tween dose and response. The default assumption, in the absence of any mechanistic or mode-of-action information, should be that a threshold or low-dose nonlinear dose-response relationship exists for health effects other than cancer; that is, that for noncancer health effects, there is a dose below which concern is not warranted. The default assumption for cancer is a linear low-dose extrapolation. This principle is important in considering the relevance of the dose of dietary supplements causing harm, as a linear dose extrapolation leads to the conclusion that any amount of the substance is a risk. The rationale for a linear or nonlinear assumption is not without its detractors, but it is an established principle used in risk assessment of other ingested substances (Rodricks et al., 2001) that should be applied to dietary supplements as well.

In addition to the pharmacokinetic differences described above, there are several well-described examples of pharmacodynamic differences between animals and humans, that is, differences in how a chemical affects the body (Klaassen, 2001). For example, while rodent carcinogenicity studies are often predictive for human carcinogenesis from chemicals (Huff, 1999; Rodricks et al., 2001; Tomatis, 2001), some target sites in rats and mice have been questioned as relevant endpoints for human risk assessment (Capen et al., 1999; Rodricks et al., 2001). Examples include kidney toxicity/carcinogenicity in male rats related to the production of alpha-2-globulin (Rodricks et al., 2001), liver toxicity/carcinogenicity in rodents related to peroxisome proliferation (Rodricks et al., 2001), thyroid toxicity/carcinogenicity in rats (Capen et al., 1999), and bladder tumors in rats caused by terephthalate acid or cyclamate (IARC, 1999). When these specific endpoints are observed, they raise significant questions regarding relevance to humans. Such findings, or others that suggest irrelevance of the particular animal study evidence to humans, should be used to reach conclusions about possible human toxicity only after careful review. In the absence of specific evidence that certain animal study findings are irrelevant to humans, animal evidence should be used to evaluate potential human risk.

When interpreting a substance’s effects or lack of effect in animal studies, it is important to remember the variability among humans in their sensitivity to toxic effects from ingested substances. Some members of the human population are more sensitive than the so-called average (Hayes, 2001), an issue best captured under the concept of “natural variability in response,” a well-documented phenomenon. Many of these differences are