Reconciling Simulated Ensembles of Apomyoglobin with Experimental HDX Data

Abstract

Hydrogen/deuterium exchange (HDX) is a powerful tool to investigate protein conformational dynamics at equilibrium. In the classic model of HDX, each of the amide hydrogens of a protein can either exist in an “open” conformation, where HDX can occur with the exchange rate kint, or in a “closed” conformation, where HDX rate, kex, gets much slower. The extent of protection for a residue i, is characterized by a protection factor, PFi = kex,i/kint,i, which is related to the effective free energy difference between open and closed states, ΔG, through ln(PFi ) = βΔGi. However, it has been challenging to make direct connections between molecular simulations of native-state protein dynamics, both in predicting HDX PF from trajectory data, and in using experimental protection factors as restraints in simulated ensembles. In this study, we propose an expression, ln(PF)= βc<Nc> + βh<Nh> + β0, to predict the protection factors based on the average number of heavy-atom contacts <Nc> and hydrogen bonds <Nh> observed in simulations. The model parameters are trained on ultra-long, all-atom trajectories of proteins with experimentally measured protection factors (BPTI and ubiquitin). Using Bayesian inference to propagate the uncertainties in model parameters, we then use this model to refine conformational ensembles of apomyoglobin at neutral pH. To minimize bias, we simulated apomyoglobin using a range of weak to strong amide exposure restraints. We applied a transition-based reweighting method, TRAM, to construct a Multi-ensemble Markov model from the combined simulation data, and then used our Bayesian Inference of Conformational Populations (BICePs) algorithm to refine a conformational ensemble that best matched the experimental structural data. The resulting ensemble shows a partially-disordered F/H helix, and atomically detailed information that was previously missing from 2D NMR studies.