In Silico Prediction of a Multimutational Stereoselective Alcohol Dehydrogenase

Three decades ago, directed evolution of enzyme stereoselectivity opened an exciting research field, requiring laborious high-throughput screening. Despite improvements in mutagenesis techniques, the final objective has not been reached: The computational prediction of an enantioselective variant without any screening during the whole iterative mutagenesis procedures. We demonstrate here that in silico prediction of an enantioselective multimutational variant is possible. A difficult goal was chosen by aiming for complete inversion of enantioselectivity of a wild-type enzyme (alcohol dehydrogenase, 99% S) to 99% (R) of the best predicted variant. This was achieved, guided by the substrate binding pocket reshaping technique, past mutational data, molecular dynamics computations, and the Rosetta algorithm for identifying residues as hotspots. Specific amino acid exchanges at these sites were then predicted. Following the experimental validation of the resulting superior variant, the deletion of residue Alanine 85 functions as a conformational proofreading switch that dictates stereochemical outcomes. Our study includes the analysis of epistatic mutational effects and molecular dynamics simulations in the quest to understand all results, including the stereoselectivity switch.