Abstract
ABSTRACT
Functional RNA structures are typically far more dynamic than most proteins solved by cryo-electron microscopy (cryo-EM). Flexible junctions and weak tertiary contacts produce broad conformational ensembles, and cryo-EM particle classification analyses discard unaligned particles, which inherently contain the information needed to quantify such broad ensembles. Here, we infer compositional and conformational distributions from unclassified particle images by Bayesian ensemble reweighting. We tested our approach on the P4-P6 domain of the
Tetrahymena thermophila
group I intron, a classic model system of Mg²⁺-dependent RNA folding. While stable monomeric and dimeric states yielded high-resolution reconstructions, analyses of abundant, flexible conformations required Bayesian ensemble inference. Our approach quantitatively captured the Mg²⁺-dependent folding transition, indicating that cryo-EM can report the abundance of coexisting RNA states in agreement with bulk solution measurements. These results showcase how quantitative studies of RNA folding can be achieved by combining cryo-EM measurements with statistical ensemble-inference tools.
SUMMARY
RNAs are remarkably versatile macromolecules that fold into dynamic three-dimensional (3D) structures to perform critical functions in cells and viruses, making them increasingly attractive target for therapeutics and as tools for biotechnology. Yet our understanding of RNA structure–function relationships remains limited, in part because structural biology tools are poorly suited to probing the dynamic ensembles that RNAs occupy. The raw 2D images collected in single-particle cryo-EM, however, inherently contain the information needed to visualize and quantify such ensembles. Here, we use the P4-P6 domain of the
Tetrahymena thermophila
group I intron, an archetypal model of Mg²⁺-dependent RNA folding, to leverage this information. While stable monomeric and dimeric states yielded high-resolution structures, flexible conformations escaped conventional 3D reconstructions—even though they dominate at physiological Mg²⁺ concentrations. To overcome this limitation, we integrated single-particle cryo-EM with a Bayesian reweighting framework that quantifies RNA ensemble distributions directly from raw 2D images. By statistically reweighting computational ensembles against experimental images via probabilistic modeling and deep learning, our approach quantitatively captures Mg²⁺-dependent population shifts between extended, compact, monomeric, and dimeric species, in agreement with fundamental principles of RNA folding and bulk solution measurements. Our framework supports the sampling and visualization of conformations within the ensemble, allows the quantification of errors, and demonstrates that the learned continuous ensemble fits the data significantly better than any set of discrete structures. These results showcase how quantitative RNA folding can be achieved by combining high-resolution structural biology measurements with statistical tools that treat RNAs as conformational ensembles.