Background:  Relationships between human genetic variants and efavirenz (EFV) pharmacokinetics, efficacy, and toxicity are incompletely understood. In previous univariate analyses involving ACTG protocol 384, CYP2B6 516G>T was associated with greater EFV plasma AUC_24h values (confirming findings from ACTG studies A5095/A5097s), but not virologic response. Conversely, MDR1 3435C>T was associated with decreased EFV virologic failure but not EFV plasma exposure. The present study explored the ability of gene-gene interactions to predict EFV pharmacokinetics, efficacy and toxicity among ACTG 384 participants.

Methods:  The present analysis involved ART-naïve subjects who received EFV (with or without nelfinavir [NFV]) plus 2 nucleoside analogues (zidovudine/lamivudine [AZT/3TC] or stavudine [d4T]+didanosine [ddI]) in ACTG 384, and contributed specimens to the ACTG Human DNA Repository under Protocol A5128. Population pharmacokinetics was estimated from a non-linear mixed effects model assuming 1 compartment and first-order absorption. Wee identified 9 single nucleotide polymorphisms (SNP) in CYP2B6, 3A4, 3A5, 2C19, and MDR1. For analyses of EFV exposure, plasma EFV AUC_24h values in the top quartile were compared with values in the lower 3 quartiles. Virologic and toxicity failure analyses were limited to subjects who received EFV without NFV. Main effects and gene-gene interactions were explored using multifactor dimensionality reduction.

Results:  The 305 study subjects included were 49% white, 33% black, and 16% Hispanic. Plasma EFV AUC_24h values in the top quartile were predicted by CYP2B6 516G>T in the total cohort, and among whites and blacks analyzed separately (accuracy = 73%, 80%, and 69%, respectively; p <0.001 for each). Among whites a 2-gene interaction between CYP2B6 516G>T and MDR1 2677G>T predicted higher plasma EFV exposure (82% accuracy, p <0.001). In blacks an interaction between CYP2B6 516G>T and CYP3A5 6896A>G predicted EFV exposure (64% accuracy, p <0.013). Among 155 subjects who received EFV without NFV, decreased virologic failure was predicted by an interaction between MDR1 2677G>T and CYP2B6 516G>T (65% accuracy, p <0.001), and increased toxicity failure by an interaction between MDR1 2677G>T and MDR1 3435C>T (71% accuracy, p <0.001). MDR1 3435C>T and 2566G>T were in linkage disequilibrium.

Conclusions:  Interactions between variant drug metabolism and transporter genes may predict EFV pharmacokinetics and treatment responses. The predictive accuracy of gene-gene interactions may differ among racial/ethnic groups.