United States
 Environmental Protection
 Agency
Office of Research and
Development
Washington. DC 20460
Chapter 8.
Dose-Response
Relationships
EPA/600/AP-92/001 h
August 1992
Workshop Review Draft
                Review
                Draft
                (Do Not
                Cite or
                Quote)
                       Notice

 This document is a preliminary draft. It has not been formally released by EPA and should not
 at this stage be construed to represent Agency policy. It is being circulated for comment on
 its technical accuracy and policy implications.

-------
DRAFT                                                              EPA/600/AP-92/001h
DO NOT QUOTE OR CITE                                                    August 1992
                                                                  Workshop Review Draft
                Chapter 8.  Dose-Response Relationships
                                 Health Assessment for
                        2,3,7,8-TetrachIorodibenzo-p-dioxin (TCDD)
                                 and Related Compounds
                                        NOTICE

THIS DOCUMENT IS A PRELIMINARY DRAFT.  It has not been formally released by the U.S.
Environmental Protection Agency and should not at this stage be construed to represent Agency
policy. It is being circulated for comment on its technical accuracy and policy implications.
                                                    I I  Q  ~-^,' ' x       :->.,•   .
                                                    u.s. i-in/noistH^'iiii (.- -•^'•ion Asency
                                                    i."3.on b, Llbr:-:.-  '-
                                                                              •
                                                                             J -^lii Hoof
                        Office of Health and Environmental Assessment
                             Office of Research and Development
                            U.S. Environmental Protection Agency
                                     Washington, D.C.
                                      Recycled/Recyclable
                                      Printed on paper that contains
                                      at least 50% recycled liber

-------
                             DRAFT-DO NOT QUOTE OR CITE



                                       DISCLAIMER


       This document is a draft for review purposes only and does not constitute Agency policy.

Mention of trade names or commercial products does not constitute endorsement or recommendation

for use.
           Please note that this chapter is a preliminary draft and as such represents work
           in progress. The chapter is intended to be the basis for review and discussion at
           a peer-review  workshop. It will be revised subsequent to the workshop as
           suggestions and contributions from the scientific community are incorporated.
                                                                                      08/27/92

-------
                           DRAFT-DO NOT QUOTE OR CITE



                                      CONTENTS


Tables  	v

Figures	vii

List of Abbreviations	  viii

Authors and Contributors	  xiii

8.  DOSE-RESPONSE MODELING FOR 2,3/7,8-TCDD	  8-1

   8.1. INTRODUCTION	  8-1

       8.1.1.  Introduction to Modeling for TCDD  	  8-14
       8.1.2.  Dosimetric Modeling  	  8-23

   8.2. TOXIC EFFECTS	  8-40

       8.2.1.  Modeling Liver Tumor Response for TCDD	  8-40
       8.2.2.  Tumor Incidence  	  8-41
       8.2.3.  Other Effects Mammary/Uterine/Anticancer Endpoints	  8-62
       8.2.4.  NonCancer Endpoints (DeVito et al., 1992)	  8-64
       8.2.5.  Neurological and Behavioral Toxicity	  8-65
       8.2.6.  Teratological and Developmental	  8-66
       8.2.7.  Immunotoxicity	  8-69
       8.2.8.  Reproductive Toxicity	  8-70

   8.4. RELEVANCE OF ANIMAL DATA FOR ESTIMATING HUMAN RISKS  	  8-76

   8.5. HUMAN MODELS	  8-80

       8.5.1.  Introduction	  8-80
       8.5.2.  Modeling Toxic Effects in the Liver	  8-81
       8.5.3.  Lung Cancer and All Cancers Combined   	  8-83

   8.6. CONCLUSIONS	8-102

   8.7. KNOWLEDGE GAPS	8-103

   8.8. REFERENCES	8-107
                                           iii                                   08/27/92

-------
                       DRAFT-DO NOT QUOTE OR CITE
                           CONTENTS (continued)
APPENDIX A:  A MECHANISTIC MODEL OF EFFECTS OF DIOXIN ON GENE EXPRESSION
            IN THE RAT LIVER

APPENDIX B:  MODELLING RECEPTOR-MEDIATED PROCESSES WITH DIOXIN:
            IMPLICATIONS FOR PHARMACOKINETICS AND RISK ASSESSMENT
                                    iv                              08/27/92

-------
                            DRAFT-DO NOT QUOTE OR CITE
                                    LIST OF TABLES
8-1     Examples of Levels of Information Available for Estimating Parameters
       in Dose-Response Modeling  	  8-19

8-2     Design Considerations for Risk Assessment Purposes	  8-22

8-3     Administered Dose, Tumor Response and Number at Risk of Hepatocellular
       Neoplasms in Male Sprague-Dawley Rats From Carcinogenicity Experiments of
       Kociba et al. (1978), Using the Pathology Review of Sauer   	  8-42

8-4     Fitting the Two-Stage (TS) Model to the Tumor Incidence Data of
       Kociba et al. (1976) Using the Pathology of Sauer et al. (1991)	  8-48

8-5     Net Parameter Estimates From the Two-Stage Model of Carcinogenesis	  8-51

8-6     Results from Two-Stage Model for Hepatocarcinogenesis in Rats   	  8-54

8-7     Using Liver Foci Parameters to Fit Tumor Incidence 	  8-59

8-8     A Comparison of Dose-Surrogates With Parameters Estimates from the Two-Stage
       Model of Carcinogenesis	  8-60

8-9     Toxic Endpoints Data	  8-74

8-10   Similiarities Between Laboratory Animals and Humans in Biological
       Effects of TCDD  	  8-77

8-11   Rat and Human Comparison of Daily TCDD Intakes and Body and Liver
       Concentration for Equitoxic Response	  8-84

8-12   Measured serum TCDD Levels and Estimated Levels at Time of Last Occupational
       Exposure to TCDD, Based on First-Order Elimination Kinetics and a Half Life
       for Elimination of 7.1 Years 	  8-91

8-13   Estimate of Total Dose Based on Adjusted Median Concentration at Test Times,
       Estimated Concentrations at Last Exposure, and Average Duration of Exposure, by
       Study Cohort  	  8-96

8-14   Estimated Lifetime Average Daily Doses and  Relative Risks by Individual Study
       Cohort	  8-97
                                                                                   08/27/92

-------
                            DRAFT-DO NOT QUOTE OR CITE
                              LIST OF TABLES (continued)
8-15   Calculation of Incremental Unit Cancer Risk Estimates and 95% Lower Limits for
       Both the Additive and Relative Risk Models Based on the Lung Cancer Deaths
       Response in the Fingerhut, Zober, and Manz Studies  	8-100

8-16   Calculation of Incremental Unit Risk Estimates and 95% Lower Limits for Both the
       Additive and Relative Risk Models Based on the Total Cancer Deaths Response in the
       Fingerhut, Zober and Manz Studies   	8-101
                                            vi                                    08/27/92

-------
                            DRAFT-DO NOT QUOTE OR CITE
                                   LIST OF FIGURES
8-1     Dose/Response Graph Showing Proportional Relationship Between Receptor
       Occupancy and Biological Response (Semilog Scale)	  8-6

8-2     Dose/Response Graph Showing Proportional Relationship Between Receptor
       Occupancy and Biological Response (Arithmetic Scale) 	  8-7

8-3     Multistage Carcinogenesis	  8-9

8-4     The Linear Multistage Model of Carcinogenesis	  8-11

8-5     Developing a Mechanistically-Based Mathematical Model	  8-16

8-6     A Two-Stage Model of Carcinogenesis	  8-44

8-7     Relative Risks of Lung Cancer and All Cancer Mortality in Three Recent Cohort
       Studies of Workers Exposed to TCDD by Estimated Lifetime Average Daily Dose
       Intake	  8-99

8-8     Biologically Based Risk Assessment Approaches for Dioxin:  Filling the Gaps	8-105
                                            vii                                     08/27/92

-------
                           DRAFT-DO NOT QUOTE OR CITE









                              LIST OF ABBREVIATIONS









ACTH             Adrenocorticotrophic hormone




Ah                Aryl hydrocarbon




AHH              Aryl hydrocarbon hydroxylase




ALT              L-alanine aminotransferase




AST              L-asparate aminotransferase




BDD              Brominated dibenzo-p-dioxin




BDF              Brominated dibenzofuran




BCF              Bioconcentration factor




BGG              Bovine gamma globulin




bw                Body weight




cAMP             Cyclic 3,5-adenosine monophosphate




CDD              Chlorinated dibenzo-p-dioxin




cDNA             Complementary DNA




CDF              Chlorinated dibenzofuran




CNS              Central nervous system




CTL              Cytotoxic T lymphocyte




DCDD            2,7-Dichlorodibenzo-p-dioxin




DHT              5a-Dihydrotestosterone




DMBA            Dimethylbenzanthracene




DMSO            Dimethyl sulfoxide




DNA              Deoxyribonucleic acid
                                          viii                                   08/27/92

-------
                            DRAFT-DO NOT QUOTE OR CITE
                            LIST OF ABBREVIATIONS (cont.)
DRE




DTG




DTK
ECOD




EGF




EGFR




ER




EROD




EOF




FSH




GC-ECD




GC/MS




GOT




GnRH




GST




HVH




HAH




HCDD




HDL




HxCB
Dioxin-responsive enhancers




Delayed type hypersensitivity




Delayed-type hypersensitivity




Dose effective for 50% of recipients




7-Ethoxycoumarin-O-deethylase




Epidermal growth factor




Epidermal growth factor receptor




Estrogen receptor




7-Ethoxyresurofin 0-deethylase




Enzyme altered foci




Follicle-stimulating hormone




Gas chromatograph-electron capture detection




Gas chromatograph/mass spectrometer




Gamma glutamyl transpeptidase




Gonadotropin-releasing hormone




Glutathione-S-transferase




Graft versus host




Halogenated aromatic hydrocarbons




Hexachlorodibenzo-p-dioxin




High density lipoprotein




Hexachlorobiphenyl
                                            IX
                                                               08/27/92

-------
                            DRAFT-DO NOT QUOTE OR CITE
                            LIST OF ABBREVIATIONS (cont.)
HpCDD




HpCDF




HPLC




HRGC/HRMS




HxCDD




HxCDF





ID*,




I-TEF
LH




LDL




LPL




LOAEL




LOEL




MCDF




MFO




mRNA




MNNG




NADP




NADPH




NK
Heptachlorinated dibenzo-p-dioxin




Heptachlorihated dibenzofuran




High performance liquid chromatography




High resolution gas chromatography/high resolution mass spectrometry




Hexachlorinated dibenzo-p-dioxin




Hexachlorinated dibenzofuran








International TCDD-toxic-equivalency




Dose lethal to 50% of recipients (and all other subscripter dose levels)




Luteinizing hormone




Low density liproprotein




Lipoprotein lipase activity




Lowest-observ able-adverse-effect level




Lowest-observed-effect level




6-Methyl-l ,3,8-trichlorodibenzofuran




Mixed function oxidase




Messenger RNA




W-methyl-/V-nitrosoguanidine




Nicotinamide adenine dinucleotide phosphate




Nicotinamide adenine dinucleotide phosphate (reduced form)




Natural killer
                                                                                   08/27/92

-------
                            DRAFT-DO NOT QUOTE OR CITE
                            LIST OF ABBREVIATIONS (cont.)
NOAEL




NOEL




OCDD




OCDF




PAH




PB-Pk




PCB




OVX




PEL




PCQ




PeCDD




PeCDF




PEPCK




PGT




PHA




PWM




ppm




ppq




ppt




RNA




SAR
No-observable-adverse-effect level




No-observed-effect level




Octachlorodibenzo-p-dioxin




Octachlorodibenzofuran




Polyaromatic hydrocarbon




Physiologically based pharmacokinetic




Polychlorinated biphenyl




Ovariectomized




Peripheral blood lymphocytes




Quaterphenyl




Pentachlorinated dibenzo-p-dioxin




Pentachlorinated dibenzo-p-dioxin




Phosphopenol pyruvate carboxykinase




Placental glutathione transferase




Phytohemagglutinin




Pokeweed mitogen




Parts per million








Parts per trillion




Ribonucleic acid




Structure-activity relationships
                                             XI
                                                                 08/27/92

-------
                           DRAFT-DO NOT QUOTE OR CITE
                           LIST OF ABBREVIATIONS (cont.)
SCOT




SGPT




SRBC




t*




TCAOB




TCB




TCDD




TEF




TGF




tPA




TNF




TNP-LPS




TSH




TTR




UDPGT




URO-D




VLDL




v/v




w/w
Serum glutamic oxaloacetic transaminase




Serum glutamic pyruvic transaminase




Sheep erythrocytes (red blood cells)




Half-time




Tetrachloroazoxybenzene




Tetrachlorobiphenyl




Tetrachlorodibenzo-p-dioxin




Toxic equivalency factors




Thyroid growth factor




Tissue plasminogen activator




Tumor necrosis factor




lipopolysaccharide




Thyroid stimulating hormone




Transthyretrin




UDP-glucuronosyltransferases




Uroporphyrinogen decarboxylase




Very low density lipoprotein




Volume per volume




Weight by weight
                                           xn
                                                               08/27/92

-------
                            DRAFT-DO NOT QUOTE OR CITE
                            AUTHORS AND CONTRIBUTORS

       The Office of Health and Environmental Assessment (OHEA) within the Office of Research
and Development was responsible for the preparation of this chapter. The chapter was prepared
through Syracuse Research Corporation under EPA Contract No. 68-CO-0043, Task 20, with Carol
Haynes, Environmental Criteria and Assessment Office in Cincinnati, OH, serving as Project Officer.
       During the preparation of this chapter, EPA staff scientists provided reviews of the drafts.

AUTHORS
       This chapter was prepared by the Dioxin Dose-Response Modeling Workgroup.
       The Workgroup is co-chaired by M.A. Gallo (Environmental and Occupational Health Sciences
Institute [EOHSI], Piscataway, NJ) and G.W. Lucier (National Institute of Environmental Health
Sciences [NBEHS], NC).  Other members are:  M. Andersen (Duke University, formerly of Chemical
Industry Institute of Toxicology [CIIT], NC); S. Bayard and P. White (U.S. EPA, Washington, DC),
K. Cooper, P. Georgopolous, and L. McGrath (EOHSI, Piscataway, NJ); E. Silbergeld  (University of
Maryland, Baltimore); M. DeVito (U.S. EPA, NC); L. Kedderis (CEM, University of North  Carolina-
Chapel Hill); J. Mills (CIIT, NC); and C. Portier (NIEHS, NC).
       The two Appendices were not prepared by the Workgroup but are included with the gracious
permission from the authors.

EPA CHAPTER MANAGER
Steven P. Bayard
Office of Health and Environmental Assessment
Washington, DC
                                           xiii                                    08/27/92

-------
                          DRAFT—DO NOT QUOTE OR CITE






                  8.   DOSE-RESPONSE MODELING FOR 2,3,7,8-TCDD







8.1.   INTRODUCTION



     Most  of  the information presented  in  the Introduction is  found  in more



extensive detail in the other background chapters.  We feel that it is useful to



summarize the salient  features of those papers which impact on the development



of dose response models so that readers of this chapter will be  able to evaluate



the scientific foundation on which our dose response models are based.



     2,3,7,8-TCDD is the most potent form of a broad family of xenobiotics that



bind to an intracellular protein known as the Ah receptor.  Other members of this



family  include  halogenated  hydrocarbons  such as  the PCBs, naphthalenes  and



dibenzofurans, as well as nonhalogenated species such as 3-methylcholanthrene and



3-naphthaflavone.   The biologic properties of dioxins  have been investigated



extensively in over 5000 publications and abstracts since the identification of



TCDD as  a chloracnegen  (Kimming and Schulz,  1957).   Much  of  the  biological



activity of TCDD  follows  the rank order binding affinity of the congeners and



analogs to the Ah  receptor (AhR).  This rank order holds for toxic responses such



as  acute toxicity  and teratogenicity, and  concentration of  several  hepatic



proteins including the up regulation of P-450IA1  and IA2, and the modulation of



the estrogen receptor  and EGF  receptor.   The carcinogenicity of TCDD has been



shown in several strains of laboratory mice and rats, and the tumor sites include



liver,  thyroid and the respiratory tract, as well as others. However, TCDD does



not interact directly with the  DNA to cause  mutations.  The study most utilized



for the cancer risk assessment of TCDD is that of Kociba et al. (1978).   These



authors reported an increase in hepatocellular carcinomas and hepatomas,  along



with decreases in several endocrine tumors in  female rats receiving TCDD at the



level of 1 ng/kg/day.   Male rats were remarkably less susceptible to TCDD action



in  these  studies.    However,   there  is  no  striking  sex  specifity  in  the



hepatocarcinogenic actions of TCDD in mice.



     The overall hypothesis of TCDD action, put forth by several  groups, has been



proposed  for  the effects  of TCDD  on the  transcriptional activities of  the



cytochrome P-450IA1 gene.  The biological basis for this approach is outlined in



the chapter by Whitlock.   Although substantial gaps in our knowledge remain,  it




                                      8-1                              08/27/92

-------
                         DRAFT—DO NOT QUOTE OR CITE






is accepted by most researchers that all cellular responses to TCDD follow upon



the interaction between TCDD and an intracellular macromolecule,  the so called



Ah or Dioxin receptor (AhR).  The binding of TCDD to AhR is similar, although not



necessarily identical,  to the interaction of many  steroid  hormones  with their



intracellular receptors, as pointed out by Gustafsson in a series of articles



drawing comparisons between the AhR and the glucocorticoid receptor (Poellinger



et al.,  1986a,  1987; Cuthill et al., 1988).  DeVito et al.  (1991) have also drawn



the analogy of  the action  of TCDD and estradiol at  their  respective receptor



protein.    Each AhR   appears  to  bind  one molecule of  TCDD,  and  at  low



concentrations of ligand (i.e., when [ligand]«[receptor]), the binding of TCDD



to AhR is linearly related to  [TCDD].  However, the  presence  of  TCDD in cells



induces an increase in the cytosolic concentration of AhR, possibly by increasing



receptor synthesis, displacement of an endogenous ligand from the  receptor or by



movement  of  "spare"  receptors from the  nucleus.   The increase  in available



receptors could amplify the signal  associated with receptor binding by increasing



opportunities for TCDD to bind to AhR.



     The binding of TCDD to AhR is reversible.  However, subsequent events seem



to reduce the  likelihood of  dissociation of the  ligand:receptor  complex.   One



such  event  that  has  been  recently  studied  is  the  association  of  the



ligand:receptor  complex  with another  macromolecule,  the  so called  ARNT  (AhR



nuclear transport)  protein  (Hoffman et al.,  1991).   There may be a  family of ARNT



proteins  that  differ  by cell types  which could  account, in part,  for  the



diversity of actions of TCDD  in different tissues.   The association of ARNT with



the ligand bound receptor  induces some physical  changes  in the complex, which



tends to reduce dissociation of the ligand and favors the movement  of the complex



into  the nucleus.   Overall,  the  relationship  between TCDD concentration and



nuclear  AhR-TCDD concentration  appears to be  linear, indicating that  at low



ligand concentrations,  ARNT is not a rate limiting  factor.  In the nucleus, the



AhR-ARNT-TCDD complex  (activated TCDD complex) associates with specific elements



in the  genome  called  the xenobiotic (or dioxin) responsive elements (XREs or



DREs).   Some of these  elements have now been sequenced and identified upstream



of promoter elements in several dioxin responsive  genes (Sutter  et al., 1992;







                                     8-2                              08/27/92

-------
                         DRAFT—DO NOT QUOTE OR CITE






Denison, 19??).  The association of the activated TCDD complex with the XRE is



also reversible, but little is known of its kinetic properties  (Gasiewicz et al.,



1991).   This  XRE binding  in  turn evokes  the  production  (or  perhaps  the



suppression) of several mRNA species.  The structure and amino acid sequence of



the AhR protein was reported by Poland and co-workers  (Burbach et al., 1992; Ema



et al., 1992).  Both the XRE(s) and the structure of the AhR are analogous to the



steroid receptors  and the respective genomic response elements.  This similarity



is important in regard  to biological models of TCDD action and risk assessment.



     The steroid hormones  and their receptors belong to a multigene family that



includes the  thyroid  hormone receptors,  oncogene products,  glucocorticoids,



mineralcorticoids, vitamin  D, retinoids,  androgens,  estrogens  and  progestins



(Evans  et  al.,  1988).   Biologically, these  are all multipotent  agents,  that



induce a range of  cellular responses in different organs, many at extremely low



concentrations.   They share  a  nuclear  location for  the  transduction  of



ligand:receptor action, and their common mechanism of action is the regulation



of gene expression (Jensen,  1991??).  Within the family of known receptors from



these  agents,  there  is considerable  sequence  homology  and  a  common  basic



structure,  consisting of a  ligand-binding  domain and a DNA-binding domain.  The



biological activity of these receptors is varyingly regulated by metals and by



phosphorylation state.   Some — but not all— hormone receptors may  interact with



chaperone-type  proteins,   subsequent  to   ligand  binding,   which  transduce



conformational changes  and other events critical to nuclear  translocation and DNA



binding.  The Ah receptor nuclear  translocator protein  (ARNT) functions in this



fashion (Hoffman et  al., 1991).  Other receptors are associated with heat shock



proteins that must be shed to transform the liganded receptor  into a DNA-binding



form, and the DNA-binding domain of some receptors contains zinc finger loops.



     The steroid hormone receptors regulate gene transcription through specific



DNA sequences near the regulated gene.   But the situation  is even more complex



because of interactions between  the liganded receptor and nuclear proteins that



function as transcription  factors  by binding to other DNA sites upstream of the



regulated gene.  These transcription factors may regulate  the binding affinity



of the steroid hormone receptor  itself  to DNA  (Muller  et al.,  1991).   As







                                     8-3                              08/27/92

-------
                         DRAFT—DO NOT QUOTE OR CITE






mentioned above, a consensus binding  site has been proposed for the Ah receptor



by Denison  et  al.  (1988).   A final  level  of complexity is  introduced  by the



interactions among steroid hormone receptors, at the genetic level, and by the



effects of hormone  upon the number, conformation, and localization of receptors.



Down- and up-regulation of receptor gene transcription and receptor synthesis,



may also  be involved  in cell-level modulation of steroid hormone  action.   As



pointed out by Muller and Renkowitz (1991), every step in the signal transduction



pathway   of   these   hormones,   from   receptor   gene   transcription   to



ligand:receptor:DNA action, is likely to be inter- and independently regulated.



     Attempts to model the steps involved in signal transduction have examined



events step-by-step as  well as  the overall set of reactions from entrance of



hormone to cellular response. Of  interest to us is the information that may be



available  concerning  the  overall   dose:response  relationship  from  steroid



hormones.  The highly complex cascade of biological events that intervenes from



hormone entrance to cell response may modulate hormone action in the following



ways:   it may  amplify  cell response, in the  way that  second  messengers for



membrane-associated receptors  (such  as neurotransmitter receptors)  appear to



amplify molecular signalling; it may transduce response in a manner proportional



to molecular signalling; it may transduce response in a manner proportional to



concentration of hormone (that is, linearly); or it may introduce dampening into



the response network.  Amplification of signal transduction implies that  at some



stage in a multistep process, more that one event  is triggered as a consequence



of one preceding event.   Dampening implies that at  some stage,  more than one



event must  occur before the  next event  is triggered.   Straight transduction



implies that the relationship of event to event, for all steps, is invariant.



     In  considering  these possible  dose:response relationships,  it  may be



important to distinguish among endogenous and exogenous  ligands  for the same



steroid  hormone receptor,  particularly if the two types of ligands differ in



rates of turnover (degradability)  or  affinity for the receptor. We are hampered



in our inferences for the dioxins  because the endogenous  ligand(s), if present,



has not yet been identified,  and thus, we are  not certain  if TCDD is more or less



stable than this ligand,  or if its affinity  is higher or lower than an endogenous







                                      8-4                              08/27/92

-------
                          DRAFT—DO NOT QUOTE OR CITE


ligand.  With  respect  to stability,  it is unlikely that  an endogenous ligand

would be as  stable  as  TCDD.   Most endogenous ligands  for steroid hormone and

other  receptors  are rapidly  cleared, either  by   compartmentation  (as  with

neurotransmitter  reuptake processess) or  by enzymatic  degradation,  as  with

peptides.  With respect  to kinetic*  of bidding of TCDD,  its  affinity for the
                                  *       I
receptor is extremely high, in the 10'" radge.  If the affinity for the natural

ligand is even higher,  then it is  likely that the overall relationships between

natural ligand and receptor are even  stronger than those  we may explicate for

TCDD; if it is lower, then it  would be an unusual member of the steroid hormone

family.  Of course,  differences in affinity,  if these exist, may not influence

the overall kinetics of the dose:response relationship as much  as differences in

the number of events required to trigger the reaction from step to step.

     Evaluation of  dose-response relationships  for  receptor-mediated events

require  information   on  the   quantitative   relationships   between  ligand

concentration, receptor occupancy and biologic response.   Roth and Grunfeld in

The Textbook of Endocrinology (1985)  state:
         "At very  low concentrations of hormone  ([H]«Kd),  receptor
         occupancy occurs but maybe trivial; i.e., the curve approaches
         0% occupancy of receptors.   But if there are 10,000 receptors
         per cell (a reasonable number for most systems), the absolute
         number of complexes formed  is respectable even at low hormone
         concentrations.  One advantage of this arrangement is that the
         system is more sensitive to changes in hormone concentration;
         at receptor occupancy  (occupied receptors/total receptors, or
         [HR]/[Ro]),   below 10%,  [HR]  is  linearly related to  [H],
         whereas at  occupancies of  10 to  90%,  [HR]  is  linear  with
         log[H]-a given increase in  [H] is more effective in generating
         HR at the lowest part of the curve than at the middle."
Figure 8-1 illustrates a  situation where  there  is  a proportional relationship

between receptor occupancy and biological response.   In this situation occupancy

of one receptor would produce a response although it would be unlikely that this

response could be detected.  It :'.s important "to note that the data in Figure 8-1

are plotted on  a semilog  scale.   If the  same data are  plotted arithmetically

(Figure 8-2),  then  the shape of  the  dose-response curve readily  conveys the

linear relationship between  receptor occupancy and  biological response at lower

concentrations and saturation at higher concentrations.

                                      8-5                              OB/21/92

-------
                              DRAFT—DO NOT QUOTE  OR CITE
    100
  S
it

ii
50
 % I
 tr «
 oS

 13

 It
   £
                                                                                 .• 10.000
                                                                       99% - 9900
             Total sites « 10000 per cell


             Affinity of Receptor K, » 10"'
                             K. - 10*
                                                            9100
                                                   Occupancy of 50% of Receptor*

                                                     « 5000 Sites Occupied
    001% - 1
    X    0 V
                                               900
                    - 10
                               100
                                                     z
                                                     c
                                                     1
                                                     a
                                                     fO
                                                                                         o
                                                                                         n
                                                                                         •a
                                                                                    5000
        10-"
                  10"1
                        10-"
    io-
-------
                     DRAFT—DO NOT QUOTE OR CITE
        0
                              FIGURE 8-2

Dose/Response Graph Showing Proportional Relationship Between Receptor
         Occupancy and Biological Response (Arithmetic Scale)
                                  8-7
08/27/92

-------
                         DRAFT—DO NOT QUOTE OR CITE






     Such a  simple  proportional relationship is  not  adequate to  explain  the



diverse biological  responses caused by  a single  hormone  utilizing a  single



receptor.    For example, low concentrations of  insulin produce much  greater



effects on fat cells than on muscle cells.  These differences are due to tissue



and cell  specific  factors  that  modulate the qualitative relationship  between



receptor occupancy  and  response.   Therefore,  it would  be expected that there



would be markedly different dose-response relationships for different effects of



TCDD.  Coordinated biological responses, such as TCDD-mediated increases in cell



proliferation, likely involve other hormone systems, which means that the dose-



response relationships for  relatively simple responses  (i.e., CYPIA1 induction)



may not accurately predict  dose-response relationships for  cancer.   As  we gain



more understanding of the entire  sequence of events responsible for TCDD-mediated



toxic effects, we  will enhance our ability  to  more  accurately  predict dose-



response  relationships.    The  mechanism(s)  responsible for  qualitative  and



quantitative diversity in receptor-mediated responses  will be discussed in more



detail in Section V "Knowledge Gaps."



     Dose-response relationships for TCDD's toxic effects have been established



for  several  endpoints  in  intact and  surgically altered animals.  In vitro



experiments have been used to determine critical concentrations and structural



relationships for TCDD effects at the cellular and molecular  level.  In the vast



majority  of  these  studies  the  role of  the AhR-ligand interaction has  been



essential  but not  necessarily  sufficient  to  evoke  a detectable biological



response.  It must  also be  kept  in mind that not all responses will necessarily



require binding of  the Ah receptor ligand complex to responsive elements on DNA.



     Cancer  is  a multistage disease using  a model composed  of four  or five



operationally defined processes:  initiation, fixation,  promotion and progression



(Figure 8-3).  In experimental systems, initiation  is  generally thought to be a



DNA  damaging or altering  event,  and  fixation  is  the  immortalization  of  the



mutation  in clonally  expanded  progeny.    Promotion  is the enhancement,  via



modifications  in growth kinetics,  of  the  initiated  cell  population by an



endogenous and/or exogenous  factors.  Progression  refers to the growth of the



tumor  to  an  end stage.    Tumor promotion in experimental  animals  is  a well







                                     8-8                             08/27/92

-------
                  DRAFT—DO NOT QUOTE OR CITE
                            M
                            (0
                            I
•o
S3
« a
r
~4
Q
£
«
01
O
*
1
c
c
                                                              0)
                                                      01
                                                      •o
                                                      «4
                                                      0)


                                                      U
                                                              g1   w
                                                         w   2   i
                                                          I    ~L   O
0
•H
JJ
(0
u
IA
4J

X
•*
fh
ID
U
8-
u
                                             CO

                                             I
                                             (9
•o
0> -i
« o>:i
»l«
O
~4
Q
^^
4J
01
Q
U

(0



cr


CD
•^
4J
                                                     •o
                                                                  h
                                                                  01
                                                                  8
                                                                  01
                                                                  u
o
«
u
fr
ec
                      0
C
01
>
U
                  &  4J
                  Q
          4J

          2
          0)
                            M
                      0)
                      a
                                 8-9
                                                             08/27/92

-------
                          DRAFT—DO NOT QUOTE OR CITE

established  paradigm that  has  demonstrated carcinogenicity  in many  tissues
including liver, breast,  bladder  and  skin.   Dioxin is a promoter  in  skin  and
liver of nice  and  rats,  respectively.  However, TCDD  promoted,  DEN-initiated
liver tumors or foci  are suppressed in ovariectomized rats.  Hence, promotion of
DEN initiated hepatocytes  by TCDD in rat* is dependent on ovarian factors or the
receptors for these factors.  Articles by Pitot et  al.  (1987) and Lucier et al.
(1991) present data on liver tumor promotion by TCDD.
     The general approach  of the U.S. EPA to regulation  of carcinogens is to use
the  the  Armitage-Doll model  of  carcinogenesis  [Linearized Multistage,  IMS]
(Figure 8-4).
     In this model, the movement of cells from one stage to the other are assumed
to be due to a sequence of mutations similar to the step of initiation/fixation
discussed above.
     As with any mathematical  model, specific forms must be chosen for the rate
constants which define the process.
     The  EPA formulation of  this model assumes the mutation rates [u,(d,t)] are
a linear  function of dose  and  are  constant over time.  These assumptions result
in a tumor incidence rate which is a polynomial function of dose.   In the low
dose region, the upper  bound on risk is  dominated by the linear term in the
polynomial (qj*).  EPA generally uses a upper confidence limit on the linear term
of this formulation of the multistage model for cancer risk assessment.  However,
this choice has not  been  predicated solely  on  the  correctness of a K-mutation
process of cancer development.  The linearized mathematical properties of the
multistage model can be  appropriate for  a  larger  class of mechanisms:   Dose
response behavior,  which is  linear at low dose,  may  have upward curvature in the
intermediate range and shows downward curvature or saturation of response at high
dose.   In  particular,  arguments  that  a  compound's  action is additive  to
background biological processes  lead  to a  linear  response at low  dose under
rather general conditions (Crump, 1976??).  Therefore, for practical modeling
purposes, it  is important  to address whether  biological knowledge  about the
action of a carcinogen can fit the general dose-response shape predicted by the
linearized multistage model.  Cross species extrapolation is carried out using

                                     8-10                             08/27/92

-------
              DRAFT—DO NOT QUOTE OR CITE
u
•o


3*
                                                            m
                                                            9)


                                                            I
                                                            3
                                                        I
                                                        00

                                                        g
                                                        D
                                                        u
                                                            «0
                                                            *J
                                                            D
                                                            ^
                                                            4J
                                                            i-(
                                                            3
                                                            10
                                                            (D
                                                            0)
   •O


   a"
          a

          0
                           8-11
                                                             08/27/92

-------
                         DRAFT—DO NOT QUOTE OR CITE

dose expressed as  daily exposure per unit  surface  area,  where an  allometric
scaling factor is  applied  to body weight to estimate  the surface area.  The
potency of a carcinogen  is related to the  slope of the dose-response curve (Q*).
     For other toxicological  endpoints such as  terata, organ toxicity,  acute
toxicity,  etc., a threshold has  been assumed as  a natter of policy.   For these
endpoints, safety factors or uncertainty  factors have been used to estimate no-
effect exposure levels (NAS/NRC,  1977).  This threshold approach is used by the
World Health Organization to  set ADIs for direct and indirect  food  additives.
EPA now uses the term reference dose.  As discussed above, EPA policy assumes the
dose-response curve for  excess carcinogenic risk  to be linear through dose zero.
Several mechanisms  could generally lead to this form of response including direct
mutational activity of the chemical agent and/or additivity to background rate
of tumor formation  (Portier, 1976??).  Since TCDD does not bind covalently to DNA
and must exert its effects through receptor action,  this default position must
be carefully reexamined.
     Since TCDD action is clearly receptor mediated,  it  is reasonable to attempt
to model the receptor kinetics using equations that  have been used for steroid
receptor mediated responses.   Recent rodent experiments  suggest the induction of
cytochrome P-450IA1 is linear through zero using long term exposure over a wide
dose range (Taitscher et al., 1992;  DeVito et al., in press).  Exposure data in
humans indicates that the general population has a body burden of  5-10 ppt (lipid
adjusted).   In occupational epidemiology,  human disease  associated  with TCDD
exposure has not been  detectable until  fat levels are one to two  orders  of
magnitude higher than the general population (100-1000  ppt).  Thus, in addition
to other Uganda, one must also be concerned with the present level of TCDD in
the general  population  and  the  long environmental and  biological  half-life of
this compound.
     The current effort to reevaluate the risk of exposure to dioxins is being
termed a Biological Basis for Risk Assessment.   The underlying premise is that
this is a  special  case  for  a nonmutagenic,  receptor mediated carcinogen.  The
goal of this reassessment is to consider more mechanistically-based models which
                                     8-12                             OB/27/92

-------
                         DRAFT—DO NOT QUOTE OR CITE






are sufficiently credible to the scientific community.   Such approaches can be



recommended for practical risk assessment needs.



     There are several models under consideration at the  present  time ranging



from very simple to complex.  What has become obvious over the past year is that



the biology governing the toxicity of TCDD, beyond a few initial critical events,



is not straightforward.  These critical events,  the first of which is binding to



the Ah  receptor,  are  generally response-independent.    The  response-dependent



events  are species, gender,  organ,  tissue and  perhaps  cell specific.   If the



binding to the AhR is essential but not sufficient for effects to occur, then the



dose-response curve for this event (as well as  the rate equations) should be a



better predictor than dose of subsequent  actions  as long  as the dose-response



curves for these subsequent actions are parallel to the receptor binding curves.



In general, the data to date indicate receptor mediation or receptor coordination



does indeed exist for most if not all low dose  actions  of  TCDD.   Since the AhR



has been detected in virtually all cells but all  cells  do  not respond to TCDD,



there must be other factors that are necessary for TCDD action.  The roles of the



other factors must be elucidated before there is a  complete understanding of TCDD



action.    However,  a  relatively complete model can be  developed  for specific



endpoints by using available data and reasonable  assumptions.



     Several important factors  have  been generally  accepted.  One,  TCDD  is a



member  of a  class  of  xenobiotics  (and  probably natural  products) that  is



nonmutagenic,  binds  to  a   cellular  receptor and  alters  cell  growth  and



development.     Two,  a  significant  amount  of  information  is  available  for



estimating risks  from  exposure  to this  compound  and  the default  position of



directly applying the linearized  multistage model (LMS) as  a  function of dose



needs to  be  reevaluated.  Three,  the biology  of  receptor mediation should be



included  in  any modeling exercise  for  TCDD  utilizing existing  mathematical



developments.  This is not to say that uncertainties do not  remain.  However, the



goal  of the modeling  is to use as  much  data  as  possible  to  reduce  these



uncertainties, and to identify the areas where  data  gaps exist.



     One difficulty with  a novel, albeit biologically based,  approach is that it



is  replacing paradigms (Safety Factors  and LMS  Models)  upon which  the  U.S.







                                     8-13                             OB/21/92

-------
                          DRAFT—DO NOT QUOTE OR CITE


government's risk assessments have been based.   There is no a priori reason to

believe that a more biologically-based model will  be more or less conservative

than IMS.  However, basing the modeling on  a mechanistic understanding of the

biochemistry of  TCDD-induced toxicity should  increase  our confidence  in the

resultant risk estimates.  As previously stated by Greenlee and collaborators:
     "Neither the position taken by U.S. EPA or by Environment Canada (and
     several other countries such as Germany and the Netherlands) is based
     on  any  detailed  mechanistic  understanding  of  receptor-mediated
     interactions between dioxin and target tissues.   Biologically-based
     strategies use  knowledge  of the mechanistic  events in the  various
     steps in the scheme  for risk  assessment.  Interspecies  extrapolation
     strategies would be  conducted  based on how these  mechanistic steps
     vary from species to species.  There are  numerous  steps that can be
     examined mechanistically,  and fairly  ambitious  programs  have  been
     proposed to examine the mechanistic details of many or  most of these
     individual steps.  More focused risk assessment  approaches are also
     being proposed based on examination of individual  steps believed to be
     critical in establishing the overall shape of the dose-response  curve
     for the  induction  of  tumors (or other toxic  endpoints) by dioxin."
     (Greenlee et al., 1991)
     This chapter presents the current thinking on TCDD mechanistic action.  It

examines several endpoints  and focuses  on dose response models  for  cancer in

laboratory animals and man.   It also evaluates for use of biomarkers  of TCDD

action as surrogates  for  modeling  receptor mediated events.    In addition the

chapter presents a section  on  "Knowledge  Gaps."  Critical  examination  of this

section  leads  to new  experiments  which  will  add to  the already  impressive

database on TCDD action. Addition of key molecular,  cellular and tissue specific

information to the current data base will be important  to establish a new risk

paradigm based  on biological  mechanism  of action of  TCDD and  perhaps  other

receptor-mediated nonmutagenic toxicants.   The  chapter reviews some of  the

critical data on noncancer endpoints  but does not attempt to model them.  These

endpoints are  clearly important when  considering the  public  health  risk of

dioxin.  However,  the lack of  key molecular information of action and molecular

dosimetry limits mathematical modeling of noncancer endpoints at this time.

8.1.1.   Introduction to  Modeling  for TCDD.   Mathematical modeling can  be a

powerful tool for understanding and combining information on complex biological

phenomenon.   The development  and use of mathematical models  is illustrated by



                                     8-14                             08/27/92

-------
                          DRAFT—DO NOT QUOTE  OR  CITE






Figure 8-5.  In the development of a mechanistically-based mathematical model,



the beginning  point  is generally  a series of experiments  aimed  at studying a



xenobiotic agent.  The experimental results (data) can indicate a mechanism which



leads  to  the  creation  of a mathematical  model.   The  model is used  to  make



inferences which are then validated against the existing knowledge base for the



effect   under  study  and the  xenobiotic agent.    This can  then lead  to new



experiments  and further laps through this model development  loop.   Each time



through the loop the model either  gains additional validation by predicting the



new experimental results or it is  modified to  encompass  the new results without



sacrificing the prediction of previous results. In either case, subsequent loops



through the model generally increase our confidence in the model (although it may



be difficult or impossible to quantify this confidence).



     There is  no one model  development loop  for  any given compound or effect.



Instead, there are always numerous exercises which lead  to the development of a



mechanistic model.  In modeling the effects from exposure to TCDD,  there are many



smaller model development circles which make up the larger overall model.   For



example, a  mechanistic approach to TCDD-induced carcinogenicity must include



models of exposure, tissue distribution, tissue diffusion, cellular biochemistry,



cellular action, tumor  incidence  and  cancer mortality.  At each stage and for



each model, data must be collected and understood in order for the model to be



valid and acceptable as a tool  for understanding the  observed effects and for



predicting the effects  of TCDD outside of the range of experimental findings.



     Confidence in  any one model  is  not only dependent upon  the information



available for that compound, but is also supported by the information available



on other  systems  which act  similarly  and for which models  have already  been



developed.  In  the case of TCDD,  the modeling of effects will be greatly enhanced



by  existing  information  on  the  receptor-based  systems,  general  work  in



physiologically-based pharmacokinetics models and in tumor incidence modeling.



     Risk assessment, however,  is  another issue.  The use of mechanistically-



based modeling in  extrapolating to exposure patterns and exposure doses outside



the range  of  the  data  is  in  its  infancy.   Even  though there may be  high



confidence in the ability of the model to predict experimental results,  there







                                     8-15                             08/27/92

-------
             DRAFT—DO NOT QUOTE OR  CITE
                       Experimental
                         E>idcncc
               /
              /
                                          Modr!
         ExcwriuieuUl                     Development
           D«l§n
                           Model
                          Inference
                       FIGURE 8-5

Developing  a Mechanistically-Based Mathematical Model


                          8-16                               08/27/92

-------
                          DRAFT—DO NOT QUOTE OR CITE

 could be low confidence in the ability of the model to predict outside the range
 of data.  Risk assessment inherently demands a careful scrutiny of the behavior
 of a mechanistic model under a variety of exposure scenarios, scrutiny which has
 not generally  been  applied  to  the   use  of  mechanistic  models  in science.
 Confidence  can only be  gained by  numerous  loops through this model development
 process for the express purposes of risk assessment.  Only then will the process
 be acceptable as  a tool for risk extrapolation.   It is important to  note that
 mechanistic modeling  has  a  role to aid  in  explaining  and  understanding
 experimental results which is devoid of its proposed use for risk assessment and
 that our confidence in the methods used in mechanistic modeling will differ upon
 the history of  it's use.    For  historical perspective,  it  is  important  to
 recognize  that  this  is  not the  first loop through the  cycle of mechanistic
 modeling for carcinogenesis.  Early exercises based upon tolerance distributions
 used  the  then  current understanding  of carcinogenesis to develop statistical
 models which could be used  for risk estimation.   Later use of the linearized-
 multistage  model  was  also  based  on an  understanding  of  the carcinogenic
 mechanism.    This  exercise   also  benefits  from  recent  attempts  to  use
 physiologically-based pharmacokinetic  (PB-PK) models in risk estimation.  Thus,
 this exercise is a  logical next step  in the continuum of mechanistically-based
 modeling for risk  assessment purposes.
     In any realistic and practical modeling exercise, the major component  of the
 model  revolve  around  the  estimation of  parameters  in  the  model  utilizing
 statistical  tools.   These  tools  range  from  very  simple techniques such  as
 estimating  a mean  to extremely complicated approaches such as estimation via
 maximizing  a statistical likelihood.  The estimation of parameters is not done
 in  a vacuum, but is intimately tied to the data  available to characterize the
 model.  The way  in which model parameters are estimated and  the data used  to
 estimate those model parameters are  the  major components  in  determining  the
 reliability and trust we will place in any mathematical model.  Fundamentally,
 sufficient biological data need to be available to convincingly show that  the
model correctly  represents in  vivo processes and that the processes modeled are
the ones that lead to toxic events.

                                     8-17                              08/27/92

-------
                         DRAFT—DO NOT QUOTE O3 CITE






     In modeling biological phenomenon, the  data  can be divided into  4  broad



categories as shown in Table 8-1.   At the  top are  effects on the whole  animal.



Examples of  data  included  in this category would be data on survival of  the



organism, its ability to reproduce and it's ability  to properly  function  (e.g.



behavioral data).   The levels of data then get increasingly specific going from



whole organism to tissue/organ system responses to  cellular responses  down to



biochemical responses in the cell.  Other  groupings  are possible and even more



detailed categories  (e.g. molecular biochemistry); the point  is  that the data



range from data at the bottom of  Table 8-1 which  is  extremely mechanistic and



deals with the interactions between molecules,  to  the data at the top of  Table



8-1, which is effectively counting bodies.  All of this information is useful and



should be  incorporated  into a  mathematical model  aimed  at  understanding some



biological response.



     Mathematical models which incorporate parameters which are  mechanistic in



nature do not automatically constitute "mechanistic models."  The available data



for characterizing the model and the method by which this data is incorporated



into the model are important in determining if a model is truly  "mechanistic" and



soundly based on the biology or is instead,  simply a curve fit to data.



     There are two basic ways in which biological effects  can be estimated.  The



first and  most  common approach is a  "top-down"  approach.   In  the "top-down"



approach,  data  on the effect of  interest (e.g.,  carcinogenicity)  is  modeled



directly by  applying  statistical  tools to  link the  observed data (e.g.,  tumor



incidence  data  from a  carcinogenicity   experiment)  to  a  model   (e.g.,  the



multistage model of carcinogenesis).  This approach is extremely powerful in it's



ability to describe the observed results and to generate hypotheses about model



parameters and the potential effects of changes in  these parameters.  Where this



modeling  approach begins  to  lack credibility  is in  the  ability  to  predict



responses outside the range of the data currently being evaluated. Even  when the



model  being  applied to the data is mechanistic  in the sense  that  the  model



parameters are tied to some mechanism for the toxic effect (e.g.,  mutation rates



and molecular  effects),  without direct evidence concerning the value  for this



parameter or even evidence supporting the particular structure of the model, one







                                     8-18                              08/27/92

-------
DRAFT—DO NOT QUOTE OR CITE
TABLE 8-1
Examples of Levels of Information Available for
Estimating Parameters In Dose-Response Modeling
Organism
Tissue
Cell
Biochemical
Morbidity
Mortality
Fertility
Hyperplasia
Hypertrophy
Carcinogenicity
Chemical Distribution/Disposition
Improper Development/Function
Mitosis
Cell Death
Gene Expression
Protein Levels
Receptor Binding
Adduct Formation
            8-19
OB/27/92

-------
                         DRAFT—DO NOT QUOTE OR CITE






is basically left with a  curve fit to the data.  The historical applications of



the "linearized" multistage model in risk assessment has been in this fashion.



     We view true mechanistic modeling in  a different fashion.  In this case, the



model structure  and the  parameters  in the model are derived  in a "bottom-up"



fashion.  The mechanistic parameters in  the model  are  estimated directly from



mechanistic data rather than from effects data or data one level higher in the



hierarchy of data illustrated  in  Table 8-1.   The  goal of  true mechanistic



modeling is to explain all or most known results relating to the process under



study in a way which is reasonable in it' s biology and soundly rooted in the data



at hand.  In this case, biological confidence in predictions from the model would



be much higher than that from the "curve fitting" approach.



     In practice, it is generally impossible to completely divorce mechanistic



modeling from curve fitting.  At some point in the modeling process,  gaps must



be  filled  relating the  modeled, mechanistic  effects  to the  observed toxic



effects.  It is generally at this point that  some amount of curve-fitting is



necessary to calibrate the mechanistic response to the toxic effect.   Although



not technically mechanistic modeling,  this combined approach is preferred to



simple curve fitting when inferences outside of the range of the toxic effects



data is desired.



     This is not intended to infer that with mechanistic modeling,  we can get  a



precise estimate of risk of  a toxic effect outside the range of the data  or even



a more  precise  estimate  of risk.  Without data, the statistical issue of the



accuracy of a prediction cannot be easily addressed.  Thus, while  there may be



a greater deal of biological confidence in extrapolated results,  it is unlikely



that an increased statistical confidence  can be demonstrated.  However,  for each



level  and  type of data,  there are  ranges  of exposure  beyond  which  it is



impossible  to  demonstrate  an  effect given the practical  constraints  on the



experimental protocols.    In general,  effects can  be  demonstrated  at lower



exposures for data at  the bottom of Table 8-1  compared  to the data at the top.



If  this is the case, there may be  both increased biological confidence in



extrapolated results   and  increased  statistical accuracy.    This  is also not



intended  to infer that models derived  through  curved fitting  should  always be







                                      8-20                             08/27/92

-------
                         DRAFT—DO NOT QUOTE OR CITE






given little weight.  Many of the advances in biology and science are due to the



application (either formally or informally) of a model  to data at a higher level



in order  to generate  hypotheses  about effects  at lower  levels.   The  major



difference between the application of a "curve-fit" model  in basic biology and



that in risk estimation is that in basic biology one is  creating hypotheses which



at some point will be tested.  In risk estimation, it  is unlikely that one will



ever be able to validate extrapolated risk estimates.



     Many side issues  are also  related to the frequency of use of  this  model



development  loop in  trying to understand a  biological  mechanism.   One  of



considerable importance is experimental design.  For mechanistic modeling aimed



at risk assessment, we  are just beginning to understand the types of experiments



which may benefit  the risk estimation process.   Because  of this, now  is the



perfect time to consider the types of designs which are  best suited to addressing



problems which are specific to risk assessment.  In general design situations one



would have a mechanism in mind,  qualitatively describe that mechanism and form



the structure of  a mechanistic model,  make educated guesses about the parameters



of this  model,  then use the  quantitative model  to locate designs which are



optimal at characterizing the mechanism. For the purpose of risk estimation, the



basic outline also holds.   There are also some simple  design rules which would



aid the extrapolation of results to doses outside the observed response range and



to humans from animals.  A few of  these design points  are  listed in Table 8-2.



     TCDD has been chosen as a prototype for exploring  and examining the ability



of mechanistic modeling to improve the accuracy of  quantitative risk assessment.



In essence,  the  data base for a mechanistic  modeling  approach  to  TCDD is very



extensive and contains a considerable amount of information on low-dose behavior.



In addition, there  is  good  human  data which is supported by the  experimental



evidence in animals.   On the other hand, some aspects  of the mechanism by which



TCDD induces  it's  effects,  such as binding  to the Ah receptor have  not been



modeled extensively and, thus, even for scientific  purposes are  in only their



first few loops through the model development cycle shown in  Figure 8-5. In this



case, several competing mechanistic theories will agree with the existing data



adding to the uncertainty  in any projected  risk  estimates.  This outcome  is







                                     8-21                             08/27/92

-------
DRAFT—DO NOT QUOTE OR CITE
TABLE 8-2
Design Considerations for Risk Assessment Purposes
Aspects of Risk Estimation
Mechanism
Species Extrapolation
Other
Design Points to Consider
Multiple Times of Observation
Multiple Doses
Multiple Ages at Exposure
Pharmacokinetics
Multiple Species
Both Sexes
Tissue Concentrations (including blood)
System Clearance
Essential interactions with endocrine systems
Record data on the level of individual animals
            8-22
08/27/92

-------
                          DRAFT—DO NOT QUOTE OR CITE






inevitable for a novel mechanism and  for the application  of  the technology of



mechanistic modeling to a new area.  To reiterate an earlier point, mechanistic



modeling has a role to  aid in explaining and understanding experimental results



which is devoid of it's proposed use for risk assessment and our confidence in



the methods used in mechanistic modeling will differ depending upon the history



of it's use.  As we know more about the limitations of current data and current



methods for the  application  of  mechanistic models to risk estimation,  we can



improve experimental designs  and significantly improve the process.  Since TCDD



is an  early attempt along these  lines, we  must  be cautious  in coming  to a



judgement concerning the overall utility of mechanistic modeling  as an important



tool for risk assessment based upon this one case  study.



8.1.2.   Dosimetric Modeling. Dose Delivery and Tissue Modeling and Biochemical



Modeling (See Appendix  for complete manuscripts by Kohn et al., 1992 and Andersen



et al., 1992)



     In the NAS report, Risk  Assessment in  the Federal Government: Managing the



Process  (NRC,  1983),  "dose-response  assessment"  referred to  the process  of



estimating the expected  incidence  of  response for various exposure  levels in



animals and humans.   Tissue response is not  always directly related to exposure.



This can be due to saturation  and activation of metabolic  pathways.  (Hoel et al.,



19??), influence of competing pathways having different  efficiencies of action



for  the parent  compound  and/or   its key metabolites   and  factors  such  as



cytotoxicity, mitogenesis or  endocrine influences which can radically modify the



homeostatic properties of the tissue.  These complex interactions can result in



markedly nonlinear  dose-response;  nonlinearities  which  could  lead  to  risk



estimates which may be greater or less  than  the risk derived from  a linear model.



Because of the potential for nonlinearities, it  is essential  to distinguish



between exposure  level and dose to critical  tissue when modeling  risks  from



exposure to  xenobiotics.     It  is  also  essential  that  we  understand  the



quantitative  relationship between  target  tissue  dose  and changes  in  gene



expression.   This  is  especially  true  when extrapolating  to  low doses  and



extrapolating across species.
                                     8-23                             08/27/92

-------
                         DRAFT—DO NOT QUOTE OR CITE






     For dioxin, the abundance of data  on many levels allows one to  create  a



collection  of  models  which  include  the  determination of  the  quantitative



relationship  between  dioxin exposure  and  tissue   concentration,   tissue



concentration and  cellular action,  cellular  action  and  tissue response  and



finally tissue  response and host survival  (Andersen and Greenlee, 1991; Portier



et al., 1984).   This portion  of  the reevaluation of  dioxin risks entails  the



description and development of mechanistically based mathematical models of  the



effects of  TCDD.   This includes a  discussion  of the extrapolation of  tissue



dosimetry and response  from high dose exposures to those expected at  much lower



exposure and the extrapolation of  dose-response  models from test  animals to



people. These extrapolations will be based upon empirical relationships used to



derive explicit,  though incomplete, biologically-based mechanistic models of the



events involved in the toxic action of dioxin.



     Biological modeling  is the process of developing mathematical descriptions



of the  interrelationships  among the mechanistic determinants of those  toxic



events resulting from  exposure to  TCDD.   Research with dioxin  has  focused on



biological  responses at  the levels  of organization shown in  Table  8-1 (i.e.,



biochemical, cellular,  tissue  and organism).  At each  level of organization, we



focus on the mechanisms responsible for these observations.  Mechanism refers to



the critical biological factors that regulate occurrence, incidence and severity



of a  particular factor and the nature  of the interrelationships among these



factors.  The details of the mechanisms of interaction differ markedly for the



various levels  of  biological organization  with  specific determinants  of  the



behavior at each level driving the creation on an appropriate quantitative model.



     For dioxin,  the mechanisms of three processes  are of  primary interest:



(1) the dosimetry  of  dioxin  throughout the body and specifically to target



tissues; (2) the molecular interactions  between dioxin and tissues, emphasizing



the  activation  of  gene  transcription  and   increases  in  cellular  protein



concentrations  of  specific,  growth  regulatory  gene  products and  specific



cytochromes; and (3)  the progressive tissue level  alterations  resulting  from



these  interactions which  lead, eventually,  to  toxicity.  The modeling process



involves identification  of the mechanistic determination of the dose-response







                                     8-24                             08/27/92

-------
                         DRAFT—DO NOT QUOTE OR CITE

continuum  through experimentation  and  the  encoding  of these  processes  in
mathematical equations.  The  extent  to which model predictions  coincide with
experimental results is a test of the validity of  the  model  structure.   After
validation, the model can be used for risk assessment.   In addition to their use
in risk assessment,  these models  have importance for aiding in  the design of
future research, both in terms of a basic understanding of dioxin toxicity and
further risk estimation.
     The  following  sections discuss the mechanistic  biological modeling for
dioxin with  regard  to dosimetry,  induction  of  gene transcription  and  tissue
response, especially those associated with hepatic carcinogenesis.  This modeling
effort follows a natural progression related to the kind  of information available
at the time at  which the model was developed.  We  will begin with a review of
tissue concentration followed by modulation of protein concentrations and tissue
response.
     Tissue dosimetry encompasses the absorption, distribution, metabolism and
elimination of dioxin from tissues within the body.  The  determinants of dioxin
dosimetry  in  the body  include  physicochemical  properties   (e.g.,  diffusion
constants, partitioning constants, kinetic constants and biochemical parameters
for metabolism) and  physiological  parameters (e.g., organ flows  and volumes).
The mathematical  structure which  describes  the interconnections  among these
determinants constitutes a mathematical model for the tissue dosimetry of dioxin.
The model  we will use in the  context of this reassessment  is a PB-PK model.
These types of  models describe the pharmacokinetics of dioxin with  a series of
mass-balance differential equations.   These  models have been validated in the
observable  response  range for numerous  compounds  in both animals  and humans
making them extremely useful for risk assessment; especially for cross-species
extrapolation.  In addition, they aid in extrapolation from one chemical to other
structurally related chemicals since many of the  components of  the model can be
deduced for structurally related compounds.  The development of  PB-Pk models for
general  use is discussed  in Galowski  and  Jain  (1983) and  for use  in risk
assessment by  Clewell and Andersen (1985).
                                     8-25                             08/27/92

-------
                          DRAFT—DO NOT QUOTE OR CITE



     In brief,  a PB-Pk model consists of a series of compartments that are based


on the anatomy and physiology of the test animals; hence, the term PB-PK (ref.


original use of term).    The  time course of behavior in each  compartment is


defined by an  equation  containing terms  for input and loss  of  chemical.   For


example, if Ct  represents the concentration of  compound in a tissue (£) and CB


the concentration of compound in blood (B), one of the simplest relationships one


might use is:
                     dC,
                         - rM CB ~ Iw ct ~ ** ct   (equation!)
where dct/dt represents the change  in the concentration in the tissue over time


(t), rjj/ is the rate (per  unit  concentration)  of  the  movement  of the compound


from blood  to tissue,  r^g the  rate  from tissue to blood  and  rm  the  rate of


metabolism in the tissue.  Equations of this form will be used in mass balance


modeling of the pharmacokinetic processing of TCDD.


     Several PB-Pk models  have been developed for dioxin and related chemicals


(see Chapter 1 for a brief overview).   PCBs have been extensively studied (Lutz


et al., 1977, 1984; Mathews and Dedrick,  1984).  King  et al.  (1983) modeled the


kinetics  of 2,3 7,8-TCDF  in several  species  and Kissel and  Rambarge (1988)


proposed a human PB-Pk model.


     The development of  PB-Pk  models for TCDD  began with work by Leung et al.


(1988)  in mice.  This model was extended  to Sprague-Dawley rats by  Leung et al.


(1990a) and to 2-iodo-3,7,8-trichlorodibenzo-p-dioxin in mice  (Leung  et al.,


1990b).  Since many of the regulatory standards for dioxin have been based upon


a finding of hepatocarcinogenicity in female  Sprague-Dawley  rats, we will focus


on the  model by Leung et al. (1990a) in this species.


     The  Leung et al.  (1990a)  PB-Pk model contains  five  tissue  compartments


including blood, liver, fat, slowly perfused  tissue and richly perfused tissue.
                                     8-26                              08/27/92

-------
                          DRAFT—DO NOT QUOTE OR CITE






This early model is blood flow  limited,  a  condition which is appropriate when



membrane diffusion is much more rapid than blood flow to the tissue.  Thus,  in



this PB-Pk model,  the tissue/tissue blood compartments are lumped together as a



single compartment in which the effluent venous blood concentration of TCDD is



equilibrated with the tissue concentration.  The model includes a TCDD binding



component in blood described by a linear process with an effective equilibrium



between the bound  and free TCDD given by a binding constant,  K^.   They also




include  binding  of   TCDD to  two classes of  protein  in  the  liver:    one



corresponding to the high affinity, low capacity Ah receptor and the other to a



lower affinity, higher capacity microsomal protein (CYP1A2) which is inducible



by TCDD.  In the  formulation of this  PB-Pk model  by Leung et al.  (1990a) both



types  of  binding proteins  are  explicitly  defined  using  an  instantaneous



correspondence between occupancy and induction using separate binding capacities



and dissociation constants for each protein.  These binding reactions are modeled



via Michaelis-Menten equations.



     In the Leung  et al.  (1990a) model, the tissue  storage capacity depends upon



the partition  coefficients (assumed  to  be linear with  concentration)  and the



specific protein  binding.  Dioxin is very lipophilic and is  found  in higher



concentrations in liver than would be  expected based on partition coefficients.



The specific binding of dioxin to a liver protein used  by Leung  et al. (19??) is



an improvement over earlier models for these  lipophilic compounds.



     In  various   studies,  dioxin   has  been  administered   by   intravenous



administration, intraperitoneal injection,  oral feeding or intubation (gavage)



or by  subcutaneous injection.   In the  PB-Pk modeling  framework,  intravenous



injection can be described by starting the integration with an  initial mass equal



to the dose in the blood compartment.  Oral intubation and sc injection can be



modeled as if they adhere to first-order uptake kinetics with dioxin appearing



in the liver blood  after oral administration and in the mixed venous blood after



sc injection.   Feeding was modeled by Leung et al.  as a zero-order input on days



that dioxin was included in the diet.  These descriptions of the  routes of uptake



are clearly not defined  in specific  physiological  terms; they are empirical
                                     8-27                             08/27/92

-------
                         DRAFT—DO NOT QUOTE OR CITE






attempts to estimate an  overall rate  of  uptake of TCDD into the  PB-Pk model.



This  is  one  area  in which  additional research  could improve  dose-response



modeling for TCDD.  Efforts to provide more biological details  concerning the



physiological basis  of  absorption across  these  various membranes,  including



intact skin, would prove valuable for exposure assessments with dioxin.  With the



iodinated  analog,  2-iodo,3,7,8-trichlorodibenzo-p-dioxin,  the estimated  rate



constant for oral absorption was considerably larger in  induced (0.15/hour) than



in naive animals (0.4/hour).  The physiological basis of this change is unknown.



     With many volatile  organic  chemicals there are convenient in vitro methods



for estimating partition coefficients (Dato and Nakajima,  1978;  Gargas et al.,



1989).  For TCDD and other highly lipophilic, essentially nonvolatile compounds,



there are no reliable in vitro methods and  these constants have to be estimated



from measurements of tissue and blood concentrations in exposed animals.  This



leads to a difficulty in differentiating between specific tissue binding and the



partitioning  to the tissue.   Leung  et al.  (1990a)  overcame this  problem by



assuming  binding occurred  only  in  the  liver  and that  the liver  partition



coefficient  was the same  as the  kidney.   This  permitted estimation of the



relative binding capacities  and affinities of specific hepatic  proteins.  The



predictions  from this  modeling exercise prompted a series  of  experiments to



examine the nature of these binding proteins in mice (Poland et al., 1989a,b).



     Metabolic clearance was modeled as a first-order process.  In the mouse with



the iodo-derivative, dioxin pretreatment  at maximally inducible  levels  caused a



3-fold increase in the rate of metabolism.  There  is no evidence to suggest an



increase  of  metabolism  in  the  rat for TCDD; however,  there is data supporting



small  increases in metabolic clearance at high  doses  for the tetrabrominated



analog (Kedderis et al.,  1991).  The identity of the enzymes responsible for TCDD



metabolism are  as yet unknown.



      Finally, Leung et al.  (19??)  kept all physiological parameters (flow rates,



tissue weights,  etc.) constant over the lifetime  of the animal.



      Dioxin  and dioxin  analogs have  dose- and time-dependent kinetics  in both



rodents  (Kociba et  al.,  1976; Poland  et al.,  1989a; Abraham  et al., 1988; Rose
                                      8-28                              08/27/92

-------
                          DRAFT—DO NOT  QUOTE OR CITE






et al., 1976; Tritscher et al.,  1992) and humans (Wolf, 19???; Carrier, 1991).



For single- and short-duration exposures, as the exposure level increases, the



proportion of total dose found in the liver increases.  For chronic exposures,



there appears to be a linear  relationship between dose and tissue concentration



in the gavage study  of  Tritscher et al.  (1992),  but this  may be  simply an



inability to  observe  nonlinearity in liver in intermediate dose ranges.   The



Leung et al. (1990a) model adequately predicts the tissue concentrations observed



by Rose et  al.  (1976), but  did  considerably worse at predicting  the results



observed by Kociba et  al.  (1976)  underpredicting concentrations at the lowest



dose by a  factor of 3.2 and overpredicting concentrations  at the highest dose by



a factor of 2.  The data  of  Abraham  et  al.  (1988)  and Tritscher et al. (1992)



were not available at  the time this  model  was  developed,  but  at least for the



data of Tritscher et  al.  (1992),  this model  has been  shown to overpredict the



tissue concentrations  (Kohn et al.,  1992).



     As mentioned earlier, the default position of  the EPA in estimating risks



from exposure to xenobiotics involves  the  use of a model  which produces risk



which  is proportionate to dose  for  low  doses  (low-dose  linearity).   Thus, in



discussing the models and submodels which form a basis for a mechanistic model



for  TCDD,   we  will   focus   on  aspects  of  the model which could  lead  to



nonproportional response for  low environmental doses.  The model of Leung et al.



(19??)  predicts slight  nonlinearity  between  administered   dose  and  tissue



concentration in the experimental  dose range.   In the low-dose range, the model



predicts a  linear  relationship  between  dose and concentration.   [They argue,



however, that tissue dose alone  should not  be used for risk assessment for TCDD



due to the large species specificity in the ability of TCDD to elicit toxicity



(Andersen,  1987)].   They instead suggest  that use of time-weighted receptor



occupancy linked with a two-stage model of carcinogenesis as a better approach



to risk estimation.   The  time-weighted  receptor occupancy predictions derived



from the Leung et al.  (1990a) model are  linear  in the low-dose region, reaching



saturation  in the  range of  high  doses  used to assess the toxicity  of TCDD.
                                     8-29                             08/27/92

-------
                         DRAFT—DO NOT QUOTE OR CITE






Application  of this  proposed  effective dose  with  the two-stage  model  of



carcinogenesis is presented later in this chapter.



     Looking at one  small  aspect of modeling  TCDD's  effects, Portier  et  al.



(1992) examined the relationship between tissue  concentration and the modulation



of the three liver proteins by TCDD in intact female Sprague-Dawley rats.   The



proteins studied included the induction of two hepatic cytochrome P-450 isozymes,



CYP1A1 and CYF1A2, and the reduction in maximal binding to the EGF receptor in



the hepatic plasma membrane.  The modulation of  these proteins  is believed to be



mediated through  TCDD binding to the Ah receptor.  Then, as described in earlier



chapters,  through a  series  of  alterations in  the receptor-dioxin  complex,



transport to the  nucleus, binding to transcriptionally active recognition sites



on DMA, activation of gene  transcription  and alterations  in gene mRNA products,



CYP1A1 and CYP1A2  are induced.  Reduction in maximal binding to the EGF receptor



requires additional protein interactions.



     General empirical models  have  been developed for the regulation of  gene



expression  (Hargrove et al.,  1990).   This modeling  approach  includes  mRNA



production  by  a  zero-order process  and  first-order degradation.   Activation



alters one or both of these rates.   The production of protein is assumed to be



directly related to mRNA concentration.  A more specific pharmacodynamic model



has  been  described  to  account for  the  induction of  TAT  activity  by  the



corticosteroid prednisolone (Nichols et al.,  1989).  In this induction model, the



input  prednisolone concentration  is  specified by the measured  time course of



prednisolone  in  plasma.   Prednisolone  binding  to receptor  is  specified by



association and dissociation rate constants.  The prednisolone receptor binds DNA



with  a specified  association rate  constant  and the bound receptor recycles to



cytosol  with a transport  time,  T  (effective compartment transport  times are



included to  account  for delays between interaction with DNA and the appearance



of TAT activity).  A power function can describe a nonlinear relationship between



the concentration of prednisolone receptor  and the production rate of protein.



The actions of prednisolone and maintenance  of its tissue concentration are much
                                     8-30                             08/27/92

-------
                          DRAFT—DO NOT QUOTE OR CITE

more short-lived than those of dioxin and the modeling period of interest is only
on the order of several hours to a day instead of days, weeks or months as with
dioxin.
     The important  relationships presented here are the association of dioxin
with the Ah receptor and the association of the dioxin receptor complex with DNA.
As described above, Leung et al. (1988) modelled the induction of CYP1A2 as due
to a basal amount of protein plus an additional amount of protein resulting from
binding of  TCDD to  the Ah  receptor.  The extent of  induction was calculated as
instantaneously related to percent occupancy of the Ah receptor via a Michaelis-
Menten type relationship.  Changes in  CYPIA1 and EGF receptor binding were not
modeled by  Leung et al.  (1990a).
     Portier et al.  (1992) modeled the rate  limiting step in the induction of
CYP1A1  and CYP1A2  following  exposure to TCDD using a  Hill equation.   Hill
equations  are  commonly used for modeling ligand-receptor  binding  data.  This
equation  allows  for  both  linear and  nonlinear response  below the  maximal
induction range.  A complete  discussion of Hill kinetics  and other models for
ligand-receptor binding is given by Boeynaems and Dumont (1980??); examples of
the use  of Hill kinetics  for ligand receptor  binding  include  the muscarinic
acetylcholine receptors (Hulme et al., 1981??), nicotinic acetylcholine receptors
(Colquhoun,  1979??),  opiate  receptors  (Blume,  1981)??  and  the Ah receptor
(Gasewicz,  1984).   As a  direct comparison  to  what was  done by Leung et al.
(1990a), it is  interesting to note that the  Hill model  can be thought of as a
very general kinetic model which includes standard Michaelis-Menten kinetics when
the Hill exponent is 1.  Portier et al.  (1992) modeled the reduction  in maximal
binding to  the EGF receptor  also as  following Hill kinetics, but  with TCDD
reducing the binding  from  the  maximum  level  when no TCDD  is present.  For all
three proteins, proteolysis was assumed to follow Michaelis-Menten kinetics.  The
proposed models fit the data in the observable response range.
     The major purpose of the paper by Portier et al. (1992)  was to emphasize the
importance  of  endogenous protein expression  on  the  shape of  the  tissue
concentration/response curve.   For each protein, they considered two separate
models.  In the first, the additional expression of protein induced by TCDD is

                                     8-31                             OB/27/92

-------
                         DRAFT—DO NOT QUOTE OR CITE






independent of the basal level expression.  Such a mechanism is similar to that



used by  Leung et  al.   Under this model protein expression  is given by  the



equation:

where P is the concentration of protein  in  the  liver,  Bp  is  the  basal rate of




production of protein,  Vm is the maximal  level of induction of protein by TCDD,




Kd is the apparent dissociation  constant for binding (in the rate-limiting step),




C  is  the  concentration  of  TCDD  in  the tissue,  V_  is the  maximal  rate  of




proteolysis,  K_ is the proteolysis  rate  constant and n is the Hill exponent.




When the Hill exponent is estimated to be an integer, the estimate of n can be



interpreted as corresponding to the effective number of binding sites which must



be occupied for the effect  of  the binding reaction to  be  expressed.  When the



Hill exponent is not  an integer, no  real  molecular meaning can be attributed to



the equation  and the  model becomes  phenomenological  (Boeynaems  and Dumont,



1980??).



     The second model they considered was one in which the basal  expression of



these proteins was  due to  an  endogenous ligand which competed  with TCDD for



binding sites.  This leads to equations of the form:
                     8P _  .„,-.-,      ,D.     (equation!)
where E refers  to  the  concentration  of the endogenous ligand in units of TCDD



binding-affinity equivalents.  Under  steady-state conditions, equations  (2) and



 (3) are simplified  (Portier et al.,  1992).



     Using these simpler formulas, they see virtually no difference between the



 independent  and additive  models  in the observable response range, even to the



point of getting almost equal Hill coefficients in the two models  for all three




                                     8-32                             08/27/92

-------
                         DRAFT—DO NOT QUOTE OR CITE

proteins.  In the low-dose range where risk extrapolation would occur, the models
differed depending upon the value of the Hill coefficient.
     In all cases, the additive model resulted in low-dose linearity.   This is
expected, since, under the additive model, each additional molecule of TCDD adds
more  ligand  to  the  pool   available  for  binding  and  thus  increases  the
concentration of protein.  Similar observations have been made with regards to
statistical  (Hoel,  19??) and mechanistic  (Portier,  1986??) models  for tumor
incidence.  For CYP1A1, the Hill exponent was estimated to be "2.  When the Hill
exponent is >1, the independent model yields a nonlinear dose-response which is
concave  (threshold looking).   For CYP1A2,  the Hill exponent was estimated to be
-0.5.   When the Hill  exponent is estimated  to be <1,  dose-response  is again
nonlinear,  but in  this case  it is  convex,   indicating  greater than  linear
increases  in response  for low doses.   Finally,  for  the EGF receptor,  the Hill
exponent was approximately 1 in which case the two models are identical.
     Thus, even though these two  basic  models show almost identical response in
the observable response region, their low-dose behavior is remarkably different.
If either CYP1A1 or CYP1A2 levels had been used as dose  surrogates for low-dose
risk estimation,  the  choice  of the  independent or  additive model would make a
difference  of  several orders of  magnitude  in the risk estimates  for humans.
Using CYP1A1 as a dose surrogate,  the independent model would predict much lower
risk estimates  than the  additive  model.  For CYP1A2, the opposite occurs.  For
EGF receptor, there would be no difference. We do not propose to directly employ
the models  of  Portier  et al. (1992) for risk estimation.
     Andersen et al. (1992) modified the model of Leung et al., (19??) to include
Hill  kinetics  in the  induction  of CYP1A1 and  CYP1A2 and to use  a diffusion
limited  approach  to the development  of a PB-Pk model as compared to the blood
flow  limited  approach  used  by Leung  et al.   Diffusion'  limited  modeling is
preferred when diffusion into a tissue is less rapid than blood flow to a tissue.
In the model used by Andersen et al. (1992) each tissue has  two subcompartments,
the tissue blood  compartment and the tissue  itself.   Free TCDD flows into the
tissue blood compartment and, from there,  diffuses into the tissue.   There is no
direct   relationship  between  effluent  venous   concentrations  and  tissue

                                     8-33                             08/27/92

-------
                         DRAFT—DO NOT QUOTE OR CITE






concentration in this  diffusion limited model.  For TCDD, the diffusion limited



approach is preferred  due to the  compounds potentially slow diffusion into the



liver from blood (Kohn et al.,  1992).



     Binding of TCDD to the Ah receptor was modeled in  a  fashion  identical to



that used by Leung et al. (1990a)  The concentration  of CYP1A2 was  modeled as



before using a steady-state model,  in this case,  with  Hill kinetics  instead of



a Michaelis-Menten model.  The resulting  equation is identical to  that used by



Portier et al. (1992)  for the independent induction of CYP1A2 except that they



related this to the concentration of Ah-receptor/TCDD  complex rather than the



concentration of TCDD  in the liver.  Since they assume binding of TCDD to the Ah-



receptor follows Hill kinetics with a Hill coefficient  of 1 (Michaelis-Menten



kinetics),  the  model  of  Andersen et  al. (1992)  approaches the  independent



induction model of Portier et al.  (19??)  for  low doses.



     The  induction  of CYP1A1 was modeled as a  time  dependent process  as in



equation (2), again utilizing TCDD bound  to the Ah receptor rather than tissue



concentration of TCDD.   Most of the  physiological constant, and many  of the



pharmacological and biochemical constants used in the Leung et al. (1990a) model



were changed for the Andersen et al.  (1992) model to correspond to Wistar rats.



The parameters in the  model  were optimized to reproduce tissue distribution and



CYPlAl-dependent enzyme activity  in a  study by Abraham et  al.  (1988), and liver



and  fat  concentrations  in  a  study by Krowke et  al.  (1989).  For  the longer



exposure regimens and observation  periods,  changes  in total body weight and the



proportion  of weight  as  fat compartment volume  were  included via piecewise



constant values  (changes occurred at 840 hours and 1340 hours).



     Andersen et al.  (19??)  noted that the liver/fat concentration ratio changes



as dose changes due to an increase in the amount of binding protein  in the liver.



For high doses in chronic exposure studies, this  introduces a nonlinearity into



the concentration of TCDD in the liver.  In the low-dose region, because the Hill



coefficient  for CYP1A2  and for  the  Ah  receptor  are  equal to  1,  the liver



concentration  as  a function of dose is  still  effectively linear  (i.e., small



doses  of TCDD  will  bind • to the  Ah receptor  increasing  the amount  of Ah-



receptor/TCDD complex which then induces additional production of CYP1A2 which







                                     8-34                             08/27/92

-------
                         DRAFT—DO NOT QUOTE OR CITE






can bind to free dioxin).   In the observable response range,  there is a slight



nonlinearity in the concentration of  TCDD  in the liver as a  function of  dose



under chronic exposure (Andersen et  al.,  1992).   This nonlinearity in the  dose



region of  1-100  ng/kg/day  does  not agree  with  the  findings of Kociba  et al.



(1976) and Tritscher et al.  (1992) for chronic exposure in Sprague-Dawley rats.



The plateau in total liver concentration predicted by the model of Andersen et



al. (10??) does occur in the data of Kociba et al. (19??) and Tritscher et al.



(19??) but in the  range  of  100  ng/kg/day rather  than the  10,000  ng/kg/day



predicted by Andersen et al.  However, the changes in liver/fat ratio observed



by Andersen et al.  (19??)  and supported by  human evidence  (Carrier, 1991) are a



necessary part of the modeling for TCDD.



     Finally, with regards to risk estimation, Andersen et al. (19??) compared



the induction of CYP1A1 and CYP1A2,  the concentration of  free TCDD in the liver



and the total concentration of TCDD in the liver to tumor incidence (Kociba et



al., 1976) and the volume of altered hepatic foci (Pitot et al., 1987).  Using



a  biweekly dosing regimen  assuming  gavage ???  exposure for  ???  weeks,  they



integrated the concentration of TCDD in the  liver over time and the concentration



of induced protein over time to get summary measures of internal exposure.  They



concluded that tumor promotion correlated more closely with predicted  induction



of CYP1A1  than the other integrated quantities.  No  formal measure was used to



support this observation.   It has been shown for the tumor incidence  data that



it  is difficult  to determine nonlinearity  or  linearity  although it  is not



inconsistent with  a linear  response  in  the low dose region  (Portier et al.,



1984).  In general, one cannot expect a  linear correlation between appropriate



dose  measures  and toxic response.  In addition,  the choice of an independent



induction model for CYP1A1 and a Hill coefficient >1 leads to nonlinear low-dose



behavior.   If  the promotional effects of TCDD follow a similar mechanism, the



risk  from  exposure at low doses  will be  negligible.   For risk  assessment,  it is



important  to know  if an additive model also fits this data and agrees with the



promotional  effects of TCDD  since  such  a model  will  have different low-dose



behavior than  the  independent model.
                                      8-35                             08/27/92

-------
                         DRAFT—DO NOT QUOTE OR CITE






     Kohn et al.  (1992) expanded upon  the model of Leung et al. (19??) to include



Hill kinetics, a restricted flow-limited PB-Pk formulation and an extensive model



of the biochemistry of TCDD in the liver.  The goal of the model was to explain



TCDD-mediated  alterations  in  hepatic   proteins  in  the   rat,   specifically



considering CYP1A1, CYP1A2, Ah receptor,  EGF receptor  and estrogen receptor over



a wide dose range.   In addition,  the model describes the distribution of TCDD to



the various tissues, accounting for both time and dose effects observed by other



researchers.  The PB-Pk models developed  by  Leung et al. (19??)  and Andersen et



al. (19??) relied on several single dose data sets (Rose et al.,  1976; Abraham



et al.,  1988)  and were validated against dosimetry  results from  longer term



subchronic and chronic  dosing regiments  (Kociba  et al., 1976,  1978;  Krowke et



al., 1989).  More  recent studies in  which female Sprague-Dawley  rats received



TCDD  (Tritscher  et al., 1992;  Sewall et  al.,  1992)  were  used by  Kohn et al.



(1992) to model the pharmacokinetics  and  induction of  gene products in this sex



and species.  Among the data reported by Tritscher et al.  (1992)  and Sewall et



al.  (1992)  were  concentrations  of  TCDD  in  blood and liver,  concentrations of



hepatic CYP1A1 and CYP1A2  and EGF receptor  on  the hepatocyte plasma membrane.



Kohn et al. (1992)  refer to their model as the Kohn model.  The tissue dosimetry



for the Kohn model were validated against  the  single dose  and chronic dosing



regimen experiments employed by Leung et al. (1990a) and Andersen et al.  (1992).



     In the biochemical effects portion of  the Kohn model,  the binding of TCDD



to  the  Ah  receptor  is  modeled using  explicit  rate constants  instead  of



dissociation equilibrium   constants  (equation 2 with  B  = 0).  However, larger




dissociation rates (kd, k;) were used leading to a formulation of the amount of




TCDD-Ah-receptor complex similar to that used by Leung et al.  (19??)  and Andersen



et  al.  (1992)   Many of the other binding reactions  in the model were handled



similarly  (e.g., TCDD binding to CYP1A2 and TCDD bound to blood).  This is simply



a numerical trick to avoid the necessity  of solving for the concentration of TCDD



in the liver using the mass conservation relationship described in  Leung et al.



(1990a).



     The physiology described in the Kohn model  is dependent upon the body weight



of the animal.  Body weight changes as a function of  dose and age were  recorded






                                     8-36                              08/27/92

-------
                          DRAFT—DO NOT QUOTE OR CITE






by Tritscher et al.  (19??) and directly incorporated into the model via a smooth



function.  Tissue volumes and flows were calculated as a allometric formula based



upon recent work by Delp et al.  (1991).   To allow the model to fit the data of



both Rose et al.  (1976) and  Tritscher  (1992),  the  Kohn  model  includes loss of



TCDD from the liver  by  lysis of dead cells where the rate of cell death (and the



resulting  lysis)  was  assumed  to  increase as a  hyperbolic  function of  the



cumulative exposure in the liver to unbound TCDD.  No information regarding the



rate of TCDD from lysed cells is available, therefore, this feature of the Kohn



model represents a net contribution of TCDD clearance by TCDD-induced cell death.



     In the biochemical effects portion of  the Kohn model, the Ah-receptor/TCDD



complex  up-regulates  four  proteins;  CYP1A1,  CYP1A2,   the  Ah  receptor  and



transforming growth factor-a  (see  Fig. 1).  For  all  four  proteins, synthesis and



degradation rates are defined explicitly.   Changes  in CYP1A1, CYP1A2 and the Ah



receptor are compared to data on these concentrations.  The induction of TGF-a



is deduced from observations on human keratinocytes (Choi et al., 1991; Gaido et



al.,  1992)  and is  quantified  based on  an assumed  interaction  with the  EGF



receptor.   However,  TCDD-mediated  induction  of  TGFa  has  not  been  clearly



demonstrated in liver  (see Appendix).   Constitutive rates of expression  for



CYP1A2, Ah receptor  and EGF receptor are assumed independent  (equation 2) at the



induced expression.   This has no effect on low-dose  rate  extrapolation since the



Hill coefficients for  the induction of these proteins by the Ah-receptor/TCDD



complex were estimated  to be 1.  Induction of CYP1A1 was assumed to be based upon



additive induction (equation  3), but again the Hill  exponent was estimated to be



1 leading to low-dose  linearity under  either model (2  or 3).  Thus,  the Kohn



model  found  that  the induction of all gene  products appear to be  hyperbolic



functions of dose without any apparent cooperativity (i.e., the value of the Hill



exponent, n, in equation 2,  is estimated  to be 1).   The discrepancy  in  the



estimates of the Hill exponents between this model and the other models discussed



(Portier et al., 1992;  Andersen et al.,  1992;  Kedderis et al., 1992) is probably



related to the inclusion of induction of  the  Ah receptor in the Kohn model.



     In the Kohn model, the Ah-receptor/TCDD complex down-regulates the estrogen



receptor.    It  is  assumed  that  the   estrogen   receptor-estrogen  complex







                                     8-37                             08/27/92

-------
                         DRAFT—DO NOT QUOTE OR CITE






synergistically reacts with the Ah-receptor/TCDD complex to  transcriptionally



activate TGF-a.   This  synergy  was  introduced  to  partially  account for  the



observation of reduced TCDD promoting potency in males and ovariectomized females



as compared to  female  rats (Lucier et al.,  1991).   This mechanism,  although



supported by some data (Clark et al., 1991; Sunahara et al., 1989) is speculative



(Kohn et al. (19??)  (see Appendix).



     There are  basically  three  levels of complexity of  Pb-Pk models  for  the



effects of TCDD.  First is the traditional PB-Pk model by Leung  et al.  (1988)



with the added complexity of protein binding in the liver.   The  next level of



complexity  is  the model  by Andersen  et al. (1992)  using  diffusion  limited



modeling and more detailed modulation of  liver proteins.  Finally, there is the



model of Kohn et al. (1992) with extensive liver biochemistry.  All three models



have biological structure and encode hypotheses about the modulation of liver



proteins by TCDD.  However, for gene expression, all  three models fall in between



curve fitting and mechanistic modeling.  In their derivation, the parameters were



estimated using dose-time-response data  for  protein concentrations and enzyme



activity which  are a direct consequence  of gene expression.   This constitutes



curve fitting at this level.  However, the structure of the  models is derived



from qualitative information on the effects of TCDD,  the PB-Pk model and even the



biochemical model of Kohn et al. (1992),  for protein concentrations using data



sets which were not included in the original derivation of the model and which



were derived from  designs other than those  used to characterize the original



model.   This constitutes  a mechanistic  validation of the original models and



places  these  exercises  in the  realm of  partial  curve  fitting  and  partial



mechanistic modeling.



     In  terms of  low-dose risk  estimation, all  three  models have limitations.



The Leung et al. (1988) model fails to reproduce the tissue concentration data



from Kociba et al. (1976)  and Tritscher et al.  (1992).  This is probably due to



the high concentration of liver binding protein (CYP1A2) predicted by this model.



     The Andersen et al. (1992)  and Kohn  et al.  (1992) models use  Hill kinetics



to describe at  least some of the binding  reactions.  Hill's equations are based



upon a molecular scheme of  interaction in which it is assumed that there are  n







                                     8-38                             08/27/92

-------
                         DRAFT—DO NOT QUOTE OR CITE






binding sites for the ligand and that the reaction  is stable in only two states:



either  completely  unoccupied  or  fully  occupied.    This  implies  that  the



association process is  a  (n+1) - molecular reaction.  In the application of this



model to experimental concentration curves, very little molecular meaning can be



attributed to the resulting model due to the flexibility of this function with



regards  to  dose-response   shape.     In   addition,  there  is  no  unequivocal



relationship between an  estimated value  of  n  and the existence  of  molecular



interactions between binding sites.   For example,  two  binding sites may exist,



but binding to  one site  produces a small effect,  while binding to both sites



produces a much  greater effect.  This would lead to a noninteger value  for n when



curve fitting.  Considering the importance of the  Hill coefficient in terms of



low-dose extrapolation  (Portier et al., 1992) and considering its limitations in



terms of biological understanding of the sequence  of molecular events involved



in induction (Andersen  et al., 1992), caution must  be used when extrapolating to



tissue dose regions outside of those examined directly  in the experiment.  It is



thus  appropriate to conclude that  independence of the  action of  TCDD would



necessarily imply a nonlinear response at low dose.



     Some of the mechanistic assumptions  in  these models are speculative.  Many



of the binding and induction equations related to  the  Ah-receptor/TCDD complex



are encoded in  equations,  but  their exact nature  and level of  control at the



molecular level are unknown.  This is true of CYP1A1,  CYP1A2,  the Ah-receptor,



the  estrogen  receptor and TGF-a.   Also,  the  reduction  in  EGF  receptor by



internalization described in the model by Kohn et al. (1992) represents just one



mechanism  for  its  depletion.    It  is  also  possible that  the  synthesis or



degradation of  this protein may be  under direct  control of  the  Ah receptor,



although TCDD does not  alter mRNA levels  in either human keratinocytes (Osborne



et al.,  1988) or  mouse liver  (Lin et al.,  1991) and EGF receptor  does seem to



move  from the plasma membrane to  the cell interior  following  TCDD exposure in



female rats (Sewall et al., 1992).



      It  should  be  noted that the  mechanistic models  have  the  advantage of



suggesting  experimental strategies for pursuing  the hypothesis  of  action of



TCDD.   These  models propose specific mechanisms, which can be tested  in the







                                     8-39                             08/27/92

-------
                         DRAFT—DO NOT QUOTE OR CITE


laboratory as a means of validation  of  this model.   For the purposes  of  risk

estimation, one must be careful to recognize that these models do not necessarily

impart added confidence in low-dose risk estimates because many molecular/

biological events that  lead to  toxic responses are  not known.  However,  the

mechanistic assertions put forward by the model can be  discussed in scientific

peer review and,  if found valid,  can provide additional confidence.

8.2.   TOXIC EFFECTS

8.2.1.   Modeling Liver Tumor Response for TCDD

     [This section has  not  been  reviewed by all members of the  Dose Response
     Committee,  so it  does not represent a consensus  section at this time. There
     was considerable  discussion regarding the appropriateness of applying dose
     response models for risk assessment in this chapter.  The decision was made
     that we would  include  progress on the development  of biologically-based
     models  using  liver cancer  in  rats  as the  example.    This  approach  is
     obviously fraught with uncertain assumptions.  Nevertheless, we feel that
     this information will be of considerable use to the peer  review panel in
     their attempts to generate  a risk characterization for dioxin based on an
     evaluation of dose response relationships.]



     Long term carcinogenicity studies in rodents have shown TCDD  to be a potent

carcinogen (Huff,  1991). The most seriously affected organ has been  the liver in

female rodents.   The lack of any detectable  DNA binding and the failure of TCDD

to produce positive findings in a battery of short term tests for genotoxicity

compels one to conclude that TCDD does not possess direct genotoxic activity.

It can only  be concluded that the  mechanism of  action  of TCDD  is by secondary

effects.  One possible mechanism of hepatocarcinogenicity from exposure to TCDD

is  via  the promotion of  previously incurred genetic  damage.    This  has been

illustrated  in several  initation-promotion studies in rat livers (Pitot, 1987;

Clark, 1991a).  There is definitely a role of estrogen in the promotional effects

of TCDD in rat liver since TCDD enhances  hepatocyte proliferation  and stimulates

the development of enzyme-altered hyperplastic foci in intact female rats but not

in  ovariectomized  rats (Lucier,  1991).   The conclusion  resulting from these

experiments  are  not definitive, but it has been suggested that  the  type of

promotional effect resulting from TCDD exposure is receptor-mediated mitogenesis.

There is also the possibility that additional CYPIA2 mediated metabolism could

lead to an increase in the activation of estrogens leading to cell damage (e.g.,

via free oxygen radicals).


                                     8-40                             08/27/92

-------
                         DRAFT—DO NOT QUOTE OR CITE






     This section will discuss modeling the tumor  response observed in rodents



and comparing  that modeled  response to  some  of  the  biomarkers of  exposure



discussed in the section on the biochemistry of TCDD.  This approach will deviate



from the pure mechanistic modelling outline discussed in the introduction.  Due



to limitations  in the data  available for characterizing  the models we will



employ, some of the parameters used in this modeling exercise had to be obtained



directly from the tumor incidence data.   These parameters  are then compared to



the relative changes we would have expected using  the  biomarkers of effect and



exposure which seem reasonable for this endpoint.  Thus, this exercise falls in



between curve fitting and pure mechanistic modeling.  At the end of this section,



we will discuss the data needed to move  this approach  into a full mechanistic



development and discuss  tumor responses  in other sexes and  species  at  other



sites.



     The carcinogenicity data we will use are from a 2-year  feeding  study in male



and female Sprague-Dawley rats (Kociba et al., 1978).  For female rats, the study



used 86  animals  in the control  group and 50  animals  per group  in  the  three



treated groups given doses of 1,  10,  100 ng/kg/day.  The original pathology of



the study recorded significant, dose-related increases  in tumor incidence  in the



lung,  nasal turbinates, hard  palate and liver.   The original liver pathology has



been reviewed several  times, most recently by a  group convened  by  the  Maine



Scientific Advisory Panel (PWG, 1990).  The data we will concentrate on is this



analysis is the incidence of  liver adenomas and  carcinomas  (combined) based upon



the most recent pathology  review.    A summary of these data  are  presented in



Table 8-3.



     There was a  substantial reduction  in survival  in  all experimental groups



(including controls) during the course of the study.   Other studies have shown



that correcting for this drop can result  in as much as  a  2-fold change in the



low-dose risk estimates ( ).   A simple correction  for  survival differences ( )



was applied to  these data to  present  the  risk summaries given in Table 8-3.  In



the analysis which follows, a more rigorous statistical approach was employed.



8.2.2.   Tumor Incidence.    In recent  years,  there  has been a resurgence in



interest in refining the mechanistic representation  of mathematical models of







                                     8-41                             OB/27/92

-------
                        DRAFT—DO  NOT  QUOTE  OR  CITE
TABLE 8-3
Administered Dose, Tumor Response and Number at Risk of Hepatocellular
Neoplasms in Male Sprague-Dawley Rats From Carcinogenicity Experiments of
Kociba et. al, 1978, Using the Pathology Review of Sauer (PWG, 1990)
Original Dose (ng/kg/day)
Number with Neoplasm
Number on Study
Survival-adjusted number at risk8
Lifetime Tumor Risk
0.0
2
86
57
0.035
1
1
50
34
0.029
10
9
50
27
0.333
100
18
50
31
0.581
aUsing the "poly-3"  survival adjustment suggested by Portier and Bailer
 (19??)
                                    8-42
08/27/92

-------
                          DRAFT—DO NOT QUOTE OR CITE


carcinogenesis.  With few exceptions, the mathematic modeling of carcinogenesis

at the  cellular level has  concentrated  on the use  of the  multistage model.

Theoretical discussions on these models began in the mid-20th  century  (Arley and

Iverson, 1952; Fisher and Holloman, 1951; Nordling, 1953).  The first practical

application of models from this class was done by Armitage  and Doll  (1954).  One

major failure of the Armitage-Doll model is  a lack of growth kinetics of the cell

populations  {Armitage  and Doll,  1957;  Neyman and  Scott,  1967;  Moolgavkar and

Venzon, 1979).  Several researchers proposed a second model, deemed the two-stage

model which is  illustrated in Figure 8-6.

     The two-stage  models assumes that  carcinogenesis is the result  of two

separate mutations,  the first resulting in an intermediate cell population and

the  second resulting  in  malignancy.    Cells  in  the  normal  and intermediate

populations are allowed to expand  in number via replication or reduce in number

due to  death or differentiation.   There  is  considerable  confusion  as to the

underlying mathematics in this model since several  groups have proposed the same

model but  used  different  mathematical  developments to predict tumor incidence

from this model (Armitage  and Doll, 1957; Neyman and Scott,  1967; Moolgavkar and

Venzon, 1979; Greenfield et al., 1984??).  In the application of the two-stage

model which follows, the mathematical development of this model by Moolgavkar and

Venzon  (1979) and subsequent development of this  model, will be used.

     The two-stage  model  (Figure 8-6)  has  six   basic rates  which  must  be

estimated.   These are:



         1.  PN    = birth rate for cells  in the normal state.

         2.  6N    = death/differentiation rate for cells  in the normal state.

         3.  JJN_J   = rate  at  which  mutations  occur  adding  cells  to  the
                     intermediate state.

         4.  Pj    = birth rate for cells  in the intermediate state.

         5.  6j    = death rate for cells  in the intermediate state.

         6*  ^I-M   s rate at which mutations occur adding cells to the malignant
                     state.
                                     8-43                             08/27/92

-------
                DRAFT—DO NOT QUOTE OR CITE
ft.
    	  .  	  A
a
                          C*H*
                                                 •*   cuft
                           FIGURE 8-6



               A Two-Stage Model  of CarcinogenesiB







                              8-44
                                   08/27/92

-------
                          DRAFT—DO NOT QUOTE OR CITE

     To apply this model to dioxin,  or any other chemical carcinogen,  requires
estimates of  these rates as they  change over dose  and  time.   A mechanistic
approach to this would be to incorporate some of the relative changes in proteins
seen in the Kohn et al.  model directly into the two stage model as rate changes
in these parameters.  Considering the complexity and novelty of this approach,
it is left as a research topic.   Instead, we will apply this model directly to
tumor incidence data,  focal lesion  data and cell  labeling data  comparing the
resulting parameter estimates to predicted dose-surrogates  from the models of
Leung et al.  (1988), Andersen et al.  (1992)  and Kohn et al.  (1992).
     This is not the first application of TCDD data to the two-stage model.  An
application of this model to TCDD was presented by Thorslund (1987).  Thorslund
treated the effects of  TCDD as  a direct  promoter  having  an  effect only on the
birth rate  of intermediate  cells  (fJj) in the two-stage model.  The number of
normal cells were  assumed constant (this  is equivalent to setting PN  =  N ~ ^
in the model  in  Figure 8-6).   Two parametric models of  the  change in Pj as a
function of dose were used, one model having a single parameter (a first-order
kinetic or exponential model)  and the second based upon two parameters (a log-
logistic  model).    The  parameters  in  the  exponential  two-stage model  were
estimated from the tumor incidence data of Kociba et al. (1978) and validated by
goodness-of-fit, cell labeling data and species/sex/strain extrapolations.  The
slope parameter in the  log-logistic two-stage model chosen to be 1,  2 or 3 based
upon slopes observed in other biological systems.   The remaining parameters in
this model were estimated from the Kociba et al. (1978) data.
     The liver tumor response from the Kociba et al. (1978)  study are given in
Table 8-3 using  the most recent pathology review of the  liver sections (PWG,
1991).  Shown are  the  number of animals  with  tumor (row  2),  number of animals
placed on study (row 3), a survival-adjusted (Portier and Bailer,  19??) number
of animals at  risk (row  4), and  the survival-adjusted lifetime tumor probability
(row 5 which equals the entry in row 2 divided by the entry in row 4).
     The data is more complex than appears in Table 8-3 since death times were
recorded.   Since  there are no interim  sacrifices which would  allow  for an
estimation of tumor-induced mortality (Portier, 1986),  the two-stage model was

                                     8-45                             OB/27/92

-------
                         DRAFT—DO NOT QUOTE OR CITE


fit to this data using the assumption of  incidental tumors  (Hoel  and Walbury,

1972??; Dinse and Lagakos, 19??).   Since liver tumors induce  little,  if  any,

mortality in male and female Fischer 344  rats  and B6C3F mice  (Portier et  al.,

1986; Portier and Bailer, 1989),  this assumption seems to  be warranted.

     There are a variety of  mathematical formulations which could be used to

derive a tumor incidence  rate under the two-stage model.  As has been assumed by

other authors (	),  we assume:



         1.  all cells act independently of all other cells;

         2.  the rates in the two-stage model (Figure 8-6)  are constant over the
             life span of the animal; and

         3.  the tumor incidence rate corresponds to the rate  of appearance of
             the first malignant cells.



All three of these assumptions  are likely  to be violated in the case of dioxin.

In most tissues,  there is a homeostatic feedback system to  control the number of

cells in the tissue.  No such system can be assumed here since it results in a

mathematic formulation which  is intractable.  For the large pool of normal cells

in the liver, this  is unlikely to have  an effect, but for the small number of

intermediate cells  (at least  for early times) this could have a small effect on

tumor  incidence.    This  issue  cannot  be  resolved  without further  research.

Assumption (2) is clearly violated based upon the behavior of the PB-Pk models

presented  earlier.    Time-dependent  changes   in  tumor  incidence   could  be

incorporated  into  the modeling  and will  be  at a  later  time.   Finally,  the

kinetics of cell growth  for  malignant  clones  has been studied (Moolgavkar and

Lubeck, 1992) and the assumption (3) was found  to have a moderate impact on the

tumor incidence rates.   This also will be  investigated at a later time.

     The exact tumor  incidence  rate was used  to avoid potential bias from the

routinely  used approximation  (Kopp  and  Portier,  1989).  Thorslund (1987) used

this approximation  in his analysis.  The methods outlined by Portier  and Kopp-

Schneider  (1991)  employing the Kolmogorov backwards equations were used to derive

the exact  tumor  incidence rate.
                                     8-46                             08/27/92

-------
                          DRAFT—DO NOT QUOTE OR CITE






     Under these assumptions,  when fitting the two-stage model to the Kociba et



al. (1976) data, there  is the potential to estimate as many as 24 parameters



(four dose groups, each with its  own two-stage model having six parameters).  The



data given in this study  make it  impractical to estimate  this many parameters



with any degree of accuracy (Kopp-Schneider and Portier,  1992).  To reduce the



number of parameters in the model, P^ and 6^ were assumed to be known without




error.  Assuming that, on  average and over finite time,  the population of normal



cells is effectively constant,  it is reasonable to  assume that PN = 6N.  This




assumption may not hold true for TCDD since the labeling index for normal cells



seems to  increase with  increasing exposure to  TCDD  (Lucier  et  al.   (19??).



However, to illustrate the use of the two-stage model,  this assumption will be



employed.   Labeling data (	) suggest that normal  hepatocytes undergo mitosis



at an average rate of one  mitotic  event per 300 days (a 2% labeling index for a



6 day labeling experiment) suggesting that (5N = 3.333xlO~3.  This value was used




in the  analysis which follows.   Other  values for  PN  =  6N  was  specified and




assuming changes in P^ do  not  imply changes in p^-i (Portier and Kopp-Schneider,




1991).  The number of normal hepatocytes was assumed to be 6xl08 for the female




rat.



     This leaves 16 parameters  to be estimated from the  tumor incidence data.



Table 8-4  shows one  set  of parameter  estimates  resulting  from  this  fitting



exercise (labeled "Unrestricted" under the model heading).  Multiple attempts at



fitting this model to the  tumor  incidence data indicated that the model was not



identifiable; this means that there are an infinite number of parameter estimates



which give effectively the same  answer for this data. The main problem seems to



be estimating pN.j and ^.M simultaneously  (for example,  in the high-dose group,




any model with ^N.j •  pj.M  eg^ial to 3.4 •  10"15,  seemed to yield equivalent fits)




and with fitting P] and 6j separately (no obvious relationship was observed).  The




main reason  for including the "Unrestricted"  model in Table  8-4  is that this



model represents the  best  fit we can possibly achieve and the likelihood  (Column



6) represents a best possible measure of goodness-of-fit.   We will compare the



likelihood for other models with this target  likelihood.






                                     8-47                             08/27/92

-------
oo
•c*
03
O
00
TABLE 8-4
Fitting the Two-Stage (TS) Model to the Tumor Incidence Data of Kociba et al. (1976) Using the Pathology of Sauer et al. (1991)
Model
Unrestricted
Unrestricted fifg.|
Unrestricted MI-M
Unrestricted B(
Unrestricted 6,
(B| in low-dose group
set to zero)
Dose
(ng/kg/day)
0
1
10
100
0
1
10
100
0
1
10
100
0
1
10
100
0
1
10
100
Parameters
"N-l
I.HxIO-10
9.48x10~10
9.85x10'10
3.33x10'5
2.55x10'8
1.37x10'8
1.36x10"7
5.80x10"7
1/31x10'5
-
-
-
2.56x10'7
-
-
-
2.56x10'7
_
-
-
Bl
6.90x10~2
4.75x10'2
5.27x10'2
2.01x10'2
1.75x10'2
-
-
-
2.86x10'2
-
-
-
1.24x10'2
5. 82x1 0'3
2.15x10"2
2.46x10'2
6.70x10"3
0.0
1.59x10'2
1.90x10"2
«l
3.Kx10'2
1.58x10'2
2.44x10'2
1.61x10"2
1.06x10'2
-
-
-
2.16x10'2
-
-
-
1.79x10'2
-
-
-
1.23x10'2
—
-
-
''l-M
1.20x10"12
9/2-x!-'15
1.12x10'12
1.03x10'10
1.31x10"9
-
-
-
2.46x10'12
1.37x10'12
1. 53x10"' '
5.81x10'11
3. 02x1 0'9



3. 02x1 0'9
-
-
-
Likelihood
-58.61
-64.96
64.96
-67.57
67.57
vO
to

-------
                          DRAFT—DO  NOT  QUOTE  OR  CITE

     The first restricted model  to be considered  is  a model  in which the effect
of TCDD is only on the mutation rate from normal cells to initiated cells in the
two-stage model.  This  model is fit  by forcing  Pj, 6j and  p/j_M  to be constant
across all dose groups  and allowing pN.j to vary  freely  as  dose changes.   The
parameter estimates for this model are  given in Table 8-4 as model (2).  There
are nine less  parameters in this model than in the unrestricted model.  Comparing
the two likelihoods, it is evident that this model  fits the data  significantly
worse than does the  unrestricted model (X29 = 12.71, p<0.01, as a  technical note,
since there is  a problem  with identifiability, the degrees-of-freedom, 9, for
this X2  random variable is  inflated  which  would inflate the p-value and the
significance of the result would remain).
     The  problem with  identifiability does  not  abate  by restricting  the
parameters in the model.  This is illustrated by the  next model (3) in Table 8-4.
In this  model,  it  is  assumed the effect of TCDD  is restricted  to the rate  of
mutation from the initiated state to the malignant state.  In this model,  P], 6j
and pN_j  are held constant  over all  groups  and  /JJ.M changes  as  dose  changes.
Three points  indicate the problem of nonidentiflability with this model.  The
first  is that the  likelihood for this  model (3)  and  the  model changing pN.j
(2) are  identical.  This  indicates  that the two models explain  the same amount
of noise in the data.  The  second point is that,  even thought the magnitude  of
the birth rate Pj has changed, the difference  Pj - 8j has not (~0.7xlO~2). Further
modeling with this data  indicates  this to be true over a wide range of Pj values.
The third point is  that,  for each dose  group,  /JN.J • A/J_M is  the  same  in the  two
models.  Again, use of  different  fixed  values for pN.j in model  (3) and /JJ_M  in
model  (2)  support this result.
     The net  result is  that, unlike the results  seen by  Moolgavkar  and Lubeck
(1992) when fitting colon  cancer  data to the  two-stage  model, the magnitude  of
the mutation  rate  pN_j relative to pj.M  cannot be addressed for the TCDD  data.
However,  the  relative change in the  product  of  these two mutation rates as a
function of dose can be studied.  This  will be done later in this document.

                                      8-49                              OB/21/92

-------
                          DRAFT—DO NOT QUOTE OR CITE






     A second restricted model is to consider that dioxin's only effect is on the



birth rate of initiated cells (Pi).  This is in fact the modeling approach used




by Thorslund  (1987).  This model assumes that A/N.J* 5j  and pj.M do not change as




a function of exposure to dioxin.  This model is given as model (4) in Table 8-4.



This model is also inconsistent  with the data (likelihood = 67.57) and provides



a fit which  worse than  the mutational effect model given by (2).   [Technical



note: it is difficult to directly  compare thee two models since they constitute



nonnested models (Becker and Wahrendorf,19??) ].  This implies  that the effect of



TCDD on  tumor incidence  in these rodents  is  more likely due  to  a mutational



effect than a mitogenic effect on initiated cells.  Caution must be taken when



interpreting this result.   Firstly,  no statistical confidence can be placed on



this statement so the observed difference in the two models may  solely be due to



random change.  Secondly, this statement is only justified within the restricted



context of this two-stage  model of carcinogenesis.   If any  assumptions of the



model are incorrect (independent cell action,  constant  rates,  two stages, etc.),



the interpretation based upon this model could be biased.



     The problem with identifiable parameters also remains with this formulation



of the model.   This  is  illustrated  by the fifth model  in Table  8-4.   In this



model, the  birth rate in  the lowest dose  group was  fixed  to be  zero.   The



resulting  likelihood  and  estimates  of p^-i  and PJ.M remained  the  same.   The




estimated birth rates (P| and death rate 6j) changed, but  the  difference between




the two in the various dose groups  did  not change.  Repeated applications of this



formulation  of  the model  confirmed this problem of nonidentifiability in this



case.



     With  this  problem   of  identifiability,  there  are basically  only  two



parameters to be  estimated for the two stage model for each  dose group.  These




are  the  mutation parameter,  given  by fi  - n^_\  '  A'j-M'  anc*  *-ne proliferation




parameter,  given by II  =  Pj  -  6j.    Table  8-5 gives  the estimates  of these




parameters for  the simple  models considered here.  The  first model corresponds



to the  case  where we allow both  the  mutation  parameter and the proliferation



parameter to change  as a  function of  dose  (model  (1)  in  Table 8-4).  For this








                                      8-50                              08/27/92

-------
CD
1
in
M
TABLE 8-5
Net Parameter Estimates From the Two-Stage Model of Carcinogenesis
Two- Stage Parameters Changed
Both
Both
(proliferation rate <0.02)
Nutation rates
Proliferation rates
Doses
Control
1.37x10'22
0.0376
2.80x10"20
0.02
3.34x10'17
-0.0055
1 ng/kg/day
0.077x10'22
0.0317
1.91x10'20
0.0196
1.79x10"17
-0.0121
10 ng/kg/day
8.86x10'22
0.0283
18.79x10'20
0.0196
17.8x10'17
0.00362
100 ng/kg/day
3.43x10'15
0.0040
3.58x10'15
0.00198
76.0x10'17
0.00669
Other Parameters


proliferation rate = 6.90x10
mutation rate = 7.73x10"16
DRAFT — DO NOT
I
8
O
M
H

c
o>
-J


VO

-------
                         DRAFT—DO NOT QUOTE OR CITE

model, as dose increases, the mutation parameter drops to one-half the control
value for the 1 ng/kg/day dose group and increases substantially in the remaining
two dose groups (6.5-fold for 10 ng/kg/day and 25xl06-fold for  100 ng/kg/day).
The hepatocyte replication parameter drops as a function  of  dose.   In the 100
ng/kg/day dose group,  the  replication parameter is small relative  to control
(10-fold smaller)  to adjust for  the much larger mutation parameter.  The reason
for this particular pattern is clear if one studies the  tumor incidence data from
the Kociba et al.  (19??) study.   Over time,  the tumor  incidence in the highest
dose group is larger than the others  (the mutation parameter and the replication
parameter combine to  control the magnitude of  the response) but  climbs  less
steeply with time (this is controlled by  the replication parameter).
     The replication rates in the control group and the two lowest dose groups
are very  high.   Assuming the death rate  is  zero,  this would correspond  to a
labeling index  of  40.9% for a 7-day labeling experiment in the control animals.
If the death rate  is >0,  the labeling  index would be even larger.  Also, if one
cell entered the  initiated state on day  1, by  730 days  (2  years),  one would
expect a  clone of size  8xlOu, "140  times the size of a  normal  rat  liver.   To
control for this problem, the same model was fit to the  data with the replication
parameter constrained to be less than  0.02 (a first day clone would be expected
to have  size  of 2xl06 by  study end).   This resulted in the same  pattern of
mutation parameters  but with a mutation parameter two orders of magnitude larger
in the control and two lowest dose  groups.  For all three groups, the replication
parameter was estimated to  be at  the  boundary  (0.02).   The  parameters for the
high dose group did not change.
     The model  in which the replication parameter is constant  across dose groups
and the mutation parameter changes for each new dose is  given in row 3 ("Mutation
Rate") of Table 8-5.   The  pattern in this case is very  clear  and matches the
observed cumulative tumor incidence will with the mutation rate dropping in the
1 ng/kg/day dose group then rising in  the  remaining groups.  This observed drop
is not statistically significant from the  control mutation parameter (likelihood
ratio test, p>0.05).  The constant replication rate (6.9xlO'3) is reasonable and
is not likely to produce unrealistically  large  clones of initiated cells.   As
                                     8-52                             08/27/92

-------
                          DRAFT—DO NOT QUOTE OR CITE

mentioned before, this model provided a significantly worse fit to the data than
did the unrestricted model.
     Finally, the estimated parameters for the model  in which  the replication
parameter changed with  dose  and the mutation parameter was  constant  over all
doses is given by the last row ("Replication Rates") in Table 8-5.  The change
in replication rates mirrors the tumor response in the same manner as was seen
for the mutation rates.  The replication rates are  of reasonable size and should
not produce impossibly large clones.  As mentioned earlier, this model does not
fit the data as well as the unrestricted model or as well as the model with fixed
replication rate over dose and varying mutation rate over dose.
     There is other evidence which can be used to examine the adequacy of this
model for  tumor incidence from exposure  to  TCDD.  Lucier  and  his colleagues
recently conducted an initiation/promotion study  in female Sprague-Dawley rats
(Tritscher et al., 1992; Kohn et al., 1992; Portier et al., 1992; Sewall et al.,
1992; Maronpot et al.,  1992).  In this study, they measured number and size of
preneoplastic foci in liver sections.  It has been suggested that the cells in
these  lesions  correspond  to  the initiated  cells  in  the two-stage  model of
carcinogenesis.   (reference).   If this  is  true,  it is  possible  to apply the
methods  of  DeWanji  et  al.   (1990)  to this  data   to   analyze  the  growth
characteristics  of these  cells  (Moolgavkar et al., 1991).  However, as before,
we run into the problem of nonidentifiability.  Since  the data in these studies
were collected at only one time  point, it may not be possible to estimate all of
the parameters in the first half of the two-stage model (i.e. MN-!'  PI and 6I> and
get a unique solution.  For each dose group,  one of these parameters should come
from outside information.
     Maronpot et al.  (1992)  measured the rate of cell proliferation in these
rodents (Tritscher et al.,  1992) using immunocytochemical detection of cells that
had  undergone  replicative DNA synthesis.  The average labeling index by dose
group is given in Table  8-6 for the uninitiated animals  (saline controls)  in this
study.  It  is seen there  is a slight drop in mean labeling  index  (p>0.05) from
control to the group given 3.5 ng/kg/day.   The labeling  index then increases with
increasing  dose.  Under the assumption of a linear  birth  death process, it is

                                     8-53                             08/27/92

-------
00
 I
U1
TABLE 8-6
Results from Two-Stage Model for Hepatocarcinogenesis in Rats*
Model
Labeling index data and
calculations
Using labeling index
(Model A)
Allowing all parameters to vary
(Model B)
Constant ratio (f1/B1 estimated
.31X10 •)
Parameter
labeling index
birth rates
relative change
(X)
scaled birth
rates
mutation rate
birth rate
ratio ((5,/B,)
birth rate
ratio (£i/Bj)
mutation rate
hepatocytes
initiated
birth rate
mutation rate

Control
3.41
2.75x1(T3
0
1.67x10-13
7.505x10"13
1.23x10'2
0.01
2.60x10"2
7.74x10'12
1.507x10'12
2. OX
2.60x10'2
1.49x10'12
TCDD Dose Administered
3.5 ng/kg/day
3.22
2.34x10'3
-14.9
1.42x10'2
13.98x10"13
1.41x10'2
0.01
3. 36x1 0'2
5.28x10'2
1.087x10'12
4.34X
3. 20x1 0'2
1.035x10'12
10.7 ng/kg/day
4.87
3.56x10'3
29.5
2.16x10'2
8.007x10'13
2.14x10"2
0.01
5. 02x1 0'2
3.75X10"1
0.881x10'12
3.32X
3.34x10'2
0.59x10'12
35.7 ng/kg/day
5.31
3.90x10'3
41.8
2.34x10'2
25.20x10'13
2.32x10'2
0.01
3.47x10"2
1.79x10'2
1.896x10'12
12.84X
3.41x10'2
1.875x10-12
1 25 ng/kg/day
7.09
5.26x10'3
91.3
3.19x10'2
9.912x10'13
2.30x10'2
0.01
3. 33x1 0'2
LOOxlO"5
1.572x10'12
8.72X
3.32x1Q-2
1.579x10'12
                                                                                                                                                                    I
                                                                                                                                                                   O
                                                                                                                                                                   O

                                                                                                                                                                   z
                                                                                                                                                                   o
                                                                                                                                                                   O
                                                                                                                                                                   G
                                                                                                                                                                   O
                                                                                                                                                                   H
                                                                                                                                                                   M
      Source:  Lucier et al.f 1991;  Tritscher et al.,  1991;  Maronpot et al., 1992
O
co
10
\o
K)

-------
                         DRAFT—DO NOT QUOTE OR CITE






possible to convert these labeling indices into estimated birth rates for these




cells using the formula:
Where f) is the birth rate,  t is  the number of days over which labeling was done



and LI is the labeling index (Moolgavkar and Lubeck, 1992).  This conversion is



given in row 2 of Table  8-6.  These  rates  are  in nonfocal hepatocytes (normal



cells) and correspond to an average of one to two births per year per hepatocyte.



TCDD  seems  to  double the  control  birth  rate for  a  dose of 125  ng/kg/day (a



relative change of almost 100% — row 3 of Table 8-6).



     Maronpot  et al.  (1992)  also  calculated  the  labeling  index  in  focal



hepatocytes in the high-dose group from this  study.  The mean LI for animals in



the noninitiated group exposed to 125  ng/kg/day of TCDD was 36% which corresponds



to a birth rate of 0.0319.  The ratio of birth rate in nonfocal cells to birth



rate in focal cells is 0.0319/0.00526 = 6.06.  Assuming this ratio is constant



over all dose groups, we can rescale the birth rates in the remaining dose groups



to correspond to birth rates for focal cells  resulting in  the rates show as row



4 in Table 8-6.



     Using the same method as that used by Moolgavkar et al.  (1991), given these



birth rates, it is possible to estimate pN.j and 6j for the focal lesion data of




Lucier et al. (1991).  The resulting parameter estimates are given as rows 5 (pN_




j)  and 6  (6(/Pj) in Table 8-6.  In all  cases, the  best estimate of the death rate




was 6j = 0.01, an arbitrary lower bound chosen by curve-fitting.  This  indicates




that these birth rates are far too small for the sizes of the lesions observed



since death rates of this magnitude would result in almost no loss of  initiated



cells.  (Technical note:  the results do not change markedly if the death rate



is allowed  to  be estimated  as  zero).   The estimated  mutation  rates  are also



extremely small and have  no set pattern to them.   Finally,  there is considerable



lack-of-fit  with these  parameters  of the  model to the  focal lesion  data



themselves.





                                     8-55                             08/27/92

-------
                          DRAFT—DO NOT QUOTE OR CITE






     To get a handle on how large the  labeling index would have to be to result



in a reasonably good fit to the  data,  the model was again fit to the liver foci



data, starting with the solution given in rows 3  and 4 of Table 8-6 and allowing



|J| to roam freely and the ratio 6j/Pj to be  >10"5 (rather than 10~6).  The resulting




parameter estimates are given in rows 5,  6 and 7 of Table 8-6.



     The first point to note in this table is that the estimated birth rate in



the high dose group  (0.0333) is approximately the same as that estimated from the



labeling data.   This  provides  some  mechanistic  support for  this  particular



parameterization of the model.   We  also see that the  death rate of initiated



cells  (6j - ratio • |Jj) is near zero  in all groups except  for the middle dose




group.  The birth rate  is virtually identical in  the 3.5,  35.7 and 125 ng/kg/day



dose groups corresponding to a  labeling index in 7 days of 37.5%.  The control



group has a birth rate  for initiated cells of 0.026 which corresponds to an LI



of "30.5% and the dose  group receiving 10.7 ng/kg/day have a birth rate of 0.050



corresponding to an Li of roughly 50%. These values suggest that TCDD may have



a small effect on the growth characteristics of  initiated cells, but the effect



is not as large as  might have been expected.  There is very  little change in the



mutation rate as a function of dose  level.  These small changes do, however,



result in changes in the expected number of cells in the initiated state; this



is shown in row 11 of Table 8-6.  This is calculated as follows:  assuming that



liver cells have a radius of 0.012 mm, the  volume  of a  single  liver cell is 4/3



n (.012)3 = 1.5079xlO'3 mm3.  Inverting this gives a total of 663 cells/mm3.  The




expected number  of  initiated  cells  per  mm3.   The  expected number of initiated




cells per mm3 is given by the formula:
where t  * 229 days (the length of  the  experiment).   The percentage is  simply



E[X,]/663 • 100%.










                                      8-56                              OB/21/92

-------
                          DRAFT—DO  NOT QUOTE OR CITE






     The differences observed in the percentage of initiated hepatocytes can be



easily explained by examining Pj  and  A