Document 3QQN4gx9rX2qV8L3nVgrr8p16

-) nlllrriiCI!llfFi !il'cmftm>I~MM IUJsltilliUJ\tm 1220 L Street, Northwest Washington, D.C. 20005 ~ .<,~... July 3, 1996 Mary Paxto~ !fi!D@[J[Jj)@ lro: Benzene Basic Research Work Group (BBRWG) I Benzene Task Force (BTFl conference Call on JulY 11 The BBRWG needs to address the items in the proposal present by Tony Cox (attached with memos summarizing earlier results) and to define the next steps that Rich Irons should take in validating '. our BBRA model for hematopoiesis with the surrogate agent, cyclophosphamide. This will also provide an opportunity to discuss our thoughts after having read Ben Thomas's revision of the naenzene Mechanism White Paper." A conference call at 2 ~-00 EDT next Thursday, July 11, is proposed '.. i~ t for these purposes. Other interested BTF members are welcome to ~' 't;'l participate. Please return the form below to 202-6B2-8270 indicating whethe-r you plan to participat.e. ***** (full name) (company) will will not (A better time would be: participate in the !S'!'F conl'E~li::'SZAca e~ll at 2:-00 :!!:li)'il' on 'i'hux-s&y 6 Ju~y 11, ~996. (Dial 703-736-7274 at that time to join the call.) -. BP-00017907 -> 21~ Sab 0314 D. SYROYMER IPa!lJe ~3 1"@: Dr. Mary P~on oIF~rom: ony Co" jgj<e: Proposed BBRA Work for 'i 996-'i 997 DaR: 6-27-1996 As we discussed, I am pleased to submit thisproposed set of tasks for the Committ~ to consider, based on our recent meeting in Denver. As always, I look forward to working with you to refine and shape the scope of work to make it as valuable as possible to the API. ' ,, . For several years. the API has suppot12d developm9nt of biologically based risk assessment (BBAA) models that offer the potential to provide more realistic health risk assessments and more informative uncertainty analyses than current regulatory risk models provide, especially at low doses. This work has resulted in insights and conclusions that. if they are accepted by the regulatory risk assessment community, have strong implications for how chemicals such as benzene, 1,3-butadiene, isoprene, and styrene are re.gulated. Some of the .key predictions and findings from the BBRA modeling and researcb_include the following 1. Hematotoxic responses at low concsmrafions ar(J significantly over-predicted from responses at h;gher r:oncentro.tions using the "equivalent" dose metrics, based on area-under curve (AUC) of administered dose. adopted in regulatory risk assessments. Specifically, low concentration, long duration (measured in months or years) exposures are far less dangerous than would be predicted by AUC e>etrapolations from algebraically "equivalent" higher concentration, briefer dura'iion e)tposures. Also, exposure e2pisodes that occupy only a fraction of the day (e.g., less than two ho~;~rs) have much- smaller hematotoJtic effects than would be e>epected based on "equivalent doses-that occupy a larger fraction of the day (e.g., 24-hour exposures). 2. When internal doses of oonzene metabolites are used as a basis for risk prediction, in preference to admonistered doses, then standard regulatory risk assessment models applied ro animal data by the California EPA yield a best- fitting dose-response relafion at low doses that is cubic, rather than linear. As a consequence, predicted low-dose risks b.ecome vanishingly small at low doses. Projected human health risks at low concentrations (e.g., below 1 ppm), using ee-standard regulatory e~rapolation procedures, can five or more orders of magniiude smaller than current California EPA risk estimates based on administered-doses. This finding is robust to plausible changes in the specific metric of internal doses (e.g., total hepatic metabolites, marrow metabolites, AUC o1 HQ or MA jo bone marrow, and so forth). BP-00017908 ...\ rw/. [ I .. ; ,. \o'' -> 3. Scientific and data uncertainties about the PBPK model structure andparameters (and about the iden~iiies of the specific metabolites involved in carcinogenesis), when properly managed, have lmle or no impact on ability fo predict the relative cumulative internal doses mc'ived per unit time at low doses from diff<m:mt dosing regimens. Indeed, the PBPK and hemato~oxiciiy components of the BBRA model ar0 only loosely coupled. From the standpoint of the hematotoxic response, only highly aggregate features of exposurn (e.g., number of hours per day and approximate magnitude of exposures) are important. Detailed descriptions of exposure ~ransients and model parametr values are not needed to predict hematotoxic effects. 4. Anlmal tumor data for various chemicals, including isoprene and benzene, show s~rong nonlinearities over a wide range of doses, suggesting that highconaenfration e)(perimental tumor data are not useful for pr~dicting tumor responses at low~r concentrations. For isoprene, tumor responses exhibit the sam~S types of dosetime-msponse "anomalies"- compared to AUCobased predictions (e.g., more weeks of exposure do not proportionally increase tumor risks. higher exposure conceniraiions can be disproportionately more hazardous, eic.) as hemaioto:xic-responses. Th~se empirical data rnay suggest thai a IBBRA-type approach, rather than AUC-based e)drapolation, is needed to obtain realistic understanding of !ow-dose risks. 5. Preliminary statistical analysis of animal data using "model-free" methods suggest that both hematoio)(ic and genotoxic dose-response relations caused by benzene exposures may have effective concentration thre~holds such that no positive .dose-response relation exists unless the threshold concentration is e:xceeded. This appears to be generally consistent with epidemiological data for benzene cancer risks. 6. The PBPK and hematoto:xiciiy components of the BBRA model indicate that human epidemiological and industrial hygiene data need to be soiled and analyzed based on factors such as fraction of the day exposed and number of years exposed, rather than being analyzed based only on AUC estimates -of cumulaiwe exposure. APt's BBRA project has permitted these and other points to be developed and presen~ed in a number of forums to both sci~mtific and risk analysis I regulatory audiences. Some of papers' and publications growing out of this API-funded work over the past si-'t years include the following: Cox, LA_, Jr., "Assessing cancer risks: From statistical to biological models, J. Energy Engineering, 'il"iil&i, 3, 189-2...1-0, 1990. "Extending biolog-ically-based cancer risk modeling to apply to benzene-inducedleukemogenesis~-- in B.J. Garrick and W.C. Gelder (eds), The Analysis, Communication, and Perception of Ri~k. Plenum Press, New York;-'i991a. BP-00017909 -> Zl& SB& 831~ D. STROI~BR i .1 . ~ '. .Biological basis of carcinogenesis: Insights from benzene. n Risk Analysis, 1 'll, 3, 453-4-M, 1991b. Reassessing benzene cancer risks using internal doses.H Risk Analysis. 12. ~. 401-410, 1992a. "Emending the stochastic two-stage model of carcinogenesis to include self-regulation of the non-malignant cell population," Risk Analysis. 12. 1, 129-136, 1992b. A biologically-based risk asssssment (BBRA) model of leukemog~nesis induced by cyclophosphamide", Poster presemed (by Dr. Mary Paxton) at the workshop on Biological Mechanisms and Quantitative Risk Assessment: From Experimental Design to Risk Characterization. Research Triangle Park, North Carolina, November 1-4, 1993&. "Coping with uncertainties in a computer simulation model of cyclophosphamide ! j ~ induced lel:l'fcemogenesis." invited talk, 1993 Annual Meeting o~ the Society for Risk Analysis, Savannah, Georgia, December 5-8, 1993b. "Machine learning for uncertainty management in complex risk models". inviied talk presented at The-ll'lstitute of Management Sciencss (TIMS) and Operations ',i ': Research Society of Amarica (ORSA) joint meeting, Boston Marriott Copley Place, Boston, Massachusetts, April24-27, 1994a. MNew frontiers_ in to:ldcol9gy information: Technologies of the information superhighway.n Keynote address, 21st Annual Toxicology Information Roundfable, IViJCROMEDEX To;<icolo.gy, Medicine, and Environmental Series, W"'stin Hotel at Tabor Center, Denver, Colorado, October 13, 1994b. "Nonlinear dose-time-response relations for chemical leukemog~ms: Computer simulation of the roles of cell kinetics and hematotoxicity." Invited talk pmsented at the Symposium on Recent Developments _in Benzene Epidemiology. and Toxicology. Villa Florence Hotel. San Francisco, CA. November ~-30, 1994<:. "Nonlinear cell kinetics can e>lplain observed anomalies in dose-time-response pattems; Poster session (presented with Dr. M.B. Paxton), 1994 Annual Meeting of the Society for Risk Analysis, Hyatt Reg~ncy at the Inner Harbor, Baltimore. I , MD, December 4-7,_ 1994d. r ! "Simple relations between administer2d and internal doses in compartmental flow ! ' models," RiskAnalysis,15,g,197-204, 19~5a. "An exact analysis of the multistage mod~! explaining dose-response concavicy." Risk Analysis, 'il~. ,a, 359-368, 1995b. "Reassessing benzene risks using in~emal doses". paper presented at---een&fJn~ '95, Rutgers, New Jersey, June. 1995. -Forthcoming in Environmental Health Perspectives. 1995c 3 BP-00017910 -> More accurate estimates of dose-response functions using Monte-Carlo uncertainty analysis: The Data Cube approach. Human and ecological Risk Assessment, 2. 1. 146~170, 1996a. Schnatter, A.A., iV'r.G. Bird, L.A. Cox, Jr., and R.F. Herrick:, "Defining optimal ~ .. ,;...... exposure assessment methods and metrics for epidemiologic studies exposures of petroleum distribution workers to benzene." Occupational Hygiene. 155-160, 1996b. in summary, the work completed to date on this project, including a substantial amount of as-yet unpublished work on BBRA model validation and comparisons to clinical data, appears to be promising in the following respects: Reasonable approaches for overcoming scisntific and data limitations and uncertainti~s have been developed that appear to be effective in making BBRA modeling practical with e)Cisti!}g data. Experimental and clinical data appear to. be consistent with model predictions I,,.. The model predicts that low-dose risks (whether of tumors in animals or of myeloto)Cic effects in animals and humans) are significantly over-estimated by current regulatory risk assessment procedures. Thus. BBRA modeling appears to have something to offer beyond what simple AUC-based extrapolations can give. Despite these accomplishments, our- BBRA work must close some important gaps to become as influential as it potentially can be. The most important next steps, .in order for the work to have its greatest possible impactwith scientists :md decision makers, wer-s- discussed briefly at our Denver meeting. They probably nnclude the following: on1. fSJMIBrdl CY~tr1/i~iUofty tllrrirrfl ~tr:(f!}pfla!f!i1tt!l9J fotr flffqqp /JjJ~If!~ fflt'Mi~U flffu g;C;o@f!iJflof!(; tt;ommMl!ilff[o/. Show the rasul~s and power of the BBRA approach to more of the scientific community, and achieve wide acceptaRce for its predictions. At present, ihe risk-analytic predictions and implications of the BBRA approB~ch are vulnerable to dismissal on the grounds that they are ou~puts of a complicated, unfamiliar model. 11 non-controversial but novel predictions from the model {e.g. its analysis of CP hematotoxicity results) are more widely understood and accepted, ~en the risk-rela~eol predictions---from the model are more likely to be accepted as well. ;:-1 2. lflfl~olfii(/(Q)rr~(!) fliJiJil!J ?@01w~lfii@ @r! fll!iJ m@cd/@U ~@f! fioiJIJIIi!il~f/P, @Bia~BJ. One of the outstanding potential- contributions of the BBAA model of CP hematotoxicity is that it ma~kes prsdictlons for humans that appear to match clinical data qui~e well 4 BP-00017911 -) Zl~ 58~ 931~ D. Sfi~t~ER and that can be extended-to low doses. This capability should be reinforced with more thorough examination of human data. perhaps using clinical results for otheli" chemotherapeutic agents in addition to CP. The goal is to demonstrate that ths BBRA approach is widely applicable to hematoto:xic (and perhaps leukemogenic chemicals) and that it makes trustworthy predictions for humans I..'. over a wide range of data. This will position it as a more credible basis for ~r: predicting benzene effec~s in humans. 3. !l@Y!fJ!@tp "e/!'D~ Cyi/g~WtriJ@fli$ SJrrvQJ ~(J;j~gmtirt; G@11i!'iJ[p8W~IfDft ~f! tt!JF~ 881{]&, m@~l The gn~atest current vulnerability of the BBRA model may be that it leaves unaddressed-the causal link that regulators care most about: the genoto)(ie (and r cytogenetic) component of health effects. This raises the possibility that all of ' our model approaches and predictions will be accepted as valid, in~eresting scientific work, but yet it will not be used to influence or replace current risk assessments, because it will be seen as too incomplete to be useful. E)(p!oratory work in 1995 and early 1996 suggest that a BBRA simulation descripftion -of tl'le dynamics of cytogenetic and genotoxic damage:! and dosetime-response is highly likely to reveal interesting and imporla.nt low-dose nonlinsarities, much as the hematoto)(icity component did. Without this component in place, the usual"d&fault" regulatory position of assuming low-dose linear genotoxic damage and an AUC basis for extrapolating risk is likely to continua. The BBRA framework expresses tumor ha2:ard as a product of two factors -- namely, the number of cells at risk (output from the hematotoxicity BBRA componen~) and the rate at which they are affected by carcinogenic metabolites {output from the genoto)(icity BBRA component). Addressing only the first component leaves open the objection that. at sufficiently low doses, the behavior of this product, and hence the age-specific and e)(posure-specific tumor hazard rate, is determined entirely or almost entirely by the second component. The BBRA model must be e)(tended to cover this gap. Ths scope of work-proposed in the foJiowing sections undertakes to fill all three of these potential gaps, positioning our BBRA work to have sirong practical impact In summary, the scientific and research caliber of the BBRA work to date appeal's to be strong. To have practical impact with decision makers and scientists in 1997 and beyond, the-work should ideally be positioned as a viable, general, well-validated approach to predicting relevant biological eff~s in humans-of several chemicals, including those of interest to regulators. To this end, I propose to e)(tend the application of the BBRA hemaio1o)(icit'y'- model to additional human da~a and to develop the genoto:xicity component of the BBRA model so that it can analyze existing dose-time-response data on cytogenstie effects in humans and animals, and so that it can make quantitative- predictions for these effects at low doses. 5 BP-00017912 ..'.t:- 1- -> i. \:" To make our BBRA model dearly relevant and trustworthy as a basis for quantitatively predicting health effec~s in humans -- including health effects though Yo -be causally relevant for chemical leukemogenesis ~- I propose to undertake the following four tasks. TASK 1: EXTEND THE HEMATOPOIESIS COMPONENT OF THE BBRA MODEL WITH ADDITIONAL HUMAN DATA AND CHEMOTHERAPEUTIC AGEN-TS r; Description: This task will reposition om BBRA hematotoxicity model, so far developed and validated for CP, as a model of hematoto>ticity (in humans and other species), rather than as -a- model of CP. Accordingly, published clinical data not only on CP, but i also on other hematoto>tic, immunosuppressive, and possibly leukemogenic chemoftherapeutic agents, will be culled from the merature, and compared to the BBRA model's predictions. I will continue to rely on Rich Irons and on the Committee for ., ~ :. ' recommendations and approval of the-speeific chemicals to be studied. I suggest that we plan to analyze no fewer than five known hematotoxic chemicals (including CP) to provJde an adequate basis for assessing and validating the _g~neral descriptive and predictive usefulness of the BBRA model. I aniicipab~ that we can select at least one of these chemicals within the first few weeks of the contract, and agree to all of then within the first iwo months. Deliverable: A technical report that .(i) Documents multiple hmnan data sets on hematotoxic effects (doseDtime-response curves for multiple periods and dosing- regimens) o'i' multiple chemicals, and (ii) Shows- the comparison between empirical ebservations and BBRA model predictions, including the results of any data-analyses performed on data obtained from F!ich's Raw model validation experim0nt for CP. This report wiU provide a basis for validating and assessing the perfo-rmance ot the BBRA hematotoxicity model as a general approach to myelotoxicity prediction. It will be backed up by (ii-i) Computer simulation software and models that can be used to run new dosing scenarios with different chemicals. These models will take chemical-specific cytotoxicity parameter values estimated from the literature, together with user-specified dosing scenarios. as inputs. They will predict effec~s on the~ time courses of different hematopoietic ceiL populations as outputs-,- 6 BP-00017913 -> Z16 566 8311 ID. SIHDI~ER '- Discussion: Only parsimoniOtJS__ changes in the CP model parameters to~anow for ~' .. differ~nt cytotoxic potencies of differ~nt chemicals, will be considered. If the BBRA model can parsimoniously predict I explain human hematoto)(icity data for a wide range of myelotoxic chemicals, it will be well positioned for future use as a source of relevant predictions for benzene and other chemicals. Estimated Resource Requirements: $26k TASK 2: DEVELOP GENOTOX~CITY COMPONENT OF THE BBRA MODEL AND ANALYZE EXISIING CYTOGENETICS I GENOTOXICITY DATA Descript;on: This task will create and validate an initial dynamic simulation model of the geno~oxic and cytogenetic effects of-the chemicals (including CP) identified in Task 1. The model will describe I predict the time courses of markers such as MN-PCE, DNA adducts, and perhaps SCE in hematopoietic populations during and following different dose regimens. ~~will also compare these time courses ~o published data. Deliv~rables: (i) A 'computer simulation- model, probably implemented in either JiH!NK or PowerSim. that implements the genoto~icity I cytogenicity component of the B~RA model. (ii) A technical report describing the model's structure, the data used to estimate its parameters, the formulas and parameter values in the model (with their biological interpretations. where applicable}, and the results of initial model validation tests based on comparisons to published da~a sets for several chemicals and genotoxic endpoints. Discussion: This effort will recapitulate for the genotoxic link the methods and modeling approach akeady developed for the hematotoxicity link. An innovative--aspect of this task will be appiDcation of appropriate methods for estimating_ and validating cytogen-etic and genotoxic potencies from published time course dats. In particular, the question of whether a low-dose nonlinear dynamic model fully explains aH --of the available data, or whether a significantly better description could be achieved by postulating a (perhaps small) linear component, will be addressed using accepted statistical methods for model selection and comparison. Such me~b.ods, ranging from ~he classical Akaike Information Criterion to modem computational methods for model selection and- hypothesis testing, have already been identified in an earfier phase of this project as part of the methodo1ogy for dealing with uncertainties in BBRA modefing. They are now ready for use in this phase. Estima~ed resource Requirem~:mis: $201< 7 BP-00017914 ,. pi i ,, l'...~ \ : .. 'lr: -> TASK 3: DOCUMENT ALL FINDINGS TO DATE IN AN A-PI TECHNICAL REPORT Description: All of the important findings and resuns achieved in the project to date should be consolidated in a technical report This includes as-yet unpublished results on the roJJowing areas, discussed at our Denver meeting: {i) Preliminary results of applying the BBRA hematotoxicity mode to ben2:ene (namely, that experimental findings and apparent anomaries for CFU-GM population time courses under various benzene dosing regimens published in previous decades are well explained I described I predicted by the BBRA model). (ii) Resulis of our model validation using CP {namely. that the experiments conducted by Rich Irons' Jab produce data that are very consistent with the BBRA model predictions) (iii) Results from-preliminary analysis of genoioxicity and cytogenetics data (namely, that model-free statistical analysis suggests threshold-like effects for any positive doseg tim~gresponse relaiion). After Tasks 1 and 2 have been completed and the results of Rich Irons' -new experimen~ have been analyzed. so that the preliminary findings now available can be extended and tested in greater depth, ii will be important to document and archive them in a form that other researchers can use. (It is also prudent project management practice to summarize all of our accomplishments to date in case the project is canceled, key personnel leave, etc.) Deliverable: A report, suitable for publication within the API technical report series. focused solely on documenting all of the Bl8Aill project's technical find~ngs and results to date (including results of model validation experiments, comparisons to da~a in the literature,. summary of the data sets used. etc.), but with minimal or no interpretation or discussion of possible implications for risk assessment. This report could furnish tr-le material for the "Methods" and n-Resufts" sections of scientific journal ~rticles that API may subsequently elect to publish based on this project. Discussion: The last API technical report produced for the BBRA modeling project was written in 1993. While numerous conference posters and presentations have reported on our progress since then, and while there has been a lot of exciting progress to report, this information is not available to others in any convenient form. Since .it- is premature to publish our interim findings on the res1:1fts of model validation and anaryses-of previousry pll..!blished da~a sets. an API tecbr:lical report appears too be-a very appropriate vehicle for documenting the pro_gress we have made since 1993. Estimated-Resource Requirements: $15k 8 BP-00017915 I .. .. ~ .. -~ -> TASK 4: DEVELOP AND APPLY STATISTICAL METHODS FOR QATA ANALYSIS-OF BBRA MODELS WITH A NON-STATIONARY CONTROL GROUP Description: This ~ask wm ~ake care of some technical issues that require attention in ligh~ of the experim~ntal results to date. In particular, we must resolve the questions of how best to (a) Estimate relative time courses (prediciedvs. actual); (b) F=ormally test hypotheses about them; and (c) Report model validation results. when the control population displays substantial interindividual variability and unexplained systema~ic variation over time. This task will develop and apply appropriate statistical methods for handling these issues, which were not anticipated in advance of the experimental data collection. Deliverables: Analyses of new experimental data from Rich Irons' lab using the statistical methods identified and implemented in this ~ask. Discussion: Promising approaches to the sia~isiical issues raised here include (i) Model roouction, in which the complex BBRA model is simplified and rsexpressed in terms of a small number of quantities ihat support clear tests of hypotheses; and (ii) Charac~eri:zation of the nonstationarities in the control group (e.g., determine whether they can be represented as the result of passing random disturbances through the nonlinear dynamic hematopoiesis system, or estimated and controlled for using time series models such as the ARIMA or ARMAX specificationsJ. Cleaning up these statistical issues will also enable us to more nearly optimize the design of BBRA parameter estimation. hypothesis testing, and model validation experiments Estimated Resource Requirements: $12k Based on my previous experience with this type of modeling and analysis, I believe that the entire scope of work proposed here can be completed within 6-8 months, depending on API's needs, for a cost of approximately $73k. As always, I look forward to working with you to re1inl9 the proposed scope ot work to make it as useful-as possible to the API and-to reflect your needs and priorities. 9 ) '- BP-00017916 57/SJ/% 12:55!2ll. VIa Fax -> lPage llllll.2 T@: Dr. Mary Pamon, API . IFirIJ'iJil: Tony Co):(, Cox Associates ~: Resul~s o~ statist-ical analysis of CP experiments IQ)tal~: 6-2-1996 This msmorandum summaruz0s ~hs r0sa.n~s of our 1W5 IBBRA Pro]ec~ ~or ac!dillio111al ~riment&D vs~udation of our CP BBAA model of hematotoJ~;oc~. Themajor rsllllts of this project were p~resooied ai ilhe lBena::ene Workshop in Napa . '.. -~ VaD~ey in February and ~~ our Denver meeting on Jl.!lne. ~ )I In bril!llf, ~11m eJtperiment was designed in 1995 to ru~her test our Cl? royhemm~otoJt:ocijy sim!J!ation mod! having it prsdict the time co!.!lrse o1 OFU~GM popuBalllions over a seven-day altpsrimsntal p0riod foijlowing administra~ion o1 a1 sin~!e olooo of CP. IPrsdociions based on the BBRA simulaUion modlsl w~rs developed in August and Septsmoor of ~ 9915. Unceliain~ies about ~lhs doss ooa!s B!Jild aooui ihs opern!itiolr'lla~~ dlefi111ition o~ ~he CFU-GM popu~&ition we11e treattsdl ll.llsing s~msitivity ~Aaly$eS, i.e., time ooursas war predicteo11or differali"i~ s~.nbpopulaioons o~ ithe mod~~ and! ror diffsrsnt doss scc:Aie 1~ctors over abon.!l~ a oi"ie t)rQ'ier o1 magnitu~ mn{lJe. {Tilis range had a~maoly Ibsen idsn~ffiisd base~ ol1'! ~R't rSVJ!is of litemtum seali"Cihes andl iha piloU s:ltpsrimoot cond!.!c;1!ed Di'11 iths 1994-iS95 projeci.) Ksy prsdlictioros made and oJocumsntsd in iteclhnicStl memoranda in a.dlv&nce oi' th eltlQeriment includsd the followin9: 1. ThriJ CFU-Gil!l level should peak around day 4 for U~e aggregate CFU-GM - .:. popuiSlitioliil. {This contrasts wii"tl a1 peaet a8 around dlmy 8 in ~he peripheral b~oocl comp~nt.) Thus. if FIIJGM data are conec~ed on days 2, 4, a1nd 7, ttl! high 'Valus Sh9Ju~d occur on day ~- This was ihe design used in our e)(priment 2. tow-dose Md high-dose scenarios IPad to markedly diffrSrent hemafofo1tic ~~ .... mspon!!;e$; how~ver, veFJI di~r-en~ "high" doses lead io very simiiUJr ~me ceumes, and similarly for "low doses". A rslatively small (s.g., fute-1oldl) mcmals in~ mdlminu~Uamdl COD11Cntratuon can rmll~ the dfii'iersncs betweell'l a "How @ossa ll'sponse mat ha!s m relmtivs~y minor depression followed by a r~SD&1ovc&1y smaU jpS~ik. lll!"'Uol Q "hig~-dlosem response ft~at Glthibits a clear depli'es@.ion On marrow CFU-GI\Id, foDiowedl by a marked over-shoot in the refoound. This parim, no~ad as Sl iheorsltical possibili~ roassd on BBRA simulations in "i 994 and "i 995, is consistent wiili wb.&t we unorot~ntiona!ly endoo up showing in the Mo pilot ~rimen~ perfo.rmed on ~995- 1996. Namely, the 10 mgilkg olose showsd little iV any impa~ct ('ihs-~low-dlosan outcome), while the 50 mglkg (Pattern coniormoo to ~he preoliclted "high-dose outcome. ,_ ,} BP-00017917 r;t:\. t.I . -> \."' 8. The value of CFU-GM- on day 7 should be b~twerm ;rs values on days 2 and . 4, wiih ihs day 2 value !being lowest atnd the day 4 value baing highsst. 1"he BIBIRA modlel gives quantitative predic~ions, allowing the appro)t:ima~~e ratios of the CIFU-GM values on days 2, 4, 21nd i ~o bs predicted for difi'sn~nt CFU-GM subpopulatiorns and 1orr the a~ggragai CFU-GM popula~non as a whole. The e:ttperimen~ was carried oui in th0 1ouruh quarter o~ 1995, att a tdlos l0vel (1 0 mg I kg) \that was ~oo ~ow to give conclusive results. It was rep~too ai 1a1 ~iSJ~rr o'los level (50 mg/kg) in the forni qual"ier of 1995. This dase isvel "iiJmsdl Oll.!lt ~o be hnglh ~~:mough to crrea~re n..mambigL!Ious resul~s. The key fi!iidings were a.s : : ~oi!ows. frffNDING 1. The CFU-GM fime cours~ peaked t?Jf around day .tt, as prrdictd. 'This finding is illu~raterOl in Figure i . Th Jefq)anel of Figurs 1, wt'lich wes one of several doz~m graphs IUJSOO ~o thoroughly eJtamine ihs da\Ra, sO"'ows a clea~~ . \, peell< a~ -day 4 compared to o'Jays 2 andl 7. ~@lUJIR!~ 11 : Th ~!F1UJ...@!Mil 'iluffl e@ll.!Jli'$ ~@~$ fil~ l%li'@twll'il@J lQli!W ~ f .G.$ ,.--,---.----...-----, lt.I'----r-!III----.---~-- ; l. f--+----l ----+--1 t---- I --<:>- A_FleMU~ --a- ~-Rl!VJliR. QL-~----L----~~ DAY G_1:z: G...,2:4 t\....3:7 DAY G_t:z: G...;l:4 G_3:7 ~ C_~~I\IJUR _.,...._ I"'EM_MEAN GROUI" GROUP ,....:' CY Control (These are time coll.!~ses in which each data poinft is an avemge tor fiv~& mice, wi~h fiv~ replicates of CFUJeGM oownts p0r mouse. Av~r~ge CIFIIJeGI\II wluss ~&rn shown ~r frhrss otf ~he five replicat~ss, to ondicat that ~hens is roo ssgnificant- dliffrenoe-across re~cm~S. The same pSlrtem hoiTds when all five mplie&~e$ ar inciu<:lled, allihough ~e II'Slphs become a imis more chBteredl. These and other plots are included in ths lhBlnJdloui grom ou1r recsnt Denver mee~ing.) 2 BP-00017918 r '' I' .: , -> _A-surprising observat!on is ihai the con~ro! group (right panel) shows a simi!Blr p~~em. For both sets of animals (control and 50 mg!kg), the peak on d~y ~ is ' consis~sll"'l~ across replicates (&nomc:ds -and bone marrow samp~es) and ffs statisiical!y signincan~ at p < 0.05 ~msed of! a Kolmogorov-Smimov ~es~ for dlinerences in froo.~uency dislliibll.!ftiolils. Th0re is a~ signi~icann uns~IS\inoo door0~se in CFU-GM on day 7 compa~reell fto day 4. Thus, an llDne:~tpecloo flii1dlin~ (whiclh lheldl Bllso in ~he 10 m~g ~r~&tmernft al!1d oon~rol groups arndl ha~d forr WBCs in tl'le pifo~ e:~tperomsnt condiUJcftecil-in tlhe ~ 9941995 prroject) is ij"Jaft fh~r(!; is statisiically significanf variation over wme in ~ha confrol group hemafopoi9?:ic cell populations. This nonsmtionauri~ iiFi ~lle control group is no~ e){p~ainoo by fri'ne BBRA mooel, which predlicts ~hat Yhe con~rol grol.ijp mBlrrow popu!e~ions sll"liouldl remain appmx:imately coiT'lstant over time. I~ is ~emp~ong to s~eeYia~e ltha't ~he significan~ changes in con1ml grol!Jp blood! cell counis over ~ims may be c!ue ~o acelima~~aiion, handling dluring treatmen~. orr o~hsr e>q>eliimenlla~ Vaeiors; however, 1he tri!J~ causs are ct.mrenily un~nown. FINDING 2: There is subst&ng-iaf in8erindividual heterogeneiiy in mouse CFU-GM counts, ~orr ~l"iie sams dl~y a~rudl ~rea~imsnt group (50 mgll..g or contro~). ThilS> caJspCt of ~oo @a~a. shown in Figura 2, can po~enti8llly roe explained by ililte!indlividh..nBt~ vaiiabili~ in IBBAA mode! paramsasrs. J;;:U@QJI~~ ~: ruo@~ [Q)~ ~W&:I~ DllllftIJ"Dii'iiJfiwfiJl\!JU IHJ~I1'@@ll'il~ . \i- Plot of Maar! CFU-GM per FG!RltJr, by Mouse. Group and Day There is signific:ant inter-inc!Mdual het~ among mice in CFU-GIVI 2eS b -o- D- ~ 1.5!5 ! 1~ !:: 50!lllOO i' b y ~ I I''r-.: I f 0 -50tlOOO ! ! ~ .! i 4- ~ ~ +I i i i 1 I ! -f<p ' --<a> "'()- 6-<7 -o- i -o- ~ l <; -;!!,- ~ -o -o -o- -ri t:r .-.-..:,.-.. j <0- j -o- ~ t;I DAY G_1:2! G..J!;4 GflOUP ~3;7 DAY 0_1:2: G_.2:~ GROUP G_3:7 ~ -o- - a( ; ' ( Control -o- r "' ...(:,- 3 BP-00017919 -> F/NDJNG--3: The rela'live magnitudes of the CFU-Gfl.!l..coun~s on days 2, 4, and 7 '. _predicfoo by 'ihe BBRA model are 6lpproxima~11ly correcf, assruJming tlhat ttl ; aggregate;> CIFU-GM population in the BBAA modsl corresponds io what ths ' CFU-GM assay measu.nres. FigiUlre 3 displays this finding using the raw dat61, which is Rlhe way -~ 1s.vomble way to sll:hibit ~he BBRA modal's prrsdieltive osaccu~cy. Even so, ~lh fiR lbeRween model predictions andl obsllllrvsd va~ues coll'ilspnc!.!lon.ns, sspscn~~l' given that ~lhs predided vBllues at days 2, ~. arndl 7 were olbtaon~OJ li'lRire!y lbibr looking at the dcn'ta. Model Pr~ictions (top) vs. Actual {middle) &nd Control (txmom) for 50 mglltg Mean Ct"'U-GM VaiW$ 1.Se6 I .~ 1.Se8 1.%6 I 1.2ea ~~"..~ J 1sS ---o- GFIOUP CY ~ \. l -o- GROUP Control --o--- GROUP G_1:2: G_2:4 G_3:7 CP-!tJlodel CAY Whsn nomlaJ.li:iledl values a~re used -(i.1., when tlle raiio o1 trsatment grroup to control gro!.!p vsluss is used instsed of the mw daial), tllle fit oo~csn ~e obseNedl ano1 predidsd rrwan vahnes improv.ss 0vero further. in ~iighi of Rhe unexpla.ii"Dadl varnatoon in ~he contf!'o! group data over ijmre, anol especially nh~ ILmex~~ainecll decline in contror group CFIUJ-GM on dlmy 7 (Figure ~ ), it is reassuring ~~"tat ~lha ll'a'lios or ihs tm81tmen~ gronJJp CFU-GM 'to OOU'lltroi group . CFU-GM display 1lhs-same paWem (i.e., peak on dlmy 4} and approximats rs~a~ive ' magnirudles as ih IBBRA mods! predicts. Thi!As, standa~rdizing or "normaVizing" ~, : Rhe tn~mamen~ group da~ (!sfft panel oi Figure 1) ~o correct for rslatolm controi group VQ\h..!les {rig&l~ panel in Fi~urs 1) leaves ~he valid!~ of this prnd!Uc!tion uschaungedL Dndeed, ~s predic~ive accur6lcy o~ ~he IBBRA model improvss when t' 4 BP-00017920 (\. . ti7/~3/9& 12:51:~9 Via Fmx -> 21& 566 B314 D. SYRD~ER I. I the 50 mglkg mouse ola~~a are e:xpressed as ratios, relative to the control grot.np- I -VB!Iues, in that ihe day 7 normalized CFU-GM leva/ lias between the day 2 and day -It normalized values, as fhe BBRA model predicts. Since ililere is significan~ irnsrindlividl.8al heRerogsm~ity in responses (Figure 2), i~ is wollih considering not only ~s ratoo .of the means (treatment group mean vs. 1:' '!<. con~rol group me:m, ~or S!Ch clay) 21s a way o~ shnnolardi2ing ~he trea~meni dstm, but &iso ~o sltamine the dosinlb~oons of these ra~oos. When aach treatment mouse is compared to eSJch oon~ro~ mouse on each day, tlls values of 'itl'ls ratios Wfpic~!ly wang~ over a- ~actor of a~bout 2. However, ahe null hypothesis of no differen~e betMreen trea~tmsni and CClntrot groups can be rsjsc'ted using a :!):, r~oopar~msaric sign ~est on days 2 (when 19 out -of 25 ratios ars less ~han i) and 1 q.w. 4 (when 21 out o~ 25 ratios a~re greater than 1, implying that ~he ~ream-tent grou.np I indleso1 has an elevated CFU-GM count compared to the con~rol group. More sophisticated ways of ta1rud~ing ttls daRa (e.g., using individual mice as trea~mem lbloc~s. with repl!cstss of CFU-GI\Il msasn.JJremems within bloc~s) and ~he tims variation in control gmup counts may be developed- as -p<u~ of a subsequent e~ort ,'I Ths da~a collec~sd ano1 &nai~ed in ~his pro]ect generally suppoii aha conclusion thcd the B~AA model IProvidss useful prediciions of obseNed hsm&toto:xic ~ims courses ~or CFU-GM in the msrrow, as well as for WBCs (previously a~al~eol). There is no reason ~o rejec~ ~e BBRA hemato~o:xicity data ~sed on the resu~s of the validation pilot sltpsriments conducted to date. However, the fiinding of siaiis-ticai!y signil1cani nonsmiionmviiy il1l th0 control group CFU-GI\Il counts does sugg0st ~ha'i V~ure modal ve.lid&tion experim~mts sho!J!Idl bs- o1esign18ol md ona~na!yzsd taking in~o aocoun~ the likelihood of unexplained varia~oons control group diita. ! .. ~ ;,.. 5 '!. BP-00017921 ~11~31'9115 12:59:11 Yid F~~ -> 2. A potentialhr useful model of !benzene hematoto:Kicuty has been deve~oped by making parsimonious changes in our CP model. (Thus, the CP moolel mighi lbe better regarded as a model of taematotoxiciiy than as a modal o~ CP.) SIPecmcaD!y, qnotol!.ic potencies for benz0n0 can bQ sstuma~ed lbasoo -on prsviously published !itsra~u~re. Erythroid! c~otolt:icitty can be mooe~ed lby increasing Rile 1rat0 of witllidrawal ~rom ih0 CfiU-S popuil:llftion in & dosdepsnolent way, ~o mimic ~he competition of ft~e elr)lthmid linGage gor CFilJ-S cells. Firn!21Uy, a dose sca~e for mice can be rov..gghly es~ima~oo from previously publisl'le<:a ~im~ course data (Green et al., 19_81; Tofft ei al, 'i 982). These oochamges l'llave crea~sd a beli'lzens hematotol!.ici'ty modle!- ~lhat c&n used ~o make ~esaalble predictions for time courses of h.ema~opooetic popull~.tions_ (e.g., wrac ano1 CFU-GM) in response to different benzene eJ(posUJrss. 3. As lasciiJssed a'l our Oenvell' meeaing, ihe simple lbana:ene hsma~otoxociW modll lhlas been compared! io several previously p~lblished o'Jata oo~s. w~h e](coftong rssul~s. Several apparent "anomalies" on previously published -aa~ awe wail olsseribsd I explaitnsdl I pradlicteo1 by ~me bem::ene hsmaio~ol!.icity modsl. Spoouficany, there ars predictable patterns of departure from Sl simplistfic "equal AUCs crr~ats e~ual c~oio}tic dammge model, which seems-to 100 &mplicia in some prsvious li$k assassments for benzene. Temalbls moolel predictions -\tt-!at agree wnth prsvoously published dlam include ~ .. (i) A smaller curvuJiative dose o<i b~S>n<~:ene can h21ve a larger mysloto}:{ic en'oots. (For S):tQml!)le, 21 ppm o1 bsn~ene for 144 hours us fl)reo'lidoo'J to cause lsss CIFUo GM suppression ~ham 50 P!Pm ~orr 48 hours, in agrsemen~ with ~he rssiiJJ!ts o\1 Toft et al., 1982.) (ii) 10 ppm ~or 6 hoOJrslday, 5 days/week for 10-wooks h~s almost no sffect on CFU-GM, whereas 100 ppm 1or s hours/dlay, 5 days/was~ for 1 week has a marlc:ed 0Mect. Ofu~er com~S~riisons to pl.!lblished aaUt stlow simi~arly proml$ing II'~SUJ!is. AnhoJYgh ws lilav~ agrsdl iha~ i~ is prma1ure 1lo publ~sh th~S$ findoli"igs, -they nli1di!CCll~S ~a~i: nn~frla be>nc,:::sns BBM model mm~ roe usefl!JJI s]{piai!i11fi'llg pasi dlat~a. and IPQien'iia~Uy in pr~dlic~nrug hemaaotoroc sff@Cts from _otller s~posure scenarios (ior ~umans) or olosins rogimens (for mics). Moreover, such findings ma!Y svemuallv hlslp4o validstRe our IBfBlAA hemato~o:~tici1ty model-as a usefful, parsomoraioYs &IPB>roach fro J>redic'ting hematoto}tic '1ffecRs in hl!JJmans andl in mice, as w.ell as for convening does ooilwean ~ese species without relying on AUCs &s an "eOJUDllalen~m dose meiric. This may i:lelp io improve aspects of ris~ assessmen~ pracaics 'ihat arre I'SBevan~ Vor ~nzene and o~r chemicals. .I I 2 TOTAL P.1B BP-00017922