A New Constant pH Method to Simultaneously Predict pH-Induced Conformational Changes and Individual PKA Values in Large Biomolecules

Abstract

We describe a constant pH MD method that integrates a new FEP replica-exchange implementation for titratable residues into the Hamiltonian pH replica-exchange routine. Using the high performance OpenMM engine, during each FEP/REMD stage, a switch of protonation probabilities is attempted between 2 adjacent replicas (for a randomly chosen set of residues with a conventional Metropolis criterion). This FEP scheme enables a gradual evolution of the force field representation of protonation states for each residue independently, leading to a more precise calculation of protonation probabilities. Any net charge discrepancies arising from switches of protonation coupling parameters in systems with explicit solvent are addressed with an adaptive technique to preserve the overall zero charge for all replicas individually after each exchange step. Each replica runs on a single GPU, enabling linear scaling of N replicas on N GPU; this parallelization strategy, combined with advanced sampling, results in a significantly faster time-to-solution than other constant pH methods. Thus, the new method yields predictions of conformational states corresponding to given pH values with considerably reduced computational expense. Notably, the implementation's constant pH and MD modules are independent, rendering the method adaptable to a variety of propagation techniques and advanced sampling methods. The results obtained for small-, medium- and large-scale pH-dependent protein systems will be discussed to illustrate the performance of the current version of the method.