ToxicologyLetters, 66/65 (1992) 631-636 Q 1992 Elsevier Science Publishers B.V., All rights reserved 03784274/92/$5.00
Risk assessment of non-genotoxic
631
carcinogens
Suresh H. Moolgavkar and E. Georg Luebeck Fred Hutchinson Cancer Research Center, Division of Public Health Sciences, Seattle, WA (VSN Key words: Somatic mutation; Carcinogenesis; Cell division; Mutation; Differentiation; Apoptosis; Hepatocarcinogenesis; Altered foci
SUMMARY Rates of cell proliferation, cell death, and cell differentiation affect the risk of cancer profoundly. An increase in cell proliferation rates leads to an increase in mutation rates per unit of time, which, in turn, leads to an increase in the risk of cancer. An increase in cell division rates relative to death or differentiation rates may lead to an increase in the population of critical target cells, which, again, leads to an increase in cancer risk These fundamental principles are well illustrated by the rodent liver model for carcinogenesis. In this paper, we shall briefly discuss some of the consequences of incorporating cell proliferation kinetics into quantitative models of cancer risk assessment. Consideration of cell kinetics can shed light on apparently paradoxical observations such as, e.g. the observation that the administration of two different promoters may lead to the same volume fraction in the rodent liver, with one promoter giving rise to a large number of small foci, and the other to a small number of large foci. Some consequences of explicitly considering cell proliferation kinetics in malignant foci are briefly discussed.
INTRODUCTION
Recent laboratory advances provide strong support for the somatic mutation theory of carcinogenesis. At the same time there is accumulating evidence that cell proliferation kinetics play an important role in the process of malignant transformation Ill in at least two different ways. First, an increase in cell division rates leads to an increase in the rates of mutation per unit of time. Thus, an increase in cell division rates may lead to an increase in the rates of mutations critical to carcinogenesis. Second, increase in cell division rates relative to cell death or differentiation rates may lead to an increase in the population of critical target cells, thus increasing the Correspondence to: S.H. Moolgavkar, Fred Hutchinson Cancer Research Center, Division of Public Health Sciences, 1124 Columbia Street, Seattle, WA 98104, USA.
632
probability that one of these target cells will sustain the mutations required for malignant transformation. Promotion, for example, is believed to cause such an expansion of initiated cells. Current risk assessment procedures are based on an obsolete paradigm, which recognizes the importance of somatic mutations in carcinogenesis, but ignores completely the role of cell proliferation. One defense of these procedures is that they are conservative and, therefore, protect the public health. However, it is now time to explore other methods based on better understanding of the process of carcinogenesis, and it seems clear that any such methods will have to take explicit account of cell proliferation kinetics and their disruption by putative carcinogenic agents. The introduction of cell proliferation kinetics into dose-response models of carcinogenesis has some very interesting consequences, which we will briefly explore in this paper. There is now clear-cut evidence that programmed cell death, apoptosis, is just as important as cell division in determining cancer risk. Some promoters may inhibit apoptosis 121, and some oncogenes may do the same [31. When considering cell proliferation kinetics in models of carcinogenesis, there is the wide-spread misconception [ll that it is sufficient to consider the net proliferation rate, i.e., the difference between cell division and cell death (apoptosis). This is false. Explicit consideration of cell death has profound implications for cancer risk estimation. Similarly, some so-called simulation models consider only the mean number of cells in each of the stages on the pathway to malignancy. This again leads to erroneous results: stochastic considerations cannot be ignored. THE MODEL
Most of our conclusions are general, and do not depend upon assuming a specific model of carcinogenesis. However, quantitative cancer risk assessment must be based on a specific model, and it is convenient to couch the discussion in terms of this model. The minimal model postulates two ratelimiting steps on the pathway to carcinogenesis. The first of these steps may be identified with initiation, the second with malignant conversion. Promotion may be identified with clonal expansion of initiated cells. The model is shown in Figure 1, and the main features of the model are described below. Initiation and Conversion The model assumes that normal target cells are transformed into cancer cells via an intermediate stage in two rate limiting, irreversible and hereditary (at the level of the cell) steps. Intermediate cells are assumed to be generated from normal cells as a non-homogeneous Poisson process with
633
.*
l:.
P t
VX
,’
@
conveFon@
progresskn
.
initiation
A70
I
promotion Fig. 1. Pictorial representation of the two-mutation model.
intensity I&)X(S), where Y(S) can be thought of as the rate of initiation per cell per unit time and X(s) is the number of normal cells at risk at time s. The second step is the conversion of intermediate cells into malignant cells with rate p(s) per cell per unit time. Precisely, an intermediate cell divides into one intermediate cell and one malignant cell with rate p(s). Promotion The salient feature of promotion is assumed to be the growth (clonal expansion) of intermediate cells as a stochastic birth-death process with cell division rate a(s) and death (differentiation) rate B(s). An agent is a promoter if it increases the net growth rate (a-(3) and an anti-promoter if it decreases it. Note that promoters or anti-promoters may affect either the cell division rate or the cell death rate or both. Details of the model can be found in recent publications [4,51. SOME CONSEQUENCES OF CONSIDERING CELL DIVISION AND DEATH [6-S]
An important property of populations of cells undergoing cell division and cell death is that the population may become extinct. Thus, if the rate of apoptosis is greater than zero, then an initiated cell will die without giving rise to a detectable lesion, such as a papilloma on the skin or an altered focus in the liver, with probability greater than zero. This conclusion may come as a surprise because the irreversibility of initiation appears to be current dogma. However, we are not asserting here that individual initiated cells revert to normal, but rather that, because initiated cells may die, initiation may be partially reversible on the level of the organ.
634
If the rates of cell division and apoptosis are constant (independent of time), then the probability that an initiated cell and all its progeny will die is given (asymptotically) by the ratio of the rate of apoptosis and the rate of cell division. If the rate of apoptosis is larger than the cell division rate, then the (asymptotic) probability of extinction is 1. However, some foci may still be visible because of the stochastic nature of the process. In recent analyses of altered hepatic foci in rodent hepatocarcinogenesis experiments, we concluded that most initiated cells (perhaps up to 90%) die without giving rise to altered foci. Some preliminary data on the placental form of Glutathione S-Transferase (GST-P) positive cells appear to support this estimate [9, Schulte-Hermann, personal communicationl. The mean number of initiated cells at any time depends upon the rate of initiation and the net rate of intermediate cell division, a+ However, there is considerable stochastic variation around this mean number, and the actual number depends upon a and l3individually, and not just upon their difference. Further, the distribution of altered cells in foci also depends upon a and p individually. Thus, for a given value of a-p, large values of a and p lead to small numbers of large foci, and small values of a and p lead to large numbers of small foci. Consider an hypothetical example. Suppose a+ = 0.01 per cell per day, and suppose that one has the following two combinations of parameters: a = 0.5, f3 = 0.49 and a = 0.1, l3 = 0.09. Both these combinations of parameters lead to a+ = 0.01 and thus to the same mean number of initiated cells provided that the rates of initiation are identical. However, the first set of parameters will lead to a small number of large foci, whereas the second set will lead to a large number of small foci. All other things being equal, the first combination of parameters carries a higher risk of malignant transformation than the second. This is because a high cell division rate implies a high mutation rate. Examples of the phenomenon described here are provided by promoters such as 4-DAB and the peroxisome proliferators, which lead to a small number of large foci, and others such as NDEOL and phenobarbital, which lead to a large number of small foci. By measuring labelling indices, it should be possible to confirm that division rates in foci associated with the former compounds are higher than the division rates in foci associated with the latter compounds. The incidence of malignant tumors depends upon both a and p, and not just on their difference. One (biological) reason for this was pointed out above: a large cell division rate implies a large mutation rate. However, even if the mutation rates are assumed to be independent of cell division rates, the incidence function depends upon both a and p individually. This is a simple mathematical consequence of the model. Thus simulations that take into account only the mean behavior of cells in the intermediate compartment 111lead to erroneous results for the incidence of malignant tumors.
635
Most cancer models used for risk assessment, including the two-stage model described here, make the assumption that the first malignant cell that is generated in a tissue gives rise to a malignant tumor after a suitable lag time. However, cell death and differentiation occur in malignant tumors, and considerations such as those described above should apply to malignant tumors as well as to pre-malignant lesions. As discussed above, one implication of considering cell death in malignant tumors is that some malignant cells must become extinct without giving rise to malignant tumors. Suppose that the cell division rate is a and that the death rate is p. Then, if lifetime is assumed to be infinite, the distribution of the number of malignant cells generated up to and including the first one that grows into a malignant tumor is geometric with parameter 1 - @/a). The average number of mahgnant cells that must be generated for one to grow into a malignant tumor is (1 - @/a)}-‘. Sin c e 1i f et ime is not infinite, this distribution is truncated, and the actual average number is smaller than the average given above. We have simulated the appearance of malignant tumors in the rodent liver. Using simulated data that take into account birth and death processes in pre-malignant and malignant lesions, we used likelihood methods to fit the two mutation models to the data (with the usual assumption that the first malignant cell developed into a malignant tumor after a suitable lag time). Although our results are still preliminary, we find that the main effect of ignoring the growth kinetics of malignant cells is to lower considerably the estimated rate of initiation. Surprisingly, the estimates of the growth kinetics of intermediate cells and the second mutation rate appear to be little affected. Thus, ignoring the proliferation kinetics of malignant cells can yield misleading results, although the extent of the bias introduced in the parameter estimates needs to be determined in more detailed simulation studies. CONCLUDING
REMARKS
Although there is little doubt that carcinogenesis is a multistage process, the number and nature of the stages involved are not completely understood for any neoplasm, with the possible exception of retinoblastoma. Recent laboratory work suggests that many specific genetic changes are associated with carcinogenesis. However, it is not known which of these changes are necessary. It is also clear that cell proliferation kinetics play a major role in carcinogenesis, and in this paper we have tried to summarize some of the important consequences of incorporating cell proliferation kinetics into models of carcinogenesis. We have emphasized, in particular, the role of apoptosis. Mechanistic models of carcinogenesis should consider cell division and apoptosis explicitly. It is not sufficient to consider only net proliferation rates.
636
Our discussion in this paper is couched in terms of a specific two-mutation model for carcinogenesis. However, our qualitative conclusions are not dependent upon the choice of this particular model. Quantitative conclusions, on the other hand, will clearly depend upon the model used for data analysis. In particular, assumptions regarding the number of stages in the carcinogenic process and the kinetics of cell division and cell death will clearly influence the estimates of parameters. Unfortunately, we do not yet have the biological information to determine how many rate limiting steps are required for specific malignancies, and how cell division and death rates may vary with time and dose of the agent under study. Until such time as this information becomes available it would seem prudent to use the simplest possible model that is consistent with the data. We believe that the two-mutation model presented here is the most parsimonious model consistent with epidemiologic and experimental data, and allows the explicit incorporation of cell proliferation kinetics into cancer risk assessment. REFERENCES Cohen, S.M. and Ellwein, L.B. (1990) Cell proliferation in carcinogenesis. Science 249, 1007-1011. Schulte-Hermann, R., Parzefall, W., Bursch, W., Timmerman-Trosiener, I. (1989) Hepatocarcinogenesis by non-genotoxic compounds. In: C. Travis (Ed.), Biologically Based Methods for Cancer Risk Assessment. Plenum, New York. pp. 155-163. Hockenberry, D., Nunez, G., Milliman, C., Schreiber, R.D. and Korsmeyer, S.J. (1983) Bcl-2 is an inner mitochondrial membrane protein that blocks programmed cell death. Nature 348,334-336. Moolgavkar, S.H. and Knudson, A.G. (1981) Mutation and cancer: A model for human carcinogenesis. J Natl. Cancer Inst. 66,1037-1052. Moolgavkar, S.H. and Luebeck, E.G. (1990) Two-event model for carcinogenesis: Biological, mathematical and statistical considerations. Risk Anal. 10,323341. Dewanji, A., Venzon, D.J. and Moolgavkar, S.H. (1989) A stochastic two-stage model for cancer risk assessment II. The number and size of premalignant clones. Risk Anal. 9, 179-187. Luebeck, E.G. and Moolgavkar, S.H. (1991) Stochastic analysis of intermediate lesions in carcinogenesis experiments. Risk Anal. 11,149-157. Moolgavkar, S.H., Luebeck, G., de Gunst, M., Port, R.E. and Schwarz, M. (1990) Quantitative analysis of enzyme altered foci in rat hepatocarcinogenesis experiments. Carcinogenesis 11,1271-1278. Satoh, K., Hatayama, I., Tateoka, N., Tamai, K., Shimizu, T., Takematsu, M., Ito, N. and Sato, K. (1989) Transient induction of single GST-P positive hepatocytes by DEN. Carcinogenesis 10,2107-2111.
Risk assessment of non-genotoxic
631
carcinogens
Suresh H. Moolgavkar and E. Georg Luebeck Fred Hutchinson Cancer Research Center, Division of Public Health Sciences, Seattle, WA (VSN Key words: Somatic mutation; Carcinogenesis; Cell division; Mutation; Differentiation; Apoptosis; Hepatocarcinogenesis; Altered foci
SUMMARY Rates of cell proliferation, cell death, and cell differentiation affect the risk of cancer profoundly. An increase in cell proliferation rates leads to an increase in mutation rates per unit of time, which, in turn, leads to an increase in the risk of cancer. An increase in cell division rates relative to death or differentiation rates may lead to an increase in the population of critical target cells, which, again, leads to an increase in cancer risk These fundamental principles are well illustrated by the rodent liver model for carcinogenesis. In this paper, we shall briefly discuss some of the consequences of incorporating cell proliferation kinetics into quantitative models of cancer risk assessment. Consideration of cell kinetics can shed light on apparently paradoxical observations such as, e.g. the observation that the administration of two different promoters may lead to the same volume fraction in the rodent liver, with one promoter giving rise to a large number of small foci, and the other to a small number of large foci. Some consequences of explicitly considering cell proliferation kinetics in malignant foci are briefly discussed.
INTRODUCTION
Recent laboratory advances provide strong support for the somatic mutation theory of carcinogenesis. At the same time there is accumulating evidence that cell proliferation kinetics play an important role in the process of malignant transformation Ill in at least two different ways. First, an increase in cell division rates leads to an increase in the rates of mutation per unit of time. Thus, an increase in cell division rates may lead to an increase in the rates of mutations critical to carcinogenesis. Second, increase in cell division rates relative to cell death or differentiation rates may lead to an increase in the population of critical target cells, thus increasing the Correspondence to: S.H. Moolgavkar, Fred Hutchinson Cancer Research Center, Division of Public Health Sciences, 1124 Columbia Street, Seattle, WA 98104, USA.
632
probability that one of these target cells will sustain the mutations required for malignant transformation. Promotion, for example, is believed to cause such an expansion of initiated cells. Current risk assessment procedures are based on an obsolete paradigm, which recognizes the importance of somatic mutations in carcinogenesis, but ignores completely the role of cell proliferation. One defense of these procedures is that they are conservative and, therefore, protect the public health. However, it is now time to explore other methods based on better understanding of the process of carcinogenesis, and it seems clear that any such methods will have to take explicit account of cell proliferation kinetics and their disruption by putative carcinogenic agents. The introduction of cell proliferation kinetics into dose-response models of carcinogenesis has some very interesting consequences, which we will briefly explore in this paper. There is now clear-cut evidence that programmed cell death, apoptosis, is just as important as cell division in determining cancer risk. Some promoters may inhibit apoptosis 121, and some oncogenes may do the same [31. When considering cell proliferation kinetics in models of carcinogenesis, there is the wide-spread misconception [ll that it is sufficient to consider the net proliferation rate, i.e., the difference between cell division and cell death (apoptosis). This is false. Explicit consideration of cell death has profound implications for cancer risk estimation. Similarly, some so-called simulation models consider only the mean number of cells in each of the stages on the pathway to malignancy. This again leads to erroneous results: stochastic considerations cannot be ignored. THE MODEL
Most of our conclusions are general, and do not depend upon assuming a specific model of carcinogenesis. However, quantitative cancer risk assessment must be based on a specific model, and it is convenient to couch the discussion in terms of this model. The minimal model postulates two ratelimiting steps on the pathway to carcinogenesis. The first of these steps may be identified with initiation, the second with malignant conversion. Promotion may be identified with clonal expansion of initiated cells. The model is shown in Figure 1, and the main features of the model are described below. Initiation and Conversion The model assumes that normal target cells are transformed into cancer cells via an intermediate stage in two rate limiting, irreversible and hereditary (at the level of the cell) steps. Intermediate cells are assumed to be generated from normal cells as a non-homogeneous Poisson process with
633
.*
l:.
P t
VX
,’
@
conveFon@
progresskn
.
initiation
A70
I
promotion Fig. 1. Pictorial representation of the two-mutation model.
intensity I&)X(S), where Y(S) can be thought of as the rate of initiation per cell per unit time and X(s) is the number of normal cells at risk at time s. The second step is the conversion of intermediate cells into malignant cells with rate p(s) per cell per unit time. Precisely, an intermediate cell divides into one intermediate cell and one malignant cell with rate p(s). Promotion The salient feature of promotion is assumed to be the growth (clonal expansion) of intermediate cells as a stochastic birth-death process with cell division rate a(s) and death (differentiation) rate B(s). An agent is a promoter if it increases the net growth rate (a-(3) and an anti-promoter if it decreases it. Note that promoters or anti-promoters may affect either the cell division rate or the cell death rate or both. Details of the model can be found in recent publications [4,51. SOME CONSEQUENCES OF CONSIDERING CELL DIVISION AND DEATH [6-S]
An important property of populations of cells undergoing cell division and cell death is that the population may become extinct. Thus, if the rate of apoptosis is greater than zero, then an initiated cell will die without giving rise to a detectable lesion, such as a papilloma on the skin or an altered focus in the liver, with probability greater than zero. This conclusion may come as a surprise because the irreversibility of initiation appears to be current dogma. However, we are not asserting here that individual initiated cells revert to normal, but rather that, because initiated cells may die, initiation may be partially reversible on the level of the organ.
634
If the rates of cell division and apoptosis are constant (independent of time), then the probability that an initiated cell and all its progeny will die is given (asymptotically) by the ratio of the rate of apoptosis and the rate of cell division. If the rate of apoptosis is larger than the cell division rate, then the (asymptotic) probability of extinction is 1. However, some foci may still be visible because of the stochastic nature of the process. In recent analyses of altered hepatic foci in rodent hepatocarcinogenesis experiments, we concluded that most initiated cells (perhaps up to 90%) die without giving rise to altered foci. Some preliminary data on the placental form of Glutathione S-Transferase (GST-P) positive cells appear to support this estimate [9, Schulte-Hermann, personal communicationl. The mean number of initiated cells at any time depends upon the rate of initiation and the net rate of intermediate cell division, a+ However, there is considerable stochastic variation around this mean number, and the actual number depends upon a and l3individually, and not just upon their difference. Further, the distribution of altered cells in foci also depends upon a and p individually. Thus, for a given value of a-p, large values of a and p lead to small numbers of large foci, and small values of a and p lead to large numbers of small foci. Consider an hypothetical example. Suppose a+ = 0.01 per cell per day, and suppose that one has the following two combinations of parameters: a = 0.5, f3 = 0.49 and a = 0.1, l3 = 0.09. Both these combinations of parameters lead to a+ = 0.01 and thus to the same mean number of initiated cells provided that the rates of initiation are identical. However, the first set of parameters will lead to a small number of large foci, whereas the second set will lead to a large number of small foci. All other things being equal, the first combination of parameters carries a higher risk of malignant transformation than the second. This is because a high cell division rate implies a high mutation rate. Examples of the phenomenon described here are provided by promoters such as 4-DAB and the peroxisome proliferators, which lead to a small number of large foci, and others such as NDEOL and phenobarbital, which lead to a large number of small foci. By measuring labelling indices, it should be possible to confirm that division rates in foci associated with the former compounds are higher than the division rates in foci associated with the latter compounds. The incidence of malignant tumors depends upon both a and p, and not just on their difference. One (biological) reason for this was pointed out above: a large cell division rate implies a large mutation rate. However, even if the mutation rates are assumed to be independent of cell division rates, the incidence function depends upon both a and p individually. This is a simple mathematical consequence of the model. Thus simulations that take into account only the mean behavior of cells in the intermediate compartment 111lead to erroneous results for the incidence of malignant tumors.
635
Most cancer models used for risk assessment, including the two-stage model described here, make the assumption that the first malignant cell that is generated in a tissue gives rise to a malignant tumor after a suitable lag time. However, cell death and differentiation occur in malignant tumors, and considerations such as those described above should apply to malignant tumors as well as to pre-malignant lesions. As discussed above, one implication of considering cell death in malignant tumors is that some malignant cells must become extinct without giving rise to malignant tumors. Suppose that the cell division rate is a and that the death rate is p. Then, if lifetime is assumed to be infinite, the distribution of the number of malignant cells generated up to and including the first one that grows into a malignant tumor is geometric with parameter 1 - @/a). The average number of mahgnant cells that must be generated for one to grow into a malignant tumor is (1 - @/a)}-‘. Sin c e 1i f et ime is not infinite, this distribution is truncated, and the actual average number is smaller than the average given above. We have simulated the appearance of malignant tumors in the rodent liver. Using simulated data that take into account birth and death processes in pre-malignant and malignant lesions, we used likelihood methods to fit the two mutation models to the data (with the usual assumption that the first malignant cell developed into a malignant tumor after a suitable lag time). Although our results are still preliminary, we find that the main effect of ignoring the growth kinetics of malignant cells is to lower considerably the estimated rate of initiation. Surprisingly, the estimates of the growth kinetics of intermediate cells and the second mutation rate appear to be little affected. Thus, ignoring the proliferation kinetics of malignant cells can yield misleading results, although the extent of the bias introduced in the parameter estimates needs to be determined in more detailed simulation studies. CONCLUDING
REMARKS
Although there is little doubt that carcinogenesis is a multistage process, the number and nature of the stages involved are not completely understood for any neoplasm, with the possible exception of retinoblastoma. Recent laboratory work suggests that many specific genetic changes are associated with carcinogenesis. However, it is not known which of these changes are necessary. It is also clear that cell proliferation kinetics play a major role in carcinogenesis, and in this paper we have tried to summarize some of the important consequences of incorporating cell proliferation kinetics into models of carcinogenesis. We have emphasized, in particular, the role of apoptosis. Mechanistic models of carcinogenesis should consider cell division and apoptosis explicitly. It is not sufficient to consider only net proliferation rates.
636
Our discussion in this paper is couched in terms of a specific two-mutation model for carcinogenesis. However, our qualitative conclusions are not dependent upon the choice of this particular model. Quantitative conclusions, on the other hand, will clearly depend upon the model used for data analysis. In particular, assumptions regarding the number of stages in the carcinogenic process and the kinetics of cell division and cell death will clearly influence the estimates of parameters. Unfortunately, we do not yet have the biological information to determine how many rate limiting steps are required for specific malignancies, and how cell division and death rates may vary with time and dose of the agent under study. Until such time as this information becomes available it would seem prudent to use the simplest possible model that is consistent with the data. We believe that the two-mutation model presented here is the most parsimonious model consistent with epidemiologic and experimental data, and allows the explicit incorporation of cell proliferation kinetics into cancer risk assessment. REFERENCES Cohen, S.M. and Ellwein, L.B. (1990) Cell proliferation in carcinogenesis. Science 249, 1007-1011. Schulte-Hermann, R., Parzefall, W., Bursch, W., Timmerman-Trosiener, I. (1989) Hepatocarcinogenesis by non-genotoxic compounds. In: C. Travis (Ed.), Biologically Based Methods for Cancer Risk Assessment. Plenum, New York. pp. 155-163. Hockenberry, D., Nunez, G., Milliman, C., Schreiber, R.D. and Korsmeyer, S.J. (1983) Bcl-2 is an inner mitochondrial membrane protein that blocks programmed cell death. Nature 348,334-336. Moolgavkar, S.H. and Knudson, A.G. (1981) Mutation and cancer: A model for human carcinogenesis. J Natl. Cancer Inst. 66,1037-1052. Moolgavkar, S.H. and Luebeck, E.G. (1990) Two-event model for carcinogenesis: Biological, mathematical and statistical considerations. Risk Anal. 10,323341. Dewanji, A., Venzon, D.J. and Moolgavkar, S.H. (1989) A stochastic two-stage model for cancer risk assessment II. The number and size of premalignant clones. Risk Anal. 9, 179-187. Luebeck, E.G. and Moolgavkar, S.H. (1991) Stochastic analysis of intermediate lesions in carcinogenesis experiments. Risk Anal. 11,149-157. Moolgavkar, S.H., Luebeck, G., de Gunst, M., Port, R.E. and Schwarz, M. (1990) Quantitative analysis of enzyme altered foci in rat hepatocarcinogenesis experiments. Carcinogenesis 11,1271-1278. Satoh, K., Hatayama, I., Tateoka, N., Tamai, K., Shimizu, T., Takematsu, M., Ito, N. and Sato, K. (1989) Transient induction of single GST-P positive hepatocytes by DEN. Carcinogenesis 10,2107-2111.
