Metabolic engineering and deep learning-driven protein engineering for N-Acetylneuraminic acid biosynthesis in Escherichia coli

Abstract

N-Acetylneuraminic acid (NeuAc) is a sialic acid valued in pharmaceuticals and infant nutrition, and microbial synthesis offers a scalable route once the pathway’s catalytic bottlenecks are relieved. We combine deep learning with metabolic engineering to build a high-titer NeuAc-producing E.coli strain. Modular pathway engineering reaches 1.25 g L−1 and pinpoints N-acetylglucosamine 2-epimerase (AGE) as the rate-limiting step. We develop DLCatalysis, a deep learning framework that predicts kcat Km−1 directly from protein sequence and substrate, and use it to mine AGEBf that lifts the titer to 6.84 g L−1. DLCatalysis-guided redesign of AGEBf and NeuBNm raises NeuAc to 9.27 g L⁻¹, and identifying and deleting exuT, a previously unannotated NeuAc transporter, blocks product reuptake. In 5-L fed-batch fermentation the optimized strain produces 85.5 g L−1, showing that AI-guided enzyme discovery can resolve the bottlenecks that have limited microbial NeuAc production. N-Acetylneuraminic acid (NeuAc) is a valuable pharmaceutical. Here the authors develop DLCatalysis, a deep learning framework that predicts kinetics from protein sequence and substrate, and use DLCatalysis to mine enzymes for NeuAc production in E. coli.