Artificial intelligence-guided MOF immobilization of Leloir glycosyltransferases for enhanced catalytic efficiency

Leloir glycosyltransferases (LGTs) are key enzymes in glycan assembly, prone to flocculation and inactivation during practical application. Unlike many other commercially available enzymes, conventional immobilization strategies often lead to the inactivation of LGTs. This study proposes the use of structurally tunable metal organic frameworks (MOFs) as immobilization carriers. To overcome the complexity of screening diverse MOF parameters, we developed an integrated molecular dynamics simulations (MD)-machine learning (ML) strategy that revealed hydrophilic ZIF-90 as the optimal carrier. We further optimized the synthesis conditions for immobilized LGTs, achieving efficient enzyme loading and high catalytic performance. Using KfoC and PmHS2 as representative model enzymes, both immobilized enzymes retained over 98% of their activities. Notably, PmHS2@ZIF-90 exhibited up to 37.2-fold and 8.0-fold activity enhancements relative to the free enzyme under alkaline and elevated-temperature conditions, respectively, demonstrating exceptional stability and tolerance. Meanwhile, the synthesized PmHS2@ZIF-90 was successfully applied in the synthesis of heparosan polysaccharide. This rational immobilization platform substantially surpasses conventional empirical strategies and offers a broadly applicable framework for the customized immobilization of LGTs in carbohydrate biomanufacturing.