Discovery and engineering of a formate dehydrogenase with enhanced bicarbonate reduction activity

The enzymatic conversion of CO₂ into value-added chemicals offers a promising strategy for carbon capture and utilization; however, the inherently low reductive activity of natural formate dehydrogenases (FDHs) limits their practical application. In this study, we employed an integrated approach combining deep learning, virtual saturation mutagenesis, and molecular dynamics simulations to discover and engineer FDH variants with enhanced bicarbonate reduction activity. Starting from ∼ 5000 homologous sequences, high-throughput k_cat prediction via the DLKcat model and molecular docking prioritized 36 candidates. Subsequent phylogenetic analysis and experimental screening identified GbFDH as a superior lead template for bicarbonate reduction. To further enhance its reductive performance, structure-guided virtual saturation mutagenesis and combinatorial mutagenesis were performed, generating the double mutant E142R/Y145K. Compared with the wild type, this mutant exhibited a markedly decreased K_m and a 5.34-fold increase in catalytic efficiency (k_cat/K_m) toward bicarbonate, while its formate oxidation activity was substantially reduced. When integrated into a three-enzyme cascade (FDH, FaldDH, and ADH), the mutant achieved a methanol production of 5.05 mg·L^-1, representing a 2.26-fold improvement over the wild-type system. Molecular dynamics simulations and structural analyses revealed that the E142R/Y145K substitutions reshape the electrostatic landscape of the substrate-recognition region. This remodeling stabilizes NADH binding and promotes a compact, catalytically competent conformation under reductive conditions. Ultimately, this work demonstrates that deep learning-guided targeted engineering can effectively invert the catalytic bias of FDHs, yielding promising biocatalysts for biological carbon capture and C₁ biotransformation.