Introduction

Biological function emerges from the ability of well-defined molecular structures to explore dynamic conformational landscapes. Function arises from transitions between states, through coordinated motions that span multiple spatial and temporal scales (Henzler-Wildman & Kern, 2007). The concept of an energy landscape, in which macromolecules continuously interconvert between alternative conformations, provides a unifying framework to describe this dynamic behavior (Frauenfelder et al., 1991). But structural biology has so far primarily focused on capturing discrete conformational states of macromolecules.

RNA molecules exemplify this principle. Beyond adopting defined secondary and tertiary folds, many RNAs populate ensembles of conformations with comparable free energy, enabling them to act as molecular switches, scaffolds, catalysts, and regulators (Ganser et al., 2019; Vicens & Kieft, 2022). This intrinsic flexibility underlies key biological processes, from ribozyme catalysis and riboswitch-mediated gene regulation to viral replication and RNA-driven assembly. However, despite its fundamental importance, RNA dynamics has remained challenging to characterize at high resolution, due to both experimental and conceptual limitations.

Recent advances in biophysical techniques are now changing this landscape. Single-molecule fluorescence resonance energy transfer (smFRET) can detect hidden RNA states and enable detailed screening of RNA conformations through increased automation, supporting RNA structure prediction (Groves et al., 2023; Hartmann et al., 2023; Weber et al., 2026). Molecular dynamics simulations (MD) and nuclear magnetic resonance (NMR) spectroscopy, especially when coupled to X-ray and neutron scattering (SAXS/SANS), can quantitatively describe the dynamics of these low-populated RNA states essential for function (Arnold et al., 2025; Bussi et al., 2024). Single-molecule chemical probing further reveals the plasticity of RNA structure in response to cellular and environmental stimuli (Arney et al., 2024; Spitale & Incarnato, 2023).

Integrating with and complementing these techniques, cryogenic electron microscopy (cryo-EM) is playing a transformative role, because it can now visualize discrete RNA conformational states and continuous structural variability at near atomic resolution (Khusainov et al., 2024). In other words, cryo-EM now visualizes in 3D RNA states that smFRET, SAXS/SANS and NMR can only infer kinetically and that chemical probing experiments can deconvolute but only at the level of secondary structure maps.

In contrast to traditional approaches that favor well-ordered, homogeneous samples, modern cryo-EM workflows – combined with innovations in sample design, vitrification, and computational analysis – are beginning to make the structural characterization of flexible and dynamic RNAs tractable (Bonilla & Jang, 2024; Bonilla & Kieft, 2022; Ma et al., 2022; Zhang et al., 2024). The opportunity to integrate cryo-EM now with biochemical, biophysical, computational and structural biology techniques marks a conceptual shift for RNA structural biology: from determining static structures toward reconstructing conformational ensembles and, ultimately, energy landscapes. This transition has been particularly rapid in recent years. The number of protein-free RNA structures determined by cryo-EM has increased sharply, driven by methodological advances that extend the technique to increasingly flexible and heterogeneous targets. At the same time, new computational tools enable the disentanglement of conformational heterogeneity and the de novo prediction of RNA 3D structures and conformational ensembles, revealing not only distinct structural states but also the motions that interconnect them. Together, these approaches provide a direct structural view of RNA dynamics that was previously inaccessible.

Here, we synthesize recent progress at this emerging interface between cryo-EM and RNA biology. We argue that cryo-EM is evolving into a central tool for decoding RNA structural-functional dynamics, enabling the visualization of both discrete states and continuous conformational trajectories. Through eight representative case studies spanning ribozymes, riboswitches, viral RNAs, and synthetic RNA assemblies, we illustrate how cryo-EM has revealed new mechanistic principles of RNA motion, including domain rearrangements, hinge-mediated flexibility, ligand-induced switching, and cooperative assembly. Building on these examples, we identify key experimental and computational bottlenecks – particularly in construct design, sample preparation, and data processing – and highlight recent solutions that have enabled the visualization of dynamic RNA behavior.

Finally, we outline future directions toward a more complete description of RNA dynamics, where cryo-EM data are integrated with complementary approaches such as molecular simulations, chemical probing, and time-resolved methods. These advances will not only deepen our understanding of RNA biology, but also open new opportunities for RNA-targeted therapeutics and the design of dynamic RNA-based nanodevices. By connecting structural snapshots into coherent dynamic models, cryo-EM is poised to move the field beyond static representations and toward a quantitative understanding of RNA energy landscapes.

Structural studies on RNA

In 1871, Swiss chemist Friedrich Miescher described DNA (isolated from pus cells and named “nuclein”) in his seminal work Über Die Chemische Zusammensetzung Der Eiterzellen (Miescher, 1871; Thess et al., 2021). Since then, it took over a century to discover messenger RNA (mRNA) (Astrachan & Volkin, 1958; Brenner et al., 1961; Gros et al., 1961; Hershey, 1953; Jacob & Monod, 1961; Monod et al., 1952; Pardee, 1954; Ycas & Vincent, 1960). Now, many RNA classes have been identified including tRNA, rRNA, microRNAs (miRNAs), small interfering RNAs (siRNAs), piwi-interacting RNA (piRNAs), small nuclear RNAs (snRNA), snoRNA (small nucleolar RNA), circular RNAs (circRNA) and long non-coding RNA (lncRNA) (Chen & Kim, 2024). Indeed, RNA comprises a major part of total cell biomass, up to 20% of a cell’s dry weight, and the number of non-coding RNAs is greater than the number of proteins (Chen & Kim, 2024; Majumder et al., 2016). Importantly, the newly discovered classes of RNAs play fundamental roles in physiological gene regulation and in pathologies such as cancer, congenital syndromes and neurodegenerative disorders (Nemeth et al., 2024).

Despite the biological significance of RNA, structural studies of RNA have historically lagged behind structural studies on proteins. While the crystal structure of an RNA double helix was determined around the same time as that of the DNA double helix and of the first protein molecules (Rich & Watson, 1954), it took another 20 years to obtain the crystal structure of a full length, biologically relevant RNA, namely yeast phenylalanine tRNA (tRNA^Phe) (Robertus et al., 1974). But the functional importance of RNA structures started to be fully appreciated only after the discovery of catalytic RNAs in the ‘80s and the consequent determination of the crystal structure of the catalytic core of the Tetrahymena group I intron - which revealed an RNA-only active site (Altman et al., 1989; Cate et al., 1996; Kruger et al., 1982). In the late ‘90s, X-ray crystal structures of the 30S and 50S ribosomal subunits emphasized the potential of RNA to form a structural scaffold for proteins. At the same time, crystal structures of riboswitches revealed the ability of RNA to recognize small molecules selectively and specifically (Kavita & Breaker, 2023).

While the number of RNA structures is still now only representing less than 1 % of the number protein structures deposited in the PDB (as of Feb 5^th 2026), recent advances in cryo-EM have enabled structure determination of monomeric, multimeric, synthetic, and chimeric RNAs. Since 2024, 126 new structures of RNAs have been determined by cryo-EM, more than doubling the number of available RNA cryo-EM structures. These studies have led to fundamental insights into the reaction mechanism of various ribozymes, such as self-splicing group I and group II introns, into gene regulation mechanisms of various riboswitches, into the potential of synthetic RNA-based nanotechnology, and into the mode of function of regulatory viral RNAs (vRNAs). Below, we briefly review the different classes of RNAs for which cryo-EM structures have been determined so far, and for each class we highlight a collection of significant recent case studies that specifically characterized RNA structure-functional dynamics.

Ribozymes

RNA enzymes catalyze two of the most fundamental reactions in biology: protein synthesis and pre-mRNA splicing. These reactions are catalyzed by RNA in the context of large ribonucleoprotein (RNP) assemblies – the ribosome and the spliceosome. Complementing decades of biochemical, single-molecule, and spectroscopic studies, recent advances in cryo-electron microscopy have provided unprecedented insight into the dynamic behavior of these complexes, revealing conformational transitions that underlie substrate recognition, catalysis, and product release. For the ribosome, time-resolved cryo-EM and heterogeneous reconstructions – also in situ – have begun to capture transient functional states along the translation cycle (Garg et al., 2025; Rickgauer et al., 2024) and during early assembly (Qin et al., 2023). Similarly, cryo-EM studies of the spliceosome coupled to computational simulations have uncovered a series of compositional and conformational rearrangements that accompany assembly and catalysis (Beusch & Madhani, 2024; Pokorna et al., 2025).

But recent work has begun to resolve at near-atomic resolution also the folding pathways and catalytic mechanisms of the evolutionary ancestors of the spliceosome, namely self-splicing introns, which function as RNA-only systems. Self-splicing introns are ribozymes present in certain pre-mRNA transcripts of bacteria, archaea, and plants and fungal organelles. They catalyze the same two successive transesterification reactions as the spliceosome, to first cleave the 5’-exon and then ligate the 5’ and 3’ exons to generate a mature mRNA (Cech, 1990; Michel & Ferat, 1995). Among self-splicing introns, group I introns use a guanosine cofactor and form a linear intron product after excision, whereas group II introns do not need cofactors and form a lariat product, i.e. a structure containing 2’-5’ phosphodiester linkage at the so-called branchpoint adenosine (Cech, 1990; Xu et al., 2023). To perform splicing, both intron classes form intricate tertiary structures, wherein the active site coordinates mono- and divalent ions essential for splicing (Cate et al., 1996; Marcia & Pyle, 2012; Toor et al., 2008). For the last 40 years, both intron classes have been used as prototypical model systems for structural and biochemical studies on RNA.

Case study 1: Dynamics of the Tetrahymena group I intron

Dynamics of the Tetrahymena group I self-splicing intron regulate its co-transcriptionally folding and mechanism of catalysis (Li et al., 2023; Luo et al., 2023; Zhang et al., 2023). 16 recent cryo-EM structures, determined in the 2.7-4.1 Å resolution range revealed that, in the first step of splicing, the so-called P1 helix formed from pairing of the 5’-exon with the intron 5’-terminus adopts three distinct states, named relaxed, intermediate and docked states, respectively (Figure 1). Quantitatively, P1 moves by 66° from the relaxed to the intermediate state, and then by 50° from the intermediate to the docked state (Luo et al., 2023) (Figure 1). This dynamic movement of P1 effectively decreases the distance between the 5’-exon and the binding site of the exogenous guanosine cofactor from 62 to 4 Å (Figure 1) The P1 helix is held closer to the active site via interactions with J4/5 and J7/8. When the exogenous guanosine binds to the active site the 5’-exon is cleaved, which is immediately followed by the movement of the 3’-exon closer to the 5’-intron and leads to the formation of so-called P10 helix and of a previously unknown pseudoknot interaction formed by the 5’-exon, 5’-intron and 3’-exon. Once the exogenous guanosine is replaced by the internal G414 nucleobase, the two exons are ligated through the second transesterification reaction.

Figure 1.

Pronounced dynamics also occurs during the group I intron folding process, as captured by another set of four recent cryo-EM structures determined at around 3.5-4.0 Å resolution (Bonilla et al., 2022; S. Li et al., 2022). These structures captured the so-called group I intron misfolded (M) state folding intermediate identifying discrete conformations and enabling a previously impossible comparison with the fully folded, native (N) state of this ribozyme. Such comparison shows that, in the transition from the M- to the N-state, the P7 helix rotates by 90° and moves by 18.9 Å to dock on to the P9.1/P9.1a internal loop, the P3 helix completely refolds, and the P8 helix rotates by 360° around its helical axis to twist and correctly position the J7/3 and J8/7 junctions (Figure 1) (Bonilla et al., 2022; S. Li et al., 2022).

Case study 2: Dynamics of group II intron folding intermediates

Analogously to group I introns, structural dynamics in group II self-splicing introns also regulate folding and catalysis (Jadhav et al., 2025; Maiorca et al., 2025). 16 recent cryo-EM structures at 2.6-7.6 Å resolution showed that during its sequential assembly process, the intron 5’- terminal domain (D1), which is the biggest domain, folds first and undergoes specific conformational changes to dock to the 3’-terminal domains (D2-D6) and form the fully folded catalytic state (Jadhav et al., 2025). A helical subdomain in D1, called D1c, rotates around a highly conserved internal loop that acts as a hinge to displace the tip of the helix by 51 Å (Figure 2) The detailed conformational changes of the hinge residues were resolved at the nucleotide level revealing the exact gating mechanism by which D1 controls intron folding and catalysis, acting as a “nutcracker” (Jadhav et al., 2025).

Figure 2.

Importantly, the structures also enabled to see active site dynamics occurring during the transition from the first to the second step of splicing, which could otherwise so far only be hypothesized on the basis of crystal structures and mutagenesis experiments (Maiorca et al., 2025).

Regulatory RNAs

Beyond catalyzing fundamental biochemical reactions, RNA also plays key roles in the regulation of gene expression at both the transcriptional and translational levels. At the interface of RNA regulation and protein complexes, the 7SK ribonucleoprotein (7SK RNP) controls transcriptional elongation by modulating P-TEFb activity, and recent cryo-EM studies have begun to elucidate its structural organization and conformational variability, highlighting the role of RNA-mediated scaffolding in dynamic regulatory assemblies (Yang et al., 2022). Furthermore, a prominent, recently discovered class of regulatory transcripts consists of long non-coding RNAs (lncRNAs), which modulate chromatin organization, transcription, and epigenetic states. Although high-resolution structures of lncRNAs are still unavailable due to their size and conformational heterogeneity, biophysical approaches such as atomic force microscopy (AFM) and SAXS, applied for instance to lncRNA MEG3, Braveheart, and XIST, have revealed highly dynamic and modular architectures, suggesting that conformational plasticity is central to their regulatory function (Aguilar et al., 2022; Kim et al., 2020; Uroda et al., 2019). In addition, a growing number of structured noncoding RNAs of still emerging function are being characterized at high resolution. For example, recent cryo-EM analysis of the bacterial RaiA noncoding RNA, which is present in 2,700 bacteria and is as abundant as rRNA and tRNA, has revealed a conserved architecture and suggested mechanisms for its interaction with the translational machinery (Badepally et al., 2024; Haack et al., 2025; He et al., 2026; Kretsch, Wu, et al., 2025).

But among RNA-only regulatory systems, riboswitches remain the best characterized. These non-coding and non-catalytic RNA elements, typically located in the 5’ untranslated regions of mRNAs, regulate gene expression by undergoing ligand-induced conformational changes that control transcription or translation (Garst et al., 2011). Using cryo-EM, various ligand-bound, ligand-free, active and inactive conformations of different riboswitches have been determined so far. The first riboswitch cryo-EM structure to be determined was that of the SAM-IV riboswitch bound to S-adenosyl-methionine (SAM) (Zhang et al., 2019). This structure was determined at 3.7 Å resolution, showed SAM bound to the riboswitch core and revealed that SAM-IV forms peripheral tertiary interactions different to other classes of SAM riboswitches e.g SAM-I and SAM-I/IV. Second, the cryo-EM structures of the glycine riboswitch from V. cholerae and F. nucleatum were determined in the resolution range between 4.8 to 10 Å (Kappel et al., 2020). These structures showed that the glycine riboswitch consists of three helical stems arranged around a three-way junction, in agreement with previously published crystal structure (Butler et al., 2011; Huang et al., 2010). Third, two cryo-EM structures of T- box riboswitches from B. subtilis and M. smegmatis were determined at 4.9 and 6.3 Å resolution, respectively (Jia et al., 2023; Li et al., 2019). The B. subtilis T-box riboswitch displays a U-shaped form capable of clamping and recognizing tRNA (Li et al., 2019). Instead, the M. smegmatis T-box riboswitch binds tRNA via its 5’-terminus and anticodon loop region only (Jia et al., 2023). Fourth, the cryo-EM structure of a trimeric form of flavin mononucleotide (FMN) riboswitch was determined at 4.9 Å resolution (Liu et al., 2022). Because this riboswitch is particularly small (35 kDa), it was induced to oligomerize to increase its size and facilitate cryo-EM structure determination. Finally, the cryo-EM structure of the cobalamin riboswitch (Vitreschak et al., 2003) – which senses vitamin B12 and thus regulates expression of genes involved in its biosynthesis – showed the structure-functional dynamics of this target as highlighted below (Ding, Deme, et al., 2023).

Case study 3: Dynamics of the cobalamin riboswitch

Structural dynamics regulate how the cobalamin riboswitch recognizes its cognate ligand (Ding, Lee, et al., 2023). This riboswitch had been previously imaged at lower resolution by AFM, visualizing the so-called P-, candy-, compact- and Y-shaped monomeric conformers formed in the absence of cobalamin, and dimeric conformers formed in the presence of cobalamin (Ding, Lee, et al., 2023) (Figure 3). Now, 5 recent cryo-EM structures at 2.9-5.3 Å resolution visualized at higher resolution the dimeric forms of the riboswitch, which are formed through intermolecular kissing loop interactions mediated by the P5 and P13 helices (Ding, Deme, et al., 2023). Dimers 1 and 2 display a slight orientation difference with a 4.7 Å RMSD difference, accompanied by a rigid body rotation of 5.6° of chain B with respect to chain A (Figure 3) (Ding, Deme, et al., 2023; Ding, Lee, et al., 2023). In dimers 3 and 4, only chain A is structured, whereas chain B is flexible. In dimer 3, only the density of P1 and P13 is visible, whereas in dimer 4, in addition to P1 and P13, the majority of P2 and the distal half of the P6 are structured (Ding, Deme, et al., 2023; Ding, Lee, et al., 2023). Comparison of apo- and holo-cobalamin riboswitch further showed the flexibility of the P6 domain (Figure 3). P6 movement is characterized by the increase in distance between P5 and P3, which sandwiches P6, from 11.6 Å (in the apo state) to 26.8 Å (in dimer 1 in the holo state) (Figure 3) (Ding, Deme, et al., 2023; Ding, Lee, et al., 2023). Only when the ligand is bound, P6 docks to the ligand-binding site and interacts with the receptor binding domain and with cobalamin (Ding, Deme, et al., 2023; Ding, Lee, et al., 2023).

Figure 3.

Viral RNAs

Viral RNA genomes harbor many regulatory and structured RNA elements that are essential for viral infection (Nalewaj & Szabat, 2022). Structural studies have started to reveal the heterogeneous organization of packaged viral RNA genome (Cai et al., 2024) and the dynamics of viral proteins during packaging of the RNA genome in the capsid (Xia et al., 2024). Meanwhile, the understanding of structure and dynamics of viral RNA-alone has remained limited. In this respect, cryo-EM was used to determine structures of viral RNA regulatory elements alone or in complex with host proteins. For example, 3.2 Å resolution cryo-EM structure of the ribosome and internal ribosome entry site (IRES) complex revealed how the viral RNA motif host ribosomal proteins and tRNAs for viral protein translation (Bhattacharjee et al., 2026). The cryo-EM structure of HIV-1 Rev protein, Rev Response Element (RRE) and Crm1 complex, which were previously only been studied in isolation by SAXS and smFRET (Fang et al., 2013), explained how HIV-1 nuclear export is mediated in the host cell (Smith et al., 2024).

Currently, there are five examples of cryo-EM studies of viral protein-free RNA elements. First, the 9.0 Å resolution cryo-EM structure of 30 kDa HIV-1 Dimer Initiation Site (DIS) RNA displayed a right-handed RNA helix with major groove features (Zhang et al., 2018). The precise base-pairing pattern of the DIS was determined through integration of cryo-EM structure with the NMR experiments (Zhang et al., 2018). Second, the cryo-EM structure of the frame-shifting stimulatory element (FSE) from SARS-CoV-2 was determined at 6.9 Å resolution and displayed an intricate three-stem pseudoknot architecture, revealing how the RNA 5’-terminus is threaded through the ring of the pseudoknot to cause ribosome slippage and thus the frame-shift (Zhang et al., 2020). Interestingly, this structure represents the lowest molecular weight RNA cryo-EM structure in PDB (88 nucleotides, ~28 kDa). Third, the cryo-EM structure of the FSE element from Rous sarcoma virus (RSV) was determined at 3.3-4.0 Å resolution and revealed both dimeric and monomeric forms for this target (Jones & Ferre- D’Amare, 2025). Finally, the cryo-EM structures of so-called tRNA-like motifs (TLS) and of stem-loop 5 elements (SL5), i.e. motifs that are crucial to regulate translation, revealed the overall topology and peculiar structural-functional dynamics of these targets, as highlighted below.

Case study 4: Dynamics of viral tRNA-like structures (TLS)

Structural dynamics of viral tRNA-like structures (TLS) regulate how this motif bind to its cognate partner, tyrosyl-tRNA synthetase (TyrRS) (Bonilla et al., 2021). Two cryo-EM structures at 4.3 Å resolution showed that the brome mosaic virus (BMV) TLS comprises 3 helical domains, which are coaxially stacked. Domain 1 comprises 3 helices (B1, B2, and C). Domain 2 comprises helix A, which is functionally important as it acts as an acceptor stem for tyrosylation and the replication initiation site, and helix D. Domain 3 comprises helix B3, which is analogous to the tRNA^Tyr anticodon, and helix E. Domain 3 displayed lower local resolution, suggesting that it is flexible (Bonilla et al., 2021). 3D variability analysis (3DVA) revealed multiple conformations for helices B3 and E revealing how these helices are formed and deformed (Figure 4) (Bonilla et al., 2021). Remarkably, domain 3 rotates by ~90° between its TyrRS-bound and unbound states (Figure 4) (Bonilla et al., 2021).

Figure 4.

Case study 5: Dynamics of the stem loop 5 (SL5) motif of the genomic RNA of Coronaviridae

Analogously, structural dynamics regulate how the SL5 motif modulates translation. First, one cryo-EM structure at 4.7 Å resolution of the SL5 motif from SARS-CoV-2 revealed the T- shaped arrangement of this viral motif (Kretsch et al., 2024). Stem loops SL5a and SL5b coaxially stack in a perpendicular direction to the main SL5 stem forming a four-way junction, while the smaller SL5c stem protrudes out from the base of this four-way junction (Manfredonia et al., 2020). The SL5 element also contains two conserved hexaloops, 5’-UUYYGU-3’ sequences, which are important for protein-RNA and RNA-RNA interaction (Chen et al., 2021). The cryo-EM structures shows that these hexaloop sequences are localized in the terminal loops of SL5a and SL5b. Cryo-EM structures of homologous SL5 elements from other viruses, e.g., SARS-CoV-1, MERS, BtCov-HKU5, and HCoV-229E, displayed a similar topological arrangement of constituent stem loops about the four-way junction and SL5 from all these viruses displayed a single homogenous conformation (Kretsch et al., 2024). However, the cryo-EM structures of the SL5 element from MERS and BtCoV-HKU5 viruses showed three and four distinct conformations, respectively (Kretsch et al., 2024). In these viruses, SL5a is bent to different extents, and therefore it is not co-axially stacked with SL5b. The different bent conformations of SL5a arises possibly from the swing-like motion around a hinge, formed by nucleotides in the internal loop of SL5a. The measurement of inter-helical angles between SL5a/b and SL5/c uncovered the angular motion of 81° to 84° and 84° to 88° for MERS and BtCoV-HKU5 SL5, respectively (Figure 4). Such hinge motion, in result, increases the intrahexaloop distance from 74 to 85 Å and 71 to 85 Å for MERS and BtCoV-HKU SL5, respectively (Figure 4) (Kretsch et al., 2024).

Synthetic RNAs

Besides naturally occurring RNAs like the ribozymes, riboswitches, and viral RNA motifs described above, synthetic RNAs are also being generated and characterized thanks to computational design or in vitro evolution experiments, for applications in nanotechnology. For instance, the ATP-TTR-3 aptamer was designed computationally to specifically bind AMP (Yesselman et al., 2019). Its structure was the first cryo-EM structure determined for a synthetic RNA (Kappel et al., 2020). The structure was solved at 10 Å resolution for AMP-bound and unbound forms of the aptamer and displayed a clothespin-like scaffold. Instead, in vitro evolution experiments lead to the emergence of an RNA polymerase ribozyme, which uses trinucleotide triphosphate (pppNNN) and linear or circular RNA templates for RNA replication (Attwater et al., 2018). Its cryo-EM structure revealed functionally important dynamics of this synthetic ribozyme, as highlighted below.

Case study 6: Dynamics of RNA polymerase ribozyme

Specifically, the cryo-EM structure at 5.0 Å resolution of an in vitro selected RNA polymerase ribozyme revealed the overall organization of this target, showing that it resembles an upturned left hand (McRae et al., 2024). The ribozyme consists of two subunits, 5TU and t1. The 5TU subunit, which corresponds to the thumbs of the left-hand-like structure, consists of a catalytic domain (helices P3-P7) and of peripheral domains (helices P1 and P8-P10). The t1 subunit, which corresponds to the fingers of the left-hand-like structure, consists of three helices (P1-P3). Both subunits are held together by kissing loop interactions between P1 (in 5TU) and junction J1/2 (in t1), and between P7 (in 5TU) and P3 (in t1) (McRae et al., 2024). The cryo-EM structure showed that, while these kissing loops are rigid, the two subunits, 5TU and t1, are internally flexible. 3DVA analysis showed that P1 (in t1) and P10 (in 5TU) move towards the active site corresponding to helices P3-P7 (in 5TU) by about 30-35° and 20°, respectively (Figure 5). Fitness landscape analysis revealed that P10 is highly sensitive to mutations around its terminal loop and its junctions with neighboring helices, suggesting that these nucleotides could be critical for its motion (McRae et al., 2024). Similarly, the motion of P1 is regulated by hinge residues located close to an A-anchor motif (McRae et al., 2024). Notably, the conformational changes of P1 and P10 are essential to establish functionally important interactions with the substrate, which is an RNA primer-template duplex, and specifically with its minor groove (McRae et al., 2024). Supporting this central role of P10 and its dynamics, in vitro enzymatic experiments showed that RNA polymerase ribozyme variants lacking P10 display significantly reduced fidelity (McRae et al., 2024).

Figure 5.

Multimeric RNAs

Both natural and synthetic RNAs have recently been shown - largely thanks to cryo-EM – to be able to form multimeric complexes.

Among biologically occurring multimeric RNAs, the cryo-EM structure of Ornate Large Extremophilic (OLE) RNA was first determined at 2.6-3.1 Å resolution (Kretsch, Wu, et al., 2025; Puerta-Fernandez et al., 2006; Wang, Xie, et al., 2025a). OLE is a bacterial RNA involved in metal ion homeostasis and cellular adaptations, but its precise molecular mechanism is unclear (Breaker et al., 2023; Wallace et al., 2012). The cryo-EM studies showed OLE’s homodimeric architecture and how this RNA exposes to the surface binding sites for its protein partners OapA and OapC. Second, the cryo-EM structure of another bacterial RNA, which may play a role in plasmid replication and is thus called Area Required for Replication in a Plasmid Of Fusobacterium (ARRPOF), was determined at 4.0 Å resolution (Wang, Xie, et al., 2025b). ARRPOF forms a homodimer complex, too, and its dimerization is mediated by kissing loops and intermolecular palindromic interactions. Third, the cryo-EM structure of Giant, Ornate, Lake- and Lactobacillales-Derived (GOLLD) RNA was solved at 2.5-6.1 A resolution (Kretsch, Wu, et al., 2025; Wang, Xie, Zhang, et al., 2025; Zhang et al., 2025). GOLLD function is still completely unknown, but its structure revealed a fascinating multimeric nanocage, formed by 14 RNA chains arranged with D7 quaternary symmetry and held together by kissing loops. Remarkably, the 380-Å-diameter cavity within the GOLLD nanocage is sufficiently big to encompass an entire bacterial ribosome. Fourth, the cryo-EM structure of Rumen-Originating, Ornate, Large (ROOL) RNA, an RNA present in bacterial prophages and phages, was determined at 1.9-3.1 Å resolution (Cousin et al., 2017; Kretsch, Wu, et al., 2025; Wang, Xie, et al., 2025a; Zhang et al., 2025). These structures showed that ROOL forms an octameric nanocage, wherein any one RNA molecule forms 8 bridges with neighboring molecules. These bridges are tertiary A-minor or kissing loop interactions that hold together the ROOL nanocage to form a hollow core large enough to encompass a ribosome, like GOLLD. These multimeric RNA structures display the potential of RNA to assemble into large scaffold structures with important implications for drug delivery.

Because of the potential importance of RNA cages in medicine and nanotechnology, synthetic multimeric RNAs, called RNA origami, have also been developed and characterized by cryo-EM, revealing unique structural dynamics, as highlighted below.

Case study 7: Dynamics of RNA origami nanostructures

Specifically, three sets of cryo-EM structures visualized interesting dynamics of synthetic RNA origami structures.

First, two cryo-EM structures at 4.9 and 5.2 Å resolution revealed the architecture of the ‘6- helix bundle with a clasp’ (6HBC) origami, which forms a hexagonal arrangement of 6 helices (H1-H6) and adopts distinct conformations during transcription and folding (McRae et al., 2023). Specifically, cryo-EM data supported by SAXS data showed that H6 rotates by 177° around its helical axis and a rotation of about 38° between its state immediately after transcription and its state ~10 h post-transcription (Figure 6). Remarkably, this rotation involves remodeling of a kissing loop formed between the two halves of the H6 (McRae et al., 2023).

Figure 6.

Second, 16 cryo-EM maps at 23-27 Å resolution showed that extensions of the 6HBC origami, such as the ‘16-helix satellite’ (16HS) origami, comprising other synthetic units called ‘5-helix tile’ (5HT) and ‘6-helix bundle’ (6HB) connected by kissing loops, form a Y-shaped antibodylike structure (McRae et al., 2023). Individual-particle cryo-electron tomography (IPET) revealed 16 different conformations of three structural modules of 16HS. These data showed that, within 16HS, the kissing loop connecting 5HT and 6HB rotates by about 150° (Figure 6) Third, the cryo-EM structure at 5.4 Å resolution of the ’14-14 traptamer’ device, displayed how the dynamics of a synthetic RNA can be used to switch on and off a fluorescent aptamer (Vallina et al., 2024). Specifically, in the 14-14 traptamer the fluorescence of iSpinach is modulated by controlling how this aptamer recognizes its cognate fluorogenic ligand. The cryo-EM structure showed that in the apo-traptamer the iSpinach motif is bent by about 68° with respect to the ligand bound state, positioning residues U22 and A45 in the ligand binding pocket and thereby inhibiting the binding of the fluorogenic dye (Figure 6) (Vallina et al., 2024). Observing the folding mechanism of RNA origami and understanding the dynamics of these synthetic RNAs is crucial to guide and design the novel assemblies of RNA nanostructures and to achieve novel functions through nanotechnology.

Chimeric RNAs

Finally, both natural and synthetic RNAs have been engineered into fusion constructs, called chimeric RNAs, to combine biological functions or facilitate structural characterization by increasing the size of the target. Using this strategy, cryo-EM structures have been determined, for example, for the Zika virus xrRNA fused to a group I intron (at 4.5 - 5.0 Å resolution) and for the thiamin phosphopyrate riboswitch or the raiA bacterial ncRNA fused to a group II intron (at 2.5 and 3.0 Å resolution, respectively) (Haack et al., 2025; Langeberg & Kieft, 2023). Negative stain images (but not molecular structure models) were also obtained for a mango-III aptamer, a pre-tRNA, or a primary microRNA (pri-miRNA) fused to ROOL (Ling et al., 2025). Finally, two new ad hoc engineered scaffolds with two- and four-fold symmetry, respectively, were used to determine the cryo-EM structures of tRNA^Asp, Mango-III, quinine and 8-oxoguanine aptamer at 3.7, 3.0, 2.9, and 2.9 Å resolution, respectively (Jones & Ferre- D’Amare, 2026). Interestingly, a completely synthetic chimeric RNA, called the D43 aptazyme, has also been developed to act as a gene regulation device. Its cryo-EM structure revealed peculiar dynamics between its functional modules, as highlighted below.

Case study 8: Dynamics of the D43 aptazyme

D43 is a synthetic RNA device, which has been developed to control the expression of programmed cell death protein (PD-1) in mammalian cells (Stagno et al., 2025). It is composed of a tetracycline-binding aptamer and a hammerhead ribozyme connected by a communication module. The communication module is sandwiched between coaxially stacked RNA helices from the aptamer (helix P1) and the ribozyme (stem II). The communication module is also connected to a non-canonical pseudoknot formed by loop L3 and junctions J1/2 and J2/3, which is part of the tetracycline-binding site. Two cryo-EM structures at 3.2 and 3.0 Å resolution, supported by NMR and SAXS data, revealed the architecture and dynamics of the D43 aptazyme in the apo- and holo-conformations (Stagno et al., 2025). Based on these studies, the D43 aptazyme appeared to be regulated at the nucleotide level, rather than involving large domain motions, as for the RNA origami discussed above. Specifically, in D43, the tetracycline-binding pocket remains largely similar (the overall RMSD difference between the two structures is only 1.4 Å), but two base pairs in the communication module dynamically change conformation (Figure 7). In the absence of the ligand, base pairs G49•C99 and G48•U100 open, breaking the coaxial stacking between helix P1 and stem II and thus inducing flexibility of the whole aptazyme (Figure 7).

Figure 7.

Challenges inherent to cryo-EM studies of dynamic RNAs and how to overcome them

The case studies highlighted above were successful thanks to specific computational, experimental, and technological developments that helped overcome the intrinsic challenges posed by dynamic RNAs, and particularly challenges in sample design, sample preparation and data acquisition and processing, as discussed below.

Challenge 1: Construct design and engineering of dynamic RNAs

Designing and engineering suitable constructs is an important step before performing a cryo-EM experiment with dynamic RNA molecules and can be based on computational predictions, biochemical information or previously solved structures.

To capture catalytic dynamics of the Tetrahymena group I intron, this ribozyme has been engineered to introduce an extended internal guide sequence (IGS), which forms dynamic helices that are important for both splicing steps (Figure 1). Moreover, a mutation (G264A) was introduced in the active site to slow down the first step of splicing, and 5’- and 3’-exon like substrates were added to capture different conformations during the second step of splicing (Figure 1). Modelling with SimRNA (Boniecki et al., 2016) was also used to guide the structure determination of a previously unseen group I intron conformation involving a pseudoknot and a four-way junction, whose relevance was confirmed biochemically, with mutations.

Informed by the architecture of full-length group II intron from previous structures, folding intermediates of this ribozyme could be obtained by engineering several constructs consisting of different domains of the RNA added in a sequential manner (D1, D1-2, D1-3, D1-4, D1-5, Figure 2). Engineering of mutations in an important hinge motif further served to validate the mechanistic folding model (Figure 2).

No prior structural information was available for viral tRNA-like structures (TLS), making it difficult to model the atomic coordinates of the wild-type RNA in cryo-EM maps (Bonilla et al., 2021). This prompted the authors to design two constructs of BMV TLS RNAs with extended/truncated helices, with the rationale that local differences between WT and mutant constructs can help in assigning helices. One construct with stem B3 extended and stem C shortened, helped identify these helices in the WT, owing to the local differences. The same strategy was applied with another construct with stem D extended and stem B2 shortened (Figure 4). A similar strategy was employed for determining the cryo-EM structures of Coronaviridae SL5 motifs (Kretsch et al., 2024). Here, an engineered extended SL5 construct possessing a helix downstream of SL5 (helix SL6) was designed and characterized by cryo-EM to help identify the 3’-terminal region of SL5 and thus correctly assign the nucleotides in the density (Figure 4).

In vitro evolution experiments were used to select a functionally active and stable RNA polymerase ribozyme (Figure 5) (McRae et al., 2024). Synthetic design was also used to generate RNA origami structures, by engineering kissing loops and crossovers using an automated design (ROAD) pipeline (Figure 6) (Geary et al., 2021; McRae et al., 2023). Iterative engineering and cryo-EM modeling informed the design of constructs with increasingly improved geometry and shape, and to identify and avoid the formation of kinetic trap during co-transcriptional folding (McRae et al., 2023).

Finally, engineering a stabilizing mutation (U100C) in the communication module allowed better visualization of overall structure and dynamics of the D43 aptazyme in both the holo and apo states (Stagno et al., 2025).

Challenge 2: Sample preparation of dynamic RNAs

After identifying and bioinformatically characterizing the target, the next critical challenging step for the cryo-EM analysis of dynamic RNAs is sample preparation. RNA sample preparation consists of three steps: production, purification and biochemical and biophysical characterization, structural characterization in solution, and cryo-EM grid preparation.

Production, purification, and biochemical and biophysical characterization

RNA synthesis is critical to yield large quantities of homogeneous RNA for grid vitrification and retain the physiological dynamic nature of the RNA target. Recent optimizations of in vitro transcription protocols have been crucial to enable successful studies on dynamic RNAs. First, screening of ionic strength (particularly physiological ions such as magnesium and potassium) and pH is crucial because buffer composition affects RNA dynamics and its stability in the vitreous ice (Chillon et al., 2015; Uroda et al., 2020). Recent studies on dynamics RNAs have reported the use of pH in the range from 6.0 to 8.0 and of 1-30 mM magnesium and 10-150 potassium ion concentrations (Chen et al., 2025). It is important to consider that magnesium ions are essential for RNA folding, ligand recognition, and ribozyme activity, but high levels of magnesium may induce RNA particle oligomerization and aggregation, exacerbating a common challenge of dynamic RNAs (Chen et al., 2025). Second, purification strategies also affect RNA dynamics and its folding into physiological, functional structures. Two strategies have been successfully used for studying dynamic RNAs, namely denaturing and non-denaturing purification protocols (Chillon et al., 2015; Nilsen, 2013). Purification using denaturing acrylamide gels and refolding has been used for cryo-EM studies of Tetrahymena group I intron dynamics (Li et al., 2023; Zhang et al., 2023) (Figure 1), TLS and SL5 viral RNA elements (Bonilla et al., 2021; Kretsch et al., 2024), and the cobalamin riboswitch (Ding, Deme, et al., 2023). In this procedure, the temperature at which refolding and reconstitution with physiological ions occur is crucial. For instance, Tetrahymena group I intron folding intermediates have been obtained at low temperature (25 °C), whereas the native states have been obtained at higher temperature (50 °C) (Bonilla et al., 2022; S. Li et al., 2022). Non-denaturing purification has been utilized for cryo-EM studies of group II intron folding intermediates (Jadhav et al., 2025), RNA polymerase ribozyme (McRae et al., 2024), group I intron (Luo et al., 2023), RNA origami scaffolds (McRae et al., 2023; Sampedro Vallina et al., 2023; Vallina et al., 2024) and RNA D43 aptazyme (Stagno et al., 2025).

Followed by purification, biochemical and biophysical characterization is necessary to confirm that the target RNA is pure and functionally active, to obtain initial insights into its dynamics, and to exclude the formation of non-specific aggregates or unfolded molecules (Chen et al., 2025). Studies on engineered ribozymes have confirmed activity using enzymatic assays (Figures 1–2). Furthermore, to confirm homogeneity, studies on group II intron folding intermediates, RNA origami (5HT, 6HB, 6HBC, 16HS, and traptamer), and the D43 aptazyme have used gel electrophoresis and size exclusion chromatography (SEC) (Jadhav & Marcia, 2025b; McRae et al., 2023; Stagno et al., 2025). It is worth noting that in some cases analytical gel electrophoresis shows high sample homogeneity (i.e. BMV TLS, Figure 4) while in other cases it displays multiple bands (i.e. group I and II introns, Figures 1–2, and D43 aptazyme, Figure 7), but these samples are nonetheless suitable for cryo-EM investigation. Studies on group II intron folding intermediates have additionally used mass photometry (MP) and size exclusion chromatography coupled to multiangle laser light scattering (SEC-MALLS) to identify unfolded or aggregated forms of the target and to gain information regarding molecular weight and hydrodynamic properties of the target (Jadhav & Marcia, 2025b). These latter techniques require significantly different amounts of RNA for analysis, i.e. micromolar concentrations for SEC-MALLS and nanomolar concentrations for MP. Thus, comparison of how the RNA behaves in these different techniques is particularly informative to judge if concentration plays a role in determining RNA homogeneity.

Finally, biochemical probing can also support studies on RNA structural dynamics, specifically maps the secondary structure of the RNA target and thus facilitates model building in the cryo-EM maps. For instance, the Ribosolve pipeline, which integrates secondary structure probing using the so-called “mutate-and-map read out through next-generation sequencing” approach (M2-seq), is now largely used in RNA cryo-EM studies (Cheng et al., 2017; Kappel et al., 2020). Such approach was used to elucidate the structure of the SL5 element of Coronaviridae (Figure 4) (Kretsch et al., 2024). Chemical probing studies has also revealed the deformation of specific tertiary motifs during the folding process of group II introns (Waldsich & Pyle, 2008). Now, the structural basis behind such deformation and its implications for group II intron dynamics and folding has been confirmed and rationalized by cryo-EM studies on group II intron folding intermediates (Figure 2) (Jadhav et al., 2025).

In-Solution RNA characterization

After purification and before cryo-EM imaging, visualizing RNA dynamics in solution is beneficial to gain initial insights into the range of conformational changes that the target undergoes. Three techniques have been used in recent cryo-EM studies of dynamic RNAs, namely SAXS, NMR and AFM.

First, SAXS has been used to study group II intron folding intermediates (Jadhav & Marcia, 2025b), 6HB RNA origami nanostructures (McRae et al., 2023), and the D43 aptazyme (Figure 7) (Stagno et al., 2025). SAXS provides insights into RNA flexibility, hydrodynamic parameters like the radius of gyration (Rg) and the maximum pairwise interatomic distance (Dmax), and the overall low-resolution shape of the molecule, which can later support RNA structure modeling into the cryo-EM maps (Chen & Pollack, 2016; Kim et al., 2020). In addition, SAXS informs about the effect of ions or ligands on RNA conformation (Pollack, 2011). For instance, the folding of domain 3 of group II introns, which forms a characteristic ‘S-turn’ motif (Marcia & Pyle, 2014), is significantly impacted by the ionic composition of the buffer and SAXS studies could capture this dependence. In the absence of potassium ions, group II intron domain 3 is elongated, but in the presence of potassium ions it docks onto the globular scaffold of the intron (Jadhav & Marcia, 2025b). Finally, SAXS helps characterize conformational ensemble states of the target and suggests if the cryo-EM data have captured all or only some of the conformational states explored by the target in solution. For instance, the back- calculated SAXS curves of apo- and holo-D43 aptazyme cryo-EM structures differ from the experimental scattering curves (Figure 7). The deviation is greater for the apo-state compared to the holo-D43 aptazyme. Additionally, cryo-EM structure-derived curves are similar to each other compared to the comparison of SAXS experimental curves. This discrepancy suggests that there are structural differences between the apo- and holo-D43 aptazyme (Figure 7) (Stagno et al., 2025).

Second, NMR provides dynamics insights at the single nucleotide level, i.e. changes in base pairing or base stacking patterns, and it was used to support cryo-EM studies on the dynamic D43 aptazyme (Figure 7) (Stagno et al., 2025). Such conformational changes may be difficult to visualize in the cryo-EM maps, for instance due to modest resolution. In the case of the D43 aptazyme, NMR helped define the conformation of nucleobases from the communication module to transduce the signal from the aptamer to the ribozyme module (Figure 7) (Stagno et al., 2025).

Finally, AFM together with SAXS and isothermal titration calorimetry (ITC) was used in support of cryo-EM studies on the dynamics of the cobalamin riboswitch (Ding, Deme, et al., 2023; Ding, Lee, et al., 2023). AFM captures the 3D shape and contour information of the target RNA immobilized on mica surfaces. For the cobalamin riboswitch, AFM specifically helped identifying distinct orientation of RNA helices and the dimerization of the RNA (Figure 3) (Ding, Lee, et al., 2023).

Cryo-EM grid preparation

After production, purification and biochemical and biophysical characterization, the RNA target is ready for cryo-EM investigation of its dynamics, at high resolution. The first step to prepare the sample for cryo-EM is grid preparation. Recent studies on dynamic RNAs have shown that two steps in grid preparation are important, namely inspection by negative staining followed by grid vitrification.

First, negative staining is rarely used nowadays, considering the easier accessibility and higher throughput of vitrification approaches. But studies on dynamic RNAs have shown that negative staining can still provide useful quick insights into the homogeneity of the target, its stability, and its behavior on the EM grids. For instance, negative staining procedure have been used to visualize dynamics of the 16HS RNA origami and of group II intron folding intermediates (Figure 6) (Jadhav & Marcia, 2025b; McRae et al., 2023). Initial views into negative staining images of RNAs may display compact, flexible, or unfolded thread-like molecules of the RNA sample. Depending on the RNA behavior, one may revisit RNA purification before proceeding with vitrification trials.

Second, grid vitrification is affected by a number of factors. The choice of the grid type is critical because it is one of the factors that determines the thickness of the vitreous ice. Different sizes of the holes in holey carbon and the metal support, typically copper or gold, can yield different ice thickness after plunge freezing and affect beam-induced particle motion, which consequently affects the resolution of the data (Russo & Passmore, 2014). The cobalamin riboswitch (Ding, Deme, et al., 2023), the D43 aptazyme (Stagno et al., 2025), the traptamer RNA origami (Vallina et al., 2024), the 16HS and 6HBC RNA origami (McRae et al., 2023), and the RNA polymerase ribozyme (McRae et al., 2024) have been vitrified using 300 mesh 1.2/1.3 Au grids. Group II folding intermediates have been vitrified using 300 mesh 1.2/1.3 Cu grids (Jadhav et al., 2025; Jadhav & Marcia, 2025b). Group I intron samples have been vitrified using 200 mesh 2/1 Au or Cu grids (Li et al., 2023; Luo et al., 2023; Zhang et al., 2023). These examples show that although the choice of the size of the holey carbon holes should in principle be directed by the length of the target RNA, 1.2 μm holey carbon layers have been used successfully for the grid vitrification of RNA samples ranging from 100 kDa ribozymes to megadalton-large RNA nanocage structures (Haack et al., 2025; Kretsch, Wu, et al., 2025; Ling et al., 2025).

Challenge 3: Cryo-EM data collection and processing of dynamic RNAs

After successfully depositing the target RNA on cryo-EM grids, the next step towards structure determination consists of data collection and processing. Here, challenges specific to dynamic RNAs can be identified at the screening, data collection, data processing, and structure reconstruction steps.

Cryo-EM grid screening

Grid screening is necessary to identify grids with optimal particle distribution and density throughout grid squares. Dynamic RNA particles often do not behave homogenously. Reports on which grid squares are most successful for structure determination of dynamic RNAs are rare, but recent studies on group II intron folding intermediates showed that dynamic RNAs prefer thicker vitreous ice squares compared to thinner ones (Jadhav & Marcia, 2025b), and this may be a generalizable feature typical of dynamic RNAs. Thus, dynamic RNAs display optimal distribution and well-folded particles in the smaller squares, which harbor thicker ice (Figure 2) (Jadhav & Marcia, 2025b).

Cryo-EM data collection

The thicker ice required by dynamic RNAs poses a challenge in data collection for two reasons.

First, thick ice squares can hamper localization of holey carbon holes. This problem can now be efficiently circumvented thanks to recent technological advances, i.e. direct alignment approaches, wherein an individual hole is centered just before image acquisition, or the newer “junk detector” tool in cryoSPARC, which can exclude micrograph images collected on carbon. However, aberration-free image shift (AFIS), which is beneficial to enable faster and larger data acquisitions, is not efficient when collecting on grids with thicker ice. Should AFIS be utilized for data collection on dynamic RNA samples, then the user should pay particular attention to exclude from the dataset images collected on the carbon support instead of in the holes, before downstream data processing.

Second, thick ice would typically require higher acquisition dose and would thus risk to induce particle radiation damage or more pronounced beam-induced motions. But the high ‘Z’ number of phosphorus, which is abundant in RNA, increases RNA’s overall electron scattering potential and allows using proportionally lower electron dose for data collection on RNA vs proteins. While a systematic study on the effect of electron dose on image quality has been performed only for RNA origami nanostructures (McRae et al., 2023), an empirical compromise used in most other recent studies on dynamic RNAs is that the same dose that is suitable for proteins in thin ice (i.e. 40-60 e-/Ų) is also suitable for dynamic RNA in thicker ice. For instance, an electron dose of 60 e-/Ų has been used to image the TLS element of BMV (Bonilla et al., 2021), the RNA polymerase ribozyme (McRae et al., 2024), and group I introns (Luo et al., 2023). An electron dose of 50 e-/Ų has instead been applied to image other group I intron samples (Li et al., 2023; Zhang et al., 2023), the cobalamin riboswitch (Ding, Deme, et al., 2023), and the PD1 aptazyme (Stagno et al., 2025). A lower electron dose of 40 e-/A2 has been used to acquire movies of group II intron folding intermediates (Jadhav et al., 2025).

Cryo-EM particle picking

During data processing, particle picking needs to ensure that aggregated or unfolded particles, which pose a particular challenge for dynamic RNAs (see above), are excluded from the dataset. To facilitate particle picking, the choice of an appropriate box size, i.e. a box significantly larger than the expected particle dimensions is beneficial, because it helps to distinguish between folded particles and unfolded thread-like particles (Kretsch et al., 2024). Of note, RNAs that possess both structured and flexible regions may appear as a comet on raw micrographs, where the head corresponds to the structured region of the RNA and the tail to its flexible domains (Figure 2) (Jadhav & Marcia, 2025b). Both CryoSPARC and EMAN2 have been used successfully for particle picking in recent studies on dynamic RNAs (Punjani et al., 2017; Tang et al., 2007). For instance, group I intron samples were processed using pipeline involving motion correction (MotionCor2 (Zheng et al., 2017)), CTF estimation (CTFFIND4 (Rohou & Grigorieff, 2015)), particle picking with EMAN2 (Tang et al., 2007), 2D classification in Relion (Scheres, 2012) and 3D classification and 3DVA analysis in CryoSPARC (Punjani & Fleet, 2021; Punjani et al., 2017). Instead, studies on the BMV TLS element (Bonilla et al., 2021), RNA polymerase ribozyme (McRae et al., 2024), the D43 aptazyme (Stagno et al., 2025), RNA origami nanostructures (McRae et al., 2023) and the SL5 viral element (Kretsch et al., 2024) used CryoSPARC for the entire data processing pipeline. Cryo-EM structures of group II intron folding intermediates were acquired using CryoSPARC and EMPROVE (Jadhav et al., 2025; Maiorca et al., 2025). Cobalamin riboswitch study used SIMPLE software (Caesar et al., 2020) for micrograph pre-processing and CryoSPARC and Relion for 3D reconstruction (Ding, Deme, et al., 2023).

Cryo-EM 3D structure reconstruction

Besides particle picking, another important step in data processing of dynamic RNAs is 3D reconstruction and the visualization of the target’s conformational heterogeneity. The first step in 3D reconstruction is ab initio reconstruction. Ab initio reconstruction provides the initial map of the RNA and therefore affects the accuracy and quality of the final, refined maps (Ma et al., 2022). Once an initial map is acquired, refinement consists of homogenous refinement steps to improve the resolution of the initial map, and heterogeneous refinement steps to gain insights into the heterogeneity of the target. Biological macromolecular structures display two types of heterogeneity, compositional and conformational heterogeneity. Compositional heterogeneity exists due to distinct molecular composition of the same target. In contrast, conformational heterogeneity, which corresponds to the presence of discrete or continuous conformational states, arises from the inherent motion of different domains within the same macromolecular complex. Resolving both types of heterogeneity is essential to determine the high-resolution structure and gain functional understanding into dynamic RNAs, but this is a challenge, partly due to unique signal to noise ratio associated with each cryo-EM sample and dataset which makes it difficult to develop algorithms that can process all cryo-EM samples (Kimanius & Schwab, 2024). Recent studies on dynamic RNAs have successfully used three different approaches to resolve RNA compositional and conformational heterogeneity. 3D classification in Relion (Kimanius et al., 2016) or CryoSPARC (Punjani et al., 2017) has been successfully used to discern distinct conformations of the cobalamin riboswitch (Ding, Deme, et al., 2023). 3D classification in Relion has also been used to reveal the dynamics that govern the progression of group I intron throughout its self-splicing reaction (Luo et al., 2023). 3DVA (Punjani & Fleet, 2021) has been successfully applied to gain functional insights into the dynamics of group I introns (Li et al., 2023), group II intron folding intermediates (Jadhav et al., 2025), the viral SL5 and TLS elements (Figure 4) (Bonilla & Kieft, 2022; Bonilla et al., 2021; Kretsch et al., 2024; Yang et al., 2025), 5HT and 6HBC RNA origami structures (McRae et al., 2023) and the RNA polymerase ribozyme (McRae et al., 2024). Finally, a recent algorithm still under development (EMPROVE), which helps identify discrete low occupancy conformational states, has also been successfully employed for resolving group II intron dynamics (Maiorca et al., 2025). Other approaches to resolve compositional and conformational heterogeneity that have been developed recently, albeit not having been used in studies on RNA dynamics, are CryoDRGN (Zhong et al., 2021) and e2gmm (Chen & Ludtke, 2021; Chen et al., 2023).

Conclusions and Future Perspectives

The study of RNA has long been constrained by static representations of inherently dynamic molecules. Here, we review recent advances in the field that collectively point to a fundamental transition: RNA structural biology is moving beyond the determination of individual conformations toward the reconstruction of dynamic ensembles and, ultimately, energy landscapes. Cryo-EM has emerged as a central driver of this transition by enabling direct visualization of both discrete structural states and continuous conformational variability across a broad spectrum of RNA systems.

RNA function is encoded not in a single structure, but in the accessibility, interconversion, and regulation of multiple conformational states. Across diverse classes of RNAs – including ribozymes, riboswitches, viral elements, and synthetic assemblies – cryo-EM is now enabling to describe the dynamics that enables RNA function. The case studies discussed in this review illustrate how cryo-EM can capture RNA dynamics states and the motions that connect them, revealing mechanisms such as hinge-mediated domain rearrangements, ligand-dependent switching, and cooperative assembly processes. Importantly, these insights are increasingly quantitative, providing spatial and mechanistic detail that links structural transitions directly to functional outcomes.

At the same time, these advances highlight the experimental and computational challenges that must be overcome to fully realize this paradigm. The intrinsic flexibility, heterogeneity, and sensitivity of RNA impose stringent requirements on construct design, sample preparation, vitrification, and data analysis. Recent progress in these areas – including improved RNA engineering strategies, optimized vitrification approaches, and powerful computational tools for disentangling conformational heterogeneity – has been essential for extending cryo-EM to dynamic and previously intractable RNA targets. Continued innovation along these axes will be critical for pushing the limits of resolution, sensitivity, and interpretability.

Looking forward, the most impactful developments are likely to arise from strengthening the integration of cryo-EM with complementary approaches. Coupling cryo-EM with molecular dynamics simulations, chemical probing, fluorescence microscopy, and other solution-based techniques is starting to make it possible to construct experimentally grounded conformational ensembles that bridge static structures and dynamic behavior. For instance, using fluorescent tags to localize specific RNAs in situ (i.e., based on CRISPR/Cas13a or MS2/PP7 systems) enables correlative light and electron microscopy studies linking the dynamics observed in vitro to how RNAs behave inside living cells (Cao et al., 2022; Hampton et al., 2017; Moser et al., 2019; Park et al., 2014). Furthermore, the observation of RNA dynamics through cryo-EM synergizes with serial femtosecond crystallography (SFX) which allows the visualization of ultrafast structural intermediates (Chapman et al., 2011; Stagno et al., 2017). In this context, advances in ultrafast vitrification hold the promise of capturing transient intermediates along functional pathways, bringing structural biology closer to a true “molecular cinematography” of RNA processes. For instance, Chameleon provides blot-free self-wicking, ultra-rapid vitrification, and automated sample handling (Levitz et al., 2022), while time-resolved cryo-plunger, Easy-grid, microfluidic chips or micro-sprayers allow for on-the-spot mixing and spray-freezing (Garg et al., 2025; Torino et al., 2023). These approaches reduce RNA denaturation and structural disturbance and increase the chance to capture short-lived states. Computational advances in the analysis of RNA sequences and in the prediction of RNA 3D structures are also starting to play a transformative role. For now, recent CASP competitions (Kretsch et al., 2026) demonstrated that physics-based models, like Vfold (J. Li et al., 2022) and SimRNA (Boniecki et al., 2016) are still generally more accurate in predicting RNA 3D structures de novo than AI-based ones, like AlphaFold3 (Abramson et al., 2024), RNAnneal (Herron et al., 2026), DynaRNA (Li et al., 2025), RoseTTAFold2NA (Baek et al., 2024). But these same CASP competitions are also promisingly showing that AI-based RNA structure prediction algorithms can powerfully integrate the prediction of RNA conformational ensembles (Dube et al., 2026) and of dynamic solvent shells surrounding RNA structures, especially when integrated with SAXS, outperforming MD predictions (Kretsch, Posani, et al., 2025; Patt et al., 2025).

But AI is not only improving RNA structural predictions. It is also starting to make an impact on RNA cryo-EM data analysis. AI-driven particle picking and classification are getting better at handling extreme heterogeneity, automatically telling folded cores apart from flexible tails and spotting rare but functionally important states in huge datasets. We can expect that nextgen algorithms will further blend the strengths of tools like cryoDRGN, 3DVA, and EMPROVE (Maiorca et al., 2025; Punjani & Fleet, 2021; Zhong et al., 2021) and directly feed into MD simulations and ensemble refinement. This combination will help reveal exactly how solvent, ions, and structural fluctuations drive function and improve the modeling of flexible or poorly resolved regions (Bernetti & Bussi, 2023; Languin-Cattoen & Bussi, 2026; Wlodarski et al., 2024).

Ultimately, the emerging dynamic view of RNA has important implications beyond fundamental biology. A detailed understanding of RNA conformational landscapes informs the development of RNA-targeted therapeutics by revealing transient binding pockets and regulatory states that can be selectively stabilized or disrupted. Likewise, the ability to design and control RNA dynamics expands the toolkit of RNA nanotechnology, enabling the creation of responsive molecular devices with programmable behavior. In this context, cryo-EM is poised to play a central role, providing the structural framework needed to connect sequence, structure, dynamics, and function.

In summary, the convergence of methodological innovation and conceptual advances is redefining how we think about RNA. By linking structural snapshots into coherent dynamic models, cryo-EM is transforming our understanding of RNA from static architecture to adaptive, functional motion. This shift toward a dynamic, ensemble-based perspective not only resolves long-standing challenges in RNA biology but also opens new avenues for discovery at the interface of structural biology, biophysics, and medicine.

Table 1.

Acknowledgements

We thank all members of the Marcia lab for helpful discussion. Work in the Marcia lab is partly funded by the Swedish National Research Council (VR, 2024-04107), by the HORIZON- MSCA-2023-DN-01 action (project: TargetRNA, n. 101168667), by Cancerfonden (project 25 4360 Pj), by SciLifeLab (Strategic initiatives and technology development call 2025), and by the Italian Association for Cancer Research (AIRC, IG 28746).

Additional information

Author contribution statement

MM conceived and supervised the work and obtained funding; MM, SJ, SS, and QS analyzed the structures; SJ and MM interpreted the data with contribution of all authors; SJ and MM produced the first draft of the manuscript; all authors approved the final version of the manuscript.

Funding

Vetenskapsrådet (VR) (2024-04107)

• Marco Marcia

EC | Horizon Europe | Excellent Science | HORIZON EUROPE Marie Sklodowska-Curie Actions (MSCA)

https://doi.org/10.3030/101168667

• Spandan Saha

• Marco Marcia

Cancerfonden (Swedish Cancer Society) (25 4360 Pj)

• Marco Marcia

Science for Life Laboratory (SciLifeLab) (Strategic initiatives and technology development call 2025)

• Marco Marcia

Fondazione AIRC per la ricerca sul cancro ETS (AIRC) (IG 28746)

• Marco Marcia