Ancestral Sequence Reconstruction Synergized with Deep Learning Significantly Enhances the Thermophilicity and Thermostability of the Highly Active Alkaline Protease AprE

Abstract Although alkaline proteases have broad industrial applications, their practical use is hindered by insufficient thermostability and limited strategies to elevate optimal catalytic temperature (Topt) while retaining catalytic activity. To enhance Topt and thermostability of Bacillus clausii AprE without sacrificing catalytic performance, this work combined ancestral sequence reconstruction (ASR) with deep-learning prediction. Ancestral amino acid states were inferred to identify candidate substitution sites for grafting onto the wild-type (WT) backbone, and an ESM-1v-based support vector regression model was built to predict Topt. Systematic screening identified AP291, which exhibited a 4 °C higher Topt and 3.46-fold longer half-life (t1/2) at 60 °C, while maintaining favorable catalytic efficiency (kcat/Km) compared with WT AprE. Molecular dynamics simulations revealed that AP291 exhibits reduced local structural fluctuations and enhanced conformational stability, clarifying the underlying stability-enhancing mechanism. This work presents a synergistic ASR-deep learning strategy for enzyme engineering, and AP291 is promising for industrial biocatalysis.