Origins and breadth of pairwise epistasis in an α-helix of β-lactamase TEM-1

The effect of mutations in a protein may depend on the presence of others-a phenomenon known as epistasis. Epistasis plays a key role in evolution and complicates predictions of mutational effects, as effects can be context-dependent. Yet, despite its importance, the mechanistic basis of epistasis remains poorly understood. To better characterize epistasis, we focused on an 11-residue α-helix in TEM-1 β-lactamase and constructed a comprehensive library of over 14,000 double mutants. Fitness and minimum inhibitory concentration, two contrasted measure of protein efficiency, reveal consistent widespread epistasis. A non-linear two-state protein stability model in which destabilizing, neutral, or stabilizing mutations contribute additively to the stability phenotype, largely explain the data. Most epistatic effects are consequently predictable from single-mutation effects. However, systematic deviations from the model occur when both mutated residues directly interact in the 3D structure-a fold conserved across distant TEM-1 homologs. We therefore investigated the predictive power of statistical models trained on distant homologous sequences and found that they could partially recover the observed epistatic interactions. Our results, built on a short structural element of a protein, shed light on multiple determinants of the epistatic landscape that have shaped the evolutionary trajectory of β-lactamase proteins over long timescales.