on 22 of 29 chemicals considered previously by Ennever (1987) for which positive rodent bioassay data is available. These chemicals are of interest in that the epidemiological data did not provide unequivocal evidence of carcinogenic effects in humans: many were in fact in category 3 (i.e., not classifiable with respect to human carcinogenicity) within the classification scheme used by the International Agency for Research on Cancer (1987), and none were in category 1 (sufficient evidence of carcinogenicity in humans). Based on their re-analysis of this data, Goodman & Wilson (1991b) argued that the excess risks observed in epidemiological studies (which may or may not have been significantly elevated) were roughly consistent with predictions based on potency values in the CPDB. Goodman & Wilson (1991a) recently reviewed interspecies comparisons of carcinogenic potency and concluded that ''there is a good correlation of the carcinogenic potencies between rats and mice, and the upper limits on potencies in humans are consistent with rodent potencies for those chemicals for which human exposure data are available."
The completion of a large number of laboratory studies of the carcinogenic potential of chemicals has afforded an opportunity to evaluate the variation in the potency of chemical carcinogens. The Carcinogenic Potency Database developed by Gold et al. (1984) provides a convenient summary not only of the data from nearly 4,000 individual experiments, but also of the potency of chemical carcinogens expressed in terms of the TD50. The TD50s in the CPDB indicate that carcinogenic potency may vary by nearly 10 million-fold.
Several investigators have reported a strong correlation between the maximum dose tested (MDT) in carcinogen bioassay and the TD50, which generally corresponds to the maximum tolerated dose (MTD). In particular, the estimate of the TD50 based on the one-hit model can be shown, using both theoretical and empirical arguments, to be restricted to lie within a factor of about 32-fold of the MTD. Empirical evidence indicates that measures of carcinogenic risk at low doses, such as the value of q1 in the linearized multistage model, are also correlated with the MTD, suggesting that preliminary estimates of low dose cancer risk
may be based on an estimate of the MTD. Specifically, Gaylor (1989) has shown that dividing the MTD by a factor of 380,000 will approximate the 10-6 RSD obtained from bioassay data using the linearized multistage model.
Carcinogenic potency has also been shown to be somewhat correlated with both acute toxicity and mutagenicity, both of which are important factors in neoplastic change. In particular, target tissue toxicity may lead to proliferation of preneoplastic cells, and hence increase the pool of cells available for malignant transformation. Travis et al. (1991) have demonstrated a strong correlation between a composite index based on toxicity and mutagenicity and carcinogenic potency as measured by the TD50. These results suggest that data on toxicity and mutagenicity may be combined to reduce the uncertainty in the carcinogenic potential of chemicals not yet subjected to long-term carcinogen bioassay.
The apparent correlation between acute toxicity and carcinogenicity does not imply a causal relationship between toxicity and carcinogenicity. The establishment of a causal relationship between toxicity and carcinogenicity presupposes a biological relationship between these two end points. In this regard, Hoel et al. (1988) noted little association between toxic tissue injury and neoplastic change in NTP studies. Clayson & Clegg (1991), however, discuss specific examples in which toxicity plays an important role in carcinogenesis. Parodi et al. 91982b) note that covalent binding with macromolecules, which can influence the mutagenic potency of chemicals, can also induce toxicity in some cases.
While these empirically derived correlations are of considerable interest, a clear interpretation of these findings in either biological or statistical terms remains to be accomplished. To be biologically meaningful, the rationale for such associations should be toxicologically plausible. While toxic, mutagenic, and carcinogenic effects do share certain characteristics in common, each of these processes is sufficiently complex to cast doubt on a causal relationship between simple measures of toxic and mutagenic potential and carcinogenic potency. Statistically, correlations between the MDT and the TD50 occur as a result of the narrow range of possible potency values within a single experiment in relation to the wide variation observed in the potency of chemical carcinogens. This has led to suggestions that the observed correlation between the MTD and the TD50 may simply be an artifact of the experimental designs currently used in carcinogen bioassay. In this regard, Reith & Starr
(1989a) concluded that "the chronic rodent bioassay, in and of itself, is altogether inadequate as a data source for estimating the risk to humans from exposure to carcinogenic agents".
In our view, correlations between the MTD and measures of cancer potency reflect the limited amount of information on cancer risks provided by carcinogen bioassay data. Once the MTD has been determined, TD50 and q1* values are somewhat insensitive to the experimental results and are constrained to lie within a narrow range, particularly when viewed in light of the eight order of magnitude variation in TD50 values for chemical carcinogens. This does not imply that estimates of carcinogenic potency based on bioassay data are not meaningful, but does demonstrate that both the TD50 and q1* represent relatively crude indicators of risk. At the same time, however, the value of q1* does represent the smallest possible linearized upper bound on low dose risk based on the multistage model which is consistent with the experimental data. The TD50, moreover, represents a dose which has been shown, often without the need for extrapolation outside of the observable response range, to reduce the proportion of tumor-free animals by one-half.
Measures of carcinogenic potency such as the TD50 have also been shown to be highly correlated between different rodent species (rats and mice). Although this appears to offer support for quantitative interspecies extrapolation of cancer bioassay data, it is possible that this correlation may be largely due to the high correlation between the MTDs for different rodent species. Kaldor et al. (1988) have suggested that because of the relationship between animal LD 50s and the doses of antineoplastic agents used in cancer chemotherapy, the apparent correlation in potency of these agents in animals and humans may be explained in part by toxicity considerations. Despite this correlation, the error associated with quantitative interspecies extrapolations of carcinogenic potency values can be 100-fold or greater.
Imperfect qualitative agreement between species also suggests the need for caution in quantitatively extrapolation between species (Freedman & Zeisel, 1988). Although all known human carcinogens are also carcinogenic in animals (Tomatis et al., 1989), concordance between rats and mice with chemicals tested in the U.S. National Toxicology Program is only about 74% (Haseman & Huff, 1987). Gold et al. (1989) subsequently reported on overall concordance between rats and mice of 76% for 392 chemicals in the CPDB. Piegorsch et al. (1992) note that