Data derived from human chemical exposure studies allow researchers to avoid many of the uncertainties and problems that are inherent in interspecies extrapolations. High-quality human data are preferred by regulatory agencies for use in assessing the potential of chemicals to cause adverse health effects in exposed populations. This is the case for the Environmental Protection Agency (EPA; 1994), the Food and Drug Administration (FDA, 2000), Health Canada (Meek et al., 1994), and the World Health Organization (IPCS, 1994). As described in Chapter 1 of this report, uncertainties associated with animal data are reflected by the routine use of a 10-fold interspecies uncertainty factor when extrapolating from laboratory animals to humans. Using existing human data for risk assessment, of course, is dependent on the quality of the data. The Food Quality Protection Act (FQPA) of 1996 specifies that there should be “reasonable certainty of no harm” occurring from pesticide residues in foods. Pertinent, scientifically valid human data should provide those assessing risk the highest degree of certainty that they are being protective but not overly conservative by relying too heavily on default approaches.

Knowledge of chemical toxicity can be gained from several types of human studies. Intentional dosing studies of humans typically involve acute or short-term administration of low to moderate doses of drugs, vaccines, cosmetics, food additives, pesticides, or occupational or environmental agents. Doses of potential therapeutic agents may be high enough to elicit adverse effects in Phase 1 clinical trials, in order to adequately characterize their tolerability. Compounds suspected or known to be toxic are commonly administered to patient volunteers rather than healthy volunteers (FDA, 2002). Doses of occupationally and environmentally encountered chemicals may also be high enough to elicit reversible biochemical, physiological, or toxicological effects.

The intentional human dosing studies with pesticides reviewed by this committee involved low doses that produce no effects or minimal, reversible changes in sensitive biomarkers, albeit in one study the effect was sufficiently large to warrant termination of the study. Although epidemiological investigations of exposed populations may identify associations of adverse effects and chemical exposures and support inferences of cause and effect, epidemiological data are nonetheless usually limited by inadequate characterization of exposures and by an inability to recognize or control confounding factors (Dourson et al., 2001). Most clinical case reports of toxicant exposures have the same limitations. Such information, however, can alert us to previously unrecognized toxicities and identify critical effects to evaluate in subsequent investigations. Human cells and tissues can be very useful for metabolism and mode of action studies (MacGregor et al., 2001). Good correlation is often found between the metabolism of chemicals in vivo and metabolism by isolated hepatocytes of the same species (Oesch and Diener, 1995). Mechanistic studies with humans and laboratory animals may identify relevant toxicity end points and bioactive moieties and facilitate development of the most pertinent animal models (Jorkasky, 1998; Gregus and Klaassen, 2001).

Toxicological data from human exposure to pesticides and other chemicals are often limited or nonexistent. Obviously, one cannot administer sufficient amounts of a chemical to characterize the dose dependency of major adverse effects that exposed individuals could experience. Long-term exposures cannot be conducted in order to elicit chronic conditions. Parallel laboratory animal-human experiments, however, can be very useful in assessing the relevance of particular animal models to humans. Ideally, toxicologists and risk assessors would like to have dose-response data from experiments in which the same parameters were monitored and in which there was overlap of the range of doses given to each species. The doses administered to humans would be relatively low, but they should produce changes in sensitive adaptive effects, or biomarkers. Pharmacokinetic (PK), metabolic, and mechanistic studies in humans and animals also provide valuable information for scientifically based interspecies extrapolations (Jorkasky, 1998). Nonetheless, comprehensive toxicology investigations in different species of laboratory animals are necessary to fully evaluate the hazard potential of most chemicals.

Evaluation of the toxicity of chemicals in laboratory animals is a cornerstone of human safety evaluation. Experimentation with animals makes it possible to learn a great deal about the toxic potential of drugs and other chemicals. Explicitly defined investigations in laboratory animals are prescribed by EPA, FDA, and other regulatory authorities for approval of pesticides and drugs. Animals can be utilized for short, intermediate, and chronic exposure studies through which scientists can characterize the spectrum of adverse effects of a compound over a wide range of doses, dosage regimens, and exposure durations. Often, the toxicologist initially will administer high doses and evaluate a broad spectrum of parameters in order to identify target organs. Focused dose-response studies employing a limited number of sensitive indices of injury can then be performed. Ideally, dosage routes and regimens will be designed to mimic actual human exposure situations. The use of laboratory animals as toxicology research subjects is advantageous for several reasons. Most rodent species are relatively inexpensive and easily maintained. Large numbers of rodents can be assessed over a wide range of doses, increasing the likelihood of detecting adverse events (Zbinden, 1991). A number of biochemical, cellular, and physiological endpoints that can be examined only in human biopsy samples or at autopsy can be evaluated in animals. In addition, considerable background information often is available on commonly used strains of mice, rats, and dogs, including their genetic makeup, their abilities to metabolize xenobiotics, and their responses to other compounds. Groups of uniform animals can be administered measured doses of chemical(s) under defined and carefully controlled conditions, circumstances under which adverse effects to a specific chemical exposure can be attributed with greater certainty. Human populations, in contrast, are much more genetically diverse (Weber, 1999), with endogenous and exogenous factors (e.g., diet, stress, health, age, personal habits, use of drugs, exposures to other chemicals) that may not be recognized or controllable. In addition, the degree and duration of an individual's exposure to the chemical of interest are often unclear in epidemiological studies and case histories.

Findings in animal toxicology studies generally are applicable to humans, although responses of laboratory animals and humans to chemicals may differ qualitatively and/or quantitatively. The most definitive study to date of interspecies concordance involved an International Life Sciences Institute-sponsored review of data supplied by 12 pharmaceutical companies (Olson et al., 2000). The database consisted of toxicity findings from preclinical (i.e., experimental animal) and clinical (i.e., human) studies of 150 compounds in 15 therapeutic classes. Interspecies concordance of toxicity was said to exist if generally severe effects on the same organ occurred in humans and in laboratory animals. There was an overall interspecies concordance for 61 percent of the compounds. Rodents alone were predictive of human toxicities for 43 percent of the agents, while nonrodents (primarily dogs) alone were predictive for 63 percent. In another comparative investigation, 43 percent of the clinical toxicities of 64 marketed drugs in Japan were not predicted from animal experiments (Igarashi, 1994). The poorest concordance in this and the Olson survey (2000) were for cutaneous hypersensitivity and endocrine and hepatic functions. Obviously, animal studies cannot reveal subjective effects such as headache, myalgias, dizziness, nausea, or mental disturbances. The Olson study described other reports of poor correspondence between animal data and human toxicities severe enough to lead to market withdrawal of drugs. Many of these cases apparently involved idiosyncratic reactions that occurred with a very low incidence in patient populations, a phenomenon that reflects the pronounced influence of exogenous and endogenous factors on interindividual responses.

An evaluation by Dourson et al. (2001) of susceptibilities to industrial and agricultural chemicals has provided some additional information on the reliability of animal toxicology findings. These investigators compared human data-based reference doses (RfDs) for 22 chemicals in EPA's Integrated Risk Information System (IRIS) database with RfDs the authors calculated from animal data in IRIS using standard uncertainty factors. Seven of the 22 compounds were pesticides, for which cholinesterase inhibition was measured in intentionally dosed research participants. The interspecies concordance rate was approximately 40 percent. The human-based RfDs were lower than the animal-based values for 7 (32 percent) of the 22 chemicals. The human values were more than three times lower for five of these seven compounds, leading the authors to conclude that exposure limits based upon animal data may not be protective of public health. The power of Dourson's analysis is somewhat limited by the modest number of chemicals that were evaluated and by the quality and applicability of some of the data.

A considerable amount of information has been published on interspecies similarities and differences in susceptibility of chemical carcinogenesis. Faustman and Omenn (2001) pointed out that all human carcinogens that have been adequately tested in animals have produced cancer in at least one animal model. However, an evaluation of National Toxicology Program cancer bioassay data for 400 chemicals revealed that only 23 percent of the carcinogenic compounds produced tumors in both mice and rats (Fung et al., 1995). Some carcinogens, such as vinyl chloride, produce tumors in humans and in both sexes of other species tested. Conversely, many other carcinogens appear to be sex, strain and/or species specific (Grisham, 1996). Unleaded gasoline-induced kidney toxicity and cancer, for example, are limited to male rats, which is attributed to binding of gasoline to ∝_2u-globulin, a male rat-specific protein. The protein is hypothesized to accumulate to toxic levels in kidney cells and thereby induce sustained cellular proliferation, with its attendant cancer risk factors (Lehman-McKeeman, 1997). It also is hypothesized that oxidative moieties produced by peroxisomal enzymes and modification of cell signaling by activation of peroxisome proliferator-activated receptor-α can elicit liver cancer (Lake, 1995). A variety of compounds, including drugs such as ciprofibrate and nafenopin and solvents such as trichloroethylene and perchloroethylene, markedly induce hepatic peroxisomes and produce hepatic tumors in mice and/or rats. Studies of humans taking clofibrate and gemfibrozil, however, reveal little peroxisome proliferation and no increased incidence of liver cancer. Pharmacodynamic differences (i.e., disparities in receptor numbers and affinities) appear to account for this phenomenon (Cattley et al., 1998).

Variances in pharmacokinetics are often responsible for pronounced interspecies differences in susceptibility to toxic agents. The term “pharmacokinetics” encompasses systemic absorption, distribution, metabolism, and elimination. Many chemicals undergo metabolic activation (i.e., are metabolized to toxic metabolites). Others are detoxified through metabolism. Aflatoxin B₁, one of the most potent hepatocarcinogens known, is metabolically activated by cytochrome P450s and subsequently detoxified by conjugation with glutathione. Mice have been found to be much more resistant to aflatoxin B₁-induced liver cancer than rats. This disparity has been attributed to very efficient conjugation of the major reactive metabolite by mice.

Interspecies extrapolations on the basis of body surface area and comparative metabolism studies with primary hepatocytes of mice, rats, and humans indicate that the susceptibility of humans to a number of compounds resembles that of rats (Hengstler et al., 1999). Tamoxifen is a nonsteroidal antiestrogen that is used to treat pre- and postmenopausal women with breast cancer. It is a full estrogen in mice, a partial estrogen/ antiestrogen in rats and humans, and an antiestrogen in chicks (Jordan and Robinson, 1987). Tamoxifen is metabolically activated to a DNA-binding metabolite by a combination of Phase I and II metabolism. Biotransformation of tamoxifen is qualitatively similar in rats and humans, but the amounts of reactive metabolites and DNA adducts formed in the human liver are much lower than those formed in rats (Hengstler et al., 1999). Knowledge of qualitative and quantitative species differences in the metabolism of a xenobiotic allows the selection of the animal strain and species that is most like the human.

There are a number of quantitative methods for extrapolation of animal toxicity data to humans. The standard uncertainty factor default approach (described in Chapter 6 of this report) is frequently used because of a paucity of data. Linear extrapolations based on body weight equivalence often are inaccurate unless species-specific conversion factors are applied (Voisin et al., 1990), while allometric scaling on the basis of body surface area is more accurate. Freireich et al. (1966) report that doses of anticancer drugs lethal to 10 percent of rodents and maximally tolerated doses (MTDs) in nonrodents correlate with MTDs in human patients, when the doses are normalized to the surface area of each species. Normalization of body weight to the 2/3 or 3/4 power results in accurate predictions of body surface area, since both size (weight) and form (height) are taken into account (Davidsohn et al., 1986). FDA (2002) describes the use of standard species-specific factors that allow conversion of animal doses in mg/kg to animal doses in mg/m³ and human doses in mg/kg. The use of PK and metabolism data, when available for each species of interest, facilitates the most reliable interspecies conversions.

FDA (2002) has published a draft guidance document that describes a strategy recommended for deriving safe starting doses of therapeutic agents for clinical trials with healthy research participants. The first step in the process involves the identification of NOAELs (no observed adverse effect levels) from animal toxicity studies. The NOAEL for the most appropriate species is selected, regardless of whether this species is the most sensitive. The selection is based on information available on relative bioavailability, metabolic profile, molecular biology, physiology, and reactions to similar compounds. Humans and marmosets, for example, have constitutive levels of hepatic CYP1A2, a P450 isozyme that activates heterocyclic amines to reactive metabolites (Hengstler et al., 1999). Cynomolgus monkeys lack constitutive CYP1A2. Marmosets are thus a more suitable animal model for heterocyclic amines than cynomolgus monkeys. For drugs, the most appropriate animal NOAEL is converted to the human equivalent dose (HED) by the body surface area normalization process described by FDA (2002). Finally, the HED is divided by a safety factor to yield the maximum recommended starting dose.

Pharmacokinetics-based conversions provide the most reliable means of extrapolating from one species to another. Such approaches require PK data for each species of interest. Optimally, animal blood and target organ time-course data and metabolic information will be available for a range of doses, including those within which toxicity occurs. Human metabolic and blood-level data for low doses also would be necessary. Blood time-course data alone allow comparison of areas under blood concentration versus time curves (AUCs) for test animals and humans. Physiologically-based PK (PBPK) models (described below) are more precise, versatile, and scientifically credible than classical compartment-based models for inter-route, interdose and interspecies extrapolations (Voisin et al., 1990). PBPK models incorporate the unique anatomical, physiological, and metabolic characteristics of each species, as well as the physicochemical properties of the toxicant. PBPK models can be utilized to predict blood and target organ peak concentrations and AUCs of toxic moieties, whether they are the parent compound or a particular metabolite (Gerlowski and Jain, 1983). Toxicant exposures required to produce target organ doses that result in toxic effects of a given magnitude in laboratory animals are determined experimentally and modeled. The PBPK model is then allometrically scaled up to humans, or human-specific physiological and biochemical parameters are utilized for the model. Low-dose PK studies in volunteers are necessary to validate (i.e., assess the accuracy of) the model's predictions. Metabolic rate constants can be obtained from these studies or from in vitro experiments with human tissues or cells. Validated models allow one to simulate the human exposure conditions that will produce a target organ dose equivalent to that previously found to be associated with toxicity in the test animal. This so-called HED approach has been used successfully for a number of chemicals including, among others, methylene chloride (Andersen et al., 1991), acrylic acid (Frederick et al., 1998), and chlorpyrifos (Timchalk et al., 2002). Sensitivity analyses can be conducted to learn which physiological or biochemical parameters have the greatest impact on the pharmacokinetics of a particular chemical. One can also determine the influence of variability (that may exist in a human population) of the key parameters on estimates of tissue doses. Monte Carlo sampling of parameter distributions generates a distribution of model-generated target organ doses for different exposure regimens. The risk assessor can assess the variability in this distribution and judge whether a 10-fold intraspecies factor is merited (Watanabe et al., 1992; Thomas et al., 1996).