Combinatorial Deep Mutational Scanning Uncovers Protein Superbinders and Molecular Determinants of Epistasis

Traditional deep mutational scanning (DMS) encodes every single amino-acid substitution from a wild-type sequence. We hypothesize that Combinatorial DMS (CDMS) libraries, incorporating all mutations in all combinations, can enable the discovery of high-affinity protein (super)binders by capturing epistatic, non-linear amino-acid interactions. Here, we introduce ORCHID, which systematically maps regions of wild-type-independent epistasis across all mutational contexts and trajectories. For benchmarking, we build a high-throughput peptide display assay measuring PIN1WW-domain affinity for a CDMS peptide library containing phosphoserine via amber codon suppression. ORCHID raises prediction accuracy 45% over non-epistatic models. We identify and structurally characterize two epistatic superbinders, SPY-tide and LYR-tide, binding 3–5-fold tighter than current optimal PIN1WW-domain binders through 'fold-and-turn' and 'register-shifted' conformational changes. We also identify two molecular determinants of epistasis, PIN1F25 and PIN1R14, that natively encode non-linear binding and, when mutated, abolish it. Hence natural proteins recognize peptides non-linearly, offering opportunities for improved binder design.