Reconciling simulated ensembles of apomyoglobin with experimental HDX data using Bayesian inference and multi-ensemble Markov State Models

Hydrogen/deuterium exchange (HDX) is a powerful technique to investigate protein conformational dynamics at amino acid resolution. Because HDX provides a measurement of solvent exposure of backbone hydrogens, ensemble-averaged over potentially slow kinetic processes, it has been challenging to use HDX protection factors to refine structural ensembles obtained from molecular dynamics simulations. This entails two dual challenges: (1) identifying structural observables that best correlate with backbone amide protection from exchange, and (2) restraining these observables in molecular simulations to model ensembles consistent with experimental measurements. Here, we make significant progress on both fronts. First, we describe an improved predictor of HDX protection factors from structural observables in simulated ensembles, parameterized from ultra-long molecular dynamics simulation trajectory data, with a Bayesian inference approach used to retain the full posterior distribution of model parameters.We next present a new method for obtaining simulated ensembles in agreement with experimental HDX protection factors, in which molecular simulations are performed at various temperatures and restraint biases, and used to construct multi-ensemble Markov State Models (MSMs). Finally, the BICePs algorithm (Bayesian Inference of Conformational Populations) is then used with our HDX protection factor predictor to infer which thermodynamic ensemble agrees best with experiment, and estimate populations of each conformational state in the MSM. To illustrate the approach, we use a combination of HDX protection factor restraints and chemical shift restraints to model the conformational ensemble of apomyoglobin at pH 6. The resulting ensemble agrees well with experiment, and gives insight into the all-atom structure of disordered helices F and H in the absence of heme.Graphical TOC Entry