Enhancing the Promiscuous Phosphorylation Activity of Acid Phosphatase via Transition-State-Stabilization-Mediated...

Acid phosphatases primarily catalyze phosphate ester hydrolysis, yet some members exhibit promiscuous phosphotransferase activity, enabling substrate phosphorylation using inexpensive inorganic phosphate. This catalytic promiscuity provides attractive biocatalysts for the sustainable production of phosphorylated food-related compounds, but substrate-dependent activity variation makes efficient enzyme identification and engineering challenging. Here, we developed TSCF, a strategy that integrates machine learning with transition-state conformational features to mine enzyme sequences and guide sequence optimization using transition-state models. For inosine 5'-phosphate formation, TSCF showed strong predictive capability, with 70.0% of 20 designed mutants exhibiting 1.2-6.4-fold higher activity than their wild-type enzymes. Incorporation of molecular dynamics simulations further increased screening accuracy to 77.8%. Application to L-ascorbic acid phosphorylation identified effective mutants with 75.0% accuracy and 71.1% conversion, demonstrating the generality of TSCF for improving acid phosphatase-catalyzed phosphotransfer.