033: Improving the Working Models for Drug-Drug Interactions: Impact on Preclinical & Clinical Drug Development

Statement of Purpose: Drug-drug interactions (DDIs) may alter the pharmacokinetics of a drug and cause unexpected changes in exposure, potentially resulting in increased off-target effects or decreased efficacy. The US Food & Drug Administration (FDA) and other health regulatory agencies have published guidelines to evaluate DDIs, which include methods to predict clinical DDIs using data from in vitro and preclinical systems. The objectives of this study were to evaluate the predictive performance of static models via an in vitro to in vivo extrapolation (IVIVE) approach using Boehringer Ingelheim (BI) proprietary compounds, and to introduce methods to improve the DDI risk assessment in preclinical and clinical development.

Description of Methods & Materials: A total of 40 clinical studies, encompassing 20 BI proprietary compound-CYP enzyme pairs, were used to assess the predictive performance of FDA static models. This clinical dataset included small molecules with a diverse range of physicochemical properties, spanning various therapeutic areas. Inhibition (Ki, IC₅₀, KI, K_inact) and induction (E_max, EC₅₀) parameters were obtained from standard metabolic assays (using human liver microsomes and/or hepatocytes). Pharmacokinetic parameters (e.g., AUC_0-t, C_max, t_1/2), determined via noncompartmental analysis, were obtained from in-house clinical studies. BI compounds were assessed as both the object drug and precipitant of reversible inhibition, time-dependent inhibition and induction of CYP1A2, CYP2C8, CYP2C9, CYP2D6 and/or CYP3A4.

Data & Results: Application of basic model to assess BI compounds as inhibitors of DDIs (n=20) resulted in true positive/negative, false positive and false negative prediction rates of 74%, 26% and 0%, respectively. In comparison, evaluation using the mechanistic static model resulted in prediction rates of 90%, 10% and 0%, respectively. Assessment of BI compounds as inducers of DDIs (n=8) using basic models resulted in prediction rates of 37%, 63% and 0%, respectively; whereas the prediction rates with the mechanistic static model were 75%, 25% and 0%, respectively. When evaluating BI compounds as the object drug in DDIs, the mechanistic static model successfully predicted the exposure ratio (AUC in the presence-to-absence of inhibitor) to within 2-fold of that observed for 60% of the compounds. Modification of the current working model to include renal excretion data from Phase 1 studies resulted in 70% of the tested compounds falling within the 2-fold range. Similar improvements were observed (79% within 2-fold) using renal excretion data recovered from pre-clinical species (i.e., rats) (Figure 1).

Interpretation, Conclusion or Significance: Although the mechanistic static model provides greater prediction accuracy, the input required to generate robust extrapolations can be time consuming and costly. The current study demonstrated that the basic models should be used in early drug discovery to “rule out” DDI risks because of the minimal inputs required and the low rate of false negative predictions. Mechanistic models can then be implemented for compounds that require further evaluation, whereas clinical studies are reserved for compounds that are predicted to have DDI liabilities in both models. This study also demonstrated that addition of renal excretion to the mechanistic static model resulted in improved DDI prediction accuracy. In summary, IVIVE is a powerful tool that can be used to minimize the need for unnecessary clinical studies and reduce the overall costs associated with drug development.

Disclosures: David Joseph, Ashish Sharma and Fenglei Huang are employees of Boehringer Ingelheim; James Nguyen was a summer intern of Boehringer Ingelheim.