Design of Protein-Protein Interactions with a Novel Ensemble-Based Scoring Algorithm

Abstract

AbstractProtein-protein interactions (PPIs) are vital for cell signaling, protein trafficking and localization, gene expression, and many other biological functions. Rational modification of PPI targets provides a mechanism to understand their function and importance. However, PPI systems often have many more degrees of freedom and flexibility than the small-molecule binding sites typically targeted by protein design algorithms. To handle these challenging design systems, we have built upon the computational protein design algorithm K * [8,19] to develop a new design algorithm to study protein-protein and protein-peptide interactions. We validated our algorithm through the design and experimental testing of novel peptide inhibitors.Previously, K * required that a complete partition function be computed for one member of the designed protein complex. While this requirement is generally obtainable for active-site designs, PPI systems are often much larger, precluding the exact determination of the partition function. We have developed proofs that show that the new K * algorithm combinatorially prunes the protein sequence and conformation space and guarantees that a provably-accurate ε-approximation to the K * score can be computed. These new proofs yield new algorithms to better model large protein systems, which have been integrated into the K * code base. K * computationally searches for sequence mutations that will optimize the affinity of a given protein complex. The algorithm scores a single protein complex sequence by computing Boltzmann-weighted partition functions over structural molecular ensembles and taking a ratio of the partition functions to find provably-accurate ε-approximations to the K * score, which predicts the binding constant. The K * algorithm uses several provable methods to guarantee that it finds the gap-free optimal sequences for the designed protein complex. The algorithm allows for flexible minimization during the conformational search while still maintaining provable guarantees by using the minimization-aware dead-end elimination criterion, minDEE. Further pruning conditions are applied to fully explore the sequence and conformation space.To demonstrate the ability of K * to design protein-peptide interactions, we applied the ensemble-based design algorithm to the CFTR-associated ligand, CAL, which binds to the C-terminus of CFTR, the chloride channel mutated in human patients with cystic fibrosis. K * was retrospectively used to search over a set of peptide ligands that can inhibit the CAL-CFTR interaction, and K * successfully enriched for peptide inhibitors of CAL. We then used K * to prospectively design novel inhibitor peptides. The top-ranked K *-designed peptide inhibitors were experimentally validated in the wet lab and, remarkably, all bound with μM affinity. The top inhibitor bound with seven-fold higher affinity than the best hexamer peptide inhibitor previously available and with 331-fold higher affinity than the CFTR C-terminus. Abbreviations used: PPI, protein-protein interaction; CFTR, Cystic fibrosis transmembrane conductance regulator; CAL, CFTR-associated ligand; DEE, Dead-end elimination; MC, Monte Carlo; CF, cystic fibrosis; NHERF1, Na + /H + Exchanger Regulatory Factor 1; GMEC, global minimum energy conformation; BLU, biochemical light unit; ROC, receiver operating curve; AUC, area under the curve.KeywordsPartition FunctionMonte CarloPeptide InhibitorReceiver Operating Curve CurvePeptide ArrayThese keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.