Accounting for\nFast vs Slow Exchange in Single Molecule\nFRET Experiments Reveals Hidden Conformational States

Proteins are dynamic systems whose\nstructural preferences\ndetermine\ntheir function. Unfortunately, building atomically detailed models\nof protein structural ensembles remains challenging, limiting our\nunderstanding of the relationships between sequence, structure, and\nfunction. Combining single molecule Förster resonance energy\ntransfer (smFRET) experiments with molecular dynamics simulations\ncould provide experimentally grounded, all-atom models of a protein’s\nstructural ensemble. However, agreement between the two techniques\nis often insufficient to achieve this goal. Here, we explore whether\naccounting for important experimental details like averaging across\nstructures sampled during a given smFRET measurement is responsible\nfor this apparent discrepancy. We present an approach to account for\nthis time-averaging by leveraging the kinetic information available\nfrom Markov state models of a protein’s dynamics. This allows\nus to accurately assess which time scales are averaged during an experiment.\nWe find this approach significantly improves agreement between simulations\nand experiments in proteins with varying degrees of dynamics, including\nthe well-ordered protein T4 lysozyme, the partially disordered protein\napolipoprotein E (ApoE), and a disordered amyloid protein (Aβ40).\nWe find evidence for hidden states that are not apparent in smFRET\nexperiments because of time averaging with other structures, akin\nto states in fast exchange in nuclear magnetic resonance, and evaluate\ndifferent force fields. Finally, we show how remaining discrepancies\nbetween computations and experiments can be used to guide additional\nsimulations and build structural models for states that were previously\nunaccounted for. We expect our approach will enable combining simulations\nand experiments to understand the link between sequence, structure,\nand function in many settings. Understanding protein dynamics is crucial\nfor understanding protein function, yet few methodologies report on\nprotein motion at an atomic level. Combining single molecule Förster\nresonance energy transfer (smFRET) experiments with computer simulations\ncould provide atomistic models of protein ensembles which are grounded\nin experiments, however, there has been limited agreement between\nthe two methods to date. Here, we present an algorithm to recapitulate\nsmFRET experiments from molecular dynamics simulations. This approach\nsignificantly improves agreement between simulations and experiments\nfor proteins across the ordered spectrum. Moreover, with this approach\nwe can confidently create atomic models for states observed during\nsmFRET experiments which were otherwise difficult to model due to\nhigh amounts of flexibility, disorder, or large deviations from crystal-like\nstates.