ABSTRACT:The prediction of in vivo drug-drug interactions from in vitro enzyme inhibition parameters remains challenging, particularly when time-dependent inhibition occurs. This study was designed to examine the accuracy of in vitro-derived parameters for the prediction of inhibition of CYP3A by erythromycin (ERY). Chronically cannulated rats were used to estimate the reduction in in vivo and in vitro intrinsic clearance (CL int ) of midazolam (MDZ) after single and multiple doses of ERY; in vitro recovery of CL int was determined at 1, 2, 3, and 4 days after discontinuation of ERY. Enzyme inhibition parameters (k inact , K I , and K i ) of ERY were estimated in vitro by using untreated rat liver microsomes. In vivo enzyme kinetic analysis indicated that single and multiple doses of ERY (150 mg/kg i.v. infusion over 4 h) reduced MDZ CL int by reversible and irreversible mechanisms, respectively. CYP3A inactivation after multiple doses of ERY treatment reflected metabolic intermediate complex formation without a significant change in hepatic CYP3A2 mRNA. A physiologically based pharmacokinetic model of the interaction between ERY and MDZ predicted a 2.6-fold decrease in CYP3A activity after repeated ERY treatment using in vitro-estimated enzyme inhibition parameters and in vivo degradation half-life of the enzyme (20 ؎ 6 h). The observed -fold decreases were 2.3-fold and 2.1-fold for the in vitro-estimated CYP3A activity and the in vivo CL int , respectively. This study demonstrates that in vivo DDIs are predictable from in vitro data when the appropriate model and parameter estimates are available.The prediction of in vivo drug-drug interactions (DDIs) from in vitro data has met with limited success, and the general applicability of such predictions is unclear (von Moltke et al., 1998;Wang et al., 2004;Obach et al., 2005Obach et al., , 2007. As drugs capable of mechanismbased inhibition (MBI) have emerged as clinically important CYP3A inhibitors (Zhou et al., 2004), the uncertainty in predictive accuracy has increased. Approaches to predicting MBI-based DDIs vary from a relatively simple method using a single-inhibitor concentration to more complicated physiologically based pharmacokinetic (PBPK) models that consider the change of inhibitor and substrate concentrations with time (Mayhew et al., 2000;Ito et al., 2003;Wang et al., 2004;Obach et al., 2007). Changes in the amount of active enzyme in the presence of a mechanism-based inhibitor are estimated by using the in vitro enzyme inactivation parameters, K I and k inact , which can be used to predict the increase in the in vivo exposure of the substrate. These predictions often suffer from poor characterization of substrate and inhibitor properties in vitro and/or in vivo and from uncertainties in predictive model parameter values. For example, DDIs may reflect simultaneous reversible inhibition, irreversible inhibition, and induction at multiple sites of enzyme expression, but the relative contribution of each type of interaction in vivo may be difficult to d...