Reevaluating Cancer Risk Estimates for Short-Term Exposure Scenarios

N. Christine Halmes*, Stephen M. Roberts{dagger},1, J. Keith Tolson{dagger} and Christopher J. Portier{ddagger}

* TERRA Inc., Denver, Colorado; {dagger} Center for Environmental and Human Toxicology, University of Florida, Gainesville, Florida; and {ddagger} National Institute for Environmental Health Sciences, Research Triangle Park, North Carolina

Received May 11, 2000; accepted August 7, 2000


    ABSTRACT
 TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
 RESULTS
 DISCUSSION
 REFERENCES
 
Estimates of cancer risk from short-term exposure to carcinogens generally rely on cancer potency values derived from chronic, lifetime-exposure studies and assume that exposures of limited duration are associated with a proportional reduction in cancer risk. The validity of this approach was tested empirically using data from both chronic lifetime and stop-exposure studies of carcinogens conducted by the National Toxicology Program. Eleven compounds were identified as having data sufficient for comparison of relative cancer potencies from short-term versus lifetime exposure. The data were modeled using the chronic data alone, and also using the chronic and the stop-exposure data combined, where stop-exposure doses were adjusted to average lifetime exposure. Maximum likelihood estimates of the dose corresponding to a 1% added cancer risk (ED01) were calculated along with their associated 95% upper and lower confidence bounds. Statistical methods were used to evaluate the degree to which adjusted stop-exposures produced risks equal to those estimated from the chronic exposures. For most chemical/cancer endpoint combinations, inclusion of stop-exposure data reduced the ED01, indicating that the chemical had greater apparent potency under stop-exposure conditions. For most chemicals and endpoints, consistency in potency between continuous and stop-exposure studies was achieved when the stop-exposure doses were averaged over periods of less than a lifetime—in some cases as short as the exposure duration itself. While the typical linear adjustments for less-than-lifetime exposure in cancer risk assessment can theoretically result in under- or overestimation of risks, empirical observations in this analysis suggest that an underestimation of cancer risk from short-term exposures is more likely.

Key Words: risk assessment; carcinogenicity; cancer bioassays; estimating cancer risks; stop-exposure studies.


    INTRODUCTION
 TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
 RESULTS
 DISCUSSION
 REFERENCES
 
Carcinogenesis is generally considered the result of a multi-step process consisting of multiple mutations occurring in some predefined sequence, selective growth of mutated cells, and the eventual progression of cells to malignancy. The basic experimental paradigm used to identify carcinogens and estimate their potency is the chronic animal cancer bioassay. Cancer bioassays conducted by the National Toxicology Program (NTP) generally involve exposure of the animal for 2 years, are accompanied by a full histopathological evaluation of over 40 tissues/organs, and include a rigorous review process. Cancer bioassays conducted by toxicologists elsewhere may differ somewhat, but all typically involve exposure to a potential carcinogen over a major portion of the life span of the animals, generally two years for rodents. The U.S. Environmental Protection Agency (EPA), the U.S. Food and Drug Administration (FDA), and other regulatory agencies use data from these lifetime exposure studies to derive cancer potency estimates that are used for risk assessment purposes.

Conventional cancer risk assessments are generally predicated on the assumption that cancer risk increases as a function of the cumulative carcinogen dose. For exposure to a carcinogen at a given rate, this means that the excess cancer risk is a linear function of exposure duration (e.g., exposure for 7 years is assumed to carry one-tenth the cancer risk of exposure for 70 years). This is analogous to the concept of toxicity as a linear function of concentration and time, known as Haber's Law, although the original work of Haber (1924) involved the acute lethality of gases rather than chemical carcinogenesis. Evidence to support the validity of this relationship in carcinogenicity comes chiefly from studies of antineoplastic agents inducing secondary cancers in humans and primary cancers in rodents (Dedrick and Morrison, 1992Go; Kaldor et al., 1988Go).

Despite this evidence, there is increasing concern that cancer risk may not be a linear function of cumulative carcinogen dose, and that the use of cancer potency estimates from lifetime studies may not adequately estimate cancer risks from less-than-lifetime exposures (Chen et al., 1988Go; Crump and Howe, 1984Go; Kodell et al., 1987Go; Murdoch et al., 1992Go; U.S. EPA, 1998Go). In part, this concern arises from inconsistency between the cumulative dose concept and current mathematical models of cancer. For example, the two-stage model of carcinogenesis predicts different risks for early exposure to initiators than for late exposures (Murdoch et al., 1992Go; Portier, 1987Go). Also, empirical data to support this assumption are extremely limited (e.g., Aylward et al., 1996; Dankovic and Staynor, 1998). Given that there are many situations such as short-term occupational or accidental exposures where the use of models and potency estimates developed from lifetime cancer bioassays might lead to biased cancer risk estimates, it is imperative that this assumption be evaluated critically.

The cumulative dose assumption predicts that a short-term carcinogen exposure, when averaged over the life span of the animal or human, will yield risk equivalent to a continuous exposure given at this averaged dose (e.g., U.S. EPA, 1996b). For example, the risk of cancer in an animal given dose d of a possible carcinogen for 1 year and necropsied at 2 years is assumed to be equivalent to the risks for an animal given dose d/2 continuously for the 2 years and then necropsied. It is possible to test this assumption by comparing observations from short-term versus lifetime-exposure cancer studies for specific chemicals. Tumorigenicity data for short-term and lifetime exposures exist in the literature for many chemicals, but, in general, the studies are poorly matched. That is, few studies incorporate both short-term and lifetime exposures, and differences in species, strain, dosing regimen, histopathological criteria, and other critical variables make it difficult to compare observations following lifetime exposure to a chemical in one study with those following less-than-lifetime administration in another. There is one data set, however, that is uniquely well suited to this kind of analysis—the stop-exposure studies conducted by the NTP. In these studies, some animals are exposed to the test chemical for the standard 2-year period, while other animals of the same strain are exposed for a more limited time, followed by an exposure-free period for the remainder of the 2-year study. The animals are the same age at sacrifice and analyzed collectively for tumor incidence. The dosing rate (in mg/kg/day) for the stop-exposure animals is not always the same as for 2-year-treated animals, but except for dose and duration (the 2 determinants of cumulative dose), animals in these treatment groups are probably as closely matched as can be reasonably achieved.

The analysis reported here uses data from the NTP stop studies to test the hypothesis that short-term exposure to carcinogens, when compared to lifetime exposure, result in proportional decreases in cancer risk. Data for 11 chemicals were found to be suitable for these analyses, and tumor incidence as a function of dose and time was evaluated through comparison of observations in stop-exposure and 2-year-exposure treatment groups.


    MATERIALS AND METHODS
 TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
 RESULTS
 DISCUSSION
 REFERENCES
 
Data
The NTP has conducted almost 500 2-year chronic carcinogenicity studies. Twelve of these studies were identified as including stop-exposure groups. Chemicals evaluated in these 12 studies were 1-amino-2,4-dibromoanthraquinone, 2,2-bis(bromomethyl)-1,3-propanediol, 1,3-butadiene, coumarin, 3,4-dihydrocoumarin, furan, hexachlorocyclopentadiene, methyleugenol, o-nitroanisole, oxazepam, pentachlorophenol, and salicylazosulfapyridine. Of the 12 studies identified, 11 were chosen for this analysis (see Table 1Go). Hexachlorocyclopentadiene was excluded because no evidence of carcinogenicity was found in either the chronic or the stop study (NTP, 1994).


View this table:
[in this window]
[in a new window]
 
TABLE 1 NTP Stop-Exposure Studies Selected for Analysis
 
All of the remaining 11 chemicals demonstrated evidence of carcinogenicity following either lifetime- or stop-exposure to the chemical agent, or both. For some of the chemicals, (e.g., coumarin) only one organ/tissue was affected, whereas for others (e.g., 2,2-bis(bromomethyl)1,3-propanediol), increased tumors occurred at numerous sites. Data for each site were analyzed separately. Ten of the 11 chemicals had stop-exposure groups only in rats, and one (1,3-butadiene) had stop-exposure data only in mice. This analysis focuses on males from these studies.

Statistical Methods
The NTP has been evaluating the statistical significance of tumor-incidence rates in their 2-year cancer bioassays for many years. The methods used have included Fisher's exact test, logistic regression, Peto's test, and the poly-3 trend test. In this reanalysis of NTP data, significance of a finding was based upon the judgement of the NTP peer-review panel as stated in the summary of the technical report for each chemical; no new analysis for significance was conducted. Cancer endpoints regarded by NTP as potentially significant are summarized for each of the 11 chemicals in Table 1Go.

The poly-3 adjusted (Portier and Bailer, 1989Go) cancer rates were fit to a Weibull model using maximum-likelihood estimation. The Weibull model has the form:

(1)
where P(d) is the lifetime probability of cancer, given continuous exposure to dose d of the test agent; {alpha}, ß and {gamma} are parameters being estimated from the data; and {alpha} >= 0, ß >= 0, {gamma} >= 0.5. The parameter {alpha} pertains to the estimated background tumor incidence rate, ß pertains to the magnitude of the risk (a potency value), and {gamma} describes the shape of the dose-response curve ({gamma} < 1 implies supralinearity or a dose-response that bends upward; {gamma} > 1 implies sublinearity or a threshold-looking dose-response). The estimates of the parameters are obtained by finding values for {alpha}, ß, and {gamma} that maximize the log-likelihood function given by:

(2)
where xi represents the number of animals in the group given dose di, and ni represents the survival-adjusted (by the poly-3 procedure) number of animals in this group; i = 0, 1,..., to k groups.

To test whether exposure groups with stopped exposure differ markedly from the continuous-exposure groups when dose is averaged over the lifetime of the animal, a likelihood ratio test was used. The test was applied by fitting the Weibull model to all of the data (continuous and stop-exposure groups) and the maximum value of the log-likelihood is obtained (L1). A modified Weibull model was fit to the same data, again using maximum likelihood estimation (L2 = maximum value of the log-likelihood) in which the stop-exposure groups each have an additional parameter of the form:

(3)
where Ii = 1 if the group given dose di is a stop exposure group, Ii = 0 if the exposure is continuous in the group given dose di, and {Delta}i is a parameter to be estimated. The test statistic {chi}k = 2(L2-L1) is distributed as a Chi-squared random variable with k degrees of freedom, where k is the number of stop-exposure groups (Bickel and Doksum, 1977Go) and tests if, simultaneously, all of the {Delta}i = 0. This test is referred to as the "averaged-dose test".

For each tumor response observed in a bioassay, the dose yielding an excess risk of 1% (ED01) was calculated. This is defined as the value satisfying the equation:

(4)
where P(d) is given by Equation 1Go. From these data, it is possible to invert the problem and find the averaging time that yields perfect agreement between the observed tumor incidence and that predicted by the model fit only to the continuous exposure data; this quantity will be referred to as the "equivalent averaging time". If P(d*E/A) is the predicted response to stop-exposure dose d (in daily dose units) given for E days and averaged for A days, and O(d) is the observed tumor response for this stop-exposure group, then the equivalent averaging time is found by locating the value of A, which solves the equation O(d) = P(d*E/A). For the Weibull model used in this analysis, the solution is:

(5)
where {alpha}, ß, and {gamma} are the estimated parameters from the Weibull model fit. If the assumption is that short-term exposures, when averaged over the life span of the animal, yield risk equivalent to a continuous exposure, A should equal 104 weeks, or 2 years. In some cases, A is less than the actual time exposed due to the risk in the stop-exposure group being higher than that seen in any chronic exposure group. In these cases, the result is noted simply as less than the time exposed.


    RESULTS
 TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
 RESULTS
 DISCUSSION
 REFERENCES
 
The objective of this analysis was to test the assumption that short-term exposures to carcinogens have the same dose-response relationships as long-term, continuous exposures when dose and duration are taken into account through dose-averaging. This assumption was tested in a number of ways. In the first, each dose of each chemical used in a stop-exposure experiment was converted to a lifetime (two-year) average dose. The "averaged-dose test" (described above) was then used to determine, for each cancer endpoint, whether the tumor responses from short-term exposures fell on the dose-response line generated by continuous, lifetime exposure for that chemical. Data for 3 tumor sites are presented in Table 2Go as examples. For heart hemangiosarcomas following exposure to 1,3-butadiene (Table 2Go), the p-value for the averaged dose test is <0.01, indicating a statistically significant difference between the stop-exposure responses and the curve predicted by fitting the dose-response model to data from continuous exposure. A significant difference (p < 0.01) was also observed for leukemia following exposure to o-nitroanisole (Table 2Go), while for histiocytic sarcomas following 1,3-butadiene exposure (Table 2Go), the responses were not significantly different (p = 0.76).


View this table:
[in this window]
[in a new window]
 
TABLE 2 Examples of Bioassay Results Illustrating Cases Where Stop-Exposure (Expressed as a 2-Year Average) Agrees or Disagrees with the Results from the Continuous Exposure
 
The responses observed for animals in the stop-exposure groups, when dose-averaged over 2 years, were significantly higher than those of the continuously-exposed animals for at least one site for 6 of the 11 chemicals we examined (Table 3Go). For furan (1 tumor site) and pentachlorophenol (2 tumor sites), tumor responses following stop exposure were significantly greater than predicted by continuous exposure data for all cancer endpoints. The other 4 chemicals demonstrated mixed responses ranging from most of the tumor sites being significantly elevated (2,2-bis(bromomethyl)-1,3-propanediol and o-nitroanisole) to chemicals for which the majority of sites did not differ significantly (methyleugenol). Two chemicals displayed responses for the stop-exposures below what was expected from the chronic exposures; o-nitroanisole had 2 significantly reduced responses (also 5 increased and 1 not significant) and 1-amino-2,4-dibromoanthraquinone had 1 reduced site (versus 4 with no difference). 1,3-Butadiene, with the most elaborate pattern of dosing including 4 stop-exposure groups, showed 5 tumor sites for which the stop-exposure responses were significantly increased; 3 sites for which the responses were significantly different from the chronic studies but with some stop-exposure responses increased and some reduced; and 5 sites with no significant differences. On a site-by-site basis, 33 of the 59 sites with increased cancer displayed a statistically significant lack of agreement between the stop-exposure group responses and the chronic responses, when the dose used for the stop exposure was averaged over the length of the chronic study. This lack of agreement is considerably greater than what would be expected by chance alone, indicating a serious problem with the common use of dose-averaging for short-term exposures.


View this table:
[in this window]
[in a new window]
 
TABLE 3 Summary of Comparison of Tumor Responses Between 2-Year Continuous Exposure and Stop-Exposure Groups from 11 NTP Carcinogenicity Studies
 
The observation of statistically significant changes in response does not necessarily imply significant changes in estimated risk. The impact of the differences was evaluated by comparing the estimated dose yielding a 1% tumor response (ED01) using only the results from the chronic-exposure groups to the ED01 estimated by including the stop-exposure group(s). Examples of the ED01`s for selected individual sites are given in Table 2Go. The differences are categorized for all chemical/tumor site combinations in Table 4Go. In a number of instances, the effect on ED01 of including stop-exposure data could not be assessed. For 4 chemicals (10 tumor sites), positive dose-response was only seen when the stop-exposure groups were included in the analysis, thereby making this comparison uninformative. Addition of the stop-exposure groups sometimes (2 responses in 2 chemicals) resulted in negative dose-response, and hence, in no estimated ED01 for comparison with estimates using only the continuous-exposure data. In one case, this occurred because the averaged dose for the stop-exposure group fell in the lower half of the experimental range, and the higher exposures, with no tumors, forced the dose-response curve downward (squamous cell papillomas of the forestomach for 2,2-bis(bromomethyl)-1,3-propanediol). In the other case (o-nitroanisole), a weak response (renal tubular cell tumors) was negated by zero response in the stop-exposure groups.


View this table:
[in this window]
[in a new window]
 
TABLE 4 Changes in ED01 Values Resulting from the Inclusion of Responses from the Stop-Exposure Groups Using the 2-Year Averaged Responses
 
Inclusion of responses from the stop-exposure groups decreased the ED01 by a factor of >2 for tumors at 1 or more sites for 6 of the 11 chemicals: 1-amino-2,4-dibromoanthraquinone (1 of 5 cancer endpoints), 2,2-bis(bromomethyl)-1,3-propanediol (12 of 15 cancer endpoints that could be evaluated), 1,3-butadiene (8 of 11 cancer endpoints), methyleugenol (one of 6 cancer endpoints), o-nitroanisole (1 of 2 cancer endpoints), and coumarin (1 of 1 cancer endpoint). An example of this magnitude of reduction is illustrated by heart hemangiosarcomas following exposure to 1,3-butadiene (Table 2Go) where the ED01 with the stop-exposure groups included was 1.25 as compared to 3.83 when they were excluded (a 3-fold decrease). Overall, of the 47 chemical/tumor site combinations with positive ED01 estimates for both the continuous exposure data and the continuous/stop-exposure data combined, 24 had ED01 values that were at least 2-fold smaller when the stop studies were included. For 9 of these chemical/tumor site combinations, the difference was greater than 10-fold, with the largest difference being a 102-fold decrease for lymphomas in the study of 1,3-butadiene. Finally, 37 of the 47 sites demonstrated a decrease in ED01 when the stop exposures are included, indicating a potential bias towards higher risk with shorter exposures.

There were only 2 cases of a 2-fold or higher increase in the ED01 (kidney adenomas in the methyleugenol study, a 3.2-fold increase; and mononuclear cell leukemia in the o-nitroanisole study, a 6-fold increase shown in Table 2Go). Overall, 10 of the 47 groups resulted in an increase in the ED01.

Another way to evaluate dose averaging is to determine the length of time one should average the stop-exposure dose so that the observed response falls exactly on the line resulting from fitting the continuous-exposure data. The method used for the calculation of "equivalent averaging times" is given as equation (5Go). An equivalent averaging time of 104 weeks would indicate that short-term exposures, averaged over the duration of the chronic study (i.e., 104 weeks), produced results consistent with continuous lifetime exposures. To illustrate the concept, equivalent averaging times for heart hemangiosarcomas for the 13-week stop-exposure group given 1,3-butadiene was 57 weeks (Table 2Go). This indicates that for this chemical the stop-exposure dose (625 ppm), averaged over 57 weeks, would yield results similar to continuous lifetime exposure. Expressing this in a different way, if the dose used for this group in the dose-response analysis was 625 ppm x (13 weeks/57 weeks) = 142.5 ppm, the observed response would perfectly match the predicted response from fitting the model (Equation 1Go) to the continuous-exposure-only data.

Table 5Go summarizes the equivalent averaging times for the tumors/sites where these could be estimated. It is clear from this table that most of the averaging times are less than the full 2 years. In some circumstances, the equivalent averaging time was quite comparable with the actual treatment duration. Examples include forestomach squamous cell papilloma or carcinoma from 27-week stop exposure to 18,000 ppm o-nitroanisole (equivalent averaging time: 32 weeks) and Zymbal's gland adenomas and carcinomas from 13-week exposure to 2,2-bis(bromomethyl)-1,3-propanediol (equivalent averaging time: 12 weeks). In these situations, to achieve comparable results between the stop-exposure and continuous exposure tumor rates, the most appropriate averaging time for the stop-exposure was the exposure duration itself. There were also cases where the equivalent averaging time was less than the actual treatment time, such as the tumor response seen with 65-week stop exposure to coumarin compared with results from either shorter or longer exposures to the same dose (equivalent averaging time: 7 weeks) and several endpoints for 1,3-butadiene. For most endpoints, however, the equivalent averaging times for the stop-exposure studies were longer than the actual exposure durations, but less than 104 weeks, suggesting that short-term exposures were generally more effective in producing tumors than continuous exposure studies would predict. In Table 5Go, the median equivalent averaging time was less than 104 weeks for 12 of the 14 dose groups for which this comparison could be made. The median equivalent averaging time for all of the groups was 62 weeks.


View this table:
[in this window]
[in a new window]
 
TABLE 5 Equivalent Averaging Times for Positive Tumor Sites in 11 NTP Studies using Stop-Exposure Groupsa
 

    DISCUSSION
 TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
 RESULTS
 DISCUSSION
 REFERENCES
 
Contemporary risk-assessment methodology estimates cancer risks using slope factors typically derived from 2-year bioassays in rodents. These cancer-slope factors are based on, and intended to represent, potency of the chemical under conditions of continuous lifetime exposure. In many situations, exposure to a potential carcinogen is not continuous over a lifetime, but rather intermittent or of short duration. When estimating risk from less-than-lifetime exposure, the assumption is usually made that cancer risk from exposure at a given rate decreases proportionally with decreasing exposure duration, and adjustments are made in the risk calculation accordingly. Although this approach represents an expedient means to develop cancer risk estimates for limited exposure duration scenarios, its validity has never been well established experimentally.

There is ample basis to suspect that short-term exposures may yield different cancer potency estimates from continuous-exposure studies, and that the use of slope factors developed from chronic bioassays to estimate cancer risks from short-term exposures may lead to errors. If a carcinogen affects a human or animal primarily in a particular life stage, a short-term exposure during that stage may be very effective in producing cancer, while the same short-term exposure during a different life stage may be ineffective. When compared to continuous exposure over a lifetime, the apparent potency of the short-term exposure may be greater or less than that of continuous exposure, depending upon when the short-term exposure occurs relative to the period of greatest susceptibility to that particular carcinogen (Crump and Howe, 1984Go). For example, carcinogens that selectively affect initial steps in the carcinogenesis process would be more effective when exposure occurs early in life, while carcinogens that affect latter stages (i.e., "progressors" or "completors") would have greatest effect when exposure occurs later in life. A short-term exposure during the most relevant life stage would result in a greater potency estimate compared with continuous exposure in which much of the dose may be administered during times when the carcinogen is relatively ineffective. Consistent with this, data available on perinatal carcinogen exposure suggest that the rate of exposure in early life may more closely describe the carcinogenic potency than average lifetime exposure (McConnell, 1992Go; U.S. EPA, 1996aGo). These studies did not address in utero exposure, making it difficult to reach general conclusions.

Continuous exposure could also produce, on the other hand, a higher estimate of cancer potency than short-term exposure, if the short-term exposure occurs during a period of relative invulnerability. Using various models of carcinogenesis, it has been predicted that use of cancer potency values based on lifetime exposures to derive risk estimates for short-term exposures could result in a several-fold under- or over-estimation of risk, depending upon the nature of the carcinogen and the timing of the short-term exposure (Chen et al., 1988Go; Kodell et al., 1987Go; Murdoch et al., 1992Go). These theoretical observations have never been systematically evaluated using actual data until now, although there are some examples of conclusions for individual chemicals (e.g., Drew et al., 1983; Melnick et al., 1990). Information with which to compare potency of carcinogens under conditions of short-term versus continuous exposure is quite limited. For the analysis presented here, data from NTP stop-exposure studies conducted concurrently with chronic bioassays were used because they are well matched in terms of a variety of tangible and intangible variables that could affect tumorigenicity. An indication of the direction and magnitude of differences in cancer potency between stop-exposures and continuous exposures was gained by estimating cancer potency with and without inclusion of stop-exposure data. Among the 11 chemicals included in this analysis, tumor response was significantly greater than expected by dose-averaging in one or more tumor sites in the stop-exposure groups for 6 of the compounds and less than expected for 2. Inclusion of stop-exposure data with continuous-exposure data resulted in a 2-fold or greater reduction in ED01 values for at least one endpoint, compared with ED01 values derived from continuous-exposure data alone, for 6 of the 11 carcinogens. Some of the carcinogens produced increased tumors at several sites (e.g., 12 sites for 2,2-bis(bromomethyl)-1,3-propanediol). When the positive tumor data for all sites from the 11 compounds are viewed collectively, inclusion of stop-exposure data more than doubled cancer potency (i.e., 2-fold decrease in the ED01) in roughly 50% of the cases. These observations suggest that the apparent cancer potency from short-term exposures was usually greater than that from continuous exposure.

The magnitude of the difference in cancer potency derived from stop-exposures versus continuous exposures is difficult to estimate precisely. Derivation of cancer potency estimates from individual stop-exposure studies was not attempted—usually only one dose was tested in each of the stop-exposure experiments, making accurate estimation of potency difficult. In most cases, inclusion of stop-exposure data in the calculation of the ED01 for a cancer endpoint resulted in less than a 10-fold change (see Table 4Go), suggesting that differences in potency between stop and continuous exposures were in the range of one order of magnitude. Seven tumors from 2,2-bis(bromomethyl)-1,3-propanediol and 2 tumors for 1,3-butadiene were exceptions to this. A greater than 50-fold decrease in ED01 values were found when stop-exposure data were included for 3 cancer endpoints for 2,2-bis(bromomethyl)-1,3-propanediol and for one cancer endpoint for 1,3-butadiene, suggesting differences in potency under stop-study conditions can be large.

For o-nitroanisole, 4 sites in the stop-exposure group had tumors, while no tumors were found in these sites in the chronic groups, and for pentachlorophenol there was no evidence of carcinogenicity in the chronic groups, while two sites in the stop-exposure group had tumors. Interpretation of these very large changes is complicated somewhat by the fact that stop-exposure studies for these particular chemicals were conducted with dosages greater than those used in the continuous-exposure studies, and that tumor response for the continuous doses was sometimes very low, making measures of proportionate change unstable. As such, these very large apparent differences in potency may not be reliable, or could be due to biological effects associated with high dosing rates in the stop-exposure studies that do not occur with lesser, continuous exposure (e.g., effects involving a threshold or saturable biological process). However, they clearly indicate the possibility that high, short-term exposures could pose a cancer hazard unrecognized in a continuous exposure setting.

The implications of these findings in terms of cancer risk assessment are illustrated by observations regarding equivalent averaging times for the stop-exposure studies (Table 5Go). For most chemicals and endpoints, comparable cancer potencies for stop- and continuous-exposure were observed when stop-exposure doses were averaged over less than a lifetime. In a number of cases, the equivalent averaging time approximated the actual exposure duration. This suggests that if cancer potency values derived from lifetime exposures are to be used to estimate cancer risks from short-term exposures, lifetime averaging of the short-term dose may be inappropriate for some carcinogens. Averaging over a lesser period, or in some cases no averaging of the dose at all (i.e., use of the dosing rate as it occurs during the exposure period), may be necessary to avoid underestimating risk for some chemicals.

Comparisons between stop-exposure and continuous-exposure are particularly interesting for 1,3-butadiene as noted previously by Melnick et al. (1990). For example, for heart hemangiosarcomas, exposure to 1,3-butadiene for 13, 26, 40, or 52 weeks produced a much larger tumor response when compared to continuous exposure for a lifetime at the same dosing rate, with a mean equivalent averaging time for the stop-exposure groups of only 34 weeks (Table 2Go). These observations suggest that only a short period of exposure is necessary to account for much or all of 1,3-butadiene carcinogenicity for this endpoint, and in fact for most cancer endpoints for this chemical. For several cancer endpoints for 1,3-butadiene, longer exposure appeared to actually decrease the tumor incidence. This possibility has been anticipated by some carcinogenesis models in which premalignant cells may be more vulnerable to toxicity from the chemical during exposures in the later stages of carcinogenesis (Portier, 1987Go). In this situation, after an initial exposure event, continued exposure to the chemical would reduce, rather than increase, tumor response. Some epidemiology studies of cancer mortality among workers exposed to 1,3-butadiene have found greater incidence among those with shorter, rather than longer exposures (e.g., Divine et al., 1993), although this has not been observed consistently (e.g., Matanoski et al., 1990) and it is possible that the shorter exposures may have involved higher concentrations (Landrigan, 1993Go). If this phenomenon truly exists, even for only a limited number of compounds, it will be a difficult but important challenge to develop tools to effectively address this in the risk assessment process.

In some cases, tumor incidence appeared to be decreased from high doses over a shorter period of time. For example, for mononuclear cell leukemia in rats exposed to o-nitroanisole at 6000 or 18,000 ppm for 27 weeks, tumor incidence was 16.8% and 0%, respectively, while rats exposed continuously for a lifetime to 666 ppm and 2000 ppm had a tumor incidence of 82.3% and 73.3%, respectively (Table 2Go). In this example, much of this decrease may be explained by the very low survival in the high-dose, stop-exposure groups.

In general, it is unclear why tumorigenicity from some chemicals appears to be a linear function of exposure duration while that from other chemicals does not. Presumably, the basis for this difference lies in their mechanisms of carcinogenicity, but there are few clues offered by the limited set of chemicals available for this analysis. The carcinogenicity of most of these 11 chemicals has received little study beyond simple bioassays, and their mechanisms of carcinogenicity are largely unknown (the clear exception being 1,3-butadiene). Boutwell (1964), using an initiator-promoter regimen to study mouse skin carcinogenesis, found tumor response to be a function of cumulative dose for the initiator (9,10-dimethyl-1,2-benzanthracene, DMBA) but not for the promoter (croton oil). If these observations could be generalized to exposure duration, then a linear relationship with tumor response would be most likely for chemicals that are genotoxic. Genotoxicity studies conducted by the NTP for these chemicals are summarized in Table 6Go. Among the chemicals that showed the best agreement between responses from continuous- and stop-exposure (1-amino-2,4-dibromoanthraquinone, coumarin, 3,4-dihydrocoumarin, oxazepam, and salicylazosulfapyridine; see Tables 3 and 4GoGo), only 1-amino-2,4-dibromoanthraquinone and coumarin shown strong genotoxic potential. On the other hand, the only commonality among the six chemicals for which short-term and lifetime exposures gave greater than 2-fold differences in cancer potency estimates for at least one endpoint (viz., 1-amino-2,4-dibromoanthraquinone, 2,2-bis(bromomethyl)-1,3-propanediol, 1,3-butadiene, coumarin, o-nitroanisole, and methyleugenol; see Table 4Go) is that all except methyleugenol were positive in at least one Salmonella mutagenicity assay and all except 1,3-butadiene, which was not tested, and 2,2-bis(bromomethyl)-1,3-propanediol were positive in the sister chromatid exchange assay in Chinese hamster ovary cells. From these observations, no obvious relationship can be deduced between genotoxicity and the influence of exposure duration on tumor response. The relationship between tumorigenicity and cumulative carcinogen dose does not appear to be a function of any particular target organ, as there are examples of both types of responses in major target organs such as kidney, liver, and intestine. Also, there are no clear correlates with chemical structure.


View this table:
[in this window]
[in a new window]
 
TABLE 6 NTP Genetic Toxicology Summary
 
In conclusion, this analysis suggests that limited duration exposures do not produce proportional decreases in cancer risk for some carcinogens. Consequently, the use of this assumption in developing cancer risk estimates for short-term exposures, such as those that might occur in occupational environments, may result in error for many chemicals. Theoretically, the use of cancer potency estimates derived from continuous exposure to estimate short-term-exposure cancer risks could result in either an under- or overestimation. Empirical data from the limited number of chemicals included in this analysis suggests an underestimation of risk appears to be more likely. Understanding the relationship between dosing rate, exposure duration, and tumor incidence is vital to producing meaningful cancer risk estimates. A major obstacle in developing this understanding is the availability of relevant data—the number of chemicals for which stop-exposure data are available are few and most have been poorly studied as to mechanism of action. There is a need for more stop-exposure studies encompassing a broader array of carcinogens. Also, cancer-risk assessments will have to address the possibility that short-term exposures may not result in proportionally lesser cancer risks and, for some chemicals, might actually pose higher risks than long-term exposure.


    NOTES
 
1 To whom correspondence should be addressed at the Center for Environmental and Human Toxicology, University of Florida, Box 110885, Gainesville, FL 32611. Fax: (352) 392-4707. E-mail: smr{at}ufl.edu. Back


    REFERENCES
 TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
 RESULTS
 DISCUSSION
 REFERENCES
 
Aylward, L. L., Hays, S. M., Karch, N. J., and Paustenbach, D. J. (1996). Relative susceptibility of animals and humans to the cancer hazard posed by 2,3,7,8-tetrachlorodibenzo-p-dioxin using internal measures of dose. Environ. Sci. Tech. 30, 3534–3543.[ISI]

Bickel, P. J., and Doksum, K. A. (1977). Mathematical Statistics: Basic Ideas and Selected Topics. Holden-Day, San Francisco.

Boutwell, R. K. (1964). Some biological aspects of skin carcinogenesis. Prog. Exp. Tumor Res. 4, 207–250.[ISI]

Chen, J. J., Kodell, R. L., and Gaylor, D. W. (1988). Using the biological two-stage model to assess risk from short-term exposures. Risk Anal. 8, 223–230.[ISI][Medline]

Crump, K. S., and Howe, R. B. (1984). The multistage model with a time-dependent dose pattern: Applications to carcinogenic risk assessment. Risk Anal. 4, 163–176.[ISI]

Dankovic, D. A., and Stayner, L. T. (1998). Comment on "Relative susceptibility of animals and humans to the cancer hazard posed by 2,3,7,8-tetrachlorodibenzo-p-dioxin using internal measures of dose." Environ. Sci. Tech. 32, 549–550.[ISI]

Dedrick, R. L., and Morrison, P. F. (1992). Carcinogenic potency of alkylating agents in rodents and humans. Cancer Res. 52, 2464–2467.[Abstract]

Divine, B. J., Wendt, J. K., and Hartman, C. M. (1993). Cancer mortality among workers at a butadiene production facility. IARC Sci. Publ. 127, 345–362.[Medline]

Drew, R. T., Boorman, G. A., Haseman, J. K., McConnell, E. E., Busey, W., and Moore, J. A. (1983). The effect of age and exposure duration on cancer induction by a known carcinogen in rats, mice, and hamsters. Toxicol. Appl. Pharmacol. 68, 120–130.[ISI][Medline]

Haber, F. (1924). Zur Geschichte des Gaskrieges (On the History of Gas Warfare). In Funf Vortrage aus den Jahren 19201923 (Five Lectures from the Years 19201923), pp. 76–92. Springer, Berlin.

Kaldor, J. M., Day, N. E., and Hemminki, K. (1988). Quantifying the carcinogenicity of antineoplastic drugs. Eur. J. Cancer Clin. Oncol. 24, 703–711.[ISI][Medline]

Kodell, R. L., Gaylor, D. W., and Chen, J. J. (1987). Using average lifetime dose rate for intermittent exposures to carcinogens. Risk Anal. 7, 339–345.[ISI][Medline]

Landrigan, P. J. (1993). Critical assessment of epidemiologic studies on the human carcinogenicity of 1,3-butadiene and styrene. IARC Sci. Publ. 127, 375–388.[Medline]

Matanoski, G., Santos-Burgos, C., and Swartz, L. (1990). Mortality of a cohort of workers in the styrene-butadiene polymer manufacturing industry. Environ. Health Perspect. 86, 107–117.[ISI][Medline]

McConnell, E. E. (1992). Comparative response in carcinogenesis bioassay as a function of age at first exposure. In Similarities and Differences Between Children and Adults: Implications for Risk Assessment (P. S. Guzelian, C. J. Henry, and S. S. Olin, Eds.), pp. 66–78. ILSI Press, Washington, DC.

Melnick, R. L, Huff, J., Chou, B. J., and Miller, R. A. (1990) Carcinogenicity of 1,3-butadiene in C57BL/6 x C3H F1 mice at low exposure concentrations. Cancer Res. 50, 6592–6599.[Abstract]

Murdoch, D. J., Krewski, D., and Wargo, J. (1992). Cancer risk assessment with intermittent exposure. Risk Anal. 12, 569–577.[ISI][Medline]

National Toxicology Program (NTP) (1993a). Toxicology and Carcinogenesis Studies of Furan (CAS No. 110009) in F344 Rats and B6C3F1 Mice (Gavage Studies). NTP TR-402.

National Toxicology Program (NTP). (1993b). Toxicology and Carcinogenesis studies of o-Nitroanisole (CAS No. 91236) in F344 Rats and B6C3F1 Mice (Feed Studies). NTP TR-416.

National Toxicology Program (NTP). (1993c). Toxicology and Carcinogenesis Studies of Coumarin (CAS No. 91645) in F344/N Rats and B6C3F1 Mice (Gavage Studies). NTP TR-422.

National Toxicology Program (NTP). (1993d). Toxicology and Carcinogenesis Studies of 3,4-Dihydrocoumarin (CAS No. 119846) in F344/N Rats and B6C3F1 Mice (Gavage Studies). NTP TR-423.

National Toxicology Program (NTP). (1993e). Toxicology and Carcinogenesis Studies of 1,3-Butadiene (CAS No. 106990) in B6C3F1 Mice (Inhalation Studies). NTP TR-434.

National Toxicology Program (NTP). (1994) Toxicology and Carcinogenesis Studies of Hexachlorocyclopentadiene (CAS No. 77474) in F344/N Rats and B6C3F1 Mice (Inhalation Studies). NTP TR-437.

National Toxicology Program (NTP). (1996a). Toxicology and Carcinogenesis Studies of 1-amino-2,4-dibromoanthraquinone (CAS No. 81492) in F344/N Rats and B6C3F1 Mice (Feed Studies). NTP TR-383.

National Toxicology Program (NTP). (1996b). Toxicology and Carcinogenesis Studies of 2,2-bis(bromomethyl)-1,3-propanediol (FR-1138®) (CAS No. 3296900) in F344 Rats and B6C3F1 Mice (Feed Studies). NTP TR-452.

National Toxicology Program (NTP). (1996c). Toxicology and Carcinogenesis Studies of Oxazepam (CAS No. 604751) in F344/N Rats (Feed Studies). NTP TR-468.

National Toxicology Program (NTP). (1997) Toxicology and Carcinogenesis Studies of Salicylazosulfapyridine (CAS No. 599791) in F344/N Rats and B6C3F1 Mice (Gavage Studies). NTP TR-457.

National Toxicology Program (NTP). (1999) Toxicology and Carcinogenesis Studies of Pentachlorophenol (CAS No. 87865) in F344/N Rats (Feed Studies). NTP TR-483.

National Toxicology Program (NTP). (in press) Toxicology and Carcinogenesis Studies of Methyleugenol (CAS No. 93152) in F344/N Rats and B6C3F1 Mice (Gavage Studies). NTP TR-491.

Portier, C. J. (1987). Statistical properties of a two-stage model of carcinogenesis. Environ. Health Perspect. 76, 125–131.[ISI][Medline]

Portier, C., and Bailer, A. J. (1989). Testing for increased carcinogenicity using a survival-adjusted quantal response test. Fundam. Appl. Toxicol. 12, 731–737.[ISI][Medline]

U.S. EPA (United States Environmental Protection Agency). (1996a). Comparison of the Effects of Chemicals with Combined Perinatal and Adult Exposures Versus Adult-Only Exposure in Carcinogenesis Bioassays, U.S. EPA Office of Pesticide Programs, Docket OPP-00449, Science Advisory Panel Meeting, October 29–30, 1996, Arlington, VA.

U.S. EPA (United States Environmental Protection Agency). (1996b). Proposed Guidelines for Carcinogen Risk Assessment. EPA 600/P-92/003C, Washington DC.

U.S. EPA (United States Environmental Protection Agency). (1998). Summary of the U.S. EPA Workshop on the Relationship Between Exposure Duration and Toxicity. Prepared by Eastern Research Group, Contract No. 68-D5–0028, Work Assignment No. 98–10, September 30, 1998.