Deep learning-guided engineering of phospholipase A2 for enhanced catalytic properties and efficient heterologous expression in Bacillus subtilis

Phospholipase A 2 (PLA 2 ) is an important biocatalyst with broad applications in food processing, oil refining, and biotechnology. However, PLA 2 enzymes from natural sources often exhibit insufficient activity and poor thermal stability, which limits their industrial use. This study aims to enhance PLA 2 thermal stability and catalytic activity by combining deep-learning-guided site screening with combinatorial assembly, with the goal of producing catalysts better suited for efficient industrial applications. We applied a deep learning-guided design strategy, specifically utilizing the MutCompute model, to identify mutation hotspots in PLA 2 . Guided by these predictions, we generated single-point and combinatorial mutants and obtained variants with markedly improved properties. The N21E/I84L/Q116M triple mutant showed a 225% increase in specific activity relative to the wild type (WT), enhanced thermal stability (the half-life at 50 °C increased from 26.69 min to 33.85 min), and a broadened active pH range (WT retained high activity only between pH 7.0 and 7.5, N21E/I84L/Q116M retained ≥71.5% of its maximal activity over pH 6.0–9.0). To further facilitate industrial production, we optimized the expression of this superior mutant in Bacillus subtilis . The optimized promoter P 43 -P aprE increased expression by 69.6%, and the optimal signal peptide SP pelB enhanced secretion by 18.4%. Finally, a whole-cell biocatalyst was developed by surface-displaying the engineered PLA 2 . In the biotransformation of soybean lecithin, this mutant whole-cell biocatalyst yielded 48.8% more product than its WT counterpart. In conclusion, this study presents a comprehensive and efficient molecular engineering strategy for PLA 2 , providing a highly active, stable, and easily expressed biocatalyst well-suited for industrial application.