- Email: [email protected]
Calculation of Benchmark Doses for Reproductive and Developmental Toxicity Observed after Exposure to Isopropanol
REGULATORY TOXICOLOGY AND PHARMACOLOGY ARTICLE NO.
28, 38 – 44 (1998)
RT981226
Calculation of Benchmark Doses for Reproductive and Developmental Toxicity Observed after Exposure to Isopropanol Bruce Allen, Robinan Gentry, Annette Shipp, and Cynthia Van Landingham1 The K. S. Crump Group, Inc., ICF Kaiser, 602 East Georgia, Ruston, Louisiana 71270 Received February 13, 1998
centrations (RfCs) using the selected NOAEL and application of uncertainty factors (USEPA, 1992a). The use of an experimentally derived NOAEL in setting standards for human exposure to environmental toxicants has been criticized by the scientific community (Crump, 1984; Kimmel and Gaylor, 1988) and by the USEPA’s Science Advisory Board (USEPA, 1986, 1988a,b). Limitations of the NOAEL approach are summarized in recent USEPA guidance (USEPA, 1995) as follows:
Reproductive, including developmental, toxicity risk assessment has typically relied on estimation of toxicity criteria values derived from no-observed-adverseeffect levels (NOAELs). The benchmark dose (BMD) approach has been proposed as an alternative that avoids problems with NOAELs. In this analysis of the reproductive and developmental toxicity observed in a multigeneration study of rats exposed to isopropanol, the BMD approach has been applied to all effects exhibiting significant dose–response relationships. The BMD estimates were very consistent across models and across end points; they were within the range of doses (100 to 500 mg/kg/day) that has been suggested as being the NOAEL. The use of the BMD approach for analysis of isopropanol reproductive toxicity is shown to avoid the experiment-specific argument of whether a particular treatment has induced statistically significant differences, compared to controls, in favor of the estimation of experiment-independent doses corresponding to risk levels of interest. The consistency of the BMD estimates, with values of about 420 mg/kg/ day, suggests that, for isopropanol, the available multigeneration study data may provide a suitable basis for considering safe exposure. © 1998 Academic Press
● Whether or not a given experimental dose actually constitutes a NOAEL is subject to scientific judgment and is often a source of controversy; ● Larger NOAELs can result from experiments involving fewer animals, that is, a poorly designed study may be ‘‘rewarded’’; ● The shape and slope of the dose response is not considered in the determination of the NOAEL; ● The NOAEL (if one exists) must be one of the experimental doses; and ● Use of a NOAEL does not provide estimates of potential risk at any exposure level.
A quantitative approach, the benchmark dose (BMD) approach, has been proposed as an alternative to the NOAEL approach for determining a toxicity value that can be used in setting exposure limits (Crump, 1984, 1995; Gaylor, 1989; USEPA, 1990, 1993). As defined in those proposals, a BMD is a statistical lower confidence bound on the dose estimated to result in a specified increase in risk or change in response; the specified response level is termed the benchmark risk (BMR) level, and doses corresponding to that level are calculated using dose–response modeling. The BMD approach has several advantages over the NOAEL approach:
INTRODUCTION
Assessment of the potential for a chemical to cause adverse noncancer health effects in populations that may be exposed to that chemical frequently involves derivation of a toxicity criteria value to which the amount of chemical exposure for a selected use pattern is compared. Toxicity values usually are derived from animal data and traditionally have been based on doses determined to be no-observed-adverse-effect levels (NOAELs). A NOAEL, in turn, is defined as the highest experimental dose at which no biologically or statistically significant changes in selected toxicity responses or end points are observed. For example, the United States Environmental Protection Agency (USEPA) has frequently calculated toxicity values termed reference doses (RfDs) and reference con1
● The BMD approach, unlike the NOAEL, takes into account the dose–response information (i.e., the shape of the dose–response curve); ● The BMD approach does not involve sometimes argumentative ‘‘all or nothing’’ decisions, such as determining whether or not a NOAEL was defined at a particular dose; ● The value of a lower confidence limit appropriately reflects the sample size of a study (smaller studies tend
To whom correspondence should be addressed.
0273-2300/98 $25.00 Copyright © 1998 by Academic Press All rights of reproduction in any form reserved.
38
39
REPRODUCTIVE BENCHMARK DOSES FOR ISOPROPANOL
to result in wider confidence limits and lower BMDs, whereas the opposite is true for NOAELs); and ● A benchmark dose can be determined even when a NOAEL has not been identified in a study. The BMD approach has been recommended as the basis for toxicity value derivation, as an approach without some of the limitations inherent in the NOAEL approach, and, therefore, as a means to improve noncancer risk assessment (Barnes et al., 1995). Indeed, several RfDs and RfCs listed on USEPA’s Integrated Risk Information System have been developed using BMD methodology, including those for methylmercury, carbon disulfide, and 1,1,1,2-tetrafluoroethane. BMD approaches have been most widely applied to, and studied in the context of, developmental toxicity end points (Allen et al., 1994a,b, 1996; Faustman et al., 1994; Kavlock et al., 1995). Dose–response modeling approaches have also been applied to other reproductive effects (Pease et al., 1991). This report presents the results of dose–response modeling and BMD estimation for both developmental and other reproductive effects observed in a multigeneration study of rats exposed to isopropanol (Exxon Biomedical Sciences, Inc., 1992; Bevan et al., 1995). Previous investigations of isopropanol, a widely used industrial solvent also found in many consumer and industrial products, have provided limited information about the potential for reproductive, including developmental toxicity (Lehman et al., 1945; FDRL, 1975; Gallo et al., 1977; BIBRA, 1988). The multigeneration study (Exxon Biomedical Sciences, Inc., 1992; Bevan et al., 1995) analyzed here included four dose groups (0, 100, 500, and 1000 mg/kg daily by oral gavage), with 30 rats of each sex initially exposed (P1). Exposure lasted at least 10 weeks prior to mating, which occurred within groups for up to 3 weeks. Exposure continued through lactation for the females and until the delivery of the last litter for the males. The P2 adults were selected from the F1 litters and were dosed for 10 –13 weeks before they were mated. Only a single F2 litter was obtained from each P2 mating. Reproductive effects, including developmental toxicity, were determined for P1, P2, F1, and F2 animals. This study provides a good basis for assessing risks of reproductive toxicity associated with isopropanol exposure.
after correcting for correlated observations (fetuses within litters), reported that significant differences in pup survival were also present at day 7 (in the F2 generation only). However, because the primary effect was observed at or before pnd 4, and because of culling that occurred on pnd 4, the 7-day survival data were not analyzed here. The BMD approach involves application of mathematical dose–response models using appropriate statistical procedures (Crump, 1984). The dose–response models use likelihood-based methods to determine best estimates and confidence limits on doses corresponding to specified response levels (benchmark response levels or BMRs). The maximum-likelihood estimates of dose estimated to yield the BMR are referred to as BMDs and the 95% lower bounds on the BMDs are referred to as BMDLs. The BMRs are expressed in terms of the probability of a response for quantal (‘‘yes/no’’ or count) data; for continuous end points the BMRs are expressed in terms of the change in a mean response. For the one effect with uncorrelated quantal response data, male mating index,2 probability of response as a function of dose was modeled using both the Weibull model, which has the form q
P~d ! 5 1 2 e ~2q02q13d !,
(1)
2
and the polynomial regression model, with form P~d ! 5 1 2 e ~2q02q13d2· · ·2q k3d !, k
(2)
where P(d) is the probability of response at dose d, and the qi parameters are estimated by methods of maximum likelihood. These models were implemented using the THRESHW and THRESH software programs of ICF Kaiser (Howe, 1984a,b), respectively. Using these models, doses corresponding to extra and additional risks of defined magnitudes (i.e., equal to the BMR) are calculated. The extra risk at a dose d is defined as P~d ! 2 P~0! , 1 2 P~0!
(3)
whereas additional risk is defined as METHODS
The end points selected for inclusion in this dose– response, BMD analysis were those for which there were statistically significant dose-related changes. The only significant nondevelopmental effect reported (Bevan et al., 1995) was a significant decrease in male mating index in P2 male rats. The significant developmental effects noted were decreased offspring survival on both postnatal day (pnd) 1 and pnd 4 for both the F1 and F2 generations (Bevan et al., 1995). Tyl (1996),
P~d ! 2 P~0!.
(4)
Allen et al., (1994a) suggested that an additional risk of 10% (or possibly greater) be used as the BMR when a generic quantal end point, as represented by the male mating index, is analyzed. The BMDs and BMDLs for 2 What was actually modeled was the proportion of males failing to mate, so that increasing proportions with increases in exposure level are indicative of an adverse reproductive effect.
40
ALLEN ET AL.
male mating index reported below are based on this suggestion. Because of the potential for correlations among the developmental toxicity effects (i.e., observations within a litter are likely to be more closely related to one another than they are to observations from other litters), it is appropriate either to treat the litter as the unit of observation or to implement a procedure that accounts for the intralitter correlation. Both approaches were used in this analysis. When the litter was used as the unit of observation, the summary of response for each litter was proportion of offspring dying. The dose–response models used to represent changes in those proportions treated them as continuous variables. Two models were considered. The polynomial mean response model is F~d ! 5 q 0 2 ~q 1 3 d 1 · · · 1 q k 3 d k!.
(5)
The power mean response model is q
F~d ! 5 q 0 2 q 1d 2.
(6)
For both models, F(d) is the mean response at dose d. Variability around the means is assumed to be normal, with group-specific variances. The parameters, the qi’s which are estimated by methods of maximum likelihood, are constrained to be nonnegative. These models were implemented using the THC and THWC programs of ICF Kaiser (Howe, 1984c,d), respectively. When using models that predict the mean response as a function of dose, BMRs are defined in terms of changes in the means. Since this approach was only used when the continuous measure for each experimental unit was a proportion, i.e., for the postnatal survival end points where each litter had some proportion of its fetuses dying in a specified time interval, the BMR was set so that F~d ! 2 F~0! 5 0.05.
(7)
This choice was based on the results of Allen et al., (1994a), who found, for a large data base of similar developmental toxicity end points, that the 5% BMR yielded BMDLs matching, on average, the end-pointspecific NOAELs. This finding is consistent with the USEPA recommendation that the BMR level should fall somewhere in the 1 to 10% range (USEPA, 1995). When the offspring from each litter were considered the experimental units, the intralitter correlation was handled by considering beta-binomial models. A betabinomial assumption entails that each of the offspring within any given litter respond independently from one another according to a binomial distribution with some underlying probability of response, but that probability of response is allowed to vary from litter to litter, within a dose group (and perhaps a litter size) accord-
ing to a beta distribution. Such a formulation induces correlations among littermates. For each dose group, the mean probability of response (the mean of the beta distribution) and the intralitter correlation coefficient are estimated. Based on the findings of Allen et al. (1994b) the model used to represent the effect of dose level is a log-logistic type model given by P~d,s! 5 a 1 ~u1s! 1
1 2 a 2 ~u1s! , 1 1 exp~b 1 ~u2s! 2 g log~d 2 d0!! 0 # a 1 u1s # 1, (8)
where d is dose and s is litter size. The parameters (a, b, u1, u2, and g) are estimated by methods of maximum likelihood. This model was fit to the data using the software package TERALOG from ICF Kaiser (Howe et al., 1990). As shown, in addition to accounting for intralitter correlation, this model allows litter size to influence the probability of response. When the beta-binomial modeling approach was used, the model predictions are in terms of probabilities of response, and so BMRs could be defined in terms of extra or additional risk, as shown in Eqs. (3) and (4). A BMR of 5% additional risk was selected, in keeping with the findings of Allen et al. (1994b), who determined that such a definition of the BMR yielded a reasonable match, on average, between the resulting BMDLs and corresponding NOAELs. For the determination of the BMDs and BMDLs, the average litter size from the control group was used. RESULTS
The only reproductive effect observed to be significantly related to dose was male mating index, with the only significant effect observed at 1000 mg/kg/day. Model fits for this end point were satisfactory (P values for goodness-of-fit were equal to 0.55 and 0.81 for the Weibull and polynomial models, respectively; see Fig. 1). The only discrepancies between model predictions and observations occurred when the observed proportions were not monotonically increasing. The polynomial and Weibull models applied to this data set yielded similar BMD estimates; the BMDL estimates from the two models were almost identical (Table 1). This BMDL corresponds to a BMR rate of 10%, a rate found to provide a relatively close match to NOAELs, on average, for a large set of quantal end points from developmental toxicity studies (Allen et al., 1994a). In this case, a BMDL between 416 and 418 mg/kg/day would be only slightly less than the NOAEL of 500 mg/kg/day. Postnatal mortality, expressed in terms of average percentage of offspring dead by pnd 1 or pnd 4, increased as a function of dose (Fig. 2). The power and polynomial models used to model those percentage changes adequately fit the data (P values for goodness-
REPRODUCTIVE BENCHMARK DOSES FOR ISOPROPANOL
41
FIG. 1. Observed and predicted percentage of males failing to mate. Model predictions (solid line for the polynomial model and dashed line for the Weibull model) are very similar to one another and both provide an adequate representation of the dose–response pattern for the observed (F) percentages.
of-fit exceeding 0.42 in all cases) and yielded BMDs and BMDLs as shown in Table 2. There is relatively little model dependence for BMD and BMDL estimates (model differences exist only for the F1 generation results). The BMD and BMDL values for mortality through pnd 4 are less than those for mortality through pnd 1, for both the F1 and F2 animals. Although the approach based on percentage mortality per litter has been used in previous studies (Allen et al., 1994a), it does not allow for the incorporation or estimation of litter size effects. The size of the litter can be an important factor in the postnatal survival rate. Moreover, because some of the observed mean percentages are close to 0% (Fig. 2), the assumption of normality of variation around those means is not supportable. Therefore, the alternative, recommended approach for the analysis of the postnatal mortality data is the beta-binomial approach. Not only does it provide a
TABLE 1 BMDs and BMDLs for the Reproductive Effect, Male Mating Index in the P2 Generation BMD (mg/kg/day) Model P2 male mating index
BMDL (mg/kg/day)
Polynomial Weibull Polynomial Weibull 754
794
416
418
better representation of data variability (including intralitter correlations), but it also accounts for litter size. The model fits to the data were satisfactory (goodness-of-fit P values ranging between 0.24 and 0.31; Fig. 3). The BMDs and BMDLs that were obtained using the alternative approach are shown in Table 3. Despite the theoretical advantages of this approach, the BMDs and BMDLs estimated here are generally consistent with those derived using the power and polynomial models (Table 2). The primary differences appear to be in the best estimates of the doses corresponding to a 5% increase in probability of postnatal death for F2 offspring at pnd 1 or pnd 4 and in the lower bounds on dose corresponding to 5% increased probability of death at pnd 4 in the F1 animals. DISCUSSION AND CONCLUSION
The most relevant BMDLs for the reproductive and developmental effects analyzed here were either 416 mg/kg/day (for male mating index) or 418 mg/kg/day (based on the preferred analysis of the F2 generation 4-day survival rate). The BMDLs are remarkably similar for the two types of effects significantly affected by isopropanol exposure. This is true even though the response rates that define the BMDs (the BMRs) for the two types of end point differ. For the male mating index (coded as a quantal response) the BMR is 10%,
42
ALLEN ET AL.
FIG. 2. Observed and predicted dose–response patterns for average offspring mortality. The polynomial mean (solid line) and power mean (dashed line) model predictions of the average percentage mortality correspond well to the observed means (F, shown 6 one standard deviation) for the F1 generation, through pnd 1.
whereas for pup survival (a nested quantal end point) the BMR is 5%. Although this appears paradoxical, it is consistent with the findings of Allen et al. (1994a,b) in which these two choices provided the closest match, on average, to corresponding NOAELs. There may be several reasons why different response rates corresponded to NOAELs depending on end point type, but one salient reason has to do with the ability to determine whether or not two groups are statistically significantly different. With the quantal end points, the tests employed tend to require at least a 10% increase in incidence before significant differences are detected. Group differences in continuous or nested quantal end points (the latter of which having much greater sample sizes since accounting for correlation allows one to TABLE 2 BMDLs for the Developmental Effects Using Litters as the Unit of Observation (Percentage Survival) Power model
Data set F1, F1, F2, F2,
1 4 1 4
day day day day
survival survival survival survival
Polynomial model
BMD BMDL BMD BMDL (mg/kg/day) (mg/kg/day) (mg/kg/day) (mg/kg/day) 807 588 1000 657
663 322 588 415
765 539 1000 657
642 343 588 415
consider the fetus or pup as the unit of observation) tend to be significant with less change from control. In this way, the end-point-specific differences in the choice of BMRs in this analysis are a by-product of common practice and accepted toxicological opinion with respect to what is meaningful and relevant to human health risk assessment, to the extent that NOAELs are considered to be a satisfactory starting point for such assessments. The BMDL values of interest lie between the study dosages of 100 and 500 mg/kg/day, for which there are conflicting no-effect interpretations. The USEPA (1992b) and Tyl (1996) interpreted the reductions in postnatal survival as treatment related and concluded that 100 mg/kg/day was the study NOAEL. Alternatively, Bevan et al. (1995) and Harris (1995) deemed the observations not to be biologically significant and concluded that the NOAEL was 500 mg/kg/day. The BMD approach provides a means to clarify the study findings and a means by which such disagreements can be circumvented. This is because it provides estimates of an experiment-independent quantity, the dose corresponding to a fixed level of response, rather than attempting to address the experiment-specific question of whether or not a treatment has caused a significantly different amount of response. It is also important to note that comparisons among NOAELs and BMDLs, such as those that formed the
REPRODUCTIVE BENCHMARK DOSES FOR ISOPROPANOL
43
FIG. 3. Observed and predicted dose–response patterns for offspring mortality. Unlike Fig. 2, this figure shows percentage mortality for each litter in each dose group (F; the number of overlapping observations is not presented, however). The log-logistic model (solid line) predicts the average value of those individual observations, adjusting for litter size and accounting for intralitter correlations among the observations for the F2 generation, through pnd 4.
basis for selecting BMR levels of 5 or 10% additional risk (Allen et al., 1994a,b), already account for uncertainties about the true shape of the dose–response curve. Such uncertainties affect estimates of the value of dose determined to correspond to any given response level. Thus, these uncertainties are reflected in the lower bounds that define the BMDLs. Moreover, by considering at least two model forms (e.g., Weibull and polynomial) a measure of model dependence can be ascertained. In the case of isopropanol, little model dependence existed. The BMR levels appropriate for this analysis were within the range of responses observed in the treatment groups, so uncertainty and model dependence were reduced. The consistency of the results obtained here for isopropanol reproductive/ developmental toxicity suggests that the set of experiTABLE 3 Recommended Approach for BMD Estimation of Developmental Effects Using a Beta-Binomial Model Data set F1, F1, F2, F2,
1 4 1 4
day day day day
survival survival survival survival
BMD (mg/kg/day)
BMDL (mg/kg/day)
911 656 1505 804
670 449 660 418
ments and accompanying BMD analyses form an appropriate basis for considering safe exposures to isopropanol. REFERENCES Allen, B., Kavlock, R., Kimmel, C., and Faustman, E. (1994a). Dose– response assessment for developmental toxicity. II. Comparison of generic benchmark dose estimates with no observed adverse effect levels. Fundam. Appl. Toxicol. 23, 487– 495. Allen, B., Kavlock, R., Kimmel, C., and Faustman, E. (1994b). Dose– response assessment for developmental toxicity. III. Statistical models. Fundam. Appl. Toxicol. 23, 496 –509. Allen, B. C., Strong, P. L., Price, C. J., Hubbard, S. A., and Daston, G. P. (1996). Benchmark dose analysis of developmental toxicity in rats exposed to boric acid. Fundam. Appl. Toxicol. 32, 194 –204. Bevan, C., Tyler, T., Gardiner, T., Kapp, R., Andrews, L., and Beyer, B. (1995). Two-generation reproduction toxicity study with isopropanol in rats. J. Appl. Toxicol. 15, 117–223. BIBRA (1988). A Single Generation Reproduction and Embryotoxicity Study with Isopropyl Alcohol in Rats. The British Industrial Biological Research Association, Carshalton, Surrey. Crump, K. (1984). A new method for determining allowable daily intakes. Fundam. Appl. Toxicol. 4, 854 – 871. Crump, K. (1995). Calculation of benchmark doses from continuous data. Risk Anal. 15, 79 – 89. Exxon Biomedical Sciences, Inc. (1992). Multi-Generation Rat Reproduction Study with Isopropanol. Project No. 259835. Prepared for Chemical Manufacturers Association, April 17, 1992.
44
ALLEN ET AL.
FDRL (1975). Toxicity Studies in Rats with 2-Butanol including Growth, Reproduction, and Teratologic Observations. Food and Drug Research Laboratories, East Orange, NJ. Gallo, M., Oser, B., Cox, G., and Bailey, D. (1977). Studies on the long-term toxicity of 2-butanol. Toxicol. Appl. Pharmacol. 41, 135. Gaylor, D. (1989). Quantitative risk analysis for quantal reproductive and developmental effects. Environ. Health Perspect. 79, 243– 246. Gaylor, D., and Slikker, W., Jr. (1990). Risk assessment for neurotoxic effects. Neurotoxicology 11, 211–218. Harris, S. B. (1995). A review of the EPA comments regarding the study entitled ‘‘Multi-generation rat reproduction study with isopropanol.’’ Report prepared for the Chemical Manufacturers Association Isopropanol Panel. Howe, R. (1984a). THRESH—A computer program to compute a reference dose from quantal animal toxicity data using the benchmark dose method. Howe, R. (1984b). THRESHW—A computer program to compute a reference dose from quantal animal toxicity data using the benchmark dose method. Howe, R. (1984c). THC—A computer program to compute a reference dose from continuous animal toxicity data using the benchmark dose method. Howe, R. (1984d). THWC—A computer program to compute a reference dose from continuous animal toxicity data using the benchmark dose method. Howe, R., and Van Landingham, C. (1990). TERALOG—A computer program for assessing risks of reproductive/developmental toxicity. Kavlock, R., Allen, B., Faustman, E., and Kimmel, C. (1994). Dose– response assessment for developmental toxicity. IV. Benchmark doses for fetal weight changes. Fundam. Appl. Toxicol. 26, 211– 222. Kimmel, C., and Gaylor, D. (1988). Issues in qualitative and quantitative risk analysis for developmental toxicology. Risk Anal. 8, 15–21. Lehman, A., Schwerma, H., and Richards, E. (1945). Acquired tolerance in dogs, rate of disappearance from the blood stream in various species, and effects on successive generations of rats. J. Pharmacol. Exp. Ther. 85, 61– 69.
Pease, W., Vandenburg, J., and Hooper, K. (1991). Comparing alternative approaches to establishing regulatory levels for reproductive toxicants: DBCP as a case study. Environ. Health Perspect. 91, 141–155. Tyl, R. W. (1996). Personal correspondence to Kathryn Rosica, Chemical Manufacturers Association, Arlington, VA, from Rochelle Tyl, Center for Life Sciences and Toxicology, Research Triangle Institute, Research Triangle Park, NC. February 12, 1996. U.S. Environmental Protection Agency (USEPA) (1986). Guidelines for the health assessment of suspect developmental toxicants. Fed. Reg. 50, 39426 –39436. U.S. Environmental Protection Agency (USEPA) (1988a). Proposed guidelines for assessing male reproductive risk. Fed. Reg. 53, 24850 –24969. U.S. Environmental Protection Agency (USEPA) (1988b). Proposed guidelines for assessing female reproductive risk. Fed. Reg. 53, 24834 –24847. U.S. Environmental Protection Agency (USEPA) (1990). Interim Methods for Development of Inhalation Reference Concentrations. EPA/600 8-90/066A. Review draft. Office of Research and Development, Washington, DC. August 1990. U.S. Environmental Protection Agency (USEPA) (1992a). IRIS. Background document (4/1/91). Office of Health and Environmental Assessment, Environmental Criteria and Assessment Office, Cincinnati, OH. U.S. Environmental Protection Agency (USEPA) (1992b). Memorandum to Keith Cronin, Chemical Testing Branch, from Jennifer Seed, Toxic Effects Section. July 1, 1992. U.S. Environmental Protection Agency (USEPA) (1993). Background and Quantitative Procedures for Derivation of Inhalation Reference Concentrations. Environmental Criteria and Assessment Office, U.S. Environmental Protection Agency, Research Triangle Park, NC. June 1993. U.S. Environmental Protection Agency (USEPA) (1995). The Use of the Benchmark Dose (BMD) Approach in Health Risk Assessment. Final report. EPA/630/R-94/007. Risk Assessment Forum, U.S. Environmental Protection Agency, Washington, DC.
28, 38 – 44 (1998)
RT981226
Calculation of Benchmark Doses for Reproductive and Developmental Toxicity Observed after Exposure to Isopropanol Bruce Allen, Robinan Gentry, Annette Shipp, and Cynthia Van Landingham1 The K. S. Crump Group, Inc., ICF Kaiser, 602 East Georgia, Ruston, Louisiana 71270 Received February 13, 1998
centrations (RfCs) using the selected NOAEL and application of uncertainty factors (USEPA, 1992a). The use of an experimentally derived NOAEL in setting standards for human exposure to environmental toxicants has been criticized by the scientific community (Crump, 1984; Kimmel and Gaylor, 1988) and by the USEPA’s Science Advisory Board (USEPA, 1986, 1988a,b). Limitations of the NOAEL approach are summarized in recent USEPA guidance (USEPA, 1995) as follows:
Reproductive, including developmental, toxicity risk assessment has typically relied on estimation of toxicity criteria values derived from no-observed-adverseeffect levels (NOAELs). The benchmark dose (BMD) approach has been proposed as an alternative that avoids problems with NOAELs. In this analysis of the reproductive and developmental toxicity observed in a multigeneration study of rats exposed to isopropanol, the BMD approach has been applied to all effects exhibiting significant dose–response relationships. The BMD estimates were very consistent across models and across end points; they were within the range of doses (100 to 500 mg/kg/day) that has been suggested as being the NOAEL. The use of the BMD approach for analysis of isopropanol reproductive toxicity is shown to avoid the experiment-specific argument of whether a particular treatment has induced statistically significant differences, compared to controls, in favor of the estimation of experiment-independent doses corresponding to risk levels of interest. The consistency of the BMD estimates, with values of about 420 mg/kg/ day, suggests that, for isopropanol, the available multigeneration study data may provide a suitable basis for considering safe exposure. © 1998 Academic Press
● Whether or not a given experimental dose actually constitutes a NOAEL is subject to scientific judgment and is often a source of controversy; ● Larger NOAELs can result from experiments involving fewer animals, that is, a poorly designed study may be ‘‘rewarded’’; ● The shape and slope of the dose response is not considered in the determination of the NOAEL; ● The NOAEL (if one exists) must be one of the experimental doses; and ● Use of a NOAEL does not provide estimates of potential risk at any exposure level.
A quantitative approach, the benchmark dose (BMD) approach, has been proposed as an alternative to the NOAEL approach for determining a toxicity value that can be used in setting exposure limits (Crump, 1984, 1995; Gaylor, 1989; USEPA, 1990, 1993). As defined in those proposals, a BMD is a statistical lower confidence bound on the dose estimated to result in a specified increase in risk or change in response; the specified response level is termed the benchmark risk (BMR) level, and doses corresponding to that level are calculated using dose–response modeling. The BMD approach has several advantages over the NOAEL approach:
INTRODUCTION
Assessment of the potential for a chemical to cause adverse noncancer health effects in populations that may be exposed to that chemical frequently involves derivation of a toxicity criteria value to which the amount of chemical exposure for a selected use pattern is compared. Toxicity values usually are derived from animal data and traditionally have been based on doses determined to be no-observed-adverse-effect levels (NOAELs). A NOAEL, in turn, is defined as the highest experimental dose at which no biologically or statistically significant changes in selected toxicity responses or end points are observed. For example, the United States Environmental Protection Agency (USEPA) has frequently calculated toxicity values termed reference doses (RfDs) and reference con1
● The BMD approach, unlike the NOAEL, takes into account the dose–response information (i.e., the shape of the dose–response curve); ● The BMD approach does not involve sometimes argumentative ‘‘all or nothing’’ decisions, such as determining whether or not a NOAEL was defined at a particular dose; ● The value of a lower confidence limit appropriately reflects the sample size of a study (smaller studies tend
To whom correspondence should be addressed.
0273-2300/98 $25.00 Copyright © 1998 by Academic Press All rights of reproduction in any form reserved.
38
39
REPRODUCTIVE BENCHMARK DOSES FOR ISOPROPANOL
to result in wider confidence limits and lower BMDs, whereas the opposite is true for NOAELs); and ● A benchmark dose can be determined even when a NOAEL has not been identified in a study. The BMD approach has been recommended as the basis for toxicity value derivation, as an approach without some of the limitations inherent in the NOAEL approach, and, therefore, as a means to improve noncancer risk assessment (Barnes et al., 1995). Indeed, several RfDs and RfCs listed on USEPA’s Integrated Risk Information System have been developed using BMD methodology, including those for methylmercury, carbon disulfide, and 1,1,1,2-tetrafluoroethane. BMD approaches have been most widely applied to, and studied in the context of, developmental toxicity end points (Allen et al., 1994a,b, 1996; Faustman et al., 1994; Kavlock et al., 1995). Dose–response modeling approaches have also been applied to other reproductive effects (Pease et al., 1991). This report presents the results of dose–response modeling and BMD estimation for both developmental and other reproductive effects observed in a multigeneration study of rats exposed to isopropanol (Exxon Biomedical Sciences, Inc., 1992; Bevan et al., 1995). Previous investigations of isopropanol, a widely used industrial solvent also found in many consumer and industrial products, have provided limited information about the potential for reproductive, including developmental toxicity (Lehman et al., 1945; FDRL, 1975; Gallo et al., 1977; BIBRA, 1988). The multigeneration study (Exxon Biomedical Sciences, Inc., 1992; Bevan et al., 1995) analyzed here included four dose groups (0, 100, 500, and 1000 mg/kg daily by oral gavage), with 30 rats of each sex initially exposed (P1). Exposure lasted at least 10 weeks prior to mating, which occurred within groups for up to 3 weeks. Exposure continued through lactation for the females and until the delivery of the last litter for the males. The P2 adults were selected from the F1 litters and were dosed for 10 –13 weeks before they were mated. Only a single F2 litter was obtained from each P2 mating. Reproductive effects, including developmental toxicity, were determined for P1, P2, F1, and F2 animals. This study provides a good basis for assessing risks of reproductive toxicity associated with isopropanol exposure.
after correcting for correlated observations (fetuses within litters), reported that significant differences in pup survival were also present at day 7 (in the F2 generation only). However, because the primary effect was observed at or before pnd 4, and because of culling that occurred on pnd 4, the 7-day survival data were not analyzed here. The BMD approach involves application of mathematical dose–response models using appropriate statistical procedures (Crump, 1984). The dose–response models use likelihood-based methods to determine best estimates and confidence limits on doses corresponding to specified response levels (benchmark response levels or BMRs). The maximum-likelihood estimates of dose estimated to yield the BMR are referred to as BMDs and the 95% lower bounds on the BMDs are referred to as BMDLs. The BMRs are expressed in terms of the probability of a response for quantal (‘‘yes/no’’ or count) data; for continuous end points the BMRs are expressed in terms of the change in a mean response. For the one effect with uncorrelated quantal response data, male mating index,2 probability of response as a function of dose was modeled using both the Weibull model, which has the form q
P~d ! 5 1 2 e ~2q02q13d !,
(1)
2
and the polynomial regression model, with form P~d ! 5 1 2 e ~2q02q13d2· · ·2q k3d !, k
(2)
where P(d) is the probability of response at dose d, and the qi parameters are estimated by methods of maximum likelihood. These models were implemented using the THRESHW and THRESH software programs of ICF Kaiser (Howe, 1984a,b), respectively. Using these models, doses corresponding to extra and additional risks of defined magnitudes (i.e., equal to the BMR) are calculated. The extra risk at a dose d is defined as P~d ! 2 P~0! , 1 2 P~0!
(3)
whereas additional risk is defined as METHODS
The end points selected for inclusion in this dose– response, BMD analysis were those for which there were statistically significant dose-related changes. The only significant nondevelopmental effect reported (Bevan et al., 1995) was a significant decrease in male mating index in P2 male rats. The significant developmental effects noted were decreased offspring survival on both postnatal day (pnd) 1 and pnd 4 for both the F1 and F2 generations (Bevan et al., 1995). Tyl (1996),
P~d ! 2 P~0!.
(4)
Allen et al., (1994a) suggested that an additional risk of 10% (or possibly greater) be used as the BMR when a generic quantal end point, as represented by the male mating index, is analyzed. The BMDs and BMDLs for 2 What was actually modeled was the proportion of males failing to mate, so that increasing proportions with increases in exposure level are indicative of an adverse reproductive effect.
40
ALLEN ET AL.
male mating index reported below are based on this suggestion. Because of the potential for correlations among the developmental toxicity effects (i.e., observations within a litter are likely to be more closely related to one another than they are to observations from other litters), it is appropriate either to treat the litter as the unit of observation or to implement a procedure that accounts for the intralitter correlation. Both approaches were used in this analysis. When the litter was used as the unit of observation, the summary of response for each litter was proportion of offspring dying. The dose–response models used to represent changes in those proportions treated them as continuous variables. Two models were considered. The polynomial mean response model is F~d ! 5 q 0 2 ~q 1 3 d 1 · · · 1 q k 3 d k!.
(5)
The power mean response model is q
F~d ! 5 q 0 2 q 1d 2.
(6)
For both models, F(d) is the mean response at dose d. Variability around the means is assumed to be normal, with group-specific variances. The parameters, the qi’s which are estimated by methods of maximum likelihood, are constrained to be nonnegative. These models were implemented using the THC and THWC programs of ICF Kaiser (Howe, 1984c,d), respectively. When using models that predict the mean response as a function of dose, BMRs are defined in terms of changes in the means. Since this approach was only used when the continuous measure for each experimental unit was a proportion, i.e., for the postnatal survival end points where each litter had some proportion of its fetuses dying in a specified time interval, the BMR was set so that F~d ! 2 F~0! 5 0.05.
(7)
This choice was based on the results of Allen et al., (1994a), who found, for a large data base of similar developmental toxicity end points, that the 5% BMR yielded BMDLs matching, on average, the end-pointspecific NOAELs. This finding is consistent with the USEPA recommendation that the BMR level should fall somewhere in the 1 to 10% range (USEPA, 1995). When the offspring from each litter were considered the experimental units, the intralitter correlation was handled by considering beta-binomial models. A betabinomial assumption entails that each of the offspring within any given litter respond independently from one another according to a binomial distribution with some underlying probability of response, but that probability of response is allowed to vary from litter to litter, within a dose group (and perhaps a litter size) accord-
ing to a beta distribution. Such a formulation induces correlations among littermates. For each dose group, the mean probability of response (the mean of the beta distribution) and the intralitter correlation coefficient are estimated. Based on the findings of Allen et al. (1994b) the model used to represent the effect of dose level is a log-logistic type model given by P~d,s! 5 a 1 ~u1s! 1
1 2 a 2 ~u1s! , 1 1 exp~b 1 ~u2s! 2 g log~d 2 d0!! 0 # a 1 u1s # 1, (8)
where d is dose and s is litter size. The parameters (a, b, u1, u2, and g) are estimated by methods of maximum likelihood. This model was fit to the data using the software package TERALOG from ICF Kaiser (Howe et al., 1990). As shown, in addition to accounting for intralitter correlation, this model allows litter size to influence the probability of response. When the beta-binomial modeling approach was used, the model predictions are in terms of probabilities of response, and so BMRs could be defined in terms of extra or additional risk, as shown in Eqs. (3) and (4). A BMR of 5% additional risk was selected, in keeping with the findings of Allen et al. (1994b), who determined that such a definition of the BMR yielded a reasonable match, on average, between the resulting BMDLs and corresponding NOAELs. For the determination of the BMDs and BMDLs, the average litter size from the control group was used. RESULTS
The only reproductive effect observed to be significantly related to dose was male mating index, with the only significant effect observed at 1000 mg/kg/day. Model fits for this end point were satisfactory (P values for goodness-of-fit were equal to 0.55 and 0.81 for the Weibull and polynomial models, respectively; see Fig. 1). The only discrepancies between model predictions and observations occurred when the observed proportions were not monotonically increasing. The polynomial and Weibull models applied to this data set yielded similar BMD estimates; the BMDL estimates from the two models were almost identical (Table 1). This BMDL corresponds to a BMR rate of 10%, a rate found to provide a relatively close match to NOAELs, on average, for a large set of quantal end points from developmental toxicity studies (Allen et al., 1994a). In this case, a BMDL between 416 and 418 mg/kg/day would be only slightly less than the NOAEL of 500 mg/kg/day. Postnatal mortality, expressed in terms of average percentage of offspring dead by pnd 1 or pnd 4, increased as a function of dose (Fig. 2). The power and polynomial models used to model those percentage changes adequately fit the data (P values for goodness-
REPRODUCTIVE BENCHMARK DOSES FOR ISOPROPANOL
41
FIG. 1. Observed and predicted percentage of males failing to mate. Model predictions (solid line for the polynomial model and dashed line for the Weibull model) are very similar to one another and both provide an adequate representation of the dose–response pattern for the observed (F) percentages.
of-fit exceeding 0.42 in all cases) and yielded BMDs and BMDLs as shown in Table 2. There is relatively little model dependence for BMD and BMDL estimates (model differences exist only for the F1 generation results). The BMD and BMDL values for mortality through pnd 4 are less than those for mortality through pnd 1, for both the F1 and F2 animals. Although the approach based on percentage mortality per litter has been used in previous studies (Allen et al., 1994a), it does not allow for the incorporation or estimation of litter size effects. The size of the litter can be an important factor in the postnatal survival rate. Moreover, because some of the observed mean percentages are close to 0% (Fig. 2), the assumption of normality of variation around those means is not supportable. Therefore, the alternative, recommended approach for the analysis of the postnatal mortality data is the beta-binomial approach. Not only does it provide a
TABLE 1 BMDs and BMDLs for the Reproductive Effect, Male Mating Index in the P2 Generation BMD (mg/kg/day) Model P2 male mating index
BMDL (mg/kg/day)
Polynomial Weibull Polynomial Weibull 754
794
416
418
better representation of data variability (including intralitter correlations), but it also accounts for litter size. The model fits to the data were satisfactory (goodness-of-fit P values ranging between 0.24 and 0.31; Fig. 3). The BMDs and BMDLs that were obtained using the alternative approach are shown in Table 3. Despite the theoretical advantages of this approach, the BMDs and BMDLs estimated here are generally consistent with those derived using the power and polynomial models (Table 2). The primary differences appear to be in the best estimates of the doses corresponding to a 5% increase in probability of postnatal death for F2 offspring at pnd 1 or pnd 4 and in the lower bounds on dose corresponding to 5% increased probability of death at pnd 4 in the F1 animals. DISCUSSION AND CONCLUSION
The most relevant BMDLs for the reproductive and developmental effects analyzed here were either 416 mg/kg/day (for male mating index) or 418 mg/kg/day (based on the preferred analysis of the F2 generation 4-day survival rate). The BMDLs are remarkably similar for the two types of effects significantly affected by isopropanol exposure. This is true even though the response rates that define the BMDs (the BMRs) for the two types of end point differ. For the male mating index (coded as a quantal response) the BMR is 10%,
42
ALLEN ET AL.
FIG. 2. Observed and predicted dose–response patterns for average offspring mortality. The polynomial mean (solid line) and power mean (dashed line) model predictions of the average percentage mortality correspond well to the observed means (F, shown 6 one standard deviation) for the F1 generation, through pnd 1.
whereas for pup survival (a nested quantal end point) the BMR is 5%. Although this appears paradoxical, it is consistent with the findings of Allen et al. (1994a,b) in which these two choices provided the closest match, on average, to corresponding NOAELs. There may be several reasons why different response rates corresponded to NOAELs depending on end point type, but one salient reason has to do with the ability to determine whether or not two groups are statistically significantly different. With the quantal end points, the tests employed tend to require at least a 10% increase in incidence before significant differences are detected. Group differences in continuous or nested quantal end points (the latter of which having much greater sample sizes since accounting for correlation allows one to TABLE 2 BMDLs for the Developmental Effects Using Litters as the Unit of Observation (Percentage Survival) Power model
Data set F1, F1, F2, F2,
1 4 1 4
day day day day
survival survival survival survival
Polynomial model
BMD BMDL BMD BMDL (mg/kg/day) (mg/kg/day) (mg/kg/day) (mg/kg/day) 807 588 1000 657
663 322 588 415
765 539 1000 657
642 343 588 415
consider the fetus or pup as the unit of observation) tend to be significant with less change from control. In this way, the end-point-specific differences in the choice of BMRs in this analysis are a by-product of common practice and accepted toxicological opinion with respect to what is meaningful and relevant to human health risk assessment, to the extent that NOAELs are considered to be a satisfactory starting point for such assessments. The BMDL values of interest lie between the study dosages of 100 and 500 mg/kg/day, for which there are conflicting no-effect interpretations. The USEPA (1992b) and Tyl (1996) interpreted the reductions in postnatal survival as treatment related and concluded that 100 mg/kg/day was the study NOAEL. Alternatively, Bevan et al. (1995) and Harris (1995) deemed the observations not to be biologically significant and concluded that the NOAEL was 500 mg/kg/day. The BMD approach provides a means to clarify the study findings and a means by which such disagreements can be circumvented. This is because it provides estimates of an experiment-independent quantity, the dose corresponding to a fixed level of response, rather than attempting to address the experiment-specific question of whether or not a treatment has caused a significantly different amount of response. It is also important to note that comparisons among NOAELs and BMDLs, such as those that formed the
REPRODUCTIVE BENCHMARK DOSES FOR ISOPROPANOL
43
FIG. 3. Observed and predicted dose–response patterns for offspring mortality. Unlike Fig. 2, this figure shows percentage mortality for each litter in each dose group (F; the number of overlapping observations is not presented, however). The log-logistic model (solid line) predicts the average value of those individual observations, adjusting for litter size and accounting for intralitter correlations among the observations for the F2 generation, through pnd 4.
basis for selecting BMR levels of 5 or 10% additional risk (Allen et al., 1994a,b), already account for uncertainties about the true shape of the dose–response curve. Such uncertainties affect estimates of the value of dose determined to correspond to any given response level. Thus, these uncertainties are reflected in the lower bounds that define the BMDLs. Moreover, by considering at least two model forms (e.g., Weibull and polynomial) a measure of model dependence can be ascertained. In the case of isopropanol, little model dependence existed. The BMR levels appropriate for this analysis were within the range of responses observed in the treatment groups, so uncertainty and model dependence were reduced. The consistency of the results obtained here for isopropanol reproductive/ developmental toxicity suggests that the set of experiTABLE 3 Recommended Approach for BMD Estimation of Developmental Effects Using a Beta-Binomial Model Data set F1, F1, F2, F2,
1 4 1 4
day day day day
survival survival survival survival
BMD (mg/kg/day)
BMDL (mg/kg/day)
911 656 1505 804
670 449 660 418
ments and accompanying BMD analyses form an appropriate basis for considering safe exposures to isopropanol. REFERENCES Allen, B., Kavlock, R., Kimmel, C., and Faustman, E. (1994a). Dose– response assessment for developmental toxicity. II. Comparison of generic benchmark dose estimates with no observed adverse effect levels. Fundam. Appl. Toxicol. 23, 487– 495. Allen, B., Kavlock, R., Kimmel, C., and Faustman, E. (1994b). Dose– response assessment for developmental toxicity. III. Statistical models. Fundam. Appl. Toxicol. 23, 496 –509. Allen, B. C., Strong, P. L., Price, C. J., Hubbard, S. A., and Daston, G. P. (1996). Benchmark dose analysis of developmental toxicity in rats exposed to boric acid. Fundam. Appl. Toxicol. 32, 194 –204. Bevan, C., Tyler, T., Gardiner, T., Kapp, R., Andrews, L., and Beyer, B. (1995). Two-generation reproduction toxicity study with isopropanol in rats. J. Appl. Toxicol. 15, 117–223. BIBRA (1988). A Single Generation Reproduction and Embryotoxicity Study with Isopropyl Alcohol in Rats. The British Industrial Biological Research Association, Carshalton, Surrey. Crump, K. (1984). A new method for determining allowable daily intakes. Fundam. Appl. Toxicol. 4, 854 – 871. Crump, K. (1995). Calculation of benchmark doses from continuous data. Risk Anal. 15, 79 – 89. Exxon Biomedical Sciences, Inc. (1992). Multi-Generation Rat Reproduction Study with Isopropanol. Project No. 259835. Prepared for Chemical Manufacturers Association, April 17, 1992.
44
ALLEN ET AL.
FDRL (1975). Toxicity Studies in Rats with 2-Butanol including Growth, Reproduction, and Teratologic Observations. Food and Drug Research Laboratories, East Orange, NJ. Gallo, M., Oser, B., Cox, G., and Bailey, D. (1977). Studies on the long-term toxicity of 2-butanol. Toxicol. Appl. Pharmacol. 41, 135. Gaylor, D. (1989). Quantitative risk analysis for quantal reproductive and developmental effects. Environ. Health Perspect. 79, 243– 246. Gaylor, D., and Slikker, W., Jr. (1990). Risk assessment for neurotoxic effects. Neurotoxicology 11, 211–218. Harris, S. B. (1995). A review of the EPA comments regarding the study entitled ‘‘Multi-generation rat reproduction study with isopropanol.’’ Report prepared for the Chemical Manufacturers Association Isopropanol Panel. Howe, R. (1984a). THRESH—A computer program to compute a reference dose from quantal animal toxicity data using the benchmark dose method. Howe, R. (1984b). THRESHW—A computer program to compute a reference dose from quantal animal toxicity data using the benchmark dose method. Howe, R. (1984c). THC—A computer program to compute a reference dose from continuous animal toxicity data using the benchmark dose method. Howe, R. (1984d). THWC—A computer program to compute a reference dose from continuous animal toxicity data using the benchmark dose method. Howe, R., and Van Landingham, C. (1990). TERALOG—A computer program for assessing risks of reproductive/developmental toxicity. Kavlock, R., Allen, B., Faustman, E., and Kimmel, C. (1994). Dose– response assessment for developmental toxicity. IV. Benchmark doses for fetal weight changes. Fundam. Appl. Toxicol. 26, 211– 222. Kimmel, C., and Gaylor, D. (1988). Issues in qualitative and quantitative risk analysis for developmental toxicology. Risk Anal. 8, 15–21. Lehman, A., Schwerma, H., and Richards, E. (1945). Acquired tolerance in dogs, rate of disappearance from the blood stream in various species, and effects on successive generations of rats. J. Pharmacol. Exp. Ther. 85, 61– 69.
Pease, W., Vandenburg, J., and Hooper, K. (1991). Comparing alternative approaches to establishing regulatory levels for reproductive toxicants: DBCP as a case study. Environ. Health Perspect. 91, 141–155. Tyl, R. W. (1996). Personal correspondence to Kathryn Rosica, Chemical Manufacturers Association, Arlington, VA, from Rochelle Tyl, Center for Life Sciences and Toxicology, Research Triangle Institute, Research Triangle Park, NC. February 12, 1996. U.S. Environmental Protection Agency (USEPA) (1986). Guidelines for the health assessment of suspect developmental toxicants. Fed. Reg. 50, 39426 –39436. U.S. Environmental Protection Agency (USEPA) (1988a). Proposed guidelines for assessing male reproductive risk. Fed. Reg. 53, 24850 –24969. U.S. Environmental Protection Agency (USEPA) (1988b). Proposed guidelines for assessing female reproductive risk. Fed. Reg. 53, 24834 –24847. U.S. Environmental Protection Agency (USEPA) (1990). Interim Methods for Development of Inhalation Reference Concentrations. EPA/600 8-90/066A. Review draft. Office of Research and Development, Washington, DC. August 1990. U.S. Environmental Protection Agency (USEPA) (1992a). IRIS. Background document (4/1/91). Office of Health and Environmental Assessment, Environmental Criteria and Assessment Office, Cincinnati, OH. U.S. Environmental Protection Agency (USEPA) (1992b). Memorandum to Keith Cronin, Chemical Testing Branch, from Jennifer Seed, Toxic Effects Section. July 1, 1992. U.S. Environmental Protection Agency (USEPA) (1993). Background and Quantitative Procedures for Derivation of Inhalation Reference Concentrations. Environmental Criteria and Assessment Office, U.S. Environmental Protection Agency, Research Triangle Park, NC. June 1993. U.S. Environmental Protection Agency (USEPA) (1995). The Use of the Benchmark Dose (BMD) Approach in Health Risk Assessment. Final report. EPA/630/R-94/007. Risk Assessment Forum, U.S. Environmental Protection Agency, Washington, DC.