Quantitative Mechanistically Based Dose-Response Modeling withEndocrine-Active CompoundsMelvin E. Andersen,1Rory B. Conolly,2 Elaine M. Faustman,3 Robert J. Kavlock,4 Christopher J. Portier,5 Daniel M.Sheehan,6 Patrick J. Wier, and Lauren Ziese8'Department of Environmental Health, Colorado State University, Fort Collins, Colorado USA; 2Chemical Industry Institute of Toxicology,Research Triangle Park, North Carolina USA; 3Department of Environmental Health, University of Washington, Seattle, Washington USA;4National Health and Environmental Effects Research Laboratories, U.S. Environmental Protection Agency, Research Triangle Park, NorthCarolina USA; 5National Institute of Environmental Health Sciences, Research Triangle Park, North Carolina USA; 6National Center forToxicological Research, U.S. Food and Drug Administration, Jefferson, Arkansas USA; 7SmithKline Beecham Pharmaceuticals, King of Prussia,Pennsylvania USA; 8California Protection Agency, RCHAS/Office of Environmental Health Hazard Association, Berkeley, California USA

A wide range of toxicity test methods is used or is being developed for assessing the impact ofendocrine-active compounds (EACs) on human health. Interpretation of these data and theirquantitative use in human and ecologic risk assessment will be enhanced by the availability ofmechanistically based dose-response (MBDR) models to assist low-dose, interspecies, and in vitroto in vivo extrapolations. A quantitative dose-response modeling work group examined the state ofthe art for developing MBDR models for EACs and the near-term needs to develop, validate, andapply these models for risk assessments. Major aspects of this report relate to current status ofthese models, the objectives/goals in MBDR model development for EACs, low-dose extrapolationissues, regulatory inertia impeding acceptance of these approaches, and resource/data needs toaccelerate model development and model acceptance by the research and the regulatorycommunity. Key words: endocrine-active compounds, endocrine disruptors, linkage modelsmechanistic dose-response modeling, pharmacodynamics, pharmacokinetics, risk assessmentextrapolations. - Environ Health Perspect 1 07(suppl 4):631-638 (1999).http.//ehpnetl.niehs.nih.gov/docs/1999/suppl-4/631-638andersen/abstract.html

The potential for various compounds to alterendocrine system function has received increas-ing public attention in the United States [e.g.,Food Quality Protection Act, 1996 (1), SafeDrinking Water Act Amendments of 1996(2)] and in other countries throughout theworld. The recognition of this potential toxic-ity has led to debate about the ability of cur-rent testing methods to identify endocrinesystem effects throughout the full gamut oflife stages. New screening assays and test pro-tocols for reproductive and developmentaltoxicity have been developed and others arecurrently being evaluated (3). In addition, alarge number of new mechanistic test systemshave been developed to evaluate interactionsof endocrine-related compounds with specifichormone receptors. Together, these effortswill provide more comprehensive characteriza-tion of the potential hazards posed by expo-sure to these compounds. Coordinate withdevelopment of these new tools for hazardidentification and new mechanistic tests is aneed to create a set of refined dose-responseassessment tools that use as much of this newdata as possible (4).

A workshop was held in May 1998 inwhich several subgroups focused on approachesfor characterizing the effects of endocrine-activecompounds (EACs) on human health at envi-ronmental exposure levels. One work groupaddressed issues related to development ofmechanistically based dose-response (MBDR)models for EACs, emphasizing the potential

role of mechanistic models in improving thescientific foundations of dose-response assess-ments for EACs. This report is the product ofthat work group.

Dose-Response ModelsCurrently, default dose-response assessmentapproaches differ for cancer and noncancerend points. Default carcinogen risk assess-ments assume that all doses of a carcinogeniccompound carry some degree of risk (5).Noncancer end points, including reproductiveand developmental toxicity, have traditionallybeen regulated by assuming that these responseshave a threshold. No observed adverse effectlevels obtained from toxicity tests are adjustedby the application of uncertainty factors toderive reference doses or reference concentra-tions. The reference concentration methodol-ogy (6) for inhaled compounds includesdefaults to calculate doses of inhaled com-pounds in specific regions of the respiratorytract. The newly proposed U.S. EnvironmentalProtection Agency guidelines for carcinogenrisk assessment emphasize the role of mode ofaction and tissue dosimetry (i.e., mechanisticdata) in supporting departure from the linearcancer defaults. Consideration of both dosime-try and mode of action is essential in producingdose-response assessments that make maximaluse of available data and reduce uncertainties.

Quantitative dose-response models fortoxicology relate adverse response outcomewith exposure duration and intensity. These

include empirical models that derive modelparameters from fitting response data, modelsthat incorporate limited mechanistic informa-tion, and finally, models that include expo-sure, dosimetry, tissue interactions, mode ofaction, and biologic responses in an integratedand more quantitative fashion. These lattermodels, explicitly incorporating mode ofaction and tissue dosimetry data, are referredto here as MBDR models. This summary ofthe dose-response work group provides back-ground information regarding mechanisticmodels for EACs. It stresses the potential forthese models to improve the precision of riskestimates below the range of sensitivity in cur-rent test methods (usually at the 5-10% inci-dence levels or at the 5-10% increase in acontinuous measure of response) and to reduceuncertainties in risk assessments with EACs.

Empirical Dose-ResponseModels in Risk AssessmentParameters derived from fitting empiricalmodels to response data do not necessarilyhave specific biologic meaning or bear one-to-one relationships with particular biochem-ical or molecular parameters. Nonetheless,these empirical approaches are still importantfor assessing the range of response behaviorsassociated with exposure to these compounds(7). Empirical models are especially useful ifthey are capable of describing a wide range of

This report was developed at the Workshop onCharacterizing the Effects of Endocrine Disruptors onHuman Health at Environmental Exposure Levels held11-13 May 1998 in Raleigh, North Carolina.

Address correspondence to M.E. Andersen, TheK.S. Crump Group, Inc., 3200 Chapel Hill-NelsonHighway, Suite 101, PO Box 14348, Research TrianglePark, NC 27709. Telephone: (919) 547-1723. Fax: (919)547-1710. E-mail: andersenme@aol.com

*Current address: Dept. of Environmental Health,CETT/Foothills Campus, Colorado State University, Ft.Collins, CO 80523. Telephone: (970) 491-8253. Fax:(970) 491-8034. E-mail: manders@cvmbs.colostate.eduWe thank S. Cecil of ICF Kaiser for technical assis-

tance in preparing the text and working with the vari-ous authors. We also thank H. Barton for reading thepaper and offering helpful comments to improve orga-nization and clarity of presentation.

Received 9 December 1998; accepted 24 March1999.

Environmental Health Perspectives a Vol 107, Supplement 4 - August 1999 631

ANDERSEN ETAL.

Biochemical linkage Physiologically

Blood concentration Receptor binding Two stagee ssue concentration Receptor-DNA cancer models

Metabolite binding BBDRconcentrations Enhanced developmental

Area underthe curve transcription modelsOther measures of Reduced Models for

tissue dose transcription enhanced cellEnzyme inhibition replication

Figure 1. Schematic of the modular components of a mechanistic model. BBDR, biologically based dose response.

curve shapes, provide a measure of a potencyof response, and are useful for assessing time-dependent aspects of the test system. Thebenchmark dose methodology (8-10) appliedto noncancer end points and discussed inthe revised carcinogen guidelines (5) is anexample of this dass of model structures.

Other useful attributes of empirical mod-els would include their ability to describebackground information, including priorexposure, background incidence, heterogene-ity, and variability, and to permit extrapola-tion to other groups of chemicals. Usuallythese empirical models are tailored to specificdata sets. The accumulation of informationfrom these independent quantitative analysespermits the integration of multiple responses,comparisons of potency, and the cataloging ofthe shapes of response curves that need to beevaluated in more detailed mechanism-basedmodeling. With limited information regard-ing the biologic system, it would be possibleto create hybrid models that induded specificbiologic parameters such as protein or recep-tor binding affinity constants, enzyme activi-ties, or receptor number (11-14). Thesebiologic parameters would be embeddedwithin the otherwise empirical model. Otherapproaches that consider the pharmacologyof ligand-receptor signaling have alsoreceived attention from the pharmacologycommunity (15).

Goals and Expectations forMechanistic Models forEndocrine-Active CompoundsGeneral CharacristicsMore comprehensive mechanistic models aredesigned to include descriptions of bothpharmacokinetics, especially the time coursefor distribution of compounds to targettissues, and pharmacodynamics, i.e., theinteractions between compounds and targettissues. Biochemical linking models connectthe pharmacokinetic and pharmacodynamicmodels. The linking models specify themanner in which the chemical alters critical

biochemical processes and initiates the seriesof steps leading to toxicity. Thus, a morecomplete mechanistic model for EACsshould consist of a number of modular ele-ments, e.g., a pharmacokinetic module, alinkage module for receptor-based interac-tions, and a module for tissue response(Figure 1). Obviously, the parameters inthese models are expected to correspond withspecific biochemical, physiologic, andanatomic characteristics of the test systemand test compound.

The advantage of a modular structure isto show clearly the manner in which datafrom multiple studies and disciplinesbecome tightly integrated into the overallMBDR model. The goal in pursuing devel-opment of these integrated models is togain insight from the quantitative mecha-nistic analysis of responses in animals topredict exposure outcomes for the healthof human and wildlife populations. In thisregard, inclusion of population characteris-tics such as genetic variability may beessential for ensuring that these mechanis-tic models become more useful for riskassessment. MBDR models with these gen-eral characteristics are intended to playspecific roles in chemical risk assessment.Five of the more important roles area) understanding the expected shape of theresponse curve based on biologic principles;b) understanding differences in response forstructurally related compounds, differentgenders, strains, organ systems, life stages,and animal species; c) applying better extrap-olations for specific exposure situations inexposed human populations; d) accountingfor nonlinearities in pharmacokinetics andpharmacodynamics; and e) accounting forgenetic variability and other factors thatcontribute to differential sensitivitiesamong subpopulations.

Although MBDR models attempt to cap-ture the salient features of a given processand/or effect, they are always simplificationsof the real world. Nonetheless, even sim-plistic mechanistic models may be useful in

summarizing information, generatinghypotheses, challenging scientific knowledge,and clarifying the key issues to considerwhen extrapolating across routes, acrossspecies, to lower exposures, and across differ-ent ages (16-18). In their initial stages,MBDR models are designed to explain lim-ited sets of data and are limited in the con-clusions they can support. They have to bemodified as new information becomes avail-able, as interest in a given effect is increasedby either scientific curiosity or regulatoryscrutiny, and most especially, as our under-standing of basic endocrinology and biologyadvances. The molecular biology revolutionof the past 20 years has provided a broadnew set of tools for analyzing molecular sig-naling and cellular control processes. Theinsights and data arising from use of thesebiologic techniques will provide continuingchallenges for refining and extending MBDRmodels at all levels of structural organization.In MBDR models the normal biology shouldbe described quantitatively and then theimpact of the xenobiotic examined as aperturbation of the normal state.

Models ofEndocrine System FunctionEndocrine organs secrete hormones thattravel throughout the body, affecting distanttarget cells (endocrine), neighboring cells(paracrine), and cells of origin (autocrine).These hormones control processes involvedin maintaining normal development andfunction of the organism. The molecularmachinery of the target organs transduces thehormonal signals to create biologic responses.EACs have the potential to alter these signal-ing, recognition, and transduction processes,leading to mild perturbations, altered bio-logic function, or overt toxicity. MBDRmodels of endocrine function eventuallyinclude all these molecular characteristics, aswell as information on the distribution of thekey hormones to active tissues, hormone syn-thesis and metabolism, hormone binding toplasma proteins, interactions of multiplereceptor isoforms (e.g., estrogen receptors aand P) and physiologic interactions that serveto maintain homeostasis. Key aspects of theseprocesses include receptor synthesis, recy-cling, and degradation; tissue-specific regula-tion and activation of receptors duringdifferent life stages; and control of enzymesthat synthesize or metabolize ligands.Homeostasis in adult organisms is regulatedby important feedback controls for manyendocrine processes. In addition, an impor-tant aspect of modeling in the endocrinesystem is consideration of multiple time-dependent phenomena. These behaviorsinclude dynamics of organism development,circadian rhythms, puberty, estrous or men-strual cycles, and reproductive senescence

Environmental Health Perspectives * Vol 107, Supplement 4 * August 1999632

DOSE-RESPONSE MODELING OF ENDOCRINE-ACTIVE COMPOUNDS

and aging. In these life stages, the coordina-tion and timing of sequential events by a vari-ety of receptors and natural ligands organizesa set of events that evolves over time.

Once the normal endocrinology has beenconsidered in a model for effects of EACs,inclusion of the major biologic processesaffected by individual EACs and theirmetabolites becomes the key focus of the nextstage of modeling. This concept of the char-acterization of the major steps deserves specialemphasis in regard to an evolving under-standing of signal transduction pathways,including gene transcription, translation, andposttranslational modification and regulation.It would be difficult or impossible to accountfor the behavior of all these factors in anycontemporary model. However, identifyingand modeling the essential characteristics thataccount for the potency of exogenous com-pounds in the mechanistic description may besufficient to significantly improve the scien-tific credibility of current risk assessments.Many biologic factors may contribute to themajor potency-determining steps in a process,and the specific step involved may vary fordifferent exposure situations. Among the fac-tors that may have to be considered are tim-ing of exposure, amplification of multistepcascades, multiple receptor isoform interac-tions, and the presence of tissue-specificaccessory proteins, such as co-activators andco-repressors.

When empirical models are applied tomultiple data sets, there is no reason to expectconsistency in the estimates of parametersacross different experiments. With MBDRmodels, there should be consistency in valuesof specific parameters. Although receptornumber varies across age, gender, and speciesand is affected by cycling and circadianrhythms, basic parameters such as receptornumber, affinities, and metabolic characteris-tics should be similar as long as the same age,gender, tissue, and species of animal are usedfor studies. In this fashion, MBDR modelshave the ability to synthesize a broad range ofscientific observation into a coherent descrip-tion of normal and altered biologic states. Assuch, these models reflect the present state ofknowledge about EAC toxicity and provide aquantitative organization of prevailing, domi-nant hypotheses regarding the modes ofaction of hormones and EACs on endocrine-regulated processes. The ability to createquantitative hypotheses of the mode of actionand the use of these hypotheses to designcritical experiments for their verification aresimply the applications of the scientificmethod to problems in toxicology and riskassessment. One of the strongest argumentsfor expanding the use of MBDR models inEAC risk assessments is that they allowdevelopment of testable hypotheses related to

mode of action and to the shape of thedose-response curve at low doses.

Endocrine systems and the impact ofEACs on these systems appear to be readilyamenable to MBDR modeling techniques.Much is known about the regulation of organsystem function by natural hormones andabout the interactions among variousendocrine systems. This body of informationcan be combined with data on the perturba-tions of the endocrine system by specificclasses of EACs to develop comprehensiveand testable hypotheses. It is important torealize that these organized mechanistic mod-els are a first-step hypothesis compilation.Optimally, these models should be used inexperimental design in confirming or refutingthe original hypothesis and for risk assess-ments based on the most plausible prevailingperception of modes of action. To accumu-late appropriate data for improving the bio-logic basis of risk assessment with theseEACs, special effort is necessary to developprotocols that simultaneously fulfill specificdata needs for modeling and specific dataneeds for regulatory testing requirements.

Model EvaluationThe ability to organize and explain a broadarray of diverse scientific findings is a primarygoal of mechanistic modeling and is a keycomponent for achieving broad scientificacceptance. A difficulty in gaining acceptancefor these models occurs because of the veryrequirement that the models organize andintegrate such a wide variety of experimentaldata. These ambitious structures may ade-quately describe the majority of studies butfail to match all observed results. It will benecessary to use a broad array of scientific evi-dence in characterizing and assessing the suc-cess of mechanistic models. The degree ofemphasis to be placed on particular studiesand the congruence of the model with theobserved results will depend on both statisticalissues and scientific judgment.

In many situations, it has been difficult tocharacterize how scientific judgment affects aregulatory decision because of the inability toexpress quantitatively the different risk esti-mates derived from differing assumptionsregarding modes of action for xenobiotics.MBDR modeling uses diverse expertise todetermine key biologic aspects of a modelsuch as causal linkages between exposurechanges and biologic effect. The explicit artic-ulation of the mechanistic assumptions inthese models shows their impact on the riskpredictions. Thus, the process of risk assess-ment becomes more transparent (objective)and the ability to test these assumptions(challenging the model and/or validating itsuse) is greatly enhanced. The integration andanalysis of hypotheses can provide a more

explicit incorporation of scientific judgmentinto the process of evaluating the impact ofspecific mechanistic assumptions in risk assess-ments. The coupling of modeling results withelicitation of expert judgment has been con-sidered a potential tool for achieving consen-sus in the use of mechanistic data andmechanistic models in risk assessment.

Mechanistic models, by their nature,attempt to describe in mathematical detail theprocesses involved in generating an adversehealth effect from an environmental expo-sure. This has several clear advantages. First,when data are available on interindividualvariation in response to any element of theprocess, this information can be directlyincorporated into the use of the model forprediction using various population-orientedmethods (19,20). Second, mechanistic mod-eling provides the potential for generalizingresults from one chemical to other agents. Forexample, key components of a model, such asthe mechanism for stimulation of folliclegrowth and ova release by follicle-stimulatinghormone and luteinizing hormone, oncecharacterized, can be used in models for otherenvironmental agents that affect the samemechanisms. Additionally, once models havebeen developed for prototypical agents in aclass of chemical agents, structure-activityrelationships can be used to determine phar-macokinetic model parameters for other con-generic compounds (21,22) or bindingparameters for other chemicals with similarbiologic activity (23,24). More simple appli-cations of mechanistic data may providereduction in experimental costs, i.e., usingexisting models to estimate toxic potency onthe basis of alterations in key parameters.Predictive models may find use in prioritizingcompounds for testing. High-exposure com-pounds that are predicted to have highpotency would be candidates for immediatetesting. Streamlined approaches become pos-sible when the components of the modeldescribe mechanisms that can be clearly iden-tified as parts of the cascade of events leadingto toxicity for other agents.

Risk Assessment ApplicationsMBDR modeling can provide scientificsupport for the shape of the dose-responsecurve in the low-dose (or low-incidence)region. A representation of the potential roleof biomarkers and mechanistic models inextending the region of dose where confidentextrapolations are possible is shown in Figure 2.

Mechanistic studies can be used todetermine elements of the system that give riseto the shape of the dose-response curve belowthe range of observation of overt toxic effects,e.g., the studies that evaluated the shape of theprotein induction curve for CYPlAl bydioxin at doses much below those that caused

Environmental Health Perspectives a Vol 107, Supplement 4 * August 1999 633

ANDERSEN ET AL.

100 -

-)

0

c0

0.0CLa-

0

Region ofextrapolation

Accessible withbiomarkers andperhaps withMBDR models

Region whereovert toxic

responses areobserved withclear increases

above background

,-

(7

NOAELDose -

Figure 2. Range of incidences where validated MBDRmodels coupled with biomarker measurements couldimprove predictions of shapes of dose-response curves.MBDR, mechanistically based dose response; NOAEL,no observed adverse effect level.

overt toxicity (25). Carefully designedexperiments and quantitative organization ofthe experimental data into a model should pro-vide a more precise determination of theexpected risks at different exposure levels.More generally, the development of mechanis-tic models can actually provide support for therelationship between responses and biomarkersat most levels of exposure.

Mechanistic modeling has other uses inaddition to dose-response assessment. Thesemodels have been used to identify data gapsin our understanding of the toxicity of EACsand to identify key experiments to fill thesegaps. In addition, by locating the critical ele-ments governing the potential potency of anagent for a given effect, mechanistic model-ing aids in identifying useful short-term,cost-effective testing strategies that directlycontribute to the prediction of risks andstrengthen the level of evidence needed todetermine if a hazard exists. Optimally, sev-eral mechanistic models with differingunderlying biologic mechanisms can becompared against various data sets to dis-criminate between competing mechanisms(25). Experimental simulations with thesecompeting models prior to acquisition ofnew data can identify the parameters withthe greatest impact on model predictions.These parameters should be investigated ini-tially to provide the greatest ability to dis-criminate between competing hypotheses.With EACs, this could involve comparingmodels, assuming receptor agonist activity,antagonist activity, or mixed activities beforedeciding on a specific experimental design totest the predictions.

Additionally, mechanistic models offer alogical framework in which to link exposureassessment and dose-response modeling.Extrapolation issues, when properly consideredprior to developing a mechanistic model, canbe more readily addressed. These include

cross-species, cross-organ, cross-route, andcross-compound extrapolations. Specific datawill be required for these model extrapolations;however, the amount of experimentationshould be considerably less than required inthe initial model development. Furthermore,issues associated with exposure to multipleagents working additively or synergisticallycan be addressed through model simulationsthat are followed by targeted experimentationto verify and/or refine these assumptions.

Other uses are in improving design oftoxicity tests and creating modules useful formultiple chemicals. Regulatory agenciesrequire that the chemical industry generatespecific types of data on many compounds.Mechanistic models could greatly improvethe value of these mandated toxicity testingresults for risk assessment. This interaction,however, would only be possible if there weremore flexibility in establishing protocols fortoxicity testing. In addition, experimentationwith prototype endogenous hormones (e.g.,estradiol, testosterone) could provide infor-mation helpful in predicting the response tomany other compounds with similar activi-ties. Such generic, natural hormone modelswould include modules containing pharma-cokinetics, receptor interaction, and tissue-response portions. This information would beimportant for future studies with agonist/antagonists of these native hormones.

Regulatory AcceptanceThere is a degree of skepticism about thewillingness of regulatory bodies to makedecisions based on mechanistic models.Sometimes these concerns are caused by theunfamiliarity of regulators and many researchscientists with these modeling techniques. Itdeserves emphasis that regulatory acceptanceof MBDR models should be secondary,following broader acceptance of thesetechniques by the scientific community.

Within the regulatory structures, there isappreciation that current risk assessmentshave many uncertainties. These uncertaintiesoften lead to legislation requesting newmechanistic data. However, the new data areonly infrequently incorporated in the riskassessment because the entire package ofresults is still regarded as incomplete orappears to have been collected in the absenceof a clear context for its use. MBDR modelsare an important avenue for providing thecontext for successful incorporation of variousnew data.

Another concern has been that the use ofmodels may provide a level of confidence thatis not justified because of the uncertainties inbiologic knowledge. In practice, the reverseappears true. These models define the individ-ual parameters that comprise the overallfunction of the endocrine system within an

organism. Each parameter in the model thenhas associated variability and uncertainty,together with uncertainty associated with thechoice of mode of action. Frequently, the mereidentification of the parameters in the modelsand the appreciation of associated uncertaintieshas provided an impression in the regulatorycornmunity of an increased level of unlcertaintywhen these models are proposed. This percep-tion was voiced in the initial attempts to incor-porate physiologically based pharmacokineticmodels in health risk assessments. However,the perception is based on an inappropriatecomparison of MBDR approaches and defaultrisk assessment models. Default methods aresimply sets of experiential rules. Although therisk assessments derived from application ofthe default procedures may lack precision indefining the true risk, there is no way to assessinherent uncertainties in the default process.The MBDR modeling approach does uncoverareas of uncertainty but explicitly definesthose areas that need the most attention inassessing uncertainties.

In addition to risk assessment applica-tions, mechanistic dose-response modelsmight also play a role in clinical diagnosis andmanagement of specific disease states. Forexample, mechanistic models might be espe-cially useful in assessing the degree of riskassociated with hormone imbalance in certainhuman endocrinopathies. An important clini-cal problem that might approached in thisway is hyperandrogenism in women.

Comparisons with ContemporaryApproachesIn their applications to risk assessment,MBDR models should be compared andcontrasted with the current default assump-tions. The hurdle for application of mecha-nistic models and mechanistic data shouldnot be set so high as to disqualify all but themost sophisticated and detailed of thesemodels from application in risk assessments.Clearly, the utility of any quantitative modelwill be derived from a clear articulation ofthe reasons for constructing the model andthe range of detail that is successfully cap-tured in the model structure. The level ofbiologic detail in mechanistic models variesdepending on the present state of knowledgeof the normal biology and for the biology ofspecific EACs. Although it is a laudable goalto include all of the details of the expectedbiologic interactions, this level of detail can-not be provided at this time or in the foresee-able future. Hence, procedures must be inplace to accept MBDR models as a meansof improving risk assessments as part of areiterative hypothesis-generating process,where continuing changes in these modelsmay occur as our understanding of complexbiologic systems improves. No risk assessment

Environmental Health Perspectives * Vol 07, Supplement 4 * August 999634

DOSE-RESPONSE MODEUNG OF ENDOCRINE-ACTIVE COMPOUNDS

should be regarded as the final word on acompound, as long as toxicologic researchand toxicity testing continues to accumulatenew findings and data. Risk assessments andMBDR models are simply recapitulations ofthe state of the science for a compound at aparticular point in time.

Information and Data NeedsMany types of information could be organizedand included within structured mechanisticmodels. Because of the breadth of studies thatcould be completed, it is important to priori-tize efforts and to remain focused on the mostpressing uncertainties and data needs. Therole of background hormone concentrationsin regulating and eliciting specific responsesneeds to be included in these MBDR modelsin order that estimates of risk above back-ground can be evaluated. A challenge inassessing normal function of the endocrinesystem is in defining the range of normalfunction and the perturbations of this range ofnormal function that would be regarded as anadverse response. This consideration is impor-tant for individual responses and for responsesof populations. Thus, the inclusion of thebehavior of endogenous hormones serves as acritical element of the overall model structure,

providing important perspective to theperturbations caused by incremental increasesin agonists or antagonists. The discovery thatcompounds serve as effectors for multiplereceptors, e.g., methoxychlor and its metabo-lites display both estrogen-agonist and andro-gen-antagonist activity, counsels some cautionin assigning single mechanisms of action tocompounds that have not been adequatelyscreened for interactions with these varioussystems. In assessing data needs for MBDRmodels, it is possible to focus on specifics inminute detail, such as the presence of hor-mone receptors in specific tissues with esti-mates of binding constants, receptor number,etc. Another approach is to broadly list thecharacteristics that appear to be important inconstructing these mechanistic models inorder to identify critical response-limitingparameters. In real-world applications, modelbuilding, data accumulation, and model test-ing and refinement will usually help guide dis-cussion about the level of detail required foruse of the models in risk assessment.

Current emphasis is on potential effects ofestrogenic, antiandrogenic, and thyroid-mimetic compounds on human health.Assessments for estrogenic compounds requireinformation on normal effects of estrogen

during different life stages (26,27). Modelingstudies with other endocrine systems includeandrogen control of spermatogenesis (4) andthyroid hormone-mediated control processes(28). The estrogenic system has multipleinteractions with other hormonal regulatorysystems, and it might be that simpler systemscould be examined to develop strategies andmodeling techniques in more sparsely con-nected hormonal systems. Blumenthal et al.(29) reported on the development of a phar-macokinetic model for the pineal hormonemelatonin. Androgen function in males isalso likely to be more tractable than the time-dependent control of cycling and pregnancyin females. A tabulation of some of thecharacteristics of available MBDR models forendocrine related toxicities appears inTable 1.

The end points of interest with agonistsand antagonists of sex-steroid action are repro-duction, development, immunologic func-tion, neurologic, and neoplasia. Models forreproduction and development are still intheir formative stages compared to risk modelsfor cancer. Despite some activities in mecha-nistic models for developmental responses(39), none of these models have beenorganized to correlate the concentrations of

Table 1. Attributes of MBDR models including some used to describe endocrine-related processes.

Dosimetry modeling

Examples

ChloroformFormaldehyde

Methylene chloride

Endocrine-relatedphysiologicprocess models

Toxicants affectingendocrine status

Melatonin

Testosterone regulation,antiestrogen

Estrus cycle atrazine

5-Fluorouracil

TCDD-NIEHS

TCDD

Endocrine-relatedphysiologic effectmodels

SAR models

Developmental models

Thyroid homeostasis(e.g., thionamide)

Leydig cell regulation(e.g., cimetidine,linuron, procymidone)QSAR

CH3HgCI and fetalgrowth

ADME

PBPKCFD/PBPK

PBPK

Natural hormone,PBPK

Feedback (T),

PBPK modelEmpirical model,

(E2) PKCompartmentalPK model

PBPK

PBPK

NA

A

PBPK model forCH3HgCI

Receptor/(ligand)interactions Al

Metabolizing enzymes CellDNA-protein cross-links, Cytctissue CH20 to

GST-mediated metabolism, PresGSH metabolite po

Putative brain T-receptor, T-LFtesticular LH receptor

Putative hypothalamic LH sE2-receptor

Enzyme inhibition Alte

Ah-receptor binding estradiol, Cellmetabolism EGF-receptor, Thyienzyme induction TSH

Ah-receptor, CYPlA1 Hepprotein induction ch.

TRH,TSH,T4,T3 TSH(interruption of normal sylinhibitory feedback dimechanism)

LH, T (interruption of normal Elevregulatory feedback loops) ce

incThyroid hormone AhR, AR, ERreceptors

Not receptor based Cell

ltered cell functions

Ikilling by metabolitesotoxicity, mutation dueCH20 itselfsumed initiation)tential

H-FSH feedback

surge

ered nucleotide pools

cycle alterationsrroid statusd levelspatic cell phenotypeianges

I-stimulated DNArnthesis and cellvision

vated LH or Leydig11 responsiveness toicreased LH

11growth

Altered biologicresponses

Toxicity/carcinogenicityClonal growth models fornasal tumorslMS cancer models

Altered hormonelevels

Persistent anovulatoryestrus, mammary tumors

Fetal growthMalformationsThyroid tumors, livertumors, and clonalgrowth

Clonal growth ofinitiated cells

Follicular cell tumors

Leydig cell tumors

Comparative evaluationsof potency

Fetal weight

Environmental Health Perspectives * Vol 107, Supplement 4 * August 1999

Reference

(30)(31)(32)

(2)

(4)

(33)

(11)

(34)

(35,36)

(37)

(38)

(13)(39)

Abbreviations: ADME, absorption, distribution, metabolism, and excretion; AhR, arylhydrocarbon receptor; AR, androgen receptor; CFD, computational fluid dynamics; EGF, epidermal growth factor; ER,estrogen receptor; E2, 17f-estradiol; FSH, follicular-stimulating hormone; GSH, glutathione; GST, glutathione S-transferase; LH, luteinizing hormone; LMS, linearized multistage; NIEHS, National Instituteof Environmental Health Sciences; PBPK, physiologically based pharmacokinetics; PK, pharmacokinetics; SAR, structure-activity relationship; TCDD, tetrachlorodibenzo-p-dioxin; QSAR, quantitativestructure-activity relationship; T, testosterone; TRH, thyroid-releasing hormone; TSH, thyroid-stimulating hormone; T3, triiodothyronine; T4, tetraiodothyronine.

635

ANDERSEN ET AL.

morphogens, including ligands and receptormolecules, with the occurrence of specificstructural abnormalities. Models for devel-opment need to consider feedback loopsand potentially include differential tran-scriptional activities of individual agonistsfor hormone receptors in different celltypes, in different regions of the embryo,and at different times during gestation.Models for developmental stages also havethe challenge of accounting for a dynamicsystem where sensitivity of the tissue to thehormone changes markedly over time (40).Low-Dose Extrapolations withEndocrine-Active CompoundsThe default position in noncancer riskassessments assumes a threshold, i.e., a dosebelow which there are negligible risks. A com-bination of empirical modeling and mecha-nistic data has been marshaled to suggest thatthere are certain developmental end pointsfor which thresholds are unlikely to occurand others for which thresholds would notbe at all unexpected (41). There do appearto be situations in which it would be realisticto consider that there is a continuum ofresponses (i.e., no threshold dose for theobserved effects) for added hormone agonists.For instance, when the natural outcome inadult reproductive system structure and func-tion is determined by the in utero exposuresto natural hormones, the addition of exoge-nous agonists should cause alterations in theresponse incidence. This expectation is con-sistent with additivity to background andwith results of studies of the effect of uterineposition on adult reproductive systemparameters in rodent species (42).

The influence of processes without clearthresholds for risk assessment must still con-sider the severity (adversity) of the responses.Some effects, such as prostate size, representchanges in specific phenotypic characteristicsthat are themselves variable in the adult pop-ulation. Risk assessment for these end pointswill revolve around definitions of adversity.The issues surrounding altered distributionof normal characteristics in the populationare complex, requiring both technical inputabout adversity and public policy input onthe level of tolerance for changes in thesedistributions by the public.

Several end points for which thresholdsare uncertain were discussed in our delibera-tions. These responses included changes inandrogen receptor number and prostateweight in adult male mice following in uteroestrogen exposures (42) and turtle sex ratiosfollowing egg painting with estrogens(43,44). Other examples were also dis-cussed, including fertility in a continuousbreeding study in females exposed to DESin utero (45) and vaginal threads in female

mice exposed to dioxin in utero (46).Although these effects appear consistentwith absence of threshold, they have notbeen examined statistically to determine theminimal threshold that would be consistentwith the data. Statistical analyses of thiskind would be informative and should beperformed routinely.

Some of the molecular characteristics ofgene transcriptional control by hormones andtheir receptors are expected to give rise tohighly nonlinear dose-response characteristicsdue to positive feedback loops, receptorautoregulation, phosphorylation cascades,and control of enzymes involved in synthesisof high-affinity ligands (47). These molecu-lar behaviors can give rise to biologicswitches, i.e., to the ability to abruptly changefrom one biologic condition to another over avery small change in ligand concentration.Examples appear to include estrogen receptorautoregulation during vitellogenesis in somefish and frog species (48) and thyroidhormone receptor upregulation in frogtadpoles during metamorphosis (49). Highlynonlinear effects were also reported in proges-terone-mediated maturation of Xenopusoocytes, a response mediated via mitogenic-activated protein kinase (50). Many of thesenonlinear switching mechanisms are expectedto produce nonlinear dose-response curvesfor the action of native ligands. However, thedose response for effects of exogenous com-pounds, even when biologic switches are pre-sent, still depends on the combination ofeffects of the native ligand and perturbationsof the EAC on the specific biologic effect.

Another aspect of the debate surroundingendocrine-active compounds is the concept ofpharmacokinetic thresholds. Even in cases inwhich there is likely to be a linear dependenceof response on added hormone-mimetic xeno-biotics, such as turtle sex determination, sto-chastic principles determine the distributionamong competing binding sites, i.e., nativereceptor, shell surface structures, nonreceptorbinding sites, etc., at low doses. Thus, not alladded hormone or xenobiotic would be avail-able for receptor binding. In addition,because many pharmacokinetic processes arenonlinear, it would be simplistic to assumethat all EACs act additively.

Recommendations* Make MBDR model development a

routine part of the risk assessmentprocess. The strongest recommendationis that appropriate MBDR models bedeveloped in concert with accumulationof data on mechanistic end points andwith data from new screening and toxic-ity testing protocols. Improved hazardidentification methods alone will, inisolation, do little to assess low-dose risk

situations. The use ofMBDR models willprovide perspective and context to hazardidentification studies and improve thequantitative significance of these studiesin contemporary risk assessment. Otherreports from the workshop appearing indifferent chapters in this monograph aidin identifying specific topical areas forpursuing compound-specific MBDRmodels for toxic responses.Organize and present contemporary exam-ples. There are examples of mechanisticmodels that have been incorporated intorisk assessment applications (Table 1).Fairly complete mechanistic models fordioxin include pharmacokinetics, geneinduction, cell proliferative responses,and tumor formation (34-36,51). Oneversion of the dioxin mechanistic modelhas been described in the dose-responsechapter of the dioxin reassessment (34).A second model structure with similarpharmacokinetic and gene inductionmodules was based on different assump-tions of the characteristics of cells at riskfor transformation in treated rats (52). Ona more limited basis, pharmacokineticmodules have been used in risk cal-culations with several halogenated hydro-carbons. Linkage models have beendeveloped and proposed for risk assess-ment use for cytotoxicity with chloroform(30) and acrylic acid (53). A model hasbeen described that explains the effects ofserum binding on estrogen potency duringdevelopment (26). A more comprehensivebiologically based dose-response modelfor developmental effects has recently beencompleted using methylmercury as theprototypical compound (39).

Mechanistic information has beenused in semiquantitative fashion forresponses such as thyroid carcinogenicity(37). Although less effort has beenfocused on mechanistic models ofendocrine system function, a number offirst-generation models have been pub-lished in recent years (4,27-29).Examples should be compiled and madeavailable in document form to indicatethe manner in which these models areconstructed and their potential applica-tions. The availability of such a docu-ment would aid in explanation of theprocess of mechanistic modeling in riskassessments and provide an opportunityto learn from past efforts (Table 2).

* Select prototype compounds/mixturesfor model development. For a limitednumber of case studies, mechanisticmodel development should be pursuedto collect appropriate data for developingeach of the modular components inthe exposure, dose, linkage, response

Environmental Health Perspectives * Vol 107, Supplement 4 * August 1999636

DOSE-RESPONSE MODEUNG OF ENDOCRINE-ACTIVE COMPOUNDS

Table 2. Characteristics for selecting prototype com-pounds for model development.

* Compound represents a human health concern* Metabolism and the potency of metabolites are

understood* Target tissue dose can be measured for

compound/metabolites* Intermediate markers of response are available for

study* Results are expected to be generalizable to other

compounds* latency from exposure to response is relatively short* A high penetrance rate for the effect can be achieved* There is a relevant animal model to study* Can make predictions of dose response for risk and

precursor states.* Predictions verifiable from intermediate responses

and biomarkers.

cascade. Criteria for selecting these pro-totype chemicals and responses must becarefully considered. Among these crite-ria are the broader applicability of indi-vidual components, e.g., modules thatprovide for the synthesis and eliminationof estrogen, progesterone, or testos-terone. The influence of in utero estra-diol exposure of males on adult prostaticfunction should be an important casestudy for examining the potential non-monotonic dose-response curves forEACs. Another potential case study forassessing threshold behavior and the roleof nonlinear positive feedback loops indose-response relationships is tempera-ture-dependent sex determination insome reptiles (43).

Several other candidate responses/ com-pounds were noted: estrogen exposure inrelation to mammary tissue neoplasia inrats (54,55) and dopaminergic prolactine-mia and amenorrhea. Other candidateexamples should also be identified basedon the deliberations of the other workgroups. These prototype examples shouldbe sketched out to indicate the connec-tions of the modular portions to the morecomplete model and data needs identifiedbased on the available data for each andthe ease with which the data can be usedin a quantitative fashion. The prototypesshould include a compound(s) with morethan one mode of action and exampleswith mixtures of compounds with differ-ent modes of action affecting a commonend point. A suggestion was provided ofthe effects of a mixture of dioxin, dibutyl-phthalate, and an androgen antagonist onmale reproduction. These compoundsappear to have very different modes ofaction expressed as the same kind of func-tional deficit. Alternatively, prototypemixtures present in the environment couldbe used (56).

* Foster intimate interdisciplinary commu-nication. The ability to develop and utilizethese quantitative models requires newteam building between individuals withtraining in such fields as toxicology,endocrinology, pharmacokinetics, statis-tics, and biomathematics. Appreciation forthe risk assessment process needs to bebroadly conveyed to groups collecting crit-ical mechanistic data. In return, theMBDR modelers need to be immersed inthe biological tools that provide the infor-mation for successfully coordinating themodular components.

* Promote education on multiple fronts.Three education-based activities need tomove ahead in tandem. They are develop-ment of case studies to demonstrate theprocess, education of scientific communityand regulatory bodies with regard to theapplication of these models, and introduc-tion of quantitative mechanistic modelingprograms as a more routine part of curric-ula in toxicology and risk assessment. Thislatter recommendation is a long-term activ-ity but important for ensuring an environ-ment that encourages the introduction ofbiologic data and mechanistic modeling inquantitative dose-response assessments.

* Encourage development and use of mod-ular components of mechanistic models.To fully benefit from the use of MBDRmodeling in science and risk assessment,there must be a forum for the review andacceptance of mechanistic models andtheir component parts. Critical, open peerreview will not only guarantee better mod-els for risk assessment but will encouragethe use of components of these models inthe development of models for numerousagents. The use of common modules formultiple compounds, in turn, will simplifythe review task, improve the quality of riskassessments, and encourage the furtherdevelopment of mechanistic models.Development of these models and moduleswas a priority area for research identified inthe Endocrine Disruptor Work Group ofthe Committee on Environmental andNatural Resources (57).

* Develop funding resources to supportMBDR model development. If efforts tocreate MBDR models as dose-responseassessment tools for EACs are to succeed,modeling activities will have to be moreexplicitly encouraged as an essential partof the present initiative to expand thetoxicity databases for these compounds.This support should come in the form offunds specifically earmarked for modeldevelopment, training, and education.With any attempt to provide theseresources, it must be recognized that theskills to collect important data and

develop model structures for hypothesistesting and risk assessment may wellreside in different locales. Attention mustbe given to encouraging multidisciplinaryactivities within single institutions andmulti-institutional activities for successfulcompletion of these types of modelinginitiatives.

Summary and ConclusionsMBDR models are promising tools forimproving risk assessments with EACs. Thetechnology and biology required to developthese models is advancing rapidly, providingmany opportunities for model-building effortswith diverse EACs. In the absence of theseMBDR models, the abundance of hazardidentification data being collected will serve toindict specific EACs without providing insightregarding the exposure conditions likely topose any significant level of risk in exposedpopulations. The successful development ofthese models requires close cooperationbetween the modelers and the laboratories andindividuals collecting biologic data. In theearly stages of model building, emphasisshould be placed on prototypical compoundsand on endogenous hormones themselves.Accurate dose-response models of endoge-nous hormones and the processes controlledand organized by these hormones are a neces-sary prerequisite for understanding the pertur-bations associated with EAC exposures. Thepotential ofMBDR models will only be ful-filled if resources are made available for modelbuilding, along with the resources for toxicitytesting and mechanistic research. A series ofrecommendations for expediting the develop-ment and application of MBDR modelsfocuses on education, fiscal and data resources,and creation of a library of examples of themodel-development process.

REFERENCES AND NOTES

1. U.S. EPA. The Federal Insecticide, Fungicide and RodenticideAct (FIFRA) and Federal Food, Drug and Cosmetic Act (FFDCA)As Amended by the Food Quality Protection Act (FQPA) ofAugust 3, 1996. 730L97001. Washington, DC:Office ofPesticide Programs, 1997.

2. 104th Congress. Safe Drinking Water Act Amendments of1996, Public Law 104-182.

3. U.S. EPA. EDSTAC 2nd Final Draft Report. Endocrine DisruptorScreening and Testing Advisory Committee. Washington, DC:U.S. Environmental Protection Agency, 1998.

4. Barton HA, Andersen ME. A model for pharmacokinetics andphysiological feedback among hormones of the testicular-pitu-itary axis in adult male rats: a framework for evaluating effectsof endocrine active compounds. Toxicol Sci 45:174-187 (1998).

5. U.S. EPA. Proposed Guidelines for Carcinogen RiskAssessment. EPA 600-P-92-003C. Washington, DC:Office ofResearch and Development, 1996.

6. U.S. EPA. Methods for Derivation of Inhalation ReferenceConcentrations and Application of Inhalation Dosimetry.EPA/600/8-90/066F. Washington, DC:Office of Research andDevelopment, 1994.

7. McGrath LF, Georgopoulos P, Gallo MA. Application of a biologi-cally-based RFD estimation method to tetrachlorodibenzo-p-dioxin (TCDD) mediated immune suppression and enzymeinduction. Risk Anal 16:549-548 (1996).

Environmental Health Perspectives * Vol 107, Supplement 4 * August 1999 637

ANDERSEN ET AL.

8. Allen B, Kavlock B, Kimmel C, Faustman E. Dose-responseassessment for developmental toxicity. II: Comparison of genericbenchmark dose estimates with no observed adverse effect lev-els. Fundam AppI Toxicol 23:487-495 (1994).

9. Allen BC, Kavlock RJ, Kimmel CA, Faustman EM. Dose-responseassessment for developmental toxicity. Ill: Statistical models.Fundam AppI Toxicol 23:496-509 (1994).

10. Kavlock RJ, Allen BC, Faustman EM, Kimmel CA. Dose-responseassessments for developmental toxicity. IV: Benchmark doses forfetal weight changes. Fundam AppI Toxicol 26:211-222 (1995).

11. Shuey DL, Lau C, Logsdon TR, Zucker RM, Elstein KH, NarotskyMG, Setzer RW, Kavlock RJ, Rogers JM. Biologically based dose-response modeling in developmental toxicology: biochemical andcellular sequelae of 5-fluorouracil exposure in the developing rat.Toxicol AppI Pharmacol 126:129-144 (1994).

12. Shuey DL, Buckalew AR, Wilke TS, Rogers JM, Abbott BD. Earlyevents following matemal exposure to 5-fluorouracil lead to dys-morphology in cultured embryonic tissues. Teratology 50:379-386(1994).

13. Leroux BG, Leisenring WM, Moolgavkar SH, Faustman EM. ABiologically-based dose-response model for developmental toxi-cology. Risk Anal 16:449-458(1996).

14. Gaido KW, Leonard LS, Lovell S, Gould JC, Babal D, Portier CJ,McDonnell DP. Evaluation of chemicals with endocrine modulatingactivity in a yeast-based steroid hormone receptor gene transcrip-tion assay. Toxicol AppI Pharmacol 143: 205-212 (1997).

15. Limbird LE, Taylor P. Endocrine disruptors signal the need forreceptor models and mechanisms to inform policy. Cell 93:157-163 (1998).

16. Yates FE. Good manners in good modeling: mathematical modelsand computer simulations of physiological systems. Am J Physiol234:R159-R160 (1978).

17. Lucier GW, Portier CJ, Gallo MA. Receptor mechanisms and dose-response models for the effects of dioxin. Environ Health Perspect101:36-44 (1993).

18. Andersen ME, Clewell HJ Jr, Frederick CB. Applying simulationmodeling to problems in toxicology and risk assessment-a shortperspective. Toxicol AppI Pharmacol 133:181-187(1995).

19. Portier CJ, Kaplan NL. Variability of safe dose estimates whenusing complicated models of the carcinogenic process. A casestudy: methylene chloride. Fundam AppI Toxicol 13:533-544(1989).

20. Clewell HJ Ill, Andersen ME. Use of physiologically based phar-macokinetic modeling to investigate individual versus populationrisk. Toxicology 111:315-329 (1996).

21. Parham FM, Kohn MC, Matthews HB, DeRosa C, Portier CJ. Usingstructural information to create physiologically based pharmacoki-netic models for all polychlorinated biphenyls. Toxicol AppIPharmacol 144:340-347 (1997).

22. Parham FM, Portier CJ. Using structural information to createphysiologically based pharmacokinetic models for all polychlori-nated biphenyls. II: Rates of metabolism. Toxicol AppI Pharmacol151:110-116(1998).

23. Tong W, Perkins R, Strelitz R, Collantes ER, Keenan S, Welsh WJ,Branham WS, Sheehan DM. Quantitative structure-activity rela-tionships (QSARs) for estrogen binding to the estrogen receptor:predictions across species. Environ Health Perspect105:1116-1 124 (1997).

24. Tong W, Perkins R, Xing L, Welsh WJ, Sheehan DM. QSARModels for binding of estrogenic compounds to estrogen receptora and ,B subtypes. Endocrinology 138:4022-4025 (1997).

25. Vanden Heuvel JP, Clark GC, Kohn MC, Tritscher AM, GreenleeWF, Lucier GW, Bell DA. Dioxin-responsive genes: examination

of dose-response relationships using quantitative reversetranscriptase-polymerase chain reaction. Cancer Res 54:62-68(1994).

26. Sheehan D, Branham W. Dissociation of estrogen-induceduterine growth and ornithine decarboxylase activity in thepostnatal rat. Teratog Carcinog Mutagen 7:411-422 (1987).

27. Andersen ME, Clewell JH, Ill, Gearhart J, Allen BC. A pharmaco-dynamic model of the rat estrus cycle in relation to endocrinedisruptors. J Toxicol Environ Health 51:189-209 (1997).

28. Kohn MC, Sewall CH, Lucier GW, Portier CJ. A mechanisticmodel of effects of dioxin on thyroid hormones in the rat.Toxicol AppI Pharmacol 165:29-48 (1996).

29. Blumenthal GM, Kohn MC, Portier CJ. A mathematical model ofproduction, distribution, and metabolism of melatonin in mam-malian systems. Toxicol AppI Pharmacol 147: 83-92 (1997).

30. ILSI-HESI. An Evaluation of EPA's Proposed Guidelines forCarcinogen Risk Assessment Using Chloroform andDichloroacetate as Case Studies: Report of an Expert Panel.Washington, DC:lnternational Life Sciences Institute, Healthand Environmental Sciences Institute, 1997.

31. Conolly RB, Andersen ME. An approach to mechanism-basedcancer risk assessment for formaldehyde. Environ HealthPerspect 101:169-176 (1993).

32. Andersen ME, Clewell HJ, Ill, Gargas ML, Smith FA, Reitz RH.Physiologically based pharmacokinetics and the risk assess-ment process for methylene chloride. Toxicol AppI Pharmacol87:185-205 (1987).

33. Andersen ME, Clewell HJ Ill, Barton HA. A Pharmacodynamicmodel of atrazine effects on estrous cycle characteristics inthe Sprague-Dawley rat. In: Triazine Herbicides: RiskAssessment (Ballantine LG, McFarland JE, Hackett DS, eds).Washington, DC:American Chemical Society, 1998;432-447.

34. Kohn MC and Portier CJ. Effects of the mechanism of -receptor-mediated gene expression on the shape of the dose-response curve. Risk Anal 13:565-572 (1993).

35. Andersen ME, Mills JJ, Gargas ML, Kedderis L, Birnbaum LS,Neubert D, Greenlee WF. Modeling receptor-mediatedprocesses with dioxin: implications for pharmacokinetics andrisk assessment. J Risk Anal 13:25-36 (1993).

36. Andersen ME, Birnbaum LS, Barton HA, Eklund C. RegionalHepatic CYPlA1 and CYP1A2 induction with 2,3,7,8-tetra-chlorodibenzo-p-dioxin evaluated with a multi-compartmentgeometric model of hepatic zonation. Toxicol AppI Pharmacol144:145-155 (1997).

37. U.S. EPA. Assessment of Thyroid Follicular Cell Tumors.EPA/630/R-97-002. Washington, DC:U.S. EnvironmentalProtection Agency, 1998.

38. Cook JC, Foster PMD, Hardisty JF, Klinefelter GR, Sharpe RM.A review of Leydig cell hyperplasia in rodents and their rele-vance to man. Crit Rev Toxicol 29:169-261 (1999).

39. Faustman EM, Lawandowski TA, Ponce RA, Bartell SM.Biologically based dose-response models for developmentaltoxicants: lessons from methylmercury. Inhal Toxicol11:101-114(1999).

40. Kavlock RJ, Setzer RW. The road to embryologically baseddose-response models. Environ Health Perspect 104: 107-121(1996).

41. Daston GP, Gooch JW, Breslin WJ, Shuey DL, Nikiforov Al, FicoTA, Gorsuch JW. Environmental estrogens and reproductivehealth: a discussion of the human and environmental data.Reprod Toxicol 11:465-481 (1997).

42. vom Saal FS, Timms GG, Montano MM, Palanza P, Thayer KA,Nagel SC, Dhar MD, Ganjam VK, Parmigiani S, Welshons WV.

Prostate enlargement in mice due to fetal exposure to low dosesof estradiol and diethylstilbestrol and opposite effects at highdoses. Proc Nati Acad Sci USA 94:2056-2061 (1997).

43. Crews D, Bergeron JM, Bull JJ, Flores D, Tousignant A, SkipperJK, Wibbels T. Temperature-dependent sex determination inreptiles: proximate mechanisms, ultimate outcomes and practi-cal applications. Dev Genet 15:297-312 (1994).

44. Crews D, Cantu AR, Bergeron JM. Temperature and non-aromatizable androgens: a common pathway in male sexdetermination in a turtle with temperature-dependent sexdetermination? J Endocrinol 149:457-463 (1996).

45. McLachlan JA, Newbold RR, Shah HC, Hogan MD, Dixon RL.Reduced fertility in female mice exposed transplacentally todiethylstilbestrol (DES). Fertil Steril 38:364-371 (1982).

46. Gray LE, Wolf C, Mann P, Ostby JS. In utero exposure to lowdoses of 2,3,7,8-tetrachlorodibenzo-p-dioxin alters reproduc-tive development of female Long Evans hooded rat offspring.Toxicol AppI Pharmacol 146:237-244 (1997).

47. Andersen ME, Barton HA. Biological regulation of receptor-hormone complex concentrations in relation to dose responseassessments for endocrine active compounds. Toxicol Sci (inpress). 48:38-50 (1999).

48. Shapiro DJ, Barton MC, McKearin DM, Chang T-C, Lew D, BlumeJ, Nielsen DA, Gould L. Estrogen regulation of gene transcriptionand mRNA stability. Recent Prog Horm Res 45: 29-64 (1989).

49. Tata JR, Baker BS, Machuca I, Rabelo EML, Yamauchi K.Autoinduction of nuclear receptor genes and its significance. JSteroid Biochem Mol Biol 46:105-119 (1993).

50. Ferrell JE Jr, Machleder EM. The biochemical basis of an all-or-none cell fate switch in xenopus oocytes. Science 280: 895-853(1998).

51. Kohn MC, Lucier GW, Clark GC, Sewall C, Tritscher A, PortierCJ. A mechanistic model of the effects of dioxin on geneexpression in the rat liver. Toxicol AppI Pharmacol 120:138-154 (1993).

52. Conolly RB, Andersen ME. Hepatic foci in rats after diethyl-nitrosamine initiation and 2,3,7,8-tetrachlorodibenzo-p-dioxin promotion: evaluation of a quantitative two-cellmodel and CYP 1A1AZ as a dosimeter. Toxicol AppIPharmacol 146:281-293 (1997).

53. Frederick CB, Bush ML, Lomax LG, Black KA, Finch L, KimbellJS, Morgan KT, Subramaniam RP, Morris JB, Ultman JS.Application of a hybrid computational fluid dynamics andphysiologically based inhalation model for interspeciesdosimetry extrapolation of acidic vapors in the upper air-ways. Toxicol AppI Pharmacol 152:211-231 (1998).

54. Tsai TL, Katzenellenbogen BS. Antagonism of developmentand growth of 7,12-dimethylbenz(a)anthracene-induced ratmammary tumors by the antiestrogen U 23,469 and effectson estrogen and progesterone receptors. Cancer Res 37:1537-1543 (1977).

55. Russo IH, Russo J. Mammary gland neoplasia in long-termrodent studies. Environ Health Perspect 104:938-967 (1996).

56. Constan AA, Yang RSH, Baker DC, Benjamin SA. A unique pat-tern of hepatocellular proliferation in F344 rats followinglong-term, low-level exposure to a chemical mixture ofgroundwater contaminants. Carcinogenesis 16:303-310(1995).

57. Reiter LW, DeRosa C, Kavlock RJ, Lucier G, Mac MJ, Melillo J,Melnick Rl, Sinks T, Walton BT. The U.S. federal framework forresearch on endocrine disruptors and an analysis of researchprograms supported during fiscal year 1996. Environ HealthPerspect 106:105-113 (1998).

638 Environmental Health Perspectives * Vol 107, Supplement 4 * August 1999