Received April 22, 1997; accepted July 8, 1997

Transcription

1 TOXICOLOGY AND APPLIED PHARMACOLOGY 146, (1997) ARTICLE NO. TO Hepatic Foci in Rats after Diethylnitrosamine Initiation and 2,3,7,8-Tetrachlorodibenzo-p-dioxin Promotion: Evaluation of a Quantitative Two-Cell Model and of CYP 1A1/1A2 as a Dosimeter Rory B. Conolly*,1 and Melvin E. Andersen *Chemical Industry Institute of Toxicology, Six Davis Drive, Research Triangle Park, North Carolina 27709; and ICF Kaiser Engineers/K. S. Crump Division, 3200 Chapel Hill Nelson Boulevard, Suite 208, Research Triangle Park, North Carolina Received April 22, 1997; accepted July 8, 1997 (1996) independently demonstrated the plausibility of the Hepatic Foci in Rats after Diethylnitrosamine Initiation and same quantitative, biologically based model by analyzing 2,3,7,8-Tetrachlorodibenzo-p-dioxin Promotion: Evaluation of a diethylnitrosamine (DEN) initiation, 2,3,7,8-tetrachlorodi- Quantitative Two-Cell Model and of CYP 1A1/1A2 as a Dosimeter. benzo-p-dioxin (TCDD) promotion experiments. In this Conolly, R. B., and Andersen, M. E. (1997). Toxicol. Appl. Pharmamodel (the one-cell model), mutations of normal cells give col. 146, rise to a single kind of mutated cell. The foci descended 2,3,7,8-Tetrachlorodibenzo-p-dioxin (TCDD) is a potent hepatic from these mutant cells by clonal expansion are consequently tumor promoter in female rats. We used a quantitative, stochastic identical in their responses to TCDD exposure. Moolgavkar initiation-promotion model based on R. B. Conolly and J. S. Kimet al. (1996) and Portier et al. (1996) found that the onebell (Toxicol. Appl. Pharmacol. 124, , 1994) to analyze initicell model provided good descriptions of the data on hepatic ation promotion results from a previously published study (H. C. Pitot et al., Carcinogenesis 8, , 1987) within the context foci when TCDD was allowed to increase the mutation rate of a negative selection model of tumor promotion. In this model, of normal hepatocytes and alter the division and death rates two types of initiated cells (called A and B cells) are produced by of the mutated cells. However, TCDD is not mutagenic in DEN initiation. Visually excellent correspondence between model a variety of assays (Shu et al., 1987; Randerath et al., 1988; predictions and data (i.e., foci/cm 3 liver and percentage of liver Turteltaub et al., 1990). We were therefore interested in occupied by foci) are obtained when TCDD is described as having determining if an alternative model, having more than one dose responsive effects on division and death (apoptotic) rates of kind of mutated cell (the two-cell model), but no effect of these two cell types. For A cells, both the division and the death TCDD on mutation rates, could describe DEN initiation rates increase while the difference between division and apoptotic TCDD promotion data. rates decreases. For B cells, the difference between division and The negative selection mechanism for hepatic tumor proapoptotic rates increases, primarily due to a decrease in the apomotion proposed for phenobarbital by Jirtle et al. (1991) ptotic rate. We also linked these alterations in cell kinetics to a pharmacokinetic model for TCDD incorporating a five subcompart- builds on earlier work by Farber and others (e.g., Farber, ment model of the liver acinus with induction of CYP1A1 and 1987; Tsuda et al., 1980) and provides a biological basis for 1A2 genes in the subcompartments. Alterations in A cell kinetics the two-cell model. In this mechanism, the promoter procorrelate with effects of TCDD in the region most sensitive to induction vides a mitogenic stimulus to hepatocytes, and the liver then (subcompartment 5 centrilobular region); B cell dynamics elaborates mitosuppressant growth factors to constrain pro- correlate with induction in subcompartments 3 5 (centrilobular and liferation. Finally, specific cells, mutated at loci that render mid-zonal regions). In summary, these modeling exercises show that them less sensitive or insensitive to the mito-inhibitory envi- (1) the two-cell model, without presuming effects of TCDD on the ronment, derive a growth advantage that allows them to grow mutation rate of normal hepatocytes, reproduces the data of Pitot out to altered foci and eventually progress to tumors. With et al. (1987) and (2) induction of CYP1A1/1A2 in different regions phenobarbital, transforming growth factor b 1 is the inhibiof the hepatic acinus can be used as a general correlate of these tory growth factor. Oberhammer et al. (1992) reported that presumed changes in cell growth kinetics Academic Press the mature form of TGF-b 1 is involved in the initiation of apoptosis in the liver and Ohno et al. (1995) demonstrated 2,3,7,8-Tetrachlorodibenzo-p-dioxin (TCDD) is a potent that pericentral-equivalent hepatocytes were more sensitive tumor promoter. Moolgavkar et al. (1996) and Portier et al. to TGF-b 1 -induced apoptosis than were periportal-equivalent hepatocytes. Mutations in the mannose-6-phosphate re- 1 To whom correspondence should be addressed. ceptor, which binds and activates TGF-b 1, appear to be re X/97 $25.00 Copyright 1997 by Academic Press All rights of reproduction in any form reserved.

2 282 CONOLLY AND ANDERSEN sponsible for rendering some cells unresponsive to TGF-b 1 (Jirtle et al., 1994). Andersen et al. (1995) have suggested that this mechanism may be broadly applicable to a diverse group of liver tumor promoters, including dioxins. With TCDD, the dose response curves described by Pitot et al. (1987) would be explained by both A and B cells being resistant to mito-inhibition and having different sensitivities to the mitogenic effects of TCDD. A preliminary version of the two-cell model was described in abstract form (Andersen et al., 1993). We also investigated the plausibility linking the two-cell model with a pharmacokinetic model for TCDD describing the hepatic dosimetry of TCDD and regional induction of expression of CYP 1A1 and 1A2 genes in the liver (Andersen et al., 1997). The goal was to determine if the dose response behavior for CYP 1A1 and 1A2 induction predicts the dose responsive effects of TCDD on the cells of the two-cell model (called A and B cells). The dose response for CYP induction was considered a surrogate for the dose response of unspecified biochemical factors which actually link TCDD exposure to changes in the growth kinetics of A and B cells. The objectives of this study were to evaluate (1) the ability of the two-cell model to replicate the TCDD data of Pitot et al. (1987) using biologically plausible parameter values, and (2) the utility of hepatic dose surrogates based on induction of CYP 1A1 and 1A2, obtained using the PBPK model of Andersen et al. (1997), to set parameter values for the twocell model. For both of these objectives, parameterization of the two-cell model was developed in the context of the negative selection hypothesis of Jirtle et al. (1991). METHODS FIG. 1. (A) One-cell clonal growth model. The number of normal cells is determined by the balance between the rates of division and death. Mutation of normal cells is linked to their division. When mutation occurs, Simulation model. The simulation model used for the analysis of the division results in one normal cell and one mutant. Each mutant cell so TCDD dose response data was modified from that described by Conolly created may be the progenitor of a clone of cells or may become extinct. and Kimbell (1994). Altered hepatocytes arise by mutation of normal cells The division and death rates of the mutant cells typically differ from those and correspond to the cells in the altered hepatic foci described by Pitot et of the normal cells. (B) Two-cell clonal growth model. The only difference al. (1987). Only the mutations that give rise to the experimentally observed from the one-cell model is that two types of mutation of normal cell are foci of mutated cells are evaluated in this type of modeling. Other mutations described, leading to distinct populations of A and B cells. Division and presumably may occur concurrently, but they are irrelevant to the endpoints death rates of A and B cells typically differ from each other as well as of interest here. There may also be more than one genetic locus where from those of normal cells. mutation can give rise to the experimentally observed foci. In this case, the mutation rates specified in the model are actually compound parameters representing mutations at all these loci. The model of Conolly and Kimbell (1994) described only one pathway Growth kinetics of hepatocytes are described by division and death rates from normal to mutated cells (Fig. 1A). The present model (the two-cell (a, b; 1/day), model) describes two pathways, leading to the formation of two types of mutated cells (Fig. 1B). dn The liver is treated as consisting of hepatocytes/cm 3 (Weibel Å N(a 0 b), dt et al., 1969). This cell density is used for both normal and mutated cells (1) and both normal hepatocytes and hepatocytes in foci behave independently of each other. The cell cycle is not described per se. These simplifying where N is the number of normal hepatocytes. Note that b represents all assumptions are used for modeling the TCDD data from Pitot et al. (1987) modes of cell death including apoptosis and necrosis. While Eq. (1) is because no data are available to support modeling at a greater level of deterministic, the mutation of normal cells is calculated stochastically. The detail. While there may be differences in cell density between normal cells expected number of mutations is linked to the number of hepatocytes divid- and cells in foci, they would not be expected to change the major conclusions ing during each time step Dt by of this work, though they could result in small changes in parameter estimates. N m Å NmaDt (2)

3 ANALYSIS OF TCDD DOSE RESPONSE FOR HEPATIC FOCI IN RATS 283 The data. Pitot et al. (1987) gave female F344 rats 70% partial hepatec- tomy, 10 mg DEN/kg 1 day later, and various doses of TCDD (0.0001, 0.001, 0.01, and 0.1 mg/kg/day) by biweekly intramuscular injection in corn oil starting 2 weeks after the DEN treatment. DEN-only and TCDD-treated rats were sacrificed after 8 and 6 months, respectively. Livers were evaluated for foci by staining for g-glutamyltranspeptidase, canalicular ATPase, and glucose 6-phosphatase. Data were reported as the three-dimensional volume fraction of the liver occupied by foci and number of foci per liver (Fig. 3). The U-shape of these dose response curves is particularly interesting. However, the longer survival of the control group (sacrifice at 8 rather than 6 months) means that the reported U-shaped curves, particularly that for volume fraction, are probably exaggerated relative to that which would have been seen had the controls been sacrificed with the TCDD-exposed groups at 6 months (H. C. Pitot, personal communication, 1994). Fitting the model described herein to the 8-month time point and backextrapolating from 8 to 6 months suggested that there is no U-shaped dose response behavior for volume fraction. However, the model predicted that number of foci per liver changes little between 6 and 8 months, suggesting that the U-shaped behavior reported for this end point may be a reproducible observation. Teeguarden et al. (1996), in a preliminary report on a repeat of the experiment by Pitot et al. (1987), did not observe U-shaped behavior for volume fraction and only an equivocal U-shaped response for number of foci per liver. For our analysis, the average focus size (cells per focus) had to be calculated. This quantity was not reported by Pitot et al. (1987). Calculation of average focus size from foci per liver and volume fraction, the quantities reported by Pitot et al. (1987), requires an estimate of liver size. We assumed where N m is the number of normal hepatocytes mutating during the time interval Dt, and m is the probability of mutation per cell division. A random deviate about N m denoting the number of mutations during Dt is drawn from a Poisson distribution using the function POIDEV (Press et al., 1989). 2 Inputs to POIDEV are the mean of the Poisson distribution (N m ) and a pseudorandom number between 0 and 1 generated with the algorithm UNIFL (Bratley et al., 1987). Once a mutated cell is created, it may either divide or die with probabilities per time step calculated from the division and death rates for altered cells according to Conolly and Kimbell (1994). Division of these single altered cells derived directly from normal cells may give rise to clones of altered cells. The simulation program keeps track of each clone over time, allowing average clone size (cells/clone) and number (clones/cm 3 )tobe described. Volume fraction is calculated directly from average clone size and number of clones/cm 3. The behavior of each cell in a clone is calculated stochastically for each time step as long as the clone is smaller than 1000 cells. When clones become sufficiently large, greater than 1000 cells in size, their growth kinetics become effectively deterministic. At this point, the program reverts to a deterministic calculation of growth for each clone to save computer time. If clone size falls below 1000 cells, the calculation switches back to the stochastic mode. A number of simulations were run showing that growth of individual clones is effectively deterministic at 1000 cells by varying the cutoff point at which the calculation switches between stochastic and deterministic modes. For the experiment analyzed, no change in program behavior could be seen as long as the cutoff was greater than 100 cells (data not shown). Some change in behavior was seen with cutoffs between 10 and 100. The cutoff of 1000 was chosen as a conservative value to ensure an accurate description of the role of stochastic behavior on clone growth. Clones smaller than 24 cells in size (in three dimensions), corresponding to a two-dimensional focal transection diameter of about 65 mm, were considered undetectable for the purpose of comparing simulations to the data reported by Pitot et al. (1987). The methods used by Pitot et al. (1987) do not reliably identify focal transections below this size. Partial hepatectomy is described as an instantaneous decrease in liver weight. The remaining liver starts to regenerate the missing tissue 24 hr later and the liver is fully regenerated 7 days (Higgins and Anderson, 1931; Tatematsu et al., 1988) after partial hepatectomy. Regeneration is achieved by increasing the division rate (a) of hepatocytes. The division rate returns to its normal value once the liver is fully regenerated. The value of a needed to regenerate the liver is calculated using lw Å lwcre t(a0b), (3) where lw is the weight of the full size liver, lwc is the weight of the liver immediately after partial hepatectomy, and t is the interval (6 days) over which the liver regenerates. The hepatocyte death rate (b) stays at its normal value during regeneration. Pitot et al. (1987) gave DEN 24 hr after partial hepatectomy. The effect of DEN is modeled as a transient increase in the probability of mutation per cell division while the liver is undergoing regenerative cellular replication. This increased probability of mutation lasts until the liver is fully regenerated, at which point the probability returns to its basal level. This part of the model is intended to approximate the actual sequence of events relating to regenerative proliferation and mutation after partial hepatectomy and DEN treatment as described by Pitot et al. (1987). The model could be refined by attempting to link time-course data on specific DNA adducts 2 POIDEV and associated algorithms found in Press et al. (1989) did not execute properly with DOS and Windows operating systems, though they worked correctly with VMS. Upgrades of these algorithms available on the World Wide Web at ran on DOS and Windows without problem. FIG. 2. Simulation of liver weight and probability of mutation per cell division during partial hepatectomy and regenerative proliferation as described by the two-cell model. This description incorporates a liver weight of 6.6 g (Y. P. Dragan, pers. commun., 1994) and 2/3 partial hepatectomy (Pitot et al., 1987). Regenerative proliferation is complete by 1 week after partial hepatectomy (Higgins and Anderson, 1931; Tatematsu et al., 1988). Probability of mutation per cell division is set to zero at all times except during regenerative replication. Overall, this description is intended to mimic the experimental protocol of Pitot et al. (1987) in which the mutations of interest are assumed to arise from the combination of partial hepatectomy and DEN treatment (see text for additional details). to the mutation probabilities. For the purposes of the analyses described here, however, a more exact description of this phase of the experiment is less important than development of a plausible model structure giving rise to the required number of foci. The time-dependent changes in liver weight due to partial hepatectomy and regenerative proliferation and the associated changes in probability of mutation are shown in Fig. 2.

4 284 CONOLLY AND ANDERSEN in the present work simulates three-dimensional quantities and does not incorporate an algorithm allowing unbiased conversion from three to two dimensions, as has been described by Moolgavkar et al. (1996). The present analysis thus takes the data as reported by Pitot et al. (1987) as is with no attempt to compare simulated three-dimensional quantities directly to the original two-dimensional data. Parameter values. Data on average size and average number per unit volume of a particular kind of altered focus do not uniquely identify the three parameters, mutation rate and division and death rates, that control its growth. One of the three must be fixed (Conolly and Kimbell, 1994). For the analyses described here, we first fixed the values of the mutation rates. The strategy for fixing the two mutation rates was as follows. First, all mutations of normal hepatocytes leading to formation of cells in hepatic foci were assumed to be due to the effect of DEN on the probability of mutation per cell division during regenerative replication. At all other times in the study, the probability of mutation per cell division (i.e., the basal probability) was assumed to be 0. (This approach is different from that taken by Moolgavkar et al. (1996) and Portier et al. (1996). They allowed TCDD to increase the probability of mutation.) In reality, the basal probability of mutation to the cell types found in foci is probably not 0 but rather some number that is small relative to that due to DEN. This can be inferred from the data on altered foci in rats given TCDD but not DEN (Fig. 3). Using zero for the basal probability is a convenience that does not significantly affect the analysis described here. From Fig. 3, we can estimate the number of detectable foci per liver at the end of the study. If b were 0, the minimum possible value for the mutation rate during regenerative replication would be that required to generate the number of foci observed at the end of the study. This minimum value would be accurate, given b Å 0, if all progenitor cells created by mutation of normal hepatocytes grow out into observable foci, which are assumed to consist of at least 24 cells (Conolly and Kimbell, 1994). We assumed, however, that b has a small nonzero value for initiated cells in rats not exposed to TCDD ( /day). This value is typical of normal hepatocyte division rates measured in mature rats. In a mature animal, the division and death rates of cells in a tissue are roughly in balance, and therefore the measured division rate must approximate the death rate. There is uncertainty associated with using this value for the basal death rate of cells in foci since the growth kinetics of cells in foci may be different from those of normal cells. However, no data that allowed a direct estimate of b for either A or B cells were available. With the basal value of b Å /day, probabilities of mutation were set sufficiently large to provide the observed number of foci at the end of the study. In all cases, estimates of the division and death rates were obtained by visually fitting the model to the data. Visual fitting does not guarantee that the parameter values provide the best possible fit to the data. This was not a significant factor for the current study, however, as its goal was to identify whether or not the two-cell model was plausible, i.e., if it was capable providing visually good fits to the data using biologically reasonable param- eter values. Use of a formal method for identifying optimal parameter values would be more important if the model were being used for quantitative risk assessment. Division and death rates were first estimated for the control group since it is the background on which the simulations of the TCDD dose groups are built. Rotstein et al. (1986) showed that cells in hepatic nodules of male F344 rats after DEN treatment responded to partial hepatectomy in a similar manner to normal hepatocytes. Although the protocol used by Rotstein et al. (1986) differed from that of Pitot et al. (1987), we used their results to guide parameterization of the two-cell model because no better data were available. We therefore configured the two-cell model so that the division rates of A and B cells increased during regenerative proliferation after partial hepatectomy to the same level as normal hepatocytes after partial hepatectomy. Once the liver had recovered from partial hepatectomy, the values of a and b giving rise to the foci in control rats were estimated. This was accomplished in two steps. First, the deterministic quantity (a 0 b) was FIG. 3. Dose response data reported by Pitot et al. (1987) for altered foci per liver (A) and volume percentage of foci in the livers (B) of female F344 rats given 70% partial hepatectomy, with or without 10 mg DEN/kg, and administered TCDD for 6 months at the dose levels noted. Additional details under Methods (redrawn from Pitot et al., 1987). a constant liver weight of 6.6 g throughout the duration of the experiment (Y. P. Dragan, pers. commun., 1994), except as associated with partial hepatectomy as described above. The assumption of time-constant liver weight was used because no data were available on changes in liver weight during the course of the experiment. Using an estimate of hepatocytes/cm 3, a liver weight of 6.6 g, and data on volume fraction, an average number of cells per focus for each dose of TCDD could be calculated. The quantities reported by Pitot et al. (1987), three-dimensional volume fraction of the liver occupied by foci and number of foci per liver (Fig. 3), were derived from data on size and number of two-dimensional focal transections. Assumptions used in the stereological transformation from two to three dimensions (Campbell et al., 1986) introduce uncertainty into quantitative values of the three-dimensional data. The model described

5 ANALYSIS OF TCDD DOSE RESPONSE FOR HEPATIC FOCI IN RATS 285 obtained by fitting the model to the volume fraction data. Finally, holding (a 0 b) constant, a and b were adjusted to fit the data on number of foci/ cm 3 (Dewanji et al., 1989). At this point, estimates of m, a, and b had been obtained, which provided a good simulation of the control group. Since m was described as being invariant with TCDD dose, parameter estimation for the TCDD dose groups simply involved fitting (a 0 b) for each dose level followed by adjustment of a and b together. Use of gene induction dose surrogates as inputs to the clonal growth simulation model. Immunohistochemical observations of rat livers treated with various doses of dioxin show a clear demarcation between cells with maximally elevated levels of CYP1A1 and CYP1A2 and cells with normal levels of these proteins (Bars and Elcombe, 1991; Tritscher et al., 1992). The elevated levels of CYP1A1 and CYP1A2 are presumed to reflect dioxinmediated activation (induction) of transcription of the corresponding genes. These immunohistochemical data have now been considered in a PBPK model for dioxin disposition and gene induction in the liver (Andersen et al., 1997). The liver in this model has five subcompartments corresponding to different regions of the hepatic acinus. This spatial description of the liver permits simulation of the regional induction of CYP1A1 and CYP1A2 in response to dioxin. Three dosimeters were derived from the PBPK model for chronic dosing with dioxin (Tritscher et al., 1992) and evaluated for their ability to predict the data of Pitot et al. (1987). The dosimeters were: (1) induction in the centrilobular zone (subcompartment 5, the region most sensitive to induction), (2) induction in the total liver (all five subcompartments combined), and (3) induction in subcompartments 3, 4, and 5 (the centrilobular and midzonal areas of the liver) (Table 1). The numerical values of the dosimeters are the fractions of maximum possible enzyme induction as determined by the model of Andersen et al. (1997). Use of these dosimeters was based on the assumption that the dose response for induction of CYP1A1 and CYP1A2 predicts the dose response for the genes whose products actually control the growth of the cells in the foci described by Pitot et al. (1987). TCDD-induced changes in the values of the CYP1A1 and 1A2 dosimeters were translated into effects on the division (a) and death (b) rates for A and B cells. Translation was not required for m because its value is fixed independently of TCDD dosing, as described above. We assumed that TCDD affects the fraction of mutated cells (A and B cells) that are either actively in the cell cycle or dying (i.e., undergoing apoptosis or necrosis). This assumption provides a unifying hypothesis linking the gene induction dosimeters to the parameters of the clonal growth model. The quantitative aspects of this hypothesis are described in the Appendix. Software and hardware. The simulation model was written in ACSL (MGA Software, Concord, MA) and run on a 133 mhz Intel Pentium (Gateway 2000, N. Sioux City, SD). Typical time for 25 runs of the stochastic program was about min. A listing of the simulation program in hard copy or by is available from RBC (rconolly@ciit.org). FIG. 4. Simulation of dose response curves for (A) altered foci/cm 3 and (B) focal volume as percentage of liver in DEN-treated, TCDD-exposed rats. The simulated data represent the sum of the separate values for A and B cells. Twenty-five consecutive simulations are shown to illustrate the role of stochastic variability. For controls, model predictions are for the 6-month time point, while control data in Fig. 3 are for 8 months (see Methods for additional details). TABLE 1 TCDD Gene Induction Dosimeters RESULTS A cell dosimeter: B cell dosimeter: Centrilobular Midzonal centrilobular Simulation of the Growth of Altered Foci TCDD induction induction The two-cell model provided accurate simulations of the (mg/kg/day) (subcompartment 5) a (subcompartments 3, 4, and 5) a data described by Pitot et al. (1987) (Fig. 4) with parameter estimates that were nondecreasing functions of TCDD dose (Table 2). The DEN-induced probabilities of mutation per cell division used to describe the creation of A and B cells from normal hepatocytes were and , a Numerical values of the dosimeters are the fractions of maximum possi- respectively. ble induction of CYP1A1 and 1A2 in specific subcompartments of the liver In these simulations (Fig. 4), essentially all the volume acinus as defined by Andersen et al. (1997). fraction is composed of A cells for controls and the 0.001

6 286 CONOLLY AND ANDERSEN TABLE 2 Division (a) and Death (b) Rates Used for Simulations Shown in Fig. 4 A cells B cells TCDD (mg/kg/day) a (1/day) a b (1/day) a a (1/day) a b (1/day) a a Parameter values estimated by visually fitting the two-cell model to the data of Pitot et al. (1987) as described under Methods. and mg TCDD/kg/day dose levels. While B cells are TCDD exerts a mitogenic stimulus on A cells at all dose created by mutation of normal hepatocytes during regenerative replication after partial hepatectomy and DEN, they do crease the apoptotic rate to maintain a normal liver size. B levels. The liver responds by expressing factors which in- not undergo significant clonal expansion as their division cells are insensitive to low doses of TCDD but respond at and death rates are the same (Table 2). At the higher dose high dose levels with an increased division rate and a delevels of TCDD, 0.01 and 0.1 mg TCDD/kg/day, B cells do make up a significant fraction of the volume fraction, while the relative contribution of A cells declines. At the highest dose level, 0.1 mg TCDD/kg/day, the volume fraction consists almost entirely of B cells. These dose-dependent growth kinetics reflect a differential sensitivity of A and B cells to TCDD. A cells respond at all dose levels of TCDD while B cells respond only at the two highest dose levels. Simulation of the time course of focus growth for TCDD dose level 0.01 mg/kg/day predicts a biphasic growth pattern for number of foci/cm 3 (Fig. 5). The early increase is due to expanding clones of A cells while the later increase is due to B clones (Fig. 5A). Surface plots allow visualization of model-predicted changes in foci/cm 3 and volume fraction as functions of both time and TCDD dose (Fig. 6). At 6 months, a U-shaped curve is predicted for foci/cm 3 (Fig. 6A) that is similar in shape to the curve described by Pitot et al. (1987) (Fig. 3A). However, the simulation model predicts a monotonically increasing curve for volume fraction (Fig. 6B), while Pitot et al. (1987) reported U-shaped behavior for this end point (Fig. 3B). This discrepancy arises because the control group was sacrificed at 8 months while the TCDD-exposed groups were sacrificed at 6 months (H. C. Pitot, pers. commun., 1994). The simulation model predicts that numbers of foci/cm 3 do not change significantly between 6 and 8 months. Therefore, the curve for foci/cm 3 reported by Pitot et al. (1987) is similar in shape to that predicted by the model. The lack of change in foci/cm 3 occurs because essentially all clones that grow large enough to be experimentally detectable do so prior to 6 months. However, the model predicts that volume fraction increases significantly between 6 and 8 months, leading to FIG. 5. Time course simulation of (A) foci/cm 3 and (B) focal volume the difference between Figs. 3B and 6B. as a percentage of liver for rats given partial hepatectomy, DEN, and 0.01 Consistency with the Negative Selection Hypothesis mg TCDD/kg/day). The data of Pitot et al. (1987) are denoted by. The upward inflection in foci/cm 3 at about 150 days reflects the growth of B The parameter estimates shown in Table 2 are consistent cells on the A cell background. Twenty-five consecutive simulations are with the negative selection hypothesis of Jirtle et al. (1991). shown to illustrate the role of stochastic variability.

7 ANALYSIS OF TCDD DOSE RESPONSE FOR HEPATIC FOCI IN RATS 287 FIG. 6. Surface plots showing (A) foci/cm 3 and (B) focal volume as a percentage of liver as functions of time and TCDD dose. These plots are based on a single run of the stochastic program and therefore do not illustrate stochastic variability. For controls, model predictions are for the 6-month time point, while control data in Fig. 3 are for 8 months (see Methods for additional details). cells from G 0 into either the cell cycle or apoptosis or necrosis (Eq. (1A)). The ability of the gene induction dose surrogates to predict the data from Pitot et al. (1987) was evaluated by deriving division and death rates for A and B cells from the surrogates as described in the Appendix (Table 3). Because the results of the fitting exercise described above had suggested that A cells respond to lower doses of TCDD than B cells, a centrilobular dosimeter (column 2 of Table 1) was used to estimate the A cell parameters. This dosimeter was based on CYP1A1 induction in the centrilobular region of the liver, the region most sensitive to this effect of TCDD. A surrogate based on induction in subcompartments 3 5 (column 3 of Table 1) was used for B cells. A good simulation of the dose response for volume fraction was obtained using parameter values based on the dose surrogates (Fig. 7B). Simulation of the dose response for foci/cm 3 was less satisfactory (Fig. 7A), failing to reproduce the U-shaped behavior for this endpoint (Figs. 3A, 4A). The reason for this discrepancy can be explained in terms of the hypothesis used to link the dose surrogates to the division and death rates of A and B cells (Appendix). The hypothesis states that TCDD increases faction (f act ) of A or B cells which are in the cell cycle or undergoing apoptosis or necrosis rather than resting in G 0. The dose response for this fraction can be calculated from both the fitting exercise and the dose surrogates (Fig. 8). There is good agreement between the dose response curves for f act for B cells (Fig. 8B). The agreement is weaker for f act for A cells, however (Fig. 8A). For this latter case, the fitting exercise predicted a response at the lowest TCDD dose, mg/kg/day, while the dose surrogate values were not affected significantly at this dose. This reflects the fact that Pitot et al. (1987) found effects of TCDD on clonal growth at lower doses than Bars and Elcombe (1991) and Tritscher et al. (1992) found induction of CYP1A1 and CYP1A2. DISCUSSION The main objectives of this study were to (1) evaluate the hypothesis that a two-cell model provides a biologically plausible interpretation of the TCDD dose response curves for altered foci described by Pitot et al. (1987) and (2) combine the two-cell model with the TCDD pharmacokinetics creased death (apoptotic) rate. B cells are thus a population enzyme induction model (Andersen et al., 1997) to evaluate of cells which is resistant to homeostatic control and expands the ability of TCDD-mediated regional induction of rapidly at high dose levels. CYP1A1 and 1A2 in rat liver to predict the dose response curves for preneoplastic foci described by Pitot et al. (1987). Consistency of Parameter Estimates with Predictions Based on Regional Induction Dosimeters The dose surrogates based on induction of expression of the CYP1A1 and CYP1A2 genes are listed in Table 1. Our interpretation of these values is that they describe a dose response curve for the ability of TCDD to move mutated The two-cell model provided good simulations of the data of Pitot et al. (1987). Biologically reasonable parameter estimates (Table 2) were obtained when the model was visually fit to the data. The rationale for choice of values for the probabilities of mutation to A or B cells per division of normal cells during regenerative replication was described

8 288 CONOLLY AND ANDERSEN TABLE 3 Division and Death Rates Calculated from the Gene Induction Dosimeters a,b A cells c B cells d TCDD dose (mg/kg/day) a (1/day) b (1/day) a (1/day) b (1/day) a Dosimeter values are given in Table 1. b See Appendix for method used to translate dose surrogates into division and death rates. c Estimated from subcompartment 5 dose surrogates. d Estimated from averaged dose surrogate for subcompartments 3 5. under Methods. Briefly, the values selected ( and mates for b on the order of /day. (We assume that significant numbers of mutations occur only during regenerative for A and B cells, respectively) were the minimum values sufficient to produce enough mutant cells to generate replication following partial hepatectomy and DEN treatment.) If the mutation probabilities were much smaller, not the observed descendant foci at 6 months given basal estienough mutant cells would be created. Much higher values of these probabilities are also unlikely. For example, if the probabilities were doubled, twice as many single A and B cells would be created by mutation of normal cells. Accurate simulation of the Pitot et al. (1987) data would then require that many of these single cells and descendant clones become extinct. This would require significantly larger values of a and b while maintaining a constant (a 0 b) (Dewanji et al., 1989). For example, the current parameterization suggests that 10 20% of A cells are in the cycle or dying at 0.1 mg TCDD/kg/day (Eq. (1A) and Tables 2 and 3). Increasing a and b together to increase the rate at which clones become extinct while maintaining the volume fraction (holding [a 0 b] constant) would require an even greater fraction of A cells to be either in the cell cycle or in apoptosis or necrosis. A recent report on TCDD initiation promotion (Stinchcombe et al., 1995) indicates that such high division and death rates are unlikely when rats are given 0.1 mg TCDD/kg/day, the highest dose used by Pitot et al. (1987). Overall, these arguments suggest that the chosen values for probability of mutation, while not exact, are fairly tightly bounded. Dragan et al. (1992) used a protocol similar to that of Pitot et al. (1987) but evaluated altered foci after 1, 3, and 5 months of dosing with 0.01 mg TCDD/kg/day. The time course of volume fraction and number of foci per liver reported by Dragan is similar to that predicted by the two-cell FIG. 7. Simulation of dose response curves for altered foci/cm 3 (A) model for this dose level (Fig. 5A). In particular, the number and focal volume (B) as a percentage of liver in DEN-treated, TCDDof foci per liver reported by Dragan shows a sharp upward exposed rats using gene induction dosimeters (Table 1) to estimate division and death rates for A and B cells (Table 3). These simulations were obtained inflection at about 3 months. The current parameterization using the subcompartment 5 (centrilobular) dose surrogate (Table 1) for A of the two-cell model suggests that this upward inflection cells and the averaged surrogates for subcompartments 3 5 (midzonal to occurs when clones of B cells grow past the minimum size centrilobular) (Table 1) for B cells. The simulated data represent the sum required for detection. The earlier portion of the curve repreof the separate values for A and B cells. Twenty-five consecutive simulasents A cells, which expanded more rapidly than the B cells tions are shown to illustrate the role of stochastic variability. For controls, model predictions are for the 6-month time point, while control data in Fig. at this dose level of TCDD (Table 2). 3 are for 8 months (see Methods and Appendix I for additional details). Taken as a whole, the results are consistent with the hy-

9 ANALYSIS OF TCDD DOSE RESPONSE FOR HEPATIC FOCI IN RATS 289 FIG. 8. Comparison of predictions from the fitting exercise or from the dose surrogates of the effect of TCDD on the fractions of A and B cells in the cell cycle or dying by apoptosis or necrosis (f act ; Appendix, Eq. (1A)). (A) f act for A cells obtained from the fitting exercise and from the averaged dose surrogate for the midzonal centrilobular regions (subcompartments 3 5) of the PBPK model of Andersen et al. (1997) (Table 1). (B) Predictions for B cells from the fitting exercise and from the dose surrogate for the centrilobular region (subcompartment 5). genotype in addition to promoting the growth of the mutated cells. This assertion is surprising in light of the large number of studies that have failed to show DNA-reactivity or mutagenic potential with dioxin. The results of the present work show that the two-cell model provides a good description of the Pitot et al. (1987) data without the need to invoke a novel mechanism of action for TCDD. The Pitot et al. (1987) data set represents the pooling of three different phenotypes of altered cells: g-glutamyltranspeptidase, canalicular ATPase, and glucose 6-phosphatase. Therefore, proposing at least two functionally distinct populations of altered cells on the basis of response to TCDD is not unreasonable. Experimental evaluation of the correlation between histochemical phenotype and growth behavior in response to TCDD exposure is desirable. If such correlations were found to exist, they could help to evaluate the validity of the two-cell versus competing hypotheses for the mechanism of TCDD promotion. (Alternatively, if the correlations do not exist, the need for new histochemical markers more predictive of changed growth behavior would be highlighted.) The present analysis did not attempt a statistical comparison of the two-cell model with the model used by Moolgavkar et al. (1996) and Portier et al. (1996). No quantitative conclusions should be drawn from the present work about the relative abilities of the different models to describe the data. We were more concerned with determining the plausibility of the two-cell model by which we mean its ability to fit the data with reasonable parameter values. As noted above, the twocell model provides a visually good description of the Pitot et al. (1987) data just as the models of Moolgavkar et al. (1996) and Portier et al. (1996) provide good fits to the data they analyzed. Discriminating between these models will require not only the use of rigorous statistical methods but also careful consideration of the biological mechanisms which they describe. Any introduction of bias by the method used for stereological transformation of the data from two to three dimensions will also need to be considered. While comparing applications of these alternative models to diverse data sets would be informative, the primary direc- tion suggested by the two-cell hypothesis is the need for better characterization of growth kinetics and dioxin-responsive genes in the clones at the low and high doses of dioxin used in the Pitot et al. (1987) studies. The results of the combined PBPK and clone growth mod- eling can be qualitatively interpreted within the framework of the negative selection mechanism of tumor promotion and of homeostatic controls on cell proliferation (Jirtle et al., 1991; Andersen et al., 1995). The portion of the dioxin dose pothesis that one action of TCDD may be to move mutated cells from G 0 into either the cell cycle or apoptosis or necrosis. This hypothesis can be stated quantitatively by saying that TCDD exposure increases the value of f act in Eq. (1A) of the Appendix. Moolgavkar et al. (1996) and Portier et al. (1996) have analyzed data on altered foci from DEN initiation TCDD promotion experiments with a model in which only a single response curve associated with P 450 induction in the mutated cell phenotype is formed. Portier et al. (1996) anaadaptation centrilobular region is associated with normal function and lyzed the Pitot et al. (1987) data set in addition to other to a minimal mitogenic stimulus. The A cells of data. To describe the data, both Moolgavkar et al. (1996) and the two-cell model respond in this portion of dioxin dose Portier et al. (1996) concluded that TCDD was increasing response curve to both the proliferative stimulus and the the rate of transition of normal hepatocytes to the mutated homeostatic response. As the dose level increases, the adap-

10 290 CONOLLY AND ANDERSEN tive, homeostatic response can be overwhelmed. In this risk assessment implications of the two-cell model as used higher dose range, more of the liver is affected, as evidenced to analyze the hepatic foci data of Pitot et al. (1987). The by the immunohistochemistry. Cells that are less responsive low dose portion of the data of Pitot et al. (1987) represents to homeostatic signals but which do respond to the highportion mainly the growth behavior of A cells while the high dose dose mitogenic stimulus of TCDD (i.e., B cells in the twocomprises is mainly B cells. Response at the middle doses cell model) have a growth advantage. a mixture of A and B cells. The dose response Several factors should be considered in evaluating the abil- for cancer risk over the dose range studied by Pitot et al. ity of the gene induction dose surrogates to provide simulatypes (1987) would thus depend on whether neither, one, or both tions of the Pitot et al. (1987) data (Fig. 7). First, the hypotherently of cells were actually preneoplastic. No data are cur- sis used to translate the surrogates into division and death available to pursue this question. rates of A and B cells (Appendix) does not have direct experithe Assuming (1) that A and B cells are preneoplastic and (2) mental support though it is consistent with data showing applicability of the negative selection hypothesis of Jirtle effects of TCDD in division and death rates of cells in altered et al. (1991) to the data, the primary carcinogenic risk would foci (e.g., Maronpot et al., 1993; Stinchcombe et al., 1995). arise from B cells. At high doses, B cells respond to a mito- In particular, the pattern of TCDD-induced changes in focal genic stimulus from TCDD but resist any homeostatic checks labeling and apoptotic indices seen by Stinchcombe et al. on their growth. Thus, at high doses of TCDD, the volume (1995) was consistent with the parameter estimates obtained fraction of B cells expands rapidly and, moreover, the B cell with the two-cell model. Second, the hepatic foci and the population has a high division rate which would presumably dose surrogates were measured in different experiments in predispose these cells to additional mutations. By contrast, different laboratories at different times (Pitot et al., 1987; A cells respond the homeostatic signals at all doses studied. Bars and Elcombe, 1991; Tritscher et al., 1992). A better test The implications of this model for extrapolation below the of the proposed linkage would be provided by simultaneous doses used by Pitot et al. (1987) are also interesting. Hepatic measurement of foci and dose surrogates in the same experilinear induction of CYP1A1 and CYP1A2 by TCDD is highly non- ment. Finally, the dose metrics used may not be optimum. (Vanden Heuval et al., 1994; Andersen et al., 1997). With a five-compartment liver, subcompartment 5 represents To the extent that the dose response for CYP induction pre- 6.8% of the liver. In the present parameterization, the dose dicts the dose response for focus growth, promotional efmetric for the low-dose segment of the dose response curves fects of TCDD will also be highly nonlinear. The question is derived from the behavior of this single subcompartment. of whether or not the degree of nonlinearity leads to an By using an acinar model with a larger number of compartmust remain open until additional data become available. effective dose threshold for promotional effects of TCDD ments, the size of this compartment would be reduced and the dose response relationship for this dosimeter, induction The significance of these comments on the risk assessment of CYP gene expression in the most sensitive liver compartpreted. It is not clear that the protocol used by Pitot et al. implications of the two-cell model should not be overinter- ment, would be shifted to lower doses. While a more systematic fitting process could be done to estimate the best parameand TCDD promotion, is a good analog of the bioassay de- (1987), consisting of partial hepatectomy, DEN initiation, ter for each part of the curve expressed as proportion of liver induced, this would be extremely computer-intensive with scribed by Kociba et al. (1978). Further, as described under the simulation model and primarily useful for hypothesis Methods, fitting the two-cell model to the data did not use testing. As more detailed studies of the molecular mechathe fitting exercise (Table 2), while biologically plausible, formal methods and the parameter estimates obtained from nisms of promotion become available, suggesting optimum dose metrics for the proportion of liver affected based on may not provide the best possible fit of the model to the data. more biological criteria may be possible. In summary, (1) the two-cell model provides a biologically The dose surrogates based on predicted induction of plausible description of the TCDD dose response curves for CYP1A1 and CYP1A2 should not be interpreted as suggesting hepatic foci described by Pitot et al. (1987), requiring only that TCDD affect the division and death rates of the cells in the that cytochrome P450 induction is necessarily part of the foci and (2) the TCDD dose response for regional hepatic mechanism linking TCDD exposure with promotion of heinduction of CYP1A1 and CYP1A2 shows promise as a quantipatic foci. Rather, the surrogates are assumed to portray typitative predictor of the dose response for putative preneoplastic cal dioxin dose-expression of gene induction response behavfoci in rodent initiation promotion experiments. iors for the as yet unknown genes directly involved in the growth of hepatic foci. Thus, the observation of induced cells in the centrilobular area is assumed to be an indicator of the APPENDIX effects of dioxin in the same dose range on other processes The fraction of cells in a tissue that are either in the cell elsewhere in the liver and, perhaps, elsewhere in the body. cycle or dying ( f act ) is given by Finally, given that TCDD is hepatocarcinogenic in female rats (Kociba et al., 1978), it is interesting to consider the f act Å cctra / aptrb, (1A)