RNA systems biology: uniting functional discoveries and structural tools to understand global roles of RNAs

Abstract

RNAs assume sophisticated structures that are active in myriad cellular processes. In this review, we highlight newly identified ribozymes, riboswitches and small RNAs, some of which control the function of cellular metabolic and gene expression networks. We then examine recent developments in genome-wide RNA structure probing technologies that are yielding new insights into the structural landscape of the transcriptome. Finally, we discuss how these RNA ‘structomic’ methods can address emerging questions in RNA systems biology, from the mechanisms behind long non-coding RNAs to new bases for human diseases.

Graphical abstract

Introduction

The ability of RNA to encode both genetic and structural information is paramount to its biological centrality. Its predominantly single-stranded nature allows RNA to serve as both the physical template of protein synthesis and adopt intricate structures that influence genetic processes. For example, catalytic RNAs (ribozymes) perform essential cellular functions including translation, tRNA maturation, and splicing. Even more diverse are the roles of non-coding RNAs in regulating gene expression. These roles are frequently mediated by cis- and trans-acting RNA structures that block or expose regulatory elements within mRNAs that control transcription, translation, or RNA degradation [1]. Further regulatory roles for RNAs include protein recruitment, molecular scaffolding, and RNA interference, with many others being discovered at an accelerating rate [2]. These advances frame an emerging picture of diverse RNAs acting together in a networked, systems-level fashion to regulate the fundamental processes of the cell (Figure 1).

Figure 1. The centrality of RNA structures in regulating cellular processes.

Non-coding RNAs (ncRNAs) play widespread and diverse roles in the regulation of cellular processes. (Center) A schematic of representative classes of the structures formed by ncRNAs including ribozymes (orange), riboswitches (green), small RNAs (sRNAs, blue), and long non-coding RNAs (lncRNAs, purple). The surrounding panels depict representative functions of each of these classes including (clockwise) concatemer cleavage in rolling circle replication [56] and group II intron splicing [57] (ribozymes); metal-ion sensing [10–12], regulation of biosynthetic operons [13], and regulation of sRNA expression [16,17] (riboswitches); sequestration of regulatory factors [16,17] and information processing in regulatory networks [22] (sRNAs); and controlling DNA methylation [53] and scaffolding for inter-chromosomal structures [52] (lncRNAs). Functional RNA motifs are highlighted in colors corresponding to the center schematic. Protein components are shaded grey.

Alongside exciting discoveries about the breadth of RNA function is the development of tools to uncover ‘omics’-level views of RNA structure. As RNA function is intimately tied to RNA structure, these technologies provide powerful strategies for elucidating RNA structure-function relationships on a systems-level scale by accessing structural information for entire transcriptomes in their native cellular context.

In this review, we unite exciting developments in the growing knowledge of systems-level RNA functions and new capabilities used to uncover the RNA structures that give rise to those functions. We start by highlighting new discoveries that expose the prevalent and varied nature of RNA functions in biological systems. Next, we discuss recent experimental developments in high-throughput RNA structure analysis at the transcriptome level, the bioinformatic advances necessary to analyze the generated data sets, and the insights these studies have provided. Finally, we highlight questions that can be asked with a systems-level knowledge of RNA structure-function relationships and consider the new role that their answers will play in the future of RNA biology.

I. Unearthing New Global Roles for RNAs in Regulating Cellular Processes

Recent efforts to identify and characterize RNA-mediated regulatory pathways have led to an appreciation for the role of RNA in governing global cellular processes such as metabolic and gene expression networks. The identification of new RNA mechanisms and functional roles suggests that others remain hidden within the transcriptome.

Twister, Twister Sister, Pistol, and Hatchet – New Ribozymes Hiding in Plain Sight

In spite of their involvement in major cellular functions such as translation and tRNA processing, as of 2013 only 10 classes of natural ribozymes had been identified [3]. Recently, Roth et al. discovered a new ‘twister’ class of self-cleaving ribozyme using a comparative genomics approach that incorporated RNA structure prediction through sequence covariation analysis to identify over 2,700 sequences that match the twister motif across diverse organisms from bacteria to eukarya [3]. This sequence and structure-based search methodology also suggested that the twister ribozyme forms a compact, double-pseudoknotted structure that is further supported by biochemical evidence and three crystallographic studies [3–6] (Figure 2).

Figure 2. New roles for non-coding RNAs.

(A) Widespread identification of novel self-cleaving ribozymes. Secondary structures for the twister, twister sister, pistol, and hatchet ribozymes are shown [3,7]. A crystal structure of a twister ribozyme is shown (PDBID: 4QJH) [5]. Pseudoknot interactions are shown in magenta and orange in the twister ribozyme secondary structure and correspond to magenta and orange nucleotides in the twister ribozyme crystal structure. (B) An AdoCbl riboswitch regulates the expression of an sRNA sponge [16,17]. In the absence of AdoCbl, the full length EutX (or Rli55) sRNA is synthesized and sequesters the transcription antiterminator EutV. In the presence of AdoCbl, premature termination of EutX/Rli55 permits EutV to antiterminate transcription of the eut operon, leading to the expression of proteins involved in ethanolamine catabolism. (C) New roles for sRNAs in the global regulation of cellular processes. (Left) In response to outer membrane stress in γ-proteobacteria, the sRNA MicL is expressed from a σ^E-dependent promoter embedded in the cutC coding sequence and is processed into MicL-S, which inhibits translation of the outer membrane protein Lpp [26]. (Right) Decay of the gltI mRNA yields the sRNA SroC, which promotes degradation of the global regulatory sRNA GcvB, which regulates translation of the gltI mRNA [24].

Despite this exciting discovery, the functional roles of twister ribozymes remain a mystery. Furthermore, by searching near genetic elements frequently associated with twister or hammerhead ribozymes, Weinberg et al. recently added the ‘twister sister’, ‘pistol’, and ‘hatchet’ classes of self-cleaving ribozymes to the list of new ribozyme classes with unknown function [7–9] (Figure 2). While the enrichment of ribozymes in specific genetic contexts could yield clues to their function, their apparent ubiquity across broad organism classes suggests new roles of ribozymes in controlling core cellular processes that have yet to be uncovered.

Riboswitches – A Network of Small-Molecule Regulators

Like ribozymes, riboswitches have functions that are closely linked to their structures. Whereas ribozyme structures enable catalysis, riboswitch structures switch between distinct conformations in the presence or absence of a ligand to modulate gene expression (Figure 2). These ligand-mediated structural changes provide a natural sensory feedback mechanism to regulate genes involved in controlling ligand concentration at the transcriptional, translational, and splicing levels.

The recent identification and characterization of new riboswitches has expanded our understanding of their role in modulating cellular state through specific regulation of key transporter genes and even small RNAs. Recent studies have identified roles for riboswitches in prokaryotic metal ion homeostasis. Specifically, an Mn²⁺-responsive riboswitch (yyb-ykoY) was shown to control the expression of an Mn²⁺ exporter [10,11] and a Ni²⁺/Co²⁺ riboswitch was shown to control expression of Co²⁺ transporters [12]. Since Mn²⁺ and Co²⁺ are cofactors of protein enzymes, but are toxic at elevated levels, these studies highlight how riboswitches provide feedback mechanisms that affect cell physiology.

In another example of expanded roles of riboswitches, Kim et al. recently identified and characterized a class of riboswitch that responds to ZMP, a purine biosynthetic intermediate, and its 5′-triphosphorylated derivative, ZTP [13]. The widespread existence of the ZTP riboswitch provides a molecular basis for a previous proposal that elevated levels of ZTP function as an alarmone to signal low levels of 10f-tetrahydrofolate, a formyl group donor in purine biosynthesis. Finally, Kellenberger et al. and Nelson et al. have identified the subclass GEMM-1b riboswitch in the bacterium Geobacter metallireducens that responds selectively to the cyclic dinucleotide cAG [14,15]. Interestingly, many genes that are regulated by the GEMM-1b riboswitch are associated with extracellular electron transfer [14,15].

Finally, two studies reported a riboswitch that controls the expression of small RNAs (sRNAs) that regulate the eut mRNAs involved in ethanolamine catabolism [16,17]. In this system, the protein EutV interacts with the 5′ untranslated region of the eut mRNAs to regulate their expression by transcription antitermination. However, an sRNA (Rli55 in Listeria monocytogenes and EutX in Enterococcus faecalis) can sequester EutV, preventing it from antiterminating the eut mRNAs. The sRNA is regulated by an adenosyl cobalamine (AdoCbl) riboswitch that terminates sRNA synthesis before the EutV binding site in the presence of the ethanolamine catabolism cofactor AdoCbl (Figure 2).

Big Roles For Small RNAs in Bacteria

Bacterial sRNAs have long been known to be important regulators of cellular state via regulating specific target genes. Recently, several groups have reported advances in understanding how the RNA-binding protein Hfq mediates these processes by presenting sRNAs for mRNA target recognition [18–21]. These structural studies are revealing key design principles for sRNA structure and function and are supporting newly discovered big roles of sRNAs across the cell, including as elements in sophisticated regulatory networks that facilitate cellular information processing. For example, Papenfort et al. showed that the sRNA RprA controls a coherent feed-forward loop with AND gate logic that regulates Salmonella plasmid conjugation by controlling expression of the ricI gene [22].

Several recent studies have reported more global roles for sRNAs in the regulation of cellular state. Duss et al. recently uncovered the molecular basis behind this capability of the Pseudomonas fluorescens sRNA RsmZ by showing that it can sequentially bind five RsmE dimers, as well as release RsmE following RNaseE cleavage of RsmZ [23]. In another example, Miyakoshi et al. showed that the sRNA SroC functions as a sponge for another sRNA, GcvB, a global regulator in Salmonella [24]. Interestingly, SroC is generated as an mRNA decay product of one of GcvB’s targets, creating an sRNA feedback loop within this regulation (Figure 2). Additionally, Chao and Vogel showed that RNaseE cleaves the 3′ untranslated region of the stress chaperone CpxP mRNA to produce the sRNA CpxQ, which represses mRNAs that encode inner membrane proteins [25]. Finally, Guo et al. reported the discovery of a new sRNA, MicL, that downregulates the most abundant protein in E. coli, the major lipoprotein Lpp, in times of membrane stress [26]. This is particularly interesting because MicL is expressed from a newly identified σ^E-dependent promoter within the coding sequence of cutC (Figure 2). This suggests a potential abundance of RNAs playing systems-level regulatory roles that remain hidden throughout the transcriptome waiting to be found.

II. RNA Structomics – A Burgeoning New Field Enabled by New Technologies

The accelerating discovery rate of new RNA functions demands methods that can provide structure-function insights at the same pace. While phylogenetic analysis of RNA structure has been immensely successful in identifying bacterial functional RNAs, the application of such methods to eukaryotes is complicated by increased genomic complexity and reduced sequence divergence [27]. New techniques that marry RNA enzymatic or chemical probing with next-generation sequencing (NGS) provide an experimental framework for the identification of functional RNAs at a transcriptome-wide level. This new ‘RNA structomics’ field [28] is already uncovering new insights into the global roles of RNA structures across cellular processes.

A Suite of New High-Throughput Methods Characterize RNA Structures Across the Entire Transcriptome

Early approaches to transcriptome-level RNA structure probing include the FragSeq and PARS techniques, which used NGS to sequence and map cleavage positions generated by ssRNA and dsRNA nucleases [29,30] (Table 1, Figure 3). In a similar spirit, PIP-seq combined nuclease-based RNA structure probing with crosslinking methods to access a transcriptome-wide profile of RNA-protein interactions [31]. Transcriptome-wide RNA structure probing techniques underwent another breakthrough with the use of small molecule chemical probes that can diffuse across the cell membrane and thereby probe RNA structures in their native environment [32,33]. Chemical probes also allow the interrogation of RNA structures at higher resolution because of their small size compared to the more bulky enzymes. Following the development of methods for coupling chemical probing with NGS [34], techniques such as DMS-Seq [35], structure-seq [36], and Mod-Seq [37] were developed to probe the structure of the transcriptome inside the cell (Table 1, Figure 3). These chemical probing-NGS methods consist of a core set of steps outlined in Figure 3. While powerful, the first versions of these techniques were limited by dimethyl sulfate (DMS), which has a strong preference for A and C positions. Incarnato et al. partially addressed this by probing with both DMS and N-cyclohexyl-N’-(2-morpholinoethyl)carbodiimide metho-p-toluenesulfonate (CMCT), which reacts primarily with G and U, although their method, CIRS-seq, did not modify the RNA until after cell lysis [38] (Table 1, Figure 3). Complete nucleotide coverage was achieved with icSHAPE, which uses a clickable version of the SHAPE reagent 2-methylnicotinic acid imidazolide (NAI) [32], NAI-N₃. In icSHAPE, after modification of RNA by NAI-N₃, a biotin moiety is added via click chemistry to enable selective purification of probed RNAs [39]. SHAPE-MaP, which uses mutational profiling to locate modification positions, has also been applied on a transcriptome-wide scale using the well-characterized SHAPE reagent 1-methyl-7nitroisatoic anhydride (1M7) [40]. Finally, a complementary technique called RNA proximity ligation (RPL) was recently developed to characterize the proximity of nucleotides in three-dimensional space using a combination of RNase cleavage and localized ligation [41].

Table 1. Characteristics of high throughput RNA structure probing methods.

These techniques use a series of structure probing, extraction, processing, next generation sequencing library preparation, sequencing and bioinformatic data analysis steps (Figure 3) to characterize RNA structures in high throughput and in some cases transcriptome-wide. Detailed differences are included below.

Name	Modifying reagent or enzyme	Probing, Extraction and Processing before Sequencing Library Preparation	Organism(s) studied	Current Analyzing Software	Reference(s)
PARS (parallel analysis of RNA structure)	V1 (dsRNase) & S1 (ssRNase)	RNA extraction, equilibrium refolding, enzyme treatment, RNA fragmentation & linker ligation	Saccharomyces cerevisiae	Bowtie2, custom scripts	Kertesz, 2010, Nature [30]
FragSeq	P1 (ssRNase)	RNA extraction, equilibrium refolding, enzyme treatment, fragmentation & linker ligation	Mus musculus	FragSeq algorithm	Underwood, 2010, Nat. Meth. [29]
PIP-Seq	RNase ONE (ssRNase) & V1 (dsRNase)	crosslinking, cell lysis, enzyme treatment, crosslink reversal, RNA extraction	Homo sapiens, Arabid opsis thaliana	Tophat, CSAR	Silverman, 2014, Genome Biol. [31]
DMS-Seq	DMS	in-cell DMS modification, RNA extraction, fragmentation & linker ligation	Saccharomyces cerevisiae	SOAP	Rouskin, 2014, Nature [35]
structure-seq	DMS	in-cell DMS modification, RNA extraction, random priming	Arabidopsis thaliana	StructureF old (as part of Galaxy suite)	Ding, Nature, 2014 [36]
Mod-Seq	DMS	in-cell DMS modification, RNA extraction, fragmentation & linker ligation	Saccharomyces cerevisiae	Mod-seeker	Talkish, RNA, 2014 [37]
CIRS-Seq	DMS, CMCT	RNA extraction, DMS or CMCT modification, random priming	Mus musculus	custom scripts	Incarnato, Genome Biol, 2014 [38]
icSHAPE (in vivo click selective 2′-hydroxyl acylation and profiling experiment)	NAI-N3 (2-methylnico tinic acid imidazolid e)	in-cell NAI-N3 modification, RNA extraction, fragmentation, biotin click & purification	Mus musculus	Bowtie2, custom scripts	Spitale, Nature, 2015 [39]
RPL (RNA proximity ligation)	Endogeno us RNases	spheroplast,endogenous RNase cleavage, RNA cross-strand ligation, fragmentation & ligation	Saccharomyces cerevisiae	STAR aligner, custom scripts	Ramani, Nat. Biotech., 2015 [41]
SHAPE-MaP (selective 2′-hydroxyl acylation analyzed by primer extension and mutational profiling)	1M7 (1-methyl-nitroisatoic anhydride)	in vitro synthesis or viral purification, equilibrium folding, in-cell 1M7 modification, RNA extraction and fragmentation, random/targeted priming & ligation	Hepatitis C & HIV, Mus musculus	ShapeMa pper	Mauger, PNAS, 2015 & Lavender, Plos Comput. Biol., 2015 [40,58,59]
In-cell SHAPE-Seq	1M7	In-cell 1M7 modification, RNA extraction, specific priming	Eschericia coli (natural and synthetic sRNAs, riboswitches and RNase P)	Spats	Watters, NAR 2016 [47]

Figure 3. Transcriptome-wide RNA structure probing technologies.

Transcriptome-wide RNA structure probing uses chemical probing or enzymatic cleavage to introduce covalent modifications or directly cleave RNA in a structure dependent fashion, respectively. Modification or cleavage positions are detected through processing steps followed by next-generation sequencing. Bioinformatic processing of the resulting sequencing reads yields a measure of chemical modification ‘reactivity’ or enzymatic cleavage frequency at each nucleotide. High reactivities correspond to flexible nucleotide positions that are not participating in RNA structures or bound by cellular factors. High enzymatic cleavage frequencies give information on structure depending on the characteristics of the nucleases used. These values can be used for several types of specific analyses, such as constrained RNA folding, averaging meta-analysis of reactivities across the entire transcriptome, and comparisons with in vitro probing data. An outline of the steps for SHAPE-Map [40] and in-cell SHAPE-Seq [47] (purple) CIRS-seq [38] (maroon), icSHAPE [39] (red), structure-seq [36] (orange), DMS-seq [35] (yellow), Mod-seq [37] (light green), PARS [30] (green), FragSeq [29] (blue), and RPL [41] (grey), is shown. Further technique details can be found in Table 1.

The complexities of the datasets generated by NGS methods have required substantial new developments in bioinformatics pipelines that can ultimately convert NGS reads into RNA structure models. To do this, NGS reads are converted into ‘reactivity’ values, broadly defined as a measure of the flexibility of a given nucleotide position [34,42,43]. Reactivities can then be used to generate RNA structural models that account for the tendency of more reactive nucleotides to be unpaired [44]. SeqFold presents one particularly interesting modeling approach to select RNA structures that are most consistent with the experimental reactivity data [45]. Seqfold’s approach is powerful because it uses reactivity information to pick from clusters of sub-optimal structures rather than relying solely on the minimum free energy structure model. The various reactivity calculation and RNA structural modeling approaches are still in their infancy and represent a challenging new frontier for computational biology to fully utilize the vast datasets generated by the new NGS structure probing techniques.

Global Insights Into the Roles of RNA Structures Across Cellular Processes

Characterization of RNA structures at the transcriptome level is revealing features of RNA structures on a genome-wide scale. Meta-analyses that average reactivities of many different RNAs have revealed structure-function trends across multiple species. Notably, a three-nucleotide periodicity of reactivity was observed within mRNA coding regions [36,38,39,46]. As another example, the Kozak sequence appeared to be highly reactive in H. sapiens [46], A. thaliana [36], and M. musculus [38,39], suggesting that it is generally unstructured to facilitate translation initiation. In addition, Wan et al. observed that nucleotides preceding sequences known to interact with miRNAs tend to be unstructured [46].

There are also interesting conclusions gained from comparing reactivities of RNAs refolded and probed in vitro to RNAs probed in vivo. For example, S. cerevisiae RNAs appear more unstructured in vivo than in vitro [35]. The data collected from icSHAPE in M. musculus also support this argument, although the degree of in vivo unfolding observed was different across different classes of RNA elements [39]. In A. thaliana, Ding et al. reported a correlation between less structured mRNAs in vivo and mRNAs annotated for stress response and suggested that reduced structure may facilitate stress-mediated RNA structural changes [36]. Spitale et al. found that Kozak sequence accessibility observed in vivo was preserved in vitro, suggesting that this and other structural features of translation regulatory regions are programmed by the mRNA sequence and not through interactions with cellular factors [39]. Smola et al. developed the ΔSHAPE analysis framework for characterizing RNA-protein interactions through comparison of RNAs probed in cellulo and ex vivo [40]. Spitale et al. also show that comparison between in vivo and in vitro reactivity data can uncover specific RNA structural changes due to protein binding [39], which has also been shown in another recent study in E. coli with in-cell SHAPE-Seq [47].

III. New Technologies Enable New Questions

Rapid advances in RNA structure characterization technologies promise to change the way we investigate the relationship between RNA structure and function at a systems-level. While many new questions can be addressed, two of the most interesting are the structural basis of long non-coding RNA (lncRNA) function and the role of RNA misfolding in human diseases.

What are the structure-function principles of long non-coding RNAs?

lncRNAs are loosely defined as RNA molecules more than 200 nucleotides long with little-to-no protein-coding capacity [48]. Despite their abundance [49], lncRNAs are one of the least understood RNA classes. While we know lncRNA structure is important [50], we have little detailed knowledge of how specific lncRNA structures mediate their broad arrays of function, although this has begun to change. Recently, Somarowthu et al. used several chemical probing techniques to determine the secondary structure of the 2,148 nt long lncRNA HOTAIR, giving structural insights into how this RNA performs the twin functions of regulating epidermal tissue development and repressing tumor and metastasis suppressor genes [51]. It will be exciting to gauge how RNA structures influence the function of newly discovered lncRNAs, such as Firre, which has been shown to act as a platform for organizing trans-chromosomal association [52] (Figure 1). Another interesting new example is the extra-coding CEBPA, which controls DNA methylation state at the CEBPA locus using RNA structures that are targeted by a DNA methyltransferase, DNMT-1 [53] (Figure 1). We anticipate this to be the tip of the iceberg, as high-throughput structural studies enable a wealth of new insight into the structure-function principles of these important global RNA regulatory molecules.

How does RNA misfolding contribute to human disease?

The growing appreciation for the role of RNA structure in cellular activities has lead to intriguing questions about the role of RNA structure in human disease. A recent focus of these studies is the “riboSNitch”, an RNA-encoded regulatory element in which a single nucleotide variant (SNV) significantly alters its structural ensemble, sometimes leading to a disease state such as β-Thalassemia or Chronic Obstructive Pulmonary Disease [54]. Following the initial computational prediction of riboSNitches and their disease associations [54], Wan et al. used PARS (described above) to structurally probe the transcriptomes of a mother, father, and child on a genome-wide scale and found that over 1,907 (15%) of identified SNVs altered RNA structures between these relatives [46]. The dataset acquired in this study was then used by Corley et al. to benchmark RNA folding algorithms to predict the effect of SNVs on RNA structure and thus accurately predict the locations of riboSNitches from primary sequence information [55]. While there is still fascinating work to be done to improve computational prediction, the link between riboSNitches and disease is one of the most exciting areas of future RNA research, both in terms of understanding the global RNA structure-function relationship and as a new frontier in human disease research.

Conclusion

Far from being a passive carrier of genetic information and an intriguing catalyst of select chemical processes of life, RNAs play diverse roles as regulators of central cellular processes. Our knowledge of these roles is expanding at an accelerated rate, with recent discoveries uncovering RNAs in unexpected places and with unexpected function. These studies suggest that we may need to rethink our view of RNA yet again and may warrant investment in a new study of ‘RNA systems biology’ that can more thoroughly uncover the roles and mechanisms of RNAs in modern biology.

Highlights.

• We review recent progress in uncovering the global roles of RNAs across the cell.

• Regulatory RNAs are more ubiquitous and play more global, networked roles than previously thought.

• ‘RNA Structomics’ allows cellular RNA structure to be characterized across transcriptomes.

• New ways in which RNA structures globally regulate cellular processes have been uncovered.

• RNA Structomics is enabling progress in understanding the role of RNA misfolding in human disease.