&EPA
   United States
   Environmental Protection
   Agency
                   QSAR/VFAR Workshop Summary
                   REPORT
                                             200 -
                                                               15
   Office of Research and Development
   National Homeland Security Research Center

-------

-------
                                                                  August 2007 EPA/600/R-07/095
          QSAR/VFAR  Workshop

          Summary Report

          By
          Smita Siddhanti and Caroline Baler-Anderson
          EnDyna, Inc.
          McLean, VA 22102

          Under subcontract to:
          Science Applications International Corporation
          11251 Roger Bacon Drive
          Reston,VA  20190

          EPA Contract No. 68-C-02-067

          Project Officers
          Chandrika Moudgal
          National Homeland Security Research Center
          U.S. Environmental Protection Agency
          Cincinnati, OH 45268
          Douglas Young
          National Risk Management Research Laboratory
          U.S. Environmental Protection Agency
          Cincinnati, OH 45268
Office of Research and Development
National Homeland Security Research Center, Threat and Consequence and Assessment Division

-------

-------
                                                                          Disclaimer
The opinions expressed within the workshop summaries do      trade names or commercial products does not constitute
not necessarily represent the views of the EPA. Mention of      endorsement or recommendation for use.

-------
Acknowledgements
This report was prepared for the U.S. Environmental
Protection Agency, Office of Research and Development,
National Homeland Security Research Center and National
Risk Management Research Laboratory. The primary authors
were Dr. Caroline Baler-Anderson and Dr. Smita Siddhanti
of EnDyna, Inc., under subcontract to Science Applications
International Corporation (SAIC), Contract No. 68-C-02-
067. Lisa Kulujian was SAIC's Work Assignment Manager.
Chandrika Moudgal and Douglas Young of EPA's National
Homeland Security Research Center and National Risk
Management Laboratory, respectively, served as EPA's
project officers.

-------
Table  of  Contents
Notice	ii
Disclaimer	iii
Acknowledgements	iv
Acronyms	vii
Executive Summary	ES-1
      Introduction	ES-1
      Specific Workshop Goals	ES-1
      Background	ES-1
      Major Themes Discussed	ES-2
      Charge to the Expert Panel and Major Considerations	ES-3
      Recommendations	ES-6
Chapter 1: Introduction	1
      1.1 Background	1
      1.2 Purpose and Goals of the Workshop	1
      1.3 Charge to the Expert Panel	2
      1.4 Organization of This Report	2
Chapter 2: Background and Opening Remarks	5
      2.1 NHSRC and NRMRL	5
      2.2 Opening Presentations	5
Chapter 3: VFAR Presentation Summaries	7
      3.1 Introduction to the VFAR Concept
          Gerard Stelma, Senior Science Advisor, NERL	7
      3.2 Using VFAR in a Risk Assessment Framework
          Joan Rose,  Homer Nowlin Endowed Chair for Water Research,
          Michigan State University	7
      3.3 VFAR Factors Related to Genomic Variability
          Syed Hashsham, Associate Professor, Department of Civil and Environmental
          Engineering and Center for Microbial Ecology, Michigan State University	8
      3.4 A Bioinformatic Approach to VFAR Analysis and Characterization
          R. Paul Schaudies, SAIC	8
Chapter 4: VFAR Charge Questions	11
      4.1 Summary of VFAR Charge Questions Discussions	11
      4.2 VFARs Closing Remarks	16
Chapter 5: QSAR Presentation Summaries	19
      5.1 From Reactivity to Regulation:  Integrating Alternative Techniques to Predict Toxicity
          Mark Cronin,  Professor of Predictive Toxicology, Liverpool
          JohnMoores University	19
      5.2 Integrated QSAR - PBPK Modeling for Risk Assessment
          Kannan Krishnan, Director of the Human Toxicology Research Group (TOXHUM),
          Universite de Montreal	19
      5.3 Weight of Evidence and Mode of Action in Predictive Toxicology
          Andrew Maier, Associate Director, TERA	20
      5.4 Novel Approaches to QSAR and VFAR Modeling
          William Welsh, Norman H. Edelman Professor in Bioinformatics and Computer-Aided
          Molecular Design, Department of Pharmacology, University of Medicine & Dentistry
          of New Jersey (UMDNJ)	20

-------
       5.5 Role of the European Chemicals Bureau in Promoting the Regulatory Implementation
          of Estimation Methods
          Andrew Worth, European Chemicals Bureau, Institute for Health & Consumer
          Protection, Joint Research Centre, European Commission	21
Chapter 6: QSAR Charge Questions	23
       6.1 Summary of QSAR Charge Questions Discussions	23
       6.2 QSAR Closing Remarks	27
Chapter 7: Major Considerations and Recommendations	29
ChapterS: References	33
Appendix A: List of Speakers	A-l
Appendix B: Biosketches of Speakers and Panelists	B-l
Appendix C: Workshop Agenda	C-l
Appendix D: List of Attendees	D-l
Appendix E: Workshop Presentation Materials	E-l
       Introduction to the VFARs Concept
       Jerry Stelma,  U.S. EPA, ORD, Cincinnati, OH	E-l
       Using VFAR in a Risk Assessment Framework
       Joan B. Rose, Department of Fisheries and Wildlife, Michigan State University,
       East Lansing, MI	E-4
       VFAR: Factors Related to Genomic Variabilities
       SyedA. Hashsham, Department of Civil and Environmental Engineering and
       Center for Microbial Ecology, Michigan State University, East Lansing, MI	E-10
       A Bioinformatic Approach to VFAR Analysis and Characterization
       R. Paul Schau dies, Scientific Applications International Corporation, Rockville, MD	E-l 4
       From Reactivity to Regulation: Integrating Alternative Techniques to Predict Toxicity
       Mark Cronin, School of Pharmacy and Chemistry, Liverpool John Moores University,
       Liverpool, England	E-l7
       Integrated QSAR-PBPK Modeling for Risk Assessment
       Kannan Krishnan, Universite de Montreal, Montreal, Canada	E-20
       Weight of Evidence and Mode of Action in Predictive Toxicology
       Andrew Maier andRaghu Venkatapathy, Toxicology Excellence for Risk Assessment,
       Cincinnati, OH	E-24
       Novel Approaches to QSAR & VFAR Modeling
       William Welsh, Robert Wood Johnson School, University of Medicine & Dentistry
       of New Jersey, Piscataway, NJ	E-27
       Role of the European Chemicals Bureau in Promoting the Regulatory
       Implementation of Estimation Methods
       Andrew Worth, European Commission - Joint Research Centre, Ispra, Italy	E-32

-------
Acronyms
BMD:  Benchmark Dose
BMDL: Benchmark Dose Lower Confidence Limit
BMC:  Benchmark Concentration
BMCL: Benchmark Concentration Lower Confidence Limit
CCL: Contaminant Candidate List
CAMRA:  Center for Advancing Microbial Risk Assessment
CTC: Computational Toxicology Center
DORIAN: Dose-Response Information Analysis System
EC50: Median Effective Concentration
ECB: European Chemicals Bureau
EPA: Environmental Protection Agency
ER:  Endocrine Receptor
EU:  European Union
FDA: Food and Drug Administration
FIGUR: Fast Identification of Genomic Unique Regions
GPCRs: G Protein-Coupled Receptors
GSH: Glutathione
LOAEL: Lowest Observed Adverse Effect Level
LD5Q: Median Lethal Dose
LC50: Median Lethal Concentration
MOA:  Mode of Action
NERL:  National Exposure Research Laboratory
NCEA:  National Center for Environmental Assessment
NHSRC: National Homeland Security Research Center
NOAEL: No Observed Adverse Effect Level
NRMRL:  National Risk Management Research Laboratory
OECD:  Organization of Economic Cooperation and Development
OPPTS: Office of Pollution Prevention and Toxic Substances
OSWER:  Office of Solid Waste and Emergency Response
PBPK:  Physiologically Based Pharmacokinetic
PCR: Polymerase Chain Reaction
PD:  Pharmacodynamics
PDB: Protein Data Bank
PK:  Pharmacokinetics
PNN: Polynomial Neural Network
QSAR:  Quantitative Structure-Activity Relationship
RE:  Risk Estimation

-------
REACH:  Registration, Evaluation, and Authorization of Chemicals
RfC:  Reference Concentration
RfD:  Reference Dose
SDWA: Safe Drinking Water Act, as amended in 1996
TCAD: Threat and Consequence Assessment Division
TERA: Toxicology Excellence for Risk Assessment
TOXHUM:  The Human Toxicology Research Group, Universite de Montreal
UMDNJ:  University of Medicine & Dentistry of New Jersey
VF: Virulence Factor
VFAR: Virulence Factor-Activity Relationship
VHTS: Virtual High-Throughput Screening
VMG: Virulence and Marker Genes
WOE: Weight of Evidence

-------
                                                                    Executive  Summary
Introduction
The U.S. Environmental Protection Agency's (EPA's) National
Homeland Security Research Center (NHSRC) and National
Risk Management Research Laboratory (NRMRL) conducted
a QSAR/VFAR Workshop, on June 20-21, 2006 in
Cincinnati, OH. The workshop's main purpose was to explore
the application of Quantitative Structure-Activity Relationship
(QSAR) and Virulence Factor-Activity Relationship (VFAR)
concepts to the risk assessment process in situations where
chemical- or biological-specific empirical data are either
inadequate or lacking.
The mission of both NHSRC and NRMRL is related to
assessment of public health and environmental risk from
harmful chemicals. Parts of the Office of Research and
Development (ORD), NHSRC and NRMRL manage and
support a variety of research and technical assistance
efforts. NHSRC focuses on enhancing the ability to detect,
contain, mitigate the effects of, and clean up after significant
emergency events, terrorist attacks, or natural disasters.
NHSRC scientists and engineers seek to identify or develop
affordable, effective technologies and methods for addressing
the risks posed by chemical, biological, and radiological
agents. NRMRL's mission is to develop ways to prevent
and reduce pollution of air, land, and water. This mission
plays a critical role in EPA's goal of achieving sustainability;
several methodologies have been developed within NRMRL
to quantify the potential environmental harm of chemical
releases.
The overarching goal of this workshop was to evaluate the
potential uses of QSAR and VFAR to advance the rapid and
efficient evaluation of chemicals and microbes of potential
concern. To achieve this goal, the QSAR/VFAR workshop
convened toxicologists, microbiologists, chemists, engineers,
biostatisticians, pharmacologists, biochemists, and risk
assessment scientists to discuss the state of the science,
opportunities for advancement, and practical applications.
Expert panel members included researchers with expertise
ranging from microbial genomics to computational toxicology
and risk assessment. The workshop also included EPA
scientists with expertise in the development and application of
QSAR and VFAR. To facilitate discussion at the workshop, a
list of charge questions was made available to the expert panel
and the workshop participants.

Specific Workshop Goals
NRMRL
NRMRL will use the outcome of the QSAR/VFAR Workshop
to inform its research in developing QSARs in a number
of ways: validating the importance of QSAR research,
providing guidance for QSAR development, and providing
a vision for the future role of QSAR in a regulatory context.
The diverse group of participants and panel members
that attended the workshop, which included researchers
from EPA Program Offices, the European Unions' (EU)
Commission involved with the Registration, Evaluation,
and Authorisation of Chemicals (REACH), other federal
agencies, nongovernmental organizations, industry, and
academia, validate the importance of continuing with QSAR
research. Based on discussions at the workshop, it is apparent
that QSAR has an important role in the future of chemical
regulation and industry both here in the U.S. and in the
EU. With development of new technological areas, such as
bioinformatics  (which includes genomics, proteomics, and
metabonomics), there has been some question as to whether
or not QSAR research has a useful future.

NHSRC
NHSRC will use the outcome of the QSAR/VFAR workshop
to streamline its current QSAR research and to initiate its
VFAR research. Currently, the Center is developing several
QSAR models for predicting acute, subacute, and subchronic
benchmarks to  address exposure durations that are of key
importance during an emergency event, terrorist attack, or
natural disaster. The VFAR method is being explored to
determine the hazards associated with exposure to highly
potent pathogens. Since little is known about the VFAR
methodology, this workshop will allow the Center to define
VFAR and to assess the state of the science. To aid the
discussion process, the Center, in collaboration with NRMRL,
developed a set of strategically and technically sound charge
questions aimed at key aspects of QSAR and VFAR methods.
The discussions on these  charge questions will set the
stage for future QSAR and VFAR research at NHSRC
and NRMRL.

Background
It has long been recognized that chemical substances with
sufficiently similar structures and chemical activities exert
similar qualitative toxicities with differing magnitudes
(Ashby and Tennant 1988; DHHS 1980; Gray and Ostby
1993; Harada et al.  1992; Lewis et al. 1993; Lowell et al.
1989; Rosenkranz and Klopman 1989; Weisburger 1979;
Weisburger and Fiala 1979). Thus, analysis of molecular
structures and physicochemical properties of chemical
substances can  provide a  rapid means of predicting and
quantifying the toxicity of minimally tested chemicals.
This fundamental observation is the basis for the qualitative
structure-activity  relationship (QSAR) method of toxicity
analysis. The QSAR method of toxicity analysis assumes
that a sufficiently strong structure-activity relationship
of chemical substances is indicative of qualitative and
quantitative similarity in toxicity (EPA 1994,  1992, 1989).
Consequently, the long-term toxicities of  minimally tested
congeners of a  chemical series or type can be estimated
from those of better-known congeners on the basis of
available information (EPA 1989, 1992; Rosenkranz and
Klopman 1989; Weisburger and Fiala 1978). Hence, the
QSAR method  provides a means by which the toxicity of a
candidate chemical substance, for which adequate toxicity

-------
data for risk assessment is not available, can be reasonable
inferred from those of a toxicologically better-known
structurally and chemically related surrogate chemical
substance or congener. This method of analysis is intended
to establish a  qualitative and quantitative association between
the  structure-activity and toxicity of a candidate chemical
substance and that of the surrogate chemical substance to
enable quantitative estimates of the toxicity of the
minimally tested candidate chemical substance.
QSAR estimates allow for the prioritization of such
chemical substances for more costly and time-consuming
toxicological  testing or for setting cleanup or media health-
based limits in the regulation of minimally tested chemical
substances. This method of toxicity assessment is especially
useful for many environmental toxicants for which there
is a critical lack of adequate toxicity and pharmacokinetics
data for risk assessment. The use  of the QSAR method for
the  evaluation and establishment of interim or provisional
toxicity references for chemical substances for which there
are  inadequate toxicity data  for use in risk assessment is an
EPA-approved practice and  the basis of the data found in the
EPA's Assessment Tools for the Evaluation of Risk (ASTER)
(EPA 1994). This method has also been used by EPA and
internationally (under the auspices of the North Atlantic
Treaty Organization's Committee on Challenges of Modern
Society [NATO/CCMS]) to  develop toxicity equivalency
factors (TEF) for chlorinated dibenzo-/>-dioxins and -
dibenzofurans (CDDs and CDFs) (EPA-TEF/87,1-TEF/89),
wherein the toxicity of 2,3,7,8-tetrachloro-dibenzo-/>-dioxin
(2,3,7,8-TCDD)  is used as the central reference with which
the  toxicity of all the other CDDs and CDFs are qualitatively
and quantitatively related (EPA 1989). EPA has proposed a
similar QSAR approach for  polycyclic aromatic hydrocarbons
(PAHs) for which there is inadequate toxicity information as
described by EPA (1992).
With thousands of chemicals representing potential
environmental contaminants, the need for a framework of
effective  prioritization for regulatory development and risk
characterization is vital. EPA's Contaminant Candidate List
(CCL) for drinking water contaminants represents one type of
framework, though the selection of chemicals for the CCL has
remained problematic due to the large number of chemicals
that must be evaluated. Similar challenges are posed by
microbes of potential concern. Methods to identify and
prioritize these microbes in  anticipation of potential health
threats from environmental and intentional releases remains a
critical unresolved dilemma.
QSARs are based on the relationship between the structure of
chemicals and their interaction with biological tissues, leading
to adverse effects, whereas VFARs extend this concept to
microbial contaminants, suggesting that the pathogenicity
of a microbial agent is directly related to the architectural
and biochemical components found in that organism. Both
QSAR and VFAR have the potential not only to facilitate the
prioritization  of chemicals and microbes of potential concern,
but  also to inform the subsequent risk assessment and risk
management process.
The practice of risk assessment, which is composed of hazard
identification, exposure assessment, dose response or toxicity
assessment, and risk characterization generally integrates
data from in vivo, in vitro, and epidemiological studies in the
characterization of human health risk assessment. However,
as previously mentioned, the chemical universe is large with
the majority of these chemicals lacking traditional toxicity
measurements. In such instances, risk characterization can
integrate data from in silica methods in combination with
in vitro, in vivo, and epidemiological studies to develop the
weight of evidence (WOE) in the characterization for human
health risk assessment. Thus, the theme that was advanced
during the workshop was that additional useful evidence can
be provided to enhance the overall hazard identification and
toxicity assessment based on QSAR and VFAR methodology.
Additionally, the analysis of microbial virulence factors
can provide critical information for identifying sources of
biological exposure and contamination.

Major  Themes Discussed
Throughout the workshop, the following themes were
discussed by experts in the VFAR and QSAR fields. The
QSAR concept, used for chemical toxicity prediction, is
more mature than the VFAR concept, which may be used for
assessing hazard from  exposure to  microorganisms. In an
introductory presentation on VFARs, Dr. Gerard Stelma of the
National Exposure Research Laboratory (NERL) explained
that the VFAR concept is related to the architectural and
biochemical components of microorganisms that are defined
by both genes and proteins. These components are related
to pathogenicity and human health risks as presented in
a series of National Research Council (NRC) meetings
(1999, 2001). As indicated by Dr. Joan Rose, Michigan
State University, and Principal Investigator of EPA and the
Department of Homeland Security (DHS)-funded Center
for Advancing Microbial Risk Assessment (CAMRA), the
challenges of integrating potential applications of VFAR
into the  risk assessment framework is under intense scrutiny.
Dr. Syed Hashsham, Associate Professor in the Department
of Civil and Environmental Engineering and the Center for
Microbial Ecology at Michigan State University, explained
how descriptors for the microbial genome and proteome can
be applied to evaluate  the impact of variability on microbial
virulence, broadly defined as the ability of a microbial agent
to infect its human host, reproduce, and/or cause disease.
Dr. R. Paul Schaudies  of Science Applications International
Corporation (SAIC) then described a rapid technique for
identifying variability  in the microbial genome. The technique
can be readily used for identification and hazard assessment.
Although the science of QSAR is more mature than that of
VFAR, there remain important challenges in the application
of QSAR to the risk assessment paradigm. One of these
challenges is the determination of how mode of action (MOA)
data can be employed  more fully to improve prediction
of toxicity benchmarks using QSAR. Dr. Mark Cronin,
Professor of Predictive Toxicology, Liverpool John Moores
University, addressed a challenge that has puzzled researchers
for years: the identification of a method for quantifying the

-------
structure-activity relationship of highly reactive electrophiles.
Dr. Kannan Krishnan, Director of the Human Toxicology
Research Group at the Universite de Montreal, described
a model that integrates QSAR with physiologically based
pharmacokinetic (PBPK) modeling to derive extrapolation
capabilities. This model can be adjusted for variations in
exposure route, rate, duration, and other factors. Dr. Andrew
Maier, Associate Director of Toxicology Excellence for
Risk Assessment (TERA), discussed how QSAR could
provide critical information in risk characterization based
on the WOE approach. In his talk, Dr. Maier emphasized
that an understanding of the chemical MOA can enhance the
applicability of QSAR in risk assessment. Dr. William Welsh,
Department of Pharmacology, University of Medicine &
Dentistry of New Jersey (UMDNJ), discussed the assemblage
of a wide variety of computational toxicology tools, including
QSAR-based methodologies that are applicable to risk
assessment. Dr. Andrew Worth, leader of the QSAR Project,
European Chemicals Bureau, Institute for Health & Consumer
Protection, Joint Research Centre, European Commission,
described how the new REACH legislation, which
incorporates a preference for alternatives to animal testing, is
serving to promote QSAR research and applications.

Charge to the Expert Panel
Following the presentations by the expert panels, a number of
questions charged to the panel were discussed.  Each charge
question under VFAR and QSAR,  given below, is followed by
highlights of considerations by the panel members.
VFAR
1. Identify selection criteria for virulence factors that should
   be considered in the VFAR approach. Should certain
   classes of virulence factors be excluded?
      The initial development of VFAR methodology and
      technology allows for a very broad array of gene
      identification. Thus, there is no need to omit any classes
      of virulence factors (VFs) from consideration and no
      reason to  rule out anything until it can be demonstrated
      that it is not relevant. The presence of a VF may be
      necessary but not sufficient for the development of
      pathogenicity. Other factors, such as those that permit
      the expression of VFs, the survival and persistence of
      the microbes, or even a particular array of microbes in
      the environment, are needed to permit the development
      of pathogenicity or the occurrence of disease. There is
      also an  urgent need to characterize background  levels of
      common VFs in organisms to better recognize a change
      in conditions that may pose a human health risk.

2. Compare and contrast the VFAR and QSAR approaches.
   Considering the similarities to QSAR, should the VFAR
   approach work with biotoxins? Viruses? Spores? Cysts?
   What are the strengths of the VFAR concept?
      Host-specific factors (e.g., individual variability in
      metabolism, sensitive subpopulations, the immune
      response of the host) alter the dose-response relationship
      in all traditional toxicity testing protocols. Therefore,
     there will always be uncertainty associated with such
     factors, which will be extended to QSAR and VFAR
     modeling efforts. Due to the variability of individual
     immune system function, host-specific factors are more
     important for microbial agents than for chemical agents.
     However, these limitations should not be a deterrent
     for using these approaches in the evaluation of the
     universe of chemical and microbial agents that need
     to be assessed using nontraditional methods. Because
     of some commonalities between biotoxins, viruses,
     spores and cysts, the VFAR approach may be useful in
     assessing the hazards associated with these different
     forms of biological agents. For the initial prioritization
     of chemicals or microbial agents, when toxicological
     or empirical data are lacking, QSAR and VFAR can be
     particularly useful.
     The data being collected and models under development
     could be critical to facilitating a rapid response in the
     event of an intentional attack. Available empirical data
     could be linked to predictions regarding virulence
     and potential adverse outcomes. QSAR and VFAR
     can provide  critical information regarding alerts to
     human health concerns, and chemical and biological
     plausibility in terms of potential human health effects,
     particularly as an input to comprehensive WOE
     approaches.

3.  Discuss how VFARs can be  used in the detection of
   recognized biothreat agents, newly emerging pathogens,
   and bioengineered pathogens.
     It is unlikely that VFs will be the focus of genetic
     engineering for the purpose of bioweapon development.
     However, the analysis of VFs can provide information
     regarding genetic engineering for both bioweapons and
     natural evolution. In addition, it should be noted that
     other characteristics, such as factors that enhance gene
     expression or environmental persistence, will also play
     a role in exposure and risk.

4.  Describe technology available for examining virulence
   factors. How can we determine the presence of such
   virulence factors  in water or air?
     There are many tools and technologies available for
     examining virulence factors, including genomics and
     gene arrays, polymerase chain reaction (PCR), and
     proteomics for the analysis  of protein products. These
     technologies are all under development in terms of
     applicability to VFARs, but there are current limitations
     in terms of the identification and characterization of
     VFs that have a hazard associated with them and the
     background occurrence of VFs. Difficulties in sample
     collection and processing still exist and must be
     addressed before these technologies can be applied to
     surveillance in water or air.

-------
5.  Discuss the positive and negative applications of using
   VFARs in bioengineering. Discuss the construction
   of highly potent pathogens inserting single gene or
   combinations of virulence genes into commensal
   organisms. Do certain classes of virulence genes lend
   themselves to genetic engineering?
     The genetic changes that occur naturally are an excellent
     example of the ingenuity of the microbial genome. Most
     notably, microbes can transfer plasmids, resulting in the
     rapid exchange of genetic material. Because virulence
     mechanisms are not completely understood, there is a
     need to look for unusual combinations of genes, as well
     as other factors, such as gene arrays and genes that are
     either up- or down-regulated. In general, a change in
     potency is accompanied by a string of changes, not just
     a single change.

6.  How can VFARs be used to determine the human toxicity
   potential of the virulent genes? Is it possible to obtain
   a quantitative estimate of the virulence along with a
   qualitative estimate?
   • For the purposes of public health protection, the goal
     is to be able to use VFARs to aid in the:
   • Identification of the presence of microbes of concern
   • Identification of accessory genes necessary for virulence
   • Identification of environmental conditions  necessary
     for virulence
   • Extrapolation from virulence gene expression to
     virulence protein expression
   • Prediction of the magnitude of the health hazard it
     represents
   • Determination of the infectivity or dose-response
     relationships to gauge the response needed to prevent
     or mitigate an outbreak
These characterizations and predictions would provide
information critical to an understanding of the magnitude of
the public health risk associated with a natural or intentional
exposure event.

7.  Can a virulence gene be altered so that it is still active but
   no longer detectable by the gene probes that are typically
   used?
     The current state of knowledge is focused on the
     identification of virulence factors and how the virulence
     factors function in the microbe to express virulence.
     The capability does not yet exist to link this information
     to health outcomes, though the potential clearly exists.
     Due to the degeneracy of the genetic code, alterations in
     the gene might not result in a change in the synthesized
     protein. Constant changes in the  microbial genome
     necessitate surveillance for these genetic mutations and
     evaluation of how virulence is affected.
QSAR
1.  In light of emerging technologies (e.g., genomics,
   proteomics, and bioinformatics), what role will QSAR
   methods play in the future with regard to EPA's risk
   assessment/risk management process?
     For the purposes of regulatory prioritization and the
     development of remedial action strategies, the universe
     of chemicals must be characterized and reduced to
     assess the chemical threats to human health. Also, in
     order for chemical characterization to be most effective,
     mechanisms of toxicity or MOA must be determined.
     This is an essential component of expert system based
     structure-activity relationships where the aspect of the
     structure of the chemical that results in a particular
     effect or outcome must be determined. This concept
     can greatly enhance QSAR model development and
     interpretation.

2.  How can genomic, proteomic, and bioinformatic data be
   used in QSAR models? Are there examples where the
   "-omics" technologies in combination with QSAR models
   have proven to be able to predict, both qualitatively and
   quantitatively, acute/chronic toxicity across multiple
   chemical classes?
     In terms of the role of -omics and QSARs in EPA's
     framework for risk assessment, any useful and valid
     information will help decrease uncertainty in the context
     of the overall weight of evidence. Some technologies
     may be better  for screening than for regulatory decision
     making in that these technologies may not be fully
     validated or accepted. -Omics technologies and QSARs
     fit into this category. Currently, -omics technologies
     serve primarily as hazard identification tools by
     providing insight into the chemical's potential MOA.
     Such knowledge can provide  informed interpretation
     of QSARs. The integration of QSARs with -omics
     technologies may allow these complementary
     technologies to reinforce each other. Computational
     toxicologists are working on this integration.

3.  Can QSAR methods be used to reduce the uncertainty in
   extrapolating from acute and short-term benchmarks (such
   as median lethal dose [LD50]) to  subchronic and chronic
   lowest observed adverse effect levels (LOAELs)? What
   are the issues that must be dealt with in order to do this?
     There are distinct challenges in using QSARs to
     inform the extrapolation from acute to chronic effects
     because the critical endpoints are different. If there is
     knowledge about the critical effects and MOA, then
     it may be possible to use QSARs to extrapolate and
     reduce uncertainty. It is possible that there are cases in
     which the critical effects and MOA are the same, such
     that extrapolation using QSARs may be helpful. If
     there is commonality in MOA, then extrapolation from
     acute to chronic is more reasonable, but the rationale
     and the uncertainties must be discussed explicitly.
     Discriminators also can be segregated by MOA.

-------
     Participants acknowledged that many approaches have
     been suggested for evaluation of chemicals that lack
     toxicity data. Some indicated it was possible to take
     the LD50 and divide by several uncertainty factors and
     use this derived dose as a substitute for chronic effects.
     Others assert that since the MOA for acute effects
     is generally different from that of chronic effects, it
     is inappropriate to extrapolate from acute to chronic
     effects for most chemicals.

4.  Since rule-based and expert models are based on
   congeneric groupings of chemicals (i.e., the training set is
   a congeneric data set), how can statistical models, which
   are generally based on non-congeneric training sets, be
   improved? Can such models incorporate MOA data if
   available? Can such statistical models provide some insight
   regarding MOA for a chemical query?
     There are several opportunities to combine QSARs and
     MOA information to better inform  risk assessment,
     and the panel noted that routine acceptance of QSAR
     predictions will likely require that they be derived
     with an underlying mechanistic understanding. As
     models become more sophisticated, they will further
     incorporate  structural features and property features
     and therefore allow for fuller evaluation of chemicals
     through the  consideration of MOA data. Several
     examples of developments in this area were described.
     The integration of QSARs with PBPK modeling
     was discussed, where MOA considerations (e.g.,
     identification of appropriate dose metrics based on
     chemical metabolism prediction) are factored into the
     PBPK model. Growing use of tools in bioinformatics
     (e.g., protein structure prediction and libraries) have
     allowed for  the use of shape signatures based on the
     comparison of surface features to integrate MOA (e.g.,
     receptor  binding) into QSAR methodology. MOA
     data can  be  applied to large groups of chemicals to
     identify clusters of closely related chemicals. This is
     the conceptual basis for decision tree and regression
     tree approaches. QSAR models can be tailored via
     selection of descriptors for each cluster to provide more
     uniform  training sets for QSAR development or to aid
     in interpreting global QSAR predictions.

5.  The toxicity of a chemical for any given health endpoint
   is in general due to an adverse interaction between the
   chemical and/or its metabolite and the tissue/organ/DNA
   associated with the endpoint. In  developing statistically
   based QSAR models for chemicals with different modes
   of action, the descriptor pool contains descriptors that are
   chemical specific (i.e., they depend on the structure of the
   chemical alone). Are there any descriptors that can describe
   the tissue/organ/DNA characteristics and its interaction
   with a chemical and/or its metabolites?
     The focus of QSARs is on describing the potential
     interaction between chemicals and  biological molecules.
     There are two basic types of chemical-biological
     interactions. Receptor-based interactions often are the
     basis of endocrine disruption effects, and covalent
     interactions occur with nonspecific macromolecular
     binding. Mechanistic QSARs for predicting receptor-
     based interactions are commonly used in drug
     development and are increasingly being used for
     toxicity prediction. Nevertheless, many chemicals act
     via relatively nonspecific covalent interactions, which
     can be quite complex even within a chemical class, as
     was highlighted in the context of phenolic electrophiles.
     To be most useful, QSARs need to account for this
     complexity  more fully. While mechanistic QSARs
     are preferred, an intermediate step in this direction is
     to focus efforts on endpoint-specific QSARs, since
     the specificity of target organs can arise based on
     local metabolism or the nature of cell/tissue response
     (toxicodynamics).

6.  Current methodology on the statistically based QSAR
   development for toxicity prediction calls for the inclusion
   of as many (classes of)  descriptors in the descriptor
   pool as possible to explain the variance  in the dependent
   variables (some measure of toxicity). In developing these
   QSARs, are there any (class of) descriptors that one should
   definitely include in the potential descriptor pool (e.g.,
   partition coefficients to  account for transfer from blood to
   tissue)?
     Although certain descriptors (i.e., molecular size
     or hydrophobicity) are more commonly used, the
     mechanistic context  must be used as  a starting point
     for the selection of descriptors. Since the mechanistic
     context varies based on chemical class, it is not possible
     to make blanket statements regarding the selection
     of descriptors. Examples of descriptors based on
     chemical mechanisms are those descriptors that describe
     accumulation or penetration through  membranes,
     reactivity with macromolecules, receptor binding with
     critical targets, and others.

7.  Qualitative SAR models (i.e., models yielding
   dichotomous or graded  responses such as yes/no or
   low/med/high) do not provide a quantitative measure
   of a chemical's toxicity while quantitative SAR models
   (i.e., models yielding numerical potency estimates) do
   not provide a qualitative measure of the activity of a
   chemical for any given  health endpoint.  How does the
   panel view the feasibility of applying hybrid QSAR models
   (i.e., capitalizing on the benefits of SAR and QSAR by
   minimizing the disadvantage, if any, of each approach)
   for toxicity prediction?  If feasible, how  does the panel
   envision EPA applying such models?
     Several approaches for hybrid SAR/QSAR analyses
     were discussed. Approaches ranged from using
     MOA descriptors as  a screening step  for the initial
     classification of chemicals to  help in  interpreting
     global QSARs to direct use of MOA  descriptors in
     developing quantitative endpoint-specific logistic
     regression models. Semiquantitative  QSARs methods
     include decision trees or modifications of this concept

-------
that use parallel sets of decision trees to improve
predictability. Binned chemicals identified through
these tools could serve as endpoint-specific QSAR
training sets or be used to identify characteristics
associated with potency levels for risk assessment
using threshold-of-concern approaches.  For chemical
risk assessment, there is often a need to  extrapolate
from dose-response data based on exposure durations
of less than a lifetime to estimate the effects of lifelong
exposure. Traditionally, for EPA risk assessments, a
default factor of 10 is applied to adjust adverse effect
levels from subchronic (i.e., exposure for roughly
10 percent of the lifetime) to chronic exposure
conditions. This default factor of 10 can be useful for
the extrapolation of subchronic to chronic toxicity;
however, it may be inappropriate for the extrapolation
from acute to chronic exposure because the critical
endpoints are often different and the MOA is different
between acute and chronic exposure. The panel
noted that several correlation approaches have been
developed to address this situation - but these  are not
necessarily QSARs. While QSARs may fully address
this application directly, they can also provide  very
important insights that are used in decisions regarding
such extrapolations. For example, they are used
to predict toxicokinetic parameters (e.g., partition
coefficients or metabolism parameters) that impact
decisions regarding the potential for increased body
burden with longer-duration exposures.  Furthermore,
QSARs can  provide understanding of both acute
and chronic  toxicity mechanisms — which impact
considerations of potential for accumulation of tissue
damage with increased exposure duration.
Recommendations
Several recommendations of near-term applications of VFAR/
QSAR models were discussed by the panel. These include:
•   To advance the applicability of VFAR in real-world
    situations, it is critical to facilitate the characterization of
    samples collected during natural outbreaks of microbial
    diseases. This will permit the identification of background
    levels of VFs and advance understanding of the natural
    evolution of VFs in addition to providing the framework
    to test hypotheses pertaining to the dose-response
    relationships of VFAR.
•   Another potential opportunity for the advancement of
    VFAR research involves the BioWatch Program, which
    consists of continuous sampling at locations across the
    country. This would be an  opportunity for researchers to
    obtain material for the characterization of background
    levels of VFs in urban environments in addition to testing
    hypotheses.
•   The state of the science regarding QSAR modeling
    is considerably more advanced than that of VFAR
    modeling; therefore, the key recommendation for near-
    term applications focused on the integration of MOA
    and PBPK with QSAR models to enhance biological
    applicability.
•   For both VFAR and QSAR, host-specific factors alter
    the dose-response relationship (e.g., human variability in
    metabolism, sensitive subpopulations, immune response
    of the host); therefore, there will always be uncertainty
    in the ability to model host factors. Due to the variability
    in human immune system function, host-specific factors
    are important considerations when evaluating responses
    to microbiological agents.  However, these limitations
    should not be a deterrent for using these approaches in the
    evaluation of the vast universe of chemicals and microbes
    that require attention. For the initial prioritization of
    chemicals or microbes, when toxicological data are
    lacking, QSAR and VFAR can be particularly useful.
•   The data being collected and models under development
    could be critical to facilitating a rapid response in the
    event of an intentional attack by linking field data to
    predictions regarding virulence and potential adverse
    outcomes. QSAR and VFAR can provide critical
    information regarding alerts to human health concern,
    and chemical and biological plausibility in terms of
    potential human health effects—particularly as an input
    to comprehensive WOE approaches.

-------
                                                                                                          1.0
                                                                                     Introduction
1.1  Background
The U.S. Environmental Protection Agency's (EPA's)
National Homeland Security Research Center (NHSRC) and
National Risk Management Research Laboratory (NRMRL)
convened this workshop on June 20-21, 2006, to explore
the development and application of Quantitative Structure-
Activity Relationship (QSAR) and Virulence Factor-Activity
Relationship (VFAR) models in the risk assessment process,
specifically as they relate to homeland security needs
and contamination associated with natural disasters and
accidental or intentional releases. To this end, the workshop
convened toxicologists, microbiologists, chemists, engineers,
biostatisticians, pharmacologists, biochemists, and risk
assessment specialists to address the goals of the workshop.
These goals included the identification of data needs for the
development of quantitative noncancer and cancer models,
that are capable of predicting commonly used toxicity
benchmarks, such as the lowest  observed adverse effect level
(LOAEL), LD50, and benchmark dose (BMD), for various
exposure durations. Of particular importance is the prediction
of benchmarks and health effects associated with acute
and short-term exposure to chemical and biological agents.
The workshop explored the development and application
of VFAR models to estimate the human health effects of
microorganisms and their biological toxins. The workshop
also focused on approaches for incorporating mode of action
(MOA) data in the development or refinement of such
models, including the incorporation of genomic, proteomic,
and metabolomic data. In addition, the workshop addressed
computational and data mining approaches, such as various
regression methods, neural networks, and expert systems for
improving QSAR and VFAR development.
The risk assessment process involves four steps as defined by
the National Academy of Sciences, National Research Council
(NRC, 1983): hazard identification, dose response or toxicity
assessment, exposure assessment, and risk characterization.
Risk management integrates the results of the risk assessment
with other considerations, such as economic or legal concerns,
to reach decisions regarding the need for and practicality
of implementing various risk reduction activities. NHSRC
and NRMRL, both part of EPA's the Office of Research and
Development (ORD), are primarily involved in dose response
or toxicity assessment, and in developing guidance for risk
management. Under these processes, an attempt is made to
understand the toxic properties of individual chemicals as well
as mixtures of chemicals, and develop appropriate guidance
documents. An important goal of research in toxicology is
the prediction of the  toxic potential of chemicals from acute
short-term and long-term chronic exposures.
Globally, the chemical industry  and regulatory agencies such
as EPA spend millions of dollars on testing and assessing the
health risks associated with chemicals. For most chemicals,
the risk assessment process is conducted using limited
experimental data. In such instances, the ability to rapidly
and accurately predict potential health hazards from chemical
exposures is needed. One approach to meeting this need is
the use of nonempirical parameters, which can be calculated
directly from a chemical structure. This can be achieved by
the application of computational toxicology or QSAR models,
which have proven to be both appropriate and useful for many
chemicals. Similar computational toxicology approaches are
also being employed to enhance risk assessment processes
for exposure to microorganisms and their toxins.  This field,
involving the methodologies for deriving VFAR, is emerging
to estimate the health hazards posed by biological agents
via the characterization of proteins, which convey toxicity,
infectivity, pathogenicity and/or virulence. The concept
of VFAR was developed as a way to relate the structural,
architectural, and biochemical components (such as biotoxins)
of a microorganism to its potential to cause human disease.

1.2  Purpose and Goals of the  Workshop
The workshop was conducted to explore the application
of these techniques to the risk assessment (RA) process in
situations where chemical-specific empirical data are either
inadequate or lacking.
The following list details the goals  and objectives of the
workshop:
   •  Identification of data needs for the development of
     quantitative noncancer and cancer models,  including
     models that are capable of predicting benchmarks such
     as LOAEL, LD50, median lethal concentration (LC50),
     BMD, and benchmark concentration (BMC) for various
     exposure durations.
   •  Prediction of benchmarks and health effects associated
     with acute and short-term exposure to chemical and
     biological agents.
   •  Exploration of the feasibility  of developing and
     applying hybrid QSAR models.
   •  Exploration of the development and application of
     VFAR models to estimate the activity of microbial
     agents.
   •  Exploration of the incorporation of genomic,
     proteomic,  and metabolomic  data into QSARs in
     order to incorporate the MOA into QSAR models.
   •  Assessment of the development of models for
     predicting the relative toxicity of the parent
     compound and metabolites for identification
     of the ultimate chemical effectors.

-------
   • Discussion of computational approaches, such
     as various regression methods, genetic algorithm
     descriptor selection techniques, data clustering methods,
     neural networks, and expert systems.

1.3 Charge to the  Expert Panel
Following the presentations by the expert panels, the
following questions were discussed:

VFAR
1.  Identify selection criteria for virulence factors that should
   be considered in the VFAR approach. Should certain
   classes of virulence factors be excluded?
2.  Compare and contrast the VFAR and QSAR approaches.
   Considering the similarities to QSAR, should the VFAR
   approach work with biotoxins? Viruses? Spores? Cysts?
   What are the strengths of the VFAR concept?
3.  Discuss how VFARs can be used in the detection of
   recognized biothreat agents, newly emerging pathogens,
   and bioengineered pathogens?
4.  Describe  technology available for examining virulence
   factors. How can we determine the presence of such
   virulence factors in water or air?
5.  Discuss the positive and negative applications of using
   VFARs in bioengineering. Discuss the construction
   of highly potent pathogens inserting single genes
   or combinations of virulence genes into commensal
   organisms. Do certain classes of virulence genes lend
   themselves to genetic engineering?
6.  How can  VFARs be used to determine the human toxicity
   potential  of the virulent genes? Is it  possible to obtain
   a quantitative  estimate of the virulence along with a
   qualitative estimate?
7.  Can a virulence gene  be altered so that it is still active but
   no longer detectable by the gene probes that are typically
   used?

QSAR
1.  In light of emerging technologies (e.g., genomics,
   proteomics, and bioinformatics), what role will QSAR
   methods play in the future with regard to EPA's risk
   assessment/risk management  process?
2.  How can  genomic, proteomic, and bioinformatics data
   be used in QSAR models? Are there examples where the
   "-omics"  technologies in combination with QSAR models
   have proven to be  able to predict, both qualitatively and
   quantitatively, acute/chronic toxicity across multiple
   chemical  classes?
3.  Can QSAR methods be used to reduce the uncertainty in
   extrapolating from acute and short-term benchmarks (such
   as LD50) to subchronic and chronic LOAELs? What are
   the issues that must be addressed in order to do this?
4.  Since rule-based and expert models are based on
   congeneric groupings of chemicals (i.e., the training set
   is a congeneric data set), how can statistical models that
   are generally based on noncongeneric training sets be
   improved? Can such models incorporate MOA data if
   available? Can such statistical models provide some
   insight regarding MOA for a chemical query?
5.  The toxicity of a chemical for any given health endpoint
   is, in general, due to an adverse interaction between the
   chemical and/or its metabolite and the tissue/organ/DNA
   associated with the endpoint.  In developing statistically
   based QSAR models for chemicals with different modes
   of action, the descriptor pool  contains descriptors that are
   chemical specific (i.e., they depend on the structure of the
   chemical alone). Are there any descriptors that can describe
   the tissue/organ/DNA characteristics and its interaction
   with a chemical and/or its metabolites?
6.  Current methodology on the statistically based QSAR
   development for toxicity prediction calls for the inclusion
   of as many  (classes of) descriptors in the descriptor pool as
   possible to explain the variance in the dependent variables
   (some measure of toxicity). In developing these QSARs,
   are there any (classes of) descriptors that one should
   definitely include in the potential descriptor pool (e.g.,
   partition coefficients to account for transfer from blood to
   tissue)?
7.  Qualitative  SAR models (i.e., models yielding
   dichotomous or graded responses such as yes/no or
   low/med/high) do not provide a quantitative measure
   of a chemical's toxicity while quantitative SAR models
   (i.e., models yielding numerical potency estimates)  do
   not provide a qualitative measure of the  activity of a
   chemical for any given health endpoint.  How does the
   panel view the feasibility of applying hybrid QSAR models
   (i.e., capitalizing on the benefits of SAR and QSAR by
   minimizing the disadvantage, if any, of each approach)
   for toxicity  prediction? If feasible, how does the panel
   envision EPA applying such models?

1.4  Organization of This Report
The remainder of this report is organized as follows:
   •  Chapter  2 presents the background of the workshop's
     sponsoring organizations, NHSRC and NRMRL.
   •  Chapter  3 provides summaries of the  VFAR
     presentations made by expert panelists, including
     ensuing discussions from panelists and other workshop
     participants.
   •  Chapter 4 provides summaries of discussion based on
     charge questions related to the VFAR concept posed to
     the expert panel.
   •  Chapter  5 provides summaries of the  QSAR
     presentations made by expert panelists, including
     ensuing discussions from panelists and other workshop
     participants.
   •  Chapter 6 provides summaries of discussion based on
     charge questions related to the QSAR concept posed to
     the expert panel.

-------
Chapter 7 includes major considerations to which
discussions of the charge questions gave rise.
Chapter 8 provides references mentioned during
presentations on the  QSAR and VFAR concepts.
Appendix A presents a list of workshop speakers.
Appendix B provides "biosketches" of the speakers and
expert panelists.
Appendix C contains a copy of the workshop agenda,
as well as the EPA-distributed flyer for the workshop.
Appendix D provides a list of all workshop attendees.
Appendix E includes copies of all presentation materials
in Microsoft PowerPoint slides.

-------

-------
                                                                                                     2.0
                                 Background  and  Opening  Remarks
This section summarizes the background of the workshop's
sponsors, National Homeland Security Research Center
(NHSRC) and National Risk Management Research
Laboratory (NRMRL). Three main speakers from these two
ORD entities set the stage for the workshop discussions,  and
their presentations are synopsized.

2.1  NHSRC and NRMRL
NHSRC, headquartered in Cincinnati, Ohio, was formed
in 2002. It manages, coordinates, and supports a variety of
research and technical assistance efforts and develops and
delivers reliable, responsive expertise and products based on
scientific research and evaluations of technology. NHSRC's
expertise and products are widely used to prevent, prepare
for, and recover from public health and environmental
emergencies arising from terrorist threats and incidents. The
center provides a management structure that ensures effective
design and oversight of research and facilitates interaction
with EPA program offices and regions, other federal agencies,
the private sector, and research partners. NHSRC's team of
scientists and engineers are dedicated to understanding the
terrorist threat, communicating the risks, and mitigating the
results of attacks. Guided by the roadmap set forth in EPA's
Strategic Plan for Homeland Security, NHSRC ensures rapid
production and distribution of security-related products.
These products include methodologies and tools to support
contaminant detection and characterization, treatment
and decontamination, physical security enhancement, risk
assessment and communication, as well as numerous papers
and technical briefs covering a variety of topics.
The mission of NRMRL is to develop ways to prevent and
reduce pollution of air, land, and water. With headquarters
in Cincinnati, Ohio, and divisions in North Carolina,
Oklahoma, and New Jersey, NRMRLs several hundred
scientists and engineers share the mission to solve a wide
range of environmental challenges in seven research areas:
drinking water protection, air pollution control, contaminated
media remediation, watershed management and protection,
environmental technology verification, technology transfer,
and technology support.

2.2  Opening Presentations
NHSRC and the Workshop Goals
AndyAvel, Assistant Center Director, NHSRC
Mr. Avel stated that in the event of a terrorist attack, both
EPA and the Department of Homeland Security will have
responsibility for cleanup. However, after first responders
leave, EPA will have the primary responsibility for remedial
activities. Since the September 11, 2001, terrorist attacks,
EPA, under a series of Homeland Security Presidential
Directives (HSPDs), has been given specific roles, including
decontamination of buildings, public infrastructure, and public
areas in the event of biological, chemical, or radiological
terror attacks, and protection of the drinking water
infrastructure. NHSRC is organized to address chemical,
biological, and radiological weapons of mass destruction
targeted toward water and the environment. Its primary
focuses include:
   •  Developing detection methods to identify an attack
   •  Developing risk  assessment methodologies to assess,
     characterize risks, and provide guidance for cleanup and
     reentry
   •  Understanding and anticipating chemical and biological
     warfare agents
   •  Incorporating the radiological component
Mr. Avel went on to discuss the need to build on what
has already been done, particularly in terms of cleanup
management. Once contamination occurs, he said, impact
has to be minimized via containment. Once contained,
impacted media must be assessed for potential human
exposure and health risk and must be handled to remove/
reduce  contamination.  Once decontaminated, the removed
hazardous materials or residues must be disposed of, using
the best available control technology (e.g., landfill, thermal
destruction), according to local, state, and federal regulations.
NHSRC has developed a technology verification program
to test claims made by industry regarding the technologies
for managing chemical, biological, or radiological agents.
NHRSC also is expanding its capability for the analysis of
these chemicals and agents and increasing lab capacity in
general. NHSRC is currently collaborating with other parts
of EPA including the Office of Solid Waste and Emergency
Response (OSWER), the National Center for Computational
Toxicology (NCCT), the National Center for Environmental
Assessment (NCEA), the National Exposure Research
Laboratory (NERL), and NRMRL.

Cindy Sonich-Mullin, NHSRC, Director of Threat and
Consequence Assessment Division (TCAD)
Ms. Sonich-Mullin discussed the mission of TCAD's research
program: to become better prepared to respond to threats and
emergency incidents. She stressed the need for rapid response
to specific threats and risks from terrorism. Among TCAD's
goals are:
   •  Adapting and developing risk assessment methods for
     homeland security
   •  Developing tools for responders to access information
   •  Developing cleanup advisory levels and methods for
     achieving cleanup goals or levels

-------
Exposure timeframes ranging from 24 hours, to 30 days, to 2
years are the focus of TCAD's risk assessment efforts. This
is in contrast to acute exposure levels (< 24 hrs) and chronic
exposure levels, which are traditionally based on reference
doses (RfDs) and reference concentrations (RfCs), and are
the traditional focus of the Agency. TC AD envisions that
QSARs will play a significant role in filling data gaps for
this timeframe. There is a need for risk assessment methods
for different exposure scenarios, higher concentrations,
and unknown agents. With very little data and capability,
innovative techniques and approaches are required to develop
a credible risk assessment.
The charge questions are the key to the success for facilitating
development of TCAD's risk assessment capability. The focus
is on short-term exposures, and TCAD is developing and
using QSARs and VFARs to:
   • Extrapolate from either acute or chronic exposures
   • Decrease default uncertainty, traditionally applied in
     the risk assessment process
   • Develop credible or sound cleanup level estimates
     for emerging chemical and biological agents,  in an
     emergency
   • Make these efforts transparent and rapid
NRMRL and the Workshop Goals
Subhas Sikdar, Acting Associate Deputy Director for Health,
NRMRL
Dr. Sikdar noted that NRMRL has been involved with the
computational toxicology initiative and QSAR methodology
from the beginning, which led to the establishment of the
National Center for Computational Toxicology (NCCT).
NRMRL's goal for QSAR methods research is to predict
the environmental outcomes of new chemicals throughout
their life cycles, while working with the NCCT to develop
analytical, computer-based models that decrease the need
for animal testing. Dr. Sikdar reiterated that the goal  of
this workshop  is to enhance QSAR and VFAR activities by
bringing together experts  in the field to discuss progress and
the path forward.

Doug Young, Clean Processes Branch Chief, NRMRL
Dr. Young indicated that NRMRL's computational toxicology
program was involved in the original development of QSARs
and has representation on the current steering  committee
for NCCT. The NRMRL engineering lab is working to
develop risk management solutions, including alternative
solutions such as Life Cycle Assessment and environmental
impact tools. Other categories of interest include quantifying
impacts on human and ecological health by developing/using
toxicological values as indicators. For example, given 2,000
chemicals to rank and prioritize while lacking toxicological
values, new tools and techniques, such as QSARs and
bioinformatics, are necessary to reduce the uncertainty of
estimations. NRMRL is in the early stages of using VFARs
and has a particular interest in having the Water Supply and
Water Resources Division develop VFAR tools to evaluate
recreational and drinking-water quality and potential risks.

-------
                                                                                                        3.0
                                           VFAR   Presentation   Summaries
The following summarizes the presentations made for the
VFAR concept, its use in a risk assessment framework, VFAR
factors related to genomic variability, and a bioinformatics
approach to VFAR. The discussion following each
presentation is also summarized.

3.1  Introduction to the VFAR Concept
Gerard Stelma, Senior Science Advisor, NERL
Summary
The VFAR concept, as Dr. Stelma noted, originated during
a National Research Council (NRC, 2001) meeting. The
resultant report recommended the further development of
the concept with a specific challenge to incorporate  it into
the drinking water program. NRC recommended the concept
of VFARs to assist in the development of the Contaminant
Candidate List (CCL) under the Safe Drinking Water
Act (SDWA), as amended in 1996. SDWArequires that
unregulated contaminants in drinking water be identified,
prioritized, and reviewed by EPA as candidates for regulation.
At the time  of SDWA's reauthorization in 1996, there were no
methods for CCL development and prioritization. The NRC
subsequently developed a framework for the selection of both
CCL and pre-CCL chemicals (2001).
Dr. Stelma stated that because priority chemicals must be
identified to meet SDWA requirements, there is a basic
need to prioritize the universe of unregulated chemical and
biological contaminants. For biological contaminants, the
goal is to explore the feasibility of using VFARs to evaluate
microbes and develop a system that would parallel the
QSAR approach for chemicals. The VFAR approach would
emphasize emerging pathogens by building on evidence
from previous research and developing a list of descriptors
tied to pathogenicity. The idea is to focus on elements tied to
virulence.
Dr. Stelma said that if the possibility exists to characterize
the descriptors (i.e., genes, surface proteins, etc.), then the
descriptors could be used to predict pathogens present in
water. However, as pathogens are dynamic, gene arrays
associated with pathogenic virulence may change over time.
Thus, the use of VFARs for pathogen indication/identification
may require constant updating to keep up with pathogen
evolution. Additionally, there are virulence genes, such as
hemolysins, that are necessary but not sufficient for virulence.
Therefore, assaying multiple genes using a gene array may
be important in determining virulence. The applicability of
VFARs may be limited because the current methodology does
not incorporate host susceptibility, the role of unexpressed
virulence genes, and the effect of virulence factors from  dead
cells. Despite these limitations, VFARs can be an important
tool in the pathogenicity assessment toolbox.
Discussion
Following Dr. Stelma's presentation, it was noted that an
array of genes is often needed for the evaluation of potential
virulence, which is a significant challenge. A workshop
attendee went on to discuss another source of uncertainty of
communal pathogenicity, that essentially some microbes may
require the presence of other microbes to express their own
virulence.

3.2  Using VFAR in a Risk Assessment
      Framework
Joan Rose, Homer Nowlin Endowed Chair for Water
Research, Michigan State University
Summary
Dr. Rose  began by discussing risk assessment as a method
to qualitatively or quantitatively evaluate the potential for
harm from exposure to contaminants or specific hazards.
There are four components to risk assessment: hazard
identification, dose response, exposure assessment, and
risk characterization. Risk assessment principles can be
applied to microbes not only to address natural outbreaks,
but also to address the needs of homeland security. Hazard
identification is the process of identifying the microbe, source
of exposure, and the associated virulence. Microbial genetics
is key to this process. For exposure assessment, the goal is
to quantify exposure concentration, duration, and frequency,
though source identification is also important.  Genetic
elements provide information regarding persistence both in
the environment and during disinfection. Monitoring data,
indicators, and models can also be used to estimate exposure
concentrations.
Dr. Rose  stressed the difficulty posed by obtaining dose-
response  information for microbial agents. Quantification is
important, she said, as is the need to extrapolate from  less
pathogenic to more pathogenic strains, from healthy adults
to sensitive populations, and from high dose to low dose. It
is necessary to measure data in the same units as they  are
measured in the environment. Infectivity, the number of
microbes needed to trigger infection, is another important
characteristic that must be quantified. Infectivity may be
related to virulence, though this is not known for certain.  The
process of risk characterization is the combination of all data
to evaluate health risks. It is in dose response that uncertainty
comes to the forefront.
Dr. Rose  noted the need to characterize the background rate
of gene occurrence. For example, to assess biohazards, the
following must be understood:
   •  Why the genome of some microbes are conserved,
     while others are variable
   •  Why some genomes are host-specific and others are not

-------
   • Why some microbes cause chronic diseases and others
     do not
Other important factors are what controls occurrence,
survival, regrowth, accumulation, attenuation, etc. Dr. Rose
concluded her remarks by suggesting that the focus needs to
be on future applications, though there will be uncertainty, and
the applications may not be appropriate for all microbes.

Discussion
The ensuing discussion focused on the difficulty in
characterizing host-organism interactions. There is a need
to focus on the mechanisms that drive differential responses
in microorganism-related host response. In the past it has
been assumed that differential response was driven by host
variables, although now the focus has been extended to
pathogen factors as well. For example, characteristics such as
the presence of housekeeping genes that enhance persistence,
the expression of specific receptors, or the production of
toxins will become the focus of classification as these
characteristics determine virulence. It also may be necessary
to categorize characteristics in different ways, such as lump
and split techniques, based on health effects.
It was also suggested that the uncertainty that is inherent
to the current application of WAR might be difficult to
accept. There is a need for  qualitative and quantitative
characterization of uncertainty. The understanding of
variables that contribute to persistence, what allows a given
microbe or family of microbes to survive and thrive in certain
environmental conditions, is limited. One meeting participant
noted that the Office of Water has an interest in the rapid
identification of biological contaminants and questioned the
application of the VFAR methodology for rapid screening
since there are inherent uncertainties associated with the
method (as discussed above). The panelists also indicated that
identifying management strategies is essential and ties in with
pathogen discovery  and subsequent application of the VFAR
method for hazard identification purposes. The potential exists
to use well-studied pathogens as a  starting point for a rapid
identification tool. A dual-pronged approach, which focuses
on reducing  uncertainty while simultaneously developing
monitoring/identification concepts, may be important for tool
development and refinement.

3.3  VFAR Factors  Related to Genomic
     Variability
Syed Hashsham, Associate Professor, Department of Civil
and Environmental Engineering and Center for Microbial
Ecology, Michigan State University

Summary
Dr. Hashsham began with a discussion of how the genome,
proteins, and toxins  of microbes can all be characterized via
descriptors and how it is possible to use these descriptors
for ranking and uncertainty analysis today. Virulence genes
are associated with function (e.g., antibiotic resistance,
virulence), and not necessarily microbial identity. Depending
on the genome, variability  can range from 1.6 percent to 20
percent. There are variable genes and variable effects, but
in general, there are correlations between specific genes and
adverse health effects that are worth exploiting. The actual
link between health effects and gene variability is undefined.
There is a need to develop gene-family training sets, which
are groups of related virulence factors (VFs). Training sets can
be used to demonstrate the applicability of models defining
VFAR,  drawing from large data sets of virulence and marker
genes (VMG) that are under development. However, the
link between health effects and gene variability and the
quantification of health effects are key to VMG rankings
and eventual pathogen prioritization. A possible approach to
defining this link may involve tying species and genomic data
to known outbreaks using historical outbreak data.
Dr. Hashsham  noted that it is possible to look at rankings
based on variability within species, length of gene, and
number of virulence genes, to determine whether a gene is
a potential marker. Some  genes are  better markers because
they are more specific than others. Dr. Hashsham explained
that the capacity to map the genome of different strains exists
and common and variable regions can currently be identified.
Genes that are constantly  changing are more likely to be on
plasmids. Fewer changes  are found on certain parts of the
chromosome. This information can be used to understand
which genes are associated with virulence and pathogenicity.
However, all of these changes in descriptors ultimately must
be related to response. Data related to response as a function
of differences in descriptors are deficient and require the
most attention.  For the purposes of  monitoring, gene chips
have been developed that contain the simultaneous genomic
sequences of up to 20 pathogens. It is possible to conduct a
high-throughput real-time polymerase chain reaction (PCR)
to amplify any number of genes of interest, using multiple
probes to ensure that specific virulence markers are identified,
in a manner that is economical. This technology is useful for
monitoring and identifying pathogens because it can target
multiple VFs from each pathogen for enhanced reliability.

Discussion
One meeting participant noted that the small volumes used
in chip development can be a problem; when only picoliter
sample volumes are used  for the chip, not all representative
organisms will be in that small  sample. More work is needed
in terms of sample processing to ensure that the samples are
adequately representative.

3.4  A Bioinformatic Approach to VFAR
      Analysis  and Characterization
R. Paul Schaudies, SAIC

Summary
Dr. Schaudies stressed that the significant challenge in
microbial risk is the rapid characterization of the microbe.
The software program Fast Identification of Genomic Unique
Regions (FIGUR) was developed to characterize microbes
within hours by identifying unique elements within the entire
chromosome. Using DNA microarray technologies, a pattern
can be obtained for an organism that can subsequently be
compared to other organisms in established databases for

-------
the purposes of identification. If the pattern is not present
in the databases, it is possible to determine whether the
pattern is similar to known strains.  From there it is possible
to accomplish empirical generation of a library of these
near-neighbor patterns. In contrast to PCR, amplification
via microarray technology is random (using random
primers, rather than gene-specific primers), which broadens
applicability.
The results, generated by computer software, are color coded
to identify unique and conserved sequences. Hybridizations
on the chip can be included to demonstrate that there is no
cross reactivity between genes. Data also can be filtered by
hybridization cutoffs to focus attention on genes that represent
an appropriate level of similarity.
Dr. Schaudies presented an example of three different species
of Yersinia that are associated with the disease plague. Though
all three species are 95 percent similar at the genomic level,
the VFs differ among the three strains. Thus, it is possible
to begin to develop profiles of VFs that define a species.
Using VF profiles as a filter increases the chance of finding
specific strains. One key feature of this approach is that it
does not require the whole genome, but a part of the genome
(e.g., 1Kb). The data can be analyzed serially to refine the
comparisons, and common factors in each subsequent analysis
can be removed to identify what is unique.

Discussion
This technology may be helpful in understanding the
association between VF and pathogenicity. The identification
of genes that are present within a broad array of genes can
be accomplished within hours rather than days. Predictive
capacity is not currently programmed into this tool; its
development was not funded in the current application.
The discussion focused on the need to test this application
in both clinical and environmental settings to help  determine
research needs. Validation is needed for VFAR in strains
with known differences in virulence. A panelist noted that it
would  be interesting to compare Bacillus cereus strains, which
exhibit variability in pathogenicity (as demonstrated by Dr.
Schaudies), with anthracis strains, which exhibit very little
variability.

-------

-------
                                                                                                        4.0
                                                           VFAR  Charge  Questions
The following summarizes the discussions on charge
questions related to VFAR. Discussion under each charge
question is summarized as themes discussed related to the
question. Each theme is a bullet point under the charge
question followed by the summary discussion of the theme.

4.1  Summary of VFAR Charge
      Questions Discussion
1.  Identify selection criteria for virulence factors that should
   be considered in the VFAR approach. Should certain
   classes of virulence factors be excluded?
   • No virulence factor, or selection criteria, omissions
     should be made at this point in the development of
     the VFAR approach.
Participants noted that the VFAR concept is still in the initial
stages of its scientific development and it is important to
collect as much data  as technologically feasible (within the
economy of scale)  as it may not be possible to go back and
retrieve that data later (e.g., following an outbreak or event).
Selection criteria should not be reduced, particularly at the
outset.
In the initial  development of the VFAR methodology,
there is no need to omit any known or potential VFs from
consideration. Technology allows for a very broad array
of gene identification, which is relatively simple and
inexpensive.  The challenge is  in determining which genes
represent critical VFs. A single virulence factor may require
the expression of multiple genes to be effective. The capacity
to make these types of determinations will come only from
the collection of a large quantity of data pertaining to the
existence and ecology of the VFs. Currently, the library of
known virulence sequences is limited; hence, there is no need
to limit the collection of VF data.
If VFARs are thought to be analogous to QSARs, where
VFARs explore whether specific microbes cause disease, then
tools that are developed should include, to the extent possible,
all known VFARs associated with disease. Subsequently, a
host of microbes can then be prioritized, although the process
of identifying and characterizing VFARs is ongoing and
remains far from the threshold of utility. To develop predictive
capabilities for EPA, Food and Drug Administration (FDA),
and other agencies, there is a need to take theoretical,
empirical, and Bayesian approaches to the analysis of VFAR-
related data,  in conjunction with other predictive techniques.
At this time,  the parameter sensitivities of VFARs are not
known.
It was pointed out that the fate of accumulated data might
depend upon the questions that are asked. Are the questions
related to basic monitoring or source identification?
Development of a database that will aid in the identification
of sources of microbes would facilitate understanding of how
genes combine from different populations (e.g., Zoonotic
transmission) and how this can lead to virulence. Ultimately,
such database development can aid in the prediction of
outcomes and the development of management options.
   • Data gaps on background occurrence of VFARs are
     a current challenge.
If a background sample from the natural environment is
analyzed for the presence of genes representing candidate
VFs, VFs will be found to be present. Therefore, background
conditions need to be better characterized and understood.
Similarly, when the genetic signatures of an organism of
interest are characterized, these sequences will also be found
in background samples.  To interpret the significance of these
genetic signatures, it will be necessary to sample and analyze
the background environment to see how those known genes
correlate. This will help to develop the database using a more
focused approach.
Another consideration raised is that any of the most deadly
bacterial toxins can be engineered into multiple species;
this actually occurs in nature in cases where certain toxins
transcend species.  The categorization of microbes may have
to be rethought. As one attendee asked, is the concern over
species identification, or should the species be characterized
based on potential health effects? For example, if symptoms
characteristic of plague were encountered, would one first
look for toxins associated with  the plague?
   • In VFAR development,  proteomics can be used to
     inform genomics.
Proteins are less conserved for  screening, but it may be useful
to start with proteins and work  back toward the genes. VFARs
are based on the understanding that function follows structure.
There is an obvious role for proteomics in VFAR analysis.
This approach has been used with viruses, where proteins
were characterized first and then characterization moved back
toward the genome. Though protein is less conserved for
screening technologies, collection of more data to identify and
characterize VFs will result in a better understanding of the
protein structures to enable their direct use.
   • A dual-pronged approach, combining short-term
     practical applications using the current knowledge
     base with ongoing research, development, and
     refinement of the methodologies, may work best to
     advance the science of VFARs.
Another consideration raised was the need for short-,
mid-, and long-term goals and approaches. In order to
achieve long-term  goals, all approaches to develop VFARs
should be considered. However, in the short term, with limited
knowledge and limited research funding, it may be helpful
to select a group of genes that are known to  be associated

-------
with pathogenicity and focus on them during development of
limited monitoring and assessment tools. For example, it may
be necessary to focus on a subset of data collected to achieve
specific short-term goals such as development of monitoring
or screening tools. While work proceeds toward short-term
goals, mid- and long-term research can be planned and means
to reduce uncertainty, expand capability and capacity, and
increase applicability can be laid out.
In the short term, one approach is to use what is known to
develop and demonstrate concepts. In the mid-term, data can
be collected to add to what is known and to determine what
works for predictability.  Such approaches can be applied or
tested on the growing body of data.
2. Compare and contrast the VFAR and QSAR approaches.
   Considering the similarities to QSAR, should the VFAR
   approach work with biotoxins? Viruses? Spores? Cysts?
   What are the strengths of the VFAR concept?
There are several common factors for the use of VFARs and
QSARs in risk assessment. For both chemical and biological
threats to human health,  the chemical and microbial universes
need to be characterized and reduced.
For both approaches to be most effective, mechanisms of
toxicity or modes  of action must be determined. This is
an essential component of expert system based structure-
activity relationships, where the  aspect of the structure of the
chemical that results in a particular effect or outcome must
be determined. This concept can greatly enhance QSAR and
VFAR model development and interpretation. In the case
of microbial virulence, the structure may refer to a physical
structure resulting from protein expression and subsequent
processing, carbohydrate metabolism, or genetic coding.
Unlike chemicals, microbes are dynamic. Chemicals may
exhibit different properties in different environments and can
be metabolized in the body, producing a range of metabolites
that may or may not be toxic. Microbes, as living organisms,
can exhibit rapid evolution. The  flexibility of microbes and
viruses, which refers to their ability to transfer genes on
plasmids or into the bacterial chromosome, as well  as their
rapid evolution over short periods of time, present unique
challenges to  the development of a VFAR methodology.
It is challenging to use VFARs and QSARs in dose-response
determinations as  they require large quantities of data derived
from multiple testing approaches. In the near term,  it may
be easier to predict hazards by identifying the potential for
adverse health outcomes and looking to VFARs and QSARs
as more robust tools for screening.
There is a deficiency of tools for rapid or instantaneous
identification of biological organisms for use in emergency
situations. The available tools, which use culture methods
and genetic techniques to identify microbes that are present,
can be used to determine whether illnesses are caused
by intentional events (e.g., Salmonella in the salad bar).
However, these techniques take time. There may be other
characteristics in addition to VFARs, such as factors that
enhance gene expression or environmental persistence, that
indicate the presence of weaponized forms of biological
species or an intentional exposure event. The application
of molecular techniques is likely to be the most sensitive,
specific, and rapid approach.
   •  There are several analogous short-, mid-, and long-
      term goals.
Panelists discussed that in regard to long-term goals, the
current state of VFARs is analogous to that of QSARs
many years ago. VFAR tools for identification are under
development; however, it is possible that scientists are
spending too much time on tool components and not enough
on tool composition. In regard to VFARs in particular, it
is necessary to move beyond the academic arena and test
hypotheses to reveal data gaps.
In the short term, the goal may be to construct a framework
for VFAR analysis rather than to focus on details. Mid- and
long-term goals could focus on details and, using an iterative
approach, make updates and modifications to the framework
of analysis as more data are collected.
It is important to articulate the questions that need to be
answered. For example, the  question of whether the intended
purpose of a VFAR tool development is monitoring,
classification, screening, or risk assessment should be
determined up front. Answering these questions may require
different levels of detail, and different techniques may be
more or less appropriate. It was suggested that proteomics
might be particularly useful for screening, followed by a
search for different genetic signatures that give analogous
structures. Bioinformatics tools can be used to solve the
question of the relevancy of genes and to predict structure-
activity relationships.
There is an advantage  in trying to develop these frameworks
now to identify areas that require research. These concepts
should work for viruses and other organisms. In fact, it
may be advantageous to work with viruses because they are
simpler organisms.
EPA emergency management staff are interested in the
practical applications of these tools and, in particular, in
opportunities for quick detection in the field, especially for
engineered organisms. There is also a need for the scientific
community to  evaluate persistence to determine appropriate
decontamination methods and develop cleanup levels.
As first conceived by the NRC, these questions define the
framework for decision making, particularly with respect to
weaponized agents, which are very different from naturally
occurring agents. Naturally occurring microbes may or may
not aerosolize, while weaponized agents, such as the agent
used in the U.S. Senate anthrax event, are readily aerosolized.
EPA's Office of Water also is particularly interested in the
rapid detection of microbes of concern. It is a challenging
problem that needs a solution. From a public health
standpoint, as one panelist illustrated, the search for a probe
capable of identifying a contaminant and signaling an alarm
prior to consumption of the water would potentially save lives
by eliminating the time lapse required by current detection
technology.

-------
3. Discuss how VFARs can be used in the detection of
   recognized biothreat agents, newly emerging
   pathogens, and bioengineeredpathogens.
   • Analysis of VFs may indicate bioengineering, but
     may not be the sole indicator or focus of engineering.
The analysis of VFs can provide information regarding
genetic engineering for both bioweaponization and for
naturally occurring genetic evolution. However, VFs may
not be the focus of genetic engineering for the purpose of
bioweapon development;  there may be other characteristics
that are altered to increase exposure and risk. For example, a
gene or genes may be altered in a way that allows a microbe
to persist in an environment, which will result in higher
human exposures and lead to increased risk of disease.
Persistence factors traditionally are not considered VFs.
VFARs can be used to identify biothreat agents, newly
emerging pathogens, and  bioengineered pathogens when
applied to a surveillance system. However, it may be difficult
to determine whether an agent was bioengineered based on
VFARs alone. The approach for determining whether an agent
has been bioengineered is classified, although, as panelists
discussed, the approach goes beyond virulence factors to look
at survival, the degree to which the agent can be cultured or
stored, and other nongenetic factors. Genes are only one part
of the equation.
The tools in use for bioengineering may have nothing to do
with virulence; for instance, the bioengineering process may
entail gene manipulation for eliciting a protein.
Similarly, bioengineering  is not necessary for an intentional
attack. For example in 1984, Salmonella was found in a salad
bar in Oregon. It was a commercially available American
Type Culture Collection (ATCC) strain, and initial efforts
focused on determining whether the contamination was
intentional. The use of VFARs or another genetic approach
would not necessarily help in this type of investigation.
   • To use VFARs, the question needs to be denned — is
     it for detection or risk characterization?
Discussion focused on the applicability of VFARs to detection
and risk characterization as well as the overlap between the
two. The emphasis on which elements are most important may
be slightly different. Specifically, for applications pertaining
to the detection of pathogens in the environment, the key
factors of interest may be  the array of VFs that are present.
Other factors that may not be directly related to virulence  are
necessary for identification of pathogenic strains or species,
based on their association with those strains or species. For
quantitative risk assessment, one participant noted the need
for a more rigorous definition of the relationship between
the VFs and health effects. For qualitative applications in
risk characterization, it may be possible to glean significant
information based on the  presence of VFs in the sample.
VFARs can be useful for the BioWatch program, which uses
a series of pathogen detectors co-located with EPA air quality
monitors. Currently, the BioWatch program is based on the
collection of airborne particles on filters, which are removed
and tested using PCR for  the presence of select pathogens
(Shea and Lister, 2003). The development of libraries of
VF markers, coupled with more rapid and economical
technologies, could facilitate the rapid identification of
airborne pathogens of concern. Part of this surveillance
could potentially be used to look for VFs as markers of
bioengineering. The  application of gene array technologies,
rather than PCR, could yield test results within hours or,
possibly, minutes. This type of technology transfer could
fit into the surveillance and characterization of background
conditions and for BioWatch applications.
Perhaps the most important role of VFARs, as has been
identified by the NRC, is to characterize microbes that
cannot be cultured and/or are novel to assess the potential for
pathogenicity. New technologies discussed in this workshop
can aid in the early detection of the presence of pathogenic
microbes.
   •  Bioengineering vs. Nature
The explanation of engineered pathogens is complicated by
the natural rapid changes that  occur in the microbial genome.
Rapid changes in this genome can come about via the transfer
of plasmids. However, in terms of genetic engineering, the
challenge is to get the specific proteins expressed, which
involves the coordination of multiple genes, and is therefore
a very complicated process. VFARs may have greatest
applicability in developing an  approach to screening genes
associated with potential health effects.
   •  What is the definition of the "VFAR Approach"?
There was additional discussion regarding the meaning of the
"VFAR approach" and whether there was a consensus as to
its exact meaning. The VFAR  approach, in its broadest sense,
implies the creation of a database of VFs (descriptors), related
health effects (response), and data analysis tools that relate
and rank the pathogens (mathematical models for VFAR).
Many different tools  are being used to construct this database,
though the VFAR approach should not, at  this time, be
limited to one technology, such as PCR or gene arrays. This
will allow maximum flexibility for developing applications
of VFARs, in terms of structure-activity relationships, that
are parallel to those used in QSARs. While clearly the major
goal is to use the structural relationship to identify and
characterize pathogenicity and subsequent health effects (e.g.,
develop dose-response scenarios), VFARs can also be used
for detection and hazard assessment through the identification
of microbes that pose a potential health risk based on VF
presence.
   •  Can "most important" VFs be defined?
VFs can provide critical information about a pathogen.
However, given the limited knowledge available to the
scientific community today and in light of data gaps, it is still
a challenge to state what the "most important" VFs might be.
As with QSARs, application to dose-response assessment
is still a major challenge. Based on what the scientific
community knows about VFs, one participant encouraged
testing hypotheses of VFAR application to various elements
of the risk assessment paradigm. The participant presented
a current example of the challenges inherent in monitoring

-------
the risks associated with the evolving strains of bird flu.
The participant asked how VFARs can be used as a tool for
understanding the way risks change with the evolution of the
virus. For example, despite all of the research on the Spanish
flu of 1918, few clear-cut answers for explaining its virulence
exist.
Another participant stressed the scientific community's need
to know the biological characteristics that make organisms
virulent. This likely involves VFs in conjunction with other
factors, such as accessory genes and housekeeping genes.
Once the characteristics of virulence are known, technologies
can be developed for an early warning system that could be
applied to both natural and terrorist events.
4. Describe technology available for examining
   virulence factors. How can the presence of such
   virulence factors in water or air be determined?
   •  Focus on technologies for identification and detection
Tools and technologies available for examining virulence
factors include genomics and gene arrays, high-throughput
real-time PCR, and proteomics for the analysis of protein
products. These technologies are constantly under
development. All may be applicable to VFARs, but currently
there are limitations in terms of sample collection and
processing. Limitations include low concentrations in the
environment,  sample processing losses, and  minimum
detection limits associated with the  molecular technologies.
Such issues must be addressed before these technologies can
be applied to surveillance in water or air.
   •  Media-specific sampling issues
There are media-specific problems with extraction of
microbial material for the purpose of analysis and detection.
Samples need to  be processed, prepared, and concentrated.
Water may be the simplest media with which to work. The
greatest challenge is extracting the sample from the media
for analysis. The amount of the sample must be sufficient
for biological and statistical analysis. One panelist stressed
that the scientific community ideally needs to be able to
identify the biological agent in any given volume. To improve
analyses, the sample may be concentrated or subjected to
processing, depending upon the media from which it is taken.
For drinking water, concentration typically is required. For
surface water, some processing is needed in addition to
concentration. Other media samples could require additional
processing due to interference by other media constituents.
The closer the scientific community gets to the source of
contamination, the easier it is to use molecular methods.
Quantitative or reverse transcriptase PCR gives robust
quantification capacity. It is useful for analyzing the presence
of microbes in sewage and ground water, particularly
those that are  nonculturable. However, it requires prior
knowledge of which VFs might be present so that appropriate
oligonucleotide primers will be selected. It is important to
solve sampling issues or look for targeted, specific genes
or organisms. Sensitivity is improved if targeting specific
organisms, but concentration or enrichment is often needed.
As with chemical contaminants, it is difficult to determine
transport, fate, and exposure concentrations. Furthermore,
background levels of genes are not well characterized and will
likely continue to be a problem until better data are available.
Background occurrence of VFs can provide an important
perspective on what constitutes a change in occurrence that
signals the presence of a potential hazard.
Once it is understood what makes organisms unique to
specific environments, it will be possible to target specific
VFs within the organisms to determine whether exposure is
occurring. Additional research and discovery is required to
target for occurrence and exposure.
   • Use of -omics technologies in VFAR development
The current focus in VFAR development is on genomics
because of its sensitivity. Sometimes genomics is overly
sensitive as the gene may be present, but not expressed.
However, if the gene is not available, neither will be the
message or protein.
It may be possible to use proteomics. Proteomics represents a
complementary approach that can be initiated with the protein
product, followed by an examination of the structural features
of importance. Researchers can then work backwards, mining
the  genome for similar genes. With current technologies,
proteomics may be more costly and time-consuming than
other -omics technologies. In addition, there may be limited
database availability for comparing and identifying proteins.
   • VF ranking and application in prioritization and risk
     assessment
Although VFARs may be used in the prioritization of
microbes and in microbial risk assessment, the databases
to support such applications are still being developed. At
the  current time,  it  is not clear how VFs will be used to
rank microbes and for application to risk assessment or
prioritization.
5. Discuss the positive and negative applications of using
   VFARs in bioengineering. Discuss the construction
  of highly potent pathogens inserting single gene or
  combinations of virulence genes into commensal
  organisms. Do certain classes of virulence genes lend
  themselves to genetic engineering?
   • Bioengineering vs. Nature revisited
The changes that occur in the natural environment are an
excellent example of how genetic factors change; however,
genetic engineering is delicate. There are many examples in
which genetic  engineering  resulted in unanticipated results.
Most notably, microbes can transfer plasmids resulting in the
rapid exchange of genetic material.  The growing presence
of antibiotic-resistant bacterial strains is an example. Also,
some members of Burkholderia  (earlier grouped under
Pseudomonads that are generally known to be benign) are
now of major concern to cystic fibrosis patients.
It has been demonstrated that Pseudomonads can be altered
in the laboratory for various engineering applications. The
simplest approach may be to co-culture  organisms to facilitate
the  transfer of plasmids. The presence of genetic material

-------
does not guarantee that it will be expressed. Additional steps,
whether bioengineered or inherent to the microorganism,
are required to translate the genetic code into proteins, and
further modification may be required to ensure that the protein
is functional. One participant recommended caution, citing
an example in which genes were inserted into mouse pox
with the intention of making a better vaccine; however, the
resulting product was lethal.

   •   Factors that change/increase potency
Increases in potency are not always understood. In general,
a change in potency is accompanied by a string of changes,
not just a single change. It might not just be VFs that change
to increase potency. There is a need to  look for unusual
combinations of genes, as well as other factors.
   •   Host-specific effects
Although extrapolation from animal studies introduces
uncertainty,  animal studies are, and will continue to be, an
important avenue of research to identify potential human
health risks. For some pathogens, outbreaks in other hosts
precede infectivity in humans. Therefore, there needs to be an
understanding of specific activity changes both in animals and
in humans. These changes could occur on either the genotypic
or phenotypic level. More importantly, the process of infecting
a host can induce changes in the microbe. For example, in
laboratory studies, passage through mice is frequently used to
increase potency. In laboratory studies, passage through the
animal is sometimes needed to identify new genes.
6. How can VFARs be used to determine the human
   toxicity potential of the virulent genes? Is it possible to
   obtain a quantitative estimate of the virulence along with
   a qualitative estimate?
   •   The predictive capability comes from
      characterization and linkage to known health effects.
Although it may currently be possible to begin to rank gene
sequences, the capacity to link gene sequences to health
effects is still being developed. For the purposes of public
health protection, where it is necessary to gauge the response
needed to prevent or mitigate  an outbreak or reduce endemic
disease, the goal is to be able to use VFARs to aid in the
identification of the presence of microbes of concern, the
prediction of the magnitude of the health hazard represented,
and the determination of the infectivity or dose-response
relationships. The scientific community needs to be able to
answer questions such as, "How many people are likely to be
affected? "
The current state of knowledge is focused on the identification
of virulence factors, and how these virulence factors function
in the microbe to explain virulence. As one panelist noted, the
scientific community does not yet have the capacity to link
this information to health outcomes, however the potential
clearly exists.
An example of how these connections  can be made is
Escherichia coli. The 0157:H7 strain carries Shigella toxin
and is much more virulent than other E. coli strains. By
analyzing the genetics of this strain and comparing it to
other E. Coli strains that lack Shigella toxin, the basis for
strain potency can be developed. The characterization of
these associations will drive the development of hypotheses
regarding VF-activity linkages.
   • Virulence of organisms, not just genes
It is not solely the virulence of genes that is important, but
also the virulence of organisms (i.e., the genes need to be
understood within the context of the organism). This is how
VFs are tied to dose response. There is clearly a relationship
between VF and dose response;  however, dose response is
more highly variable for biological agents than for chemicals.
Factors that contribute to the definition of the dose response
of microbes include:
   • Factors that control infectivity
   • The evasion of the host immune system
   • The ability to  colonize within the host
   • The initiation  of the disease process
For example, poliovirus, in comparison to other disease
viruses, requires high concentrations of the virus to initiate
infectivity. In the case of poliovirus, the disease is not
perpetuated at the site of infection,  as may be the case with
other viruses. The scientific community needs to understand
the relationship between what happens at the site of infection
and where the microbe exerts its health effects.
There are genetic factors that control all of these processes.
The goal is to illustrate the relationship between VFs and dose
response, recognize the complexity in this relationship among
different organisms, and use the relationship as a proxy for the
virulence of the organisms.
As with chemical exposures, variability of individual factors
such as sex, age, and the presence or absence of chronic
conditions, can be an important factor in host response.
Furthermore, the genetic diversity of the immune system
among individuals, which involves somatic mutations in
the development of the specific immune response, increases
the variability. Therefore, individual variability in terms of
host response to biological agents is much broader and more
challenging to characterize than it is for chemical agents.
The process of weaponization can be targeted at altering
factors controlling dose response, including infectivity,
evasion strategies, colonization, and pathogenicity. Successful
bioengineering is not just a matter of altering genes alone.
Gene expression and protein synthesis within the context of
the organism are critical challenges.
   • Importance of exposure pathway and the
     relationship between exposure pathway and
     dose response
As with chemicals, the exposure pathway is an important
determinant of potential health effects. Anthrax exposure
pathways, for example, include dermal absorption,
ingestion, and inhalation, the last of which is the most
potent. However, there is insufficient information on dose-
response relationships via direct dermal and ingestion routes
to  determine the health impacts of anthrax via these routes.

-------
Due to the lack of such dose-response data, anthrax inhalation
dose-response relationship data are extrapolated to produce
estimates for dermal and ingestion exposures. In addition to
naturally pathogenic microorganisms, weaponized forms of
microbes have the potential to alter exposure pathways and
the associated dose-related response.
   • Pathogenicity is based on a complex set of factors,
     some related to the microbe, some related to the host.
One panelist raised the need to define VFs more broadly
because of related factors that confer or enhance virulence,
pathogenicity, and persistence.
Further discussion from participants included several
examples that provide insight into dose-response
relationships, though it was acknowledged that in each
case critical information was missing. For the anthrax
contamination that occurred at the Washington, D.C., post
office in 2001, the mortality rate was 1/20,000 (based on the
exposed population), and only a small number of individuals
became ill. Many of those exposed were treated prior to
showing signs of infection,  so it is not possible to measure
infectivity. In Boca Raton, Florida, in 2001, many locations
within  a building tested positive for anthrax. One person died;
however, no one else became ill even though the anthrax
spores  were presumed to have been present for many days.
One woman in Connecticut died from exposure to anthrax-
contaminated mail. In Philadelphia (1976), Legionnaires'
disease was spread through the ventilation system and many
people died.
In summary, the  relationship between the organism and the
host is  extremely complex. As with chemical contaminants,
there may be a threshold below which infectivity does not
occur, while for others the threshold may be so low that it is
negligible. In general, there is a lack of dose-response data;
hence it is difficult to predict dose response, particularly at
low-level exposures.

7. Can a virulence gene be altered so that It Is still active but
   no longer detectable by the gene probes that are typically
   used?
   • VFs can be altered, but expression is not always
     predictable.
It is possible for  VFs to be altered so that they are still active
but no  longer detectable; however, oligonucleotide primers
can be  made for  PCR and microarrays when alterations cannot
be made without changing function.
Because of the degeneracy of the genetic code, alterations
in the gene may be possible while preserving activity. With
constant changes in the microbial genome, it is necessary to
maintain surveillance for these changes and determine how
they will affect virulence.
It is also possible that subtle changes over time will eventually
affect the protein. The point at which activity actually changes
depends on the organism, the protein, the specific function of
the protein, and its biological interaction.
   •  Specific tools have advantages and disadvantages in
      the identification of VFs.
With carefully designed gene arrays using large numbers
of probes, it may be easy to detect changes in the genome.
Because microarray technology allows for multiple markers
and probes, the probability of detecting genetic changes is
increased as compared to  PCR, which normally detects one
gene target at a time.
To advance the use of VFs in the evaluation of health risks,
discussion indicated the need for multiple VF descriptors.
Gene occurrence and expression should be the initial
descriptor. In addition, participants pointed out a need to
characterize exposure routes (i.e., ingestion, inhalation, and
dermal contact), survival and persistence, and attenuation
in the environment. In addition, algorithms that relate genes
to function need to be developed. With an initial focus on
the use of VFARs to conduct quick screening, available
data can be used to test the applicability of known VFs.
However, researchers will need to develop  computer models
to determine the sensitivity of specific descriptors and the
correlation of the descriptors to endpoints of concern.
It is possible that the scientific community has sufficient
data to begin to develop a proof of concept that would take
available data and demonstrate its applicability to detection,
hazard identification, dose-response assessment,  and risk
characterization. For detection, a collection of VFs could be
applied to predict the presence of pathogenic organisms in
unknown samples. The analysis could also  include predictions
regarding potential sources. For risk assessment, the VFs
could be used to qualitatively predict  pathogenicity or health
effects from the unknown samples. Although the results may
not be fully accurate, these types of exercises could identify
data gaps and help prioritize research to advance the field.

4.2  VFAR Closing Remarks
Factors that relate the virulence of microbes to adverse
health effects should be determined. There  are factors that
control the ability of the microorganism to persist in a given
environment, infect a host, evade the  host's immune system,
colonize within the host, and then initiate the disease process.
These factors should be characterized. Factors may include
receptor proteins, binding proteins, invasion capability, or
toxin production. They may include components that aid in
survival under different circumstances (e.g., in the presence
of ultraviolet light or commonly used disinfectants such as
chlorine). Many tools are available to characterize these
factors, though more exist in the area  of genomics than in
proteomics.
The challenge lies in evaluating genes and proteins within
the context of the organisms and their ecology. There are
also important considerations regarding the manipulation of
genes for the purposes of bioterrorism. In developing this
understanding, the scientific  community will be better able
to identify and prioritize microbes to ensure the protection of
human health.

-------
VFARs also have the potential to provide important
information for risk assessment. The identification of
factors, genes, or proteins that confer an advantage to
the microbe, which impacts pathogenicity, can assist with
hazard identification and priority ranking, as well as the
characterization of dose-response relationships under
different exposure scenarios. Although panel members
presented impressive current advances, there is a need to
collect more data and develop analytical  algorithms as the
concept moves forward.
To advance the understanding of VFARs within the
context of the microbes' ecology, researchers need to
make an attempt to collect data during outbreak conditions.
Doing so will help identify factors that were important in the
outbreak, who will be affected by illness and why, the dose-
response relationship, etc.
Given what is known now, there are opportunities to begin to
test the concept of VFAR application. Although initial efforts
will be challenging, they will help to identify critical data
gaps for a more comprehensive study. Future efforts should
begin with a broad definition of VFARs as factors that confer
an advantage to organisms for their survival and success,
identify background levels of known VFs, and track changes
in the microbial community.
The focus should continue to be on the development of
a set of tools, based on molecular techniques that can be
used in the short- or medium-term to facilitate scanning for
VFs. The level of stringency can be varied to collect a large
amount of information in a short period of time, resulting in
algorithm generation and analysis of the data to understand
pathogenicity.
While it is expected that VFARs can help to prioritize
microbes for the CCL, the existing datasets are not sufficiently
robust for this application at the present time. However, since
the concept is sound, development and testing of hypotheses
to advance the science should be initiated.

-------

-------
                                                                                                        5.0
                                          QSAR   Presentation   Summaries
The following summarizes the presentations made for
the QSAR concept. These include the integration of
physiologically based pharmacokinetic (PBPK) modeling
with QSAR models to reduce uncertainties in the chemical
risk assessment process, the use of MOA and WOE in
predictive toxicity, the application of reactivity as a
descriptor in the development of more accurate QSAR
models, a discussion of innovative and varied approaches
to QSAR model development, and the role of a regulatory
agency in advancing the development and implementation
of QSARs. The discussion following each presentation is
also summarized at the end of each presentation summary.

5.1  From Reactivity to Regulation: Integrating
      Alternative Techniques to  Predict Toxicity
Mark Cronin, Professor of Predictive  Toxicology, Liverpool
JohnMoores University

Summary
Dr. Cronin began by describing the challenge reactive
electrophilic compounds have posed to toxicologists in
terms of identifying descriptors that accurately define their
parameters and quantify their characteristics. Electrophilic
chemicals are highly reactive and extremely toxic.
Conventional QSAR methods consistently under-predict
toxicity for this group of chemicals. Dr. Cronin stated that
by using an enzyme assay, it is now possible to quantify
electrophilicity to predict reactivity in biological systems.
The assay is based on the chemical reaction with  glutathione
(GSH). There is a strong correlation between cytotoxicity and
GSH reactivity. Quantification is based on the measurement
of the reactivity index. Reactivity works well as a descriptor
to rank a group of related chemicals based on this mode of
action, however, it is still a challenge to translate  this into
a usable tool. As there is a spectrum of electrophiles, the
first step is to define the domain, correlate it with toxicity,
and model it. Dr. Cronin said that this process is expected
to be particularly valuable under REACH and has direct
application to regulatory issues. However, it will still require
the use of multiple tools to characterize risk, and  as with
all chemicals, Dr. Cronin conceded, it is still a challenge to
quantify uncertainty. The initial focus in the development of
this process is to develop a model for fish toxicity and skin
sensitization.

Discussion
It was noted that reactivity works well as a descriptor in
ranking a group of related chemicals based on this MOA,
but challenges remain to translate this into a usable tool.
Other chemicals with different modes of action will require
different descriptors.
Reactive chemicals may be metabolites, although, as
workshop participants discussed, this is not currently the
focus of the research. It should be possible to include a model
of metabolism prior to GSH reaction.

5.2  Integrated QSAR - PBPK Modeling for
      Risk Assessment
Kannan Krishnan, Director of the Human Toxicology
Research Group (TOXHUM),  Universite de Montreal.

Summary
Dr. Krishnan stated that based on the risk assessment
paradigm, animal toxicity testing is evaluated to determine no
observed adverse effect levels (NOAELs) for the derivation
of risk-based criteria. QSARs can be used to predict the
differential responses based on variation of chemical
substitutents, but they are context specific and dependent
on exposure route, rate, duration, etc. When conditions are
varied, different QSARs need to be derived or extrapolations
need to be made. The goal is improving derivation or
extrapolation capabilities, and as Dr. Krishnan emphasized,
integrating QSARs with PBPK modeling can do just that.
PBPK models facilitate extrapolations of one of the two
key components for the exposure-response relationship:
pharmacokinetics (PK), representing external dose to internal
dose; and pharmacodynamics (PD), representing tissue
dose to effect. As components of dose response, PK and
PD both can be related  to QSARs to enhance extrapolation
capability. Since there are more data available for PK, the
focus of Dr. Krishnan's research is on the development of
QSARs for PK profiles that change as a function of species,
exposure route, dose, and duration. In the QSAR, given a set
of related chemicals, the model begins with an administered
dose and calculates changes in blood concentrations with
chemical substituent changes. The model uses an easy-to-
use spreadsheet to test how kinetics change with the related
class of chemicals (using VOCs as a test case). The user
enters chemical structure and duration of exposure into the
spreadsheet to estimate tissue  exposures, which will aid
in the estimation of toxicity. The program will also allow
modifications of exposure concentrations, routes, and
exposure scenarios to evaluate how these changes impact
tissue dose.

Discussion
Following Dr. Krishnan's presentation, discussion focused
on the development of the PD component that incorporates
MOA, which is in the early development stage. With the
inclusion of MOA, prediction of effects should be possible.
Gene microarray data, or other data relating to gene
expression,  cannot yet be incorporated.

-------
For the physiological component the input variables are
volumes, only. Partition coefficients in various tissues have
been derived via in vitro testing. Tissue cultures, such as
data from a liver slice can be used for QSAR development,
though the scale of tissue levels must be increased for dose
considerations.

5.3  Weight of Evidence and Mode  of Action
      in Predictive Toxicology
Andrew Maier, Associate Director,  TERA
Summary
Dr. Maier began by stating that the importance of QSARs
is growing, in part, due to the incorporation of the  concept
into risk assessment methods using a WOE approach.
WOE emphasizes decision making based on the totality of
toxicological evidence. The WOE concept is being driven
by improved biological understanding, such as knowledge
of MOA, the increasing sophistication and validation of
alternative study designs, and several quantitative tools,
including SARs and QSARs. In the QSAR field, the
consensus modeling concept embodies WOE principles.
While WOE approaches use QSARs as an input for decision
making, application of the results of the WOE can also be
used as feedback in an iterative way to enhance the SAR  and
QSAR models. Another possibility for enhancing the QSAR
concept is to link SARs and QSARs via the integration of
MOA data. For many chemicals, the  detailed mechanism  of
toxicity is not known, though the MOA data are available
for a number of chemical classes. In lieu of waiting for a  full
mechanistic understanding, which will rarely be available,
research should capitalize on the degree of biological
understanding available to  refine QSAR approaches. Several
approaches for accomplishing this objective are available. On
the simplest level, MOA data (including -omics data) provide
a tool for interpreting the outputs of global QSAR methods.
In addition, MOA data can be used to separate chemical
groups using qualitative or quantitative decision-analysis
approaches as an initial step in developing endpoint (or MOA-
specific) QSAR models. MOA parameters can be used as
chemical descriptors in building logistic regression models.

Discussion
Participants concluded that there is a need to consolidate  what
is known about chemical MOA to allow researchers to rank
and prioritize their ability to integrate biology with QSAR.
Biomarkers, particularly early effect biomarkers, can be useful
in understanding the MOA for enhancing QSAR development.
Both genomics and proteomics can be used as tools for MOA
identification to aid in QSAR development or interpretation.
These -omics technologies are also complementary with
SARs and QSARs for reaching WOE conclusions for risk
assessment.
5.4  Novel Approaches to QSAR and
      VFAR Modeling
William Welsh, Norman H. Edelman Professor in
Bioinformatics and Computer-Aided Molecular Design,
Department of Pharmacology, University of Medicine &
Dentistry of New Jersey (UMDNJ)
Summary
Dr. Welsh stated that no one QSAR fits all, and that one
way of dealing with this is by integrating consensus
modeling, experimental data, bioinformatics, and -omics
into WOE decision making.  The New Jersey Environmental
Bioinformatics and Computational Toxicology Center
is developing computational toxicology tools such as
Dose-Response Information Analysis System (DORIAN).
There also are numerous chemical toxicology tools under
development, including the following QSAR-based
approaches:
   • Decision forest, which makes predictions and evaluates
     prediction confidence
   • Shape signatures, used for large-scale screening based
     on similarity in three-dimensional shape and bio-
     relevant surface properties
   • Polynomial Neural Network (PNN), developing
     linear and nonlinear QSAR models
   • Virtual High-Throughput Screening (VHTS) to assess
     the binding affinity of small-molecule compounds inside
     the positive binding pocket of protein receptors
The goal is  to develop new methods that work in concert
with established ones, while developing a hierarchy of
strategies. The hierarchy will begin with fast, easy-to-use
tools, such as structural filters and alerts, and then proceed to
more computationally demanding tools such as classification
models, followed by segregation using  chemical activity.  If
the compounds are active, they will be selected for additional
study. For example,  within the decision forest, each tree
includes a series of descriptors that segregate chemicals
into active or inactive compounds. As the descriptors are
independent, this results in consensus predictions. Each
branch of the decision tree represents an "if-then" formatted
query, thereby allowing for rapid evaluation. The shape-
signature model begins with the molecule or receptor pocket.
Shape and biorelevant features are converted into compact
shape signatures for comparison. A data bank containing the
shape signatures of greater than 5 million small-molecule
compounds is then used to compare and contrast these
features. A separate  data bank for screening contains the
shape signatures of more than 5,000 ligands extracted from
the high-resolution X-ray crystal structures of proteins found
in the publicly available protein data bank (PDB). The data
bank is a repository  for  protein  crystalline structures, and
there is a library for screening. This process allows for the
explanation of mechanistic clues of a molecule with an
unknown MOA through comparison with chemicals in this
PDB-extracted data  bank of protein ligands. In theory, this
process can also be applied to chemicals or proteins from
bacteria of interest, such as Escherichia coli.

-------
Discussion
The challenge is to move from structure to function. The next
generation of shape signatures will tackle this problem, for
example, distinguishing a receptor antagonist from an agonist.
The traditional application of three-dimensional QSARs
requires subjective molecular alignment for the comparison
of structures. With the new shape signatures program,
the comparison is rotationally invariant, which removes
subjectivity.
Other properties besides shape that can be used in the
comparison include surface charge, polarity, hydrogen
bonding capability, or any property that is mappable on the
external structure  of a molecule. Shape coupled with polarity
has proven to work well.
In this schematic,  there is a conformation generator and
clustering tool that can generate multiple  conformers and
compare them using shape signatures. It may be more efficient
to compare the shape signatures of clusters of conformers for
a single molecule  with multiple degrees of freedom. Shape
signature works for a wide variety of molecular entities,
including organic, inorganic, and organometallic molecules;
neutral or  charged species; proteins; and even nanoparticles.

5.5  Role of the European Chemicals
      Bureau in Promoting the Regulatory
      Implementation of  Estimation Methods
Andrew Worth, European Chemicals Bureau, Institute for
Health & Consumer Protection, Joint Research Centre,
European  Commission

Summary
The implementation of REACH legislation will depend on
the efficient evaluation of chemicals of concern, using QSARs
and methods for grouping chemicals. Authorities require that
companies demonstrate the safe use of their chemicals. The
WOE approach is needed, and animal testing is used only as
a last resort. The focus is on developing the WOE approach
by means of integrated testing strategies. If the model is
scientifically validated and applicable to substances of
interest, QSARs can be used for the purposes of classification
and labeling and/or risk assessment, provided there is
adequate and reliable documentation. The category approach
can be used to group chemicals according to chemical
similarity (e.g., structural properties, three-dimensional
structure) to avoid the need to test every member of the group
for every endpoint. Certain conditions apply; if categories
are too large, it may not be applicable for every chemical,
but the concept of subcategories is foreseen. The European
Chemicals Bureau (ECB) is currently developing a guidance
document on the use of grouping methods, including insights
from the current practices of EU regulators, and introducing
new approaches, such as computational toxicology and other
new methods. All of the ECB's guidance development (which
includes many other guidance documents for REACH) is
conducted to be transparent to regulated industry, thereby
permitting access to and use of the most advanced state of the
science  in preparing submissions for new chemicals.  ECB is
also building an online inventory of publicly available models,
intended to  be useful to EU industry and the future Chemicals
Agency. The current emphasis is on model validation,
documentation, consensus building, and capacity building.

Discussion
The adaptation of standard information  requirements and the
replacement of traditional test data using QSARs, reactivity
data, -omics, etc. is a priority under REACH as a means of
reducing animal testing. Integrated testing strategies based
on a WOE approach will be used to combine the use  of
multiple approaches. Gaining consensus among industry
organizations and 25 EU members on methods and
approaches for risk assessment is extremely challenging.
To this end, QSARs must be scientifically validated and
applicable to substance (s) of interest for the purposes
of classification and labeling and/or risk assessment.  In
addition, adequate and reliable documentation must exist.

-------

-------
                                                                                                        6.0
                                                          QSAR  Charge  Questions
The following summarizes the discussions on charge
questions given to the expert panel. Discussion under each
charge question is summarized as themes related to the
question. Each theme is a bullet point under the charge
question, followed by the summary discussion of the theme.

6.1  Summary of QSAR Charge  Questions
      Discussions
1.  In light of emerging technologies (e.g., genomics,
   proteomics, and bioinformatics), what role will QSAR
   methods play in the future with regard to EPA's risk
   assessment/risk management process?
   • It is important to have multiple tools for the
     evaluation of chemical toxicity.
Participants expressed that any useful and valid information
obtained through the application of emerging technologies
will help to decrease uncertainty in the context of the overall
weight of evidence. Genomics can aid in the identification
of the MOA. For chemical reactivity, it is useful to have a
genomic and proteomic fingerprint of the chemical since
the genomic fingerprint may offer insight into a chemical's
MOA. Some technologies may be better for screening than for
regulatory decision making because they  may be more readily
validated, accepted, etc. Currently, the integration of QSARs
with -omics technologies will result in an iterative approach,
whereby these complementary technologies reinforce each
other. Computational toxicologists are working on this
integration to serve primarily as a hazard identification tool
by providing insight  into the potential chemical's MOA. Such
knowledge can provide informed interpretation of QSARs.
There are several opportunities to combine QSARs and
MOA information to better  inform risk assessment, and
members of the panel noted that routine acceptance of QSAR
predictions will likely require that they be derived with an
underlying mechanistic understanding. As models become
more sophisticated, they will incorporate  nontraditional
structural features and property features and, therefore, allow
for evaluating chemicals completely through the consideration
of MOA data. Several examples of developments in this
area were described.  The  integration of QSARs with PBPK
modeling was discussed, wherein MOA considerations (e.g.,
identification of appropriate dose metrics based on chemical
metabolism prediction) are factored into the PBPK model.
The growing use  of tools  in bioinformatics (e.g., protein
structure prediction and libraries) has allowed for the use
of shape signatures based on the comparison of surface
features to integrate MOA (e.g., receptor binding) into QSAR
methodology. MOA data can be applied to larger groups
of chemicals to identify clusters of more closely related
chemicals. This is the conceptual basis for decision tree and
regression tree approaches.  QSAR models can be tailored
via selection of descriptors for each cluster to provide more
uniform training sets for QSAR development or aid in
interpreting global QSAR predictions.
It was noted that -omics data have been applied in
pharmacology and toxicology for the purposes  of drug
discovery, prognostic and diagnostic methods, biological
pharmacological activity, and the toxicity-based landmark
studies of John Weinstein (e.g., Bussey et al. 2006, Nishizuka
et al. 2003, Blower et al.  2002). The foundation paper on this
subject (Blower et al. 2002) reviewed the linkage between
chemical and -omics technologies. However, there are
inherent uncertainties in -omics technologies as well, in terms
of interlaboratory variability and chip-to-chip errors. Judicious
interpretation remains important in the use of -omics data
as a supplement to or as an input into QSAR development.
The field of single nucleotide polymorphisms (SNPs) is an
exciting area of development that could provide information
for QSARs. QSARs can also be used to understand -omics
and focus on critical variables. This would, in turn, promote
development of QSARs for critical molecules.
Currently, -omics data are not used directly as the primary
basis for EPA risk assessment decisions, though they can lend
support to the overall descriptions of toxicity mechanisms and
are part of the Agency's risk assessment documents. In the
EU, there is a placeholder in REACH legislation for the use of
alternative methods, such as -omics technologies, either alone
or in combination with other methods. -Omics approaches
have yet to be standardized so that they are reproducible, and
the need exists currently to categorize, document, and define
these approaches.
   •  One size does not fit all.
Although the identification of a single technology for all
chemical evaluation would greatly streamline the risk
assessment process, no single technology can provide the
necessary information for all chemicals. REACH requires
consensus building and acceptance among industry and
regulators. Toxicologists are often in a position in which
they must explain that although QSARs may be easy to use,
expertise and judgment are needed in the interpretation of the
results. Increasingly, the concepts of consensus modeling and
WOE are being incorporated into risk assessment guidance in
recognition that no single technology is likely to provide all
the answers.
   •  Although QSARs can play an important role in risk
     assessment, there is a need to consider the WOE to
     evaluate chemicals.
Panelists noted that the reliability of nontraditional risk
assessment methods needs to be quantified. Even with
95 percent accuracy,  the consequence of using incorrect
predictions needs to be carefully assessed considering the

-------
large number of chemicals that must be evaluated. When
applied to human health outcomes, tolerance for uncertainty
is very small and accuracy must increase to higher levels,
such as >99 percent. All technologies have limitations, and
no single method will provide all the data needed. With
-omics, as with QSARs, there are layers of uncertainty —
measurement error, unexpected patterns — and it may be very
difficult to interpret the results. Altered gene expression does
not necessarily mean that there is an effect. Most up and down
regulated changes in genes are attributable to housekeeping
genes. A fusion of technologies is needed and is occurring.
Another panelist stressed the need to work together, maintain
skepticism for all technologies, and verify the results, by
considering the WOE, rather than focusing on one technology.
There is a need to be transparent when communicating
how the conclusions of a hazard/risk assessment depend on
underlying results and the methods used to generate those
results.

2.  How can genomics, proteomics, and bioinformatics data
   be used in QSAR models? Are there examples where the
   -omics technologies in combination with QSAR models
   have proven to be able to predict, both qualitatively and
   quantitatively, acute/chronic toxicity across multiple
   chemical classes?
   •  QSARs and -omics technologies are complementary
     and can be used to reinforce or refine estimates of
     toxicity.
It was reiterated that -omics  data have been applied along
with QSARs in pharmacology and toxicology for the purposes
of drug discovery, prognostic and diagnostic methods,
biological pharmacological activity, and toxicity assessment
for many years. The integration of QSARs with -omics
technologies will result in an iterative approach, whereby
these complementary technologies reinforce each other.
Computational toxicologists are working on this integration.
One example of such iterative use of these technologies is
that -omics data can help explain MOA and mechanisms of
toxicity, which can then serve as inputs for defining QSAR
parameters, building more closely aligned training sets or
explaining variability in model predictions. Furthermore, data
from -omics can be used as descriptors  in QSARs. In theory,
it should be both possible and useful to use data from -omics
research as descriptors in QSARs. MOA descriptors may be
informed by genomics and proteomics. Caution is needed in
the use of genomics because genes that are transcribed may
not necessarily be translated into functional proteins (e.g.,
post-translational modification). Proteomics data may provide
more directly relevant information, but the experimental
methods are more cumbersome. Metabolomics  may fit more
readily with the use of QSARs, but this growing area has not
yet been fully explored in the context of QSAR application.
   •  QSARs and -omics technologies can be particularly
     useful for hazard assessment.
QSAR models can more readily predict a potential toxic
outcome, which is equivalent to predicting hazard. If
researchers are trying to develop correlations between
exposure and hazard using -omics as an endpoint or outcome,
this is a feasible approach. In other words, -omics can be
used as biomarkers of exposure to identify hazard, which will
feed into other elements of the risk assessment paradigm. A
recent publication (Ekins et al. 2005), reviewed the use of
Absorption, Distribution, Metabolism, and Excretion (ADME)
and drug metabolism software to build in toxicogenomics,
proteomics, metabonomics, and pharmacogenomics, using a
systems biology approach. This is an example of integration
that may work for chemical toxicology hazard assessment. To
date, potency estimates (i.e., dose-response estimates) based
on -omics have not yet been defined; hence, they have not
been widely used in QSARs. -Omics data, therefore, remain
largely a tool for MOA or hazard identification.

3.  Since rule-based and expert models are based on
   congeneric groupings  of chemicals (i.e., the training set
   is a congeneric data set), how can statistical models,
   which are generally based on noncongeneric training set,
   be improved? Can such models incorporate MOA data
   if available? Can such statistical models provide some
   insight regarding MOA for a chemical query?
   • Examples where MOA can be integrated into QSARs
Panelists noted that, as discussed in an earlier presentation,
QSARs can be integrated with PBPK modeling,  where MOA
is factored into the PBPK  model. In addition, the use of shape
signatures allows for  the comparison of surface features and
integrates MOA (e.g., receptor binding) into the  methodology.
QSAR models are sophisticated, incorporating structural
and property features. However, they should be sufficiently
flexible to add MOA considerations directly into the model
for chemical evaluation. Alternatively, given a large group
of chemicals, one approach is to develop and apply MOA-
based tools to subdivide chemicals into clusters.  Global
QSAR models can be developed for a variety of chemicals,
or QSAR models can be tailored via selection of chemical
clusters belonging to  a certain chemical classification.
Expertise is needed to make these decisions. For developing
class- or cluster-based models, strict descriptor definitions
are required for that class  or cluster. A mechanism or MOA-
based approach can be used to define these descriptors,
but this can be challenging. Nevertheless, this approach
has been successfully applied. For example, Knaak et al.
(2004) integrated physicochemical and biological data for
the development of predictive QSARs and PBPK models
for organophosphate pesticides. In ecotoxicology, there have
been examples where mechanisms of toxicity were generated
from QSAR data. In addition, work has been published on
the cytotoxicity of phenols, assigning modes of action and
mechanisms of toxicity on the basis of QSARs (Schultz et al.
1997, Croninetal. 2002).
   • QSARs for ecotoxicology are more widely accepted
     than in human health.
It is easier to validate QSAR descriptors by experimentation
in ecotoxicology than in human toxicology. There are existing
databases (e.g., from  studies in the EPA laboratories in
Duluth) that facilitate QSAR development for aquatic toxicity

-------
endpoints. Comparable databases for prediction of human
health effects are sparse and not readily available. In addition,
there is a lack of mechanistic data for the often more complex
human health endpoints than for ecotoxicology endpoints.
This is due, in part, to the existence of more mechanisms of
action in human health  endpoints. In the EU, an attempt was
made to use a simpler classification of chemicals based on 17
different modes of action; however, the  classification proved
to be quite complex.
   •  Development of  appropriate and meaningful
      chemical grouping techniques requires knowledge
      of the model's purpose.
There are thousands of  descriptors available for each
chemical. These descriptors, such as molecular weight and
number of carbon atoms, can be physically meaningful or they
can be physically uninterpretable constructs based on graph
theory. To enhance the biological meaning for analysis, it is
important to select methods that identify descriptors that are
biologically meaningful and defensible. Developing QSARs
based solely on statistical identification carries the potential
risk of developing circumstantial correlations that may be
highly predictive but biologically meaningless.
   •  With the advent of toxicogenomics, the transfer of
      this technology to computational toxicology should
      help us understand the potential effects of chemicals
      on sensitive populations.
Efforts are under way in pharmacology  and toxicology to
understand the interaction between variations in the  human
genome and variability  in response to understanding how
individual variability impacts chemical toxicity and risk
assessment. Mechanistic QSARs can help define variations in
chemical structure or properties that impact interactions with
polymorphic receptors or xenobiotic metabolizing enzymes.

4. The toxicity of a chemical for any given health endpoint
   is, in general, due to an adverse interaction between the
   chemical and/or its metabolite and the tissue/organ/DNA
   associated with the endpoint. In developing statistically
   based QSAR models for chemicals with different modes
   of action, the descriptor pool contains descriptors that
   are chemical specific (i.e., they depend on the structure
   of the chemical alone). Are there any descriptors that
   can describe the tissue/organ/DNA characteristics and its
   interaction with a chemical and/or its metabolites?
   •  QSARs focus on describing the potential interaction
      between chemicals and biological molecules.
There are two basic types of chemical-biological interactions.
Receptor-based interactions often are the basis of endocrine
disruption effects, and covalent interactions occur with
nonspecific macromolecular binding. The latter are relatively
nonspecific, but it is useful to focus on covalent interactions
and characterize their diversity. This illustrates why  endpoint-
specific QSARs are useful. The specificity of target organs,
where metabolism generally occurs, or the nature of cell/tissue
type, provokes a reaction. Tissues introduce repair capacity,
buffer capacity, etc., which modulates effects.
   • Metabolism is one of the keys to predictive success.
Ideally, descriptors should relate to the toxic moiety (parent or
metabolite). In cases where the toxic moiety is a metabolite,
consideration of tissue characteristics related to metabolism
(e.g., the presence of relevant metabolizing enzymes) can
enhance model predictivity. For most endpoints, descriptors
are not available to include relevant tissue characteristics.
Statistical QSARs may implicitly include metabolism;
however, metabolism will be correlated to various structural
features. Software has been developed  (Madden and Cronin,
in press) to aid in the prediction of metabolites, although
there are limitations inherent in the software. However, when
applying such metabolism prediction models, biological
understanding is still required to identify the metabolites
associated with toxicity.
Complementary Ligand Based  Receptor Interaction is a
type of descriptor that considers the ligand and the receptor
docking or binding.  Researchers can use this descriptor and
then screen potential ligands  against known ligands. There are
also models that account for the electron properties that map
to the surface of DNA to model the binding of transcription
factors to DNA.
   • Advancement of models that incorporate MOA and
     health effects data
Although the pharmaceutical industry has  been using
mechanistic QSARs for years, these  often  have limited
applicability outside the specific receptor or molecular
endpoint being studied. Furthermore, much of the advanced
work is proprietary. In terms  of global QSARs, commercial
software is available, but  in many  cases the underlying
algorithms or databases are not transparent. Currently, there
are research initiatives in  chemical informatics (e.g., at
Rensselaer University) to improve public domain data and
modeling as well as software techniques. There are also
nonpharmaceutical industry models available.  Therefore,
alternative approaches are needed  to advance QSAR model
applications that incorporate  MOA and health effects data.
There are published examples of QSAR development in the
literature pertaining to organic chemicals and human health.
(Beliveau et al. 2005, Beliveau and Krishnan 2003, Waller et
al. 1996). There are also examples in ecotoxicology; however,
the endpoints are not likely to be highly relevant to human
health  (e.g., lethality).
Tissue  microarrays are used in medical diagnostics to
determine anticancer therapies  and in the testing of drug
cocktails, and these data are transferable to toxicological
applications. Access to tissues from repositories would be
required to generate experimental data from which QSAR
models could be developed.
Physical and chemical descriptors  can be used to predict
interactions with a biological target as a pharmacodynamic
(PD) approach. The scientific community needs to identify
additional PD descriptors, although they may already be
correlated, resulting in unnecessary redundancy. For example,
given a QSAR model for  breast cancer, researchers can add
PD factors, including endocrine receptor (ER)  binding and

-------
prolactin release. The goal, essentially, is to model a series
of steps that define a complex event.
   • Ratio of descriptors to compounds
As a rule of thumb, the number of descriptors should be
limited to 1 descriptor for 5 compounds. Thus, given 40
compounds, there should be no more than 8 descriptors. In
the selection of descriptors, less is better. The QSAR equation
describes a mathematical relationship that maps the target
based on the descriptors. Descriptors may be correlated to the
endpoint being predicted, but this does not indicate a causal
relationship. In other words, a statistically derived QSAR may
not be related to the pertinent MOA but may still accurately
describe the relationship. To derive meaning from these types
of descriptors may result in over-interpreting the model. In
addition, since the QSAR models are mathematical equations
— regardless of the chemical structure — the equations will
predict some response. This is inconsistent with biological
knowledge, where many chemicals will have no meaningful
effect on certain endpoints. To overcome these problems
in model parameter definition, approaches for selecting a
few descriptors that may be most relevant from the MOA
standpoint have been suggested.

5. Current methodology on the statistically based QSAR
   development for toxicity prediction calls for the inclusion
   of as many (classes of) descriptors in the descriptor pool
   as possible to explain the variance in the dependent
   variables (some measure of toxicity). In developing these
   QSARs, are there any (classes of) descriptors that one
   should definitely include in the potential descriptor pool
   (e.g., partition  coefficients to account for transfer from
   blood to tissue) ?
   • When selecting descriptors, start with the
     mechanistic context.
Although certain descriptors are commonly used, the use of
the mechanistic context as a starting point for the selection of
descriptors is advisable. Since the mechanistic context varies
based on chemical class, it is not possible to make blanket
statements regarding the selection of descriptors. Examples
of descriptors based on chemical mechanisms are factors that
describe accumulation in a certain tissue (hydrophobicity),
reactivity, receptor binding, etc. As described in an earlier
presentation, modeling reactivity  is difficult, and it is easy
to miss subtleties. For  example, given a reactive group on
an aliphatic compound, if a stearic group is added near
the reactive site, there  will be stearic hindrance that is not
captured using conventional descriptors. However, novel
types of descriptors (e.g., atom-based fragments and certain
fingerprint-based descriptors) may capture this information.
To develop QSAR models, branching of groups that
incorporate mechanistic-based descriptors may be needed
to ensure that the molecule geometry has been adequately
interpreted.
   • Graph theory as an alternative to define QSARs
The use of graph theory to define QSAR descriptors is an
important alternative when information about a chemical
is lacking. These descriptors are easy to compute and are
not subject to variability. One of the attractions in the use
of descriptors derived from graph theory is that they are not
conformation dependent, so researchers do not need to know
anything about the conformation. However, these descriptors
may not have any obvious mechanistic interpretation.
   • Integration of ligand-receptor interactions in QSAR
     models
A recent evaluation of ligand-receptor interactions found
significant differences in the properties of ligands. A recent
article in the Journal of Medicinal Chemistry (Vol. 49,
3451-3453 [2006]) by a group of researchers from Ely-Lilly
evaluated ligands that bind to different classes of proteins,
such as kinases, nuclear receptors, and G protein-coupled
receptors (GPCRs). They found differences in the properties
of ligands. It would be interesting if the query were posed,
"What is the target tissue?" and then have the software
determine what descriptors have been successful for similar
applications. There are instances where compounds were run
against a panel of receptor proteins at single concentrations,
creating a biospectrum of binding affinity. This can be used to
characterize ligands.
Another option is to develop LOAEL models that are specific
for specific endpoints, leading to the development of a suite of
QSARs based on mechanistic considerations.
   • What is the status of the QSARs field (exploration
     vs. comparability and refinement)?
Panelists noted that the answer to this question depends on
where the researcher is in the life cycle of a given QSAR
model. In regard to the development of QSARs, the initial
stage can be characterized as exploratory — the gathering of
data to develop correlations between chemical structure and
outcome. As the field matures, models that are developed
for different sets of related chemicals can be compared and
refined. Most of the available models are in the comparability/
refinement stage. As the available models continue to be
advanced, the expectations for QSARs are very high. As more
ideas are developed, models will be able to incorporate more
complex biological considerations.

6.  Qualitative SAR models (i.e., models yielding
   dichotomous or graded responses such as yes/no or
   low/med/high) do not provide a quantitative measure
   of a chemical's toxicity while quantitative SAR models
   (i.e., models yielding numerical potency estimates) do not
   provide a qualitative measure of the activity of a chemical
  for any given health endpoint. How does the panel
   view the feasibility of applying hybrid QSAR models
   (i.e., capitalizing on the benefits of SAR and QSAR by
   minimizing the disadvantage, if any, of each approach)
  for toxicity prediction? If feasible, how does the panel
   envision EPA applying such models?
   • Qualitative analyses can be useful for the purpose of
     comparing chemicals.
Qualitative SAR comparisons may be useful for the hazard
identification of chemicals with very little toxicity data.

-------
Other qualitative analyses of SARs require the subjective
classification of toxicity (e.g., low, medium, high), and
there may be no biological significance. The context must
be considered. It is better to have biologically meaningful
classifications based on measurable biological events.
Hybrid analyses can also be applied; one type of hybrid
analysis could begin with the initial classification based on
MOA, followed by the application of the QSAR model.
Semiquantitative QSARs can take  the form of regression
trees,  using decision logic to inform the interpretation, such as
binning or identifying the threshold of concern. There are also
models based on quantitative relationships.
As experimental techniques improve, very low levels
of toxicity (e.g., ER activity) can be measured. In some
instances, QSARs are expected to quantify activity at such a
low level that it is beyond the sensitivity of the model.

7.  Can QSAR methods be used to reduce the uncertainty
   in extrapolating from acute and short-term benchmarks
   (such as LD5) to subchronic and chronic LOAELs?
   What are the issues that must be addressed in order to do
   this?
   •   There are distinct challenges in using QSARs to
      inform the extrapolation from acute to chronic
      effects because the critical endpoints are different.
The extrapolation of subchronic to chronic exposure
is frequently based on Haber's law, which states that
concentration times duration is a constant, and this gives
a ratio of exposure duration of about 10 (e.g.,  in rodents
800 days/90 days, which provides a rational definition for
the extrapolation value of 10). This can be useful for the
extrapolation of subchronic to chronic toxicity;  however, it
is inappropriate for the extrapolation from acute to chronic
exposure because the critical endpoints are often different.
Also,  the MOA is different between acute and chronic
exposure.
   •   If there is knowledge about the critical effects, and
      MOA, then it may be possible to use QSARs for
      extrapolation and reducing uncertainty.
It is possible that there are cases where the critical effects
and MOA are the same, such that extrapolation using
QSARs may be helpful. If there  is  commonality in MOA,
then extrapolation from acute to chronic is more reasonable,
but the rationale and the uncertainties must be discussed
explicitly. Discriminators also can  be segregated by MOA.
In particular, this may work for noncumulative reversible
effects. If the target tissue and MOA are the same, this forms
a basis to build an extrapolation  algorithm. This information
can also help decrease uncertainty. For instance, the default
duration uncertainty factor in the EU is 100 (whereas it is
10 in the U.S.). Information from QSARs has been used to
decrease the uncertainty factor. In other cases, it has been
found that a safety factor of 100 is not adequately protective
(research by Jan Ahlers, German Environmental Protection
Agency).
There have been attempts to build models for PCBs,
which tend to bioaccumulate, and the model incorporates
the accelerating effects of the chemical over time. These
models have had mixed success.
Ultimately, extrapolating data may be more of a science
policy decision. Health Canada will not use acute data to
derive subchronic or chronic values.
   •  There is a critical need to evaluate thousands of
     chemicals that have no toxicity data, and all options
     should be evaluated.
Participants acknowledged that many approaches have been
suggested for evaluation of chemicals that lack toxicity
data. Some think it is possible to take LD5Qs and divide by
some number and use this derived dose as a substitute for
chronic effects. Others assert that since the MOA for acute
effect is generally different from that of chronic effects, it is
inappropriate to extrapolate from acute to chronic effects for
most chemicals. In addition, communicating risk to the public
can be challenging when acute to chronic extrapolations are
performed.
The process of determining which chemicals should be on
the CCL requires consideration of not only potency, but
also severity. Using LD50s does not seem to fit well into this
paradigm.
Correlations have been done using regression analysis,
primarily as a first-tier approach.  This must be followed by
an assessment of what is known about the chemical of interest
and whether there are characteristics that can be used to make
predictions. In short, expert judgment is required.
Regardless of the methods used, transparency, communication
of assumptions, domain of applicability, and communication
of uncertainties are critical components of any risk
characterization.
   •  The probabilistic derivation of QSARs could make
     better use of the available data.
The application of probabilistic techniques for QSARs
is feasible, using a range of data for each input. This may
actually represent the best use of the available data. Monte
Carlo approaches can then be used to generate a range of
QSARs.
Also, in some databases, there are a number of measurements
for any given endpoint. To develop a QSAR, decisions must
be made on the selection of input values. Some may take the
most conservative value, while other approaches will consider
using the average. Guidance is extremely limited, and
transparency is critical.

6.2  QSAR Closing Remarks
It is important to have multiple tools to use for the evaluation
of chemical toxicity. The characterization of MOA can
provide critical information regarding chemical toxicity. For
chemical reactivity or cytotoxicity, it can be useful to have
a genomic fingerprint of the response to the chemical to
determine whether the observed effects represent different

-------
gene-based responses. Some technologies may be better for
screening than for regulatory decision making because they
may be more readily validated, accepted, etc. An iterative
approach between QSARs and -omic technologies can be
used to reduce the uncertainties in each. In effect, a validated
QSAR can be used to reduce uncertainty in -omic approaches
and vice versa.
In theory, it should be possible to use -omics research data
as descriptors in QSARs. MOA descriptors may be informed
by genomics and proteomics. Caution is needed in the use of
genomics because upregulated genes may not be expressed
or functional (e.g., post-translational modification, etc.).
Proteomics may provide more relevant information, and
metabolomics may fit more readily with QSARs, but this has
not yet been attempted. In addition, participants pointed out a
need to consider dosing issues.
The integration of QSARs with PBPK modeling, in which
MOA is factored into the overall framework, is possible and
useful. The use of shape signatures allows for the comparison
of surface features and integrates MOA (e.g., receptor
binding) into the methodology. Models are much more
sophisticated, incorporating structural features and property
features; therefore, they should allow for more flexibility in
evaluating chemicals by adding MOA considerations. A large
group of chemicals can be subdivided into clusters. QSAR
models can be developed globally for all chemicals in all
clusters, or they can be tailored via selection of descriptors for
each cluster.
The QSAR equation describes  a mathematical relationship
that maps the health endpoint to descriptors. Descriptors may
be circumstantial. They may not be related to MOA, but they
may be able accurately describe the relationship between
a chemical structure and health endpoint. The use of graph
theory, which is not dependent on conformation or biological
interactions, to define QSAR descriptors is an important
alternative when MOA information about the chemical is
lacking. Nonmechanistic descriptors allow the problem of
data gaps to be bypassed. However, to derive MOA meaning
from these types of descriptors is to risk over-interpreting the
model.
In some instances, the toxicity of metabolites has been
incorporated in the original development of the equation.
Although certain descriptors are more commonly used, a
mechanistic context,  if known, must be used as a starting
point for the selection of descriptors. Since the mechanistic
context varies based on chemical class, it is not possible
to make blanket statements regarding the selection of
descriptors.
A panelist pointed out that a few descriptors could be selected
that may be most relevant from the MOA standpoint. Tissue
characteristics, essentially static or defined, can be built in as
a constant. These descriptors should relate to the metabolite,
and tissue characteristics can be an important factor,
particularly in terms of the concentrations of metabolizing
enzymes, etc. For most endpoints, descriptors are not
yet available to build in tissue characteristics, although
some approaches do implicitly include metabolism. The
pharmacology industry must routinely make predictions
about metabolism in order to predict toxicity and the possible
cellular targets (i.e., DNA, protein, etc.) when selecting
descriptors.
Qualitative applications of QSAR analysis are possible and
can provide important information for hazard identification.
One type of hybrid analysis could be the initial classification
based on MOA, followed by the application of a QSAR
model. Semiquantitative QSARs can take the form
of regression trees, using decision logic to inform the
interpretation, such as binning and identifying the threshold
of concern.

-------
                                                                                                       7.0
           Major  Considerations  and   Recommendations
Discussion of the VFAR and QSAR charge questions gave
rise to the following major considerations:
   • Because technology allows for a very broad array of
     gene identification, there is no need to omit any classes
     of VFs from consideration in the initial development
     of VFAR methodology. Such elimination should
     occur only when the irrelevance of the VF can be
     demonstrated. In addition, the presence of a VF may
     be necessary but not sufficient for the development
     of pathogenicity. There are other factors, such as
     those that permit the expression of VFs, the survival
     and persistence of the microbes, or even a particular
     array of microbes in the environment, that permit the
     development of pathogenicity. There also is an urgent
     need to characterize background levels of VFs in the
     environment to better recognize a change in conditions
     that may  pose a human health risk.
   • The analysis of VFs can provide information regarding
     genetic engineering for changes occurring due to both
     bioweapons and naturally occurring genetic evolution.
     However, VFs may not be the focus of genetic
     engineering for the purpose of bioweapon development.
     There may be other characteristics that are altered to
     increase exposure  and risk.
   • There are many tools and technologies available for
     examining VFs, including genomics and gene arrays,
     PCR, and proteomics  for the analysis of protein
     products. These technologies are all under development
     in terms of their applicability to VFARs, but there are
     limitations due to sample collection and processing
     issues that must be addressed before these technologies
     can be applied to surveillance in water or air.
   • Genetic changes that occur naturally are an excellent
     example of the flexibility of the microbial genome.
     Most notably, microbes can transfer plasmids, resulting
     in the rapid exchange of genetic material. Increases  in
     potency are not always understood. There is a need to
     look for unusual combinations of genes as well as other
     factors. In general, a change in potency is accompanied
     by a string of changes, not just a single change.
   • For the purposes of public health protection, the goal is
     to be able to use VFARs to aid in the:
     ° Identification of the presence of microbes of concern
     0 Identification of accessory genes necessary for
       virulence
     0 Identification of environmental conditions necessary
       for virulence
     0 Extrapolation from virulence gene expression to
       virulence protein expression
     0  Prediction of the magnitude of the health hazard
        represented
     0  Determination of the infectivity or dose-response
        relationships to gauge the response needed to prevent
        or mitigate an outbreak
These characterizations and predictions would provide
information critical to understanding the magnitude of the
public health risk associated with a natural or intentional
exposure event.
   •  The current state of knowledge is focused on the
     identification of VFs and how these VFs function in the
     microbe to express virulence. The scientific community
     does not yet have the capability to  link VF information
     to health outcomes, though the potential exists. Because
     of the degeneracy of the genetic code, it is possible
     for there to be alterations  in the gene while its  activity
     is preserved. With constant changes in the microbial
     genome, it is necessary to maintain surveillance for
     these changes and evaluate how they affect virulence.
     It is possible to make primers for areas where changes
     cannot be made without changing function, thereby
     minimizing the chance of missing known VFs.
   •  For both chemical and biological threats to human
     health, the universe of microbes and chemicals needs
     to be characterized and narrowed for the purposes of
     regulatory prioritization and development of remedial
     action strategies. Also, for both approaches to be
     effective, either the MOA or mechanism of toxicity
     must be determined. This is an essential component
     of expert system based structure-activity relationships
     where the aspect of the structure of the chemical
     that results in a particular effect or outcome must be
     determined. This concept can greatly enhance QSAR
     model development and interpretation.
   •  In terms of the  role of -omics and QSARs in EPA's
     framework for risk assessment, any useful and valid
     information will help decrease uncertainty in the context
     of the overall WOE. Some technologies may be better
     for screening than for regulatory decision making
     because they may not be fully validated or accepted.
     QSARs and -omics technologies fit into this category.
     Currently, genomics technologies primarily serve as
     hazard identification tools by providing insight into the
     potential MOA by which a chemical is acting.  Such
     knowledge can inform the interpretation of QSARs. The
     integration of QSARs with -omics  technologies may
     allow these complementary technologies to reinforce
     each other. Computational toxicologists are working on
     this integration.

-------
There are several opportunities to combine QSARs and
MOA information to better inform risk assessment,
and the panel noted that routine acceptance of QSAR
predictions will likely require that they be derived
with an underlying mechanistic understanding. As
models become more sophisticated, they will further
incorporate structural features and property features
to allow for evaluating chemicals more fully through
the consideration of MOA data. Several examples of
developments in this area were described. Dr. Kannan
Krishnan discussed the integration of QSARs with
PBPK modeling, where MOA considerations (e.g.,
identification of appropriate dose metrics based on
chemical metabolism prediction)  are factored into the
PBPK model. Dr. Welsh discussed the growing use of
tools in bioinformatics (e.g., protein structure prediction
and libraries). Such tools have allowed for the use of
shape signatures based on the comparison of surface
features to integrate MOA (e.g., receptor binding) into
QSAR methodology. MOA data can be applied to larger
groups of chemicals to identify clusters of more closely
related chemicals — this is the conceptual basis for
decision tree and regression tree approaches. QSAR
models can be tailored via selection of descriptors for
each cluster to provide more uniform  training sets for
QSAR development or aid in interpreting global QSAR
predictions.
The focus of QSAR is on describing the potential
interaction between chemical and biological molecules.
There are two basic types of chemical-biological
interactions. Receptor-based interactions often are the
basis of endocrine disruption effects, and covalent
interactions occur with nonspecific macromolecular
binding. Mechanistic QSARs for predicting receptor-
based interactions are  commonly used in drug
development  and are increasingly being used for
toxicity prediction. Nevertheless, many chemicals
act via the disruption of membranes. The latter are
relatively nonspecific, but it is useful to focus on
covalent interactions, which can be quite complex,
even within a chemical class, as was highlighted in the
context of phenolic electrophiles. To be most useful,
QSARs need  to account for this complexity more
fully. While mechanistic QSARs  are preferred, an
intermediate step in this direction is to focus efforts on
endpoint-specific QSARs since the specificity of target
organs can  arise based on adsorption and distribution
(toxicokientics) or the nature of cell/tissue response
(toxicodynamics).
Although certain descriptors (i.e., molecular size
and hydrophobicity) are more commonly used, the
mechanistic context must be used as a starting point
for the selection of descriptors. Since  the mechanistic
context varies based on chemical class, it is not possible
to make blanket statements regarding the selection
of descriptors. Examples of descriptors based on
chemical mechanisms are those descriptors that describe
accumulation at or penetration through the membrane,
     reactivity toward cellular macromolecules, or receptor
     binding with critical targets, and others.
   •  Several approaches for hybrid SAR/QSAR analyses
     were discussed. Approaches ranged from using
     MOA descriptors as a screening step for the initial
     classification of chemicals to help in interpreting global
     QSARs to direct use of MOA descriptors in developing
     quantitative endpoint-specific logistic regression
     models. Semiquantitative QSAR methods included
     decision trees or modifications of this concept that use
     parallel sets of decision trees to improve predictability.
     Binned chemicals identified through these tools could
     serve as endpoint-specific QSAR training sets or be
     used to identify characteristics associated with potency
     categories for risk assessment using threshold of
     concern approaches.
   •  For chemical risk assessment, there is often a need to
     extrapolate from dose-response data based on exposure
     durations of less than  a lifetime to estimate the effects
     of lifelong exposure. Traditionally, for EPA risk
     assessments, a default factor of 10 is applied to adjust
     adverse effect levels from subchronic (i.e., exposure for
     roughly 10 percent of the lifetime) to chronic exposure
     conditions. This can be useful for the extrapolation
     of subchronic to chronic toxicity; however, it is
     inappropriate for the extrapolation from  acute to chronic
     exposure because the  critical endpoints are often
     different and the MOA is different between acute and
     chronic exposure. The panel noted that while several
     correlation approaches have been developed to address
     this situation, these are not QSARs per se. While
     QSARs may address this application directly, they can
     provide important insights.  For example, QSARs are
     used to predict toxicokinetic parameters  (e.g., partition
     coefficients or metabolism parameters) that impact
     decisions regarding the potential for increased body
     burden with longer-duration exposures. Furthermore,
     QSARs can provide information pertaining to both
     acute and chronic toxicity mechanisms, which impacts
     considerations of potential for accumulation  of tissue
     damage with increased exposure duration.

From the discussion of these charge questions came
the following major recommendations:

1.  Several recommendations on near-term applications of
   VFAR/QSAR models were discussed. To advance the
   applicability of VFARs in real-world situations, it is critical
   to facilitate the analysis of samples collected during natural
   outbreaks of microbial diseases. This will permit the
   identification of background levels of VFs and advance the
   understanding of the natural evolution of VFs in addition
   to providing the framework to test predictions of VFARs.
   Another potential opportunity for the advancement of
   VFAR research involves  the BioWatch Program, which
   consists of continuous sampling at locations across the
   country. This would be an opportunity for researchers to
   obtain material for the characterization of background
   levels of VFARs in urban environments, in  addition to

-------
   testing hypotheses. The state of the science regarding
   QSAR modeling is considerably more advanced than that
   of VFARs, therefore the key recommendation for near-term
   applications focused on the integration of MOA and PBPK
   with QSAR models to enhance biological applicability.

2.  For both VFARs and QSARs, host-specific factors alter
   the dose-response relationship (e.g., individual variability
   in metabolism, sensitive subpopulations, and host immune
   response); therefore, there will always be uncertainty in
   the ability to model host factors. These limitations should
   not be a deterrent from using these approaches in the
   evaluation of the universe of chemicals and microbes that
   require attention. For the initial prioritization of chemicals
   or microbes, when toxicological data are lacking, QSARs
   and VFARs can be particularly useful. Similarly, the
   databases and models under development could be critical
   to facilitating a rapid response in the event  of an intentional
   attack. QSARs and VFARs can provide critical information
   regarding alerts to human health concerns,  and chemical
   and biological plausibility in terms of potential human
   health effects, particularly as input to comprehensive
   WOE approaches.

3.  For the initial prioritization of chemicals or microbes,
   when toxicological data are lacking, QSARs and VFARs
   can be extremely useful. Similarly, the databases and
   models under development could be critical to facilitating
   a rapid response in the  event of an intentional attack.
   However, as noted by the expert speakers, these methods
   may not be sufficient for all chemicals or all microbes.
   One panelist charged that all the tools available should
   be used to begin to address these urgent public health
   concerns. QSARs and VFARs can be important tools in
   characterizing human health risks based on the weight
   of the evidence. Both QSARs and VFARs can be used to
   advance understanding of potential human health effects
   as well as in the regulatory context to help prioritize
   chemicals of concern. How EPA applies those concepts
   will likely vary by EPA division. A similar process is
   occurring in the EU.

4.  Other panelists urged that to move this discipline
   forward, single QSAR or VFAR predictions should not
   be considered an  answer. Rather, consensus or WOE
   approaches result in a more robust analysis. It is critical
   to be able to demonstrate how QSAR and VFAR tools can
   contribute to an understanding of health risks by providing
   information on hazard assessment and dose-response
   relationships.

5.  As a result of the discussions, participants noted
   the creation of more questions. It is becoming more
   common to  develop handbooks and guidelines to derive
   the necessary components. The field is very dynamic
   and needs virtual and enhanced screening in addition to
   genomics. QSARs and VFARS are, and will always be,
   two tools among  many.

-------

-------
                                                                                           8.0
                                                                          References
Ashby J, Tennant RW. Chemical structure, Salmonella mutagenicity and extent of carcinogenicity as indicators
of genotoxic carcinogenesis among 222 chemicals tested in rodents by the U.S. NCI/NTP. MutatRes. 1988 Jan;
204(1): 17-115. Review, http://www.ncbi.nlm.nih.gov/sites/entrez?Db=pubmed&Cmd=ShowDetailView&Term
ToSearch=3277047&ordinalpos=382&itool=EntrezSystem2.PEntrez.Pubmed.Pubmed ResultsPanel.Pubmed
RVDocSum

Beliveau M, Lipscomb J, Tardif R, KrishnanK. 2005. Quantitative structure-property relationships for
interspecies extrapolation of the inhalation pharmacokinetics of organic chemicals. Chem Res Toxicol.
18(3): 475-85.

Beliveau M and KrishnanK. 2003. In silica approaches for developing physiologically based pharmacokinetic
(PBPK) models. In: H Salem and S Katz, eds. Alternative ToxicologicalMethods. CRC press, NY, 479-532.

Blower PE, Yang C, Fligner MA, Verducci JS, Yu L, Richman S, Weinstein JN. 2002. Pharmacogenomic analysis:
correlating molecular substructure classes with microarray gene expression data. Pharmacogenomics J. 2002:
2(4): 259-71.

Bussey KJ, Chin K, Lababidi S, Reimers M, Reinhold WC, Kuo WL, Gwadry F, Ajay, Kouros-Mehr H, Fridlyand
J, Jain A, Collins  C, Nishizuka S, Tonon G, Roschke A, Gehlhaus K, Kirsch I, Scudiero DA, Gray JW, Weinstein
JN. 2006. Integrating data on DNA copy number with gene expression levels and drug sensitivities in the NCI-60
cell line panel. Mol Cancer Ther. 2006 Apr; 5(4): 853-67.

Cronin MTD, Aptula AO, Duffy JC, Netzeva TI, Rowe PH, Valkova IV, Schultz TW. 2002. Comparative
assessment of methods to develop QSARs for the prediction of the toxicity of phenols to Tetrahymena pyriformis.
Chemosphere 49: 1201-1221.

Ekins, Nikolsky and Nikolskaya in Trends in Pharmacological Sciences Vol. 26, No 4, April 2005.

Gray LE Jr. and Ostby J. 1993. The effect of prenatal administration of azo dyes on testicular development in the
mouse: A structure activity profile of dyes derived from benzidine, dimethylbenzidine, or dimethoxybenzidine.
Fundamental and Applied Toxicology. 20, 177-183.

Harada A, Hanzawa M, Saito J, Hashimoto K.  1992. Quantitative analysis of structure-toxicity relationships of
substituted anilines by use of Balb/3T3 Cells. Environmental Toxicology and Chemistry, Vol. 11: 973-980.

Knaak JB, Dary CC, Power F,  Thompson CB, Blancato JN. 2004.  Physicochemical and biological data for the
development of predictive organophosphorus pesticide QSARs and PBPK/PD models for human risk assessment.
CritRev Toxicol.  34(2): 143-207.

Lewis DFV, loanides C, and Parke DV 1993. Validation of a novel molecular orbit approach (COMPACT)
for the prospective safety evaluation of chemicals, by comparison  with rodent carcinogenicity and Salmonella
mutagenicity data evaluated. MutatRes. 291: 61-77.

Lowell HH, Maynard EL, Kier LB. 1989. Structure-activity relationship studies on the toxicity of benzene
derivatives: III. Predictions and extension to new substituents. Environmental Toxicology and Chemistry,
Vol. 8: 431-436.

JC Madden, MTD Cronin. 2006. Structure-based methods for the prediction of drug metabolism. Expert Opinion
on Drug Metabolism and Toxicology 2'. in press.

National Research Council (NRC). 1999. Identifying Future Drinking Water Contaminants Based on the
1998 Workshop on Emerging Drinking Water Contaminants. Water Science and Technology Board, Board on
Environmental Studies and Toxicology. National Academy Press, Washington, DC.

-------
National Research Council (NRC). 1999. Setting Priorities for Drinking Water Contaminants. Committee on
Drinking Water Contaminants, Water Science and Technology Board, Board on Environmental Studies and
Toxicology. National Academy Press, Washington, DC.

National Research Council (NRC). 2001. Classifying Drinking Water Contaminants for Regulatory Consideration.
Committee on Drinking Water Contaminants, Water Science and Technology Board, Board on Environmental
Studies and Toxicology. National Academy Press, Washington, DC.

Nishizuka S, Charboneau L, Young L, Major S, Reinhold WC, Waltham M, Kouros-Mehr H, Bussey KJ, Lee
JK, Espina V, Munson PJ, Petricoin E 3rd, Liotta LA, Weinstein JN. 2003. Proteomic profiling of the NCI-60
cancer cell lines using new high-density reverse-phase lysate microarrays. Proc Natl Acad Sci USA. 2003 Nov 25;
100(24): 14229-34.

Rosenkranz HS and Klopman G. 1989. Structural basis of the mutagenicity of phenyazoaniline dyes.
MutatRes.22\: 217-234.

Schultz TW, Sinks GD, Cronin MTD. 1997. Identification of mechanisms of toxic action of phenols to
Tetrahymena pyriformis from molecular descriptors. In: Chen F and Schuurmann G (Eds) Quantitative
Structure-Activity Relationships in Environmental Sciences - VII. SETAC Press, Pensacola, USA, 329-342.

Shea, DA and Lister S. 2003. The BioWatch Program: Detection of Bioterrorism. Congressional Research Service
Report No. RL 32152, November 19, 2003. http://www.fas.org/sgp/crs/terror/RL32152.html

U.S. Environmental Protection Agency (EPA). 1994. Assessment Tools for the Evaluation of Risk (ASTER).
On-line Database. Environmental Research Laboratory-Duluth.

U.S. Environmental Protection Agency (EPA). 1992. Provisional Guidance for the Qualitative Risk Assessment
of Polycyclic Aromatic Hydrocarbons. Prepared by the Environmental Criteria and Assessment Office, Office
of Health and Environmental Assessment, Cincinnati, OH, for the Office of Research and Development,
Cincinnati, OH.

U.S. Environmental Protection Agency (EPA). 1989 Update to the Interim Procedures for Estimating Risk
Associated with Exposures to Mixtures of Chlorinated Dibenzo-/>-Dioxins and -Dibenzofurans (CDDs and CDFs).
Risk Assessment Forum, Washington, DC.

Waller CL, Evans MV, McKinney JD. 1996. Modeling the cytochrome P450-mediated metabolism  of chlorinated
volatile organic compounds. DrugMetab Dispos. 24(2): 203-10.

Weisburger EK. 1979. TV-Substituted aryl compounds in carcinogenesis and mutagenesis. Presented at the
International Conference on Carcinogenic and Mutagenic TV-substituted Aryl Compounds, Rockville, Maryland.

Weisburger JH and Fiala ES. 1979. Mechanisms of species, strain, and dose effects in arylamine carcinogenesis.
Presented at the International Conference on Carcinogenic and Mutagenic N-substituted Aryl Compounds,
Rockville, Maryland.

-------
                                                                                 Appendix A
                                                                       List  of  Speakers
Andy Avel
Assistant Center Director
U.S. EPA, Office of Research and Development
National Homeland Security Research Center
26 West Martin Luther King Drive (MS 163)
Cincinnati, OH 45268-1320
Phone:  513-569-7951
Email: avel.andy@,epa.gov

Mark Cronin, Ph.D.
School of Pharmacy and Chemistry
Liverpool John Moores University
Byrom Street
Liverpool, England L3 3 AF
Phone (from UK): 0151 231 2402
Phone (from outside UK): + 44 151 231 2402
FAX (from UK):  0151 231 2170
FAX (from outside UK): + 44  151 231 2170
Email: m.t.cronin(@,ljmu.ac.uk

Syed A. Hashsham, Ph.D.
Edwin Willits Associate Professor
Department of Civil and Environmental
Engineering Center for Microbial Ecology
Michigan State University
A126 Research Complex-Engineering
East Lansing, MI 48824
Phone: 517-355-8241
FAX: 517-355-0250
Email: hashshani(@,egr.msu.edu

Jonathan Herrmann, P.E., DEE
Center Director
U.S. EPA, Office of Research and Development
National Homeland Security Research Center
26 West Martin Luther King Drive (MS 163)
Cincinnati, OH 45268-1320
Phone: 513-569-7839
Email: herrmann.jonathan(@,epa.gov

Kan nan Krishnan, Ph.D.
Professeur titulaire et Directeur TOXHUM
Universite de Montreal
2375 Cote Ste. Catherine, Room 4105
Montreal, PQ, Canada, H3T 1A8
Phone: 514-343-6581
FAX: 514-343-2200
Email: kannan.krishnan(@,umontreal.ca
Andrew Maier, Ph.D., CIH, DABT
Associate Director
Toxicology Excellence for Risk Assessment
2300 Montana Avenue, Suite 409
Cincinnati, OH 45211
Phone: 513-542-7475x23
FAX: 513-542-7487
Email: maier(@.tera.org

Chandrika Moudgal, M.S.
U.S. EPA, Office of Research and Development
National Homeland Security Research Center
Threat and Consequence Assessment Division
1001 SW5thAvenue, Suite 1510
Portland, OR 97204
Phone: 503-326-3541
FAX: 503-326-4005
Email: moudgal.chandrika@,epa. gov

Joan B. Rose, Ph.D.
Homer Nowlin Chair in Water Research
Department of Fisheries and Wildlife
Michigan State University
13 Natural Resources
East Lansing, MI 48824
Phone: 517-432-4412
Fax: 517-432-1699
Email: rosejo(@,msu.edu

R Paul Schaudies, Ph.D.
Assistant Vice President
Science Applications International Corporation
Biological and Chemical Defense
9700 Great Seneca Highway, Suite 220
Rockville, MD 20850
Phone:240-453-6312
FAX: 240-453-6208
Email: schaudiesr@,saic.com

Subhas Sikdar, Ph.D.
Acting Associate Director for Health
U.S. EPA, Office of Research and Development
National Risk Management Research Laboratory
26 West Martin Luther King Drive (MS 235)
Cincinnati, OH 45268
Phone: 513-569-7528
Email: sikdar.subhas(@,epa.gov

-------
Cindy Sonich-Mullin, M. En.
Division Director
U.S. EPA, Office of Research and Development
National Homeland Security Research Center
Threat and Consequence Assessment Division
26 West Martin Luther King Drive (MS 163)
Cincinnati, OH 45268-1320
Phone: 513-569-7923
Email: sonich-mullin.cynthia@epa.gov

Gerard Stelma, Ph.D.
U.S. EPA, Office of Research and Development
National Exposure Research Laboratory
26 West Martin Luther King Drive (MS 593)
Cincinnati, OH 45268-1320
Phone: 513-569-7384
Email: stelma.gerard@epa.gov

William J. Welsh, Ph.D.
Norman H. Edelman Professor in Bioinformatics
Department of Pharmacology Robert Wood Johnson Medical
School University of Medicine & Dentistry of New Jersey
(UMDNJ)
Director
UMDNJ Informatics Institute
Director
UMDNJ Environmental Bioinformatics
& Computational Toxicology Center
675 Hoes Lane
Piscataway, NJ 08854
Phone: 732-235-3234
FAX:  732-235-3475
Email: welshwi@umdni.edu
Andrew Worth, Ph.D.
Scientific Officer
European Commission - Joint Research Centre
Via Enrico Fermi 1
21020Ispra (VA), Italy
Phone: +39 0332 789566
FAX:+39 0332 786717
Email:  andrew.worth@jrc.it

Douglas Young, Ph.D.
Branch Chief
U.S. EPA, Office of Research and Development
National Risk Management Research Laboratory
Clean Processes Branch
26 West Martin Luther King Drive
Cincinnati, OH 45268-1320
Phone: 513-569-7624

-------
                                                                                    Appendix  B
                          Biosketches  of  Speakers  and   Panelist
Mr. Andy Avel started his career in 1972 as an engineering
geologist for the U.S. Tennessee Valley Authority and
over the following ten years was assigned in Chattanooga,
Kingsport, and Knoxville, TN. He joined the Clinch River
Breeder Reactor Plant Project in Oak Ridge, TN, in 1982, as
a geotechnical engineer. Upon cancellation of the Breeder
Reactor, Andy moved to the Department of Energy's Office
of Civilian Radioactive Waste Management in Columbus,
OH, where he served as a licensing engineer. He returned to
Oak Ridge as a project manager in the Formerly Utilized Sites
Remedial Action Program and then moved to the Fernald
Feed Materials Production Plant, near Cincinnati, where he
managed the CERCLA cleanup program.
Mr. Avel joined ORD in 1991 as the Director of the Office of
the Senior Official in Cincinnati. Following the reorganization
of 1996, he was assigned as special assistant to the Director
of NRMRL and then as the Acting Lab Director (ALD) for
Pesticides and Toxic Substances. In November of 2002, Andy
joined the National Homeland Security Research Center as the
Deputy Director for Management. In January 2005, he was
named Acting Director of NHSRC.

Dr. Mark Cronin is Professor of Predictive Toxicology in the
School of Pharmacy and Chemistry at Liverpool John Moores
University, England. He was previously a lecturer (from 1994)
and reader (from 2001) in that department. In addition to a
full teaching load on the Master of Pharmacy degree  course,
he maintains an active research focus on the development of
computational methods to predict toxicity. Particular emphasis
at the moment is on the prediction of reactive toxicity (e.g.,
skin sensitization) and the use of quantitative structure-
activity relationships (QSARs) for regulatory purposes. He
has over 150 publications in these areas and has co-organized
a number of conferences in predictive toxicology. Mark
obtained his degree in Biology and Ph.D. in ecotoxicological
QSAR from Liverpool Polytechnic.

Dr. Syed A. Hashsham is Edwin Willits Associate Professor
of Civil  and Environmental Engineering at Michigan State
University (MSU). He is also a Co-Principal Investigator (PI)
in the Center for Microbial Ecology and CAMRA, the U.S.
EPA/DHS Center for Advancing Microbial Risks Assessment.
Syed's expertise is in the area of environmental genomics
and modeling of molecular data with a focus on microbial
issues related to drinking water and wastewater. His research
work is sponsored by the NIH, EPA, DHS, DoD, NSF, and
state agencies. He has published on DNAbiochip-based
parallel microbial detection (Biosensors & Bioelectronics,
2004), VFAR (Water Science and Technology, 2004),
microbial community dynamics (Applied and Environmental
Microbiology, 2000), probe design (Nucleic Acids Research,
2006) and dehalo-respiration (Science, 2002). Syed earned his
Ph.D. in Environmental Engineering and Science from
the University of Illinois at Urbana-Champaign and conducted
post-doctoral research at the Center for Microbial Ecology at
MSU and Stanford University.

Mr. Jonathan Herrmann has been with EPA since 1975.
He first worked in the Agency's Region VIII office in Denver,
Colorado. He came to the EPA's Office of Research and
Development (ORD) in 1978 and, except for a brief time in
the private sector in the early 1980s, has been with ORD in
Cincinnati, OH. Mr. Herrmann holds a Bachelor's Degree
in Civil Engineering from Youngstown State University and
a Master's Degree in Business Administration from Xavier
University. He is a Registered Professional in Engineering in
the State of Ohio. He is a member of the American Society
of Civil Engineers, the American Academy of Environmental
Engineers, and the American Water Works Association.
Mr. Herrmann's career has spanned many areas. He
has worked in mined land reclamation, Superfund site
remediation, land disposal of hazardous and household
wastes, and environmental technology testing and evaluation.
In the mid-1990s he was a strategic planner for the National
Risk Management Research Laboratory and lead the
development of ORD's Pollution Prevention Research
Strategy and Mercury Research Strategy.
Mr. Herrmann joined NHSRC in September 2002 as the
Water Security Team Leader and with a group of scientists
and engineers developed the Water Security Research and
Technical Support Action Plan in cooperation with the
Agency's Office of Water (OW). He is currently serving as
the Center Director for NHSRC. As such, he is responsible
for the day-to-day personnel, funding, and product delivery
aspects of the Center.

Dr. Kannan Krishnan received his Ph.D. in Public Health
from Universite de Montreal, Canada, and postdoctoral
training from the Chemical Industry Institute of Toxicology
(CUT), Research Triangle Park, North Carolina. He is
currently Professor of Occupational and Environmental Health
and Director of the Human Toxicology Research Group
(TOXHUM) at Universite de Montreal. He has been the
leader of the risk assessment methodologies theme team of the
Canadian Network of Toxicology Centers (1994-2001), and
Vice President of the Biological Modeling Specialty Section
of the Society of Toxicology (2001-2002). He has also
been a member of the U.S. National Academy of Sciences
(NAS) Sub-committee on Acute Exposure Guideline Levels
(2001-2004), member of the U.S.  EPA's Human Studies
Review Board (2006-), president of the Risk Assessment
Specialty Section of the Society of Toxicology (2005-2006),
and a temporary advisor for the World Health Organization
for developing a scientific document on the principles for

-------
evaluating health risks in children associated with chemical
exposures.  His expertise is in the areas of mixture toxicology,
health risk assessment methods, and the development of
quantitative structure-pharmacokinetic relationships. He
has been a peer reviewer of several IRIS updates, risk
assessments, mixture risk assessment supplemental guidance,
and efforts on interactions for U.S. EPA. He has also been
actively involved as a reviewer of ATSDR documents on
toxicological profiles and interaction profiles. He  has been on
the editorial boards of Toxicological Sciences, the International
Journal of Toxicology, the Journal of Applied Toxicology
and the Journal of Child Health.  An author of a textbook
on environmental pollution, Dr. Krishnan has authored or
coauthored over 100 full-length publications and 250 abstracts
in the general areas of toxicology, PBPK modeling, QSARs,
and risk assessment.  His research team received the Best
paper award (2003) from the Board of Publications of the
Society of Toxicology (U.S.A.) for a publication on a novel
risk assessment methodology (Haddad S, Beliveau M, Tardif
R, and Krishnan K. [2001]) and a PBPK model-based approach
for the risk assessment of chemical mixtures {Toxicological
Sciences 63: 125-135) and more recently received  recognition
for a publication on QSAR modeling (Beliveau M, Lipscomb
J, Tardif R, and Krishnan K [2005]), Quantitative structure-
property relationships for interspecies extrapolation of the
inhalation pharmacokinetics of organic chemicals  (Chemical
Research in Toxicology 18: 475-485) was part of a "top ten"
list of publications advancing the Science of Risk Assessment.
Dr. Krishnan was honored with the  Veylian Henderson Award
in 2000 by the Society of Toxicology of Canada for significant
contributions to the field of toxicology.

Dr. Andrew Maier currently serves as the Associate Director
for the nonprofit organization Toxicology Excellence for Risk
Assessment (TERA). In his capacity as a toxicologist and risk
assessor, he has coauthored technical reports, human health
risk assessment documents, and toxicity summaries covering
more than 100 individual substances for government and
private sponsors. He has led a variety  of efforts for developing
and applying methods in preventive toxicology and hazard
screening that make use of QSAR approaches. Dr. Maier
completed his M.S. in industrial health at the University
of Michigan and his Ph.D. in toxicology at the University
of Cincinnati. He has research interests in the molecular
mechanisms of toxicity and has conducted basic research
in the areas of metal and polycyclic aromatic hydrocarbon
mixtures, environmentally relevant genetic polymorphisms,
and risk assessment methods. His recent research efforts
have focused on using early biological effect markers and
MOA information to reduce uncertainties in chemical risk
assessment. Dr. Maier remains active in communicating his
findings through participation in professional societies such as
the Society of Toxicology. He is a Diplomate of the American
Board of Toxicology.

Ms. Chandrika Moudgal is currently PI and technical lead
on four projects related to the development of end point-
specific QSAR models, PI and technical lead to explore
the state of VFAR science and develop a case study using
cyanotoxins, and PI and technical lead on a project to
develop a Web-based "Data Dictionary" for agents of
concern to NHSRC. In addition, she supports TCAD's
Provisional Advisory Level (PAL) guidance documents.
She also lends support to other NHSRC  divisions by
reviewing technical documents.
Chandrika earned her M.S. in Toxicology from the University
of Cincinnati and her B.S. in Chemistry  from the University
of Gujarat in Ahmadabad, India. She has also completed
course work for an M.S. in Environmental Science at the
Ohio State University. Prior to joining NHSRC, Ms. Moudgal
served as an environmental health scientist at the National
Center for Environmental Assessment (NCEA), ORD, U.S.
EPA for approximately seven years. In this position she gained
experience and expertise in the development and application
of QSARs to fill experimental data gaps and expertise in the
application of the Agency's risk assessment methodology. She
also served as chemical manager and reviewer of documents
for the IRIS database. Additional previous experience includes
serving  as an  Organic Chemistry Section Supervisor with
R.D. Zande & Associates in Columbus for seven years:
working as a water research analyst for the City of Columbus,
drinking water treatment plant for one year: and working
as a laboratory scientist for the State of New Hampshire for
three years. Chandrika has published several papers related
to QSAR research and has presented various papers both
nationally and internationally on the topic.

Dr. Joan Rose serves as the Homer Nowlin Chair  in Water
Research at Michigan State University and is currently
Director of the Center for Water Sciences.  Dr. Rose received
her Ph.D. in Microbiology from the University of Arizona
in 1985. She served as a Professor in the College of Marine
Science, University of South Florida (USF) from 1998 to
2002.
Dr. Rose's professional experience includes environmental
virology, environmental parasitology,  drinking water
treatment and disinfection, microbial risk assessment,
wastewater treatment and reuse, water pollution microbiology,
mycology, and food microbiology. Dr. Rose is an international
expert in water microbiology, water quality, and public
health safety,  publishing more than 200 manuscripts. She has
been involved in the investigation of numerous waterborne
outbreaks worldwide. Her work has examined new molecular
methods for waterborne pathogens and zoonotic agents
such as Cryptosporidium and enteric viruses and source
tracking techniques. She has been involved  in the study of
water supplies, water used for food production, and  coastal
environments as well as water treatment, wastewater treatment,
reclaimed water and water reuse, and quantitative microbial risk
assessment. She is specifically interested in microbial pathogen
transport in coastal systems and has studied the impact of
wastewater discharges and climate on water quality. She was
named as one of the 21 most influential people in water in the
21st Century by Water Technology Magazine (2000) and won
the Clarke Water Prize (one of five international awards for
contributions to water science and technology).

-------
Current service on advisory committees includes Chair of the
Drinking Water Committee for the Science Advisory Board
for the U.S. Environmental Protection Agency; the Science
Advisory Board of the International Commission of the Great
Lakes, 2003-08; Vice-Chair of USA National Committee for
the International Water Association (IWA), 2002-06, Member
of the Strategic Council for IWA 2005-08, Chair of the
Specialist Group Health-Related Water Microbiology (IWA)
2004-07; Research Advisory Board, National Water Research
Institute, 2002-06, and Council Policy Committee for the
American Society of Microbiology, 2001-06.
Sources of recent grant and/or contract support include
NOAA, U.S.  EPA, Water Environmental Research
Foundation, NSF, and AWWARF. She was recently
awarded as PI as $10 million grant for directing the Center
for Advancing Microbial Risk Assessment funded by EPA
and the U.S. Department of Homeland Security.

Dr. R Paul Schaudies, Assistant VP at Science Applications
International  corporation (SAIC), heads a diverse team of
technologists who conduct contract biomedical research,
scientific analyses, and technical support. Dr. Schaudies is an
internationally recognized expert in the fields of biological
and chemical warfare defense. He served as a primary Science
and Technology Consultant to the Incident Commander,
Sergeants-at-Arms for the House and Senate, and U.S. EPA
On-Scene Coordinator in response to the October 2001
anthrax incident in Washington D.C. He has served on five
National Academy committees in the areas of biological
defense and nanotechnology. He has served on numerous
national level advisory panels for the Defense Intelligence
Agency, the Defense Advanced Research Projects Agency,
and the Department  of Energy. Dr. Schaudies served 12 years
as a U.S. Army officer. While on active duty, Dr. Schaudies
served as Chief of the General Support Laboratory in the
Department of Clinical Investigation at Walter Reed Army
Medical Center, a Senior Researcher at the Walter Reed Army
Institute for Research, and  a Program Manager for Biological
and Chemical Defense Research at the Central Measurement
and Signature Intelligence Office at the Defense Intelligence
Agency. Dr. Schaudies received his Bachelor's degree in
Chemistry from Wake Forest University and his doctoral
degree in Biochemistry from Temple University School of
Medicine.

Ms. Cynthia Sonich-Mullin is the Director of the Threat and
Consequence Assessment Division (TCAD) at the National
Homeland Security Research Center. She has provided
program leadership, focusing on rapid risk assessment and
support to the entire NHSRC team and ORD, since March
2003.
Prior to this assignment,  Ms. Sonich-Mullin worked in the
National Center for Environmental Assessment in a number of
capacities.  Most recently, she served concurrent details as the
Acting Deputy Director,  Cincinnati Division, and the Acting
Center Director for Human Health Research.
Since 1993, Ms. Sonich-Mullin has worked on behalf of the
International Programme on Chemical Safety (IPCS), a joint
program of the World Health Organization (WHO), United
Nations Environment Programme, and the International
Labour Organization. In October 1993, Ms. Sonich-
Mullin worked with IPCS to initiate the IPCS Project:
Harmonization of Approaches to the Assessment of Risk
from Exposure to Chemicals, on behalf of the WHO. In
this capacity, she worked as an IPCS/WHO staff member in
Geneva, Switzerland for three years. Upon returning to the
U.S., Ms. Sonich-Mullin (as part of U.S. EPAs contribution
to the WHO) has continued to work  on various aspects of the
Harmonization Project.
Prior to the detail, Ms. Sonich-Mullin was a scientist with
U.S. EPA's Environmental Criteria and Assessment Office
(now the National Center for Environmental Assessment),
serving a number of roles  including:
   •  Acting Deputy Director, Cincinnati Division
   •  Chief, Chemical Mixtures Assessment Branch
   •  Chief, Systemic Toxicants Assessment Branch
In these capacities, she led and participated in projects
related to the assessment of chemicals in air, drinking water
and ambient water, municipal solid waste disposal options,
and on issues related to Superfund sites. She worked on
the development of Agency risk assessment guidelines and
served on numerous task groups and research committees
including the Agency's Water Research Committee, Air Risk
Information Support Center, and as Director of the Superfund
Technical Support Center, a center designed to provide
risk assessment support, guidance, and advice on issues
specifically pertaining to Superfund  sites. In a concurrent
assignment, she was selected to serve on Vice President Al
Gore's Commission to Reinvent Government.
Ms. Sonich-Mullin began  her career at EPA working as
an environmental health scientist with the Health Effects
Research Laboratory.  In this capacity, she designed
and conducted epidemiological  studies related to water
contamination and has published in this area. Some specific
issues studied included the health effects associated with
drinking water chlorination, health effects of sodium
in drinking water, and the  health effects of the carbon
tetrachloride spill into the Ohio  River in the late  1970s.
Ms. Sonich-Mullin holds a Master of Environmental Sciences
degree, specializing in Applied Biology/Zoology from the
Institute of Environmental Sciences, Miami University,
Oxford, Ohio. She has also completed doctoral course
work in Epidemiology and Biostatistics at the University of
Cincinnati, College of Medicine, Cincinnati, Ohio.

Dr.  Gerard N. Stelma Jr. is a Senior Science Advisor for
the Microbiological and Chemical Exposure Assessment
Research Division (MCEARD)  of the National Exposure
Research Laboratory (NERL), which is part of U.S. EPA's
Office of Research and Development. In this role, he provides
expert advice regarding microbiological issues, principally

-------
those pertaining to bacterial pathogens, to the division's
microbiologists and to various EPA program offices. Dr.
Stelma served as the Chief of MCEARD's Microbial
Exposure Research Branch for 13 years and as Acting
Director of MCEARD for nearly 3 years. Prior to his arrival at
EPA, Dr. Stelma was a research microbiologist for the FDA.
He holds a B.S. in Biology from the University of Michigan
and a Ph.D. in Microbiology from Michigan State University.

Dr. Subhas K. Sikdar is the Acting Associate Director
for Health for NRMRL. As the Director of the Sustainable
Technology Division until Jan 9, 2004, he was the primary
spokesman for U.S. EPA's R&D on clean technologies and
pollution prevention. He directed research, both intramural
and extramural, on tools and methods for pollution
prevention, cleaner process technologies, and demonstration
and verification of cleaner technologies. Before joining EPA
in 1990, Dr. Sikdar held managerial positions at the National
Institute of Standards and Technology in Boulder, Colorado,
and General Electric Corporate Research &  Development
Center in Schenectady, New York.  He began his professional
career as a Senior Research Engineer with Occidental
Research Corporation in Irvine, California, in 1975. Dr.
Sikdar earned his B.S. in Chemistry, a B.Tech in Chemical
Engineering, and an M.Tech in Polymer Science from
Calcutta University in India. He received his M.S. and Ph.D.
in Chemical Engineering from the University of Arizona.
Dr. Sikdar is a Fellow of the American Association for the
Advancement of Science (AAAS), Fellow of the American
Institute of Chemical Engineers, Honorary Fellow of the
Indian Institute of Chemical Engineers, winner of three EPA
bronze medals, an R&D 100 award (1990), AIChE's Larry
Cecil Award for Environmental Chemical Engineering (2002),
and University of Arizona's Distinguished Engineering
Alumnus Award (2003). In the past he was a member of the
Vision 2020 Steering Committee for the chemical industry,
an action network leader of the Council for Chemical
Research. He is a member of the Board of Governors of the
Council for Chemical Research (CCR) and of the Green
Chemistry Institute, a member of AIChE's Research and New
Technology Committee, and the Chair of the Sustainable
Engineering Forum. For some years he has been championing
the concepts and methods for clean products and processes
through a NATO pilot project, two NATO workshops, and
an Engineering Foundation conference. He is a current
member of the Industrial Advisory Board of the University
of Arizona's College of Engineering and of the Department
of Chemical and  Environmental Engineering, and of the
Department of Chemical and Environmental Engineering of
the Illinois Institute of Technology. Dr. Sikdar is the leader
of the technical expert group for the Center of Excellence on
Environmental Engineering and Hazardous Wastes, which
is composed of several universities in Thailand. He is the
founder and co-Editor-in-Chief of the international journal,
Clean Technologies and Environmental Policy,  published
quarterly by Springer Verlag of Germany. Dr. Sikdar has
published more than 60 technical papers in reputed journals,
holds 22 U.S. patents, and has edited 13 books.
Dr. Sikdar was instrumental in developing the highly
successful Occidental Hemihydrate process for phosphoric
acid manufacture. His other technical achievements include
developing several membrane processes for pervaporative
separation of VOCs from aqueous effluents and for highly
selective sorption of heavy metals, masterminding the
development of a waste reduction algorithm for process
design (the WAR algorithm), a solvent design algorithm
(PARIS II), and a data portal for life cycle assessment
(LCAccess).

Dr. William J. (Bill) Welsh holds the Norman H. Edelman
Professorship in Bioinformatics and Computer-Aided
Molecular Design in the Department of Pharmacology at
the Robert Wood Johnson Medical School (RWJMS) in
Piscataway NJ,  University of Medicine and Dentistry of New
Jersey (UMDNJ). Concurrently, he serves as Director of the
UMDNJInformatics Institute (http://informatics.umdni.edu)
that coordinates university-wide initiatives in bioinformatics,
clinical informatics, and computer-aided molecular design.
He is  also PI and Director of the EPA-supported New Jersey
Research Center for Environmental Bioinformatics and
Computational Predictive Toxicology, the first of its kind in
the nation. He is a member of various centers and institutes
of excellence at UMDNJ and Rutgers University, including
the Cancer Institute of New Jersey, the New Jersey Center for
Biomaterials, Rutgers University School of Pharmacy, and
the Environmental & Occupational Health Sciences Institute
(EOHSI).
Dr. Welsh earned a B.S. degree (magna cum laude) in
Chemistry from St. Joseph's University (Philadelphia,
PA) and a Ph.D. degree in Theoretical Physical Chemistry
from the University of Pennsylvania (Philadelphia, PA).
He conducted postdoctoral research in the laboratory of Dr.
James E. Mark, Distinguished Professor of Polymer Science
at the University of Cincinnati (Cincinnati, OH). In 1985,
Dr. Welsh joined the University of Missouri (St. Louis) as
an Associate Professor of Chemistry and rose through the
ranks to Distinguished Professor in 1998. During this period
he was appointed Director, Laboratory for Computer-Aided
Molecular Design, at the University of Missouri. In 2001, Dr.
Welsh joined UMDNJ-Robert Wood Johnson Medical School
to assume his present role.
Dr. Welsh's laboratory specializes in the development and
application of computational tools for drug discovery.
Promising candidates emanating from these rational
design approaches are synthesized and tested as potential
therapeutic or diagnostic agents. His laboratory is widely
reputed for its innovation, such as the development of the
Shape Signatures tool and the discovery of potential drug
candidates for the treatment of cancer, severe and chronic
pain, neurodegenerative diseases, and heart conditions.
Dr. Welsh's publication record includes over 350 articles
in peer-reviewed books and journals, 600 abstracts from
presentations at professional scientific meetings, and several
patents  and patent applications. He is the recipient of
numerous awards and honors, including the Teacher of the
Year Award (1983 and  1985), the St. Louis Research Award

-------
(1998), the University ofMissouri-St. Louis Chancellor's
Research and Creativity Award (2001), the University of
Missouri Entrepreneur of the Year Award (2001), the Norman
H. Edelman Endowed Professorship in Bioinformatics at
UMDNJ-RWJMS (2003), and most recently the John C.
Krantz, Jr. Award (2004). He serves on the advisory boards
of several scientific journals. Spanning the last twenty years,
over 125 students (postgraduate and graduate students,
undergraduates, and research associates) have trained in his
laboratory.
Dr. Andrew Worth works at the European Chemicals Bureau
(ECB) within the European Commission's Joint Research
Centre (JRC) in Italy. He joined the JRC with degrees in
Physiological Sciences and Linguistics from the University
of Oxford (UK) and with post-graduate experience in the
fields of biochemistry and toxicology. He subsequently
gained a Ph.D in Computational Toxicology from Liverpool
John Moores University (UK). His research interests have
focused on the development of QSAR models and methods
and on the development of Integrated Testing Strategies for
chemical toxicity based on the use of physicochemical and
in vitro data. Since 2003 he has been leading the JRC Project
on Computational Toxicology. In addition to coordinating
QSAR-related work within the JRC, Dr. Worth also chairs the
EU Working Group on QSARs and represents the European
Commission in several OECD working groups.
Dr. Douglas Young leads the Clean Processes Branch (CPB)
that resides in the Sustainable Technology Division (STD) in
the Office of Research and Development within EPA. STD
is home to EPA's in-house research in the areas of Green
Chemistry and Sustainability. Dr. Young's research is in the
areas of environmental impact assessment as it pertains to
the chemical processing industry and the estimation of acute
toxicity measurements. He has been intimately involved in the
creation of the Computational Toxicology Research Program
and the National Center for Computational Toxicology
within  the EPA. He was instrumental in the development
and commercialization of the generalized Waste Reduction
(WAR) algorithm. Dr. Young received his Ph.D. from the
University of Arizona where his dissertation focused on the
bioremediation of high-energy explosive waste generated at
the Los Alamos National Laboratory. He received his M.S.
from the University of Notre Dame and his B.S. from the
University of Michigan. All three of Dr. Young's degrees are
in  Chemical Engineering.

-------

-------
                                                                   Appendix  C
                                                      Workshop Agenda
Tuesday, June 20,  2006
8:00 am  Welcome (Chandrika Moudgal, NHSRC)
8:10 am  Opening remarks (Andy Avel, NHSRC: Jonathan Herrmann, NHSRC: Subhas Sikdar, NRMRL)
8:30 am  Background on NHSRC and NRMRL (Cindy Sonich-Mullin, NHSRC: Douglas Young, NRMRL) and
        introduction of expert panel members
8:50 am  QSAR/VFAR program and charge to expert panel members (Chandrika Moudgal, NHSRC)
9:00 am  Introduction to the WAR concept (Dr. Gerald Stelma, NERL)
9:20 am  Using VFAR in the Risk Assessment Framework (Dr. Joan Rose, MSU)
9:50 am  VFAR: factors related to genomic variabilities (Dr. Syed Hashsham, MSU)
10:20 am Break
10:50 am A bioinformatic approach to VFAR analysis and characterization (Dr. Paul Schaudies, SAIC)
11:20 am VFAR charge questions 1 and 2 (Discussion)
12:00 pm Lunch
1:00 pm  VFAR charge questions 3 and 4 (Discussion)
3:00 pm  Break
3:30 pm  VFAR charge questions 5, 6, and 7 (Discussion)
5:00 pm  VFAR closing remarks from panel and EPA

Wednesday, June 21, 2006
8:00 am  From reactivity to regulation: integrating alternative techniques to predict toxicity
        (Dr. Mark Cronin, TOXHUM)
8:20 am  Integrated QSAR-PBPK modeling for risk assessment applications (Dr. Kannan Krishnan, TOXHUM)
8:40 am  Integration  of MOA and WOE concepts in predictive toxicology (Dr. Andrew Maier, TERA)
9:00 am  Activities at the new UMDNJ Computational Toxicology Center: advanced QSAR-based methods of
        rapid hazard identification, prediction, and characterization (Dr. William Welsh, RWJMS)
9:20 am  The role of the European Chemicals Bureau in promoting the regulatory implementation of
        estimation methods  (Dr. Andrew Worth, JRC)
9:40 am  Break
10:10 am QSAR charge questions 1,2, and 3 (Discussion)
12:00 pm Lunch
1:00 pm  QSAR charge questions 4 and 5 (Discussion)
3:00 pm  Break
3:30 pm  QSAR charge questions 6 and 7 (Discussion)
4:30 pm  Workshop closing remarks (Panel, NHSRC and NRMRL management,
        Douglas Young, Chandrika Moudgal)
5:00 pm  Adjourn

-------

-------
                                                                              Appendix  D
                                                                   List  of Attendees
Femi Adeshina, Ph.D.
1200 Pennsylvania Avenue, NW (8801R)
Washington, DC 20460
Phone: 202-564-1539
Email: adeshina.femi@epa.gov

Caroline Baier-Anderson, Ph.D.
EnDyna, Inc.
7925 Jones Branch Drive, Suite 5300
McLean, VA 22102
Phone: 410-610-1737
FAX: 703-873-4372
Email: canderson@endyna.com

Irv Baumel, Ph.D.
USEPA/NHSRC/TCAD
Phone: 202-564-2338
Email: baumel.irwin@epa.gov

Dominic L. Boccelli, Ph.D.
Environmental Engineer
USEPA/ORD/NHSRC/WIPD
26 West Martin Luther King Drive (MS  163)
Cincinnati, OH 45268-1320
Phone: 513-569-7654
FAX: 513-487-2555
Email: boccelli.dominic@epa.gov

Kathyrn Boyle
Chemist
USEPA/OPP
1200 Pennsylvania Avenue, NW (7506P)
Washington, DC 20460
Phone: 703-305-6304
Email: boyle.kathryn@epa.gov

Nichole Brinkman
Biologist
USEPA/NERL
26 West Martin Luther King Drive (MS  320)
Cincinnati, OH 45268
Phone: 513-569-7315
FAX: 513-569-7117
Email: brinkman.nichole@epa.gov

Karen Burgan
Senior Policy Advisory
USEPA/OSWER/OEM/NPPD
1200 Pennsylvania Avenue, NW (5104A)
Washington, DC 20460
Phone: 202-564-1978
FAX: 202-564-2620
Email: burgan.karen@epa.gov
Dan Chappie
Battelle
10300 Alliance Road, Suite 155
Cincinnati, OH 45242
Phone: 513-362-2600
FAX: 513-362-2610

Kathy Clayton
USEPA/ORD/NHSRC
26 West Martin Luther King Drive (MS 163)
Cincinnati, OH 45268-1320
Phone: 513-569-7046
Email: clayton.kathy-ci@epa.gov

Maura J. Donohue, Ph.D.
Chemist
USEPA/ORD/NERL/MCEARD/CERB
26 West Martin Luther King Drive
Cincinnati, OH 45268
Phone: 513-569-7634
FAX: 513-569-7757
Email: donohue.maura@epa.gov

Anthony Fristachi, M.S.
Exposure Analyst
USEPA/ORD/NCEA
26 West Martin Luther King Drive  (MS A110)
Cincinnati, OH 45268
Phone: 513-569-7144
FAX: 513-487-2539
Email: fristachi.anthony@epa.gov

Bernard Gadagbui, Ph.D.
Toxicology Excellence for Risk Assessment
2300 Montana Avenue, Suite 409
Cincinnati, OH 45211
Phone: 513-542-7475 ext.  27
FAX: 513-542-7487
Email: bgadagbui@tera.org

Robert Goble, Ph.D.
Research Professor and Director
George Perkins Marsh Institute, Clark University
950 Main Street
Worcester, MA 01610
Phone: 508-751-4612
FAX: 508-751-4600
Email: rgoble@clarku.edu

-------
Paul Harten, Ph.D.
Physical Scientist
USEPA/ORD/NRMRL
26 West Martin Luther King Drive
Cincinnati, OH 45268
Phone: 513-569-7045
Email: harten.paul@epa.gov

Stephanie Hines
OSU Extension - Clermont County
1000 Locust Street, P.O. Box 670
Owensville, OH 45160
Phone: 513-732-7070
Email: hines.180@osu.edu

Sheldon Jobe
EnDyna, Inc.
7925 Jones Branch Drive
Suite 5300
McLean, VA 22102
Phone: 703-873-4367
FAX:  703-873-4372
Email: sjobe@endyna.com

Barbara Klieforth
Biologist
USEPA/ORD/OSA
1300 Pennsylvania Avenue, NW RM B26J
Washington, DC 20004
Phone: 202-564-6787
FAX:  202-565-2431
Email: klieforth.barbara@epa.gov

Steven S. Kueberuwa, Ph.D.
Toxicologist
USEPA/OW/OST/HECD
1200 Pennsylvania Avenue, NW
Washington, DC 20460
Phone: 202-566-0233
FAX:  202-566-1139
Email: kueberuwa.steven@epa.gov

Jason C. Lambert, Ph.D.
ORISE Fellow
USEPA/ORD/NCEA
26 West Martin Luther King Drive (MS A110)
Cincinnati, OH 45268-1320
Phone: 513-569-7078
Email: lambert.jason@epa.gov

Josh Larson
Biosecurity Analyst
Sandia National Laboratories
P.O. Box 5800 MS 1371
Albuquerque, NM 87185
Phone: 505-844-0357
FAX:  505-284-8870
Email: iilarso@sandia.gov
Todd Martin, Ph.D.
Research Chemical Engineer
USEPA/NRMRL/CPB
26 West Martin Luther King Drive (MS 443)
Cincinnati, OH 45268
Phone: 513-569-7682
Email: martin.todd@epa.gov

Deborah McKean, Ph.D.
Toxicologist
USEPA/OSWER/OEM/NDT
26 West Martin Luther King Drive
Cincinnati, OH 45268
Phone: 513-487-2435
FAX:  513-487-2537
Email: mckean.deborah@epa.gov

Leroy Michelsen
Engineer
OSWER/OEM/NDT
26 West Martin Luther King Drive
Cincinnati, OH 45268-1320
Phone: 513-487-2431
FAX:  513-487-2537
Email: mickelsen.leroy@epa.gov

Matthew D. Miller, Ph.D.
Post-doctoral research associate
University of Missouri - Kansas City
7543 Terrace Street
Kansas City, MO 64114-1637
Phone: 816-277-8264
FAX:  816-235-6543
Email: mdma95@umkc.edu

H.A. Minnigh
RCAP Solutions, Inc./CECIA, UIPR
P.O. Box 48
Lajas, PR 00667
Phone: 787-392-7186
FAX:  787-892-2089
Email: hminnigh@compuserve.com

Vlasta Molak, Ph.D.
President and CEO
GAIA Foundation, Inc.
8987 Cotillion Drive
Cincinnati, OH 45231
Phone: 513-521-9321
Email: drmolak@gmail.com

Tonya Nichols, Ph.D.
USEPA
26 West Martin Luther King Drive (MS 163)
Cincinnati, OH 45268-1320
Phone: 513-569-7805
Email: nichols.tonva@epa.gov

-------
                                                                     Appendix  E
                            Workshop  Presentation  Materials
       Introduction to  The
         VFARs Concept
              Jerry Stelma
             June 20, 2006
Although this work was reviewed by EPA and approved for
presentation, it may not necessarily reflect official Agency policy
                                                            Research and
                                                        Development at EPA
                   1,950 employees
                   $700 million budget
                   $100 million extramural
                   research grant program
                   13 lab or research facilities
                   across the U.S.
                   Credible, relevant and timely
                   research results and
                   technical support that inform
                   EPA policy decisions
         Making decisions with sound science
           requires..
            • Relevant, high quality, cutting-edge research in
             human health, ecology, pollution control and
             prevention, economics and decision sciences
            • Proper characterization of scientific findings
            • Appropriate use of science in the decision
             process
         Research and development
           contribute uniquely to..
            • Health and ecological research, as well as
             research in pollution prevention and new
             technology
            • In-house research and an external grants
             program
            • Problem-driven and core research
High Priority Research Areas
                    Human Health
                    Particulate Matter
                    Drinking Water
                    Clean Water
                    Global Change
                    Endocrine Disrupters
                    Ecological Risk
                    Pollution Prevention
                    Homeland Security
          The Contaminant
        Candidate List(CCL)
    Developed as a result of the 1996
    amendments to SDWA
     • EPA must periodically develop a list
      of currently unregulated
      contaminants
     • EPA must select 5 contaminants for
      regulatory decisions per 5 years
        The Contaminant
      Candidate List(CCL)
   Method for developing the lists
   not specified by SDWA
   Methods for selecting the five or
   more contaminants not specified
   by SDWA

-------
  NRC Workshop Results

  "Identifying future Drinking Water
  Contaminants" Recommendations
  • A process was developed to narrow,
    focus and prioritize contaminants.
  • Explore the feasibility of using
    virulence factor activity relationships
    (VFARs) for microbial contaminants
   Origin of the concept

 Structural activity relationships (SARs)
 found in chemicals
    • Compares newly identified chemical
     structures to known chemical structures
    • Toxicity is predicted by the
     comparisons
 Premise
 • Architectural and biochemical components
   of microorganisms that cause disease are
   also structurally related
       Central Concept

Ability to predict virulence by
microbial characteristics
 • Microbial VFARs should function much
  the same as QSARs do in chemistry
 • Research has shown certain common
  characteristics among pathogens
 • "Descriptors" have been tied to specific
  genes
    Why would WG expect
   structural relationships
       among genes?

  Parallel evolution
  Horizontal gene exchange
  • Common occurrence within a genus
  • Has been observed beyond genus
    boundaries
  Genetic engineering
 Examples of Descriptors

 • Genetic elements
 • Surface proteins
 • Toxins
 • Attachment Factors
 • Metabolic pathways
 • Invasion factors
 Current Challenges to use
    of VFARs/Microarrays

QSARs vs VFARs:  Does the
biological  universe parallel the
chemical universe?
 • Chemicals are static
 • Microbes are dynamic
 • Examples of parallel VFs
   • Cholera toxin and E. coli LT
   • Pyrogenic toxins of Strep, and Staph.

  ;	....... i 	...,•..  • .,.

-------
 Current Challenges to use
    of VFARs/Microarrays


 • Examples of parallel VFs
    • Cholera toxin and E. coli LT
    • Pyrogenic toxins of Strep, and Staph.

 • Examples of unique VFs
    • Salmonella invA gene
    • Legionella mip gene
 Current Challenges to use

    of VFARs/Microarrays

Host susceptibility factors and dosages

DMA Variability among structural genes

Effect of unexpressed virulence genes?
 • Genes can be present but not expressed

Are VFARs valid for viruses and protozoa?
 • All are obligate parasites
 • Factors leading to species specificity?

Effect of DMA from dead cells
     Current Challenges to use
       of VFARs/Microarrays


Too many unknown virulence  genes

 • Individual virulence genes are necessary for virulence

 • Individual virulence genes are not sufficient for virulence

 • Entire arrays of virulence genes are needed
     Current Challenges to use
       of VFARs/Microarrays

  "The message is that there are known knowns
  - there are things that we know that we know.
  There are known unknowns - that is to say,
  there are things that we now know we don't
  know. But there are also unknown unknowns -
  there are things we do not know we don't
  know. And each day we discover a few more
  of those unknown unknowns".
  Rumsfeld 2003

-------
           Using VFAR in a
     Risk Assessment Framework
                Joan B. Rose
               rosejo@msu.edu
 Homer Nowlin Endowed Chair for Water Research

     Risk assessment is a method to
        examine qualitatively or
     quantitatively the potential for
         harm from exposure to
   contaminants or specific hazards.
     Monitoring and data are some of the keys to
     establishing risks and therefore safety goals.
                                                      antitative Risk AssessnK
^Tool used to estimate adverse health effects
  associated with specific hazards.
^Elicits a statistical estimate or probability of
  harm.
^Used for risk management decisions.
   NATIONAL ACADEMY OF SCIENCES
      RISK ASSESSMENT PARADIGM
  HAZARD IDENTIFICATION
  Types of microorganisms and disease end-points
IDOSE-RESPONSE
  Human feeding studies, clinical studies, less
  virulent microbes and health adults
^EXPOSURE
  Monitoring data, indicators and modeling used to
  address exposure
^RISK CHARACTERIZATION
  Magnitude of the risk, uncertainty and variability

-------
              PROBLEM FORMULATION
   ANALYSIS
                  CHARACTERIZATION

                 ^^o, ,r= I    of Human
             of Exposure
                           Health Effects
             RISK CHARACTERIZATION
           RISK MANAGEMENT OPTIONS
Hazard ID
Dose-response
Exposure
Characterization
Source; Identification;
virulence; potential for severe
Source; persistence
(in nature, during
disinfection)


Sensitive populations
(receptors);  Evolution of Pathogens
                                                                                                 ewage  (Mark Wong, MSU)
                      "J atlorf iht btaitty 
-------
              Biological Hazards
 Viruses, prions, bacteria, and protozoa are more likely than fungi
 or helminths to be associated with emerging infections.
 Zoonotic pathogens comprise 75% of emerging infectious
 diseases.
 Pathogens which are subject to relatively frequent mutation or
 genomic reassortment events (e.g. RNA viruses and viruses with
 segmented genomes) are more likely to emerge.
 Pathogens which infect multiple hosts or pathogens that infect
 species that can harbour multiply closely related agents providing
 an opportunity for reassortment or recombination (e.g. SARS in
 cats) are likely to emerge.
   Dose-response data sets have been developed
in human feeding studies for
  Dose measurements were by PFU/or by
infectious titer, CPU or cysts or oocysts.
>End points of measurements were excretion of
the pathogen and/or antibody response, rarely
disease.
^Mathematically address the shape of the ratio of
those affected/exposed.
>Need minimum of three doses. Must have doses
which elicit effects different from 0% and 100%

-------
                                                             Strain Differences
                                                  Human volunteers,
                                                  C. parvum,

                                                  DuPont et al.
                                                  (1995)

                                                  Okhuysen et al.
                                                  (1999)

                                                  Potential for probabilistic
                                                  modeling of inter-strain
                                                  variability (Teunis and
                                                  colleagues)
                            Emunotbt
                            PUolftHnor)
                            PoUoHUpow)
                          —Mil
                            Echo 12
                          -Rotjvlria
        	-Qtartlt
   io-»  lo-4   10-'   10-*   10-'   io-°    t
                  Dose
                                                           Low Dose Extrapolation of Risks
                                                            /
        4
                                                               ,
           'f

                                                                u
                                                                    w
                                                                   'hi
                                                                                  - •- -nil III 1 KMI1IJAOH1AH
                                                                                  -o- •*«..,». no*— w
 EXPOSURE ASSESSMEN1

Route of Exposure
Duration of exposure
 - Seconds, hours, minutes
Number of exposures
 - How many times in a day, month, year
Degree of exposure
 - Liters of water ingested
 - Liters of air inhaled
 - Grams of food ingested
                                                         Microbial Source Tracking
•Tools are now available
to determine the
molecular fingerprint of
the fecal pollution.
•Health risks
•Remediation
                   .
•Prioritization
•Responsibility

-------
Host Specific Markers are Key to
      Source Tracking Future
Bacteroides (genetic approaches PCR)
4/4 sewage; 4/4 human; 4/5 cow (lowest concentration
missed) 4/4 dogs however no marker for Birds: Missed
2 samples with dog and 2 with cow that were mixed.

E.coli Toxin genes able to detect sewage (4/4).

Enteroviruses and Adenoviruses found in 3 of 4 sewage
samples.

Enterococc; T
 EXPOSURE ASSESSMENT

Occurrence
Survival
Regrowth
Accumulation
Transport

-------
                   KNOWN
                ;ENE FUNCTIONS
                (certainty of knowledge)
                   PRIMER SETS AND PROBES

                   WATERBORNE OUTBREAKS
       New Tools and Data bases for
     Assessing Occurrence and Safety
                                                         Understanding
                                                         Genetic
                                                         Detection in
                                                         Water
                                                         For disinfection.
                                                         Removal capabilities
                      Treatment vs. Influent: Endemic Risk
                           Log Reduction (or Aft cpUblc Risk
                                                                                OJB   0.1   1    10   100
                                                                                        InfluMif f/lOOL
                 QSARs
Quantitative Structure-Activity Relationship used by EPA for
over 13 years for hazard risk evaluation of chemicals part of the
new chemicals program.
First explored in 1950s by Hansch to correlate the molecular
structure to biological activity.
4500 citations
Software program (PBT profiler) developed just released 10
years in the making, (enter by drawing the structure, entering the
identifying # or written chemical linear structure.
Persistence (1/2 lives predicted ambient conditions.
Bioaccumulation
Toxicity (acute and chronic fish toxicity)
Predictive, some uncertainty, limitations (does not do metals,
endocrines).
Can place them into chemical categories.
Defines high, medium and low risk.
       WATERBORNE DISEASE
         GENOMICS PROGRAM
A Long-term Commitment to Developing the Data,
    the Technology, Supporting Analyses,
    Algorithms and Research projects including a
    Program in Functional Genomics is necessary.
Recognition that this is a predictive approach to
    examining risk and uncertainty will be part of
    the program.
May not work for all classes of Microbes equally.
                                                        THANK YOU
                                                        Science for

                                                        Societal Benefits.

-------
                          MICHIGAN s|.\Il
                                      i
     VFAR: Factors Related to Genomic Variabilities
                    US EPA QSAR/VFAR Workshop
                           Cincinnati, OH

                            June 20,2006
                              9:50 AM
                       Syed A. Hashsham
                                                              FACTORS RELATED TO DEVELOPMENT
                      ie Overall Concepi
         Tourlousse et al., Water Environment Research, 79(2007)




                                                                                                        Depth: Strain
Strain
KI2
W3I10
EDL933
                Pathotype
  * : '"' :'
  . I   .
EHEC
                EHEC
                UP EC
         fa 197   EIEC
   . .  . •• /       EIEC
EHEC: Enterohemorrhagic E. coh', UPEC: Uropathogenic K colt EIEC: Enteroinvasive E. coh
                                                                        toxR of V. parahaemolyticus     tap of L. monocytogenes    eae of-S- co
                                                                                                     .AF5322G9
                                                                                                      AF5322B
                                                                                                   	I-AF532235
                                                                                                    "-AF532302
                                                                                                   I

-------

Wide (and Dynamic) Range of Genetic Variability

Pathogen
Eml
Hpyton
H pylori
E. cot
H pylon
L monocyfoowiM
L monocytoywos
V pinhaemolytKus
C perinnyens
V cMfrae
V.cbohrae
V patahaemolyticus


analyzed length
<
MM
ytcA
cagA
stxIA
ureA
up
flcB
ah
pic
SxA
cbcB
toxR
Dynamic
^«i
No.)
76
77
89
41
36
42
116
20
18
30
33
20
: Chan
, 2006. l¥ate
(°P>
2822
3896
3585
948
717
1440
870
570
1197
777
375
879
ges
Emir
diversity
P>P)
356,4
306.6
2189
22.7
17.0
29.5
202
108
14.9
1.7
2.6
61
as the

<*>
126
79
61
2.4
2.4
20
23
12
0.2
07
0.7
database
vch. Special Issue
variability
(M (%)
564 200
602
388
77
37
67
I
28
17
7
14
grows!
(Accepted')
155
08
81
52
47
43
35
23
22
19
16


                                                                 Ranking

                                                                                          !
                                        Pipeline
                                                                                              *«««7
                                                                                              FCTI'
                                                                                          UK
                                                                                     -•»   u»
                                                                                 -„•_;       JIM   .
                                                                                 .»*»
        Hundreds of genomes are now available!
Activities (as in VFAR) are not always available;
Virulence and Marker Genes (VMG) Database: Dynamic
                                          18 to 50-merx
                                                    5S1
 Tow 47   90
I^CL Organisms Compared to the VMG Database
                                                                     Tourfoasse el al. \\ ater Environment RurCHf^li. ' y (2007)
                             10
                             Number of hits

-------
FACTORS RELATED TO MONITORING
1. Aeromonas hydrophila
2. Burkholderia pseudomallei, mallei
3. Campylobacterjejuni
4. Clostridium perfringens
5. Enterococcus faecalis, faecium
6. Escherichia coli, Shigella
7. Helicobacter pylori
8. Klebsiella pneumoniae
9. Legionella pneumophila
10. Leptospira interrogans
11. Listeria monocytogenes
12. Mycobacterium avium, paratuberculosis, tuberculosis, leprat
13. Pseudomonas aeruginosa
14. Salmonella typhimurium DT104
15. Staphylococcus aureus
16. Vibrio cholerae, mimicus, vulnificus
17. Vibrio parahaemolyticus
18. Yersinia enterocolitica, pestis, pseudotuberculosis
19. Cryptosporidium parvum, hominis
20. Giardia lamblia, intestinalis
                       Probe Resolution
                   Ease of Universal Amplification
                                                                                      Multiplex Amplification- A Must!
                                                                          Multiplex PCR-amplification followed by DNAchip-based amplicon identification
                                                                          Without Multiplex Amplification
                                                                             ~1 % of the population
                   With Multiplex Amplification
                          DMA
                             multiplex PCR-ampllflcatloi


                     mixture of amplicons


                           t
                  identification of amplicons using DNAchip


                       0.01 to 0.0001%
                      Wick et al., 2006  Nurtrir ArJi ftfxarr*. 20O*. \ »•/. J4. Ai>. f ttt
                  iWPfe,
                 ru  *7 42 « 47 44 43 43 «7 44 43 41 50 SI 43 39 47 40 i
?!U^^''^^^7r?ur^H^
                                                                                  Replication of Complex Target Mixtures
                                                                                             Sample Size vs Signal Strength

-------

Screening for All Known VMGs
Tourlousse et al. Water Environment Research, 79 (2007)
VFAR D«tnba»fi
^-On*. \
• Vmj- HK:KM SOM
. tAotetfr, mof«T/
-Sunmvinw
^ff • ~iiirt»tr in fci-*M» ^^.
j^ •, .n»-u ,' >.
Development of VFAR
High Thr.»igh|>u zer nwvtrmn
^ L. lf^*« J










"TuLswTk^ Ecolichi Protocols B
Jean Marie Rouillard Probe design
Yongmei Xia Target synthesis . vV
Trinh Pham Goal Labeling ~\jt> N1^^
Doctoral candidates: '&
Robert Stedtfeld Pathogen chip/
Sam Baushke Bioinformatics/PCR-chip /^^^
Dieter Tourlousse Functional genes \PfPE'
Ruifang Xu Protocol optimization
Yu Yang Target gene
Munir Ahsan Chip modeling
Research Associates/MS:
Sarah Miller Time optimization
Vidya Srinivasan Gold laebeling
Un dergradu ates:
Amanda Herzog Hybridization
Pis:
James Tiedje Syed Hashsham Erdogan Gulari
James Cole Joan Rose Thomas Whittam
                Funding
    National Institutes of Health-NCRR
Michigan economic Development Corporation
            MSU Foundation
          Department of Defense

-------
       VFAR Analysis and
       Characterization
 /lolecular Radar™ Biological
  " Jentification Technology

 Highly multiplexed nucleic acid
 hybridization based approach
 Target unique and virulence related
 genetic regions
 Micrqarray format allows for
 identification of tens of thousands
 of individual sequences  in parallel
 "Complete" genetic
  1  iracterization within 4-24 hours
      Capabilities Offered

  Simultaneously identify multiple pathogen,
      in level resolution
  Identify signs of genetic engineering
  Characterize unknown organisms
  Virulence factors and antibiotic resistance
  characterization
  Functional equivalent of a 10,000-fold
  multiplexed PCR reaction
  Technology is adaptable to multiple
     forms and applications
  System Concept
sav Develooment   Routine Samole Analvsis
Computer
identification of
informative
DNA/RNA
sequences
Identification of
candidate
oligonucleotides
On-chip synthesis
of oligos
Whole-genome
amplification with
label incorporation
Hybridization on
chip
Spot profile
identifies
sequences present
in original sample

nique sequences ueneratei
       FIGUR Software
   1C VER 1 Pathogen Array

 Sequences selected following
    ial screening arrays with
 JAIC funding
 Organisms arrayed in groups to
 aid rapid visual analysis
 Bioinformatics required for
 detailed strain level analysis

-------
          Bacillus anthracis Amei
         Sterne on  SAIC VER 1 Arra
B anthracis Sterne
                               B anthracis Ames
               pXO1
               pXO2
                       Spot Color
                       Purple
                                                       F.tulartnsts r.pestls Y.pseudotuburojioaisYentwocolittca hutian




      xampies o
      available
      Arrays

-------
             ... v.-
              ,
    ..icroarray Design
   Various Pathogen;
          ummary
Molecular Radar™ provides high fidelity
identification and virulence factor
characterization of microorganisms
We have achieved resolution down to
the level of strain for pathogens and
near-neighbor organisms
We can design and validate arrays for
any DMA or RNA containing organism
at desired level of resolution
Array can be tailored to different levels
of fidelity
Capability exists today to analyze
 sample!

-------
From Reactivity to Regulation:
    Integrating Alternative
Techniques to Predict Toxicity

           Mark Cronin

   School of Pharmacy and Chemistry
    Liverpool John Moores University
             England
             Models

          Mechanisms

             Modes

            Madness
 Cytotoxicity vs Hydrophobicity for
   Approximately 500 Chemicals

     c
     H
      J	
       •2-10123456

               Log Kow
 An Unspecific Mechanism (Non-
Polar Narcosis) is Easily Predicted
     3 '


     2 •

     >-!
     .-SB '



     H
       -2-1012345
              log Kow
 Unspecific Bioreactive Compounds
     H
     -1 -
       -2-10123456

               log Kow

    Toxicity = 0.65 log Kow - 0.34 Elumo - 1.11

        n = 353 r2 = 0.86 s = 0.35
    Toxicity of Specifically Acting
    Electrophiles is Underpredicted
                                           H
        -2-1012345
                                                     log Kow

-------
  The Toxicity of Specifically Acting
 Electrophiles is Poorly Modelled by
          QSAR Approaches
  we are not very good at
      parametrising
        reactivity

   reactivity is poorly
       quantified
Qvavri^vxfxtiov oty
        LO 5u|x|>ixuXT
 quantification of
reactivity is difficult

 reactivity is  not
 well parametrised
Quantitative Assessment of Reactivity:
     Glutathione Reactivity Assay

 • An olefin conjugated to a carbonyl group, is
  inherently electrophilic
 • Potential to act by Michael-type nucleophilic
  addition to macromolecules
 • Measured GSH reactivity is related directly to
  cytotoxicity

       log Toxicity = 0.95 log GSHreactivity + 0.54
         n=46   r2 = 0.91 s = 0.27 F = 460

       SchultzTWetal(2005) SAR QSAR Environ. Res. 16: 313-322
 Reactive Mechanistic Domains:
    Electrophiles in Toxicology
   • Michael acceptor
   • SNAr
   •SN2
   • Schiff base
   • Acyl transfer
   • Metabolically activated compounds
                          In Chemico Assays for Reactivity:
                       Spanning the Electrophilic Mechanisms
    Other Toxicity Endpoints with
      Electrophilic Mechanisms
  Skin sensitisation
  Respiratory sensitisation
  Carcinogenicity/ mutagenicity
  Skin irritation
  Inhalation irritation
  Liver toxicity
  Idiosyncratic drug toxicity
                        Are they the Same Mechanisms?
                        • Chemically the mechanisms are the
                          same, the target and endpoint differ
                        • Useful information may be obtained if
                          we can extrapolate this information

-------
Application to Regulatory Problems

• New chemicals legislation will require
  - Increased risk assessment
  - Potential increase in animal testing
  - Increase in cost
• There is an incentive for the greater
  use of alternative methods
• We know we have a problem
  predicting "reactive toxicity"
• How can we implement our
  knowledge of reactive toxicity across
  endpoints
     Alternative Methods:
 Integrated Testing Strategy
Existing Data for the Compound
   or Similar Compounds
  Computational Methods
     QSAR, In Silica
    Further Assessment
Prediction Models for Reactive Toxicity:
Application of in Chemico Measurement


  Using gutathione reactivity as a model soft nucleophile:

If Michael addend, plGC50 = 1.01 pEC50(GSH) + 0.57
   Schultz TW et al (2005) SAR QSAR Environ. Res. J6: 313-322

If R(GSH) > -0.55, chemicals are Skin Sensitisers
   Aptula AO et al (2006) Toxicol. in l/«ro20: 239-247
  Can we go in chemico to in silico?
      Defra LINK Project:
            Work Plan
           Conclusions

• Specific reactivity is poorly
  parametrised in toxicology, but
  underpins many endpoints
• Measuring reactivity in chemico has
  been shown to assist in predicting
  reactive toxicity better
• Needs for more reactivity data,
  computational capability and
  strategies for implementation

-------
     Integrated QSAR-PBPK
 modeling for risk assessment
           Kannan Krishnan
      Universite de Montreal, Canada
                                             Outline
   Introduction
   QSAR-PBPK: Development
   Risk assessment applications
   Conclusion
QSARs - Current Paradigm

  NOAELs vs chemical structure or props.
  Context-specific QSAR
  Duration of exposure (short-term)
  Oral route
  Species of interest (Rat)
m For a different route, species & duration
    Develop new sets of QSARs
    Develop "extrapolable" QSARs
QSARs - An alternative paradigm
DOSE


PK Tissue dose
* or Blood
QSAR Conn.


PD
QSAR


EFFECT

H QSAR
• Relative contribution of the TK and TD processes
• Extrapolations based on TK determinants
QSAR: PK-TK
  QSAR models are based on response-
  specific dose level for each species
• No efforts on the relationship between
  structure and internal dose
  Can we develop QSARs for pharmacokinetic
  profiles ? (changing as a function of route,
  dose and species)
  Inhalation, steady-state, rats...
Blood Concentration at Steady-state
                                CH,-CH,-X
                           | [Steady-state]
                                CH,-X

-------
  Blood concentration vs structure
                                                 Structure vs Blood concentration

                                                    C = [2x-3.25] +[lx6.8]
                                                      = 3.25
  PBPK Models
                    Inhaled ehamieal
                                 	i
                                   B
                            Brain     I

                          •*•—I
                            Kktn*y»
                          Rest oHlM txrty ^  I
      Physiology, partition coefficients, metabolic clearance

QSARs for PBPK Parameters

• Fragment constant approach

  Multilinear regression (SPSS®)
  46 VOCs, Fragments: CH3, CH2, CH, C, C=C,
  H, Cl, Br, F, B-ring, 2 E1 substrates
• Cross-validation, external validation
  QSAR-PBPK Modeling
Physiology
          Simulations
                                                  QSAR/PBPK modeling - Rat

                                                       Toluene
                                                    Methyl chloroform
                                                                           DCM
                     Trichloroethylene

-------
  Chemicals in the application domain
  Trifluoromethane
  Dichlorofluoro methane
  Bromodichloromethane
  Bromoform
  Dibromofluoro methane
  Bromoethane
  1,1,1-Tribromoethane
• 2,2-Dichloro-1,1,1-trifluoroethane
  1,2-Dibromo-1,1,2-trifluoroethane
  1-Chloropropene
  1,2-Dichloropropene
t" 1,3-Dichloropropene
~~ 1,1-Dibromopropene
  1 -Bromo-2-chloropropene
Pentane
Tribromoethylene
Tetrabromoethylene
1 -Bromo-2-chloroethylene
m-Dichloro benzene
Propylbenzene
1,2,4-trimethylbenzene
m-chloromethylbenzene
Ethyl benzene
    QSAR/PBPK model - Ethyl benzene
                                  QSAR/PBPK modelrDichloromethane
   Interspecies extrapolation of tissue:air
     partition  coefficients using QSARs
                                  Risk Assessment
                                                         	  Risk = q* • d

-------
Risk Assessment
^tissue = Human PBPK model
q* = Animal PBPK model

QSAR-PBPK Models in
 risk assessment
 QSAR-PBPK models facilitate internal dose based
 risk assessment (lethal and non-lethal effects)
 Influence of exposure concentrations, routes and
 scenarios can be examined
 Effects on specific sub-populations can be
 evaluated
 Modeling of multiroute exposures for risk
 assessment applications
                 Fragor?

-------
  TERA
Toxicology EitvllcMi
     Weight of Evidence
    and Mode of Action
 in Predictive Toxicology

                Dr. Andrew Maier
                      and
             Dr. Raghu Venkatapathy  j
                  June 21,2006
                                                                   1"
                                                                  TER.-1


                                                                 Weight of Evidence (WOE)

                                                                      in Risk Assessment
                                                              Risk Assessment Initiatives
                                                               - U.S. EPA Cancer risk assessment - requires addition of
                                                                 a "weight of evidence narrative"
                                                               - Increasingly used in Hazard Screening Algorithms
                                                                 (e.g., Health Canada ComHaztool)
                                                               - Weight of evidence characterized by use of "totality of
                                                                 the evidence" in making decisions about causality
                                                               - Emphasis on "Totality" has opened door for
                                                                 predictive toxicity tools

                                                              Evolving concept driven by
                                                               - Improved biology understanding (understanding of the
                                                                 mode of action or MO A)
                                                               - Increased sophistication and validation of alternative
                                                                 study designs and consideration of study design (e.g.,
                                                                 gene knock-outs)
                                                               - Improved quantitative tools (including toxicogenomics
                                                                 and QSAR)
  TER-i
Role of Predictive Toxicity


       Empirical Data Confidence
          Inadequate    Adequate
                    r«
                   a
Collect
Data
Collect
Data
Characterize
Risk
Resolve
Conflict
                            = Candidates for Predictive Toxicity
                                                                    J
                                                                  TER-l
                                                                        WOE and QSAR
    J
  TKR.'i
 ikdoty Eitrll*iK«
 M- Rltk A»*f rawM
  , ,,. .. ,. ...

Tools for Evaluating WOE

 Hill criteria for causality
 Expert j udgment
  - Peer review/consultation
  - Expert elicitation techniques
     • Survey approaches
     • Software tools
 Quantitative tools
  - Decision and Uncertainty Analysis
  - Bayesian Analysis
J
TERA
TtUkdDtt r ici-Urnf r
to, Rbk A»tinneM
Biology understanding is needed for
interpreting results

Type of Damage
Point mutation
Oligonucleotide
insertion or
deletion
Allele Loss
Small
Chromosome
alteration
Large
Chromosome
alteration
Aneuploidy
Genotoxicity QSAR Modules
Mouse
Lymphoma
Yes
Yes
Yes
Yes
Yes
?
Chromosome
Aberrations in
CHO cells
No
No
No
?
Yes
Yes
Ames Bacterial
Mutagenicity
Yes
Yes
No
No
No
No
Adapted from M Moore (2004)

-------
TKRA
   Biological Understanding


Our level of understanding of the underlying
biological basis of toxic responses represents a
continuum.
For risk assessment we often distinguish between
knowing the mechanism of toxicity versus the mode
of action.
Mechanism of toxicity refers to a detailed
understanding to the cellular and subcellular level of
the basis for toxicity.
Mode of action  refers to a less detailed level of
understanding, but ability to identify key precursor
steps in the  pathway to toxic response.
                                                             TKRA
              Defining Mode of Action

           A critical challenge in integrating mode of
           action data in global QSARs is defining
           appropriate predictors:
           What does mode of action mean?
            - Target organ? (liver toxicity)
            - General cellular response (necrosis)
            - Subcellular target (ATP synthesis disruption)
            - Potential presence of reactive moiety
              (electrophiles, oxygen radicals)
  J
TER.A
       Problem Statement

 Currently SARs and QSARs are often used as
 independent tools, a practice that does not
 optimize what can be learned when the
 varying approaches are used in a coordinated
 manner.
 Approaches for developing consensus
 modeling approaches that use biology
 understanding (MO A) for integrating SAR
 and QSAR models are needed.
TERA     Goal - Maximizing Use of Biology
                                                                          Mechanism of action known. Develop
                                                                          mechanistic QSARs - excellent
                                                                          predictivity - but limited applicability
                                                                          Mode of action data available, develop
                                                                          hybrid or MOA-informed QSAR -
                                                                          balance of predictivity with applicability
                                                                          Biology unknown - use global statistical
                                                                          QSAR - decreased predictivity - but
                                                                          broad applicability
TER.\
Using MOA to Refine Statistical
                QSARs
                            Experimental

            Basic principles would indicate that correlations of similar
            chemicals would improve prediction.
                                                             TER-i

              MOA in Logistic Regression

           Endpoint specific SAR models are often
           designed as either expert system-based
           models or statistical models
           A hybrid approach that uses logistic
           regression analysis with a dummy dependent
           variable coded 0 and 1 (for negatives and
           positives,  respectively) can allow the input of
           key data derived from MOA decision rules.
           Probability of end point toxicity is:
                                                             lot

-------
  TER.-1
Toxicology EitvllcMi
Using "Omics" for Binning
              Toxicogenomics,
              proteomics,
              metabonomics already in
              use for hazard
              identification
              Used to identify MOA for
              hypothesis testing
              Public databases will be
              increasingly populated for
              data mining
              These data can serve as
              sorting variables to
              enhance QSAR
              development
                                 I
                                   u.
                                                             1"
                                                             TEK.I
 Regression Tree Approach
     D-2nod«
 .     |
        ~
                                 i
                                > 20
  TEJ£I      Role of MOA based hybrids
 riMto0bHMan
 r RUk Alsrumrnt
 iXSlrr^     Inform the interpretation of global QSARs -
 »»C.*U*.'14iMl».
             e.g., identifying critical endpoints.
             Serve as sorting variables to bin chemicals for
             development endpoint or MOA-specific
             QSARs.
             If MOA biomarkers are used as dependent
             variable, then serve as QSAR endpoint
             verifiable by relatively non-invasive tests.
             Binned (or nodes) can be used for assigning
             potency using group average or "Threshold of
             Concern" approach.
                                                 J
                                               TER-l
          Conclusions
WOE evaluation represents a maturation in
chemical risk assessment
Critical use in resolving conflicting data - are
assays or predictive tools testing the same
thing? Can differences be explained by the
MOA understanding?
Advances in basic biology (molecular and
cellular biology), chemistry (computational
chemistry), and mathematics (better statistical
and dose-response tools) should be used by
the risk assessment community
Tool developers should make full use of our
mode of action understanding

-------
                                         Host RT WOOD IOHNSON
                                         MEDICAL SCHOOL
            Novel Approaches to
         QSAR & VFAR Modeling

                 William (Bill) Welsh
                UMDNJ CompTox Center
                  welshvvj@umdnj.edu
       New Jersey Environmental Bioinformatics
           & Computational Toxicology Center

                      elCTC
                http://www.ebCTC.org

                   Funded with support from the
           U.S. EPA, National Center for Environmental Research
               Science to Achieve Results (STAR) Program

                 William Welsh, Center Director
                                                                  June 20-21, 2006
                                                                                     QSAR/VFAR Workshop, EPA-Cincinnati
             Consortium Members
                      ROBIRI WOOD (OHNi
                      MEDICAL SCHOOL
 " ; '•'< L LMNBH . •-•' NBA . • t
                                          University
                   Center for Toxicoinformatics, NCTR  wnl
             Major Research Thrusts
• DORIAN Computational Toxicology System that spans the Source->Dose->Outcome continuum
• The Environmental Bioinformatics Knowledge Base (ebKB)
• ebTrack, a toxicological bioinformatics platform to process genomics, proteomics and
 metabonomics data
• Hepatocyte Metabolic Model for Xenobiotics
> ChemTox, a suite of chem- informatics tools for toxicant identification, prioritization,
 characterization
  June 20-21, 2006
                     QSAR/VFAR Workshop, EPA-Cincinnati
                                                                                I  1  IT
    ebTrack System: Extension of Array Track
      - Bioinformatics Analysis of Microarray Data -
               S
           I •HI        '"'  .       "•
                             i,
  June 20-21, 2006
                     QSAR/VFAR Workshop, EPA-Cincinnati
   Overview of QSAR-based Approaches

• Decision Forest (DF)
  - fast consensus modeling technique that quantifies prediction confidence
• Shape Signatures
  - enables fast large-scale screening of query chemicals against databases
   based on similarity in shape and other biorelevant molecular features
• Polynomial Neural Network (PNN)
  - generates optimal linear or nonlinear QSAR models in parametric form
• Virtual High-Throughput Screening (vHTS)
   - predict & quantify ligand binding affinity to proteins
   - provide insights into mechanism of action (toxicity pathways)
   - assess validity of cross-species extrapolation (e.g., rat vs. human)
                                                                  June 20-21, 200I
                                                                                     QSAR/VFAR Workshop, EPA-Cincinnati

-------
                  Integrated Approach
Receptor-based Approaches
                                          Predictive
                                          Molecular
                                          Toxicology
                                                       I
                                         . Virtual Screening
   June 20-21, 2006
                      QSAR/VFAR Workshop, EPA-Cincinnati
                                  Computational Screening Paradigm
                                              - Priority Setting -
                                                                     June 20-21, 2006
                                                                                         QSAR/VFAR Workshop, EPA-Cincinnati
Schematic of Hierarchical Screening Framework
- addresses the need to minimize false negatives and uncertainties -










l^^j^^B Rejection Filters 1
1 1
Fn T^^M Active / Inactive Assignment f
1 I
K~ '' f^^l Quantitative Predictions 1
i I
Kr "t': /T^^BI Knowledge- Base Approach I
-MW
• Structural properties
• Structural alerts
- Pharmacophores
- Decision Forest
Shape Signature*


• PNN QSAR model

• Other priority setting 1



^^H



actors
• Modifying Tiers 1, 11. HI
* Human expert knowledge











; ov.rv.w
                                                                                 Decision  Forest  (DF)
                                                                       - improve classification by combining individual models -
                                                                           c           j
                                                                   Tree 1   Tree 2   Tree 3   Tree 4
                                                                      I	I	I       I
                                                                             Consensus
                                                                             Prediction
                                                                     Key Features
                                                                     *  Combining several independent yet predictive trees improves performance
                                                                     *  DF structure permits assessment of prediction confidence, reduces uncertainty
                                                                     *  Each tree consists of simple 'If-Then' branches, hence the DF is extremely fast
                                                                     June 20-21, 2006
                                                                                         QSAR/VFAR Workshop, EPA-Cincinnati
     Schematic of Hierarchical  Framework
                 - based on USFDA's EDKB -
     Tier I
     Tier II
     Tier III
     Tier IV
                     Rejection Filters
                Active / Inactive Assignment
                  Quantitative Predictions
• Other priority setting factors
• Modifying Tiers I. II. Ill
• Human expert knowledge
   June 20-21, 2006
                      QSAR/VFAR Workshop. EPA-Cincinnati
                                      Shape Signatures Tool
                                                                         START
                                                                                            PROCESSING
                                                                                                                    OUTPUT
  Small molecule or         Ray tracing to             1D and 2D
Protein binding pocket    generate the raw data      Shape Signatures
                                                                     June 20-21, 2006
                                                                                         QSAR/VFAR Workshop, EPA-Cincinnati

-------
              Shape Signatures Tool
     molecules are compared by subtracting their histograms
    1?p-estradiol
                                              Diff=0.082
       DES
Small Diff value means that two molecules have similar shape and polarity
  June 20-21, 2006
                     QSAR/VFAR Workshop, EPA-Cincinnati

                                                                    Shape Signatures User Interface


                                                                   June 20-21, 2006
                                                                 Shape Signature Databases
                                                                General Database >4 million chemicals
                                                                Kinase, GPCR, NR ligand databases
                                                                PDB-extracted ligand database
                                                                Receptor binding sites of 30,000 proteins (BWAs)
                                                                Hazardous Chemicals (EDCs, H20 CCLs, DSSTox,
                                                                      R Worksn[)p EPA-Cincinnati              14
Chemical T> Target Protein -> Mechanisms
        Protein Data Bank (PDB): World Repository of -35,000
      Protein-Ligand Crystal Structures (http://www.rcsb.org/pdb/)


      Shape Signatures of PDB-extracted ligands
   Hn* BC UK R«*ulL»
   Protein Structure
                      y
                  0"
  June 20-21, 2006
                  Species/Protein Family
QSAR/VFAR Workshop. EPA-Cincinnati	15
                                               Molecules ^^Target Protein ^* Mechanism
                                                                           Shape Sigs PDB Ligands
                                                                                                              Protein's Binding Sib
                                                                                                        '
                                                                                              Links to
                                                                                              Biological
                                                                                              Pathways
                                                                   June 20-21. 2006
                                                                                      OSAR/VFAR Workshop. EPA-Cincinnati
        Identifying Problem Chemicals
              & Possible Surrogates
   surrogate
   chemicals
  June 20-21, 2006
     QUERY CHEMICAL
              — Shape Signatures Libraries
                     QSAR/VFAR Workshop, EPA-Cincinnati
                                               Discovery of Previously Unrecognized EDCS
                                                                                t                       """-	 ,V,S
                                                                     amoxifen    !  -       -'i..|.-R,H.«,nM>  «	 ;£T
                                                                     (query)
                                                                   June 20-21, 200i
                                                                                      QSAR/VFAR Workshop. EPA-Cincinnati

-------
  Shape Signatures: Discovery of Anthrax LF Inhibitors
  June 20-21, 2006
                                                S503428 docked in the
                                            ligand binding pocket of anthrax LF.
                       QSAR/VFAR Workshop, EPA-Cincinnati
                                                                                Key Features of Shape Signatures
                                                                        >  Innovative: Encodes molecular shape and other biorelevant features in a
                                                                           single entity
                                                 > Non-congeneric: Finds hits missed by techniques that search on chemical
                                                   (sub)structure

                                                 > User Oriented: fast, simple, expandable

                                                 > Versatile: works for any number or type of molecular species (organics,
                                                   organometallics, ions, etc.)

                                                 > Applicable in ligand-based mode (ligand-ligand similarity) and receptor-
                                                   based mode (ligand-receptor complementarity)
                                                                           June 20-21, 2006
                                                                                                 QSAR/VFAR Workshop, EPA-Cincmnati
Schematic of Hierarchical Framework
- based on USFDA's EDKB -










Rejection Filters
|
| Active / Inactive Assignment ,'
I
; ~ | Quantitative Predictions >
I
'^H^B I Knowledge-Base Approach
•MW
• Structural properties




• Structural alerts
• Decision Forest


- PLS QSAR models
- PNN QSAR model <



<•— i

• Other priority setting factors
• Modifying Tiers I, II. Ill
• Human expert knowledge










June 20-21, 2006 QSAR/VFAR Workshop, EPA-Cincinnati 21
                                                                                Polynomial Neural Network  (PNN)
                                                                         combines the parametric form of PLS and the nonlinearity of ANNs -
                                                                                  Polynomial Neural Network

                                                                                                                Produces linear or non-linear
                                                                                                                QSAR models in parametric form
                                                                                                                User control of model complexity
                                                                                                                Insensitive to irrelevant variables
                                                                                                                and outliers
                                                                                                                Yields predictive models, even for
                                                                                                                 sparse or noisy data sets
                                                                                                                Trains rapidly, thus amenable to
                                                                                                                 large data sets
                                                                                                                Automatically selects best indep.
                                                                                                                variables; no preprocessing requirec
                                                                                                                Customizable to fit user's needs
                                                                           June 20-21, 2006
                                                                                                 QSAR/VFAR Workshop. EPA-Cincinnati
   Biological Data -> VFAR Databases & Models
Biological Input Data
* Regulated species
• Virulence factors (VF)
   Knowledge Base
 ' Candidate species
     • Taxonomy
     • Ecology
 • Candidate data
  June 20-21, 2006
    VFAR Models
• VF input sequence
 and structures
• Virtual Screening
    * Sequences(BIAS
    * Domains (BLINK)
    * Structures (VAST
• Dynamic Updating

                       QSAR/VFAR Workshop, EPA-Cincinnati
                                                                         Bacterium -» VF -» VF Structure -» Candidate Structure
                                                                          Picking one
                                                                          EcoliVF
                                                                              T
                                                                                     5 co/i Cytotoxic
                                                                                    Necrotizing Factor
                                                                                      Typel (1HQO)
                           Conserved hypothetical protein
                            from Caulobacter crescentus
       Vector-aligned Domains:
       • Red - identical AAs
v-1-     • Blue - non-identical AAs
                                                                                                                           Vector alignment of
                                                                                                                           the two structures
                                                                           June 20-21, 2006
                                                                                                 QSAR/VFAR Workshop, EPA-Cincinnati

-------
            Thank You!



          welshwl@umdty.edu
June 20-21, 2006
             QSAR/VFAR Workshop, EPA-Cincinnati

-------
• (UtOMAN COMMISSION
I Joint Reuvdi Centre

  Role of the European Chemicals Bureau in Promoting
  the Regulatory Implementation of Estimation Methods
           US EPA QSAR / VFAR Workshop, 21 June 2006
                       Andrew Worth
                  European Chemicals Bureau
         Institute for Health & Consumer Protection (IHCP)
        Joint Research Centre (JRC), European Commission
                     21020 Ispra(Va), Italy
http://ecb.jrc.it/QSAR                   E-mail: andrew.worth@jrc.it

                                                                        I lUROPfAN COMMISSION
                                                                        IjojnttaarcliCinln


                                                                                                Outline
                                                                           1. The Joint Research Centre (JRC) & the European
                                                                             Chemicals Bureau (ECB)
                                                                           2. Use of estimation methods under REACH
                                                                           3. ECB research in computational toxicology
                                                                           4. ECB assessment of methods and models
                                                                           5. Promoting (regulatory) acceptance and implementation
                                                                           6. Training & capacity building
•
a>
o>
o
u
CO
0»
CO
ec
•*"•*
"5
• EUROPEAN COMMISSION
1 Joint Research Centre
The European Commission's Joint Research
| European Commission
x^s.
| Directorates-General < JRC
XN "
Directorates or Institute";
XN
[Units | European Chemicals B
European Centre for th
Physical & Chemical E
tre •;-„,
''CX
-1 f
'/A- vis
E£B
/
4
-!/[/''


ureau (ECB)
e Validation of Alternative Methods (ECVAM)
iposure(PCE)Unit "llffP

•
O
o
CO
CO
CO
co
"£
"o
• lUCOPf AN COMMISSION
1 Joint Wirch Ceinn
The European Ch
Assessment of
chemicals
Existing Substances
New Substances
Biocides
Frnort / /mnort
lA IUCI.IOB
REACH IT &
Informatics
REACH-IT for Chemicals
Agency
IUCUD5
emicals Bureau: http://ecb.jrc.it
• f REACH Support
Guidance & tools for industry
& authorities
F^R
MkP
"|E"
—

Computational
Toxicology
Development, validation,
acceptance and implementation
of estimation methods
IUROPtAN COMMISSION

I Joint noeirch Craln                                           ECB
         Information requirements under REACH

 Standard information requirements for chemicals are largely
 tonnage dependent, however:
 •  Annex VI Specific requirements are context-dependent
 •  Annexes VII-X Standard information requirements
 •  Annex XI "Adaptation" of standard information requirements:
   - replacing traditional test data with predictions or equivalent data
   - providing standard information at lower or higher tonnages
   - exposure-based waiving (Annexes VII & VIII, a100 tonnes)
   - providing additional information (if necessary)
 =>"Intelligent" rather than box-ticking approach to information gathering
                                                                      • EUROPEAN COMMISSION
                                                                      I Joint Research Cenlre
                                                                                   Integrated Testing Strategies (ITS)
                                                                                       Endpoint-specific
                                                                                             strategy
                                                                                               T
                                                                            p&L, risk assessment, PBT (vPvB) assessment!

-------


I EUVOPf AN COMMISSION
I Joint Rtsnirch Conln                                               FC R
              Annex XI of REACH - (Q)SARs

 Results  obtained  from valid  qualitative or quantitative structure-
 activity relationship models ((Q)SARs)  may indicate the presence or
 absence of a certain dangerous property. Results of (Q)SARs may be
 used instead of testing when the following conditions are met:

 •  results are derived  from a (Q)SAR model whose scientific validity
   has been established
 •  the  substance falls within the applicability domain of the (Q)SAR
   model
 •  results are adequate for the  purpose of classification and labelling
   and/or risk assessment, and
 •  adequate  and reliable  documentation of the applied method  is
   provided
                                                                            I ElMOPf AN COMMISSION
                                                                            I Joint Rniiirck Cmln                                               FC B
                                                                                       Annex XI of REACH - Categories (1)
                                                                             Substances    whose    physicochemical,    toxicological    and
                                                                             ecotoxicological properties are likely to be similar or follow a regular
                                                                             pattern as a result of structural similarity may be considered as a
                                                                             group or "category" of substances.
                                                                             Application  of the  group concept requires that  physicochemical
                                                                             properties,  human  health  effects  and  environmental  effects or
                                                                             environmental  fate  may  be  predicted from data for a  reference
                                                                             substance within the group by interpolation to other substances in the
                                                                             group (read-across  approach). This avoids  the need to  test every
                                                                             substance for every endpoint.
                                                                             ... If the group concept  is applied, substances shall be classified and
                                                                             labelled on this basis.

I EUCOPf AN COMMISSION
I Joint Be'seircri Centre                                               f(
           Annex XI of REACH - Categories (2)

 In all cases results should:

 •   be adequate for the purpose of classification and labeling and/or
    risk assessment
 •   have adequate  and reliable coverage  of  the  key parameters
    addressed in the corresponding test method referred to in Article
    12(2)
 •   cover an  exposure duration comparable to or longer  than the
    corresponding test method referred to in Article 12(2) if exposure
    duration is a relevant parameter, and
 •   adequate and  reliable  documentation  of  the  applied  method
    shall be provided
• ElttOPtAN COMMISSION
Joint R«eirch Centre ECB

g
CD
o

08
CD
CO
OP

'o
—5
Chemical category - administrative view











Property 1
>roperty 2
>roperty 3
>roperty 4


ftctivity 1
ftctivity 2
ftctivity 3
ftctivity 4

^
* -^^

**'
| •


I °
.
o
0

•v
0
™%
•


o
.
o
0

— ,.
.
jf"
•


0
.
o
0

o
.

•


o
.
o
0

SAR/read-
frfiBrjfelation
extrapolation
D — i
T
QSAR
^ ^ 	 1

• reliable data point
O missing data point
I fUtOff AN COMMISSION
I Joint Research Centre
                  Chemical category - QSAR
                              view


                               Principal Components Analysis based on
                                      connectivity indices
        95% confidence interval
        (possible boundary)

                                                                            I EUKOPiAN COMMISSION
                                                                            I Joint Research Centre
                                                                                        Read-across assessment of ETBE
HT CjHjjO MF CtHuOi ^^•Mj^EuO 1
SMILES swms sanrn^. i

Sywnxymt EVopue. Sw-wipm 2J- Syn»*ya» Methyl f
2-tortiwxj^- 4aieth]fl-prop«)03c lcrt-«myl dbn-. 2- !
ractliyl-. radii?] ( Mid, methyl ester. M«byl 2 i
(ut^l niicf. Mtt^l Unkvi piviliw. mntixytutiK, ]
Physicochemical properties,
mutagenicity, sensitisation, aquatic
2-ethoxy-2-methylpropane
H^C 0
CH3 V
-CH3

-------
 03
o
I lUfiOPf AN COMMISSION
I Joint Reseirch Centra
      ECB research on computational toxicology (1)

 1.  QSARs for aquatic toxicity (& modes of action)
 2.  QSARs for bioaccumulation
 3.  QSARs for sensitisation
 4.  QSARs for endocrine disruption
                                                             ECB
                                                                            • fUROMAN COMMISSION

I Joint Reselrch Contra                                            FC F
      ECB research on computational toxicology (2)

 5. Methods for chemical similarity analysis and grouping
    ECB workshop on TTC and grouping methods (Nov 05)
    ED / OECD Guidance on grouping
 6. Methods for descriptor-based ranking of chemicals
    Workshop on Ranking Methods with Italian Chemometrics Society &
    Milan Bicocca University (2-4 October 06)
 7. Methods for defining (Q)SAR  applicability domains
 8. Weight-of-evidence in  hazard  & risk assessment
    ECB workshop on consensus modeling (Sept 05)
 9. Computational nanotoxicology
    I lUOOPf AN COMMISSION
    I Joint Reseirch Centra
 03
CJ

                  Chemometric Ranking Tools (1)
    Possible
    subcategories
    based on PBT
    ranking
                         Total Order Ranking of 323 phthalate
                           esters based on Utility Function
                               N. observations
                                                                        I EUROPEAN COMMISSION
                                                                        I Joint Research Centre
                                                                                       Chemometric Ranking Tools (2)
                                                                                                       Total Order Ranking of 323 phthalate
                                                                                                       esters based on Dominance Function
                                                                                                      Possible subcategories
                                                                                                      based on predicted PBT
                                                                                                           profile
    I EUROPEAN COMMISSION
 O3
O
     Joint Retard! Cmtri
            ECB assessment of methods and models
 1.  QSAR validation studies (2005-2006)
    Acute fish toxicity, skin penetration, skin sensitisation, steroid
    hormone receptor binding
 2.  Validation of BfR rulebases
    Skin and eye irritation / corrosion
    Eye irritation
 3.  Validation of TerraQSAR™ FHM model  -I Jj . -jb ,
 4.  Beta testing of AIM
 5.  Beta testing of AMBIT software for QSAR applications
    http://ambit.acad.bg
I EUROPEAN COMMISSION
I Joint Research Centra
           Promoting (regulatory) acceptance

 1.  EU Working Group on QSARs
    Capacity building among regulators and industry
    Scientific & technical preparations for REACH
 2.  OECD ad hoc Group on QSARs
    Principles for QSAR validation (adopted)
    Practical guidance on QSAR validation (under review)
    Case studies on regulatory acceptance (completed)
    ECB hosted meeting on 8-9 June 06, Stresa, Italy
 3.  OECD Validation Management Group for Non-Animal
    Methods
    ECB coordinates QSAR Task Group
                                                                                                                                    -tip

-------
    I EUffOKAN COMMISSION

    I Joint Research Centre
 Oi
CJ
        Promoting implementation: ECB QSAR Inventory
ECB is designing a QSAR
Inventory
Oracle database
Will be available via ECB
Website
Integration with  ESIS
(European chemical
Substances Information
System)
Uploading of models via
ECB website
(QSAR Reporting Formats)
Quality check of models by
ECB
                                                                               I EUtOPf AN COMMISSION
                                                                         I Joint neiurcft Centre
                                                                              Structure-searchable Interface to ESIS and QSAR
                                                                                                  Inventory
                                                                                                                          Structure- searchable
                                                                                                                                interface
    I [WOMAN COMMISSION
    I Joint Research Centre
            Promoting implementation: Estimation tools
                                       ToxTree estimates toxic hazard by
                                       applying the Cramer classification
                                       scheme (33 structural rules)
                                       Groups chemicals according to
                                       structure for Threshold of
                                       Toxicological Concern estimation
                                       Developed by Nina Jeliazkova
                                       (Ideaconsult Ltd, Sofia, Bulgaria)
                                       (http://ambit.acad.bg) under ECB
                                       contract
                                       Flexible - can be adapted to
                                       encode different structural rules
                                       Freely available from ECB website
                                                                        I [IMOPf AN COMMISSION
                                                                        I Joint flesiiitli Cralra
                                                                            Where to find ECB QSAR tools:  http://ecb.jrc.it/QSAR
    I EUROPEAN COMMISSION
    I Joint Research Centra
                         Capacity building
    1.  Training on (Q)SARs for regulatory and industry end-users
       1st ECB course: 19-21 October 2005. Sofia, Bulgaria
       2nd ECB course: 24-25 July. Ispra, Italy
    2. Training on decision analysis
      ECB / INERIS workshop planned
      (Nov-Dec 06)

    3. Information tools via ECB website
      Danish QSAR database
                                        Dubfe(Q)5ARDti*biM

I EUtOPf AN COMMISSION
I joint Research Centra
                 Challenges for the future


 1.  Need for capacity building (stepping stone to acceptance)
    Training courses, workshops, learning by doing

 2.  Establishing the basis for REACH-implementation
    Chemical databases and tools for property estimation
    Guidance and criteria for use of (Q)SARs and grouping methods:
    "Manual of Experience"

 3.  Research to fill the information gaps
    ITS and its component parts, e.g. new, tailor-made (Q)SARs
    New methods for applicability domain assessment and chemical
    similarity analysis

-------
  • lUflOPtAN COMMISSION
  IjolntResiircliCentri                                             ECB
                    The ECB QSAR Team
  Arianna Bassan
     Chemoinformatics, QSAR tools, computational nanotoxicology
  Ana Gallegos Saliner
     Chemical similarity, skin irritation
  Tatiana Netzeva
     Environmental QSAR, consensus modeling, training and enlargement
j Grace Patlewicz
     Human health QSAR, decision analysis
  Manuela Pavan
     Ranking methods, environmental QSAR
  Ivanka Tsakovska
     3D QSAR modeling, eye irritation
  Andrew Worth
     Human health QSAR, Integrated Testing, regulatory applications--. On

-------

-------
&EPA
     United States
     Environmental Protection
     Agency
     Office of Research and Development
     National Homeland Security Research Center
     Cincinnati, OH 45268


     Official Business
     Penalty for Private Use
     $300
             Recycled/Recyclable
             Printed with vegetable-based ink on
             paper that contains a minimum of
             50% post-consumer fiber content
             processed chlorine free
PRESORTED STANDARD
 POSTAGES FEES PAID
         EPA
   PERMIT NO. G-35

-------