Loop engineering and activity improvement of TEV protease by a phagemid-based selection system

Loop engineering of enzymes remains challenging due to high flexibility and conformational complexity, posing a bottleneck for deep-learning-based design. Here, we constructed mutant libraries for three loops of TEV protease to assess combining directed evolution with deep learning. Using an M13 phagemid-based selection system, the three libraries were screened, resulting in a Loop 1 variant (HyperTEV60/L1) that significantly enhanced the Michaelis constant (Km) of the HyperTEV60 scaffold, a highly active mutant identified by ProteinMPNN. Structural modeling suggested that a single-residue deletion and substitution in Loop 1 expands the substrate binding pocket, accounting for the improved Km. Although the catalytic efficiency kcat/Km of HyperTEV60/L1 was only marginally higher than HyperTEV60, due to a kcat decrease, our results reveal that the phagemid-based selection system tended to find variants optimizing Km. This study demonstrates that combining deep-learning-based global optimization with localized directed evolution maximizes the probability of discovering distinct, high-performance enzyme variants.