Guiding Double Electron Electron Resonance Experiments using Information Theory and Molecular Dynamics Simulations

Abstract

Flexible proteins present a unique set of challenges for refinement with either computational or experimental techniques: because these proteins occupy a large conformational space, standard molecular dynamics (MD) simulations can be prohibitively expensive, and high-resolution experimental techniques like x-ray crystallography intentionally capture only small set of low energy states. A new approach is therefore needed to capture the conformational ensembles of flexible proteins. Double electron-electron resonance (DEER) spectroscopy, an experimental technique that measures distance distributions between pairs of spin-labeled residues on a protein, provides quantitative information about less populated but still potentially important conformational states. However, because DEER experiments are costly and time consuming, only a small subset of pairs can be measured, often chosen arbitrarily. If instead of selecting an arbitrary or random subset, we select a subset of pairs such that they provide the maximum amount of information about the conformational ensemble, DEER could become an even more powerful tool for refinement. We have therefore developed 1) a new information-theoretic metric that ranks pairs of residues based on how well they refine a conformational ensemble and 2) a new methodology which identifies a set of highly informative pairs that perform well under this metric. Specifically, we incorporate DEER distance distributions into restrained-ensemble MD simulations, utilize the data from these simulations to identify a set of maximally-informative and minimally-redundant pairs, measure the pairs with DEER, and demonstrate the use of these pairs in refining the conformational ensemble of a flexible protein.