Below is the uncorrected machine-read text of this chapter, intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text of each book. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.
8 Risk Assessment The Safe Drinking Water Committee evaluated a number of approaches to assessing the risks of a variety of health effects, including cancer, reproductive and developmental impairments, and neurological diseases. Among the major considerations in any of these approaches are the variety and extent of the variables that will be encountered. For example, different sources of data may be used in the risk-assessment process, ranging from short-term tests for mutagenicity to long-term epidemiological studies of humans. Other variables that make risk assessment difficult and sometimes subject to substantial variation include metabolic differences between spe- cies and the variety of ways that exposures can be delivered, e.g., multiple, sporadic, or peak. Other considerations include route of exposure (e.g., inhalation, dermal, or ingestion) and source of exposure (e.g., drinking water, workroom air, or food). In some circumstances, risk assessment can be enhanced when there is an opportunity to evaluate pharmacokinetic mechanisms involved in responses to environmental agents. In this chapter, the committee first reviews the discussions of risk as- sessment set forth by previous Safe ~ng Water Committees. It then examines three topics common to all risk assessment: estimation of exposure from various sources and by different routes, pharmacokinetics, and inter- species extrapolation. The remainder of the chapter is devoted to a discussion of risk assessment for different categories of health effects. Cancer is ex- amined first, because risk assessment in this area has the longest history and, therefore, the methods are more highly developed than those for over adverse health effects. However, many of these methods are germane to assessing the risk for noncancer end points as well. The last two sections cover risk 250
Risk Assessment 251 assessment for developmental and reproductive effects and for neurotoxic effects. Conclusions and recommendations drawn from all these discussions are presented at the end of the chapter. To assess risks associated with exposure to toxic chemicals, agencies have developed a systematic scientific and administrative framework (NRC, 1983~. The process commonly begins with the identification of a hazard often brought to light by data from studies on laboratory animals, from other laboratory procedures, or sometimes from case reports involving humans. The next step is determination of the dose-response relationships between specific quantities of a substance and associated physical re- sponses, such as the development of tumors, birth defects, or necrologic deficits. Then follows exposure estimation and assessment. At that time, a search is made for an answer to the question: To what dose levels or range of dose levels will human populations most likely be exposed? Finally, the dose-response model is applied to the expected exposure levels to produce a quantitative estimate of risk. To date, quantitative risk assessment (QRA) has been used largely for estimating the risk of developing or dying from cancer, and has been used little in evaluating or estimating noncancer health effects of exposure to materials in the environment. The earliest approaches to QRA for cancer are probably those of Mantel and Bryan (19611. Modifications to these approaches have been published through the years. Recently, the U.S. Environmental Protection Agency (EPA) issued proposed guidelines for assessing the risk of exposure to carcinogens, mutagens, and develop- mental toxicants (EPA, 1984a,b,c). A developing, interdisciplinary field, risk assessment is not without unresolved issues, including the extrapolation of results to doses outside the range of observations in the experiments; the choice of the appropriate dose-response model for extrapolating from nigh-dose animal data to the anticipated low levels of exposure of humans in the ambient environment; the appropriate translation of data from the laboratory animal to humans, which in turn involves questions concerning the influence of body size, life span, and possible metabolic differences; and the potential for con- founding (e.g., whether there is synergistic or antagonistic response to exposures to other materials). Problems that lie at the interface of science and policy include selection of test method, selection of animal data and bioassay results for use as the basis for extrapolation to humans, deter- mination of how one should use data on so-called benign as well as malignant tumors, selection of appropriate safety factors for developing standards, and selection of the mathematical model to be used for ex- trapolation. QRA methods have been the subject of a number of publications (An- derson and CAG, 1983; California Department of Health Services, 1982;
252 DRINKING WATER AND H"LTH OSTP, 1985), which will not be reviewed here. They can be used to provide a broad range of estimated risks for the purpose of setting priorities for regulatory action or to identify a specific acceptable or permissible level of human exposure to a carcinogen. Most experts and policymakers agree that current QRA techniques at best indicate a range of risks rather than a precise number. Since 1977, when the first Safe Drinking Water Committee conducted QRAs on waterborne carcinogens, National Re- search Council committees have consistently highlighted limitations of the methodology (NRC, 1977, 1980, 1983~. This committee endorses attempts to develop, validate, and apply QRA techniques to the evaluation of potential noncancer toxic responses, such as neurotoxicity, reproductive toxicity, and developmental toxicity, as well as liver damage, kidney damage, and respiratory responses other than lung cancer. Dunng the last decade, cancer has been regarded as a nonthreshold phenomenon, whereas most noncancer responses are be- lieved to require a minimum (i.e., threshold) dose before any toxic man- ifestation will appear. Some recent research has suggested, however, that this distinction may not be a clear one. PREVIOUS SAFE DRINKING WATER COMMITTEES' VIEWS ON RISK ASSESSMENT Earlier Safe Drinking Water Committees considered the problems of risk assessment in the first and third volumes of Drinking Water and Health (NRC, 1977, 19801. The present committee affirms many of the views expressed by those groups. Issues Concerning Threshold After reviewing the many problems inherent in assessing the risks of carcinogenesis and mutagenesis, the 1977 committee acknowledged that many scientists distinguish between injuries produced by chemicals likely to have a threshold dose and effects for which there is likely to be either no threshold (e.g., carcinogenesis and mutagenesis) or no way known to estimate one for large, heterogeneous populations. On the basis of this observation, the 1977 committee concluded, "It is more prudent to treat some kinds of toxic effects that may be self-propagating or strictly cu- mulative, or both, as if there were no threshold and to estimate the upper limits of risk for any given exposure" (NRC, 1977, p. 254. The report included among self-propagating or strictly cumulative effects those that result from early, chemically induced alterations in cellular DNA that are transmitted by cell propagation and irreparable injuries, such as destruction
Risk Assessment 253 of neurons, noting that "destruction of enough neurons leads to a decrease in central nervous system function." Carcinogenic and Mutagenic Effects The same committee outlined the following principles that underlie efforts to assess the irreversible effects of long-continued exposure to carcinogenic substances at low dose rates: 1. "Effects in animals, properly qualified, are applicable to man." 2. "Methods do not now exist to establish a threshold for long-ter effects of toxic agents." 3. "The exposure of experimental animals to toxic agents in high doses is a necessary and valid method of discovering possible carcinogenic hazards in man." 4. "Material should be assessed in terms of human risk, rather than as 'safe' or 'unsafe' " (NRC, 1977, pp. 53-561. Noncarcinogenic Effects For presumably reversible noncarcinogenic and nonmutagenic effects, the committee advised, "For noncarcinogens for which it seems likely that there are thresholds for toxic effects, the acceptable dose should be below the threshold. If a threshold cannot be shown, the acceptable dose must be related to the data from animal experimentation and consideration of the seriousness of the toxic effects, as well as the likelihood and ease of reversibility, the variability of the sensitivity of the exposed population, and the economic and health-related importance of the material" (NRC, 1977, pp. 57-58~. The present committee appreciates that there is not a clear distinction between cancer and other toxic responses. Safety, or Uncertainty, Factors To this general statement the committee added specific advice on the use of safety, or uncertainty, factors (NRC, 1977, p. 8041. An uncertainty factor of 10 was recommended when there are valid results from studies on prolonged ingestion by humans and no indication of carcinogenicity. An uncertainty factor of 100 was recommended when there are few or no toxicological data on ingestion by humans, but there are valid results from long-term studies in animals and no indication of carcinogenicity. An uncertainty factor of 1,000 was recommended when there are no long- term or acute data on humans, only scanty data on animals, and no indication of carcinogenicity.
254 DRINKING WATER AND HEALTH The Use of An emal Data to Predict Human Risk The first Safe Drinking Water Committee remarked, "Current knowl- edge of the proper principles for extrapolating toxicological data from high dose to low dose, and from one species to another, is inadequate" (NRC, 1977, p. 251. In Volume 3 of Drinking Water and Health (NRC, 1980), the reconstituted committee reexamined the prediction of risks to human health using acute and chronic toxicity data on laboratory animals. The entire third chapter of Volume 3 was devoted to this issue.The com- mittee summarized its conclusion as follows: The concentration of most potentially toxic chemicals in drinking water is usually so low that it is difficult to predict potentially adverse effects from drinking the water. In cases of noncarcinogenic toxicity, the preferred procedure would be to make a risk estimate based on extrapolation to low dose levels from ex- perimental curves obtained from much larger doses for which effects can be readily measured. In most instances, such data are not available, and the acceptable daily intake (ADI) approach should be used until better data are obtained. In the ADI approach, "safety factors" based on the quality of the data are applied to the highest no-observable-effect dose found in animal studies. The [Safe Drinking Water Committee's] Subcommittee on Risk Assessment believes that the ADI approach is not applicable to carcinogenic toxicity, and that high dose to low dose extrapolation methods should be used for known or suspected carcinogens. Six models were evaluated for low dose carcinogenic risk estimation. They were the dichotomous response model; linear, no-threshold model; tolerance distribution model; logistic model; "hitness" model; and time-to-tumor- occurrence model. Because of the uncertainties involved in the true shapes of the dose-response curves that are used for extrapolation, a multistage model was judged to be the most useful. Such a model has more biological meaning than other models, e.g., the probit or logistic model. Moreover, it tends to be con- servative in that at low doses it will give higher estimates of the unknown risk than will many others. More confidence would be placed in mathematical models for extrapolation if they incorporated biological characteristics such as pharmacokinetic data and time- to-occurrence of tumors. Until such data are available, the extrapolation from animals to humans should be done on the basis of surface area (NRC, 1980, pp. 2-34. As the discussion in that volume indicates, the computation of risk depends upon several assumptions, ranging from the choice of mathe- matical model for low-dose extrapolation to the minor operating assump- tions made within the model itself to implement a specific computer program. The present committee noted that within the multistage model one can compute the dose necessary to develop a given level of risk by using either an experimentally restricted model or a generalized model. The restricted model limits the number of possible stages in the multistage
Risk Assessment 255 TABLE 8-1 Unit Riska Depending on Assumed Stages of Carcinogenesis Stages Experimentally Stages Generalized Estimate Restricted to Data Maximum likelihood estimate 1.2 x 1o-s 8.8 x 10-6 Upper 95% confidence limit estimate 2.2 x 1o-s 1.7 x 1o-s aAssurrung a daily consumption of 1 liter of water containing the compound in a concentration of 1 ,ug/liter. process to the number of doses at which the experiment was conducted minus 1. The generalized model places no such limit on the possible number of stages but, rather, permits computation of the best fit to the data without the constraint of doses used in the experiment. For example, by using data on response to acrylamide for male A/J mice with lung tumors and the two different computational assumptions concerning stages of carcinogenesis, one can estimate two slightly different risks, as shown in Table 8-1. A similar small difference was found in the risks estimated for female mice with lung tumors. Where the data are highly curvilinear, the exper- imentally restricted approach can assign less of the effect to the linear term, and may even estimate it to be zero. In contrast, the unrestricted form of the computation is more likely to identify a nonzero linear term, which may be more in keeping with current biological knowledge and assumptions. Computations in the third volume were based on both ap- proaches. The differences were only slight and, thus, not important. Fol- lowing the recommendations of the risk-assessment panel of the Consensus Workshop on Formaldehyde (1984), the computations presented in the present volume are based on the generalized form of the model. In considering the appropriate scaling for extrapolating data from the laboratory to humans, the 1980 committee remarked: The practice among cancer chemotherapists of basing dose on body surface area is useful, particularly for extrapolation from small animals to humans, and is supported by a sizeable body of experimental evidence. Since body surface area is approximately proportional to the two-thirds power of body weight, the anticancer drugs are relatively more toxic to the larger animals than to the smaller ones (NRC, 1980, p. 29~. Combined Exposures The 1980 committee considered information on the combined action of materials found in drinking water, noting first that the joint action could be additive, synergistic, or antagonistic, and that "in general, there is not
256 DRINKING WATER AND H"LTH likely to be sufficient information on [the action of] mixtures.... Con- sequently, estimates will . . . have to be based on an . . . assumption of additivity" (NRC, 1980, p. 27~. The committee pointed out that assuming independence of action of a material in relation to background exposure instead of assuming additivity in dose could, at a low-dose level, easily lead to a 100-fold difference in estimated risk. In a recent review of the possible effects of exposure to a mixture of materials, Berenbaum (1985), arguing by analogy from the behavior of combinations of antibiotics, remarked, "It is therefore not unreasonable to assume that carcinogens will prove to behave similarly [i.e., will show synergism] . . . and it appears sensible to assume this until proved oth- erwise." Berenbaum also noted that the "effect of a marked antagonism is to produce a threshold in the curve." Reliability of Risk Estimation Considering all the variables encountered in the process of estimating risks, the 1980 committee remarked, "If the estimates of risk from low doses of carcinogens are made with reasonable data and reasonable models, [there will be] a precision of 1 or 2 orders of magnitude in the estimates" (NRC, 1980, p. 601. It has since been pointed out that maximum likelihood estimates (MLEs) are extremely sensitive to the data in that very small differences can lead to large differences in the MLEs (Cohn, 1986; T. W. Thorslund, EPA Office of Health and Environmental Assessment, Wash- ington, D.C., personal communication, 1985~. The 95% upper confidence limit estimates are much more stable. Noncarcinogens The 1980 committee noted that determination of no-effect levels for noncarcinogens depends upon both data interpretation and the number of animals in the bioassay. It stated, "The likelihood of observing a no- adverse effect at a given dose is statistically greater for experiments with few animals than for larger experiments" (NRC, 1980, p. 311. Thus, if these and other details are not described in published studies, the use of more formal (dose-response) risk-assessment procedures is impeded and it is more difficult to interpret whatever data are at hand. Among the important matters included are whether best-fit or 95% upper confidence limit curves should be used in expressing risk and whether the dose- response model, log-normal model, or log-logistic model should be applied to the dose-response curves. The 1980 committee concluded, "The po- tential utility of dose-response extrapolation methodology for noncarci-
Risk Assessment 257 nogenic human risk assessment does exist but has been found to be of limited value for contaminants in drinking water" (NRC, 1980, p. 351. Crump (1984) has suggested an alternative to the no-observed-effect level (NOEL) for determining an acceptable daily intake (ADI). If it is possible to define as "acceptable" some very small increase over the response level in the control group, then a benchmark dose can be estab- lished as the lower (95% or 99%) confidence limit on an exposure that would produce small (or smaller) increases over the control level. Crump's proposal makes more effective use of all the experimental data than does the NOEL approach, taking into account the slope of the exposure-response curve and the size of the experiment. Thus, there is a greater efficiency in the use of the data, and larger experiments should, in general, lead to higher ADIs (rather than the reverse, which the 1980 committee noted is characteristic of the NOEL approach). This procedure is similar to the one used by the committee in developing a recommended range of ex- posures to aldicarb (see Chapter 91. The 1980 committee also provided a detailed discussion on the estab- lishment of suggested no-adverse-response levels (SNARLs) for acute 24- hour or 7-day exposures to noncarcinogenic materials. It based its cal- culations on assumptions that " 100% of the exposure to the chemical was supplied by drinking water during either the 24-hour or the 7-day period" (NRC, 1980, p. 681. To calculate chronic SNARLs, the committee ar- bitrarily assumed that drinking water provided 20% of the intake of the chemical of concern. Estimates of Exposure from Different Sources and Various Routes All sources of exposure must be considered when assessing risk or setting safety factors for estimating ADIs of chemicals in water, regardless of the biological end point under consideration. The population may be exposed to chemicals through air and food as well as through water. For some population subgroups and for certain toxic chemicals, the air and the workplace may be the major sources of exposure. For others, drinking water may be the primary source. The various routes of exposure from these sources must also be con- sidered. Systemic absorption resulting from simultaneous exposure via multiple routes has received little attention to date. Shehata (1984) pre- sented a modeling approach that may be useful in estimating the relative contribution of multiple exposure routes to body burdens of volatile or- ganics. However, further research is required to validate these multiple route models. The present committee finds it worthwhile to distinguish between dose response and exposure response. Dose response may be defined as the .
258 DRINKING WATER AND H"LTH response to the dose of toxicant actually delivered to the target site. Generally, determination of dose response requires knowledge of the fate and distribution of the administered material, including pharrnacokinetic behavior and metabolic activation or deactivation. Exposure response is easier to determine, since information about exposure or nominal dose is more accessible and exposures are subject to potential regulation. Ade- quate understanding and elucidation of the mechanisms of toxicity will require knowledge of dose-response relationships. For regulatory pur- poses, an adequate description of the exposure-response relationship is more important. WATER For chemicals with a threshold dose, the ADI is based on animal tox- icology data by applying a safety factor to a NOEL. To determine the maximum exposure from water that should be allowed, contributions to total exposure made by sources other than drinking water should be sub- tracted from the permissible exposure (maximum permitted intake is usu- ally the ADI times assumed adult body weight of 70 kg). The maximum permissible level in water can then be determined from the permissible exposure through drinking water and the standard volume of daily water consumption, which is normally assumed to be 2 liters in a 70-kg adult male and 1 liter in a 10-kg child (Kelly, 1980; NRC, 1977~. Where the standard volume of consumption is set high in the range of that consumed in a population, an action level of exposure may be safely set without any increase in risk. In fact, actual water intake varies considerably with physical activity, environmental temperature, and relative humidity, which in turn vary by region of the country. At this time, however, the committee was unable to estimate the contributions of these other factors and sources other than drinking water and encourages the development of better data and models on these variables. Valid arguments may be constructed for setting the standard daily con- sumption volume for drinking at the population mean, the median, the ninth docile, or some higher standard volume. If set at the population mean or median, roughly half the population will be expected to exceed this level. This choice must be related to recommended maximum per- missible levels in water. Although ingestion is the chief route of exposure to chemical contam- inants in drinking water, inhalation and dermal exposure may also con- tribute to systemic absorption of waterborne contaminants. For example, large amounts of volatile organic chemicals may be inhaled from boiling water or hot showers. Since many organic compounds are poorly soluble
Risk Assessment 259 in water and are quite volatile, a substantial portion of the contaminants in water may evaporate under certain usage conditions in the home, e.g., from washing machines, dishwashers, sinks, bathtubs, and showers. Dur- ing showers with chemically contaminated water, the combined action of the spray and high temperature could result in the generation of relatively high vapor levels in confined areas. Seasonal changes in household ven- tilation may also greatly influence the extent to which volatilized chemicals are retained in the household air. Brown et al. (1984) proposed that absorption through the skin may contribute significantly to the total dose of volatile organic compounds received during normal daily use of contaminated water. Their calculations were based on skin absorption rates for toluene, xylene, styrene, and ethylbenzene, which were measured in human exposure studies by Dut- kiewicz and Tyras (1967, 19681. The reliability of the models and the accuracy of such predictions must, of course, be verified through exper- imentation. For less volatile water contaminants, there are no definitive experimental data on inhalation and dermal exposures. Thus, doses from these routes of exposure cannot be reliably estimated at this time. It may be prudent to assume that in addition to the standard 2 liters of water consumed orally each day, daily exposure to another 2 liters results from inhalation and skin absorption during bathing, showering, cooking, washing, and other activities involving water usage. The adoption of this concept would, of course, have the effect of lowering to one-half the permissible concen- ~ation of a chemical in water. FOOD Traces of pesticides and other chemicals that contaminate drinking water also contaminate foods. Exposure through food (not including any con- tribution from the water used in cooking) varies considerably, depending on diet. The estimated tolerances to a pesticide or recommended action levels for other chemicals in foods may be used to calculate a theoretical maximum residue contribution (TMRC) through food (Ariens and Si- monis, 1982; Hathcock et al., 19834. These TMRC values constitute an upper limit on the exposure through food. They are difficult to use, how- ever, because they usually exceed the measured intakes by one to several orders of magnitude. Moreover, estimates of the proportion of the diet accounted for by specific foods are often not current and there is little information on variation in diets by region, ethnicity, age, sex, or other factors (NRC, 19821.
260 DRINKING WATER AND HEALTH AIR The background level of exposure to many toxic chemicals in the am- bient air is usually even more variable and difficult to determine than the exposure through food. The extent of both indoor and outdoor exposures through air depends strongly on occupation, region (urban levels exceeding rural levels for many substances), climate, and life-style (e.g., whether sedentary or active, length of time spent indoors and outdoors) (NRC, 1985~. WORKPLACE In occupational settings, workers may be exposed to toxic chemicals by routes other than ingestion of water, including inhalation, direct contact with skin, or ingestion after contamination of hands, cigarettes, or food. Contaminants in ambient air and in settled dust may account for much of the total workplace exposure. Contaminants may come into contact with skin from clothing contaminated by vapors, dust, or spills. HOUSEHOLD The Total Exposure Assessment Measurement (TEAM) study conducted by the EPA (Pellizzari et al., 1984) has shown tremendous variation in personal indoor exposures to toxic chemicals. Until there is a better un- derstanding of the doses provided by all the different indoor sources of pollutants, estimates of average national exposures to drinking water con- taminants in the home will not be useful in determining total exposures to the chemical of interest. Upper limits of likely ranges of exposure may be more useful than the boundaries selected. PHARMACOKINETICS An understanding of pharmacokinetic principles is necessary to the successful extrapolation of data from high to low doses. (See Chapter 6 for a detailed discussion of models based on pharmacokinetics.) The Office of Science and Technology Policy (OSTP) (1985), elaborating on a con- clusion reached by Hoel (1980), noted, "Even if only a small portion of the background incidence [of the toxic responses is associated with the same mechanistic process as the study chemical, linearity will tend to prevail at sufficiently low doses" (OSTP, 1985, p. 81~. Some arguments to the contrary have been raised. An unresolved issue that may have a bearing on the possible existence of a threshold is the dissimilarity of the kinetics of toxification-detoxification at high in com-
Risk Assessment 261 parison to low doses. Some scientists believe that chemical carcinogens can be crudely classified into two groups: · those that attack DNA directly in the nucleus (e.g., as assayed by adduct formation or mutation) and · those that do not appear to modify DNA directly. There is substantial disagreement about how to make this distinction, as discussed later in this chapter and in Chapter 5. Most of the substances in the first category discovered to date are not carcinogenic per se, but are instead activated enzymatically in the cell to produce unstable, chemically reactive intermediates (e.g., polycyclic hy- drocarbons, aLkanes, aLkyl halides, and aryl amines). Mixed-function ox- idases (i.e., the cytochrome P450 system and the monoamine oxidase system) are responsible for such activation (Miller and Whitlock, 1982; Nebert and Gelboin, 19681. The primary function of both systems appears to be the detoxification of cells exposed to lipid-soluble molecules. Such molecules are metabolized (typically to form epoxides or acetoxy com- pounds), and then condensed catalytically with water-soluble substrates, such as glutathione (Gelboin, 19809. This metabolic pathway has three important features: · The oxidases are inducible by a mechanism that directly increases the level of oxidase-specific messenger RNA, resulting in a 10- to 100- fold increase in the activity of the enzyme (Bresnick et al., 1981~. · The inducibility and the absolute basal activity of the oxidases and transferases that are fundamental enzymes for cell detoxification vary as much as 100-fold among different tissues and species (Miller and Whit- lock, 1981~. · In such a coupled activation condensation pathway, the steady-state concentration of activated, potentially toxic carcinogen (d*) is very sen- sitive to the relative concentration and the relative activity of the various enzymes in the coupled pathway (Hoer et al., 1983~. As a result, there is a fundamentally nonlinear relationship between the administered dose (d ~ and the effective dose (at*) at the molecular targets, although under certain circumstances it can become linear at low doses (Hoer et al., 1983~. Figure 8-1 depicts four consequences of coupling between activators (oxidases) and condensing enzymes. Part A of the figure shows a noninduced oxidase, where the transferase reaction has a high Km compared to the oxidase. The dose d* available to attack DNA displays ordinary Michaelis-Menten enzyme kinetics. When the transfer- ase reaction has a low Km' the reaction will become saturated at a relatively low applied dose d (part B). Under those conditions, the relationship between d and d* is very flat at low doses but converges at a higher dose
262 DRINKING WATER AND H"LTH A 1 -- - CO o UJ > LU , LL I ~ , 1 / l l 1/ - DOSE ADM I N ISTER ED (d) C _ * - UJ o C, LL > , - UJ 11 . I 1 1 ~ dj DOSE ADM I N ISTE R E D (d) B * - in o LL > Cot LL 11 LL * - LL CO o > LU 11 - - DOSE ADMINISTERED (d) D / / /'/ I ,,' / / / / DOSE ADM I N ISTER ED (d) FIGURE 8-1 Four consequences of coupling between activators (oxidases) and condensing enzymes. to behavior that is more like that of a simple enzyme, with an apparent Km corresponding to that of the oxidase alone. The inducibility of the oxidase (and perhaps the transferase) in these kinetic schemes further complicates the picture. Part C of the figure shows the simplest behavior expected for an inducible oxidase, corresponding to the enzyme kinetics in part A, except that at a dose di, the cell responds to the inducer by producing more enzyme. As seen in part C, the applied dose to d* response becomes biphasic with a characteristic inflection point at di. Part D is a combination of parts B and C. To gain an understanding of dose extrapolation for carcinogens in terms of these figures, it is important to recognize Hat long-term animal testing
Risk Assessment 263 must generally be performed at doses within a factor of 10 of the level that produces immediate tissue damage (the maximum tolerated dose). Given that the hydroxylase-transferase system is the principle mechanism by which cells are detoxified, such dose schedules are almost certainly positioned on the right side of the dose responses described in Figure 8-1. In any biological or biochemical process, an experimentalist looks for specific characteristics. In enzymology, such a characteristic would be the substrate concentration at half maximal velocity Km. Below that critical concentration, the reaction rate is roughly linear with respect to substrate concentration. Above Km, reaction velocity becomes level and independent of the substrate. By contrast, at the limit of high substrate concentration (such as those that might be expected at common environmental expo- sures), the enzyme reaction rate depends only upon rate constants for processing bound substrate. At the limit of low substrate concentration, the measured rate depends upon substrate-binding and enzyme-processing rates. An important experimental fact should be evident from these consid- erations: the information necessary to extrapolate from high to low doses cannot be obtained from observations of enzyme behavior at high doses, even if the formal relationship between dose and response is clearly un- derstood. The reason for this is that the behavior of an enzyme system at high doses is (or may be) dominated by factors different from those that determine behavior at low doses. Consequently, by analogy with a simple enzyme kinetics process, it is a distinct possibility that the experimental induction of a tumor at high doses may be dominated by system characteristics that differ from those directing behavior at low doses. Again, a functional form can always be fit to high-dose data. However, even if that function precisely describes the relationship between tumor and dose, it will not yield a meaningful low-dose prediction if the system has become saturated, or approached saturation, with respect to crucial enzymatic processes. The dose responses described in parts A and B of Figure 8-1 appear to be very different, but they result from exactly the same coupled enzyme pathway; only the enzyme parameters have been altered. Given the sub- stantial tissue-specific (Nebert and Jensen, 1979), species-specific (Gregus et al., 1983), and individual-specific (Nebert et al., 1975) differences that have been described for procarcinogen oxidases, it is likely that dose- dependent behavior as different as that shown in parts A and B of Figure 8-1 may occur in different organs in the same person, or in different people, or at different stages in cell differentiation. As a class, carcinogens that do not bind to DNA are even less well understood. The best studied of these compounds is dioxin, which is a potent activator of aryl hydrocarbon hydroxylase activity and therefore
264 DR'NK! NG WATER AND H "LTH increases the rate at which procarcinogens are transformed to a carcino- genic state (Miller et al., 1983~. Such activation could substantially alter the active dose of carcinogen in the nucleus, resulting in tumor formation. A second type of indirect carcinogenic effect has recently been suggested for promoters of carcinogenesis such as the phorbol esters, e.g., 12-0- tetradecanoyl-phorbol- 1 3-acetate (TPA). These compounds appear to be- have as mitogens, disrupting the ordinary pattern of cell-cycle control, and in turn possibly disrupting the expression of genes crucial to tumor formation (Michell, 1984; Nishizuka, 1984~. Although the dose response for dioxin activation or TPA modification of cell-cycle control may be linear at low applied doses, the events are complicated enough that the dose response may depart substantially from linearity under experimental conditions. As for DNA-binding carcinogens, experiments conducted at high doses may be affected by saturation and for that reason may not be extrapolated to low doses in a simple fashion. Given the data at hand, extrapolation for carcinogens such as dioxin or for promoters such as TPA cannot be made with certainty at this time. Experimental data could be fit empirically by equations describing a mul- tistage model or a Weibull model for generating a low-dose extrapolation. At present, however, there is no firm experimental evidence to support the selection of any particular extrapolation protocol, although the mul- tistage model was derived as a way to explain or describe the age patterns of cancer in humans. As indicated earlier, the OSTP does not consider "goodness of fit" as an adequate basis for choice among models (OSTP, 1985~. INTER- AND INTRASPECIES EXTRAPOLATION Calabrese (1983) has reviewed both qualitative and quantitative differ- ences between species. These include differences in physiology (e.g., rats are obligate nasal breathers and have a nonglandular stomach but humans do not, and the volume of blood flowing to various organs in rats is different than in humans); biochemistry (e.g., differences in basal meta- bolic rates, pharmacokinetics, enzyme activity, and receptors); size; life span; and the nature, routes, and duration of exposures. Although it may be possible to identify test species both qualitatively and quantitatively similar to humans in some respects, it has not been possible to identify which similarities are most important for the comparison of long-term chronic effects such as carcinogenesis. For the near future, chronic studies in laboratory animals are likely to remain the primary means of evaluating chronic toxicity and carcinogenicity. Enzyme function is highly efficient among vertebrates. However, there can be enormous quantitative differences in enzyme activity, even among
Risk Assessment 265 evolutionarily conserved pathways. There are three basic types of diversity in enzyme activity: differences among species, differences among indi- viduals, and organ- or tissue-specific differences. Gregus and colleagues (1983) have catalogued differences in the activity of liver enzymes. Among eight different vertebrates tested, the activating and condensing enzymes can vary 10- to 30-fold. More importantly, there is no systematic evolutionary relationship: trout display an enzymatic fingerprint that is as similar to that of dogs as that of cats. Thus, in the absence of quantitative enzymology, it is not necessarily valid to presume that higher vertebrates will be more similar to humans in terms of their dose response. Substantial intraspecies differences in sensitivity to carcinogens have been reported (Mohrenweiser and Neel, 19821. The extent of such variation has been elucidated in a dramatic fashion by Miller and Whitlock (1982), who identified two natural variants within a Hepa-lclc7 mouse clonal hepatocyte population: · 1% to 2% of the hepatocytes had no basal aryl hydrocarbon hydrox- ylase (AHH) activity (they cannot metabolize procarcinogens); · 1% to 2% of them had basal enzyme activity about 40 times greater than the population average. Differences of that sort can also be identified in intact animals. At the biochemical level of analysis, for example, C57BL/6 mouse strains have a highly inducible liver cytochrome P450 AHH system, but DBA/2 mice do not (Gielen et al., 1972~. This variation has been localized genetically to the Ah locus, which appears to encode a gene for a cell surface receptor (Tukey et al., 1981~. To date, there is no equivalent genetic information on humans, but it is very unlikely that genetic diversity in the carcinogen- processing enzymes Is specific to mice. When there are results from more than one valid and well-conducted bioassay, data are usually selected from the bioassay in which the most sensitive species was used (California Department of Health Services, 19821. This decision is based in part on the widely accepted assumption that in the absence of data to the contrary, humans should be considered as sensitive as the most sensitive species. Some evidence suggests that humans are roughly as sensitive or in some cases more sensitive than experimental animals (California Department of Health Services, 1982; Crouch and Wilson, 1979; NRC, 1975, 19771. To predict carcinogenic risk to humans from data on animals, one should consider all the available data on the type of tumor produced, number of tumors, and time-to-tumor induction. Attention also needs to be. paid to interspecies differences in body size and life span; the duration, nature, and route of exposure; and possible variations in metabolic and pharrna-
266 DRINKING WATER AND HEALTH cokinetic patterns and rates as well as in inherent susceptibility. In practice, conversion factors have been generally applied for size and exposure differences. For the other variables, humans and experimental animals are generally assumed to be similar; for example, a lifetime of exposure of a laboratory animal is considered equivalent to a lifetime of exposure of humans. Conversion factors for size are usually based on surface area or body weight. When surface area is used to scale dose rates from experimental animals to humans, the projected human risks in rats and mice are 6- and 14-fold higher, respectively, than rates derived from body weight (OTA, 19811. In assessing the risk of waterborne carcinogens, different com- mittees of the National Research Council have used surface area rather than weight (NRC, 1977, 1983), which ensures the provision of risk estimates with the greatest potential for protecting health. CARCINOGENESIS Because of the complexity of carcinogenesis, the selection of the most appropriate risk-assessment method is not a simple matter. As a general rule, nonthreshold models should be used. It is reasonable to use models that incorporate background additivity and, hence, low-dose linearity, unless there are convincing experimental and human data showing that such an assumption is inappropriate. Long-term tests designed to determine the response of laboratory ani- mals to possible carcinogens are performed on relatively small populations of approximately 100 treated animals per sex per species. Because of inherent sampling error, a suspected carcinogen must be administered in doses usually many times greater than those expected in the environment to obtain a statistically significant response in such small test groups. For that reason, the process by which a low-dose relationship is inferred has been (and will continue to be) the subject of intense debate. The NOEL Volume 1 of Drinking Water and Health contained a detailed discussion of the relationship between classical toxicological testing and the extrap- olation to low doses. In a classical testing protocol for a small sample, it is always possible to identify an applied dose do small enough that the elicited response cannot be distinguished from that detected in an unex- posed control population. That dosethe NOEL should be interpreted carefully. If the probability of a toxic event occurring at some dose do is no greater than background, an upper limit to the real excess probability of that toxic
Risk Assessment 267 response can be calculated by using sampling theory (Young, 19621. First, an upper confidence limit is specified. This is the chance that the calculated response probability is not a true upper limit to that specified by the data. Typically, a confidence limit of 95% is selected. That is, given the ob- served data, the true excess toxic response probability will be at or below the calculated upper probability 95% of the time. When there are no responders in a population of N test animals (i.e., at the NOEL), the foal relationship between parameters is (1 - Pr)N= 0.05, where Pr is a calculated upper limit to the toxic response probability at dose do (Young, 19621. For a population of 100 test animals (N = 100), Pr = 0.03, i.e., a dose do produced zero toxic responses in 100 animals, implying that there is a SO chance the true response at that dose is greater than 0.03. If this kind of computation were applied to cancer, we recognize at once that a 3% incidence rate is very high (between a thousand and a million times greater than many spontaneous tumor rates). Therefore, in addition to the detailed considerations required in relating human sensi- tivity to that of small test animals, estimation of cancer risk from envi- ronmental exposures requires that experimental data (i.e., estimates of risk in animals) be extrapolated to doses far lower than the usual NOEL determined for threshold-type responses. Quantitative Assessment A set of experimental data relating applied dose d to tumor probability Pr~d[) can be fit mathematically by many functional forms nearly equally. But calculations of probability of tumor occurrence at low doses based on those equations can vary widely. In Volume 3 of Drinking Water and Health (NRC, 1980), the committee noted that "goodness of fit" was not an appropriate criterion for the selection of an extrapolation model. There- fore, data from chronic studies in animals cannot be used to determine the correct relationship between dose and response at low-dose levels. Models need to be evaluated in relation to basic knowledge of the tumor formation process. Epidemiological (ICPEMC, 1983), clinical (Lee and O'Neill, 1971), and biochemical (Land et al., 1983) evidence combined suggests that tumors occur after a cell has experienced two or more lesions, or hits. These lesions need not be environmentally induced. For example, one or both may be the consequence of an inherited genetic trait. In many in- stances, the crucial hits may occur in DNA or in the cellular machinery responsible for gene control. In such a genetic model for tumor growth,
268 DRINKING WATER AND HEALTH it is presumed that transformation follows one or more hits, after which a single cell can proliferate to form a tumor. These experimental findings limit the selection of the model to be used for low-dose extrapolation. In addition to the consequences of the existence of some background level of incidence, if two or a few hits within a single cell are sufficient to initiate tumor growth, then there can be no absolute threshold in the dose-response relationship; i.e., at any dose, there will be some probability that a cell can be transformed, thereby initiating tumor growth (Crump et al., 19761. A multistage model of carcinogenesis pro- posed several years ago (Moolgavkar and Knudson, 1981) provided a way for taking into account the birth and death of cells in any preneoplastic compartment. Thus, the model appears to be consistent with current bio- logical knowledge and should be able to elucidate whether an agent affects transition rates, tissue growth, or tissue differentiation. By including the differential birth and death process, the model should be able to account for nonlinearities seen in some dose-response data. A major problem in understanding the etiology of cancer lies in the choice of the dose metameter. If the probability of the development of cancer is linearly related to dose, then the appropriate measure to consider would seem to be lifetime cumulative exposure. Most risk-assessment models use this cumulative exposure as the dose. However, there are data that do not conform to such an explanation, implying some nonlinear mechanisms. For example, in reviewing the dose-times-rate effects of the administration of 2-acetylaminofluorene (2-AAF), Littlefield and Gaylor (1985) noted that the higher dose rates for shorter periods led to a higher incidence of tumors, although total dose was equivalent. These results strongly suggest that for some materials dose rate may be a more important measure than total dose. This, in turn, raises questions about the age of the animals (or persons) at the time of exposureperhaps suggesting that the administration of a sufficient dose at a particularly sensitive time in the life of an animal may lead to a higher incidence of cancer. This possibility then raises questions concerning the design of experiments for testing to identify or elucidate the effects of erratic or sporadic exposures of the kind that humans experience. The Multistage Dose-Response Mode' Crump et al. (1976) devised a mathematical approach for fitting a linear dose-response curve at low doses, which is consistent with a no-threshold model. This model presumes that the probability of tumor occurrence is related to dose d, as a product of exponential terms, each in some way related to the probability of the occurrence of a necessary stage in car- . . clnogenesls.
Risk Assessment 269 Equations such as the one given below have several features that make them attractive to policymakers. The equations become very close to linear when the probability of cancer incidence, Prod ), is within a factor of two of the background, independent of the number of steps k built into the formalism or the value of the coefficients qk (Crump et al., 19761. The model leads to a no-threshold extrapolation procedure (predicting a finite excess risk of cancer at any dose): Prods = 1 e -Qua) where Cod) = qO + q~d~ + q2d2 + qkdk; qi > 0, and qk are coefficients to be fit to the data and dk is applied dose raised to the kth power. When applied to data, k can be chosen arbitrarily as being equal to (or one less than) the number of independent data points to be fit or can be allowed to be unrestricted and, thus, be determined by the data. Formally, each of the k terms in Q are believed to be equivalent to a transition between individual steps in a multistep pathway leading to tumor growth. The equation presented above formally corresponds to the multistage model of carcinogenesis, such as that of Armitage and Doll (19541. Threshold Models One of many alternative threshold models that have some heuristic, but less biological, appeal is the Weibull model: Prods = 1 expE(a + Admit, where Prod ~ is the lifetime probability of developing cancer when exposed at a dose rate, d; ax is the natural background incidence; ~ is a slope or potency coefficient (corresponding roughly to the qi in the multistage model); and m is a shape parameter. Thus, m < 1 implies a response curve concave down (hyperlinear), m = 1 implies a linear dose-response curve, and m > 1 implies a convex (rising at high doses more rapidly than linear) dose-response curve (hypolinear). The shape parameter m for the dose indicates the severity of the end point s. The more severe the end point, the steeper the dose-response curve (Carlborg, 19821. The Weibull model has several advantages: · It often fits the laboratory data better than does the multistage or linearized multistage model in the high-dose region where most laboratory experiments are conducted.
270 DRINKING WATER AND H"LTH · It can be modified to permit consideration of time-to-tumor occurrence and time to death. · It coincides with earlier (Druckrey, 1967) formulations of the car- cinogenic process relating median time-to-tumor occurrence and dose (i.e., higher dose leading to lowered median time-to-tumor occurrence). · It can be modified in a manner similar to the multistage model to give an upper limit (i.e., 95% or 99%) to the estimated risk at some low dose. When the only coefficient of consequence in the multistage model is A, and the shape parameter m of the Weibull model equals unity, the mul- tistage and Weibull models give the same low-dose estimates of risk. As discussed by Crump (1985), an estimating or predicting equation is built on three basic premises: · Tumor production is likely to be a multistage, or multistep, process. · Chemical carcinogens do not behave uniquely. Many natural and synthetic substances appear to act in a similar manner. Therefore, the background probability of tumor incidence is never zero. · Tumors can arise from a single transformed cell. In effect, these premises preclude the existence of a dose-response threshold. However, there is substantial evidence suggesting that in mam- malian cells, or in whole animals, there can be a dose response that is experimentally indistinguishable from that which would be predicted if there were a threshold. For instance, in the largest study performed in mice to date (635,000 mice overall), Russell et al. (1982) have shown that testicular mutations produced in response to doses of ethylnitrosourea (ENU) vary in a sigmoidal fashion with dose. In their report, they cited unpublished observations indicating that at each dose, the amount of ENU reaching the testes remains constant. Such behavior resembles a classical threshold dose-response model and suggests that if carcinogen-induced mutation is a necessary first step in carcinogenesis, and external exposure is the only source of the mutation, then there may be a dose-response threshold for tumor formation in mouse testes in response to ENU. Ehling and colleagues (1983) reviewed evidence concerning the exis- tence of thresholds for carcinogens. They found that among the eucaryotic systems studied to date, a threshold or apparent threshold dose response appeared in fewer than 23% of the tested cases. The log-probit model described by Mantel et al. (1975) is one widely discussed threshold model for low-dose extrapolation. This model is based on a classical toxicological description of the dose response and is one of the tolerance models de- scribed earlier. It presumes that each member of a population has a personal carcinogen dose threshold: at a dose above the threshold, they will develop
Risk Assessment 271 one or more tumors, and at an individual subthreshold dose, they are unaffected. Such a model is appealing because it considers individual diversity within a population and is consistent with the traditions of clas- sical reversible toxicology (e.g., many toxicants have a well-defined threshold dose: two aspirin tablets are a mild analgesic, whereas 200 aspirin tablets are lethal). The model, however, has been severely criticized as having little or no biological basis in carcinogenesis (NRC, 1980~. There is some philosophical equivalence between the tolerance distri- bution models and the no-threshold models if one adds the concept of individual differences in probability of response. In the no-threshold mod- els there is no dose level at which there is a zero probability of response for all members of the population. This probability of response, however, may be larger or smaller for specific individuals. Current developments in cancer research related to the presence and distribution of oncogenes in individuals may be expected to lead to cancer incidence models that take into account both the initiator-promoter concepts and the potential for response in individuals, as a consequence of the distribution of numbers and kinds of oncogenes. In general, models that include thresholds for individuals predict a dose response that may become very flat at low doses. After being fit to high- dose data, such a model may predict low-dose risk values substantially lower than those calculated from a linear or linearized multistage model (Swartz et al., 1982~. DEVELOPMENTAL AND REPRODUCTIVE EFFECTS As indicated in Chapters 2 and 3, reproductive impairments of one kind or another are frequent and widespread. Nonetheless, formal risk assess- ment has seldom included measurements of adverse effects on reproduc- tion. Under the Toxic Substances Control Act, information on reproductive effects is not required before a product is marketed. Koeter (1983) recently showed that, in comparison to data from sub- chronic studies, data on reproductive toxicity produced lower estimates of the lowest-observed-effect level (LOEL) for 35% of the compounds tested, the same estimates for another 35% of the compounds, and higher estimates for the remaining 30%. For 8 of the 37 compounds tested (>20%), fertility or reproduction toxicity end points were the most sensitive and thus solely determined the LOEL. Thus, the inclusion of reproductive toxicity effects in a standard test battery would have produced a lower "safe" dose in one-fourth to one-third of these materials. Koeter con- cluded that reproductive function is highly sensitive to impairment and should be examined at earlier stages of safety testing.
2 7 2 OR ~ N K' NG WATER AN D H "LTH At present, there is limited agreement about how to apply the results of animal reproductive and developmental toxicity studies to assess the risk of exposure to a compound. This is partly because an understanding of the underlying events leading to reproductive toxicity is usually missing. There are no agreed-upon standards or quantitative methods for cross- species extrapolation for reproductive or developmental effects. However, there are some conditions under which reproductive toxicity data from animal studies can be used to estimate risk to humans. In the following sections, consideration is given to concepts of susceptibility, timing of exposure, patterns of dose response, interpretation of animal data, and possible methods for cross-species extrapolation. The NOEL If the data are of sufficient quality and quantity, it should be possible to identify a NOEL or LOEL, the maternally toxic dose levels, and the specific types and incidences of adverse effects. Data sets that fail to identify these responses or dose levels are inadequate for quantitative risk assessment. Even without these data, however, a qualitative ranking of agents as having high, moderate, or low potential for developmental tox- icity in humans can at times be made. Agents that selectively induce irreversible developmental toxicity in animals at low, nonmaternally toxic doses are assumed to have the highest potential for causing developmental toxicity in humans. If a high-level, maternally toxic exposure to an agent causes irreversible developmental toxicity, or if a low-level exposure that is not maternally toxic causes reversible variants or minor malformations in animals, the agent should be considered a moderate risk to humans. A low risk to humans can be anticipated if prolonged exposure to high levels of the agent in a well-conducted experiment of sufficient size does not result in any developmental toxicity in animals. The decision to use a NOEL or LOEL approach to risk assessment is based largely on which value can be most accurately identified from the data base. It is usually easier to identify a LOEL from experimental data, because this value can be observed directly, whereas the NOEL can be orders of magnitude below the lowest experimental exposure level ob- served to induce developmental toxicity. The LOEL is not restricted to a dose at which responses are statistically significantly different from control responses. Trends in the data indicating biologically relevant increases in the incidence of adverse effects at low doses can be used to establish a LOEL if there is also an increased incidence of the same effects at high doses or if there is a statistically significant dose-response relationship. LOELs are most accurately identified under conditions where the response is minimal and the end point involves reversible developmental toxicity,
Risk Assessment 273 indicating that a NOEL is being approached. In the absence of statistical significance, which may often be the case because of sample size con- siderations, it is sometimes possible to define a minimal response such as a doubling of the low background rate of the particular response. Protection against doubling of an adverse background rate can be achieved by using a large safety factor for the NOELs selected under these conditions, rec- ognizing that doubling may produce an unacceptably large increase. For major malformations, a doubling would represent an unacceptable increase in the United States from approximately 60,000 malformed infants to 120,000 per year. Quantitative Assessment Some work is under way to develop a quantitative index for comparing developmental toxicity across species taking concurrent maternal toxicity into account (Fabro et al., 1982; Johnson, 19801. This approach derives from the perceived need to distinguish between compounds that are uniquely toxic to the embryo and those that induce developmental toxicity only at exposure levels that are also toxic to the mother. Maternally toxic materials would need to be regulated on the basis of their adult toxicity, whereas regulations for compounds toxic to the embryo only at low doses would be based on that unique embryotoxicity. In attempting to provide a quan- titative index to separate these two types of compounds, Johnson (1980) has defined a "teratogenic hazard potential," which is the log of the ratio of doses that produce adult and developmental toxicity: lowest adult toxic (lethal) dose g lowest developmental toxic dose The higher this so-called AID ratio, the more likely it is that the material has special embryotoxicity. Johnson calculated this ratio for more than 70 compounds using data from an in vitro system of adult and embryonic tissues from Hydra attenuata. (It has been reported that the AID ratio from the hydra assay has ranged from one-tenth to 10 times that of the mam- malian AID ratio.) Most compounds had ratios near 1. Several of them had ratios larger than 5, but very few had ratios larger than 10 (Johnson and Gabel, 19831. The use of this ratio has been proposed as a method for setting priorities for further testing of agents in mammalian develop- mental toxicity studies. Fabro et al. (1982) have begun to explore the quantitative characteristics of a similar type of index in mammalian studies. Dose-response data for adult lethality and fetal malformations were fitted separately (probit of response against log of dose) for eight compounds. The observed log-
274 DRINKING WATER AND HEALTH probit dose-response lines for lethality and teratogenicity were not parallel, nor was there a constant ratio between the slopes for the two lines. Con- sequently, a simple ratio between the median effective lethal and terato- genic doses (i.e., LDso tDso) could not be used. To calculate a relative teratogenic index, Fabro et al. chose one point (i.e., a dose corresponding to some arbitrary percent response) from each dose-response line and generated a ratio from these two points. The Loo value was chosen to represent adult lethality based on the argument that using a low LD value would help guard against compounds that have a shallow dose-response curve for adult lethality. The tDos value, which is the dose causing a 5% increase in the malformation rate above background, was chosen to rep- resent teratogenicity. It was believed that the tDos value could be estimated with confidence for most teratogens because frequency of induced mal- formations often ranges between 1% and 20% in animal studies. This approach appeared to be satisfactory for ranking the candidate compounds according to teratogenic potency provided the dose-teratogenic response relationship was not complicated by significant adult lethality. The selec- tion of other dose levels for computing the ratio could change the relative ranking of the compounds evaluated. This ranking system was developed to assist in evaluating the structure- teratogenicity relationships between structurally related compounds. For this purpose, the Fabro index may be adequate. The usefulness of this index for interspecies comparisons and risk estimation, however, has not been established. In their evaluation of the index, Hogan and Hoel (1982) argued that due to the lack of parallelism between the fitted probit lines for lethality and teratogenicity, the index will not be invariant to the selection of other LD and tD values; e.g., if a ratio of LD~o to tDos were selected instead of an Loo to tDos ratio, a different ranking of the com- pounds could occur. In addition, the index would be subject to the estab- lished deficiencies of the probit model, which tends to be insensitive in the low-dose region. Therefore, until the index is more extensively applied and evaluated, it should not be used for formal risk assessment. If a uniform method for ranking agents according to their embryotoxic potential were based on selective toxicity to the conceptus, it might provide a yardstick for comparing all agents and, thus, possibly standardize a procedure for the selection of the appropriate NOEL or LOEL for risk-assessment pur- poses. The safety factor could then be selected to reflect the severity of the end point. Existing models for quantitative risk assessment do not appear to be appropriate for data on developmental toxicity, for which there are prob- ably threshold doses (Wilson, 19731. To establish safe levels for poly- chlorinated biphenyls, EPA (1983) examined a number of models based on developmental toxicity data. The safe dose for one set of data varied
Risk Assessment 275 by a factor of 7,000, depending on the model used. Rai and Van Ryzin (1985) have proposed a dose-response model for teratological quantal response data where litter size is considered in evaluating the probability of response for one animal in a litter. Their model extends the one-hit model of Hoel et al. (1975), while recognizing that the probability of response for an offspring varies among females exposed to the same dose. Litter size is used as a measure of the female-to-female variability, based on the assumption that the more sensitive animals will have smaller litters. An interagency governmental group concluded that existing mathematical models were inappropriate for assessing developmental toxicity data and that the safety factor approach is appropriate for establishing exposure levels expected to yield acceptable levels of risk (EPA-ORNL, 19821. Nonetheless, the development of models for assessing reproductive tox- icity should be encouraged. The Food and Drug Administration has also indicated that it will use the safety factor approach in developmental toxicity risk assessment, but it has not given specific details on how safety factors will be chosen. The agency has reported, however, that it will apply safety factors ranging from 100 to 1,000 to NOELs identified in developmental toxicity animal testing for drug residues in human food. According to Norcross and Set- tepani (1983), smaller factors will be used when the prenatal effect can be ascribed to nonspecific maternal toxicity. In the absence of other widely accepted approaches, the use of safety factors seems to be a reasonable approach for the establishment of safe levels of exposure to materials that may result in developmental toxicity. The considerations discussed above suggest the following criteria for selecting safety factors for developmental toxicity data: · A minimum quality and quantity of data are required to perform a quantitative risk assessment. Thus, compounds not having a sufficient data base should only be qualitatively assessed for high, moderate, and low potential to cause developmental toxicity in humans, and should be assigned high, moderate, and low safety factors accordingly. · Human populations should be considered to include individuals who are at least 50 times more sensitive than laboratory animals to agents causing well-defined developmental toxicity. · Compounds that lead to developmental toxicity at levels lower than those causing maternal toxicity constitute a greater potential hazard than compounds that cause developmental toxicity only at maternally toxic doses. This potential hazard implies the need for a larger safety factor. · The potential hazard associated with a compound is related to the severity of response and both the time and route of exposure. The greatest
276 DRINKING WATER AND H"LTH potential hazard is presented by compounds causing serious effects under conditions of exposure that may be encountered by humans. NEUROTOXICITY With the exception of lead, there has been little effort devoted to quan- titative risk assessment of chemical neurotoxicants. Risk assessment is substantially more complex for neurotoxicants than for carcinogens be- cause of the many end points associated with neurotoxicity. Moreover, numerous chemicals affect the nervous system of animals when admin- istered in sufficiently high doses. Clinical experience with therapeutic substances has demonstrated that hundreds of pharmaceutical products can induce a variety of neurological and psychiatric disorders at prescribed therapeutic dose levels. By contrast, there are only 30 chemicals or in- dustrial processes for which the International Agency for Research on Cancer (IARC) considers there is "sufficient" evidence of carcinogenicity in humans (IARC, 19821. Few of the substances that produce neurotoxic effects in humans were studied in the experimental laboratory before they were marketed and subsequently observed to produce these effects in humans. This experience stands in marked contrast with that of experimental chemical carcinogen- esis, where several hundred substances have been reported as carcinogens in experimental animals. Although toxicologists have long recognized the susceptibility of the nenous system to toxic perturbation, there has been no coordinated program of animal bioassay to seek out and regulate en- vironmental neurotoxicants other than organophosphorus pesticides. When administered in large doses, many chemical substances produce marked changes in human behavior and neurological function, such as depression of the central nervous system, that disappear rapidly and with no apparent sequelae after exposure has ceased. The same compounds may produce another type of neurotoxic effect in persons chronically exposed to lower doses for long periods. The mechanisms underlying these two (or more) neurotoxic effects may be entirely different. Some of these compounds, such as n-hexane and toluene, cause various types of neurodegenerative diseases after prolonged exposure; others, such as methyl ethyl ketone and acetone, do not appear to induce such chronic effects. Evidence that environmental agents produce adverse effects on the hu- man nervous system can be provided through clinical evaluations of ex- posed humans and experimental animal studies. Definitive demonstration that a substance is neurotoxic comes from complementary investigations showing that the suspect agent produces the same type of disorder in humans and in one or more appropriate test species. In the absence of data on humans, convincing demonstration that a substance is neurotoxic
Risk Assessment 277 in an appropriate animal species can be taken as evidence that the agent is probably neurotoxic in humans. For example, certain well-studied clas- ses of chemicals, such as the organophosphates, reliably produce human- like disorders in fowl. Chronic neurotoxicity may develop many years after exposure to certain chemical substances (e.g., phenothiazines). Negative results obtained from studies conducted over shorter periods should be interpreted carefully, since chronic (lifetime) studies to evaluate agents for neurotoxic properties have rarely been undertaken. Short-term in vitro methods for assessing neurotoxic activity are under development. Tissue culture systems, especially those that exploit a com- plex cellular structure and function comparable to that found in parts of the human nervous system, may ultimately provide surrogates for in viva animal bioassays of potential neurotoxicants. Organotypic tissue cultures composed of structurally and functionally coupled spinal cord, dorsal root ganglia, peripheral nerve, and muscle develop specific types of patholog- ical change (e.g., sensory neuronopathy, axonopathy, myelinopathy) that are also seen in humans and animals challenged with the same substances. These tissue-culture systems also have been used to study alterations of neurotoxic response to one agent (n-hexane) by concurrent exposure to a second (e.g., methyl ethyl ketone, ethanol, toluene). In some cases, structure-activity considerations may be useful in as- sessing whether a substance may pose a hazard to the human nervous system. Examples of compounds with structural similarities that constitute neurotoxicants include some organophosphorus esters (phosphates, phos- phoramidates, phosphonates), pyrethrins, and 1,4-dicarbonyl aliphatic hy- drocarbons. Factors likely to modify human neurotoxicity also may be predicted on those rare occasions when the mechanism is largely under- stood. Examples include the protective action of sulfonates, phosphonates, and carbamates against organophosphorus neuropathy. The NOEL and LOEL Information on general dose response for neurotoxicity in human pop- ulations exposed to environmental chemicals is extremely limited. For the one exception, lead, responses have been well characterized with reliable biological markers of toxicity that may be readily determined by assay of blood or other tissues. The LOELs for different neurological effects are shown in Table 8-2. For populations with special susceptibilities, there are uncertainties about the applicability of general epidemiological findings. For example, chil- dren and women of reproductive age were found to be most susceptible to the toxicity leading to spastic paraparesis during an outbreak in Mo-
278 DRINKING WATER AND H"LTH TABLE 8-2 Lowest-Observed-Effect Levels for Lead-Induced Neurological Effects in Children and Hematological Markers of Exposurea Lowest-Observed- Effect Level Neurological Hematological (PbBb in ,ug/dl) Effects Markers 80-100 Encephalopathic signs and symptoms 70 Frank anemia 50 Peripheral neuropathies 40 Reduced hemoglobin synthesis Elevated coproporphynn Increased urinary b- aminolevulinic acid 30 CNS cognitive effects (e.g., IQ deficits) Peripheral nerve dys- function (slowed nerve conduction velocities) 15 Erythrocyte protopor- phyr~n elevation 10 Altered CNS electro- Inhibition of b-amino- physiological levulinic acid dehy- responses dratase py 5 Nc activity inhibi- tion aAdapted from EPA, 1986, p. 13-34. bPbB = blood lead concentrations. CPy-S-N = pyrimidine-S'-nucleotidase. zambique, which was attributed to subacute cyanide intoxication from the consumption of raw cassava (a cyanogenic plant) during a famine (Ministry of Health, Mozambique, 1984~. The extent to which these findings can be extrapolated to a well-nourished population composed of males and females of all ages is unknown. When there are no data on human populations, animal bioassay data are useful for estimating dose response for substances with toxic properties. Accurate models of human neurological disorders develop in most adult animals challenged with selected chemical substances administered at a suitable dose for an appropriate period. However, extreme caution is necessary when using data on animals since some animal species are relatively refractory to potent human neurotoxicants. For these substances (many of which are probably unknown), evidence of dose response in two or more quite different species (e.g., the rat and chicken) is desirable.
Risk Assessment 279 Since neurotoxic disorders usually increase in severity as a function of dose and duration of exposure, the earliest change directly linked to the disorder under scrutiny represents the most appropriate toxic end point from which to estimate dose response for neurotoxic substances. This assumes that by protecting the target most sensitive to the agent in Question. one would be protecting all important functions. Because of the availability of continuously measurable (as opposed to categoric) end points, risk assessment for neurotoxicity has the potential for greater precision and accuracy than is currently attainable in risk assessments for carcinogenicity. Whenever possible, therefore, continu- ously measurable end points should be selected for study. From a regulatory viewpoint, it is simpler to set exposure limits for noncarcinogens with a clear threshold dose below which no adverse neu- rotoxic response is observed (the NOEL). At low doses of a given chem- ical, for example, the response may lie within the body's ability to maintain homeostasis so that no overt effect occurs. In the presence of additional insults, however, the body's ability to maintain homeostasis may be re- duced, which should be considered as a secondary and possibly unmea- surable toxic effect of the chemical. If there is a threshold dose for this secondary effect, it would most likely be lower than the threshold for the primary effect. Thus, although there may be a threshold dose (LOEL) for a given effect in a particular animal, it is not likely that there will be a single threshold dose for all end points, nor is it likely that a single dose level will constitute the threshold for every animal in a population. When the measurement device, scale, or experiment is insufficiently sensitive, a threshold may be incorrectly inferred where none exists. Conversion of a continuous effect to a categoric scale (e.g., no effect, tremor, convul- sions) can lead to such a false threshold. There appears to be no reason why a neurotoxic response is unlikely to follow common sigmoidal dose-response relationships; however, em- pirical data on the dose response of environmental neurotoxicants are urgently needed to support or refute this conclusion. Quantitative Assessment Since there are no uniformly accepted methods for assessing risks of exposure to neurotoxicants, the following section discusses the theoretical requirements of possible methods. Two intermediate goals of risk assess- ment are to learn whether a given chemical produces a neurotoxic effect and, if so, to determine the nature of the toxicity and the risk associated with various doses. Carcinogenicity studies can be optimally designed to evaluate whether neurotoxic as well as carcinogenic effects occur. How- ever, the calculation of a dose-response relationship for low doses of
280 DRINKING WATER AND H"LTH neurotoxicants is problematic, as indicated in the preceding discussion. Ideally, separate experiments could be conducted to detect neurotoxicity and to estimate risks at specific doses. When resources are limited, con- sideration should be given to developing a statistical design that is a compromise between optimal detection of an effect and estimation of a dose response. In the case of continuous dose-response curves for neu- rotoxicity, exposing small groups of animals to a wide range of doses (e.g., 4 animals exposed to each of 12 doses rather than 12 animals to each of 4 doses) is likely to impair detection only slightly while possibly producing a large amount of model-specific information for use in risk estimation. Dose-response relationships can be regarded in two ways: responses of a more serious nature in a person exposed at increasingly higher doses or a greater percentage of a population having a specified adverse effect of a given intensity as dose is increased. Biologically plausible mathematical models of an individual animal's dose-response relationships can be developed. These, coupled with fairly general assumptions, will lead to log-normal or normal distributions of response severities across a population of animals. When it is inappropriate to use such distributions, distribution-free (nonparametric) or other robust statistical procedures can be used. The fidelity with which animals can serve as models for neurotoxic responses in humans suggests that the same class of mathematical functions for modeling dose-response relationships for humans would also be ap- propriate for animals. Experiments on two or more animal species could be conducted to provide information on the species-to-species variability of the parameters for a specific neurotoxicant. This knowledge could then be used to assess the uncertainty surrounding the extrapolation of animal data to determine the dose-response relationship in humans, and any safety factor (or confidence limit) could be adjusted accordingly. A biologically plausible mathematical model (relating each possible dose to a corresponding level of response severity) for a particular animal is dependent upon a comprehensive understanding of the mechanism whereby a specific neurotoxicant causes a given response or of data from experi- ments specifically designed to help choose among appropriate models. It is reasonable to anticipate that a small class of mathematical functions (characterized by one or possibly two animal-specific parameters) will suffice to describe the dose-response relationship for each animal in a population. Once such a class of functions has been proposed, the vari- ability across the population can be modeled by assuming a probability distribution for the parameters. (Thus, random selection of an animal for an experiment or of a human for exposure to a pollutant corresponds to
Risk Assessment 281 random selection of one dose-response curve from the small class of mathematical functions.) This combining of the two types of dose-response relationships into a single model creates a random coefficients model, a current topic of research published in the biostatistics and econometrics literature. Application of such models to a data set requires estimation of different aspects of the probability distributions. A class of mathematical functions (involving possibly as few as three or four parameters) should suffice to model the animal-specific dose- response relationships for a broad class of neurotoxic substances. Different types of neurotoxic end points may require different classes of mathe- matical models. If resources are dedicated to experiments designed to identify a class of functions sufficient for modeling dose-response relationships for a group of neurotoxicants, benefits for future risk assessments for neurotoxins in the group will include the replacement of two major sources of uncertainty: (1) the problems involved in selecting a model for low-dose extrapolation will be lessened by the development of a procedure with measurable (via confidence limits, for example) uncertainty of lesser magnitude and (2) the uncertainties surrounding species-to-species extrapolation will be re- duced. The concentration of the chemical to which humans are exposed may be measured directly. More typically, however, exposure data are incom- plete and must be estimated. Among the neurotoxicants for which there are rather accurate estimates are lead, mercury, pesticides, and certain biological agents in drinking water. Chemical type, effect, route, duration of exposure, and exposure to possibly interacting materials are other im- portant considerations in exposure assessment. The intake by groups especially susceptible to neurotoxic agents must also be determined. Some compounds interfere selectively with the de- velopment of the nervous system; others only affect adult populations, especially aged individuals with normal deterioration of the nervous sys- tem. Other persons may show genetic susceptibilities to certain compounds (e.g., slow acetylators exposed to isoniazid). Those with metabolic, psy- chiatric, or other disorders associated with neurological changes might have their disease unmasked or exacerbated by concurrent exposure to a neurotoxic agent. Susceptibility to neurotoxicity might also be heightened by medical (drug) treatment, alcohol abuse, or renal or hepatic compro- mise. Malnutrition is another susceptibility factor that alters the risk of neurotoxic disorders both during development and in adult life. People concurrently exposed to two or more neurotoxic agents (e.g., streptomycin and noise) are sometimes at a greater risk for developing neurological disorders (i.e., ototoxicity).
282 DRINKING WATER AND H"LTH CONCLUSIONS AND RECOMMENDATIONS Carcinogenicity Carcinogenesis is a complex process involving multiple steps of initi- ation, promotion, and progression as a cell proceeds from the normal to the tumor state (see Chapter 51. Initiating agents are genotoxic and exhibit cumulative, nonreversible effects. Certain promoting agents also exhibit genetic toxicity, but the precise contribution of this effect to promotion is not yet known. Promoting agents are not directly genotoxic and can produce reversible effects. Data from studies of radiation-induced muta- tions indicate that there is no threshold for initiation. No matter how low the dose of a mutagen, there remains the finite possibility of producing the mutational event. Thus, because cancer derives from the multiplication of a single cell and because there is background disease, the risk assessment models chosen for cancer assume no threshold. Multiple applications of experimental promoters are generally required to produce activity, suggesting that some, but not necessarily all, of the steps in promotion may be reversible (Slaga et al., 19821. Some com- pounds induce cell replication as a result of regenerative growth following severe cell toxicity, whereas others induce hyperplasia directly (Miyazawa et al., 1980; Recknagel, 19671. Increased DNA synthesis following in- creased regenerative growth may result in an increase in spontaneous mutational events, promotional effects, or alteration of expression of those genes controlling cell division. Thus, forced cell proliferation may play a contributory, but not necessarily sufficient, role in the process of car- cinogenesis. Mouse liver cells may be particularly sensitive to this type of action, which may explain the observation that mouse liver tumors have been induced with agents that did not produce tumors at other sites or in other rodent species and which apparently exhibit minimal or no genetic toxicity (Doull et al., 1983; Ward et al., 1979~. These observations have led some to propose that nongenotoxic agents should be regulated because there are likely to be thresholds for these compounds and because the assumptions inherent in linear mathematical risk models do not apply (Kroes, 1979; Stott et al., 1981; \Veisburger and Williams, 1980, 19811. Included in this category are diethylstilbestrol (DES), azathioprine, pheno- barbital, chlorinated hydrocarbons such as dichlorodiphenyltrichloroethane (DDT), and the phorbol esters (Weisburger and Williams, 1980, 1981~. The committee identified four major objections to the proposed dichot- omous approach to risk assessment for carcinogens (IARC, 1983; Perera, 1984; Weinstein, 1983~: 1. The terms genetic, genotoxic, epigenetic, and nongenotoxic have been used in the recent past as operational terms implying that different chemicals
Risk Assessment 283 act via separate, distinctive, and well-defined mechanisms. However, the lines are increasingly blurred as substances previously characterized as nongenotoxic are being shown to alter the structure or sequence of the genetic material. For example, carbon tetrachloride is a weak hepatocar- cinogen (IARC, 1979), but it is extremely hepatotoxic and produces a dramatic increase in cell turnover as a result of regenerative growth (Mir- salis et al., 1982; Recknagel, 19671. Carbon tetrachloride has also been reported to be nonmutagenic in bacteria (McCann et al., 1975) and in mammalian cells (Dean and Hodson-Walker, 1979; Stewart, 1981), and it fails to induce DNA repair in the hepatocytes of treated animals (Mirsalis et al., 19821. Yet, other studies have shown that carbon tetrachloride binds to DNA, RNA (Cunningham et al., 1981; DiRenzo et al., 1982; Rocchi et al., 1973), and protein (Bolt and Filser, 1977) and induces mutation in yeast (Caller et al., 19801. Other substances that were at one time believed to be acting by so-called epigenetic mechanisms, including TPA, DDT, DES, asbestos, saccharin, and phenobarbital, have since been re- ported to exhibit genetic toxicity in some assays. A distinction has been made between primary genotoxicity and secondary- genotoxicity resulting from another activity. This distinction may only be academic, however, because the latter also damages the genome (Becker et al., 1981; Birnboim, 1982~. Furthermore, new information suggests that the promoter TPA has a memory effect lasting for at least 8 weeks (Furstenberger et al., 1983~. Thus, a lack of adequate testing or a limited knowledge of the mechanisms involved could lead to the misclassification of a substance (see Chapter 51. 2. Somatic cell mutation is not necessarily the most important mechanism in carcinogenesis, nor are genetic and nongenetic mechanisms mutually ex- clusive. Many concurrent factors (both genetic and epigenetic) may take part in He process of tumorigenesis. These include chromosome abnormalities, gene rearrangements, oncogene activation, disorders of differentiation, DNA damage, and disruption of DNA repair (see Chapter 51. 3. The threshold issue concerns the shape of He dose-response curve at increasingly small doses where little or no information is available. At present, we do not know He shapes of the low-dose curves or if Here is or is not a true threshold for an animal or human population for any carcinogen (Ehling et al., 1983~. There are no convincing data to support a nonlinear dose- response curve or threshold at very low doses for any carcinogen (Hoer et al., 1983; Weinstein, 19831. This information is not available even for DES and 2,3,7,8-tetrachlorodibenzo-p-dioxin (TCDD), two of He better-studied compounds on He conventional lists of epigenetic agents (Hertz, 1977; Wein- stein, 1983~. 4. Experimental data demonstrating that agents such as TCDD can be carcinogenic by themselves and may have greater carcinogenic potency
284 DRINKING WATER AND H"LTH than many initiating agents contradict the implied assumption that epi- genetic agents carry lower risks. The striking reductions in human cancer risk following decreases in exposure to promoters, e.g., by cessation of smoking and by reduction or elimination of some estrogen therapy (Day and Brown, 1980), strongly support rigorous control of epigenetic agents. In general, quantitative biochemical information is not sufficient for low-dose extrapolation. For carcinogenic risk assessment, the data suggest that a multistage model is consistent with certain qualitative aspects of cancer biology. This model is attractive because for most experimental data, the curve becomes linear at low doses. However, the biochemistry also suggests that regulatory agencies should not be complacent about such a dose-response model, despite its simplicity and its apparent con- servative approach to extrapolation to low doses. The dose response may be fundamentally nonlinear at low doses, and a linear extrapolation may underestimate risk for certain individuals, species, or tissues. Even more importantly, basic biochemical rate concepts suggest that when experi- mental dose-response data are accumulated at doses near the maximal tolerated dose, carcinogen activation, detoxification, and repair pathways may become saturated. Under those circumstances, measured dose- response data might not contain the information required to make a low- dose extrapolation, even if the precise mathematical relationship between dose and response were known independently. In light of such considerations, a generalized multiparameter-fitting protocol may be a reasonable mechanism for generating a low-dose ex- trapolation. However, it is impossible to determine if this is a consistently conservative procedure, which is the type most generally favored by reg- ulatory agencies. The agencies prefer the risk-assessment approach with the greatest potential for protecting human health, i.e., treating all car- cinogens in a similar manner (EPA, 1976; NTP, 1984; Perera, 1984; Weinstein, 1983~. Although it may not be possible to use mechanistic data by extrapolation at this time, the committee hopes that as our un- derstanding of the carcinogenic process increases so will our ability to make better risk assessments. For now, any information on mechanisms of cancer induction that bears on the risk-assessment process should at least be noted by those doing the evaluation. Developmental and Reproductive Toxicity Assessing developmental and reproductive toxicity is especially com- plex due to the great variety of possible toxic end points and the likely involvement of threshold doses. At present there is limited agreement about how to apply the results of animal reproductive and developmental
Risk Assessment 285 toxicity studies to assess the risk of exposure to a compound. This is partly because an understanding of the underlying events leading to re- productive toxicity is usually missing. There are no agreed-upon standards or quantitative methods for cross-species extrapolation. Nonetheless, there are some conditions under which the committee believes that reproductive and developmental toxicity data could be used to estimate risk in humans. When there are sufficient data, the NOEL approach is the most reasonable approach at this time (EPA-ORNL, 19821. It is usually easier to identify a LOEL from experimental data because this value can be observed di- rectly, whereas the NOEL can be orders of magnitude below the lowest experimental exposure level observed to induce a toxic effect. LOELs are most accurately identified under conditions where the response is minimal and the end point involves reversible effects, indicating that a NOEL is being approached. There are several considerations in the selection of an appropriate safety factor to be used with NOELs or LOELs for reproductive toxicity. The committee recommends that humans should be considered to be at least 50 times more sensitive than laboratory animals to agents causing well- defined developmental and reproductive toxicity. Substances that lead to developmental toxicity at levels lower than those causing maternal toxicity constitute a greater potential hazard than substances that cause develop- mental toxicity only at maternally toxic doses. The size of the safety factor should reflect the potential hazard and should consider not only the severity of the response but also the time and route of exposure. The greatest potential hazards are presented by substances causing serious effects under conditions of exposure that may be encountered by humans. Existing models for quantitative risk assessment have not been suffi- ciently well developed to be applied to reproductive toxicity data. How- ever, some encouraging work in progress may result in the production of acceptable models for some types of data. The advantage of modeling reproductive toxicity data over the use of NOELs is that the modeling approach takes into account the slope of the exposure-response curve and the size of the experiment. An example of modeling reproductive data is shown in Chapter 9, where tolerance distribution models were applied to the developmental toxicity data for nitrofen. When insufficient data are available for the NOEL approach or possibly the modeling approach, a ranking system or quantitative index may be used. Underlying this approach is the need to distinguish between sub- stances that are uniquely toxic to the embryo and those that induce de- velopmental toxicity at exposure levels that are also toxic to the mother. Agents in the latter category should be regulated on the basis of their adult toxicity, whereas those in the former would be regulated on the basis of their unique toxicity to the embryo.
286 DRINKING WATER AND H"LTH Neurotoxicity In the four-step process concluding with a risk assessment described by a National Research Council committee (NRC, 1983), the work eval- uating neurotoxicity appears to present serious difficulties that are a natural consequence of the complexities surrounding neurotoxicity. The many neurotoxic effects that can be induced by different exposures range from barely perceptible sensory deficits to gross behavioral or functional ab- normalities. With increasing dose and duration of exposure, one specific effect may be manifested in a larger and larger proportion of the popu- lation, or the number of affected people may not increase but the effects may become more and more serious and incapacitating. Intense short- term effects may occur, but no residual effects may be detectable after exposure ends. At the other extreme, an exposure may produce no ob- servable consequences yet may leave the exposed person highly vulnerable to a subsequent exposure to the same, or even an unrelated, neurotoxicant. Nutritional status has been found to strongly affect responses. A very large number of substances are known to produce neurotoxic effects in humans, in contrast to cancer, for which 30 causative materials or industrial pro- cesses have been implicated (IARC, 19821. Because of the many neurotoxic end points, implying many different mechanisms of action, there are essentially no general mathematical mod- els of neurotoxicity leading to quantitative risk assessment. Models may have to be constructed on a material-by-material basis. In some evaluations for lead, levels of exposure were related to necrologic effects, many of which can be associated with hemolytic markers (Table 4-21. The difficulty in measuring exposures is illustrated by the need to use secondary markers of exposure such as blood lead levels. Regulatory actions are usually based on what have been identified as nominal exposure levels that are objective, measurable external quantities such as parts per million in an air sample. Exposure to these nominal levels, of course, is not the same as the dose, at least not in the biological sense. Lead found in the blood can be regarded as an internal dose (the amount of the substance or its active metabolites in body tissues) or as the biologically effective dose (the amount of the active material that interacts with the tissue or organ). These biological doses are rarely measured; very likely vary with age, sex, and genetic background; and at times may even be unmeasurable. Because quantitative dose-response models or even adequate measures of exposures do not exist, safety factors must be used. A NOEL or a LOEL must be identified and the observed doses divided by an appropriate safety factor. The intensity or seriousness of the response, age-sex vari-
Risk Assessment 287 ation, experiment size, and similar factors need to be taken into account in setting the appropriate safety factor. Use of these safety factors is intended to lead to exposure levels that will be safe for the most sensitive individuals. In general, however, they are usually determined on a maters by-mater~al basis and do not take into account aspects of bioac- cumulation, sensitization, or multiple exposures all factors that need specific consideration. The committee's approach to a quantitative estimate of a safe exposure level through a modified LOEL approach is described for the cholinesterase inhibitor aldicarb in Chapter 9. The approach is based on the argument that if no clinical manifestation of cholinesterase inhibition is observed unless cholinesterase levels are reduced by at least 20% or 30%, then the lower 95% confidence limit on the dose that produced such an inhibition could be looked upon as the acceptable, or maximum permissible, ex- posure level. To assess risk for neurotoxic end points, research must be conducted to develop measures of exposure, including biological markers; better laboratory techniques, including short-term tests for identifying neurotox- icants; and quantitative models for low-dose extrapolation, reflecting the different types of effects and species-to-species variability. Studies must also be undertaken to learn whether or not s~ucture-function relationships can be predictive. The effects of interactions and modified (or sporadic) exposures must also be examined. Furthermore, the relationship of mor- phological changes to neurological or neurotoxic responses should be explored. REFERENCES Anderson, E. L., and CAG (Carcinogen Assessment Group of the U.S. Environmental Protection Agency). 1983. Quantitative approaches in use to assess cancer risk. Risk Anal. 3:277-295. Ariens, E. J., and A. M. Simonis. 1982. General principles of nutritional toxicology. Pp. 17-80 in J. N. Hathcock, ed. Nutritional Toxicology. Vol. 1. Academic Press, New York. Armitage, P., and R. Doll. 1954. The age distribution of cancer and a multi-stage theory of carcinogenesis. Br. J. Cancer 8:1-12. Becker, R. A., L. R. Barrows, and R. C. Shank. 1981. Methylation of liver DNA guanine in hydrazine hepatotoxicity: Dose-response and kinetic characteristics of 7-methylguanine and O6-methylguanine formation and persistence in rats. Carcinogenesis 2:1181-1188. Berenbaum, M. C. 1985. Consequences of synergy between environmental carcinogens. Environ. Res. 38:310-318. Birnboim, H. C. 1982. DNA strand breakage in human leukocytes exposed to a tumor promoter, phorbol myristate acetate. Science 215: 1247-1249.
288 DRINKING WATER AND H"LTH Bolt, H. M., and J. G. Filser. 1977. Irreversible binding of chlorinated ethylenes to macromolecules. Environ. Health Perspect. 21: 107-112. Bresnick, E., M. Brosseau, W. Levin, L. Reik, D. E. Ryan, and P. E. Thomas. 1981. Administration of 3-methylcholanthrene to rats increases the specific hybridizable mRNA coding for cytochrome P-450c. Proc. Natl. Acad. Sci. USA 78:4083-4087. Brown, H. S., D. R. Bishop, and C. A. Rowan. 1984. The role of skin absorption as a route of exposure for volatile organic compounds (VOCs) in drinking water. Am. J. Public Health 74:479-484. Calabrese, E. J. 1983. Principles of Animal Extrapolation. John Wiley & Sons, New York. California Department of Health Services. 1982. Carcinogen Identification Policy: A State- ment of Science as a Basis of Policy. Section 2: Methods for Estimating Cancer Risks from Exposures to Carcinogens. State of California Department of Health Services, Sacramento, Calif. [88 pp.] Callen, D. F., C. R. Wolf, and R. M. Philpot. 1980. Cytochrome P-450 mediated genetic activity and cytotoxicity of seven halogenated aliphatic hydrocarbons in Saccharomyces cerevisiae. Mutat. Res. 77:55-63. Carlbort, F. W. 1982. Speculations on an extended dose-response model for carcinogenesis. Food Chem. Toxicol. 20:319-323. Cohn, M. S. 1986. Estimated carcinogenic risks due to exposure to formaldehyde released from pressed wood products. Tab E in February 1986 Briefing Package to the Com- missioners of the Consumer Product Safety Commission on Formaldehyde Emissions from Urea-Formaldehyde Pressed Wood Products. [53 pp.] (Available from Consumer Product Safety Commission, Washington, D.C. 20207.) Consensus Workshop on Formaldehyde. 1984. Report on the Consensus Workshop on Formaldehyde. Environ. Health Perspect. 58:323-381. Crouch, E., and R. Wilson. 1979. Interspecies comparison of carcinogenic potency. J. Toxicol. Environ. Health 5:1095-1118. Crump, K. S. 1984. A new method for determining allowable daily intakes. Fund. Appl. Toxicol. 4:854-871. Crump, K. S. 1985. Mechanisms leading to dose-response models. Pp. 235-277 in P. Ricci, ed. Principles of Health Risk Assessment. Prentice-Hall, Englewood Cliffs, N.J. Crump, K. S., D. G. Hoel, C. H. Langley, and R. Peto. 1976. Fundamental carcinogenic processes and their implications for low dose risk assessment. Cancer Res. 36:2973-2979. Cunningham, M. L., A. J. Gandolfi, K. Brendel, and I. G. Sipes. 1981. Covalent binding of halogenated volatile solvents to subcellular macromolecules in hepatocytes. Life Sci. 29: 1207-1212. Day, N. E., and C. C. Brown. 1980. Multistage models and primary prevention of cancer. J. Natl. Cancer Inst. 64:977-989. Dean, B. J., and G. Hodson-Walker. 1979. An in vitro chromosome assay using cultured rat-liver cells. Mutat. Res. 64:329-337. DiRenzo, A. B., A. J. Gandolf~, and I. G. Sipes. 1982. Microsomal bioactivation and covalent binding of aliphatic halides to DNA. Toxicol. Lett. 11:243-252. Doull, J., B. A. Bridges, R. Kroes, L. Golberg, I. C. Munro, O. E. Paynter, H. C. Pitot, R. Squire, G. M. Williams, and W. J. Darby. 1983. The Relevance of Mouse Liver Hepatoma to Human Carcinogenic Risk. A Report of the International Expert Advisory Committee to the Nutrition Foundation. The Nutrition Foundation,~Inc., Washington, D.C. 34pp.
Risk Assessment 289 Druckrey, H. 1967. Quantitative aspects in chemical carcinogenesis. Pp. 60-78 in R. Truhaut, ed. Potential Carcinogenic Hazards from Drugs: Evaluation of Risks. UICC Monograph Series. Vol. 7. Springer-Verlag, New York. Dutkiewicz, T., and H. Tyras. 1967. A study of the skin absorption of ethylbenzene in man. Br. J. Ind. Med. 24:330-332. Dutkiewicz, T., and H. Tyras. 1968. Skin absorption of toluene, styrene, and xylene by man. Br. J. Ind. Med. 25:243. Ehling, U. H., D. Averbeck, P. A. Cerutti, J. Friedman, H. Greim, A. C. Kolbye, Jr., and M. L. Mendelsohn. 1983. Review of the evidence for the presence or absence of thresholds in the induction of genetic effects by genotoxic chemicals. ICPEMC Publi- cation No. 10. Mutat. Res. 123:281-341. EPA (U.S. Environmental Protection Agency). 1976. Health risk and economic impact assessments of suspected carcinogens. Interim procedures & guidelines. Fed. Regist. 41:21402-21405. EPA (U.S. Environmental Protection Agency). 1983. Quantitative Risk Assessment of Re- productive Risks Associated with Polychlorinated Biphenyl (PCB) Exposure. Office of Pesticides and Toxic Substances, U.S . Environmental Protection Agency, Washington, D.C. EPA (U.S. Environmental Protection Agency). 1984a. Proposed guidelines for carcinogen risk assessment. Fed. Regist. 49:46294-46301. EPA (U.S. Environmental Protection Agency). 1984b. Proposed guidelines for mutagenicity risk assessment. Fed. Regist. 49:46314-46321. EPA (U.S. Environmental Protection Agency). 1984c. Proposed guidelines for the health assessment of suspect developmental toxicants. Fed. Regist. 49:46324-46331. EPA (U.S. Environmental Protection Agency). 1986. Air Quality Criteria for Lead. Report No. EPA-600/8-83-028C. Environmental Criteria and Assessment Off~ce, U.S. Envi- ronmental Protection Agency, Research Triangle Park, N.C. (final draft) EPA-ORNL (U.S. Environmental Protection Agency-Oak Ridge National Laboratory). 1982. Pp. 99, 111 in Assessment of Risks to Human Reproduction and to Development of the Human Conceptus from Exposure to Environmental Substances. Report No. EPA- 600/9-82-001. U.S. Environmental Protection Agency, Washington, D.C. Fabro, S., G. Shull, and N. A. Brown. 1982. The relative teratogenic index and teratogenic potency: Proposed components of developmental toxicity risk assessment. Teratogen. Carcinogen. Mutagen. 2:61-76. Furstenberger, G., B. Sorg, and F. Marks. 1983. Tumor promotion by phorbol esters in skin: Evidence for a memory effect. Science 220:89-91. Gelboin, H. V. 1980. Benzo[a]pyrene metabolism, activation, and carcinogenesis: Role and regulation of mixed-function oxidases and related enzymes. Physiol. Rev. 60:1107- 1166. Gielen, J. E., F. M. Goujon, and D. W. Nebert. 1972. Genetic regulation of aryl hydro- carbon hydroxylase induction. II. Simple Mendelian expression in mouse tissues in vivo. J. Biol. Chem. 247:1125-1137. Gregus, Z., J. B. Watkins, T. N. Thompson, M. J. Harvey, K. Rozman, and C. D. Klaassen. 1983. Hepatic phase I and phase II biotransformations in quail and trout: Comparison to other species commonly used in toxicity testing. Toxicol. Appl. Phar- macol. 67:430-441. Hathcock, J. N., A. Z. Vary, S. Berger, and A. Brzozowska. 1983. Evaluation of FAO/ WHO pesticide standards in relation to Polish and Honduran diets. Regul. I~oxicol. Pharmacol. 3:216-223.
290 DRINKING WATER AND H"LTH Hertz, R. 1977. The estrogen-cancer hypothesis with special emphasis on DES. Pp. 1665- 1673 in H. H. Hiatt, J. D. Watson, and J. A. Winsten, eds. The Origins of Human Cancer. Book C. Human Risk Assessment. Cold Spring Harbor Laboratory, Cold Spring Harbor, N.Y. Hoel, D. G. 1980. Incorporation of background in dose-response models. Fed. Proc. 39:73-75. Hoel, D. G., D. W. Gaylor, R. L. Kirschstein, U. Saff~otti, and M. A. Schneiderman. 1975. Estimation of risk in irreversible delayed toxicity. J. Toxicol. Environ. Health 1: 133-151. Hoel, D. G., N. L. Kaplan, and M. W. Anderson. 1983. Implication of nonlinear kinetics on risk estimation in carcinogenesis. Science 219:1032-1037. Hogan, M. D., and D. G. Hoel. 1982. Extrapolation to man. Pp. 724-727 in A. W. Hayes, ed. Principles and Methods of Toxicology. Raven Press, New York. IARC (International Agency for Research on Cancer). 1979. Carbon tetrachloride. Pp. 371-399 in IARC Monographs on the Evaluation of the Carcinogenic Risk of Chemicals to Humans. Vol. 20. Some Halogenated Hydrocarbons. International Agency for Re- search on Cancer, Lyon, France. IARC (International Agency for Research on Cancer). 1982. Table 1. Summary evaluations of carcinogenic risk to humans from chemicals, industrial processes and industries based on evidence for carcinogenicity to humans and to animals and for activity in short-term tests. Pp. 17-22 in IARC Monographs on the Evaluation of the Carcinogenic Risk of Chemicals to Humans: Chemicals, Industrial Processes and Industries Associated with Cancer in Humans. IARC Monographs, Vols. 1 to 29. Supplement 4. International Agency for Research on Cancer, Lyon, France. IARC (International Agency for Research on Cancer). 1983. General principles for eval- uating the carcinogenic risk of chemicals. Pp. 14-20 in IARC Monographs on the Eval- uation of the Carcinogenic Risk of Chemicals to Humans. Vol. 30. Miscellaneous Pesticides. International Agency for Research on Cancer, Lyon, France. ICPEMC (International Commission for Protection Against Environmental Mutagens and Carcinogens). 1983. Committee 4 Final Report: Estimation of genetic risks and increased incidence of genetic disease due to environmental mutagens. Mutat. Res. 115:255-291. Johnson, E. M. 1980. A subvertebrate system for rapid determination of potential terato- genichazards. J. Environ. Pathol. Toxicol. 4(5):153-156. Johnson, E. M., and B. E. G. Gabel. 1983. An artificial 'embryo' for detection of abnormal developmental biology. Fund. Appl. Toxicol. 3:243-249. Kelly, H. E. 1980. A survey of water consumption in the New Haven area. M.P.H. thesis. Yale University, New Haven, Conn. Koeter, H. B. W. M. 1983. Relevance of parameters related to fertility and reproduction in toxicity testing. Am. J. Ind. Med. 4:81-86. Kroes, R. 1979. Animal data, interpretation and consequences. Pp. 287-302 in P. Emmelot and E. Kriek, eds. Environmental Carcinogenesis: Occurrence, Risk Evaluation and Mechanisms. Proceedings of the International Conference on Environmental Carcino- genesis held in Amsterdam, May 8-11, 1979. Elsevier/North-Holland, Amsterdam. Land H., L. F. Parada, and R. A. Weinberg. 1983. Cellular oncogenes and multistep carcinogenesis. Science 222:771-778. Lee, P. N., and J. A. O'Neill. 1971. The effect both of time and dose applied on tumour incidence rate in benzopyrene skin painting experiments. Br. J. Cancer 25:759-770. Littlefield, N. A., and D. W. Gaylor. 1985. Influence of total dose and dose rate in carcinogenicity studies. J. Toxicol. Environ. Health 15:545-550.
Risk Assessment 291 Mantel, N., and W. R. Bryan. 1961. "Safety" testing of carcinogenic agents. J. Natl. Cancer Inst. 27:455-470. Mantel, N., N. R. Bohidar, C. C. Brown, J. L. Ciminera, and J. W. Tukey. 1975. An improved Mantel-Bryan procedure for "safety" testing of carcinogens. Cancer Res. 35:865-872. McCann, J., E. Choi, E. Yamasaki, and B. N. Ames. 1975. Detection of carcinogens as mutagens in the Salmonella/microsome test: Assay of 300 chemicals. Proc. Natl. Acad. Sci. USA 72:5135-5139. Michell, B. 1984. Oncogenes and inositol lipids. Nature 308:770. Miller, A. G., and J. P. Whitlock, Jr. 1981. Novel variants in benzo(a)pyrene metabolism: Isolation by fluorescence-activated cell sorting. J. Biol. Chem. 256:2433-2437. Miller, A. G., and J. P. Whitlock, Jr. 1982. Efficient metabolism of benzo(a)pyrene at nanomolar concentrations by intact murine hepatoma cells. Cancer Res. 42:4473-4478. Miller, A. G., D. Israel, and J. P. Whitlock, Jr. 1983. Biochemical and genetic analysis of variant mouse hepatoma cells defective in the induction of benzo(a)pyrene-metabo- lizing enzyme activity. J. Biol. Chem. 258:3523-3527. Ministry of Health, Mozambique. 1984. Mantakassa: An epidemic of spastic paraparesis associated with chronic cyanide intoxication in a cassava staple area of Mozambique. I. Epidemiology and clinical and laboratory findings in patients. Bull. W.H.O. 62:477- 484. Mirsalis, J. C., C. K. Tyson, and B. E. Butterworth. 1982. Detection of genotoxic car- cinogens in the in vivo-in vitro hepatocyte DNA repair assay. Environ. Mutagen. 4:553- 562. Miyazawa, S., S. Furuta, T. Osumi, and T. Hashimoto. 1980. Turnover of enzymes of peroxisomal ,B-oxidation in rat liver. Biochim. Biophys. Acta 630:367-374. Mohrenweiser, H. W., and J. V. Neel. 1982. Models to man: Establishment of reference points for estimating genetic risk in man. Pp. 471-486 in B. A. Bridges, B. E. Butter- worth, and I. B. Weinstein, eds. Indicators of Genotoxic Exposure. Banbury Report 13. Cold Spring Harbor Laboratory, Cold Spring Harbor, N.Y. Moolgavkar, S. H., and A. G. Knudson, Jr. 1981. Mutation and cancer: A model for human carcinogenesis. J. Natl. Cancer Inst. 66:1037-1052. Nebert, D. W., and H. V. Gelboin. 1968. Substrate-inducible microsomal aryl hydroxylase in mammalian cell culture. I. Assay and properties of induced enzyme. J. Biol. Chem. 243:6242-6249. Nebert, D. W., and N. M. Jensen. 1979. The Ah locus: Genetic regulation of the metabolism of carcinogens, drugs, and other environmental chemicals by cytochrome P-450-mediated monooxygenases. CRC Crit. Rev. Biochem. 6:401-437. Nebert, D. W., J. R. Robinson, A. Niwa, K. Kumaki, and A. P. Poland. 1975. Genetic expression of aryl hydrocarbon hydroxylase activity in the mouse. J. Cell. Physiol. 85:393-414. Nishizuka, Y. 1984. The role of protein kinase C in cell surface signal transduction and tumour promotion. Nature 308:693-698. Norcross, M. A., and J. A. Settepani. 1983. Current guides to establishing safe levels of animal drug residues. Paper presented at the International Symposium on the Safety Evaluation of Animal Drug Residues, Berlin, Federal Republic of Germany, 27-29 September, 1983. NRC (National Research Council). 1975. P. 95 in Principles for Evaluating Chemicals in the Environment. National Academy of Sciences, Washington, D.C. NRC (National Research Council). 1977. Drinking Water and Health. National Academy of Sciences, Washington, D.C. 939 pp.
29 2 OR ~ N. K! NG WATER AN D H "LTH NRC (National Research Council). 1980. Drinking Water and Health. Vol. 3. National Academy Press, Washington, D.C. 415 pp. NRC (National Research Council). 1982. Cancer incidence and mortality. Pp. 16-1 to 16-19 in Diet, Nutrition, and Cancer. National Academy Press, Washington, D.C. NRC (National Research Council). 1983. Risk Assessment in the Federal Government: Managing the Process. National Academy Press, Washington, D.C. 191 pp. NRC (National Research Council). 1985. Epidemiology and Air Pollution. National Acad- emy Press, Washington, D.C. 224 pp. NTP (National Toxicology Program). 1984. Report of the NTP Ad Hoc Panel on Chemical Carcinogenesis Testing and Evaluation. U.S. Department of Health and Human Services, Washington, D.C. 280 pp. OSTP (Office of Science and Technology Policy). 1985. Chemical carcinogens; a review of the science andits associated principles, February 1985. Fed. Regist. 50:10371-10442. OTA (Office of Technology Assessment). 1981. Assessment of Technologies for Deter- mining Cancer Risks from the Environment. Office of Technology Assessment, Congress of the United States,Washington, D.C. 240 pp. Pellizzari, E. D., T. D. Hartwell, C. M. Sparacino, L. S. Sheldon, R. Whitmore, C. Leininger, and H. Zelon. 1984. Total Exposure Assessment Methodology (TEAM) Study: First Season Northern New Jersey. Interim Report RTI/2392/03-03 S. Research Triangle Institute, Research Triangle Park, N.C. 321 pp. Perera, F. P. 1984. The genotoxic/epigenetic distinction: Relevance to cancer policy. Environ. Res. 34:175-191. Rai, K., and J. Van Ryzin. 1985. A dose-response model for teratological experiments involving quantal responses. Biometrics 41:1-9. Recknagel, R. O. 1967. Carbon tetrachloride hepatotoxicity. Pharmacol. Rev. 19:145-208. Rocchi, P., G. Prodi, S. Gnlli, and A. M. Ferren. 1973. In viva and in vitro binding of carbon tetrachlonde with nucleic acids and proteins in rat and mouse liver. Int. J. Cancer 11 :419-425. Russell, W. L., P. R. Hunsicker, G. D. Raymer, M. H. Steele, K. F. Stelzner, and H. M. Thompson. 1982. Dose-response curve for ethylnitrosourea-induced specific-locus mutations in mouse spermatogonia. Proc. Natl. Acad. Sci. USA 79:3589-3591. Shehata, A. T. 1984. An assessment of the contribution of chemically contaminated drinking water to routes of exposures. Paper presented at EPA Symposium on Advances in Health Risk Assessment for Systemic Toxicants and Chemical Mixtures, October 23-25, 1984, Cincinnati. Slaga, T. J., S. M. Fischer, C. E. Weeks, A. J. P. Klein-Szanto, and J. Reiners. 1982. Studies on the mechanisms involved in multistage carcinogenesis in mouse skin. J. Cell. Biochem. 18:99-119. Stewart, B. W. 1981. Generation and persistence of carcinogen-induced repair intermediates in rat liver DNA in vivo. Cancer Res. 41:3238-3243. Stott, W. T., R. H. Reitz, A. M. Schumann, and P. G. Watanabe. 1981. Genetic and nongenetic events in neoplasia. Food Cosmet. Toxicol. 19:567-576. Swartz, J. B., C. R. Riddiough, and S. S. Epstein. 1982. Analyses of carcinogenesis dose- response relations with dichotomous data: Implications for carcinogenesis risk assess- ment. Teratogen. Carcinogen. Mutagen. 2:179-204. Tukey, R. H., D. W. Nebert, and M. Negishi. 1981. Structural gene product of the [Ah] complex: Evidence for transcriptional control of cytochrome P~-450 induction by use of a cloned DNA sequence. J. Biol. Chem. 256:6969-6974. Ward, J. M., R. A. Griesemer, and E. K. Weisburger. 1979. The mouse liver tumor as an endpoint in carcinogenesis tests. Toxicol. Appl. Pharmacol. 51:389-397.
Risk Assessment 293 Weinstein, I. B. 1983. [Letter: Carcinogen policy at EPA.] Science 219:794-796. Weisburger, J. H., and G. M. Williams. 1980. Chemical carcinogens. Pp. 84-138 in J. Doull, C. D. Klaassen, and M. O. Amdur, eds. Casarett and Doull's Toxicology: lithe Basic Science of Poisons, 2nd ed. Macmillan, New York. Weisburger, J. H., and G. M. Williams. 1981. Carcinogen testing: Current problems and new approaches. Science 214:401-407. Wilson, J. G. 1973. Environment and Birth Defects. Academic Press, New York. Young, H. D. 1962. Statistical Treatment of Experimental Data. McGraw-Hill, New York. 172pp.
