genetic polymorphisms in assessing interindividual ... · received april 20, 2001 introduction...

Regulatory Toxicology and Pharmacology 35,177-197 (2002) (",doi:l0.l006/rtph.2001.1517, available online at http://www.idealibrary.com on II E ~ L '-'

Genetic Polymorphisms in Assessing Interindividual Variabilityin Delivered Dose

L. T. Haber,*,l A Maier,* P. R. Gentry,t H. J. Clewell,t andM. L. Dourson**7bricology Excellence for Risk Assessment, 1757 Chase Avenue, Cincinnati, Ohio 45223;and tEnviron, 602 East Georgia Avenue, Ruston, Louisiana 71270 .

Received April 20, 2001

INTRODUCTION

Treatment of human variability (also called intra-species or interindividual variability) in risk assess-ment, including protection of sensitive subpopulations,has traditionally been semiquantitative. For noncancerrisk assessment, a default uncertainty factor (UFH)of 10 has been used to account for human variability(Barnes and Dourson, 1988; U.S. EPA, 1994). This fac-tor considers both toxicokinetic and toxicodynamic vari-ability. A number of researchers have evaluated humandata on variability in the context of evaluating whetherthis 10-fold factor accurately accounts for the variabi-lity between the average and sensitive human in res-ponse to chemicals (Dourson and Stara, 1983; Hattiset ai., 1987; Kaplan, 1987; Sheenan and Gaylor, 1990;Calabrese et ai., 1992; Calabrese and Gilbert, 1993;Hattis and Silver, 1994; Renwick and Lazarus, 1998;Burin and Saunders, 1999). In general, data from allof these studies indicate that the default value of 10for intraspecies variability is protective when startingfrom a median response, or by inference, from a no-observed-adverse-effect level assumed to be from an av-erage group of humans. Although some of these analy-ses (Calabrese et ai., 1992; Hattis et ai., 1987; Kaplan,1987) noted a range of variability greater than 10-fold,it is because these authors evaluated the total rangeof human variability, rather than considering that theuncertainty factor of 10 is applied to account for the de-gree of variability from the population average to thesensitive human.

By contrast, variability in the human population hasnot been addressed explicitly in traditional cancer as-sessment for genotoxic carcinogens. There are two rea-sons for this difference. First, a different assumptionis used for the origin of the dose-response curve fornoncancer endpoints and classical (genotoxic) carcino-gens. For noncancer endpoints, the dose-response curveis assumed to be due to differences in sensitivity in thetest population. At low doses, only the most sensitivemembers of the population are expected to respond, ifa response is observed. As the dose increases, both the

Increasing sophistication in methods used to ac-count for human variability in susceptibility to toxi-cants has been one of the success stories in the conti-nuing evolution of risk ~ssessment science. Geneticpolymorphisms have been suggested as an importantcontributor to overall human variability. Recently,data on polymorphisms in metabolic enzymes havebeen integrated with physiologically based pharma-cokinetic (PBPK) modeling as an approach to deter-mining the resulting overall variability. We presentan analysis of the potential contribution of polymor-phisms in enzymes modulating the disposition of fourdiverse compounds: methylene chloride, warfarin,parathion, and dichloroacetic acid. Through these casestudies, we identify key uncertainties likely to be en-countered in the use of polymorphism data and high-light potential simplifying assumptions that might berequired to test the hypothesis that genetic factors area substantive source of human variability in suscep-tibility to environmental toxicants. These uncertain-ties include (1) the relative contribution of multipleenzyme systems, (2) the extent of induction/inhibitionthrough coexposure, (3) allelic frequencies of majorethnic groups, (4) the absence of chemical-specificdata on the kinetic parameters for the different al-lelic forms of key enzymes, (5) large numbers of low-frequency alleles, and (6) uncertainty regarding differ-ences between in vitro and in vivo kinetic data. Oureffort sets the stage for the acquisition of critical dataand further integration of polymorphism data withPBPK modeling as a means to quantitate populationvariability. (0 2002 Elsevier Science (USA)

Key Words: polymorphism; risk assessment; PBPKmodel; warfarin; parathion; dichloromethane; methy-lene chloride; dichloroacetic acid.

1 To whom correspondence and reprint requests should be ad-

dressed. Fax: (513) 542-7487. E-mail: [email protected].

1'r1 0273-2300/02 $35.00@ 2002 Elsevier Science (USA)

All rights reserved.

:ir)\/1ri"-"

TERA

This material has been published in Regulatory Toxicology and Pharmacology 35, 177-197 (2002), the only definitive repository of the content that has been certified and accepted after peer review. Copyright and all rights therein are retained by Academnic Press. This material may not be copies or reposted without explicit permission. Copyright 1988 by Academic Press www.idealibrary.com

HABER ET AL.178

severity of the response and the percentage of the popu-lation affected are assumed to increase. As risk as-sessOrs characterize the dose response at the lowerend of the curve, variability is addressed directly innoncancer assessments. In contrast, according to thesimplest version of the cancer paradigm, cancer is as-sumed to be a stochastic process. Although the stochas-tic nature of the response generally refers to the prob-ability of the chemical reacting with DNA, it can alsobe interpreted as meaning that the individuals who getcancer at lower doses are simply unlucky, not more sen-sitive (e.g., they happened to have a DNA damage eventaffecting an oncogene, rather than a noncoding DNAregion).

In reality, both assumptions are likely to be a bitoversimplified. For example, toxicokinetic variabilitywill still lead to differences in tissue dose for a givenexposure, and different people may have different ratesof repair of DNA damage on top of the random nature ofthe cancer process. Nonetheless, the approach for geno-toxic carcinogens has generally considered that themodels used are sufficiently conservative that sensitivepopulations are protected (U.S. EPA, 1986, 1996, 1999).Sources of conservatism include (1) dose-response as-sessments for cancer generally rely on chronic animalbioassays done at high doses (where metabolism maybe saturated), so toxicokinetics and other parametersmay not be representative of those at low doses; (2) thelinear extrapolation is described as a "plausible upperbound estimate of risk at low dose where true risk maybe lower, including zero" (U.S. EPA, 1999); and (3) itis assumed that no threshold exists for nongenotoxiccarcinogens. Another source of conservatism for bothcancer and noncancer risk assessment is that the mostsensitive species, strain, and sex is often used, unlessthere is evidence that the data are not applicable tohumans.

The traditional factor of 10 for human variabilityin noncancer assessment has typically been replacedwhen adequate data exist to do so, such as when dataare found on known sensitive subgroups (e.g., nitrateRfD, U.S. EPA, 2001). Efforts are now under wayto formalize this replacement on a more systematicbasis with a factor that more accurately representshuman variability. "Data-derived" uncertainty factorswere developed (IPCS, 1994; Meek, 1994; Renwick andLazarus, 1998) that divide the intrahuman UFH intoequal factors of 3.16 for toxicokinetics and toxicodyna-mics, based on the earlier work of Renwick (1991, 1993).When chemical-specific data are available, the toxicoki-netic or toxicodynamic components may be replacedwith factors derived from the data (e.g., IPCS, 1998;Dourson et al., 1998; Murray and Andersen, 2001). Aninternational effort under the auspices of the interna-tional Programme on Chemical Safety has defined thedata requirements for development of data~erived un-certainty factors, now referred to as chemical-specific

adjustment factors (CSAFs)2 to replace default uncer-tainty factors for interspecies differences and humanvariability (Meek et al., 2001; IPCS, 2001).

Another approach used to account for human vari.ability is to use probability distributions of uncertaintyfactors to characterize the population and hence UFH(Baird et al., 1996; Slob and Pieters, 1998; Price et al.,1997; Swartout et al., 1998). One approach to character-ize the human distribution (and UFH) is based on toxi-cological "first principles," using data on heterogeneityin animals and assumptions about the relationship be.-tween animal and human heterogeneity (Baird et al.,1996; Baird, 2001). Another approach is to estimate adistribution of UFH based on the U.S. EPA definitionof the Rffi (Swartout et al., 1998; Price et al., 1997). Inbrief, a log-normal distribution is assumed, with the dis-tribution parameters set such that the median is 10°.5and the 95th pereentile is 10. Slob and Pieters (1998)used similar assumptions about the shape and widthof the distribution to estimate a distribution of UFH.Hattis and colleagues have been collecting data for anumber of years on human variability in parametersrepresenting steps in the pathway from external expo.sure to production of biological response (e.g., Hattisand Silver, 1994; Hattis and Barlow, 1996; Hattis et al.,1999a,b). Using this database, they estimated that ifthe population distribution is normal out to the extremetails, a dose 1/10 that corresponding to a 5% effect levelwould be associated with an effect incidence rangingfrom slightly less than 1/10,000 (for a median chemical/response) to an incidence of a few per thousand (forchemicals with high interindividual variability) (Hattiset al., 1999b).

Other authors have used physiologically based phar-macokinetic (PBPK) modeling, sometimes in combina-tion with Monte Carlo analysis, to evaluate the com-posite effect of variability in a number of physiologicalparameters. Dankovic and Bailer (1993) used a PBPKmodel for methylene chloride to evaluate how exer-cise and variability in metabolism by glutathione s.transferases (GST) affects the calculated tissue doseand therefore the cancer risk. Clewell and Andersen(1996) reviewed the use of Monte Carlo analysis withPBPK modeling to determine distributions of risks andeffect levels due to parameter variability and uncer-tainty in PBPK models.

Genetic variability can make an important contribu-tion to human variability, such as in the form of poly-morphic genes for metabolism or repair. Although it haslong been recognized that genetic polymorphism plays

2 Abbreviations used: CYP, cytochrome P450; PBPK, physiologi-cally based pharmacokinetic; Vmax, maximal enzyme velocity; Km,Michaelis-Menten constant; CSAF, chemical-specific adjustmentfactors; GST, glutathione S-transferase; COHb, carboxyhemoglobin;DPX, DNA-protein cross-links; AChE, acetylcholinesterase;DFP, diisopropylfiuorophosphate; DCA, dichloroacetic acid; GSTZ,glutathione transferase~; MTBE, methyl tert-butyl ether.

179POLYMORPHISM IN RISK ASSMSMENT

It is also important to note what this paper does notaddress. Consideration of the effects of polymorphismson toxicodynamics (defined for these purposes as theaspects that affect the body's response to the chemi-cal) is beyond the scope of this paper. The impact of ge-netic polymorphisms on toxicodynamics may be morecomplex than the effects of polymorphisms on toxico-kinetics.

METHODS

Potential candidate compounds were identifiedthrough a multiple-step screening process. In the firststep, recent reviews (Puga et al., 1997; Daly et ai.,1998; Eaton and Bammler, 1999; Ingelman-Sundberget al., 1999; Omiecinski et ai., 1999; Tanaka et al.,2000; Wormhoudt et al., 1999) were used to identify alist of metabolizing genes with demonstrated polymor-phisms affecting activity. Data from these review ar-ticles and studies identified through supplemental lit-erature searches in Medline and Toxline were used tocompile an initial list of 17 candidate toxicologically sig-nificant chemicals that are substrates for polymorphicenzymes (Table 1).

This initial list of candidate compounds was furtherculled, based on the strength of the database for eachsubstance, as evaluated by applying the following fourcriteria. First, the metabolic pathway had to be well-characterized, including identification of the isozymeinvolved in all major steps. For example, it was not

TABLE 1Data Availability for Candidate Chemicals

Key Allelicenzymes frequency data Phenotype PBPKidentified for key enzyme data modelChemica!

an important role in driving the variability in xenobi-otic metabolism, this awareness has typically not trans-lated into the use of these data in a quantitative sensefor risk assessment. Instead, CSAFs are based on ratiosof a critical metric for the mean of the main group andpercentiles for the whole population (Meek et al., 2001;IPCS, 2001). Although few assessments have been de-veloped to date using CSAFs, this approach acknowl-edges the applicability of incorporating data on key pa-rameters, such as polymorphism data, in the context ofa PBPK model to estimate population variability in thedose metric of interest.

The pharmaceutical field has recognized for years theimportance of genetic polymorphism of genes that mod-ulate drug kinetics, resulting in the growing field ofpharmacogenetics. In the environmental area, a con-tribution of genetic variability was recognized as earlyas 1985 (Calabrese, 1985), but researchers have only re-cently begun to investigate the quantitative effect of ge-netic variability on tissue dose for individual chemicals.Renwick and Lazarus (1998) examined classical phar-macokinetic parameters (e.g., clearance or area underthe curve) for genetically different populations exposedto a number of pharmaceuticals and calculated the frac-tion of the exposed population that would not be cov-ered by a 3.16-fold factor for toxicokinetic variability.One approach used by several groups (e.g., Heijmanset al., 2000; Katoh et ai., 2000; London et al., 2000)has been to conduct epidemiology studies evaluatingthe association between the presence of polymorphismsand increased risk of an adverse outcome, such as car-diovascular disease or cancer associated with a spe-cific chemical exposure. Another approach is to evaluatehow genetic polymorphisms affect the tissue dose of thetoxic agent and incorporate that determination into thetraditional risk assessment paradigm (Dankovic andBailer, 1993; El-Masri et ai., 1999).

In this paper, we use a case study approach to iden-tify critical issues and data needed for the quantitativeuse of data on polymorphisms in metabolic enzymesin risk assessment. This effort outlines the extent towhich existing data can lead to informative applica-tion of genetic polymorphism for quantitative risk as-sessments by identifying areas in which simplifying as-sumptions are likely to be needed and by identifyingminimum data requirements that are likely to be re-quired. An additional purpose of this work is to identifychemicals amenable to a more quantitative analysis.Currently ongoing work, to be described in a follow-uppaper, will use data from these case studies, togetherwith appropriate PBPK models, to evaluate how thepolymorphisms affect predicted tissue dose, and the im-plications for UFH and for cancer risk assessment. Inparticular, the follow-up work will explore the quanti-tative relationship between the enzyme variability ob-served in vitro and the resulting variability in tissuedose.

AnalineArsenicAtrazine

YesYesYes

PartialN/AYes

NoYesAnimal

onlyNoYesNoNov-

PartiaJONoPartial

YesYesYesYesYes

Yes

Partial

Yes

Yes

Yes

Yes

Partial

Yes

Yes

Yes

YesYesYes

PartialPartialYes

PartialPartialYes

Benzidine1,3-ButadieneChlorpyrifosDiazinonDichloroacetic

acidMethoxychlorMTBEMethyl chloride

NoYesAnimalonlyYesy~ Yes YesMethylene .J --- --;

chlorideParathion Yes Yes Yes YesPhenol Yes Partial Partial PartialStyrene Yes Partial Partial YesTrichloroethylene Yes Yes Yes YesWarfarin Yes Yes Yes Partial

a Partial indicates that some, but not all, of the key information isavailable.

180 HABERETAL.

weighted average of the groups for which data wereavailable.

RESULTS

Methylene Chloride (Dichloromethane)

Methylene chloride is used extensively as an indus-trial solvent and paint stripper. The liver is the primarytarget of chronic oral and inhalation exposure. Neuro-toxicity is the primary effect following acute high-levelexposure, although liver toxicity is also observed (re-viewed in ATSDR, 1999; IARC, 1999). The productionof carboxyhemoglobin (COHb) from methylene chlo-ride metabolism can also result in toxicity, particu-larly acute effects. Methylene chloride has not beenshown to cause tumors in humans, but NTP (1986)concluded that there was "clear evidence" for carcino-genicity in male and female mice, based on increases inalveolarlbronchiolar adenomas and carcinomas and inhepatocellular adenomas and carcinomas. There wasalso "clear evidence" of carcinogenicity in female ratsand "some evidence" for carcinogenicity in male rats,based on increases in benign neoplasms of the mam-mary gland in both sexes.

As shown in Fig. 1, methylene chloride is metabolizedvia two major pathways (Gargas et al., 1986). A high-affinity, low-capacity pathway is mediated by CYP2El,producing carbon monoxide, which forms COMb in theblood. A lower affinity, higher capacity pathway occursin the cytosol via GST, producing formaldehyde andcarbon dioxide. Pharmacokinetic modeling studies havefound that tumorigenicity correlates with productionof metabolites in the lung and liver via the GST path-way and that the production of metabolites via the CYPpathway does not affect tumorigenicity (Reitz, 1990;Andersen and Krishnan, 1994; Casanova et ai., 1996).The lung carcinogenici ty of methylene chloride in mice,but not in rats, has been attributed to the greater de-gree of metabolism occurring via the GST pathway inmice.

sufficient to know that a cytochrome P450 (CYP) cata-lyzed a specific step; identification of the enzyme asCYP3A4 or CYP2C9, for example, was necessary. Sec-ond, allelic frequency data had to be available for allenzymes playing a major role in the metabolism of thecompound. Although the initial selection of the chemi-cal was based on a polymorphic enzyme being involvedin the chemical's metabolism, this second criterion re-quired that other enzymes involved in the chemical'smetabolism not have significant uncharacterized poly-morphism. Third, at least some phenotype data (ie.,kinetic parameters such as the Vmax and Km) had to beknown for the proteins encoded by each major variantallele. At this step, we did not require that these ki-netic parameters be known for the chemical of interest,since that was part of the literature review in the casestudy. Fourth, an existing PBPK model had to be avail-able, or an existing model for a related compound hadto be readily adaptable. The list was further culled byeliminating chemicals that are primarily metabolizedby CYP2E1, since there is considerable variability inthe activity of this enzyme that is not due to polymor-phism, and focusing on polymorphism data would ig-nore a major source of variability. In addition, an at-tempt was made to avoid chemicals with very complexmetabolic pathways involving a number of polymorphicenzymes, although we were unable to completely avoidchemicals metabolized by multiple enzymes. Three ofthe candidate chemicals (chlorpyrifos, diazinon, andparathion) had similar mechanisms of toxicity. Of these,parathion was chosen, since it is the chemical with thebest-characterized metabolism. Based on this screen-ing process, we chose four diverse compounds for thisstudy, methylene chloride (dichloromethane), warfarin,parathion, and dichloroacetic acid. As described in therest of this paper, the literature on the toxicity and toxi-cokinetics was reviewed in detail for each chemical, inorder to determine the feasibility of conducting a morequantitative analysis using a PBPK model to determinevariability in tissue dose.

Overall genotype frequencies were calculated fromallele or genotype frequency data provided in thereferences cited within the respective table. Where al-lele frequency data were provided, the correspondinggenotype frequencies were calculated assuming Hardy-Weinberg equilibrium, ie., using the equation 1 = (fre-quency of allele 1 + frequency of allele 2 . . . + frequencyof allele n)2. The U.S. average genotype frequency wascalculated as the sum of the genotype frequency foreach ethnic group multiplied by the percentage of theU.S. population represented by that ethnic group. Thepercentage of the U.S. population represented by eth-nic groups was used as reported in El-Masri et at.(1999): Caucasians 72.5%, African Americans 12.2%,Hispanics 11.4%, and Asian Americans 3.9%. If no datawere available for a particular group, then the contri-bution of that group was considered to be equal to the

0

.~, ;c~'o

+ H+ + CIH' CI

GSH

co! + CI

FIG. 1. Summary of methylene chloride meta1X>lism. Adaptedfrom Gargas et oJ. (1986).

181POLYMORPHISM IN RISK ASSESSMENT

TABLE 2Population Distributions of GS1'T Genotypesa

Genotype frequency

Population +/+ ,-if/'"'" "'/~

0.190.220.100.620.20

0.490.500.430.330.48

CaucasianAfrican AmericanHispanicAsian AmericanU.S. averageb

0.310.280.470.050.32

a Adapted from data presented in EI-Masri et al. (1999) and Nelsonet al. (1995) and assuming Hardy-Weinberg equilibrium, where "+"and "-" refer to the wild-type and null alleles, respectively.

b U.S. average calculated as for Table 1.

The GST isozyme that metabolizes methylene chlo-ride has been identified as GSTT1-1 (Meyer et al.,1991); other human liver GSTs appear to be unableto substitute for GSTr1-1 if this isozyme is absent(Bogaards et al., 1993). GST is also involved in a sec-ondary reaction of the first pathway (see Fig. 1), re-sulting in the production of carbon dioxide. There arespecies-specific differences in lung GSTr1-1 activity,with significant levels seen in the rat, but low expres-sion in the human lung, where activity is approximately1 order of magnitude lower than that in the human liver(Mainwaring et al., 1996; Sherratt et al., 1997).

A polymorphism in GSTrl has been well-eharacter-ized, and the population distribution of genotypesis shown in Table 2. "Nonconjugators" lack theGSTT1-1 enzyme (Their et al., 1991; Bogaards et al.,1993; Hallieretal., 1994), and the incidence of the poly-morphism has been characterized in a number of differ-ent ethnic groups (Warholm et al., 1994; Nelson et al.,1995; Bruhn et al., 1998) as shown in Table 2. Heterozy-gotes, who have one positive allele and one null allele("low conjugators"), have half the activity of the homozy-gous "high conjugators" (Wiebel et al., 1999), indicatinga significant gene-dosage effect.

Methylene chloride forms an interesting case studybecause an extensive amount of toxicokinetic data areavailable, and these data have been used to develop anumber of PBPK models of increasing sophistication(e.g., Andersenetal., 1987; Reitzetal., 1988, 1989, 1997;Andersen and Krishnan, 1994; Dankovic and Bailer,1994; Casanova et al., 1996). Even the earliest model(Andersen et al., 1987) predicted both parent chemicallevels and production of metabolites by both pathways.Several authors have used PBPK models to evaluate theeffect of the GSTT1 null polymorphism on tissue dose,and on cancer risk, either as point estimates (Dankovicand Bailer, 1994) or as population distributions(EI-Masri et al., 1999; Jonsson and Johanson, 2001a).

EI-Masri et at. (1999) used the model of Casanovaet at. (1996) to estimate the amount of DNA-proteincross-links (DPX) formed from exposure to methylene

chloride. The cancer potency of methylene chloride, us-ing DPX as a dosimeter, was calculated based on theNTP (1996) mouse bioassay. Monte Carlo modeling wasused to calculate the distribution of human risk, takinginto account the distribution ofGSTT1 null in differentethnic groups in the United States, the ethnic distribu-tion of the U.S. population, and variability in a numberof physiological parameters. Variability in enzyme ac-tivity was included only as the presence or absence ofthe GSTT1 activity; reduced activity of heterozygotesand variability in activity among those with the GSTT1activity were not taken into account. Including the dataon the polymorphism resulted in a biphasic risk distri-bution, with a peak at 0 risk (for the GSTr1 null group)and a second, broad peak for the GSTr1 positives, rep-resenting the variability in other physiological param-eters. Furthermore, at exposure levels of 1-1000 ppm,including the GSTr1 polymorphism resulted in averageand median population risk estimates 23-30% lowerthan when the polymorphism was not included.

Jonsson and Johanson (2001a) enhanced the ap-proach of EI-Masri et at. (1999) in a number of waysand applied their model to a Swedish population. ThePBPK model was enhanced and variability in themetabolic parameters was determined by collectingdata on 27 lean and obese males (Jonsson et at., 2001;Jonsson and Johanson, 2001b). A Bayesian technique,Markov-chain Monte Carlo simulation, was used tosimultaneously fit a population PBPK model to all ofthe toxicokinetic data. In addition to the GSTT1 nulland high-activity groups, the intermediate activity ofheterozygotes was taken into account. Variability in theprotective CYP2E1 pathway was also included. MonteCarlo simulations were used to predict the distributionof cancer risks. This study found that, for the Swedishpopulation, inclusion of the polymorphism decreasedthe mean population risk by 30% or more.

The work of EI-Masri et at. (1999) and Jonsson andJohanson (2001a) demonstrates that Monte Carlo ana-lysis can be productively combined with PBPK mod-eling to evaluate the effect of human polymorphismson risk from environmental chemical exposures. Bothstudies used cancer risk as the endpoint of concern. Fur-ther analysis could evaluate human variability in theproduction of CORb from acute exposure to methylenechloride. While the GSTT1 null allele decreases can-cer risk from methylene chloride exposure, it could in-crease CORb production, since all methylene chloridemetabolism would proceed via the CYP2E 1 pathway.However, the magnitude of the increase would be small,since the CYP2E1 pathway predominates at low expo-sure levels. Instead, variability in CYP2E1 metabolismwould be expected to dominate variability in CORb pro-duction, and this variability could also be incorporatedinto the modeling.

Although polymorphisms in CYP2E1 have been iden-tified (Daly et at., 1998; Wormhoudt et ai., 1999),

182 HABER ET AL.

the functional significance of CYP2El polymorphismsis unclear (Inoue et al., 2000; Tanaka et al., 2000).Inoue et al. (2000) found that three CYP2El polymor-phismB had no effect within ethnic group (Japanese vsCaucasians) on CYP2El protein expression or activity,but significant differences in enzyme activity betweenJapanese and Caucasians indicated that some polymor-phism (as yet unidentified) affecting CYP2El expres-sion or activity differs between these two groups. Moreimportantly, CYP2El expression is inducible by a va-riety of chemicals, including ethanol and chlorinatedsolvents, and individual activity of CYP2El is knownto vary widely (Snawder and Lipscomb, 2000). Thus,variability in CYP2El activity may be dominated byenvironmental or other factors that regulate CYP2El,rather than genetic polymorphism ofCYP2El itself.

Thus, the results with methylene chloride demon-strate that the approach envisioned for the otherchemicals is feasible, at least for chemicals with asingle well-defined polymorphism. Additional modifi-cations to the approach could extend the analysis fromthe carcinogenic effects of methylene chloride to itsnoncancer effects.

WarfarinWarfarin is a cownarin derivative used clinically as

an anticoagulant; its anticoagulant properties also formthe basis for its application as a rodenticide. The mech-anism of action is well understood and involves inter-ference with vitamin K metabolism (a necessary cofac-tor for the synthesis of clotting factors) (reviewed inSutcliffe et ai., 1987; Haustein, 1999; Redman, 2001).Warfarin's hwnan toxicity is also related to its an-ticoagulant activity. The u.S. EPA's warfarin RiD of0.3 .ug/kg-day is based on this endpoint, using a UF of10 to account for hwnan variability (U.S. EPA, 2001).In addition to hematological effects, warfarin is also adocumented human teratogen (Hall et ai., 1980; Abbottet ai., 1977; Whitfield, 1980). Warfarin can exist as the(R) and (8) enantiomers (Fig. 2). Although warfarin is

administered clinically (and is used as a pesticide) asthe racemic mixture, approximately 75% of its activityis attributable to the (B) enantiomer (Choonara et ai.,1986). The parent compound is the active form of war-farin, and it is inactivated by metabolism.

Warfarin is metabolized in the liver. Each CYP en-zyme involved in the metabolism of warfarin acts specif-ically to hydroxylate certain carbons, with a stereose-lective preference for a specific enantiomer, as shownin Fig. 2 (Wang et ai., 1983; Rettie et at., 1992;Kunze and Trager, 1996; Kaminsky and Zhang, 1997).More than 80% of (B)-warfarin metabolism is catalyzedby CYP2C9 (Black et ai., 1996; Rettie et at., 1992),primarily through 7 -hydroxlation, but also through6-hydroxylation (Rettie et ai., 1992; He et ai., 1999). Inan elegant series of experiments, Rettie et at. (1992)expressed 11 human CYP genes (but not CYP2C18or CYP2CI9) in human liver HepG2 cells, in orderto identify the specific enzymes involved in warfarinmetabolism and to identify the products of each en-zyme. They found that CYP2C9, CYP1A2, and CYP3A4all act on (B)-warfarin, but that the CYP2C9 accountsfor the majority of the in vitro activity. CYP2C9 gen-erated the (B)- 7 -hydroxy and (B)-6-hydroxy metabo-lites in a ratio of approximately 3.5:1. The Km valuesfor 7-hydroxylation of (S)-warfarin in HepG2 trans-fected with human CYP2C9 and in human liver mi-crosomes were very similar, as were the Km valuesfor 6-hydroxylation of (B)-warfarin and the ratio of theVmax for (B)-7-hydroxylation and (B)-6-hydroxylation.Together, these data suggest that CYP2C9 is the prin-cipal enzyme involved in metabolizing (B)-warfarinin vivo.

The metabolism of (R)-warfarin is more complex andnot quite as well-characterized. (R)- Warfarin is metabo-lized by CYP1A2, 2C18, 2C19, and 3A4, with CYP2C19identified as a high-affinity 8-hydroxylase (Wienkerset at., 1996). CYP3A4 has a similar affinity for (R)-warfarin, but acts as a 10-hydroxylase (Kunze andTrager, 1996; Kaminsky and Zhang, 1997); the affinityof CYP1Al and 1A2 for (R)-warfarin is approximately

CYP2C19

'"CYP1A1 ---CVP1A2

'",~ci~\ .:

/CYP1A1CYP1A2CYP2C19

~CVP3M

(8)

FIG. 2. Enzymes involved in warfarin metabolism and site of oxidation. Adapted from data in Rettie et aI. (1992); Kaminsky and Zhang1997); Kunze et aI. (1996); Sullivan et aI. (1996); and Wienkers et at. (1996). Metabolism occurs priIDari1y via the enzymes shown in bold.


TABLE 3Kinetic Data for (B)-Warfarin of CYF2C9 Alleles Using 7-Hydroxylation as the Prototype and

Selected Data for (H)-Warfarin, Using 4'-Hydroxylation

Kinetic parameter

Vmax/ Km relativeto wt in same

study (%)

Vmax/Km(pmol/min/nmol

P450//LM)

Km

(/LM)

Vmax(pmol/min/nmol

P450)P450 form

( S)- Warfarin212:i: 7.4280:i:l

133.3:i: 6.1421 :i: 21421:i:3

0.35 (turnover No.)0.097 (turnover No.)

165.6 :i: 5.227:i: 1.247:i: 1.1

145:i:40.57 (turnover No.)0.060 (turnover No.)

41:i:3.967:i: 15

180.5 :i: 8.0246 :i: 58

100100100100100N/AN/A11515.418.1N/AN/AN/A3.J5.9

17.4

34.2

CYP2C9*1 (wt) 6.0:f: 0.43112.6 :f: 0.3b

11.6 :f: 0.9"4.1:f:0.~3.4 :f: 0.0"

4.0f4.OK

12.5 :f: 0.7"1.7:f: O.ad2.1:f:0.~

6:f: 1"2.7f3.8'

30 :f: 4.3410.4:f: 1.7b92.3 :f: 5.5c6.6 :f: 0.4b

3510811.5

103124N/AN/A13.215.922.424.2N/AN/A1.36.42.037

CYP2C9*2 (Rl44C)

CYP2C9*3 (I359L)

CYP2C9*1/CYP2C9*3 (50/50)

(R)-Warfarin212.8:1: 203.2

82.6 :1:37.9384 :I: 311

CYP2C9*1 (wt)CYP2C9*2 (Rl44C)CYP2C9*3 (l359L)

20.2 :J: 1.0"3.4:J: 0.5c

10.6:J: O.~

10.524.336.3

100231346

Note. N/A, not applicable; wt, wild-type.d Haining et al. (1996), baculovirus/insect cell system, purified enzyme.b Takahashi et al. (1998b), yeast expression, microsomes.C Sullivan-Klose et al. (1996), yeast expression, microsomes.d Rettie et al. (1994), HepG2 cells, cell lysate.e Rettie et al. (1994), HepG2 cells, particulate preparation.r Crespi and Miller (1997), baculovirus/insect cell microsomes.g Crespi and Miller (1997), lymphoblast microsomes.h Rettie et al. (1999), expressed in insect cells, purified enzyme.

fivefold lower (Zhang et ai., 1995). (R)-Warfarin alsobinds to CYP2C9 with an affinity similar to that of(B)-warfarin (Kunze et ai., 1991), but it is poorly me-tabolized by this enzyme, with a Km more than 3 or-ders of magnitude higher than the Km for (S)-warfarin(Sullivan-Klose et ai., 1996). The resulting Ki for theinhibition of (B)-warfarin metabolism by (R)-warfarin,as measured in this system was 6.0-6.9 .uM (Kunzeet ai., 1991). Downstream metabolism of warfarin in-cludes reductive metabolism of the acetonyl side chain,glucuronidation, and sulfation (reviewed by KaminRkyand Zhang, 1997; Redman, 2001), but the rate-limitingstep is inactivation of warfarin by the CYPs.

Polymorphisms have been identified for several ofthe enzymes in warfarin metabolism, and the clinicalrelevance of the polymorphisms has been investigated(Aithal et ai., 1999; Furuya et ai., 1995; Steward et ai.,1997; Takahashi et ai., 1998a,b; Taube et ai., 2000).These studies have shown that the polymorphisms dis-

cussed below have clinical significance in identifyingsafe and efficacious doses of warfarin. Table 3 presentsthe kinetic differences in the metabolism of(S)-warfarinand (R)-warfarin by the three human CYP2C9 alleles,as characterized from cDNA expression systems. In or-der to characterize the warfarin kinetics of the indi-vidual isozymes, the cloned human genes have beenexpressed in a baculoviruslinsect cell system (Haininget ai., 1996; Rettie et ai., 1999) or in recombinant yeastmicrosomes (Sullivan-Klose et al., 1996). In the studythat identified the greatest effect of the variant protein(Haining et al., 1996), the Km for the CYP2C9* 3 variantis about five times the wild-type value (lower affinity),and the Vmax is about 5-fold lower, resulting in a 26- to27 -fold reduction in the Vmax/ Km ratio, a measure of thecatalytic efficiencY (also termed the substrate-limitedintrinsic clearance) of that allele.

Table 4 presents allelic frequencies for CYP2C9. Thetwo variant alleles are CYP2C9*2, in which a cysteine

184 HABER ET AL.

TABLE 4Population Distributions of CYP2C9 Genotypes"

Genotype frequencyb

~,-Vl ~~ ~td ~2 ~ *3/*3Population

0.010.740.97ND0.960.78

0.140.02ND

0.100.01ND0.050.09

0.01

ND ND ND

CaucasianAfrican AmericanHispanicAsian AmericanU.S. averaged 0.12 0.01 0.01

Note. ND, no data available.a Data adapted from allelic frequency data reported in Wormboudt

et at. (1999). Allele *1 is the wild-type encoding arginine at position144; *2 encodes cysteine at position 144; and *3 encodes leucine in-stead of isoleucine at position 359.

b Calculated from allelic frequencies assuming Hardy-Weinberg

equilibrium.C Frequency less than 0.01.d U.S. average genotype frequency was calculated as the sum of the

individual products of the genotype for each ethnic group and thepercentage of that ethnic group in the U.S. population. Contributionsof each group to the total U.S. population were done as reported in EIMasri et at. (1999).

replaces an arginine at position 144, and CYP2C9*3, inwhich a leucine replaces the isoleucine residue at posi-tion 359. There is general agreement in the literaturethat the CYP2C9*3 allele is associated with the poormetabolizer phenotype (Aithal et al., 1999; Stewardet al., 1997; Takahashi et al., 1998a,b; Taube et al.,2000), and these in vivo data are supported by in vitroenzyme kinetic data (Haining et al., 1996; Sullivan-Klose et al., 1996).

Although the data for the CYP2C9*2 allele are lessclear, this allele appears to result in a smaller decreasein warfarin metabolism. Furuya et at. (1995) foundthat patients with one allele of the CYP2C9*2 varianttended to require a lower maintenance dose of warfarinthan patients with two wild-type alleles, but the differ-ence was relatively small and there was considerableoverlap in the distribution of required doses. Taube et at.(2000) found that patients heterozygous for CYP2C9*2tended to require a lower maintenance dose of war-farin, and the few patients homozygous for CYP2C9*3required even lower doses. The in vitro data are mixed.Using cloned genes in a HepG2 expression system,Rettie et at. (1994) found that both the Km and the Vmaxwere decreased compared to wild type, resulting in a5.5-fold decrease in the catalytic efficiency (Table 3). Alater study by the same group (Rettie et al., 1999) foundno effect on Km and a marginal decrease in Vmax for thepurified CYP2C9*2 enzyme expressed in insect cells, incomparison with the results of Haining et at. (1996) forthe wild-type allele in the same expression system. Us-ing a yeast expression system, Kaminsky et at. (1993)found that the rate of metabolite formation for the vari-ant was approximately twice that of the wild type, but

the variant did not affect regioselectivity or stereose-lectivity. Together, the in vivo and in vitro data indicatethat clearance of(S)-warfarin would be markedly slowerin people with the CYP2C9*3 allele and that clearancein people with the CYP2C9*2 allele would be slightlyslower, if at all affected at environmental exposure lev-els. A minimal effect of the CYP2C9*2 allele makesbiological sense, since the affected amino acid (num-ber 144) is not in any known substrate recognition site,while residue 359 (affected in the CYP2C9*3 variant) islocated close to the proposed substrate recognition site 5(Gotoh, 1992). Crespi and Miller (1997) investigated thedifferences between the wild-type and CYP2C9*2 vari-ant in greater detail by comparing kinetics of cDNA-expressed microsomes in a baculoviruEVinsect cell sys-tem and in human lymphoblas~ (Table 3). They foundthat the Km of the variant was similar to that of thewild type, but slightly lower, and that the Vmax of thewild type was higher. They also found that the ratio ofactivity between the wild type and variant varied withthe expression system and that the activity of the wildtype was strongly dependent on the amount ofNADPH-cytochrome P450 oxidoreductase present, while the ac-tivity of the CYP2C9*2 allele was essentially unaffectedby the amount of oxidoreductase.

Although functional polymorphisms have been iden-tified for CYP2C19 (e.g., Roh et ai., 1996), the lim-ited available data indicate that these polymorphismsare not clinically significant for warfarin metabolism.Takahashi et at. (1998a) determined the CYP2C19genotype (wild type or containing one or both variantalleles) for 76 Japanese patien~ receiving warfarin andmeasured the unbound oral warfarin clearance. No sig-nificant difference in (R)-warfarin clearance was foundbetween the 56 patien~ with at least one wild-type al-lele and the 10 patien~ with only variant alleles. Thestudy authors concluded that the CYP2C19 polymor-phism plays an insignificant role in variability in (R)-warfarin metabolism in vivo. This conclusion is limitedby the small sample size. In addition, sensitivity to poly-morphic differences is reduced, since the authors pooledheterozygotes containing one wild-type and one variantallele together with the wild-type homozygotes. In sup-port of this idea, data for the CYP2C9*3 allele showthat clearance in the heterozygote is intermediate be-tween that of the wild type and the homozygous vari-ant (Takahashi et ai., 1998b). Another reason that theCYP2C19 polymorphism is likely to have a relativelysmall impact on (R)-warfarin metabolism is that theCYP 1A1 and CYP1A2 isozymes also act at the samesites; polymorphisms in CYP1A1 have been reported tohave a minimal effect on (R)-warfarin metabolism (re-viewed in Redman, 2001).

Although a fully developed PBPK model for warfarinthat includes metabolism has not been published, itappears that the data are available to develop such amodel. Blacket at. (1996) evaluated the toxicokinetics of


warfarin in six human subjects and calculated the half-life, clearance, AUC, and volume of distribution for (R)-and (8)-warfarin. Lueckeetal. (1994) developed aPBPKmodel for human pregnancy that took into account howbody parameters change with time and used warfarin asa test case. Warfarin metabolism in the liver was mod-eled as a single compartment, using rat metabolism pa-rameters from Luecke et al. (1980). The human modelhas not been validated, and it appears that no allomet-ric scaling was done for the metabolic parameters. Arat PBPK model for intravenous dosing has been de-veloped, validated, and used to model the decrease ofwarfarin elimination by bromosulfophthalein (Lueckeand Wosilait, 1979). Mungall et al. (1985) also developedan empirical toxicokinetic model to describe warfarinmetabolism, and Chan et al. (1994) described an em-pirical toxicokinetidtoxicodynamic model for warfarin.However, these models were not based on physiologi-cal parameters and so cannot be adapted for the cur-rent project. Kunze and Trager (1996) determined thekinetic parameters (Vmax and Km) for the formationin human liver microsomes of (8)-7 -hydroxywarfarinfrom (8)-warfarin and for the formation of (R)-6,(R).7, (R)-8, and (R)-10-hydroxywarfarin from (R)-warfarin.

An interesting aspect of warfarin metabolism is thedifference in activity of the (R) and (8) enantiomers andtheir differences in metabolism. This means that expo-sure to a racemic mixture of warfarin can be consid-ered quantitatively as exposure to two different chem-icals, which are metabolized by different enzymes andwhich have the same effect, but with potencY differ-ing by threefold. The inhibition of the metabolism ofthe (8) enantiomer by (R)-warfarin will also need to beaddressed, since the enzyme kinetics were determinedwith single enantiomers, but exposure occurs to theracemic mixture. This can be done quantitatively us-ing the Kj calculated by Kunze et al. (1991).

A second aspect to consider is that the 2C9 enzymehas three major polymorphic forms. Theoretically, thisincreases the possible allelic combinations from three,for a gene with a wild-type and a single variant allele,to six. Although this makes the resulting mathemat-ical calculations of overall variability more complex,the principle of the evaluation is unaffected. The pres-ence of multiple alleles with differing activities is alsolikely to drive the population distribution toward beingone broad distribution, rather than a bimodal distribu-tion with clearly separated peaks. For this case study,it may be possible to reduce the analysis to an eval-uation of the CYP2C9*1 and CYP2C9*3 alleles. TheCYP2C9*2 allele has a Km in the same range as that ofthe wild type, although the Vmax is somewhat reduced(Table 3). This suggests that at environmental expo-sure levels, the CYP2C9*2 allele will behave very sim-ilarly to the wild-type allele. Quantitatively predictingthe metabolic parameters for different genotypes is also

likely subject to some imprecision in the heterozygotes.Takahashi et at. (1998b) measured the enzyme kineticsof purified wild-type and variant CYP2C9*3, as well asa 1: 1 mixture of the two proteins, and found that theKm of the mixture was the average of the two pure en-zymes, but the Vmax of the mixture was close to that ofthe wild type.

A third aspect of the analysis is that both CYP2C19and CYP3A4 play significant roles in the metabolismof (R)-warfarin. Since warfarin metabolism is an inac-tivating step, metabolism by either enzyme is function-ally equivalent, suggesting that one enzyme can com-pensate for low levels of the other. This is consistentwith the finding of Takahashi et at. (1998a) thatthe CYP2C19 variants did not affect the average un-bound oral clearance. In this case, functional polymor-phisms in CYP3A4 have not been sufficiently well-characterized to use in a quantitative manner (seethe parathion case study, below) and metabolism byCYP3A4 would be expected to compensate for low activ-ity ofCYP2C19. Overall, the effect of multiple enzymescatalyzing the same reaction is to dilute the potentialimpact of any single gene polymorphism.

Finally, warfarin kinetics are altered by a number ofdrugs that induce enzymes that metabolize warfarin orthat are direct inhibitors of warfarin metabolism. Forexample, CYP2C9 is inducible by rifampicin, ethanol,and phenobarbital, while a number of drugs, includ-ing fluoxetine, phenytoin, and tolbutamide, inhibit (8)-warfarin metabolism (reviewed in Miners and Birkett,1998). The inhibition kinetics for some of these chemi-cals have been characterized in vitro (Kunze and Trager,1996) and in vivo (Choonara et al., 1986; Chan et al.,1994; Black et al., 1996). Competitive inhibition wouldnot affect the Vrnax for a chemical, but would affectthe apparent Km. This means that long-term pharma-ceutical exposure (which would generally be at muchhigher daily doses than environmental exposure tochemicals) could have a substantial effect on the kinet-ics for metabolism of environmental chemicals. Whilethis issue is not directly related to the effects of poly-morphisms, enzyme induction and inhibition both con-tribute to overall human variability in enzyme activityand would need to be considered in a complete descrip-tion of human variability.

In summary, warfarin would be a useful case study tocharacterize the contribution of genetic polymorphismto variability in tissue dose. The mechanism of actionis well understood. Phenotype data and allelic frequen-cies are available for alleles of the key metabolic en-zyme, CYP2C9, although other enzymes playing lesssignificant roles have not been as well-characterized.It also appears that a PBPK model can be developedto integrate the polymorphism data on warfarin withother data on human variability. Several issues willneed to be addressed (either quantitatively or by con-sidering alternative assumptions) in developing a more

186 HABER ET AL.

quantitative approach using a PBPK model. These in-clude consideration of the different activities of the (R)and (8) enantiomers, taking into account the three ma-jor allelic forms of CYP2C9, accounting for the con-tributions of CYP2C9 and CYP3A4 in the metabolismof (R)-warfarin, and consideration of the effect onmetabolism of other coexposures that can alter warfarinkinetics.

crosomes prepared from human liver microsomes fromthree individuals. Kinetic parameters for paraoxon for-mation for the individual livers were Vmax values of143.1, 167.8, and 313.5 pmol/min/mg protein and cor-responding Km values of 13.7, 9.0, and 15.9 11M. Ki-netic parameters for p-nitrophenol formation were notcalculated, but following treatment with 20 and 20011M parathion, the median ratios of paraoxon to p-nitrophenol formation were 0.143 and 0.247, respec-tively. Butler and Murray (1997) reported that medianrates of paraoxon and p-nitrophenol formation were3.03 and 2.26 nmol/min/mg, respectively, in human livermicrosomes incubated with 250 11M parathion. Thus,the relative importance of oxidative desulfuration anddearylation reactions remains uncertain.

The identity of CYP isoforms contributing to oxida-tive metabolism of parathion has been investigated inseveral recent studies (Butler and Murray, 1997; Mutchet al., 1999; Sams et al., 2000). Experiments using hu-man liver microsomes cotreated with CYP inhibitors,and parathion metabolism experiments in human celllines expressing specific CYP isoforms, demonstratethat CYP3A4 activity accounts for the greatest extentof parathion metabolism (Butler and Murray, 1997;Mutchet al., 1999). Other CYP isoforms thought to playa lesser role include CYPlA, CYP2B6, CYP2D6, andCYP2C. Sams et al. (2000) measured the abilityofCYPinhibitors to block AChE inhibition, as a measure ofparaoxon formation, and reported that both CYP2D6and CYP3A4 playa major role in the bioactivation ofparathion.

CYP isoforms capable of metabolizing parathion maynot have identical tendencies to result in desulfu-ration versus dearylation of parathion. For example,Butler and Murray (1997) measured the proportion of

Parathion

Parathion is a prototypical member of the classof organophosphate pesticides. Parathion and otherprominent members of this class, such as chlorpyrifosand diazinon, exert their toxicity primarily through theinhibition of acetylcholinesterase (AChE) in the centraland peripheral nervous system. Chronic syndromes as-sociated with acute high-dose exposures to organophos-phates might also be due to inhibition of the neu-ronal carboxylesterase, neuropathic target esterase.As shown in Fig. 3, parathion is oxidized by CYP tophosphooxythiran, a reactive intermediate. This reac-tive intermediate can spontaneously degrade, either tonontoxic compounds (p-nitrophenol and either diethylphosphate or diethyl thiophosphate) via dearylation orto paraoxon via oxidative desulfuration (Mutch et ai.,1999). Essentially all the AChE inhibition seen follow-ing parathion exposure is ascribed to its metaboliteparaoxon (reviewed in Ecobichon, 1996). Thus, the bal-ance of these alternative end products ofCYP-mediatedoxidation reactions can be an important determinant ofparathion toxicity.

The liver is an important site for the oxidativemetabolism of parathion. Mutch et at. (1999) deter-mined paraoxon formation rates in incubations with mi-

"

~sI O~NO.:p-O- - ~~i~ ..."

P--OH~+ HOONO2

p-Ntrophenol

CH3CH.O .

Diethyt PhO$~eParalhkx1

I 'CYP3A4 PON1t

f~11i 9'

~,: a48CHP. I DesulfuralJon

\

"'" IIO ~ -- Carboxylesterase

/p~o, - NO2 Phosphorylation

Paraoxon

~

~,;T (> \'. ~p.o-. ~ NO~

"""'- \ Dealy\ation

phosphooxyllWran

FIG. 3.


ef 0

IIc,OH

Glycolate

GlycineCO2

~ACI -C~

-""""

~

~1::

HO +t'""'"

,~:' 0H: c."!;#C,~C

CI 0

MCAQ~&J'",~,,~~3",~~

HO OH

Oxalic acid

OHH

1

Thlodiglycolate

FIG. 4. Summary of dichloroacetic acid metabolism. Adaptedfrom Bull (2000) and Lash et al. (2000). Reproduced with permissionfrom Environ. Health Perspect.

American and European American men by measuringplasma concentrations of midazolam. The clearance ofmidazolam following intravenous dosing was 15% lowerin African Americans (4 heterozygous; 10 homozygousfor variant) compared to European Americans (all wild-type homozygous). The individuals homozygous for thevariant allele had 30% lower systemic midazolam clear-ance than subjects homozygous for the wild-type allele.However, no difference in oral clearance was observed,suggesting that variability in systemic metabolism dueto the polymorphism was overshadowed by other fac-tors affecting oral clearance (e.g., intestinal absorptionand metabolism).

Recent papers have identified additional variant alle-les for C¥P3A4. Sata et at. (2000) identified a serine toproline change at codon 222 (referred to as CYP3A4 *2),at a frequency of 2.7% in a Caucasian population, butabsent in African or Chinese subjects. In vitro ana-lysis revealed a six- to nine-fold lower intrinsic clear-ance (Vmax/ Km) of nifedipine for this variant, but nokinetic differences in testosterone 6f3-hydroxylase ac-tivity, demonstrating a clear substrate dependence forfunctional differences.

Several additional variant alleles have been reportedat low frequencY in Chinese populations. Sata et at.(2000) reported a methionine to threonine change atcodon 445 (CYP3A4*3) in a single Chinese subject.Hsieh et at. (2001) identified three variant alleles in agroup of 102 Chinese subjects: CYP3A4 *4 (lle118Val),CYP3A4 *5 (Pro218Arg), and CYP3A4 *6 (A17776 inser-tion). The CYP3A4 activity in these individuals wasmeasured by 6f3-hydroxycortisol/cortisol ratio and wascompared to the general Chinese population. Individ-uals harboring variant alleles (all heterozygous) hadlower metabolic ratios, providing preliminary evidencefor a functional effect of these variants. Further studieswill be needed to confirm the presence of these variantalleles in larger populations.

Taken together, evidence is increasing for functionalpolymorphisms in CYP3A4. However, the data are notsufficient for evaluation of their impact on parathionmetabolism for two reasons. First, in all cases exceptCYP3A4 *lB, which probably has limited if any func-tional significance, allele frequencies are low, and thusthe population frequencies of the variants need addi-tional study. In addition, functional implications ofvari-ants are substrate-dependent, limiting the appropri-ateness of generalizing to untested substrates such as

parathion.Paraoxon can be detoxified through competing mech-

anisms, enzymatic hydrolysis by paraoxonase (PON1)and phosphorylation of carboxylesterases lacking hy-drolysis activity. Like, the carboxylesterase, the classicAChE inhibition by paraoxon occurs via its phospho-rylation by paraoxon. Early observations of bimodaland trimodal distributions of paraoxonase activity, aswell as pedigree studies, suggested a genetic basis for

paraoxon and p-nitrophenol in microsomes from hu-man cells expressing various human CYP450 isoforms.Based on these data, CYP2D6 and 2B6 appeared tofavor paraoxon formation over p-nitrophenol forma-tion as a percentage of total oxidative metabolitesformed, while CYPSA4 and CYP1A2 formed nearlyequal amounts of each metabolite.

A high degree of variability in CYPSA4 activity hasbeen well-documented, but only recently have dataon a potential genetic basis for this variability begunto be characterized. Although evidence is building forfunctional polymorphism in CYPSA4, the evidence re-mains too preliminary to predict with confidence thedegree to which polymorphism might impact parathionmetabolism.

The first variant allele to be characterized(CYP3A4 *lB) results in a single nucleotide changein the CYP3A4 gene promoter. The functional sig-nificance of this variant allele remains an area ofdebate. Ball et at. (1999) reported no difference inCYPSA4-dependent enzymatic activities, includingerythromycin demethylation or nifedipine kinetics inindividuals homozygous for the wild-type and variantalleles. Westlind et at. (1999) reported interindividualvariation of CYPSA4 activity of SI-fold in 46 humanlivers, but two individuals heterozygous for the vari-ant allele did not have responses different from themedian value for the whole group. von Moltke et al.(2000) reported low clearance values of two CYPSA4substrates, alprazolam and trazodone, in an individualhomozygous for the wild-type allele. Since low clear-ance could be observed even in individuals homozygousfor the wild-type allele, these studies suggest thatmechanisms other than CYP3A4 *lB are responsiblefor variability in CYPSA4 activity.

On the other hand, Wandel et at. (2000) tested thefunctional effect of the CYP3A4 *lB allele in Mrican

188 HABER ET AL.

TABLE 5Population Distributions of PONt Genotypes

Genotype frequency"

Population Q/Q QtR R/R

0.470.350.150.170.41

0.430.480.480.480.45

0.100.170.370.350.15

CaucasianbAfrican AmericancHispani&Asian AmericaneU.S. average'

a Genotype frequency was calculated for codon 192 variant, where

Q refers to glutamine at this position and R refers to arginine. Fre-quencies were based on population or allele frequency data presentedin the cited references and assuming Hardy-Weinberg equilibrium.

b Adapted from Eckerson et al. (1983); Mueller et al. (1983).C Adapted from Diepgen and Geldmacher-von Mallinckrodt (1986).d Adapted from Davies et al. (1996)." Adapted from Sanghera ct al. (1998).f U.S. average calculated as for Table 1.

can contribute to differences in the catalytic efficiencyof PON1.

The amino acid change at position 192 does not ap-pear to completely account for the degree of human vari-ability in human PON1 activity. Although earlier stud-ies had identified as much as a 40-fold variability inparaoxonase activity, remaining variability among in-dividuals with the same low-metabolism genotype stillranged over 13-fold. Furlongetai. (1993) tested whetherthis variation in paraoxonase activity among individ-uals having the same genotype could be explained bydifferences in protein levels. Based on Western blots ofserum from five homozygous low-activity individuals,the level of activity correlated well with the amount ofPON1 present in the serum, providing strong evidencethat both genotypc and protein level will likely affectin vivo levels ofPONl activity. Another possible expla-nation for variability in PON1 activity is the possibil-ity that uncharacterized genetic polymorphisms mightalso playa role. Recently two human genes, PON2 andPON3, have been characterized that are structurallyrelated to PONl (Mochizuki et at., 1998; Primo-Parmoet ai., 1996). No data on the ability of these genes tometabolize organophosphates were identified.

Human sensitivity to organophosphates depends oninteractions with their metabolites and multiple pro-teins. For example, Mutch et at. (1992) measuredvariability in several enzyme activities related toparaoxon kinetics and dynamics. PON1 activity rangedover approximately 6-fold, but effects on AChE andcholinesterase activities were less variable, on the orderof 2-fold. Variability in neuropathy target esterase (apotential marker for delayed neuropathy) was 6.5-fold.

The potentially important role of enzymes other thanPON1 as a determinant of paraoxon toxicity was welldemonstrated by Li et at. (2000). Ponl gene knockoutmice were not more sensitive to paraoxon than wild-type mice, suggesting that in vivo, PON1 is not a majordetoxification enzyme. This finding was supported bythe absence of a protective effect from injections ofrabbit PON1 or human PON1 and by observationsof paraoxon-trea~ transgenic mice overexpressingthe human PON1 gene. In contrast to the resultsfor paraoxon, PON1 was important in modulatingthe toxicity of the chlorpyrifos-oxon and diazoxon inmice, perhaps due to the greater catalytic efficiencyof PON1 for these organophosphate metabolites thanfor paraoxon. This finding underscores the need forsubstrate-specific metabolism data when assessing therole of gene polymorphisms on target dose or effect.Similar results were not observed in rats, in whichPON1 injections did protect against parathion toxicity(Costa et ai., 1990). Thus, there may be additionalspecies-dependent factors that modulate the relativeimportance of PON1 in parathion toxicity.

The lack of an effect of PON1 status in mice sug-gests that other pathways play an important part in

this variability. Adkins et at. (1993) and Humbert et at.(1993) determined that a glutamine (QI92) at position192 of paraoxonase confers a low-activity phenotype,while arginine (RI92) results in high paraoxonase ac-tivity. Heterozygotes have an intermediate activity phe-notype. The distribution ofPONl genotypes in the U.S.population is shown in Table 5.

Several studies have provided quantitative estimatesof human paraoxonase activity. Mueller et at. (1983)determined the rate of formation of p-nitrophenolfrom paraoxon in the serum of high and low paraox-onase activity individuals. They found that the Vmaxwas 200 Ji,mol/min for the low-activity phenotype and420 Ji,mol/min for the high-activity phenotype. The Kmvalues were not significantly different for the twogroups (0.46 roM in the low-activity group and 0.42 mMin the high-activity group). Smolen et at. (1991) char-acterized enzymatic activities of the purified isozymes.Km values for low-activity and high-activity variantswere 0.503 and 0.265 roM, respectively. Paraoxonaseturnover numbers at Vmax concentrations were 344 and659 min-l for low- and high-activity variants, respec-tively. In this same paper, the activity in serum was0.001 Ji,mol/min/mg for low-activity individuals ver-sus 0.005 Ji,mol/min/mg for high-activity individuals.Davies et at. (1996) reported levels of paraoxonase ac-tivity in serum from individuals of varying genotype.The mean paraoxonase activity was 328:t: 79 U/L(Ji,mol/min/L) for low-activity homozygous, 977:t: 171U/L in heterozygous, and 1769:t: 354 U/L in high-activity homozygous individuals. In a recent paper, Liet at. (2000) reported kinetic parameters for purifiedhuman PONI proteins. The Krn was 0.81 mM and theVrnax was 0.57 Ji,mol/min/mg for the low-activity iso-form, while the Krn was 0.52 roM and the Vmax was3.26 Ji,mol/min/mg for the high-activity allele. Togetherthese data suggest that both changes in Km and Vmax

189POLYMORPillSM IN RISK ASSESSMENT

modulating parathion toxicity. Similarly, it has beensuggested that paraoxonase may playa limited role inparathion metabolism in humans at low exposure lev-els, due to its low catalytic efficiency toward parathioncompared to activity toward other organophosphate ox-ons (Li et al., 2000). In support of this idea, Mutchet at. (1999) reported that EDTA-mediated inhibition ofPON1 in human liver microsomes had no effect on theformation of paraoxon or p-nitrophenol from parathionand only a limited effect on p-nitrophenol formationwhen microsomes were treated with saturating (1 mM)concentrations of paraoxon.

One mechanism for removal of paraoxon is throughparaoxon phosphorylating carboxylesterases, ratherthan the phosphorylating target for paraoxon toxicity,AChE. In this way, carboxylesterases serve as a nonen-zymatic sink for the paraoxon, serving to reduce thechance for interactions of paraoxon with the target en-zyme. The capacity of the carboxylesterases to removeparaoxon is therefore a function of the amount of pro-tein available. Tang and Chambers (1999) investigatedthe role of this pathway by measuring p-nitrophenolactivity in rat liver homogenate or rat serum, withor without addition of esterase. At low concentrationsor short incubation times, paraoxon removal via car-boxylesterase phosphorylation was greater than byPON1-mediated hydrolysis, although this nonspecificactivity was rapidly saturable. The importance ofcarboxylesterases was confirmed in vivo, based on thefinding that brain AChE levels were significantly in-hibited after pretreatment with a carboxylesteraseinhibitor. The relative contribution of nonspecific car-boxylesterases and enzymatic hydrolysis by paraox-onase is an important consideration in evaluating thepotential role of PONI polymorphism on susceptibil-ity to parathion toxicity. Sweeney and Maxwell (1999)recently developed a quantitative model to estimatethe relative roles of hydrolase and carboxylesterase ac-tivities in protection against organophosphates, whichmight be useful for further determinations of the rel-ative importance of these two pathways for paraoxondetoxification.

Gearhart et at. (1990) developed a PBPK modelfor the organophosphate diisopropylfluorophosphate(DFP), which was chosen as a model compound to rep-resent the class of highly toxic organophosphate com-pounds, including parathion. In addition to data forDFP, the model was designed to incorporate disposi-tion of parathion as well. Metabolic parameters derivedfrom data in the rat were used to develop the modeland included parameters to describe the metabolismof parathion to paraoxon (liver and kidney compart-ments), paraoxon hydrolysis by paraoxonase (blood,brain, liver, kidney, rapidly perfused tissue), and activ-ity of esterases for nonhydrolytic binding of paraoxon(all tissue compartments except fat). Model simulationsfor AChE inhibiton in humans were not reported for

parathion, although the authors noted that prelimi-nary simulations resulted in general agreement withhuman kinetic data from the literature, after modifi-cation of some of the enzyme parameters. The modelpredictions generally agreed with tissue concentrationsof parathion and paraoxon observed experimentally forthe rat. Opportunities for further optimization of themodel that were discussed included modeling the ef-fects of prolonged exposure on AChE synthesis ratesand assumptions regarding cross-species scaling of en-zyme parameters. This latter issue may be of particu-lar importance, as significant interspecies differencesto parathion have been reported (Johnson and Wallace,1987; Veronesi and Ehrich, 1993).

In summary, parathion provides another example of acompound for which integration of polymorphism datawith PBPK modeling would provide insight into the rel-ative contribution of genetic factors to overall variabil-ity. Parathion is particularly useful as an example ofa chemical for which there are multiple detoxificationroutes. Although the toxic mechanism and metabolismof parathion are well-studied, new information on func-tional CYP3A4 polymorphisms and any genetic factorsaffecting PON 1 expression that become available willallow even greater opportunity to apply this case study.Parathion was chosen as a well-studied member of thefamily of organophosphate pesticides. However, the rel-ative contributions of organophosphate detoxificationmechanisms to the disposition of related compounds,such as chlorpyrifos and diazinon, are not the same asfor parathion. Therefore, it will also be of interest to ex-pand this work to look at other members of this class of

compounds.

Dichloroacetic Acid

Dichloroacetic acid (DCA) is a common drinkingwater disinfectant by-product, resulting in low-level ex-posure of a large proportion of the population (Boorman,1999). DCA has been employed as a therapeutic agentto treat lactic acidosis, diabetes, and familial hyperlipi-demia in humans for many years, based on its apparentability to enhance the activity of pyruvate dehydroge-nase; reported adverse effects are limited to mild seda-tion and mild peripheral neuropathy (Stacpoole et al.,1998). In animals, targets for the noncancer toxicity ofDCA include metabolic alterations, liver toxicity, testic-ular effects, neurotoxicity, and, at higher doses, devel-opmental toxicity (Cicmanec et al., 1991; Linder et al.,1997; Moser et al., 1999; Smith et al., 1992; Epsteinet al., 1992). In animal studies, DCA treatment inducesliver tumor formation at high doses (reviewed in IPCS,2000). Although epidemiological data support a link be-tween ingestion of chlorinated drinking water and riskof rectal, colon, and bladder cancer, chlorinated wateris a complex mixture, and an association with DCA hasnot been established (Mills et al., 1998; Boorman, 1999).

190 HABER ET AL

Studies on the toxicokinetics of DCA (Crabb et al.,1981; Evans and Stacpoole, 1982; Larson and Bull,1992; Lin et al., 1993) have shown that DCA is metabo-lized by oxidative dechlorination to glyoxylate, which isin turn metabolized to oxalate (Fig. 4). This metabolicpathway also occurs in humans, as evidenced by the ap-pearance of oxalic acid in the urine of DCA-treated hu-mans. There has been no evidence ofthiodiacetic acid ormonochloracetic acid excretion in DCA-treated humans(Stacpoole, 1989; Stacpoole et al., 1998; Stacpoole andGreene, 1992)j suggesting that the reductive dechlori-nation pathway is not likely to be important.

Lipscomb et at. (1995) found that DCA metabolismresults from GSH-dependent pathways, rather than ox-idation by CYP. Tong et at. (1998a) determined thatglutathione transferase ~ (GSTZ) is responsible forDCA metabolism in both humans and rodents. Basedon species differences in metabolic capacity (Vmax/Km),GSTZ-dependent metabolism of DCA is highest in themouse, with intermediate rates in rats, and the slowestmetabolism of DCA occurring in humans.

Genetic polymorphism of GSTZ may engender differ-ences in DCA susceptibility within human populations.Blackburn et at. (2000) reported on polymorphism ofGSTZ in a population of 141 Caucasians. Three alleleswere identified, GSTZ*lA, GSTZ*lB, and GSTZ*lC.Based on in vitro experiments with purified proteins,GSTZlA had a 3.6-fold higher activity toward DCA thanthe other two allelic variants. The activities for thedifferent allelic variants were 1610::1:: 20 nmol/min/mgfor GSTZlA, 450:!::20 nmol/min/mg for GSTZ1B, and450:!:: 30 nmol/min/mg for GSTZ1C. The Km valueswere not reported. The effect of this polymorphismon in vivo DCA metabolism in humans has not beenreported. Blackburn et at. (2000) calculated allelicfrequencies of 0.09, 0.28, and 0.63 for GSTZ*1A,GSTZ*lB, and GSTZ*lC, respectively. Thus, assumingHardy-Weinberg equilibrium, the segment of the pop-ulation homozygous for the high-activity allele mightbe small (i.e., 0.8%). No data were identified on allelicfrequencies in other ethnic groups.

Unlike many other chemicals, DCA can inhibit itsown metabolism. Several studies in rodents havedemonstrated that pretreatment with DCA inhibits themetabolism of subsequent doses (Gonzalez-Leon et al.,1997; James et al., 1997). The plasma elimination half-life is also increased in humans following repeated dos-ing with DCA (Curry et al., 1991), suggesting that au-toinhibition also occurs in humansj but the supportingin vitro data are mixed. Cornett et at. (1999) foundin in vitro studies that DCA pretreatment markedlyinhibited its own metabolism in rat cytosol, but theydid not observe any inhibition in human cytosol. Incontrast to this result, Tzeng et at. (2000) observeda dose-dependent inhibition of GSTZ1 using humanliver cytosol. Species differences in DCA-induced inhi-bition were noted, with relative rates of DCA-induced

inactivation of GSTZ reported as rat> mouse> human.Tzeng et at. (2000) also found differences in the rate ofinactivation of the different allelic forms of GSTZ, sug-gesting that these differences may be as important, ormore important, than the differences in catalytic effi-ciency. The rate constant for inactivation of GSTZlAwas 3.0:!:: 0.1 min-l, a rate approximately half that ofthe proteins encoded by the other variant alleles.

Several PBPK models have been developed for DCAor have incorporated elements of DCA metabolism aspart of larger models designed to evaluate the dis-position of trichloroethylene, of which DCA is a mi-nor metabolite. Clewell et at. (2000) included a sub-model for DCA in their human trichloroethylene model.Metabolism and excretion of DCA were considered,but autoinhibition of metabolism was not included inthis model. Mouse models of DCA metabolism also ex-ist (Abbas and Fisher, 1997; Greenberg et at., 1999;Barton et ai., 1999), all developed as part of largertrichloroethylene models. Chen (2000) proposed a bio-logically based dose-response model for liver tumorsinduced by TCE and DCA.

Thus, DCA is a chemical to which many people areexposed and for which there is a polymorphism in thekey metabolic enzyme. The implications of this polymor-phism are likely to be complex and depend on whetherthe key toxic agent is DCA or its metabolite(s) (or both).Comparing different PBPK dose metrics for tumori-genicity and DCA tumor formation, Barton et at. (1999)suggested that the prevalence of DCA-induced tumorswas correlated with the amount metabolized, while thetumor multiplicity appeared to correlate better witharea under the curve of DCA in the liver. These re-sults suggest that there may be different roles in thetumorigenic process for the parent compound and DCAmetabolites. The toxic agent for the noncancer effects ofDCA is also unknown. Cornett et at. (1999) suggestedthat DCA-mediated inhibition of GSTZ might be an im-portant factor in the toxicity of DCA GSTZ is a keyenzyme for tyrosine metabolism, and inhibition of ty-rosine metabolism may result in increased levels of re-active tyrosine metabolites that may adversely affectthe liver and nerves, targets of DCA toxicity. Data fromMoser et at. (1999) suggest that DCA-induced neuro-toxicity (at low doses, evidenced by gait abnormalitiesand impaired righting reflex) may be due to the par-ent compound, rather than a DCA metabolite. Over-all, these data suggest that noncancer effects of DCAat low doses may be attributed primarily to the par-ent compound, while DCA tumorigenicity at low dosesis primarily due to metabolites, and tumorigenicity athigher doses may be enhanced by the parent. However,these data, particularly the data addressing the toxicmoiety for noncancer effects, are preliminary.

An issue to consider in evaluating the implicationsof the GSTZ polymorphism is how to accoUnt for DCAinhibiting its own metabolism. The activity of purified


GSTZlA is about 3.5-fold that of the other forms, andGSTZI is inactivated by DCA at a rate about half that ofthe other forms. Both the differences in activity and thedifferences in the rate of inactivation work in the samedirection, so that people with the GSTZlA form wouldhave higher production of DCA metabolites than peo-ple with other forms (all other things being equal). Thismeans that the GSTZlA form would increase the risk ofeffects resulting from DCA metabolite(s), but decreasethe risk of effects resulting from DCA itself. Determina-tion of the quantitative implications of the increased ac-tivity of GSTZlA form would require a validated PBPKmodel that accounts for autoinhibition from long-termexposure to DCA

In summary, it appears that proceeding with theDCA case study is feasible. Although there are severaloutstanding issues, these issues can be addressed byconsidering the implications of multiple alternatives.For example, since it is not known whether DCA orone of its metabolites is responsible for the observedtoxic effects, the implications of polymorphisms can bedetermined under both the scenario that the parentand that a metabolite is the toxic moiety. The overallmetabolic pathway for DCA is well-characterized, andthis case study provides an opportunity to study theeffects of genetic polymorphism on autoinhibition aswell as altered catalytic efficiency of the key enzyme.A PBPK model can be developed that describes DCAmetabolism sufficiently to evaluate the implication ofthe GSTZ polymorphism, although developing a PBPKmodel that fully describes DCA metabolism might notcurrently be feasible, due to the complexity of the totalmetabolic process.

DISCUSSION

In this paper we present a summary of the toxi-cokinetics and genetic polymorphism data for methy-lene chloride, warfarin, parathion, and DCA to high-light data requirements for the meaningful applicationof genetic polymorphism data in human health riskassessments. We show that meaningful application ofpolymorphism data requires that several key issues beaddressed (Table 6).

Several key points affect whether information onvariability due to genetic polymorphism is likely to be

TABLE 6Common Data Gaps Increasing Uncertainty

E~nt of induction/inhibition through coexposureRelative contribution of multiple enzyme systemsAllelic frequencies for major ethnic groupsLarge numbers of low-frequency allelesAbsence of chemical-specific phenotype dataUncertainty regarding differences in in vitro and in vivo kinetic data

informative for a given chemical. The spectrum of cur-rently identified polymorphisms constitutes only one ofthe many sources of human variability. For example,the degree of variability in CYP2El and PONI activ-ities is not adequately explained by known polymor-phisms. Additional variability may result from as-yet-unidentified alleles in these genes, as well as from thecontribution of polymorphism in other genes that en-code similar enzymes or that regulate their activity.Clearly, there are also environmental contributions tothe degree of variability that is observed. Perhaps one ofthe best examples is the ability of coexposures to induceor inhibit these metabolizing genes, as seen with drug-drug interactions for CYP2C metabolism of warfarin(Miners and Birkett, 1998). Finally, the metabolism ofa chemical at low doses may be determined by nonen-zymatic factors, such as blood flow to the liver. Becauseflow-limited metabolism would mean that a large dif-ference in enzyme activity would result in a minimaldifference in tissue dose at low exposure levels, PBPKmodeling to determine the effect on tissue dose is animportant tool in evaluating the effect of enzyme poly-morphisms.

The relative contribution of known genetic polymor-phisms and environmental factors to the total variabil-ity in tissue dose from a given external dose is an im-portant consideration that highlights the advantagesand disadvantages of alternative approaches for us-ing human variability data in risk assessment. A thor-ough evaluation of variability (and the contribution ofthe polymorphism to the overall variability) needs toaccount for variability in multiple parameters. PBPKmodels combined with Monte Carlo modeling provide aconvenient approach for evaluating the implications ofmultiple sources of variability, as was done by El-Masriet al. (1999) and Jonsson and Johanson (2001a) formethylene chloride. Other authors have directly eval-uated variability in parameters affecting tissue dose,such as variability in enzyme kinetics (Lipscomb et al.,1999) or variability in physiologic parameters (Hattiset al., 1999b).

The genetic approach (consideration of allelic fre-quencies of identified polymorphisms) may have the ad-vantage of more easily providing a broad picture of thepopulation distribution. The enzyme activity of differ-ent alleles can be obtained relatively easily from clonedgenes, and allelic frequencies can be determined us-ing relatively noninvasive sampling methods (i.e., fromblood samples). This means that information on allelicfrequencies in different ethnic groups can be used todevelop an overall population distribution of allelic fre-quencies, without having to measure enzyme activityfrom a large number of human samples. By contrast, ob-taining a sample of liver tissue requires invasive meth-ods, making it more difficult to obtain a sufficientlylarge number of samples to be representative of the di-versity in the population.

192 HABERFrAl

PBPK approach. Our results highlight a number of crit-ical data needs and simplifying assumptions that arelikely to be required. One area in need of development isto define the minimal set of polymorphism data for im-plementation of the approach. One of the criteria for se-lection of the case study chemicals was the availabilityof allelic frequency data for the key enzymes involvedin its metabolism. While each chemical met this cri-terion, the data were not always optimal for studyingthe contribution of genotype frequencies in a diverseU.S. population. For example, PON1 genotype frequen-cies in Asian Americans were calculated from a studyof Chinese subjects (Sanghera et ai., 1998), and esti-mates for African Americans were based on African pop-ulations (Diegpen and Geldmacher-von Mallinckrodt,1986). In some cases, no data were available, even forthe most populous ethnic groups. For CYP2C9, no esti-mate of genotype frequency was available for Hispanics(Wormhoudt et ai., 1999), and for GS'n, data wereavailable only for Caucasians (Blackburn et ai., 2000).Thus, an important decision for the analysis is the de-gree to which reliable data on each ethnic group is avail-able for making a reasonable estimate of total humanvariability.

We also found that simplifying assumptions may needto be made in the common situation where multipleenzymes are involved in a chemical's metabolism. Insuch cases, it may be appropriate to focus on the en-zyme for the rate-limiting step. It may also be appro-priate to downplay the impact of enzymes that carryout metabolic steps for which other, higher efficiencyenzymes also playa role.

Another criterion for inclusion in the case studies wasthe availability of phenotype data for each allelic vari-ant. Our cases were of minimal complexity. However,for substrates metabolized by N-acetyltransferase orCYP2D6, two genes having a large number of variantalleles (Wormhoudt et ai., 1999), the task of includingall variant alleles would be overwhelming. Four spe-cific issues regarding the source of the phenotype databecame apparent in our review of the data. First, it be-came clear that general phenotyping data are not likelyto be sufficient for chemical-specific risk assessments.The importance of phenotyping data using the specificchemical of interest is clearly demonstrated by the sub-strate specificity of CYP2C9 for warfarin isomers anddifferences in catalytic efficiency of PON1 for relatedorganophosphate pesticides.

Second, the implications of using enzyme kinetic datafrom purified variant proteins versus tissues from in-dividuals having known genotypes need to be consid-ered. As described above for the genetic approach versusthe overall variability approach, data from purified pro-teins (or cloned proteins) allow activity to be attributedto a single enzyme but remove from consideration en.zyme regulation and other factors affecting enzymeactivity.

On the other hand, measuring enzyme activity usingcloned genes cannot take into account other factors un-related to the polymorphism, such as variability in tran-scriptional regulation, enzyme synthesis, and degrada-tion or other factors associated with the cellular milieuthat regulate the amount or activity of the enzyme. Inthis way, measuring activity using liver biopsy tissueprovides a more accurate measure of the actual variabil-ity in expression of the enzymes. For example, evaluat-ing variability in actual enzyme activity is preferablewhen most of the variability is due to environmental,rather than genetic sources, such as for CYP2E1. In ad-dition, while in vitro analysis can assess the catalyticefficiency of the polymorphic gene product, its impor--tance relati ve to other enzymes is not easily weighed. Ashighlighted for warfarin (Wang et at., 1983; Rettie et at.,1992; Kunze and Trager, 1996; Kaminsky and Zhang,1997) and parathion (Butler and Murray, 1997; Mutchet ai., 1999), there are multiple CYPs that catalyze theoxidation of these chemicals, and the overall variabilityin activity due to the combination of these enzymes isperhaps best estimated using the human liver samples.In the event that data from purified variant proteinsare used, the choice of which of the minor pathwaysto include in the analysis requires a balance betweenmodel accuracy and overparameterization. As experi-ence in application of polymorphism data in the PBPKapproach increases, simplifying assumptions such asthe degree to which minor pathways can be ignored willbe needed. Overall, the two approaches have a comple-mentary utility, with the genetic approach providinga broader picture of variability in the general popu-lation, but the liver biopsy approach perhaps provid-ing a more accurate determination of the variability inenzyme activities.

Regardless of the approach used, appropriately eval-uating the impact of polymorphisms on variability intissue dose requires the integration of the variabil-ity in multiple parameters (or at least in the rate-limiting parameter), a task which is conveniently ap-proached using PBPK modeling combined with MonteCarlo analysis. It is also important to recognize thepotential for differences between the kinetics at phar-macologic doses (comparatively high doses designed toconsistently elicit an effect) and at lower doses morecharacteristic of environmental exposures. High dosesare more likely to result in saturation of metabolic ca-pacity, with metabolism limited by the Vmax of key en-zymes. By contrast, metabolism at lower exposures maybe limited by the Km of key enzymes or even by bloodflow rate to the liver. PBPK models are well-suited tosimulate the kinetics at a variety of exposure levels, sothat the impact of variability at environmentally rele-vant exposures can be quantitatively evaluated.

One goal of our analysis of the case studies was toidentify the extent to which the data on a series of well-studied chemicals would support an analysis using the

POLYMORPHISM IN RISK ASSESSMENT 193

Third, the suitability of general phenotyping tests forevaluating disparity in allelic variants must be consid-ered. Phenotyping tests are often developed to maxi-mize the difference between the variant proteins, andtherefore the degree of variability observed in an invitro assay may overestimate in vivo differences. In ad-dition, for data-rich chemicals, a choice might need tobe made regarding which of multiple inconsistent val-ues should be used for kinetic parameters. For exam-ple, for warfarin, not only did the values differ consid-erably between studies, but the interstudy differenceswere often larger than the differences between alleles.The different studies were not even consistent in ra-tios of key parameters of the different alleles. Theseobserved differences are likely to be due to differencesin experimental conditions. For example, Crespi andMiller (1997) noted the role of salt concentration andof oxidoreductase levels in determining the activity ofCYP2C9. Other differences may be related to varyinglevels of protein expression in different systems. Onealternative for addressing the inconsistent kinetic pa-rameters is to determine which values are the best pre-dictors of the in vivo data. Another alternative would beto choose the set of biologically meaningful parametersthat gives the most conservative results.

most chemicals this is a long-term project before suchan approach would be ready for use in the regulatoryarena.

ACKNOWLEDGMENTS

The authors thank Sheri Lawson for assistance with editing of themanuscript and Jacqueline Patterson and Jay Zhao for technical re-view. This work was funded by a research grant from the AmericanChemistry Council.

REFERENCES

CONCLUSIONS

l11e case studies presented Un thds paper are a firststep toward developing an approach for quantitativelyincorporating polymorphdsm data Unto development oftoxicity values for environmental chemicals. Based onthe data presented in thds paper, it is apparent thatenough quantitative data exist to productively evalu-ate the degree to which variability in enzyme kineticsfor these chemicals affects the delivered tissue dose.Although several issues have been identified for eachchemical, addressing the implications of the issues re-quires working through the case studies, applying theappropriate PBPK model, and quantitatively evaluat-ing the implications of different assumptions.

Taken together, these data provide guidance regard-ing the critical uncertainties that are likely to be facedUn the application of polymorphdsm data for quanti-tative risk assessments. l11ere are important advan-tages in couplUng genetic polymorphdsm data and PBPKmodeling, particularly for beginning to estimate therange in tissue dose variability for large populations.We evaluated the data sets for a series of diverse chem-icals to identify key areas of uncertainty. There is aclear need to put thds method to use, as a means tobetter describe the overall human variability in dose oftoxic agent resulting from genetic variability. l11e de-velopment of standard guidelines on assumptions forintegrating genetic data with PBPK models will lead toimproved incorporation of data on human variability,and ultimately improved risk assessments, although for

Abbas, R, and Fisher, J. W. (1997). A physiologically based pharma-cokinetic model for trichloroethylene and its metabolites, chloralhydrate, trichloroacetate, dichloroacetate, trichloroethanol, andtrichloroethanol glucuronide in B6C3F1 mice. 7bxicol. Appl. Phar-macol. 147,15-30.

Abbott, A, et at. (1977). Chondrodysplasi punctata and maternalwarfarin treatment. B/: Med. .l1, 1639-1640.

Adkins, S., et al. (1993). Molecular basis for the polymorphic formsof human serum paraoxonase/arylesterase: Glutamine or arginineat position 191, for the respective A or B allozymes. Am. .l Hum.Genet. 52, 598-608.

Aithal, G. P., et at. (1999). Association of polymorphisms in the cy-tochrome P450 CYP2C9 with warfarin dose requirement and riskof bleeding complications. Lancet 353, 717-719.

Andersen, M. E., et al. (1987). Physiologically based pharmacokinet-ics and the risk assessment process for methylene chloride. Toxicol.Appl. Pharmacol. 87, 185-205.

Andersen, M. E., and Krishnan, K (1994). Physiologically basedpharmacokinetics and cancer risk assessment. Environ. HealthPerspect. 102(Suppl. I), 103-108.

ATSDR (1999). Toxicological Profile for Methylene Chloride, draft forpublic comment. Agency for Toxic Substances and Disease Reg-istry, Research Triangle Park, NC.

Baird, S. J. S., (2001). Probabilistic noncancer risk estimation. Comm.Toxicol. 7,541-572.

Baird, S. J. S., et al. (1996). Noncancer risk assessment: A probablis-tic alternative to current practice. Hum. Ecol. Risk Assess. 2, 79-102.

Ball, S. E., et al. (1999). Population distribution and effects ondrug metabolism of a genetic variant in the 5' promoter regionof CYP3A4. Clin. Pharmacol. The/: 66, 288-294.

Barnes, D. G., and Dourson, M. (1988). Reference dose mffi): Descrip-tion and use in health risk assessments. Regul. Toxicol. Pharmacol.8,471-486.

Barton, H. A., et at. (1999). Dichloroacetate (DCA) dosimetry: Inter-preting DCA-induced liver cancer dose response and the potentialfor DCA to contribute to trichloroethylene-induced liver cancer.Toxicol. Lett. 106, 9-21.

Black, D. J., et al. (1996). Warfarin-fluconazole. II. A metabolicallybased drug interaction: In vivo studies. Drug Metab. Dispos. 24,422-428.

Blackburn, A. C., et at. (2000). Discovery of a functional poly-morphism in human glutathione transferase zeta by ex-pressed sequence tag database analysis. Pharmacogenetics 10,49-57.

Bogaards, J. J., et al. (1993).lnterindividual differences in the in vitroconjugation of methylene chloride with glutathione by cytosolicglutathione S-transferase in 22 human liver samples. Biochem.Pharmacol. 45,2166-2169.

194 HABER ET AL

Daly, A K., et al. (1998). Recent advances in understanding themolecular basis of polymorphisms in genes encoding cytochromeP450 enzymes. Toxicol. LeU. 1M-1~ 143-147.

Dankovic, D. A., and Bailer, A. J. (1994). The impact of exerciseand intersubject variability on dose estimates for dichloromethanederived from a physiologically based pharrnacokinetic model.Fundam. Appl. 7bxicol. J2, 20--25.

Davies, H. G., et ai. (1996). The effect of the human serum paraox-onaae polymorphism is reversed with diazoxon, soman and sarin.Nature Genet. 14, 334-336.

Diepgen, T. 1.., and Geldmacher-von Mallinckrodt, M. (1986). in-terethnic differences in the detoxification of organophosphates:The human serum paraoxonase polymorphism. Arch. 7bxicol.Suppl. 9, 154-158.

Dourson, M., et ai. (1998). Boron tolerable intake: &-evaluationof toxicokinetics for data-derived uncertainty factors. Biol. Tra«Elem. Res. 86, 453-463.

Dourson, M. L., and StarR, J. F. (1983). Regulatory history and ex-perimenta1support of uncertainty (safety) factors. ReguL 7Micol.Phannacol. 3, 224-238.

Eaton, D. 1.., and Bammler, T. K. (1999). Concise review of theglutathione S-transferases and their significance to toxicology.7Micol. Sri. 49, 156-164.

Echobichon, D. (1996). 7hric Effects of Pesticides:. Casarett & Doull'R7bxicology-The Basic Science of Poisons (c. Klaassen, Ed.).pp. 643-690. McGraw-Hill, New York.

Eckerson, H. W., et ai. (1983). The human serum paraoxonase/arylesterase polymorphism. Am. .l Hum. Genet. 35, 1126-1138.

El-Masri, H. A, et ai. (1999). Effects of glutathione transferase thetapolymorphism on the risk estimates of dichloromethane to hu-mans.1bxicol. Appl. Pharmacol. 158,221-230.

Epstein, D. L., et ai. (1992). Cardiopathic effects of dichloroacetatein the fetal Long-Evans rat. Teratology 48, 225-235.

Evans, O. B., and StaCJ)OOle, P. W. (1982). Prolonged hypolactatemiaand increased total pyruvate dehydrogenase activity by dichloroac-etate. Biochem. Pharmacol. 31, 1295-1300.

Furlong, C. E., et ai. (1993). Human and rabbit paraoxonases: Purifi-cation, cloning, sequencing. mapping and role of polymorphism inorganophosphate detoxification. Chern. Bioi. Interact. 87, 36-48.

Furuya, H., et ai. (1995). Genetic polymorphism ofCYP2C9 and itseffect on warfarin maintenance dose requirement in patients un.dergoing antia)agulation therapy. Pharmacogenetics 5, 389-392.

Gargas. M. L., et ai. (1986). Metabolism of inhaled dihalometbanesin vivo: Differentiation of kinetic constanta for two independentpathways. Toxicol. Appl. Pharmacal. 82. 211-223.

Gearhart, J. M., et ai. (1990). Physiologically based pharmacoki-netic and pharmacodynamic model for the inhibition of acetyl-cholinesterase by diisopropyUluoropbosphate. 7b%jcol. Appl. Phar-macal. 106, 295-310.

Gonzalez-Leon, A, et al. (1997). Pharmacokinetics and metabolismof dichloroacetate in the F344 rat after prior administration indrinking water. 7bricol. AppL Pharmacol. 148, 189-195.

Gotoh, O. (1992). Substrate recognition sites in cytochrome P450 fam-ily 2 (CYP2) proteins inferred from comparative analyses of aminoacid and coding nucleotide sequences. .l Biol. Chern. 267, 83-90.

Greenberg. M. S., et ai. (1999). Physiologic:aJly based pharmacoki.netic modeling of inhaled trichloroethylene and ita oxidativemetabolites in B6C3F1 mice. 1bxicol. Appl. Pharmacol. 164, 264-278.

Raining, R. 1.., et al. (1996). Allelic variants of human cytochromeP450 2C9: Baculovirus-mediated expression, purification, struc-tural characterization, substrate stereoselectivity, and procm-ral selectivity of the wild-type and 1359L mutant forms. Arch.Biochem. Brophy&. 383,447-458.

Boorman, G. A (1999). Drinking water disinfection byproducts: Re-view and approach tA> tA>xicity evaluation. Environ. HealthPerspect.l07(Suppl. I), 207-217.

Bruhn, C., et al. (1998). Concordance between enzyme activity andgenotype of glutathione S-transferase theta (GSTT1). Bi«hem.Pharmacol. H, 1189-1193.

Bull, R. J. (2000). Mode of action of liver induction by trichlo-roethylene and its metabolites trichloroacetate and dichloroac-etate. Environ. Health PerBpect. 1000Suppl. 2), 241-259.

Burin, G. J., and Saunders, D. R. (1999). Ad~6ing human variabil-ity in mk a88e88ment--Tbe robmtneu oftJie intra8pecies uncer--tainty factor. Regul. 7bxicol. Phannacol. 30, 209-216.

Butler, A M., and Murray, M. (1997). Biotransformation of parathionin human liver: Participation of CYP3A4 and its inactivation dur-ing microsomal parathion oxidation. .l Pharmacal. Exp. 7'he1: 280,966-973.

Calabrese, E. J. (1985). U~rtainty factors and interindividual vari-ation. Regul. 7bxicol. Phamuu:ol. i, 190-196.

Calabrese, E. J., et oJ. (1992). Does the animal-to-human uncertaintyfactor inOOrJM)rate intenpecies difference6 in 6urface area? Regui.Thxicol. Pharmacol. Ii, 172-179.

Calabre6e, E. J., and Gilbert, C. E. (1993). Lack oftA>taI independenceof uncertainty factors (UPs): Implicati!m8 for the mze of the totaluncertainty factor. Regul. ~ Phamuu:ol. 17,44-51.

Casanova, M., et oJ. (1996). DNA-protein CI'O88-links (DPX) andcell proliferation in B6C3F1 mice but not Syrian golden ham-6ten exposed tA> dichloromethane: Pharmaeokinetics and mk as-8e88ment witJi DPX as dosimeter. Fundam. Appl. Taxicol. 31,103-116.

Chan, E., et oJ. (1994). Stereochemical aspects of warfarin drug in-teractions: Use of a combined pharmacokinetic-pharmacodynamicInodel Clin. P~ 77le1: H, 286-294.

Chen, T. L., et oJ. (2000). Modulation of cytochrome P-450 dependentDK)DOOxygenaaes in streptozotAlcin-induced diabetic h8Jn6ter. I. Ef-fects ofpropofol on deftuorination and cytochrome P-450 activities.Acta Anae.tIIeBiol. Sin. 38, 15-21.

Choonara, I. A., et al. (1986). Stereoselective interaction between theR-enantiomer of warfarin and cimetidine. BI: .l Clin. Phamuu:ol.21,271-277.

Cicmanec, J. L., et al. (1991). 9O-day tA>xicity study of dichloroacetatein dogIi. Fundam. Appl. 7bxicol. 17, 376-389.

Clewell, H. J., et oJ. (2000). Development of a physiologically basedpharmacokinetic model of trichloroethylene and its metabolite6 foruse in risk asaeament. Environ. Health PerBpect. 1000Suppl 2),283-305.

Clewell, H. J., and Andersen, M. E. (1996). UBe of physiologicallybased pharmaeokinetic modeling tA> investigate individual venuspopUlation risk. To%icolOIY 111,315-329.

Cornett, R., et oJ. (1999). Inhibition of glutathione S-tran6ferase zetaand tyroeine metabolism by dichloroacetate: A potential unifyingmechanism for its altered biotransformation and tA>xicity. Biochem.Biophy& Res. Commun. J82, 752-756.

Costa, L. G., et al. (1990). Serum paraoxonue and its influenceon paraoxon and chlorpyrifos-oxon toxicity in rata. Thxicol. Appl.Phamuu:ol. 108,66-76.

Crabb, D. W., et oJ. (1981). The metabolic effects of dichloroacetate.Metaboli8m SO, 1024-1039.

Crespi, C. L., and Miller, V. P. (1997). The Rl44C change in theCYP2C9.2 allele alten interaction of the cytochrome P460 witJiNADPH: Cytochrome P450 oxidoreductase. Pharmacogenetics 7,203-210.

Curry, S. H., et oJ. (1991). Disposition and pharmacodynamics ofdichloroacetate (DCA) and oxalate following oral DCA doBes.Biopharm. Drug DiBpos. 12, 375-390.


tion in hepatic ~l from humans and rats, is reduced indichloroacetate-treated rats. Drug Metab. Dispos. 25, 1223-1227.

Johnson, J. A, and Wallace, K. B. (1987). Species-related differencesin the inhibition of brain acetylcholinesterase by paraoxon andmalaoxon. 7bxicol. Appl. Pharmacol. 88, 234-241.

Jonsson F., and Johanson, G. (2001a). A Bayesian analysis of thein1Iuence of GSTTI polymorphism on the cancer risk estimate fordichloromethane. 7bxicoL AppL Pharmacol. 174, 99-112.

Jonsson, F., and JobanBOn, G. (2001b). Bayesian estimation ofvari-ability in adipose tissue blood flow in man by physiologicallybased pharmacokinetic modeling of inhalation exposure to toulene.7b%icology 1&7, 177-193.

Jonsson, F.,etal. (2001). Physiologically based pharmacokinetic mod-eling of inhalation exposure of humans to dichloromethane duringmoderate to heavy exen:ise. 7bxicol. Sci. 58, 209-218.

Kaminsky, L. S., et al. (1993). Correlation of human cytochromeP4502C substrate specificities with primary structure: Warfarinas a probe. MoL Pharmacol. 43, 234-239.

Kaminsky, L. S., and Zhang, Z. Y. (1997). Human P450 metabolismof warfarin. PharmaL'Ol. Ther: 73, 67-74.

Kaplan, N. (1987). An evaluation of the safety factor approach in riskassessment. Branbury Rep. 28, 335-347.

Katoh, T., et al' (2000). Inherited polymorphism in theN-acetyltranaferase 1 (NATl) and 2 (NAT2) genes and sus-ceptibility to gastric and colorectal adenocarcinoma. Int. .l Cancer85, 46-49.

Kunze, K. L., et al. (1991). Metabolic enantiomeric interactions: l1leinhibition of human (S}-warfarin-7-hydroxylase by (R}-warfarin.Chirality 3, 24-29.

Kunze, K. L., and Trager, W. F. (1996). Warfarin-fluconazole. m. Arational approach to management of a metabolically based droginteraction. Drug Metab. Di8pos. 24, 429-435.

Larson, J. L., and Bull, R. J. (1992). Metabolism and lipoperoxidativeactivity of trichloroacetate and dichloroacetate in rats and mice.Tozicol. Appl. Pharmacol. 116, 268-277.

Lash, L. H., et al. (2000). Metabolism of trichloroethylene. Environ.Health Perspect. 1000Supp. 2),177-200.

Li, W. F., et al. (2000). Catalytic efficiency determines the in-vivoefficacY of PONI for detoxifying organophosphorus compounds.Pharmarogenetics 10, 767-779.

Lin, E. L., et al. (1993). Tissue distribution, excretion, and urinarymetabolites of dichloroacetic acid in the male Fischer 344 rat..l Thxicol. Environ. Health 38, 19-32.

Linder, R. E., et al. (1997). SpermatotoDcity of dichloroacetic acid.Reprod. 7bxicol. 11,681-688.

Lipscomb, J. C., et aL (1995). Dichloroacetic acid: Metabolism incytoeol. Drug Metab. Dispos. 23, 1202-1205.

London, S. J., et al' (2000). Isothiocyanates, glutathione S-transferase Ml and Tl polymorphisma, and lung-cancer risk: Aprospective study of men in Shanghai, China. Lancet 358, 724-729.

Luecke, R. H., and Wosi1ait, W. D. (1979). Drug elimination inter-actions: Analysis using a mathematical model. .l Pharmacokinet.Biopharm. 7,629--641.

Luecke, R. H., et aL (1994). A physiologically based pharma-ookinetic computer model for human pregnancy. Teratology 49,90-103.

Mainwaring, G. W., et al. (1996). Isolation of a mouse theta glu-tathione S-transferase active with methylene chloride. Biochem..l 314, 445-448.

Meek, M. E., et al. (1994). Approach to assessment of risk to hu-man health for priority substances under the Canadian Environ-mental Protection Act. Environ. Carcinogen. Ectoxicol. Rev. C12,105-143.

Hall, J. G., et al. (1980). Ma~rnal and fetal sequalae of anticoagula-tion during pregnancy. Am. .l Med. 68, 122-140.

Hallier, E., et al. (1994). Metal>olism of dichloromethane (methylenechloride) w formaldehyde in human erythrocytes: Influence of poly-morphism of glutathione transferase theta (GST Tl-l). Arch. 1bx-icol. 68, 423-427.

Hattis, D., et al. (l999a). Distributions of individual susceptibilityamong humans for wxic effects: How much protection does thetraditional ~nfold facWr provide for what fraction of which kindsof chemicals and effects? AIm. N. y: Acad. Sci. 895,286-316.

Hattis, D., et at. (1999b). Human in~rindividual variability in pa-rame~rs related w health risks. Risk Anal. 19,711-726.

Hattis, D., and Barlow, K. (1996). Human in~rindividual variabilityin cancer risks-Technical and management challenges. Hum. Ecol.Risk A8se8& 20, 194-220.

Hattis, D., et at. (1987). Human variability in susceptibility w wnccbemicais-A preliminary analysis of pharmacokinetic data fromnormal volunteers. Risk Anal. 7, 415-426.

Hattis, D., and Silver, K. (1994). Human in~rindividual variability-A ~r source of uncertainty in aB8e88ing risks for noncancerhealth effects. Risk AnaL 14, 421-481.

Haustein, K. O. (1999). Pharmacokinetic and pharmacodynamicproperties of oral anticoagulants, especially phenprocoumon.Semin. Thromb. HemOBt. 26, 5-11.

He, M., et al. (1999). Structural forms of phenprocoumon and war-farin that are metabolized at the active si~ of CYP2C9. Arch.Biochem. Biophys. 371, 16-28.

Heijmans, B. T., et al. (2000). Common paraoxonase gene variants,mortality risk and fatal cardiovascular events in elderly subjects.Atherosclerosis 149, 91-97.

Hsieh, K. P., et at. (2001). Novel mutations of CYP3A4 in Chinese.Drug Metab. Dispos. 29, 268-273.

Humbert, R., et al. (1993). The molecular basis of the human serumparaoxonase activity polymorphism. Nature Genet. 3,73-76.

!ARC (1999). IARC Monogmphs on the Evaluation of Carcino-genic Risks to Humans: Reevaluation of Some Organic Chemicals,Hydrazine and Hydrogen Peroxide, Vol. 71, part one. In~rnatjonalAgency for Reeeareb on Cancer, Lyon, France.

Ingelman-Sundberg, M., et al. (1999). Polymorphic human cy-tochrome P450 enzymes: An opportunity for individualized drugtreatment. Trends Pha~ Sci. JO, 342-349.

Inoue, K., et al. (2000). Characterization of liver microsomal7 -ethoxycoumarin O-deethylation and chlorzoxazone 6-hydroxyla-tion activities in Japanese and Caucasian subjects genotyped forCYP2El gene. Arch. 1bxicol. 74, 372-378.

IPCS (2001). Guidance Document for the Use of Data in Develop-ment of Chemical-Specific Adjustment Factors (CSAF) for Inter-species Differences and Human Variability in Dose I ConcentrationResponse AB8e88ment. In~mational Programme on ChemicalSafety, World Health Organization, Geneva. [available online atwww.ipcsharmonize.org]

lPCS (2000). Environmental Health Criteria 216: Disinfectants andDisinfectont Byproduct&. International Programme on ChemicalSafety, World Health Organization, Geneva.

IPCS (1998). Environmental Health CriUria 204: Boron. In~rna-tional ProgramJne on Chemical Safety, World Health Organization,Geneva

IPCS (1994). Environmental Health CriUria 170--A88e88ing Hu-man Health Risks of Chemicals: Derivation of Guidance Val-ues for Health-Based Exposure Limits. In~matjonal Pr0-gramme on Chemical Safety, World Health Organization,Geneva.

James, M. 0., et al. (1997). Glutathione-dependent conversion wgiyoxylate, a major pathway of dichloroaceta~ biotransfonna-

196 HABER ~ AL.

Renwick, A. G. (1993). Data derived safety faclA>rs for the evaluationof food additives and environmental contaminants. Food AddU.Contam. 10, 275-305.

Renwick, A. G., and Lazarus, N. R. (1998). Human variability andnoncancer risk assessment-An analysis of the default uncertaintyfaclA>r. Regul. 7bxicol. Pharmacol. 27,3-20.

Rettie, A. E., et at. (1999). A common genetic basis for idiosyncratictoxicity of warfarin and phenytoin. Epilepsy Re& 31, 253-255.

Rettie, A. E., et at. (1992). HydroxyJation of warfarin by humancDNA-expressed cytochrome P-450: A role for P.4502C9 in theetiology of (S}-warfarin-drug interactions. Chern. Re& 7bxicol. 6.54-09.

Rettie, A. E., et 01. (1994). Impaired (S}-warfarin metabolism catal-ysed by the Rl44C a11elic variant ofCYP2C9. Pharmacogenetics 4,39-42.

Roh, H. K., et 01. (1996). CYP2C19 genotype and phenotype deter-mined by omeprazole in a Korean population. Pharmacogenetics6, 547-651.

Barns, C., et 01. (2000). Evidence for the activation of organoph~phate pesticides by cytochromes P450 3A4 and 2D6 in human livermicrosome&. 7bxicol. Lett. 116, 217-221.

Banghera, D. K., et al. (1998). The codon 55 polymorphism in theparaoxonase 1 gene is not associated with the risk of coronaryheart disease in Asian Indians and Chinese. Atherosclerosis 136,217-223.

Sata, F., et 01. (2000). CYP3A4 allelic variants with amino acid sub-stitutions in exons 7 and 12: Evidence for an a11elic variant withaltered catalytic activity. Clin. Phannacol. The!: 87, 48--56.

Sheenan, D. M., and Gaylor, D. W. (1990). Analysis of the adequacyof safety factors. Thratology 41, 590-591.

Sherratt, P. J., et 01. (1997). Evidence that human class Thetaglutathione S-transferaae T1-1 can catalyse the activation ofdichloromethane, a liver and lung carcinogen in the mouse: Com-parison of the tissue distribution of GST T1-1 with that of classesAlpha, Mu and Pi GST in human. Biochem. .l 326, 837-846.

Slob, W., and Pieters, M. N. (1998). A probabilistic approach for deriv-ing acceptable human intake limits and human health risks fromtoxicological studies: General framework. Risk Anal. Is, 787-798.

Smith, M. K, et 01. (1992). Developmental toxicity of dichloroacetatein the rat. Thratology 48, 217-223.

Smolen, A., et al. (1991). Characteristics of the genetically deter-mined aUozymic forms of human serum paraoxonase/arylesteraae.Drug Metab. Dispos. 19, 107-112.

Snawder, J. E., and Lipscomb, J. C. (2000). Interindividual variance ofcytochrome P450 forms in human hepatic microsomes: Correlationof individual forms with xenobiotic metabolism and implicationsin risk assessment. Regul. Toxjcol. Pharmacol. 32, 200-209.

Stacpoole, P. W. (1989). The pharmacology of dichloroacetate.Metabolism 38, 1124-1144.

Stacpoole, P. W., and Greene, Y. J. (1992). Dichloroacetate. Diabet.Care 15. 785-791.

StaCPOOle, P. W., et at. (1998). Pharmacokinetics, metabolism andtoxicology of dichloroacetate. Drug Metab. Rev. 30, 499--539.

Steward, D. J., et 01. (1997). Genetic association between sensitivity towarfarin and expression ofCYP2C9*3. Pharmacogenetics 7, 361-367.

Sullivan-Klose, T. H., et 01. (1996). The role of the CYP2C9-Leu359a11elic variant in the tolbutamide polymorphism. Pharmarogenet-ics 6, 341-349.

Sutcliffe, F. A., et al. (1987). Aspects of anticoagulant action: A reviewof the pharmacology, metabolism and toxicology of warfarin andcongeners. Rev. Drug Metab. Drug Interact. 5, 225-272.

Swartout, J. C., et al. (1998).,A probabilistic framework for the refer-ence dose (probabilistic ROJ). Risk Anal. 1s, 271-282.

Meek, M. E., et al. (2001). Guidelines for application of compoundspecific acljustment factors (CSAF) in dose/concentration responsea88e88ment. Comments 7bxicol. 7,575-590.

Meyer, D. J., et al. (1991). Theta, a new class of glutathione trans-ferases purified from rat and man. Biochem. .l 274, 409-414.

Mills, C. J., et al. (1998). Workshop report-Health risks of drinkingwater chlorination by-products: Report of an expert working group.Chronic Di& Can. 19,91-102.

Miners, J. 0., and Birkett, D. J. (1998). CytA)Chrome P4502C9: Anenzyme of major importance in human drug metabolism. B7: J:Cun. Pharmacol. 40, 526-538.

Mochizuki, H., et al. (1998). Human PON2 gene at 7q21.3: Cloning,multiple mRNA forms, and missense polymorphisms in the codingsequence. Gene 213, 149-157.

Moser, V. C., et al. (1999). Behavioral evaluation of the neuroWxicityproduced by dichloroacetic acid in rats. Neurotoxicol. Teratol. 21,719-731.

Mueller, R. F., et al. (1983). Plasma paraoxonase polymorphism: Anew enzyme assay, population, family, biochemical, and linkagestudies. Am. J: Hum. Genet. 36, 393-408.

Mungall, D. R., et al. (1985). Population pharmacokinetics of racemicwarfarin in adult patients. J: Pharmacokinet. Biopharm. 13, 213-227.

Murray, F. J., and Andersen, M. E. (2001). Data-derived uncertaintyfactors: Boric acid (BA) as a case study. Hum. Ecol. Risk A88e3& 7,125-138.

Murray, M., et al. (1997). Induction of cytochromeB P450 2B and 2Elin rat liver by isomeric picoline N-oxides. 7bxicol. Lett. 98, 196-203.

Mukh, E., et al. (1992). Interindividual variations in enzymes con-trolling organophosphate toxicity in man. Hum. Exp. 7bxicol. 11,109-116.

Mukh, E., et al. (1999). The role of metabolism in determiningsusceptibility to parathion toxicity in man. 7bxicol. Lett. 107,177-187.

Nelson, H. H., et al.' (1995). Ethnic differences in the prevalence of thehomozygous deleted genotype of glutathione S-transferase theta.Carcinogenesis 18, 1243-1245.

NTP (1986). 1b%it'Ology and Carcinogenesis Studies of Dichlo-romethane (Methylene Chloride) (CAS No. 75-09-2) in F344/ NRats and B6C3Fl Mice (Inhalation Studies), TR-306. NTP, Re-search Triangle Park, NC.

Omiecinski, C. J., et al. (1999). Concise review of the cytochromeP450s and their roles in toxicology. 7bxicol. Sci. 48, 151-156.

Price, P. S., et al. (1997). An approach for modeling noncancer doeeresponses with an emphasis on uncertainty. Risk Anal. 17,427-437.

Primo-Parmo, S. L., et al. (1996). The human serum paraox-onase/arylesterase gene (PON1) is one member ofa multigene fam-ily. Genomics 33, 498-507.

Puga, A., et al. (1997). Genetic polymorphisms in human drug-metabolizing enzymes: Potential uses of reverse genetics toidentify genes of toxicological relevance. Crit. Rev. Toxicol. 27,199-222.

Redman, A R. (2001). Implications of cytochrome P450 2C9 polymor-phism on warfarin metabolism and dosing. Pharmacotherapy 21,235-242.

Reitz, R. H. (1990). Quantitating the production of biological reactiveintermediates in target tissues: Example, dichloromethane. Adv.Exp. Med. Bioi. 283, 649-655.

Reitz, R. H., et aI. (1989). In vitro metabolism of methylene chloridein human and animal tissues: Use in physiologically based phar-macokinetic models. Toxicol. Appl. Pharmacol. 97, 230-246.

Renwick, A. G. (1991). Safety factors and establishment of acceptabledaily intakes. Food Addit. Contain. 8, 135-149.

POLYMORPHISM IN RISK ASSESSMENT 4.97

U.s. EPA (1996). Proposed Guidelines for CarcilUJgen Risk ABBeSs.mem. Office of Research and Development, U.S. EnvironmentalProt«tion Agency, Washington, DC.

Veronesi, B., and Ehrich, M. (1993). Differential cytotoxicsensitivity in mouse and human cell linee exposed toorganophosphate insecticides. Toxicol. Appl. Pharmacol. 120,240-246.

von Moltke, L. L.,et al. (2000). Unusually low clearance of two CYP3Asubstrates, alprazolam and trazodone, in a volunteer subject withwild-type CYP3A4 promoter region. .l Clin. Pharmacol. 40, 200-204.

Wandel, C., et al. (2000). CYP3A activity in African American andEuropean American men: Population differences and functionaleffect of the CYP3A.*1B5'-promoter region polymorphism. Cun.Pharmacol. 77ler: 88, 82-91.

Wang, P. P., et al. (1983). Purification and characterization of six cy-tochrome P-450 isozymes from human liver microsomes. Biochem-istry 22, 5375-6383.

Warholm, M., et al. (1994). Polymorphic distribution of glutathionetransferase activity with methyl chloride in human blood. Phar-macogenetics 4, 307-311.

Westlind, A, et al. (1999).lnterindividual differences in hepatic ex-pression of CYP3A4: Relationship to genetic polymorphism in the5' -upstream regulatory region. Biochem. Biophys. Res. Commun.269, 201-205.

Whitfield, M. F. (1980). Chodrodysplasi punctata after warfarin inearly pregnancy: Cue report and summary of the literature. Arch..l Dis. Child 55, 139-180.

Wiebel, F. A, et al. (1999). l1le hereditary transmission of the glu-tathione transferase hGSTrl-l conjugator phenotype in a largefamily. Pharmacogenetics 8, 251-256.

Wienkers, L. C., et al. (1996). Formation of (R}-8-hydroxywarfarinin human liver micnleOmes: A new metabolic marker for the (8)-mephenytoin hydroxylase, P4502CI9. Drug Metab. Dis. 24, 610-614.

Wormhoudt, L. W:, et al. (1999). Genetic polymorphisms ofhuman N-acetyltransferaae, cytochrome P450, glutathione-s.transferase, and epoxide hydrolase enzymes: Relevance to xeno-biotic metabolism and toxicity. Crit. Rev. Tozicol. 28,59-124.

Sweeney, R. E., and Maxwell, D. M. (1999). A theoretical model of thecompetition between hydrolase and carboxylesterase in protectionagainst orgaDOphosphorua poisoning. Math. Biosci. 180, 175-190.

Takahashi, H., et ai. (l998a). Metabolism of warfarin enantiomersin Japanese patients with heart diseaae having different CYP2C9and CYP2C19 genotypes. Clin. Pharmacol. ~7: 83,519-528.

TakAhAAhi, H., et ai. (1998b). Comparisons between in-vitro andin-vivo metabolism of (S}-warfarin: Catalytic activities of cDNA-expressed CYP2C9, its Leu359 variant and their mixture versusunbound clearance in patients with the corresponding CYP2C9genotypes. Pharmacogenetics 8, 365-.173.

Tanaka, E., et at. (2000). CytA)Chrome P450 2El: Its clinical and tox-icological role..l Clin. Pharm. ~7: 25, 165-175.

Tang, J., and Chambers, J. E. (1999). Detoxication ofparaoxon by ratliver homogenate and serum carboxylesterases and A~sterases..l Biochem. Mol. 7b%icol. 18, 261-268.

Taube, J., et al. (2000). Influence of cytochrome P-450 CYP2C9 poly-morphiams on warfarin sensitivity and risk of over-anticoagulationin patients on long-term treatment. Blood 96,1816-1819.

Tong, z., et ai. (19988). Glutathione transferase zeta catalyses theoxygenation of the carcinogen dichloroacetic acid to glyoxylic acid.Biochem. .l 331, 371-374.

Tong, z., et ai. (l998b). Glutathione transferase zeta-catalyzed bio-transfonnation of dichloroacetic acid and other aIpha-haIoacids.Chem. Rea. 7b%icol. 11,1332-1338.

TUng, H. F., et at. (2000). Polymorphism- and species-dependentinactivation of glutathione transferase zeta by dichloroacetate.Chem. Res. 7bxicol. 13, 231-236.

U.s. EPA (1994). Methods for Ikrivation ofinhalationRefe~nce Con-centrations and Application ofInha1ation Dosimetry. U.S. Environ-mental Protection Agency, Washington, DC.

U.S. EPA (1999). Guidelines for Carcinogen Risk A88e88ment. U.S.Environmental Protection Agency, Washington, DC.

U.S. EPA (2001). Integrated Risk Information System aRIS). U.S.Environmental Protection Agency, Washington, DC. [available athttp://www.epa.gov/iris/]

U.8. EPA (1986). Gui4elines for Careinogen ~ Aa8e88menLOffice of Research and Development, U.S. Environmental Protec-tion Agency, Washington, DC.

genetic polymorphisms in assessing interindividual ... · received april 20, 2001 introduction...

Documents