Tobacco Documents Online

FUNDAMENTAL AND APPLIED TOXICOLOGY 37, 16--36 (1997) ARTICLE NO. FA972305 WORKSHOP OVERVIEW Implementation of EPA Revised Cancer Assessment Guidelines: Incorporation of Mechanistic and Pharmacokinetic Data N. P. Page,* D. V. singh,~" W. "Farland,? J. I. Goodman,~: R. B. Conolly,§ M. E. Andersen,II H. J. Clewell,I C. B. Frederick,** H. Yamasaki,tt and G. Lucier~ *ToxaChemica. International P.O. Box 10547. Gaithersburg. Ma~'land 20849: *US EPA. NCEA-DC. Washington. DC 20460: .~Michigan State Universit.'. East Lansing. Michigan 48824: §CilT. Research Triangle Park, North Carolina 27709: tlCF/Kaiser Engineers. Research Triangle Park. North Carolina 27709: llCF/Kaiser Engineers. Ruston. Louisiana 71270: **Robin & Haas Co.. Spring House. Penno.lvania 19477: **International Agency for Research on Cancer. Lyon. Cedex 08. France: +,~National Institute for Environmental Health Sciences. Research Triangle Park. North Carolina 27709 Received January 27. 1997 Implementation of EPA Revised Cancer Assessment Guidelines: Incorporation of Mechanistic and Pharmacokinetic Data. Page, N. P., Singh, D. V., Farland, W., Goodman, J. I., Conoily, R. B., Andersen, M.E., Cl~well, FL J., Frederick, C. B., Yamasaki, H., and Lucier, G. (1996). Fundam. Appl. Toxicol. 37, 16-36. A workshop entitled "Implementation of EPA Revised Cancer Assessment Guidelines: Incorporation of Mechanistic and Phar- macokinetic Data" was held in Anaheim, California, in 1996 at the 35th Annual Meeting of the Society of Toxicology (SOT). This workshop was jointly sponsored by the Carcinogenesis, Risk As- sessment, and Veterinary Specialty Sections of the SOT. The thrust of the workshop was to discuss the scientific basis for the revisions to the EPA Guidelines for cancer assessment and EPA's plans for their implementation. This is the first revision to the original EPA guidelines which have been in use by EPA since 1986. The princi- pal revisions are intended to provide a framework for an increased ability to incorporate biological data into the risk assessment pro- cess. Two cases were presented, for chloroform and trichioreethyl- ene,.that demonstrated the use of the revised guidelines for specific cancer risk assessments. Using these new guidelines, nonlinear margin of exposure analyses were proposed for these chemicals instead of the linearized multistage model previously used by the EPA as the default method. The workshop participants generally applauded the planned revisions to the EPA.guidelines. For the most part, they considered that the revised guidelines represented a positive step which should allow for and encourage the use of biological information in the conduct of cancer risk assessments. Several participants cautioned however that the major problem gith cancer risk assessments would continue to be the inadequacy of available data on which to conduct more scientific risk assess- ments. ¢ I~? Socie~ of Toxkology. INTRODUCTION (N. P. Page and D.V. Singh) The revised EPA Guidelines for cancer assessment were released for public comment in the Federal Register on April 23. 1996, and simultaneously made available via the Internet on the EPA Office of Research and Development home page. This public release provided for a 120-day comment period. After that, the EPA Science Advisory Board was requested to again review the guidelines along with the public com- ments. The final step will be for the EPA staff to redraft as needed and then publish the guidelines in final form. It is expected that the process to finalize the guidelines will be completed in mid- 1997. These revised guidelines are intended to replace the EPA' s guidelines on carcinogenic risk assessment that were formu- lated in 1986 (U.S. EPA, 1986) and have been in use since that time by EPA as well as other national and international agencies. In the process of evaluating various issues that have been of considerable concern and debate within the scientific community, the EPA has undertake.n a rather ex- haustive review and revision process. This has included the sponsorship of several workshops to address specific guide- line issues (U.S. EPA, 1989a,b, 1994). The Society of Toxi- cology (SOT), at the 1994 annual meeting in Dallas, held a workshop on the "'Evolution of the Methodology for Quanti- tative Cancer Risk Assessments." That workshop specifi- cally addressed dose-response and quantitation methodol- ogy which was being considered for revision. The SOT membe~hip provided to the EPA administrators and scien- tific staff very useful ideas at that time. In addition to input from the SOT, the Society for Risk Analysis also held a 02"2-05q0/t~7 $25 t~) Cop.~nght ~? It,~7 h.~ the St~:~et) of Toxicolt~g) All tagh[~ of reprt~uctlon in any form reserved. 16

REVISED EPA CANCER ASSESSMENT GUIDELINES 17 workshop pertaining to the EPA cancer assessment guide- . lines at its 1992 annual meeting (Anderson et al., 1993). While the EPA has reached a significant milestone in its guidelines revision process, they have announced that the • guidelines will continue to evolve as the scientific knowledge of carcinogenesis increases. Indeed. it is anticipated that ad- ditional changes will evolve as some aspects continue to remain contentious. The revision to the EPA guidelines has set forth principles and procedures that the Agency will use to guide its future carcinogenicity risk assessment ap- proaches in the future. Indeed, EPA has already begun imple- mentation of these revi~ed-guidelines with procedures in place to institute the final guidelines when the time arrives. The following summaries of workshop presentations in- clude an overview of the main scientific issues of the guide- lines revision and a paper describing the types of information needed to utilize mode of action or mechanisms in the carci- nogenesis assessment process. These were followed by pre- sentations of two case studies (for chloroform and trichloro- ethylene) in which the revised cancer guidelines were used to conduct cancer risk assessments. The speakers also com- pared the results obtained with the revised approaches with results obtained with the previous default model (linearized multistage). A panel consisting of scientists representing in- ternational agency (IARC), industry (Rohm & Haas), and non-EPA U.S. agencies (National Institute for Environmen- tal Health Sciences) discussed issues of concern in imple- menting the revised guidelines. Finally, comments from the open discussion with other workshop participants are sum- marized. INCREASING THE USE OF SCIENTIFIC INFORMATION IN RISK ASSESSMENT: REVISIONS OF THE EPA CANCER ASSESSMENT GUIDELINES (William Farland) The revised EPA guidelines for cancer assessment were released in the Federal Register on April 23, 1996. and made available on the Interact via the EPA Office of Research and Development home page. This was followed by a 120-day comment period and review by the EPA Science Advisory Board is anticipated. Several workshops will be held to fur- ther evaluate specific aspects of the proposed guidance, pri- marily through the use of case studies. R is expected that the process to finalize the guidelines will take about 18 months from their release date. These guidelines set forth principles and procedures that the Agency will use to" guide its carcinogenicity risk assess- ment approaches in the future. EPA scientists will use new approaches which have been evolving over the last few years. This guidance also provides an opportunity to inform decision makers and the public about how EPA conducts its work. We know from past experience that there has been a wide usage of these guidelines by other Federal and State FIG. 1. Overview of revised EPA cancer assessment guidelines. Agencies. The final Agency guidelines are likely to benefit from revisions based on comments from those agencies as well as the public and industry. The proposal reflects the fact that new information has evolved on the earcinogenieity processes and the develop- ment of tools that can be incorporated in the conduct of risk assessments. The new guidelines attempt to take these developments into account. The guidelines should be flexible so that they can change as new information and develop- merits evolve. New information should be incorporated into the guidelines more easily than in the past. People have asked whether or not once we release these guidelines we will start the revision process again, as has happened in the past. We think that we have built flexibility into the guide- lines and that they are written in such a way that we can take into account emerging science. However. in the future we will publish a series of technical documents that will address significant issues which may bring about a revision of the overall guidelines. , How have the guidelines been revised and how will they increase our use of scientific information in risk assessment? First, we recognized the need for the risk assessment process to provide information useful for risk management decision making. As can be seen in Fig. 1, risk characterization is the product of the process of risk assessment keying into decision making, along with discussions of non-risk analyses and various control options that are available. Risk character- ization is obviously not the whole story for decision making in risk management, but it is an important part. But what is often not discussed is the process for identifying research needs through the risk assessment process, seeing that those needs are addressed in the laboratory, and assuring that the scientific data and research finds its way back into the risk assessment process. The iterative nature of the process is the important part in our ability to advance the state of the

18 PAGE ET AL. science of risk assessment and the update of our perspectives on what is known and not known about various agents. Major Revisions The revision direction of these cancer guidelines can be summarized in four major points. First, these guidelines em- phasize the importance of full characterization. Second, they expand the role of mechanistic information. This is a goal that is written in the guidelines explicitly, and we think that there is a real opportunity to take advantage of this type of data. The implication of the use of these data will fundamen- tally change the way we evaluate carcinogenesis, but we must also understand the uncertainties that come with the use of mechanistic information. Third, they suggest how to use all the information that is available todesign dose- response approaches. Finally, these new guidelines take the approach of a two-step, dose-response assessment. Full characterization examines hazard assessment, dose- response assessment and exposure assessment. It involves a knowledge as to what we know and what we don't know, what the assumptions and the uncertainties are, and how confident we are about the ability to extrapolate from animals to humans and from high dose to low dose in each step of the process. This full characterization allows one to get away from the linear process of going from hazard to dose re- sponse to exposure to characterization and to begin to realize that exposure has a bearing on how one considers hazard and hazard has the ability to influence dose-response assess- ments. These types of interrelational issues come out in this enhanced characterization approach. Expanding the role of mechanistic information in the haz- ard characterization step involves the use of mechanistic information to inform hazard assessment, particularly as re- lated to the relevance of animal data to humans. This is captured in the guidance to assure that this is part of all hazard characterization activities and pursue the question of how we look at conditions of expression of this hazard as part of the hazard characterization. What we do know about are route-specific effects or levels of exposure-specific ef- fects with regard to potential for hazard, not just risk, but hazard. These are somewhat controversial points that are specifically highlighted in the notice of the guidelines. In terms of dose-response characterization, we want to try to bring mechanistic information into dose-response assess- ment. We will use a decision logic for approaching dose response with mechanistic information contributing strongly to the output of that decision logic. As we move toward hazard characterization of the car- cinogenic endpoint, we are seeing evolution of scientific thought. We are seeing an evolution that has moved us from hazard identification, as black and white decisions, through traditional toxicologic testing, to hazard characterization through evaluation of mechanisms, modes of action, and biologically based models of cancer assessment. This is a fundamental change: it is a change that has been readily accepted by most of the scientific community to really focus on this idea of moving beyond just traditional approaches to toxicologic testing. An additional issue that one begins to address with hazard characterization is weight of evidence versus strength of the evidence. Again, the idea is to consider all studies that are available, both positive and negative studies, understanding the power of those studies to be positive or negative. We must understand what the studies tell us about the responses in animals or in humans. Using all of the evidence in hazard characterization goes way beyond the idea of just animal or human tumor responses. We are really examining key data for clues to inform our assessment. This information with regard to dose level, metabolism, and mechanism improve our insights on hazard and risk does in a very important way. In addition, this is a big departure from the way we used the 1986 guidelines. In 1986, we classified chemicals on the basis of animal or human tumor response but said we would then modify them if we found other types of information that would make us "'change our mind." Now our concept is to bring all of the information together with an integrated approach for discussing hazard. In terms of relevance of animal response, certainly some of the species- specific responses that have to do with the r,-2/a-globulin mechanism in male rat kidneys and issues related to over- coming homeostasis in thyroid carcinogenesis, and other such insights are particularly important to focus on. We can use some examples to illustrate this approach. We have limited experience thus far in dealing with mechanistic information risk assessment for several agents. This experi- ence has led to the discussion contained in these cancer guidelines and represents an evolution of the approach that we had been using. Dioxin is an ongoing activity that looks at receptor-mediated toxicity and we are clearly tryirfg to bring the huge literature on dioxin to bear on this question of both hazard and risk. D-Limonene is an example of a natural compound that produces cancer in animals by a male rat kidney response; however, these make rat kidney re- sponses are the only response seen in a 2-year bioassay. We think that we understand the mechanism of that response, and we feel that it is probably not relevant to human cancer response. That's not to say that all male rat kidney responses axe irrelevant, but, for those where we believe we understand the underlying mechanism of action, we ought to take that into account. Thyroid tumodgens and the question of T3-T4-thyroxin ratios and the overcoming of homeostasis clearly is an issue that needs to be dealt with. The Agency has developed a report that is going to be available soon. As yet another example, science has come a long way in understanding the carcinogenic mechanisms for methylene chloride. Certainly

REVISED EPA CANCER ASSESSMENT GUIDELINES 19 the metabolism and pharmacokinetics of methylene chloride are big points of discussion now. This information suggests that we probably should take another look at methylene chlo- ride and try to bring more of that information to bear on the risk assessment. The same thing goes for formaldehyde. Again. many of you have followed the story of the use of DNA-protein crosslinks as a dosimeter for formaldehyde and have looked at the issue of the new cell separation techniques in evaluating the risks tbr formaldehyde. These are all important issues to take into account as we improve our cancer risk assessments. In addition to these discussions that have gone on over the last few years, the guidelines have a discussion of several cases where we examine the available data on hypothetical cases. The compounds in these cases are not necessarily identified, but most of you know the data well enough to recognize that we used real cases or just slightly altered them to illicit our point. There are seven of those cases discussed in the guidelines to try to illustrate how these guidelines will be applied to a set of information to charac- terize hazard. In addition to that, there are five cases where the hazard characterization paragraphs or discussion that would be evoked by this type of database are actually illus- trated in example form in the back of the guidelines. These are meant to stimulate thought, for people to comment on, and to see how well they track with the guidance we are giving and how useful they are in terms of making some decisions. Now, as we move into dose response, the important ques- tion to ask is how do you use all the information to conduct dose-response assessment, not only the tumor data, but pharmacokinetics and metabolism data to begin to look care- fully at both dose and response issues and to deal with data on effects of the carcinogenic process or processes in select- ing an approach to dose response and to characterizing. The approach that we are going to use is a two-step dose- response assessment. The first step is to evaluate and model the data in the range of observation, to focus on the informa- tion that is available to us from the animal or human studies or perhaps both, and to do the modeling at that point only, within the range of observation. Then we will explicitly move into a discussion in the second step of how one evalu- ates dose response below that range, into the range of human exposure, basically what we call extrapolation. So, being explicit as we move from the observed range into the range of extrapolation is one of the features of these new guide- lines, one that we think is going to be very important. In this first step, evaluation of the range of observation, we use tumor data and other effects data to first choose a biologi- cally based model. If biological models are not available or feasible, is it possible to conduct curve-fitting using standard statistical approaches and models that will fit the data to the best of its ability? To explore the lower range, we use served data as our point of departure and identify some benchmarks for comparison with noncancer health effects assessment. There are some controversial issues associated with benchmark dose assessment and selection of points of depar- ture. But the point is to be able to bring the cancer and the noncancer guidance together so that one can look at both of these issues across individual chemicals. In the case of carcinogens, we want to identify the lowest reliable part of the dose-response curve as a point of departure for extrapo- lation. The second step attempts to make use of approaches that are indicated by the mechanism of action. We choose a biologically based model to extrapolate if at all possible and, hopefully, link it to the range of observation in a way that makes biological sense. If not, we find ourselves back to a default situation where ~ne needs to look at how one deals with dose response in the range of extrapolation in the ab- sence of data that helps to inform us of the likely shape to the point of confidence. Certainly, one would look to choose a linear approach if there are data that suggest that linear is the appropriate way to go or a nonlinear approach if that is the approach that the data suggest. Perhaps here one will look to the "'margin of exposure" approach, where one asks the question "how far is the ob- served range in a particular human situation from the ob- served range from animals or humans?" We look at this "'margin of exposure" as a way to get an idea of how far we exposed humans are from where we actually observe effects in animals. We characterize results and use alterna- tive approaches when it is appropriate. In some cases we conduct both a model approach and a margin of exposure assessment if that seems like the most appropriate thing to do. There is some guidance in the guidelines to suggest how one would do that. As we begin to use mode-of-action data to inform dose- response assessment, we hope to be able to construct a bio- logically based or case-specific model as we have been trying to do for dioxin, as we are evaluating right now for trichloro- ethylene and for other cases that you have heard about and will hear more about today. We will use dose response for other effects in lieu of tumor effects if they are judged to be a better measure of potential risk. For those of you who know the formaldehyde story or know some of the stories where you do have a good opportunity to look at induced cell proliferation, we can put more of an emphasis on cell proliferation as a modulator of carcinogenic activity to help to understand the shape of the dose-response curve. And, finally, to use this to conduct the assessment of risk in the range of extrapolation, to ask whether or not those 'would carry down below the range of observation. So, the decision logic that we are using is to suggest that one would use a linear approach to dose response for DNA-reactive i+ •

20 PAGE ETAL. or other evidence that supports a linearity of response. If chemicals are not DNA-reactive, but if there are insufficient data to characterize a nonlinear mode of action, then the default will be made to a linear approach. This is a public health protective approach, and that linear extrapolation that we will use is not going to be the typical linearized multistage default model as used in the past. But, in fact. it is going to be what we term "truth in advertising." which is to say that we will characterize the point of depar- ture at the lower end of the dose-response curve and extrap- olate simply down to zero. Basically, one can lay a ruler on the page and determine the slope of the dose-response curve if one assumes a linear response from that point of departure. If does not model the response, it does not suggest more precision than it ought to. and it gives us a conservative, what we think will be an upper bound, estimate of risk in those low-dose regions. Clearly, depending on how close you are to the range of observation, you may have more or less confidence in that estimate of risk and that should be part of the uncertainty discussion. In terms of nonlinear responses, if chemicals are not DNA- reactive or otherwise linear and sufficient data are there to characterize a nonlinear mode of action, we would certainly want to see that as a nonlinear dose-response curve. In some cases, we are going to suggest that perhaps both should be done. depending on whether there is differing activity at different sites. We heard just earlier that there may be three or four different modes of action called into play for chemi- cals like trichloroethylene. If there is complex activity need- ing both approaches to describe, certainly we would want to have both of those included in the assessment. In the case of the nonlinear dose-response curve, we think it is going to be quite unlikely that in the near future we are going to see biologically based dose-response curves for nonlinear carcinogenesis below the range of observation. And in that case, we,would argue that the default should be a "margin of exposure" approach. It should not be linear and it should not be a half-baked attempt to try to describe dose response in the face of great uncertainty. Instead, look at this from the standpoint of how close human exposures of concern are to the range of observation to gain some insights on risk. So, risk characterization is the bottom line. We are try. ing to integrate and summarize the hazard characterization, dose responses, and exposure assessment characterizations that have gone on to develop a public health risk estimate where that is possible. We will also develop a framework to define the significance of the risk in every assessment that we do and to present the assumptions and certainties and scientific judgments that went into that to provide alternative ap- proaches and their implications. We think that biologically based dose-response models and biologically based risk as- sessments are possible and that they will improve hazard characterizations by the use of biomarkers of response with some mechanistic linkage to endpoints of concern and that that's going to be a real advance in the way we do business. This will strengthen some of the inferences regarding the shape of the dose-response curves outside the range of ob- servation and this is a goal. This will also help to identify targets of opportunity for further study of potentially sensi- tive human populations and allow us to do a better job of collecting epidemiology data in the future. Now, this approach is not without some risk itself. Biolog- ically based risk assessment presents increased complexity, which complicates simple approaches to risk management, some of which you heard about in the discussions of trichio- roethylene. It complicates already difficult risk communica- tion about difficult scientific issues, but we think it is abso- lutely critical for scientific integrity in the risk assessment process and we hope to be able to further that aim with these new guidelines. USE OF MODE OF ACTION INFORMATION IN CANCER RISK ASSESSMENT (J. I. Goodman) The carcinogen bioassay is a qualitative test (Goodman, 1990). However, our purpose is not to simply identify chemi- cals that can be labeled as carcinogens. On the contrary, the overall goal is to provide a reasonable estimate of the possi- ble hazard that a chemical might pose to people under realis- tic conditions of exposure. The key issues center around dose selection, dose-response relationships, and species-to- species extrapolation (Counts and Goodman, 1995a: Bucher et aL. 1996; Counts et aL. 1996). Therefore, the bioassay should be approached more like a research project than sim- ply a test and a rational approach to risk assessment requires the use of biological information (Goodman, 1994). Dose influences mechanism and. over a wide range of doses, mechanism may be expected to change with dose. Thus, a carcinogenic effect observed at a high dose is not necessarily expected to occur at lower doses (Counts and Goodman, 1995a), especially when dealing with nongeno- toxic :chemicals (McClain. 1994). For example, a report on the rdationship between use of the MTD and study sensitiv- ity for detecting rodent carcmogemc~ty concluded "... [an] important limitation of our analysis is that the range of doses used in NCI/NTP long-term rodent studies is generally rather narrow, typically extending from ¼ MTD to MTD or from ½ MTD to MTD. Thus, it could be argued that carcinogenic effects that are present at even the "lowest' of these doses are due to the same cell killing and compensatory mitogen- esis effects that occur at the MTD" (Haseman and Lockhart, 1994). Increased cell proliferation may facilitate carcinogen- esis due to the fact that mitogenesis can facilitate mutagene- sis (Ames and Gold, 1990). However, this is a complex process and it is not surprising that a one-to-one relationship between cell proliferation and carcinogenesis is not apparent (Melnick. 1992). ~o

REVISED EPA CANCER ASSESSMENT GUIDELINES 21 A consideration of dose selection for the bioassay entails two primary questions: Ca) What is an appropriate range of doses, including the maximum tolerated dose (MTD), to use for chronic exposure'? (b) After the assay has been com- pleted, what are the appropriate doses to employ in order to estimate the possible effect(s) the agent in question may produce in people? The two key principles that underlie dose selection are (a) it is not correct to make an assumption a priori that these doses are the same: and (b) any high dose. no matter how high, that permits the test animals to survive long enough to develop cancer is not necessarily an appro- priam dose to employ for the purpose of a risk assessment (Goodman, 1995; Counts and Goodman, 1995a). Nongenotoxic compounds appear to produce cancer either in a species- and/or dose-specific fashion (Cohen, 1995). We have made progress in discerning a basis for this observation by focusing on the'~.ariety of roles that alterations in DNA methylation may play in carcinogenesis (Counts and Good- man, 1995b) and placing an emphasis on the role that hypo- methylation may play as an epigenetic, nongenotoxic mecha- nism involved in tumor promotion (Counts and Goodman, 1994; Counts and Goodman, 1995c). Tumorigenesis in mouse liver is employed as our model system (Vorce and Goodman, 1991; Ray et al., 1994; Counts et al., 1996). The results of our research indicate that the B6C3FI mouse is deficient with regard to is ability to maintain normal DNA methylation and this appears to, in part, underlie its uniquely high susceptibility towards development of liver tumors (Counts and Goodman, 1995c,d: Counts et al.. 1996). We believe that hypomethylation is relevant to tumorigenesis in both rodents and humans (Counts and Goodman, 1995c), and that humans may be less susceptible than rodents due, in part, to a better ability to maintain nascent patterns of DNA methylation (reviewed in Counts and Goodman, 1995c: Counts et al., 1996). It appears to be inappropriate to make human risk assess- ment decisions based upon a mouse liver tumor response. However, in those situations where the results of a bioassay indicate that the mouse liver is one of several sites where an increased tumor incidence occurs the mouse fiver tumor response, as well as the other target sites for tumorigenesis, should be evaluated with regard to the mode of action of the chemical in question. Safety assessment for those chemicals (especially nongenotoxic chemicals) acting through a thresh- old-exhibiting mode of action should be based on a safety factor, or multiplicity of exposure, approach (Counts et aL, 1996). In this regard, it should be noted that the high propen- sity of the B6C3F1 mouse to develop liver tumors indicates that there is a "built-in" safety factor when this end point is viewed in comparison to the human, even at equal doses of the chemical in question. The maintenance of nascent DNA methylation status should be regarded as a fundamental homeostatic mecha- nism. Accordingly, it is appropriate to invoke the secondary mechanism concept (Scheuplein. 1995) in this context. Al- tered DNA methylation can be viewed as a secondary mech- anism involved in carcinogenesis and assessment of this pa- rameter may provide insight leading to a more rational inter- pretation of animal studies for human risk assessment (Counts and Goodman, 1995c). Therefore, an examination of DNA methylation status should be considered for inclu- sion as an ancillary component (e.g., in addition to assess- ments of cell proliferation in vivo and in vitro tests for the potential to cause genotoxicity) of both subchronic studies. and the carcinogen bioassay (Counts and Goodman, 1994; Counts and Goodman, 1995c.d: Counts et aL, 1996). This can aid in discerning the mechanism of action (e.g., a possi- ble nongenotoxic, threshold exhibiting mechanism) of the chemical being evaluated. In addition, this information could facilitate a rational approach to dose setting and the selection of appropriate doses for risk assessment; e.g., if toxicity occurs only at doses above those which cause altered DNA methylation in the target organ(s), these data could aid in providing the basis for placing a proper emphasis on lower doses (Counts and Goodman. 1995c: Counts et aL, 1996). However, it would not be appropriate to consider measure- ment of alterations in DNA methylation as a short-term test for carcinogens (Counts and Goodman, 1994). In summary, emphasis should be placed upon research that may discern probable thresholds for the carcinogenic effect of carcinogenic agents, especially nongenotoxie chemicals (Counts and Goodman. 1994: Counts and Good- man. 1995a,c,d). This has to involve hypothesis-driven re- search and be based upon insight regarding the mode of action of the chemical of interest (McClain. !994). The practical significance here is that the proposed strategy can provide the basis for a safety factor, or multiplicity of expo- sure, approach to risk assessment for those chemicals for which a likely threshold can be demonstrated (Williams and Whysner, 1995). This is a rational approach to risk assessment. The Environmental Protection Agency's (EPA) new Pro- posed Guidelines for Carcinogen Risk Assessment (EPA, 1996) represent a major positive step in the correct direction. These proposed Guidelines permit, indeed encourage, the use of biological information to make rational risk assess- ment decisions. Specific classification groups A through E have been discarded in favor of placing emphasis on a narra- tive description of the carcinogenic potential of a chemical under particular conditions of exposure. The linear multi- stage model has been discarded in favor of permitting the use of linear and nonlinear dose-response models as justified by the experimental data available. Terms like mode of action, threshold, secondary mechanism, and genotoxic (and by in- ference nongenotoxic) are employed. Additionally. the pro- posed Guidelines recognize explicitly that a carcinogen may

22 PAGE ET AL. (Upper bound on risk) overwhich actual dsk X: Pradic~ed dsk at may exceed predicted ~ upper end of range risk even though dsk of uncertainty due assessment is to consewative dsk conservalive, assessment. (Lower bound on risk) FIG. 2. Range of uncertainty about predicted risk (x). True risk lies somewhere between the upper and lower bounds, which are not known with certainty. Since conservative assumptions are typically used. the pre- dicted risk is likely to be near the upper limit of the range of uncertainty. However, even with the use of conservative assumptions, the actual risk may exceed the predicted risk. may be overly stringent in some cases and insufficiently stringent in others are direct reflections of the uncertainty associated with risk assessment. If there were no uncertainty, i.e., if risk assessment were known to accurately estimate actual risk, such concerns would be moot. The need to im- prove risk assessment to minimize the possibilities of over- and underregulation can therefore be restated as a need to reduce uncertainty in risk assessment. In this report, I briefly describe a biologically based expo- sure response (BBER) model for the hepatotoxicity and he- patocarcinogenicity of chloroform and compare the uncer- tainty of dose-response assessment performed with this model with an assessment based on default approaches as described in the 1986 U.S. Environmental Protection Agency (U.S. EPA) Guidelines for Carcinogen Risk Assessment (U.S. EPA, 1986). The Proposed Guidelines for Carcinogen Risk Assessment (U.S. EPA, 1996) specifically encourage the use of biologically based models for cancer risk assess- ment. Thus the new chloroform model described herein also illustrates some aspects of how dose-response assessment motivated by the new Guidelines is likely to differ from older approaches. act through a nonmutagenic mode of action, including per- turbation of DNA methylation patterns which my contribute to carcinogenesis by alteri.ng gene expression. I applaud the EPA for proposing these new Proposed Guidelines for Car- cinogen Risk Assessment and I look forward to opportunities to work with the Agency in order to implement them. CASE STUDIES IN CANCER RISK ASSESSMENT: CHLOROFORM (Rory B. Conoily) Risk assessment influences regulations intended to protect the public health. These regulations incur economic and so- cial costs associated with remediation, product substitution, and other aspects of compliance. Because risk assessment is intentionally conservative in the face of scientific uncer- tainty, some of these costs are prolqgb~ly unnecessary, i.e., the conservatism of risk assessment leads to regulations more stringent than needed to protect the public health. Uncer- tain~y in risk assessment also means, at least in theory, that risk assessment intended to be conservative may not, in fact, be protective of the public health (Fig. 2). These two fac- torsmthe likelihood that unnecessary costs are incurred and the possibility that in some cases the public health is not protected--motivate research to improve the scientific basis of risk assessment. While both exposure assessment and dose-response assessment can be pursued using policy- based default options, reduction of uncertainty for these ac- tivities depends on improved scientific understanding of the relevant biology, chemistry, and exposure conditions. The concerns that regulations based on risk assessment Exposure-Dose Response What is commonly called dose-response modeling en- compasses both the pharmacokinetic and pharmacodynamic components of the overall linkage between exposure and response. The BBER model for chloroform consists of linked submodels for pharmacokinetics (PK submodel) and phar- macodynamics (PD submodel) and can readily provide quan- titative descriptions of the relationships between (1) expo- sure to chloroform in air, water, or any other medium and tissue dose (PK submodel) and (2) tissue dose and response (PD si~bmodel). This approach to partitioning the exposure- response relationship leads to an expansion of the standard definition of dose-response (NRC, 1983: U.S. EPA, I996) with exposure-dose-response in which "dose" refers to the dose to the target tissue. This expanded definition is used throughout this report and is in general useful when describ- ing BBER models. Description of the BBER Model for Chloroform The basic structure of the chloroform model has been previous.ly described (Conolly and Butterworth, 1995) and -is only briefly reviewed here. The PK submodel corresponds to the physiologically based pharmacokinetic model for chloroform described by Corley et al. (1990). This submodel allows for definition of an hepatic dose surrogate that is the interface to the PD submodel. The PD submodel links tissue dose with a tissue response, in this case the killing of hepato- cytes by chloroform. The PD submodel also describes regen- erative hepatocyte replication in response to chloroform-in- duced cytolethality.

REVISED EPA CANCER ASSESSMENT GUIDELINES 23 The exposure-response curve for chloroform-induced cy- tolethality and regenerative replication is taken to be a suit- able surrogate for the exposure resl~mSe curve for chloro- form induced hepat¢~.'ellular carcinoma for the reasons dis-. cussed in detail by Larson et aL (1994c) and Butterworth et aL (1995). Briefly. I1) there is a large body of evidence showing that chloroform is not a directly genotoxic carcino- gen in mammals; (2) chloroform is a potent cytotoxicant to human and rodent liver and kidney; and (3) in all cases studied to date. carcinogenic effects of chloroform are pre- ceded by cytolethality and regenerative replication. Thus the available data. which are substantial, suggest that the mode of carcinogenic action of chlorotbrm is cytolethality. Under this mode of action, the mutations that lead to chloroform- induced hepatocellular carcinoma are assumed to arise as consequences of chronic cytolethality and induced cellular regeneration, not as direct effects of chloroform or of its metabolites on hepatocyte DNA. The exposure response curve for cancer under this mode of action is expected to be sublinear (U.S. EPA, 1996). In fact. the chloroform expo- sure-cancer response curve probably has a true exposure threshold since it seems unlikely that a single molecule of chloroform is capable of killing a hepatocyte, thereby initiat- ing the chain of events that ultimately leads to formation of procarcinogenic mutations. Definition of the target tissue dose surrogate is a key feature-of any BBER model. The dose surrogate is the transi- tion point between the PK and PD submodels and can be thought of as the pharmacokinetic driving force for subse- quent PD events. Before the advent of PBPK modeling, this driving force was typically an exposure metric such as the concentration of chemical in inhaled air or in drinking water. PBPK modeling allows for definition of a tissue dose that incorporates nonlinearities in the relationship between.expo- sure and dose to tissue and which is consistent with the roles in toxicity of bioactivation and metabolite kinetics. Chloroform is metabolized to phosgene and hydrochloric acid (Pohl. 1979). Phosgene has a short half-life in aqueous systems. In the liver, phosgene can be conjugated with gluta- thione or form macromolecular adducts, but any phosgene that does not enter one of these two pathways is rapidly hydrolyzed. Chloroform-derived hydrochloric acid should similarly be short-lived since it is removed by intracellular buffering. It follows that the pharmacokinetic driving force for hepatocyte killing by chloroform is more closely, related to the rate of chloroform metabolism, i.e., to the rate of production of phosgene and hydrochloric acid than to, for example, the area under the curve for metabolite production. The latter definition would b,e appropriate for stable metabo- lites. Ideally. the dose surrogate is defined without reference to the biochemical mechanism of the tissue response, i.e.. the pharmacodynamic component of the exposure-response re- 0.3O 0.~$ tO.~O ~ 0.'15 ~0.10 0.05 0.00 I 81 - o.oo ~.oo a.oo 3.oo 4.oo s.oo e.oo Liver dose surrogate (~rnol metabol/zed/g Iiver/60 m/n) FIG. 3. Empirical function describing the relationship between target tissue dose surrogate, as defined in the pharmacokinetic (PK) submodel. and ceil killing inferred from the pharmacodynamic (PD) submodel using 3.5-day labeling index data. During a simulated chloroform exposure, the value of the dose surrogate is calculated at each time point, t. and the function spe~ifics the corresponding amount of cytolethality. See text for further details. lationship. Limitations in the available data. however, make it difficult to completely maintain this. distinction in devel- oping the BBER for chloroform. The dose surrogate .used for the current model probably represents not only a pharma- cokinetic driving force but also some of the kinetic aspects of the pharmacodynamic response. Because of this. the quan- titative definition of the dose surrogate is provided within the following description of the PD submodel. The PD submodel consists of difference equations that track time-dependent changes in the number of hepatocytes in the liver as a function of basal rates of cell division arid death, chloroform-iladuced cytolethality, and regenerative replications. Note that tracking of both basal and regenera- tive replications allows the labeling index to be simulated. The model is constructed so that for each cell killed due to chloroform there is a corresponding regenerative replication. This one-for-one correspondence allows the number of cells killed at any level of chloroform exposure to be calculated from labeling index data (Larson et aL, 1995a, b) and corre- lated with the dose surrogate (Fig. 3). The dose surrogate is the only adjustable parameter in this approach to defining the tissue dose-cytolethality-regenerative replication rela- tionship. The amount of chloroform metabolized over the interval (t, t - 60 min), where t is the current time, was found to provide a good simulation of the labeling index data. As noted above, this definition of tissue dose is proba- bly not a purely pharmacokinetic tissue dose surrogate but also embodies to some extent the kinetics of the tissue re- sponse that leads to cytolethality and regenerative replica-

tion. Given that the half-lives of chloroform-derived phos- gene and hydrochloric acid inside hepatoc~tes are expected to be brief, a purely pharmacokinetic dose surrogate would be expected to be defined for an interval considerably shorter than 60 rain. Integration of the rate of metabolism over 60 rain probably reflects kinetic aspects of the tissue response ending with cell death and regenerative replication. This response has not, to date, been characterized at the biochemi- cal level. Refinement of the dose surrogate definition re- quires data on tissue half-lives of chloroform-derived phos- gene and hydrochloric acid that are currently unavailable. lnterspecies Scaling Conolly and Butterworth (1995) described interspecies scaling of the chloroform model where different scaling strategies were used for the PK and PD submodels. The PK submodel was scaled from rodents to humans using a combination of allometric relationships and human data as described by Corley et al. (1990). Since PD scaling is not as well understood as PK scaling and since no data were available to guide scaling of the PD submodel, it was not changed when used to describe the human response to a liver dose of chloroform. The human version of the chloro- form BBER model thus uses a PK submodel scaled from that used for rodents but the same PD submodel used for the rodent models. This hybrid scaling strategy takes advan- tage of the different levels of sophistication in our under- standing of PK and PD for chloroform. Such a strategy is likely to be useful for other chemicals since PK is, in general, better understood than PD. Preliminar3." Simulation of the Human Hepatoxicity of Chloroform Wallace (I 950) described a case of hepatitis and nephrosis in a 50-year-old man who consumed about 12 ounces/day of a.cough syrup containing chloroform and 3% alcohol. After a prolonged period of cough syrup consumption at this rate, the individual experienced signs of both hepatic and renal toxicity. He had been consuming about 33.4 mg chloro- form/kg/day. The PK submodel was configured for this indi- vidual according to Corley et al. (1990). including the speci- fication of his body weight at 78 kg, as reported by Wallace (1950). It is worth noting that Corley et aL (1990) had used human liver microsomes to estimate the capacity of human hepatocytes to metabolize chloroform (V~, of 15.7 mg chlo- roform/kg body wt/hr, multiplied by BW°'7 to adjust for body weight differences either between or within species). With this configuration, a simulation of 7 days of cough syrup consumption showed no effect on labeling index, i.e., label- ing index stayed at its basal value (Fig. 4). However. increas- ing Vm~, by a factor of two to 31.4 mg/kg/hr resulted in an increase in simulated labeling index for the same exposure scenario. Under these conditions, the model predicted chlo- 1.0 0.9 1.0 FIG. 4. Simulation of hepati~ concentration of chloroform (solid line) and hepatic 3.5-day labeling index (dashed line) in a 78-kg man consuming a bolus dose of 33.4 mg chioroform/kg/day for 7 days. Labeling index is increased (upper dashed line) when the capacity to metabolize chloroform (V,,~) is set to 31.4 mg chioroform/kg BW/day. With V,~, at 15.7. the value obtained by Corley et al. (1990). no increase in labeling index over the basal level was seen (horizontal dashed line). roform hepatotoxicity. The change in V~,., by a factor of two that was required to simulate liver damage is not unreason- able given (1) the coexposure to alcohol in the cough syrup and (2) the possibility that prolonged exposure in this case may have caused some induction of the capacity to metabo- lize chloroform. This result, while preliminary, suggests that the model development process to date is valid. More work, however, is needed before the model can be used with confidence for quantitative prediction of the hu- man hepatotoxicity of chloroform. Particularly important is the specification of drinking behavior. For the simulation shown (Fig. 4), cough syrup consumption was modeled as a single bolus dose to the gastrointestinal tract, once a day, with a first-order absorption rate constant of 1.0 (1/hr). Pre- diction of hepatotoxicity is sensitive to changes in this de- scription. For example, splitting the single daily bolus dose into 12 equal doses taken 1 hr apart over a 12-hr period (probably a more realistic description of drinking behavior) decreases the value of the dose surrogate obtained at the dose rate of 33.4 mg/kg/day and eliminates the predicted hepatotoxic response. Wallace (1950) provides no informa- tion on the details of the drinking behavior associated with chloroform toxicity. Without this detail, description of the drinking behavior as a single bolus dose per day m. aximizes the value of the dose surrogate calculated by the PK submo- del and thereby provides a worst-case estimate of the contri- bution of chloroform absorption kinetics from the gastroin- testinal tract to the likelihood of toxicity. One of the advantages of mechanism-based modeling is its ability to identify specific sources of uncertainty such as the description of drinking behavior. In default approaches to chloroform risk assessment (see below), this factor could not be explicitly considered, and its associated uncertainty would be hidden. It is easy to infer that uncertainty hidden

REVISED EPA CANCER ASSESSMENT GUIDELINES TABLE 1 Comparative Uncertainty of Default and Mechanism-Based Dose-Response Modeling for Ch/oroform 25 Default model Mechanism-b~ed model Component 11980 Guidelines)~ (1996 Proposed Guidelines)~ Equivalent dose calculation C "< T with duration adjustment Target tissue dose from PBPK model based on physiology and mechanistic considerations. (1~ uncertainty) (~, uncertainty) BWu' PBPK scale-up based on data. allometric relationships, and physiology. (~ uncertainty) (~ uncertainty) n/a Single bolus vs sips Irate dependence) if uncenain~ O, uncen.aimy) LMS Cell replication response surrogate (1" uncertninty) (~ uncertainty) Drinking ~havior Cancer model a U.S. EPA (1986). b U.S. EPA (1996). in this way does not exist, leading to the mistaken impression that default models are less uncertain than mechanism-based models. Relative Uncertain~. of Default and Mechanism-Based Models for Chloroform The cu~.ent noncancer risk assessment for chloroform by the U.S. EPA is in the form of an oral reference dose(RfD) calculated from a study by Heywood et al. (1979). In this study, fatty cysts developed in beagle dogs given chloroform in a toothpaste base in gelatin capsules 6 days/week for 7.5 years. The lowest-observed-adverse-effect level (LOAEL) for the study was 12.9 mg/kg/day, and the RfD was set at 0.01 mg/kg/day (U.S. EPA, 1995). Uncertainty factors of 10 were applied to the LOAEL to account for interspecies conversion, protection of sensitive human subpopulations, and concern that the effect seen was a LOAEL and not a no-observed-adverse-effect level. The RfD was calculated without use of either pharmacokinetic or nontumor pharma- codynamic data specific to chloroform. In contrast, the mech- anism-based, exposure-response model described here makes maximum use of available data to construct the PK and PD submodels (Table 1). As a result, predictions of the exposure-regenerative cell replication relationship obtained with the mechanism-based model are less uncertain than the RfD calculation. A further advantage is the explicit identifi- cation of sources of model uncertainty obtained with the mechanism-based approach (Table 1). This contrasts with the hidden or implicit nature of much of the uncertainty in the current default models. The standing U.S. EPA cancer risk assessment for chloro- form (U.S. EPA, 1995) uses the LMS model with tumor data from NCI (1976) and Jorgensen et aL (1985). As with the noncancer risk assessment described above, the cancer risk assessment uses no pharmacokinetic or nontumor pharmaco- dynamic data. The lack of direct, mammalian genotoxicity of chloroform (Rosenthal, 1987; Larson et aL, 1994c) and the correlation between chloroform-induced cytolethality and regenerative proliferation and carcinogenicity (Eschenbren- her and Miller, 1945; Larson er aL. 1994a.b: Larson er aL, 1995a.b) are consistent with the hypothesis that cytolethality and regenerative proliferation are necessary precursors to chloroform-induced cancer. These factors suggest that predic- tions of cancer risk obtained with a mechanism-based model for chloroform cytolethality are less uncertain than risk as- sessments based on the LMS model (Table 1). Application of the mechanism-based model for chloro- form in formal risk assessment should include use of Monte Carlo analysis to approximate variability in the human popu- lation. Confidence in the model could also be increased by more extensive use of human data for its parameterization. Studies designed to support these improvements are under- way at the Chemical Industry Institute of Toxicology. When these studies are complete, the model will be used for predic- tion of human risk of hepatotoxicity and bepatocarcinogeni- city from chloroform exposure. Under the 1996 Proposed Guidelines for Carcinogen Risk Assessment (U.S. EPA, 1996) a margin of exposure analysis would be used, re- flecting the nonlinear mode of action'of chloroform as both a cytotoxicant and a carcinogen. CONSIDERING PHARMACOKINETIC AND MODE2OF- ACTION INFORMATION IN CANCER RISK ASSESSMENTS FOR ENVIRONMENTAL CONTAMINANTS: EXAMPLE WITH TRICHLOROETHYLENE (Melvin E. Andersen and Harvey J. Clewell) The principal challenge facing cancer risk assessors today is to realistically consider the implications of the chemical's