All atom molecular dynamics simulations enable ensemble refinement of flexible and mismodelled cryo-EM derived RNA structures

The advent of single particle cryogenic electron microscopy (cryo-EM) allowed to gather structural information on large biological macromolecules at near atomic-level resolution and thus to advance their functional understanding. Yet, building RNA structures from cryo-EM maps remains challenging owing to their inherent plasticity. Standard cryo-EM refinement tools assume that all the single-particle images used for an individual three-dimensional reconstruction can be associated to a single structure, but this may lead to mis-modeled structural motifs in case of highly flexible molecules such as RNAs. We target this problem by integrating molecular dynamics simulations and experimental cryo-EM density maps to improve the model quality and, thus, to obtain better predictions about the structural and functional properties of the system under investigation. Specifically, we apply metainference, a Bayesian-inference-based method, to characterize the group II intron ribozyme, a complex and biologically relevant RNA macromolecule. By reconstructing an ensemble of structures that best matches the experimental cryo-EM density, we identify modeling inaccuracies of flexible helical regions linked to their artificial representation into a single structural model. An analysis of all RNA-containing structures deposited in the PDB databases reveals that this problem affects most cryo-EM structures in the 2.5–4 Å resolution range. Hence, RNA-containing structures determined by cryo-EM must be handled with care, and our approach may be broadly applicable to many other biologically-relevant RNA systems.