may be narrowed or modified
based on predictive value, derived from associations between
various data domains and the known toxicological properties of
the reference chemicals. From this proof-of-concept, a broader
strategy for identifying toxicity potential, minimizing false
negatives, and prioritizing subsequent testing can be developed
for larger number of environmental chemicals having limited
toxicological data. This proof-of-concept will be especially
important because of the challenges of ToxCast, as compared to
conventional drug discovery, attributable in part to the diversity
of environmental chemicals and issues relating to solubility,
volatility, or confounding cytotoxicity.
Working from EPA databases, 826 conventional chemical
pesticide actives that are currently registered or undergoing
registration were identified. Of these 826, at least 270 are food-
use pesticides that have the most extensive testing require-
ments. Table 3 presents EPA's Office of Pesticide Programs
(OPP) use categories and chemical classes for the majority of
these pesticides. Table 5 lists the general selection criteria that
were used for ranking chemical pesticide actives as candidates
for the ToxCast proof-of-concept. Structural annotation was
added to these pesticide actives, and further chemoinformatic
analysis was conducted using LeadScope Enterprise (http://
www.leadscope.com; Table 6). These 785 chemicals were
characterized into 101 structural classes, with 28 of these
classes being singletons. For proof-of-concept, the chemicals
were prioritized based on several criteria. High priority was
generally given to those chemicals in common with the 1408
chemicals that the National Toxicology Program (NTP) has
provided NCGC in early 2006 for HTS. Compatibility with
standard HTS assays was also considered; thus, low prior-
ity was given to inorganics, organometallics, high ALogP
TABLE 6
Chemical Pesticide Actives Properties and Structural Classes
Relative to use in ToxCast Proof-of-Concept
Physical-chemical properties"
# Unique chemicals
Common to NTP-1408"
MW > 150, solubility > 2, ALogP < 5
100 6
Total
92
328
219
146
785
Note. MW, molecular weight.
"Physical-chemical properties and structural classes calculated using
Leadscope Enterprise v2.3.6-2 (Leadscope Inc., Columbus, OH).
*NTP provided 1408 chemicals to NCGC for HTS.
(octanol/water partitioning), and molecular weights < 150. The
328 prioritized chemicals were secondarily ranked in descend-
ing order of representation in other toxicological databases
annotated in the EPA DSSTox Structure Data File collection
(http://www.epa.gov/ncct/dsstox/index.html), or in other EPA
programs (e.g., industrial HPV chemicals) that correlate in
some fashion to ToxCast. A small minority of inorganics and
organometallics are included in this set of 328 chemicals
because of their relevance to other toxicological programs. The
remaining chemicals included an additional 219 chemical
pesticides that might be suited to HTS.
INTEGRATING CHEMICAL AND BIOLOGICAL DATA TO
FORECAST TOXICITY
Within ToxCast, data will be generated on an environmental
chemical library using numerous types of assays evaluating
a broad spectrum of bioactivities (Fig. 3). These data will need
to be relationally linked within the ToxCast database to other
physical-chemical, toxicological, and in silico information, and
a structured strategy developed to predict toxicity based on this
entire data set. This structured strategy will be forged upon the
known toxicities of the proof-of-concept chemical pesticides.
We are currently in the process of collecting toxicological
data on pesticides and working with the OPP on how to ac-
curately and precisely capture this information into a relational
database. The OPP evaluates submitted toxicological studies in
a standardized review process, which is captured in a Data
Evaluation Record (DER). Information is being culled from
DERs on endpoints, dose-response, and critical effects in mam-
malian test species for approximately 400 chemical pesticides.
Further
Screening
And
Testing
HIGH
MEDIUM
FIG. 4. The application of ToxCast data to the process of prioritizing
environmental chemicals based on hazard prediction: chemicals given a low
priority may enter into no further testing, medium priorities may be recycled
into ToxCast for further evaluation, and high priorities recommended for
further screening and testing.
Previous
TOC
-------
TOXCAST CHEMICAL PRIORITIZATION
11
The DERs being used are primarily from neurotoxicity,
developmental, reproductive, subchronic, chronic, and cancer
guideline toxicology tests. The OPP conventional toxicology
for the proof-of-concept pesticides will complement the
chemoinformatic, HTS, and toxicogenomic information in
the ToxCast database, allowing us to develop and validate
ToxCast's predictive power. In addition, toxicological data
from other EPA Programs (e.g., HPV Challenge) and the NTP
will also be helpful in developing ToxCast. Throughout the
course of methods and data development, our goal is to keep
ToxCast a public and transparent enterprise.
Another ongoing informatics effort is aimed at generating,
collating, reviewing, and organizing unambiguous definitions
of chemical identity and structure for the various environmen-
tal chemical domains relevant to ToxCast and EPA. To ac-
complish this, we are building on the DSSTox project. This will
also aid in the identification of other potentially useful sources
of data relative to the ToxCast candidate chemical list, as well
as help identify structurally similar chemicals for which
toxicity or bioassay data might be available.
Figure 4 presents a flowchart for applying ToxCast data and
predictions to the process of prioritizing chemicals. Hazard
prediction represents both the primary goal and the key bio-
informatics and chemoinformatics challenge of this approach,
and the value of such an approach is self-evident so long as false
negatives are minimized. Over the past several years, a number of
studies have been published presenting alternative, in vitro, and
in some cases HTS methods for integrated testing of chemicals
for bioactivity and associations with toxicity or side effects. One
example relevant to ToxCast was an integrated, tiered approach
using computational and experimental in vitro data for hazard
assessment, although limited to only 10 environmental chemicals
(Gubbels-van Hal et al, 2005). The hazard assessment for these
10 substances was performed on the basis of available nonanimal
data, quantitative structure activity relationship, physiologically
based pharmacokinetic modeling, and additional new in vitro
testing. Based on these data, predictions of various toxicities
were made and then compared with prior in vivo testing to
demonstrate at least a partial success. However, the limited
number of chemicals included in the study of Gubbels-van Hal
et al. did not allow conclusions to be drawn for the thousands of
chemicals subject to REACH. It is apparent that methods
compatible with larger numbers of chemicals, which do not lead
to substantially higher costs for industry, need to be developed.
We suggest that HTS technologies, larger chemical libraries, and
expanded data analysis techniques may accomplish these
broader goals within ToxCast.
CONCLUSIONS
The strategy of ToxCast encompasses a diverse range of data
types. No single assay or endpoint will have a large impact on
interpretation of the fingerprint or bioactivity profile. It will be
the overall pattern across many assays and data types that will be
the predictor of toxicity used for prioritizing chemicals. This
will be the main goal of ToxCast, taking advantage of HTS and
toxicogenomic technologies for bioactivity profiling of envi-
ronmental chemicals related in structure or mechanism of
action. Although the primary purpose is not to identify mech-
anisms of action of environmental toxicants per se, this might be
a future benefit of the program. The availability of a biologically
and chemically based system to categorize chemicals of like
properties and activities will provide EPA Program Offices with
a valuable tool that heretofore has been seriously lacking.
In late 2005, EPA organized the Chemical Prioritization
Community of Practice (CPCP) to provide a forum for dis-
cussing the utility of computational chemistry, HTS, and
various toxicogenomic technologies for chemical priori tization
and Agency use. The CPCP has brought together experts and
interested parties to discuss chemical prioritization research.
This has afforded various groups the opportunity to consider
the ToxCast concept, from EPA Program Offices, to external
stakeholders such as the American Chemistry Council, the Cen-
ter for Alternatives to Animal Testing, CropLife America and
Environmental Defense. In addition, the CPCP has been
helpful in building partnerships and communicating with the
NTP, the NIEHS, and the NCGC.
Many hurdles remain to be cleared by ToxCast as it transits
from concept to proof-of-concept and ultimately to a useful
prioritization tool, including (1) accessing a chemical library
providing coverage of sufficient chemical space, (2) identifying
an upper limit on the per chemical cost of obtaining screening
level data, (3) selecting assays within available resources that
produce predictive bioactivity profiles, (4) evaluating the im-
pact of metabolism on the efficiency and accuracy of assays,
(5) developing a bioinformatic approach to mining ToxCast
data and identifying predictive signatures, and (6) carrying out
a prospective prioritization for chemicals currently entering
a traditional testing process, in such a way that minimizes false
negatives. These hurdles will be the focus of the ToxCast
program over then next few years.
SUPPLEMENTARY DATA
Supplementary data are available online at http://toxsci.
oxfordj ournals .org/.
ACKNOWLEDGMENTS
We thank the many EPA, NTP, and NCGC colleagues who have supported
our initial and ongoing efforts to develop ToxCast. Particular thanks are due to
Maritja Wolf (Lockheed Martin, contractor to the EPA) for chemoinformatics
assistance; Raymond Tice, Cynthia Smith, and Kristine Witt for coordination
with the NTP HTS program; Chris Austin and Jim Inglese for coordination
with the NCGC; Tina Levine, Elizabeth Mendez, Elissa Reaves, and Jess
Rowland, for assistance with EPA/OPP toxicological data; and to Vicki
Dellarco (EPA/OPP) and Phil Sayre (EPA/OPPT) for guidance and helpful
comments in the review of this article.
Previous
TOC
-------
12
DIXETAL.
REFERENCES
Allen, J. W., Khetani, S. R., and Bhatia, S. N. (2005). In vitro zonation and
toxicity in a hepatocyte bioreactor. Toxicol. Set. 84, 110-119.
Austin, C. P., Brady, L. S., Insel, T. R., and Collins, F. S. (2004). NIH Molecular
Libraries Initiative. Science 306, 1138-1139.
Berg, E. L., Kunkel, E. J., Hytopoulos, E., and Plavec, I. (2006). Character-
ization of compound mechanisms and secondary activities by BioMAP
analysis. J. Pharmacol. Toxicol. Methods 53, 67-74.
Bhogal, N., Grindon, C., Combes, R., and Balls, M. (2005). Toxicity testing:
Creating a revolution based on new technologies. Trends Biotechnol. 23,299-307.
Borchert, K. M., Galvin, R. J., Frolik, C. A., Hale, L. V., Halladay, D. L.,
Gonyier, R. J., Trask, O. J., Nickischer, D. R., and Houck, K. A. (2005).
High-content screening assay for activators of the Wnt/Fzd pathway in
primary human cells. Assay Drug Dev. Technol. 3, 133-141.
Bredel, M., and Jacoby, E. (2004). Chemogenomics: An emerging strategy for
rapid target and drug discovery. Nat. Rev. Genet. 5, 262-275.
Coecke, S., Ahr, H., Blaauboer, B. J., Bremer, S., Casati, S., Castell, J., Combes, R.,
Corvi, R., Crespi, C. L., Cunningham, M. L., et al. (2006). Metabolism: A
bottleneck in in vitro toxicological test development. The report and
recommendations of ECVAM workshop 54. Altern. Lab Anim. 34,49-84.
Ekins, S., Berbaum, J., and Harrison, R. K. (2003). Generation and validation of
rapid computational filters for cyp2d6 and cyp3a4. Drug Metab. Dispos. 31,
1077-1080.
Fliri, A. E, Loging, W. T, Thadeio, P. E, and Volkmann, R. A. (2005a).
Analysis of drug-induced effect patterns to link structure and side effects of
medicines. Nat. Chem. Biol. 1, 389-397.
Fliri, A. E, Loging, W. T, Thadeio, P. E, and Volkmann, R. A. (2005b).
Biological spectra analysis: Linking biological activity profiles to molecular
structure. Proc. Natl. Acad. Sci. USA. 102, 261-266.
Fliri, A. E, Loging, W. T, Thadeio, P. E, and Volkmann, R. A. (2005c).
Biospectra analysis: Model proteome characterizations for linking molecular
structures and biological response. J. Med. Chem. 48, 6918-6925.
Griffith, L. G., and Swartz, M. A. (2006). Capturing complex 3D tissue
physiology in vitro. Nat. Rev. Mol Cell Biol. 7, 211-224.
Gubbels-van Hal, W. M., Blaauboer, B. J., Barentsen, H. M., Hoitink, M. A.,
Meerts, I. A., and van der Hoeven, J. C. (2005). An alternative approach for
the safety evaluation of new and existing chemicals, an exercise in integrated
testing. Regul. Toxicol. Pharmacol. 42, 284-295.
Janzen, W. P., and Hodge, C. N. (2006). A chemogenomic approach to
discovering target-selective drugs. Chem. Biol. Drug Des. 67, 85-86.
Kikkawa, R., Fujikawa, M., Yamamoto, T, Hamada, Y., Yamada, H., and
Horii, I. (2006). In vivo hepatotoxicity study of rats in comparison with
in vitro hepatotoxicity screening system. J. Toxicol. Sci. 31, 23-34.
Klekota, J., Brauner, E., Roth, F. P., and Schreiber, S. L. (2006). Using high-
throughput screening data to discriminate compounds with single-target
effects from those with side effects. J. Chem. Inf. Model 46, 1549-1562
Macarron, R. (2006). Critical review of the role of HTS in drug discovery. Drug
Discov. Today 11, 277-279.
Melnick, J. S., Janes, J., Kim, S., Chang, J. Y, Sipes, D. G., Gunderson, D.,
Jarnes, L., Matzen, J. T, Garcia, M. E., Hood, T. L., et al. (2006). An efficient
rapid system for profiling the cellular activities of molecular libraries. Proc.
Natl. Acad. Sci. USA. 103, 3153-3158.
O'Brien, S. E., and de Groot, M. J. (2005). Greater than the sum of its
parts: Combining models for useful ADMET prediction. J. Med. Chem. 48,
1287-1291.
O'Brien, P. J., Irwin, W., Diaz, D., Howard-Cofield, E., Krejsa, C. M.,
Slaughter, M. R., Gao, B., Kaludercic, N., Angeline, A., Bernardi, P., et al.
(2006). High concordance of drug-induced human hepatotoxicity with in
vitro cytotoxicity measured in a novel cell-based model using high content
screening. Arch. Toxicol. 80, 580-604.
Poroikov, V V, Filimonov, D. A., Ihlenfeldt, W. D., Gloriozova, T. A., Lagunin,
A. A., Borodina, Y. V, Stepanchikova, A. V, and Nicklaus, M. C. (2003).
PASS biological activity spectrum predictions in the enhanced open NCI
database browser. J. Chem. Inf. Comput. Sci. 43, 228-236.
Richard, A. M., Gold, L. S., and Nicklaus, M. C. (2006). Chemical structure
indexing of toxicity data on the internet: moving towards a flat world. Curr.
Opin. Drug Discov. Dev. 9, 314-325.
Scherf, U., Ross, D. T, Waltham, M., Smith, L. H., Lee, J. K., Tanabe, L., Kohn,
K. W., Reinhold, W. C., Myers, T. G., Andrews, D. T, et al. (2000). A gene
expression database for the molecular pharmacology of cancer. Nat. Genet.
24, 236-244.
Schwartz, D. A., Freedman, J. H., and Linney, E. A. (2004). Environmental
genomics: A key to understanding biology, pathophysiology and disease.
Hum. Mol. Genet. 13, R217-224.
Sivaraman, A., Leach, J. K., Townsend, S., lida, T, Hogan, B. J., Stolz, D. B.,
Fry, R., Samson, L. D., Tannenbaum, S. R., and Griffith, L. G. (2005). A
microscale in vitro physiological model of the liver: Predictive screens for
drug metabolism and enzyme induction. Curr. Drug Metab. 6, 569-591.
Smith, S. C., Delaney, J. S., Robinson, M. P., and Rice, M. J. (2005). Targeting
inputs and optimising HTS for agrochemical discovery. Comb. Chem. High
Throughput Screen. 8, 577-587.
Tietjen, K., Drewes, M., and Stenzel, K. (2005). High throughput screening
in agrochemical research. Comb. Chem. High Throughput Screen. 8, 589-
594.
Walum, E., Hedander, J., and Garberg, P. (2005). Research perspectives for pre-
screening alternatives to animal experimentation—On the relevance of
cytotoxicity measurements, barrier passage determinations and high
throughput screening in vitro to select potentially hazardous compounds in
large sets of chemicals. Toxicol. Appl. Pharmacol. 207, 393-397.
Xing, J. Z., Zhu, L., Jackson, J. A., Gabos, S., Sun, X. J., Wang, X. B., and
Xu, X. (2005). Dynamic monitoring of cytotoxicity on microelectronic
sensors. Chem. Res. Toxicol. 18, 154-161.
Yang, Y, Abel, S. J., Ciurlionis, R., and Waring, J. F. (2006a). Development of
a toxicogenomics in vitro assay for the efficient characterization of
compounds. Pharmacogenomics 7, 177-186.
Yang, C., Richard, A. M., and Cross, K. P. (2006b). The art of data mining the
minefields of toxicity databases to link chemistry to biology. Curr. Comput.
Aided Drug Des. 2, 1-19.
Previous
TOC
-------
A Balanced Accuracy Fitness Function Leads to Robust
Analysis using Grammatical Evolution Neural Networks in
the Case of Class Imbalance
Nicholas E. Hardison
Bioinformatics Research Ctr.
Department of Statistics
North Carolina State University
Raleigh, NC 27606
nhardis@ncsu.edu
Theresa J. Fanelli
Ctr. for Human Genetics Research
Department of Molecular Physiology &
Biophysics; Vanderbilt University
Nashville, TN 37232
tjf5004@psu.edu
Scott M. Dudek
Ctr. for Human Genetics Research
Department of Molecular Physiology &
Biophysics; Vanderbilt University
Nashville, TN 27232
dudek@chgr.mc.vanderbilt.edu
David M. Reif
National Ctr. for Computational
Toxicology; U.S. Environmental
Protection Agency
RTP, NC 27711
Marylyn D. Ritchie
Ctr. for Human Genetics Research
Department of Molecular Physiology &
Biophysics; Vanderbilt University
Nashville, TN 37232
reif.david@epa.gov ritchie@chgr.mc.vanderbilt.edu
Alison A. Motsinger-Reif
Bioinformatics Research Ctr.
Department of Statistics
North Carolina State University
Raleigh, NC 27606
motsinger@stat.ncsu.edu
ABSTRACT
Grammatical Evolution Neural Networks (GENN) is a
computational method designed to detect gene-gene interactions
in genetic epidemiology, but has so far only been evaluated in
situations with balanced numbers of cases and controls. Real data,
however, rarely has such perfectly balanced classes. In the current
study, we test the power of GENN to detect interactions in data
with a range of class imbalance using two fitness functions
(classification error and balanced error), as well as data re-
sampling. We show that when using classification error, class
imbalance greatly decreases the power of GENN. Re-sampling
methods demonstrated improved power, but using balanced
accuracy resulted in the highest power. Based on the results of this
study, balanced error has replaced classification error in the
GENN algorithm.
Categories and Subject Descriptors
Genetics-Based Machine Learning and Learning Classifier
Systems.
General Terms
Algorithms
1. INTRODUCTION
Grammatical Evolution Neural Networks (GENN) uses
grammatical evolution to evolve neural networks to detect gene-
Permission to make digital or hard copies of all or part of this work for
personal or classroom use is granted without fee provided that copies are
not made or distributed for profit or commercial advantage and that
copies bear this notice and the full citation on the first page. To copy
otherwise, or republish, to post on servers or to redistribute to lists,
requires prior specific permission and/or a fee.
GECCO '08, July 12-16, 2008, Atlanta, Georgia, USA.
Copyright 2008 ACM 1-58113-000-0/00/0004...$5.00.
gene interactions in studies of complex human diseases [1].
GENN has shown initial successes in both real and simulated
data, and while these results are encouraging, previous simulation
studies have used datasets with balanced numbers of cases and
controls. Unfortunately, when using standard classification error
as the fitness function, many machine learning methods are not
robust to class imbalance.
To try to solve this problem, investigators have tried techniques
such as re-sampling [2] or altering the fitness metric. One metric
that has been shown to be highly successful is balanced
error/accuracy [3]. This metric has been shown to solve the class
imbalance problem for another approach designed to detect
epistasis-Multifactor Dimensionality Reduction (MDR) [4].
We assessed the performance of GENN on data with varying
levels of class imbalance and show that the power of GENN using
classification error decreases as the controlxase ratio departs from
unity. We compared three methods for addressing this concern:
re-sampling methods (over- and under-sampling) and balanced
accuracy as a fitness function.
2. METHODS
2.1 Grammatical Evolution Neural Networks
The steps of GENN have been previously described in detail [1].
For the purposes of the current study, an option was added to the
configuration file to specify the fitness function used:
classification error (CE) or balanced error (BE). BE is the inverse
of balanced accuracy, defined as the mean of sensitivity and
specificity [3]:
Balanced Accuracy = (sensitivity + specificity)/2 =
'/2 [TP/(TP+FN) + TN/(TN+FP)]
where TP represents true positives, TN represents true negatives,
FP represents false positives, and FN represents false negatives.
This formula equally weights the errors within each class. In the
case of balanced data, this is equivalent to standard CE.
-------
2.2 Data Simulation
The intention of the data simulations for this power study was to
mimic gene-gene interaction, or epistasis, in case-control genetic
data to evaluate GENN using penetrance functions. Penetrance
defines the probability of disease given a particular genotype
combination by modeling the relationship between genetic
variations and disease risk. We used two well-described purely
epistatic models, where the heritability (the proportion of trait
variance due to genetics) -5%. The first is referred to as the XOR
model, and the second is referred to as the ZZ model [5]. Both are
nonlinear models with no marginal main effects. Software
described by Moore et al [5] was used to simulate the data.
For both models, we simulated data with a range of controlxase
ratios and sample sizes. For the first set of simulations, the total
number of individuals in the dataset was held constant, at two
different total sample sizes: 600 and 1200. For each sample size,
three control:case ratios were simulated: 1:1, 2:1, and 4:1. To
ensure the results seen were due to class imbalance instead of
decreasing numbers of cases, a second set of simulations was
done, holding the number of cases constant at 300 and 600.
Again, for each number of cases, three control:case ratios were
simulated. For each set of parameters, 100 replicates were
simulated. Each dataset had a total of 100 SNPs, two of which
were functional in predicting disease. For the models with
imbalanced control:case ratios, re-sampling was performed. In the
case of under-sampling (US), controls were randomly removed
until a ratio of 1:1 was achieved. In the case of over-sampling
(OS), cases were randomly re-sampled until a 1:1 ratio was
achieved.
2.3 Data Analysis
GENN was used to analyze all epistasis models with classification
error, balanced error, or classification error in combination with
data re-sampling. Parameter settings remained identical between
the analyses and included: 4 denies, migration every 25
generations, population size of 200 per deme, 400 generations,
crossover rate of 0.9, and a reproduction rate of 0.1. Power for
all analyses is reported as the number of times GENN correctly
identified the correct loci with no false positives over 100
datasets.
3. RESULTS
Tables 1 and 2 show the results for all analyses, with several
apparent trends. Using classification error (CE), increased
imbalanced ratios greatly decreases the power of GENN. The
power of GENN greatly improves when OS is used. With US, a
Table 1. Results for constant sample size simulations for
different control: case ratios (CCR).
Table 2. Results for constant case number simulations.
Total
Samples
600
1200
CCR
1:1
2:1
4:1
1:1
2:1
4:1
XOR Power (%)
CE
100
74
3
99
88
6
BE
100
100
100
100
100
100
US
100
87
62
100
98
85
OS
100
98
97
100
99
99
ZZ Power (%)
CE
100
100
63
100
100
59
BE
100
100
99
100
100
100
US
100
96
74
100
99
94
OS
100
100
99
100
100
100
Case
Count
300
600
CCR
1:1
2:1
4:1
1:1
2:1
4:1
XOR Power (%)
CE
100
80
2
99
83
3
BE
100
100
100
100
99
100
US
100
91
86
100
93
71
OS
100
96
95
100
99
98
ZZ Power (%)
CE
100
100
49
100
100
36
BE
100
100
100
100
100
100
US
99
99
90
100
98
92
OS
100
99
96
100
99
97
marked decrease in power in smaller datasets with large class
imbalance is seen. This trend is ameliorated somewhat in larger
datasets, as well as the datasets with fixed numbers of cases.
Most significantly, for all models analyzed, power recovers
completely when using balanced error (BE).
4. DISCUSSION
From these results, we conclude that balanced error should be
used as the fitness metric in GENN instead of classification error,
as it outperforms standard classification error and re-sampling
methods. Additionally, since balanced error and classification
error are mathematically equivalent in when data is balanced,
there is no disadvantage to using balanced error in balanced data.
5. ACKNOWLEDGMENTS
This work was supported by National Institutes of Health grants
HL65962, GM62758, and AG20135. We would also like to thank
Jason H. Moore and Digna R. Velez for helpful discussions on
class imbalance. This paper has been reviewed and approved for
publication according to US EPA policy but does not necessarily
represent the views of the Agency.
6. REFERENCES
[1] Motsinger-Reif A.A., Dudek S.M., Hahn L.W., Ritchie M.D.
Comparison of Approaches for Machine Learning
Optimization of Neural Networks for Detecting Gene-Gene
Interactions in Genetic Epidemiology. Genet. Epidemiol.,
Epub ahead of print. (2008)
[2] Japkowicz N., Stephen S. The Class Imbalance Problem: A
Systematic Study. Intelligent Data Analysis Journal, 6
(2002)429-450.
[3] Powers R., Goldszmidt M., Cohen I. Short term performance
forecasting in enterprise systems. Hewlett-Packard
Development Company Technical Reports, Computer
Science Department, Stanford University, Stanford, CA.
(2005)
[4] Velez D., White B.W., Motsinger A.A., Bush W.S., Ritchie
M.D., Moore J.H. A Balanced Accuracy Metric for Epistasis
Modeling in Imbalanced Datasets using Multifactor
Dimensionality Reduction. Genet. Epidemiol. 4 (2007) 306-
15.
[5] Moore, J., Hahn, L., Ritchie, M., Thornton, T., White, B.
Application of genetic algorithms to the discovery of
complex models for simulation studies in human genetics. In
Proceedings of the Genetic and Evolutionary Computation
Conference. (New York, USA, July 9-13, 2002). Morgan
Kaufman, San Francisco, CA, 2002, 1150-1155.
-------
FY2004 "New Start" Award Bibliography
Project Title: Linkage of Exposure and Effects Using Genomics, Proteomics, and
Metabolomics in Small Fish Models
Peer Reviewed Publications:
Ankley, G.T., K.M. Jensen, E.J. Durban, E.A. Makynen, B.C. Butterworth, M.D. Kahl,
D.L. Villeneuve, A. Linnum, L.E. Gray, M. Garden, V.S. Wilson. 2005. Effects of two
fungicides with multiple modes of action on reproductive endocrine function in the
fathead minnow (Pimephales promelas). Toxicol. Sci. 86, 300-308.
Ankley, G.T., K.M. Jensen, M.D. Kahl, E.A. Makynen, L.S. Blake, K.J. Greene, R.D.
Johnson and D.L. Villeneuve. 2007. Ketoconazole in the fathead minnow (Pimephlaes
promelas): reproductive toxicity and biological compensation. Environ. Toxicol. Chem.
26, 1214-1223.
Ankley, G.T., D.H. Miller, K.M. Jensen, D.L. Villeneuve and D. Martinovic. 2008.
Relationship of plasma sex steroid concentrations in female fathead minnows to
reproductive success and population status. Aquat. Toxicol. 88, 69-74.
Ankley, G.T., D. Bencic, M. Breen, T.W. Collette, R. Connolly, N.D. Denslow,
S. Edwards, D.R. Ekman, K.M. Jensen, J. Lazorchak, D. Martinovic, D.H. Miller, E.J.
Perkins, E.F. Orlando, N. Garcia-Reyero, D.L. Villeneuve, R.-L.Wang , and K.
Watanabe. 2009. Endocrine disrupting chemicals in fish: Developing exposure
indicators and predictive models of effects based on mechanisms of action. Aquat.
Toxicol. 92, 168-178.
Breen, M.S., D.L. Villeneuve, M. Breen, G.T. Ankley and R.B. Conolly. 2007.
Mechanistic computational model of ovarian steroidogenesis to predict biochemical
responses to endocrine active compounds. Ann. Biomed. Engin. 35, 970-981.
Ekman, D.R., Q. Teng, K.M. Jensen, D. Martinovic, D.L. Villeneuve, G.T. Ankley and T.
W. Collette. 2007. NMR analysis of fathead minnow urinary metabolites: a potential
approach for studying impacts of chemical exposures. Aquat. Toxicol. 85, 104-112.
Ekman, D.R., Q. Teng, D.L. Villeneuve, M.D. Kahl, K.M. Jensen, E.J. Durban, G.T.
Ankley and T.W. Collette. 2008. Investigating compensation and recovery of fathead
minnow (Pimephales promelas) exposed to 17a-ethynylestradiol with metabolite
profiling. Environ. Sci. Tecnhol. 42, 4188-4195.
Ekman, D.R., Q. Teng, D.L. Villeneuve, M.D. Kahl, K.M. Jensen, E.J. Durban, G.T.
Ankley and T.W. Collette. 2009. Profiling lipid metabolites yields unique information on
gender- and time-dependent responses of fathead minnows (Pimephales promelas)
exposed to 17a-ethynylestradiol. Metabolomics 5, 22-32.
Previous I TOC
-------
Garcia-Reyero, N., D.L. Villeneuve, K.J. Kroll, L. Liu, E.F. Orlando, K.H. Watanabe,
M.S. Sepulveda, G.T. Ankley and N.D. Denslow. 2009. Expression signatures for a
model androgen and antiandrogen in the fathead minnow ovary. Environ. Sci. Technol.
43,2614-2619.
Garcia-Reyero, N., K.J. Kroll, L. Liu, E.F. Orlando, K.H. Watanabe, M.S. Sepulveda,
D.L. Villeneuve, E.J. Perkins, G.T. Ankley and N.D. Denslow. 2009. Gene expression
responses in male fathead minnows exposed to binary mixtures of an estrogen and
antiestrogen. BMC Genomics, In Press.
Johns, S.M., M.D. Kane, N.D. Denslow, K.H. Watanabe, E.F. Orlando, D.L. Villeneuve,
G.T. Ankley and M.S. Sepulveda. 2009. Characterization of ontogenetic changes in
gene expression in the fathead minnow (Pimephales promelas). Environ. Toxicol.
Chem. 28, 873-880.
Martyniuk, C.J., S. Alvarez, S. McClung, D.L. Villeneuve, G.T. Ankley and N.D.
Denslow.2009. Quantitative proteomic profiles of androgen receptor signaling in the
liver of fathead minnows (Pimephales promelas) J. Proteome Res. In Press.
Martinovic, D., L.S. Blake, E.J. Durhan, K.J. Greene, M.D. Kahl, K.M. Jensen, E.A.
Makynen, D.L. Villeneuve and G.T. Ankley. 2008. Characterization of reproductive
toxicity of vinclozolin in the fathead minnow and co-treatment with an androgen to
confirm an anti-androgenic mode of action. Environ. Toxicol. Chem. 27, 478-488.
Miller, D.H., K.M. Jensen, D.L. Villeneuve, M.D. Kahl, E.A. Makynen, E.J. Durhan and
G.T. Ankley. 2007. Linkage of biochemical responses to population-level effects: a
case study with vitellogenin in the fathead minnow (Pimephlaes promelas). Environ.
Toxicol. Chem. 26, 521-527.
Perkins, E.J., N. Garcia-Reyero, D.L. Villeneuve, D. Martinovic, S.M. Brasfield, L.S.
Blake, J.D. Brodin, N.D. Denslow and G.T. Ankley. 2008. Perturbation of gene
expression and steroidogenesis with in vitro exposure of fathead minnow ovaries to
ketoconazole. Mar. Environ. Res. 66, 113-115.
Villeneuve, D.L., P. Larkin, I. Knoebl, A.L. Miracle, M.D. Kahl, K.M. Jensen, E.A.
Makynen, E.J. Durhan, B.J. Carter, N.D. Denslow and G.T. Ankley. 2007. A graphical
systems model to facilitate hypothesis-driven ecotoxicogenomics research on the brain-
pituitary-gonadal axis. Environ. Sci. Technol. 40, 321-330.
Villeneuve, D., L. Blake, J. Brodin, K. Greene, I. Knoebl, A. Miracle, D. Martinovic and
G.T. Ankley. 2007. Transcription of key genes regulating gonadal steroidogenesis in
control and ketoconazole- or vinclozolin-exposed fathead minnows. Toxicol. Sci. 98,
395-407.
Villeneuve, D.L., L.S. Blake, J.D. Brodin, J.E. Cavallin, E.J. Durhan, K.M. Jensen, M.D.
Kahl, E.A. Makynen, D. Martinovic, N.D. Mueller and G.T. Ankley. 2008. Effects of a
Previous I TOC
-------
3(3-hydroxysteroid dehydrogenase inhibitor, trilostane, on the fathead minnow
reproductive axis. Toxicol. Sci. 104, 113-123.
Villeneuve, D.L., N.D. Mueller, D. Martinovic, E.A. Makynen, M.D. Kahl, K.M. Jensen,
E.J. Durhan, J.E. Cavallin, D. Bencicand G.T. Ankley. 2009. Direct effects,
compensation and recovery in female fathead minnows exposed to a model aromatase
inhibitor. Environ. Health Perspect. 117, 624-631.
Villeneuve, D.L., R.-L. Wang, D.C. Bencic, A.D. Biales, D. Martinovic, J.M.
Lazorchak, G. Toth and G.T. Ankley. 2009. Altered gene expression in the brain and
ovaries of zebrafish exposed to the aromatase inhibitor fadrozole: microarray analysis
and hypothesis generation. Environ. Toxicol. Chem. In Press.
Wang, R.-L., A. Biales, D. Bencic, D. Lattier, M. Kostich, D. Villeneuve, G.T. Ankley, J.
Lazorchak and G. Toth. 2008a. DMA microarray application in ecotoxicology:
experimental design, microarray scanning, and factors impacting transcriptional profiles
in a small fish species. Environ. Toxicol. Chem. 27, 652-663.
Wang, R.-L., D. Bencic, A. Biales, D. Lattier, M. Kostich, D. Villeneuve, G.T. Ankley, J.
Lazorchak and G. Toth. 2008b. DMA microarray-based ecotoxicological discovery in a
small fish specbs. Environ. Toxicol. Chem. 27, 664-675.
Watanabe, K.H., K.M. Jensen, E.F. Orlando and G.T. Ankley. 2007. What is
normal? A characterization of the values and variability in reproductive endpoints of the
fathead minnow, Pimephales promelas. Comp. Biochem. Physiol. 146, 348-356.
Watanabe, K.H., Z. Li, K. Kroll, D.L. Villeneuve, N.J. Szabo, E.F. Orlando, M.S.
Sepulveda, T.W. Collette, D.R. Ekman, G.T. Ankley and N.D. Denslow. 2009. A
physiologically-based model of endocrine-mediated responses of male fathead
minnows to17a-ethinylestradiol. Toxicol. Sci. In Press.
Project Title: Simulating Metabolism of Xenobiotic Chemicals as a Predictor of
Toxicity
Peer Reviewed Publications:
Mazur, C. S.; Kenneke, J. F. 2008. Cross-species comparison of conazole fungicide
metabolites using rat and rainbow trout (Onchorhynchus mykiss) hepatic microsomes
and purified human CYP 3A4. Environmental Science and Technology, 42:947-954.
Mazur, C. S.; Kenneke, J.F.; Tebes-Stevens, C. Okino, M. S.; Lipscomb, J. C. 2007. In
vitro metabolism of the fungicide and environmental contaminant trans-bromuconazole
and implications for risk assessment. Journal of Toxicology and Environmental Health,
Part A, 70:1241-1250.
Kenneke, J. F. 2006. Environmental fate and ecological risk assessment for the
reregistration of antimycin A (PC Code 006314), Appendix D: In vitro mammalian
Previous I TOC
-------
metabolism and Appendix G: Summary of antimycin hydrolysis research, U.S. EPA,
Office of Pesticide Programs, Reregistration Eligibility Decision (RED) on Antimycin A
Project Title: Risk Assessment of the Inflammogenic and Mutagenic Effects of
Diesel Exhaust Particulates: A Systems Biology Approach
Peer Reviewed Publications:
Cao, D., Bromberg, PA and Samet, JM (2007). Diesel-induced Cox-2 expression
involves chromatin modification via degradation of HDAC1 and recruitment of p300. Am.
J. Respir. Cell. Mol. Biol.37:232-239.
Cao, D., Tal, T., Graves, L, Gilmour, I., Linak, W., Reed, W., Bromberg, P., and Samet,
J., Diesel Exhaust Particulate (DEP)-lnduced Activation of StatS Requires Activities of
EGFR and SRC in Airway Epithelial Cells, American Journal of Physiology: Lung, Cell,
& Molecular Physiology, 292, L422-L429 (2007).
Cho, S.-H., Yoo, J.-l., Turley, AT., Miller, C.A., Linak, W.P., Wendt, J.O.L., Muggins,
F.E., and Gilmour, M.I., Relationships between Composition and Pulmonary Toxicity of
Prototype Particles from Coal Combustion and Pyrolysis, Proceedings of the
Combustion Institute, 32, in press (2008).
Ciencewicki, J., Gowdy, K., Krantz, Q.T., Linak, W.P., Brighton, L., Gilmour, M.I., and
Jaspers, I., Diesel Exhaust Enhanced Susceptibility to Influenza Infection is Associated
with Decreased Surfactant Protein Expression, Inhalation Toxicology, 19, 1121-1133
(2007).
DeMarini, D.M., Brooks, L.R., Warren, S.H., Kobayashi, T., Gilmour, M.I., and Singh P.,
Bioassay-Directed Fractionation an dSalmonella Mutagenicity of Automobile and Forklift
Diesel Exhaust Particles, Environmental Health Perspectives, 112, 814-819 (2004).
Gottipolu, R.R., Wallenborn, J.G., Karoly, E.D., Schladweiler, M.C., Ledbetter, A.D.,
Krantz, Q.T., Linak, W.P., Nyska, A., Johnson, J.A., Thomas, R., Richards, J.E., Jaskot,
R.H., and Kodavanti, U.P. (2009)., One-month Diesel Exhaust Inhalation Produces
Hypertensive Gene Expression Pattern in Healthy Rats, Environmental Health
Perspectives. 117:38-46.
Gowdy, K., Krantz, Q.T., Daniels, M., Linak, W.P., Jaspers, I., and Gilmour, M.I.,
Modulation of Pulmonary Inflammatory Responses and Anti-microbial Defenses in Mice
Exposed to Diesel Exhaust, Toxicology & Applied Pharmacology, 229, 310-319 (2008).
Linak, W.P., Yoo, J.I., Wasson, S.J., Zhu, W., Wendt, J.O.L, Huggins, F.E., Chen, Y.,
Shah, N., Huffman, G.P., and Gilmour, M.I., Ultrafine Ash Aerosols from Coal
Combustion: Characterization and Health Effects, Proceedings of the Combustion
Institute, 31, 1929-1937 (2007).
Previous I TOC
-------
Reed, W., Gilmour, I., DeMarini, D., Linak, W. and Samet, J. (2008). Gene Expression
Profiles of Human Airway Epithelial Cells Exposed to Diesel Exhaust Particles of
Varying Composition. In Preparation.
Saxena, RK, Williams, W& Gilmour, Ml. (2007) Suppression of basal and cytokine
induced expression of MHC, ICAM 1 and B7 markers on mouse lung epithelial cells
exposed to diesel exhaust particles. Am J Biochem Biotech. 3(4). 187-192.
Saxena, RK., Gilmour, Ml., & MD Hayes. Uptake of diesel exhaust particles by lung
epithelial cells and alveolar macrophages. Biotechnology, 2007, 3(4). 187-192
Singh, P., DeMarini, D.M., Dick, C.A.J., Tabor, D., Ryan, J., Linak, W.P., Kobayashi, T.,
and Gilmour, M.I., Bioassay-Directed Fractionation, Physiochemical Characterization,
and Pulmonary Toxicity of Automobile and Forklift Diesel Exhaust Particles in Mice,
Environmental Health Perspectives, 112(8) 820-825 (2004).
Stevens, T., Krantz, Q.T., Linak, W.P., Hester, S., and Gilmour, M.I., Increased
Transcription of Immune and Metabolic Pathways in Naive and Allergic Mice Exposed
to Diesel Exhaust, Toxicological Sciences, 102(2), 359-370 (2008).
Stevens, T., Linak, WP., Gilmour, Ml. Differential potentiation of allergic lung disease in
mice exposed to chemically distinct diesel samples. Tox Sci. 107(2), 522-534.
Stevens,T, Hester, S, & Gilmour Ml. Differential transcriptional changes in mice
exposed to chemically distinct diesel samples. Submitted.
Tal., T., Bromberg, P.A., Kim, Y. and Samet, J.M. (2008). Tyrosine phosphatase
inhibition induces epidermal growth factor receptor activation in human airway epithelial
cells exposed to diesel exhaust Toxicol. Appl. Pharmacol. 233:382-388.
Project Title: Development of Microbial Metagenomic Markers for Environmental
Monitoring and Risk Assessment
Peer Reviewed Publications:
Lamendella R, Santo Domingo JW, Yannarell AC, Ghosh S, Di Giovanni G, Mackie Rl,
Oerther DB. Evaluation of swine-specific PCR assays used for fecal source tracking and
analysis of molecular diversity of Bacteriodales-swine specific populations. Appl Environ
Microbiol. 2009 Jul 24. [Epub ahead of print]
Lu J, Santo Domingo JW, Hill S, Edge TA. Microbial Diversity and Host-
specificSequences of Canadian Goose Feces. Appl Environ Microbiol. 2009 Jul 24.
Santo Domingo, J.W. and T.A. Edge. 2009. Identification of primary sources of faecal
pollution. In Safe Management of Shellfish and Harvest Waters. G. Rees., K. Pond, D.
Kay and J. Santo Domingo. IWA Publishing, London, UK.
Previous I TOC
-------
Lee, Y.-J., M. Molina, and J.W. Santo Domingo, J.D. Willis, M. Cyterski, D.M. Endale,
and O.C. Shanks. 2008. A temporal assessment of cattle fecal pollution in two
watersheds using 16S rRNA gene-based and metagenome-based assays. Appl.
Environ. Microbiol. 74:6839-6847.
Lu, J. and J.W. Santo Domingo. 2008. Turkey fecal microbial community structure and
functional gene diversity revealed by 16S rRNA gene and metagenomic sequences. J.
Microbiol. 46:469-477.
Lu, J., J.W. Santo Domingo, R. Lamendella, T.Edge, and S.Hill. 2008. Phylogenetic
diversity and molecular detection of gull feces. Appl. Environ. Microbiol. 74: 3969-3976.
Lamendella, R., Santo Domingo J.W., Kelty C, and Oerther DB. 2008. Occurrence of
bifidobacteria in feces and environmental waters. Appl. Environ. Microbiol. 74:575-584.
Santo Domingo, J.W., D.G. Bambic, T.A. Edge, and S. Wuertz. 2007. Quo vadis source
tracking? Towards a strategic framework for environmental monitoring of fecal pollution.
Water Res. 41:3539-3552.
Lu, J., J.W. Santo Domingo, and O.C. Shanks. 2007. Identification of chicken-specific
fecal microbial sequences using a metagenomic approach. Water Res. 41:3561-3574.
Shanks, 0., J.W. Santo Domingo, J. Lu, C.A. Kelty, and J. Graham. 2007. PCR Assays
for the identification of human fecal pollution in water. Appl. Environ. Microbiol. 73:
2416-2422.
Vogel, J.R., D.M. Stoeckel, R. Lamendella, R.B. Zelt, J.W. Santo Domingo, S.R.
Walker, and D.B. Oerther. 2007. Identifying fecal sources in a selected catchment
reach using multiple source-tracking Tools. J. Environ. Qual. 36:718-729.
Lamendella, R., J. W. Santo Domingo, D. Oerther, J. Vogel, and, D. Stoeckel. 2007.
Assessment of fecal pollution sources in a small northern-plains watershed using PCR
and phylogenetic analyses of Bacteroidetes 16S rDNA. FEMS Microbiol. Ecol. 59:651-
660.
Santo Domingo, J.W., Lu, J., Shanks, 0., Lamendella, R., Kelty, C. A., and Oerther,
D.B. "Development of host-specific markers for source tracking using a novel
metagenomic approach," Water Environment Federation, Proceedings of Disinfection
2007, Pittsburg, PA, February 4-7, 2007.
Shanks, 0., J. W. Santo Domingo, R. Lamendella, C.A. Kelty, and J. Graham. 2006.
Competitive metagenomic DMA hybridization identifies host-specific genetic markers in
cattle fecal samples. Appl. Environ. Microbiol. 72:4054-4060.
Previous I TOC
-------
Shanks, 0., J. W. Santo Domingo, and J. Graham. 2006. Use of competitive DMA
hybridization to identify differences in the genomes of two closely related fecal indicator
bacteria. J. Microbiol. Methods. 66:321-330.
Project Title: A Systems Approach to Characterizing and Predicting Thyroid
Toxicity Using an Amphibian Model
Peer Reviewed Publications:
Sternberg, Thoemke, Hornung, Tietge, and Degitz. Regulation of thyroid-stimulating
hormone release from pituitary by t4 during metamorphosis in Xenopus laevis (In
review)
Serrano, Higgins Witthuhn, Korte, Hornung, Tietge, and Degitz In vivo assessment and
potential diagnosis of xenobiotics that perturb the thyroid pathway: Part I. Differential
protein profiling of Xenopus Laevis brain tissues by two-dimensional polyacrylamide gel
electrophoresis and peptide-labeling with isobaric tags for relative and absolute
quantification (iTRAQ) following exposure to model T4 inhibitors. (In review)
Conners, K. , Jorte, J.J., Anderson G., Degitz SJ. Charaterization of thyroid hormone
transporting protein expression during tissue-specific metamorphosis in Xenopus
tropicalis (In review)
Hornung, M.W. Degitz, S.J., Korte, L.M., Olson, J., Kosian, P.A., Linnum, A.L., Tietge,
J.E. Inhibition of thyroid hormone release from cultured amphibian thyroid glands by
methimazole, 6-propylthiouracil, and perchlorate. (Completed, NHEERL In-House
review. To be submitted with the following Hornung et al. paper).
Degitz, S.J., Hornung, M.W., Korte, J.J, Holcombe, G.W, Kosian, P.A., Thoemke, K.R.,
Helbing, C., Tietge, J.E. In vivo and in vitro regulation of genes in the thyroid gland
following exposure to the model T4 synthesis inhibitors methimazole, 6-propylthiouracil,
and perchlorate. (In preparation. To be submitted with above paper).
Hornung, M., Burgess, E. Tandem in vitro and ex vivo thyroid gland assays to screen
xenobiotic chemicals for thyroid hormone synthesis inhibition. (In preparation).
Nichols systems model paper (In preparation).
Tietge Butterworth Kosian Hammermeister Hornung Haselman Degitz Analysis of
Thyroid Hormone and Related lodo-Compounds in Complex Samples by Inductively
Coupled Plasma Emission/Mass Spectrometry. (In preparation)
Tietge, Butterworth, Haselman, Holcombe, Korte, Kosian, Wolfe, Degitz. Early Temporal
Effects of Three Thyroid Hormone Synthesis Inhibitors in Xenopus laevis. (In
preparation)
Previous I TOC
-------
Project Title: Mechanistic Indicators Of Childhood Asthma (Mica): A Systems
Biology Approach For The Integration Of Multifactorial Environmental Health Data
Peer Reviewed Publications:
Kim SJ, Dix DJ, Thompson KE, Murrell RN, Schmid JE, Gallagher JE, Rockett JC.
Effects of storage, RNA extraction, genechip type, and donor sex on gene expression
profiling of human whole blood . Clin Chem. Jun;53(6): 1038-45. (2007)
Vesper, S.,McKinstry C., Haugland., R., Neas, L, Hudgens, E., Heidenfelder, B., and
Gallagher J. Environmental Relative Moldiness Index (ERMIsm) as a Tool to Identify
Mold Related Risk Factors for Childhood Asthma Sci Total Environ. May 1;394(1): 192-6
(2008)
Johnson M, Hudgens E, Williams R, Andrews G, Neas L, Gallagher J, Ozkaynak H. "A
Participant-Based Approach to Indoor/Outdoor Air Monitoring in Community Health
Studies" Journal of Exposure Science and Environmental Epidemiology. (2008), 1-10
(2008).
Cohen Hubal E, Richards A., Shah I, Edwards S, Gallagher J, Kavlock R, Blancato, J
Exposure Science and the US EPA National Center for Computational Toxicology J
Expo Sci Environ Epidemiol. November (2008).
Heidenfelder B,. Reif D, Harkema, JR, Cohen Hubal E, Hudgens,E. Bramble L G.
Wagner G, Harkema JR, Morishita M, Keeler G , Edwards,SW and Gallagher J.
Comparative Microrarray Analysis and Pulmonary Changes in Brown Norway Rats
Exposed to Ovalbumin and concentrated Air Particulates Tox Sci. volume 1082009
March 2 (2009)
Heidenfelder B, Johnson M, Hudgens E, Inmon J, Hamilton R, Neas L, and Gallagher
J, Increased plasma reactive oxidant levels and their relationship to blood cells, total
IgE, and allergen-specific IgE in asthmatic children Journal of Asthma accepted (2009)
Williams AH, Gallagher JE, Hudgens E, Johnson MM, Mukerjee S, Ozkaynak H, Neas
LMN. EPA Observational studies of children's respiratory health in Detroit and
Dearborn, Michigan. Proceedings of AWMA 102nJune 16-19; Detroit, Michigan.(2009)
J. E Gallagher, E A Cohen Hubal, S.W. Edwards Invited book Chapter "Biomarkers of
Environmental Exposure" "Biomarkers of toxicity: A New Era in Medicine Editors Vishal
S. Vaidya and Joseph V. Bonventre Publisher: John Wiley and Sons, Inc. October 1,
(2009)
Markey M. Johnson, Ron Williams, Zhihua Fan, Lin, Edward Hudgens, Jane Gallagher,
Alan Vette, Lucas Neas, Haluk Ozkaynak Indoor and outdoor concentrations of nitrogen
dioxide, volatile organic compounds, and polycyclic aromatic hydrocarbons among
MICA-Air households in Detroit, Michigan submitted AWMA (2009)
Previous I TOC
-------
Gallagher, J Reif, D; Heidenfelder, B Neas, L; Hudgens, E Williams, A Inmon, J;
Rhoney, S, Andrews G., Johnson, M Ozkaynak, H; Edwards, S, Cohen-Hubal, E
Mechanistic Indicators of Childhood asthma ( MICA); A systems biology approach for
the integration of multifactorial environmental health data submitted: Journal of
Exposure Science and Environmental Epidemiology (2009)
In preparation
David M. Reif, Jane E. Gallagher, Brooke L. Heidenfelder, Ed E. Hudgens, Wendell
Jones, ClarLynda Williams-DeVane, Lucas M. Neas, Elaine A. Cohen Hubal, Stephen
W. Edwards Elucidating Asthma Phenotypes via Integrated Analysis of Blood Gene
Expression Data with Demographic and Clinical Information ( Nature Genetics) 2009
David M. Reif*, ClarLynda Williams-DeVane*, Elaine A. Cohen Hubal, Wendell Jones,
Ed E. Hudgens, Brooke L. Heidenfelder, Lucas M. Neas, Jane E. Gallagher, Stephen
W. Edwards
*Authors contributed equally. Systems Modeling of Gene Expression, Demographic and
Clinical Data to Determine Disease Endotypes PLOS Comp Bio 2009
Previous I TOC
-------
s
c
July 20, 2005
Mr. E. Timothy Oppelt
Acting Assistant Administrator
Office of Research and Development
U.S. Environmental Protection Agency
Washington, DC 20460
Dr. Robert Kavlock
Director
National Center for Computational Toxicology
U.S. Environmental Protection Agency
Research Triangle Park, NC 27711
Re: National Center for Computational Toxicology Review
Dear Mr. Oppelt and Dr. Kavlock:
This is a letter report from the Board of Scientific Counselors (BOSC)
reviewing the progress of the new National Center for Computational
Toxicology (NCCT). Dr. Kavlock and his staff at the NCCT presented an
overview of the Center's structure, activities, goals, and progress on April 25-
26, 2005, to a Subcommittee of the BOSC. The Subcommittee consists of
Drs. George Daston (Chair), James Clark, Richard DiGiulio, Michael Clegg,
and Ken Ramos. Dr. Clegg was unable to attend the briefing and Dr. Ramos
recused himself because of a potential conflict of interest.
Because the NCCT is so new, becoming operational in February 2005, this
report is a prospective one, and is intended to be the first of several
consultative reviews of the Center's progress. In particular, we concentrate on
NCCT's strategic goals; its collaborations, and connectedness to the rest of the
Agency and to outside scientists; its staffing plan; and its thematic choices.
We addressed a number of charge questions intended to focus on each of these
areas. Those charge questions and the Subcommittee's responses are listed
below, following some general comments about the Center.
The Subcommittee was extremely impressed with the progress NCCT has
made in the few short months of its existence. NCCT's mission is to serve as
a focal point for the U.S. Environmental Protection Agency (EPA) in the
application of mathematical and computational tools to all facets of the risk
assessment process. To be successful at this, the NCCT must: (1) provide a
critical mass of expertise in computational, mathematical, and statistical
modeling; (2) develop research collaborations and partnerships with a large
number of groups within and outside the Agency; and (3) have a clear
understanding of and regular interactions with its customers in the rest of the
Office of Research and Development (ORD), the program offices, and the
regions. The Center already has made considerable progress on all three
fronts.
Because its staffing is limited, NCCT has made the appropriate choice of
concentrating on gathering staff with biological, chemical, and statistical
modeling expertise rather than on a particular biological or chemistry
Previous
TOC
Next
-------
Computational Toxicology Subcommittee Letter Report
Page 2 of 6
specialty. This is an appropriate choice, as the staff is strongly aligned to the mission of
the Center. The composition of the staff is impressive; it includes some of EPA's most
accomplished biological modelers, chemists, and statisticians. Most of these individuals
have strong track records of collaboration with multiple laboratories and already are
sought after as research partners. This choice of personnel automatically leverages
NCCT's potential well beyond what would normally be expected of a group of 19.
NCCT has created a virtual organization that brings these people together in a way that
allows them to synergize and form ad hoc groups to make progress on multiple fronts
simultaneously.
The Center already has collaborations and programmatic augmentations via internal and
Science To Achieve Results (STAR) grants. These partnerships cover a large number of
areas of modern biology and chemistry that require high-powered computational and/or
modeling expertise, such as genomics, proteomics, and metabonomics, with coverage of
mammalian toxicology, ecotoxicology, microbiology, exposure assessment, and
quantitative risk assessment. NCCT has a steering committee—the Computational
Toxicology Implementation and Steering Committee (CTISC)—that represents ORD
laboratories/centers, program offices, and regions. The role of the CTISC is still evolving
and it will be an important avenue for communication and for identifying possible
partnerships.
NCCT's strategic plan includes deliverables with short- and longer-term time horizons.
Emphasis on Information Technologies, Prioritization Tools, Biological Models, and
Cumulative Risk will take advantage of the Center's strengths and provide much of the
Agency with technology that will improve its ability to fulfill its mission. It is clear that
the Information Technologies (especially DSSTox) and Prioritization Tools (especially
ToxCast) have the potential to address significant issues in toxicology data management
and prioritization for testing.
Charge Questions and Responses
Questions for the Center as a whole:
1. Success of the NCCT will depend upon establishing effective collaborations with the
other ORD laboratories and centers. What advice can you provide to ensure that
operations remain integrated with the other laboratories and centers within ORD?
The Subcommittee members believe that NCCT has been set up in an optimal way to
maximize interactions, by concentrating expertise in modeling within the Center,
rather than the toxicologists, risk assessment specialists, etc., who populate the other
laboratories and centers. This provides a natural focus area on which the other
laboratories and centers will seek collaboration. Furthermore, most of the staff at the
Center are highly experienced and have a long history of successful collaborations,
including a number of active collaborations that they bring with them. The staff is a
natural magnet for collaborations.
-------
Computational Toxicology Subcommittee Letter Report
Page 3 of 6
One challenge will be to transition the Center from a collection of experts in various
fields to a center of excellence in applying the broad tools of computational
toxicology to address the human health and environmental health issues under the
purview of EPA. Experts will need to develop procedures to capture the essence of
thought processes and computational tools that can be applied to the diversity of
challenges the Agency addresses. Many of the Center staff will be required to shift
their focus from finding computational approaches to address a set of specific issues
to developing robust tools and procedures that provide computational frameworks
that support ORD and Agency programs.
Not all the modeling expertise within EPA resides within NCCT, let alone the
disciplines that rely on computational toxicology. The Center should consider
forming an informal "community of practice" within EPA that can serve a networking
function for interested scientists. This community of practice would not be an
administrative unit, but a virtual professional society within the Agency. Most of its
business can be conducted via electronic media, with occasional meetings. The
Subcommittee endorses the Center's concept of trying to develop various personnel
alignments and management tools (e.g., appointing agency/federal/academic
scientists as adjunct or associate faculty of the Center) to help recruit or gain input
from a broader number of scientists. Those individuals with technical expertise
aligned with the Center's activities can be encouraged to contribute to NCCT
activities while being housed in other organizations within ORD, EPA, or outside of
the Agency; they will form the nucleus of the community of practice.
The CTISC should be explicitly tasked with identifying possible partnerships and
collaborations (and of prioritizing them, if necessary). ORD should continue to hold
regular meetings of its Laboratory and Center Directors, at which partnerships among
centers, including NCCT, can be explored.
The internal grant program that supports many of the NCCT collaborations is
important and likely to be highly successful. Future grant programs should provide a
preference for projects that collaborate with the Center.
Finally, NCCT should develop a communications plan to share its accomplishments
and capabilities with the rest of EPA and those external to the Agency.
2. In terms of anticipated staffing, are there particular areas that should receive greater
or lesser attention?
NCCT may wish to consider adding one or two staff who have expertise in
bioinformatics. The planned grant for an external bioinformatics center will cover
most of the Center's needs in this area, but having some internal expertise would
complement the external bioinformatics efforts and provide a natural point of contact
between the external group and NCCT. The Center also should consider whether
there are social science applications to computational toxicology, and if so, whether
there is a social science expertise that should be represented on the staff.
-------
Computational Toxicology Subcommittee Letter Report
Page 4 of 6
NCCT also may wish to consider hiring one or two leading scientists in the field of
ecological modeling. Of the many competencies that could be targeted, fields such as
modeling large-scale ecological processes, population and community dynamics,
tissue dynamics in ecological receptors (PB/PK, bioaccumulation processes, and
lethal/adverse effects of body burdens), and environmental fate and effects of
chemicals (including microbial biodegradation and bioavailability) would be
particularly useful. Although it is unlikely that one individual will have expertise
spanning all these areas, having an individual with modeling expertise will serve as a
focal point for collaborations with EPA scientists outside the Center who have
complementary expertise. During the review, the Center staff demonstrated the
importance of obtaining insight from social scientists in developing technically sound
and meaningful studies. NCCT should consider including this area of expertise
among the core competencies of the Center.
3. As we find ourselves in the post-genome era, science is progressing at a rapid pace.
This makes it difficult to stay abreast with the current state-of-the-science. Clearly,
being cognizant of and understanding the technologies and advanced methods in the
areas of the omics, modeling, and statistics is a considerable vested interest to NCCT
for several reasons, such as being able to make decisions about which technologies
are best for the Center to pur sue and most beneficial to the Agency. Can the BOSC
provide any suggestions on how best to keep apace with new technologies and
methodologies ?
This is a problem that we all face, but is perhaps more severe for an integrating group
such as NCCT. Partnerships with other organizations with similar/complementary
interests may be the best way to facilitate keeping current. Active collaborations,
which already are the stock-in-trade for the Center, publication, and participation in
professional meetings will keep the Center staff fresh and well informed. These
efforts also will serve to attract the brightest students and post-doctoral fellows, who
will bring with them the latest technologies.
Questions concerning the areas of emphasis, or "Concept Topics":
4. Has the Center articulated a clear rationale for each topic area, and has it provided
evidence that the contemplated approaches will be able to address the major goals
stated in A Framework for a Computational Toxicology Research Program?
The Subcommittee members believe that NCCT is on track. It will be important for
the Center to prepare a synthesized set of goals/milestones for the numerous projects
in which the Center is involved, explaining how each fulfills a need, and how each
topic area will provide tools for the Agency. The prioritization process that the
Center leadership has developed is a good one, which works well in selecting
program areas that are consistent with the Center's mission.
-------
Computational Toxicology Subcommittee Letter Report
Page 5 of 6
5. To be successful in addressing the Concept Topics, can you help identify potentially
fruitful partnerships with others outside the Agency?
The review provided plenty of evidence that the Center is reaching out to find
potential collaborators among a diverse set of U.S. government and private
institutions. Many of the collaborations discussed should be formalized in
Memoranda of Understanding (MOU), Interagency Agreements (lAGs), and other
formal commitments to demonstrate the degree of cooperation, leverage, and interest
generated with other partners. Also, NCCT will need to have opportunities to work
with scientists and regulatory authorities from countries around the world, as
computational toxicology is an area of evolving science with expertise in Europe,
Canada, Asia, perhaps Russia, as well as the United States.
One approach to broaden international contacts would be to consider development of
ties with U.S.-based academic centers and institutions that have liaisons with
international scientists and organizations. Also, Center management may want to
specifically reserve some travel allocations to allow attendance at conferences,
workshops, or technical exchanges and site visits at leading international sites and
organizations around the world. A world-class center will need worldwide
perspectives in computational toxicology.
NCCT already is doing a good job of establishing liaisons with other organizations
involved in aspects of computational toxicology, such as the National Center for
Toxicogenomics at the National Institute of Environmental Health Sciences (NIEHS).
Efforts should be continued to partner with private industry in areas of mutual
interest.
6. Given the mission, staffing, and resources of the Center, what is your view of the
depth and breadth of the areas currently selected for emphasis? Are there additional
areas that should be considered?
The Subcommittee members believe that the Center is doing a good job of
maintaining broad coverage through its collaborations with multiple laboratories.
Depth will come from the other laboratories and programs with which NCCT
collaborates.
The Center's goal to take advantage of opportunities to broaden and generalize the
technical approaches to the diverse scope of Agency issues is an admirable goal, and
one that will require a disciplined approach among the technical and managerial team
to implement. The Subcommittee realizes that the endocrine disrupter studies offer
many concrete examples of the kind of molecular and cellular work the NCCT can
provide in the future. It will be important that the Center quickly provides similar
services and value to EPA programs that can benefit from these tools applied to non-
endocrine disruption issues. Plans to broaden program office representation in the
CTISC (to include the Offices of Solid Waste and Emergency Response and
Homeland Security, and possibly others) should quickly bring these opportunities to
-------
Computational Toxicology Subcommittee Letter Report
Page 6 of 6
the forefront. Discussions should proceed with Agency programs and offices dealing
with waste management and issues surrounding remediation of contaminated sites;
applications of environmental models to total maximum daily loads (TMDLs);
environmental health monitoring programs such as the Environmental Monitoring
and Assessment Program (EMAP), various regional Bay programs (Chesapeake Bay,
Great Lakes Program, Florida Everglades), as well as the air and water monitoring
programs conducted by the states with federal assistance. Understanding the
chemical and biological stressors encountered in these environmental health studies
will broaden the types of contaminants and thus computational tools that must be
considered by NCCT. It also will challenge applications of the Center's tools to
issues with a broad temporal and spatial scale and provide opportunities to assess
some dynamic aspects of human and animal populations.
In conclusion, the BOSC Subcommittee believes that NCCT has made great progress and
is on the right track to deliver against its mission. We are pleased to provide advice on
this important Center and look forward to our continuing oversight of NCCT.
Sincerely yours,
James H. Johnson, Jr.
Chair, Board of Scientific Counselors
Previous I TOC
-------
UNITED STATES ENVIRONMENTAL PROTECTION AGENCY
WASHINGTON, D.C. 20460
September 8, 2005
OFFICE OF
RESEARCH AND DEVELOPMENT
Dr. James H. Johnson, Jr.
Chair, Board of Scientific Counselors
Dean, College of Engineering, Architecture, and Computer Sciences
Howard University
2366 6th Street NW
Washington, DC 20059
Dear Dr. Johnson:
The Office of Research and Development (ORD) would like to take this
opportunity to thank you and the members of the Board of Scientific Counselors (BOSC)
for the April 2005 progress review of the new National Center for Computational
Toxicology (NCCT). We especially thank the members of the Computational
Toxicology Subcommittee who conducted the review, Drs. George Daston (Chair),
James Clark, and Richard Di Giulio.
We are pleased that the BOSC was very supportive of the NCCT and the direction
we are taking in this very important research program. Enclosed with this letter is our
response to the comments in your Letter Report of July 20, 2005. Please feel free to
contact me if further information is needed.
Again, thank you for your advice to ORD.
Sincerely yours,
William H. Farland, Ph.D.
Acting Deputy Assistant Administrator
for Science
Enclosure
cc: Dr. George Daston (Computational Toxicology Subcommittee, Chair)
Dr. James Clark
Dr. Richard Di Giulio
Internet Address (URL) • http://www.epa.gov
Recycled/Recyclable • Printed with Vegetable Oil Based Inks on Recycled Paper (Minimum 20% Poslconsumer)
-------
OKD Response to Board of Scientific Counselors (BOSC) Review of the National
Center for Computational Toxicology (NCCT) in April 2005
The following is a narrative response to the comments and recommendations of
the BOSC review of ORD's National Center for Computational Toxicology. The review
was held April 25 - 26, 2005, in Research Triangle Park, NC. The committee considers
this to be part of a series of consultative reviews. For this review, the committee
concentrated on the NCCT's strategic goals; its collaborations, and connectedness to the
rest of the Agency and to outside scientists; its staffing plan; and its thematic choices.
They addressed a number of charge questions intended to focus on each of these areas.
Generally, the committee was very favorable to the formation of the NCCT and the
progress the Center has made since its inception a few months ago. The committee
recognized the unique and important role for the Center because of its small size and
ability to establish strong collaborations with other groups within and outside of the
ORD. The committee emphasized the importance of collaborations and positively
commented on the number of collaborations that are already taking place. The committee
also commented favorably on the Center's four focus areas of Information Technologies,
Prioritization Tools, Biological Models, and Cumulative Risk. The committee
highlighted that the first two have the potential to address "significant issues in
toxicology..." The committee felt the NCCT has made appropriate choices in bringing
together expertise from several related disciplines to fulfill the Center's mission.
Following are specific comments related to the charge questions made by the
committee. The committee's comments are written in italics and ORD's response
follows in regular type. Attached to this document is a summary table which provides a
summary of BOSC comments and proposed ORD actions.
I. The first charge question asked for advice on collaboration with other ORD
laboratories and centers and asked for advice to ensure that operations remain integrated
with those laboratories and centers. The committee responded:
One challenge will be to transition the Center from a collection of experts in
various fields to a center of excellence in applying the broad tools of
computational toxicology to address the human health and environmental health
issues under the purview of EPA. Experts will need to develop procedures to
capture the essence of thought processes and computational tools that can be
applied to the diversity of challenges the Agency addresses. Many of the Center
staff will be required to shift their focus from finding computational approaches
to address a set of specific issues to developing robust tools and procedures that
provide computational frameworks that support ORD and Agency programs.
ORD strongly agrees with this comment. The NCCT is writing an
implementation plan that will outline the research being conducted over
the next several years. In this plan there is a strong commitment to
conduct work that addresses specific Agency needs. The plan will
recognize the need for providing generic tools that will facilitate the
-------
incorporation of computational methods into the hazard and risk
assessment processes. An example of such a current activity is assisting in
the compilation and web hosting of a database of parameters that can be
used for physiologically based pharmacokinetic modeling across the life
stages. The NCCT is also committed to the long term goal of conducting
two annual training workshops on topics that will help promote the use of
computational approaches in ORD and the EPA. The plan will also
discuss the interactions between activities within the Center and other
components of ORD. For example, on-going work on using the newest
"omics" technologies and bioinformatics to better identify and
characterize pathways of toxicity are being conducted in collaboration
with scientists from the National Health and Ecological Effects Laboratory
(NHEERL). Compounds include endocrine disrupting compounds and
pesticides, all of interest and concern to Agency program offices.
Similarly, NCCT scientists are working with scientists from the National
Exposure Laboratory (NERL) to use computational chemistry methods to
better quantify the rates of key biochemical processes that are important in
pharmacokinetic and pharmacodynamic modeling being conducted by
scientists with the Center and other parts of ORD. Often the models being
developed and applied are to address specific needs of the Program
Offices such as for pyrethroid and N-methylcarbamate pesticides.
Because of its small size, the Center staff has a good rapport and
meets regularly to discuss and share their work and progress. In these
meetings there is a free exchange and collaboration and interaction is
easily promoted. Similar meetings have also been initiated with
colleagues at the National Institute of Environmental Health Sciences
(NIEHS). Because of the varied expertise within these groups, problems
can be addressed with solutions as the goal rather than from only a
specific narrow discipline.
It should also be noted that NCCT scientists work closely with
others within the Agency on projects specifically relevant and important to
Program Offices. Agency scientists work directly on assessments being
performed by the Office of Pesticide Programs (OPP) in support of the
Food Quality Protection Act mandated re-registration program, for
example. NCCT scientists serve on the Agency Risk Assessment forum as
well. These kinds of activities help assure the NCCT will address a broad
array of problems relevant to Agency needs.
Not all the modeling expertise within EPA resides within NCCT, let alone
the disciplines that rely on computational toxicology. The Center should consider
forming an informal "community of practice " within EPA that can serve a
networking function for interested scientists. This community of practice would
not be an administrative unit, but a virtual professional society within the Agency.
-------
Most of its business can be conducted via electronic media, with occasional
meetings.
The NCCT recognizes the need for integration of computational
efforts across ORD and has provided leadership for the formation of two
Communities of Practice (CoP) within the EPA. One is on
chemoinformatics and one is on biological modeling. The goal of the
chemoinformatics CoP is to facilitate, coordinate and integrate efforts to
address the challenges of chemical structure annotation (or indexing),
retrieval, and mining of chemically-related data and documents, including
newer toxicogenomics and metabonomics data, across EPA Program
Offices, Labs and Centers. The goal of the biological modeling
community of practice is to advance the principals for development and
application of dosimetry and other biologically based models within the
Agency. Dosimetry modeling includes multiple forms of toxicokinetic
modeling (e.g., physiologically based toxicokinetic (PBTK) modeling,
compartmental modeling), respiratory tract dosimetry modeling (e.g.,
computational fluid dynamics), and related modeling (e.g., dermal
absorption modeling). The working group will also focus on biologically
based response modeling with special emphasis on using the newest
"omics" information in biologically based models. A further goal is to
foster adoption of modeling science by Agency clients in regulatory
decision making.
Membership in these CoPs has initially been solicited from within
ORD. In the near future this will be extended across the Agency. ORD is
considering gaining endorsement for these groups from the Agency's
Office of Science Policy. It is ORD's belief that this will extend the
expertise and more importantly, assure that the CoPs will focus on
important issues relevant to current and future Agency problems. These
CoPs will operate as the committee has suggested, i.e. electronic media
and occasional meetings.
The Subcommittee endorses the Center's concept of trying to develop
various personnel alignments and management tools (e.g., appointing
agency/federal/academic scientists as adjunct or associate faculty of the Center)
to help recruit or gain input from a broader number of scientists. Those
individuals with technical expertise aligned with the Center's activities can be
encouraged to contribute to NCCT activities while being housed in other
organizations within ORD, EPA, or outside of the Agency; they will form the
nucleus of the community of practice.
The NCCT endorses this recommendation and welcomes the
possibility of individuals from outside the NCCT doing rotational details
to acquire training and skills in computational toxicology. The CoPs
mentioned above will help create networks of individuals working towards
-------
similar objectives. Besides the CoPs, a number of informal alignments
have already occurred and resulted in fruitful endeavors. One example is
a working relationship between NHEERL scientists and NCCT scientists
in working contractually with a private company to investigate the
feasibility of that company's capabilities in genomic signature
development for screening and prioritization and for toxicity pathway
identification. Another example is the aforementioned collaboration
between NCCT scientists in computational toxicology contributing to a
pharmacokinetic modeling project with NERL scientists. Finally,
interactions with the NTP/NIEHS are developing, as both groups have
similar goals as identified in the NTP Roadmap and the EPA's
Computational Toxicology Framework.
The CTISC [Computational Toxicology Implementation Steering
Committee] should be explicitly tasked with identifying possible partnerships and
collaborations (and of prioritizing them, if necessary). ORD should continue to
hold regular meetings of its Laboratory and Center Directors, at which
partnerships among centers, including NCCT, can be explored.
The CTISC had been expanded to include the NHSRC and
Program Office Staff; it now contains delegates from OPPTS (2), OW (2),
OAQPS (1) and the Regions (1). The topic of interactions between NCCT
scientists and those in other components of ORD has been forwarded as an
upcoming agenda item for the CTISC.
ORD Laboratory and Center Directors meet on a regular basis,
approximately once a month. Many topics including partnerships are
regularly addressed at these meetings. In addition, the management team
of the NCCT has scheduled monthly meetings with their counterparts in
NHEERL and NERL and will hold quarterly meetings with the NPDs for
Safe Pesticides and for Human Health Research. A Memorandum of
Agreement (MOA) has been established between the NCCT, NHEERL,
and NERL to provide administrative support, which has provided a strong
partnership amongst the co-located units in RTP.
The internal grant program that supports many of the NCCT
collaborations is important and likely to be highly successful. Future grant
programs should provide a preference for projects that collaborate with the
Center.
The ORD agrees that a program in which the NCCT works with
other scientists helps coordinate a great deal of the computational
toxicology research is very fruitful in terms of promoting collaborations.
While another round of request for proposals has yet to be planned, the
NCCT has committed to reserve at least 10% of its available extramural
resources in the coming year to be used to augment or initiate
-------
computational toxicology research within other laboratories and centers.
Strong preference will be given to those projects in which NCCT staff will
be involved.
Finally, NCCT should develop a communications plan to share its
accomplishments and capabilities with the rest of EPA and those external to the
Agency
A formal communication plan will be prepared in the coming year.
At this time, NCCT continues to look for opportunities to informally
communicate their capabilities and accomplishments. Recently we
established an internet homepage for the research program and have
initiated discussions with the communications team in HQ about a broader
scale communications effort.
II. The second charge question asked for advice on anticipated staffing. The committee
responded:
NCCT may wish to consider adding one or two staff who have expertise in
bioinformatics. The planned grant for an external bioinformatics center will
cover most of the Center's needs in this area, but having some internal expertise
would complement the external bioinformatics efforts and provide a natural point
of contact between the external group and NCCT. The Center also should
consider whether there are social science applications to computational
toxicology, and if so, whether there is a social science expertise that should be
represented on the staff.
The NCCT has already advertised for two staff positions with
expertise in bioinformatics. Final selection is anticipated within the next
month. We have also requested approval to use the new Title 42 hiring
authority to attract a more senior level bioinformatatist. In addition, two
Environmental Bioinformatic Research Centers are being established
through the STAR program to bolster the state of the science of informatic
analysis in environmental health sciences. A senior position seeking
expertise that can help develop high through-put screening and
prioritization methods has also been advertised. Likewise, final selection
is expected within the next month.
The NCCT is considering the possibility of hiring a social science
expertise in the future. Our short term plan is to provide postdoctoral
support to the visual analytic effort looking at children's exposure issues
as a first foray into this area. Additionally, ORD may choose to hire such
expertise in other programs within the other laboratories or centers. This
will be considered in ORD's overall work force planning activities.
-------
Finally, we note that the NCCT recently hired a well known senior
scientist as an ST to take leadership in the area of systems biology.
III. The third charge question sought advice on how to best keep apace with new
technologies and methodologies. The committee response:
This is a problem that we all face, but is perhaps more severe for an integrating
group such as NCCT. Partnerships with other organizations with
similar/complementary interests may be the best way to facilitate keeping current.
Active collaborations, which already are the stock-in-trade for the Center,
publication, and participation in professional meetings will keep the Center staff
fresh and well informed. These efforts also will serve to attract the brightest students
and post-doctoral fellows, who will bring with them the latest technologies.
The ORD and NCCT agree fully with this comment. The staff are and
have been actively engaged in such activities both nationally and internationally
(e.g., ILSI, WHO, OECD). They also look for training opportunities. Resources
are maintained for travel and training. Recently the Center selected a candidate
for the cross-ORD post-doctoral program. This highly qualified candidate will be
working with a senior scientist from the NCCT and one from the NERL on a
research very relevant to computational toxicology.
IV. The fourth charge question asked if the NCCT articulated a clear rationale for its
concept topic areas of research. The committee response:
The Subcommittee members believe that NCCT is on track. It will be
important for the Center to prepare a synthesized set of goals/milestones for the
numerous projects in which the Center is involved, explaining how each fulfills a
need, and how each topic area will provide tools for the Agency. The
prioritization process that the Center leadership has developed is a good one,
which works well in selecting program areas that are consistent with the Center's
mission.
The NCCT appreciates the comments and the staff is currently
preparing a research implementation plan that will address goals,
rationale, and milestones over the next three years. This plan is expected
to be ready for review during September, 2005. An important component
of this implementation plan is the launching of the ToxCast program,
which is being designed to establish a process for the prioritization of
chemicals for toxicological testing, one of the key driving forces for the
inception of the computational toxicology program.
V. The next charge question asked the committee to help identify potentially fruitful
partnerships with others outside the Agency. The response:
-------
The review provided plenty of evidence that the Center is reaching out to
find potential collaborators among a diverse set of U.S. government and private
institutions. Many of the collaborations discussed should be formalized in
Memoranda of Understanding (MOU), Interagency Agreements (lAGs), and other
formal commitments to demonstrate the degree of cooperation, leverage, and
interest generated with other partners. Also, NCCTwill need to have
opportunities to work with scientists and regulatory authorities from countries
around the world, as computational toxicology is an area of evolving science with
expertise in Europe, Canada, Asia, perhaps Russia, as well as the United States.
One approach to broaden international contacts would be to consider
development of ties with U.S.-based academic centers and institutions that have
liaisons with international scientists and organizations. Also, Center
management may want to specifically reserve some travel allocations to allow
attendance at conferences, workshops, or technical exchanges and site visits at
leading international sites and organizations around the world. A world-class
center will need worldwide perspectives in computational toxicology.
NCCT already is doing a good job of establishing liaisons with other
organizations involved in aspects of computational toxicology, such as the
National Center for Toxicogenomics at the National Institute of Environmental
Health Sciences (NIEHS). Efforts should be continued to partner with private
industry in areas of mutual interest.
Since the review in April the NCCT staff has visited programs in
Russia seeking opportunities for collaboration. Although not yet fruitful
several promising areas were identified and are being pursued.
The recommendation to establish ties with U.S. based academic
centers that have liaisons with international scientists is a good one and the
Center will investigate such possibilities. Also, the NCCT staff will be
working closely with the STAR program Bioinformatics Centers. A one
day workshop was held in May 2005 with DOEs Pacific Northwest
National Laboratory to develop communication links and begin to identify
areas of collaboration between the two organizations.
The NCCT management, as mentioned previously, has been and
will continue to be careful about reserving sufficient resources to allow
staff the ability to attend and present at conferences, workshops, etc. In
addition, the Center is planning a program of specific topic workshops to
be conducted at the EPA and at national and international meetings of
professional societies. The NCCT scientists are considering formulating
and teaching courses in relevant areas. This will serve to show Center
capabilities and extend the exchanges between experts from throughout
the world and Center staff.
-------
Center staff are also actively involved with a number of activities
of ILSI, the WHO and OECD and will have made more than a dozen
presentations this year at international meetings specifically related to
various aspects of computational toxicology. These presentations have
helped communicate the formation of the NCCT to the international
scientific community.
VI. The final charge question asked the committee to comment on the depth and breadth
of the emphasis areas and whether they recommended other areas for consideration. The
responses:
The Subcommittee members believe that the Center is doing a good job of
maintaining broad coverage through its collaborations with multiple
laboratories. Depth will come from the other laboratories and programs with
which NCCT collaborates.
Collaboration with other laboratories and centers is a centerpiece
of NCCT's mode of operation and has been discussed in responses to
previous comments.
The Center's goal to take advantage of opportunities to broaden and
generalize the technical approaches to the diverse scope of Agency issues is an
admirable goal, and one that will require a disciplined approach among the
technical and managerial team to implement. The Subcommittee realizes that the
endocrine disruptor studies offer many concrete examples of the kind of
molecular and cellular work the NCCT can provide in the future. It will be
important that the Center quickly provides similar services and value to EPA
programs that can benefit from these tools applied to non-endocrine disruption
issues. Plans to broaden program office representation in the CTISC (to include
the Offices of Solid Waste and Emergency Response and Homeland Security, and
possibly others) should quickly bring these opportunities to the forefront.
Discussions should proceed with Agency programs and offices dealing with waste
management and issues surrounding remediation of contaminated sites;
applications of environmental models to total maximum daily loads (TMDLs);
environmental health monitoring programs such as the Environmental
Monitoring and Assessment Program (EMAP), various regional Bay programs
(Chesapeake Bay, Great Lakes Program, Florida Everglades), as well as the air
and water monitoring programs conducted by the states with federal assistance.
Understanding the chemical and biological stressors encountered in these
environmental health studies will broaden the types of contaminants and thus
computational tools that must be considered by NCCT. It also will challenge
applications of the Center's tools to issues with a broad temporal and spatial
scale and provide opportunities to assess some dynamic aspects of human and
animal populations.
-------
As noted above, membership in the CTISC has been expanded.
The Center in particular has extensive work dealing with pesticides and
high production volume chemicals that include substances other than
endocrine disrupters. Discussions with parts of the Agency recommended
by the committee have begun and will be expanded and continued in the
coming year. At this time however it is not clear what role the NCCT
itself will take in some of the more ecologically related areas mentioned
by the Committee. Given the small size of the NCCT and the fact that
some of those activities are well represented in the other laboratories and
centers these areas may be addressed through collaborative efforts and
temporary assignments of those scientists to the NCCT. However, this
needs further discussion within ORD.
Based upon the urgent needs to develop a prioritization and
categorization process for evaluating the large numbers of chemicals for
which standard toxicological studies are not available, the NCCT will
soon be launching the ToxCast project. This effort, to some extent, builds
on the activities of the EDC proof of concept projects in that it will be
using a variety of computational and molecular tools to collect biological
activity patterns using high throughput screening devices. If successful,
this concept will provide multiple programs offices with a solution to a
vexing problem. We are now involved in a number of briefings and
presentations across ORD, the Program Offices (e.g., OPPTS and OW) in
order to build a consensus about the overall program and to fine tune the
directional details. Completion of our staffing targets that are scheduled
for this year will greatly facilitate our ability to broaden beyond the efforts
presented to the BOSC during the April review.
Recognizing that this review was a progress review early in the life of the NCCT
it is expected that subsequent reviews by the Committee will take place. The next
progress review is expected late 2006 or early 2007.
-------
Computational Toxicology Program
Summary of BOSC Comments From July 2005 Letter Report and Proposed ORD
Actions
Recommendation
Action Items
Timeline
Charge Question 1, Advice on Collaboration:
Shifting focus from finding
computational approaches to
address a set of specific issues
to developing robust tools and
procedures that provide
computational frameworks that
support ORD and Agency
programs
An Implementation Plan for the
NCCT is being written that
incorporates all facets of the
Computational Toxicology Program
including research, outreach, and
operations. The plan will recognize
the need for providing generic tools
that will facilitate the incorporation of
computational methods into the
hazard and risk assessment processes.
The NCCT is also committed to the
long term goal of conducting two
annual training workshops on topics
that will help promote the use of
computational approaches in ORD
and EPA.
September,
2005
Current and
on-going
Form an informal "community
of practice " within EPA that
can serve a networking function
for interested scientists
The NCCT recognizes the need for
integration of computational efforts
across EPA. Two such Communities
of Practice have been initiated - one
for chemoinformatics and one for
biological modeling.
Expect first
meetings by
October 30,
2005
Develop various personnel
alignments and management
tools to help recruit or gain
input from a broader number of
scientist
The NCCT welcomes the opportunity
for staff from other units of ORD or
EPA to have rotational details for the
purpose of acquiring training and
experiences in computational
methods. The Communities of
Practice offer other means of gaining
input from a broader range of
scientists. The weekly work in
progress meeting with the
NTP/NIEHS offers yet another input
function
Current and
on-going
The CTISC should be explicitly
tasked with identifying possible
partnerships and collaborations
(and of prioritizing them, if
The CTISC had been expanded to
include the NHSRC and Program
Office staff. The topic of interactions
between NCCT scientists and those in
Late FY
2005/early FY
2006
Page 1 of4
-------
Recommendation
Action Items
Timeline
necessary
other components of ORD has been
forwarded as an upcoming agenda
item for the CTISC.
ORD should continue to hold
regular meetings of its
Laboratory and Center
Directors, at which
partnerships among centers,
including NCCT, can be explore
ORD Laboratory and Center
Directors meet on regular basis
approximately once a month. In
addition, the management team of the
NCCT has scheduled monthly
meetings with their counterparts in
NHEERL and NERL on a monthly
basis, and has agreed to hold at least
quarterly meetings with the NPDs for
Safe Pesticides and for Human Health
Research.
On going
Future [internal}grant programs
should provide a preference for
projects that collaborate with
the Center.
The NCCT has committed to reserve
at least 10% of its available
extramural resources in the coming
year to be used to augment or initiate
computational toxicology research
within other Laboratories and
Centers. Strong preference will be
given to those projects in which
NCCT staff will be involved.
FY 2006 and
beyond
NCCT should develop a
communications plan to share
its accomplishments and
capabilities with the rest of EPA
and those external to the
Agency
Formal communications plan for
NCCT to be developed and
implemented; an internet homepage
has been established.
FY 2006
Charge Question 2, advice on anticipated staffing:
NCCT may wish to consider
adding one or two staff who
have expertise in
bioinformatics
Two such positions have been
advertised and selection process is on-
going. We have also requested
approval to use the new Title 42 hiring
authority to attract a more senior level
bioinformaticist.
Final
selection by
October 1,
2005
The Center also should
consider whether there are
social science applications to
computational toxicology,
and if so, whether there is a
social science expertise that
should be represented on the
staff
The NCCT recognizes the importance
of this research activity and it will be
considered by NCCT and others in
ORD's overall work force planning
activities. The short term plan is to
provide postdoctoral support to the
visual analytic effort looking at
children's exposure issues as a first
FY 2006
Page 2 of4
-------
Recommendation
Action Items
foray into this area.
Timeline
Charge Question 3, advice on how to keep apace with new technologies and
methodologies:
Consider partnerships with
other organizations with
similar /complementary
interests to facilitate keeping
fresh; , publication, and
participation in professional
meetings will also keep the
Center staff fresh and well
informed
Staff are actively engaged in National
and International activities (ILSI,
WHO, OECD) and meetings -
resources are set aside for such
activities
On-going
Charge Question 4, has NCCT articulated a clear rationale for topic research areas:
Prepare a synthesized set of
goals/milestones for the
numerous projects in which
the Center is involved,
explaining how each fulfills a
need, and how each topic
area will provide tools for the
Agency
NCCT is developing research
implementation plan. This plan will
articulate the particular directions and
expected milestones of the research
program over the next three years.
First draft -
September,
2005
Charge Question 5, identification of fruitful partnerships with others outside the Agency
NCCT will need to have
opportunities to work with
scientists and regulatory
authorities from countries
around the world, as
computational toxicology is
an area of evolving science
with expertise in Europe,
Canada, Asia, perhaps
Russia, as well as the United
State
Consider development of ties
with U. S. -based academic
centers and institutions that
have liaisons with
international scientists and
organization
Center management may
The NCCT staff are involved in a
number of ongoing international
efforts, including those with ILSI, the
WHO and OECD. In addition,
NCCT staff recently visited Russia to
develop potential working
partnerships and we are now working
through OSP/ORD and ISTC/Russia
to develop several research proposals
in computational toxicology
The newly established STAR Centers
for Environmental Bioinformatics
should provide a logical starting place
for interactions with academic
institutions. A one day workshop was
held in May 2005 with scientists from
PNNL looking to develop a
collaborative relationship.
NCCT agrees and has and will
Ongoing
Continued
discussions
with
collaborative
proposals
developed
early in FY
2006
Immediate
and
continuing;
selection for
new staff for
bioinformatics
expected by
Oct 1,2005.
On-going
Page 3 of4
-------
Recommendation
Action Items
Timeline
want to specifically reserve
some travel allocations to
allow attendance at
conferences, workshops, or
technical exchanges and site
visits at leading international
sites and organizations.
continue to reserve sufficient
resources to allow staff participation
at conferences, workshops, etc. With
our current budgetary situation, we do
not foresee any difficulty in
supporting this function.
NCCT is also preparing a program of
specific topic workshops
Expected in
2006 and then
continuing
Charge question 6, depth and breadth of emphasis areas and other possible areas of
consideration:
The Subcommittee members
believe that the Center is
doing a good job of
maintaining broad coverage
through its collaborations
with multiple laboratories.
Depth will come from the
other laboratories and
programs with which NCCT
collaborates
The NCCT appreciates the positive
feedback and will continue to develop
collaborations that will allow delivery
of important products to the Agency
over the next 3-5 years.
Ongoing
The Subcommittee realizes
that the endocrine disruptor
studies offer many concrete
examples of the kind of
molecular and cellular work
the NCCT can provide in the
future. It will be important
that the Center quickly
provides similar services and
value to EPA programs that
can benefit from these tools
applied to non-endocrine
disruption issues.
Launching of the ToxCast concept
that builds on the activities of the
EDC proof of concept projects in that
it will be using a variety of
computational and molecular tools to
collect biological activity patterns
using high throughput screening
devices to prioritize and categorize
chemicals for more standard
toxicological evaluation.
FY 2006
Page 4 of4
-------
BOARD OF SCIENTIFIC COUNSELORS
December 12, 2006
Dr. George Gray
Assistant Administrator
Office of Research and Development
U.S. Environmental Protection Agency
1200 Pennsylvania Avenue, NW
Washington, DC 20460
Dr. Robert Kavlock
Director
National Center for Computational Toxicology
Office of Research and Development
U.S. Environmental Protection Agency
Research Triangle Park, NC 27711
Dear Drs. Gray and Kavlock:
This is a letter report from the Board of Scientific Counselors (BOSC) reviewing
the Computational Toxicology Research Program conducted by the National
Center for Computational Toxicology (NCCT). The Computational Toxicology
Subcommittee of the BOSC reviewed NCCT's progress and plans during a 2-
day meeting on June 19-20, 2006, at the EPA facility in Research Triangle Park,
North Carolina. The BOSC Subcommittee consists of George Daston (Chair),
James Clark, Michael Clegg, Richard DiGiulio, Muiz Mumtaz, and John
Quackenbush.
,,! This is the second review of the NCCT. The first review of the Center was
conducted in May 2005. The Subcommittee was very pleased with the progress
that the NCCT has made towards its goals since that first review. The Center
first became operational in February 2005; during the 16 months between its
establishment and this review, the NCCT has made substantial progress in:
(1) establishing priorities and goals; (2) making connections within and outside
EPA to leverage the staffs considerable modeling expertise; (3) expanding its
capabilities in informatics; and (4) significant contributions to research and
decision-making throughout the Agency.
Many of the recommendations made by the BOSC during its first review have
been acted on by NCCT. This includes expanding its capabilities in
bioinformatics through the funding of two external centers and through staff
hires, expansion of its technical approaches to even more programs within the
A Federal Advisory Committee for the U.S. Environmental Protection Agency's Office of Research and Development
Previous I TOC I Next
-------
December 2006 BOSC Computational Toxicology Letter Report
2
Agency, and the development of communities of practice (CoPs) throughout the EPA
research community in chemoinformatics, biological modeling, and chemical
prioritization. CoPs are cross-organizational groupings of experts who share an interest in
a common technology.
The Subcommittee addressed a number of charge questions during its review, the
responses to which provide a basis for comments on progress as well as specific
recommendations.
Question 1: What progress has been made in the last year in developing/maximizing
connections and collaborations within ORD and the Agency, through communities of
practice and other interactions? Are there notable examples of collaborations that have
been established to increase the reach and effectiveness ofNCCT? Are there additional
collaboration opportunities NCCTshould explore?
During the review, the Subcommittee members heard reports on three active CoPs:
(1) Chemoinformatics; (2) Biological Modeling; and (3) Chemical Prioritization. They
also heard about one proposed CoP, Cumulative Risk.
All active CoPs have formal memberships and are chaired by NCCT staff. The Center
also has observed active participation among numerous EPA laboratories and centers and
several program offices. The Chemoinformatics and Chemical Prioritization CoPs
already have demonstrated outreach to outside agencies, such as the National Institutes of
Health (NIH), National Institute of Environmental Health Sciences (NIEHS), and
National Toxicology Program (NTP). Some are working with or soliciting international
and private sector collaboration. The CoPs have been effective in focusing on defining
problems and suggesting solutions, agreeing on modeling approaches and database
issues, and setting up forums and workshops for discussions. They will be responsible
for leading a better coordinated effort within EPA and among agencies.
The Subcommittee believes that establishing a Cumulative Risk CoP is worthy of pursuit.
Such a CoP would provide significant opportunities to define areas for improvement in
risk assessment practices and could provide inventory tools and other benefits. NCCT
should consider whether it would like to provide a facilitator role or leadership role in
this area.
With regard to other opportunities for exploration, the Subcommittee suggests that NCCT
seek broader program office input. Additionally, CoPs covering areas such as Mixtures,
Cross-Species Extrapolation, Population/Systems Dynamic Models, and Multimedia Fate
and Effects Modeling should be considered for either NCCT use or ORD's broader use.
Question 2: How does the work of the new extramural bioinformatics centers
complement the intramural program, and how should the outputs best be integrated into
NCCT strategic direction?
ORD funded two extramural Bioinformatics Centers, one at the University of North
Carolina directed by Fred Wright and a second at the University of Medicine and
Dentistry of New Jersey headed by William Welsh. The Centers are used to extend the
-------
December 2006 BOSC Computational Toxicology Letter Report
3
capabilities of the intramural program. Individually, the Bioinformatics Centers were
viewed to be excellent choices, each providing expertise and resources largely
complementary to each other and to the NCCT with little overlap. Although both Centers
are just beginning their work with EPA, there is great opportunity for synergy in
developing new approaches for the analysis of toxicogenomics data and the integration of
diverse information necessary to place these data into an appropriate context. In addition,
each Center has existing links to risk managers and risk management groups, providing
additional potential avenues for outreach to link the research programs of the NCCT and
the Centers to real problems.
Integration of the external Bioinformatics Centers and the programs within NCCT will
occur following the hiring of one senior and one junior bioinformatics scientist. This
may not represent sufficient personnel, however, to allow NCCT to fully support its
overall mission. Much of NCCT's program is focused on development of predictive
models using systems biology approaches. Although this is a laudable approach, it
ultimately will be driven by the availability of high-quality, well-annotated data and their
integration with a wide range of other information. This will require significant effort.
Although there are efforts underway under the direction of various NCCT personnel to
begin this process, a more integrated approach is needed.
Consequently, NCCT needs to develop a more comprehensive strategic plan for data
collection, management, and integration through creation of databases that model the
structure of the underlying information and its potential use. This will require a careful
assessment of the capabilities extant in each center so that necessary components, as well
as areas for future development, can be identified. Addressing these issues will provide
the structured data needed by NCCT's Systems Modeling and Computational Chemistry
groups.
It also was noted that there exists a need within the field for trained personnel in
computational toxicology. In addition to the existing postdoctoral program, one feasible
approach would be to institute a career development award similar to the NIH "K"
awards that would provide mentored training and research to more senior personnel.
Question 3: Although the intent is not to review individual research programs, do the
research programs highlighted during this review offer the promise of increasing the use
and effectiveness of computational methods in Agency research? Do the efforts fulfill the
goal of leveraging the resources of NCCT to increase effectiveness?
The long-term goals (LTGs) of the Computational Toxicology Research Program are to
provide risk assessors with: (1) improved methods to understand the source-to-response
continuum, (2) advanced hazard characterization tools for prioritization and screening,
and (3) methods that enhance dose-response assessment and quantitative risk assessment.
The research efforts that were highlighted as part of the review cover each of these LTGs,
and have the potential to be broadly used within and outside the Agency. This included
efforts in high-throughput screening (HTS), modeling of molecular interactions with
biological targets, modeling of complex pharmacokinetic and pharmacodynamic
behaviors of small molecules, and database development and management, among others.
The portfolio provided a mix of short- and long-term deliverables. Many of the former
stand a good chance for application within program offices or other parts of ORD within
-------
December 2006 BOSC Computational Toxicology Letter Report
4
months. The research programs included those from external institutions. The
Subcommittee found that NCCT has effectively leveraged its limited resources.
One of the major aims of NCCT is to develop useful relational databases. This also
presents a significant challenge in managing the information. The Center should develop
a strategic plan for data integration and for constructing databases that should be
considered as information models.
Question 4: Because a large part of the mission of NCCT is to accelerate the use of
computational tools in the mission of the Agency, please comment on:
-Y- Part A: Do the proposed computational models have the potential to identify and
reduce uncertainties associated with the risk assessment process?
Yes, proposed computational models have the potential to identify and reduce
uncertainties associated with risk assessment. Additional opportunities outside the
mechanistic models (especially in biomarkers that indicate exposure but that are not
immediately or directly linked to toxicological response) may exist to fulfill NCCT's
mission.
-Y- Part B: Will these models be able to help identify susceptible populations and
compare potential risks to those populations with risks to the general and less
susceptible population?
Ultimately, these and other models within NCCT and outside the Agency can help
identify susceptible populations. Appropriately, models currently are being developed
for use in computational toxicology. Within 3-5 years, some of these models likely will
be sufficiently developed and validated to address susceptibility. "Susceptible
populations" may be defined to include life stages, gender, race, socioeconomic group,
species, and geographic distribution.
-Y- Part C: Is the coordination between model development and associated data
collection sufficient to avoid problems with the models being either over- or under-
determined?
Overall, data collection appears appropriately coordinated with model development. It
will be important to validate models based on genomic methodologies given the inherent
constraints in sample sizes, and other challenges, with these approaches.
Question 5: Please comment on the Computational Toxicology Implementation Plan,
focusing on the NCCT and Science To Achieve Results (STAR) components. Does it set
an achievable road map for accomplishing NCCT's major goals over the next 3 years, as
described in "A Framework for Computational Toxicology Research Program "? Does it
set realistic and relevant milestones, and clearly articulate projected program outputs
that will result in environmental outcomes?
The Implementation Plan consists of five research tracks that are intended to fulfill three
long-term goals:
-------
December 2006 BOSC Computational Toxicology Letter Report
5
1. EPA risk assessors use improved methods and tools to better understand and describe
linkages across the source-to-outcome paradigm;
2. EPA program offices use advanced hazard characterization tools to prioritize and
screen chemicals for toxicological evaluation;
3. EPA risk assessors and regulators use new models based on the latest science to
reduce uncertainties in dose-response assessment, cross-species extrapolation, and
quantitative risk assessment.
The research tracks that will support these long term goals are: (1) development of data
for advanced biological models; (2) information technologies development and
application; (3) prioritization method development and application; (4) providing tools
and system models for extrapolation across dose, life stage, and species; and
(5) advanced computational toxicology approaches to improve cumulative risk
predictions.
Each of the research areas is active. Tables 1 and 2 of the Center's Implementation Plan
provide details of projects and the outputs/outcomes and expected impacts of the projects.
NCCT has a core strength in modeling, and is expanding its expertise in informatics. The
Center is leveraging its position by outreach to other EPA laboratories and programs via
internal research funding and communities of practice, and externally via STAR grants,
including the external bioinformatics centers. The addition of the informatics centers in
particular strengthens NCCT's research in information technologies. This will be
strengthened further through the hiring of NCCT staff with informatics expertise. The
STAR grants greatly expand NCCT's capacities in the generation of high-information-
content data sets that will be needed to support model development.
Some challenges remain that will need to be overcome in the areas of database
development and management. More details are provided in our response to Question 2.
This will be especially important in the development and demonstration of biological
models derived from complex data sets. The Center is encouraged to do whatever it can,
within the boundaries of the grant process, to foster coordination of efforts between the
two external bioinformatics centers and NCCT's internal program.
The research has milestones with nearer term and longer term time horizons, which is
appropriate. It is clear that chemoinformatics tools and prioritization tools are well
underway and are likely to be applied by risk assessors and regulators within the next few
years. The timelines are realistic and the milestones will provide practical tools and
methods to program offices. In the shorter term, information databases such as DSSTox
and prioritization models such as ToxCast will be important tools for the pesticides and
toxic substances programs, and will demonstrate the utility of computational toxicology
in an applied setting. In the longer term, biological models such as the virtual liver, will
improve mechanistic understanding of toxicological response and provide support for
mechanism-based risk assessment.
-------
December 2006 BOSC Computational Toxicology Letter Report
6
The BOSC recommends that the NCCT develop a more detailed work plan for the virtual
liver model, and that this plan be more extensively reviewed by the Computational
Toxicology Subcommittee during its next annual review.
Question 6: Please comment on the progress made in the five major research track
thematic areas of the Computational Toxicology Research Program, and whether the
current/planned research will address the major goals in the framework. The Center has
made staffing additions and initiated new research over the past year. Based on these
changes, what is the Subcommittee 's view of the depth and breadth of the areas selected
for emphasis?
The Subcommittee believes that the research program covers the range of thematic areas.
Some areas, however, have deeper coverage than others. The areas of cumulative risk
assessment and cross-species extrapolation are still under-represented, but given the state-
of-the-science, it is appropriate to place limited emphasis and continue to leverage
research outside the Agency in these areas for the next 3-5 years. The staffing additions
in HTS, toxicogenomics, and biological modeling are all strong and have improved the
strength and breadth of NCCT. The planned staff additions in bioinformatics will be
critical to the continuing success of the Center. One of these additions should have
strong skills in data management systems.
Question 7: What evidence exists that NCCT is responsive to program office and
regional needs?
Most of the presentations addressed program office input in planning priorities and
approaches.
Some projects formed to support program office issues, such as carbamate cumulative
risk, DSSTox, and RefTox DB. The Subcommittee noted program office and regional
office staff as co-principal investigators on various projects. The Implementation Plan
references a role for the Computational Toxicology Implementation and Steering
Committee (CTISC), which could be useful, if sustained.
Question 8: Please comment on how effectively NCCT is communicating its research
program to EPA program offices, regional offices, and other stakeholders to inform their
environmental decision making.
NCCT has components of both a research and service center—it both initiates and
receives new ideas. For a young organization, NCCT has done very well in establishing
communication with its collaborators, contractors, and some stakeholders. The
establishment of CoPs and participation of internal clients is a good start to
communication within the Agency. Also of note is NCCT's establishment of monthly
videoconference presentations. Within the past year, NCCT has commendably given 21
presentations to various offices within EPA to raise awareness. Most of the other
communication activities seemed to be investigator-initiated. Given that the Center plans
to develop tools and methods that will be used by ORD and other EPA staff, NCCT
should establish a regularly scheduled plan for communication and updates. This process
will convey the sense that new ideas are welcomed by NCCT and allow the Center to
accept ideas and be aware of the needs of the program offices, regional offices, and
-------
December 2006 BOSC Computational Toxicology Letter Report
7
stakeholders. The establishment of such a process will enhance the marketing of tools
and methods developed by NCCT. One way to give Agency clients part ownership in the
Center is to invite them to BOSC reviews, such as this one, and ask them to share how
they are using NCCT's methods, tools, and information. The Subcommittee recommends
that NCCT communicate with the Regional Risk Assessor's Office and seek its
representation.
Question 9: Is the current research program designed to achieve environmental
outcomes? Please provide recommendations on how the NCCT can best measure these
outcomes.
The current program is designed to achieve environmental outcomes that are appropriate
to the Agency. Potential measures to determine these outcomes include:
-Y- Use of screening models for chemical prioritization.
-Y- Validation and use of genomics-associated biomarkers in field studies.
-Y- Use of computational models in the risk assessment process in the long term.
-Y- Success of databases (DSSTox, pesticides) in cleaning up and organizing disparate
databases and making them widely useful to environmental science and regulatory
communities.
-Y- Use of specific models (such as virtual liver, pyrethroid metabolism, macromolecular
modeling, physiologically based pharmacokinetic (PBPK) models, steroidogenesis
models, cumulative risk models, and so forth, by broader environmental science and
risk assessment communities.
In conclusion, the Computational Toxicology Subcommittee of the BOSC believes that
NCCT is making exceptional progress towards its mission. We are pleased to provide
advice on this important Center and look forward to future opportunities to offer
suggestions for improving the NCCT.
Sincerely,
James R. Clark
Chair, Board of Scientific Counselors
Previous I TOC
-------
UNITED STATES ENVIRONMENTAL PROTECTION AGENCY
WASHINGTON, D.C. 20460
JAN 1 9 2007
Dr. James R. Clark OFFICE OF
m. • r> J fv •*•<-,-. 1 RESEARCH AND DEVELOPMENT
Chair, Board of Scientific Counselors
Exxon Mobil Research & Engineering Co.
3225 Gallows Road, Room 3A412
Fairfax, VA 22037
Dear Dr. Clark:
The Office of Research and Development (ORD) would like to take this
opportunity to thank you and the members of the Board of Scientific Counselors
(BOSC) for the June, 2006 progress review of the National Center for
Computational Toxicology (NCCT). We especially thank the members of the sub-
committee who conducted the review, George Daston (Chair), James Clark,
Michael Clegg, Richard DiGiulio, Muiz Mumtaz, and John Quackenbush.
Enclosed with this letter is ORD's response to the comments and
recommendations on the NCCT in your letter report of December 12, 2006. Please feel
free to contact me if further information is needed.
We are pleased that the BOSC was very supportive of the NCCT and the direction
we are taking in this very important research program.
Again, thank you for your advice to ORD.
Sincerel
Teichman, Ph.D.
Acting Deputy Assistant Administrator
for Science
Enclosure
cc: Dr. George Datson
Dr. Michael Clegg
Dr. Richard DiGiulio
Dr. Muiz Mumtaz
Dr. John Quackenbush
Internet Address (URL) • http://www.epa.gov
Recycled/Recyclable • Printed with Vegetable Oil Based Inks on Recycled Paper (Minimum 50% Postconsumer content)
-------
•
~r
u
\
"
Office of Research and Development's (ORD) January 2007
Response to the Board of Scientific Counselors (BOSC) December
2006 Final Letter Report that Reviews ORD's National Center for
Computational Toxicology
BOSC Computational Toxicology Subcommittee:
Dr. George Daston (Chair)
Dr. James R. Clark
Dr. Michael Clegg
Dr. Richard DiGiulio
Dr. M. Moiz Mumtaz
Dr. John Quackenbush
Submitted by:
Dr. Robert Kavlock
Director
National Center for Computational Toxicology
Office of Research and Development
Previous I TOC I Next
-------
January 2007 ORD Response to BOSC December 2006 Computational Toxicology Final
Letter Report
ORD Response to Board of Scientific Counselors (BOSC) June 2006 Review of the
National Center for Computational Toxicology (NCCT)
The following is a narrative response to the comments and recommendations of the
BOSC review of ORD's National Center for Computational Toxicology that was held on
June 19-20, 2006, in Research Triangle Park, NC. The review was conducted by a
standing subcommittee of the BOSC. The subcommittee had previously reviewed the
NCCT on April 25-26, 2005, and ORD had responded to that review on September 8,
2005. In the second review, the BOSC noted that in its 16 months of existence, "NCCT
had made substantial progress in (1) establishing goals and priorities; (2) making
connections within and outside EPA to leverage the staff s considerable modeling
experience; (3) expanding its capabilities in informatics; and (4) significant contributions
to research and decision-making throughout the Agency." Furthermore, they noted,
"many of the recommendations made by the BOSC during its first review have been
acted on by NCCT."
Following are specific comments related to the charge questions made by the committee.
The charge questions are summarized in bold text, followed by the BOSC's comments in
italics, and ORD's response to the comments in regular type. Attached to this document
is a summary table of the BOSC comments and proposed ORD actions.
1. The first charge question asked for an evaluation of progress the Center made
during the past year in developing and maximizing connections and
collaboration within ORD and the rest of the Agency. Specifically, the
committee was asked about interactions, including the established Communities
of Practice (CoPs) and other notable examples and if there are other
opportunities that NCCT should explore.
All active CoPs have formal memberships and are chaired by NCCT staff. The Center
also has observed active participation among numerous EPA laboratories and
centers and several program offices. The Chemoinformatics and Chemical
Prioritization CoPs already have demonstrated outreach to outside agencies, such as
the National Institutes of Health (N1H), National Institute of Environmental Health
Sciences (N1EHS), and National Toxicology Program (NTP). Some are working with
or soliciting international and private sector collaboration. The CoPs have been
effective in focusing on defining problems and suggesting solutions, agreeing on
modeling approaches and database issues, and setting up forums and workshops for
discussions. They will be responsible for leading a better coordinated effort within
EPA and among agencies. The Subcommittee believes that establishing a Cumulative
Risk CoP is worthy of pursuit. Such a CoP would provide significant opportunities to
define areas for improvement in risk assessment practices and could provide
inventory tools and other benefits. NCCT should consider whether it would like to
provide a facilitator role or leadership role in this area.
Response: ORD appreciates the Committee's recognition of NCCT's current efforts.
Also, ORD agrees on the importance of pursing the formation of other relevant CoPs.
-------
January 2007 ORD Response to BOSC December 2006 Computational Toxicology Final
Letter Report
Due to the small size of the NCCT staff, ORD is concerned these could over tax the
staff. NCCT is committed to supporting the three existing CoPs (Chemoinformatics,
Chemical Prioritization, and Biological Modeling) and will take steps to ensure their
vitality. For other CoP ideas, ORD will look across all the Labs and Centers to see if
relevant similar work groups and committees are already established and could be
amended to address such issues, or encourage the establishment of ones for which no
precedent can be formed. In particular, we have further considered a CoP centered on
Cumulative Risk, as proposed during the review. While still favoring the idea, we
have realized from the activities of the existent CoPs that they function best when
aligned along a well defined issue and have a commonly identified goal. For
Cumulative Risk, our current opinion is the issue and goal of a dedicated CoP must be
better refined, and we hope to work with other Agency scientists to foster this
refinement and development.
With regard to other opportunities for exploration, the subcommittee suggested
NCCT seek broader program office input. Additionally, CoPs covering areas such as
Mixtures, Cross-Species Extrapolation, Population/Systems Dynamic Models, and
Multimedia Fate and Effects Modeling should be considered for either NCCT use or
ORD's broader use.
Response: ORD also agrees with the Committee's recommendation of broader
program office input. The Center is in the process of increasing the number and
frequency of contacts and meetings with Program and Regional Offices. Formal
presentations are often part of those contacts. For example, we have recently given
overview presentations of the program to EPA's Science Policy Council and Regional
Risk Assessors, both of which drew considerable interest. In addition, we are
scheduled to present a detailed overview of the ToxCast program to Office of
Prevention, Pesticides, and Toxic Substances (OPPTS) on February 1, 2007, in
Washington, D.C. Finally, the upcoming International Forum on Computational
Toxicology being organized by NCCT on behalf of ORD is expected to provide
opportunities for contact within and outside the Agency. NCCT is preparing and
executing several means to better communicate progress, outputs, and abilities to the
rest of the Agency, and in particular we are working on ways to improve the content
of our internet site. NCCT staff are also organizing a series of short courses in the
field of computational biology for Agency staff and others as well.
2. The second charge question dealt with interactions with the two newly funded
STAR Environmental Bioinformatics Seminars.
Individually, the Bioinformatics Centers were viewed as excellent choices, each
providing expertise and resources largely complementary to each other and to the
NCCT with little overlap. Although both Centers are just beginning their work with
EPA, there is great opportunity for synergy in developing new approaches for the
analysis of toxicogenomic data and integration of diverse information necessary to
place these data into an appropriate context. Integration of the external
Bioinformatics Centers and the programs within NCCT will occur following hiring of
-------
January 2007 ORD Response to BOSC December 2006 Computational Toxicology Final
Letter Report
one senior and one junior bioinformatics scientist. This may not represent sufficient
personnel, however, to allow NCCT to fully support its overall mission.
Response: NCCT is continuing to work with ORD's National Center for
Environmental Research (NCER) to ensure the Bioinformatic Centers enhance the
current state of the science in this critical research area. The hiring of Dr. Richard
Judson as a Title 42 Senior Bioinformatician in NCCT has greatly facilitated the
interactions with the Centers. Dr. Judson coordinates a monthly EPA-wide seminar
program (Info on Informatics), which features one of the project areas from each of
the Centers. The goal of the series is to promote EPA awareness of the objectives of
the Centers and to help facilitate development of interactions with them. Drs. Judson
and Kavlock, together with staff from NCER, performed a site visit to the University
of Medicine & Dentistry of New Jersey (UMDNJ) Center in December, which
resulted in very rewarding discussions concerning future interactions. A second site
visit is planned for early 2007 by Dr. Judson and several other EPA scientists to
further develop ties. Due to the geographical closeness, interactions with the
University of North Carolina (UNC) Center have been more frequent and targeted. A
predoctoral student has been identified to interact on matters related to genomic data
storage, analysis and interpretation, and several interactions have developed in
conjunction with the chemical prioritization efforts of the ToxCast program.
Consequently, NCCT needs to develop a more comprehensive strategic plan for data
collection, management, and integration through creation of databases that model the
structure of the underlying information and its potential use.
Response: Dr. Judson, working in conjunction with Dr. Imran Shah, our Title 42
Computational Systems Biologist (both joined the Center in September), has also
taken the lead in developing an overall framework for information management
within NCCT. In response to the strong recommendation of the BOSC related to the
need to adequately address this topic, we propose a targeted briefing on our approach
to information management and information technology for the BOSC sometime in
the May-June 2007 time frame.
It was noted that there exists a need within the field for trained personnel in
computational toxicology. In addition to the existing postdoctoral program, one
feasible approach would be to institute a career development award similar to the
NIH "K" awards that would provide mentor ed training andresearch to more senior
personnel.
Response: We appreciate the recommendation to strengthen our training component,
as we view this as one of our three critical functions (in addition to providing a
service function to other ORD researchers and conducting innovative research on the
use of computational models in risk assessment). We will work with appropriate
human resource components within EPA to explore options for career development
training of other scientists. We have also engaged advanced discussions within
-------
January 2007 ORD Response to BOSC December 2006 Computational Toxicology Final
Letter Report
NCCT on hosting several advanced training courses for EPA staff. Lead topics are
Physiologically Based Pharmacokinetic Models and Chemical Prioritizations Tools.
3. The third charge question was designed to promote discussion about the
potential of NCCT research programs making impacts on Agency function, and
how well the NCCT was leveraging its resources in this regard.
The portfolio provided a mix of short- and long-term deliverables. Many of the
former stand a good chance for application within program offices or other parts of
ORD within months. The research programs included those from external
institutions. NCCT has leveraged its limited resources to good effect.
Response: The NCCT program was designed with the goal of having some short-to
intermediate- term deliverables, as well as some projects with longer timelines, and
we appreciate the recognition of the value of this by the BOSC. We have continued
to work to best leverage our resources, and present three examples of related efforts
since the review. The first is the establishment of an Interagency Agreement with the
National Chemical Genomics Center (NCGC) of the NIH to conduct quantitative,
high throughput screening analysis of ToxCast chemicals against a number of nuclear
receptor assays. This TAG provides NCCT with a direct link to NIH's Molecular
Library Initiative. This Initiative will be providing extremely cost effective data to us
over the next 5 years, as it taps into a well established infrastructure geared to running
these types of assays. NCCT has also started a series of high level meetings with the
management of the National Health and Environmental Effects Research Laboratory
(NHEERL), the National Environmental Research Laboratory (NERL), and the
National Risk Management Research Laboratory (NRMRL) to best define working
relationships between these groups and how to target the computational toxicology
resources available in those laboratories. The first of these meeting was held with
NHEERL on January 18, 2007. Finally, we have been working closely with staff in
NCER to define the next Request for Applications (RFA) in computational
toxicology. The objective of this RFA will be to establish several academic centers
working in areas of computational systems biology, and we are excited about the
prospect of this activity to move us forward more rapidly in programs such as the
Virtual Liver, as well as in developing the computer infrastructure and computational
approaches to systems biology from a toxicological viewpoint.
One of the major aims of NCCT is to develop useful relational databases. This also
presents a significant challenge in managing the information. The Center should
develop a strategic plan for data integration and for constructing databases that
should be considered as information models.
Response: As noted in the response to Q2, NCCT is developing a strategic plan for
data information and management, and is prepared to bring its plan to the BOSC for
comment within the next 6 months.
-------
January 2007 ORD Response to BOSC December 2006 Computational Toxicology Final
Letter Report
4. The fourth charge question consisting of three parts focused on NCCT's mission
to accelerate the use of computational tools in the Agency mission. The first part
more specifically asked the committee to comment on whether the proposed
computational models have the potential to identify and reduce uncertainties
associated with risk assessment.
Yes, proposed computational models have the potential to identify and reduce
uncertainties associated with risk assessment. Additional opportunities outside the
mechanistic models (especially in biomarkers that indicate exposure but that are not
immediately or directly linked to toxicological response) may exist to fulfill NCCT's
mission.
Response: ORD is pleased the committee endorses the selection of computational
models and the planned approach to develop, test, and use these models. ORD agrees
there are other opportunities that can use other than mechanistic models including
exposure biomarkers. Since the review, new expert staff have come on board with
systems modeling expertise. Plans are being formulated for an extensive liver model
that can simulate its molecular processes and predict the possible toxic effects of
chemicals on liver function. As part of this effort, modules at many different levels
and complexities will be formulated, including those that relate data to tissue outcome
without detailed specific knowledge of mechanism. Further, work has now begun on
computational approaches to apply advanced statistical and machine learning
methods to evaluate human exposure and environmental health data. Target data
include multiple types of biomarker and environmental exposure information.
The second part of the charge question addresses the models' ability to help identify
susceptible populations and compare the risks to those populations with the risks to
the general population.
Ultimately, these and other models within NCCT and outside the Agency can help
identify susceptible populations. Appropriately, models currently are being
developed for use in computational toxicology. Within 3-5 years, some of these
models likely will be sufficiently developed and validated to address susceptibility.
"Susceptible populations " may be defined to include life stages, gender, race,
socioeconomic group, species, and geographic distribution.
Response: ORD accepts this endorsement and will continue in its computational
modeling activities to consider this an important goal.
The last part of this question asked whether there was sufficient coordination between
model development and associated data to avoid having the models being either over-
or under-determined.
Overall, data collection appears appropriately coordinated with model development.
It will be important to validate models based on genomic methodologies given the
inherent constraints in sample sizes, and other challenges, with these approaches.
-------
January 2007 ORD Response to BOSC December 2006 Computational Toxicology Final
Letter Report
Response: ORD agrees and recognizes the importance and challenge of validating
and testing of all models. NCCT modelers have established close working
collaborations with laboratory biologists and chemists who are conducting many of
the experiments or gathering and using existing data for model building and testing.
5. The fifth charge question addressed whether the Computational Toxicology
Implementation Plan described an achievable roadmap and set forth realistic
milestones and outputs.
Each of the research areas is active. NCCT has a core strength in modeling, and is
expanding its expertise in informatics. The Center is leveraging its position by
outreach to other EPA labs and programs via internal research funding and
communities of practice, and externally via STAR grants and the external
bioinformatics centers. The addition of the informatics centers in particular
strengthens NCCT's research in information technologies. This will be strengthened
further through the hiring of NCCT staff with informatics expertise. The STAR grants
greatly expand NCCT's capacities in the generation of high-information-content data
sets that will be needed to support model development.
There are still some challenges that will need to be overcome in the areas of database
development and management. More details are provided in our response to
question 2. This will be especially important in the development and demonstration
of biological models derived from complex data sets.
The research has milestones with nearer term and longer term time horizons, which is
appropriate. It is clear that chemoinformatics tools andprioritization tools are well
underway and are likely to be applied by risk assessors and regulators within the next
few years.
Response: Some challenges remain that will need to be overcome in the areas of
database development and management. More details are provided in our response to
Question 2. This will be especially important in the development and demonstration
of biological models derived from complex data sets. The Center will do whatever it
can, within the boundaries of the grant process, to foster coordination of efforts
between the two external bioinformatics centers and NCCT's internal program.
The BOSC recommends the NCCT develop a more detailed work plan for the virtual
liver model, and that this plan be more extensively reviewed by the Computational
Toxicology Subcommittee during its next annual review
Response: Development of the virtual liver model has gained momentum with the
hiring of Dr. Imran Shah, a Title 42 Computational Systems Biologist. He has been
leading biweekly discussions with relevant staff members from NCCT, NHEERL,
NERL and NCEA to articulate reasonable goals and expectations for this effort.
NCCT proposes we schedule a teleconference with the BOSC in the third quarter of
-------
January 2007 ORD Response to BOSC December 2006 Computational Toxicology Final
Letter Report
2007 to present a briefing and lead a discussion on development of the Virtual Liver
activity.
6. The sixth charge question addressed the depth and breadth of the resources
directed at fulfilling the Implementation Plan.
The subcommittee believes that the research program covers the range of thematic
areas. Some areas, however, have deeper coverage than others. The areas of
cumulative risk assessment and cross-species extrapolation are still under-
represented, but given the state-of-the-science, it is appropriate to place limited
emphasis on these areas for the next 3-5 years. The staffing additions in HTS,
toxicogenomics, and biological modeling are all strong and have improved the
strength and breadth ofNCCT. The planned staff additions in bioinformatics will be
critical to the continuing success of the Center. One of these additions should have
strong skills in data management systems.
Response: We agree and the recently hired Title 42 scientists who are filling two
critical gaps in our expertise. Their contribution will be evident when we brief the
BOSC on our information management and virtual liver programs. Together, they
provide expertise in informatics and advanced computational methods, and are
working in key areas for the NCCT. We have reserved a more junior level position to
support the programming needs of these two members, and are in the processing of
re-orienting the support provided to us by the Environmental Modeling and
Visualization Laboratory of the Office of Environmental Information, which has been
supplying a variety of support activities to the Computational Toxicology Program
for the past two years. Finally, we are in advanced discussions with a senior level
scientist in the area of toxicogenomics. We will know shortly whether this additional
Title 42 position within the NCCT will provide senior leadership in genomics.
7. The seventh charge question asked about evidence that NCCT is being
responsive to program and regional office needs.
Most of the presentations addressed program office input in planning priorities and
approaches.
Some projects formed to support program office issues, such as carbamate
cumulative risk, DSSTox, andRefToxDB. The Subcommittee noted program office
and regional office staff as co-principal investigators on various projects. The
Implementation Plan references a role for the Computational Toxicology
Implementation and Steering Committee (CTISC), which could be useful, if sustained.
Response: ORD thanks the committee for its response and encouragement. While
ORD recognizes the usefulness of the CTISC, its role is being reevaluated to
determine if, in its current state, this is the most effective manner to insure wide
involvement and support from the Program and Regional Offices as well as others.
As mentioned in our discussion under Question 1, we are engaging many other
-------
January 2007 ORD Response to BOSC December 2006 Computational Toxicology Final
Letter Report
opportunities, including the Communities of Practice, to this end. We are also taking
the opportunity to brief various EPA groups about the research program, with recent
presentations to the Science Policy Council, the Regional Risk Assessors (which
consists of both EPA and state risk assessors involved in Superfund sites), and an
upcoming presentation to the Office of Pesticide Programs.
8. The eighth charge question dealt with communication issues.
NCCT has components of both a research and service center—it both initiates and
receives new ideas. For a young organization, NCCT has done very well in
establishing communication with its collaborators, contractors, and some
stakeholders. The establishment ofCoPs and participation of internal clients is a
good start to communication within the Agency. Also of note is NCCT's
establishment of monthly videoconference presentations. Most of the other
communication activities seemed to be investigator-initiated. Given that the Center
plans to develop tools and methods that will be used by ORD and other EPA staff,
NCCT should establish a regularly scheduled plan for communication and updates.
This process will convey the sense that new ideas are welcomed by NCCT and allow
NCCT to accept ideas and be aware of the needs of the program offices, regional
offices, and stakeholders. The establishment of such a process will enhance the
marketing of tools and methods developed by NCCT. One way to give Agency clients
part ownership in the Center is to invite them to BOSC reviews, such as this, and ask
them to share how they are using NCCT's methods, tools, and information. The
Subcommittee recommends that NCCT communicate with the Regional Risk
Assessor's Office and seek its representation. Within the past year, NCCT has
commendably given 21 presentations to various offices within EPA to raise
awareness.
Response: We agree and NCCT is paying close attention to this. Staff are regularly
looking for and finding opportunities to interact with other scientists, organizations,
and Agency programs. In addition to scientific publications and presentations,
feature articles are often written, such as one in the January, 2007 issue of EM
highlighting the research activities of NCCT. NCCT is in the process of enhancing
the Computational Toxicology website to communicate more effectively and in a
more timely fashion. A senior ORD communications staff member and an intern in
the communications office are working with NCCT to develop, publish, and
disseminate appropriate messages. We expect to be releasing periodic updates on
progress in implementing the ToxCast program, and we just completed a fact sheet
describing the Interagency Agreement we just signed with NCGC/NIH. This is the
first tangible component of the ToxCast program. As the various supporting
contracts are awarded over the next six months, we will be posting updates on our
website. We also will be using the upcoming International Science Forum on
Computational Toxicology to engage a large number of Agency scientists. Finally, as
noted above, at their request we briefed the Regional Risk Assessors on the program,
and received a number of emails following the presentation asking for additional
details.
-------
January 2007 ORD Response to BOSC December 2006 Computational Toxicology Final
Letter Report
9. The ninth and final charge question asked if the current research program was
designed to achieve environmental outcomes and how those outcomes could be
measured.
The current program is designed to achieve environmental outcomes that are
appropriate to the Agency. Potential measures to determine these outcomes include:
? Use of screening models for chemical prioritization.
? Validation and use of genomics-associated biomarkers infield studies.
? Use of computational models in the risk assessment process in the long term.
? Success of databases (DSSTox, pesticides) in cleaning up and organizing
disparate databases and making them widely useful to environmental science
and regulatory communities.
? Use of specific models (such as virtual liver, pyrethroid metabolism,
macromolecular modeling, physiologically basedpharmacokinetic (PBPK)
models, steroidogenesis models, cumulative risk models, and so forth, by
broader environmental science and risk assessment communities.
Response: ORD thanks the committee for these suggestions. NCCT is looking to
develop specific ways to regularly gather information to apply to those measures.
Progress and results will be shared with the committee at future meetings. Our
current thinking is it would be best to engage the BOSC over the next year on specific
projects, particularly ToxCast, the Virtual Liver, and our Information Management
plans. As these are programs still in rapid phases of evolution, dialogue with the
BOSC would be beneficial to use in refining their approaches. At the discretion of
the BOSC, these could be done either in individual teleconferences over the next 6-9
months, or at a face-to-face meeting, focusing on the three topic areas.
We suggest the next all encompassing review of the program be held in the first half
of 2008. At that time, we would have made considerable progress on a number of
research fronts that would allow us to change the main purpose of the review from
reviewing strategic directions to analyzing the research outcomes.
-------
Computational Toxicology Program
Summary of BOSC Comments and Recommendations from December 2006 Letter
Report and Proposed ORD Actions
Recommendation
Action Items
Timeline
Establish a Community of Practice
(CoP) for Cumulative Risk
We favor the idea. We have realized from the
activities of the existent CoPs that they function
best when aligned along a well defined issue and
have a commonly identified goal. Upon further
reflection since the BOSC review, our current
opinion is that we need to better refine the issue
and goal of a dedicated Cumulative Risk CoP, and
we hope to work with other Agency scientists to
foster its conceptualization and development.
2007 and 2008
NCCT seek broader program office
input. Additionally, CoPs covering
areas such as Mixtures, Cross-
Species Extrapolation,
Population/Systems Dynamic
Models, and Multimedia Fate and
Effects Modeling should be
considered for either NCCT use or
ORD's broader use.
We are committed to an increased number and
frequency of contacts and meetings with Program
and Regional Offices with formal presentations;
other noteworthy communication events include
the International Forum on Computational
Toxicology; short training courses and an
improved Web site communication. Regarding
the establishment of additional CoPs, ORD is
concerned that these could over tax the small staff
of NCCT. NCCT remains committed to
supporting the three existing CoPs) and taking
steps to ensure their vitality. For other CoP ideas,
ORD will look across all the Labs and Centers to
see if relevant similar work groups and
committees are already established that could be
amended to address such issues, and encourage
establishment of ones for which no precedent can
be found.
2007
Integration of the external
Bioinformatics Centers and the
programs within NCCT will occur
following hiring of one senior and
onejuniorbioinformatics scientist.
This may not represent sufficient
personnel, however, to allow NCCT
to fully support its overall mission.
NCCT continues to work with NCER to ensure
that the Bioinformatic Centers enhance the current
state of the science in this critical research area.
The hiring of Dr. Richard Judson as a Title 42
Senior Bioinformatician in NCCT has greatly
facilitated the interactions with the Centers. Dr.
Judson coordinates a monthly EPA wide seminar
program (Info on Informatics), which features one
of the project areas from one of the Centers.
Regular visits between the Centers and
EPA/NCCT are scheduled and have begun.
Initiated and on-going
NCCT needs to develop a more
comprehensive strategic plan for data
collection, management, and
integration through creation of
databases that model the structure of
the underlying information and its
potential use.
Dr. Judson, working in conjunction with Dr. Imran
Shah, our Title 42 Computational Systems
Biologist, has taken the lead in developing an
overall framework for information management
within NCCT. hi response to the strong
recommendation of the BOSC related to the need
to adequately address this topic, we propose a
targeted briefing on our approach to information
management and information technology to the
BOSC sometime in mid 2007
On-going; proposed
briefing for sub-
committee in mid-2007.
-------
Recommendation
Action Items
Timeline
There is a need within the field for
trained personnel in computational
toxicology, hi addition to the
existing postdoctoral program, one
feasible approach would be to
institute a career development award
similar to the MH "K" awards that
would provide mentored training and
research to more senior personnel.
We view this as one of our three critical functions
(in addition to providing a service function to
other ORD researchers and conducting innovative
research on use the use of computational models
in risk assessment). We will work with
appropriate human resource components within
EPA to explore options for career development
training of other scientists. We are also having
advanced discussions within the NCCT on hosting
several advanced training courses for EPA staff.
Lead topics are Physiologically Based
Pharmacokinetic Models and Chemical
Prioritizations Tools.
Start in 2007 and
continue beyond
Additional opportunities outside the
mechanistic models (especially in
biomarkers that indicate exposure but
that are not immediately or directly
linked to toxicological response) may
exist to fulfill NCCT's mission.
ORD is considering these and all types of model
structures in its developing model program. This
is particularly true in NCCT virtual liver project
currently under design.
On-going
It will be important to validate
models based on genomic
methodologies given the inherent
constraints in sample sizes, and other
challenges, with these approaches.
We agree and recognize the importance and
challenge of validating and testing of all models.
The NCCT modelers have established close
working collaborations with laboratory biologists
and chemists who are conducting many of the
experiments or gathering and using existing data
for model building and testing.
On-going
NCCT develop a more detailed work
plan for the virtual liver model, and
that this plan be more extensively
reviewed by the Computational
Toxicology Subcommittee during its
next annual review
Development of the Virtual Liver has gained
momentum with the hiring of Dr. Imran Shah, a
Title 42 Computational Systems Biologist. He has
been leading biweekly discussions with relevant
staff members from NCCT, NHEERL, NERL and
NCEA to articulate reasonable goals and
expectations for this effort. ORD is committed to
briefing the BOSC on the status of this project in
mid 2007.
3M Quarter of 2007 for a
teleconference on plans
and progress of the
virtual liver project with
committee.
NCCT should establish a regularly
scheduled plan for communication
and updates. NCCT should invite
Agency clients to BOSC reviews,
such as this, and ask them to share
how they are using NCCT's methods,
tools, and information. The
Subcommittee recommends that
NCCT communicate with the
Regional Risk Assessor's Office and
seek its representation.
Staff are regularly looking for and finding
opportunities to interact with other scientists,
organizations, and Agency programs. NCCT is in
the process of enhancing the Computational
Toxicology web site to communicate more
information in a more timely fashion. A senior
ORD communications staff member and an intern
in communications are now working with NCCT
to develop, publish, and disseminate appropriate
messages. We expect to be releasing periodic
updates on progress in implementing the ToxCast
program, and we just completed a fact sheet
describing the Interagency Agreement we just
signed with NCGC/NIH. As the various
supporting contracts are awarded over the next six
months, we will be posting updates on our
website. We also will be using the upcoming
International Science Forum on Computational
Toxicology to engage a large number of Agency
On-going and continuous
-------
Recommendation
Action Items
Timeline
scientists. Finally, as noted above, at their request
we briefed the Regional Risk Assessors on the
program, and received a number of emails
following the presentation asking for additional
details.
Committee suggested several
potential measures to determine
outcomes
NCCT is looking to develop specific ways to
regularly gather information to apply to those
measures. Progress and results will be shared with
the committee at future meetings. Our current
thinking is that it would be best to engage the
BOSC over the next year on specific projects,
particularly ToxCast, the virtual liver, and our
information management plans. As these are
programs still in rapid phases of evolution,
dialogue with the BOSC would be beneficial to
use in refining their approaches. At the discretion
of the BOSC, these could be done either in
individual teleconferences over the next 6-9
months, or at a face-to-face meeting focused on
the three topic areas.
2007 and beyond
-------
S
BOARD OF SCIENTIFIC COUNSELORS
September 16, 2008
Dr. George Gray
Assistant Administrator
Office of Research and Development
U.S. Environmental Protection Agency
Dr. Robert Kavlock
Director
National Center for Computational Toxicology
U.S. Environmental Protection Agency
Dear Drs. Gray and Kavlock:
This is a letter report from the Board of Scientific Counselors (BOSC)
reviewing the National Center for Computational Toxicology (NCCT). The
Computational Toxicology Subcommittee of the BOSC Executive
Committee reviewed NCCT's progress and plans during a 2-day meeting
held December 17-18, 2007, at the EPA facility in Research Triangle Park,
North Carolina. The BOSC Subcommittee consists of George Daston
(Chair), James Clark, Richard DiGiulio, Muiz Mumtaz, John Quackenbush,
and Cynthia Stokes.
This is the third review of NCCT conducted by the BOSC. The Subcom-
mittee was very pleased with the progress that the Center has made towards
its goals. NCCT first became operational in February 2005; during the 2.5
years between its establishment and this review, NCCT has made substan-
tial progress in establishing priorities and goals; making connections within
and outside EPA to leverage the staffs considerable modeling expertise;
expanding its capabilities in informatics; and making significant contribu-
tions to research and decision-making throughout the Agency. We are
pleased to see that informatics tools developed by the Center already are
being used by program offices, and that the program offices are taking
advantage of the expertise of the Center in developing critical elements for
risk assessment, such as a biologically-based dose-response (BBDR) model
for arsenic, an environmental contaminant of considerable public health
importance. Many of the recommendations made by BOSC during its
earlier reviews have been acted on by NCCT. This includes improved
capabilities in bioinformatics through the funding of two external centers
and in informatics and systems biology through staff hires; expansion of its
A Federal Advisory Committee for the U.S. Environmental Protection Agency's Office of Research and Development
Previous
TOC
Next
-------
September 2008 BOSC Computational Toxicology Letter Report
technical approaches to even more programs within the Agency; and the formation of an
extensive collaboration with the National Institute of Environmental Health Sciences (NIEHS)
and the National Human Genome Research Institute (NHGRI) for its ToxCast project.
The purpose of the December 2007 review was to continue to provide NCCT with advice on the
progress the Center has made, in the past year, in fulfilling its mission and strategic goals. In
particular, the Subcommittee addressed six charge questions for five NCCT activities (ToxCast,
Informatics Technology/Information Management (IT/IM) activities, Virtual Liver,
Developmental Systems Biology, and Arsenic BBDR). The Subcommittee's responses to these
questions follow.
Charge Question 1: Does the scope and involvement of expertise in the project reflect activities
consistent with the function of a Center?
The NCCT was founded only a few years ago and has been achieving a critical mass of expertise
through selective hiring, external grants, and the formation of connections with other groups of
experts within EPA. The purpose of this question was to gauge the progress of the Center in
achieving the level of expertise needed to pursue its mission.
The staff working in NCCT and those scientists involved from outside the Agency who are
working as collaborators are highly qualified in various aspects of computational toxicology. The
Center's effort to solidify formal agreements in terms of memoranda of understanding (MOUs),
cooperative research and development agreements (CRADAs), etc., with various organizations
has opened up a diversity of quality opportunities to leverage and enhance Office of Research
and Development (ORD) efforts. A timely example is the February 14, 2008, announcement of
the collaboration between NIEHS, NHGRI, and EPA's NCCT. As described in the press release,
this collaboration leverages the strengths of each group to use high-speed, automated screening
robots to test suspected toxicants using cells and isolated molecular targets.
The staff and collaborators at the center have the appropriate expertise and insights. The utility of
the tools and deliverables can be enhanced if the staff moves toward being more explicit on how
the tools under development support EPA risk assessments. Some of the ORD researchers seem
to be searching for an application for their sophisticated tools, and discussions with Agency staff
practicing risk assessments (Office of Pollution Prevention and Toxics [OPPT]; Office of Water,
Office of Wastewater Management; Office of Prevention, Pesticides, and Toxic Substances
[OPPTS], etc.) could provide direction as to the appropriate milestones and deliverables for these
efforts. The BOSC reviews and the Center would benefit if representatives from these Agency
offices attended BOSC reviews to ensure that all parties understand how NCCT's efforts address
the most relevant needs of the Agency. The BOSC wants to ensure that this advice is seen as
encouragement to reach out to risk assessment practitioners. The ongoing work in developing the
analytical approaches and information databases is of high technical quality, as the Center staff
and collaborators are working on many new and exciting approaches. By holding research
planning discussions with risk assessment practitioners, the applications of the computational
toxicology tools and resources can be directed to ensure the most relevant and efficient use of
data and models.
Previous I TOC
-------
September 2008 BOSC Computational Toxicology Letter Report
One challenge for the center staff involved in developing informatics datasets will be to develop
efficient and effective ways to handle the wealth of data available in some areas to avoid
redundancies of data entries and to focus on the most informative data. Again, interactions with
various program offices and their risk assessment activities should provide a basis to set the
long-term goals for the Informatics/Data management team. This will allow the development of
structured short-term and mid-term activities needed to meet the long-term goals.
The BOSC noted that it remains somewhat unclear how the Center intends to use ToxCast and
associated analyses to approach risk assessment. For instance, species-to-species translation was
mentioned, and the data are being obtained from multiple species, not just humans, but how the
different species data will be reconciled was not discussed. Although the primary goal of the
ToxCast project is prioritization of chemicals for detailed risk assessment, not the risk
assessment itself, it is interesting to contemplate how the projected database and analysis might
be directly relevant. Similarly, it was noted that an early decision regarding ToxCast was that
ecology and paths of exposure were not going to be addressed in this project (at least not
initially). Nonetheless, at several points, paths of exposure arose during the review because of
their obvious relevance. The Subcommittee is prompted to ask how it might be addressed in
future work.
The Subcommittee also noted that the means of using the eventual Virtual Liver models for
actual risk assessment at EPA is unclear. The BOSC encourages additional thought and efforts
along these lines, in collaboration with the appropriate EPA program office personnel. This is not
a criticism of the current project vision by any means, but because direct or indirect application
to risk assessment would be a fantastic result, it seems prudent to consider the possibility earlier
rather than later.
Charge Question 2: Are the goals and milestones suitably described, ambitious, and
innovative?
The purpose of this question was to determine whether NCCT is performing its mission of
providing novel approaches to the practical problems of toxicology and risk assessment that are
needed by EPA.
For the Center overall, the answer to this question is "yes." In particular, the goals of the Center
are well-described, very ambitious and innovative, as well as important for the future of research
at EPA. The issue of "milestones" is somewhat more complex, in part due to the varying levels
of maturity for Center components. In most cases, previous accomplishments and current
activities are well described, but more detail concerning projected future milestones would be
helpful. It is recognized, however, that these projects are very innovative and substantial
flexibility is appropriate. This is particularly true for less mature but highly creative projects such
as the Virtual Liver and Virtual Embryo. Also, in considering goals and milestones, it may be
appropriate to consider the timely integration of each project's accomplishments into the
Agency's risk assessment activities. In the following paragraphs, Charge Question 2 is addressed
in the context of the five major Center activities discussed at the review meeting.
Previous I TOC
-------
September 2008 BOSC Computational Toxicology Letter Report
ToxCast. ToxCast is the most mature of the Center's projects, which is appropriate considering
that it is most central to NCCT's overall mission. The goal of this project, to provide a cost and
time-efficient methodology for screening and prioritizing chemicals of concern to the Agency
(~11,000 by current estimates) is suitably ambitious, and is well described. Progress on this
project has been very strong and these accomplished milestones are very well described. Future
plans for the project also are well described, although a more detailed time table for milestones
past 2008 would be helpful.
IT/IM Activities. This project is highly symbiotic with ToxCast, and the success of one is highly
dependent on the success of the other. By extension, the success of the NCCT overall depends
upon the success of these two projects. The comments above for ToxCast also pertain to the
Informatics project. The project is highly and suitably ambitious, and its goals and substantial
progress are well described. Again, future plans are described well in a general way, but more
detail concerning future milestones (beyond 2008, which is well described) would be
appropriate.
Virtual Liver. Although narrower in scope than the foregoing projects, the Virtual Liver project
is very ambitious; it also is relatively young, apparently becoming fully operational with the
arrival of Dr. Imran Shah in September 2006. Its fit with the goals of NCCT is perhaps less clear
than the previous two projects; it is more "visionary" in nature, and less directly applicable to
risk assessment, as described by one of the EPA scientists involved. The goals of the project and
the nature of research to be performed to achieve those goals are clearly described. There is some
concern that this project may be overly ambitious. It may be helpful if key objectives were
delineated and prioritized, perhaps indicating achievements that are critical to the success of the
project and those that are highly desirable. Milestones for tracking the project's progress are not
apparent, particularly in later years (3-5). This relatively young and very innovative project
requires considerable flexibility, however, so the lack of detailed milestones in later years is very
reasonable.
Developmental Systems Biology (Virtual Embryo). This project is at a substantially earlier
stage than the Virtual Liver project; it is led by Dr. Thomas Knudsen who joined NCCT in
September 2007. The issues of goals and milestones are essentially the same as for the Virtual
Liver, that is, strong on the former, but understandably weaker on the latter. It is the
Subcommittee's expectation that a more concrete research plan with goals and milestones will be
developed over the coming months.
Arsenic BBDR. This also is a relatively new effort, with planning beginning in 2006. This
project is unusual among NCCT projects in that it is oriented toward a specific chemical with a
specific issue (Safe Drinking Water Act revisions) rather than an approach developed with
diverse chemicals in mind. However, this project is likely to inform the eventual development of
other biologically-based dose-response models and their application to risk assessments by the
Agency. Thus, in addition to informing the controversial issue of arsenic risk assessment, the
project is more broadly relevant to the mission of the NCCT. The goals of the project are very
clear and well described. Milestones, however, are not stated, and may be particularly important
for this project, which has a clear deadline (2011) in order to be useful for the 2012 Safe
Drinking Water Act review cycle.
Previous I TOC
-------
September 2008 BOSC Computational Toxicology Letter Report
Charge Question 3: Are there significant gaps in the approach that can be pointed out at this
point in the evolution of the project?
ToxCast. Dr. Dix and the ToxCast project contributors are commended for their progress in this
activity in terms of specification of desired data and the contracting of various entities to obtain
these data. The data acquisition is clearly well under way. The main gap noted is relevant to both
the ToxCast and IT/IM activities. Specifically, the Subcommittee notes that the structural
specification of the database for compilation and rigorous quantitative analysis of the ToxCast
data remains unclear. Because the data types are highly heterogeneous and the dataset is very
large, developing these structural specifications will be a challenge that the Subcommittee
suggests should be addressed as soon as possible. The IT/IM team acknowledges that this area is
a significant challenge (e.g., the description in the write-up provided to the Subcommittee prior
to the review meeting). One suggestion is that the ToxCast team compile a list of some specific
use cases, for example, specific questions that they intend to address with the database. This will
help make concrete the needed database attributes that will allow the analysis for the chemical
prioritization that is the end goal of the ToxCast project
IT/IM Activities. The IT/IM activity group has clearly made significant progress since the last
BOSC Subcommittee meeting in terms of specification and development of various software and
database tools for storing and accessing various toxicology data in existence as well as being
generated (e.g., in the ToxCast project). The fact that their ToxRef database and utility are being
used already by the Office of Pesticide Programs to retroactively explore its own data
demonstrates early utility and applicability beyond tNCCT itself. The major gap noted for this
activity was described in the ToxCast project section above. In addition, finding an efficient and
effective methodology for extracting data from text sources was a concern for the Subcommittee.
A trial of natural language processing (NLP) for pulling information into some of the databases
was described. The Subcommittee notes that this method has been attempted rather
unsuccessfully by various research groups over probably 2 decades and thereby encourages the
exploration of other possible approaches as well.
Virtual Liver. Dr. Shah and his group are commended for having good command of the
significant breadth of biology, toxicology, and modeling that impact the project. In addition, the
"big picture" vision described is useful—there are many important questions in the field and not
limiting the vision too early is appropriate. The Subcommittee believes that this should be
balanced, however, with some very specific goals, milestones, and timelines for the next few
years that are clearly attainable with the resources at hand in order to assure some useful concrete
outcomes. In a project with this possible magnitude, it can be tempting to try to do everything,
both in terms of the various project approaches (knowledgebase (KB), biological modeling,
dosimetry modeling, etc.) as well as the scope within any one approach (breadth of the KB,
breadth and detail of every model, etc.), and thereby end up with little actually completed. One
suggestion is that Dr. Shah and the group develop a short prioritized list of specific scientific
research questions relevant to EPA's goals that they desire to address as soon as possible, and
use this to focus first iterations of development of both the KB and model(s). More explicit
milestones and goals for these highest priority questions then can be developed. Later iterations
of KB development and modeling can add scope (breadth/depth) to allow NCCT to address
Previous I TOC
-------
September 2008 BOSC Computational Toxicology Letter Report
additional research questions. The Virtual Liver activity will result in models of parts of the
biology being developed simultaneously and presumably by different individuals. Because the
idea is to integrate these models eventually to predict effects from molecular function to
physiologic outcome, the compatibility of the models is paramount. Dr. Shah indicated that he is
cognizant of and planning to manage this issue, for instance, by looking into the efforts of the
international Physiome Project. The Subcommittee members note that, to their knowledge, the
issue of common coding language, which has been addressed quite extensively by the Physiome
Project, does not appear to have addressed more subtle but critical compatibility issues
concerning biological and mathematical specifications among models, such as compatibility of
assumptions, equilibrium approximations, time scales, and so forth. Hence, beyond managing
compatible coding, the activity group is encouraged to actively plan for and manage on an
ongoing basis the specifications that must be shared among models so as to produce
compatibility when it is needed.
Developmental Systems Biology (Virtual Embryo). This project is very early in its
development, and already shows interesting progress based on the continuation of the earlier
work by Dr. Knudsen. Because the data needs of the proposed models may be significant, the
Subcommittee notes that it will be critical to identify and enlist appropriate supporters and
collaborators to provide such data. The track record of the principal investigator suggests that
this will develop naturally.
Arsenic BBDR. No specific technical gaps in the approach were noted for this activity. Because
the goal is to use the project's resulting model(s) for the 2012 review cycle of the Safe Drinking
Water Act, the Subcommittee encourages continuous communication with the appropriate
program office personnel so that concerns, objections, and skepticism can be addressed early and
on an ongoing basis. The group is commended for having such communication already in place
and it is encouraged to maintain that communication to the greatest degree possible.
Charge Question 4: Does the work offer to significantly improve environmental health impacts
and is the path toward regulatory acceptance and utilization apparent?
This question was included so that the Subcommittee could provide an opinion as to the potential
for NCCT's research to impact decision-making by the programs and offices that administer
regulations that are important for public and environmental health.
The Subcommittee believes that the work being reviewed has the potential to significantly
improve a number of aspects of the risk assessment process, and in so doing will lead to
substantial improvements in environmental health. As noted in the responses to previous
questions, the programs under review are at different levels of maturity and will deliver results at
different time points. The potential to improve the public and environmental health protection
role of the Agency, however, is enormous. These improvements will come in the form of better
tools for the prioritization of chemicals to evaluate and assess, early insight into the potential
toxicity of new substances by improved capabilities of searching for structural analogs for which
data already exist, better understanding of the fundamental molecular processes that underlie
toxicity and variability in response, and better methods for incorporating that information into
risk assessment. As with the other responses, each project that was reviewed will be discussed in
Previous I TOC
-------
September 2008 BOSC Computational Toxicology Letter Report
separate paragraphs below (with the exception of the Arsenic BBDR, which will be addressed
under Charge Question 4a).
ToxCast. This project already has begun to generate a considerable database on the cellular and
biochemical effects of approximately 300 well-studied chemicals, mostly pesticides. The
advantage of choosing this set of chemicals, nominated by the Office of Pesticide Programs, is
that they already have been assessed for their potential to cause toxicity using a comprehensive
set of toxicity tests. This will provide the phenotypic anchoring for responses that are observed in
the high-throughput and other test methodologies that ToxCast is employing. ToxCast has the
goal of providing a scientific foundation for predicting the potential hazards of chemicals by
evaluating the responses of relevant molecular and cellular markers in simpler experimental
systems. This will lead to an improved ability to prioritize testing, better test methods, testing
strategies that are tailored to the chemical being tested, and perhaps ultimately to the replacement
of existing test methods with ones that are not encumbered with much of the uncertainty inherent
in traditional toxicity tests. In order to reach this potential ToxCast will need to generate a lot of
new data. The recently announced collaboration between NCCT, NIEHS, and NHGRI will
accelerate progress in this area and is a wise use of limited resources.
IT/IM Activities. The database and software development has been outstanding. ToxRef
already is in use at the Office of Pesticide Programs and is allowing toxicologists and risk
assessors to query large databases of chemical structures for common toxicological properties.
Relational databases of this type provide novel opportunities for risk assessors to consider the
potential biological activity of new chemicals instead of just production volume (or other
surrogates of potential exposure) in prioritizing them for further evaluation and testing. This
already is a major achievement with practical applications.
Virtual Liver and Virtual Embryo. These programs have longer time horizons but have
significant potential to improve risk assessment. The liver is a common target organ for toxic
agents, and is a primary site of metabolism of xenobiotic compounds. Adverse effects on
embryonic development usually are irreversible, and the economic and emotional consequences
of adverse developmental outcome are significant. Therefore, the choice of these two systems for
intense investigative and modeling approaches is appropriate for an agency interested in the
public health consequences of toxicant exposure. As noted in previous responses, these programs
will need to progress a little farther before enough of a scientific foundation is created to
accurately determine how they will be incorporated into the risk assessment process. It already is
clear that the information being generated will be important in reducing the uncertainty
associated with determining which chemicals pose hazards, variability in susceptibility in a
heterogeneous population, and other critical questions.
-Y- Charge Question 4a: In addition, specifically for the Arsenic BBDR project:
Does the proposed computational model have the potential to identify and reduce
uncertainties with the risk assessment process?
The answer to this question is yes, depending on data gaps identified and resources made
available. This study might not give all the answers but will get us halfway there. EPA
recognizes that developing a universal arsenic model describing several cancer endpoints is a
7
Previous I TOC
-------
September 2008 BOSC Computational Toxicology Letter Report
formidable challenge. Hence a step-wise research project with an eye for the future is proposed.
Initially, a generic model for cancer will be developed that will incorporate key steps of the
mode of action commonly shared for multiple cancer types such as oxidative stress. This model,
in turn, will serve as an engine to develop specific cancer models as the need arises and resources
become available. To ascertain whether appropriate steps are being incorporated, a thorough
literature review of experimental and epidemiological data and expert consultation has been
proposed. It also is acknowledged that even though there is a lot of data, they are somewhat
weak to generate exposure time course response curves. Appropriate experiments have been
proposed to fill the research needs to develop a realistic model.
-Y- Charge Question 4b: Will the model be able to help identify susceptible populations and
compare potential risks in those populations with less susceptible populations?
Yes, the initial generic model development exercise will allow identification of issues such as
mechanisms that operate in general versus subpopulations, such as susceptible populations with
varying degree of arsenic methylation. Such issues could be the subject of workshops to explore
the issue of the extent of polymorphism in the human population.
The short-term (1-2 years) goal is the establishment of a coordinated program of laboratory
research to generate essential data needed to develop a BBDR model that will increase
confidence in the predictions. To start with, the model development will be initiated with
available data. Work proposed includes multistage clonal growth modeling, target tissue
dosimetry, and methylated metabolites of arsenic.
The long-term (3-5 years) goal of developing a robust version might be too optimistic. As the
project gets underway, new questions and issues might be identified that will require additional
laboratory research and continued resources. The project has a good future as it can be easily
adapted to the latest (2007) National Academies toxicity testing report that recommends a
systems biology and computational tool integration.
-Y- Charge Question 4c: Is coordination between model development and associated data
collection sufficient to avoid problems with models being either over- or under-determined?
Yes, it is desirable to see what health effects are caused at lower doses to avoid the potential of
compromise in setting an arsenic standard based on cost-benefit analysis.
Charge Question 5: Have appropriate data management and analysis tools been incorporated
into the project?
Previous reviews have highlighted the importance, as well as the challenges, of developing
useful relational databases. This question was included so that the Subcommittee could evaluate
the Center's progress in developing and implementing strategies for data management and
analysis.
ToxCast. With regard to ToxCast, NCCT has made great progress in the past 18 months in
hiring bioinformatics and computational biology scientists and staff members to establish the
Previous I TOC
-------
September 2008 BOSC Computational Toxicology Letter Report
infrastructure necessary to begin meeting the needs of the program. The challenges here also are
the strengths of the ToxCast: the diversity of the data that it will generate and the need to
effectively organize that information to facilitate its analysis and interpretation. The approach
taken by Dr. Richard Judson and his group is a sensible one given the state of the field:
information from each technology with which data will be generated will be captured in a
technology-specific database, and this information ultimately will be collected in a central data
warehouse linking the information together. The advantage of this solution is that it allows the
data from each assay to be stored in a rational format while deferring the question of how the
information will be combined to address questions relevant to the mission of EPA.
The construction of the warehouse remains an open question. Ultimately, a database is a model
of the interactions that exist in the underlying data and the relationships relevant to the analysis
that will be performed. The diversity of the data, representing a wide range of in vivo and in vitro
assays from multiple species, makes building such a model a significant challenge. The project
seems to be lacking a set of analytical objectives necessary for building the relevant use cases
that ultimately will inform the process of database construction, and this ultimately will
determine its utility. At this stage, ToxCast needs to begin to define analytical outcomes in order
to set goals and milestones with regard to developing and validating analytical protocols. This is
an essential step at this point as it will help to anchor future development and make it relevant.
This also will help to define the requirements of the interfaces that are built to access the data.
Further, the ToxCast group should be encouraged to release the data and databases at the earliest
possible time and to consider a "CAMDA-like" workshop in which the research community is
offered access to the data with the challenge of using the data to effectively predict end points.
At least three advantages to the program will be derived from these efforts. First, public release
will help to drive the creation of relevant use cases that will further database development.
Second, it will assist in evaluating data access protocols and tools to assure the greatest utility to
the research and regulatory community. Third, it will accelerate the development of predictive
algorithms to combine the data to make predictions about relevant phenotypic outcomes.
Virtual Liver. The Virtual Liver is a very ambitious project designed to simulate molecular,
cellular, physiological, and organ-level computational models that ultimately can be used to
make predictions regarding the toxic effects of various compounds. To limit the scope of the
project to something that might be manageable, its initial focus will be nuclear receptor-mediated
non-genotoxic liver cancer. The group should be applauded for this decision as it will give staff
the opportunity to focus enough to make progress.
The starting point and first challenge will be the construction of a liver KB. In any domain, this
is a nontrivial problem and ultimately will require linking information in the literature and a host
of public data resources. The use of publicly available resources and tools and the commitment
to making the KB available are commendable not only because it will be widely useful to the
broader community, but also because it will accelerate the development and curation of the
information within the KB.
With regards to populating the KB, the use of NLP probably is not the best solution. NLP does
not work well with the scientific literature, and its application in this domain remains an area of
Previous I TOC
-------
September 2008 BOSC Computational Toxicology Letter Report
active research. Application of NLP has the potential to introduce a great deal of noise in the
system, leading to many potential false associations that could lead to more problems than it
solves. Consequently, other methods, including expert or community curation, should be
explored.
On a larger scale, the greatest potential problem will be linking each of the domain-specific
models to build a predictive system. Again, this remains an area of active research and one that
may present significant barriers to developing verifiable solutions. The greatest challenges will
be to validate any models that emerge from the analysis.
Finally, there is a need to develop standards for interactivity and try to interface with developing
standards within the community.
Virtual Embryo. This project is in its early stages, with Dr. Knudsen only having arrived 3
months prior to this review. As such, it is still not well integrated with the overall NCCT
program, and in particular ToxCast. It remains to be seen how well it will eventually integrate
with the overall program, and its integration with other internal and external initiatives needs to
be resolved. Nevertheless, it appears that this project could provide an opportunity to explore the
results emerging from ToxCast, and it may help direct selection of the next generation of
compounds for analysis in ToxCast.
Charge Question 6: How would you assess the outreach to other groups in executing the
projects?
Because of NCCT's limited size, it is vitally important that the Center be connected in
meaningful ways to other groups of experts who can augment the Center's capabilities. The
purpose of this question was to determine the Center's progress in making and leveraging
connections.
NCCT has done an admirable job in reaching out to other groups, both inside and outside the
Agency. Because of the relatively small size of the Center, outreach is important as a way of
augmenting its productivity. Outreach also is important in engaging others in understanding the
capabilities of computational toxicology, which will be crucial in convincing program offices to
use the tools developed by the Center. NCCT is doing a good job on both counts.
NCCT has been successful at developing partnerships at several levels. Within ORD, NCCT has
developed successful partnerships with the National Health and Environmental Effects Research
Laboratory (NHEERL) and the National Exposure Research Laboratory (NERL), which can
conduct experiments and supply data for analysis and modeling by NCCT scientists. The Center
is tied into a number of the research activities in these laboratories, including the Endocrine
Disrupting Chemicals, Drinking Water, Safe Pesticides/Safe Products, and Human Health
Research Programs. NCCT has allocated a fraction of its resources toward the achievement of
goals within those ORD programs.
NCCT also has developed three Communities of Practice (CoP) in chemi-informatics, biological
modeling, and categorization and prioritization. The purpose of the CoPs is to unite scientists
10
Previous I TOC
-------
September 2008 BOSC Computational Toxicology Letter Report
who have a common interest in an area in which NCCT is a center of excellence. The CoPs are
becoming a means of coordinating activity and communicating progress on an informal, grass-
roots level. Outreach also has taken place to program offices within EPA, especially the Office
of Pesticide Programs. This office is supplying data that are being used as part of the ToxCast
project, and is likely to be an early adopter of the predictive and priority-setting tools being
developed by the Center.
NCCT is doing a good job of joining forces with others outside the Agency, particularly at
NIEHS. The arm of the National Toxicology Program that operates at NIEHS has a strong
interest in high-throughput methods for predicting toxicity, a project that is complementary to
activity at the Center. NCCT and NIEHS have done a good job of information sharing and have
developed a constructive working partnership in which data and analysis methods will be shared.
NCCT also is establishing collaborations internationally, coordinated through the Organization
for Economic Co-operation and Development (OECD). OECD's project entitled "Molecular
Screening for Characterizing Individual Chemicals and Chemical Categories" has similar goals
as the ToxCast project. OECD has recognized that ToxCast can serve as a foundation for its
project and is developing an international consortium that will build on ToxCast. It is likely that
a number of nations and private companies will join this consortium in the coming year.
Furthermore, the recent MOU among NCCT, NIEHS, and NHGRI promises to be the most
important and extensive collaboration yet for ToxCast and NCCT. In summary, NCCT is doing
an excellent job at outreach, which in turn is enhancing its ability to fulfill its mission.
In conclusion, the BOSC Computational Toxicology Subcommittee believes that NCCT is
making exceptional progress toward its mission. We are pleased to provide advice on this
important Center and look forward to future opportunities to provide timely advice to guide and
improve NCCT and its programs.
Sincerely,
Gary S. Sayler, Ph.D.
Chair, BOSC
11
Previous I TOC
-------
UNITED STATES ENVIRONMENTAL PROTECTION AGENCY
WASHINGTON, D.C. 20460
OFFICE OF
RESEARCH AND DEVELOPMENT
Dr. Gary S. Sayler
Chair, Board of Scientific Counselors
The University of Tennessee
676 Dabney Hall
Knoxville. Tennessee 37996
Dear Dr. Sayler:
The Office of Research and Development (ORD) would like to take this opportunity to
thank you and the members of the Board of Scientific Counselors (BOSC) for the December,
2007, progress review of the National Center for Computational Toxicology (NCCT). We
greatly appreciate the efforts of the members of the subcommittee who conducted the review,
Drs. George Daston, James Clark, Richard DiGiulio, Muiz Mumtaz, John Quackenbush, and
Cynthia Stokes.
The subcommittee was requested to provide advice on the progress made by the NCCT,
and specifically on the ToxCast7 project, IT/IM activities, the virtual tissues projects, and
biologically based does-response modeling for arsenic.
Enclosed is our response to the comments and recommendations in your letter of
September 16, 2008. Please feel free to contact me if further information is needed.
We are pleased that the BOSC was very supportive of NCCT and the direction we are
taking in this very important research program. The guidance of the BOSC has been of great
assistance to NCCT in crafting the second generation Implementation Plan. This plan, which
covers FY09-12, will be presented at the fourth BOSC review later this year.
Again, thank you for your efforts on behalf of ORD.
Sincere
^evnp'eichfnan, Ph.D.
Deputy Assistant Administrator for Science
Enclosure
Cc: Dr. George Daston
Dr. James Clark
Dr. Richard DiGiulio
Dr. Muiz Mumtaz
Dr. John Quackenbush
Dr. Cynthia Stokes
Recycled/Recyclable • Printed with Vegetable Oil Based Inks on 100% Recycled Paper (40% Postconsumer)
-------
Office of Research and Development's Response to the
Board of Scientific Counselors Report on
ORD's National Center for Computational Toxicology
(final report received September 2008)
February 2009
BOSC Computational Toxicology Subcommittee
Dr. George Daston (Chair)
Dr. James Clark
Dr. Richard DiGiulio
Dr. Muiz Mumtaz
Dr. John Quackenbush
Dr. Cynthia Stokes
Submitted by:
Dr. Robert Kavlock
Director, National Center for Computational Toxicology
Office of Research and Development
Previous
TOC
-------
ORD Response to BOSC Computational Toxicology Letter Report
February 2009
The following is a narrative response to the comments and recommendations of the
BOSC review of ORD's National Center for Computational Toxicology (NCCT), held
December 17 and 18, 2007, in Research Triangle Park, NC. The review was conducted
by a standing subcommittee of the BOSC. The subcommittee had previously reviewed
the NCCT in April 2005 and June 2006 and ORD responded to those reviews. In this
third review, the BOSC noted, ".. .during the 2.5 years between its establishment and this
review, NCCT has made substantial progress in establishing priorities and goals; making
connections within and outside EPA to leverage the staff's considerable modeling
expertise; expanding its capabilities in informatics; and making significant contributions
to research and decision-making throughout the Agency." Furthermore they noted,
".. .many of the recommendations made by BOSC during its earlier reviews have been
acted on by NCCT. This includes improved capabilities in bioinformatics through the
funding of two external centers and in informatics and systems biology through staff
hires; expansion of its technical approaches to even more programs within the Agency;
and the formation of an extensive collaboration with the National Institute of
Environmental Health Sciences (NIEHS) and the National Human Genome Research
Institute (NHGRI) for its ToxCast™ project."
Each charge question is shown below in bold, followed by the BOSC's comments in
italics and ORD's response to the comments in regular type. A summary of the BOSC
recommendations and ORD's responses is provided in Table 1 at the end of this report.
Charge Question 1: Does the scope and involvement of expertise in the project
reflect activities consistent with the function of a Center?
The NCCT was founded only a few years ago and has been achieving a critical mass of
expertise through selective hiring, external grants, and the formation of connections with
other groups of experts within EPA. The purpose of this question was to gauge the
progress of the Center in achieving the level of expertise needed to pursue its mission.
The staff working in NCCT and those scientists involved from outside the Agency who are
working as collaborators are highly qualified in various aspects of computational
toxicology. The Center's effort to solidify formal agreements in terms of memoranda of
understanding (MOUs), cooperative research and development agreements (CRADAs),
etc., with various organizations has opened up a diversity of quality opportunities to
leverage and enhance Office of Research and Development (ORD) efforts. A timely
example is the February 14, 2008, announcement of the collaboration between NIEHS,
NHGRI, and EPA 's NCCT. As described in the press release, this collaboration
leverages the strengths of each group to use high-speed, automated screening robots to
test suspected toxicants using cells and isolated molecular targets.
The staff and collaborators at the center have the appropriate expertise and insights.
The utility of the tools and deliver able s can be enhanced if the staff moves toward being
more explicit on how the tools under development support EPA risk assessments. Some
of the ORD researchers seem to be searching for an application for their sophisticated
tools, and discussions with Agency staff practicing risk assessments (Office of Pollution
Prevention and Toxics fOPPTJ; Office of Water, Office of Wastewater Management;
Previous I TOC
-------
ORD Response to BOSC Computational Toxicology Letter Report
February 2009
Office of Prevention, Pesticides, and Toxic Substances [OPPTS], etc.) could provide
direction as to the appropriate milestones and deliverables for these efforts. The BOSC
reviews and the Center would benefit if representatives from these Agency offices
attended BOSC reviews to ensure that all par ties understand how NCCT's efforts address
the most relevant needs of the Agency. The BOSC wants to ensure that this advice is seen
as encouragement to reach out to risk assessment practitioners. The ongoing work in
developing the analytical approaches and information databases is of high technical
quality, as the Center staff and collaborators are working on many new and exciting
approaches. By holding research planning discussions with risk assessment
practitioners, the applications of the computational toxicology tools and resources can be
directed to ensure the most relevant and efficient use of data and models.
(Recommendations #1 and #2 in Table 1)
ORD Response: ORD appreciates this recommendation. As noted in the report, NCCT
regularly meets with program offices, risk assessors, and other potential practitioners in
planning and conducting this research. A priority action item of the NCCT for FY2009 is
to improve connectivity with NHEERL, NERL and NCEA relative to building the
foundation for a transformation in the conduct of evaluating the toxicity of chemicals.
We are continuing to engage Communities of Practice to help achieve this end. In
previous reviews, some of these stakeholders were invited and attended meetings of this
BOSC subcommittee. The next review will be a broad review of the computational
toxicology program and the new implementation plan. For this and future meetings,
Agency stakeholders will be invited to attend the meeting and enter discussions as
appropriate. Further, the NCCT will ask such stakeholders to review and comment on the
new implementation plan prior to the next BOSC meeting.
Charge question 1 continued:
One challenge for the center staff involved in developing informatics datasets will be to
develop efficient and effective ways to handle the wealth of data available in some areas
to avoid redundancies of data entries and to focus on the most informative data. Again,
interactions with various program offices and their risk assessment activities should
provide a basis to set the long-term goals for the Informatics/Data management team.
This will allow the development of structured short-term and mid-term activities needed
to meet the long-term goals. (Recommendation #3 in Table 1)
ORD Response: To address this important issue the NCCT has five main database-
related, data-intensive projects: ACToR, ToxRefDB, ToxMiner, the ToxCast™ chemical
registry, and DSSTox. ACToR (http://actor.epa.gov/actor) is the global repository of data
that is relevant to environmental chemicals. It is populated from more than 200 public
repositories of toxicity data to provide a broad, but in many cases shallow view of the
universe of data available on chemicals of interest to the NCCT and the EPA. ToxRefDB
is focused on extracting high quality in vivo toxicology data on chemicals in the
ToxCast™ program, capturing study information down to the treatment group level, and
extracting these into a relational database well-suited to predictive modeling. ToxRefDB
is also being developed into a web-accessible resource that can be queried to derive
Previous I TOC
-------
ORD Response to BOSC Computational Toxicology Letter Report
February 2009
treatment related toxicity effects directly from the database. ToxMiner is a compilation
of statistical tools capable of analyzing relationships between ToxCast™ and ToxRefDB
data, and performing predictive signatures. The ToxCast™ chemical registry is used to
track nominations for ToxCast™ screening, to track chemical procurement, sample
identity and sample QC, and finally to link actual samples to ToxCast™ data. DSSTox is
adding the quality reviewed chemical structure layer to data sets of interest to NCCT, and
publishing additional inventories and toxicity data sets of interest to EPA and external
groups. The underlying data model and database tables for all but DSSTox are being
consolidated to remove data redundancy and to reduce the effort required to manage
multiple related systems. DSSTox is primarily a file-based system, and as data are
curated, they are entered into the ACToR system for further use. We are actively
working with other partners (ORD, OPP, OPPT, OW, OHS, NCEA, the Tox21 partners)
to prioritize chemicals to be entered into the system and to obtain and enter data. We
believe this compilation of information on the toxicity of chemicals provides a solid
foundation for the NCCT to not only understand the extent of public information on
chemicals, but also to provide public access to this increasingly data rich repository of
information on chemicals.
Charge question 1 continued:
The BOSC noted that it remains somewhat unclear how the Center intends to use
ToxCast and associated analyses to approach risk assessment. For instance, species-to-
species translation was mentioned, and the data are being obtained from multiple
species, not just humans, but how the different species data will be reconciled was not
discussed. Although the primary goal of the ToxCast project is prioritization of
chemicals for detailed risk assessment, not the risk assessment itself, it is interesting to
contemplate how the projected database and analysis might be directly relevant.
Similarly, it was noted that an early decision regarding ToxCast was that ecology and
paths of exposure were not going to be addressed in this project (at least not initially).
Nonetheless, at several points, paths of exposure arose during the review because of their
obvious relevance. The Subcommittee is prompted to ask how it might be addressed in
future work. (Recommendation #4 in Table 1)
ORD Response: The NCCT has recognized the opportunity to address the full source-to-
outcome continuum of risk assessment, and has recently done this in several ways. This
need is reflected in the FY2009 priorities for NCCT that include increased connectivity
with other components of ORD. Thus, NCCT has organized an ORD-wide workgroup to
expand an overarching strategy for developing a high throughput approach to risk
assessment-building from the example and lessons from ToxCast™ and expanding on
applications to exposure and mode of action assessment. One part of this approach will
be to develop exposure predictions on the thousands of chemicals relevant to ToxCast™,
in a Center project tentatively called ExpoCast. Finally, the translation of ToxCast™
predictions directly to humans is being accomplished by direct comparison of results for
rodent and human targets and pathways interrogated by complementary assays. In
addition, a proposal has been accepted for consideration by the HESI Emerging Issues
Program at its annual meeting in January 2009 to establish collaborations with the
Previous I TOC
-------
ORD Response to BOSC Computational Toxicology Letter Report
February 2009
pharmaceutical industry to supply chemicals with identified human toxicity for use in
Phase lib of ToxCast™. This phase would include at least 100 pharmaceutical
Tl\/f
compounds with known human toxicities and would extend ToxCast predictive
signatures from Phase I of rodent toxicity endpoints, to similar toxicity endpoints in
humans.
Charge question 1 continued:
The Subcommittee also noted that the means of using the eventual Virtual Liver models
for actual risk assessment at EPA is unclear. The BOSC encourages additional thought
and efforts along these lines, in collaboration with the appropriate EPA program office
personnel. This is not a criticism of the current project vision by any means, but because
direct or indirect application to risk assessment would be a fantastic result, it seems
prudent to consider the possibility earlier rather than later. (Recommendation #5 in
Table 1)
ORD Response: The Virtual Liver (v-Liver) is being developed in conjunction with
NHEERL research activities. A detailed plan for v-Liver will be presented to the BOSC
at the 2009 review. The objective of the v-Liver project is to coordinate an integrated in
vitro and in virtuo program in the long-term for toxicity testing that is efficient, relevant
to humans and less dependent on animals, with the ultimate goal of use in risk
assessment. We agree that stakeholder involvement from EPA program offices is a
critical requirement for the success of the v-Liver project. Although program office
personnel were not directly involved in the early v-Liver research planning phase, senior
scientists from NCEA/RTP, NHEERL and NCCT who have a good grasp of risk
assessment needs for fulfilling EPA's mission, are part of the core team. Their collective
insight into key challenges facing risk assessment and the requirement for future toxicity
testing have been vital for shaping the vision for the v-Liver system. Therefore we
believe the v-Liver project is poised to actively engage program office personnel to
address challenges in mode of action (MOA) elucidation and quantitative dose-response
prediction for chronic liver injury.
Program office personnel will be engaged in the design, development, and utilization of
the system. This is being accomplished through a few practical use cases that demonstrate
the value of Virtual Tissues for developing a proof of concept (PoC) for assessing the risk
of environmental chemicals to liver physiology and human health. Over the next two
years, the v-Liver PoC will define a subset of hepatic effects, apical endpoints, and
relevant environmental chemicals which will be developed in close collaboration with
program office personnel to ensure application to risk assessment and relevance to the
EPA mission. In addition to developing a Virtual Tissues platform that will contribute in
the long-term to the future of toxicity testing, the short-term milestones of the v-Liver
PoC will also aim to address current client needs.
The v-Liver project plan (please see Appendix for outputs) outlines how stakeholders will
be involved. Currently, the project is aligned closely with the ToxCast™, ToxRefDB and
ToxMiner projects to develop methods to select environmental chemicals for the v-Liver
Previous I TOC
-------
ORD Response to BOSC Computational Toxicology Letter Report
February 2009
PoC focusing on nuclear receptor (NR) mediated hepatocarcinogenesis. Analyzing data
from ToxCast™ and ToxRefDB has identified a range of pesticides and persistent toxic
chemicals that match these criteria. Around ten chemicals will be used for the PoC and
these will be selected in collaboration with program office personnel who are actively
involved in their risk assessment and/or have substantial expertise in their MOA. We
plan to develop these collaborations with stakeholders by providing them two main types
of computational tools. In the short-term (FY09), interactive tools to aid hepatic MOA
organization and analysis will be developed. In the medium (FY10) to long-term, these
will be extended with prototype tissue-level simulation tools that will aid in investigating
the quantitative relationships between MOA(s) and adverse effects.
The first deliverable for risk assessors is the v-Liver Knowledgebase (v-Liver-KB),
which formally organizes information on normal hepatic functions and their perturbation
by chemical stressors into pathophysiologic states. Information about hepatic physiology
relevant for MOA analysis is dispersed across scores of public domain repositories as
well as the biomedical literature and the v-Liver-KB will leverage semantic approaches,
which are being increasingly adopted by the biomedical community, to provide effective
tools that fill the gaps toxicity MOA organization and inference. The v-Liver-KB will be
deployed as an interactive web-based and desktop tool to intuitively browse and query
physiologic knowledge on PoC chemicals, to derive MOA(s) and to link assay results
from ToxCast™, species-specific effects from ToxRefDB, and other evidence curated
from the literature. We believe this system will provide computable information on key
events that transparently indicate the uncertainties and data gaps and that make inferences
on MOA from experimental data. In addition, we will work closely with risk assessors to
customize the system for specific requirements. The v-Liver-KB will be deployed over
the next two years and updated quarterly with any new information on the PoC
chemicals.
The second deliverable (FY10), the v-Liver Simulator (v-Liver-Sim), dynamically
simulates the key molecular and cellular perturbations leading to adverse effects in
hepatic tissues. Initially, it will focus on modeling MOA leading to proliferative and
neoplastic liver lesions at a hepatic lobular scale. The v-Liver-Sim is being developed as
a cellular systems model of the hepatic lobule that will use MOA information from the v-
Liver-KB to initially provide two outputs: the visualization of tissue changes at a
histological scale and the assessment of lesion incidence. A version of this system will
also be provided as a web-based/desktop tool to enable risk assessors to perform
interactive and quantitative simulation of chemical induced perturbations of physiologic
processes leading to toxic histopathologic effects. Eventually, the liver simulator will be
integrated with PBPK models to model alternative exposure scenarios. Over the course
of the project, the system will be evaluated in collaboration with risk assessors using PoC
chemicals in vitro data from ToxCast™ and published in vivo data from rodents and
humans.
Previous I TOC
-------
ORD Response to BOSC Computational Toxicology Letter Report
February 2009
Charge Question 2: Are the goals and milestones suitably described, ambitious, and
innovative?
For the Center overall, the answer to this question is "yes. " In particular, the goals of
the Center are well-described, very ambitious and innovative, as well as important for
the future of research at EPA. The issue of "milestones " is somewhat more complex, in
part due to the varying levels of maturity for Center components. In most cases, previous
accomplishments and current activities are well described, but more detail concerning
projected future milestones would be helpful. It is recognized, however, that these
projects are very innovative and substantial flexibility is appropriate. This is particularly
true for less mature but highly creative projects such as the Virtual Liver and Virtual
Embryo. Also, in considering goals and mile stones, it may be appropriate to consider the
timely integration of each project's accomplishments into the Agency's risk assessment
activities. In the following paragraphs, Charge Question 2 is addressed in the context of
the five major Center activities discussed at the review meeting.
ToxCast: Future plans for the project also are well described, although a more detailed
time table for milestones past 2008 would be helpful. (Recommendation #6 in Table 1)
ORD Response: ORD agrees with this recommendation and has a more detailed
timetable for ToxCast™ milestones which centers around the release of data, validation
of predictive signatures, and generation of data as chemicals are tested in Phase II. With
considerable data and experience now in hand from Phase I contractors and additional
collaborations on the Phase I chemical library with laboratories within NHEERL and
outside EPA, it will be possible to better articulate the directions for Phase II of
ToxCast™. In addition, activities of the Tox21 consortium between NTP/NIEHS,
NCGC/NHGRI and NCCT/ORD are maturing and beginning to identify near-term and
medium-term goals. These activities will be described in greater detail in the second
generation Implementation Plan, which we are now developing and will present to the
BOSC at the next NCCT review. Please see appendix for detailed listing of milestones.
Charge question 2 continued:
IT/IM Activities: The project is highly and suitably ambitious, and its goals and
substantial progress are well described. Again, future plans are described well in a
general way, but more detail concerning future milestones (beyond 2008, which is well
described) would be appropriate. (Recommendation #6 in Table 1)
ORD Response: Again, ORD agrees and has a detailed time table which emphasizes the
deployment and continual upgrade of ACToR, integration of ToxCast™ and ToxRefDB
in-vivo toxicology data, importation of available exposure, neurotoxicity, and
reproductive toxicity data. A detailed listing of ACToR and ToxRefDB related
milestones can be found in Appendix I.
Regarding ToxMiner, the first goal in FY09 is to incorporate all of the ToxCast™ Phase I
data into ToxMiner. This involves processing the many individual data sets to eliminate
Previous I TOC
-------
ORD Response to BOSC Computational Toxicology Letter Report
February 2009
faulty data, to perform scaling and normalization, and to extract computationally useful
parameters such as maximum effect levels and IC50 values. The second main task is to
integrate the ToxMiner database with analysis tools for statistical analysis and machine
learning. A third task is to integrate other biological information to help interpret the
results of statistical analyses. In particular, we are incorporating pathway information
and using this as an organizing principle to make sense of the results from the hundreds
'PA/I Tl\/f
of individual ToxCast assays. The major goal of ToxCast Phase I is to develop a
series of "signatures" linking in vitro data with in vivo toxicology. The related ToxMiner
goal for FY09-FY10 is to produce and store these signatures and have them ready for
Tl\/f TA/f
validation on ToxCast Phase II chemicals. Planning is well underway for a ToxCast
Data Summit in May 2009, which will provide a forum for external scientists to come
and discuss alternatives for deriving predictive signatures of ToxCast™ HTS date
relative to ToxRef identified phenotypes.
DSSTox will increase its interactions and alignment with major NCCT projects
(ToxCast™, ToxRefDB, ACToR) and broader Agency and outside projects (NHEERL,
OPPT, NTP, CEBS, EU REACH), providing key cheminformatics support, expanding
DSSTox data file publications of toxicological data in support of predictive modeling,
and enhancing linkages to resources such as PubChem for disseminating EPA,
ToxRefDB and ToxCast™ bioassay results to the broader modeling community.
Detailed milestones are found in Appendix I.
Charge question 2 continued:
Virtual Liver: Although narrower in scope than the foregoing projects, the Virtual Liver
project is very ambitious; it also is relatively young, apparently becoming fully
operational with the arrival of Dr. Imran Shah in September 2006. Its Jit with the goals
of NCCT is perhaps less clear than the previous two projects; it is more "visionary" in
nature, and less directly applicable to risk assessment, as described by one of the EPA
scientists involved. The goals of the project and the nature of research to be performed
to achieve those goals are clearly described. There is some concern that this project may
be overly ambitious. It may be helpful if key objectives were delineated and prioritized,
perhaps indicating achievements that are critical to the success of the project and those
that are highly desirable. Milestones for tracking the project's progress are not
apparent, particularly in later years (3-5). This relatively young and very innovative
project requires considerable flexibility, however, so the lack of detailed milestones in
later years is very reasonable. (Recommendation #7 in Table 1)
ORD Response: The importance of developing and applying computational system level
models of key phenotypic outcomes is reflected in the second goal of the new EPA
Strategic Plan for Evaluating the Toxicity of Chemicals that is currently working its way
through final concurrence by the Agency. NCCT recognizes the need to better delineate
the goals and milestones of the v-Liver project, and we have made this a key activity in
response to the comments of the BOSC. NCCT is convinced the future of toxicology will
be heavily dependent upon the development of computational systems level models and
has played a key role in the development of this plan and its execution through this
Previous I TOC
-------
ORD Response to BOSC Computational Toxicology Letter Report
February 2009
project. Current and additional details will be provided at the next review of the BOSC.
The short-term goals for the v-Liver project are to identify environmental chemicals for
the PoC system. Once there is buy-in from EPA stakeholders (program offices and
NCEA) on these chemicals, the team will begin populating the v-Liver-KB with relevant
mechanistic and MOA information on these chemicals including in vitro data from
ToxCast™ and in vivo data from the literature. Concurrently, the team will develop a
prototype virtual hepatic lobule to understand the key cellular responses necessary for
modeling cancer progression beginning with nuclear receptor activation. Data generated
by ToxCast™ as well as external collaborators/new contracts will be used to begin
quantitative parameterization of the cellular and molecular responses, and their
evaluation using published in vivo rodent data. The detailed milestones for the project
are described in Appendix I.
Charge question 2 continued:
Developmental Systems Biology (Virtual Embryo). This project is at a substantially
earlier stage than the Virtual Liver project; it is led by Dr. Thomas Knudsen who joined
NCCTin September 2007. The issues of goals and milestones are essentially the same as
for the Virtual Liver, that is, strong on the former, but understandably weaker on the
latter. It is the Subcommittee 's expectation that a more concrete research plan with
goals and milestones will be developed over the coming months. (Recommendation #8 in
Table 1)
ORD Response: A formal research plan for the Virtual Embryo, including goals and
milestones, has been developed. The long-term goal will provide a computational
framework that enables predictive modeling of prenatal developmental toxicity. The
project is motivated by scientific and regulatory needs to understand how chemicals
affect biological pathways in developing tissues, and through this knowledge a more
ambitious undertaking to predict developmental toxicity. The research plan is built on an
expanded outlook of experimental-based techniques that aim to identify 'developmental
toxicity pathways' and an expanded scope of computational search-based techniques that
apply such knowledge into models for chemical dysmorphogenesis. Dr. Knudsen, the
lead scientist for this program, was recently invited to NCEA where he provided an
overview of the project. This has led to close coordination between the computational
models and the risk assessment priorities.
Virtual Embryo's short-term goals address the knowledgebase (VT-KB) and simulation
engine (VT-SE) to enable in silico reconstruction of key developmental landmarks that
are sensitive to environmental chemicals. Initial research focuses on early eye
development. Proof-of-principle (2yrs) will be measured by high fidelity simulation
models to demonstrate several generalized principles, including the ability to reconstruct
genetic defects in silico, classify abnormal developmental trajectories from genetic
network inference, and predict teratogen-induced defects from pathway-level data. A
much more detailed research plan will be provide to the BOSC in its 2009 review of the
NCCT, and detailed examples of current envisioned milestones are found in Appendix I.
Previous I TOC
-------
ORD Response to BOSC Computational Toxicology Letter Report
February 2009
Charge question 2 continued:
Arsenic BBDR: This project is unusual among NCCTprojects in that it is oriented
toward a specific chemical with a specific issue (Safe Drinking Water Act revisions)
rather than an approach developed with diverse chemicals in mind. However, this
project is likely to inform the eventual development of other biologically-based dose-
response models and their application to risk assessments by the Agency. Thus, in
addition to informing the controversial issue of arsenic risk assessment, the project is
more broadly relevant to the mission of the NCCT. The goals of the project are very
clear and well described. Milestones, however, are not stated, and may be particularly
important for this project, which has a clear deadline (2011) in order to be useful for the
2012 Safe Drinking Water Act review cycle. (Recommendation #9 in Table 1)
ORD Response: At the time of the BOSC review in December, 2007, considerable effort
had been devoted to planning the development of a BBDR model for carcinogenic effects
of inorganic arsenic (iAs). The initial focus of the planning process was a literature
review to identify data needs. This review had shown that the pharmacokinetics (PK) of
iAs were relatively well-studied, though there were some significant remaining PK
uncertainties. The literature was not, however, sufficient to identify with any confidence
the relevant mode or modes of action (MoA) of iAs responsible for its carcinogenic
effects. We therefore developed a generic experimental design that focused on: (1) the
description of a potential MoA as a sequence of key events; and (2) experimental
characterization of the dose-time response surfaces for the key events. For any given
candidate MoA, it was anticipated that this experimental approach would have provided
sufficient data to allow ranking of candidate MoAs by dose and time course. The MoA
or MoAs acting at the lowest doses and earliest time points would be considered to be the
drivers for the apical cancer outcomes.
The next step in the process was to elicit research proposals from NHEERL iAs
researchers that were to be based on the suggested experimental approach for
characterizing candidate MoAs. The literature is consistent with a relatively large
number of MoAs for iAs. These include (among others) oxidative stress, cytolethality
and regenerative cellular proliferation, altered patterns of DNA methylation, altered DNA
repair, and DNA damage. Receipt of the proposals was followed by an external peer
review meeting. The outside experts judged that the proposals received did not
adequately represent plausible modes of action, which caused NHEERL management to
markedly reduce the planned BBDR modeling effort and focus on-going research on iAs
PK, with particular emphasis on evaluation of the arsenic 3-methyl transferase knockout
mouse. The NCCT involvement in the arsenic mode of action BBDR models has been
redirected to stronger interactions with existing NCCT projects in ToxCast™ and the v-
Liver, and will be presented to the BOSC at its next review of the Center.
10
Previous I TOC
-------
ORD Response to BOSC Computational Toxicology Letter Report
February 2009
Charge Question 3: Are there significant gaps in the approach that can be pointed
out at this point in the evolution of the project?
ToxCast: Specifically, the Subcommittee notes that the structural specification of the
database for compilation and rigorous quantitative analysis of the ToxCast data remains
unclear. Because the data types are highly heterogeneous and the dataset is very large,
developing these structural specifications will be a challenge that the Subcommittee
suggests should be addressed as soon as possible. The IT/IM team acknowledges that
this area is a significant challenge (e.g., the description in the write-up provided to the
Subcommittee prior to the review meeting). One suggestion is that the ToxCast team
compiles a list of some specific use cases, for example, specific questions that they intend
to address with the database. This will help make concrete the needed database
attributes that will allow the analysis for the chemical prioritization that is the end goal
of the ToxCast project 1). (Recommendation #10 in Table 1)
ORD Response: Over the last several months, these issues have become clearer, mainly
due to the fact that we now have access to large parts of the ToxCast™ data. With the
exception of the microarray genomics data, which has been delayed due to lack of
consensus on the most appropriate bioasssy conditions, the results of all of the assays can
be reduced to a small number of summary parameters. In most cases, one of these will be
a characteristic concentration for each chemical in the assay (EC50, IC50, lowest
observed concentration at which a significant effect is seen). The second parameter will
often be a magnitude of response. For all of the assays, we can extract a relevant
concentration and for many, a response magnitude. Related to this, the endpoint data we
will be predicting from ToxRefDB are characteristic concentrations, which are the lowest
doses at which a particular effect was seen with statistical significance. A third variable
in some assays is time - cell based assay data in some cases is provided at 2-3 time points
(e.g. 6, 24 and 48 hours). We track these times, but treat each of the times as separate
assays. Finally, most assays can be linked to biological pathways, either directly through
the gene or protein, or through a higher-order processing being probed. Although
ToxCast™ was envisioned to support chemical prioritization efforts of Agency regulator
offices, it has since been viewed as a source of ancillary information that can be used in
evaluating risks. Examples of this include interest of the toxic substances office on the
effects of perfluoroacids, NCEA with phthalates, and the pesticide office with conazoles.
Such interest demonstrates the multiple values the information emerging from ToxCast™
is having on the regulatory programs of EPA beyond chemical prioritization. We
anticipate continued interest in the use of ToxCast™ in risk assessment considerations
and are engaging NCEA in optimal ways to bridge the applications.
As already stated, the goal of ToxCast™ Phase I, as supported by the ToxMiner system,
is to find links between in vitro assays and in vivo toxicity as captured in ToxRefDB.
These can be statistical correlations or more biologically-based toxicity pathway
linkages. Given this, the ToxMiner database has been organized into five main pieces:
11
Previous I TOC
-------
ORD Response to BOSC Computational Toxicology Letter Report
February 2009
1. Chemical information - this holds chemical identity and structure
2. Assay information - this holds the summary values extracted from in vitro assays
and from ToxRefDB (concentrations, response magnitude), as well as other
related quantitative and qualitative information on chemicals such as physico-
chemical properties and chemical class information.
3. Data preparation - for many of the data sets, several pre-processing steps need to
be undertaken to map raw data onto the canonical chemical and assay data
structure. These tables and data structures enable these steps to be carried out in
well-controlled manner
4. Statistical analysis workflow - many calculations need to be carried out to find
signatures and the results need to be tracked and made available to the ToxCast™
team on the web. We are implementing specific data tables and code to carry out
these steps.
5. Pathway information - this set of data tables and tools are being designed to allow
the analysis of the ToxCast™ data in terms of biological pathways.
Charge Question 3 continued:
IT/IMActivities: The major gap noted for this activity was described in the ToxCast
project section above. In addition, finding an efficient and effective methodology for
extracting data from text sources was a concern for the Subcommittee. A trial of natural
language processing (NLP) for pulling information into some of the databases was
described. The Subcommittee notes that this method has been attempted rather
unsuccessfully by various research groups over probably 2 decades and thereby
encourages the exploration of other possible approaches as well. (Recommendation #11
in Table 1)
ORD Response: NCCT agrees and is developing two main uses for literature mining, for
which we believe current technology is suitable. In the first case, we need to extract
tabular data for use in ToxCast™ and the virtual tissue project. These are, for instance,
quantitative values associated with in vivo toxicity or in vitro assays. Here we are using
text mining as a sophisticated version of a PubMed search to prioritize documents for
data extraction and to do an initial automated data extract. The results are then presented
to an analyst to do manual quality control and data cleaning.
The second task is to generate hypotheses about biological processes such as the co-
occurrence of gene expression changes and the observation of higher-order phenotypes.
The lack of success that the reviewer alludes to, we would argue, is in taking these
hypotheses and assigning some truth value to them based on statistical arguments. We
are using these simply as starting points for building representations of pathways and
processes that will be tested through further experiments and analyses. A more detailed
explanation of our approach to literature mining and evidence of utility will be presented
at the next BOSC review.
12
Previous I TOC
-------
ORD Response to BOSC Computational Toxicology Letter Report
February 2009
Charge Question 3 continued:
Virtual Liver: Dr. Shah and his group are commended for having a good command of
the significant breadth of biology, toxicology, and modeling that impacts the project. In
addition, the "bigpicture " vision described is useful—there are many important
questions in the field and not limiting the vision too early is appropriate. The
Subcommittee believes that this should be balanced, however, with some very specific
goals, milestones, and timelines for the next few years that are clearly attainable with the
resources at hand in order to assure some useful concrete outcomes. In a project with
this possible magnitude, it can be tempting to try to do everything, both in terms of the
various project approaches (knowledgebase (KB), biological modeling, dosimetry
modeling, etc.) as well as the scope within any one approach (breadth of the KB, breadth
and detail of every model, etc.), and thereby end up with little actually completed. One
suggestion is that Dr. Shah and the group develop a short prioritized list of specific
scientific research questions relevant to EPA 's goals that they desire to address as soon
as possible, and use this to focus first iterations of development of both the KB and
model(s). More explicit milestones and goals for these highest priority questions then
can be developed. Later iterations of KB development and modeling can add scope
(breadth/depth) to allow NCCT to address additional research questions.
(Recommendations #6 and #12 in Table 1)
ORD Response: The question, "How can in vivo tissue level adverse outcomes in
humans be predicted using in vitro data? " is the "Grand Challenge" scientific problem
in toxicology that motivates the v-Liver project. This is a very ambitious goal and
infeasible to achieve in the broad sense in just a few years. Hence, the v-Liver project
will take a few steps towards realizing this long-term objective by focusing on a tractable
proof of concept (PoC) system using ten environmental chemicals that activate nuclear
receptors and cause a range of apical effects in cancer progression (non-proliferative
lesions, pre-neoplastic lesions, and neoplastic lesions). The project will engage program
office personnel to ensure relevance to EPA's mission and provide deliverables for risk
assessment within the first two years. These deliverables focus on two main scientific
questions:
a) How can tissue level adverse effects be modeled to enable extrapolation? The v-Liver
leverages the Mode of Action Framework and public sources of mechanistic information
to formalize the description of key events leading to adverse hepatic outcomes. Our
claim is that MOA knowledge can be universally described across species, organs,
chemicals and doses, using genes, their interactions, pathways and cellular responses that
lead to toxic effects. This claim will be tested in the PoC by: (a) organizing sufficient
information about the 20 nuclear receptor-activators to demonstrate that key events in the
MOA(s) can be described generally for extrapolation across chemicals and species, and
(b) using semantic methods to build an ontology for the physiologic processes, a
knowledgebase to integrate this information, and inference tools for extrapolation. The
result of this exercise will be delivered as the v-Liver-KB.
13
Previous I TOC
-------
ORD Response to BOSC Computational Toxicology Letter Report
February 2009
b) How can the tissue level effects be extrapolated across doses and time? Our claim is
that quantitative tissue level effects can be generated from qualitative logical descriptions
of the MOA(s), chemical-specific data for key events and simulation of the tissues as a
cellular system. The rationale for the v-Liver Simulator is to implement a virtual hepatic
lobule as a complex cellular system to investigate emergent tissue-level effect due to
alternative MOA(s) at very low environmentally relevant doses. To extrapolate between
species, chemicals and doses, the v-Liver team is collaborating across ORD and
extramural funding to develop in vitro models and assays to relevant quantitative data
key events. In addition, to estimate internal dose and to model alternative exposure
scenarios the project is working closely with PBPK modeling efforts across ORD. The
deliverable for this part of the project will be the v-Liver Simulator.
Charge Question 3 continued:
Virtual Liver: The Virtual Liver activity will result in models of parts of the biology
being developed simultaneously and presumably by different individuals. Because the
idea is to integrate these models eventually to predict effects from molecular function to
physiologic outcome, the compatibility of the models is paramount. Dr. Shah indicated
that he is cognizant of and planning to manage this issue, for instance, by looking into the
efforts of the international Physiome Project. The Subcommittee members note that, to
their knowledge, the issue of common coding language, which has been addressed quite
extensively by the Physiome Project, does not appear to have addressed more subtle but
critical compatibility issues concerning biological and mathematical specifications
among models, such as compatibility of assumptions, equilibrium approximations, time
scales, and so forth. Hence, beyond managing compatible coding, the activity group is
encouraged to actively plan for and manage on an ongoing basis the specifications that
must be shared among models so as to produce compatibility when it is needed.
(Recommendation #13 in Table 1)
ORD Response: This is indeed a difficult and very important issue to consider. To this
end NCCT is beginning to address the issue on two fronts:
1. NCCT plans to raise this issue for discussion by multi-scale modeling experts at
the NCCT organized International Workshop on Virtual Tissues, to be held in
Research Triangle Park, NC, April 21-23, 2009. This workshop will have
representation from the Physiome project and the SBML project and is co-
sponsored by the European Union.
2. In addition, the NCCT is actively collaborating with PBPK modelers in the
Agency to develop a formal specification that will ease the integration with v-
Liver-Sim. The effort is using semantic technology to define physiologic models
at the organism level that can interface with existing tools in NERL.
These two integrated efforts will be important early steps for addressing this problem.
14
Previous I TOC
-------
ORD Response to BOSC Computational Toxicology Letter Report
February 2009
Charge Question 3 continued:
Virtual Embryo: Because the data needs of the proposed models may be significant, the
Subcommittee notes that it will be critical to identify and enlist appropriate supporters
and collaborators to provide such data. The track record of the principal investigator
suggests that this will develop naturally. (Recommendation #14 in Table 1)
ORD Response: With successful proof-of-principle (2 yrs), the computational model of
early eye development will be used to create general models of morphogenesis during
subsequent years. Any proposed model of chemical dysmorphogenesis must be
sufficiently abstract to be computationally feasible and yet detailed enough to enable the
realistic expression of developmental defects across chemicals, doses, tissues, stages, and
species. The data needs of the proposed models will be significant as noted by the
Subcommittee. Preliminary computational models can attach existing data from in vitro
studies and semi-arbitrary parameters from in silico resources. These models will be
calibrated across species (zebrafish, mouse, rat, human) and tested for predictive
capacity. In this regard, the Virtual Embryo will leverage data generated by NCCT's
high-throughput chemical screening and prioritization research program (ToxCast™,
ToxRefDB) to model developmental toxicity pathways.
Importantly, to stimulate research in this area, NCER released a funding opportunity
under its Science To Achieve Results (STAR) research program, "Computational
Toxicology Research Centers: in vitro and in silico models of developmental toxicity
pathways" (EPA-G2008-STAR-W). Collaboration with future STAR center(s) can
provide experimental data to identify developmental toxicity pathways and computational
models for developmental defects.
Because conservation of cell signaling is a founding principle of early development
across species and stages, the in silico toolbox is likely to be extensible across
morphoregulatory responses. As such, in silico models built from scratch can be
generalized to other systems (neural tube, cardiac, urogenital) and alternative models
(embryonic stem cell assays, zebrafish embryos) for chemical-pathway interactions. In
this regard, the Virtual Embryo has begun to identify and enlist collaborators at NHEERL
to help provide such data.
High-throughput platforms now offer a powerful means of data gathering to discover key
biological pathways leading to apical endpoints of toxicity, and computational model
structures our ability to integrate these data across biological scales to build predictive
models that address mode-of-action. Successful computational models can become
increasingly important in EPA efforts to translate pathway-level data into risk
assessments, and in that regard the Virtual Embryo has also begun to identify and enlist
support from NCEA. A web-site has been developed to communicate publically about
the project (http://www.epa.gov/ncct/v-Embryo/).
15
Previous I TOC
-------
ORD Response to BOSC Computational Toxicology Letter Report
February 2009
Charge Question 3 continued:
Arsenic BBDR: The Subcommittee encourages continuous communication with the
appropriate program office personnel so that concerns, objections, and skepticism can be
addressed early and on an ongoing basis. The group is commended for having such
communication already in place and it is encouraged to maintain that communication to
the greatest degree possible. (Recommendation #15 in Table 1)
ORD Response: As discussed in the response to charge question 2, this project was
largely terminated in 2008, with the exception of a few smaller efforts on
pharmacokinetics of arsenic. NCCT efforts are being redirected to incorporate concepts
of BBDR in the virtual tissue models, particularly from the viewpoint of dose-response
extrapolation. Additional NCCT efforts are being directed at interpreting the results of
ToxCast™ in vitro concentration responses relative to the range of potential external
exposures that could provide equivalent tissue level responses (i.e., reverse
toxicokinetics). As we move forward in these areas, we will ensure adequate discussion
with client offices in EPA takes place on a routine basis.
Charge Question 4: Does the work offer to significantly improve environmental
health impacts and is the path toward regulatory acceptance and utilization
apparent?
ORD Response: ORD is very appreciative of the committee's affirmation of work and
progress in ToxCast™, Informatics, and the virtual tissues. The NCCT will present
further updates on progress at the next committee review.
Charge Question 4a: In addition, specifically for the Arsenic BBDR project:
Does the proposed computational model have the potential to identify and reduce
uncertainties with the risk assessment process?
The answer to this question is yes, depending on data gaps identified and resources made
available. This study might not give all the answers but will get us halfway there. EPA
recognizes that developing a universal arsenic model describing several cancer
endpoints is a formidable challenge. Hence a step-wise research project with an eye for
the future is proposed. Initially, a generic model for cancer will be developed that will
incorporate key steps of the mode of action commonly shared for multiple cancer types
such as oxidative stress. This model, in turn, will serve as an engine to develop specific
cancer models as the need arises and resources become available. To ascertain whether
appropriate steps are being incorporated, a thorough literature review of experimental
and epidemiological data and expert consultation has been proposed. It also is
acknowledged that even though there is a lot of data, they are somewhat weak to
generate exposure time course response curves. Appropriate experiments have been
proposed to fill the research needs to develop a realistic model.
ORD Response: Please see earlier response regarding the arsenic BBDR project.
16
Previous I TOC
-------
ORD Response to BOSC Computational Toxicology Letter Report
February 2009
Charge Question 4b: Will the model be able to help identify susceptible populations
and compare potential risks in those populations with less susceptible populations?
Yes, the initial generic model development exercise will allow identification of issues
such as mechanisms that operate in general versus subpopulations, such as susceptible
populations with varying degree of arsenic methylation. Such issues could be the subject
of workshops to explore the issue of the extent of polymorphism in the human population.
The short-term (1-2 years) goal is the establishment of a coordinated program of
laboratory research to generate essential data needed to develop a BBDR model that will
increase confidence in the predictions. To start with, the model development will be
initiated with available data. Work proposed includes multistage clonal growth modeling,
target tissue dosimetry, and methylated metabolites of arsenic.
The long-term (3-5 years) goal of developing a robust version might be too optimistic. As
the project gets underway, new questions and issues might be identified that will require
additional laboratory research and continued resources. The project has a good future as
it can be easily adapted to the latest (2007) National Academies toxicity testing report
that recommends a systems biology and computational tool integration.
ORD Response: Please see earlier response regarding the arsenic BBDR project.
Charge Question 4c: Is coordination between model development and associated
data collection sufficient to avoid problems with models being either over- or under-
determined?
Yes, it is desirable to see what health effects are caused at lower doses to avoid the
potential of compromise in setting an arsenic standard based on cost-benefit analysis.
ORD Response: Please see earlier response regarding the arsenic BBDR project.
Charge Question 5: Have appropriate data management and analysis tools been
incorporated into the project?
ToxCastFM: The construction of the warehouse remains an open question. Ultimately, a
database is a model of the interactions that exist in the underlying data and the relationships
relevant to the analysis that will be performed. The diversity of the data, representing a wide
range of in vivo and in vitro assays from multiple species, makes building such a model a
significant challenge. The project seems to be lacking a set of analytical objectives
necessary for building the relevant use cases that ultimately will inform the process of
database construction, and this ultimately will determine its utility. At this stage, ToxCast
needs to begin to define analytical outcomes in order to set goals and milestones with regard
to developing and validating analytical protocols. This is an essential step at this point as it
will help to anchor future development and make it relevant. This also will help to define the
requirements of the interfaces that are built to access the data.
17
Previous I TOC
-------
ORD Response to BOSC Computational Toxicology Letter Report
February 2009
Further, the ToxCast group should be encouraged to release the data and databases at the
earliest possible time and to consider a "CAMDA-like " workshop in which the research
community is offered access to the data with the challenge of using the data to effectively
predict end points. At least three advantages to the program will be derived from these
efforts. First, public release will help to drive the creation of relevant use cases that will
further database development. Second, it will assist in evaluating data access protocols and
tools to assure the greatest utility to the research and regulatory community. Third, it will
accelerate the development of predictive algorithms to combine the data to make predictions
about relevantphenotypic outcomes. (Recommendations #6, 10 and 16 in Table 1)
ORD Response: The first part of this question (database design and construction) was
addressed in the response to charge question 3. The ToxMiner database is able to capture
and provide all of the summary information which we believe is going to be useful for
statistical and pathway-based analysis of the ToxCast data sets.
The second question relates to analytical outcomes. By this we assume the reviewer
means the desired outcomes of analyses of the ToxCast data. We believe that the
outcome of ToxCast will be a series of well-defined procedures that take as input the
results of a set of in vitro assays run on a chemical and give a result which is a statement
about the likelihood that the chemical will lead to a particular toxicity phenotype. The
simplest procedure is a formula (e.g. a logistic regression model) that uses the IC50
values for several assays and gives a binary prediction for a particular toxicity. More
complex procedures would use the results from a set of assays to predict whether a
particular pathway is activated. Then we could have a function that predicts the
likelihood of the outcome, given the activation of one or more pathways. The current
database has been designed to hold both the numerical data required to test these models,
and the model parameters and outcomes. In summary, we feel that this issue has in
general been resolved over the last several months although many details still need to be
worked out, particularly regarding the best statistical approaches to be used, and the
precise way that pathway information will be incorporated.
With regard to the last comment by the reviewer, a recommendation that we hold a
CAMDA-like workshop, we are currently planning such a meeting to be held in May
2009. We plan to make all of the ToxCast M data available to analysis partners in early
2009. By having a larger community trying many analysis techniques on this data, we
will maximize our chances of success.
Charge Question 5 continued:
V-liver: With regards to populating the KB, the use of NLP probably is not the best
solution. NLP does not work well with the scientific literature, and its application in this
domain remains an area of active research. Application of NLP has the potential to
introduce a great deal of noise in the system, leading to many potential false associations
that could lead to more problems than it solves. Consequently, other methods, including
expert or community curation, should be explored.
On a larger scale, the greatest potential problem will be linking each of the domain-
18
Previous I TOC
-------
ORD Response to BOSC Computational Toxicology Letter Report
February 2009
specific models to build a predictive system. Again, this remains an area of active
research and one that may present significant barriers to developing verifiable solutions.
The greatest challenges will be to validate any models that emerge from the analysis.
Finally, there is a need to develop standards for interactivity and try to interface with
developing standards within the community. (Recommendation #17 in Table 1)
ORD Response: Linguistic resources have several applications in the Virtual Embryo
although an important challenge noted by the Subcommittee is to unambiguously code
unstructured (text) data in a form that can be processed by a computer to derive
interesting relationships and causality. Querying within the proper context can make
these more precise and less noisy. NLP enhances the coarse semantic search for specific
concepts and then provides a way to automatically extract the key facts, relationships and
quantitative information. The results are then presented to an analyst to do manual
quality control and data cleaning. As such NLP extends, but does not replace the need
for a formal concept model (ontology) to organize the relevant information about
developmental processes and toxicities that is often present in literature in an
unstructured format.
Also noted by the Subcommittee, a broader network of expertise within the
developmental toxicology community may be useful to building the information network.
Virtual Embryo has incorporated two open ontologies to arrange information, one for
embryology and the other developmental toxicology, and implemented this ontology in
Protege (http://obofoundry.org/). This formal ontology will be available for community
participation in linking each of the domain-specific models to build a predictive system
for the embryo as a whole. Furthermore, informal ontologies that include less explicit
information about a pattern of malformations and underlying embryology can make a
useful contribution when the end-user is knowledgeable about the field. Hence, Virtual
Embryo is piloting a Wiki-space (http://v-embryo.wikispaces.com/) to generate
hypotheses about the co-occurrence of specific malformations to common embryology,
or the relationship of genetic defects to higher-order phenotypes, for building
representations of pathways and processes that can be tested through further experiments
and analyses.
Charge Question 5 continued:
V-Embryo: It remains to be seen how well it will eventually integrate with the overall
program, and its integration with other internal and external initiatives needs to be
resolved. Nevertheless, it appears that this project could provide an opportunity to
explore the results emerging from ToxCast, and it may help direct selection of the next
generation of compounds for analysis in ToxCast. (Recommendation #18 in Table 1)
ORD Response: Although still early in its development Virtual Embryo has begun to
integrate with other activities, especially ToxCast M and the Virtual Liver. Since its
inception last December and the review addressed here, the v-Embryo has been:
1. integrated into NCCT's Computational Toxicology Research second
generation Implementation Plan;
19
Previous I TOC
-------
ORD Response to BOSC Computational Toxicology Letter Report
February 2009
2. presented at five seminars at EPA (including NCEA) and six seminars
outside EPA (including a Gordon Research Conference);
3. introduced at NCCT's Computational Toxicology education course, at two
presentations describing the implementation of prenatal developmental
studies in ToxRefDB (manuscript in preparation), and one presentation on
ToxCast™'s NovaScreen assay (manuscript in preparation);
4. the topic of one book chapter (in print) and seven abstracts (five in print
and two accepted);
5. reflected in one submitted abstract in collaboration with Virtual Liver, and
three
6. submitted abstracts in collaboration with ToxCast™; and
7. presented in the virtual tissue research section at the Human Health
Program Review (BOSC, January 2009).
20
Previous I TOC
-------
ORD Response to BOSC Computational Toxicology Letter Report
February 2009
Appendix I: Summary Action Items
Detailed Milestones in response to Charge Question 2
ToxCast™:
FY09
• First initial publications and public access to ToxCast™ in vitro assay data
• Completion of generating all of the ToxCast™ Phase I data
• Sharing of ToxCast™ Phase I data with data analysis partners and hosting of the
first "ToxCast™ Data Analysis Summit"
• Develop a series of "signatures" linking ToxCast™ in vitro data with ToxRefDB
in vivo toxicology.
• Initiate generation of ToxCast™ Phase II data
• Quarterly public releases of new ToxCast™ data of various study types
FY10
• Quarterly public releases with new ToxCast™ data
• Completion of generating all of the ToxCast™ Phase II data
• Sharing of ToxCast™ Phase II data with data analysis partners and hosting of the
second "ToxCast™ Data Analysis Summit"
• Validation of predictive "signatures" linking ToxCast™ in vitro data with
ToxRefDB in vivo toxicology
FY11
• Quarterly public releases with new ToxCast™ data
• Application of toxicity predictions from Phases I and II of ToxCast™ to chemical
prioritizations in EPA Program Offices
• Initiate generation of T
requiring prioritization
ACToR:
FY09
• Initial public deployment
• Significant version 2, including refined chemical structure information
• Develop workflow for tabularization of data buried in text reports
• Integrate all ToxCast™ and ToxRefDB data
• Quarterly releases with new ToxCast™ data
Quarterly releases with new ToxCast™ data
Initiate generation of ToxCast™ Phase III data on chemicals and nanomaterials
FY10
• Implementation of a process to gather tabular data on priority chemicals from text
reports
• Perform survey of sources of exposure data and import any remaining sources
• Develop flexible query interface and data download process
• Develop process to extract data from open literature
FY11
• Quarterly releases with new ToxCast™ data
Previous I TOC
-------
ORD Response to BOSC Computational Toxicology Letter Report
February 2009
ToxRefDB:
FY09
• Initial public deployment of chronic toxicity data
• Public deployment of reproductive and developmental toxicity data
• Develop flexible query interface and data download process
• Develop workflow for curation of similar, but non-guideline chronic, reproductive
and developmental study types
• Public deployment of developmental neurotoxicity data
• Quarterly public releases of new data of various study types
FY10
• Quarterly releases with new ToxCast™ data
• Implementation of a process to curate data on ToxCast™ Phase II chemicals from
multiple sources
FY11
• Quarterly releases with new ToxCast™ data
DSSTox:
FY09
• Publish paper and property files on ToxCast 320 chemical inventory, with
guidance for SAR modeling study
• Publish DSSTox ToxCast 320 categories file and DSSTox ToxRef summary data
files
• Coordinate efforts to structure-annotate and provide effective linkages to
microarray data for toxicogenomics
• Compile and publish public genetic toxicity data and SAR predictions for
ToxCast 320
• Restart Chemoinformatics Communities of Practice using EPA's Science Portal;
FY10
• Publish new DSSTox database and doc
• Explore new approaches to SAR modeling based on feature categories within
existing DSSTox files and ToxCast™ data
• Expand CEBS collaboration to incorporate DSSTox chemical content, create
chemical linkages to external projects;
• Separately publish DSSTox structure inventory with various chemical
classifications for use in modeling using publicly available tools
FY11
• In collaboration with ACToR, establish procedures and protocols for automating
chemical annotation of new experimental data submitted to CEBS or NHEERL
• Document and employ PubChem analysis tools in relation to published DSSTox
and ToxCast™ data inventory in PubChem
• Collaborate with SAR modeling efforts to predict ToxCast™ endpoints using in
vitro data
Previous I TOC
-------
ORD Response to BOSC Computational Toxicology Letter Report
February 2009
• Continue expansion of DSSTox public toxicity database inventory for use in
modeling with co-publication and linkage to ACToR and PubChem
v-Liver:
FY09
• Prioritize proof of concept (PoC) environmental chemicals with clients. Using
toxicity data from ToxRefDB and bioactivity data from ToxCast™, a subset of
Phase I chemicals will be selected for the PoC, which will be finalized in
collaboration with program offices to ensure relevance to EPA needs.
• Begin deployment of v-Liver KB on physiologic processes perturbed by PoC
chemicals. The first version of the KB will focus on the PoC chemicals and
populated mostly with their molecular activity data from ToxCast™, and cellular
and tissue level outcomes from ToxRefDB and the literature.
• Deploy KB visualization tool for client interaction. Access to the KB will be
provided using open source tools for biological data analysis.
• Simulate of liver lesions for alternative MOA/toxicity pathways. The prototype of
the lesion simulator implementing the main MOA for hepatocarcinogenesis.
FY10
• Evaluate simulator using PoC chemicals and ToxCast data to predict outcomes.
• Quarterly update of v-Liver KB
• v-Liver KB inference tool for analyzing MOA for new chemicals/mixtures
Extend v-Liver Simulator to liver and integrate with PBPK model
• Evaluate impact of genomic variation on cellular responses and lesion formation
• Evaluate v-Liver for simulating human pathology outcomes using clinical data
Most milestones will also include manuscript submissions describing the computational
methods and their biological/toxicological relevance.
v-Embryo:
• Literature-mining tools to index relevant facts about early eye development and
concept model (ontology) to support this knowledge representation [2];
• Ocular gene network schema specified by gene-gene and gene-phenotype
associations and subjected to dynamical network inference analysis;
computer program of early eye development that reconstructs lens vesicle
induction in silico using cell-based simulators and system-wiring diagrams of
perturbation analysis of the computational (in silico) model with pathway-level
data for normal and abnormal (toxicological) phenotypes in vitro and in vivo.
FY09
• Project plan and quality assurance plans for VT-KB and VT-SE
• Recruit: student contractor and postdoctoral fellow
Previous I TOC
-------
ORD Response to BOSC Computational Toxicology Letter Report
February 2009
• Manuscript: application of VT-KB to analyze ToxRefDB developmental toxicity
studies
• Model: VT-KB based qualitative (structural) model of self-regulating ocular gene
network
• Model: VT-SE based cell-based computational model of lens-retina induction
• Manuscript: ocular morphogenesis, gene network inference, analysis and
modeling
FY10
• Project plan: extend lens-retina model to other stages and species
• Model: incorporate pathway data from ToxCast™, mESC and ZF embryos
• Manuscript: sensitivity analysis for key biological pathways
• Manuscript: analyze developmental trajectories and phenotypes in computational
models
• Project plan: integrate with other morphogenetic models
FY11
• Manuscript: test model against predictions for pathway-based dose-response
relationship
• Manuscript: uncertainty analysis of models for complex systems model: computer
program of early eye development using rules-based architecture, cell-based
simulators and systems-wiring diagrams
Previous I TOC
-------
Table 1. Summary of Recommendations and Proposed Actions.
Recommendation
Action Item
Time Line
1. Involve
stakeholders in future
BOSC meetings.
Identify and invite key stakeholders to attend and
participate in BOSC meetings. Further, these
stakeholders will also be asked to review and
comment on new implementation plan.
Early 2009
2. Hold discussions
with risk assessment
practitioners.
Discussions with NCEA and others are underway,
and regular Communities of Practice are also an
ongoing activity to help achieve this end.
Ongoing
3. Develop effective
ways of dealing with
wealth of data and
interact with program
offices on this issue.
Extensive suite of interactive databases is under
development and prioritization of data input is in
consultation with program offices and others.
Ongoing
4. Relevance of
ToxCast™ beyond
prioritization to risk
assessment, including
exposure paths, and
ecology.
Expanded workgroups to address exposure
pathways through ExpoCast. Partnering with
NHEERL and NERL for HTS for ecological
species other than human. Testing of
2009-
2012
.TM .
Pharmaceuticals in Phase II of ToxCast to
compare results to known human toxicities.
5. Involve risk
assessors and others in
program offices for
planning on eventual
application of v-Liver
to risk assessment.
Project team includes NCEA and will be
expanded to include others. Consultations with
program office to identify practical use cases that
demonstrate utility of virtual tissues.
Begin in
2009
6. Detailed time table
for milestones:
ToxCast
IT/IM.
TM
Developed and put in place (See Appendix I for
details).
Ongoing
v-tissues,
7. Identify and
prioritize key
objectives for v-Liver
with milestones.
TM
Short-term goals: Identify environmental
chemicals for proof of concept (PoC) in
consultation with stakeholders; use of ToxCast
data to begin quantitative parameterization of
cellular and molecular responses - see Appendix
I for detailed milestones.
Long-term goals - use in RA with details being
developed and to be presented at next BOSC
review.
2009
8. Develop more
detailed plan for v-
Embryo with
milestones.
Plan has been developed with long-term goal of a
computational framework enabling predictive
modeling of prenatal developmental toxi city;
Please see Appendix I for milestones.
2009
9. Develop milestones
for Arsenic BBDR.
Please see narrative for significant change in
plans for this work
N/A
Previous
TOC
-------
Table 1. Summary of Recommendations and Proposed Actions
10. Compile a list of
specific use cases for
specific questions that
will be addressed with
the database of
ToxCast™ data.
A goal of Phase I of ToxCast is to find links
between in vitro and in vivo toxicity as captured
in ToxRefDB. To achieve this, the ToxMiner
database is being organized into five main pieces
- please see narrative for details.
2009
11. Exploration of
alternatives to natural
language processing
(NLP).
Approaches for improving on previous uses of
NLP are underway - greater dependence upon
further testing and analyses for starting points
derived through NLP.
Ongoing
12. Develop explicit
milestones and
research questions for
addressing EPA's
goals, and use this to
focus first iterations of
development of both
the KB and model(s).
Questions and associated milestones have been
developed:
1) Modeling of tissue level adverse effects to
enable better extrapolation by formalizing
the description of key events leading to
adverse outcomes;
2) Extrapolating the tissue level effects
across doses and time.
Ongoing
13. Delineate model
specifications for
sharing between
models of different
scales that can then be
interconnected when
appropriate.
International workshop in April 2009 will include
multi-scale modeling experts to consider this
issue and NCCT is collaborating with PBPK
modelers to develop formal specifications to ease
model integration.
Ongoing
14. Enlist appropriate
supporters and
collaborators to gain
necessary data for
developing v-embryo.
NCCT has worked with NCER to develop STAR
funding opportunities that, through collaboration,
can provide key data. In addition, collaborators
in NHEERL have been identified and discussions
have begun.
Ongoing
15. Continuous
communication with
program office
personnel regarding
Arsenic BBDR.
Please see memo regarding the suspension of this
project. ORD's decision on this project was in
consultation with program office and based on the
program office's plans for this chemical and their
reduced need for this modeling effort.
2008
16. ToxCast needs to
define analytical
outcomes to develop
and validate analytical
methods
The outcome of ToxCast will be a series of well-
defined procedures that take as input the results
of a set of in vitro assays run on a chemical and
give a result which is a statement about the
likelihood that the chemical will lead to a
particular toxicity phenotype.
Ongoing
17. Limitations of
natural language
processing (NLP) for
v-embryo and using
alternatives.
The NCCT proposes using NLP as a starting
point and then presenting the results to an analyst
for manual quality control. NLP is used to
extend, but not replace, the need for formal
concept modeling (ontology) to organize relevant
Ongoing
Previous
TOC
-------
EPA CompTox Research Program FY2009-2012 BOSC Review Draft- 24 August, 2009
Synopsis: ACToR is a web-based informatics platform, organized at the top level by chemical
and chemical structure that is indexing, collecting, and organizing many types of data on
environmental chemicals. Environmental chemicals are defined as those likely to be in the
environment, including all chemicals regulated or tracked by the EPA, as well as related
chemicals, such as Pharmaceuticals that find their way into water sources. ACToR is indexing
and linking to data from hundreds of sources, including the EPA, FDA, CDC, NIH, academic
groups, other governmental agencies (state and national) and international organizations, such as
the WHO. Information being indexed and gathered includes in vivo toxicity, in vitro bioassay
data, use levels, exposure information, chemical structure, regulatory information and other
descriptive data. Planning for the project began in mid-FY07; beta versions were available inside
the EPA since early FY08, and a public version became available in December 2008. ACToR
consists of a back-end database and a front-end web interface built on low-cost, publicly
accessible applications and tools. Over the next 3 years, ACToR will expand to include more
publicly available resources and data, including more information extracted from text reports and
tabularized, and more information on chemical use and exposure. The latter effort will be
coordinated with the efforts of ExpoCast™ and NERL to identify, index and extract data from
exposure-related resources of highest interest and importance to EPA programs. In planned
upgrades to ACToR, the ability of users to perform flexible searches across different layers of
data will be enhanced, and customized data downloads will be implemented. ACToR will serve
as the primary vehicle to aggregate and publicly disseminate all published data associated with
the ToxCast1 , ToxRef, and Tox21 research projects. Additionally, the ToxMiner and NCCT
Chemical Repository systems are being developed as part of ACToR. These are data repositories
and data analysis engines for the ToxCast/Tox21 projects.
Partnerships/Collaborations (Internal & External):
1. EPA ToxCast™ program - provide data for use in selecting chemicals and providing
toxicology data for validation; provide route for publication of data
2. Tox21 partnership - provide data for use in selecting chemicals and providing toxicology
data for validation; provide route for publication of data
3. DSSTox coordination - align methods for registering high-interest chemical inventories
(ToxCast™, ToxRef, Tox21, DSSTox published data files), utilizing DSSTox chemical
information quality review and structure-annotation within ACToR
4. EPA Centers and Offices (OPPT/OPP/NCEA/OW) - provide data on chemicals of
interest
Milestones/Products:
FY09
1. Initial public deployment.
2. Significant version 2, including refined chemical structure information.
3. Develop workflow for tabularization of data buried in text reports.
4. Integrate all ToxCast™ and ToxRefDB data.
5. Quarterly releases with new data.
2a
Previous I TOC
------- |