Computational Stabilization of a Non-heme Iron Enzyme Enables Efficient Evolution of New Function

Abstract

Biocatalysis is a promising and sustainable alternative to conventional chemical manufacturing, which is responsible for a significant portion of global energy consumption. A major gap, however, is engineering enzymes for industrially relevant novel reactivity outside of their biological contexts. Combining the power of C-H functionalization with the superfamily of non-heme iron enzymes can address this gap and open the door to a suite of new-to-nature biocatalysts that can be readily incorporated into synthetic chemistry workflows. This thesis provides an overview of the conserved non-heme iron enzyme mechanism, which gives context to both substrate promiscuity and novel reactivity. Because wild-type enzymes often exhibit low reactivity for promiscuous substrates or reactions, protein engineering is necessary to optimize activity. Recent engineering and directed evolution efforts with non-heme iron enzymes are described, as well as general engineering challenges within this superfamily. One challenge that we explored in depth is how to address activity-stability tradeoffs during directed evolution. Emerging deep learning methods for enzyme stabilization offer a potential solution to these tradeoffs, though their effectiveness for industrially relevant enzymes remains to be fully evaluated. Here, we evaluate the use of the deep-learning tool ProteinMPNN for redesigning a non-heme amino acid iron hydroxylase with promiscuous C-H hydroxylation activity towards a carboxylic acid substrate. We describe the process by which critical residues in our candidate enzyme were fixed prior to design to maintain catalytic function, and compare directed evolution trajectories of both wild-type and a stabilized redesign. In alignment with previous findings on how stability can promote evolvability, we found that a stabilized starting point leads to more efficient directed evolution. Together, our results suggest that user-friendly deep-learning tools like ProteinMPNN could be readily incorporated into enzyme engineering workflows to generate novel and robust biocatalysts.