Balanced regulation of UvsX activity and thermostability based on active learning

Recombinase polymerase amplification (RPA) enables isothermal nucleic acid detection. However, the key enzyme UvsX of RPA exhibits suboptimal efficiency at low-temperature conditions and low template concentration. Here, we employed a directed engineering strategy combining energy-based screening and protein language models (PLMs). Through semi rational design and iterative optimization with EVOLVEpro, we obtained a triple mutant, TIG (S233T/A311I/A341G), which achieved 1.89-fold RPA relative activity of the wild-type (WT) at a template concentration of 1 × 10⁵ copies reaction^-1, and showed higher amplification activity to WT across 37.5-44.3 °C at a template concentration of 1 × 10⁴ copies reaction^-1. Molecular dynamics simulations revealed that TIG enhances residue flexibility and remodels the ATP-binding pocket, thereby improving DNA-binding pathway connectivity and strengthening the coupling between ATP hydrolysis and DNA strand exchange. Per-residue energy decomposition and protein structure network analyses further elucidate how key mutations modulate DNA-binding pathways. Finally, under optimized expression conditions (20 °C, 12 h, 0.1 mM IPTG), the yield reached 206.87 ± 3.56 mg L^-1 of TIG. Collectively, this study provides a strategy for UvsX engineering and establishes a foundation for the development of more robust RPA assays.