Deep learning-driven semi-rational design in phenylalanine ammonia-lyase for enhanced catalytic efficiency

Phenylalanine ammonia-lyase (PAL) possesses significant potential in agriculture, industry, and the treatment of various diseases, including cancer. In particular, PAL derived from Anabaena variabilis (AvPAL) has been successfully utilized in clinical settings as an enzyme replacement therapy for phenylketonuria (PKU). Nonetheless, enhancing the catalytic efficiency of enzymes continues to be a formidable task. Herein, a deep learning-guided strategy was employed to identify potential sites in AvPAL that require modification to address current challenges. In conjunction with high-throughput screening and enzymatic assays, 26 out of 33 mutants were validated to exhibit enhanced activity. Notably, the probability of identifying mutants with increased activity at each targeted site was 100 %. Through multiple rounds of combinatorial mutagenesis, the catalytic efficiency (k_cat/K_M) was improved up to 3.4-fold (M222N/N36S) and the activity was enhanced up to 2.4-fold (M222N/I149D) compared to the wild type. Molecular dynamics simulations revealed that the stabilization of the proximate attack conformation is closely associated with the enhancement of catalytic activity. Furthermore, structural superposition, residue interaction network analysis, and dynamic cross-correlation matrices indicate that the most active mutant, M222N/I149D, induces local fluctuations and distal effects in loop 81-94, leading to a more favorable conformation and thus increased activity. Collectively, this study provides a feasible approach for engineering an efficient PAL.