Pharmacogenetics of efavirenz and central nervous system side effects: an Adult AIDS Clinical Trials Group study
AIDS: Volume 18(18) 3 December 2004
Haas, David Wa; Ribaudo, Heather Jb; Kim, Richard Ba; Tierney, Camlinb; Wilkinson, Grant Ra; Gulick, Roy Mc; Clifford, David Bd; Hulgan, Todda; Marzolini, Catiaa; Acosta, Edward Pe
From the aVanderbilt University School of Medicine, Nashville, Tennessee, the bStatistical Data Analysis Center, Harvard School of Public Health, Boston, Massachusetts, the cWeill Medical College of Cornell University New York, New York, the dWashington University School of Medicine, St. Louis, Missouri and the eUniversity of Alabama at Birmingham, Birmingham, Alabama, USA.
Note: These data were presented at the Eleventh Conference on Retroviruses and Opportunistic Infections, February 2004, San Francisco, CA, USA.
Objectives: Efavirenz is an effective antiretroviral agent, but central nervous system side effects occur commonly, and population (racial) differences in pharmacokinetics and response have been reported. Efavirenz is metabolized by cytochrome P4502B6 (CYP2B6). We investigated whether polymorphisms in CYP2B6, CYP3A4, CYP3A5, and MDR1 were associated with efavirenz central nervous system side effects and pharmacokinetics.
Design: Twenty-four week cohort from a randomized study.
Methods: Adult AIDS Clinical Trials Group study A5097s examined relationships between central nervous system side effects and efavirenz plasma concentration-time profiles in HIV-infected subjects. Efavirenz plasma pharmacokinetics were estimated by a population-based method. Central nervous system symptoms were assessed by questionnaires and neuropsychological testing.
Results: Study subjects included 89 (57%) European-Americans, 50 (32%) African-Americans, and 15 (10%) Hispanics. The CYP2B6 T/T genotype at position 516 (Gln172His) was more common in African-Americans (20%) than in European-Americans (3%), and was associated with greater efavirenz plasma exposure (P < 0.0001). The median efavirenz [area-under-the-curve] (0-24 h) according to G/G, G/T, and T/T genotype was 44 (n = 78), 60 (n = 60), and 130 (n = 14) μg.h/ml, respectively (P < 0.0001). The CYP2B6 G516T genotype was also associated with central nervous system symptoms at week 1 (P = 0.036). Analysis of DNA from other subjects confirmed population differences in frequency of the G516T variant. No associations were apparent with the other polymorphisms studied.
Conclusions: A CYP2B6 allelic variant that is more common in African-Americans than in Europeans-Americans was associated with significantly greater efavirenz plasma exposure during HIV therapy. Inter-individual differences in metabolism may, in part, explain susceptibility to efavirenz central nervous system side effects.
In this study, we observed that a polymorphism in the gene that encodes the hepatic enzyme primarily responsible for efavirenz metabolism (CYP2B6), namely, G516T is associated with greater plasma efavirenz exposure during the first 24 weeks of antiretroviral therapy, and with increased central nervous system side effects during the first week. Interestingly, tolerance to central nervous system side effects occurred despite persistently greater plasma efavirenz exposure. Increased frequency of this polymorphism in African-Americans may explain lower clearance of efavirenz in this population. In multivariate analysis, after adjusting for the CYP2B6 genotype, differences in efavirenz pharmacokinetic profiles between European-Americans and African-Americans were not evident, suggesting that genotype may explain this racial disparity. This study demonstrates that a genetic polymorphism that occurs more commonly in African-Americans than European-Americans is associated with both reduced clearance of efavirenz and with a greater incidence of early central nervous system side effects. The small sample size for Hispanics prevents us from reaching conclusions regarding this population.
Our findings may explain previously reported population differences in efavirenz pharmacokinetics or efficacy. Among 334 healthy volunteers enrolled in phase I studies, efavirenz plasma clearance was lower in African-Americans than in European-Americans. In 139 HIV-infected individuals, efavirenz clearance was 21% lower in African-Americans than in European-Americans. In an observational study of HIV-infected subjects who at the time of enrollment were receiving efavirenz, indinavir, or nelfinavir, the median time to virologic failure on efavirenz was 422 days for African-Americans but greater than 1400 days for European-Americans. There was no such observed difference between populations for indinavir or nelfinavir. A larger, prospective clinical trial did not confirm this difference, but suggested earlier treatment failure in Hispanics as compared to blacks. Excessive efavirenz plasma concentrations resulting in increased adverse events might explain more rapid treatment failure among African-Americans  and/or Hispanics in previous reports.
The hepatic enzyme CYP2B6 is one of the less well characterized CYP isoforms, and the list of known substrates is growing. In addition to efavirenz it metabolizes nevirapine, methadone, cyclophosphamide, ifosfamide, tamoxifen, bupropion, diazepam, temazepam, and midazolam. It also metabolizes nicotine, methylenedioxymethamphetamine (`Ecstasy'), methylenedioxy-ethamphetamine (`Eve'), and several procarcinogens. Hepatic CYP2B6 protein and activity levels vary considerably between individuals. At least ten CYP2B6 single nucleotide polymorphisms have been identified, of which six result in altered amino acid sequence. In the only previous publication that examined associations between CYP2B6 genetic variants and drug metabolism in humans, an A to G substitution at position 785G was weakly associated with higher plasma clearance of the antidepressant bupropion. The association between efavirenz metabolism and a CYP2B6 polymorphism that is more frequent in African-Americans should prompt investigations of other commonly used prescription and recreational drugs that are CYP2B6 substrates.
The predominant CYP isoform subfamily in liver, CYP3A, may contribute to efavirenz metabolism. Our results suggest that CYP3A4 A-392G and CYP3A5 A6986G may be weakly associated with plasma efavirenz exposure. An association with CYP3A5 A6986G is not completely unexpected since this variant allele encodes essentially inactive CYP3A5. The MDR1 gene encodes P-glycoprotein which is a drug efflux transporter in the intestine, liver, kidney and blood-brain barrier. Efavirenz is not known to be a P-glycoprotein substrate. Not surprisingly, therefore, we found no apparent independent association between MDR1 polymorphisms and plasma efavirenz exposure or central nervous system side effects, and no significant relationship with virologic or immunologic responses, consistent with recent, but not earlier reports.
Of the various neuropsychological measures in study A5097s, the symptom questionnaire most clearly differentiated patients receiving efavirenz from other subjects in this blinded study. We observed an association between neurologic symptoms and CYP2B6 at week 1, but not week 24, consistent with previous observations that these symptoms lessen or resolve over time. These findings support the association between CYP2B6 genotype and central nervous system side effects. In separate analyses involving A5097s subjects, central nervous system symptoms were greater at week 1 than at baseline (P < 0.001), and this increase in symptoms was associated with higher plasma efavirenz concentrations at week 1 but not at later weeks. However, factors in addition to efavirenz plasma concentration may be important, since central nervous system side effects resolved despite persistently greater plasma efavirenz exposure in subjects with CYP2B6 position 516 T/T genotype. Such tolerance could develop by various mechanisms. Interestingly, neurons and astrocytes in human brain express CYP2B6, suggesting potential intracranial influences on efavirenz metabolism that may not be apparent by assaying peripheral plasma.
The present findings may have important implications for antiretroviral therapy. Efavirenz is uniformly prescribed at 600 mg once daily. The possibility that a lower efavirenz dose in many such individuals may reduce side effects without compromising efficacy should be studied. Although we did not detect an association with virologic or immunologic response, only 14 study subjects had the position 516 T/T genotype. Additional large studies with longer follow-up are needed to determine whether a sustained three-fold difference in plasma efavirenz exposure affects long-term efficacy or toxicity, including the risk of developing reverse transcriptase efavirenz-associated resistance mutations.
This study should stimulate further investigation of associations between CYP2B6 genotype, pharmacokinetics, and treatment response. The genotype-phenotype associations we identified may not be causal. Additional polymorphisms in CYP2B6 or other genes may be very important, and could explain the greater plasma efavirenz exposure in two subjects with CYP2B6 position 516 G/T heterozygosity. This seems likely given the increasing number of CYP2B6 polymorphisms and haplotypes recently identified. In addition, if it is found that that CYP2B6 G516T is also associated with delayed clearance of nevirapine, this will have even greater implications. Prolonged nevirapine exposure would likely increase selection for drug-resistant virus when single dose nevirapine is used to prevent mother to child transmission. Importantly, our findings must be confirmed in other datasets.
With more than 90% of HIV-infected people worldwide living in resource-limited countries, major initiatives are beginning to provide antiretroviral drugs to such individuals. Current World Health Organization guidelines recommend first line therapy for HIV-infected individuals with non-nucleoside reverse transcriptase-based regimens (including either efavirenz or nevirapine). The recommendations are supported by the results of clinical trials, most of which have been conducted in developed countries with homogeneous patient populations, and there are relatively few results available from developing world populations. Our study identified population differences in the frequency of a polymorphism in the gene that encodes the hepatic enzyme CYP2B6 that was associated with differences in antiretroviral drug concentrations and side effects. Generalizing the results of clinical trials from one population to another may be problematic if substantial genetic differences among populations lead to differences in antiretroviral drug concentrations, tolerability and outcomes. Future studies should address genetic markers for drug metabolism, side effects and treatment responses in diverse populations.
The availability of highly active antiretroviral therapy has reduced mortality associated with acquired immunodeficiency syndrome (AIDS) by at least 70%. A frequently prescribed initial therapy for HIV-infection is the non-nucleoside reverse transcriptase inhibitor efavirenz plus two nucleoside reverse transcriptase inhibitors. However, during the first weeks of therapy up to one-half of the patients who receive efavirenz experience central nervous system side effects including dizziness, insomnia, impaired concentration, somnolence, and abnormal dreams. Severe depression, aggressive behaviour and paranoid or manic reactions may also occur and such side effects may reflect varying efavirenz plasma concentrations.
Plasma clearance of efavirenz appears slower in African-Americans than in European-Americans. In addition, one study has suggested earlier virologic failure on efavirenz for African-Americans than Caucasians. A subsequent study did not confirm this difference, but demonstrated earlier treatment failure in Hispanics. It is important to identify the basis for such population differences as AIDS increasingly and disproportionately affects minorities in the United States and diverse populations worldwide.
Genetic variants in drug metabolism pathways have been shown to alter the safety and efficacy of other commonly prescribed medications. A cytochrome P4502C9 (CYP2C9) variant reduces the anticoagulant effect of warfarin, variants of CYP2D6 are associated with variable analgesic effects of codeine and with antidepressant effects of tricyclics, and a thiopurine methyltransferase variant is associated with bone marrow aplasia in persons prescribed 6-mercaptopurine or azathioprine.
Efavirenz is metabolized primarily by hepatic CYP2B6 with some involvement of CYP3A. Several single nucleotide polymorphisms of CYP2B6 yield numerous haplotypes. Large interindividual variability in the amount of CYP2B6 protein and catalytic activity in human liver have been reported, and functional differences between genetic variants in vitro have been described. Their consequences in vivo, however, are unknown. Although efavirenz is not a substrate for the multidrug transporter P-glycoprotein, one report suggested associations of an MDR1 polymorphism with lower plasma efavirenz plasma concentrations and greater CD4+ T-cell increases.
We examined possible associations between genetic variants in CYP2B6, CYP3A4, CYP3A5, and MDR1 and efavirenz plasma concentration-time profile, central nervous system side effects, tolerability, and short-term virologic and immunologic responses.
Of 190 efavirenz recipients with pharmacokinetic data from study A5097s, 157 had DNA available for analysis. Baseline characteristics are shown in Table 1. Although Hispanics were under-represented in A5097s in comparison with the parent study, A5095, subjects who underwent genetic analyses resembled other A5097s and A5095 participants.
The frequencies of CYP2B6, CYP3A4, CYP3A5, and MDR1 variants are presented in Table 2. Allelic frequencies differed between African-Americans and European-Americans as follows: CYP2B6 G516T (38.0 and 21.9%, P = 0.005), CYP2B6 C1459T (1.0 and 11.8%, P < 0.0001), CYP3A4 A-392G (63.0 and 3.4%, P < 0.0001), CYP3A5 A6986G (36.0 and 95.5%, P < 0.0001), MDR1 G2677T (11.0 and 44.0%, P < 0.0001), and MDR1 C3435T (22.0 and 55.6%, P < 0.0001), respectively. There were no significant deviations from Hardy-Weinberg equilibrium genotype frequency expectations when evaluated among African-Americans and European-Americans separately. As expected, there were strong associations between CYP3A4 A-392G and CYP3A5 G6986A, and between MDR1 G2677T and C3435T.
Plasma efavirenz exposure
Plasma efavirenz exposure was greater in African-Americans and Hispanics than in European-Americans, with median 24-h area under the curve (AUC0-24 h) and quartile values of 58.3 (40.9 to 99.0) μg h/ml, 66.1 (52.1 to 79.0) μg h/ml and 46.4 (37.7 to 58.7) μg h/ml, maximum concentration (C max) values of 3.14 (2.31 to 4.72) μg/ml, 3.18 (2.78 to 3.80) μg/ml and 2.53 (2.21 to 3.13) μg/ml, and C 24h values of 1.86 (1.17 to 3.47) μg/ml, 2.01 (1.57 to 2.76) μg/ml and 1.31 (1.00 to 1.80) μg/ml, respectively (overall P < 0.0001 for all comparisons). Values for C max, C 24h, and AUC0-24 h were highly correlated, and there was considerable overlap in the distribution of parameters between racial/ethnic populations.
Genetic variants and plasma efavirenz exposure
Among all subjects, CYP2B6 G516T, CYP3A4 A-392G, and CYP3A5 A6986G were associated with plasma efavirenz exposure by univariate analysis (P < 0.0001). The median AUC0-24 h of efavirenz was approximately three-fold higher in CYP2B6 position 516 T/T homozygotes as compared to G/G homozygotes, and was intermediate in G/T heterozygous suggesting a gene-dose effect. Similar differences for CYP2B6 were maintained in African-Americans and European-Americans analyzed separately. Although not statistically significant, associations for CYP3A4 A-392G, CYP3A5 A6986G and plasma efavirenz exposure within these populations were not inconsistent with observations for the entire cohort. For each subject, relationships between CYP2B6 G516T and AUC0-24 h according to population are shown in Fig. 1a. The Hispanic subject with T/T at CYP2B6 position 516 had higher plasma efavirenz exposure than all other Hispanics (not shown). The relationship between genotype and plasma efavirenz exposure was apparent throughout the 24 weeks.
Population pharmacokinetic modeling that adjusted for weight suggested that both CYP2B6 G516T and CYP3A5 A6986G were independently associated with efavirenz clearance. Relative to CYP2B6 position 516 G/G, efavirenz plasma clearance was 23% lower with G/T [95% confidence interval (CI) = -12 to -33%; P = 0.0001] and 54% lower with T/T (95% CI = -42 to -64%; P < 0.0001). Relative to CYP3A5 position 6986 A/A homozygotes, clearance was 10% higher (95% CI = -11 to 36%, P = 0.37) with A/G and 27% higher with G/G (95% CI = 5 to 53%, P = 0.014). After adjusting for CYP2B6 G516T and CYP3A5 A6986G there was evidence of lower efavirenz clearance for subjects who were neither European-Americans nor African-Americans (primarily Hispanics) in comparison with African-Americans (P = 0.05), with no apparent difference between European-Americans and African-Americans (P = 0.43).
Recursive partitioning for AUC0-24 h yielded splits on CYP2B6 G516T that partitioned the data according to the three genotypes. The CYP2B6 G516T T/T and G/T genotypes were further split according to CYP3A4 A-392G or CYP3A5 A6986G. Bootstrap results on the recursive partitioning indicated that a first split on CYP2B6 was robust in that 95% of the bootstrapped datasets had their first split on CYP2B6, and 87.5% were partitioned in the first two splits with one split on CYP2B6 and another on CYP2B6 (51.0%), CYP3A5 (21.6%) or CYP3A4 (14.9%). There were 154 datasets with the first three splits on the same variables as in the original dataset.
Because efavirenz can induce hepatic cytochrome P450 expression, we explored relationships between raw efavirenz plasma concentrations (without modeling) and study week. Median efavirenz plasma concentrations remained stable from study week 1 to 24. In addition, the gene-dose relationship between CYP2B6 position 516 genotype and efavirenz plasma concentrations was apparent at each study week.
Central nervous system side effects
Among all subjects, CYP2B6 G516T genotype was associated with efavirenz adverse central nervous system symptoms at 1 week (P = 0.036) but not significantly at week 24 (P = 0.76). At 1 week, median symptom questionnaire score changes from baseline associated with G/G, G/T, and T/T genotypes were +3.0, +8.0, and +8.0, respectively (Fig. 2). Attenuation of adverse experiences over time did not appear to be an artifact of premature study discontinuations. There were 14 evaluable subjects with the T/T genotype at each week. No subdomain of the questionnaire (vestibular symptoms, altered dreams, or difficulty sleeping) appeared to be specifically associated with CYP2B6 G516T genotype at week 1. There were no apparent associations with other genetic variants and adverse experience responses (P > 0.1 for each), or with other neuropsychological measures.
Tolerability, immunologic and virologic responses
Forty-seven subjects prematurely discontinued efavirenz-based regimens (n = 23) and/or experienced high-grade or treatment-limiting central nervous system toxicities (n = 31). There were no statistically significant relationships identified between CYP2B6, CYP3A4, CYP3A5, or MDR1 polymorphisms and efavirenz tolerability (including the 23 subjects who prematurely discontinued efavirenz) and immunologic or virologic responses over 24 weeks. Among subjects with CYP2B6 G/G, G/T and T/T genotypes, plasma HIV-1 RNA was less than 200 copies/ml at 24 weeks in 96, 96, and 100%, respectively (P = 0.744). Median (interquartiles) CD4+ T-cell increases from baseline were 121 (76 to 192) × 106cells/l, 126 (56 to 187) × 106cells/l, and 97 (37 to 101) × 106cells/l, respectively (P = 0.153).
CYP2B6 G516T in anonymized DNA
The frequency of the CYP2B6 G516T polymorphism was studied in 191 additional anonymized DNA samples from a reference population. Among 96 African-Americans CYP2B6 position 516 was G/G, G/T, and T/T in 40.6, 42.7, and 16.7%, respectively. In 95 European-Americans these percentages were 64.2, 29.5, and 6.3%, respectively. Thus, G516T allelic frequencies in the study subjects were consistent with those in the separate populations - African-Americans (38.0%) and European-Americans (21.1%).
Materials and methods
Study design and patients
This exploratory analysis included subjects who had been randomized to receive efavirenz while co-enrolled in Adult AIDS Clinical Trials Group (AACTG) studies A5095 and A5097s. Study A5095 is an ongoing placebo-controlled trial in which antiretroviral naïve subjects were randomized to efavirenz (600 mg once daily), abacavir (300 mg twice daily), or both, together with zidovudine and lamivudine. Efavirenz and abacavir were double-blinded. Subjects in two of the three arms received efavirenz. Subjects from both efavirenz arms were included in the present analysis, and abacavir assignment remains blinded because the A5095 study continues in active blinded follow-up. A total of 1147 subjects were enrolled in the United States and Puerto Rico, of whom 765 were randomized to receive efavirenz-based regimems.
Study A5097s, a substudy of A5095, characterized the central nervous system adverse effects and pharmacokinetics of efavirenz during the first 24 weeks of treatment. Of the 303 subjects in A5097s, 202 (26% of A5095 efavirenz recipients) were randomized to receive efavirenz. Human DNA was obtained under AACTG protocol A5128. These studies complied with the Helsinki Declaration, were approved by institutional review boards for each site, and subjects gave written informed consent. The Vanderbilt Committee for the Protection of Human Subjects and AACTG approved the use of DNA. Subjects self-identified as black not Hispanic, or white not Hispanic, in accordance with National Institutes of Health standards are, hereafter, referred to as African-American or European-American, respectively.
Central nervous system adverse effects
Central nervous system side effects were assessed by questionnaires at baseline and at weeks 1, 4, 12, and 24. A symptom questionnaire, designed for A5097s, included eighteen questions exploring common presumptive efavirenz side effects graded from zero (not at all) to four (extremely) and yielded total scores that could range from 0 to 72. Additional neuropsychological measures included NPZ3 (Trailmaking A and B and Wechsler Adult Intelligence Scale-Revised Digit Symbol), Pittsburgh Sleep Quality Index, Spielberger State-Trait Anxiety Inventory for Adults, Center for Epidemiologic Studies-Depression Scale, and Maryland Addictions Questionnaire B Short Form.
Identification of genetic variants
Genetic polymorphisms were identified by real-time polymerase chain reaction (PCR). Primers were synthesized by MWG Biotech, Inc. (High Point, North Carolina, USA). The following wild-type (w), variant (v), and common (c) primers were used, with sequences shown 5′ to 3′ and mismatched nucleotides bolded: CYP2B6 G516T (223 bp amplicon), w- GACCCCA CCTTCCTCTTCTAG, v-GACCCCACCTTCCTC TTCTAT, c-GGTCATCCTTTTCTCGTGTG; CYP2B6 C1459T (232 bp amplicon), w-AATACCCCCAACA TACCAGATTC, v-AATACCCCCAACATACCAG ATTT, c-TAATTTTCGATAATCTCACTCCTGC; CYP3A4 A-392G (73 bp amplicon), w-TCTATTA AATCGCCTCTCACT, v-TCTATTAAATCGCCT CTCACC, c-TGGGATGAATTTCAAGTATTTTG; CYP 3A5 A6986G (238 bp amplicon), w- TCCAAA CAGGGAAGAGAAAT, v-TCCAAACAGGGAAGA GAAAC, c-ACTGCCCTTGCAGCATTTAG; MDR1 G2677T/A (216 bp amplicon): w-AGTTTGACTCAC CTTCCCATC, v1-AGTTTGACTCACCTTCCCA TA, v2-AGTTTGACTCACCTTCCCATT, c-GCT ATAGGTTCCAGGCTTGCT; MDR1 C3435T (134 bp amplicon): w-GTGGTGTCACAGGAAGAGAAC, v-GTGGTGTCACAGGAAGAGAAT, c-ACTATAG GCCAGAGAGGCTGC. No other genes or polymorphisms were analyzed.
Incubation volumes were 25 μl with a 1 : 2 dilution of iQ SYBR Green Supermix (Bio-Rad Laboratories, Inc., Hercules, California, USA), 160 nmol/l of each primer, and 13 ng of DNA template. Genomic DNA digested with Xho1 (New England Biolabs, Inc., Beverly, Massachusetts, USA) was used for all reactions except CYP2B6 C1459T. Diluted PCR product from amplification of CYP2B6 exon 9 was used for CYP2B6 C1459T. Real-time PCR used an iCycler Thermal Cycler with Optical Module, and analyses were performed using iCycler iQ software (Bio-Rad Laboratories, Inc.). Assay validity was confirmed by sequencing, and controls of known sequence were run with each assay. Anonymized DNA was obtained from Coriell Institute for Medical Research (Camden, New Jersey, USA).
Plasma for determination of efavirenz by high-performance liquid chromatography was collected at weeks 1, 4, 12, and 24 of treatment. Sampling times were unspecified. Time of prior dose was obtained by patient report, and samples drawn greater than 96 h post-dose or in which efavirenz was not detected were excluded [13 (1.9%) of 684 samples]. Population pharmacokinetic modeling was performed assuming a one-compartment, open model with first-order absorption using NLME version 3.2 library. The model was parameterized in terms of drug clearance, volume of distribution and absorption rate, and assumed complete absorption of the dose. To facilitate convergence, and because few concentration-time points were obtained during the absorption phase the absorption rate constant (K a) was fixed at 1.4/h, which is equivalent to an absorption half-life of approximately 0.5 h. Sensitivity analysis of this assumption was shown to be reasonable. Briefly, for all pairs of models using values of K a ranging from 0.4/h to 1.8/h the Spearman's rank correlation of the subject estimates of clearance, area under the curve (AUC) and concentrated at 24 h post dose (C24) was > 0.99. Additionally, with the exception of the magnitude of two outlying observations, the distributions of the subject estimates of clearance, AUC and C24 were invariant to the choice of K a over the range 0.4/h to 1.8/h. The pharmacokinetic parameters analyzed here were estimated from the A5097s efavirenz cohort with evaluable pharmacokinetic data (n = 190). Twelve subjects lacked evaluable pharmacokinetic data.
Only data collected while subjects were receiving efavirenz were included. Deviation of genotype frequencies from Hardy-Weinberg equilibrium expectations was evaluated with a χ2 test (1 d.f.). Baseline characteristics were compared using χ2 or Kruskal-Wallis tests. For each outcome and gene, two-sided Jonkheere-Terpstra or Cochran-Armitage tests for trend were performed. Tarone's exact test for ordered k-sample censored outcomes evaluated time-to-event outcomes. Plasma HIV-1 RNA concentrations below limits of quantification were imputed at the assay detection limit (50 copies/ml). Recursive partitioning on pharmacokinetic parameters was carried out in Splus, Version 6.0, using the RPART library. Standard regression analyses were not used because these approaches are not valid when data are estimates. Recursive partitioning recursively splits the sample into groups wherein each group of subjects is split into two subgroups in which subjects are more similar. The procedure examined all possible binary splits of the data along each predictor variable that most reduced the deviance. Optimal splits minimized cross validation error, and robustness was evaluated by bootstrapping, using 1000 re-sampled datasets. Subjects with an MDR 1 exon 21 A allele were excluded because only six subjects had this allele, and all genes and baseline demographic variables were considered as predictors, in the recursive partitioning. There was no adjustment for multiple comparisons. Allelic frequency was estimated from total copies of individual alleles divided by all alleles in the population.