United States
Environmental Prot
Agency
. I 00i 0 0 I 00 i 0
1001! 0110000001 1001 I i
0110 1 I 1 0 1 1 1 1 0 00 I I 0 11.
00 I 001 100 000 00 00
2 01110010010 oo oi11oo
- O 000 11011 0 OOOO 1
OlO OlIOOOiOMO
:oi I 00 "001 001 I
fl I \ 1 /> *VA • 1 0 0 I 1
0 00
, 0 1 1 1 00
0000
0010110010 0 I l 000
1 00 1 1 0 1 I 0 0 0 0 0 1 O i
0 0111000 0
1000 1000 0 10010
10011 01 I 0000001 100M 0
110 1110111 i 0 00 1 I 0
iooiooiooi OOIOOIQOIOO
1111111111 11111111111
rO 100100*00 i o o i o o i o i
rriiiiiili 11111111111
O O 1 0 0 1 0 0 1 0 01001001001
1011011011 011011O11O1.
JT011O1101 1011011O110I
1001001001 OQIOOlOOir
0010010 01001001'
•trfo~o 100 1 o o 1 o~
1101101 10110...
'* 0*70110 flOl 10110 110110110110110110%
a 010010010010010010 0100100100100100100
1 i 1 i 1 i 1 1*1 fj J11111 11111111111A1111A1\
Program
ccfccr^Crc ~ iCWruurutJiu^Tcuv^v^x
AC T A C T A C TAG T A C TAG TAC T AC TACTACT KCl KOT
\CCACCACCACCACCACCACC ACCACCACCACCKCC
ACTACTACTACTACTACJAC TACTACTKCT <\CT K1
^^AGGAGGAG GAGGAGGKGGKG
C>trC>irC>4rCATCATCA TCATCATCKTCK
RESEARCH AND DEVELOPMENT
-------
-------
EPA/600/R-03/065
November 2003
A FRAMEWORK FOR A COMPUTATIONAL TOXICOLOGY
RESEARCH PROGRAM IN ORD
US Environmental Protection Agency
Office of Research and Development
Washington, DC 20460
-------
DISCLAIMER
This document has been subjected to internal and external review for clearance. The Agency's
Science Advisory Board provided comments which have been addressed in this draft. This
report does not constitute an Agency position or policy concerning computational toxicology.
Any mention of trade names does not constitute Agency endorsement.
-------
FOREWORD
The 2003 Framework for a Computational Toxicology Program in ORD addresses
research needed to apply novel technologies derived from computational chemistry,
molecular biology, and systems biology-collectively known as computational toxicology-
to improve the Agency's prioritization of data requirements and risk assessments. The
Framework identifies three strategic objectives: (1) to improve linkages across the
source-to-outcome continuum, (2) to develop approaches for prioritizing chemicals for
subsequent screening and testing, and (3) to produce better methods and predictive models
for quantitative risk assessment. The Framework provides a conceptual basis for building
on current research in ORD to develop a research program on computational toxicology.
The Framework outlines the development of a multidisciplinary, integrated
research program that will use computational approaches to link chemical transformation
and metabolism, exposure indicators, dose metrics, toxicity pathways, systems biology,
and modeling frameworks. The research program will improve the Agency's ability
to screen and test for chemical hazards and will address uncertainties associated with
dose-response assessment, cross-species extrapolation, and the assessment of chemical
mixtures. The program also aims to predict aggregate and cumulative risk, protect
susceptible subpopulations, and provide principles for the use of mechanistic data in
human health-risk assessments, which are research needs identified in the ORD 2003
Human Health Research Strategy. The achievements of the computational toxicology
research program will improve the Agency's understanding of the links between human
activities, natural dynamics, ecological stressors, and ecosystem condition and, in doing
so, will fulfill significant goals of the Agency's 2003 Strategic Plan.
The Framework is intended to identify the research needs and unique capabilities
of ORD laboratories to support a more focused and integrated research program in
the future. This document was reviewed by the Agency's Science Advisory Board
(SAB) in September 2003 and was the focus of discussion at the ORD Computational
Toxicology Workshop: Framework, Partnerships and Program Development held in
Research Triangle Park, North Carolina, on September 29-30, 2003. Comments from the
SAB and workshop participants were used to revise the Framework and will be used to
guide the development of a multi-year plan that will outline research goals and measures
for the next 5-10 years.
Paul Oilman
Assistant Administrator
-------
TABLE OF CONTENTS
Page
List of Figures vi
Authors vii
Peer Review viii
Acronyms ix
Executive Summary E-l
I. Introduction 1
A. The Computational Toxicology Research Program 3
B. Application of Computational Toxicology to Risk Assessment and Research 4
C. Overall Goal and Strategic Objectives 5
II. Research Needs and Applications of Computational Toxicology to Goals 7
A. Improve Linkages in the Source-to-Outcome Paradigm 9
1. Chemical Transformation and Metabolism 9
a. Chemical Transformation in Ecosystems 9
b. Chemical Metabolism 10
2. Exposure Indicators 10
3. Dose Metrics 12
4. Characterization of Toxicity Pathways 13
5. Metabonomics 15
6. Systems Biology 16
7. Modeling Frameworks and Uncertainty Analysis 17
B. Provide Predictive Models for Hazard Identification 18
1. QSAR and Other Computational Approaches 18
2. Pollution Prevention Strategies 19
3. High Throughput Screening 20
C. Enhance Quantitative Risk Assessment 20
1. Applying Computational Toxicology in Quantitative Risk Assessment 20
2. Examples of Applications of Computational Toxicology to Quantitative Risk
Assessment 21
a. Dose-Response Assessment 21
b. Cross-Species Extrapolation 22
c. Chemical Mixtures 24
-------
Page
TTT Current Activities 25
A. Proof-of-Concept: Endocrine Disrupting Chemicals (EDCs) 27
B. Internal Linkages 29
1. Human Health Research ^jSr 30
2. Ecological Research 31
C. External Linkages 31
1. Chemical Industry Institute for Toxicology (CUT) Centers for Human Health 32
2. Department of Energy (DOE) 32
3. National Institute of Environmental Health Sciences (NLEHS) 33
4. Science to Achieve Results (STAR) 33
IV. Next Steps 35
A. Review of the Framework 37
B. Priorities for Research on Computational Toxicology in ORD 37
C. Process for a Research Program on Computational Toxicology 38
Appendix A. Examples of Current ORD Projects Associated with Computational Toxicology Al
-------
LIST OF FIGURES
Page
Figure 1 The Source-to-Outcome Continuum 5
Figure 2 An Example of a Toxicity Pathway 15
-------
AUTHORS
Robert Kavlock Chair, National Health and Environmental Effects Research Laboratory
Gerald T Ankley National Health and Environmental Effects Research Laboratory
Jerry Blancato National Exposure Research Laboratory
Tim Collette National Exposure Research Laboratory
John "Jack" Fowle IJJ National Health and Environmental Effects Research Laboratory
Elaine Francis National Center for Environmental Research
Earl Gray National Health and Environmental Effects Research Laboratory
Karen Hammerstrom National Center for Environmental Assessment
Lawrence Reiter National Health and Environmental Effects Research Laboratory
Jeff Swartout National Center for Environmental Assessment
Hugh Tilson National Health and Environmental Effects Research Laboratory
Greg Toth National Exposure Research Laboratory
Gilman Veith National Health and Environmental Effects Research Laboratory
Eric Weber National Exposure Research Laboratory
Doug Wolf National Health and Environmental Effects Research Laboratory
Douglas Young National Risk Management Research Laboratory
-------
PEER REVIEW
Peer review is an important component of evaluating research documents in the Office of
Research and Development. The peer review strategy history for this document follows:
ORD Executive Council: June 7, 2003
External Peer Review: Science Advisory Board/ Framework for a Computational Toxicology
Program in ORD, September 12, 2003
External Peer Review Panel Members:
Science Advisory Board Members:
Dr. George W. Lucier, Panel Chair, Consulting Toxicologist, Private Consultant, Pittsboro, NC
Dr. Charles A. Pittinger, Principal, The Cadmus Group Inc., Cincinnati, OH
Consultants:
Dr. Melvin Anderson, Director, Department of Biomathematics and Physical Science, CUT Centers for
Health Research, Research Triangle Park, NC
Dr. John Balbus, Director, Environmental Health Program, Environmental Defense Fund, Washington,
DC
Dr. Richard Becker, American Chemistry Council, Arlington, VA
Ms. Patricia Billig, Environmental Toxicologist, Waterstone Environmental Hydrology and Engineering,
Boulder, CO
Dr. Stuart Cagen, Toxicology Advisor, Shell Chemical LP, Houston, TX
Mr. Harvey Clewell, Principal, ENVIRON Health Sciences Institute, Ruston, LA
Dr. Darrell Donahue, Associate Professor, Chemical and Biological Engineering, University of Maine,
Orono, ME
Dr. B. Alex Merrick, Toxicologist, National Institute of Environmental Health Sciences, Research
Triangle Park, NC
Dr. Clifford P. Weisel, Professor, Environmental and Community Medicine, EOHSI, Robert Wood
Johnson Medical School-UMDNJ, Piscataway, NJ
Dr. Angela Wilson, Assistant Professor, Department of Chemistry, University of North Texas, Denton,
TX
Dr. Andrew Worth, Scientific Officer, Joint Research Center, Institute for Health and Consumer
Protection, European Chemicals Bureau, European Commission, Ispra, Italy
Science Advisory Board Staff:
Dr. James Rowe, Designated Federal Official, 1200 Pennsylvania Avenue, NW, Washington, DC
-------
ACRONYMS
AR
BBDR Models
CEBS
CNS
CUT
DBFs
DOE
EDCs
EDSTAC
ERDEM
FQPA
HPG/T Axis
HTPS
IT
JGI
LC
LD
MENTOR
MOA
MOU
NCEA
NCER
NCT
NERL
NHEERL
NIEHS
NMR
NRMRL
OEI
ORD
PBPK Models
PC
PD
PFOS
PK
QSAR
SAB
SAR
SHEDS
SNP
STAR
Androgen Receptor
Biologically Based, Dose-Response Models
Chemical Effects in Biological System
Central Nervous System
CUT Centers for Health Research
Disinfectant By-Products
Department of Energy
Endocrine Disrupting Chemicals
Endocrine Disrupter Screening and Testing Advisory Committee
Exposure Related Dose Estimating Model
Food Quality Protection Act
Hypothalamic-Pituitary-Gonadal/Thyroid Axis
High Throughput Screening
Information Technology
Joint Genome Institute
Lethal Concentration
Lethal Dose
Modeling Environment for Total Risk Studies
Mode or Mechanism of Action
Memo of Understanding
National Center for Environmental Assessment
National Center for Environmental Research
National Center for Toxicogenomics
National Exposure Research Laboratory
National Health and Environmental Effects Research Laboratory
National Institute of Environmental Health Sciences
Nuclear Magnetic Resonance
National Risk Management Research Laboratory
Office of Environmental Information
Office of Research and Development
Physiologically Based, Pharmacokinetic Models
Personal Computer
Pharmacodynamic
Perfluorooctane Sulfonate
Pharmacokinetic
Quantitative Structure Activity Relationships
Science Advisory Board
Structure Activity Relationship
Stochastic Human Exposure and Dose Simulation Model
Single Nucleotide Polymorphism
Science to Achieve Results
-------
EXECUTIVE SUMMARY
The mission of the U.S. Environmental Protection Agency (the Agency) is to safeguard
public health and the environment from adverse effects that may be caused by exposure to
pollutants in the air, water, soil, and food. Protecting human health and the environment carries
with it the challenge of assessing possible hazardous effects for tens of thousands of chemicals.
The large number of chemicals that the Agency must consider under many different regulations,
together with the large cost of conducting test batteries, limits the full use of standard toxicity
test methods to only a small number of chemicals. The Agency is also faced with reducing
uncertainties associated with performing quantitative risk assessments on chemicals for which
data have been submitted by the chemical industry.
Over the last several years, there have been increased opportunities to utilize novel
technologies derived from computational chemistry, molecular biology, and systems biology in
toxicological risk assessment. These new areas have been referred to collectively as "Computa-
tional Toxicology," which is denned in this document as the application of mathematical and
computer models and molecular biological approaches to improve the Agency's prioritization of
data requirements and risk assessments. This document describes a framework for the develop-
ment of a research program within the Agency's Office of Research and Development (ORD)
to utilize computational toxicology to address the questions of "when and how" to test specific
chemicals for hazard identification and to improve quantitative dose-response assessment.
In assessing risk associated with exposure to a chemical or other environmental stressor,
there are a number of uncertainties associated with detecting and quantifying the presence of
the chemical in the environment, the uptake and distribution of the chemical in the organism or
human or environment, the presence of the active chemical at a systemic target site, and under-
standing the series of biological events that lead to the manifestation of an adverse outcome.
The overall goal of ORD's Computational Toxicology Research Program is to use emerging
technologies to improve quantitative risk assessment by reducing uncertainties in this source-to-
outcome continuum. The three strategic objectives of the Computational Toxicology Research
Program are (1) to improve linkage across the source-to-outcome continuum, (2) to develop
approaches for prioritizing chemicals for subsequent screening and testing, and (3) to produce
better methods and predictive models for quantitative risk assessment. The tools developed
in the first objective will be critical for the research conducted in the remaining two
objectives. The use of computational toxicological approaches is discussed for a number of links
along the source-to-outcome continuum, including chemical transformation and metabolism,
better exposure indicators, improved dose metrics, characterization of toxicity pathways,
metabonomics, system biological approaches, modeling frameworks, and uncertainty analysis.
Computational toxicological approaches are also needed to develop better predictive models
for screening and testing including quantitative structure activity relationship (QSAR) models,
-------
improved pollution prevention strategies, and approaches to high throughput screening.
Computational toxicological approaches will also be used to address a number of research needs
associated with dose-response assessment, cross-species extrapolation, and the assessment of the
effects of chemical mixtures.
The research program at ORD currently uses many computational and biological
approaches that fall under the general area of computational toxicology, and examples of such
work are described in this document. Other research agencies such as the National Institute
of Environmental Health Sciences and the Department of Energy have significantly greater
capabilities for research on computational toxicology than the Agency. ORD has initiated
discussions with these agencies in order to facilitate the development of a national approach to
the use of computational procedures in toxicology.
This document is intended to identify the research needs of the Agency and the unique
capabilities of ORD laboratories to provide the basis for a more focused and integrated research
program in the future. To accomplish this, ORD has consulted the Agency's Science Advisory
Board (SAB) on this framework and held a workshop with scientists from across ORD to discuss
the content and intent of this document. Based on comments from the SAB and the workshop,
ORD will develop an implementation plan to guide research on computational toxicology over
the next 5-10 years.
-------
ODD
JOo
JOn
-------
-------
i. INTRODUCTION
A. THE COMPUTATIONAL TOXICOLOGY RESEARCH PROGRAM
Computational Toxicology is the application of
mathematical and computer models and molecular
biological approaches to improve the Agency's
prioritization of data requirements and risk
assessments.
The overall objective of this document
is to describe a framework for the
development of a Computational
Toxicology Research Program by the
Environmental Protection Agency's (the
Agency) Office of Research and
Development (ORD). Computational
toxicology involves the application of various mathematical and computer models to predict
effects and understand the cascade of events (sometimes referred to as mode or mechanism of
action) that result in an adverse response.
The Computational Toxicology Research Program is a technology-based, hypothesis-driven
effort to increase the soundness of risk assessment decisions within the Agency. It is designed to
increase the capacity
to prioritize, screen,
and evaluate chemicals
by enhancing the
ability of the Agency
to predict chemicals'
toxicities. Success
will be measured by
the ability to improve
risk assessments by
understanding the
potential of chemicals
to affect molecular and
biochemical pathways
of concern, i.e., their
toxicity pathways.
Computational Toxicology Involves:
Computational chemistry, which refers to the physical-chemical mathematical
modeling at the molecular level and includes such topics as quantum chemistry.
force fields, molecular mechanics, molecular simulations, molecular modeling,
molecular design, and cheminformatics;
Molecular biology, which allows for the characterization of genetic constituency
and the application of wide coverage technologies such as gcnomics, protcomics,
and melabonomics to provide the key indicators of cellular and organismal response
to stressor input;
Computational biology or bioinformalics, which involves the development of
molecular biology databases and the analysis of the data; and
Systems, biology, which refers to the application of mathematical modeling and
reasoning to the understanding of biological systems and the explanation of
biological phenomena.
In the area of computational biology, recent advances in "omic" technologies make this
a particularly appropriate time for such a program. Current research in this area focuses
on sequencing whole genomes and understanding the complexity of cellular biology at the
molecular level. The development of "omic" technologies has evolved into three scientific
disciplines: genomics. which is defined as the study of genes and their function; proteomics.
which is defined as the study of the full set of proteins encoded by a genome; and metabonomics.
which is defined as the study of the total metabolite pool. The recent technological advances
in these areas have led to the development of the field of toxicogenomics in which the effects
that chemicals have on living organisms and/or the environment can be examined using genomic,
proteomic, and metabonomic methods. Although the technology continues to change and
improve, conducting these types of analyses is no longer a question of capability. The Agency
-------
has traditionally used the term "mode of action" to refer to the key events and processes that
lead to an adverse outcome and the term "mechanism of action" to refer to a more detailed
understanding and description of events than is meant by mode of action. The potential provided
by "omic" technologies is that molecular profiling at multiple levels of biological organization
will lead to a depth of understanding that was not possible in the past. This document uses
the term "toxicity pathway" to denote this deeper level of understanding which toxicogenomics
may supply.
Parallel to efforts in computational biology, there have been major advances in computa-
tional speed and access to data. Less than a decade ago, describing the complexity of
chemical behavior in biological systems was severely limited because realistic models presented
combinatorial and other problems beyond the capabilities of most computers. In the field
of bioinformatics. for example, major advances were made not from faster statistical analysis
of data after its acquisition, but from the integration of computational and data acquisition
technologies. It is now possible to consider how to evaluate the vast amounts of information
generated by "omic" technologies using data-mining tools made possible by rapid advances in
computational storage capacity and speed.
One area where computational toxicology has shown promise is in the discipline of
physical organic chemistry known as Quantitative Structure Activity Relationships (QSAR).
Application of QSAR has resulted in the development of novel predictive capabilities for
representing chemical structures as a distribution of conformations and properties rather than
discrete structures. Another promising area brought about by the joining of computer science,
biology and medical programs is an emerging discipline known as systems biology. Systems
biology has the potential to lead to the development of virtual biological systems via the
development of computational models of a cell, organ, or an organism's function based upon an
understanding of the component parts.
In developing this document and its subsequent implementation, it is recognized that the
various technologies and tools that form the current state-of-the-science are in varying stages of
maturity. How quickly the technologies are established and validated will affect the development
of critical paths to solving particular problems and the timeframe in which these solutions are
put into place.
B. APPLICATION OF COMPUTATIONAL TOXICOLOGY TO RISK ASSESSMENT
AND RESEARCH
ORD's research programs support the Agency's regulatory decision making by providing
scientific information for human health and ecological risk assessment. Risk assessment is the
process used to evaluate the potential hazards of and exposures to environmental stressors to
produce estimates of the probability that populations or individuals will be harmed by chemical
exposure and to what degree. It is one component of the process by which the Agency and many
other organizations recognize a potential risk and decide how to respond.
The Agency's risk assessments of chemicals rely primarily on laboratory testing on a
chemical-by-chemical basis to obtain data about adverse effects and the quantitative relationship
between dose level and likelihood of response. In human health risk assessment, these laboratory
data are extrapolated to humans to estimate human risk. The large number of chemicals in
commerce, coupled with the expense of laboratory testing, limits the application of extensive
standard toxicity testing to relatively few chemicals. ORD will explore the feasibility of
-------
using computational approaches to improve quantitative dose-response assessment and the
development of sensitive and specific tests for hazard identification. Preliminary efforts in this
area began in FY02 with a congressional reprogramming action which directed the Agency
to explore the use of alternative methods to animal testing for hazard identification. ORD
interpreted this as an opportunity to evaluate genomic and computational tools for screening
purposes, and it initiated several research projects on endocrine disrupting chemicals (EDCs)
as a "proof-of-concept" effort. Endocrine disrupters were selected because it was felt that a
considerable amount of knowledge concerning mechanisms of actions and toxicity pathways
existed for this class of environmental pollutants. This feasibility effort is described in greater
detail in HLA.
Environmental
Release
Fate/Transport
Models/Data
Environmental
Concentration
Exposure
Models/Data
Exposure
Concentrations
PBPK
Models/Data
Target Organ
Dose
BBDR
Models/Data
Early Biological
Effects
Systems
—NModels/Data
Figure 1 The Source-to-Outcome Continuum
Adverse Outcome
C. OVERALL GOAL AND STRATEGIC OBJECTIVES
It is useful to envision the risk assessment paradigm as a continuum of events
leading from release in the environment to adverse effect. Figure lisa simplification
of this concept showing points
along the continuum where a
measurement or an observation
can be made. The arrows
between the boxes represent a
cascade of events that lead from
one measurable event to the
next. ORD's research program
focuses on learning more about the processes that lead from exposure to adverse outcome in
order to allow the Agency to perform better risk assessments.
Objectives of the Computational Toxicology
Research Program
Improve Linkages in Source-to-Outcome Continuum
Provide Predictive Models for Hazard Identification
Enhance Quantitative Risk Assessment
-------
The overall goal of ORD's Computational Toxicology Research Program is to
use the tools of modern chemistry, biology, and computing to provide the Agency with
approaches to improve quantitative risk assessments and to reduce the uncertainties in the source-
to-outcome continuum. To meet this goal, ORD has identified three strategic objectives for the
Computational Toxicology Research Program. First, research is needed to develop improved
linkages across the source-to-outcome continuum. Understanding those linkages will decrease
uncertainties in assessing risk to human health and the environment. The tools developed in the
first objective will be important for the second objective which is to conduct research to develop
strategies for prioritizing chemicals for subsequent screening and testing. The current approach
for screening and testing chemicals requires extensive resources. Therefore, an approach must be
developed to determine which chemicals or classes of chemicals should be screened and tested
first. Finally, research is needed to develop better methods and predictive models for quantitative
risk assessment because current approaches take too long and are too costly. Benefits of this
program include the identification of molecular indicators of exposure and toxicity that can
be applied to other areas such as epidemiology, the harmonization of cancer and non-cancer
risk assessments, and the integration of human and ecological risk assessments. The following
sections of this document describe how ORD is currently using, and how it proposes to use,
emerging technologies associated with computational toxicology to address the Agency's needs
for approaches to screen and test more efficiently and to improve quantitative risk assessment.
-------
-------
SEARCH
-------
ii. RESEARCH NEEDS
AND APPLICATIONS OF COMPUTATIONAL TOXICOLOGY TO GOALS
A IMPROVE LINKAGES IN THE SOURCE-TO-OUTCOME PARADIGM
1. CHEMICAL TRANSFORMATION AND METABOLISM
At several points along the source-to-outcome continuum, it is critical to accurately
model the fate of chemical stressors to determine the level of exposure to an organism. It is
also crucial to accurately model the metabolism of a chemical inside the target organism because
it is often a metabolite of the original stressor that induces a biological event. In many cases,
the reaction processes controlling the fate of chemicals outside of the organism are similar if not
identical to those reaction processes controlling metabolism within the organism. For example,
enzyme-mediated processes such as redox reactions and hydrolysis that result in the formation
of reactive intermediates (i.e., radicals, carbenes) that react irreversibly with biological receptors
(e.g., DNA) are often the rate-determining processes controlling the fate of these chemicals in
natural aquatic ecosystems. Consequently, the process of developing and refining simulators for
environmental transformation and metabolism will have many commonalities with understanding
metabolism within organisms.
A CHEMICAL TRANSFORMATION IN ECOSYSTEMS
The state of chemical fate and transport modeling for exposure assessment has advanced
significantly in recent years. For example, it is now possible to forecast many of the
physicochemical properties that ultimately govern chemical transformation. Nonetheless,
many unknowns and uncertainties
remain, and ORE) continues to conduct
research aimed at reducing them.
There are several key areas of
uncertainty, however, that can be
reduced greatly by informing and
validating fate models with molecular
indicators of exposure. The rapid
advances in "exposure genomics" (see Section II.A.2) will provide early signs of chemical
exposure based on changes in gene expression, which will lead to the development of a new array
of molecular indicators that can guide chemical fate and metabolism studies. The integration
of genomics and molecular indicators into chemical fate studies may improve linkages in the
source-to-outcome continuum.
Chemical Fate Models
Determine minimal concentrations at which biological events occur
Identify biologically relevant chemical(s) in mixtures
Identify crucial biotransformations in the environment
The application of molecular indicators to chemical fate studies is several-fold. For
chemicals that trigger biological events of concern, molecular indicators can be used to determine
the minimal concentration at which biological events occur. This approach will narrow the task
of the exposure models to answering only the question of whether the toxicant (i.e., parent or
reaction product) is above or below this minimal concentration (i.e., it will "bound" the model).
Narrowing the model requirements can significantly increase certainty in the risk assessment
process. Another area to address is the elucidation of the biologically relevant components
-------
in chemical mixtures by measuring changes in gene expression in exposed organisms. This
application of molecular indicators can focus exposure models on a much smaller subset of
candidate chemicals, including those potentially linked to initiation of adverse effects. Finally,
molecular exposure indicators can be used to advance our understanding and ability to model
biotransformations of chemicals in ecosystems. Biotransformation is widely recognized as
the largest uncertainty in exposure modeling, and accurate models to predict and describe
biotransformation have eluded scientists because the universe of enzyme reactions is so large.
Gene expression tools will be used to narrow down this universe only to those that are
biologically relevant, and this will enable more accurate and meaningful prediction of significant
chemical transformations.
B. CHEMICAL METABOLISM
Many toxic effects result from metabolic activation of parent chemicals that forms
metabolites which are much more toxic than the parent. Moreover, many cross-species
differences in toxic effects are the result of differences in detoxification. Consequently, an
accurate computerized simulator of metabolism in the liver and other target tissues (e.g., kidney)
is essential to meet the objectives of the program. The primary goal of this research is the
development of a computational system that will predict and prioritize metabolic pathways for
liver metabolism.
Metabolic Simulator
Build libraries of relevant metabolic transformation
pathways
Develop high quality data metabolic maps
Provide probability indices for substructure! units
biotransformations
The first step in the development of
a metabolic simulator is to create a library
of all known metabolic transformations
which are nested according to the
substructural elements being transformed.
Algorithms will then be used to recognize
the relevant substructural units in a
chemical of concern that can undergo metabolism. The transformation products from each
possible reaction are then stored as a list of first-level metabolites. Each metabolite, in turn,
is subsequently submitted to the substructural matching routine in order to generate a set of
second-level metabolites from each first-level metabolite. The process is continued until the
metabolic map is completed.
This approach to simulating metabolism tends to identify many metabolic candidates that
are ultimately improbable because of kinetic considerations. This problem can be overcome
by associating a probability with each substructural transformation process in the library. Such
transformation probabilities can be derived statistically from a library of high quality metabolic
maps. Unfortunately, currently available metabolism libraries have significant gaps in relative
rates for many important metabolic reactions. Identification of these gaps will direct the
generation of new high quality data on metabolism. These data will be generated using
traditional experimental methods and new advanced analytical techniques [e.g., wide-bore, high-
resolution nuclear magnetic resonance (NMR)] for measuring metabolic rate constants and
identifying metabolites in vivo and in vitro. Mechanism-based predictive methods can also be
used to fill some important data gaps.
2. EXPOSURE INDICATORS
Exposure assessment has historically been based on the use of chemical analysis data
to generate exposure models. While the biological activity of chemicals has been recognized
-------
as important for exposure assessments, the measurement of such activity has largely been
limited to whole organism toxicity tests. Considerably less research has been done using
in vitro tests to assess specific types of biological activity present in whole or fractionated
environmental samples. Current in vitro capabilities are not sufficiently validated to address
exposure assessment.
There is a need to develop cellular and molecular indicators of exposure that can be
used to assess the vulnerability of humans and wildlife to single and multiple pathways of
exposure to chemicals in the environment. However, correlation of such indicators will require
a greater understanding of the
linkage between the cellular
and molecular indicators with
specific cellular and tissue-level
effects (e.g., reproductive or
neurologic toxicity) or
outcomes (e.g., fertility or
neurological disease).
Exposure Indicators
Few environmental stressors have Specific or sensitive indicators
Exposure indicators are poorly correlated with effects
Molecular indicators could validate fate and transformation models
Crucial for mixtures risk assessment
Essential for integrated approach to risk assessment
The indeterminate condition of exposure indicator research stands to change remarkably
with attempts to link molecular biological technologies with cellular or tissue effects and
outcomes. The Computational Toxicology Research Program aims to develop a platform
or sequence of approaches through which "the earliest recognizable signatures of exposure"
(i.e., unique patterns of up- and down-regulation of genes) can be identified for scores of
different stressors; become user-friendly procedures; are demonstrated in case studies; and are
incorporated into Agency, State, and Regional studies supported by the Agency's Environmental
Monitoring and Assessment Program and other programs. Acting on the tenet that "any
response to" or "effect from" a stressor will involve changes in the expression of some genes,
it is hypothesized that gene discovery and DNA microarray synthesis and use will provide a
window on hundreds of changes that may or may not be linked to downstream cascades of
activity responsible for adverse effects. Bioinformatic tools will be used to discriminate unique
signatures and families of signatures indicative of stressors or groups of stressors. The scope
of the Computational Toxicology Research Program is moving past the use of few genes in
an organism, such as the ecotoxicology model fathead minnow (Pimephalespromelas), to the
use of hundreds of genes and gene homologues acquired by less direct alternate molecular
methods. Ultimately, the scope of the approach will move to the level of 25,000 to 30,000
genes associated with the complete genome of selected organisms. The existence of hundreds
of signal transduction pathways in cells of higher organisms, which will be elucidated through
traditional biochemistry studies, will heighten the likelihood of unique exposure signatures for a
great number of individual chemical stressors and families of stressors and will provide tools for
future exposure studies. These studies will be both adverse effect-driven (initially, development
of molecular indicators for early molecular events in sex steroid-mediated mechanisms of
toxicity) and empirically based (clusters of activity identified from watershed or regional stream
surveys). The development of molecular diagnostic indicators of exposure will present the
opportunity for the simultaneous, near real-time measurement of biologically relevant exposures
of organisms to multiple stressors in mixtures. Nanotechnological instrumentation and robotics
offer the promise of extremely high throughput analysis of indicators that can allow for larger-
scale exposure studies at the watershed and regional levels to be undertaken. Once discriminated,
these molecular events can potentially be linked to toxicity pathways as described in Section
II.A.4.
-------
In addition to chemicals, humans and wildlife are exposed to other environmental
stressors such as microorganisms. Genomic technologies offer unique opportunities to
discriminate changes that occur within cells that define a particular microorganism's
pathogenicity and to develop strategies for microbial source tracking. Exposure to
microorganisms is also likely to be a factor in characterizing exposure of humans and wildlife
to other stressors. Prediction of the outcome of chemical exposures stands to be significantly
enhanced by an understanding of changes in cellular responses contributed to by both biotic and
abiotic influences. This information will be especially important for the development of criteria
for drinking water.
3. DOSE METRICS
Dose Metrics
Dose is often inferred from stressor uptake
Dose models stand to be enhanced with specific data on stressor
interactions with molecules initiating toxicity pathways
Genetic polymorphism data will reduce uncertainly stemming tram
assumptions of homogeneous populations
Susceptibility indicators will be developed for input into exposure
models
Qualitative and quantitative evaluations of the relationship between dose and response
are key components of the quantitative risk assessment process. ORD has developed several
modeling systems to assess exposure and dose. The Stochastic Human and Dose Simulation
Model (SHEDS), for example, supports efforts to better understand exposure to multimedia and
multi-pathway pollutants. The Exposure Dose Estimating Model (ERDEM) is a physiologically
based pharmacokinetic (PK) modeling system that estimates relevant toxic doses within the
body after actual and realistically simulated exposures. ERDEM can be applied to single or
multiple chemical exposure scenarios.
ORD also supports the development of
the Modeling Environment for Total
Risk Studies (MENTOR) with one
of its university partners. MENTOR
uses models that help to quantitatively
account for pollutants along the entire
source-to-outcome continuum. With
new information and data from
research on biological indicators, these
models will be significantly enhanced. Increased knowledge of the biology of potentially toxic
processes will enable ORD to enhance these models so it can better assess the effects of
various and detailed exposure conditions such as co-exposure, varied patterns of exposure, and
intermittent exposure.
The choice of the chemical species and the actual dose metric used for the risk assessment
process depends upon the particular mode or modes of action being assessed. Because the
biological steps between the external exposure and various internal toxicologically relevant doses
are often non-linear, PK models are often applied to estimate the effect of an exposure of interest
to an observed adverse outcome. PBPK models depend on knowledge of anatomy, physiology,
and biochemistry. Thus, in order for PBPK models to be used effectively, several pieces of
key information are needed. Some of this information (e.g., body size, organ volumes, and blood
flows) is known for several species including the human. Other important pieces of information,
such as metabolic transformation rates, are chemical specific and may vary from species to
species. The expense and time required to gain this information in laboratory studies has limited
the use of this modeling technology, as has the lack of detailed knowledge concerning the
molecular events that lead to toxicity (i.e., the toxicity pathway). The Computational Toxicology
Research Program should enable broader use of PBPK models by overcoming these limitations
by providing better indicators of the relevant doses and receptors within the target organism.
-------
Through application of modern molecular tools, this research program will provide better
definition of the most relevant dose metric for chemicals entering the body. For example, specific
binding to a particular part of the DNA, RNA, receptors, or enzymes might be a much more
relevant dose metric than simply the amount of chemical in a particular tissue. In addition, it
might be possible to identify related biomarkers in easily obtained biological fluids.
It is also expected that developments in the science of genomics will greatly help define
and characterize sensitive subpopulations, such as children and older individuals. It has long
been recognized that not all members of the population are equally sensitive to the same
environmental pollutant. While not all the factors leading to susceptibility or resistance are
understood, advances in genomics and clinical medicine are showing that, for many adverse
processes, a key component of the genome increases an individual's vulnerability to a specific
clinical disease. Additionally, the field of metabonomics offers new opportunities to characterize
variation in metabolic processes at the cellular and tissue levels. Integration of studies of
stressor dose, transformation, and metabolic sequelae have the potential to provide the clearest
perspective yet on relevant dose and its variation across organisms and populations.
Advances in molecular biology enable the characterization of the genetic variation
(polymorphisms) within populations. Much of the current activity in molecular biology has been
directed at identifying single nucleotide polymorphisms (SNPs) of key xenobiotic metabolizing
enzymes. Once the metabolic significance of these SNPs is understood, they can be readily
factored into PK models. In turn, these polymorphisms are beginning to be used to define the
contribution of genetic variation to the overall level of variation in dose-response in populations.
Ultimately, integrated research in genetics and genomics has the potential to elucidate specific
altered molecular processes associated with genotypes representative of sensitive or vulnerable
subpopulations. Dosimetry models such as PBPK and pharmacodynamic (PD) models can
incorporate these data to reduce the uncertainties associated with assuming populations are
homogeneous regarding toxic response to stressors. Indicators of susceptibility obtainable
from body fluids may also be developed to provide simple methods to characterize population
composition, thus refining the exposure characterization for human and ecological risk
assessments.
In summary, technologies developed and applied as a result of this program have the
potential to better define lexicologically relevant doses. Approaches developed by this program
will also help provide the information needed to develop mechanistically based quantitative
models to estimate the relevant doses and to realistically assess their impact.
4. CHARACTERIZATION OF ToxicITY PATHWAYS
Computational toxicology techniques offer the potential to reduce uncertainties in both
ecological and human health-risk assessment. However, use of key predictive toxicology
tools/approaches, including PBPK and QSAR models and/or alterations in gene (or protein)
expression profiles, is useful only in the context of a thorough understanding of toxicity pathways
of concern (i.e., the mechanism or mode of action). Specifically, for these types of predictive
methods to be useful, it is necessary to link adverse outcomes (e.g., reproductive or develop-
mental changes or cancer) to initiating events, ideally through a cascade of biochemical and
physiological changes that occur as a result of the initial interaction(s) of xenobiotics with
biological molecules (e.g., receptor binding, enzyme inhibition). A particularly key aspect of this
linkage is identification of the proximal (often initial) biological alteration associated with any
particular toxicity pathway. For example, chemicals which bind to and activate specific nuclear
-------
Understanding Toxicity Pathways
Identification of discrete molecular initiating events
Linking adverse outcomes to molecular alterations
Elucidating linkages across biological levels of organization
Biological basis for cross-species extrapolation
Prediction of possible interactions for untested chemicals and mixtures
receptors elicit a relatively predictable suite of biochemical and physiological responses that
are species/class-specific but culminate in veiy similar adverse reproductive and developmental
effects across numerous vertebrate species. Identification of common initiating events, such
as receptor activation, can enable the successful use of models or gene expression assays
to deal with xenobiotics as classes of compounds rather than individual chemicals. Further,
through understanding the cascade of events that occur as a result of receptor activation, in
conjunction with accurate dosimetry predictions, it will become possible to predict adverse
outcomes associated with exposure to, as yet, untested chemicals. For this to be feasible, an
understanding of toxicity pathways based on discrete initiating events is needed.
Definition of toxicity pathways associated with discrete initiating events has a variety of
direct benefits and implications germane to the risk assessment process. For example, the ability
to associate endpoints to one another through a continuum of biological organization (i.e., across
molecular, cellular, target organ, and apical endpoints) would be powerful, both for prospective
and for diagnostic risk assessments. In the former case, it would be possible to better link
responses at intermediate biological levels of organization to both the initiating event and
the adverse outcome. In this way, it is expected that the biological effects of molecular
level changes will be understood
and, therefore, their use in risk
assessment facilitated. In the
case of diagnostic assessments,
delineation of toxicity pathways
would contribute directly to an
understanding of the toxicological
significance of alterations in
markers of exposure based on
changes in gene expression. In addition, as the key initiating events are identified, polymorph-
isms in the genes can be identified; and this information will provide insight into individual
susceptibility on a dynamic level, just as how understanding SNPs in biotransformation genes
provide insight into susceptibility at the kinetic level. From another perspective, knowledge of
key initiating events relative to alterations in endpoints at higher levels of organization could
enable a direct assessment of the technical validity of using mixture models based on similar
versus dissimilar initiating events. In addition, identification of these events via alterations in
gene expression could help in species extrapolation. Demonstration that toxic initiating events
are similar across species would reduce uncertainty associated with extrapolation across species
as knowledge of common initiating events for a chemical or class of chemicals would focus the
challenge of extrapolation across species on comparative dosimetry. An underlying assumption
of this approach is that there is a relatively finite number of key initiating events and that
these can be understood and characterized using wide coverage molecular biological techniques
combined with bioinformatic processing. One must always be aware, however, that species
can display unique responses to the same perturbation; and while the initiating event may be
identical across species, responses can diverge significantly such that different genes and tissues
are affected in different species. Thus, the research focus may be on concordance of the initiating
event rather than on the effect or site of response.
Approaches used in computational toxicology will significantly improve our ability
to understand and predict how xenobiotics can interact with biological systems. Figure 2
illustrates components of a toxicity pathway for some commonly used adverse outcomes in
human health and environmental risk assessment. This schematic demonstrates the linkages
between biologically effective concentrations of a chemical at a receptor/1 igand site that lead to
-------
cellular and organ responses associated with an adverse effect at the individual level. These
approaches can then be used for identifying and utilizing profiles of gene expression linked to
cellular alterations and adverse effects or outcome.
Linking Observations Across Levels of Biological Organization
Interaction
DNA Binding
3rotein Oxidation
In vitro Assays
In vivo Assays
-Gene Activation
-Protein
Production
-Altered Sgnaling
-Protein Depletion
-Altered Physiology
-Disrupted
Homeostasis
-Altered Tissue
Development
or Function
-Lethality
-Impaired
Development
-Impaired
Reproduction
•Cancer
Figure 2 An Example of a Toxicity Pathway
5. METABONOMICS
Genomics and proteomics allow for the measurement of response to chemicals on the
genetic and cellular protein level, respectively; however, neither provide a complete description
of metabolism and chemical toxicity. For example, in some instances a xenobiotic may elicit
changes in gene and protein expression that are compensated for elsewhere and result in no net
change to the organism (i.e., no change in endogenous metabolite profile). To fully understand
xenobiotic metabolism and toxicity in the context of genomics and proteomics, it is crucial to
understand the metabolic status of the whole organism. The use of metabonomics, the multi-
parametric measurement of metabolites in living systems due to physiological stimuli or genetic
modification, provides such a means by augmenting and complementing genomic and proteomic
responses to xenobiotic exposure and by
providing a connection between genomics
and proteomics with tissue function. The
ability to conduct metabonomic studies
depends on the application of advanced
analytical techniques such as high-
resolution NMR spectroscopy and multi-
variable statistical programs. ORD is in the ™^^^^^^^^^^^^^^^^^^^^^^^^^^~
process of purchasing a wide-bore 600 MHZ NMR for metabonomic analysis in support of the
Computational Toxicology Research Program.
Metabonomics
Elucidate changes in chemical-induced metabolic
patterns for range of endogenous metabolites
Generate NMR spectral profiles for chemicals
Build models to evaluate effect of novel chemicals on
endogenous metabolites
-------
The application of metabonomics to toxicity testing involves the elucidation of changes
in metabolic patterns associated with chemical toxicity based on the measurement of component
profiles in biofluids (i.e., urine), cells, or tissues, and enables the generation of spectral profiles
for a wide range of metabolites. NMR pattern-recognition technology associates target organ
toxicity with NMR spectral patterns and enables the generation of spectral profiles for a wide
range of endogenous metabolites. Metabolite profiles (e.g., endogenous metabolites such as
creatine, lactate, glutathione) could provide a measure of the real outcome of potential changes
as the result of xenobiotic exposure.
The application of metabonomics can provide mechanistic information that could have
significant implications for the risk assessment process. Assuming that groups of compounds
induce similar changes in gene, protein, and metabolite profiles, it should be possible to classify
compounds based upon their profiles. Assessment of the profiles should also help scientists
understand the key initiating events that are involved in activating critical toxicity pathways.
By including a large number of compounds with known toxicity in databases, ORD can build
predictive models for comparing and evaluating expression profiles of novel compounds. For
example, association of a given toxic endpoint with a characteristic shift in cellular metabolites
could provide a fingerprint that is characteristic of a specific mechanism of toxicity. Once
diagnostic fingerprints are defined for different mechanisms, the metabolite pattern for a toxic
chemical having an unknown mechanism could be compared to the database. This could provide
a very powerful tool for categorizing toxicants according to a toxicity pathway.
6. SYSTEMS BIOLOGY
Conventional molecular biology strives to examine key events at increasingly finer levels
of detail. The Computational Toxicology Research Program, combined with the work being
conducted by a number of outside organizations, will provide a wealth of information on the
effects of toxicants by using genomic, proteomic, and metabonomic techniques. In order to be
most useful, this information must be integrated into a coherent picture. Systems biology is a
new field of science that uses computational methods to reconstruct an integrated physiologic and
biochemical model of an organism's or cell's biology. The approach of systems biology is similar
to developing a wiring diagram for a complicated electrical system or an engineering diagram
of how a vehicle is put together and how the
different parts interact and function together.
Analogously, systems biology is targeted at
studying how normal biological processes are
governed and how alterations can lead to
diseases or other unwanted outcomes.
Systems Biology
Computational models that reconstruct a cell, organ or
organism's function from component parts
Allows validation and simulator experiments that build
confidence in predictive ability of adverse effects
Understanding how a normal cell or organism works is key to understanding how
toxicants can cause changes in the cell. For example, in developing a biologically based dose
response (BBDR) model for the developmental effects of 5-fluorouracil, ORD researchers were
able to describe the effect of this chemotherapeutic on thymidylate synthetase activity (its target
enzyme), on subsequent nucleotide pool perturbations, on alterations in cell cycle times, and
ultimately on the size of the fore-limb bud. However, the investigators could not describe why
the fourth digit was the most affected because not enough was known about the normal biology
of limb development to understand how the preceding events altered the developmental program-
ming. In this example, a systems biology approach is necessary to understand the underlying
biology of limb development in order to better understand how the outcome of concern actually
resulted from the precedent biochemical and cellular events.
-------
A systems biology approach will enable the integration of disparate data developed by
biologists, computer scientists, chemists, engineers, mathematicians, and physicists to construct
models of organismal function and how organisms respond to a toxic insult. A choice has to
be made about the scale and level of detail for each systems biology model, and it is likely that
models useful to the Agency will be built by integrating individual subcomponents into a larger
system. Once these models are developed, hypotheses can be developed and tested through
virtual simulations prior to designing targeted experiments to validate and inform the models.
An integral part of the Computational Toxicology Research Program will be the use of relevant
model organisms to expand our understanding of the regulation of biological processes and
how toxicants can perturb these processes. In particular, cell signaling systems are receiving
considerable attention in systems biology, and this might be a promising approach. An initial
step to designing ORD's systems biology efforts might be to organize a workshop to help identify
promising areas for research and development. To supplement its intramural efforts, ORD
expects that the extramural grants [Science to Achieve Results (STAR)] program will be able to
make contributions in filling the research needs presented by systems biology.
7. MODELING FRAMEWORKS AND UNCERTAINTY ANALYSIS
The Computational Toxicology Research Program requires a modeling framework to
develop a functional tool for prioritizing chemicals for subsequent screening/testing and enhanc-
ing quantitative risk assessment. Modeling frameworks are the software infrastructure required
to facilitate modern environmental modeling. Modeling solutions to regulatory-based assessment
needs require the development and application of science-based models and databases that span
the source-to-outcome continuum. Modeling frameworks contain and manage the coordinated
execution and data exchange of numerous science-based models. They also facilitate access to
external data sources, model output data analysis, and user interfaces.
Establishment of a modeling framework (based on existing and proven technologies) that
will standardize the format and interchange protocols for all information generated
via computer simulation for the Computational Toxicology Research Program will
be a critical undertaking. The
technology will contain (or access via
Internet) models for simulating the
environmental fate and transport of
chemicals (transformation
simulators); human and ecological
exposure; the fate and transport
of chemicals within human and
ecological receptors (metabolic
simulators and PBPK); toxicity
pathways (QSARs); and adverse "^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^™
outcomes (systems biology models). In support of this modeling, the technology will include
database-connectivity tools for linkage to databases unique to the research program such as
sequence databases, libraries of metabolic and toxic pathways, metabonomic profiles, and bio-
and chemo-informatics.
Modeling Frameworks and Uncertainty Analysis
Requires science-based models and databases
Standardize format and interchange protocols for information
generated by computer simulation
Develop technology for linking required databases
Develop uncertainty analysis methods
Operating system managing numerous models
Facilitate access to other databases and user interfaces
In addition to providing the necessary modeling technology, ORD will also need to target
the development of uncertainty analysis methods. As the scope of science needed to answer
the broad questions posed by modern regulatory initiatives expands dramatically, our ability to
quantify the accuracy of our model-based estimates decreases. Consequently, a major focus of
-------
emerging scientific inquiry is related to the characterization and quantification of uncertainty
in "high order systems." Efforts will be undertaken to increase awareness of the full range of
potential health effects, especially non-cancer endpoints, and to fully characterize variability and
reduce overall uncertainty. A supercomputer (a cluster of 150 PCs) has already been configured
to facilitate the distributed computing (parallel processing) necessary to execute sophisticated
modeling simulations involving numerous models and databases. This hardware infrastructure
is specifically designed to support the development of a wide range of uncertainty analysis
methods.
B. PROVIDE PREDICTIVE MODELS FOR HAZARD IDENTIFICATION
As discussed previously, the Agency needs to develop predictive models for hazard
identification. This section describes three areas where computational approaches are being
considered or where such methods could prove fruitful.
1. QSAR. AND OTHER COMPUTATIONAL APPROACHES
Like its physico-chemical properties, the biological activities of a chemical are the result
of molecular interactions between the chemical and its immediate environment. When models
for specific molecular interactions are developed, the activity of chemicals with respect to those
interactions can be estimated directly from chemical structure. QSAR models could serve as
important tools to screen untested chemicals for their potential to interact with hundreds of
different environments using only the chemical structure and a virtual library of chemical and
toxicological models. As such, QSARs could be used to optimize laboratory testing when the
number of untested chemicals exceeds the resources available for testing. QSAR may also
be used to provide estimates of missing data in lower tier risk assessment. For example,
in the case of the initial screening
QSAR and Other Computational Approaches
Quantifying physico-chemical parameters to predict fate
Identification of potential hazard in absence of empirical data
Prioritizing large groups of chemical for later testing
Framework for optimized use of "omic" data
Estimate missing parameters for untested chemicals
of chemicals for their potential to
disrupt the endocrine system, models
of molecular interactions with critical
receptors and enzymes could be used
to develop a series of computational
methods to classify each chemical
based on its likelihood of binding to a
receptor or inhibiting a crucial enzyme. The intent of screening using QSAR is to offer a list
of chemicals most likely to test positive in standard toxicity screening assays. It is recognized
that these models will have to undergo extensive validation in terms of the descriptors associated
with structure, the chemical space that they are applicable to, and the quality of the data in
the training sets.
Another application of QSAR is to estimate the toxicity of untested chemicals directly
from chemical structure. Some chemical properties and reactions can be directly related to
structural parameters and, for some of these, QSAR has been used as a cost-effective surrogate
for routine laboratory experiments. For certain applications such as fate and effects modeling
which require expensive laboratory measurements of chemical properties, the structure-property
relationships are now sufficiently reliable that QSAR estimates rather than measured values are
widely used. QSAR can also be useful in estimating potency within a class of chemicals relative
to acute toxicity endpoints.
-------
Emerging "omics" technologies offer excellent potential to generate information that will
inform and improve the QSAR modeling process. Specifically, alterations in gene expression
can be used to identify toxicity pathways and associated key molecular initiating events. Once
initiating events are established for any given toxicity pathway/adverse outcome, predictive
modeling can form the basis for dealing with large numbers of chemicals in a relatively rapid
fashion. Specifically, there is little question that, given high-quality datasets, current QSAR
modeling methods can effectively predict discrete biological phenomenon at the molecular level
in terms of interactions of chemicals with different classes of lipids, nucleotides, or proteins.
Therefore, the key is defining the biological phenomenon for which data should be collected. If
molecular initiating events can be identified relative to specific adverse outcomes, appropriate in
vitro or in vivo assay systems (e.g., receptor binding assays) can be identified and developed to
serve as a basis for generating the data needed for robust models.
The relationship between the generation of genomics information and QSAR modeling
can be depicted in a linear fashion. It is a process, however, that is iterative in nature with
multiple potential points of "entry" in terms of genomics or modeling components. For example,
if the initiating event through which a chemical elicits adverse effects is completely unknown,
genomic approaches in which large numbers of expressed genes are assessed can be used in a
"discovery" mode. This information may effectively identify the unknown chemical as similar
to other previously tested chemicals for which there is an understanding of toxicity pathways
and associated initiating events. Alternatively, genomic information may serve as the basis for
defining previously unknown initiating events and may serve as the starting point for delineating
new or alternate toxicity pathway(s). As key initiating events are identified, appropriate data
"generation" assays can then be developed to provide data for QSAR and other computational
models capable of predicting either acute or chronic toxicity.
2. POLLUTION PREVENTION STRATEGIES
In support of pollution prevention strategies, ORE) is developing methods to estimate
the potential environmental impact of chemicals that are released into the environment. These
methods are used to evaluate chemicals for potential harm to both humans and the environment in
a life-cycle assessment framework. These chemical evaluations are performed over a wide range
of environmental impact categories, including human health (acute, chronic, and carcinogenic
indicators), aquatic health (acute and chronic indicators), terrestrial health, global warming,
ozone depletion, smog formation, acid rain, eutrophication, and natural resource depletion. To
this end, several pollution prevention tools have been developed for a variety of uses. Depending
on the level of analysis desired, the complexity of the model for evaluating human and ecological
health concerns will dictate the type of data
required. These data may be collected from
acute toxicity studies, detailed systems biology
models, fate and transport models, exposure ,
j i i • x • -X. x. j- r, ji release into environment
models, or chronic toxicity studies. Regardless f f^} Mic&t(>rs to compare large numbers of
of the level of sophistication in the models,
the final impact indicators (e.g., a broad range
of mid-point effects or final outcomes, such as
human deaths, human illnesses, crop damage, water quality issues, air quality issues) could be
used to compare a large number of chemicals. QSAR have also been developed by ORE) to
provide estimations for toxicity values for those chemicals which have no experimental toxicity
data. For example, QSAR can be used to estimate endpoints, such as the 96-hr LC5I1 values for
fathead minnows and the oral rat LD5n values. These estimated values can then be incorporated
Pollution Prevention Strategies
Methods to estimate potential impact after
release inu
Final indie
chemicals
-------
into various environmental models and used in pollution prevention strategies. Such tools will
greatly benefit from increased certainty in the fate and transport models (as discussed in Section
II. A.I.a), and they will be improved by a better understanding of metabolic maps of chemicals
(as discussed in Section II.A.l.b). These tools will also benefit from the enhancement of the
quantitative risk assessment process in areas such as increased knowledge of dose-response
assessments (as discussed in Section ELC.2) and cross-species extrapolations (as discussed in
SectionII.C.2).
3. HIGH THROUGHPUT SCREENING
Applications of new molecular and other technological advances hold promise for the
development of high throughput screens (HTPS). It has been suggested that HTPS be used as
a rapid, efficient means to provide preliminary endocrine-effects data on chemicals considered
in the endocrine disrupters screening and testing program. In view of the estimated 87,000
chemicals under consideration, it would be beneficial if rapid, HTPS systems could be developed
to assist in the prioritization of chemicals for further testing. Because all processes are automated
and can be programmed to run continuously in HTPS, large numbers of samples can be
screened in a relatively short period of time using this technology. New approaches have
the potential to make significant advances over
existing EDC screens in terms of speed, high-
throughput capability, sensitivity, reproducibility,
and reduction in animal usage in a screening
and testing program. HTPS will be a valuable
tool to help elucidate and characterize toxicity
pathways (see section HA.4). Approaches under
development could focus on classic ligand-steroid
receptor-coregulator/cofactor interactions; non-genomic mechanisms of steroid hormone action;
or mechanisms involving synthesis, metabolism or degradation of estrogens, androgens, and
thyroid hormones. Furthermore, some HTPS approaches may be flexible and versatile enough to
allow for screening to be carried out across vertebrate classes. This will help scientists address
cross-species extrapolation issues (see section II.C.2.b).
C. ENHANCE QUANTITATIVE RISK ASSESSMENT
Computational toxicology has the potential to enhance the Agency's current risk
assessment methods and to contribute to the development of new methods that are consistent
across endpoints and species. One aspect of the ORD program will be to develop broadly
applicable risk assessment methods that take advantage of new technologies and the data
generated by the Computational Toxicology Research Program.
1. APPLYING COMPUTATIONAL TOXICOLOGY IN QUANTITATIVE RISK
ASSESSMENT
High Throughput Screening
Vast chemical inventory not tested
Rapid, efficient means to provide preliminary data
Recommended for Endocrine Disrupters
ORD has ongoing research in two areas that can be related to risk assessment: computa-
tional chemistry (i.e., QSAR) and mathematical biology (i.e., PBPK/BBDR modeling). The
Agency's Program and Regional Offices are responsible for hundreds of site-specific risk
assessments which are essential to inform Agency priority-setting. Many of the site-specific
environmental issues involve chemicals about which there are insufficient data. The Agency is
currently exploring the use of QSAR to estimate toxicity benchmarks for such chemicals, as well
as the application of PBPK and BBDR models to risk assessment. It is likely that a systems
-------
biology approach to dose-response modeling will facilitate the integration of PK and PD models.
In addition, QSAR can be used to estimate parameters of PBPK models and cellular response,
and such models are being developed to link QSAR to systemic (whole-organism) dynamics.
Approaches are also being explored to determine how the use of mechanistic information
can lead to human health risk assessment approaches that are consistent across all endpoints
and replace the assumptions that some endpoints are non-threshold effects (cancer) and that
other endpoints are threshold effects (non-cancer). The Agency is also exploring data-based
approaches to adjusting default uncertainty factors in current non-cancer risk assessments. As
a first step, PBPK data and models are being used to adjust default uncertainty factors used in
inter- and intraspecies extrapolation and for developing an assessment of target organ dose for
use in dose-response assessment of chemical mixtures. Similar efforts for the PD uncertainties
would be the next step.
In the field of computational biology, the Agency is just beginning to consider application
of genomic/proteomic data to risk assessments. The Agency's Interim Genomics Policy outlines
the current state of the application of genomics data (which is defined to include proteomics
and transciptomics) in risk assessment. Genomics data may be considered in Agency decision-
making; but they are insufficient, standing alone, to inform decisions about environmental risk.
It is essential, therefore, that the Agency consider how the information generated by this new
technology will be utilized in human health and ecological risk assessment. While there are
many issues that need to be considered before routinely adopting and/or replacing existing data
requirements with expression data for Agency risk assessments, an overarching goal of the
Computational Toxicology Research Program is to address these questions in order to develop
ways to apply computational toxicology in quantitative risk assessment.
2. EXAMPLES OF APPLICATIONS OF COMPUTATIONAL TOXICOLOGY TO
QUANTITATIVE RISK ASSESSMENT
There are many potential applications of computational toxicology in quantitative risk
assessment. The following sections discuss three areas: dose-response assessment, interspecies
extrapolation, and toxicity of chemical mixtures.
A DOSE-RESPONSE ASSESSMENT
Genomics/proteomics technologies have important implications for health and ecological
risk assessment, and there is a need to develop methods that use these data to improve
quantitative dose-response assessment. One important area is the use of emerging technologies
to determine the shape of the dose-response curve in the low dose range based upon in vivo
and in vitro data that can be shown to be correlated with low dose adverse effects. Studies
using emerging technologies may also result in the identification of useful (simple, sensitive, and
relevant) biomarkers of effect so that they can be used in dose response studies, not just hazard
identification, to more accurately diagnose effects in the low dose range and for compounds
with weaker potencies.
Research from the Computational Toxicology Research Program may also lead to the
identification of biological effects that could be used as the adverse effect for risk assessment.
For example, if fetal testis endocrine function is altered such that steroid hormone production
and ins!3 gene expression are reduced by 50% or greater, it is possible that this change could be
considered an adverse effect. Chemicals producing such effects might then be regulated on this
information alone. The assumption of such an approach is that reductions in ins!3 are always
-------
associated with developmental malformations.
Chemicals could then be regulated on the
basis of fetal endocrine effects alone. The
Agency has used endocrine data in this
manner on occasion, but the approach would
be strengthened by including genomic or
proteomic information to support the
hypothesis that a specific pathway had been
sufficiently disrupted by a chemical such that
adverse effects would definitely result later
in life. Hopefully, genomic information will
prove useful in assessing population level
effects in ecosystems, as well as in predicting
risk to individuals in human risk assessment.
Another application of emerging
technologies from the Computational
Toxicology Research Program would be a determination of the relevance of the mechanism or
mode of action of low dose adverse effects to humans and other species using in vitro and in vivo
approaches. Such research could address the critical pathway initially involved in chemically
induced adverse effects and how well the mechanism is conserved between mammals and other
vertebrates.
Although scientists both inside and outside of the Agency have been proposing the
application of BBDR models in risk assessment to reduce uncertainties in the process for several
years, this goal has not been realized in spite of several long-term research efforts due to
the complexity of the models. One question that could be addressed by the Computational
Toxicology Research Program is how quantitative mechanistic genomic data could be used to
develop BBDR models that produce realistic values for risk assessment.
B. CROSS -SPECIES EXTRAPOLATION
Computational Toxicology and
Quantitative Risk Assessment
Defining the shape of the dose response at low
exposures using molecular indicators of response
Developing biomarkers for use in analysis of low
dose responses
Validating the interpretation of molecular
indicators of response
Defining the relevance of modes of action for
risk assessment
Constructing BBDR models of high priority
outcomes
Assessing population level effects in ecosystems
Cross-species extrapolation
Assessing toxicity of chemical mixtures
Integrated human and ecological risk assessment
One of the major challenges in regulatory toxicology is the prediction of toxicity of a
chemical(s) or classes of chemicals across species. In its risk assessments, the Agency often
predicts possible effects in humans from studies in rodents and other mammalian test species;
while in ecotoxicology, extrapolations to literally thousands of other species are typically based
upon results of assays with a handful of surrogate test organisms. In addition, it is generally
assumed that adverse effects seen in vertebrate wildlife are relevant to other species, including
humans. For example, when fish display evidence of exposure to estrogens or androgens
in the environment or frogs exhibit limb malformations, there are immediate concerns about
potential effects on humans and other wildlife. The concept of interspecies extrapolation is
based upon the knowledge that all species arise from common evolutionary ancestors and that
there is a great deal of homology among animal
species with regard to basic biological pathways.
However, the ability to extrapolate from species
to species is not trivial for two reasons. First,
exposures to chemicals vary as a function of an
animal's physiology and its environment. For
example, humans are not exposed to estrogenic
or androgenic materials in effluents to the same
Cross-Species Extrapolation
Understand differences in exposure,
uptake, and metabolism across target
species
Define toxicity pathways for
common chemicals in model species
-------
degree as fish living in or adjacent to an effluent. In a more phylogenetically similar
comparison, the Florida panther is at greater risk for exposure to (and effects of) bioaccumulating
contaminants than are other populations of panthers because it feeds at a higher trophic level than
other populations not because of genetic differences.
Another aspect of differential dosimetry among animals is the metabolism of xenobiotics.
Pathways of metabolism can differ significantly across species, even in closely related animals.
In fact, in many cases, failure to metabolically activate a chemical, or conversely to rapidly
detoxify it, is a primary basis for the lack of response of an animal, not species-specific
differences in gene expression and subsequent toxicity. For example, the anti-androgenic
action of vinclozolin or the estrogenic activity of methoxychlor requires metabolism to active
metabolites. Testing either of these chemicals in a species that does not produce "active"
metabolites could lead to the mistaken assumption that they would not affect endocrine function.
This component of species extrapolation, i.e., comparative dosimetry, can be addressed using
BBDR models that focus both and on concentrations of parent chemicals and metabolites in
target tissues. This type of modeling is an integral part of the Computational Toxicology
Research Program.
The second challenge to species extrapolation involves how an animal actually responds
to a given dose of chemical(s) of concern. Specifically, although there is remarkable similarity
in basic biology among animals, there are also significant species-specific differences in genes,
proteins, biochemistry, and physiology. These differences lead to uncertainties in interspecies
extrapolation. Consequently, toxicologists are generally more comfortable extrapolating among
closely related species than among those that have been separated phylogenetically for a longer
period. To address this aspect of species extrapolation, definition of toxicity pathways from a
comparative perspective is critical; as such, the Computational Toxicology Research Program
will focus on extrapolation. Specifically, characterization of toxicity pathways in well-defined
animal models serves as the basis for identifying key control points (e.g., receptor-mediated
signaling) that, quite possibly, would be conserved across species. This type of conservation
would be expected for many receptor-based processes (as illustrated by estrogen- and androgen-
controlled pathways). When control points are known, state-of-the-art molecular biology (e.g.,
genomic) techniques can be used to assess the degree to which extrapolation of effects associated
with chemicals with specific mechanisms of action can be supported. For example, once it
has been demonstrated that receptor activation by xenobiotics is key to eliciting toxicity, it
will be possible to focus on comparative receptor binding studies across species as a basis for
extrapolation. The critical point here is that, if pathways can be reliably defined in representative
model species, it is not imperative to know a comparable amount of toxicological information
in untested species, only whether the latter possess key components of pathways of concern and
how these components are affected by xenobiotics.
-------
c. CHEMICAL MIXTURES
The Agency's guidance on risk assessment of mixtures indicates that multi-chemical
exposures are ubiquitous, including air and soil pollution from municipal incinerators, leakage
from hazardous waste facilities and uncontrolled waste sites, and drinking water containing
chemical substances formed during disinfection. The guidance also indicates that exposure
scenarios are very diverse. Two approaches are
generally used to assess exposure to and effects
from mixtures, i.e., the study of whole mixtures
or of individual mixture components. The Food
Quality Protection Act (FQPA) specifies that
cumulative effects be addressed for exposure
to multiple chemicals acting by the same
mechanism. However, a number of uncertainties
Chemical Mixtures
FQPA specifies that cumulative effects must be
addressed for exposure to multiple chemicals
acting by a similar mechanism
Lack of information regarding mechanism or
mode of action for risk assessment of mixtures
exist regarding the level of mechanistic similarity among chemicals. Considerable complexity
arises even when examining interactions among chemicals that have "estrogenic" activity. While
whole mixture studies offer environmental relevance, it has been recommended that whole
mixture screening go forward only after screening had been conducted for a number of individual
chemicals. Mixture research over decades is poised to take advantage of unifying principles from
shared chemical and biological mechanistic research. New tools developed in the "omics" and
computational fields show huge potential for employing molecular profiling to understand the
complexity of mixture exposure and elucidating the mechanisms underlying biotransformation,
uptake, distribution, and response. Technological advances now enable study of the joint and
interactive properties of mixtures and definition of those characteristics that are sufficiently
similar to allow extrapolation of data from one mixture to another. They have the
potential to allow identification of emergent properties of real-world mixtures of xenobiotics
at environmental exposure concentrations rather than from defined mixtures at high (toxic)
concentration. Hypotheses can be tested to identify mixture classes (by activity or structure)
that are amenable to component approaches. Mixtures assessment guidelines stand to become
much more uniform, relevant, and easily applicable to risk assessments of real world exposure
scenarios.
-------
-------
CURRENT ACTIVITIES
-------
The previous section described several areas in which a program on computational
toxicology could provide methods and models that would lead to more efficient ways to assess
chemicals for screening and testing, as well as improve quantitative risk assessment. Section El
demonstrates that ORD currently possesses the capability to utilize approaches necessary for the
ultimate development of a Computational Toxicology Research Program in the future. The work
described below is divided into three categories, including ORD's initial research to demonstrate
the feasibility of the computational toxicology concept, examples of on-going research, and
linkages to external research groups.
A PROOF-OF -CONCEPT: ENDOCRINE DISRUPTING CHEMICALS (EDCs)
Initiated by an FY02 congressional mandate to explore alternatives to the use of animals
in toxicological testing, ORD began a research effort to explore the use of emerging technologies
and computational approaches to better prioritize chemicals for screening and testing. Because
much is known about how EDCs interact with biological systems to cause adverse health,
it was decided to focus the program on this class of chemicals and to conduct several
proof-of-concept experiments to determine the feasibility of using computational toxicology
approaches to meet an immediate Agency need. Understanding the key biological pathways
affected by endocrine disrupting chemicals affords
the opportunity to design approaches that are
more efficient in terms of resource utilization. It
also allows the Agency to extrapolate findings
from a smaller set of chemicals to the broader
chemical universe using the tools of computational
chemistry. Thus, projects using in silica, in vitro,
Proof-of-Concept Studies with EDCs
Refine existing QSAR models
Develop In vitro models
Characterize toxicity pathways
and in vivo approaches could facilitate the prioritization of chemicals for screening, reduce
the need for some in vivo assays, and provide in vivo assays that have a greater breadth of
coverage of endocrine alterations and/or provide better predictiveness of potential adverse health
outcomes. The overall program has short-term, intermediate, and long-term goals. The effort
in this proof-of-concept activity is directed at developing better tools and assays to monitor
selected aspects of endocrine disruption. Success will be measured against the recommendations
put forth by the Endocrine Disruptor Screening and Testing Advisory Committee (EDSTAC)
and will provide us confidence that the approaches are applicable to other pathways of toxicity
where the underlying biology is not so well understood at the present time. Descriptions of
the activities follow.
Receptor Binding Models - The EDSTAC-recommended approach to screening
chemicals for endocrine activity proposed the use of QSAR models of receptor binding to
help prioritize chemicals for further screening. EDSTAC also recommended that the Agency
undertake a study to evaluate the a priori predictions of available QSAR models for estrogen
receptor interaction by obtaining data on competitive binding affinity for the estrogen receptor
from a single laboratory using a standardized protocol on approximately 300 chemicals. This
demonstration exercise found the models needed additional work to improve their sensitivity,
specificity, and predictive capabilities before they could be used in a regulatory context. The lack
-------
of complete agreement between binding information and QSAR models appears to be related to
inconsistent receptor-binding information in the training sets used to initially construct the QSAR
models. The values to derive the model were derived from multiple laboratories and were based
on IC$Q values, which potentially can introduce errors because non-competitive binding might be
present. Therefore, under the proof-of-concept effort, ORD is talcing 70 of the 300 chemicals that
showed some evidence of receptor interaction and generating K. values for each of them. This
will provide an unbiased and unequivocal measure of receptor binding. Following acquisition of
these data, the QSAR models will be rederived using the new training set and their predictability
will again be assessed. This effort will demonstrate how international criteria for transparency,
domain coverage, and model acceptance evaluations can be met while reducing the number
of chemical tests needed as compared to random chemical testing approaches. Through this
iterative process, ORD will test the hypothesis that given a robust data set of high quality, reliable
QSAR models can be developed and used in hazard identification. Assuming success in this
effort, the research will progress to accomplish similar goals for the androgen receptor (AR).
In vitro models - Knowing the toxicity pathway of concern in whole animals should allow
development of simpler in vitro systems that can provide quick and inexpensive evaluation of the
potential for chemicals to interact with that pathway. EDSTAC raised concerns about the need to
evaluate the effects of chemicals on the steroidogenic pathways in Tier I screening approaches.
EDSTAC addressed this data gap by recommending a combination of studies in pubertal rodents
and enzyme inhibition studies on either placenta or minced testes preparations from male rats.
Both of these approaches are limited in their ability to study the synthesis of the key steroids
(estradiol and testosterone) from cholesterol, including the fact that they are more targeted at
detecting inhibition of synthesis rather than enhanced synthesis. A human cell line (H295R) has
been identified that maintains ability to synthesize estrogen from cholesterol and that addresses
many of the limitations of the EDSTAC approach. Under the proof-of-concept effort, ORD is
developing standard operating procedures for use of the H295R cell line to evaluate each step
of steroidogenesis at the genomic, proteomic, and metabonomic level. ORD will compare the
results from a set of chemicals with the EDSTAC-recommended assays to determine if this cell
line affords more powerful, yet easier to achieve, answers to the potential to alter steroidogenesis.
Toxicity Pathway Characterizations - These studies are integral to the proof-of-concept
effort and will consist of studying thyroid gland functioning in ecologically relevant species and
by examining the integrating function of the vertebrate hypothalamic-pituitary axis in responding
to the presence of EDCs.
The thyroid gland was selected as a target for the effort because it represents an endocrine
organ whose function is disrupted, not by direct xenobiotic interaction with the thyroid receptor,
but by interactions elsewhere in the endocrine loop (e.g., iodine uptake processes, hormone
synthesis, hormone modification, and hormone metabolism). Given the fact that thyroid function
can be perturbed at different points in the thyroid pathway, research is being directed at
developing a suite of endpoints that could ascertain which toxicity pathway is initiated by a
specific chemical. Each chemical interaction with the thyroid signaling axis would be considered
a separate toxicity pathway given that the chemical structural requirements for interaction with
each site are likely unique. This could facilitate the development of QSAR models for each
toxicity pathway. Therefore, it is important to realize that there may be multiple pathways for
any given toxicological response. It would also be advantageous to maintain the multiple
toxicity pathway distinction when assessing aggregate versus cumulative chemical risk. A
common mechanism of action cannot be assumed unless it is shown that mixtures of chemicals
across these different toxicity pathways exhibit additive behavior when equal toxic units are
-------
combined for joint toxicity assessments. Success here would demonstrate that a more global
focus on endocrine function using genomic approaches and critical life stages can provide more
meaningful information than the EDSTAC-recommended amphibian metamorphosis test, which
is neither highly sensitive nor diagnostic of disturbances in the function of the thyroid.
The last area covered included in research on toxicity pathways is more exploratory in
nature; i.e., it involves exploring the possibility that, through the use of genomic and proteomic
evaluations, a single in vivo test for endocrine-disrupter activity can be developed that will meet
the essential requirements for an EDC screening assay. This activity is based on the rationale
that the central nervous system (CNS) contains all of the relevant receptor and enzymatic target
sites of interest. Talcing advantage of the fact that there is a wealth of information available
concerning the physiological regulation of the thyroid and adrenal and gonadal axes by the
hypothalamus and the pituitary glands, it may be possible to empirically test the extent to
which changes in these endocrine systems are sensed and responded to by the CNS. It is
anticipated that the development of a genomic response profile following exposure to EDCs
of known action will provide the means to identify the target pathways that lead to altered
reproductive/thyroid/adrenal function. This approach, if successful, would be superior to the
current proposed male and female pubertal assays because it could predict all relevant target
pathways whereas the current assays can not necessarily identify specific CNS target sites.
Furthermore, this approach has the potential to advance beyond screening and may incorporate
elements important to establishing efficient Tier 2 tests meant to characterize dose response
relationships for endpoints of utility in risk assessment.
It is expected that the lessons learned from these EDC proof-of-concept projects will be
immediately transferable to future studies on the potential of non-endocrine disrupting chemicals
to affect other biological systems and signaling pathways. It is already apparent that expansion
beyond the evaluation of EDCs to other toxicity pathways will encompass several key steps:
(1) elucidation of pathways of chemical toxicity, from initiating event to adverse outcome in
individuals or populations; (2) identification of key assays indicative of toxicity pathways that
provide a means to efficiently, wisely, and with minimal animal testing, extrapolate across
chemicals and species; and (3) the application of an iterative QSAR strategic test design,
where models are developed and strategically improved, with minimal testing, until criteria for
regulatory acceptance are met. This strategy will maximize the likelihood that predictive models
developed under this approach will have a solid foundation in scientific principles and will
provide the Agency with defensible approaches to priority setting.
While not currently underway, a likely second proof-of-concept project would be one
that transcends all aspects of the source-to-outcome continuum and would therefore be able to
demonstrate the advantages/long term or far reaching benefits of a computational toxicology
approach throughout the entire risk assessment spectrum. Experience gained in the EDC proof-
of-concept research will be invaluable in helping to formulate such a project.
B. INTERNAL LINKAGES
There are several on-going research projects in ORD's core and problem-driven research
program that are supportive of, or which will benefit from, efforts in computational toxicology.
ORD's core research program aims to provide broad, fundamental scientific information
to improve understanding of human health issues. In particular, research associated with
ORD's Human Health Research Strategy will support major components of the Computational
Toxicology Research Program. For example, research to develop a common approach for
-------
the use of mechanistic data in cancer and non-cancer risk assessment will provide significant
information concerning toxicity pathways for high priority environmental chemicals. Likewise,
efforts to harmonize approaches between human and ecological risk assessment will also provide
a conceptual basis for identifying linkages in the source-to-dose paradigm. It is expected
that advances in computational toxicology will lead to reductions in uncertainty associated
with extrapolation across species and will aid in protecting susceptible subpopulations, such as
children and older individuals.
In order to illustrate the complementary nature of the various research programs within
ORD, we have chosen to describe projects that have a high regulatory impact (i.e., cancer,
reproductive, pulmonary, and neurotoxicity) and that are associated with specific environmentally
relevant contaminants [e.g., particulate matter, arsenic, disinfectant by-products (DBFs), EDCs]
or risk assessment issues. As examples, the two following sections describe on-going research
on the human health effects of DBFs and the assessment of an aquatic species in ecological
risk assessment. Appendix A provides additional examples of ORD projects that support the
Computational Toxicology Research Program.
l. HUMAN HEALTH RESEARCH
Under the Information Collection Rule, DBF occurrence data for major water systems
across the country are becoming available. Data are being collected on DBFs that have been
predicted to have an adverse health effect but which have little or no previous quantitative
occurrence information. In addition to this drinking water, utility-based information, data on
individual water use and biomarker development are being developed to increase the precision
of exposure assessment. These data will help to identify classes of DBFs and candidate model
compounds that are present at the greatest concentration. Because DBF concentrations at the tap
can vary widely and be dramatically different
from reported averages from water utilities, a Examples of Chemicals Under Study by ORD
series of measurements on drinking water at
Arsenic
Disinfection By-Producls
Endocrine Disruptors
Particulate Matter
Pesticides
different points in the distribution system is
needed for better informed exposure models.
Current exposure modeling from ingestion of
disinfected drinking water reflects only the
general characteristics of the source water
and disinfection process. To provide more
refined exposure assessments, improved exposure models are currently being developed using
more quantitative data on DBF component analysis and to more accurately predict of the
concentration of the most toxic DBFs at the tap. This work will then be combined with response
models to develop a computational approach for predicting potential adverse effects based on the
predicted and actual characteristics of the mixture of DBFs.
Computational chemistry methods and molecular modeling approaches are being
employed to compute a variety of structural, electronic, and reactivity characteristics of DBFs;
their postulated metabolites; and/or adducts for input and consideration in the development of
structure-activity models. Efforts currently focus on differences based on the bromine content
within a class of DBFs. The central issue, in this case, is the role of bromination in determining
and modulating biological activity within these classes. Toxicity information, in conjunction
with known principles of the organic chemistry of halogen-facilitated reactions, will be applied
to understand the possible range of mechanism-based reactivity and how such reactivities are
modulated by chlorination and bromination. These modulations include DNA adduct formation
-------
and carcinogenicity. Computational chemistry and SAR modeling will be used to prioritize
chemicals for testing and aid in the preliminary hazard assessment of DBFs for which little
or no toxicity data are available. These approaches can also be used to generate mechanism-
based SAR hypotheses pertaining to classes of halogenated drinking water contaminants and
to help guide the design of experimental studies to most productively address areas of greatest
uncertainty.
2. ECOLOGICAL RESEARCH
ORD has made a significant commitment to genomic and proteomic research on model
aquatic vertebrates. For example, ongoing work in this area has focused on the use of genomics
and proteomics to delineate toxicity pathways associated with disruption of key elements of the
thyroid system in the amphibian model, Xenopus tropicalis. Other work in this area is focused
on small fish models, including the sheepshead minnow, medaka, and the fathead minnow. A
primary goal of this research is elucidation of toxicity pathways induced by perturbation of
processes controlled by the hypothalamic-pituitary-gonadal/thyroid (HPG/T) axes. The fathead
minnow, Pimephalespromelas, is an important model for this research because (1) it is the
Agency standard for teleost aquatic toxicity testing; (2) the fathead minnow 21-day reproductive
toxicity test is the priority Agency-recognized screen for endocrine disrupters in teleosts; (3)
significant progress has been made in gene discovery and initial cDNA microarray synthesis via
core microarray facilities; (4) high-throughput sequencing of cDNA libraries by the Department
of Energy (DOE) is likely; and (5) a multi-laboratory effort is enabling integration of exposure
and effects research. Ultimately, identification of initiating events or other critical pathway
elements will enable the development of QSAR models, initially for estrogens and androgens
in fish, and subsequently for other targets on the HPG/T axes. As research focused on toxicity
pathways associated with these axes develops, classes of chemicals that operate via other
modes/mechanisms of action also will be considered. An important group of chemicals currently
under consideration in this regard is the polyfluorinated surfactants.
Besides the primary objective of elucidating toxicity pathways, this collaborative research
aims to significantly support other goals of the Computational Toxicology Research Program.
For example, the discovery of genes specifically induced/repressed by hormone agonists and
antagonists will enable the development of specific diagnostic molecular indicators of exposure
and the linkage of PB/PK models to mechanistically-based QSAR models (see section I-C). In
addition, genomic, proteomic, and metabonomic research will provide the basis for comparison
of toxicity pathways across species. For example, the present collaborations within ORD have
already produced highly significant data on the conservation and function of the estrogen receptor
across genera.
C. EXTERNAL LINKAGES
ORD is actively seeking to establish links with groups outside the Agency that have
expertise and capabilities that could complement and augment the emerging program on
computational toxicology. Some of these linkages have already been initiated, and they are
listed below. Other partnerships, such as with other governments or private consortiums, will
likely develop over time as the program matures and specific research needs in the intramural
program are identified.
-------
l. CHEMICAL INDUSTRY INSTITUTE FOR TOXICOLOGY (CUT) CENTERS
FOR HEALTH RESEARCH
ORD and CUT have agreed to a memorandum of understanding (MOU) to advance
the state-of-the-science of computational toxicology. Among the goals of the agreement is the
utilization of the complementary capabilities and expertise of ORD and CUT in developing
and applying computational toxicology approaches to human health risk assessment. Research
activities common to both organizations focus on the use of computational methods and
molecular biology toward the characterization of risks of environmental contaminants. CIIT
focuses on the development of PBPK and PD while ORD contributes to the joint effort by
studying relevant toxicant-induced effects in various target organs (i.e., research on toxicity
pathways). Improving the risk assessment process has been an emphasis of both groups for
many years, and approaches involved in computational toxicology are compatible with the
goals of both organizations. This collaboration will foster exchanges of information, training
opportunities, and technologies. Initial research focuses on examining the toxicity pathways for
the effects of dibutyl phthalate in the developing testes.
2. DEPARTMENT OF ENERGY (DOE)
In the fall of 2002, ORD scientists visited the Sandia and Pacific Northwest National
Laboratories of DOE. These visits led to the development of 23 possible project areas thought
to be suitable for collaboration. These were subsequently narrowed down to five projects worthy
of further exploration. The high priority projects of mutual interest to DOE and the Agency
include a range of areas such as the following: (1) research to develop computational screening
techniques beginning with traditional QSAR screening of chemicals to predict toxicity of EDCs;
(2) the development of new techniques to specifically identify molecular structures associated
with toxicity; and (3) the use of innovative and proteomic techniques to identify, characterize,
and classify sensitive subpopulations based on biological factors to reduce uncertainty in risk
assessment. Additionally, both DOE and the Agency are interested in the characterization of
atmospheric pollutants to better understand the processes influencing human risks to atmospheric
pollutants, to identify and characterize the casual agents associated with these effects, and to
better understand the systems biology and mechanisms associated with the exposure-to-dose-
to-effects parts of the continuum. An MOU and an Interagency Agreement have also been
developed between DOE and ORD to provide high performance computing consulting and/or
access to DOE non-classified computing equipment.
In January 2003, ORD scientists also visited the U.S. DOE Joint Genome Institute (JGI),
which resulted in JGFs agreement to sequence the genome of selected species currently being
used for testing and screening of environmental pollutants (i.e., Pimephalespromelas, Xenopus
tropicalis and Daphniapulex). Sequencing the genome of these test species could lead to
molecular-based models for predictive toxicology. Another visit to DOE laboratories was made
in December 2003.
Other collaborations with genomics and bioinformatics experts at the University of
Cincinnati Children's Hospital Research Center have begun to yield synthesis of complementary
DNA microarrays with fathead minnow gene sequences and expressed sequence tags. These
researchers represent the conduits for synthesis of large-scale microarrays using sequences
gleaned from the JGI collaborations.
-------
3. NATIONAL INSTITUTE OF ENVIRONMENTAL HEALTH SCIENCES
(NIEHS)
The National Center for Toxicogenomics (NCT) at NIEHS was formed to facilitate
development of gene expression and proteomic methodology, to create a public database relating
environmental stressors to biological responses, to collect information relating environmental
stressors to biological responses, to develop improved use of computational mathematics in
understanding responses to environmental stressors, and to identify biomarkers of disease or
exposure to enhance environmental health. Many of these objectives overlap with current
core and problem-driven research at ORD. Preliminary discussion has been held with NCT
concerning the Computational Toxicology Research Program at ORD. One possibility for
collaboration is that the genomic, proteomic, and metabonomic information from ORD's
Computational Toxicology Research Program will be made available to the NCT's Chemical
Effects in Biological Systems (CEBS) database. CEBS is a relational and descriptive compendia
of lexicologically important genes, groups of genes, and mutants and their functional phenotypes.
This platform allows searching for information about the biological effects of chemicals
and other agents and their toxicity pathways based on information from the literature and
contributions from intramural and extramural sources. The database utilizes standardized
procedures, protocols, data formats, and assessment methods to ensure that data meet a uniform
level of quality. It is expected that ORD will help build a publicly accessible toxicological
database that will be capable of predictive toxicology and that will serve as a major resource for
researchers to pose and test mechanistic hypotheses.
4. SCIENCE TO ACHIEVE RESULTS (STAR)
Through its external grants program, ORD is developing a series of requests for assistance
that will lead to the support of scientists in academic and not-for-profit institutions in research
areas that complement the in-house research activities. Most of the support will be in the form
of grants; but in some cases where ORD scientists would like to work more closely with the
awardees, cooperative agreements may be awarded. The selection of both awarded grants and
cooperative agreements will be through a competitive process. All awardees will be asked to
participate in periodic progress reviews in which intramural and extramural scientists are brought
together to share and review data.
-------
-------
-------
TEPS
-------
A REVIEW OF THE FRAMEWORK
The next step in the development of a Computational Toxicology Research Program in
ORD was to obtain external peer review of the Framework for a Computational Toxicology
Research Program in ORD. On September 12, 2003, the Framework was reviewed by the
Computational Toxicology Framework Consultation Panel of the Agency's Science Advisory
Board. The current document includes several revisions based on the results of that consultation.
In addition, ORD hosted a workshop on September 29-30, 2003, to discuss how ORD will use the
data and experience from other research organizations in the development of a research program.
A summary of the results of that workshop can be found on the Agency's computational
toxicology Web site fhttp://www.epa.gov/comptox ). In FY04, ORD will issue Requests for
Proposals to identify themes and intramural and extramural research approaches in the area of
computational toxicology based on the priorities and process described in the following two
sections. Eventually, an ORD Multi-Year Plan on Computational Toxicology will be developed
to describe the critical milestones that this area of research will accomplish over the next five
to eight years.
B. PRIORITIES FOR RESEARCH ON COMPUTATIONAL TOXICOLOGY IN ORD
The following criteria will be used to set priorities for research:
1. Risk-Based Planning: Research should address an element in the source-to-outcome
continuum, and it should be designed to improve quantitative risk assessment or to
facilitate the development of screening or testing strategies for chemicals.
2. Utilizes New Technology: Research that employs emerging proteomic, genomic, and
metabonomic methods.
3. Hypothesis-Driven: Research that tests hypotheses.
4. Scientific Excellence: The quality of the planned science must be able to withstand
rigorous peer review.
5. Programmatic Relevance: Research that addresses a key Agency-related mandate
concerning protection of human health and/or the environment.
6. Other Sources of Data: Research that involves partnerships and collaborations with
other organizations outside of the Agency.
7. Capabilities and Capacities: Research that can be completed within a reasonable period
of time using available material and human resources.
8. Sequence of Research: The critical path of the research dictates those efforts which
must be started before other elements are initiated. This again refers to the relative
state of maturity of different technologies included in the Framework.
9. Balance: There should be a balance of short-term versus long-term results.
-------
C. PROCESS FORA RESEARCH PROGRAM ON COMPUTATIONAL
TOXICOLOGY
The development of an ORD program on computational toxicology will require
coordination and communication between ORD managers and investigators to ensure that the
research is relevant, timely, and defensible. Engagement of personnel from Agency regulatory
offices will be an important component of the implementation process in order to ensure that the
activities are targeted towards where they can have the greatest impact. Efforts will be made to
optimize existing intramural research by seeking out complementary extramural research efforts.
A Web site has been established (www.epa.gov/comptox) that will serve as a repository of
information and progress in the program. This will serve as an important tool for communication
both within and outside the Agency.
To be successful, it is apparent that increased emphasis on bioinformatics and database
management on a scale not generally available within ORD is critically needed. In this regard,
ORD is collaborating with the Office of Environmental Information (OEI) to develop a plan
for upgrading the Agency's information technology (IT) capabilities and making those resources
available to Agency scientists and risk assessors. Through this plan, the Agency envisions
establishing a state-of-the-science IT framework and architecture network that would liken the
Agency's computing capabilities and applications to those of other federal agencies (e.g., DOE,
Centers for Disease Control and Prevention) and non-governmental organizations (e.g., the
University of Chicago Statistical Center). This enhanced IT infrastructure will enable Agency
scientists to utilize sophisticated computational techniques to develop virtual biological systems
(e.g., simulated liver metabolism models, virtual neuroendocrine system) and models to predict
toxicological effects (e.g., QSAR approaches for endocrine disrupters).
Following final release of the Framework, the technical writing team that authored this
document has been replaced by the ORD Computational Toxicology Implementation Steering
Committee. The mandate for this group is to provide oversight of the process to implement
a research program on computational toxicology in ORD. This includes the development of
an ORD Implementation Plan for computational toxicology to guide and coordinate research in
this area for the next five to eight years. Scientist-to-scientist meetings involving intramural
and extramural researchers will also be sponsored by the Steering Committee to help integrate
research with intramural and extramural groups working in this area. A first step in this
process occurred on September 29-30, 2003, at a workshop entitled "Computational Toxicology:
Framework, Partnerships and Program Development" at which nearly 200 participants heard
detailed descriptions of the Framework and began discussions to help identify research areas
for future emphasis. These discussions centered around two approaches: needs based on an
existing regulatory driver and needs that would be most compatible with the application of
emerging computational tools and approaches. The purpose of these discussions was to take
maximum advantage of emerging technologies that would be of immediate relevance to Agency
risk assessors and risk managers. Additional workshops are being planned that will provide
a greater depth of discussion for various topics in the Framework, e.g., use of computational
approaches for susceptible subpopulations, models addressing cumulative risk, or approaches
for prioritization of chemicals for Agency data requirements. Periodic review of on-going
research by the Steering Committee will be crucial to demonstrating progress and designing
future research directions. Systematic communication with the Agency's Program and Regional
Offices through the planning process will be important for providing them with an understanding
of the nature and extent to which the applications of computational toxicological approaches
meet the Agency's data requirement needs and improve quantitative risk assessment.
-------
APPENDIX A
Examples of Current ORD Projects Associated with Computational Toxicology
A. Improve Linkages in the Sourcc-to-Oiitcome Continuum
1. Chemical Transformation and Metabolism
Fale models-Core research to elucidate and mode! the behavior of organic contaminants in
natural and impacted ecosystems and in complex biological systems (NliRL)
Fate modcls-Pmblcm-drivcn research to develop the methods, tools, and databases to forecast
the fate of pesticides and toxic chemicals during the drinking water process (NERL)
2. Development of Diagnostic/Prognostic Molecular Indicators
Developmental Biomarkers-core research that focuses on identifying molecular markers of
developmental toxicity related to growth and maturation of organ systems (NHEERL)
Virulence Potential-core research to develop molecular methods to measure virulence of
microbial pathogens (NER.L)
3. Dose Metrics
Cumulative Risk for Drinking Water Contaminants-problcm-drivcn research to use QSAR,
mixtures toxicity approaches and PBPK modeling to develop a cumulative risk method
based on doses in target tissues (NCEA)
PBPK Models in Fish-core research to use genomic data to validate output of PBPK models
(NERL)
4. Characterization of Toxicity Pathways
Cell Signaling-core research to determine role of signal iransduction pathways in toxicity
pathways for high priority environmental chemicals (NHEERL)
Framework for Defining Model and Mechanism of Action for Cancer and Noncancer
Endpoints-research to explore using genomics and proteomics as an approach for
understanding mechanisms and the implications for risk assessment (NCEA)
5. Systems Biology
None
B. Provide Predictive Models for Screening and Testing
1. QSAR Approaches
Application of QSAR and modeled exposure estimates in risk-based chemical ranking model
(NCEA)
Perfluorooclane sultanate (PFOSj-problem-driven research on mode or mechanism of action of
PFOS, a breakdown product of several widespread and persistent chemicals in the
environment (NHEERL)
2, Pollution Prevention Strategies
Endocrine Disrupting Chemicals Replacement Program - problem-driven research to develop a
software tool that will allow users to quickly identify possible replacements tor
chemicals that are known or projected to have endocrine disrupting potential (NRMRL)
Pollution Prevention Tools - core-driven research to incorporate results of other ctwnputaliona]
toxicology projects into software tools used in applying pollution prevention strategies
NRMRL)
3. High Throughput Screening
Endocrine Disruptors-problem-driven research designed to identify endocrine-mediated effects
using rapid high throughput protein fingerprinting techniques (NHEERL)
-------
C. Enhance Quantitative Risk Assessment
Examples of Current ORD Projects Associated with Computational Toxicology
1. Dose-Response Assessment
Acute-to-Chronic listiniate-problem-driven research to develop regression and accelerated lite
testing models to predict long-term chronic loxicily from short-term acute responses and
determine uncertainty at low concentrations (NHEERL)
Develop and Apply Unified Modeling Procedure to Cancer and Noncancer Risk Assessment to
support biologically based dose response modeling (NCEA)
2. Cross Species Extrapolation
Cross-species Extrapolation in Birds-problem-driven research to develop PBPK models to
extrapolate reproductive and neurological elTccis of metals among bird species
(NHEERL)
Value-of-lnformation Approach to Motivate Uncertainty Factors with Mechanistic Data:
Chlorine Human Health Risk Case Study: Experimental and computational e(Torts
aimed at obtaining and integrating mechanism of action data to develop a biologically
based risk assessment for chlorine. The results will be generalized to develop a formal
framework for departing from default uncertainty factors based on PK/PD data (NCEA)
3. Chemical Mixtures
Chemical Mixtures-core research to apply genomic analyses of exposure of fathead minnows to
binary and tertiary chemical mixtures (NERL)
Interactions and Mechanism of Pesticide Mixtures: PBPK/BBDR modeling for immunotoxicity
risk assessment of chemical mixtures (NCEA)
-------
-------
-------
------- |