A unified in vivo modeling approach for quantitative prediction of the impact of gene polymorphism and drug interactions on drug exposure
Sylvain Goutelle (1,2,3), Michel Tod (1,3), Laurent Bourguignon (1,2), Nathalie Bleyzac (1), Johanna Berry (1), Fannie Clavel-Grabit (1), and the Genophar II working group
(1) University Hospitals of Lyon; (2) UMR CNRS 5558, University of Lyon 1; (3) Department of Pharmacology, School of Pharmacy, University of Lyon 1, Lyon, France
Objectives: We propose a unified approach to predict in vivo variation in drug exposure due to gene polymorphism or drug-drug interactions (DDI). An application to drugs metabolized by cytochrome 2C19 (CYP2C19) is presented.
Methods: The approach is based on frameworks proposed by Ohno  and ourselves  for drug interactions and gene polymorphism, respectively. The metrics used is the ratio of altered drug AUC*, which may be caused by gene polymorphism or drug interaction, to reference AUC measured in patients with no mutation or no interaction (AUC*/AUC, denoted RAUC). For CYP2C19 gene polymorphism, prediction of the RAUC is based on a model with two parameters: the contribution ratio of the drug CR (fraction of oral clearance dependent of CYP2C19), and the fraction of activity FA of allele combinations (relative activity of CYP2C19 in mutants compared with wild-type extensive metabolizers). For drug interactions, the two-parameter model includes CR of the victim drug and the inhibition ratio of the inhibitor (IR) which is a measure of inhibitor potency.
First, initial estimates of CRs and FAs were obtained from the literature, for 30 CYP2C19 drug substrates and 5 genotypes. Then, these values were used to predict RAUC which were compared with observed RAUC from another set of published data (external validation). Third, all data from step 1 and 2 were used to estimate posterior distributions of CRs, FAs, and AUCs by using Bayesian orthogonal regression in the Winbugs software. For drug interactions, previously estimated CRs and published data were used to estimate IRs of 10 inhibitors, by use of similar Bayesian approach. Final estimates of RAUC were compared with observed values from all genotype and DDI published studies.
Results: Published data were available for 111 (drug, genotype) and 23 (victim drug, inhibitor) pairs. The mean prediction errors of RAUC were -0.142 and -0.60, while the mean absolute prediction errors were 0.58 and 1.02 for genotype and DDI data, respectively. Overall, only 5 out of 134 predicted RAUC were outside the 50-200% range of observed RAUC.
Conclusions: This approach showed good predictive performance. It also provides unpublished prediction of RAUC corresponding to rare genotypes (e.g., ultrametabolizers *17/*17) and DDI for 30 drugs metabolized by CYP2C19, including the widely prescribed proton-pump inhibitors and clopidogrel.
 Ohno et al. Clin Pharmacokinet 2007;46:681-96
 Tod et al. Clin Pharmacol Ther 2011;90:582-87