Chemoinformatic-guided engineering of polyketide synthases

Polyketide synthase (PKS) engineering is an attractive method to generate new molecules such as commodity, fine and specialty chemicals. A significant challenge in PKS design is engineering a partially reductive module to produce a saturated β-carbon through a reductive loop exchange. In this work, we sought to establish that chemoinformatics, a field traditionally used in drug discovery, could provide a viable strategy to reductive loop exchanges. We first introduced a set of donor reductive loops of diverse genetic origin and chemical substrate structures into the first extension module of the lipomycin PKS (LipPKS1). Product titers of these engineered unimodular PKSs correlated with atom pair chemical similarity between the substrate of the donor reductive loops and recipient LipPKS1, reaching a titer of 165 mg/L of short chain fatty acids produced by Streptomyces albus J1074 harboring these engineered PKSs. Expanding this method to larger intermediates requiring bimodular communication, we introduced reductive loops of divergent chemosimilarity into LipPKS2 and determined triketide lactone production. Collectively, we observed a statistically significant correlation between atom pair chemosimilarity and production, establishing a new chemoinformatic method that may aid in the engineering of PKSs to produce desired, unnatural products.