Evaluating noncancer effects of trichloroethylene: dosimetry, mode of action, and risk assessment.

Alternatives for developing chronic exposure limits for noncancer effects of trichloroethylene (TCE) were evaluated. These alternatives were organized within a framework for dose-response assessment--exposure:dosimetry (pharmacokinetics):mode of action (pharmacodynamics): response. This framework provides a consistent structure within which to make scientific judgments about available information, its interpretation, and use. These judgments occur in the selection of critical studies, internal dose metrics, pharmacokinetic models, approaches for interspecies extrapolation of pharmacodynamics, and uncertainty factors. Potentially limiting end points included developmental eye malformations, liver effects, immunotoxicity, and kidney toxicity from oral exposure and neurological, liver, and kidney effects by inhalation. Each end point was evaluated quantitatively using several methods. Default analyses used the traditional no-observed adverse effect level divided by uncertainty factors and the benchmark dose divided by uncertainty factors methods. Subsequently, mode-of-action and pharmacokinetic information were incorporated. Internal dose metrics were estimated using a physiologically based pharmacokinetic (PBPK) model for TCE and its major metabolites. This approach was notably useful with neurological and kidney toxicities. The human PBPK model provided estimates of human exposure doses for the internal dose metrics. Pharmacodynamic data or default assumptions were used for interspecies extrapolation. For liver and neurological effects, humans appear no more sensitive than rodents when internal dose metrics were considered. Therefore, the interspecies uncertainty factor was reduced, illustrating that uncertainty factors are a semiquantitative approach fitting into the organizational framework. Incorporation of pharmacokinetics and pharmacodynamics can result in values that differ significantly from those obtained with the default methods.

The literature regarding noncancer effects due to trichloroethylene (TCE) exposure is extensive (reviewed in [1][2][3][4][5]. Studies of exposed humans and experimental animals are available reporting a wide range of effects including biochemical, cellular, and target-organ alterations. A large number of organs and organ systems have been reported to be targets at some dose in at least one study, including most prominently the nervous system, liver, and kidney. Due to the breadth of this database, some of the most significant challenges for evaluating options for noncancer risk assessment for TCE are the selection and interpretation of potential critical studies. This challenge increases as efforts are made to incorporate approaches for biologically based dose-response approaches because mode-of-action and dosimetry information are needed to analyze each end point. Many studies of exposed humans have been published including epidemiological studies of workers and the general population, controlled experimental exposures, and medical case studies of workers, overdose cases, and others. There is a substantial literature for acute effects in humans due to the use of TCE as an anesthetic and controlled human experiments to study potential neurological effects at occupational exposure limits. In contrast, efforts to determine potential chronic effects of exposure have confronted the problems that are typically associated with epidemiological studies including exposure to mixtures, difficulty demonstrating cause and effect, and limited characterization of exposure. There do not appear to be human studies that would be considered adequate for developing reference doses (RfDs) and reference concentrations (RfCs), subsequently referred to as toxicity values, but human data play an important role for cross-species comparisons of dose and toxic effects. Experimental studies generally used mice and rats. Dosing regimens ranged from single doses to lifetime oral or inhalation exposures. While these studies generally involve wellcharacterized exposures, their use for risk assessment is critically dependent upon interpretation of the toxicological significance of the effects and interspecies extrapolations. Several factors increase the difficulty of interpreting study findings with TCE. Other than neurological and kidney toxicity, effects are rarely observed in multiple species. Few studies are done under good laboratory practices guidelines. Few end points in tissues not associated with cancers have been studied in multiple experiments or by multiple laboratories. Studies report apparently contradictory results in several areas (e.g., developmental toxicity), but the exposure routes, methods, animal strains, or other factors vary, which may explain the variable results.
The relationship between the doses used for cancer and noncancer end points in the animal studies with TCE is worth noting. Chronic effects, particularly carcinogenicity, are often thought to occur at lower concentrations than effects arising from shorter exposures (except for developmental effects). For TCE, all oral lifetime studies have used relatively high doses: 500-1,000 mg/kg/day for rats and 1,000-2,000 mg/kg/day for mice in oil gavage studies. Inhalation studies have covered a wider range of doses, 50-600 ppm. These relatively high exposure doses and concentrations have been used because they generally have been well tolerated by the animals, although in some cases the maximum tolerated dose was exceeded. Similar and lower doses have been used in studies of noncancer end points in efforts to study dose-response relationships.
A consistent framework has been evolving for analyzing dose-response information that reflects relevant biological processes (6,7). Depending on the availability of information, different methods can be used within the overall exposure-dosimetry modeof-action response framework ( Figure 1). All dose-response assessment methods begin with the identification of a toxic effect and then estimate acceptable exposure levels protective of human health (7)(8)(9)(10). The default approach, in the absence of information, identifies a no-observed adverse effect level (NOAEL) or lowest-observed effect level (LOAEL) and then makes assumptions about mode of action and dosimetry embodied in standard uncertainty factors and adjustments to continuous or daily exposure (11)(12)(13)(14). An Options for dose-response assessment methods. While biological processes flow from exposure to response, dose-response assessment begins with the response and works backward. The preferred method uses quantitative pharmacokinetic and pharmacodynamic models to incorporate scientific data in a biologically based dose-response assessment. A more limited approach uses qualitative mode-of-action data to guide quantitative pharmacokinetic modeling and needed extrapolations. The low information approach relates exposure and response data using default assumptions.
alternative to the NOAEL is the benchmark dose (BMD), a dose associated with a specified risk of response determined using statistical curve fitting to dose-response data (15)(16)(17)(18). Additional scientific information can be incorporated in dose-response analyses by using a combination of qualitative modeof-action information with quantitative pharmacokinetic analysis (6,19,20). This approach has begun to be incorporated into noncancer dose-response assessment in the RfC process (11). Use of mode-of-action information helps to inform the pharmacokinetic analysis (i.e., selection of an appropriate dose metric) and the extrapolations accomplished with uncertainty factors (e.g., extrapolation of less-than-chronic data or between species). Opportunities to use these approaches were evaluated here for noncancer effects of TCE. A more complete biologically based dose-response assessment would use quantitative descriptions of the mode of action (i.e., pharmacodynamics) and dosimetry (i.e., pharmacokinetics) in animals and humans. Without relevant pharmacodynamic models, this approach is not feasible for any noncancer effects arising from exposure to TCE.
The focus of the remainder of this article will be a brief review of the toxicity database for TCE, evaluation of options for the selection of potential critical studies upon which to base dose-response values, and comparisons of the alternative methods for developing toxicity values. An extended version of this analysis provides additional details and quantitative alternatives (21).

Identification of Potential Critical Studies
Selection of potential critical studies is aided by well-designed studies using multiple doses (11). Unfortunately, many studies with TCE use only one or two doses, and data are often unavailable at other doses. When only high-dose data are reported, the studies are relatively easy to exclude from further consideration as potential critical studies because it is clear that other studies report effects at lower doses; these studies may then be considered further as supporting data. When the exposure doses used are relatively low, there is no similarly easy criterion for including or excluding the study. Rather, these decisions require consideration of the strengths and weaknesses of the overall database for the effect and the scientific design and implementation of the study. The following section describes the general state of the database for end points from which potential critical studies might be selected with a particular focus on the potential critical studies.

Methods for Selection ofCritical Studies
This article attempts to present the full range of options but ultimately also reflects the best professional judgments of its authors of where to focus their efforts. Therefore, an attempt has been made to document the chioices made so that, while others may agree or disagree, the fact that those choices were made is explicit and the reasoning is presented.
Potential critical studies were identified in several ways. Existing literature reviews were used extensively (3,4), particularly the Toxicological Profile for Trichloroethylene (1), which includes reporting of the doses used and lists key studies for a wide range of effects. An analysis of oral toxicity studies had been prepared previously with a similar focus on risk assessment options (2). Computerized searching of MEDLINE and TOXLINE (National Library of Medicine, Bethesda, MD) identified newer literature.
Based on these sources, original literature was obtained for review to determine its suitability as the basis for developing noncancer dose-response values.
Selection of the critical study cannot simply be based on the doses used in the study because different choices in methods (e.g., NOAEL vs BMD) and uncertainty factors can alter which study results in the lowest dose-response value. The choice was made to evaluate the toxicological significance of studies prior to calculating the dose-response values rather than taking every study through the quantitative analysis.

Selection ofOral Studies
End points evaluated in studies using oral exposure to TCE include neurotoxicity (and developmental neurotoxicity), immunotoxicity, reproductive toxicity, developmental malformations, kidney toxicity, and liver toxicity. Studies selected for further evaluation are summarized in Table 1.
Lifetime studies with TCE have predominantly focused on cancer and used high doses (. 500 mg/kg/day for rats and 2 1,000 mg/kg/day for mice) (22)(23)(24). These studies consistently report noncancer kidney toxicity in both species. A 6-month drinking water study using a wider range of doses (20-700 mg/kg/day) reported small changes in gross pathology, hematology, and alteration in immunological end points (25,26).
TCE is known to be neurotoxic in humans and animals, particularly at high oral and inhalation doses. The human studies include medical reports from use of TCE as an inhalation anesthetic and inadvertent or intentional acute consumption of large quantities of TCE. Epidemiological studies have been performed on workers exposed by inhalation and populations drinking TCEcontaminated water (1). Studies have reported changes in varied measures of neurophysiology such as blink reflex (indicating changes in the functioning of cranial nerves) and neuropsychology (27,28). These human studies were not considered satisfactory for developing toxicity values, although they may be considered supportive data.
Neurobehavioral effects are readily observed in rodents following acute, subchronic, or chronic dosing with 500 mg/kg/day or more (24,(29)(30)(31)(32). Studies of alterations in behavior and nervous system tissue histopathology or biochemistry following developmental exposure present mixed results (33)(34)(35). Options for potential critical studies are limited (32,36). One study exposed neonatal mice for 6 days and reported changes in one of three neurobehavioral measures at one of two later time periods (36); this study was considered too limited for quantitative analysis but is supportive of similar Environmental Health Perspectives * Vol 108, Supplement 2 * May 2000 LOAELs/NOAELs for other effects. The best of the oral neurobehavioral studies exposed adult rats to TCE for only 14 days and its NOAEL (150 mg/kg/day) is relatively high compared to others, so it was not considered further (32).
Limited data from human and animal studies are available on immunological effects of TCE (1). A few potentially immunologically related effects have been reported in humans, although the associations with TCE are speculative (4). Immunotoxicity was studied in outbred CD-1 mice exposed for up to 6 months to TCE in drinking water with 1% emulphor (approximately 20, 200, 400, and 700 mg/kg/day) (26). Ten different measures of humoral and cell-mediated immune function were measured in males and females at 4 and 6 months. Overall, effects were more frequently observed in females, suggesting they are more sensitive, but the difference is unlikely to be pharmacokinetic because males metabolize more TCE. The positive findings for the antibody-forming (or plaque-forming) cells and the delayed hypersensitivity assay are notable in light of recent analyses showing the antibody-forming assay alone and the two assays together are highly predictive of immunotoxicity (37,38). The authors conclude that in females, measures of antibodydependent immune function were affected at the two highest doses, while measures of cell immunity were affected at all four doses, but males were relatively unaffected. A similar study with chloral hydrate reported some effects in females, suggesting metabolites of TCE may be the active agents (39). The Sanders et al. study was evaluated further as a potential critical study (26).
No adverse reproductive effects have been reported for humans exposed orally to TCE (1). Several animal reproductive studies have been reported in the literature including twogeneration studies with mice and rats using microencapsulated TCE in feed (33,(40)(41)(42)(43)(44). These studies have generally been negative for a wide range of reproductive end points except at very high doses (approximately 1,000 mg/kg/day). In some cases, there are inconsistent results (e.g., for sperm malformation), although these may reflect differences in dosing, species, or strains of animals. Due to the limited positive findings (e.g., delayed parturition or whole-litter resorption with TCE or trichloroacetate (TCA) (33,45,46), difficulties in dose estimation, and the availability at similar or lower doses of better documented effects, none of these studies were addressed further.
As with other end points, there have been limited findings of developmental effects from TCE in human studies, but they are not adequate for quantitative analysis (1). Some animal studies focused upon specific end points for developmental malformations using oral dosing (2). The two effects occurring at the lowest oral doses are eye and cardiac malformations. The cardiac malformations studies are particularly difficult to interpret because of the lack of clear dose-response and temporal relationships (i.e., exposures prior to pregnancy, during pregnancy, or for both periods) (42). Developmental studies with TCA and dichloroacetate (DCA) also observed cardiac malformation (47)(48)(49)(50), but estimates of maternal blood TCA levels do not appear consistent between the TCA and TCE studies (21). No effects were observed with other TCE metabolites, notably chloral hydrate and dichlorovinylcysteine (DCVC) (49,50). Clearly, additional research is required to determine if this effect is repeatable in other laboratories and to better characterize its dose-response behavior. Several studies of eye malformations have evaluated offspring of pregnant mothers given a broad range of corn oil gavage doses of TCE (10-1,500 mg/kg/day) (31,45). Sprague-Dawley rats, a strain susceptible to this effect, were observed to have eyes of reduced size. Oral gavage with TCA or DCA also has been reported to lead to eye malformation (46,48). A positive dose-response was observed in these studies and they were analyzed quantitatively (45).
Studies of humans exposed to TCE either in drinking water or through accidental ingestion provide no evidence for kidney disease (1). A study reporting increased urinary tract infections in children included no direct measures of kidney function (51). Kidney toxicity in male and female rats (50-1,000 mg/kg/day) and mice (1,000 and 2,000 mg/kg/day) has been observed in chronic corn oil gavage studies (22)(23)(24)52). Subsequent to observing effects at the end of the 2-year study, a 90-day study in F344 rats was reevaluated and very mild indications of toxic nephrosis were observed (23). Although short exposures produced increased kidney weight, it is unclear if this represents a reliable indicator of chronic toxicity (53,54). A chronic study in rats reporting kidney toxicity was evaluated quantitatively (52).
Very limited data are available regarding liver toxicity in humans from oral exposures (1). Case studies of ingestion include one that reports liver damage and several that do not. Liver effects in animals are the best-characterized noncancer end point associated with TCE (2). Numerous measures of effects have been reported including alterations in liverto-body weight ratio (LW/BW) (25,(54)(55)(56)(57)(58)(59)(60)(61), largely due to hypertrophy and some hyperplasia (54,58), peroxisome proliferation, altered serum levels of liver enzymes, and histopathologically observable changes including necrosis. Some of these effects (e.g., LW/BW, peroxisome proliferation) are not typically considered adverse effects by themselves, while others are (e.g., histopathology). Effects in rats were less pronounced than those in mice (54,58). No histopathology was reported in chronically exposed rats (22,24), which is notable because the doses used (500 and 1,000 mg/kg/day) cause increased LW/BW in shorter exposures; e.g., Berman et al. (53). Data for noncancerous liver changes in mice exposed for a lifetime are not available because the studies only reported liver cancers (22,23). The 6-month drinking water study in mice reported that gross pathology was unremarkable, although some animals had fatty infiltration (25).
Exposure to TCE by oil gavage appears to produce greater and more severe liver effects than exposures without oil (61). Peroxisomal proliferation is one factor contributing to the increase in LW/BW (57,58). Alterations in histopathology or leakage of liver enzymes into serum have been reported at higher doses in mice and rats exposed for relatively short periods (53,54,56,58,59,61). The altered oxidative environment occurring with peroxisome proliferation may also play a role in the liver damage indicated at high doses by measures such as leakage ofserum enzyme levels.
LW/BW is a reasonable candidate as a sensitive early indicator for subsequent toxicity for risk assessment purposes. However, there are significant questions about the effects of corn oil, the role of peroxisomal proliferation, the linkage between LW/BW and other measures of liver toxicity, and interspecies comparisons. Additional data would be desirable for longer duration exposures to lower doses to further demonstrate a linkage between this early response and later liver toxicity. The 6-month drinking water study, which might be considered the most relevant exposure and duration, unfortunately only reported the doses at which liver effects were observed, not the LW/BW values; it was considered further (25). Two other studies reporting altered LW/BW ratios also were evaluated quantitatively (53,56).

Selection ofInhalation Studies
Most of the toxicity end points observed in oral studies have also been observed with inhalation exposures. The end points discussed include neurotoxicity, immunotoxicity, reproductive toxicity, developmental malformations, kidney toxicity, and liver toxicity. A summary table lists the studies that were selected for further evaluation (Table 2). Most chronic inhalation studies reported only cancer end points, but Maltoni et al. also reported noncancer effects (52,62).
Inhalation of TCE is well known to have neurological effects both in humans and in animals, hence its use as an anesthetic (1,63). Acute and occupational studies of behavioral effects in humans exposed at moderate concentrations (100-200 ppm) report no effects or changes in a variety of neurological measures (1,63). At higher concentrations, effects become obvious, with anesthesia occurring around 3,000 ppm. Studies of workers chronically exposed to TCE report a range of neurological effects. A few case studies have reported cardiac arrhythmia or other cardiac effects after unspecified or high inhalation exposures, while a brief exposure to 200 ppm for 2.5 hr had no effects (1).
A study of heart rate and electroencephalographic responses during wake and sleep periods reports alterations at 50, 100, and 300 ppm in rats exposed for 6 weeks (75). Measurements were made during the exposure and during a 22-hr postexposure period. Statistically significant changes were observed at several doses during or postexposure for time spent in wakefulness, slowwave sleep, and heart rate. This study is evaluated quantitatively below because it reports effects following subchronic exposure.
Short-duration studies at high concentration (near 1,000 ppm) have focused on end points including effects on trigeminal nerves and hearing loss (1). Animal studies using moderate concentrations and exposure durations of 5 months or less report effects at approximately the same range as reported with humans. These studies provide supporting evidence for effects following subchronic exposure at doses similar to and higher than the Arito et al. (75) study. No neurotoxicity data are available from chronic inhalation studies.
A few studies report decreases in mortality from respiratory infections and other diseases partially indicative of human immune status, though none include biochemical or cellular measures of immune system functions (1,4,76). Acute studies of immune function demonstrated effects in animals exposed by inhalation (77)(78)(79). Bacterial challenge of CD-1 mice following 3-hr TCE exposures (2.5-200 ppm) produced a dose-related increase in mortality due to compromised pulmonary immune functions including decreased phagocytosis by lung macrophages (77)(78)(79). A repeated 5-day exposure produced an increase that was substantially less than predicted from the assumption that the  (52) PK-LOAEL, PK-BMD week, 104 weeks concentration-time product would be constant (77). In the absence of data for longer exposures and given the lack of a constant concentration-time product, it is difficult to extrapolate these acute effects for evaluating potential chronic toxicities, although these data might be considered supporting data when evaluating options for RfCs derived from other studies. No reproductive studies in humans are available (1). Studies in animals reported effects at concentrations of 500 ppm or higher; they will not be evaluated further because other end points are reported to occur at lower concentrations. Data for developmental toxicity in humans are limited and inconsistent (1,(80)(81)(82). An increase in delayed ossification and whole-litter resorptions was observed following exposure of rats to 100 ppm TCE (83). Other animal studies reported no statistically significant increases in malformations with concentrations ranging from 100 to 1,800 ppm (83)(84)(85)(86).
Several studies of workers have reported limited findings of renal toxicity (1). Kidney toxicity was reported in an inhalation study exposing Sprague-Dawley rats to 0, 100, 300, or 600 ppm of TCE (62). A dose-related trend for renal megalonucleocytosis was observed in male rats. This finding is in contrast to the oral gavage data that showed effects in both sexes of rats and in mice (23,24). Maltoni et al. (62) was evaluated quantitatively.
There are little human data reporting pulmonary toxicity (1). A study in rats exposed for up to 90 days to 700 ppm TCE reported no histopathological changes (87). Vacuolation of Clara cells and decreased cytochrome P450 activity were reported in female mice exposed at concentrations of 20-2000 ppm, though no quantitative data were presented (88). This effect was not observed in rats. It is believed to be specific to mice because of accumulation of chloral in Clara cells. This study will not be analyzed further because of the limited documentation, the apparent species specificity, and the availability of studies at similar doses for other effects.
Although liver toxicity is frequently reported in animal studies with TCE, there is more limited evidence for liver effects in humans (1,3). No gross pathological liver effects were observable in several species (rats, guinea pigs, rabbits, dogs) exposed to 730 ppm (8 hr/day, 5 days/week, for 6 weeks) (87). Dose-related (37-300 ppm) increases in LW/BW in mice were reported without more direct measures of liver toxicity (89)(90)(91)(92). The effect in mice was largely reversed in 30 days following the 30-day exposure. Ethanol exposure increased liver toxicity because of TCE exposures of 500 ppm or greater, reflecting, in part, increased metabolism by cytochrome P4502E1 at this high concentration (93,94). The dose-response data of Kjellstrand et al. (91) were evaluated quantitatively based upon the interpretation that LW/BW changes were sensitive early indicators of potential liver toxicity.

Quantitative Dose-Response Analysis
Each study previously identified as a potential critical study was evaluated using one or more methods for developing toxicity values (6,1 1). These methods ranged from the default approach used in the absence of any information except a LOAEL to approaches that incorporate mode-of-action and dosimetry data to assist in extrapolating between exposure regimens, dose levels, study durations, and species.

Dose-Response Analysis Methods
All the studies selected for dose-response analysis evaluated effects of TCE exposure on laboratory animals. For each study the NOAEL or LOAEL was determined and the BMD calculated, all based on the exposure doses used in the studies. These values were then adjusted to daily dosing or continuous inhalation exposure based on an assumption that the concentration-time product would be constant (11). To calculate the human equivalent concentration (HEC) for the NOAEL or BMD, the value adjusted to continuous exposure was multiplied by 1.0 because the value for the blood-to-air partition coefficient for animals was greater than that for humans.
Uncertainty factors were then applied to obtain the RfD or RfC. These uncertainty factors are intended to account for estimating a NOAEL from a LOAEL (referred to as L), extrapolating to chronic exposure from subchronic exposures (referred to as S), extrapolating from animals to humans (referred to as A), accounting for the variability of the human population and potential sensitive subpopulations (referred to as H), and limitations of the available toxicity database (referred to as D). These uncertainty factors are assumed to be independent, so they are multiplied together. Each uncertainty factor usually has a value up to 10, though this appears to be overly conservative when multiple factors are used (13). Therefore, policy choices have been made to limit the total uncertainty factor to 3,000 when four factors are used rather than 10,000 (11).
Exposure dose is frequently complexly related to responses, so it is advantageous to use mode-of-action information to select appropriate internal dose metrics reflecting the biologically effective dose. The NOAEL or BMD in terms of internal exposure (i.e., in units of the dose metric) was estimated using the animal physiologically based pharmacokinetic (PBPK) model. The human PBPK model was used to estimate the external doses for continuous or repeated human exposure (i.e., inhalation or drinking water, respectively) that would result in the appropriate internal dose metrics. Because the human model was linear for all the dose metrics over very broad dose and concentration ranges, essentially identical results would be obtained whether the uncertainty factors were applied to the internal dose metric or the associated exposure dose estimated using the PBPK model.
Uncertainty factors also largely fit into the exposure:dosimetry:mode-of-action:response framework, though the database uncertainty factor is a policy response to data limitations and essentially falls outside this framework. When mode-of-action information was used to select a dose metric, it also helped inform issues of extrapolation from a LOAEL to a NOAEL, responses following shorter exposure to anticipated responses with chronic exposure and between animals and humans. There is less frequently information relevant to estimating human variability (95). In the RfC methodology, calculation of the HEC is assumed to account for the pharmacokinetic differences between species, so the uncertainty factor for interspecies extrapolation (A) has, at most, a value of 3. Absent additional mode-of-action information, a similar assumption was used here when internal dose metrics were estimated for oral studies (i.e., A = 3). Specific examples of how this might be done are discussed below in the evaluation of specific end points for TCE. In general, this approach attempts to maximize the use of scientiflc information in a semiquantitative manner by adjusting the uncertainty factors to reflect data that are not captured in a more extensive biologically based modeling approach.

PBPK Model for TCE and Metabolites
The PBPK model described the pharmacokinetics in mice, rats, and humans of TCE and its toxicologically important metabolites: TCOH, TCA, DCA, and DCVC (96,97). It describes oral and inhalation exposures. A range of possible dose metrics can be obtained including: area under the curve (AUC) in blood for TCE (designated AUCTCE), TCA (designated AUCTCA), and TCOH (designated AUCTCH). Other dose metrics can also be estimated such as total metabolites normalized to body weight (designated AMET), peak concentrations for TCE (designated CTCE), TCA (designated CTCA) and TCOH (designated CTCOH), and an estimate of metabolism through the DCVC pathway (designated KTOX). The model was exercised using Advanced Continuous Simulation Language (ACSL, Mitchell & Gauthier Associates, Concord, MA).
Several notable kinetic differences between the species have been observed in the experimental literature and were included in this model. Mice metabolize TCE much more effectively than either rats or humans. The two major metabolites of TCE are TCA and TCOH. In humans, TCOH appears to undergo extensive enterohepatic recirculation of its glucuronide conjugate resulting in a much longer half-life for TCOH in humans compared to the rodents. Because TCOH can be metabolized back to TCA, this also results in a much longer half-life for TCA in humans compared to rodents.

Benchmark Dose Methods
An alternative to the identification of a NOAEL is to calculate a BMD from the dose-response data (15)(16)(17). Three different kinds of data were used in BMD analyses: quantal, continuous, and nested quantal (i.e., litter) data. For each type of data, different methods were used. The quantal data were evaluated using two programs, THRESH and THRESHW (KS Crump Group, ICF Consulting, Ruston, LA). THRESH fits a polynomial model and THRESHW fits a Weibull model to the data (15). The continuous data were analyzed with BENCH_C (KS Crump Group, ICF Consulting, Ruston, LA), which fits the Power and Weibull models (16). In addition, a quadratic model for analyzing continuous data was used (98) for comparative purposes with the liver data of Buben and O'Flaherty (56). This program was generously provided by R. L. Kodell (National Center for Toxicology Research, Jefferson, AK). Finally, the litter incidence data were analyzed using TERAMOD and TERALOG (KS Crump Group, ICF Consulting, Ruston, LA) (99); these models account for possible extra-binomial variation associated with nested responses. Regardless of the type of data, the models considered express probability of response as a function of dose.
The BMD analysis for each data set used two models to estimate the maximum likelihood estimate or best fit (MLE) and its statistical lower bound (BMDL) for a 10% benchmark response (BMR). For continuous data, the 1% region of the tails of the distribution of control responses was assumed to be abnormal, i.e., P0 was 0.01. Other values for BMR and PO were investigated and are presented in an extended version of this analysis (21). Lower values for PO result in higher estimates for the MLE and BMDL. As the value of PO decreases, a smaller portion of the control distribution is considered abnormal and the acceptable variation for the continuous end point (e.g., LW/BW) is larger.
Environmental Health Perspectives * Vol 108, Supplement 2 * May 2000 limited here to explaining judgments involved in this process; absent data, typical U.S. Environmental Protection Agency (U.S. EPA) default assumptions were used, as indicated in Table 4.

Eye Malformation
Comparing mean responses, only the highest dose group, 1,125 mg/kg/day, was statistically different from the controls (45). Trend testing to find the no-statistical-significance-of-trend    bmg-hr/L internal dose metric -AUCTCA. cmg/kg internal dose metric -KTOX. dRfD is obtained by multiplying NOAEL/BMD by the adjustment for exposure and dividing by uncertainty factor and the dose metric adjustment.
(referred to in the literature as NOSTASOT) dose found a NOAEL of 32 mg/kg/day and a LOAEL of 101 mg/kg/day (2). These values are well below the dose of 1,125 mg/kg/day that was the only dose group statistically significantly different from controls, as reported by the authors using a test for comparison of means. BMDs obtained with two models were very similar using exposure doses (Table 3, Figure 2). The BMDLs are much higher than the NOAEL because the observed response was so small. The developmental exposures were daily, so no dose-averaging adjustments were made (Table 4).
LW/BW While a NOAEL of 18 mg/kg/day was reported in the drinking water study (25), only LOAELs were reported in the other two studies (  (25,53). The gavage study used 5-day/week dosing; the LOAEL was adjusted by 5/7 to obtain an RfD (56). Selection of changes in LW/BW as a potential critical end point was based on its role as an early event in the toxicity process and a sensitive indicator of potential liver effects observed at later times (2). Therefore, based upon the mode-of-action argument that this early event is an indicator of toxicities that develop later, no adjustments for the duration of exposure (uncertainty factor S) would be needed, regardless of the study duration. Similarly, the value of 3 was used for L, based on the LOAEL being a minimal change in an effect that is not actual toxicity but a sensitive indicator. The size of the changes was small, just 12% increase in LW/BW in the Buben and O'Flaherty (56) and 7% in the Berman et al. study (53). Immunotoxicity Impaired immune functions were observed in mice, particularly in females, exposed to TCE in drinking water for 4 months (26). The antibody-forming (plaque-forming) cell assay had a LOAEL of 400 mg/kg/day in females at 4 months and males at 6 months; Environmental Health Perspectives *Vol 108, Supplement 2 *May 2000 Oral Studies: Dose-Response Based on Exposure Dose Metric Six studies reporting four different toxic effects were evaluated quantitatively ( Table 1). The NOAELs, LOAELs, and BMDs for each study are reported in Table 3. Other adjustments (i.e., exposure regimen, dose metric, and uncertainty factors) and the RfDs derived are summarized in Table 4  the NOAELs were 200 mg/kg/day. The delayed hypersensitivity assay had a LOAEL of 20 mg/kg/day in female mice at 4 months compared to the vehicle-treated control, but there was a large difference between the vehicle-treated control and naive control. At 6 months, the LOAEL for this assay was 800 mg/kg/day and the NOAEL was 400 mg/kg/day in females. No effects were seen in males. Taken together, these assays are supportive of a LOAEL of 400 mg/kg/day and a NOAEL of 200 mg/kg/day ( Table 3). The BMD method was not used for the evaluation of the data from the immunotoxicity study. The dose-response relationships varied for each assay, and it was felt that no single assay should be used alone to determine the dose-response relationship for estimating a BMD. There was no dose-averaging adjustment because exposure was daily (Table 4). No adjustments for the duration of exposure were made. There appeared to be a greater effect at 4 months than at 6 months, suggesting that the mode of action may not be cumulative over longer periods of time as is assumed in the use of the subchronic-to-chronic uncertainty factor. This may partly reflect changes in immune system function over time.

Kidney Toxicity
Male Sprague-Dawley rats developed kidney toxicity when dosed with 250 mg/kg/day, 5 days/week for 52 weeks but not at the NOAEL of 50 mg/kg/day [ Table 3, (52)]. This data set is marginally acceptable for the BMD method because there is only a single positive dose (Figure 4). However, it was evaluated because based upon studies in five other rat strains (23,24), a response of nearly 100% is potentially obtainable; i.e., the 46.7% response at 250 mg/kg is a valid estimate of the percent of total response. The polynomial model appears to provide a more reasonable fit, approximating a straight line between the NOAEL and LOAEL. The Weibull model predicts no response until a relatively high dose; response then increases very rapidly. The Weibull model might be considered inadequately health protective given that there is no mode-of-action information to justify assuming virtually no response until doses near those giving a 50% response.
The NOAEL and BMD were multiplied by 5/7 to adjust to daily dosing ( Table 4). The study dosed for a significant portion of lifetime, so no adjustment was made for duration. Animals were observed till the time of natural death (intended to allow about a year for postexposure expression of tumors), so some recovery from kidney toxicity might have occurred in some animals.

Oral Studies: Incorporating Pharmacokinetics and Mode of Action
The results of the dose-response analysis using the internal dose metrics selected below are summarized in Table 4.

Eye Malformation
No hypotheses for modes of action have been proposed for the eye malformation end point that would assist in selecting the appropriate internal dose metric. The teratogenicity studies with TCA and DCA report increases in microphthalmia and anophthalmia, suggesting one or both chloroacids may contribute to the effects observed with TCE (46,48). TCE metabolism in the fetus would be limited because cytochrome P4502E1 levels are very low prior to birth and increase greatly shortly after (100); circulating maternal levels of TCA or DCA would appear to be the relevant chemical form. Comparison of the AUCTCA at doses of TCE and TCA, which produced about 10% incidence of eye malformations in pups, found drastically different results. The AUCTCA estimated for dosing with the LOAEL dose of 1,200 mg/kg/day of TCA (46) was about 20,000 mg-hr/L, while TCE at 1,125 mg/kg/day (45) produces an AUCTCA of 405 mg-hr/L. This discrepancy raises doubt that the levels of TCA produced from TCE would be adequate to cause the effect, although the rat strains were different. Although most toxicities associated with TCE arise from metabolites, it is possible that changes in membrane characteristics due to TCE itself are responsible for the developmental effects because development is highly dependent upon cell-cell communications. It is also undetermined if the effect results from the concentration of chemical or a time-integrated measure of dose, such as AUC. Finally, in the absence of models for both the developing rat and human fetus, only dose metrics in the maternal blood perfusing the placenta would be feasible. Absent strong mode-of-action hypotheses, no evaluations with internal dose metrics are reported here. Variable results were obtained when many potential pharmacokinetic NOAELs were evaluated (21). LW/BW Alterations in the liver are believed to arise from metabolites of TCE. Buben and O'Flaherty found that total urinary metabolites linearized their LW/BW data, so AMET was one potential dose metric (56). This dose metric would be consistent with effects resulting from metabolism of TCE, with no adjustments for the pharmacokinetics of the individual metabolites (i.e., short-lived vs long-lived, etc.). However, analysis of the data on LW/BW and enzyme changes following dosing with TCE and perchloroethylene suggests that AMET is not a satisfactory dose metric because it would predict that TCA from perchloroethylene (60-90% of urinary metabolites) was more potent than TCA from TCE (about 10% of urinary metabolites) (56). Oral exposure to TCA produced increased LW/BW, peroxisome proliferation, and other effects associated with TCE, so it is a candidate as the active metabolite (101)(102)(103)(104)(105). A potential dose metric for TCA is AUCTCA. This dose metric was obtained by running the PBPK model for 336 or 1,008 hr and dividing the total dose metrics for that time by 14 or 42 days, for the rat and mouse studies, respectively, to obtain the daily average AUC (53,56).
Finally, several other dose metrics might be relevant for the effects in mice or rats including those for DCA. There was a lack of data supporting human DCA production from TCE, so it was not possible to use PBPK modeling to obtain an equivalent human dose. In addition, estimates of DCA production in mice have been excessively high due to analytical chemistry artifacts (106).
For liver effects there is a database from which to develop information about the mode of action in animals and humans. Only information for TCE or its metabolites was explicitly considered here. TCE has been tested in mice, rats, gerbils, guinea pigs, rabbits, and dogs by oral or inhalation exposures. It produces limited noncancer liver toxicity in any of these species until doses approaching the LD50 (lethal dose at 50%) are reached.
Similarly, the available data indicate that TCE is not a potent liver toxicant in humans, as would appear to be predicted by AUCTCA if equal or greater pharmacodynamic sensitivities were assumed.
Because LW/BW alterations and other liver effects are believed to involve the peroxisome proliferator-activated receptor (PPAR), extrapolation of the response to humans is not quantitatively straightforward. One approach to estimating the interspecies pharmacodynamic (PD) extrapolation was to compare rats and mice. Data for LW/BW increases and alterations in palmitoyl-coenzyme A oxidase activity (a marker for peroxisomal proliferation) following drinking water exposure to TCA for 10 days showed rats to be much less responsive than mice based upon estimated AUCTCAs (101). Another study reports rats to be more sensitive than mice for palmitoyl-coenzyme A oxidation following 10 days dosing with TCA in corn oil. When AUCTCA is estimated for this study, the rats are about 1.5 times more responsive. DeAngelo et al. found that corn oil increased response in rats compared to an aqueous vehicle (101). Therefore, there are two issues for extrapolating to humans, their relative sensitivity and the effects of corn oil in the rodents versus drinking water in humans. The available molecular and pharmacological data have not identified humans with a fully active PPAR as found in the mouse. It is likely that humans are more like rats than mice, so these data do not support the standard assumption that humans are more sensitive than the most sensitive rodents. Analysis of the interspecies extrapolation for other peroxisomal proliferators, particularly the hyperlipidemic drugs would assist in further determining the appropriate interspecies extrapolation. These data indicate that the value of the uncertainty factor for interspecies extrapolation should be no greater than 1 and potentially less than that.

Kidney Toxicity
Kidney toxicity is believed to develop from metabolites formed through the glutathione conjugate pathway (107). Therefore, the model was used to estimate a dose metric for DCVC in the kidney, KTOX. KTOX represents the total production of the thioacetylating intermediate from DCVC divided by the volume of the kidney. The daily value for KTOX was obtained averaged over a 1-week (7-day) period. Limited information is available about the pharmacodynamic processes involved in the kidney toxicity. Both mice and rats developed kidney toxicity in corn oil gavage assays and the mice appeared somewhat more sensitive (22)(23)(24). In the absence of other information, the default assumption was made that humans are 3-fold more sensitive than animals for the PD processes leading to kidney toxicity.
Inhalation Studies: Dose-Response Based on Exposure Dose Metric Three studies reporting different toxic effects were evaluated quantitatively ( Table 2). The NOAELs, LOAELs, and BMDs for each study are reported in Table 5, as are other adjustments (i.e., exposure regimen, dose metric, and uncertainty factors) and the RfCs derived. Description here is limited to explaining judgments involved in this process; otherwise standard U.S. EPA default assumptions were used, as indicated in Table 5.

Neurological Effects
Exposure dose-based NOAELs, BMDs, and RfCs were derived from measurements of electroencephalographic activity and heart rate made in a 32-hr period during the second, fourth, and sixth weeks of exposures in freely moving male rats implanted with electrodes (75). Three measures were derived from electroencephalographic measurements-wakefulness, slow-wave sleep, and paradoxical sleep; heart rate was measured independently. All four measures were statistically different from control levels at the lowest dose used (LOAEL = 50 ppm); a NOAEL was not determined. The exposure dose-based BMDs vary by as much as 17-fold using the two models (Weibull and Power) ( Figure 5A, Table 5). The exposure-based LOAEL and BMD values were lowered to estimate continuous exposure from the 8-hr/day, 5-day/week exposures. The LOAEL was considered to reflect a minimal, though statistically significant, effect, so the uncertainty factor (L) was 3 instead of 10. LW/BW Increases in LW/BW, a liver toxicity end point, were statistically significant in both male and female mice at all dose levels tested (91). The LOAEL was used without adjustment because the exposure was to 37 ppm continuously for 30 days. Although this is a LOAEL, it is for a minimal effect, so the uncertainty factor (L) was 3 instead of 10. In contrast to the identical LOAELs for female and male mice, the BMDs differ because of the different shapes of the dose-response curves ( Table 5). The females showed a basically linear increase with dose in LW/BW, while the male dose response was concave.

Kidney Toxicity
Kidney toxicity was reported in rats at the two higher exposure concentrations, and a NOAEL was found at the lowest concentration of 100 ppm (52). For the inhalation data sets from the kidney toxicity study, the polynomial and Weibull models gave very similar fits and BMD estimates. To be consistent with the oral data, the results from fitting the polynomial model were used here. The NOAEL or BMDL were multiplied by 7/24 and 5/7 to adjust to continuous dosing.

Inhalation Studies: Incorporating Pharmacokinetics and Mode of Action
The results of the dose-response analysis using the internal dose metrics selected below are summarized in Table 5.

Neurological Effects
There are two major hypotheses for the mode of action of TCE in the causation of neurological effects, activity of parent TCE, or the metabolite, TCOH. Chloral hydrate might also be active, but it is generally believed that TCOH is the active species in choral hydrate anesthesia (108). Either AUC or peak concentrations might be reasonable dose metrics, but as illustrated in Figure 5C, the CTCOH linearized the data, suggesting that TCOH is the active agent. Therefore, results are reported with this dose metric. The dose metric was obtained by estimating with the model the daily concentration during the 5 exposure days of the sixth week of the experiment. Evaluation of other dose metrics for TCE and TCOH are reported elsewhere (21).
Although mode-of-action data were limited for the neurological effects, there was some comparable data in humans and rodent studies. These data provided a basis for comparing the pharmacodynamic responsiveness of the two species. Dose metrics at the LOAELs or NOAELs of the human and rat studies were compared (Table 6). One potential limitation of the human data was that all the studies use exposures lasting 1 day or, in one study, 5 days.
Whether this is a significant limitation is undear because Arito et al. report some effects show a dependence on repeated exposures, while others do not, and other effects were dependent upon repeated exposures but not on exposure concentration (75). The internal dose metrics modeled for the human studies are generally equal to or greater than those in the rat study, particularly for TCOH. Exceptions occur in the dose metrics for TCE. Because of the slower dearance ofTCOH in humans arising from enterohepatic recycling, the modeled peak concentration and AUCTCH is consistently higher than those found for the rat.
These data do not support the default assumption that humans are more sensitive to TCE exposure than the rat. The studies show little or no effect in humans with modeled internal dose metrics equal to or greater than those modeled for the rat exposed at a concentration producing minimal effects. Therefore, this combined pharmacokinetic and mode-ofaction analysis support assuming the two species equivalently sensitive (i.e., the value of A would be 1.0) for these effects when using CTCOH as the internal dose metric.

LW/BW
The AUCTCA derived from the data set of the liver toxicity study was obtained by estimating with the model the daily average AUCTCA during the 30-day continuous exposure. As described previously for oral exposures, information on the mode of action does not support the default assumption that humans are more sensitive than animals for liver effects, so a value of 1 was used for the interspecies extrapolation based upon internal dose metrics. A 10-fold adjustment for H was also made, although the mice may essentially be equivalent to a sensitive subpopulation with a fully active PPAR that has not yet been identified in humans.

KidneyToxicity
KTOX was evaluated for inhalation exposures as it was for oral exposures. The average daily value for KTOX was obtained by averaging over a 1-week (7-day) period.

Discussion and Summary
The process of developing dose-response values is an iterative one that typically is repeated as additional scientific information becomes available. The exposure:dosimetry: mode-of-action:response framework described in this paper organizes the process and promotes consistency between end points. This framework facilitates incorporating scientific data and assists scientists conducting research by defining methods by which scientific data can be used in risk assessment. These methods include: BMDs derived by empirical curve fitting, pharmacokinetic models reflecting the processes important for different chemicals and their metabolites, use of uncertainty factors to semiquantitatively adjust for incompletely described pharmacokinetic and pharmacodynamic processes, and eventually more complete pharmacodynamic models that quantitatively describe critical steps in the mode of action leading to toxicity. The choice of methods used varies in response to the availability of data as well as differences in the relevant biological processes. Appropriate dose-response analysis requires a consistent framework for organizing information and analyses, not a single universally applied analytical method. The initial step in the process is to review the literature and choose the studies that might be used as critical studies for each toxic end point. The supporting literature must be reviewed to determine the availability of hypotheses for the mode of action leading to the various effects. The NOAELs or BMDs can be determined using exposure doses or internal dose metrics; absent mode of action and pharmacokinetic information, analyses based on exposure dose represent an appropriate default approach. When feasible, pharmacokinetic analyses should be undertaken to obtain internal tissue dose metrics in the animals used in the studies. Next the mode of action must be evaluated to determine the relationship of the internal dose metric to the effect and the proper extrapolation to humans. These results are then extrapolated to humans using both dosimetry and mode-of-action information. Finally, the various possible RfDs and RfCs must be compared with each other, as well as any other relevant literature, to determine if any other factors should be considered. Overall, RfDs and RfCs developed using the BMD method with pharmacokinetic dosimetry and consideration of mode of action appear the most desirable because they incorporate the greatest amount of the scientific database into the regulatory process.
One measure of the utility of this approach is the consistency obtained for systemic effects regardless of whether the exposure was oral or by inhalation. The BMDs and toxicity values derived from them (i.e., RfDs and RfCs) for both liver and kidney effects are very similar by these two routes. For example, the maximum likelihood estimates for KTOX at the same response level (BMR = 0.1) are very similar for the oral and inhalation routes, 673 and 623 mg/L respectively, demonstrating good doseroute extrapolation using the model. The AUCTCA underlying the RfC and RfD for liver effects are also similar, differing by a factor of only 2-3, though based upon studies using different rodent strains, durations, and routes ofexposure.
The analyses presented here have included end points with widely varying databases. Effects included those for which virtually no data other than exposure and response were available (i.e., eye malformations) and others with varying amounts of data on mode of action and pharmacokinetics that inform extrapolations between exposure regimens (e.g., less than chronic-to-chronic liver effects) or between species (e.g., neurological effects). Overall, the analyses suggest that an RfD in the range of 0.06-0.12 mg/kg/day based on liver effects analyzed with AUCTCA as the internal dose metric would also be protective for the other end points evaluated. Similarly, an RfC in the range of 0.4-1.0 ppm based on slow-wave sleep analyzed with CTCOH, as the internal dose metric would be protective for the other end points as well. Modifications of these numbers might arise from the use of alternative approaches for deriving the BMDs or from different interpretations of the mode-of-action and pharmacokinetic considerations informing selection of uncertainty factor values.