Computationally guided enzyme engineering for regioselective synthesis of fucosylated human milk oligosaccharides

Human milk oligosaccharides (HMOs) are key bioactive components of human milk that support infant health and microbiome development. Prevalent HMOs include the internally fucosylated pentasaccharides lacto-N-fucopentaose II (LNFP II) and lacto-N-fucopentaose III (LNFP III), which are absent from infant formula. Their enzymatic synthesis from simpler HMOs such as 3-fucosyllactose (3FL), lacto-N-tetraose (LNT) and lacto-N-neotetraose (LNnT) represents an important step towards bridging this gap, especially now that these simpler HMOs are available on an industrial scale. We evaluated the use of the GH29B α-1,3/4-L-fucosidase SpGH29C from Streptococcus pneumoniae for transfucosylation at equimolar donor-to-acceptor ratio and applied the computational pipeline BindScan to design variants with reduced hydrolytic activity to avoid undesirable product hydrolysis. Guided by these predictions, we generated and tested 21 variants of SpGH29C, achieving significantly reduced hydrolysis and enhanced transglycosylation yields. Variants W264F and D257N reached LNFP II yields of up to 73% and 68%, respectively, while A173H improved LNFP III formation to 53%. Importantly, the product levels remained stable over 24 h as the variants displayed significantly decreased product hydrolysis as intended. Further binding analyses with BindScan enabled rational targeting of regioselectivity, identifying W211 as a key position influencing branched vs. linear product formation for LNFP II synthesis, while F202 and D257 variants improved regioselectivity in LNFP III synthesis. This study demonstrates that computationally guided protein engineering can optimize glycosidase-catalyzed transglycosylation and provides a framework for designing regioselective biocatalysts for complex oligosaccharides synthesis.