PERSPECTIVE – RNA DYNAMICS today and tomorrow

Graphical Abstract

This is the best of times for RNA science. While some RNAs have been known for decades, like the ribosomal RNAs (rRNA), the small nuclear RNAs (snRNAs), and tRNAs, others are new – long noncoding RNAs (lncRNAs), circular RNAs (cRNAs), pUG RNAs, to name but a few. These RNAs are essential in cells, where they can be regulatory (miRNAs, riboswitches, CRISPR), enzymes (rRNA, snRNA, Group I/II introns), or scaffolds (stress granules). We know that most of these RNAs undergo substantial conformational changes to accomplish their function, in part simply a consequence of starting life as a single strand. Most RNAs in a cell pass from a single strand to secondary structures (e.g. hairpins) and some to a tertiary fold (1D to 2D to 3D).

RNAs can fold because they are flexible – they have intrinsic molecular motions; that is, their dynamics. The ability of an RNA to adopt different folds is dramatically illustrated in riboswitches [1]. During transcription of a (prokaryotic) mRNA, the riboswitch will have an ‘off’ state and an ‘on’ state, two mutually exclusive conformations that can interconvert. For example, a riboswitch bound to its specific ligand will shut down transcription or translation. Ligand binding is accompanied and facilitated by a conformational change. RNAs in the ribosome [2] and spliceosome [3] undergo massive conformational changes (facilitated by proteins, ATP, and GTP), and do chemistry. Here is the challenge – how are the conformational dynamics of an RNA measured, predicted, or even imagined?

What are the timescales and magnitudes of these conformational dynamics? Within any RNA, there is a plethora of dynamic motions that are ongoing (Figure 1). These motions range from the ps-ns ribose repuckering to the minutes/hours of folding a single strand into a pUG quadruplex (Figure 2). Flexible elements in an RNA molecule can bind proteins, small molecules, ions, and each other. The paradigm of conformational selection in biological interactions might have been developed for RNA molecules as they interact with their ligands. Induced fit, driven by those ligands, is also a binding mechanism. Frequently, both mechanisms are present.

Figure 1.

Cartoon of RNA conformations, from single strands to secondary structure. Examples of biological RNAs containing a structure are listed. Timescales of dynamics of the conformational changes are based on experimental values. NMR – nuclear magnetic resonance. smFRET - single molecule Förster resonance energy transfer. Fluorescence is time correlated single photon counting fluorescence. REMD – replica exchange molecular dynamics. CG-coarse grain.

Figure 2.

A schematic of the pUG folding pathway and a quartet structure. Phosphodiester backbone is purple, nucleobases are green. Pink dots K⁺ ions. Adapted from Peterson [40].

Any single RNA structure must be considered as one of an ensemble of structures that are sampling conformations in time and space. The ensemble composition will change with temperature, ions, and ligands. There is no theoretical partition function that can quantify the population distribution. The biological function of an RNA will depend on the composition of the ensemble. The field needs new tools to describe the ensemble, where the RNA scientist can identify specific states to be investigated.

Describing the molecular motions of any RNA requires experiments that can probe timescales from ps to days, and can capture conformational changes throughout the molecule. Then, not every RNA molecule can be studied in vitro. Computational methods that elucidate the timescales and structural changes are necessary. For this to happen, there must be synergy between benchtop and computational approaches to arrive at a robust description of RNA molecules [4]. The goals of such a collaboration are lofty – to be able to design new RNAs that bind small molecules (therapeutics), to design small molecules that bind a flexible RNA, to predict the energetics of protein:RNA interactions, and to design new proteins and RNAs that together have biological activity. What is inherent in all these goals is the fundamental properties of RNA dynamics that allow interactions and chemistry to happen.

Figure 1 imagines the path of an RNA molecule from single-strand to secondary structures. I’ll start this Perspective from the ‘simple’ single strand and end with a folded RNA, with an emphasis on the dynamics of the RNA and its relation to function.

Single-stranded regions of RNAs are critical, as they allow the twists and turns of secondary structures to adopt tertiary structures. It’s worth pointing out that there are seven degrees of freedom per nucleotide, which allows single strands of RNAs to fold. Its3-way junctions, 4-way junctions, and hairpin loops are flexible, where they can form sites for interactions with ligands. They are also sites for protein binding, because they allow a protein to ‘read’ the nucleobase moieties. Unlike proteins binding to a DNA duplex, an RNA duplex does not allow a protein to make sequence-specific contacts (unless there is a deformation of the A-form duplex), so a protein uses single-stranded sites. Despite their essential contributions to an RNA molecule, there are surprisingly few molecular descriptions of single stranded RNA in solution.

The sequence of nucleobases in an ssRNA can confer unique local conformational flexibility [5]. Uridine is essential for ssRNA flexibility, since it has no propensity to stack, as seen in NMR experiments of the r(UUUU) tetramer [6]. Adenines and Cytidines stack, and NMR experiments on r(AAAA), r(GACC), and r(CAAU) [6] [7] show these tetramers also form an A-form helix. Guanine self-associates, often forming quadruplexes but worse.

Using a combination of SAXS and FRET-FCS, Wang et al. [8] studied the conformational dynamics of a 30-mer RNA 5′rAAGAAUAAAAGAGAAGCCACCCCACCCAGA-3′ and compared it to r(U)₃₀. In their 100 mM NaCl solution (10 mM MOPS, pH 7), the radius of gyration (by SAXS) was 22.40 ±0.48 Ǻ; rU₃₀ values were somewhat larger (~4 Ǻ). These data complement measurements of the chain conformational dynamics that can be measured by attaching a FRET pair to either end, as in the single-molecule nanosecond FCS experiments of Nuesch et al.[9] with rU₁₉, rC₁₉, and rA₁₉. Those investigators reported a correlation time for the chains: 17± 3 ns for rA₁₉ and 10± 2 ns for rU₁₉ and rC₁₉ (in 150 mm NaCl, HEPES pH 7). These studies focused on the global properties of a single-stranded RNA chain, but have implications for single-stranded regions in a long RNA molecule.

It is important to consider that nucleobases in a single-strand A-form helix are in equilibrium with a dynamically disordered state. In a time-resolved fluorescence study of 2-aminopurine (2AP, a fluorescent analog of adenosine) located at four sites in ^5’UAUACUUUUUAACUCUUAUCA^3’ we found that every 2AP had three lifetime decays of 150 ps to 5 ns, in addition to a dark state (in preparation). These lifetimes represent conformational states of the 2AP in the strand, and the dynamics of nucleobase motions in a single strand are rapid and complex. The conformational exchange of the nucleobases is one of the features that is most difficult to quantitatively describe in solution experiments.

The pre-Q1 riboswitch is the simplest known example of riboswitches (Figure 1), including both a single-stranded tail and a hairpin. The hairpin is effectively the aptamer where the ligand (pre-Q1) is bound, and the single stranded tail is the expression platform that regulates transcription. Base pairing of the preQ1 riboswitch 3′ tail with the aptamer loop produces a pseudoknot. The dynamics of that tail and the loop are key properties of its activity.

NMR studies of the 3′ ssRNA tail of the preQ riboswitch [10][11] showed how the tract of A’s in ^5′rAUAAA₅AAACUAA^3’ stacked to form a single-stranded A-form helix. Using NMR relaxation methods to calculate order parameters for the dodecamer (a measure of the ps-ns dynamics of the nucleotides), they found that nucleobases at the 5′ and 3′ ends of the strand were far more disordered than the A’s in the middle. None of the nucleotides was rigid, however. Remarkably, the single replacement of rA₅ with rC disrupted the stack and increased the conformational exchange of flanking nucleobases. Would an MD simulation reproduce this effect?

Molecular dynamics (MD) simulations are a promising method for describing the conformational states (the ensemble) and the timescale of nucleotide dynamics in an RNA single strand. All-atom MD looks at the nucleobase, the ribose, the phosphate, and ions, with timescales from ps to μs. However, MD force fields are still a work in progress [12][13], as experiments with the ‘simple’ tetranucleotide r(GACC) have revealed [14]. Several features of this small RNA have challenged force fields which need to balance base stacking, ribose repuckering, dihedral fluctuations, phosphate and base electrostatics and their interactions with water (nucleobases are not hydrophobic), which means that the water model used for RNA is important. Some experimental parameters for MD of RNA have been established, though.

The r(GACC) tetranucleotide has been a model system for MD experiments that compare the structural ensemble to NMR data. NMR experiments showed that it samples an A-form stack (75%) with a terminal C flip (25%). If there were other structures, they were too few to be detected [7]. In a comprehensive comparison of its structural ensembles in different MD force fields, none of the force fields produced the thermodynamically most stable (NMR) structure despite converged conformational ensembles [15]. These experiments need to be repeated with different single-stranded RNAs, combining modern NMR methods with the latest MD force fields.

To effectively sample conformational space, those simulations used REMD (replica exchange molecular dynamics), running either independent temperature or Hamiltonian simulations and allowing exchanges between them at specified intervals, maintaining thermodynamic balance. Conclusions from these and later MD experiments show that enhanced sampling is required for MD simulations of RNA.

Ideally, an MD simulation could describe a weighted conformational ensemble of the RNA and some indication of the lifetimes of the states. The dynamics of structural changes could be directly determined by fluctuations during a conventional simulation (replica exchange will lose such information but provide the ensemble). The outcome would be molecular details of the RNA dynamics, producing a movie of RNA in motion. However, MD is based on thermodynamics, and when a simulation does not include the thermodynamically most stable conformation, then accepting the ensemble as an accurate representation of its conformational space is not wise.

Hairpins.

Virtually all RNA molecules contain hairpins, which can form as the single strand is being synthesized by RNA polymerase. Since these are local secondary structures, they can form relatively quickly even as transcription is ongoing. Hairpins have base-paired duplexes (stems) and unpaired sequences (loops). They have been model systems for measurements of RNA thermodynamics, folding kinetics, and folding pathways, and they have unique dynamic properties.

STEMS.

In most RNA molecules, hairpin stems are short, 6–8 base pairs. RNA duplexes are A-form, prohibited by the ribose 2′OH from adopting the B-form of DNA. Base stacking, the enthalpic driving force for duplex formation, stabilizes even short A-form stems. In a duplex, there are two categories of base-pairs that are important determinants of stability, and they have different dynamics.

The terminal base pair in a duplex ‘frays’; that is, it samples conformations between hydrogen-bonded and not. A terminal A:U pair is less stable than a G:C pair, but either will fray. NMR experiments can observe the imino protons of a base pair if that proton is protected from exchange with water (each Watson-Crick base pair has one hydrogen-bonded imino proton that is observed in H₂O). NMR experiments to measure base pair lifetimes via the intensity of the imino proton signal [16] were unable to measure either terminal rA:rU or rC:rG lifetimes at 15° C, in 100 mM NaCl,1 mM EDTA, indicating that their lifetimes were less than 1 ms. As expected, the timescale of fraying varies with temperature and salt.

Base pairs in the stem are not static either, for they also make excursions out of the duplex. Again, NMR imino proton data have been used to measure the lifetimes of the stacked states [16]. The results revealed that base pair lifetimes are independent of flanking base pairs, and that they open independently. The timescale of base pair dynamics ranged from 50 ms for an internal G:C pair to less than 1 ms for an internal A:U pair.

NMR data do not provide any structural details of the excursions out of the helix. For example, do both terminal nucleobases unstack or only one? Since helices propagate from the end of a strand, such information on the terminal base pairs can inform synthetic constructs but also provide insight into the RNA:RNA duplexes that form in miRNA regulation.. Does a nucleobase flip out of a duplex? Does it move to the minor groove (shallow) where it can exchange its imino proton with water? Do both nucleotides move out of the duplex?

An MD simulation that starts with a perfect duplex would not be able to report the ms dynamics of base pair opening, but it was able to model terminal base pairs. Zgarbova et al. [17] ran 1 μs all atom molecular dynamics (MD) simulations with TIP3P waters in Na⁺, comparing fraying in RNA and DNA duplexes. Using AMBER ff99bsc0χOL3, the RNA terminal rG:rC remained stacked and base-paired for the entire 1 μs, while using ff99bsc0, the Χ torsion angles of both rG and rC moved from −120 to 0 degrees. For terminal rA:rU pairs, both nucleobases had significant conformational changes with both force fields. The authors were quick to note that populations were not converged, and suggested that longer simulations would be necessary.

A perfect RNA duplex is A-form, but in RNA molecules, the stems of hairpins often contain internal bulges, unpaired nucleotides, noncanonical nucleobase pairings, and nucleotide modifications. Combined with a phylogenetic comparison to identify mutable sites, the imperfections in the duplex could be linked to biological activity. They could be identified by secondary structure predictions, but solving their solution structures is difficult. What do we know about how these deviations affect RNA dynamics?

The now classic example of duplex imperfections is the TAR RNA hairpin from HIV, first solved by NMR [18]. It contains two helices connected by an asymmetric bulge, the bulge acting as a hinge. That internal bulge is a binding site for Tat protein, and the apical loop is required for binding by the Tat-Cyclin T1 complex. Three nucleotides on one side of the bulge change their structure dramatically when the Tat protein is bound.

Subsequent NMR experiments measured the dynamics of the bulge and the loop [19] leading to identification of a low probability ‘excited state’ structure of the internal bulge with a 2 ms lifetime [20]. This internal UCU bulge was extrahelical in an x-ray crystal structure, and could not be mapped into a single structure [21]. Introduction of the fluorescent 2-aminopurine (2AP) at four positions in the hairpin allowed measurements of time-resolved fluorescence of its duplex and bulges [21]. In the three-nucleotide internal bulge, the longest lifetime of 2AP was 4.4 ns, while in a single-nucleotide bulge, the longest lifetime was 7.7 ns. These lifetimes are consistent with an unstacked/stacked equilibrium within the large bulge, providing an indication of ns dynamics.

LOOPS. Loops are often sites where ligands bind, especially when they are unstructured. Such loops are expected to sample conformations, creating an ensemble. The dynamics of the sampling will depend on conditions such as ions and temperature. When such a loop is bound by a ligand, the dynamics of its conformational ensemble will facilitate interactions. Its dynamics are an important facet of its characterization.

The dynamics of the TAR apical loop, ^5’C₁U₂G₃G₄G₅A₆, were probed with NMR relaxation techniques [19]. G3 nucleobase and ribose had ps-ns local dynamics, and in other NMR structures, G3 was looped out. C₁ and G₅ form a base pair, although G₅ nucleobase has motions on the 30 μs timescale, indicating that the pairing is not stable (for reference, in the TAR construct for those experiments numbering is C₃₀ – A₃₅). The nucleobase A₆ is flipped out into solution, with dynamics on timescales from ps-ns to 37.3 ± 10.2 μs. This loop is extremely flexible, which could facilitate its interaction with proteins

Another loop with an unusual base-pairing scheme is the cU₁U₂C₃G₄g tetraloop, which is now the most widely studied model RNA hairpin. It was first identified as an interesting RNA by the realization that it was statistically overabundant in phage mRNA [22]. Its aberrantly high thermodynamic stability is due to its unique structure, which was first determined by NMR [23]. The c:g loop-closing base pair is an important feature of its structure and stability, as the U:G base pair stacks on it. That U:G pair is a trans wobble pair, where the G is syn. The C₃ nucleobase makes hydrogen bonds to a phosphate oxygen and a ribose 2′ OH. The highest resolution NMR structure of the cUUCGg tetraloop also summarized the dynamics of those nucleotides [24]. There is only one nucleotide that has significant dynamics: U₂ nucleobase and ribose and phosphate all undergo rapid conformational exchange (~300 ± 200 ps nucleobase), but also with syn/anti sampling on the ns timescale. In fact, this tetraloop family is classified as UNCG, where N is any nucleobase. It appears that N does not disrupt the interactions of the other nucleotides, and so retains its dynamic character.

The UUCG tetraloop has become the standard model for evaluation of molecular dynamics force fields, evaluated by their accuracy at reproducing its unusual structure. Significantly, the UUCG tetraloop structure does not require Mg²⁺ since no force field can handle it as an ion or a hexahydrate. A massive comprehensive comparison of UUCG tetraloop simulations used different MD force fields (ff) [13]. Three of ten ff were able to preserve the UUCG structure for 10×20 μs in standard MD. Most allowed excursions into different folds, but two ff failed completely for they quickly and irreversibly unfolded the hairpin.

Comparing the ensemble of structures from successful MD simulations with the NMR structures could provide an atomic description of the dynamics of the loop. Using enhanced sampling (RESP2) MD, Bottaro et al. [25] found two dominant structures of the UUCG tetraloop, starting from an extended 14-mer single strand. The predominant one was the consensus NMR structure of the loop. The second structure put C₃ and G₄ into solution and was identified by using eNOE data to reweight the ensemble from the MD simulations. Another integration of NMR with MD was particularly successful at producing an ensemble of structures for the CUUG tetraloop [23]. This tetraloop also has a well defined folded structure at low temperature, but loses a C:G base pair above 308 K. Simulations that were reweighted using NMR data provided an ensemble of loop structures that could correspond to conformational dynamics in solution. Bottaro et al. [25] and Reisser et al. [27] discuss the difficulties with comparisons of MD and NMR for RNA.

Often, the metric for success in an MD simulation is stabilizing the lowest free energy structure of the RNA. However, some simulations start from an unfolded or A-form single strand. The folding path is also of great interest, as this conformational change is difficult to measure in solution. Pathways could be compared to measurements, such as that of Sarkar et al., [28] who used temperature-jump fluorescence experiments to report cUUCGg folding. They found four states with three kinetic phases on the 1–5 μs timescale. The reversible folding pathways of CUUG, UUCG, and GCAA tetraloops were described in REMD simulations [29]. The two base pair stems formed in < 100 ps, after which the loop nucleotides sampled conformations (and energies). Viegas et al., [30] used energy landscape visualization to watch the folding of a gcGCAAgc tetraloop, reporting four transition states from unfolded to folded. The general conclusion is that folding of these unusually stable tetraloops is not a two-state process, and MD offers a chance to map the pathways.

Returning to the pre-Q1 riboswitch, it also contains a hairpin, which is the binding site for the pre-Q1 ligand. When it’s folded and bound to ligand, it forms a pseudoknot. It can fold with or without Mg²⁺ but the preQ1 binding mode differs. In the absence of Mg²⁺, ligand binding was necessary to effect the conformational change, while when Mg²⁺ was present, the riboswitch formed the pseudoknot without ligand [31]. The transition from hairpin and disordered 3’ tail to the pseudoknot was measured by single molecule FRET. Kinetics of docking (k_dock = 3.44 s⁻¹) and undocking (k_undock = 0.5–0.4 s⁻¹) give dynamic timescales of the conformational changes of 290 ms and 2 s, respectively. Sarkar et al. [32], using 2D fluorescence lifetime correlation spectroscopy, detected an initial μs folding/unfolding of the binding site. This conformational change of the pre-Q1 riboswitch appears to be best described as a 3-state transition with timescales μs to hundreds of ms. Since even enhanced sampling of MD is unlikely to capture the transitions of the preQ1 riboswitch, it would be an excellent choice for testing new coarse grain (CG) simulations that can incorporate nucleobase stacking and ribose puckering and electrostatics. A CG model for RNA, based on Martini 3, is currently benchmarked for dsRNA [33], but hopefully will be available soon for single strands and loops. A new CG model [34] has promise for RNA single strands and structures.

3-way junctions Combining secondary structure motifs to form larger structures is the hierarchical approach that RNAs use to make larger and more complex structures. As Figure 1 shows, joining two hairpins and adding a third stem gives a 3-way junction that can fold. The junction might start out as disordered, but it can serve as a site for binding ions (typically Mg²⁺) that collapse the structure, or specifically bind ligands. The collapse of a 3-way junction has been followed by single molecule FRET or modeled by NMR (ideally by both).

The purine riboswitch uses a 3-way junction in its aptamer to bind adenine or guanine [35]. There is a crystal structure of the bound state, but no structures of the free states. By introducing a Cy3/Cy5 FRET pair in the hairpin loops, Lemay et al., [36] used single molecule FRET (smFRET) to monitor the Bacillus pbuE adenine aptamer’s folding transition. The folded state forms via base pairing between the two loops, and is Mg²⁺-dependent. The stability of the folded state (the tertiary interaction) also depends on binding of adenine. Monitoring the duration of folded and unfolded states from the smFRET time traces +/− adenine clearly showed that the distribution of rates depended on the bound ligand. The conclusion is that the junction is flexible and sampling conformations when the adenine is absent. The rates of folding are on the order of k_fold = 0.7 s⁻¹ (1/(0.7 s⁻¹) = 1.3 s lifetime) with no ligand, and 1.5 s⁻¹ (0.67 s) with adenine; unfolding k_unfold = 1.3 s⁻¹ (0.77 s lifetime) with no adenine and 0.84 s⁻¹ (1.2 s) with adenine. These experiments revealed three states of the structure, however; there was very rapid interconversion (within the experimental 16 ms resolution) between (at least?) two conformations in a low Mg²⁺/no adenine state.

A subsequent study of the Vibrio vulnificus adenine riboswitch combined NMR and smFRET to study the folding transition of the aptamer (now with its 3′ expression domain) [37]. Without adenine, NMR exchange experiments found a folding rate of k_fold ~ 0.7 s⁻¹ (1.3 s lifetime) and the unfolding rate k_unfold ~ 2 s⁻¹ (0.5 s) The analogous smFRET experiments gave a more nuanced picture, identifying additional conformations of the aptamer in the absence of adenine. There was excellent agreement with NMR data in the transition rates from those experiments k_fold ~ 0.7 s⁻¹ (1.3 s) and k_unfold ~ 1 s⁻¹ (1 s). In order to allow the hairpin loops to make productive hydrogen bonding interactions, the junction must be sampling conformations that juxtapose the hairpins. There are few data on the structures and dynamics of the junctions.

The 23S/28S rRNA contains an intricately folded 3-way junction which is the binding site for the L11 protein. This 58 nucleotide GTPase center (GAC) is highly conserved, and seen in crystal structures of the ribosome as well as an independent crystal structure [38]. Within the tertiary fold is a triloop, a U-turn, and a T-loop held in place by base triples, chelated ions, and loop:loop interactions. Welty & Hall [39] used stopped-flow fluorescence to monitor tertiary structure formation, observing single 2-aminopurine sites. Starting from the secondary structure, folding was initiated by introduction of Mg²⁺, which resulted in an initial < 1 ms conformational transition. Subsequently, globally fitting the data from six sites around the RNA resulted in three transitions. The first global transition had a relaxation time of 18 – 30 ms (depending on Mg²⁺ concentration), the second ~1 s, and the third ~23 s at 20° C (100 mM KCl, 10 mM sodium cacodylate pH 6.5). The internal dynamics of the junction, the hairpin loops, and the internal bulge were not determined. However, NMR experiments [38] with the GAC + L11 protein showed that the protein stabilizes the folded conformation of the RNA, which otherwise is sampling conformations (at equilibrium) even in the presence of Mg²⁺. Modeling RNA folding events on these timescales will need new computational tools.

pUG RNAs.

Larger RNA molecules are full of hairpins, 3-way junctions and 4-way junctions that fold in contorted unpredictable ways (but there are attempts at 3D predictions for RNA Puzzles [40] and now using Alphafold 3). One of the most intriguing structures has been the pUG fold, a recent discovery that in C. elegans is essential for gene silencing but seems to potentially be present in almost all long RNAs [41].

Poly(UG) sequences are ubiquitous in eukaryotes, and often associated with pathology. The pUG fold is a left-handed G quadruplex with 12 (UG) repeats that requires K⁺ ions. There are stacked G quartets, with hydrogen bonding between the G’s while the U’s are flipped out and solvent exposed on the outside of the quadruplex (Figure 2). The K⁺ ions are in the middle of the quartets. Its structure was determined by both NMR [42] and crystallography [41]. Where this fold occurs in an RNA cannot be predicted [42], since the flanking sequences also contribute to its formation. A sequence variant (GA)₁₂ also adopts this fold but is less stable. pUG is included here as an example of dynamics on a different timescale [43], as well as a reminder that RNA structures are amazing.

In vitro transcripts of (UG)₁₂ to (UG)₁₈ were used in folding experiments [44]. Folding buffer contained 150 mM K⁺ at 25° C. The transition was followed by circular dichroism, as this structure has a distinctive spectrum. The RNAs fold with single exponential kinetics in these conditions, with a folding half-life of 17 minutes for (UG)₁₂, 13 minutes for (UG)₁₅, and 30 minutes for (UG)₁₈. Accessibility of guanosine imino protons to exchange with water found that the internal quartets had half-lives from 2.5 to 5 days, reflecting the slow opening of those quartets. The finding that (UG)₂₄ can also form the pUG fold was exciting, given that there are known RNA sequences (such as in the lncRNA NEAT1) that contain these long (UG) repeats. Quite intriguing, the p(UG)₂₄ was examined in 3 μs MD simulations (in GROMACS with AMBERΧOL3). Starting with two covalently linked p(UG)₁₂ structures with K⁺, they transitioned to a stable stacked structure after ~500 ns. That stacked structure is also seen in solution for the p(UG)₂₄ RNA.

SUMMARY. Not every RNA structure can be solved at atomic detail and certainly its conformational ensemble cannot be determined. NMR is an intensive method that can be applied to small RNAs (typically less than 100 nucleotides); its strength is that it can measure dynamics. X-ray crystals of small and large RNAs provide a structure, but unless there are several structural variants in the unit cell, they give no information on the ensemble and no glimpse of the dynamics. CryoEM of large RNAs has potential to provide a conformational ensemble of the RNA, since molecules in different states are trapped in solution/ice [47, 48, 49]. Single molecule FRET experiments can reveal timescales of molecular motions, with the potential for studying small RNAs and large ones, with a judicious choice of fluorophores. In an ideal ergotic world, an RNA in water with ions can be watched as it undergoes conformational changes in all-atom molecular dynamics simulations, and so its conformational ensemble can be determined. It is essential to incorporate enhanced sampling methods (REMD) to find the conformational ensemble. There are new methods to try (e.g. Viegas et al., [30], Röder [45], Love [46]) with a goal of producing an ensemble of structures that mimic the RNA in solution. Bench and computational experiments must be compared for progress in measuring and modeling the dynamics of RNA conformational changes.

• Any RNA molecule samples conformations in time and space

• The timescales of RNA dynamic motions span ps-ns-μs-ms-s-min-hrs-days

• Experiments in solution and in computations work together to define the ensemble