Towards Fast and Accurate Calculation of Protein pKa Values Exploiting Various Degrees of Conformational Flexibility

Abstract

Virtually every biological process is pH dependent. Many fundamental biological phenomena including protein folding, enzyme catalysis, protein-protein interactions and pathological conditions are profoundly influenced by the pH of their environment. Accurate prediction of the pKa values of residues that can adopt variable protonation states would be a significant step towards probing the effects of pH on proteins. Despite the progress in pKa prediction algorithms, developing a method that can incorporate extensive protein conformational flexibility while retaining a relatively small computational resource footprint remains a significant challenge. We developed a fast and accurate method to predict pKas of residues that commonly exhibit variable protonation states in proteins. The algorithm utilizes Metropolis Monte Carlo strategy to incorporate conformational flexibility through extensive amino-acid side-chain sampling and is based on Rosetta score function with an explicit term to account for protonation probabilities of individual amino-acids. We tested the technique by predicting pKas for an assembled dataset comprising 306 residues from 44 proteins resulting in a root-mean-square deviation (RMSD) of 0.81 from experimental values. We analyzed the effects of employing increasing levels of conformational flexibility by (1) sampling the side-chains of neighboring residues and (2) using a generated ensemble of 50 diverse backbone conformers resulting in 77% and 79% predictions < 1 pH unit from experimental values respectively. Using additional degrees of freedom allowed capture of vital hydrogen-bonding and charge-charge interactions, but resulted in structural rearrangements negating pKa shifts in some cases. The method yielded good results when used to predict large pKa shifts in point mutants from Staphylococcal nuclease. Finally, we employed our method to dynamically alter the ionization states of residues during protein-protein docking simulations to seek improvements in accuracy of prediction of complexes.