Quantitative Prediction of CYP3A4 Induction: Impact of Measured, Free, and Intracellular Perpetrator Concentrations from Human Hepatocyte Induction Studies on Drug-Drug Interaction Predictions

Typically, concentration-response curves are based upon nominal inducer concentrations for in-vitro-to-in-vivo extrapolation of CYP3A4 induction. The limitation of this practice is that it assumes the hepatocyte culture model is a static system. We assessed whether correcting for: 1) changes in perpetrator concentration in the induction medium during the incubation period, 2) perpetrator binding to proteins in the induction medium, and 3) nonspecific binding of perpetrator can improve the accuracy of CYP3A4 induction predictions. Of the seven compounds used in this evaluation, significant parent loss and nonspecific binding were observed for rifampicin (29.3–38.3%), pioglitazone (64.3–78.6%), and rosiglitazone (57.1–75.5%). As a result, the free measured EC₅₀ values (EC_50u) of pioglitazone, rosiglitazone, and rifampicin were significantly lower than the nominal EC₅₀ values. In general, the accuracy of the induction predictions, using multiple static models, improved when corrections were made for measured medium concentrations, medium protein binding, and nonspecific binding of the perpetrator, as evidenced by 18–29% reductions in the root mean square error. The relative induction score model performed better than the basic static and mechanistic static models, resulting in lower prediction error and no false-positive or false-negative predictions. However, even when the EC_50u value was used, the induction prediction for bosentan, which is a substrate of organic anion transporter proteins, was overpredicted by approximately 2-fold. Accounting for the ratio of unbound intracellular concentrations to unbound medium concentrations (K_{puu,in vitro}) (0.5–7.5) and the predicted multiple-dose K_{puu,in vivo} (0.6) for bosentan resulted in induction predictions within 35% of the observed interaction.