Improving expression of Photobacterium sp. JT-ISH-224 α2,6-sialyltransferase using ProteinMPNN and machine learning...

Industrial-scale production of 6'-sialyllactose (6'-SL), a sialylated human milk oligosaccharide, has recently attracted increasing attention. As a crucial enzyme for both microbial synthesis approaches and enzymatic catalysis, several α2,6-sialyltransferases (α2,6-SiaTs) have been reported and characterized. However, the low expression levels and poor stability of α2,6-SiaTs remain major bottlenecks in the construction of efficient 6'-SL production systems. In this study, we applied a deep-learning-based protein generative model, ProteinMPNN, to redesign an α2,6-SiaT from Photobacterium sp. JT-ISH-224 (pst6-224), resulting in a marked improvement in its heterologous expression in Escherichia coli BL21(DE3). Furthermore, machine learning-guided combinatorial mutagenesis was employed to restore catalytic activity in the redesigned enzyme. The expression level of the final variant was increased by approximately tenfold compared with that of the wild-type pst6-224, and a higher 6'-SL titer was achieved in a 1 L cascade catalytic system starting from CMP. This work demonstrates the potential of machine learning-assisted protein engineering and provides an α2,6-SiaT variant with high expression levels for the biological synthesis of 6'-SL.