Reprogramming Enzyme Specificity through Multi-substrate Co-evolution

Abstract

Understanding and manipulating enzyme specificity are critical to drug development. In the past two decades, directed evolution has been proven a successful methodology to obtain enzyme variants with a desired and oftentimes new-to-nature function. However, most directed evolution strategies aim at a single trait. As a result, even for similar favorable specificities, siloed and repeated evolution efforts in lab are required. Meanwhile, there remains a lack of understanding of how new specificities emerge in evolution and how different specificities trade off. Here, we reviewed protein sequence-activity relationship studies with diversified phenotypic measurements. We tied our studies around R. trifolii MatB, a malonyl-CoA synthetase. We developed multiagent screening, a novel directed evolution strategy that efficiently evolves enzymes toward multiple specificities. Analysis of mutations identified revealed that distant specificity-altering mutations are destabilizing and dissociating side-chain interactions between remote residues. Moreover, we generated a multi-substrate fitness landscape of MatB. The data revealed distinct patterns of substrate-specific effects between active site and surface mutations, which elucidate the mechanism of how MatB accommodates structurally diverse substrates. A comprehensive mapping of evolutionary trajectories also indicated that structurally distinct substrates are more synergistic in multiagent screening. Lastly, we evaluated a few protein language models as variant fitness predictors and sequence representation methods on our data. We highlighted the difficulty of obtaining a model that effectively leverages information from multiple specificities. Together, our study improves understanding of enzyme promiscuity and paves the way for future protein sequence-activity studies with multiple specificities.