Abstract

Methionine γ-lyase (MGL) is a pyridoxal-5′-phosphate–dependent enzyme with strong potential as an anticancer therapeutic, yet its poor catalytic efficiency, low stability, and high immunogenicity limit clinical application. In this study, molecular dynamics (MD) simulations were employed to identify highly flexible regions of Pseudomonas deceptionensis MGL (PdMGL), revealing the C-terminal loop near the catalytic pocket and the N-terminal region as hotspots of conformational instability. These flexible segments, together with global fitness landscape predictions, were further optimized using the deep-learning-based protein language model ESM-2 to guide rational mutagenesis. Activity screening revealed several beneficial single mutations, including H368A, K4S, Y190F, and N163T, and subsequent iterative recombination produced the quadruple mutant K4S-N163T-Y190F-H368A. This optimized variant exhibited a 3.8-fold increase in activity, over fourfold improvement in catalytic efficiency, enhanced plasma stability, and superior cytotoxicity against colorectal cancer cells, with IC 50 values reduced by more than an order of magnitude compared to wild type. These findings validate the synergy of MD simulations and protein language models in therapeutic enzyme engineering and establish K4S-N163T-Y190F-H368A as a promising candidate for anticancer applications.