Energetic frustration dictates cognate specificity of two-component signaling system

Proteins perform their biological functions through specific recognition of their interaction partners, but approaches for pinpointing specificity-determining sites and elucidating how they govern binding specificity remain limited. Here, using two-component signaling systems (TCSs) as a model, we combine local energetic frustration analysis with ProteinMPNN-based sequence generation to identify two specificity-determining sites. We find that cognate specificity in TCSs is intrinsically marginal and can be reconfigured by single mutations at these sites. The order in which mutations arise shapes evolutionary intermediates and tunes specificity through epistatic effects. We further show that mutational degeneracy is widespread and contributes to maintaining cognate specificity while limiting crosstalk. The critical roles of these two sites in tuning specificity are supported by AF3-predicted structural confidence and bacterial two-hybrid assays. This work establishes a generalizable strategy for identifying specificity-determining sites and provides mechanistic insight into how evolution shapes protein binding specificity. Protein interactions are crucial for biological functions, yet pinpointing specificity-determining sites remains challenging. Here, the authors integrate local energetic frustration analysis with deep learning-based sequence generation method ProteinMPNN to identify these sites using two-component signaling systems, revealing how mutations regulate specificity and offering a strategy with broad implications for understanding protein evolution and interaction specificity.