high doses of chemicals perturb the body or its defensive mechanisms in a qualitatively different manner from lower doses of the same substances. The MTD is deliberately designed to be close to the lower border of toxicity, so it is logical to consider whether some aspect of toxicity promotes carcinogenicity. Few studies have been conducted to address that issue directly, i.e., by testing proposed mechanistic hypotheses in long-term animal studies. In two analyses of NTP studies, Hoel et al. (1988) and Tennant et al. (1991) reported only partial correlation between the sites and types of toxic effects measured in conventional toxicologic studies and the development or lack of development of cancer. It is possible that a less conspicuous component of toxic responses, such as changes in mutation rates, or toxic responses measured much earlier in the bioassay, such as induced cell proliferation, will provide supporting evidence that toxicity provokes cancer, but much more research will be needed.
In addition to indicating whether a chemical is a carcinogen in rodents at high doses, a test performed at the MTD yields information about the carcinogenic potency of the chemical in rodents. Potency is a function of both the dose and the magnitude of the observed response. One chemical is judged to have a higher potency than another if the percentage of animals that develops tumors at a given dose is higher than for the other chemical. Current procedures typically involve testing about 50 animals per sex per species per dose. If the background incidence of tumors were 10% (5/50) in controls, the minimum statistically significant response at the MTD would be 20% (10/50). The highest response would, of course, be 100% (50/50). Clearly, a 100% response would indicate a higher potency for the test chemical than a 20% response. Knowledge about the quantitative response obtained in the bioassay enables scientists to make estimates of relative carcinogenic potency and adds information beyond the simple identification of a substance as a likely carcinogen or noncarcinogen.
Current bioassays that use the MTD and one or two lower doses provide some limited information about the shape of the high dose portion of the dose-response curve. The shape of the curve at high doses
might or might not have relevance to low dose exposures, however, depending on the reliability of the qualitative assumptions that have been made. The current standard regulatory practice of estimating ''plausible upper bounds" on risk often applies the linearized multistage model (Crump, 1984) to bioassay data obtained with the MTD. That procedure is based on the assumption of a linear relationship between tumor response and dose at low doses. Because of the linearity assumption, estimates of low dose risk obtained with this procedure are often not very sensitive to the observed dose-response shape in the experiment. It is important to note that the linear relationship cannot be verified directly, nor can it be verified that the estimate so computed provides a true upper bound on risk at very low doses. Nevertheless, the linearized multistage procedure has been widely used by regulatory agencies and is thought to provide an objective basis for decisions concerning regulation of chemicals found, in high dose bioassays, to increase tumor frequencies in animals. In particular, the procedure allows a crude rank ordering of animal carcinogens from most potent to least potent, which might then provide a basis for priorities in regulation and pollution prevention.
One of the reasons that MTD bioassays provide little information on the shape of the dose-response curve is that it uses a small number of doses. The same number of groups and numbers of animals per group tested at lower doses would probably yield even less information about the shape of the dose-response curve or about carcinogenicity, however, because testing at lower doses decreases the likelihood of response in a small experiment or the likelihood that a response will be distinguishable from background. If the response cannot be distinguished from background, no information useful for defining the shape of the curve is obtained. Although most earlier NTP and NCI studies exposed animals only at the MTD and MTD/2, more recent studies have also exposed animals at MTD/4. Among 38 positive responses in sex- and species-specific studies, 23 (61 %) would have been positive if MTD/4 had been the largest dose. Seven of the 23 would have detected all site-specific responses, and the other 16 would have detected some, but not all, site-specific effects (R. Griesemer, NIEHS, pers. comm., 1991); even the use of low experimental doses in standard bioassays can provide useful dose-response information (i.e., result in responses that are distinguishable from background). Information on the shape of the dose-response curve below the range of the chronic bioassay is more likely to be ob-
tained from experiments targeted at elucidating biologic mechanisms of action and characterizing the dose-response characteristics of the critical events, such as DNA-adduct formation or induced cell proliferation.
There are cases, of course, where extrapolating from high to low doses can be irrelevant. Some human exposures levels can be directly compared with dosing at the MTD; and in some cases, the dose rate might be comparable, especially with high dose occupational exposures (Gold et al., 1987b) and many pharmaceutical exposures. In these cases, the human dose rates are very similar to those used in rodent bioassays, and the focus of supplementary testing would be to elucidate biologic mechanisms to determine the human relevance of bioassay results, not the validity of low dose extrapolation.
Another problem with the use of the MTD is that, although it is now included in most carcinogenicity bioassays, criteria for selecting the EMTD and evaluating the selection vary among laboratories. In bioassays conducted by the NTP, the highest dose tested is the EMTD estimated from a 90 day study, and sufficient data are presented to determine whether the MTD was achieved. However, other published bioassay results might not state the rationale for dose selection and might insufficiently report toxicity data and other data needed to determine whether the MTD was achieved.
The usefulness of reports of bioassay results for regulators and risk managers could be increased by including several pieces of information: a clearly stated rationale for dose selection and a summary of the toxicity information important in evaluating the dose selection, especially whether the animals could have tolerated a higher dose and whether the high dose used elicited toxicity. Such considerations should be included in a risk assessment of a substance. The potential for reduced statistical power of studies in which the MTD was not achieved, compared with studies in which the MTD was achieved, should be noted and accounted for, or at least acknowledged in a risk assessment.