RNA Structure: Historical and Future Perspectives

1. RNA as an Organic Code

The seven-note musical alphabet played with variations in number, tempo, intensity, rhythm, pitch, or instrument has produced myriads of musical melodies over millennia. Similarly, the organic codes of nucleic acid, protein, polysaccharide, and lipid biomolecules [1-3] use relatively few symbols to generate vast information-storing combinations with context-dependent meanings, underlying the diversity and complexity of Earthly life [3-5]. Underscoring the analogy between the human-made music code and organic codes, sonification tools provide auditory displays of individual and collective biomolecule sequence information as an adjunct to visual and analytical bioinformatics tools [6-8].

The hypothesis of the emergence of overlapping organic codes heralding the living organisms (biotic) era is in line with the intricate interdependence between deoxyribonucleic acid (DNA), ribonucleic acid (RNA), proteins, carbohydrates, lipids, and metabolites in cell- or capsid (viruses)-based organisms [9-12].

The genetic code is the core of life [13], and DNA is its blueprint [14]. The 1800s saw crucial developments in this field. Justus Liebig reported an acidic material in a beef muscle filtrate. Friedrich Miescher discovered ‘nuclein’ in leukocyte nuclei as a protein-degradation-resistant, phosphorus-rich, natural living system’s chemical and chromosome structural component [15]. Richard Altmann coined the term nucleic acid [16].

By 1910, two kinds of nucleic acids were distinguished based on sources and isolation methods: the thymonucleic or zoonucleic acid, now termed DNA, from thymus or animals, and the phytonucleic acid, now termed RNA, from yeast and plants [17,18]. Later, both were found ubiquitously in living organisms. Levene and collaborators [19-21] identified the planar aromatic ring structures of DNA’s constituent purine (adenine [A] and guanine [G]) and pyrimidine (thymine [T] and cytosine [C]) nitrogenous bases, and that RNA has the pyrimidine uracil [U], instead of thymine, and pentose instead of hexose as carbohydrate.

In 1944, Oswald Avery [22] proposed that DNA is the genetic information carrier. In 1950, Erwin Chargaff [23] deciphered the consistent proportions of DNA’s constituent bases and the A-T and C-G base pairing rules. In 1953, Maurice Wilkins [24,25] and Rosalind Franklin with Raymond Gosling [26,27] conducted the X-ray diffraction and crystallography studies that led James Watson and Francis Crick to discover the double-helical structure of DNA in 1953 [28-30] as the foundation for the DNA theory of inheritance [31]. Each nucleotide interacts with water, ions, amino acids, small molecules, and every other nucleotide, stabilizing the structure [32].

The Watson-Crick right-handed helical B-DNA is the native form of DNA in cells. However, DNA’s helical structure and biological properties can vary transiently along short repetitive tracts, as in left-handed Z-DNA [33-36], or reversibly en masse, as in the transition between B- and A-DNA in microorganisms in extreme temperatures and pH [37,38]. Other higher-order variations include supercoils (double helix ends join in bacterial genomes), bubbles, hairpins and cruciforms (when palindromes are present), slipped loops, three-stranded triple helices (H-DNA), and tetrameric i-motifs (over 50,000 in the human genome) and related four-stranded G-quadruplexes [39-41]. Tertiary DNA structures vary from person to person in critical genes like the insulin gene, constituting therapeutic targets [42].

First proposed by Mitsui et al in 1970 [43] and later proven by Wang AH et al [33], Z-DNA is a left-handed helix in equilibrium with the lower energy right-handed B-DNA. Flipons, typically involving an alternating purine/pyrimidine motif, can flip between B- and Z-DNA conformations under physiological conditions aided by binding proteins, introducing diversity to transcriptomes, particularly in immunity and transcription functions [35,36,44].

Discovered by Franklin and Gosling in 1953 [45] in DNA crystals after dehydration, A-DNA, also derived from protein binding to DNA, is a right-handed double helix but with a shorter and more compact helical structure than B-DNA, resulting in slightly more base pairs per turn, a smaller twist angle, and a shorter rise per base pair. The major groove of A-DNA is deep and narrow, the minor groove is wide and shallow, and the base pairs are not perpendicular to the helix-axis as in B-DNA. A-DNA can occur in DNA-RNA hybrid double helices and double-stranded RNAs. RNA can only form an A-type double helix because of the steric restrictions imposed on ribose by the 2’ hydroxyl residue [40].

After Z-DNA was discovered and named after its sugar-phosphate backbone’s zig-zag course as an alternative to the more common Watson-Crick B-DNA, nuclear magnetic resonance, and other studies showed that the common A-RNAs, particularly those with higher Guanine/Cytosine content, could similarly undergo the right-to-left-handed conformational change to the higher energy Z-RNA [44,46-48]. Z-binding proteins specifically recognize and bind Z-DNA [49-51] and Z-RNA [52,53]. Z-DNA and Z-RNA encoded by flipons under physiological conditions are implicated in various biological processes, including transcription and immunity [44]. Z-RNA has been studied less than Z-DNA and both are challenging to detect in vivo.

Structural similarities between RNA and DNA allow the formation of RNA-DNA hybrids, such as the R loops, which also include a displaced single-stranded DNA [54]. Antisense noncoding RNAs may form R lops. R loops accumulate throughout the genome in pericentromeric DNA, telomeres, ribosomal DNA, or transcription termination regions, among others, and are involved in transcription and chromatin structure. Because they can also adversely affect genome stability and replication, several DNA and RNA metabolism factors, such as ribonucleases, RNA-DNA helicases, RNA processing factors, and topoisomerase I, degrade R-loops or prevent their formation [54].

As an example of a protein that interacts with DNA and RNA, the topoisomerase I enzyme prevents genomic instability by alleviating DNA torsional strain. Topoisomerase I introduces transient single-strand breaks that prevent the accumulation of supercoiling and torsional stress, which could otherwise lead to damage and instability of DNA, and cell death [55]. Interactions between RNA and Topoisomerase I regulate DNA during transcription by modulating Topoisomerase I-mediated relaxation. In cancer cells, for instance, DNA transcription is often elevated, necessitating increased levels of Topoisomerase I activity to relax the DNA and maintain proper gene expression. RNA opposes Topoisomerase I activity. Inhibiting RNA binding of Topoisomerase I may work similarly to antineoplastic Topoisomerase I inhibitors like camptothecin by increasing Topoisomerase I catalytic complexes on DNA [55].

Beyond cancer, dysfunction of R loop-interacting factors in several genetic diseases leads to replication stress, genome instability, chromatin alterations, or gene silencing [54]. Furthermore, many chromatin-associated complexes, including histone modifiers, transcription factors, and DNA methyltransferase, interact with RNA [56]. RNA can also promote the repair of double-strand breaks in DNA, by helping position and holding the broken DNA ends in place and guiding the cellular repair machinery, thereby contributing to genome integrity [57].

2. RNA Has Many Functions Through Its Intricate, Ubiquitous, Diverse, And Dynamic Structure

RNA has emerged as a central biomolecule in the multidirectional flow of genetic information for phenotype and biological diversity generation [12,58]. RNA is no longer considered simply an intermediary between the data stored in DNA and that functionally displayed in proteins. Although about 70–90% of the human [59] and 85–90% of the yeast genome [60] is transcribed into RNA, much remains unknown about RNA functions in cells [61].

RNA structure plays critical roles in every step of RNA’s lifecycle, including transcription, splicing, localization [62,63], translation [64,65], and RNA decay [66]. However, RNA structure differs among individual cells and provides an additional layer of information in defining cellular identities by, for instance, informing RNA-binding protein binding and gene regulation [67]. To this end, overall RNA structure profiles better discriminate cell type identity and differentiation stage than gene expression profiles alone. For instance, RNA structure is more homogeneous in human embryonic stem cells than differentiating neurons, with the greatest homogeneity found in coding regions. More extensive heterogeneity is found within 3’ untranslated regions and is determined by specific RNA-binding proteins. Moreover, the cell-type variable region of 18S ribosomal RNA is associated with cell cycle and translation control. It is therefore important to systematically characterize RNA structure-function relationships at single-cell resolution using approaches such as single-cell structure probing of RNA transcripts [67].

RNA accomplishes its many functions through various structural levels beyond its primary and secondary structural ones, which are defined by nucleotide sequence and Watson-Crick pairing-based folding, respectively [68-71]. Along with the staggering number of noncoding RNA genes, RNA structural versatility underlies biological diversity from the organismal to population levels.

RNA’s bases closely stack on each other like ‘coins in a roll’ via noncovalent interactions, exposing their charged exocyclic groups to water molecules and ions, underlying RNA’s solvability and helical conformations unrelated to Watson-Crick pairing [72]. RNA’s conformation also varies with environmental changes, liquid-liquid phase separation, or interactions with other biomolecules[73-75].

Underlying its compactness, RNA intrinsically tends to form A-U, G-C, and G-U Watson-Crick base pairs in short and long-range structures, higher-order architectures, and RNA-RNA interactions in picoseconds to seconds [76-79], which are fundamental to its diverse functions [80]. As many as 40% of the nucleotides of an RNA molecule can be part of hairpins and multi-helix junction loops [81], and 30% to 40% of RNA duplexes in living cells involve sequences over 200 nucleotides apart [80].

3. RNA's Structure Is Defined at Primary, Secondary, Tertiary, Quaternary, And Quinary Levels

3.1. Primary, Secondary, and Tertiary RNA STRUCTURES

The primary structure is the RNA’s linear nucleotide sequence (Figure 1). The secondary structure describes the paired and unpaired elements of stems, loops, and bulges that form as the single-stranded RNA molecule folds back on itself via Watson-Crick pairs and interacts via hydrogen bonding and stacking as soon as it is synthesized [74,78] (Figure 1).

The tertiary or 3D structure, which typically compacts the RNA, is achieved by longer-distance Watson-Crick and non-Watson-Crick interactions of elements within the preformed secondary structures [70] (Figure 1). These interactions give rise to structural elements, including pseudoknots, which lock together two stem-loops by base pairing and sugar-phosphate interactions, often in a so-called kissing interaction.

Many RNA loops are characterized by structural modules with highly organized networks of noncanonical interactions comprising ordered non-Watson-Crick base pairs embedded between Watson-Crick base pairs [82]. Non-Watson–Crick pairs are key for folding and binding to proteins or other ligands [83-85].

RNA three-dimensional (3D) motifs occupy places in structured RNA molecules corresponding to the hairpin, internal, and multi-helix junction loops of their 2D structure representations [81]. These 3D structural RNA modules, with specific loop geometries, contribute to structural stability, have central roles as architectural organizers of catalytic activity and ligand binding sites in RNA molecules, and are recurrently observed in RNA families throughout phylogeny [81,86-91].

Among RNA 3D motifs are pseudoknots, which are minimally composed of two helical segments connected by single-stranded regions or loops (Figure 2). Pseudoknots form the catalytic core of various ribozymes, self-splicing introns, and telomerase, and alter gene expression by inducing ribosomal frameshifting in many viruses (reviewed in [92]). The best characterized is the H-type pseudoknot (Figure 2).

Guanine-rich regions in RNA and DNA can form noncanonical G-quadruplex structures encompassing stacked guanine tetrads, a square planar structure formed by four guanine residues [93] (Figure 3). RNA G-quadruplexes participate in translation, splicing, RNA stability, and cellular stress responses, among other functions mediated by the RNA binding proteins with which they interact [93].

3.2. Quaternary and Quinary RNA Structures

Similar to the interactions of DNA and histones, RNA’s quaternary structures result from a folded RNA's interaction with other biopolymers, such as proteins and RNAs. The quinary structure of RNA results from its weak and nonspecific interaction with cellular metabolites, such as osmolytes, accumulated by cells in response to osmotic stress [75]. Understanding the effects of osmolytes on RNA's tertiary structure, whether stabilizing or destabilizing, is crucial to comprehend the intricacies of RNA [94]. For instance, as hydrated magnesium ions neutralize a notable fraction of the negative charge of an RNA tertiary structure, the RNA becomes less responsive to stabilizing osmolytes and may even be destabilized [94].

4. Determining RNA Tertiary and Beyond Structures Remains Challenging

The complex biological functions of RNA molecules are underpinned by their specific sustained 3D structures, with or without the help of proteins or other RNAs in multimolecular complexes [95]. However, the study of RNA 3D structure is often hindered by the scarcity of atomic coordinates, a significant challenge in the field. These determinations are typically low-resolution or miss atoms due to the limitations of the low-throughput and costly structure determination methods, i.e., X-ray crystallography, nuclear magnetic resonance, and cryo-electron microscopy [96], which also creates a significant gap between the number of RNAs sequenced and the number of structures defined. Moreover, RNA’s shifting into diverse forms according to environmental conditions renders structural studies challenging. Traditional imaging methods, such as cryo-electron microscopy single-particle averaging analysis, rely on averaging data from thousands of selected molecules with common shapes, making it difficult to capture the unique shapes of individual RNA molecules.

Developed during the last two decades [97], some RNA 3D structure prediction computational tools use high-resolution homologs' more precise structural information to annotate the base-pairing interactions in low-resolution structures in coarse-grained models/simulations [98,99] or in imaging data missing atoms [100]. Moreover, a machine-learning approach identifies accurate structural models without assumptions about their defining characteristics despite being trained with the atomic coordinates of only 18 known RNA structures [101].

Structural imaging studies are complemented by gel or capillary electrophoresis based on in-line probing, i.e., structural sensitivity to spontaneous degradation, nucleases targeting either single- or double-stranded regions, or chemical probes, such as dimethyl sulfoxide (DMS). DMS, for instance, is used to probe unpaired adenines and cytidines, and 1- metho-p-toluenesulfonate (CMCT) to probe unpaired uridines in chemical inference of RNA structures sequencing (CRIS-seq) [102-104]. DMS is also used in RNA structure sequencing (Structure-seq and STRucture-seq2) [105-108], dimethyl sulfate-modified RNA sequencing (DMS-seq) [109], dimethyl sulfate mutational profiling with sequencing (DMS-MaPseq) [110], and transfer RNA structure sequencing (tRNA structure-seq) [111]. Pyrdiostatin, the chemical probe in RNA GQ sequencing (rG4-seq) [112], and selective 2’-hydroxyl acylation analyzed by primer extension (SHAPE) [113-118] have also extended chemical probing to the entire transcriptome [70,119-122].

Even if RNA structures are accurately determined, they may not represent the one(s) relevant in vivo. Many factors influence RNA structure in the living cell, including variations in organelle environments and interactions with proteins or other macromolecules, which render the elucidation of RNA structure in vivo particularly challenging. For instance, in silico modeling provides the most thermodynamically stable structure of an RNA sequence, while RNAs can become trapped in vivo in alternative structures [70,123]. Moreover, processing the low abundance, long nascent, or precursor RNAs, including splicing and polyadenylation, entails pathway networks that determine mature isoform composition and control gene expression, further adding to the complexity of studying RNA structure [124].

The success of Alphafold [125,126] in predicting protein 3D structures has not yet extended to RNA [127]. This is due to differences in building blocks (amino acids vs. nucleotides), diversity of sequence range (up to tens of thousands of nucleotides for RNA vs. a few hundred amino acids for proteins), number of available structure data (orders of magnitude greater for proteins), and folding stability (multiple conformations for RNA vs. usually one for proteins) [97].

Readily available RNA 3D structural prediction tools often rely on the primary sequence and canonical 2D structures formed by A-U, G-C, and G-U Watson-Crick pairs to detect structural RNA modules from primary sequence data and identify recurrent interaction networks [128-133]. Several databases contain RNA structural information [91,134-136].

During the last decade, computational RNA structure predictions have evolved from the earliest thermodynamic and molecular dynamic-based approaches to deep learning-based conformation approaches [137]. Earlier deep learning models for RNA structure have been competitive but not consistently better than traditional 3D structure prediction methods, including ab initio physics-based methods using various levels of granularity in nucleotide representation, template-based methods that try to map sequences to structural motifs before merging them into a whole structure, or hybrid methods, combining ab initio and template-based methods [97,138]. However, platforms such as the RNA3DB dataset [139], which arranges the RNA 3D chains into distinct non-redundant groups (Components), and Dfold, which combines an autoregressive Deep Generative Model, Monte Carlo Tree Search, and a scoring model [140] have been developed to improve RNA 3D structure prediction.

An innovative technique to study the 3D structure of individual molecules without averaging builds on advanced Individual-Particle cryo-Electron Tomography (IPET) to focus on single-molecule 3D imaging in cryopreserved samples. IPET captures a snapshot of RNA's folding landscape by capturing molecules in various stages of folding, from immature states to their optimal shape. This approach may allow engineering the folding of more effective RNA vaccines and dynamic sensors for molecular medicine [141].

5. Conclusions

Since the central molecular biology dogma was formulated almost seven decades ago, mounting evidence revealed that RNA is the main determinant of biological diversity. This is driven by RNA’s abundance, modifications, and structural versatility of its coding and noncoding versions, which occasionally overlap. RNA structure-function relationships also vary among cells in an organism, determining cellular identity. However, understanding RNA’s tertiary, quaternary, and quinary structures and their functional relationships, remains challenging. Advances in approaches including the use of machine learning and artificial intelligence [101] and data gatherings, such as the United States National Institutes of Health’s and National Academies of Sciences’ RNA sequencing initiative, provide an optimistic outlook for the continued development and refinement of disruptive RNA-based approaches for medical therapy, diagnosis, and prevention, and agriculture and industrial applications [142]. To this end, novel imaging methods enable detailed RNA analysis within single cells and intact tissues [143].