Combined Enhanced Conformational Sampling and Protein-Ligand Dockings on CYP3A4 Improves in Silico Prediction of Drug Metabolism

Cytochrome P450 3A4 (CYP3A4) is the primary drug-metabolizing enzyme in humans. This is largely due to its highly flexible structure, which enables it to bind a broad spectrum of xenobiotics with a wide variety of shapes and physicochemical properties. This plasticity, however, presents challenges for in silico substrate prediction methods, in particular those that consider only a limited conformational space of the receptor or that rely on static crystal structures. In this study, we aim to explore the conformational landscape of CYP3A4 using Gaussian accelerated Molecular Dynamics (GaMD) and then to assess the viability of using this method to identify pro-reactive conformations through protein-ligand docking. GaMD simulations were initiated from distinct X-ray structures in triplicate 500 ns runs. The results demonstrate that the simulations cover a very large conformational space, including all reported X-ray structures as well as conformations with significantly expanded active-site geometries unseen in the experimental data. Protein-ligand docking of known CYP3A4 substrates onto the main cluster representatives from GaMD simulations allows the identification of metabolically active poses of the substrates. This is true even in cases where substrate-bound X-ray structures or docking against a single structure fails to yield productive results, as in the case of CYP3A4–Erythromycin. Furthermore, a classifier trained on ensemble docking scores discriminates CYP3A4 substrates from non-substrates better than any single conformation (ROC-AUC 0.735 vs 0.697), with performance increasing with ensemble size. Overall, the study shows that enhanced conformational sampling is crucial for accurately modelling highly flexible enzymes like CYP3A4 and can significantly improve metabolic predictions.

Preprint