AI-Guided Protein Engineering Unlocks DON-degrading Enzyme Thermal Stability via a Loop-to-β-Sheet Conformational Switch

Abstract

Abstract Deoxynivalenol (DON) is a prevalent mycotoxin that poses a serious threat to human and animal health. Although enzymatic biodegradation offers an effective detoxification strategy, the poor thermal stability of current DON-degrading enzymes severely limits their industrial application. In this study, the key residue associated with thermal stability of the artificial DON-degrading enzyme (ADDE) was identified through the integration of the protein language model SaProt and three-dimensional structural analysis. Site-saturation mutations of proline (Pro) at position 103 to any of the other 19 amino acids significantly enhanced thermal stability to varying degrees while maintaining catalytic activity. The best-performing mutant, P103H, exhibited a half-life of 185.78 ± 5.85 min at 50 ℃, which is 5.22 times that of the wild-type. Further three-dimensional structural simulation and molecular dynamics analysis revealed that the unique five-membered ring structure of proline at position 103 eliminates potential hydrogen bond donors, thereby restricting the formation of local hydrogen bond networks. In contrast, substitution with other amino acids introduces new hydrogen bonds or enhances hydrophobic interactions, which collectively drive the conformational transition of the L94-L105 region from a flexible loop into two antiparallel β-sheets. This structural rearrangement significantly increases the overall rigidity of the protein, ultimately leading to enhanced thermal stability. This study not only provides a reliable computational framework for enzyme protein engineering but also represents the first successful enhancement of the thermal stability of a DON-degrading enzyme using a protein language model, thereby opening a new pathway for the development of industrial-grade biocatalysts.