Substrate-specific screening for mutational hotspots using biased molecular dynamics simulations

Prediction of substrate-specific mutational hotspots for enzyme engineering is a complex and computationally intensive task. This becomes particularly challenging when the available crystal structures have no ligand, bind a distant homologue of the desired substrate, or hold the ligand in a nonproductive conformation. To address that shortcoming, we present a combined molecular dynamics simulation and molecular docking protocol to predict the conformation of catalytically relevant enzyme-ligand complexes even in the absence of a ligand-bound structure. We applied the adaptive biasing force method to predict the ligand-specific path of diffusion of a fatty acid substrate from the bulk media into the active site of cytochrome P450 CYP102A1 (BM3). Starting with a ligand-free crystal structure, we successfully identified all residues known to be involved in palmitic acid binding to BM3. The binding trajectory also revealed a yet unknown binding residue, Q73, which we confirmed experimentally. Building the free-energy landscape illustrates that, similar to human cytochrome P450s, binding is multistep and does not follow simple Michaelis-Menten kinetics. We confirmed the robustness of the method using a structurally distinct substrate, the small aromatic indole. We then applied the predicted BM3:palmitate complex to molecular docking of a library of 29 palmitate analogues. This produced catalytically relevant binding poses for the entire library, while docking directly into ligand-free and ligand-bound crystal structures gave poor results. This fast and simple computational method is broadly applicable for predicting binding hotspots in a substrate-specific manner and has the potential to drastically reduce the experimental screening effort to tailor an enzyme to substrates of interest.