Computationally supported inversion of ketoreductase stereoselectivity

Whereas directed evolution and rational design by structural inspection are established tools for enzyme redesign, computational methods are less mature but have the potential to predict small sets of mutants with desired properties without laboratory screening of large libraries. We have explored the use of computational enzyme redesign to change the enantioselectivity of a highly thermostable alcohol dehydrogenase from Thermus thermophilus in the asymmetric reduction of ketones. The enzyme reduces acetophenone to (S)-1-phenylethanol. To invert the enantioselectivity, we used an adapted CASCO workflow which included Rosetta for enzyme design and molecular dynamics simulations for ranking. To correct for unrealistic binding modes, we used Boltzmann weighing of binding energies computed by a linear interaction energy approach. This computationally cheap method predicted four variants with inverted enantioselectivity, each with 6-8 mutations around the substrate-binding site, causing only modest reduction (2- to 7-fold) of kcat/KM values. Laboratory testing showed that three variants indeed had inverted enantioselectivity, producing (R)-alcohols with up to 99% enantiomeric excess. The broad substrate range allowed reduction of acetophenone derivatives with full conversion to highly enantioenriched alcohols. The results demonstrate the use of computational methods to control ketoreductase stereoselectivity in asymmetric transformation with minimal experimental screening.