Transcription regulation through nascent RNA folding

Abstract

Folding of RNA into secondary structures through intramolecular base pairing determines an RNA’s three-dimensional architecture and associated function. Simple RNA structures like stem loops can provide specialized functions independent of coding capacity, such as protein binding, regulation of RNA processing and stability, stimulation or inhibition of translation. RNA catalysis is dependent on tertiary structures found in the ribosome, tRNAs and group I and II introns. While the extent to which non-coding RNAs contribute to cellular maintenance is generally appreciated, the fact that both non-coding and coding RNA can assume relevant structural states has only recently gained attention. In particular, the co-transcriptional folding of nascent RNA of all classes has the potential to regulate co-transcriptional processing, RNP (ribonucleoprotein particle) formation, and transcription itself. Riboswitches are established examples of co-transcriptionally folded coding RNAs that directly regulate transcription, mainly in prokaryotes. Here we discuss recent studies in both prokaryotes and eukaryotes showing that structure formation may carry a more widespread regulatory logic during RNA synthesis. Local structures forming close to the catalytic center of RNA polymerases have the potential to regulate transcription by reducing backtracking. In addition, stem loops or more complex structures may alter co-transcriptional RNA processing or its efficiency. Several examples of functional structures have been identified to date, and this review provides an overview of physiologically distinct processes where co-transcriptionally folded RNA plays a role. Experimental approaches such as single-molecule FRET and in vivo structural probing to further advance our insight into the significance of co-transcriptional structure formation are discussed.

Graphical Abstract

Introduction

Transcription, the process by which a single-stranded RNA is synthesized from a DNA template, serves as the rate-limiting step in gene expression and is catalyzed by RNA polymerases. The single-stranded RNA, unlike its double-stranded DNA parent, can fold into all sorts of secondary and tertiary structures through base-pairing and other interactions. While many bacteria possess a single RNA polymerase (Pol) that transcribes both coding and non-coding RNA, eukaryotes have several RNA polymerases: Pols I, II, and III in most organisms as well as Pol IV in plants [1–3]. These holoenzymes are responsible for transcription of distinct gene types, and these gene types vary in the degree to which their RNA products are folded. The single archaeal RNA polymerase is closely related to Pol II, which transcribes eukaryotic protein-coding genes as well as most microRNA (miRNA) and small nuclear RNA (snRNA) genes. Mature mRNAs generally exist in less structured states [4] than precursor and mature miRNAs as well as snRNAs, which require extensive folding for their function [5]. This is in part due to the genetic triplet code, which constrains structure evolution whereas noncoding RNAs and intronic sequences can evolve freely [6,7]. Pol II is the only RNA polymerase with an unstructured C-terminal domain (CTD) on its largest subunit. Because the CTD is crucially involved in transcription regulation as well as coupling with RNA processing steps, it is possible that this domain integrates co-transcriptional RNA folding and processing status with Pol II elongation behavior. Unlike Pol II, Pols I and III are specialized RNA polymerases that synthesize highly structured non-coding RNAs: ribosomal RNA (rRNA) by Pol I; transfer RNA (tRNA) and some snRNAs and small nucleolar RNAs (snoRNAs) by Pol III. Similarly, Pol IV in plants synthesizes stably folded siRNA precursors. The observation that Pols I-IV synthesize RNAs that are folded differently and to different extents raises the possibility that some or all of these folding events regulate transcription by each of these co-evolved eukaryotic polymerases.

Transcription regulation occurs at several levels starting with initiation, wherein an RNA polymerase is guided towards a gene’s promoter region and assembles with general transcription factors to access the DNA. The enzyme is then released into active elongation, which is characterized by rapid synthesis—15–75 nt/second in eukaryotes—through sequential addition of ribonucleotides to the nascent RNA 3’-end [2]. Finally, transcription is terminated once the nascent RNA is cleaved. A combination of events including nascent RNA degradation and signaling within the polymerase likely lead to polymerase release from the DNA template. Indeed, a range of processes take place during transcription in eukaryotes, including 5’-end capping of Pol II transcripts, RNA editing, base modification, endonucleolytic processing of Pol I transcripts, and 3’-end cleavage (Fig 1a). Promoter-proximal pausing and pausing just downstream of 3’-end cleavage at polyA sites likely involve the activity of proteins rather than nascent RNA folding [8]. In some cases, co-transcriptional RNA processing events have been imaged by electron microscopy, which shows the formation of protein-containing processing complexes on the nascent RNA attached to the DNA template. For example, translation itself can be a co-transcriptional process in prokaryotes: Escherichia coli chromosome spreads show nascent RNA covered with ribosomes chasing the polymerase (Fig. 1b, top), where translational activity can either promote transcription or termination ([9] and references therein). Pol II transcripts are often spliced co-transcriptionally by the spliceosome (Fig 1a), and EM imaging similarly identifies spliceosomes assembled on nascent RNA in electron micrographs of actively transcribed Drosophila genes (Fig. 1b, bottom). Each of these modification events come with their own set of regulatory interactions that often depend on RNA folding (Fig 1a). For example, RNA editing and histone pre-mRNA 3’ end cleavage depend on stem-loop formation [10,11]. So, eukaryotic RNA must fold within the timeframe of transcription [12].

Fig. 1 : Co-transcriptional RNA processing events are regulated by nascent RNA structure.

(a) Eukaryotic RNA processing involves protein and RNA/protein complexes binding sequence or structure features on the elongating transcript. The RNA 5’-end is co-transcriptionally capped and occupied by the Cap-binding complex. Components of the spliceosome (snRNPs) recognize intron boundaries, the 5’ splice site (5’SS) and branch site (BS) to facilitate co-transcriptional splicing. The exon junction complex (EJC) marks successfully spliced exon-exon boundaries. RNA editing factors such as adenosine deaminases (ADAR) bind to and act on double-stranded RNA. Small proteins specifically recognize stem loops and regulate e.g. termination of histone gene transcription. The signal for polyadenylation is read by the cleavage and polyadenylation specificity factor (CPSF), recruiting additional proteins such as cleavage stimulation factor (CstF) to release and polyadenylate the nascent transcript. Not to scale. (b) E. coli chromosome spread [16] showing ribosomes covering the nascent RNA (top); chromatin spread of a Drosophila chorion gene [15] showing spliceosome complexes assembled on nascent RNA (bottom).

This review explores the hypothesis that co-transcriptional RNA folding regulates transcription more broadly than previously known, including basic mechanisms that impact eukaryotic gene expression. Several lines of evidence have recently emerged highlighting how structures within the nascent RNA chain itself can act as modulators of transcription elongation as well as co-transcriptional processes. Control of gene expression by nascent RNA folding is a well-recognized phenomenon in bacteria, where for example hairpin loops serve as transcription termination signals during rho-independent termination [13]. Riboswitches are an additional class of RNA structural elements that fold during transcription and exert transcriptional or translational control [14]. How nascent RNA folding, processing and synthesis occur and contribute to transcription regulation is an emerging topic in eukaryotes, where very little is known. This review provides an overview of existing evidence for co-transcriptional formation of functional RNA structure and highlights experimental techniques with the potential to identify yet unknown mechanisms.

RNA folds co-transcriptionally

In the 1980s, co-transcriptional folding was first recognized with the realization that truncated tRNA fragments fold into the expected secondary structure that is present in the mature RNA, without the need of long-range or tertiary interactions that only the fully synthesized sequence could provide [17]. With the emergence of the first efficient algorithms for RNA folding [18], Nussinov and Tinoco simulated the folding pathway of a simian virus 40 (SV40) transcript and found sequential folding of local stem loop structures before gross rearrangements of single-stranded regions concluded the folding reaction. The local structures persisted in the final fold. This observation yielded the hypothesis that the sequential nature of RNA folding can limit the number of total pathways and therefore might be faster and less error-prone than folding after transcription. These experiments, however, were based on artificially or computationally truncated constructs, omitting interactions with RNA polymerase, effects of transcription speed or pauses. That RNA must start folding during transcription in vivo has been established by considering the timescales for transcription and folding. Mammalian RNA Polymerases transcribe at a speed of 10–20 nt/s, and faster bacterial homologues can reach speeds up to 80 nt/s. Small RNA hairpin structures can fold on the microsecond timescale, far faster than the addition of nucleotides to the elongating 3’-end, and therefore occurs already as transcription is still ongoing [19].

Specific RNA structure formation can depend on transcription

Through intramolecular base pairing, linear RNA molecules can fold first into secondary and then tertiary structures and exert specialized functions. Secondary structures arise from the base pairing ability between canonical G•C and A•U and the weaker G•U wobble pairs, forming double-stranded stems and single-stranded loop regions [20]. Three-dimensional arrangement of these features then determines tertiary structure with a myriad of possible topologies, analogous to protein structure [21]. RNA folding is driven by hydrogen bonding during base pairing, base stacking, and electrostatic interactions. As opposed to protein folding, no hydrophobic collapse supports formation of the native state, which is why the folding pathway is critical for an RNA to assume the correct structure [22]. Additionally, the formation and resolution of stable base pairs in folding intermediates plays a central role for native state formation, again distinct from proteins that lack such regular, specific pairings. Most of our understanding of RNA structure formation comes from in vitro studies, where RNA is denatured and subsequently re-folded. This experimental approach stands in contrast to the in vivo scenario, where the linear RNA molecule emerges sequentially from the RNA polymerase exit channel (Fig. 1a). Nucleotides that are released can already engage in base pairing interactions with upstream sequence towards the 5’-end, promoting formation of local structures which might not appear in a re-folding scenario. Only as transcription progresses, long-range structural rearrangements are permitted. In this way, potentially less stable local interactions might be favored over long-range but more stable associations, in case the energy to resolve the primary structure is not provided. Additionally, the speed of transcription can influence folding, and polymerase pausing at distinct sites is sometimes necessary to achieve efficient folding. Many of these concepts have been reviewed earlier [23,24], and are based on experiments with structured RNAs that serve as model systems for folding.

Evidence for co-transcriptional folding playing a role in native state formation first came from studies of the Tetrahymena self-splicing intron, a ribozyme that resides in the gene for the large ribosomal subunit of the ciliate and catalyzes its own removal from the transcript. Intron excision is limited by folding of the RNA and can be easily monitored by detecting the joined exons. In vivo folding was an order of magnitude faster and occurred within seconds, whereas re-folding in vitro takes several minutes [25,26]. Heilman-Miller and Woodson later revealed the origin of this phenomenon—as predicted earlier, local stem loops formed already during transcription, preventing entry into dead-end folding pathways [27]. Additionally, the order of structural domains altered folding efficiency, as has been shown for the ribozyme RNase P [28], indicating a more complicated synergy between folding and transcription that is not purely explained by sequential structure formation. Importantly, Pan et al. identified another variable: transcriptional pausing, the temporary stalling of RNA Polymerase during elongation. Pausing occurs site-specifically in bacteria and is regulated by proteins like the bacterial elongation factor NusA. The authors showed that by NusA-dependent pausing at specific favorable sites, co-transcriptional folding of RNase P is accelerated [28].

RNA polymerases pause during transcription

A major step in eukaryotic RNA Polymerase transcription regulation is to overcome pausing 30–60 nucleotides from the initiation site (promoter-proximal pausing). While the mechanism of this stall is not entirely clear, genomic methods such as chromatin immunoprecipitation (ChIPseq) and global/precision run-on sequencing (GROseq/PROseq) that can map RNA Polymerase position onto genes revealed that promoter-proximal pausing is a widespread feature of Pol II transcription. This regulatory step is difficult to reconstitute in vitro, since many factors are thought to be involved (reviewed by Core & Adelman [29]). It has been illustrated especially through single-molecule in vitro studies that even after RNA Polymerase is released into active elongation, transcription is not at all a smooth process with constant velocity. Instead, experiments with optical tweezers have shown extensive intermittent pausing [30]. In these experiments, Pol II is attached to a polystyrene bead held in place by an optical trap. Double-stranded DNA is tethered between the Pol II bead and another bead in a second optical trap. Transcription can be monitored with single base pair resolution by monitoring the force generated by Pol II pulling DNA during elongation. With such a setup, Galburt et al. found that backtracking was the likely cause for the majority of RNA pausing events (Fig. 2a), and that recovery from pausing was facilitated by the elongation factor TFIIS. Although nucleotide addition follows a ratchet-like mechanism since pyrophosphate release is practically irreversible, according to the backtracking model, pauses are initiated by backwards movement of the polymerase on template DNA and maintained by diffusion along upstream DNA [31,32] (Fig. 2d, inset). The DNA/RNA duplex follows the movement; the ratchet remains intact [33]. A power law distribution of pause times confirmed the diffusive nature of backtracking. TFIIS assists in backtracking release [34] by stimulating cleavage of the 3’-exposed RNA and thereby realigning a free 3’-hydroxyl group with the Pol II catalytic center [35]. Although functions of the short cleavage products have been proposed [36,37], their fate still remains largely unclear.

Fig. 2: Interplay of nascent RNA structure close to the polymerase and transcription speed.

(a) Single-molecule transcription assay with optical tweezers, measuring force generated by transcribing Pol II over time. Force is proportional to polymerase position along the template. The inset shows a Pol II backtracking event. Without TFIIS, the polymerase recovers from backtracked states less efficiently. Data from [30]. (b) Same transcription assay as in (a), but DNA templates with different base compositions are used. Data from [38]. (c) Metaplot for 400 nt region around Pol I occupancy peaks in the 5’ external transcribed spacer region of rDNA repeats. Left axis: Energy of folding, where each data point corresponds to a 65 nt window ending 14 nt upstream of the respective Pol I position. ΔG at 30°C was calculated with ViennaRNA [39]. Right axis: RNA crosslinking and Pol I immunoprecipitation profile (fraction of reads ×10³). Data from [40]. (d) Model for inhibition of backtracking by nascent RNA structure forming close to the polymerase. This leads to net acceleration of transcription across regions with high structure formation potential.

Transcriptional pausing can be modulated by structured RNA

A separate line of experimental evidence suggests that the RNA product itself is involved in regulating pausing. Synthesis of the initial RNA stretch has been found inefficient until about 50 nt have been transcribed [41], of which at least 14 nt are inside the polymerase (Fig. 1). Újvári, Pal and Luse showed that pauses also arise when the nascent RNA is truncated during active elongation and proposed that secondary structures that form close to the Pol II exit channel might prevent backwards movement. This hypothesis was subsequently backed by a physical model accounting for RNA structure formation close to the exit channel, effectively shortening the space in which Pol II can undergo diffusive backtracking [42]. The model suggests that due to rapid base-pairing interactions as soon as the nascent chain exits, the length of unpaired RNA upstream of Pol II remains small. Since backtracking requires re-entry of single-stranded RNA into the polymerase, double stranded regions close to the polymerase suffice to explain the 50 nt limit for efficient elongation. It follows that promoter-proximal pauses might be of the same origin as intermittent pauses, with the difference that the latter are prevented by nascent RNA structure, dependent on the sequence of the transcribed gene and associated folding propensity. Indeed, using the same optical tweezer setup as described above, Zamft et al. found experimental evidence for this theory [38]: when comparing transcription of a GC-rich (high structure stability) with an AT-rich DNA stretch, less frequent and shorter pauses were detected for high-GC templates (Fig. 2b), whereas the pause-free velocity remained constant. This GC-dependence was abolished when RNase A was added to the reaction mix to digest the nascent chain, again implicating nascent RNA in pausing regulation and supporting the model shown in Fig. 2d. Recently, the interplay between nascent RNA folding and transcription elongation has been characterized further in Saccharomyces cerevisiae [40]. Pol I occupancy on rDNA appeared uneven in electron microscopy images of chromatin spreads and in Pol I (Rpa190) RNA crosslinking and immunoprecipitation profiles, which map the elongating 3’-end. While Pol II distributions have previously been found to be uneven [43], the Pol I profiles showed unusually distinct occupancy peaks in the rDNA 5’ region with 80 nt spacing. This region of the transcript folds into a series of stem loop structures [44]. A high free energy of RNA folding (ΔG) in a 65 nt window coincided with Pol I occupancy peaks (Fig. 2c), suggesting transcriptional deceleration when structures are weak. This correlation held for evolutionarily distant Schizosaccharomyces pombe Pol I, as well as S. cerevisiae Pol II and Pol III, therefore supporting that backtracking inhibition by nascent RNA folding indeed occurs in vivo.

Riboswitches are folded nascent RNA sensors

RNA emerging from the polymerase is immediately exposed to the intracellular environment, including a multitude of RNA-interacting proteins, but also small molecules and ions. Such ligands can intersect with the co-transcriptional folding pathway of a transcript, leading to alternative conformations where the fold depends on ligand presence in a concentration-dependent manner. Prokaryotes make use of RNA elements termed riboswitches that fold in conjunction with a ligand to control genes associated with the biochemical pathway involving that ligand. Riboswitches are evolutionarily ancient gene regulatory elements [45] and bypass the need for protein production altogether, as all information necessary for ligand recognition is encoded in the short RNA stretch. The riboswitch usually resides directly in the 5’ untranslated region (UTR) of the regulated gene. Once a folding decision (“switch”) is made, the RNA structure can regulate gene expression in several ways, including transcription termination, transcriptional interference, translation initiation, or alteration of mRNA stability [14]. For example, when fluoride ions are present, the crcB riboswitch folds into an aptamer domain and prevents the formation of a terminator harpin. Transcription proceeds and the CrcB fluoride transporter is expressed, exporting fluoride ions and reducing toxicity. In the absence of fluoride, the terminator folds and recruits NusA, which causes RNA polymerase to briefly stall and subsequently terminate due to the combination of pausing and the weak DNA-RNA hybrid in the catalytic center of the U-rich tract following the terminator hairpin (Fig. 4a). Since the U-rich tract is directly downstream of the terminator hairpin, the RNA must fold rapidly after being transcribed. The importance of the interplay between transcription and riboswitch folding has therefore been investigated extensively [46–48], but emerging methods to study RNA structure allow more detailed insights into this relationship.

Fig. 4: Co-transcriptionally formed RNA structures regulate gene expression.

(a) Riboswitches fold in conjunction with small molecule ligands. If the ligand (e.g. Fluoride) is present, formation of a terminator hairpin is prevented. In the absence of fluoride the hairpin folds, recruits NusA and the transcription complex is destabilized, leading to termination. (b) A stem loop within the intron of yeast APE2 occludes the cryptic splice site AAG under normal conditions. Under heat shock the stem is melted and alternative splicing is favored [93]. (c) Histone pre-mRNA 3′-end processing complex [11,94,95]. Pol II not to scale. (d) Alternative splicing of the HIV-1 primary transcript at the A3 splice site is regulated by coexistence of two alternative conformations involving the 3’ splice site [96], leading to regulation of viral Tat protein production. The stretch pairing with A3 is in red. D: 5’ splice sites, A: 3’ splice sites.

Co-transcriptional RNA chemical probing to study the kinetic control of folding

How folding, ligand binding and transcription of the crcB fluoride riboswitch are coordinated to make the switch decision has been studied using co-transcriptional SHAPEseq [49]. The SHAPE method (selective 2′-hydroxyl acylation analyzed by primer-extension) relies on the exclusive chemical modification of conformationally flexible nucleotides [50]. The adducts can block reverse transcription of RNA to cDNA or cause mutations in the cDNA opposite of modified positions. Reverse transcription stops or mutations can be detected traditionally by primer extension or by high-throughput sequencing, respectively (Fig. 3a). Watters et al. prepared a crcB riboswitch library that contained truncations at every nucleotide position, thus representing the entire nascent RNA landscape. After in vitro transcription and SHAPE probing, conformational flexibility of each nucleotide was mapped in relation to template length (Fig. 3b). Structural transitions manifest in interrupted vertical lines in the SHAPE reactivity heatmap and helped to reconstruct the folding pathway and explain how fluoride binding prevents terminator formation: The aptamer domain is pre-folded and then stabilized through ligand binding, which leads to delays in the early folding stages of the terminator hairpin, so that it does not nucleate until RNAP has escaped the U-rich tract. If there is no aptamer stabilization, the terminator can quickly fold. Such kinetic coupling between RNA folding and polymerase position is likely not unique to riboswitches, but affects all RNA structures that form on nascent RNA. Time—often equated with polymerase position along the gene—is therefore an important variable when studying co-transcriptional RNA folding. Though this in vitro approach only reports on structures present long after transcription has been stopped and is limited by the length of the investigated RNA, it provides nucleotide-resolution data rich enough to inform detailed simulations of RNA folding pathways [51], which help to understand how final folds can depend on the directionality of transcription. A similar experiment has also been realized in vivo: Instead of artificially producing truncated transcripts, Incarnato et al. took advantage of the naturally distributed 3’-ends of E. coli nucleoid-associated RNA to observe polymerase-position dependent RNA structure [52]. Though this DMSseq experiment provides lower resolution, parts of the in vivo folding pathway of nascent pre-rRNA could be reconstructed (Fig. 3c), providing a powerful platform for investigating co-transcriptional RNA folding in the native environment.

Fig. 3: Methods to study co-transcriptional RNA folding.

(a) Principle of RNA chemical probing using SHAPE reagents or DMS (dimethyl sulfate) combined with mutational profiling. (b) Co-transcriptional SHAPEseq shows structural transitions of the crcB riboswitch during transcription. Raw data from [49]. (c) Co-transcriptional folding pathway of E. coli 23S rRNA helix 23 determined by in vivo DMS probing of nascent RNA [52]. (d) Monitoring co-transcriptional RNA folding using single-molecule FRET. Rep-X or the 10-fold slower PcrA-X helicase are loaded onto an immobilized RNA/DNA hybrid. Upon addition of missing buffer components, the helicase unwinds the duplex and releases RNA in analogy to an elongating 3’-end during transcription. Folding is monitored in real time using FRET [59].

Single-molecule methods to study co-transcriptional folding

While SHAPE and other structural probing methods rely on averaging the signal over thousands of sequencing reads, techniques that directly detect single molecules can resolve heterogeneity between individual folding events, albeit with lower resolution. Single folding events of an adenine riboswitch (pbuE) were first observed in an optical tweezer setup, where transcribing RNA polymerase was tethered to one bead in an optical trap, and the transcript 5’-end linked to another [53]. From these experiments, it was obvious that riboswitch control occurs far from equilibrium and is strongly dependent on transcription, offering only a small time window in which the ligand can stabilize the aptamer conformation and prevent timely terminator folding. Like optical tweezers, FRET (Förster resonance energy transfer) provides accurate temporal resolution and can help determine the fidelity of a folding pathway, or the longevity of a potentially functional intermediate state. FRET relies on the non-radiative transfer of energy from an excited donor fluorophore to a spectrally red-shifted acceptor, which is highly distance-dependent [54]. Since it reports on distances in the low nanometer range, it has been used extensively to study structured RNAs such as the ones found in the spliceosome [55,56], as well as riboswitch folding [57,58]. Recently, an elegant way to study co-transcriptional RNA folding on the single-molecule level was introduced [59], based on the engineered, highly processive helicase Rep-X that unwinds an RNA/DNA heteroduplex (Fig. 3d). Rep-X is loaded onto an immobilized DNA/RNA duplex in absence of MgCl₂ and ATP, which are required for RNA folding and helicase activity, respectively. These components are then added to the buffer and the helicase unwinds the duplex analogous to transcription by RNA polymerases (Fig. 3d) and successively releases RNA towards the 3’-end. The RNA that is released during this process is probed for structural changes using FRET between the two attached fluorophores.

The engineered Rep-X helicase is similar in speed to bacterial RNA polymerases, but the slower variant PcrA-X [60] can be used instead; in fact, experiments with both helicases can be compared to infer differences in folding as a function of unwinding speed [61]. Interestingly, a considerable fraction of many riboswitches is ligand unresponsive. Co-transcriptional FRET measurements implicate folding intermediates that prevent proper folding of the aptamer domain (leaky termination) or delayed folding of the terminator (leaky expression). Slower transcription reduced this fraction, demonstrating again the kinetic nature of this process. Though such studies are far from recapitulating a cellular environment and interactions between artificial components of the system and the target RNA cannot be prevented, reaction conditions can be tightly controlled and contributions from many variables disentangled. Notably, the in vivo structure of a riboswitch can indeed depend on cellular conditions [62]—live observation of RNA folding directly in cells is therefore an important milestone yet to be reached. Although riboswitches are attractive models for co-transcriptional regulation through RNA due to their length and distinct switch behavior, single-molecule techniques as discussed above are expected to reveal mechanistic detail for other co-transcriptional RNA folding processes in the future. As the mechanism of gene regulation through riboswitches is rapid and requires minimal cellular resources, one would expect to find them throughout all domains of life. Yet, the only riboswitch known so far in eukaryotes is unique to some filamentous fungi and regulates the process of alternative RNA splicing [63].

Transcription and Splicing are interdependent

Splicing, the process of removing non-coding regions (introns) from pre-mRNA, begins during transcription. Spliceosomal components associate with nascent RNA as soon as cognate splice sites exit the polymerase, allowing the spliceosome to assemble and become active before termination [64]. The proximity of nascent RNPs to the DNA axis give the spliceosome, chromatin, and Pol II opportunities to interact [65], resulting in close coordination of the two processes [66,67]. Intron-containing genes are among the most highly expressed, and insertion of introns into transgenes promotes transcription [68,69]. Promoter-proximal introns enhance transcription and determine active chromatin profiles in human cells, and inhibitions of splicing drastically reduces Pol II transcriptional output [70]. Observations that the speed of elongating Pol II influences the recognition of alternative splice sites supports a hypothesis that “kinetic competition” between splicing and transcription determines splicing patterns [71,72]. In addition, Pol II elongation responds to splicing cues. Average elongation rates within introns are faster than those within exons [73,74]; possibly local differences in elongation rates depend on gene architecture or structure formation potential. Recent transcriptional analysis in mammalian cells indicates that Pol II elongation is constant across splice sites and that the spliceosome often removes introns while Pol II is still in the downstream exon [34,75]. Our lab discovered a specific site of pausing – terminal exon pausing that occurs in short last exons in wild-type S. cerevisiae [76], and several labs have reported Pol II pausing at 5’ and/or 3’ splice site [77,78]. These observations suggest that transcription and splicing rates are matched under normal conditions of cell growth [79,80]. How pausing in last exons may occur, potentially to ensure the completion of splicing, remains to be discovered. Previously, pronounced transcriptional pausing was only known to occur near promoters and at polyA cleavage sites [81,82]; these pause sites are also the locations of two other RNA processing events: 5’-end capping and polyA cleavage, respectively [8], consistent with the intimate relationships between the initiation and termination of transcription elongation with co-transcriptional RNA processing events.

Nascent RNA structure can regulate alternative splicing

Alternative splicing can produce a set of different exon combinations from a single gene. This process can be regulated by the occupation of certain splice sites through proteins to prevent or promote access by the spliceosome. SR proteins (serine, arginine rich) and hnRNPs (heterogeneous nuclear ribonucleoprotein particles) recognize splicing regulatory elements (SREs) in splice site vicinity and can promote or inhibit usage of the site [83–88]. This regulatory principle in higher eukaryotes expands the proteome a single gene can encode and contributes drastically to genome complexity. Components of the spliceosome, specifically the RNA unit of U1 and U2 snRNPs recognize 5’ and branch sites by forming base pairs with the transcript. If those sequences are involved in intramolecular base pairing instead, alternative sites that remain single stranded might be favored by the splicing apparatus, or splicing might be inhibited altogether (Fig. 1). Conversely, splicing can be aided by intronic structures guiding spliceosome assembly or exposing specific signals. This RNA structure-based modulation of splicing represents additional means to SR proteins and hnRNPs. Many instances of such regulation have been found throughout a diverse set of species, and have been previously reviewed [64,85,89]. Genome-wide profiling of Arabidopsis nuclear RNA indicated widespread regulation of alternative splicing by RNA structure [90]. In addition to directly interacting with splice signals, structures in nascent RNA have recently been shown to also affect SR protein binding. In mouse cells, retained introns were found to feature decreased free energy of RNA folding in the exon upstream when compared to efficiently spliced introns. A plausible explanation is that SR protein binding sites that would usually enhance splicing are occluded, but protein binding can also remodel structures that formed [91]. Increased folding potential just upstream of 5’ splice sites has also been observed in plants [92].

In some cases alternative splicing regulation through RNA structure can be coupled to environmental sensing, reminiscent of the instant feedback bacterial riboswitches provide. In S. cerevisiae, splice site choice of the APE2 gene is influenced by the temperature-dependent formation of a stem loop [93]. The structure occludes an alternative splice site under physiological temperature, but is melted under heat shock (Fig. 4b).

S. cerevisiae possesses only ~300 intron-containing genes, most of which contain a single intron that is spliced co-transcriptionally as Pol II is just downstream of the 3’ splice site [79]. Systematic analysis of the folding potential within these yeast introns suggested that shortening of the effective distance between branch site and 3’ splice site through structure formation assists in splice site choice [93], yet folding of intronic regions in vivo remains to be experimentally verified. Cryptic 3’ splice sites reside in regions with lower predicted average free energy of folding when compared to the competing ones within the same gene [40,93], possibly facilitating correct splice site choice through at least two mechanisms: inaccessibility of the cryptic sites directly due to base pairing, and transcriptional acceleration of Pol II due to a reduction of backtracking (see above), which in turn shortens the kinetic window in which co-transcriptional splicing can occur.

Structure signals in nascent RNA can be specifically recognized

Apart from a direct effect of structures in nascent RNA, motifs such as small stem loops can serve as recognition signals for specific target proteins. A well-studied process for such signaling is the 3’-end processing of histone genes. Although also transcribed by Pol II, histone genes are terminated through a distinct mechanism that requires stem loop binding protein (SLBP) to recognize a hairpin structure on the nascent transcript. This interaction combined with recruitment of additional factors and base pairing of U7 snRNP with unstructured sequence downstream of the cleavage site (Fig. 4c) concludes formation of the termination complex [94]. Similar to the kinetic control of riboswitches described above, hairpin folding is tied to elongation. When transcription is slowed down in human cells by UV treatment or Pol II mutant R749H, the stem loop cannot fold—likely because surrounding sequence has the opportunity to compete for folding given the increased time [97]. Similarly to SLBP, the S. cerevisiae ribosomal protein L30 binds to a structured RNA motif within its own transcript: RPL30 forms a kink-turn motif structure mimicking the L30 rRNA binding site near the 5’ splice site. Upon binding, the structure is stabilized while transcription is still ongoing [98] and U2 association with the branch site blocked, thus targeting the transcript for degradation. This co-transcriptional negative feedback loop regulates L30 expression numbers. Whether an RNA-interacting protein binds to a specific sequence or a structure motif is difficult to discern using classical RNA-protein crosslinking experiments.

Therefore, it can be expected that more instances of such regulation will be discovered in the future, as new combinatorial techniques emerge (see Perspectives).

In vivo chemical probing identifies co-transcriptional strategies tied to RNA folding

In the past decade, high-resolution in vivo structural probing methods, including SHAPE and DMSseq (Fig. 3a), have enhanced our understanding of RNA in the cellular context. A recent study reported genome-wide RNA structure maps for mammalian cellular compartments [99]. Intriguingly, Sun et al. found an increased structure signature in RNA associated with the chromatin fraction, particularly intronic regions. The same phenomenon has been observed in Arabidopsis nuclear RNA [90]. This stands in contrast to the in vitro re-folding scenario, where introns and exons were similar in folding potential [99]. It follows that cellular conditions contribute to more structured introns. Interactions between nascent RNA, RNA polymerase, spliceosome and processing factors are likely to play a role, but the details remain poorly understood. Introns are mostly unconstrained by the genetic code, which is why evolutionary emergence of sequences favoring particular secondary structures in these regions might be facilitated. Supporting this notion, a study showed that RNA structure elements emerged within introns to regulate splicing in an artificially evolved system [100].

Importantly, in vivo structural probing can be carried out in a targeted fashion, which is especially useful since genome-wide experiments usually suffer from overrepresentation of abundant transcripts. Targets of interest can be enriched either before [101], during or after reverse transcription [102]. Some experimental designs involve enrichment of targets before chemical probing, a practice that should generally be avoided since RNA folds are susceptible to buffer change, temperature, and extended incubation periods. Extremely high sequencing read depths in targeted structural probing experiments can be used to extract sub-populations of folded RNAs to detect alternative conformations [96,103]. With a new clustering algorithm, Tomezsko et al. discovered an alternative conformation in the HIV-1 proviral RNA which was previously undetected due to population averaging (Fig. 4d). Similarly to examples described above, an alternative splice site is differentially occupied, contributing to Tat protein level regulation [96]. Another algorithm that is applicable to detecting structurally heterogeneous RNAs even in transcriptomic chemical probing datasets was developed recently [104]. Alternative RNA conformations therefore add another layer of regulatory complexity to the folding landscape of nascent RNA.

Perspectives

Despite these exciting advances, the question of what RNA does while it is transcribed remains largely open. Single-molecule studies require separation of target RNA and cellular environment; in turn, in vivo methods often leave important questions unanswered. Was a structural transition protein-induced? Is an unstructured region occupied by RNA-binding proteins? Is a particular RNA structure of functional relevance? Indeed, ~30% of any random sequence will form intramolecular base-pairing interactions. The scarcity of nascent compared to mature RNA, as well as time-dependence of transcription additionally hamper the investigation of co-transcriptional events. For example, it is unknown to what extent RNA splicing depends on the structure of the substrate. However, considering the directionality of transcription when computationally predicting RNA structure can enhance the interpretation of experimental findings [105], and new strategies combining computational modeling and experimental data are promising [51]. New means of detecting both protein binding and RNA structure are being developed [106,107], and will finally enable studies of coordination between RNA folding and transcription.

Research Highlights.

• Nascent RNA starts folding as soon as it exits RNA polymerase.

• Transcription kinetics modulate the folding pathway.

• RNA Structures close to the polymerase can reduce backtracking.

• Structures serve as recognition signals or can inhibit recognition of sequence motifs.

• Co-transcriptionally folded RNA regulates splicing in eukaryotes.

• Chemical structure probing and single-molecule methods reveal new functions of nascent RNA.