SHAPE Directed Discovery of New Functions in Large RNAs

Conspectus

RNA lies upstream of nearly all biology and functions as the central conduit of information exchange in all cells. RNA molecules encode information both in their primary sequences and in complex structures that form when an RNA folds back on itself. From the time of discovery of messenger RNA in the late 1950s until quite recently, we had only a rudimentary understanding of RNA structure across vast regions of most messenger and non-coding RNAs. This deficit is now rapidly being addressed, especially by SHAPE chemistry, mutational profiling (MaP), and closely related platform technologies that, collectively, create chemical microscopes for RNA. These technologies make it possible to interrogate RNA structure, quantitatively and at nucleotide resolution and at large scales, for entire mRNAs, non-coding RNAs, and viral RNA genomes. By applying comprehensive structure probing to diverse problems, we and others are showing that control of biological function mediated by RNA structure is ubiquitous across prokaryotic and eukaryotic organisms. Work over the past decade using SHAPE-based analyses has clarified key principles. First, the method of RNA structure probing matters. SHAPE-MaP, with its direct and one-step readout that probes nearly every nucleotide by reaction at the 2’-hydroxyl, gives a more detailed and accurate readout than alternatives. Second, comprehensive chemical probing is essential. Focusing on fragments of large RNAs or using meta-gene or statistical analyses to compensate for sparse datasets misses critical features and often yields structure models with poor predictive power. Finally, every RNA has its own internal structural personality. There are myriad ways in which RNA structure modulates sequence accessibility, protein binding, translation, splice-site choice, phase separation, and other fundamental biological processes. In essentially every instance where we have applied rigorous and quantitative SHAPE technologies to study RNA structure-function interrelationships, new insights regarding biological regulatory mechanisms have emerged. RNA elements with more complex higher-order structures appear more likely to contain high-information-content clefts and pockets that bind small molecules, broadly informing a vigorous field of RNA-targeted drug discovery. The broad implications of this collective work are two-fold. First, it is long past time to abandon depiction of large RNAs as simple noodle-like or gently flowing molecules. Instead, we need to emphasize that nearly all RNAs are punctuated with distinctive internal structures, a subset of which modulate function in profound ways. Second, structure probing should be an integral component of any effort that seeks to understand the functional nexuses and biological roles of large RNAs.

Key References

Siegfried, N. A.; Busan, S.; Rice, G. M.; Nelson, J. A. E.; Weeks, K. M. RNA Motif Discovery by SHAPE and Mutational Profiling (SHAPE-MaP). Nature Methods 2014, 11, 959–965.¹

Introduces mutational profiling (MaP) for one-step detection of chemical adducts in RNA, the low SHAPE-low Shannon entropy metric for discovery of potential regulatory motifs, and emphasizes the now-ubiquitous result that large RNAs are punctuated by extensive internal structure.

Mustoe, A. M.; Busan, S.; Rice, G. M.; Hajdin, C. E.; Peterson, B. K.; Ruda, V. M.; Kubica, N.; Nutiu, R.; Baryza, J. L.; Weeks, K. M. Pervasive Regulatory Functions of mRNA Structure Revealed by High-Resolution SHAPE Probing. Cell 2018, 173, 181–195.²

The first large-scale RNA structure probing study to focus specifically on transcripts with comprehensive, high-quality, per-nucleotide probing data. RNA structure is shown to be a pervasive regulator of gene expression, effectively comprising a distinct level of the genetic code.

Dethoff, E. A.; Boerneke, M. A.; Gokhale, N. S.; Muhire, B. M.; Martin, D. P.; Sacco, M. T.; McFadden, M. J.; Weinstein, J. B.; Messer, W. B.; Horner, S. M.; Weeks, K. M. Pervasive Tertiary Structure in the Dengue Virus RNA Genome. Proc Natl Acad Sci U S A 2018, 115, 11513– 11518.³

Provides a glimpse into the future emphasizing that large RNAs form (of course!) extensive internal base paired structures. But, distinctively, a subset of these structures also fold to create more complex higher-order structures that govern overall RNA compaction and function.

The Ubiquity and Fundamentalness of RNA Structure

As soon as they are synthesized, cellular and viral RNAs fold back on themselves to form (typically extensive) internal base pairs and (rarer) higher-order interactions. These structures may have intrinsic function, bind proteins, or occlude functional structures and thereby be switch-like (Figure 1). mRNAs, long non-coding RNAs (lncRNA), and viral RNAs therefore encode biological information both in sequence and in structure^4–10.

Figure 1.

Architecture and potential internal structures of RNA molecules. Important classes of motifs are shown, a subset of which might occur in any given RNA.

Until recently, rigorous methods for comprehensive examination of the long-range, in-solution structure of RNA molecules did not exist. First, it was difficult to measure RNA structure with nucleotide resolution and in a biophysically quantitative way. Second, it was difficult to model RNA structure accurately, especially for long RNAs. Third, many data-directed RNA structure models yield so many plausible structures that it was unclear where to look to identify the most impactful biology.

Advances in chemical probing have yielded powerful approaches for identifying, modeling, and characterizing the functional roles of RNA structure. Chemical probing is a venerable method¹¹ that has been revolutionized by melding new chemistries, readouts by massively parallel sequencing, and innovative data analysis strategies. Especially impactful have been selective 2’-hydroxyl acylation analyzed by primer extension (SHAPE)^12–14, mutational profiling (MaP) (including both SHAPE-MaP^1,15 and DMS-MaP^16–19 approaches), and pseudo-free energy change strategies for data-driven secondary structure modeling^20,21. Secondary structure landscapes can be now profiled for large RNAs efficiently and for many RNAs simultaneously in living cells. SHAPE and closely related technologies are now used in innovative ways by research labs and biotechnology companies worldwide^22–31.

In this communication, we will accept that SHAPE-based methods work as advertised and focus on examples from our lab emphasizing how comprehensive, per-nucleotide examination of RNA structure can identify functional elements in complex RNAs and define mechanisms by which RNA structure governs biological function. This overview emphasizes the importance of incorporating structural information into all efforts to understand RNA biology and provides a jumping off point for further investigations.

SHAPE Chemistry and Mechanism

SHAPE chemistry interrogates free RNA 2’-hydroxyl groups, which nearly all nucleotides contain. SHAPE probing involves two basic steps: (1) treatment of an RNA with a hydroxyl-selective reagent and (2) read-out of the pattern of chemical modification in the RNA. From the earliest experiments^12,13, it was clear that SHAPE measures local nucleotide flexibility (Figure 2A). Analyses of the geometric conformations of nucleotides with high reactivities, as visualized crystallographically, subsequently revealed that SHAPE reactivity at 2’-hydroxyl groups is facilitated by intramolecular general base catalysis and by specific conformations adopted by the proximal 3’-phosphodiester group³² (Figure 2B). SHAPE thus measures local nucleotide flexibility because conformationally dynamic nucleotides transiently adopt states that facilitate reactivity.

Figure 2.

Mechanism of RNA SHAPE chemistry. (A) Reaction scheme. (B) Nucleotide conformations that enhance SHAPE reactivity involve (top) general base catalysis by the pyrimidine O2 (or purine N3), pro-S oxygen, or through-space groups and (bottom) stabilization of the 2’-oxyanion via conformations that direct nonbridging oxygen groups away from the 2’-OH group. (C) Correlation between the model-free generalized order parameter, S², and SHAPE reactivity, exemplified by the U1A-binding RNA element³³. (D) SHAPE reactivities superimposed on the TPP riboswitch aptamer domain showing (left) absolute SHAPE reactivities and (right) changes in SHAPE reactivities (and thus local nucleotide dynamics) upon ligand binding (gray). (E) Classes of SHAPE reagents.

Local nucleotide flexibility, as measured by SHAPE, correlates well with a fundamental biophysical measurement, the model-free generalized order parameter³³, S², which varies from 0.0 (disordered) to 1.0 (fully ordered). The correlation between S² and SHAPE emphasizes that SHAPE is a biophysical measurement of spatial disorder at single-nucleotide resolution (Figure 2C). SHAPE distinguishes well-structured from conformationally dynamic regions of an RNA and detects changes in RNA structure as a function of perturbants, including binding by small molecules^29,34–36 (Figure 2D). The biophysically rigorous relationship between SHAPE and local RNA structure becomes impressively meaningful because SHAPE can interrogate thousands of nucleotides at a time, including in living cells, for a significant fraction of a cellular transcriptome.

Three broad classes of SHAPE reagents have been developed based on isatoic anhydride^12,14,37, imidazolide^38,39, and benzoyl cyanide⁴⁰ scaffolds (Figure 2E). The most important considerations for selecting a SHAPE reagent are half-life and solubility. Among the most useful reagents are 1M7 for fast, quantitative probing of RNAs under simplified conditions, in viruses, and in bacterial cells¹⁴; 5NIA, NAI, and 2A3 for probing in eukaryotic cells^37–39, and benzoyl cyanide for fast, time-resolved experiments^40–42. Differential reactivities between reagents sharing the isatoic anhydride scaffold can be used to “fingerprint” nucleotides that show distinct one-sided stacking patterns or that form the rarer C2’-endo ribose pucker and undergo slow conformational changes^35,41,43.

The Mutational Profiling (MaP) Readout

Massively parallel sequencing⁴⁴ has transformed characterization of the genome, RNA and DNA modifications and protein interactions, and (of course!) RNA structure^15,25. Sequencing-detected genomics technologies involve cleavage, covalent modification, or enrichment of an RNA or DNA and then acquiring thousands to millions of “measurements” by sequencing. There are significant challenges to sequencing-based experiments, notably their complexity, the inability to validate but a small fraction of the results, and, usually, the lack of pre-genomics ground-truth reference experiments.

Within this context, RNA structure probing occupies a special niche because RNA chemical probing experiments conducted over the past 40 years have been read out by well-validated methods^11,20,45,46 and RNAs with known structures can be used to evaluate new methods. In our work, it has proven invaluable to create a continuous trail of validation from simple biochemical to whole-transcript experiments.

SHAPE chemical adducts have been read out by several generations of technology. SHAPE 2’-O-adducts induce stops to primer extension by conventional reverse transcription, which converts an RNA to a complementary DNA (cDNA) such that the 3’-end of the cDNA indicates the site of a chemical adduct. Such “STOP-RT” SHAPE adducts were initially detected by gel electrophoresis and then by capillary electrophoresis^20,47,48, which ultimately resolved RNA structures over a few thousand nucleotides^20,47 including in cells^47,49,50.

To read out SHAPE data by massively parallel sequencing, we initially (like many other laboratories), used the STOP-RT approach and prepared libraries for sequencing by the now-standard approach of ligating DNA adapters on to the initial cDNAs. However, our lab immediately noticed that these data correlated poorly with prior well-validated capillary electrophoresis methods, a feature now widely recognized^51–54. The multiple complex biochemical steps required for library preparation, after the STOP-RT step, bias the final reactivity profiles.

To avoid throwing away the biophysical rigor of SHAPE technologies (Figure 2) upon integration with massively parallel sequencing, we invented the MaP strategy^1,15. MaP leverages relaxed fidelity reverse transcription, whereby the polymerase reads through chemical adducts in an RNA, resulting in a mutation or short deletion. cDNAs are aligned and sites of mutation are counted to yield per-nucleotide reactivity profiles (Figure 3A). The MaP strategy has several impactful features: (1) Chemical adducts are recorded in a single, direct step¹. (2) Sites of chemical adducts are recorded internally in the cDNA and thus can be amplified without biasing internal structural information. (3) Almost any sequencing library preparation approach can be used, without biasing the MaP detection step. (4) MaP readouts correlate strongly with validated capillary electrophoresis readouts^15,55. (5) Each per-nucleotide measurement originates from multiple discrete mutation events, thus defining a standard error for each measurement. And (6) multiple chemical adducts can be detected in single RNA molecules, which laid the foundation for a new category of technologies involving single-molecule correlated chemical probing^{3,16,17,19,56–59} (Figure 3A, bottom). The term MaP emphasizes the independence of the strategy from library sequencing (-seq) steps; in this view, names that link MaP and seq are oxymorons. This focus on maintaining a quantitative relationship between probing and readout has motivated diverse research groups to adopt MaP-based strategies. For example, most work investigating SARS-CoV-2 RNA genome structure at nucleotide resolution has used SHAPE-MaP or related methods^60–65.

Figure 3.

Overview of SHAPE, MaP, and RNA structure modeling. (A) Mutational profiling. RNA is treated with a reagent that reports secondary or tertiary structure; relaxed fidelity reverse transcription records chemical adducts as mutations relative to the original sequence (in red) internally in the cDNA; cDNAs are sequenced; and reads are aligned and used to create reactivity profiles. Data may be interpreted on a per-nucleotide basis or as through-space internucleotide correlations. (B) The ∆G_SHAPE pseudo-free energy change relationship that enables SHAPE-directed structure modeling. (C) Per-nucleotide reactivity profile for a domain of the STMV RNA genome^56,91. (D, E) Secondary structure models (based on data in panel C) shown as (D) probability arc plots and (E) minimum free energy secondary structure diagram. In probability arcs and lines connecting base pairs, colors indicate the likelihood of unique pairing for a given nucleotide. (F) Representative secondary structure modeling, showing the SAM-I riboswitch²¹ (left) without and (right) with SHAPE data. SHAPE-directed modeling often yields dramatic improvements, including for RNAs containing pseudoknots. Arcs indicate correct (green), incorrect (purple), and missing (red) pairs, relative to accepted structure.

SHAPE-Directed RNA Structure Modeling

SHAPE data are approximately inversely correlated with the probability that a nucleotide forms a base pair. As the logarithm of a probability corresponds to an energy, SHAPE data can be converted into a pseudo-free-energy change term, ∆G_SHAPE, and used to modify the free energy terms in nearest-neighbor RNA folding algorithms^20,21 (Figure 3B). There are two parameters in ∆G_SHAPE, the slope, m, is the penalty for base pairing that increases with SHAPE reactivity; the b term is negative and reflects a favorable term for pairing at nucleotides with low SHAPE reactivities. SHAPE information (Figure 3C) is used to create data-directed structure models that can be visualized in several ways. The most comprehensive is as probability arcs^15,66–68 (Figure 3D), which illustrate the structural complexity of an RNA ensemble. Alternatively, the most probable, minimum free energy structure can be shown in a base-pairing diagram, which simplifies the underlying complexity of an RNA (Figure 3E). SHAPE-directed secondary structure modeling achieves good-to-excellent agreement with accepted structures for diverse RNAs^1,20 including RNAs containing pseudoknots²¹ (Figure 3F).

Finding Functional Motifs: The Low SHAPE-Low Shannon Entropy Metric

Not all structures adopted by long RNAs are well-defined or functionally important. We have found that regions with both low SHAPE reactivities and low Shannon entropy (lowSS) – the former reflecting constrained nucleotides and a highly structured RNA region and the latter indicative of a well-defined structure – are much more likely to be functional (Figure 4, center box). Shannon entropy is calculated from the probability of formation of each base pair across all possible structures in the ensemble compatible with the SHAPE data^1,69.

Figure 4.

Identification of well-determined RNA structures. (box) Equation for Shannon entropy (H) and illustration of overlap of low SHAPE and low Shannon entropy (lowSS) regions. (A) Representative analysis illustrating that functional RNA elements tend to overlap with lowSS regions (blue shading), here corresponding to the 5’-UTR and frameshift elements of HIV-1. (B and C) lowSS analysis of (B) native-like and refolded dengue RNA genomes⁷² and (C) human rRNAs⁷⁴. (D) Illustration of the enormous diversity in lowSS regions, extending from the TPP riboswitch (79 nts, 100% lowSS) to the HIV-1 RNA genome (9,173 nts, 40% lowSS), and including bacterial and human rRNAs^15,74. All RNAs shown on the same length scale.

The usefulness of the lowSS metric first became clear in studies of the HIV-1 genome (Figure 4A). Most known functional motifs were found in lowSS regions, and the metric was predictive of novel, and validatable, pseudoknot-containing elements¹. Functional motifs have since been shown to be overrepresented among lowSS regions in RNA viruses^1,3,61,70, lncRNAs⁷¹, and mRNAs². The lowSS metric also allows de novo identification of well-folded and recapitulatable structural elements in synthetic or refolded RNAs studied under simplified conditions. For example, refolding the dengue virus RNA genome (~10 kb nucleotides) yielded an RNA that was more highly structured and sampled fewer conformations than the more native-like RNA extracted from virions⁷². The lowSS metric identified the subset of elements that folded similarly in both states (Figure 4B).

Bacterial^1,20,21 and archaeal⁷³ ribosomal RNAs (rRNAs) have been widely used to validate SHAPE-directed RNA structure modeling. In strong contrast, when we obtained SHAPE profiles for the human 18S and 28S rRNAs, as gently extracted from cells, numerous regions were incompatible with accepted structures⁷⁴. This deficit was not a limitation of SHAPE-directed modeling itself, as many motifs could be modelled accurately using idealized reactivities (Figure 4C). Instead, only about 40% of human rRNAs formed well-determined secondary structures. Prior studies using STOP-RT readouts had also obtained structure probing data for eukaryotic rRNAs, but the discrepancy between modeled and accepted structures was not noted, likely due to the non-quantitative nature of STOP-RT methods. Eukaryotic rRNAs thus have less well-determined structures than prokaryotic rRNAs and are extensively unfolded when gently extracted from cells (Figure 4C, 4D).

The lowSS criterion reveals extensive differences between and across RNAs. The base-paired secondary structures of the bacterial TPP riboswitch and human U1 snRNA are extremely well determined^15,74 (Figure 4D). Bacterial rRNAs similarly contain large expanses of lowSS regions, whereas human rRNAs contain large regions of high-entropy regions^15,74 (Figure 4D). Large RNAs like the HIV-1 RNA genome¹ and the Xist noncoding RNA⁷¹ have heterogeneous profiles, with high-entropy regions punctuated by regions of well-determined structure (Figure 4D). The extent of lowSS regions is one way every cellular RNA has its own “structural personality”.

The Importance of Comprehensiveness

A dominant lesson from large-scale RNA structure analysis is that RNA molecules should be studied comprehensively. For example, early models for the CAG repeat sequence in the HTT mRNA, implicated in Huntington’s disease, and for the translational frameshift element in the HIV-1 genome emphasized simple helices that form when short RNAs spanning these regions are studied^75,76 (Figure 5A, B). In the context of native flanking sequences, however, much more complex structures are observed: For the HTT mRNA, short CAG repeats primarily pair with flanking poly-CCN sequences; whereas, only long, disease-associated repeats form CAG stem-loop structures⁷⁷ (Figure 5A). For the HIV-1 frameshift element, SHAPE-based analyses of full-length genome transcripts identified a second frameshifting sequence and revealed the functional importance of RNA helices that only form in long transcripts^78,79 (Figure 5B). These complex structures may support selective binding by small-molecule ligands⁸⁰. These results emphasize the importance of analyzing RNA structure in the context of full-length RNAs.

Figure 5.

Comprehensiveness in understanding RNA structure and RNA-protein interactions. (A) Secondary structure models for CAG repeat-containing sequences from human HTT mRNA based on analysis of (left) local structures and (right) the full-length first exon (360–500 nts, depending on number of CAG repeats)⁷⁷. Start codon is boxed in green. (B) Secondary structure models of the HIV-1 RNA frameshift element based on analysis of (left) local structure and (right) the entire HIV-1 genome⁷⁸. (C) Time-resolved folding of a retroviral RNA packaging domain in the absence and presence of nucleocapsid chaperone⁸³. Per-nucleotide reactivities are shown on a scale from red (high) to black (low). Nucleotides are grouped by k-means clustering; kinetic profiles are shown for each cluster. (D) SHAPE-detected protein-RNA contacts 7 seconds after protein addition reveals interactions at guanosine (blue). (E) Model for chaperone-mediated facilitation of RNA folding by destabilizing interactions involving guanosine⁸³.

Another class of comprehensiveness is the ability to interrogate all four ribonucleotides. Folding of many large RNAs is facilitated by proteins called RNA chaperones that accelerate formation of RNA-RNA interactions, catalyze conformational changes and acquisition of unique functional structures, and act over long sequence distances^81,82. We used time-resolved SHAPE – employing the fast reacting benzoyl cyanide reagent⁴⁰ (Figure 2E; half-life 0.25 seconds) – to interrogate the structure of a retroviral RNA element that undergoes a complex dimerization reaction essential for correct packaging of two RNA genomes in a virion⁸³. The RNA-alone dimerization reaction is slow, showing seven distinct kinetic behaviors in four rate processes. In the presence of the viral nucleocapsid chaperone protein, dimerization occurs rapidly, in a single step (Figure 5C). The chaperone specifically destabilizes RNA interactions involving guanosine (Figure 5D). Guanosine forms highly stable base pairs with cytidine and promiscuously pairs with cytidine and uridine, features that likely create a complex RNA folding landscape. Our SHAPE-directed model emphasizes that the nucleocapsid RNA chaperone acts over large distances via a simple mechanism: weakening interactions involving guanosine⁸³ (Figure 5E).

Effects of the Cellular Environment on RNA Structure

The cellular environment is vastly different from conditions in test-tube experiments. Cells are crowded and contain numerous classes of RNA binding proteins. Diverse processes including translation, splicing, transport, and macromolecular assembly are likely to affect RNA structure. When we compared the structure of the aptamer domain of the adenine riboswitch under simplified test-tube conditions versus in Escherichia coli cells, our SHAPE analysis revealed that the riboswitch RNA is much more highly structured in cells than under simplified conditions⁵⁰ (Figure 6A). Even addition of Mg²⁺ to 30 mM in test tube experiments, a highly stabilizing condition for RNA, does not yield an RNA that is as structured as in cells. Molecular crowding agents, like PEG, stabilized long-range loop-loop interactions but, overall, failed to stabilize the structure of the riboswitch as well as the in-cell state⁸⁴. In cells, ligand binding only altered SHAPE reactivities of nucleotides close to the ligand pocket, indicating that the RNA is essentially fully folded in cells (Figure 6A). The intracellular environment thus has a large effect on RNA structure that is difficult to replicate in test-tube experiments.

Figure 6.

Cellular environment and RNA structure. (A) SHAPE analysis of the adenine riboswitch aptamer domain under simplified conditions, in cells (the reference state, box), and in the presence of ligand⁵⁰. Higher and lower SHAPE reactivities relative to in-cell RNA are red and blue, respectively. (B) Model for the structure of the 16S rRNA in free 30S subunits in cells⁴⁹. SHAPE reactivity pattern in helices 28 and 44 is incompatible with the structure visualized in high-resolution structures. (C) Movement of helix 44 in the in-cell state, emphasizing a large-scale conformational switch. (D) The ∆SHAPE framework for identifying significant changes between two states. Structure shows ∆SHAPE sites in the human U1 snRNP complex (green spheres) and their proximal proteins. (E) Protein interactions across the mouse Xist lncRNA mapped using large-scale difference analysis. (top) Effects of the in-cell environment categorized by absolute differences in SHAPE reactivity (50-nt sliding window). (middle) Ratio of positive to negative differences is suggestive of protein binding and RNA structural rearrangement. (bottom) Positive and negative ∆SHAPE sites are blue and red, respectively. (F) Effect of translation on RNA structure in E. coli cells. SHAPE reactivities increase, relative to the cell-free state, specifically in highly translated genes; kasugamycin treatment partially abrogates this increase.

It had been known for decades that the small subunit of the ribosome can interconvert between “active” and “inactive” states^85,86. We found that in cells, 16S rRNA in 30S subunits exists primarily in an inactive conformation that has a SHAPE profile in helices 28 and 44 incompatible with high-resolution data⁴⁹ (Figure 6B). Binding by the antibiotic paromomycin switches the 16S rRNA into the active conformation, implying that the energy barrier between active and inactive states is low. Mutants that inhibit interconversion between states compromise translation in cells. Modeling revealed that the inactive and active structures differ in the position of helix 44, which likely has large effects on interactions between the 30S subunit and mRNA and for assembly with the large subunit⁴⁹ (Figure 6C). Thus, SHAPE revealed that the classic 30S ribosomal subunit inactive state is an abundant in-cell structure that regulates ribosome function.

SHAPE-MaP is unique among structure probing strategies because counting mutation events enables both measurement of local nucleotide flexibility and quantification of the uncertainty in each measurement (Figure 3A, bottom)^1,53. These features enable a statistical framework, ∆SHAPE^53,87, that tests for significant local structural differences between two states (Figure 6D). SHAPE reactivity differences can be summed over larger windows and used to identify positive and negative differences reflecting transitions between structured and unstructured as a result of cellular interactions, especially protein binding.

We used the ∆SHAPE framework to examine the effects of the cellular environment on the 18-kb Xist lncRNA⁷¹. Roughly one-half of all Xist nucleotides are changed by the cellular environment (Figure 6E). Xist contains several partial tandem repeat sequences that show large differences in SHAPE reactivity under simplified conditions as compared to in cells. Tandem repeats are among the few motifs that are consistently unstructured over large regions, and we showed that repeat regions function as landing pads for protein binding (Figure 6E). Other regions appear to either modulate protein binding or undergo conformational changes upon forming protein interactions. Our studies of Xist demonstrated that in-cell SHAPE allows efficient identification of important functional regions of large RNAs⁷¹ that can then be the focus of mechanism-based studies.

Ongoing cellular processes also affect RNA structure. SHAPE probing revealed that translation disrupts RNA structure in coding regions in E. coli cells and that this effect is larger specifically in highly translated coding regions². This in-cell RNA destabilization effect is reduced when translation is inhibited, and disappears for cell-free RNA (Figure 6F). In sum, RNA structure is heavily modulated by the cellular environment and can increase, decrease, or rearrange dramatically due to crowding, structural rearrangements, or interaction with proteins, including in surprising ways. SHAPE enables direct discovery and modeling of these effects.

mRNAs Have Distinct Structural Personalities

Data from most “transcription-wide” studies are sparse, noisy, and often analyzed in bulk statistical or meta-gene frameworks. As emphasized above, focusing analysis on such data is often misleading. For the E. coli transcriptome, meta-gene analysis would suggest that mRNAs are featureless (Figure 7A), but comprehensive, high-quality data reveal a strikingly different picture: E. coli transcripts show enormous structural diversity² (Figure 7B). Highly structured lncRNAs, like the RNase P RNA, have low SHAPE reactivities. By comparison, SHAPE reactivities for coding regions vary dramatically (Figure 7B). Each transcript, and indeed individual regions within transcripts, have distinct “structural personalities”.

Figure 7.

“Structural personalities” of bacterial and human RNAs. (A) Meta-gene representation of averaged mRNA structure in E. coli based on SHAPE reactivity and A/U nucleotide content. (B) Per-nucleotide SHAPE reactivities for one lncRNA and six mRNAs from E. coli². (C) Model for how RNA structure modulates accessibility to regulatory sequences. Unfolding an RNA motif (∆G^‡) imposes an energetic penalty on translation initiation (and likely many other processes). (D) RNA structures that tune translation initiation in E. coli. Brown box indicates AUG translation start site; arcs illustrate RNA base pairs. (E) Example of SERPINA1 mRNA structure containing a primary ORF and three (α, β, γ) competing upstream open reading frames (uORFs). Minimum free energy structures are shown, nucleotides are colored by SHAPE reactivity, and Kozak sequences are boxed.

In one example, mRNAs vary substantially in translation efficiency (TE), the amount of protein made from a given mRNA. Classical studies had shown that mRNA structure can enhance or impede the accessibility of regulatory sequences in an mRNA, influencing TE⁸⁸. We used high-quality SHAPE data from both bacterial and eukaryotic mRNAs to develop a general model for how RNA structure regulates TE⁸⁹. For E. coli mRNAs, a non-equilibrium model in which translation requires local unfolding of 30 nucleotides of RNA structure at the site of the Shine-Dalgarno sequence and start codon explains TE well² (Figure 7C). The best relationship between RNA structure and translation was obtained when we calculated the cost of disrupting pre-existing structure without considering refolding, suggesting that TE is governed by the cost of disrupting pre-existing structure, with unfolding rapidly followed by ribosome binding⁸⁹ (Figure 7D).

We extended these results to the 11 spliced isoforms of the human SERPINA1 mRNA. Each isoform encodes exactly the same protein in a primary open reading frame (ORF) downstream of different combinations of up to three upstream open reading frames (uORFs), each with their own Kozak sequence start site sequence of differing strengths⁹⁰. Use of any of the uORFs reduces translation efficiency at the primary ORF. When we used SHAPE data to model the structures of the 11 isoforms, we found that the same non-equilibrium model (Figure 7C), developed for bacterial RNAs, explained the relative accessibility of the uORFs and the primary ORF. Our model yielded a dramatic improvement over the so-called leaky-scanning model for translation of SERPINA1 isoforms⁹⁰ (Figure 7E). In sum, these results emphasize the importance of obtaining high-quality, per-nucleotide data across entire RNA transcripts and reveal that apparently unremarkable RNA structures can tune RNA functions by modulating the interactive accessibility of RNA to diverse ligands and proteins.

Structure-based Discovery of RNA Regulatory Elements in E. coli

In high-quality E. coli transcriptome data (Figure 8A), we identified 58 lowSS motifs, situated in 51 of 147 (35%) non-coding RNA regions² (Figure 8B). Of these, 49 motifs were uncharacterized, 80% showed evolutionary conservation, and roughly half overlapped regions with literature evidence for function (Figure 8C). Several were validated by our lab. For example, we identified a three-helix junction in the 5’-UTR of the operon that encodes the ribosomal proteins L28 and L33. Remarkably, the SHAPE-directed structure of this motif is similar to that of the large subunit rRNA in the region that binds L28 (Figure 8A). We showed that L28 and L33, plus L9 encoded in another operon, bind this three-helix junction element: L9 and L28 bind together, and L9 binding competes with L33. This motif, identified de novo by SHAPE-informed discovery, exhibits remarkable regulatory complexity². Every E. coli transcript we studied, for which we could obtain high-quality SHAPE data, appears to have some element of its expression regulated by RNA structure² (Figure 8D).

Figure 8.

De novo discovery of functional RNA elements as lowSS regions. (A) A lowSS region (gray box) in the rpmB mRNA forms an autoregulatory element that mimics L9/L28 binding sites in the 28S rRNA. (B) Novel RNA regulatory elements identified in E. coli. Structures are annotated by SHAPE reactivity and evidence for conservation. (C) Conservation of lowSS structures identified in enterobacteria and evidence of function based on literature. (D) Mechanisms by which RNA structure regulates gene expression across the E. coli transcriptome based on identification of well-determined secondary structures². Arcs indicate base pairs.

The Frontier in Motif Discovery: Higher-order and Tertiary RNA Structure

The next frontier opportunity is to extend large-scale RNA structure analysis to more complex structures. We recently probed the structure of the entire dengue serotype 2 virus (DENV2) RNA genome by SHAPE and identified 24 lowSS elements³ (half are shown in Figure 9A). These elements overlap with previously identified functional elements or show evidence of evolutionary pressure to maintain the modeled secondary structure, consistent with functional roles in viral replication. The existence of numerous higher-order structures in a large RNA is now an utterly unremarkable observation. A critical goal now is to determine which of these elements form higher-order functional structures.

Figure 9.

Discovery of higher-order RNA structure. (A) Well-determined secondary structure elements across the first half of the DENV2 genome based on the lowSS metric. (B) RING correlated chemical probing analysis of three RNA elements with well-determined structures and higher-order folds. Elements are named by the gene in which they occur. (C) Effect of mutations in regions with significant RING correlations on (top) global compaction of the DENV2 RNA genome (measured by dynamic light scattering) and on (bottom) replication fitness (visualized by immunostaining of DENV2 envelope protein and nuclei). (D) Modeled three-dimensional folds of three elements in the DENV2 RNA based on RING correlations used to restrain discrete molecular dynamics simulations.

We used a new chemical probing strategy called RING-MaP^16,17 to assess the likelihood that individual elements in the DENV2 RNA genome form higher-order structures. RING-MaP exploits the ability of MaP reverse transcription to measure multiple chemical adducts in a single RNA strand. Nucleotides that experience structural communication, because they are in the same helix^19,57 or because they are linked via through-space interactions^16,17,59, show correlated internucleotide reactivities more often than expected by chance (Figure 3A, bottom). At present, RING uses the conventional structural probing reagent dimethyl sulfate^16,19. RINGs can be approximately separated into those corresponding to secondary versus tertiary interactions by the number of nucleotides between sites after excluding internal stem loops³; longer contact distances reflect tertiary structure. Eight regions in the DENV2 RNA showed RING correlations supportive of higher-order tertiary structure (three are shown in Figure 9B).

Mutating these RNA elements with higher-order structure, located in the Env and NS2A genes (although not the NS2B gene), increased the hydrodynamic radius of RNA genome transcripts and compromised viral fitness³ (Figure 9C). The mutant viruses replicated poorly even after 60 days of passage in cell culture, suggesting that disrupting RNA tertiary structure is not easily revertible by the virus. We used RING measurements as restraints in molecular dynamics simulations to model the overall folds of regions with dense RINGs. Most elements fold into compact, well-determined structures (Figure 9D). The precise mechanisms by which these higher-order RNA structures affect viral fitness are not fully known.

In sum, RING experiments revealed that RNA motifs with complex, higher-order architectures are pervasive across the DENV2 RNA, are functionally integrated with viral replication, and – we posit – are harbingers of complex, functional structures likely to be found in the future in numerous viral, messenger, and non-coding RNAs.

Long-term Opportunities

RNAs fold back on themselves to form substantial internal structures, many of which play direct roles in function. With the advent of experimentally concise, comprehensive, quantitative, and high-throughput RNA structure probing technologies, we are now poised to identify and focus on those structures most likely to impact function and regulatory mechanisms. In essentially every instance where my laboratory, and many other research groups, have applied SHAPE to interrogate biological systems, new insights regarding RNA structure-function interrelationships have emerged. RNA structure likely regulates expression and biological roles of every class of RNA, is sufficiently ubiquitous to comprise a distinct level of the genetic code, and likely creates higher-order folds broadly targetable with small molecules.

Although additional innovations will certainly be developed, per-nucleotide RNA structure probing by SHAPE is now a mature, well validated, and robust technology. SHAPE, mutational profiling (MaP), and related strategies should be applied to essentially all projects involving RNA. The RNA structure probing and analysis field is poised to undergo a transition. The next frontier is to devise experimentally concise, information-rich, and accurate strategies for measuring higher-order and tertiary interactions at large scales.

Biography

Kevin Weeks is Kenan Distinguished Professor of Chemistry at the University of North Carolina. His laboratory leverages chemical principles to devise definitive and quantitative technologies for understanding how RNA structure governs biological function and uses these technologies to define new principles of RNA-mediated regulation, spanning virus replication, mRNA and lncRNA biology, and RNA-targeted drug discovery. The overarching vision of the Weeks laboratory is to prepare team members for long-term leadership at the interfaces of chemistry, biology, technology, and genomics.