Cancer Risk Estimates for Tetrachloroethylene
The Environmental Protection Agency (EPA) draft Integrated Risk Information System (IRIS) assessment of tetrachloroethylene provides the agency’s assessment of the potential human health effects of exposure to the chemical. For cancer, EPA provides a characterization of the weight of evidence of human carcinogenicity and quantitative estimates of inhalation unit risks and oral slope factors. A unit risk is the upper bound of excess lifetime cancer risk estimated to result from continuous exposure to an agent at a concentration of 1 μg/m3 in air. An oral slope factor is the upper bound, approximating a 95% confidence limit, of the increased cancer risk from lifetime exposure to an agent; it is usually expressed as a proportion (of a population) affected per milligram per kilogram per day. For tetrachloroethylene, EPA proposes a range of inhalation unit risks of 2 × 10-6 to 2 × 10-5 per microgram per cubic meter and a range of oral slope factors of 1 × 10-2 to 1 × 10-1 per milligram per kilogram per day. These ranges reflect the application of three physiologically-based pharmacokinetic (PBPK) models to the same data. This chapter discusses how those cancer risk estimates were determined by EPA.
EPA asked for an evaluation of whether conclusions it has drawn in the draft IRIS assessment are consistent with its cancer guidelines (EPA 2005a), specifically with regard to its characterization that tetrachloroethylene is “likely to be a human carcinogen by all routes of exposure.” Box 11-1 presents EPA’s guidelines for determining such a classification.
The committee considered those guidelines, and guidelines for the other descriptors, and concluded that EPA has documented that its conclusion has been drawn from the results of bioassays that found increased incidences of hepatocelluar tumors, hemangiosarcomas, mononuclear-cell leukemia (MCL),
and renal tumors in laboratory animals and to a lesser extent from epidemiologic evidence. EPA’s decision to characterize tetrachloroethylene as likely to be a human carcinogen as opposed to “carcinogenic to humans” appropriately reflects that there could be deficiencies or potential inaccuracies in interpretation of the data. Some of those possible deficiencies and inaccuracies are discussed below for each of the data sets.
EPA Cancer Guidance for Concluding a Chemical Is Likely to Be Carcinogenic to Humans (EPA 2005)
This descriptor is appropriate when the weight of evidence is adequate to demonstrate carcinogenic potential to humans but does not reach the weight of evidence for the descriptor “Carcinogenic to Humans.” Adequate evidence consistent with this descriptor covers a broad spectrum. As stated previously, the use of the term “likely” as a weight of evidence descriptor does not correspond to a quantifiable probability. The examples below are meant to represent the broad range of data combinations that are covered by this descriptor; they are illustrative and provide neither a checklist nor a limitation for the data that might support use of this descriptor. Moreover, additional information, e.g., on mode of action, might change the choice of descriptor for the illustrated examples. Supporting data for this descriptor may include:
Mononuclear Cell Leukemia
An increased incidence of MCL in F344 rats has been reported in two bioassays. The biological significance of these increases was debated by the committee because increases were observed only in one strain of rat, which is known to have a high background incidence of MCL. Control data on F344 rats indicate background rates of MCL ranging from 7.9-52.5% in males and 2.1-24.2% in females (Thomas et al. 2007), which make it difficult to interpret the biological significance of increases observed in two studies from different laboratories (NTP 1986; JISA 1993) because of the lack of information on mode of action. Statistical methods, such as survival data analysis, which incorporate data from multiple dose groups simultaneously for dose-response analysis rather than pair-wise comparison should be explored to aid in interpretation. For example, as noted in Chapter 8, Thomas et al. (2007) have made a case that using life-table analysis to examine the MCL data provide an improved approach for interpreting the significance of a dose-response for a possible carcinogenic effect. They judged that there was a positive association between tetrachloroethylene and MCL in the NTP study when such criteria were applied, but recommended a weight-of-evidence evaluation be performed before drawing conclusions. The committee observed that the data showed inconsistency in statistical significance between sexes and uncertainty about the shape of the dose-response curve, especially in the lower range of the NTP study. There is some support from epidemiologic studies which suggest an association between lymphoma and tetrachloroethylene, but the data were relatively weak and inconsistent. A difficulty with interpreting the findings is differences in opinion about the biological concordance between MCL and human lymphohematopoietic cancer. Some members judged that similarities between human natural killer large granular lymphocyte leukemia and rat MCL, as well as mechanistic studies the committee recommended be added to EPA’s assessment, are adequate to assume human relevance, whereas other believe more research is needed to establish the relevance. In addition, there was little information on a mode of action for how tetrachloroethylene increases MCL, so it was not possible to distinguish whether exposure to tetrachloroethylene results in initiation of new tumors or enhances the ongoing expansion or promotion of existing tumors.
Evidence for a statistically significant increase in hepatic tumors was observed in male and female mice after oral or inhalation exposure. Like MCL, the biological significance of these increases was debated by the committee because B6C3F1 mice have a high background incidence of hepatic cancer (about 20%). However, the findings were reproduced among several studies and conducted in different laboratories and showed a dose-response relationship. There is also fairly substantial information for characterizing potential modes of action for hepatic tumor formation relative to the data available on MCL and renal cancer.
(See Chapters 6 and 7 and the section below on Mode of Action Analysis.) While the committee recommended that EPA revise its presentation of the mode of action evidence for tetrachloroethylene-related hepatic cancer to clarify its position, the majority of the members agreed with EPA that the mode of action is complex and remains to be established. These members also agreed there was insufficient evidence to rule out human relevance. One member objected to these conclusions and the committee’s support of using liver cancer to quantify risk. He concluded that in the absence of evidence of other contributing modes of action, the evidence is sufficient to conclude that the mode of action in mice is predominantly through activation of the peroxisome proliferator activated receptor α, a mode of action he considered to be of little relevance to humans. His arguments are presented in a dissenting statement in Appendix B of the report.
Tetrachloroethylene caused a low rate of induction of renal tumors in rats. Although the increases were not statistically significant when compared with concurrent controls, EPA has used historical controls to calculate the chances of two of these rare carcinomas to occur by chance to be less than 0.001. Furthermore, a dose-response trend was shown against the low background and the tumors in the treated rats were malignant whereas the tumors in the controls were not. EPA provided a strong evaluation of the potential modes of action evaluation for tetrachloroethylene-induced kidney cancer. The committee agrees with the agency that mode of action of tetrachloroethylene tumorigenesis is not understood, but that a mutagenic mode of action cannot be ruled out. Thus, kidney tumors observed in tetrachloroethylene-treated rats was considered relevant to humans, even thought the epidemiologic evidence of an association is weak (see Chapter 7).
SELECTION OF TUMOR TYPE FOR QUANTITATIVE ASSESSMENT
The committee was unable to reach consensus on the selection of the critical cancer end point. The majority of members judged that the uncertainties associated with MCL (particularly the high background incidence, uncertainty about the dose response, and poor understanding of mode of action) were too great to support using the data over that of hepatic or renal cancer for determining quantitative estimates of risk. These members judged that the use of the MCL data could only be justified if it is EPA’s policy to choose the most conservative unit risk when considering a range of options, but that such justification should be distinguished as a policy decision and not a scientific one. They believe that a more scientifically defensible approach would be to use the data set with the least uncertainty, rather than the data set that yields the most conservative estimate of risk. In their estimation, the hepatic cancer data would have
the least uncertainty associated with it, followed by kidney cancer and MCL. The comparison of risk estimates presented in the draft IRIS assessment indicates that a unit risk based on hepatic cancer would be approximately eight-fold less than the estimate based on MCL. A unit risk based on kidney cancer would be five-fold less.
Other members judged that the MCL data should be used for cancer risk estimation. Their opinions were based on the observation that reproducible, statistically significant increases in MCL in male and female rats above the background incidence of MCL were found, and that MCL was the cancer end point with the highest magnitude of response. These members believe that use of the most sensitive response to quantify cancer risk decreases the uncertainty associated with potential differences in metabolism and susceptibility to tetrachloroethylene across exposed populations. They concluded that additional statistical analyses of the dose-response data and the addition of supporting mechanistic information identified by the committee would strengthen existing support for MCL in the draft assessment.
EPA included mode of action (MOA) analyses for cancer in its draft IRIS assessment (Section 4.4.4, pp. 4-16 to 4-35, for the liver and Section 4.5.4, pp. 4-42 to 4-51, for the kidney). EPA’s cancer guidelines present a framework for judging whether available data support a hypothesized MOA of an agent. The application of the framework is best demonstrated in EPA’s MOA analysis for renal cancer (see Chapter 7). For hepatic cancer, the committee found that the assessment relies too heavily on generic information on peroxisome proliferators and needs to be focused on tetrachloroethylene and its metabolites.
Chapters 6 and 7 provide more specific guidance on how to improve the presentation of the MOA evidence on tetrachloroethylene-induced hepatic and renal cancer. In general, the committee observes that discussion of MOA1 analy-
ses would be improved by including the proposed temporal sequence of hypothesized tetrachloroethylene-associated key events (possibly as a diagram). Transparency would be improved by presenting the details of experimental results in tabular form, including the chemical (tetrachloroethylene or specific metabolite), species, strain, sex, dose, route and duration of exposure, and experimental outcome or end point. That would allow the reader to follow the evaluation of the relative potency of tetrachloroethylene, or its metabolites, in inducing both key events and tumors and to consider species and strain differences with respect to the events and tumor formation. Other data relevant to the evaluation of hypothesized MOAs should be included. The advantage of such a presentation is that it makes explicit the consideration of the timeline of key events in the context of dose, concordance or lack of concordance between early and late events, and the relative contribution of chemical-specific data compared with generic information on categories of chemicals. That should be done for each hypothesized MOA. Even if the data are insufficient to support hypothesized MOAs, the exercise can be used to identify critical data gaps and to inform the direction of future research.
A general difficulty that the committee encountered in reviewing the MOA analyses is the presentation of conclusions without sufficient supporting evidence or reference to prior discussions of the evidence. Much of the experimental evidence was evaluated in other sections of the draft and presumably formed the basis of statements in the MOA considerations. To make the analyses clear, some reiteration of the evidence is needed in discussions of strength, consistency, and specificity of association of the tumor response with key events; dose-response relationships; temporal associations; and biologic plausibility. Coherence of the database is necessary for characterizing the evidence supporting a MOA. The analysis needs to take into account concordance of dose-response relationships between hypothesized key events and end events and to recognize that key events are necessary but might not be sufficient (in their own right) to induce the adverse outcome.
AGE-DEPENDENT ADJUSTMENT FACTOR
Section 126.96.36.199 of the EPA draft (p. 6-24) states that “age adjustment factors for early life exposures as discussed in the Supplemental Guidance for Assessing Susceptibility for Early-Life Exposure to Carcinogens (U.S. EPA 2005b) are not recommended because little evidence exists to indicate that tetrachloroethylene or its oxidative metabolites directly damage DNA, information about genotoxicity of gluthathione (GSH) metabolites in cell assays other than Salmonella or in in vitro experiments are lacking, and the MOA for tetrachloroethylene has not been established.” In addition, the assessment reasons that “although a mutagenic MOA would indicate increased early-life susceptibility, there are no data exploring whether there is differential sensitivity to tetrachloroethylene carcinogenicity across life stages.” The committee’s recommendations for amending sections on
genotoxicity and MOA considerations would also strengthen the arguments made by EPA with regard to the need for age-adjustment and low-dose extrapolations. The committee concluded that several metabolites of tetrachloroethylene are clearly genotoxic: S-(1,2,2-trichlorovinyl) glutathione (TCVG), S-(1,2,2-trichlorovinyl)-L-cysteine, N-acetyl-S-(1,2,2-trichlorovinyl)-L-cysteine (N-Ac-TCVC), tetrachloroethylene oxide, dichloro-acetic acid (DCA), and chloral hydrate (if it is formed). However, it is questionable whether those metabolites play an important role in the MOA of tetrachloroethylene carcinogenesis in view of their presence in tetrachloroethylene-exposed animals at low or undetectable concentrations and in the absence of convincing evidence of mutagenic and tumor-initiating activity of tetrachloroethylene in vivo. In addition, the committee supports EPA’s conclusion that the MOA of tetrachloroethylene is unclear and probably complex. Thus, although the committee agrees that age-dependent adjustment factors for cancer risk should not be applied, given uncertainties with regard to the overall MOA and the biologic relevance of the data on genotoxicity of metabolites of tetrachloroethylene, the rationale for this conclusion should be revisited.
For cancer risk assessment, EPA relied on the default option of low-dose linear extrapolation to estimate inhalation unit risks and oral slope factors for tetrachloroethylene. EPA describes low-dose linear extrapolation in detail (in Section 5.4.4 of the draft). It entails three steps. First, a dose-response model, typically a mathematical function in the absence of MOA information, that appropriately fits observed data within the experimental data range must be identified. Second, a point of departure (POD) (a benchmark dose or benchmark concentration) along the fitted dose-response model is determined; it corresponds to an exposure that typically induces about 5-10% extra risk above the control’s response rate. Then the associated extra cancer risk is divided by the POD to yield a unit risk or a slope factor.
In the draft IRIS assessment, EPA illustrates low-dose extrapolation with six datasets, hepatocellular adenoma or carcinoma in male and female mice (JISA 1993), hemangiosarcoma in male mice (JISA 1993), MCL in male and female rats (JISA 1993), and renal tumors in male rats (NTP 1986). EPA considered multistage models as well as multistage Weibull models for doseresponse modeling in conjunction with the dose metric of total metabolism and administered concentration but presented results only of multistage models. It justified the use of the multistage model (p. 5-50) on the basis that MOA information is lacking and that the model has “some parallelism to the multistage carcinogenic process and it fits a broad array of dose-response patterns. Occasionally the multistage model does not fit the available data, in which case an alternate model should be considered.” In the case of hepatocellular adenoma and carcinoma in male mice, hemangiosarcoma in male mice, and MCL in female rats, the multistage model does not fit the data at lower doses, as acknowl-
edged by EPA (Figures 5-8a, 5-10a, and 5-12a). EPA did not explain the possible underlying reasons for low-dose nonlinearity and potential adjustment. EPA considered those poor-fit models acceptable solely on the grounds that statistical tests for goodness of fit were not significant (p > 0.10). The committee notes that the lack of significance of goodness-of-fit tests can be the result of a small number of animals in each dose group. For example, by doubling the number of animals per dose group while keeping the incidences of tumor the same as in the original dataset of hepatocellular adenoma and carcinoma in male mice (JISA 1993), we can fit the same model (Table 5-11) to the “larger” experiment. The goodness-of-fit test would reach a p value of 0.04, which suggests a poor fit. Alternatively, if we were to fit a multistage model with a (polynomial) degree of 2 to the original data, the goodness-of-fit test would have a p value of 0.25, which would suggest a better fit than the model chosen by EPA (Table 5-11). Thus, using the goodness-of-fit test to justify a selection of a dose-response model can be misleading. Furthermore, contrary to the statement that “dose-response modeling of the candidate data sets presented no particular difficulties” (EPA 2008, p. 5-69), the benchmark dose software automatically fixed some parameters to zero to obtain convergence in model fitting. For example, in the case of hepatocellular adenoma and carcinoma in male mice, the second order coefficient (q2=0) is fixed but the third order coefficient (term q3) is not. The criteria under which EPA selected parameters and fixed them was unclear. Also, the parameter q0 reported in Tables 5-10 and 5-11 should be reported as 1-exp(q0) to be consistent with the specification of multistage model in section 188.8.131.52. The committee also notes that the polynomial order used in the multistage dose-response models is limited by the number of dose groups in each experiment; only lower-order multistage models can be fitted, and they are forced to be nearly linear in the low dose range. Therefore, the similarity between the slope of the models and the unit risk taken from the models reflects more on the nearly linear model imposed on the data than the true shape of the dose-response curve. The questionable fitting of a multistage model to some candidate datasets and the insufficient consideration of alternative models in these situations appear to be inconsistent with EPA’s cancer-risk guidelines and can contribute to underestimation of the overall uncertainties.
Once a dose-response model was chosen, EPA carried out the estimation of benchmark concentration with its lower confidence limit (BMCL) at a 10% extra risk (5% in one case). The BMCL is used as a POD for unit risk or slope factor. EPA’s choice, estimation, and presentation of PODs are adequate and clear.
EPA adopted linear low-dose extrapolation, the default option, with several justifications. First, MOA information is insufficient, and support for dynamic models unavailable. Therefore, nonlinear mechanistic models are unavailable for dose-response modeling. Second, because mathematical models are subject to uncertainties for low-dose extrapolation beyond the experimental dose range, linear extrapolation is more conservative than all sublinear (curvilinear) dose-response models. When individual thresholds in the human population are
plausible, wide variation in threshold values implies a curvilinear shape of the dose-response relationship on the average. Thus, linear extrapolation protects susceptible subpopulations (NRC 2009). Third, a few of the candidate datasets, especially EPA’s preferred male-rat MCL data, exhibit a linear pattern of dose-response relationships. Whereas those arguments are consistent with EPA’s Guidelines for Carcinogen Risk Assessment, there is evidence in the candidate datasets that the underlying dose-response relationship can be even supralinear (for example, in female-rat MCL). When that is the case, low-dose linear extrapolation is not conservative. The full range of variation and uncertainty in relation to model choice is not presented, in part because EPA did not consider the possibility of other forms of nonlinear dose-response models, including supralinear, for all candidate datasets.
PHYSIOLOGICALLY BASED PHARMACOKINETIC MODELS, DOSE METRICS, AND INTERSPECIES SCALING
The draft IRIS assessment appears to do a thorough job of reviewing the pertinent scientific literature on the toxicokinetics of tetrachloroethylene. EPA considered several independent efforts to develop physiologically based pharmacokinetic (PBPK) models for tetrachloroethylene and used them to estimate human equivalent doses in terms of environmental exposure and to perform route-toroute extrapolation. In the sections below, the committee reviews EPA’s decisions about what PBPK models to use, its choice of dose metrics, and approaches to species extrapolation.
The committee reviewed the original papers describing the selected PBPK models and supporting studies, which in some cases provided the experimental data used to validate model predictions. Evaluation of dose metrics was based on two primary criteria: the ability of the PBPK models to provide discrete estimates of a metric (such as peak blood levels or AUC of the parent chemical or metabolite in blood or target tissue) and the relevance of the parent compound or metabolites to the toxic end point. For cancer, the available evidence suggested that various tetrachloroethylene metabolites were involved or responsible, depending on the end point.
Physiologically Based Pharmacokinetic Modeling Approaches
There have been an unusually large number of independent efforts to develop PBPK models for tetrachloroethylene. EPA is to be commended for its willingness to use the PBPK modeling approach and to explore or test the various published PBPK models for tetrachloroethylene in its risk assessment. EPA used three PBPK models (Rao and Brown 1993; Reitz et al. 1996; Bois et al. 1996). However, there is a notable lack of critical evaluation of the models. Because the most important differences between the models is in prediction of tetrachloroethylene metabolism, there should be more discussion of the pros and
cons of using a population-modeling approach as in the Bois et al. (1996) study vs the other models, which rely more directly on animal in vitro and in vivo data. In particular, there seems to be a divergence between the two approaches particularly in estimating the fraction metabolized after smaller tetrachloroethylene exposures. For example, the recent paper by Chiu and Bois (2006) suggests that much higher fractions (23% of the dose) of tetrachloroethylene are metabolized in humans after low exposure (less than 1 ppm).
Reading the descriptions of previous PBPK modeling efforts makes it clear that it would have been preferable for EPA to pursue development of a “harmonized” PBPK model (as was done for trichloroethylene), which synthesized important aspects of the various models (the use of multiple exposure routes and inclusion of all relevant tissue compartments) into a single model. In connection with this recommendation, it is important to recognize that most PBPK models of tetrachloroethylene (and trichloroethylene) are highly derivative of the PBPK model for methylene chloride published by Andersen et al. (1987). The differences between the tretrachloroethylene models are associated with inclusion or exclusion of routes of exposure and the use of experimental data to select parameters for models and validate model predictions. The approach pursued by EPA, using three PBPK models, is a reasonable alternative for the tetrachloroethylene risk assessment for which the goal is to estimate tetrachloroethylene dosimetry related to inhalation exposure. The population pharmacokinetic modeling approach used in the Bois et al. model empirically estimates metabolism parameter values to provide an adequate fit of observed tetrachloroethylene exposure data. Initial estimates (prior distributions) in the Bois et al. model were obtained from the literature by using many sources, and the final estimates (posterior distributions) were obtained by using a MarkovChain-Monte Carlo approach.
It would have been preferable to use a single PBPK model. All three of the selected models are adequate for characterizing parent-compound (tetrachloroethylene) dosimetry, but they are not equivalent in characterizing tetrachloroethylene metabolism. There is inadequate justification for the selection of dose metrics for tetrachloroethylene metabolism, particularly in the use of total metabolites as the overall dose metric for cancer. The risk assessment would be improved if more effort were devoted to estimating the fraction of an absorbed tetrachloroethylene dose that enters the GSH pathway and the fraction entering the cytochrome P-450 pathway, which leads to the formation of trichloroacetic acid (TCA). That would permit development of more discrete, rational, and defensible dose metrics (for example, total GSH metabolites vs P-450 metabolites) for cancer end points.
The committee recommends that EPA pursue development of a single “harmonized” PBPK model that includes all routes of exposure (inhalation, oral, and dermal) and all relevant tissue compartments. With regard to metabolic dose metrics, the initial goal should be to predict the fraction of an absorbed tetrachloroethylene dose that enters the GSH pathway (initially forming TCVG) and the fraction that enters the P-450 pathway (eventually leading to TCA forma-
tion). That would permit development of more discrete dose metrics (such as total GSH metabolites vs P-450 metabolites) and should lead to a more rational and defensible selection of dose metrics for the various cancer end points.
Given the incomplete data available for characterizing the GSH pathway, several approaches may need to be adopted that rely on rodent in vitro data, human in vitro data where available, and allometric scaling as needed. For some key reactions, a parallel approach with trichloroethylene metabolism might be considered; in this respect, the approach and recommendations described by Lash and Parker (2001) should be considered and tested with appropriate model simulations. If modeling the GSH pathway is determined to be infeasible, total metabolism can be used as a reasonably conservative dose metric.
The PBPK model could then be built and tested around a combination of blood tetrachloroethylene and TCA concentrations, in vitro metabolism data, and urinary-excretion data for various metabolites (such as TCA, N-Ac-TCVC). With a single harmonized PBPK model, the population modeling approach could be used more effectively to estimate a range of Vmax and Km values and compare these posterior distributions with a more robust dataset of blood, in vitro, and urinary-excretion data.
The rationale for selection of most dose metrics is clearly explained in the draft IRIS assessment. However, the committee is concerned about the selection of the dose metrics chosen for tetrachloroethylene metabolism. As thoroughly reviewed in the draft, tetrachloroethylene metabolism can be separated into cytochrome P-450-derived oxidative metabolites produced primarily by the liver (the P-450 pathway) and metabolites derived from the initial formation of a GSH conjugate (the GSH pathway) and later reactions in several tissues, including the kidney. The P-450 pathway produces several metabolites, including the biologically persistent metabolite TCA. The P-450 pathway is more closely linked to hepatic cancer in rodent models whereas the GSH pathway appears to be associated more with renal tumors and perhaps leukemia. EPA has chosen not to estimate the flux of metabolism through the GSH pathway and summarizes the rationale for that decision as follows (p. 5-48): “However, the measurements of glutathione-dependent metabolism are from in vitro studies or they are measures of urinary excretion products and are, therefore, not representative of the toxic species in vivo.” Instead, the dose metric of total metabolism is used for all cancer end points in which tetrachloroethylene metabolites are implicated. That approach has created several potential problems that are not adequately addressed in the draft. The rationale for excluding the GSH pathway is inconsistent with the use of the three PBPK models, which also use in vitro data (the Reitz model) or urinary-excretion data (the Rao and Brown model) to estimate total metabolism. A fair question to ask is why the use of in vitro data and measures of urinary excretion products was acceptable for the P-450 pathway
but not the GSH pathway. The use of total metabolism as a dose metric reflects primarily the P-450 pathway because of large differences between the pathways in the flux of metabolism. The approach used by the different PBPK models to estimate metabolism and specifically estimation of the key metabolic parameters Vmax and Km varies substantially. Estimation of total metabolite formation in humans with the Reitz model relies primarily on in vitro hepatic metabolism data (microsomal metabolism, hence only the P-450 pathway) whereas the Rao and Brown model is validated by urinary excretion of nontetrachloroethylene radioactivity and TCA (also reflective primarily of the P-450 pathway). Although there is less experimental information on the GSH pathway, there are in vitro data from two studies that characterize the formation of TCVG in rodents (Dekant et al. 1998; Lash et al. 1998). The Dekant et al. (1998) study also attempted to measure TCVG in human tissues but was unable to detect it. However, their analytic methods appear to be rigorous and to allow estimation of the highest formation rate that could have occurred (still producing undetectable concentrations of TCVG), which would be helpful for risk assessment. A summary of the rates of TCVG formation in the liver in those studies is presented in Table 11-1. These values could be used to estimate the in vivo formation clearance of TCVG in the liver and kidney (data available but not included in Table 11-1) with an approach outlined by Houston and Carlile (1997). It would have been valuable if an attempt had been made to estimate the flux of tetrachloroethylene metabolism through TCVG in rodents and compare it with that in humans by using the results of Dekant et al. (1998) as an upper limit of formation rate. The modeling exercise could be strengthened by integrating the human urinary-excretion data reported by Volkel et al. (1998), for example, on detection of N-Ac-TCVC but not DCA in tetrachloroethylene-exposed volunteers.
With respect to hepatic cancer, it is debatable whether it is preferable to use a trichloracetic acid dose metric as opposed to total metabolites. The recent paper by Sweeney et al. (2009) makes a strong argument for the former. However, given the potential role of other P-450 pathway-derived tetrachloroethylene metabolites (discussed in Chapter 6) in hepatic cancer, the use of total metabolites as the dose metric appears justified. In addition, experimental evidence suggests that the toxicity of a directly administered metabolite does not reflect that of the “formed” metabolite (TCA in the case of tetrachloroethylene) even when blood concentrations are comparable (Pang 2009).
The use of total metabolites as a dose metric for renal cancer is not well supported. According to the available data (see Chapter 7), tetrachloroethylene metabolites derived from the GSH pathway are most likely to be the causative agents. Thus, a dose metric that more accurately reflects the flux of metabolism through the GSH pathway would be preferred. For reasons discussed previously, total metabolites constitute essentially a dose metric for the P-450 pathway. The committee encourages EPA to put forth a more thorough effort to develop a TCVG-based dose metric for rodents and possibly humans by using the available data summarized in Table 11-1.
TABLE 11-1 Summary of Data on Hepatic Metabolism of Tetrachloroethylene and Urinary Excretion of Glutathione-Pathway Metabolites
The approach used for interspecies scaling is presented in a reasonably clear manner. Figure 5-7 of the draft and the discussion on pp. 5-53 to 5-55 are particularly helpful. The committee’s main concern in this regard is how errors in estimating the metabolized fraction affect the extrapolation process.
Extrapolation from Route to Route
EPA has chosen to use the venous-blood area under the curve (AUC) as the route-to-route dose metric for extrapolating an inhalation exposure to a corresponding oral dose. The rationale for this approach is sound and adequately explained in the draft document. However, its implementation raises serious methodologic concerns based on inappropriate use of the selected PBPK models and uncertainties in the fraction of an oral dose of tetrachloroethylene that is metabolized. The three PBPK models used by EPA were specifically formulated and validated against inhalation exposures. There was no attempt to validate model predictions against blood tetrachloroethylene concentrations after oral dosing. To use the PBPK models, EPA has empirically assumed a value of the rate of oral absorption of tetrachloroethylene, which is entered as a constant. That approach is inferior to direct estimation as used in other published PBPK models, such those by Gearhart et al. (1993) and Dallas et al. (1995) (the latter only for rats and dogs). These PBPK models would have been better choices to begin the extrapolation exercise. Better still, a harmonized PBPK modeling approach (recommended earlier in this chapter) would have provided the greatest confidence in the route-to-route extrapolation.
Aside from the use of an appropriate PBPK model (for example, one specifically formulated and validated against oral-dosing data), uncertainty is associated with the dose dependence of tetrachloroethylene metabolism. EPA has assumed that a person will have nine drinking-water events during a day at roughly 2-hour intervals (excluding nighttime). The calculated oral equivalent dose of tetrachloroethylene is 1.1 mg/kg per day or 0.122 mg/kg per dose (that is, the discrete tetrachloroethylene dose received in each drinking-water episode). That oral dose is an order of magnitude lower than those previously used in toxicokinetic studies of tetrachloroethylene. The data from past studies clearly suggest that the fraction of a tetrachloroethylene oral dose that is metabolized is progressively reduced as the dose increases (Pegg et al. 1979; Frantz and Watanabe 1983; Schumann et al. 1980; Dallas et al. 1995). The issue of uncertainty in fractional tetrachloroethylene metabolism and dose was also raised by Reitz et al. (1996), whose PBPK model was used by EPA for route-to-route extrapolation of total metabolites. That raises the serious concern that a much greater fraction of the 0.122-mg/kg dose of tetrachloroethylene is being metabolized than was predicted by the PBPK models used in the risk assessment. The impact of the probable error is that the estimates of the venous-blood AUC of tetra-
chloroethylene shown in Figure 5-3 of the draft are probably overpredicted (that is, a higher oral dose is needed) and the estimates of total metabolites are underpredicted and may affect cancer assessments.
Cancer risk assessment results in an overarching summary of cancer risk by using a unit risk or a slope factor. In the process of deriving the unit risk or slope factor, uncertainty at every step is propagated into the final estimate. Because of the quantitative nature of the final risk estimates, it is critical to understand the effects of uncertainties on risk estimates both qualitatively and quantitatively. EPA has clearly identified key sources of variation and uncertainty in the process of risk assessment, including human population variation (susceptibility in exposure, metabolism, and response to exposure), low-dose extrapolation (including choice of dose-response models), choice of dose metric, extrapolation from animals to humans (cross-species scaling), and the use of PBPK models for route-to-route extrapolation. EPA’s investigation of the effects of uncertainties on risk estimates is qualitative except in dealing with such issues as the choice of dose-response models, the use of PBPK models, and, to a small degree, variation between studies. The following is an appraisal of EPA’s uncertainty analysis.
EPA’s presentation of the uncertainty analysis is generally transparent and includes sufficient detail. The tabular presentation of choices of study, end points, the approach (models) to extrapolation, and their effects on risk estimates is especially informative and easy to follow. For example, Table 6-3 of the draft summarizes key characteristics of the candidate rodent experiments and associated tumor types. Whereas that form of presentation is helpful, the committee does not agree with all characterizations presented in the table (see earlier discussion about the different cancer end points).
Similarly, Table 6-5 highlights EPA’s choices and their effects on the determination of the upper bound of the risk estimate at many critical steps of the risk-estimation process. It also lists EPA’s decision and the corresponding justification. Such presentation is effective and should be fully used. In some instances, however, the justification of EPA’s choice is debatable. As discussed in Chapter 6, for example, the committee believes that the hepatic-tumor data on male and female mice should also be given strong weight for consideration on the basis of dose-response data. In the case of the choice of dose-response model from among the multistage, Weibull, log-probit, and log-logistic models, one justification for using a multistage model was the relatively small variation in unit risk among the four models (a factor of 1.4). However, that narrow variation was shown only in the male-rat MCL data. The MCL data exhibit a nearly linear dose-response relationship and hence attenuate the difference among the four models. If EPA would consider other bioassay or tumor sites (such as hepatic tumors in female mice or MCL in female rats) that show a somewhat more
nonlinear shape of the dose-response relationship, the variation in unit risk calculated by the models would be much greater. Even in the case of MCL in male rats, the risk obtained by linear extrapolation to 1.5 × 10-5 mEq/kg per day varied by up to several orders of magnitude among the same four models (Table 5B-2). Therefore, choosing a multistage model on the basis that risks with other models at a POD are similar is difficult to justify.
More detail would have been helpful in a few of EPA’s analyses of uncertainties. For example, EPA’s assessment of uncertainties under different model forms (multistage, Weibull, log-probit, and log-logistic) used bootstrap simulations. The results show variation in extra risk spanning orders of magnitude at the low dose of 1.5 × 10-5 mEq/kg per day (bootstrap mean, 9.172 × 10-7 to about 1.078 × 10-3 in Table 5B-2) among the models despite their comparable goodness of fit to the dataset on MCL in male rats. Details about the bootstrap methods and scheme would facilitate appropriate understanding of the bootstrap distributions. For example, what was the number of bootstrap replications? What bootstrap method was used to simulate the distribution of extra risk? The committee views EPA’s consideration of uncertainty due to different forms of the dose-response relationship highly valuable, and it encourages EPA to extend such quantitative evaluation to all candidate datasets so that a fuller array of uncertainties can be assessed.
The committee notes that EPA discusses uncertainties in detail. However, the discussion typically focuses on individual sources without an in-depth illustration of the propagation of the uncertainties and their cumulative effect on the final risk estimate. That limitation is in part the result of qualitative treatment of uncertainties in many instances, notably concerning MOA, the choice of bioassay, and human variations. New methods that allow probabilistic quantification of the overarching uncertainty and of the variation in the final risk estimate are emerging (see Chapter 12). The capability to quantify the full range of overarching uncertainties associated with risk estimates facilitates separation of the science of risk assessment from risk-management decision-making. The committee encourages EPA to consider recommendations in Science and Decisions (NRC 2009) regarding uncertainty and variability.