- Email: [email protected]
Use of benchmark dose and meta-analysis to determine the most sensitive endpoint for risk assessment for dimethoate
Regulatory Toxicology and Pharmacology Regulatory Toxicology and Pharmacology 43 (2005) 55–65 www.elsevier.com/locate/yrtph
Use of benchmark dose and meta-analysis to determine the most sensitive endpoint for risk assessment for dimethoate Richard Reiss *, David Gaylor Sciences International, Inc., 1800 Diagonal Road, Suite 500, Alexandria, VA 22304, USA Received 5 April 2005 Available online 15 August 2005
Abstract A meta-analysis of several rat toxicity studies for dimethoate was conducted to determine the most sensitive endpoint for use in risk assessment. The analysis was motivated by a recent developmental neurotoxicity (DNT) study, which identified the same no observed adverse effect level (NOAEL) for pup mortality and cholinesterase inhibition. The pup mortality NOAEL was lower than that determined in a range-finding study for the DNT and other reproduction studies, and was highly influenced by a single total litter loss in the middle dose group, which made interpretation difficult. First, a meta-analysis was conducted of four recent studies by gavage dosing with very similar designs, including the DNT. Benchmark dose (BMD) modeling was used to determine the appropriate point of departure for regulatory purposes, the lower limit of the BMD for a 5% incidence for pup mortality (BMDL5) and the lower limit of a 10% inhibition of brain cholinesterase ðBMDL10 Þ, the asterisk denotes that the BMD is based on continuous response variable as opposed to an incidence level. For pup mortality, the BMDL5 for post-natal days (PND) 1–4 was 0.64 mg/ kg/day. For cholinesterase inhibition, the lowest BMDL10 was 0.19 mg/kg/day for the dams at gestation day 20. These results show that the regulatory point-of-departure for cholinesterase inhibition is more than threefold lower than pup mortality. Thus, risk assessments protecting against cholinesterase inhibition are likely to also be protective of pup mortality. In addition, cholinesterase inhibition and pup mortality were evaluated in two 2-generation reproduction studies by dietary exposure. Also, cholinesterase inhibition was evaluated in a 28-day dietary study. Dietary exposure is more relevant than gavage exposures for many human risk assessment scenarios. There was no consistent pup mortality at the highest doses of the two 2-generation dietary studies (6.0 and 6.5 mg/kg/day). The average BMD10s for brain cholinesterase inhibition for the 2-generation studies was 0.65 mg/kg/day, with a range of 0.49–0.96 mg/kg/day. This suggests that cholinesterase inhibition is at least a 10-fold more sensitive endpoint than pup mortality for dietary exposures. For the 28-day dietary study, the BMD10 for brain cholinesterase inhibition was 1.1 mg/kg/day for males and 0.70 mg/kg/day for females. The exposure duration in the 28-day dietary study is closest to the durations in the gavage studies. Compared to the dams in the gavage studies, which had a BMDL10 of 0.19 mg/kg/day, the animals were more than threefold more sensitive to cholinesterase inhibition by gavage compared to dietary exposure. Ó 2005 Elsevier Inc. All rights reserved. Keywords: Dimethoate; Insecticide; Benchmark dose; Cholinesterase; Developmental effects; Meta-analysis
1. Introduction Dimethoate (O,O-dimethyl S-[N-methylcarbamoylmethyl]phosphorodithioate) is an organophosphate insecticide with numerous uses on field and agricultural crops *
Corresponding author. Fax: +1 703 684 2223. E-mail address: [email protected] (R. Reiss).
0273-2300/$ - see front matter Ó 2005 Elsevier Inc. All rights reserved. doi:10.1016/j.yrtph.2005.06.012
and ornamentals. Like most organophosphates, dimethoate has been traditionally regulated based on its potential to cause inhibition of the acetylcholinesterase enzyme (Mileson et al., 1998). However, a recent developmental neurotoxicity study (DNT) with gavage exposure identified the same no observed adverse effect level (NOAEL) for pup mortality and cholinesterase inhibition (Myers, 2001a). The result in this study was highly
56
R. Reiss, D. Gaylor / Regulatory Toxicology and Pharmacology 43 (2005) 55–65
influenced by a single total litter loss at the middle dose group. The mortality at the middle dose group was not statistically significant. Nonetheless, these results raised questions about whether cholinesterase inhibition or pup mortality is the most sensitive endpoint on which to base a risk assessment for dimethoate, and whether dosing by gavage was a factor in the DNT study results. To address these questions, the extensive database of cholinesterase inhibition and reproductive and developmental tests for dimethoate were analyzed in a metaanalysis to determine the most appropriate endpoint for dimethoate risk assessment. First, the DNT data and several other studies with gavage exposure were analyzed in a meta-analysis using benchmark dose (BMD) modeling to ascertain the most sensitive endpoint by gavage exposure. In addition, studies assessing reproduction and cholinesterase inhibition by dietary exposure were analyzed to ascertain the difference in the relative sensitivity of cholinesterase inhibition and pup mortality by the dietary exposure route.
2. Materials and methods 2.1. Toxicity Studies The analysis in this article relies on studies conducted by dimethoate registrants in the US and Europe. There were four separate, but related, gavage studies. These studies were conducted according to international Good Laboratory Practice (GLP) standards, audited, and then submitted to the US EPA in support of the reregistration of dimethoate. The US EPA prepared summaries and evaluations of the studies which are available on the EPA website. The principal study was the DNT, and the other three studies followed a similar design. In the DNT, dimethoate was administered to 24 parent female Crl:CD BR rats by gavage dosing at dose levels of 0, 0.1, 0.5, and 3.0 mg/kg/day (Myers, 2001a). The exposure began on gestation day (GD) 6 and continued through postnatal day (PND) 10. The offspring were also exposed through gavage dosing at the same levels as the dams from PND 4 to 10. The study included functional observation battery (FOB) tests on the dams. For the offspring, the testing included FOB, automated motor activity, auditory startle response, and assessment of learning and memory (Morris Water Maze). For these tests, there were either no effects with dosage or possible effects at the highest dosage. Therefore, the focus of the analysis is on the pup mortality data. The DNT was conducted under US EPA Test Guideline OPPTS 870.6300 (US EPA, 1998a). In addition to the DNT, there were three other studies conducted with virtually identical designs with the same types of rats, and were done in the same laboratory during similar time periods. First, there was a compan-
ion cholinesterase study, which included cholinesterase measurements and pup mortality data (Myers, 2001b). The dose groups in the companion cholinesterase study were identical to the main DNT study. Plasma, red blood cell (RBC), and brain cholinesterase measurements were made for dams on GD20, for pups on PND4, PND11, PND21, and PND60, and for adult animals on days 1 and 11 of treatment. The cholinesterase analysis for all of the studies (including the dietary studies discussed below), was conducted with a modified Ellman method (US EPA, 1996). Although, each study separately measured plasma, RBC, and brain cholinesterase, dimethoate is regulated with the brain cholinesterase inhibition endpoint. Therefore, only the brain cholinesterase data are considered herein. There is also a range-finding study to the DNT, which included both cholinesterase and pup mortality measurements (Myers, 2001c). The range-finding study had the following dose groups: 0, 0.2, 3.0, and 6.0 mg/ kg/day. Finally, a cross-fostering study was conducted to further investigate the cause of the pup mortality (i.e., from exposure to the dams during gestation or exposure to the dams or pups during lactation) (Myers, 2004). There were six dose groups in the cross-fostering study that included varying dosages to the birth dam and the dam that raised the pups. However, the metaanalysis only relied on the two groups in the study that were not cross-fostered, including the control group and the 6 mg/kg/day dose group, because these are the only groups that can be considered replicate data with the other studies. Additionally, there are two dimethoate 2-generation reproduction studies with dietary exposure. The first study (Brooker and Stubbs, 1992) was conducted under US EPA Test Guidelines OPPTS 870.3800 (US EPA, 1998b, most recent version). Dimethoate was administered in the diet to Crl:CD BR rats at 0, 0.09, 1.3, and 6.0 mg/kg/day. The F0 animals (24/sex/group) were treated for 10 weeks prior to the first mating. The F1 generation was selected from the F1A litters (24/sex/ group) and was first mated at week 16. There was also a second mating of the F1A generation, followed by a partial third mating involving animals which had not been successful at either of their first two pairings. The second 2-generation reproduction study (Mellert et al., 2003) was conducted under OECD Guideline no. 416 (OECD, 1983) and US EPA Test Guidelines OPPTS 870.3800 (US EPA, 1998b). Dimethoate was administered via the diet to Wistar rats (CrlGlxBrlHan:WI) at 0, 0.2, 1.0, and 6.5 mg/kg/day. The F0 animals (25/sex/ group) were treated for at least 10 weeks prior to the first mating. The F0 animals were mated to produce a first litter (F1A) and subsequently remated (after about 20 additional weeks of treatment) to produce a second litter (F1B). A group from the F1A generation (25/sex/group) was selected as the F1 parental generation and were
R. Reiss, D. Gaylor / Regulatory Toxicology and Pharmacology 43 (2005) 55–65
treated post-weaning for about 10 weeks before producing an F2A generation. Following an additional 10 weeks of exposure, the F1 parents produced the F2B generation. Brain cholinesterase was measured at sacrifice for both matings in both generations. Finally, there was also a 28-day range-finding study to the 2003 2-generation reproduction study (Kasper et al., 2004). Dimethoate was administered to five male and five female Wistar rats (CrlGlxBrlHan:WI) in the diet at dose levels of 0, 1, 3, and 12.5 mg/kg/day. Brain cholinesterase was measured at sacrifice at 28 days. 2.2. Benchmark dose modeling methods Benchmark doses for pup mortality and cholinesterase inhibition were estimated using EPAÕs BenchMark Dose Software (BMDS) package (US EPA, 2001). The BMDS package includes algorithms for fitting many common dose–response curves, and allows users to estimate BMDs and lower limits on the BMDs. For the developmental toxicity endpoints, the nested logistic model in the BMDS software was used to model the dose–response and estimate BMDs. The nested logistic model was developed by Kupper et al. (1986) and has the following form: PrðresponseÞ ¼ a þ h1 rij þ
1 a h1 rij ; 1 þ exp b h2 rij q ln dosei ð1Þ
where the probability of response is the incidence of pup mortality in the individual litters, rij is the litter-specific covariate (litter size) for the ith dose group for the jth litter, and, h are constants with a P 0, b P 0, and q P 0. In addition, there is the option to include intralitter correlation coefficients, with 06 /i 6 1. The default of the BMDS software is to restrict q P 1 because if <1 the dose–response slope becomes infinite at the control dose, which is considered biologically implausible (US EPA, 2000b, p. 15). If the litter-specific covariate terms (h1 and h2) were close to zero and not statistically significant, the model was rerun without them. Also, if the terms were close to zero and not statistically significant, then intralitter correlations were not assumed. Allen et al. (1994) evaluated the nested logistic model and two other developmental toxicity dose–response models (Rai and Van Ryzin, 1985; Kodell et al., 1991) using a database of 607 developmental endpoints from 141 separate studies. The researchers found that the nested logistic model generally gave the best fit of the dose–response data, compared to the two other models. The authors also found that the ratio of the NOAEL (determined from the proportion of affected fetuses) and the 95% confidence lower limit of the BMD5 (or BMDL5) was approximately one (median from all end-
57
points). This result shows that the BMDL5 for this model is approximately equal to the NOAEL for a developmental study. It also suggests the limit of detection for typical developmental tests is about a 5% incidence. EPA typically uses 10% as a benchmark response for BMD modeling (US EPA, 2000a), but given that the limit of detection for developmental tests is closer to 5% and the severity of this effect, a 5% response level if often used, as in this study. Dimethoate has been regulated by the US EPA using the inhibition of brain cholinesterase as the key endpoint. In its recent cumulative risk assessment for the organophosphates, US EPA used the dose that causes a 10% inhibition of brain cholinesterase as a point-ofdeparture for risk assessment, and developed a dose–response model to estimate the BMD10 (US EPA, 2000a). EPA found that a 10% response was generally the limit of detection for brain cholinesterase inhibition; and made a policy decision that a BMD10 is an appropriate point-of-departure for risk assessment. Because the BMDs for cholinesterase are based on a percentage inhibition compared to control animals, as opposed to the percentage of animals responding like the pup mortality, the asterisk (*) is used when identifying the cholinesterase BMDs to differentiate them from the different definition of a BMD used for pup mortality. While the benchmark response levels that were chosen for pup mortality and cholinesterase inhibition are different (5% for pup mortality and 10% for cholinesterase inhibition), these levels reflect the policy decisions made by EPA. Therefore, these are appropriate levels for a regulatory risk assessment. For the organophosphate (OP) cumulative assessment, US EPA developed an exponential dose–response curve to describe cholinesterase inhibition as follows (US EPA, 2000b): CheðdoseÞ ¼ Che0 ½P b þ ð1 P b Þ expðm DoseÞ;
ð2Þ
where Che(dose) is the cholinesterase level at a given dose level, Che0 is the cholinesterase level at a zero dose, Pb represents a limiting value of the minimum cholinesterase activity at high doses, and m is the slope estimate of the dose response curve. This model was found to provide an excellent fit to the cholinesterase data for most of the organophosphates. The EPA also developed a more complicated version of this model that accounted for a low-dose flat region (or shoulder) and provided generally higher BMDs for OPs that exhibited this behavior. However, this more complicated model was not found to provide better model fits for dimethoate, so it was not used in this analysis. When appropriate constant variance was assumed, but in some cases the assumption of constant variance was rejected and the variance was allowed to vary among groups. As formulated, Eq. (2) cannot be explicitly modeled in EPAÕs BMDS software. Therefore, a modified proce-
58
R. Reiss, D. Gaylor / Regulatory Toxicology and Pharmacology 43 (2005) 55–65
each dose group of each study and the mean percent pup mortality across the litters in each group. With all of the studies combined, there are a total of 220 litters. However, the litters were not culled at PND4 in the cross-fostering study, as they were in the other studies. Hence, the cross-fostering study data are only used up to PND4. After PND4, these data cannot be considered replicates of the other studies. Therefore, for the PND4– 11 period the meta-analysis includes 169 litters. In the DNT study, there was an elevated mortality compared to controls at 0.5 mg/kg/day (8.0% for PND1–4) and 3.0 mg/kg/day (18.7% for PND1–4). These values were heavily influenced by one total litter loss (TLL) at 0.5 mg/kg/day and three TLLs at 3.0 mg/kg/day. The mortality at 0.5 mg/kg/day was not statistically significant, and the mortality at 3.0 mg/kg/day was statistically significant. By contrast, the mortality in the DNT range-finder was not significantly elevated at 3.0 mg/ kg/day (3.8% for PND1–4 compared to 3.5% for the controls), and the mortality was low in the comparative cholinesterase study at 0.5 mg/kg/day (1.5% for PND1– 4) and 3.0 mg/kg/day (0.7%). Given the virtually identical designs of the studies, the reason for the differences in pup mortality across the studies is most reasonably explained as statistical variability. Except at the highest dose groups, most litters have no mortality. The percent mortality is significantly driven by a few litters, particularly when there is a TLL. For example, for the 0.5 mg/ kg/day dose group in the DNT study, 9 of 24 litters had one or more mortality, one litter had 17 of the 32 deaths (53%) (the TLL), and two litters had 22 of the 32 deaths (69%). After the cull at day 4, there was elevated mortality at the highest dose groups (3.0 and 6.0 mg/kg/day), but the mortality was lower than during PND1–4.
dure was used to estimate the BMDs. First, Pb and m were estimated from the dose response data by an iterative least squares procedure. Initial estimates of Pb were zero or negative for all cases. Since Pb cannot be negative, Pb was set equal to zero. Hence, Eq. (2) can be linearized by taking the natural logarithm of each side of the equation: CheðdoseÞ ln ¼ m Dose. ð3Þ Che0 With Eq. (2) linearized, the BMD10 and lower 95% confidence limit ðBMDL10 Þ were estimated using the BMDS program with a linear dose–response model.
3. Results 3.1. Pup mortality by oral gavage exposure As noted above, there are four recent developmental studies with gavage exposures for dimethoate that were conducted at the same laboratory during the general time period. Therefore, this dataset is ideally suited for a meta-analysis. In many cases, a meta-analysis involves a joint analysis of data from several studies to derive an overall mean response, weighting the individual studies by sample size and variability. However, given the similarity of the gavage data, the data were simply combined together to conduct the meta-analysis in this study. In other words, the individual litter data from the four studies were considered replicates and combined. Therefore, each of the individual studies is essentially weighted by their sample sizes for the metaanalysis. Table 1 summarizes the number of litters in
Table 1 Pup morality data by dose group for four dimethoate studies by gavage dosing Study
Dose group
Number of litters
Mean pup mortality for days 1–4 (%)a
Mean pup mortality for days 4–11 (%)a
Main DNT (Myers, 2001a)
0 0.1 0.5 3.0
24 23 24 24
2.7 2.4 8.0 18.7
1.0 1.1 2.9 10.2
DNT range-finder (Myers, 2001c)
0 2 3.0 6.0
10 9 9 10
3.5 0.7 3.8 27.2
0.0 0.0 0.9 11.1
Comparative cholinesterase (Myers, 2001b)
0 0.1 0.5 3.0
10 10 10 10
0.0 1.5 1.5 0.7
0.0 0.0 0.0 0.8
Cross-fostering (Myers, 2004)
0 6.0
25 22
2.6 8.2
n/ab n/ab
a
Average percentage mortality across the litters. Because there was no cull at day 4 of the cross-fostering study, unlike the other gavage studies, the mortality from PND4-11 cannot be compared with the other studies. b
R. Reiss, D. Gaylor / Regulatory Toxicology and Pharmacology 43 (2005) 55–65
59
mated by the method described by Rao and Scott (1992). The model fit was relatively poor (p = 0.029) and below the EPAÕs recommended goodness-of-fit threshold (p = 0.10) for dose–response models. The BMD5 and BMDL5 estimates were 1.5 and 0.96 mg/ kg/day, respectively. The poor fit of the data was mostly the result of the poor agreement at the high dose (6 mg/ kg/day), where the mortality is similar to the level at the next highest dose (3 mg/kg/day). The dose–response model does not provide for a threshold effect where the response levels off at high dosages, so the poor fit at the high dose for these data is not unexpected. It is possible that there are different biological processes affecting the mortality at the high dose. A common technique in dose–response modeling that is used to achieve better model fits is to rerun the model without the highest dose group (Gold and Zeiger, 1997). This appears to be particularly appropriate in this case given the poor fit at the high dose. Fig. 2 displays the model fit for PND1–4 using the data from the four gavage studies but dropping the highest dosage. After dropping the highest dose, there
Therefore, the mortality during PND1–4 is the driver for risk assessment. Table 2 summarizes the results of the BMD modeling for pup mortality, including the BMD5 and BMDL5. Indications of the model fit are provided including the AkaikeÕs information coefficient (AIC) and goodnessof-fit probability values from a chi-square test. Lower values of the AIC and higher values of the goodnessof-fit p values indicate better fits. Both the AIC and goodness-of-fit p values are provided by the BMDS software, and are routinely used to assess model fits. Models were fit with and without the litter-specific covariate, and the model with the lowest AIC was chosen, except as noted. For all models, the regression terms for the intralitter correlation were significant, so intralitter correlations were used in all of the models. Fig. 1 displays the model fit for PND1–4 using the data from the four gavage studies. The plot shows the mean response and 95% confidence interval in each dose group, the fit of the dose–response model, and the upper bound of the model. The confidence intervals were esti-
Table 2 Summary of dimethoate BMD model results for pup mortality for gavage dosing (recommended point-of-departure for risk assessment in bold) Time period
Days 1–4 Days 4–11
Dataset
AICa
Goodness-of-fit p value
Used litter-specific covariate
BMD5 (BMDL5) (mg/kg/day)
Combined gavage Combined gavage, without highest dosage Combined gavage, without highest dosage, and without total litter loss at 0.5 mg/kg/day Combined gavage Combined gavage without highest dose group
1113.92 790.70 767.91
0.029 0.49 0.43
No Yes Yes
1.5 (0.96) 1.1 (0.64) 1.7 (1.0)
0.0007 n/ac
No No
2.3 (1.2) 2.3 (1.1)
252.02b 243.71b
Note. Intralitter correlations were found to be important for all of the models, and were included in the final model runs. Low values of the AIC and higher values of the p value of the goodness-of-fit test indicate better model fits. a The AIC is a useful measure for assessing model fits between different models using the same dataset, but the AIC cannot be usefully compared between models with different datasets. b The model with the litter-specific covariate has a lower AIC, but a visual inspection of the fit showed that not including the litter-specific covariate gave a more realistic model fit. The BMD for the model with the litter-specific covariate was higher; therefore, this choice was conservative. c The BMDS program could not calculate a goodness-of-fit.
Fig. 1. BMD model fit for dimethoate pup mortality during days 1–4 using all of the gavage data.
Fig. 2. BMD model fit for dimethoate pup mortality during days 1–4 using all of the gavage data without the highest dose group.
60
R. Reiss, D. Gaylor / Regulatory Toxicology and Pharmacology 43 (2005) 55–65
are still 188 litters. The goodness-of-fit was considerably better (p=0.49) than when the highest dose was included (p = 0.029). The BMD5 and BMDL5 estimates were 1.1 and 0.64 mg/kg/day, respectively. These values are lower than the BMD estimates when including the highest dose. Given the better fit of the data and the need to be conservative in conducting a regulatory risk assessment, Fig. 2 is considered the best fit model, and the most appropriate point-of-departure for risk assessment is 0.64 mg/kg/day, the BMDL5 from this model. It is notable that, even after combining the gavage data for the meta-analysis, the single total litter loss in the 0.5 mg/kg/day group still has a very large impact on the BMD results. With the combined dataset, the average percentage mortality at 0.5 mg/kg/day with the total litter loss included is 6.1%, while the average percentage mortality without the total litter loss was 3.2%. Therefore, the single total litter loss at 0.5 mg/ kg/day doubles the percentage mortality in the 0.5 mg/ kg/day dose group. Given the impact of the single total litter loss at the 0.5 mg/kg/day dose group, it is appropriate to examine the model fit without the total litter loss. Fig. 3 displays the model fit for PND1–4 using the data from the four gavage studies but not including the highest dosage and not including the single total litter loss at 0.5 mg/kg/day in the DNT study. The goodness-of-fit was adequate (p = 0.43), and the model predicts the mortality at the 0.5 mg/kg/day dose group better than in Fig. 2. The BMD5 and BMDL5 estimates were 1.7 and 1.0 mg/kg/day, respectively. The values are significantly higher than the estimates for Fig. 2. The rationale for dropping the TLL is that it may be an outlier, although there is not a good outlier test for these non-normally distributed data. Therefore, while this analysis is interesting, the conservative approach for regulatory risk assessment is to use the BMD estimates with the model including the TLL.
Fig. 3. BMD model fit for dimethoate pup mortality during days 1–4 using all of the gavage data without the highest dose group and without total litter loss in 0.5 mg/kg/day dose group.
The pup mortality was greater for the period from PND1–4 compared to PND4–11, but the mortality during PND4–11 was still elevated compared to the controls. The results for PND1–4 are the most appropriate to use for risk assessment given the greater mortality during PND1–4. However, modeling was conducted for PND4–11 for comparison purposes. The dose–response models did not fit the data as well for PND4– 11 as for PND1–4 due to a weaker response during PND4–11. 3.2. Pup mortality by dietary exposure The pup mortality data for the 1992 2-generation study are summarized in Table 3, and the pup mortality data for the 2003 2-generation study are summarized in Table 4. For the 2-generation studies, the greatest amount of mortality occurred during the first four days, so only the mortality for days 1–4 is reported in the tables. Both of the 2-generation studies included two separate matings for each generation. There were no statistically significant increases in mortality for any dose group in any of the studies. There was a non-statistically significant elevated mortality in the first mating of the F1 generation in the 1992 study, but this result was caused by two total litter losses in the middle and high dose groups. BMD modeling cannot be conducted for these data given that most of the responses are below a 5% risk level, and there is not a clear dose–response pattern. However, it is clear that there is not a 5% risk at the highest dosage when all of the data are considered. Therefore, the only possible estimates of the BMD5s are >6.0 mg/kg/day for the 1992 study and >6.5 mg/kg/day for the 2003 study (i.e., the highest Table 3 Pup mortality for dimethoate by dose group for a 2-generation rat reproduction dietary study in 1992 Group
Dosage to dam (mg/kg/day)
Pregnant dams
F0 generation—first mating
0 0.09 1.3 6.0
26 27 24 25
6.6 3.9 2.1 2.2
F0 generation—second mating
0 0.09 1.3 6.0
25 26 25 20
2.7 3.3 2.9 3.4
F1 generation—first mating
0 0.09 1.3 6.0
23 17 17 15
3.2 3.0 10.7 11.1
F1 generation—second mating
0 0.09 1.3 6.0
16 16 14 12
1.7 1.2 1.5 1.4
a
Average percentage mortality across the litters.
Pup mortality for days 1–4a
R. Reiss, D. Gaylor / Regulatory Toxicology and Pharmacology 43 (2005) 55–65
61
Table 4 Pup mortality for dimethoate by dose group for a 2-generation rat reproduction dietary study in 2003
Table 5 Samples sizes for dimethoate brain cholinesterase measurements by dose group for gavage studies
Group
Dosage to dam (mg/kg/day)
Pregnant dams
Pup mortality for days 1–4a
Study
Subpopulation
Number of animals/ dose group
F1 generation—first mating
0 0.2 1 6.5
21 24 25 23
1.2 1.3 2.2 2.6
Comparative cholinesterase study
F1 generation—second mating
0 0.2 1 6.5
24 22 21 25
1.2 1.1 0.74 4.5
F2 generation—first mating
0 0.2 1 6.5
22 24 24 25
3.2 7.1 3.0 1.5
F2 generation—second mating
0 0.2 1 6.5
22 25 22 25
3.6 5.4 2.5 4.2
GD20 dams GD20 fetuses PND4 male offspring PND4 female offspring PND11 male offspring PND11 female offspring PND21 male offspring PND21 female offspring PND60 male offspring PND60 female offspring Adult males, Day 1 Adult females, Day 1 Adults males, Day 11 Adult females, Day 11 GD20 dams GD20 male fetuses GD20 female fetuses PND21 male offspring PND21 female offspring
8 8 14–19 13–16 8 8 8 8 8 8 8 8 8 8 5 5 5 15–20 14–20
a
Average percentage mortality across the litters.
doses tested). The best estimate of the BMD5 for pup mortality for the gavage studies was 1.1 mg/kg/day (a lower limit of the BMD can not be estimated for the dietary studies for comparison, so the central estimate of the BMDs were compared instead). Comparing the BMD5 for the gavage studies of 1.1 mg/kg/day, rats appear to be at least sixfold more sensitive for pup mortality after exposure to dimethoate when the dose is given by gavage compared to a dietary dose. 3.3. Cholinesterase Inhibition by oral gavage exposure Of the four dimethoate gavage studies that included pup mortality measurements, two of the studies, the comparative cholinesterase study (Myers, 2001b) and the DNT range-finding study (Myers, 2001c), included measurement of brain cholinesterase inhibition. The sample sizes are summarized in Table 5, including separate values for adults, fetuses, and pups at different time points of the study and different exposure regimens. For the offspring at PND4 and PND11, and adults at days 4 and 11, there are only measurements from the comparative cholinesterase study. Therefore, the BMDs were calculated with these data only. The PND4 offspring represent pups with exposure only from the milk of the dam and previous in utero exposure, as the pups are not being directly dosed at this time point. The PND11 pups were dosed once. For the GD20 dams and fetuses and the PND21 offspring, measurements were made in both the comparative cholinesterase and DNT range-finding studies. Since the study designs were identical, these represent replicate data. Therefore, these data were combined to estimate the BMDs. Furthermore, there were not significant differences in the
DNT range— finding study
cholinesterase measurements between the males and females at PND21. Therefore, the male and female data were combined to make a larger dataset. The PND21 offspring represent animals that were dosed 11 consecutive days (from PND11 to PND21). The estimated BMDs for brain cholinesterase are summarized in Table 6. The table shows the BMD10 , BMDL10 , and the p value of the goodness of fit for the dose–response model. The BMDs did not change dramatically when the range-finding data were added to the comparative cholinesterase dataset. Some of the goodness-of-fit p values in Table 6 are lower than EPAÕs acceptable level of 0.10. However, a review of the dose– response curves show that the fit of the exponential model is excellent in almost all cases. Figs. 4–6 show example model fits to the data for GD20 dams (Fig. 4), GD20 fetuses (Fig. 5), and adults on days 11 (Fig. 6). The BMD10s ranged from 0.29 mg/kg/day (GD20 dams) and 0.64 mg/kg/day (PND21 offspring) to 2.6 mg/kg/ day (adults males, day 1) and 4.0 mg/kg/day (PND4 females). The plots show the mean response, the 95% confidence interval on the mean response, and the model fit. 3.4. Cholinesterase inhibition by dietary exposure There are several dietary studies which measured cholinesterase inhibition for dimethoate, including the 2-generation rat reproduction study in 1992, the 2-generation reproduction study in 2003, and a 28-day dietary study. Table 7 summarizes the BMD estimates for the dietary studies. The BMDs are all for adult animals that were sacrificed at the end of the measurements periods in the studies. Fig. 7 shows the dose–response fit for the males in the 28-day dietary study.
62
R. Reiss, D. Gaylor / Regulatory Toxicology and Pharmacology 43 (2005) 55–65
Table 6 Summary of dimethoate BMD estimates for brain cholinesterase inhibition for gavage studies Subpopulation
Dataset
Length of exposure (days)
BMD10 (mg/kg/day)
BMDL10 Goodness-of(mg/kg/day) fit p valueb
GD20 dams
Comparative cholinesterase and DNT range-find Comparative cholinesterase and DNT range-find Comparative cholinesterase Comparative cholinesterase Comparative cholinesterase Comparative cholinesterase Comparative cholinesterase and DNT range-find Comparative cholinesterase Comparative cholinesterase Comparative cholinesterase Comparative cholinesterase
15
0.29
0.19
0.99
15 in utero
1.0
0.65
0.72
16 in utero + 4 via milk 16 in utero and 4 via milk 1 direct dose + 16 in utero and 10 via milk 1 direct dose + 16 in utero and 10 via milk 11 direct doses + 16 in utero and 10 via milk
3.8a 4.0a 1.8 1.5 0.64
2.2a 2.4a 1.5 1.2 0.58
0.025 0.03 0.41 0.02 0.27
1 1 11 11
2.6 2.1 0.47 0.36
2.0 1.7 0.36 0.32
GD20 fetuses PND4 males PND4 females PND11 males PND11 females PND21 offspring Adults males, Day 1 Adult females, Day 1 Adults males, Day 11 Adult females, Day 11
0.83 0.19 <0.01 0.09
a
Uncertain because there was less than 10% inhibition at the highest dosage. Some of the p values below 5% were for dose–response curves that appeared to fit the data average quite adequately; therefore, it is not clear that these p values are representative of the data fit. The lack of fit may be due to the large variability within dose groups. b
Fig. 4. BMD model fit for dimethoate brain cholinesterase for GD20 dams for gavage dosing.
Fig. 6. BMD model fit for dimethoate brain cholinesterase for adult females on day 11 for gavage dosing.
4. Discussion 4.1. Most sensitive subpopulation for cholinesterase inhibition by gavage
Fig. 5. BMD model fit for dimethoate brain cholinesterase for GD20 fetuses for gavage dosing.
Before comparing the pup mortality and cholinesterase BMDs to determine the most sensitive endpoint for risk assessment, the cholinesterase BMDs for gavage were compared to determine the most sensitive subpopulation for cholinesterase inhibition. From a cursory inspection, the dams have a lower BMDL10 (0.19 mg/ kg/day) compared to the adult females (0.32) (or nonpregnant females). However, the 90th percentile error bounds overlap so there is not a statistical difference. When only using the values from the comparative cholinesterase study, the comparison is much closer (BMD10 of 0.33 mg/kg/day for dams and 0.36 mg/kg/
R. Reiss, D. Gaylor / Regulatory Toxicology and Pharmacology 43 (2005) 55–65
63
Table 7 Summary of dimethoate BMD estimates for brain cholinesterase inhibition in the dietary studies Study
Subpopulation
BMD10 (mg/kg/day)
BMDL10 (mg/kg/day)
1992 2-generation rat reproduction study
F0 males F1 males F0 females F1 females F0 males F1 males F0 females F1 females Males Females
0.62 0.58 0.62 0.49 0.96 0.77 0.56 0.62 1.2 0.95
0.59 0.50 0.51 0.46 0.77 0.67 0.50 0.58 1.1 0.70
2003 2-generation rat reproduction study
28-day study
adults are lower than the BMD values for PND 21 juvenile offspring indicating that the adults are more sensitive to cholinesterase inhibition than are these offspring. This result is consistent with an analysis by Sheets (2000) of multi-generational reproduction studies with five organophosphate insecticides. To summarize, the relative sensitivity of these subpopulations to repeated gavage exposure to dimethoate is: dams = day 11 adults > fetuses > offspring on PND 21 > nursing offspring on PND 4 (most sensitive to least sensitive). 4.2. Most sensitive endpoint for gavage exposure Fig. 7. BMD model fit for dimethoate brain cholinesterase for males in 28-day dietary study.
day for adult females). When the range-finder study data were added for the dams, the BMDs were reduced (there was no range-finder data for non-pregnant females). Therefore, the pregnant dams were not more sensitive than non-pregnant females. The BMD values for dams and day 11 adults are all lower than the BMD values for fetuses. This point is further underscored by comparing the dose–response curves, which shows that the cholinesterase inhibition for the fetuses has a lower plateau than for the adults. For the fetuses, there is only a 39% inhibition at the highest dosage of 6 mg/kg/day, whereas there is 88% inhibition for the dams. As discussed above, the cholinesterase inhibition for pregnant dams and adult females on day 11 was very similar. The cholinesterase inhibition for adult males on day 11 was lower than for adult females on day 11. Therefore, dams and adult males and females (after consecutive dosing) were more sensitive to cholinesterase inhibition than the fetuses (assuming the dose to the dam as the dose to the fetus). The BMD values for dams and day 11 adults are lower than the BMD for PND 4 offspring indicating that nursing infants are less sensitive to cholinesterase inhibition than are the adults. The BMD values for dams and day 11
The BMD estimates provide an ideal method to compare the relative sensitivity of the animals to cholinesterase inhibition and pup mortality from gavage doses of dimethoate. The comparison is made on the basis of the appropriate points-of-departure for risk assessment. Most BMD analyses are based on a 10% incidence. However, as discussed above, for developmental endpoints, the appropriate point-of-departure is the lower limit estimate of the 5% incidence, or a BMDL5. For cholinesterase inhibition, EPA has determined that a 10% inhibition of brain cholinesterase is the most biologically relevant measurement. Therefore, the BMDL10 is the appropriate point-of-departure. Fig. 8 provides a graphical comparison of the estimated BMDs for pup mortality and cholinesterase inhibition. The graph shows both the central tendency estimates (BMD5 and BMD10 ) and the lower and upper limits of the estimates. The PND4 offspring are not included on the graph because the BMD estimates are so much larger than the other estimates. The pup mortality BMD5 is relatively similar, within the error bounds, to the cholinesterase inhibition BMD10 for fetuses and PND11 pups. For the PND21 pups, adult males and females on day 11, and the dams, the BMD10 estimates were lower than the BMD5 for pup mortality. There was a small overlap on the statistical bounds of the BMD10 for cholinesterase for PND21 pups and the BMD5 for pup mortality. There was no
64
R. Reiss, D. Gaylor / Regulatory Toxicology and Pharmacology 43 (2005) 55–65
Fig. 8. Comparison of dimethoate gavage bmds for pup mortality and brain cholinesterase inhibition.
overlap of the statistical bounds of BMD10 for cholinesterase for adult males and females and the dams, and the BMD5 for pup mortality. These results indicate that the most sensitive endpoint for dimethoate is cholinesterase inhibition for pregnant dams and adult males and females. Using central estimates, the ratio of the BMD5 for pup mortality and the BMD10 for cholinesterase inhibition for the dams is 3.8. Comparing the lower limits, the ratio is 3.4. Therefore, the risk assessment points-of-departure for cholinesterase inhibition is more than threefold lower than for pup mortality. This demonstrates that the use of points-of-departure for cholinesterase inhibition is also likely to be protective of pup mortality. 4.3. Most sensitive endpoint for dietary exposure Dietary exposure is more relevant than gavage exposures for many human risk assessment scenarios. Therefore, it is important to analyze the dietary toxicity data separately. First, the dietary BMD estimates can be compared with BMD estimates for adults in the comparative cholinesterase study. The day 11 adult values for gavage provide the best basis for comparison, as these more closely compare with the dietary studies which included multiple exposures. For males, the average BMDL10 for the four 2-generation reproduction dietary measurements was 0.63 mg/kg/day, compared to 0.36 mg/kg/day for the gavage study, suggesting about a twofold greater sensitivity for gavage exposures (the rats in the dietary study were exposed longer). For females, the average BMDL10 for the four 2-generation reproduction dietary measurements was 0.51 mg/kg/day, compared to 0.32 mg/kg/ day for the gavage study, suggesting about a 1.5-fold greater sensitivity for gavage exposures. In the comparative cholinesterase study, the females were exposed for 11 days, while the exposures in the dietary studies ranged from 17 to 44 weeks. Thus, even with the shorter duration
exposures, the rats were more sensitive to cholinesterase inhibition by gavage. The closest comparison in regard to duration is with the 28-day dietary study. For males, the BMDL10 value for dietary exposure was threefold higher than the gavage value. For females, the BMDL10 value for dietary exposure was more than twofold higher than the gavage value. These comparisons show that the animals were more sensitive to gavage exposures in regard to cholinesterase inhibition. One uncertainty in this analysis is that the 1992 study used a different strain of rats than the 2003 study. For pup mortality, there was less than a 5% risk at the highest dosages for the 2-generation reproduction studies. Therefore, the BMD5s were defined as greater than the highest dosage (>6 mg/kg/day for the 1992 study, and >6.5 mg/kg/day for the 2003 study). The average BMDL10s for cholinesterase for the 1992 study was 0.52 mg/kg/day, and 0.63 mg/kg/day for the 2003 study. This comparison suggests that cholinesterase inhibition is at least 10-fold more sensitive an endpoint than pup mortality for dietary exposures.
5. Conclusion This article presents a meta-analysis of several rat toxicity studies for dimethoate. Compared to the most sensitive subpopulation for brain cholinesterase inhibition (dams) for gavage exposure, the points-of-departure for risk assessment for cholinesterase inhibition is more than threefold lower than for pup mortality. Therefore, the use of cholinesterase inhibition as an endpoint for risk assessment is likely to be protective of pup mortality. A comparison of brain cholinesterase inhibition and pup mortality in the dietary studies shows that cholinesterase inhibition is about a 10-fold more sensitive endpoint than pup mortality. By route of exposure, the animals were about 2–3-fold more sensitive for brain cholinesterase inhibition by gavage than by dietary exposure. For pup mortality, the animals were at least sixfold more sensitive by gavage exposure compared to dietary exposure.
Acknowledgments This work was sponsored by Cheminova A/S. The authors wish to thank Paul Whatling, Diane Allemang, Dorrit Søndergaard, Judith Hauswirth, Don OÕShaughnessy, and Abby Li for their helpful comments.
References Allen, B.C., Kavlock, R.J., Kimmel, C.A., Faustman, E.M., 1994. Dose–response assessment for developmental toxicity: III. Statistical models. Fundamental and Applied Toxicology 23, 496–509.
R. Reiss, D. Gaylor / Regulatory Toxicology and Pharmacology 43 (2005) 55–65 Brooker, A., Stubbs, A., 1992. Dimethoate: dietary range finding study in mature male and female rate and their juvenile offspring. Report by Huntingdon Research Center, Ltd conducted under International GLP. Gold, L.S., Zeiger, E., 1997. Handbook of Carcinogenic Potency and Genotoxicity Databases. CRC Press, Boca Raton, FL, p. 559. Kasper, U., Kaufmann, W., Deckardt, K., van Ravenzwaay, B., 2004. Dimethoate range-finding study in Wistar rats, administration via the diet over 4 weeks. Study conducted at BASF Aktiengesellchaft under international GLP. Kodell, R., Howe, R., Chen, J., Gaylor, D., 1991. Mathematical modeling of reproductive and developmental toxic effects for quantitative risk assessment. Risk Analysis 11, 583–590. Kupper, L., Portier, C., Hogan, M., Yamamoto, E., 1986. The impact of litter effects on dose–response modeling in teratology. Biometrics 42, 85–98. Mellert, W., Hellwing, J., Gembardt, C., Deckardt, K. Van Ravenzwaay, B. 2003. Dimethoate two-generation reproduction toxicity study in Wistar rats— adminstration in the diet. Study conducted at BASF Aktiengesellchaft under international GLP. Mileson, B.E., Chambers, J.E., Chen, W.L., et al., 1998. Common mechanism of toxicity: a case study of organophosphorus pesticides. Toxicological Sciences 41, 8–20. Myers, D.P., 2001a. Dimethoate developmental neurotoxicity study in the CD rat by oral gavage administration. Study conducted by Huntingdon Life Sciences, Ltd under international GLP. Myers, D.P., 2001b. Dimethoate effects on cholinesterase in the CD rat (adult and juvenile) by oral gavage administration. Study conducted by Huntingdon Life Sciences, Ltd under international GLP. Myers, D.P., 2001c. Dimethoate dose finding study in CD rats by oral gavage administration preliminary to developmental neurotoxicity
65
study. Study conducted by Huntingdon Life Sciences, Ltd under international GLP. Myers, D.P., 2004. Dimethoate cross fostering study in CD rats. Study conducted by Huntingdon Life Sciences, Ltd under international GLP. OECD Test Guideline No. 416. Two Generation Reproduction Toxicity. Dated May 26, 1983. Paris, Organisation for Economic Cooperation and Development. Rai, K., Van Ryzin, J., 1985. A dose–response model for teratological experiments involving quantal responses. Biometrics 41, 1–9. Rao, J.N.K., Scott, A.J., 1992. A simple method for the analysis of clustered binary data. Biometrics 48, 577–585. Sheets, L.P., 2000. A consideration of age-dependent differences in susceptibility to organophosphorous and pyrethroid insecticides. Neurotoxicology 1–2, 57–63. US EPA, 1996. Standard operating procedure for measuring cholinesterase inhibition in laboratory rats and dogs exposed to nonreversible cholinesterase inhibitors. Federal Register, April 26, 1996, vol. 61, p. 18593. US EPA, 1998a. Health effects test guidelines OPPTS 870.6300 developmental neurotoxicity study. Report No. EPA-712-C-98239. US EPA, 1998b. Health effects test guidelines OPPTS 870.3800 reproduction and fertility effects. Report No. EPA-712-C-98-208. US EPA, 2000a. Benchmark dose technical guidance document. EPA Report No. EPA/630/R-00/001. US EPA, 2000b. Revised OP Cumulative Risk Assessment. Prepared by the Office of Pesticide Programs:
Use of benchmark dose and meta-analysis to determine the most sensitive endpoint for risk assessment for dimethoate Richard Reiss *, David Gaylor Sciences International, Inc., 1800 Diagonal Road, Suite 500, Alexandria, VA 22304, USA Received 5 April 2005 Available online 15 August 2005
Abstract A meta-analysis of several rat toxicity studies for dimethoate was conducted to determine the most sensitive endpoint for use in risk assessment. The analysis was motivated by a recent developmental neurotoxicity (DNT) study, which identified the same no observed adverse effect level (NOAEL) for pup mortality and cholinesterase inhibition. The pup mortality NOAEL was lower than that determined in a range-finding study for the DNT and other reproduction studies, and was highly influenced by a single total litter loss in the middle dose group, which made interpretation difficult. First, a meta-analysis was conducted of four recent studies by gavage dosing with very similar designs, including the DNT. Benchmark dose (BMD) modeling was used to determine the appropriate point of departure for regulatory purposes, the lower limit of the BMD for a 5% incidence for pup mortality (BMDL5) and the lower limit of a 10% inhibition of brain cholinesterase ðBMDL10 Þ, the asterisk denotes that the BMD is based on continuous response variable as opposed to an incidence level. For pup mortality, the BMDL5 for post-natal days (PND) 1–4 was 0.64 mg/ kg/day. For cholinesterase inhibition, the lowest BMDL10 was 0.19 mg/kg/day for the dams at gestation day 20. These results show that the regulatory point-of-departure for cholinesterase inhibition is more than threefold lower than pup mortality. Thus, risk assessments protecting against cholinesterase inhibition are likely to also be protective of pup mortality. In addition, cholinesterase inhibition and pup mortality were evaluated in two 2-generation reproduction studies by dietary exposure. Also, cholinesterase inhibition was evaluated in a 28-day dietary study. Dietary exposure is more relevant than gavage exposures for many human risk assessment scenarios. There was no consistent pup mortality at the highest doses of the two 2-generation dietary studies (6.0 and 6.5 mg/kg/day). The average BMD10s for brain cholinesterase inhibition for the 2-generation studies was 0.65 mg/kg/day, with a range of 0.49–0.96 mg/kg/day. This suggests that cholinesterase inhibition is at least a 10-fold more sensitive endpoint than pup mortality for dietary exposures. For the 28-day dietary study, the BMD10 for brain cholinesterase inhibition was 1.1 mg/kg/day for males and 0.70 mg/kg/day for females. The exposure duration in the 28-day dietary study is closest to the durations in the gavage studies. Compared to the dams in the gavage studies, which had a BMDL10 of 0.19 mg/kg/day, the animals were more than threefold more sensitive to cholinesterase inhibition by gavage compared to dietary exposure. Ó 2005 Elsevier Inc. All rights reserved. Keywords: Dimethoate; Insecticide; Benchmark dose; Cholinesterase; Developmental effects; Meta-analysis
1. Introduction Dimethoate (O,O-dimethyl S-[N-methylcarbamoylmethyl]phosphorodithioate) is an organophosphate insecticide with numerous uses on field and agricultural crops *
Corresponding author. Fax: +1 703 684 2223. E-mail address: [email protected] (R. Reiss).
0273-2300/$ - see front matter Ó 2005 Elsevier Inc. All rights reserved. doi:10.1016/j.yrtph.2005.06.012
and ornamentals. Like most organophosphates, dimethoate has been traditionally regulated based on its potential to cause inhibition of the acetylcholinesterase enzyme (Mileson et al., 1998). However, a recent developmental neurotoxicity study (DNT) with gavage exposure identified the same no observed adverse effect level (NOAEL) for pup mortality and cholinesterase inhibition (Myers, 2001a). The result in this study was highly
56
R. Reiss, D. Gaylor / Regulatory Toxicology and Pharmacology 43 (2005) 55–65
influenced by a single total litter loss at the middle dose group. The mortality at the middle dose group was not statistically significant. Nonetheless, these results raised questions about whether cholinesterase inhibition or pup mortality is the most sensitive endpoint on which to base a risk assessment for dimethoate, and whether dosing by gavage was a factor in the DNT study results. To address these questions, the extensive database of cholinesterase inhibition and reproductive and developmental tests for dimethoate were analyzed in a metaanalysis to determine the most appropriate endpoint for dimethoate risk assessment. First, the DNT data and several other studies with gavage exposure were analyzed in a meta-analysis using benchmark dose (BMD) modeling to ascertain the most sensitive endpoint by gavage exposure. In addition, studies assessing reproduction and cholinesterase inhibition by dietary exposure were analyzed to ascertain the difference in the relative sensitivity of cholinesterase inhibition and pup mortality by the dietary exposure route.
2. Materials and methods 2.1. Toxicity Studies The analysis in this article relies on studies conducted by dimethoate registrants in the US and Europe. There were four separate, but related, gavage studies. These studies were conducted according to international Good Laboratory Practice (GLP) standards, audited, and then submitted to the US EPA in support of the reregistration of dimethoate. The US EPA prepared summaries and evaluations of the studies which are available on the EPA website. The principal study was the DNT, and the other three studies followed a similar design. In the DNT, dimethoate was administered to 24 parent female Crl:CD BR rats by gavage dosing at dose levels of 0, 0.1, 0.5, and 3.0 mg/kg/day (Myers, 2001a). The exposure began on gestation day (GD) 6 and continued through postnatal day (PND) 10. The offspring were also exposed through gavage dosing at the same levels as the dams from PND 4 to 10. The study included functional observation battery (FOB) tests on the dams. For the offspring, the testing included FOB, automated motor activity, auditory startle response, and assessment of learning and memory (Morris Water Maze). For these tests, there were either no effects with dosage or possible effects at the highest dosage. Therefore, the focus of the analysis is on the pup mortality data. The DNT was conducted under US EPA Test Guideline OPPTS 870.6300 (US EPA, 1998a). In addition to the DNT, there were three other studies conducted with virtually identical designs with the same types of rats, and were done in the same laboratory during similar time periods. First, there was a compan-
ion cholinesterase study, which included cholinesterase measurements and pup mortality data (Myers, 2001b). The dose groups in the companion cholinesterase study were identical to the main DNT study. Plasma, red blood cell (RBC), and brain cholinesterase measurements were made for dams on GD20, for pups on PND4, PND11, PND21, and PND60, and for adult animals on days 1 and 11 of treatment. The cholinesterase analysis for all of the studies (including the dietary studies discussed below), was conducted with a modified Ellman method (US EPA, 1996). Although, each study separately measured plasma, RBC, and brain cholinesterase, dimethoate is regulated with the brain cholinesterase inhibition endpoint. Therefore, only the brain cholinesterase data are considered herein. There is also a range-finding study to the DNT, which included both cholinesterase and pup mortality measurements (Myers, 2001c). The range-finding study had the following dose groups: 0, 0.2, 3.0, and 6.0 mg/ kg/day. Finally, a cross-fostering study was conducted to further investigate the cause of the pup mortality (i.e., from exposure to the dams during gestation or exposure to the dams or pups during lactation) (Myers, 2004). There were six dose groups in the cross-fostering study that included varying dosages to the birth dam and the dam that raised the pups. However, the metaanalysis only relied on the two groups in the study that were not cross-fostered, including the control group and the 6 mg/kg/day dose group, because these are the only groups that can be considered replicate data with the other studies. Additionally, there are two dimethoate 2-generation reproduction studies with dietary exposure. The first study (Brooker and Stubbs, 1992) was conducted under US EPA Test Guidelines OPPTS 870.3800 (US EPA, 1998b, most recent version). Dimethoate was administered in the diet to Crl:CD BR rats at 0, 0.09, 1.3, and 6.0 mg/kg/day. The F0 animals (24/sex/group) were treated for 10 weeks prior to the first mating. The F1 generation was selected from the F1A litters (24/sex/ group) and was first mated at week 16. There was also a second mating of the F1A generation, followed by a partial third mating involving animals which had not been successful at either of their first two pairings. The second 2-generation reproduction study (Mellert et al., 2003) was conducted under OECD Guideline no. 416 (OECD, 1983) and US EPA Test Guidelines OPPTS 870.3800 (US EPA, 1998b). Dimethoate was administered via the diet to Wistar rats (CrlGlxBrlHan:WI) at 0, 0.2, 1.0, and 6.5 mg/kg/day. The F0 animals (25/sex/ group) were treated for at least 10 weeks prior to the first mating. The F0 animals were mated to produce a first litter (F1A) and subsequently remated (after about 20 additional weeks of treatment) to produce a second litter (F1B). A group from the F1A generation (25/sex/group) was selected as the F1 parental generation and were
R. Reiss, D. Gaylor / Regulatory Toxicology and Pharmacology 43 (2005) 55–65
treated post-weaning for about 10 weeks before producing an F2A generation. Following an additional 10 weeks of exposure, the F1 parents produced the F2B generation. Brain cholinesterase was measured at sacrifice for both matings in both generations. Finally, there was also a 28-day range-finding study to the 2003 2-generation reproduction study (Kasper et al., 2004). Dimethoate was administered to five male and five female Wistar rats (CrlGlxBrlHan:WI) in the diet at dose levels of 0, 1, 3, and 12.5 mg/kg/day. Brain cholinesterase was measured at sacrifice at 28 days. 2.2. Benchmark dose modeling methods Benchmark doses for pup mortality and cholinesterase inhibition were estimated using EPAÕs BenchMark Dose Software (BMDS) package (US EPA, 2001). The BMDS package includes algorithms for fitting many common dose–response curves, and allows users to estimate BMDs and lower limits on the BMDs. For the developmental toxicity endpoints, the nested logistic model in the BMDS software was used to model the dose–response and estimate BMDs. The nested logistic model was developed by Kupper et al. (1986) and has the following form: PrðresponseÞ ¼ a þ h1 rij þ
1 a h1 rij ; 1 þ exp b h2 rij q ln dosei ð1Þ
where the probability of response is the incidence of pup mortality in the individual litters, rij is the litter-specific covariate (litter size) for the ith dose group for the jth litter, and, h are constants with a P 0, b P 0, and q P 0. In addition, there is the option to include intralitter correlation coefficients, with 06 /i 6 1. The default of the BMDS software is to restrict q P 1 because if <1 the dose–response slope becomes infinite at the control dose, which is considered biologically implausible (US EPA, 2000b, p. 15). If the litter-specific covariate terms (h1 and h2) were close to zero and not statistically significant, the model was rerun without them. Also, if the terms were close to zero and not statistically significant, then intralitter correlations were not assumed. Allen et al. (1994) evaluated the nested logistic model and two other developmental toxicity dose–response models (Rai and Van Ryzin, 1985; Kodell et al., 1991) using a database of 607 developmental endpoints from 141 separate studies. The researchers found that the nested logistic model generally gave the best fit of the dose–response data, compared to the two other models. The authors also found that the ratio of the NOAEL (determined from the proportion of affected fetuses) and the 95% confidence lower limit of the BMD5 (or BMDL5) was approximately one (median from all end-
57
points). This result shows that the BMDL5 for this model is approximately equal to the NOAEL for a developmental study. It also suggests the limit of detection for typical developmental tests is about a 5% incidence. EPA typically uses 10% as a benchmark response for BMD modeling (US EPA, 2000a), but given that the limit of detection for developmental tests is closer to 5% and the severity of this effect, a 5% response level if often used, as in this study. Dimethoate has been regulated by the US EPA using the inhibition of brain cholinesterase as the key endpoint. In its recent cumulative risk assessment for the organophosphates, US EPA used the dose that causes a 10% inhibition of brain cholinesterase as a point-ofdeparture for risk assessment, and developed a dose–response model to estimate the BMD10 (US EPA, 2000a). EPA found that a 10% response was generally the limit of detection for brain cholinesterase inhibition; and made a policy decision that a BMD10 is an appropriate point-of-departure for risk assessment. Because the BMDs for cholinesterase are based on a percentage inhibition compared to control animals, as opposed to the percentage of animals responding like the pup mortality, the asterisk (*) is used when identifying the cholinesterase BMDs to differentiate them from the different definition of a BMD used for pup mortality. While the benchmark response levels that were chosen for pup mortality and cholinesterase inhibition are different (5% for pup mortality and 10% for cholinesterase inhibition), these levels reflect the policy decisions made by EPA. Therefore, these are appropriate levels for a regulatory risk assessment. For the organophosphate (OP) cumulative assessment, US EPA developed an exponential dose–response curve to describe cholinesterase inhibition as follows (US EPA, 2000b): CheðdoseÞ ¼ Che0 ½P b þ ð1 P b Þ expðm DoseÞ;
ð2Þ
where Che(dose) is the cholinesterase level at a given dose level, Che0 is the cholinesterase level at a zero dose, Pb represents a limiting value of the minimum cholinesterase activity at high doses, and m is the slope estimate of the dose response curve. This model was found to provide an excellent fit to the cholinesterase data for most of the organophosphates. The EPA also developed a more complicated version of this model that accounted for a low-dose flat region (or shoulder) and provided generally higher BMDs for OPs that exhibited this behavior. However, this more complicated model was not found to provide better model fits for dimethoate, so it was not used in this analysis. When appropriate constant variance was assumed, but in some cases the assumption of constant variance was rejected and the variance was allowed to vary among groups. As formulated, Eq. (2) cannot be explicitly modeled in EPAÕs BMDS software. Therefore, a modified proce-
58
R. Reiss, D. Gaylor / Regulatory Toxicology and Pharmacology 43 (2005) 55–65
each dose group of each study and the mean percent pup mortality across the litters in each group. With all of the studies combined, there are a total of 220 litters. However, the litters were not culled at PND4 in the cross-fostering study, as they were in the other studies. Hence, the cross-fostering study data are only used up to PND4. After PND4, these data cannot be considered replicates of the other studies. Therefore, for the PND4– 11 period the meta-analysis includes 169 litters. In the DNT study, there was an elevated mortality compared to controls at 0.5 mg/kg/day (8.0% for PND1–4) and 3.0 mg/kg/day (18.7% for PND1–4). These values were heavily influenced by one total litter loss (TLL) at 0.5 mg/kg/day and three TLLs at 3.0 mg/kg/day. The mortality at 0.5 mg/kg/day was not statistically significant, and the mortality at 3.0 mg/kg/day was statistically significant. By contrast, the mortality in the DNT range-finder was not significantly elevated at 3.0 mg/ kg/day (3.8% for PND1–4 compared to 3.5% for the controls), and the mortality was low in the comparative cholinesterase study at 0.5 mg/kg/day (1.5% for PND1– 4) and 3.0 mg/kg/day (0.7%). Given the virtually identical designs of the studies, the reason for the differences in pup mortality across the studies is most reasonably explained as statistical variability. Except at the highest dose groups, most litters have no mortality. The percent mortality is significantly driven by a few litters, particularly when there is a TLL. For example, for the 0.5 mg/ kg/day dose group in the DNT study, 9 of 24 litters had one or more mortality, one litter had 17 of the 32 deaths (53%) (the TLL), and two litters had 22 of the 32 deaths (69%). After the cull at day 4, there was elevated mortality at the highest dose groups (3.0 and 6.0 mg/kg/day), but the mortality was lower than during PND1–4.
dure was used to estimate the BMDs. First, Pb and m were estimated from the dose response data by an iterative least squares procedure. Initial estimates of Pb were zero or negative for all cases. Since Pb cannot be negative, Pb was set equal to zero. Hence, Eq. (2) can be linearized by taking the natural logarithm of each side of the equation: CheðdoseÞ ln ¼ m Dose. ð3Þ Che0 With Eq. (2) linearized, the BMD10 and lower 95% confidence limit ðBMDL10 Þ were estimated using the BMDS program with a linear dose–response model.
3. Results 3.1. Pup mortality by oral gavage exposure As noted above, there are four recent developmental studies with gavage exposures for dimethoate that were conducted at the same laboratory during the general time period. Therefore, this dataset is ideally suited for a meta-analysis. In many cases, a meta-analysis involves a joint analysis of data from several studies to derive an overall mean response, weighting the individual studies by sample size and variability. However, given the similarity of the gavage data, the data were simply combined together to conduct the meta-analysis in this study. In other words, the individual litter data from the four studies were considered replicates and combined. Therefore, each of the individual studies is essentially weighted by their sample sizes for the metaanalysis. Table 1 summarizes the number of litters in
Table 1 Pup morality data by dose group for four dimethoate studies by gavage dosing Study
Dose group
Number of litters
Mean pup mortality for days 1–4 (%)a
Mean pup mortality for days 4–11 (%)a
Main DNT (Myers, 2001a)
0 0.1 0.5 3.0
24 23 24 24
2.7 2.4 8.0 18.7
1.0 1.1 2.9 10.2
DNT range-finder (Myers, 2001c)
0 2 3.0 6.0
10 9 9 10
3.5 0.7 3.8 27.2
0.0 0.0 0.9 11.1
Comparative cholinesterase (Myers, 2001b)
0 0.1 0.5 3.0
10 10 10 10
0.0 1.5 1.5 0.7
0.0 0.0 0.0 0.8
Cross-fostering (Myers, 2004)
0 6.0
25 22
2.6 8.2
n/ab n/ab
a
Average percentage mortality across the litters. Because there was no cull at day 4 of the cross-fostering study, unlike the other gavage studies, the mortality from PND4-11 cannot be compared with the other studies. b
R. Reiss, D. Gaylor / Regulatory Toxicology and Pharmacology 43 (2005) 55–65
59
mated by the method described by Rao and Scott (1992). The model fit was relatively poor (p = 0.029) and below the EPAÕs recommended goodness-of-fit threshold (p = 0.10) for dose–response models. The BMD5 and BMDL5 estimates were 1.5 and 0.96 mg/ kg/day, respectively. The poor fit of the data was mostly the result of the poor agreement at the high dose (6 mg/ kg/day), where the mortality is similar to the level at the next highest dose (3 mg/kg/day). The dose–response model does not provide for a threshold effect where the response levels off at high dosages, so the poor fit at the high dose for these data is not unexpected. It is possible that there are different biological processes affecting the mortality at the high dose. A common technique in dose–response modeling that is used to achieve better model fits is to rerun the model without the highest dose group (Gold and Zeiger, 1997). This appears to be particularly appropriate in this case given the poor fit at the high dose. Fig. 2 displays the model fit for PND1–4 using the data from the four gavage studies but dropping the highest dosage. After dropping the highest dose, there
Therefore, the mortality during PND1–4 is the driver for risk assessment. Table 2 summarizes the results of the BMD modeling for pup mortality, including the BMD5 and BMDL5. Indications of the model fit are provided including the AkaikeÕs information coefficient (AIC) and goodnessof-fit probability values from a chi-square test. Lower values of the AIC and higher values of the goodnessof-fit p values indicate better fits. Both the AIC and goodness-of-fit p values are provided by the BMDS software, and are routinely used to assess model fits. Models were fit with and without the litter-specific covariate, and the model with the lowest AIC was chosen, except as noted. For all models, the regression terms for the intralitter correlation were significant, so intralitter correlations were used in all of the models. Fig. 1 displays the model fit for PND1–4 using the data from the four gavage studies. The plot shows the mean response and 95% confidence interval in each dose group, the fit of the dose–response model, and the upper bound of the model. The confidence intervals were esti-
Table 2 Summary of dimethoate BMD model results for pup mortality for gavage dosing (recommended point-of-departure for risk assessment in bold) Time period
Days 1–4 Days 4–11
Dataset
AICa
Goodness-of-fit p value
Used litter-specific covariate
BMD5 (BMDL5) (mg/kg/day)
Combined gavage Combined gavage, without highest dosage Combined gavage, without highest dosage, and without total litter loss at 0.5 mg/kg/day Combined gavage Combined gavage without highest dose group
1113.92 790.70 767.91
0.029 0.49 0.43
No Yes Yes
1.5 (0.96) 1.1 (0.64) 1.7 (1.0)
0.0007 n/ac
No No
2.3 (1.2) 2.3 (1.1)
252.02b 243.71b
Note. Intralitter correlations were found to be important for all of the models, and were included in the final model runs. Low values of the AIC and higher values of the p value of the goodness-of-fit test indicate better model fits. a The AIC is a useful measure for assessing model fits between different models using the same dataset, but the AIC cannot be usefully compared between models with different datasets. b The model with the litter-specific covariate has a lower AIC, but a visual inspection of the fit showed that not including the litter-specific covariate gave a more realistic model fit. The BMD for the model with the litter-specific covariate was higher; therefore, this choice was conservative. c The BMDS program could not calculate a goodness-of-fit.
Fig. 1. BMD model fit for dimethoate pup mortality during days 1–4 using all of the gavage data.
Fig. 2. BMD model fit for dimethoate pup mortality during days 1–4 using all of the gavage data without the highest dose group.
60
R. Reiss, D. Gaylor / Regulatory Toxicology and Pharmacology 43 (2005) 55–65
are still 188 litters. The goodness-of-fit was considerably better (p=0.49) than when the highest dose was included (p = 0.029). The BMD5 and BMDL5 estimates were 1.1 and 0.64 mg/kg/day, respectively. These values are lower than the BMD estimates when including the highest dose. Given the better fit of the data and the need to be conservative in conducting a regulatory risk assessment, Fig. 2 is considered the best fit model, and the most appropriate point-of-departure for risk assessment is 0.64 mg/kg/day, the BMDL5 from this model. It is notable that, even after combining the gavage data for the meta-analysis, the single total litter loss in the 0.5 mg/kg/day group still has a very large impact on the BMD results. With the combined dataset, the average percentage mortality at 0.5 mg/kg/day with the total litter loss included is 6.1%, while the average percentage mortality without the total litter loss was 3.2%. Therefore, the single total litter loss at 0.5 mg/ kg/day doubles the percentage mortality in the 0.5 mg/ kg/day dose group. Given the impact of the single total litter loss at the 0.5 mg/kg/day dose group, it is appropriate to examine the model fit without the total litter loss. Fig. 3 displays the model fit for PND1–4 using the data from the four gavage studies but not including the highest dosage and not including the single total litter loss at 0.5 mg/kg/day in the DNT study. The goodness-of-fit was adequate (p = 0.43), and the model predicts the mortality at the 0.5 mg/kg/day dose group better than in Fig. 2. The BMD5 and BMDL5 estimates were 1.7 and 1.0 mg/kg/day, respectively. The values are significantly higher than the estimates for Fig. 2. The rationale for dropping the TLL is that it may be an outlier, although there is not a good outlier test for these non-normally distributed data. Therefore, while this analysis is interesting, the conservative approach for regulatory risk assessment is to use the BMD estimates with the model including the TLL.
Fig. 3. BMD model fit for dimethoate pup mortality during days 1–4 using all of the gavage data without the highest dose group and without total litter loss in 0.5 mg/kg/day dose group.
The pup mortality was greater for the period from PND1–4 compared to PND4–11, but the mortality during PND4–11 was still elevated compared to the controls. The results for PND1–4 are the most appropriate to use for risk assessment given the greater mortality during PND1–4. However, modeling was conducted for PND4–11 for comparison purposes. The dose–response models did not fit the data as well for PND4– 11 as for PND1–4 due to a weaker response during PND4–11. 3.2. Pup mortality by dietary exposure The pup mortality data for the 1992 2-generation study are summarized in Table 3, and the pup mortality data for the 2003 2-generation study are summarized in Table 4. For the 2-generation studies, the greatest amount of mortality occurred during the first four days, so only the mortality for days 1–4 is reported in the tables. Both of the 2-generation studies included two separate matings for each generation. There were no statistically significant increases in mortality for any dose group in any of the studies. There was a non-statistically significant elevated mortality in the first mating of the F1 generation in the 1992 study, but this result was caused by two total litter losses in the middle and high dose groups. BMD modeling cannot be conducted for these data given that most of the responses are below a 5% risk level, and there is not a clear dose–response pattern. However, it is clear that there is not a 5% risk at the highest dosage when all of the data are considered. Therefore, the only possible estimates of the BMD5s are >6.0 mg/kg/day for the 1992 study and >6.5 mg/kg/day for the 2003 study (i.e., the highest Table 3 Pup mortality for dimethoate by dose group for a 2-generation rat reproduction dietary study in 1992 Group
Dosage to dam (mg/kg/day)
Pregnant dams
F0 generation—first mating
0 0.09 1.3 6.0
26 27 24 25
6.6 3.9 2.1 2.2
F0 generation—second mating
0 0.09 1.3 6.0
25 26 25 20
2.7 3.3 2.9 3.4
F1 generation—first mating
0 0.09 1.3 6.0
23 17 17 15
3.2 3.0 10.7 11.1
F1 generation—second mating
0 0.09 1.3 6.0
16 16 14 12
1.7 1.2 1.5 1.4
a
Average percentage mortality across the litters.
Pup mortality for days 1–4a
R. Reiss, D. Gaylor / Regulatory Toxicology and Pharmacology 43 (2005) 55–65
61
Table 4 Pup mortality for dimethoate by dose group for a 2-generation rat reproduction dietary study in 2003
Table 5 Samples sizes for dimethoate brain cholinesterase measurements by dose group for gavage studies
Group
Dosage to dam (mg/kg/day)
Pregnant dams
Pup mortality for days 1–4a
Study
Subpopulation
Number of animals/ dose group
F1 generation—first mating
0 0.2 1 6.5
21 24 25 23
1.2 1.3 2.2 2.6
Comparative cholinesterase study
F1 generation—second mating
0 0.2 1 6.5
24 22 21 25
1.2 1.1 0.74 4.5
F2 generation—first mating
0 0.2 1 6.5
22 24 24 25
3.2 7.1 3.0 1.5
F2 generation—second mating
0 0.2 1 6.5
22 25 22 25
3.6 5.4 2.5 4.2
GD20 dams GD20 fetuses PND4 male offspring PND4 female offspring PND11 male offspring PND11 female offspring PND21 male offspring PND21 female offspring PND60 male offspring PND60 female offspring Adult males, Day 1 Adult females, Day 1 Adults males, Day 11 Adult females, Day 11 GD20 dams GD20 male fetuses GD20 female fetuses PND21 male offspring PND21 female offspring
8 8 14–19 13–16 8 8 8 8 8 8 8 8 8 8 5 5 5 15–20 14–20
a
Average percentage mortality across the litters.
doses tested). The best estimate of the BMD5 for pup mortality for the gavage studies was 1.1 mg/kg/day (a lower limit of the BMD can not be estimated for the dietary studies for comparison, so the central estimate of the BMDs were compared instead). Comparing the BMD5 for the gavage studies of 1.1 mg/kg/day, rats appear to be at least sixfold more sensitive for pup mortality after exposure to dimethoate when the dose is given by gavage compared to a dietary dose. 3.3. Cholinesterase Inhibition by oral gavage exposure Of the four dimethoate gavage studies that included pup mortality measurements, two of the studies, the comparative cholinesterase study (Myers, 2001b) and the DNT range-finding study (Myers, 2001c), included measurement of brain cholinesterase inhibition. The sample sizes are summarized in Table 5, including separate values for adults, fetuses, and pups at different time points of the study and different exposure regimens. For the offspring at PND4 and PND11, and adults at days 4 and 11, there are only measurements from the comparative cholinesterase study. Therefore, the BMDs were calculated with these data only. The PND4 offspring represent pups with exposure only from the milk of the dam and previous in utero exposure, as the pups are not being directly dosed at this time point. The PND11 pups were dosed once. For the GD20 dams and fetuses and the PND21 offspring, measurements were made in both the comparative cholinesterase and DNT range-finding studies. Since the study designs were identical, these represent replicate data. Therefore, these data were combined to estimate the BMDs. Furthermore, there were not significant differences in the
DNT range— finding study
cholinesterase measurements between the males and females at PND21. Therefore, the male and female data were combined to make a larger dataset. The PND21 offspring represent animals that were dosed 11 consecutive days (from PND11 to PND21). The estimated BMDs for brain cholinesterase are summarized in Table 6. The table shows the BMD10 , BMDL10 , and the p value of the goodness of fit for the dose–response model. The BMDs did not change dramatically when the range-finding data were added to the comparative cholinesterase dataset. Some of the goodness-of-fit p values in Table 6 are lower than EPAÕs acceptable level of 0.10. However, a review of the dose– response curves show that the fit of the exponential model is excellent in almost all cases. Figs. 4–6 show example model fits to the data for GD20 dams (Fig. 4), GD20 fetuses (Fig. 5), and adults on days 11 (Fig. 6). The BMD10s ranged from 0.29 mg/kg/day (GD20 dams) and 0.64 mg/kg/day (PND21 offspring) to 2.6 mg/kg/ day (adults males, day 1) and 4.0 mg/kg/day (PND4 females). The plots show the mean response, the 95% confidence interval on the mean response, and the model fit. 3.4. Cholinesterase inhibition by dietary exposure There are several dietary studies which measured cholinesterase inhibition for dimethoate, including the 2-generation rat reproduction study in 1992, the 2-generation reproduction study in 2003, and a 28-day dietary study. Table 7 summarizes the BMD estimates for the dietary studies. The BMDs are all for adult animals that were sacrificed at the end of the measurements periods in the studies. Fig. 7 shows the dose–response fit for the males in the 28-day dietary study.
62
R. Reiss, D. Gaylor / Regulatory Toxicology and Pharmacology 43 (2005) 55–65
Table 6 Summary of dimethoate BMD estimates for brain cholinesterase inhibition for gavage studies Subpopulation
Dataset
Length of exposure (days)
BMD10 (mg/kg/day)
BMDL10 Goodness-of(mg/kg/day) fit p valueb
GD20 dams
Comparative cholinesterase and DNT range-find Comparative cholinesterase and DNT range-find Comparative cholinesterase Comparative cholinesterase Comparative cholinesterase Comparative cholinesterase Comparative cholinesterase and DNT range-find Comparative cholinesterase Comparative cholinesterase Comparative cholinesterase Comparative cholinesterase
15
0.29
0.19
0.99
15 in utero
1.0
0.65
0.72
16 in utero + 4 via milk 16 in utero and 4 via milk 1 direct dose + 16 in utero and 10 via milk 1 direct dose + 16 in utero and 10 via milk 11 direct doses + 16 in utero and 10 via milk
3.8a 4.0a 1.8 1.5 0.64
2.2a 2.4a 1.5 1.2 0.58
0.025 0.03 0.41 0.02 0.27
1 1 11 11
2.6 2.1 0.47 0.36
2.0 1.7 0.36 0.32
GD20 fetuses PND4 males PND4 females PND11 males PND11 females PND21 offspring Adults males, Day 1 Adult females, Day 1 Adults males, Day 11 Adult females, Day 11
0.83 0.19 <0.01 0.09
a
Uncertain because there was less than 10% inhibition at the highest dosage. Some of the p values below 5% were for dose–response curves that appeared to fit the data average quite adequately; therefore, it is not clear that these p values are representative of the data fit. The lack of fit may be due to the large variability within dose groups. b
Fig. 4. BMD model fit for dimethoate brain cholinesterase for GD20 dams for gavage dosing.
Fig. 6. BMD model fit for dimethoate brain cholinesterase for adult females on day 11 for gavage dosing.
4. Discussion 4.1. Most sensitive subpopulation for cholinesterase inhibition by gavage
Fig. 5. BMD model fit for dimethoate brain cholinesterase for GD20 fetuses for gavage dosing.
Before comparing the pup mortality and cholinesterase BMDs to determine the most sensitive endpoint for risk assessment, the cholinesterase BMDs for gavage were compared to determine the most sensitive subpopulation for cholinesterase inhibition. From a cursory inspection, the dams have a lower BMDL10 (0.19 mg/ kg/day) compared to the adult females (0.32) (or nonpregnant females). However, the 90th percentile error bounds overlap so there is not a statistical difference. When only using the values from the comparative cholinesterase study, the comparison is much closer (BMD10 of 0.33 mg/kg/day for dams and 0.36 mg/kg/
R. Reiss, D. Gaylor / Regulatory Toxicology and Pharmacology 43 (2005) 55–65
63
Table 7 Summary of dimethoate BMD estimates for brain cholinesterase inhibition in the dietary studies Study
Subpopulation
BMD10 (mg/kg/day)
BMDL10 (mg/kg/day)
1992 2-generation rat reproduction study
F0 males F1 males F0 females F1 females F0 males F1 males F0 females F1 females Males Females
0.62 0.58 0.62 0.49 0.96 0.77 0.56 0.62 1.2 0.95
0.59 0.50 0.51 0.46 0.77 0.67 0.50 0.58 1.1 0.70
2003 2-generation rat reproduction study
28-day study
adults are lower than the BMD values for PND 21 juvenile offspring indicating that the adults are more sensitive to cholinesterase inhibition than are these offspring. This result is consistent with an analysis by Sheets (2000) of multi-generational reproduction studies with five organophosphate insecticides. To summarize, the relative sensitivity of these subpopulations to repeated gavage exposure to dimethoate is: dams = day 11 adults > fetuses > offspring on PND 21 > nursing offspring on PND 4 (most sensitive to least sensitive). 4.2. Most sensitive endpoint for gavage exposure Fig. 7. BMD model fit for dimethoate brain cholinesterase for males in 28-day dietary study.
day for adult females). When the range-finder study data were added for the dams, the BMDs were reduced (there was no range-finder data for non-pregnant females). Therefore, the pregnant dams were not more sensitive than non-pregnant females. The BMD values for dams and day 11 adults are all lower than the BMD values for fetuses. This point is further underscored by comparing the dose–response curves, which shows that the cholinesterase inhibition for the fetuses has a lower plateau than for the adults. For the fetuses, there is only a 39% inhibition at the highest dosage of 6 mg/kg/day, whereas there is 88% inhibition for the dams. As discussed above, the cholinesterase inhibition for pregnant dams and adult females on day 11 was very similar. The cholinesterase inhibition for adult males on day 11 was lower than for adult females on day 11. Therefore, dams and adult males and females (after consecutive dosing) were more sensitive to cholinesterase inhibition than the fetuses (assuming the dose to the dam as the dose to the fetus). The BMD values for dams and day 11 adults are lower than the BMD for PND 4 offspring indicating that nursing infants are less sensitive to cholinesterase inhibition than are the adults. The BMD values for dams and day 11
The BMD estimates provide an ideal method to compare the relative sensitivity of the animals to cholinesterase inhibition and pup mortality from gavage doses of dimethoate. The comparison is made on the basis of the appropriate points-of-departure for risk assessment. Most BMD analyses are based on a 10% incidence. However, as discussed above, for developmental endpoints, the appropriate point-of-departure is the lower limit estimate of the 5% incidence, or a BMDL5. For cholinesterase inhibition, EPA has determined that a 10% inhibition of brain cholinesterase is the most biologically relevant measurement. Therefore, the BMDL10 is the appropriate point-of-departure. Fig. 8 provides a graphical comparison of the estimated BMDs for pup mortality and cholinesterase inhibition. The graph shows both the central tendency estimates (BMD5 and BMD10 ) and the lower and upper limits of the estimates. The PND4 offspring are not included on the graph because the BMD estimates are so much larger than the other estimates. The pup mortality BMD5 is relatively similar, within the error bounds, to the cholinesterase inhibition BMD10 for fetuses and PND11 pups. For the PND21 pups, adult males and females on day 11, and the dams, the BMD10 estimates were lower than the BMD5 for pup mortality. There was a small overlap on the statistical bounds of the BMD10 for cholinesterase for PND21 pups and the BMD5 for pup mortality. There was no
64
R. Reiss, D. Gaylor / Regulatory Toxicology and Pharmacology 43 (2005) 55–65
Fig. 8. Comparison of dimethoate gavage bmds for pup mortality and brain cholinesterase inhibition.
overlap of the statistical bounds of BMD10 for cholinesterase for adult males and females and the dams, and the BMD5 for pup mortality. These results indicate that the most sensitive endpoint for dimethoate is cholinesterase inhibition for pregnant dams and adult males and females. Using central estimates, the ratio of the BMD5 for pup mortality and the BMD10 for cholinesterase inhibition for the dams is 3.8. Comparing the lower limits, the ratio is 3.4. Therefore, the risk assessment points-of-departure for cholinesterase inhibition is more than threefold lower than for pup mortality. This demonstrates that the use of points-of-departure for cholinesterase inhibition is also likely to be protective of pup mortality. 4.3. Most sensitive endpoint for dietary exposure Dietary exposure is more relevant than gavage exposures for many human risk assessment scenarios. Therefore, it is important to analyze the dietary toxicity data separately. First, the dietary BMD estimates can be compared with BMD estimates for adults in the comparative cholinesterase study. The day 11 adult values for gavage provide the best basis for comparison, as these more closely compare with the dietary studies which included multiple exposures. For males, the average BMDL10 for the four 2-generation reproduction dietary measurements was 0.63 mg/kg/day, compared to 0.36 mg/kg/day for the gavage study, suggesting about a twofold greater sensitivity for gavage exposures (the rats in the dietary study were exposed longer). For females, the average BMDL10 for the four 2-generation reproduction dietary measurements was 0.51 mg/kg/day, compared to 0.32 mg/kg/ day for the gavage study, suggesting about a 1.5-fold greater sensitivity for gavage exposures. In the comparative cholinesterase study, the females were exposed for 11 days, while the exposures in the dietary studies ranged from 17 to 44 weeks. Thus, even with the shorter duration
exposures, the rats were more sensitive to cholinesterase inhibition by gavage. The closest comparison in regard to duration is with the 28-day dietary study. For males, the BMDL10 value for dietary exposure was threefold higher than the gavage value. For females, the BMDL10 value for dietary exposure was more than twofold higher than the gavage value. These comparisons show that the animals were more sensitive to gavage exposures in regard to cholinesterase inhibition. One uncertainty in this analysis is that the 1992 study used a different strain of rats than the 2003 study. For pup mortality, there was less than a 5% risk at the highest dosages for the 2-generation reproduction studies. Therefore, the BMD5s were defined as greater than the highest dosage (>6 mg/kg/day for the 1992 study, and >6.5 mg/kg/day for the 2003 study). The average BMDL10s for cholinesterase for the 1992 study was 0.52 mg/kg/day, and 0.63 mg/kg/day for the 2003 study. This comparison suggests that cholinesterase inhibition is at least 10-fold more sensitive an endpoint than pup mortality for dietary exposures.
5. Conclusion This article presents a meta-analysis of several rat toxicity studies for dimethoate. Compared to the most sensitive subpopulation for brain cholinesterase inhibition (dams) for gavage exposure, the points-of-departure for risk assessment for cholinesterase inhibition is more than threefold lower than for pup mortality. Therefore, the use of cholinesterase inhibition as an endpoint for risk assessment is likely to be protective of pup mortality. A comparison of brain cholinesterase inhibition and pup mortality in the dietary studies shows that cholinesterase inhibition is about a 10-fold more sensitive endpoint than pup mortality. By route of exposure, the animals were about 2–3-fold more sensitive for brain cholinesterase inhibition by gavage than by dietary exposure. For pup mortality, the animals were at least sixfold more sensitive by gavage exposure compared to dietary exposure.
Acknowledgments This work was sponsored by Cheminova A/S. The authors wish to thank Paul Whatling, Diane Allemang, Dorrit Søndergaard, Judith Hauswirth, Don OÕShaughnessy, and Abby Li for their helpful comments.
References Allen, B.C., Kavlock, R.J., Kimmel, C.A., Faustman, E.M., 1994. Dose–response assessment for developmental toxicity: III. Statistical models. Fundamental and Applied Toxicology 23, 496–509.
R. Reiss, D. Gaylor / Regulatory Toxicology and Pharmacology 43 (2005) 55–65 Brooker, A., Stubbs, A., 1992. Dimethoate: dietary range finding study in mature male and female rate and their juvenile offspring. Report by Huntingdon Research Center, Ltd conducted under International GLP. Gold, L.S., Zeiger, E., 1997. Handbook of Carcinogenic Potency and Genotoxicity Databases. CRC Press, Boca Raton, FL, p. 559. Kasper, U., Kaufmann, W., Deckardt, K., van Ravenzwaay, B., 2004. Dimethoate range-finding study in Wistar rats, administration via the diet over 4 weeks. Study conducted at BASF Aktiengesellchaft under international GLP. Kodell, R., Howe, R., Chen, J., Gaylor, D., 1991. Mathematical modeling of reproductive and developmental toxic effects for quantitative risk assessment. Risk Analysis 11, 583–590. Kupper, L., Portier, C., Hogan, M., Yamamoto, E., 1986. The impact of litter effects on dose–response modeling in teratology. Biometrics 42, 85–98. Mellert, W., Hellwing, J., Gembardt, C., Deckardt, K. Van Ravenzwaay, B. 2003. Dimethoate two-generation reproduction toxicity study in Wistar rats— adminstration in the diet. Study conducted at BASF Aktiengesellchaft under international GLP. Mileson, B.E., Chambers, J.E., Chen, W.L., et al., 1998. Common mechanism of toxicity: a case study of organophosphorus pesticides. Toxicological Sciences 41, 8–20. Myers, D.P., 2001a. Dimethoate developmental neurotoxicity study in the CD rat by oral gavage administration. Study conducted by Huntingdon Life Sciences, Ltd under international GLP. Myers, D.P., 2001b. Dimethoate effects on cholinesterase in the CD rat (adult and juvenile) by oral gavage administration. Study conducted by Huntingdon Life Sciences, Ltd under international GLP. Myers, D.P., 2001c. Dimethoate dose finding study in CD rats by oral gavage administration preliminary to developmental neurotoxicity
65
study. Study conducted by Huntingdon Life Sciences, Ltd under international GLP. Myers, D.P., 2004. Dimethoate cross fostering study in CD rats. Study conducted by Huntingdon Life Sciences, Ltd under international GLP. OECD Test Guideline No. 416. Two Generation Reproduction Toxicity. Dated May 26, 1983. Paris, Organisation for Economic Cooperation and Development. Rai, K., Van Ryzin, J., 1985. A dose–response model for teratological experiments involving quantal responses. Biometrics 41, 1–9. Rao, J.N.K., Scott, A.J., 1992. A simple method for the analysis of clustered binary data. Biometrics 48, 577–585. Sheets, L.P., 2000. A consideration of age-dependent differences in susceptibility to organophosphorous and pyrethroid insecticides. Neurotoxicology 1–2, 57–63. US EPA, 1996. Standard operating procedure for measuring cholinesterase inhibition in laboratory rats and dogs exposed to nonreversible cholinesterase inhibitors. Federal Register, April 26, 1996, vol. 61, p. 18593. US EPA, 1998a. Health effects test guidelines OPPTS 870.6300 developmental neurotoxicity study. Report No. EPA-712-C-98239. US EPA, 1998b. Health effects test guidelines OPPTS 870.3800 reproduction and fertility effects. Report No. EPA-712-C-98-208. US EPA, 2000a. Benchmark dose technical guidance document. EPA Report No. EPA/630/R-00/001. US EPA, 2000b. Revised OP Cumulative Risk Assessment. Prepared by the Office of Pesticide Programs: