Protein language model-assisted directed evolution of cyclodextrinase Enables Precision α-O-Oligosaccharide synthesis

Stereoselective synthesis of α-O-oligosaccharides remains a key challenge in glycobiology. While glycoside hydrolase-mediated transglycosylation is promising, current methods yield excessive byproducts and show low specificity. Here, we establish a glycoside hydrolase-based system for precise oligosaccharide synthesis using cyclodextrin as donor. Guided by an extra sugar binding space (ESBS) motif probe, a cyclodextrinase from Paenibacillus sp. MY03 was identified with a naturally high transglycosylation-to-hydrolysis (T/H) ratio. Using Pro-PRIME, a protein language model, we optimized three enzymatic properties-enhancing transglycosylation, reducing hydrolysis, and improving regioselectivity-based on minimal beneficial mutation data. Among 68 screened variants, the top mutant showed a 12-fold higher T/H ratio and improved 4-nitrophenyl-α-d-maltoheptaoside (pNP-G7) yield from 63 % to 98 %. The engineered enzyme also showed broad substrate promiscuity, underscoring its utility for diverse biotechnological applications. This study advances α-O-oligosaccharide synthesis and demonstrates the power of language model-guided enzyme engineering to balance competing catalytic activities.