Machine-learning guided engineering of fatty acyl-ACP reductases

Tuesday October 4th, 4-5pm EST | Jonathan Greenhalgh, PhD | A-Alpha Bio

Alcohol-forming fatty acyl reductases (FARs) catalyze the reduction of thioesters to alcohols and play a key role in synthetic microbial pathways for producing fatty alcohols. Though many metabolic engineering strategies utilize FARs to produce fatty alcohols from intracellular acyl-CoA and acyl-ACP pools, FAR enzyme activity, especially on acyl-ACPs, is limiting and remains a bottleneck to high-flux alcohol production. Due to the low throughput methods needed to characterize FAR activity in vivo, FAR activity is also difficult to improve by traditional directed evolution approaches. Using a machine learning (ML)-driven approach, we iteratively search the FAR protein fitness landscape for improved proteins, and over the course of ten design-test-learn rounds, we engineer FAR enzymes with enhanced activity acyl-ACPs. Our best engineered FARs produce over twofold the titer of fatty alcohols than the best natural sequences in vivo. Further we characterize the top designed FAR in vitro and show that it has a significantly enhanced catalytic rate on palmitoyl-ACP compared to the wild-type enzyme. Finally, we analyze the sequence-function data to identify features that correlate with the in vivo enzyme activity. This work demonstrates the power of ML methods for searching fitness landscapes of traditionally difficult-to-engineer proteins.

Paper: https://www.nature.com/articles/s41467-021-25831-w

Recording Link: https://www.youtube.com/watch?v=2yS3jydA8Ls