(T1130-10-58) Predictability of In Vitro Hepatocyte Induction Data in a PBPK Framework for CYP3A4 Induction Risk Assessment

Purpose: Cytochrome P450 3A4 (CYP3A4) metabolizes more than 50% of drugs on the market. The induction of CYP3A4 can lead to the accelerated metabolism of a drug, altering its intended therapeutic effect. Therefore, an important aspect of drug development is to determine the risk for a compound to be a precipitant of induction on other compounds. As clinical drug-drug interaction (DDI) trials are time intensive and expensive, quantitative in vitro to in vivo approaches have been utilized to predict the induction risk, assisting in the decision-making process for a clinical trial. This work assesses the accuracy of induction predictions using in vitro hepatocyte induction data used in a physiologically based pharmacokinetic (PBPK) framework.

Methods: Twelve internal compounds and eight commercial compounds were selected as precipitants that ranged from no CYP3A4 induction effect to a strong CYP3A4 inducer. PBPK models with first order absorption for these compounds were built in the commercially available Simcyp Simulator using physiochemical data and in vitro hepatocyte induction data. Clinical data was used to estimate the first order absorption rate, clearance, and volume of distribution. From the in vitro hepatocyte induction studies, induction titration curves were fit and the induction parameters EC₅₀ and E_max for a compound were determined. These were then calibrated to rifampin, a strong CYP3A4 inducer that was evaluated in the same experiment, and input into the PBPK model. The model was validated using multiple dose observed concentration-time profiles. Using the validated models, a DDI trial with the compound as the precipitant and a CYP3A4 substrate as the object was simulated in Simcyp and the AUC ratio (ratio of the CYP3A4 substrate coadministered with the inducer to the object alone) was calculated. After the initial assessment of model predictability, the predictions were recalculated with a model using a mechanistic absorption model that describes the drug concentration and CYP3A4 expression level across various intestinal segments, called the Advanced Dissolution, Absorption, and Metabolism (ADAM) model.

Results: In our preliminary assessment, 14 out of 20 compounds were correctly predicted as an inducer or not an inducer. All the misclassifications predicted a risk of induction when no induction was observed in the clinic. The predicted versus observed plot can be seen in Fig 1. When including the ADAM model for absorption, an improvement is seen in the predictions, reducing the number of overpredicted DDIs.

Conclusion: While the initial PBPK model with first order absorption shows an overprediction of induction for some compounds, it correctly classified all compounds that showed induction in the clinic as inducers. By including the mechanistic modeling of the gut into the PBPK model, there was a decrease in the overprediction of induction risk, something which could be further improved with other model refinements.


Figure 1: Assessing the initial predictiveness of the PBPK model