Co-transcriptional gene regulation in eukaryotes and prokaryotes

Abstract

Many steps in RNA biogenesis occur during synthesis by RNA polymerases. Co-transcriptional activities are deemed commonplace in prokaryotes, where the lack of membrane barriers allows mixing of all gene expression steps. In the past decade, an extraordinary level of coordination between transcription and RNA processing has emerged in eukaryotes. Here, we discuss recent developments in our understanding of co-transcriptional processing in both prokaryotes and eukaryotes, comparing methodologies and mechanisms. We point out striking parallels in how RNA polymerases interact with the machineries that act on nascent RNA. The development of RNA sequencing and imaging strategies that detect transient transcription and RNA processing intermediates has facilitated discoveries regarding the coordination of transcription with splicing, 3′ end cleavage, dynamic RNA folding, as well as physical contacts among processing machineries and RNA polymerase. Such studies have revealed that intron retention across an individual nascent transcript can prevent 3′ end cleavage, resulting in readthrough transcription that, along with other changes, is a hallmark of eukaryotic cellular stress responses. We highlight how co-transcriptional coordination drives fundamental aspects of gene expression in both prokaryotes and eukaryotes.

Introduction

The lack of a nuclear envelope in prokaryotes enables co-transcriptional translation, in which ribosomes bind to the 5′ end of nascent bacterial mRNA as soon as the ribosome binding site exits RNA polymerase. Direct electron microscopic visualization of co-transcriptional translation showed this basic mechanism that is restricted to prokaryotes¹. In the intervening decades, our presumptions about co-transcriptional availability and processing of nascent RNA have been underpinned by seminal studies in prokaryotes. The technique of native chromatin visualization, or “Miller spreads”, also facilitated numerous advances in eukaryotic gene expression by imaging nascent RNA attached to its DNA template through the stability of ternary complexes between RNA polymerases, DNA, and nascent RNA². These discoveries included the co-transcriptional association of the pre-rRNA processing machinery on nascent pre-rRNA in frog and yeast cells^3,4. Images of spliceosomes on nascent transcripts of the highly transcribed chorion genes of Drosophila melanogaster^5,6, highlighted another difference between evolutionary kingdoms by pointing to a co-transcriptional process that is unnecessary in prokaryotes, which lack introns. That said, there are many reasons to reconsider our dogmas and, additionally, to expect similarities in co-transcriptional processes between prokaryotes and eukaryotes.

The chemistry of RNA is the same across all forms of life, and RNA’s tendency to engage in base-pairing interactions and form structures is universal. As such, one of the most exciting emerging themes in co-transcriptional gene regulation is the question of how RNA structures form during transcription and later mature, how they sense temperature changes or other physicochemical cues, and how helicases remodel them to influence processing steps (Box 1). The impact of RNA structure on co-transcriptional RNA processing steps is currently the subject of intense research and can be separated into two basic modes: (i) specific, evolutionarily conserved RNA elements can directly influence transcription and processing, or (ii) the tendency to form non-specific structures through base-pairing of even random sequence and therefore “structure formation propensity” can feedback on RNA polymerase elongation⁷. Both of these concepts first arose from investigations into prokaryotic systems. For example, riboswitches are co-transcriptionally formed, conserved RNA elements that can directly intervene in the process of transcription and are widespread among prokaryotic lineages⁸. They can fold into two distinct conformations depending on the availability of specific small molecule ligands and thus sense cellular environments. Eukaryotes make use of similar structure-based regulation. For example, termination of histone gene transcription requires co-transcriptional stem-loop formation and recognition by stem-loop binding protein (SLBP); U7 snRNA then base pairs with the histone pre-mRNA 3’ end and recruits the termination complex^9,10 (Box 1). This folding event is sensitive to the rate of transcription elongation, demonstrating the kind of feedback RNA structure formation can provide to both transcription and processing¹⁰.

Box 1 – Local nascent RNA structure formation and base modification is co-transcriptional across kingdoms of life.

Understanding dynamic Watson-Crick base-pairing is a current major goal in the field because structures like riboswitches or stem-loops impact transcription elongation and termination as well as RNA processing⁷. In eukaryotes and prokaryotes (lower left and right panels, respectively), nascent RNA stem-loop formation leads to 3′ end cleavage and termination of replicative histone genes and operons dependent on intrinsic terminators^10,231. In bacteria, nascent RNA can fold into stable secondary and tertiary structures through intramolecular base-pairing as soon as it emerges from the polymerase, and nascent RNA folding affects polymerase speed^{156,232–234}. Nascent RNA compaction through local folding would be a general mechanism for limiting the availability of splice sites or PASs to the splicing and cleavage machinery as transcription elongation proceeds. To measure in vivo base-pairing, Dimethyl Sulfate (DMS) and SHAPE reagents modify unpaired nucleobases at the Watson-Crick face or the 2′-hydroxyl backbone position, respectively, thereby monitoring base flexibility; sites of modification are identified by incorporation of a mismatched base during reverse transcription of the RNA to cDNA²³⁵. In DMS mutational profiling (DMS-MaPseq) and SHAPE-MaP, transcriptome-wide or target-specific mutational profiles can be used to inform RNA structure prediction algorithms^236,237. Like structure-probing, some RNA modifications can be specifically detected by chemical treatments. For example, CMC forms stable, covalent adducts with the N1 and N3 positions of pseudo-uridine, enabling identification of modification sites in RNA-seq data²³⁸. BID-seq maps pseudo-U by bisulfite treatment, which induces deletions during RT that can be mapped²³⁹. Direct RNA sequencing is also a promising alternative approach to reading endogenous RNA modifications²⁴⁰ and chemical changes sensitive to base-pairing^241,242. Recent studies in eukaryotes have concentrated on isolating bulk RNA from different cellular compartments or refolding nascent RNA after purification from cells^243,244. Future methods may be able to capture local nascent RNA structure formation in eukaryotes.

This review focuses on recent discoveries on the coordination of pre-mRNA processing with transcription, mostly in eukaryotic organisms. Unlike other reviews on this topic, we aim to awaken interest in this field to all biologists by considering similarities and differences with prokaryotic RNA processing (Figure 1). Some readers may be surprised to learn that translation is not always tightly coupled to transcription in bacterial species (Figure 1a,c)^11,12 and that RNA stability is now known to play a prominent role in tuning the transcriptome in some conditions and/or species of bacteria^13,14. Prokaryotic gene architectures are more complicated than previously believed, contributing more complexity to transcriptomes^15–18. Issues of RNA polymerase initiation, elongation, and pausing are topical in eukaryotes and prokaryotes^19,20. RNA processing in prokaryotes and eukaryotes both involve 5′ end capping: cap-like elements, such as NAD⁺/NADH and 3′-dephospho-coenzyme A (dpCoA), can be incorporated at 5′ ends during transcription initiation in prokaryotes, whereas caps made of non-templated GTP are added enzymatically and methylated exclusively in eukaryotes (Figure 1c,d). Despite these differences, the regulatory significance of 5′ end caps – protection from degradation and feedback to downstream processing events (see gray arrows, Figure 1c,d) – reveals commonalities between prokaryotes and eukaryotes. For the latter, many of these effects are mediated by the nuclear cap-binding complex²¹.

Figure 1. Organization of gene expression machineries and co-transcriptional processes in pro- and eukaryotic cells.

(a-b) Schematic of prokaryotic and eukaryotic cells drawn to scale (e.g., E. coli 2 μm long, HeLa cell 30 μm long²¹⁵), showing cellular organization with membrane-bound (nuclear envelope in eukaryotes) and membrane-less compartments (e.g., nucleolus, speckle, Cajal and P bodies, and stress granule). Gene expression is spatially organized in prokaryotes, with the nucleoid containing DNA and associated proteins in the center. E. coli cell and line plot based on imaging data^{28,44,46,47,216}. Black arrows indicate molecular exchange by diffusion (a) and nuclear export (b). (c-d) Co-transcriptionality of gene expression processes in pro- and eukaryotes enables cross-regulation, indicated by light arrows. Multiple (possibly overlapping) coding sequences (CDS) can be encoded within one operon in prokaryotes. In eukaryotes, a single transcription unit includes exons and introns, yielding a single CDS after pre-mRNA splicing. Alternative 5′ and 3′ ends, as well as cap(-like) moieties and RNA folding are common in both systems, mechanistic differences/commonalities given as bullet points within the figure. Translation initiates co-transcriptionally in prokaryotes, but the proximity of the ribosome to RNA polymerase varies: 1^st RNA polymerase and ribosome resemble traditional view in E. coli, 2^nd RNA polymerase and ribosome reflect emerging view in different species and conditions^11,12. In eukaryotes, pre-mRNA splicing often occurs co-transcriptionally.

Eukaryotic precursor mRNA (pre-mRNA) splicing faces the challenge of removing 90% of the transcript in the form of introns^22–24, while prokaryotic transcription units have the challenge of vastly overlapping gene architectural features and the need to express proteins from operons that contain multiple coding sequences. 5′ end capping, splicing, and 3′ end cleavage (Figure 1b,d) are the best understood eukaryotic co-transcriptional RNA processing steps, the regulation of which still harbors enormous mystery. To enrich background knowledge of splicing, several excellent reviews have been published within the year^22–24. Here, we will discuss how recent sequencing (Box 2) and imaging approaches (Box 3) have identified highly efficient co-transcriptional splicing as well as delayed splicing and the lingering presence of the longest introns in metazoans. In the past five years, specific strategies for sequencing the transient intermediates of co-transcriptional RNA processing in vivo have revealed the position of RNA polymerase II (Pol II) where processing – chiefly pre-mRNA splicing and 3′ end formation – occurs; these experiments have revealed both timing and coordination among events. We have learned how the efficiency of co-transcriptional RNA processing can impact overall gene output and how these steps are modulated by environmental stresses. Emerging knowledge about folding, modifications, and editing in eukaryotes adds to this already complex landscape. Evidence that processing events are coordinated with each other as well as transcription, suggests that these regulatory steps combine to determine the fate of each transcript and the overall shape of the transcriptome. Co-transcriptional processes, including transcription-translation coupling, are also prominent in Archaea and the comparison to current knowledge in bacteria was reviewed previously²⁵. This review mainly considers results in bacterial species for simplicity.

Box 2 – Long-read sequencing approaches to study co-transcriptional splicing kinetics.

Illumina is the most prevalent short-read sequencing platform for transcriptomics due to its high accuracy and reproducibility²⁴⁵. However, its maximum read length (300 bp) is shorter than the typical human mRNA length (~2.5 kb), preventing capture of full-length mRNAs²⁴⁵. Two recently developed long-read sequencing (LRS) technologies overcome this limitation by achieving read lengths up to 50 and 10 kb, respectively: Pacific Biosciences (PacBio) and Oxford Nanopore Technologies (ONT)²⁴⁵. In PacBio, immobilized DNA polymerases read circularized templates and incorporate fluorescent nucleotides. Fluorescence emissions are recorded over time and used to determine sequence^245,246. In ONT, molecular motor proteins pass single-stranded DNA or RNA molecules through nanopores embedded within a membrane, resulting in changes to an electric current that can be interpreted into sequence^245,247. While PacBio reads have greater accuracy in terms of per-base quality and splice site detection, making PacBio better for detecting novel isoforms, ONT offers higher yields, making ONT better for estimating expression²⁴⁸. LRS has mainly been used for transcriptome diversity analysis in prokaryotes^69,249–251; in eukaryotes, it has developed as a strong tool to study co-transcriptional splicing kinetics.

Three LRS-based approaches to study co-transcriptional splicing kinetics include LRS of nascent RNA^103,252, nano-COP^109,110, and POINT-nano¹⁰⁴. Each method begins with nascent RNA isolation followed by LRS to capture full-length nascent transcripts, allowing splicing status to be evaluated relative to read 3′ ends, which mark Pol II position. Although each workflow incorporates chromatin fractionation, they take different approaches to extract nascent RNA; the nascent RNA LRS method uses depletion of ribosomal and polyadenylated RNA^103,252, nano-COP uses metabolic labeling^109,110, and POINT-nano and mNET-seq use immunoprecipitation with anti-Pol II antibodies^104,253,254. In principle, each method can be applied to all living systems with the limitation that nucleotide analogs are difficult to introduce into prokaryotes and eukaryotes with cell walls.

Box 3 – Microscopy-based methods to study nuclear organization and co-transcriptional RNA splicing kinetics.

Our understanding of co-transcriptional RNA processing has been largely informed by imaging methods that provide spatial and temporal context to pre-mRNA cleavage. Early micrographs of chromatin spreads depict RNA emerging from Pol II already dotted with protein complexes, and shorter than expected due to intron removal⁵. Now, 35 years later, highly sophisticated fluorescence microscopy and single-molecule techniques have been employed to explore how nuclear architecture affects pre-mRNA processing and vice versa, spatiotemporal dynamics of splicing factors binding to nascent transcripts, and how splicing kinetics impact final mRNA products^{107,222,224,228,255–257}. Fluorescence in situ hybridization (FISH) has been used extensively to visualize RNA and DNA localization in fixed cells, revealing how nuclear organization affects pre-mRNA processing. In recent work investigating the DNA structure of long genes in mouse cells, the use of FISH probes tiled over DNA within and flanking genes revealed stiff loop structures that are decorated with RNA-binding proteins in genes undergoing active transcription²⁵⁸. This work highlights a situation in which transcription and the persistence of intronic RNA could alter the surrounding nuclear environment (left panel, open looped conformation vs. closed conformation). Studies using live cell imaging have also provided key insights into the heterogeneity of splicing efficiency in single cells^222,228. This work primarily utilizes tandem tagged viral RNA-binding proteins that bind selectively to their stem-loop RNA counterparts, facilitating direct visualization of pre-mRNA. Combining PP7-mCherry-binding stem-loops and GFP-MS2-binding stem-loops along a gene allowed for direct measurement of intron removal in cells, revealing a mixture of co- and post-transcriptional splicing. Interestingly, the rate of intron removal was higher when transcripts were freely-diffusing vs. chromatin-bound²²⁸, indicating that the nuclear microenvironment can influence pre-mRNA processing speeds. More recently, time-resolved methodology was more systematically employed in the context of multiple endogenous gene sequences¹⁰⁷. Again, wide variations in splicing rates of individual long introns (indicated by the appearance and subsequent disappearance of fluorescent signal) were observed and the authors report recursive splicing in multiple introns (right panel). APEX2-sequencing and 3D study of genome organization have further shown that long introns tend to associate with the periphery of the nucleus, while short introns associate with the center^259,260. Microscopy-based methods offer the advantage of measuring pre-mRNA processing steps in single, living cells and can provide spatial and allelic context to biophysical measurements. Images adapted with permission from Refs^107,258.

Co-transcriptionality in prokaryotes

The lack of a nucleus and small cell size in prokaryotes brings all steps of the central dogma into close proximity, which allows for “co-transcriptionality” of translation, RNA processing, and interplay between 5′ and 3′ end formation (Figure 1a,c). Compared to eukaryotic mRNA metabolism, prokaryotic mRNA production and mRNA decay are substantially faster – on the order of seconds to minutes^13,26–31 – raising the question of how prokaryotes ensure sufficient mRNA production for growth and proliferation. Typical bacterial mRNA synthesis time is shorter than mRNA half-lives^27–29, and consistently most translation occurs on full-length mRNAs and is thus post-transcriptional¹⁴. Translation of mature full-length mRNAs requires the completion of transcription across the whole gene. This process is strongly facilitated by co-transcriptionally initiating ribosomes, resulting in transcription-translation coupling that prevents access of RNA decay enzymes to the nascent RNA, deleterious R-loop formation, and premature transcription termination^13,31,32. For example, experiments using bacterial T7 phage-derived RNA polymerase that is substantially faster than Escherichia coli’s RNA polymerase to transcribe E. coli genes showed strong destabilization of transcripts, arguing for the necessity of transcription-translation coupling to delay mRNA decay³³. Tight transcription-translation coupling, including physical contacts between RNA polymerases and ribosomes³⁴, is considered important to ensure effective translation (Figure 1c). This is based almost exclusively on studies of the gram-negative gut bacterium E. coli^34–37. However, millions of prokaryotic species exist³⁸ that could differ in their gene expression strategies. Indeed, dynamics can vary in different environments and among species, revealing that tight transcription-translation coupling is not universal across prokaryotes or growth conditions^{11,12,30,39,40} (Figure 1c). By using a ribozyme-assay distinguishing the first round from subsequent translation rounds, transcription-translation coupling was found to be variable and stochastic for different genes even in E. coli⁴¹.

Distinct subcellular organization of prokaryotic cells enables variations in gene expression crosstalk. In E. coli, the circular chromosomal DNA localizes to the cell center as a nucleoid with extensions to the periphery that can enable transertion, the co-transcriptional translation and insertion of membrane proteins^42,43. RNA polymerases localize to the nucleoid (Figure 1a). The degree of nucleoid compaction varies among bacterial species, changing cellular biophysical properties, macromolecular mobilities, and the three-dimensional distribution of ribosomes and chromosomal DNA towards the cell periphery and center, respectively⁴⁴. In addition, the RNA degradosome (Figure 1a), a multiprotein complex with decay-associated enzymes, localizes to the cell membrane or in cellular foci^28,45–47. Transcripts co-localizing with high concentrations of RNA decay enzymes exhibit shorter half-lives^28,45. Taken together, synthesis- and decay-intermediates, the extent of transcription-translation coupling, and sub-cellular organization are emerging regulatory mechanisms that diversify prokaryotic transcriptomes.

Generation of RNA isoforms from operons

Prokaryotic operons often encode multiple protein-coding genes in polycistronic transcription units that can yield multiple isoforms with a complexity resembling eukaryotic genes (Figure 1b,d). Transcript isoform diversity in prokaryotes is shaped by alternative transcription initiation, termination, and RNA processing mechanisms (Figure 1c). Less than 20% of operons have one defined transcript start and end in E. coli¹⁴, the rest harboring multiple transcription start and end sites^15–17. The first and most 5′ nucleotide in the nascent transcript carries a triphosphate that can stabilize the transcript as does the cap on eukaryotic transcripts⁴⁸. Enzymatic conversion to a monophosphate destabilizes the transcript by creating a suitable substrate for the decay-initiating RNA endonuclease that preferentially binds and cleaves monophosphorylated RNAs^49,50 (Figure 1c, arrow 1). Recently, several other non-canonical 5′ end moieties have been described in pro- and eukaryotes^51,52, such as NAD⁺/NADH and 3′-desphospho-coenzyme A, that are all introduced as a first nucleotide during transcription and implicated in downstream RNA metabolism^53–56. Stable 5′ ends can also be created by RNA processing in prokaryotes, where the same enzymes that can initiate mRNA decay generate cleavage products that have enhanced stability (Figure 1c, arrow 2), e.g. by stabilizing secondary structures and protein binding^16,57–59. Alternative transcript 3′ ends can be formed by transcription termination mediated by the hexameric helicase Rho, intrinsic terminators, and/or RNA processing through RNA endo- or exonucleases^60–65 (Figure 1c, Box 1). Especially transcription termination events can be preceded by RNA polymerase pausing⁶⁶. Taken together, individual operons can have multiple 5′ and 3′ ends, often associated with different mRNA stabilities, RNA secondary structure and number of CDSs within polycistronic operons^15,18,57,67. Because many operons span several kilobases, synthesis time can exceed mRNA half-life yielding co-transcriptional mRNA decay. This is common for long operons in E. coli, where measurements of both processes have been made, and few full-length mRNAs are produced^14,27,68. General rules for differential 5′ and 3′ end definition, possible cross-talk between 5′ and 3′ ends, and the precise effects of cis- and trans-regulatory factors currently remain unclear⁶⁹. These revelations highlight the need to rigorously characterize the degree to which different mRNA processing steps are co-transcriptional in prokaryotes, as has been done for rRNA processing⁷⁰.

Co- and post-transcriptional translation

The fate of nascent transcripts is influenced by the binding of ribosomes, local RNA structures, small RNA (sRNA) binding, transcription pausing and protein binding to RNA polymerase. The leading ribosome closely trails RNA polymerase once the required cis-regulatory elements⁷¹ become accessible and when translation and transcription elongation rates are matched and synchronized, e.g. by RNA polymerase pausing^29,31,66,72. In tight coupling, the ribosome can physically contact RNA polymerase and leave no space for RNA structure formation, sRNA or protein binding, RNA cleavage or Rho helicase binding^34,35,37 (Figure 1c, 1^st RNA polymerase). Recent work demonstrated that Rho helicase initiates transcription termination either by binding cis-elements in the nascent RNA (rut sites) and then “catching up” to RNA polymerase with its ATP-dependent translocase activity or by direct RNA polymerase interactions, followed by binding of nascent RNA through rut sites. Both Rho engaging ways antagonize contacts between RNA polymerase and ribosomes and point to multiple Rho-dependent mechanisms to transcription termination^{63–65,73–75}. Interestingly, these distinct mechanisms can occur to different extents at any given transcription termination site^76,77.

In Bacillus subtilis, a gram-positive soil bacterium and prokaryotic model system, RNA polymerase is about twice as fast as ribosomes, thus outpacing co-transcriptionally initiating ribosomes by “runaway transcription” and decoupling transcription and translation^11,12 (Figure 1c, 2^nd RNA polymerase). Hence, B. subtilis and E. coli are inherently different with respect to the co-transcriptionality of translation; this is accompanied by a less important role of the Rho helicase as a transcription terminator⁷⁸. Riboswitches, RNA-binding proteins and other transcription factors, such as NusA, take a more prominent role aiding transcription attenuation in that species^11,79. Also, the identity and activities of RNA decay enzymes differ among species^80,81. For example, RNA exonucleases can degrade mRNA from both ends in B. subtilis, whereas only 3′−5′ exonucleolytic mRNA decay has been detected in E. coli. Variations in RNA decay mechanisms together with faster transcription elongation could explain why B. subtilis transcriptomes are made up of fewer nascent and/or decaying RNAs than E. coli transcriptomes¹⁴. Furthermore, how closely ribosomes follow RNA polymerase affects 3′ end formation (Figure 1c, arrow 3). For example, intrinsic terminators evolved to be positioned further downstream from stop codons in E. coli than in B. subtilis, possibly ensuring that closely trailing terminating ribosomes do not occupy nascent RNA regions required for the formation of intrinsic transcription terminators^11,82. Crosstalk between ribosomes is exemplified by intricate spatial organization of polysomes and ribosome collisions that trigger translational quality control^83,84 (Figure 1c, arrow 4). Lastly, co-transcriptional gene regulation by sRNAs can enable downregulation by Rho termination or provide a kinetic advantage over post-transcriptional regulation in prokaryotes, for example^85,86. Specifically, co-transcriptional binding of repressive sRNA complexed with Hfq to the rpoS transcript is more efficient and faster than post-transcriptional binding. This is due to competition with structure formation in cis, which upregulates translation by enabling access to the ribosome binding site⁸⁶.

Taken together, co-transcriptional processes are a major determinant of mRNA fate in prokaryotes. Co-transcriptional translation can be seen as a prerequisite to enable post-transcriptional translation that constitutes the largest portion of prokaryotic translation. The prevalence of post-transcriptional translation is also evident from the short time a transcript spends during synthesis relative to its half-life^14,27–29. Additionally, an understanding of co- and post-transcriptional regulation, the interplay between translation and RNA decay, as well as cellular metabolism in different conditions reflecting the microbial lifestyle is needed to fully grasp the prokaryotic gene expression logic and diversity^14,87,88. For example, riboswitches are RNA sensors of chemical signals, such as metabolites and ions, to enable bacterial cells to respond to the availability of nutrients⁸⁹. The stability of RNA structure is enhanced during cold shock, creating a need for and RNA chaperones, RNases and DEAD-box helicases to maintain translation and open RNA states as cells acclimatize to the colder temperature⁹⁰. RNA sensors of oxidative stress, involving chemical changes in RNA nucleobases and/or RNA polymers are emerging as mediators of translational stalling and ribosome disassembly⁹¹. As we will see below, co-transcriptional processes also integrate eukaryotic stress.

Co-transcriptionality in eukaryotes

Instead of polycistronic operons, eukaryotic genes are complicated by abundant, non-coding introns that interrupt coding sequences in pre-mRNA and require removal by the spliceosome. The early discovery of co-transcriptional splicing in fruit flies⁶ placed the assembling spliceosome adjacent to DNA, chromatin, and RNA polymerase II (Pol II), suggesting that interactions among these elements of the nuclear landscape could contribute to gene regulation in vivo. One could imagine how post-translational modifications on histone tails or Pol II pausing might be associated with the efficiency, fidelity, and regulation of splicing. Cleavage of the nascent transcript’s 3′ end is required to initiate termination in eukaryotes. In addition, alternative cleavage and polyadenylation (APA) describes the use of different polyA cleavage sites that can also be defined at the ends of alternative exons. Co-transcriptional RNA structure formation has been implicated in histone 3′ end formation (Box 1) and the regulation of splicing^92,93; the only known eukaryotic riboswitches affect splicing⁹⁴. Spliceosomal RNAs recognize their target sequences through direct base-pairing interactions with the 5′ splice site (5′SS) and branchpoint at the intron’s 3′ end, thus competing with intramolecular base pairs. Such competition can promote the usage of canonically less favorable splice sites, resulting in alternative splicing. RNA base editing (e.g., A-to-I) and modification (e.g., m6A and pseudouridylation) also occur co-transcriptionally and regulate splicing by altering RNA folding and/or the affinity with which key regulators bind RNA sequences (e.g., U2AF65 binding of the polypyrimidine tract at the 3′SS^23,95–97. Thus, feats of eukaryotic RNA processing are functionally intertwined with transcription and base-pairing.

Co-transcriptional pre-mRNA splicing

Over the past decade, the field has sought to answer the questions (i) How frequently are introns removed co-transcriptionally? (ii) Is there a specific location in the gene where splicing occurs? (iii) How rapidly do spliceosomes assemble and catalyze splicing? And (iv) does pre-mRNA processing in eukaryotes feed back on Pol II as co-transcriptional translation feeds back on prokaryotic RNA polymerase? To do so, methods quantifying the progression of splicing relative to transcription have been established in a variety of systems, including yeast, flies, and mammalian tissue culture cells (Figure 2; Boxes 2 and 3). These approaches either associate the detection of splicing (ligated splice junction) with the 3′ end of the nascent read (Pol II position) or infer splicing timing from metabolic labeling data, by calculating intron half-lives to obtain a measurement in units of time. While earlier work relied heavily upon chromatin-based assays, fluorescence imaging of model genes and short-read sequencing of nascent RNA, as reviewed in Refs^98,99, the rise of long-read sequencing (LRS) methods (Box 2) has facilitated the development of new methods to estimate splicing timing. Two leading themes, discussed below, have emerged: (i) rapid co-transcriptional splicing, detected in all species examined, may indeed involve physical contact between the splicing machinery and Pol II or chromatin (Figure 3a) and (ii) delayed splicing may enable any number of gene regulatory events, such as RNA folding, editing, modification, decay and/or alternative splicing. These two modes of co-transcriptional splicing may resemble the two modes of prokaryotic co-transcriptional translation – tightly coupled vs runaway transcription – wherein physical contact between RNA polymerase and processing machinery could determine timing and flexibility.

Figure 2. Splicing estimates measured in terms of (a) Pol II position or (b) time.

Distance-based methods measure the position of Pol II where upstream exon-exon ligation is detected (distance “0” is the position of the 3′ splice site, which is required for the second step of splicing)^{5,6,100,101,103–106,109,217–220}. Time-based methods calculate intron half-life or the time required for splicing completion^{107,108,167,221–230}. The values along the gene ruler or timeline represent the half-maximum or a similar value if the half-maximum was not reported. Dashed lines in panel (a) separate rapid and delayed co-transcriptional splicing and post-transcriptional splicing. Rapid co-transcriptional splicing occurs while Pol II is transcribing early in the gene (<1.5 kb), delayed co-transcriptional splicing occurs while Pol II is transcribing late in the gene (>1.5 kb), and post-transcriptional splicing occurs after Pol II has completed transcription of the gene. Each study is color-coded according to the organism used to obtain the estimate (see key). This figure has been updated and also modified (with permission) from Refs^98,99 in recognition of the fact that splicing estimates obtained in terms of distance or time cannot yet be directly compared, due to varying transcription elongation rates within gene bodies (see text).

Figure 3. Interplay between RNA processing and Pol II transcription.

(a) Components of the RNA processing machinery interact with the Pol II C-Terminal Domain (CTD) (capping enzymes, cyan; cleavage and polyadenylation factors, green) or other areas of Pol II (U1 snRNP, orange; spliceosome, purple). CTD phosphorylation is marked with an “x”. (b) In addition to telescripting, in which U1 snRNP binding to AU-rich nascent RNA prevents usage of premature cleavage and polyadenylation (PCPA) sites¹³⁹, U1 snRNP binding to the Pol II large subunit helps to propel Pol II elongation through introns six times faster¹⁴¹. Transcriptional pausing is associated with promoter-proximal regions where capping occurs as well as with the PAS where 3′ end cleavage occurs. Pause sites must be distinguished from changing elongation rates that are faster over AT-rich introns than over CpG-rich regions and GC-rich exons. (c) These variations in elongation rates and pausing at specific sites creates overall higher Pol II densities (PRO-seq) over exons compared to introns, although more nascent RNA is produced (TT-seq) over introns due to faster RNA synthesis. (d) GC content is greater in exons and lower in introns, allowing intron/exon structure to be interpreted as a function of GC-richness¹⁴¹.

Rapid co-transcriptional splicing

Recent support for rapid co-transcriptional splicing comes from sequencing-based approaches designed to capture nascent transcripts. Single-molecule intron tracking (SMIT) relies on short-read sequencing and shows that the fraction of transcripts spliced reaches a plateau by the time Pol II reaches ~130 nt downstream of introns in budding yeast, which has many single-intron genes¹⁰⁰. LRS of nascent RNA (Box 2) demonstrated similar rapid splicing in fission yeast, which has multi-intron genes^100,101. More recent studies have focused on mammalian cell lines with more complex gene architecture. A typical mammalian gene contains 8–10 introns, with ~150-nt-long internal exons¹⁰². Given the length and abundance of introns in mammals relative to yeast, we might expect splicing to follow different kinetics in mammals. Surprisingly, a nascent LRS study in murine erythroleukemia (MEL) cells revealed rapid splicing with most (75%) distances between Pol II and the nearest splice junction being less than or equal to 300 nt¹⁰³. “Polymerase intact nascent transcript nanopore sequencing” (POINT-nano, Box 2) – another LRS approach – supports both immediate splicing, in which exon ligation takes place before Pol II can reach the downstream 5′ splice site, and delayed splicing, in which exon ligation occurs after Pol II transcribes ~2 kb of the downstream intron in human cell lines¹⁰⁴. A related method, native elongating transcript sequencing (dNET-Seq) in D. melanogaster embryos also supports immediate splicing for some transcripts and delayed splicing for others¹⁰⁵. Taken together, these studies make it clear that the spliceosome can operate quickly on 3′SSs when they emerge from Pol II, and that this is common among species.

Additional support for rapid co-transcriptional splicing comes from alternative methods, including co-transcriptional lariat sequencing (CoLa-Seq)¹⁰⁶ and live cell imaging¹⁰⁷. Using CoLa-Seq, which involves the capture and sequencing of nascent lariat intermediates, most introns were found to be spliced before Pol II transcribed 1.5 kb downstream of the 3′ splice site in human K562 cells¹⁰⁶, similar to the data from mouse and other human cells^103,104. Rapid splicing times have also been reported by a live cell imaging study in human (HBEC-kt) cells that used a single-cell, single-molecule system¹⁰⁷. Relevant limitations to consider for rapid co-transcriptional splicing measurements include possible mRNA contamination or nascent RNA fragmentation that could cause splicing to appear faster, but most of the studies discussed here control well for this. Reverse transcription limits the length of nascent transcripts that can be observed to ~6 kb, and library preparations that involve polymerase chain reaction (PCR) amplification will introduce a size bias towards shorter products, favoring rapid splicing. Nevertheless, methods that directly measure Pol II position relative to splicing are more appropriate for estimating splicing kinetics than indirect measurements, such as those that measure intron reads. This is because intron-retained transcripts remain in the nucleus and can dwell there until they are degraded or processed post-trancriptionally^106,108. These reads have unknown fates and can extend the apparent half-life of introns without representing actual co-transcriptional splicing intermediates or splicing kinetics.

Delayed co-transcriptional splicing

Importantly, work in human tissue culture cells¹⁰⁴ and in fruit flies¹⁰⁵ identified two populations of co-transcriptional splicing kinetics, both immediate and delayed splicing, indicating the two populations can be biologically relevant. Delayed splicing had been previously reported in metabolic labeling studies (Figure 2b). In addition, Nanopore-analysis of co-transcriptional processing (nano-COP, Box 2) detected abundant introns in human (K562 and BL1184) and D. melanogaster (S2) cell lines; spliced exon-exon junctions were observed when Pol II had transcribed 2 to 4 kb downstream of the intron, respectively^109,110. Prior work showing that even some of the longest introns are co-transcriptionally spliced¹¹¹, in agreement with detection of rapid splicing even among long introns by CoLa-seq¹⁰⁶, suggests that more work with greater coverage may be required to make general statements about long introns. Longer estimates of splicing times in budding yeast (2 min) have also been inferred based on RNA labeling with a uracil analog to determine half-lives for the chemical reactions of splicing¹⁰⁸. Limitations to consider for delayed co-transcriptional splicing measurements include potential bias from metabolic labeling and differences in bioinformatics analysis approaches. Metabolic labeling with uracil analogs (Box 2) enriches for U-rich transcripts, such as introns, and incorporation of analogs can impede splicing, causing the process to appear slower¹¹². Differences in bioinformatics approaches to estimate splicing dynamics can also have dramatic effects on interpretation. For example, by cumulative analysis of spliced reads¹⁰³, it was observed that splice junctions occur at similar Pol II positions in the nano-COP dataset¹⁰⁹ compared to the MEL dataset¹⁰³, both of which represent nascent RNA long reads. The disadvantage of this analysis, however, is that it does not quantify unspliced reads and likely underestimates the delayed Pol II positions¹⁰³. This highlights the need to unify analyses as much as possible to enable better generalization across genes and introns. In summary, some introns are delayed or even blocked in their removal despite the capability of the spliceosome to act immediately on other 3′SSs. The mechanisms that regulate such timing differences are currently unknown.

Transcription rates affect splicing

Strong evidence for co-transcriptional regulation of alternative splicing and polyadenylation encompasses associated promoter usage, transcription factors, elongation rate, and/or histone post-translational modifications^22–24. Early work on promoters set the stage for the recent discoveries that transcription start sites are associated with APA site choice and that inclusion of “new” alternative internal exons promoted the evolution of alternative TSSs^113–117. The model of “kinetic competition” between elongation rate and splice site recognition attained early support by linking high histone acetylation levels in the gene region with alternative exon exclusion¹¹⁸. Single amino acid mutations in Pol II can modify elongation rates and also show changes in alternative splicing, in directions as predicted¹¹⁹. In budding yeast, the completion of splicing is substantially delayed when the fast Pol II mutant is expressed¹⁰⁰, just as faster transcription outpaces co-transcriptional translation in some bacterial species¹¹. More physiological examples of how slower elongation rates are associated with recognition of weaker splice sites and cassette exon inclusion include responses to UV damage in human cells and light regulation in plants^120,121. Since the light cycle is also known to determine alternative splicing patterns in algae¹²², it is tempting to speculate that changes in elongation rate and alternative splicing patterns go hand in hand when organisms respond to these and other environmental conditions.

Despite these compelling arguments that co-transcriptionality plays a role in determining splicing alternatives, tackling co-transcriptional mechanisms is not easily done. The main reason is that global long-read sequencing does not yield enough reads to enable fractional measurements of alternative 5′SS, 3′SS, and cassette exon usage. CoLa-seq has greater read depth and provides insights into the question of how exon definition plays a role in alternative splicing^106,123. Given the short average internal exon size of 150 nt and the tendency for exon definition when Pol II has completely transcribed the downstream exon, it is important to know the position of Pol II relative to the exon-intron boundary when splicing begins; the authors predict most splicing events occur after Pol II has left the downstream exon creating an opportunity for classical “exon definition” through recognition of the 3′SS and downstream 5′SS encompassing one exon¹²³. That said, exon definition must be experimentally tested, since introns can be committed to splicing even if they have not yet been removed¹²⁴.

Currently, major efforts are directed at applying LRS to describing all mRNA isoforms – from 5′ to 3′ end – in cells and tissues^125,126, because this will also inform on functional differences in RNA and protein outputs. A recent study used LRS to characterize polyA+ mRNA associated with chromatin or nucleoplasm, finding evidence for post-transcriptional alternative splicing in the high proportion of partially spliced polyA+ transcripts¹²⁷. Given the insights CoLa-seq provides into the frequency of exon definition in co-transcriptional splicing and the likelihood that introns may already be committed to splicing, drawing conclusions about alternative splicing decisions based on partially spliced polyA+ transcripts with indeterminate final outcomes is risky. Nevertheless, post-transcriptional splicing is widely believed to occur in nuclear speckles and likely favors delayed splicing of alternative internal exons with characteristically weak 5′ and 3′SSs^128,129.

3′ end cleavage and polyadenylation

In eukaryotes, 3′ end cleavage and polyadenylation (CPA) triggers transcription termination and constitutes one of the last co-transcriptional steps of pre-mRNA processing prior to export to the cytoplasm (Figure 1d). Work over the last decade has uncovered dozens of proteins that carry out efficient cleavage that generally occurs around consensus motifs of AAUAAA¹³⁰. The mechanisms and factors that govern polyadenylation and cleavage have been recently reviewed¹³¹. Here we focus on the growing body of literature that highlights the role of transcription elongation in regulating poly(A) site (PAS) choice. APA site usage fundamentally alters the 3′ untranslated regions (UTRs) of mRNAs by adding or subtracting cis-regulatory elements important for the remaining lifetime of the mRNA. Indeed, APA is carefully regulated throughout developmental programs and stress responses because it impacts protein expression levels, protein–protein interactions, mRNA localization, and mRNA half-life¹³². Interestingly, APA during D. melanogaster nervous system development was recently linked to alternative promoter usage¹¹³ and alternative splicing¹³³, suggesting crosstalk between the two ends of the gene, as well as introns and reinforcing the co-transcriptional nature of regulatory mechanisms^113,116.

Transcription rates effect APA

Transcription elongation could serve as a regulatory point in which local transcription rates could dictate the amount of time a given cleavage site is available before another one is synthesized. In S. cerevisiae, S. pombe, and HEK 293T cells, transcription speed has been shown to play an important role in tandem PAS choice, with “slow” Pol II mutants favoring the use of upstream PASs^134–136 and “fast” Pol II somewhat favoring the use of downstream PASs^134,135. Moreover, Pol II transcription rate appears to primarily affect CPA within clusters of APA sites rather than shifting between distant clusters, suggesting that CPA likely occurs shortly after the emergence of the PAS from the Pol II exit channel¹³⁷. Several trans-acting factors that block PASs have now been identified. An example is the nuclear poly(A) binding protein, PABPN1, which has been shown to block the use of weak upstream PASs in U2OS cells¹³⁸. It is not yet known whether base-pairing in nascent RNA could block cleavage sites from being recognized by the CPA machinery. It is likely that APA site choice is a result of combined effects from elongation rates (e.g. how quickly the PAS becomes available), the presence of trans-acting factors, and cis-regulatory elements that may prevent CPA at certain locations within the mRNA.

Intronic CPA alters gene output

Although tandem APA sites and alternative last exons are the most common types of APA, CPA can also occur prematurely within the gene body¹³⁰. This most frequently occurs in AU-rich introns, which harbor many sequences that resemble the PAS signal, AAUAAA. U1 snRNP binds co-transcriptionally to the AU-rich sequences, including 5′SSs, and suppresses CPA in nascent pre-mRNAs¹³⁹ (Figure 3a). Bound U1 snRNP prevents cleavage at intronic non-consensus PASs, an activity referred to as telescripting (Figure 3b). More recently, it was reported that U1 at 5′SSs can interact with cleavage factors already bound to intronic cleavage sites¹⁴⁰. This interaction suppresses cleavage and polyadenylation complex (CPC) catalytic activity, thereby preventing premature cleavage within introns^139,141. RNA-binding proteins, like ELAV in neurons¹⁴², and transcription factors, like PGC-1α in muscle cells through binding to nuclear cap-binding complex²¹, can act similarly. Because intronic CPA produces truncated mRNAs, it serves to downregulate expression of many genes through RNA decay. For example, depletion of the cleavage stimulating factor PCF11 upregulates long-intron genes, including the gene itself, and leads to defects in cell proliferation¹⁴³. Premature CPA has been associated with disease, e.g. in the context of chronic lymphoblastic leukemia by driving cancer progression by premature CPA in tumor suppressor genes¹⁴⁴.

The concept that U1, which should ultimately stimulate spliceosome assembly, would suppress polyA cleavage in an intron is an elegant example of how co-transcriptional coupling works; transcription must persist through the whole intron to the 3′SS if U1 snRNP binding is to be productive. New methods to probe protein–RNA complexes and RNA base-pairing in the context of transcription elongation will undoubtedly reveal additional principles that govern APA. More broadly, intronic CPA is reminiscent of transcription polarity effects caused by the termination factor and helicase Rho that is facilitated by cis-regulatory elements (rut sites) and decoupled transcription and translation in bacteria, resulting in downregulation of gene expression as well^145,146.

U1 snRNP is a Pol II elongation factor

Remarkable new discoveries concerning U1 snRNP’s role in transcription that are distinct from the phenomena described above have emerged recently. The visualization by cryo-electron microscopy (cryo-EM) of a structure containing human Pol II, DNA, nascent RNA with a 5′SS, and U1 snRNP (Figure 3a) revealed specific binding interactions between the U1 snRNP protein U1–70K and surface residues on the RPB2 and RPB12 subunits of Pol II¹⁴⁷. This quaternary structure did not require the presence of a 5′SS in the nascent RNA, indicating that the contacts between U1 and Pol II are potentially strong enough to support binding in vivo independent of 5′SS recognition and/or splicing. Instead, U1 may associate with Pol II during transcription and base-pair with the nascent RNA as soon as it emerges. Moreover, evidence for nascent RNA looping when it is extended after the 5′SS suggests that Pol II can maintain contact with the 5′SS as RNA synthesis continues, as proposed previously¹⁴⁷, potentially helping the 5′SS meet the 3′SS when subsequently synthesized. This dovetails with work showing that U2 snRNP is able to form the next spliceosome intermediate, the A complex, with U1 in a conformation that could still accommodate Pol II binding¹⁴⁸. Indeed, U1, U2, and other components of the spliceosome were recovered by pulldown with antibodies specific for the Pol II C-Terminal Domain (CTD) phosphorylated on Serine 5, which is associated with splicing intermediates^149,150. Although the association may be partially RNA-mediated, these findings suggest binding is stable, suggesting that such contacts impinge on Pol II elongation, analogous to ribosomes affecting bacterial RNA polymerase elongation. Recent comprehensive analyses of transcription elongation and splicing regulation have uncovered a large, overlapping network of determinants in budding yeast^151,152. Importantly, one of the major CTD kinases in mammals, CDK11, phosphorylates and regulates the activity of SF3B1, a critical component of the U2 snRNP that selects the branchpoint defining the 3′SS and initiating the chemistry of splicing¹⁵³. Thus, regulation of splicing and transcription elongation are relentlessly intertwined.

A role for U1 snRNP in promoting Pol II elongation was suggested by the observation that Pol II pauses when it encounters stable nucleosomes early in elongation, which can lead to termination in the absence of U1 snRNP¹⁵⁴. In the absence of U1, AT-rich sequences are more slowly transcribed, because the weakened RNA-DNA and RNA-RNA hybrids destabilize the elongation complex and lower nascent RNA folding needed to reduce backtracking^155–157. The bona fide elongation factor SPT5 can play a similar role to U1¹⁵⁸. Importantly, the presence of U1 effectively propels Pol II forward through AT-rich but not GC-rich sequences, yielding faster elongation rates in introns as compared to GC-rich exons¹⁴¹ (Figure 3b). Accordingly, the density of Pol II is lower over introns than exons^159,160 (Figure 3d). Previously, Pol II and promoter-associated histone post-translational modifications were shown to be enriched at or near the first 5′SS¹⁶¹, raising the question of how the presence of the 5’SS is sensed as a landmark in chromatin. In this new model, GC-richness is the most important feature that creates an “edge” when GC-transitions to AT-richness (Figure 3d); thereafter, U1 snRNP is first needed to maintain elongation by preventing backtracking¹⁴¹. The study carries out both TT-seq and PRO-seq, creating an “elongation index” that reflects Pol II nucleotide incorporation frequency and position, respectively. This allows for assessment of the dependence of elongation rate on the underlying GC-content of the DNA template and explains that the 6-fold range in transcription rates measured in mammalian cells and previously attributed to intron/exon structure¹⁶² can more accurately be viewed as a function of GC-rich versus AU-rich sequence (Figure 3d).

U1 snRNP’s ability to stimulate elongation offers an exciting explanation for why Pol II density decreases over introns (Figure 3c). Early studies analyzing Pol II density at 5′ and 3′SSs, suggested Pol II pausing there. The presence of first step splicing intermediates with 3′ ends at the end of each exon may have been misleading when some nascent RNA-Seq libraries were first sequenced. Recent studies using mNET-seq and PRO-seq have also rigorously documented and statistically evaluated Pol II density without detecting pausing at splice sites per se^103,104,163. Nevertheless, several studies provide evidence of Pol II accumulation downstream of 3′SSs, including in terminal exons, suggesting additional slowing before termination that may permit processing steps to complete before cleavage^164,165. Pol II pausing at 5′ and 3′SSs has been reported in the context of splicing inhibition and gene activation, respectively^166,167. Thus, Pol II pausing at splicing-related landmarks under conditions, such as stress, cell cycle, and disease should be revisited using the precision methodologies emerging today.

All-or-none (pre-)mRNA processing

Studies of co-transcriptional processing events in budding and fission yeasts and murine erythroleukemia (MEL) cells using long-read sequencing revealed an unexpected subpopulation of genes exhibiting “all-or-none” pre-mRNA processing that impacts how many functional mRNAs are exported and available for translation^101,103,151. Nascent transcripts of these genes are either fully spliced and cleaved at the PAS or fully unspliced and uncleaved, despite the overall high splicing and cleavage efficiency in the tested cells (Figure 4a). All-or-none pre-mRNA processing reveals coordination among individual introns of the same gene. Specifically, rapid removal of the first intron stimulates the removal of the downstream introns. Likewise, if the first intron is not spliced, downstream introns are retained. All-or-none pre-mRNA processing is reminiscent of the ‘first come, first served’ model of intron removal in which introns are spliced or committed to splicing in order of transcription and as soon as they are fully transcribed¹²⁴.

Figure 4. Splicing outcomes involving transcriptional readthrough and/or stress.

(a) In all-or-none (pre-)mRNA processing, transcripts synthesized by RNA polymerase II (Pol II) are either all spliced with efficient 3′ end cleavage carried out by the cleavage and polyadenylation complex (CPC) (left) or all unspliced with transcriptional readthrough (right). Recruitment of positive splicing factors may contribute to “all” transcripts, while prolonged binding of splicing inhibitory factors may contribute to “none” transcripts. (b) In tomatoes, two temperature-dependent RNA structures can form at the 3′ splice site (3′SS) of the second intron in HsfA2 pre-mRNA¹⁸³. Under non-stress conditions (left), an elongated RNA structure forms, which orients the polypyrimidine tract (PPT) and 3′SS for U2 snRNP recognition and results in a fully spliced product. Under heat stress (right), a smaller RNA structure is favored, which exposes the PTT and potentially allows for binding of a splicing silencer or polypyrimidine tract-binding protein (PTB). This results in retention of intron 2¹⁸³. (c) Under non-stress conditions (top), Pol II transcribes until it reaches the transcription termination site, releasing spliced and polyadenylated mRNAs. Under stress conditions (bottom), Pol II continues transcribing into intergenic regions and downstream “read-in genes”, producing downstream-of-gene (DoG) transcripts¹⁸⁶ in which the read-in genes are unspliced^151,190.

Inter-dependencies among neighboring introns have been observed in other single-gene and genome-wide studies; however, they demonstrate that splicing frequently occurs outside of the order of transcription^{109,127,168–170}. A pairwise splicing analysis of human introns reported introns that are always spliced before one of their neighbors but only about half of those are the upstream introns¹⁷⁰. This corresponds to later direct RNA nanopore sequencing studies also showing splicing often occurring in a specific order within a gene that does not follow the direction of transcription^109,127. The phenomenon of intron retention is another example where splicing does not proceed in the order of transcription. Thus, ‘first come, first served’ splicing as in all-or-none pre-mRNA processing does not appear to be a general mode of splicing.

The all-or-none processing demonstrates that CPA is promoted by splicing and blocked in unspliced transcripts, which indicates a crosstalk between the spliceosome and the CPC. This finding was tested using a human β-globin minigene, as β-globin is highly enriched for all-or-none processing¹⁰³. The minigene carried a known β-thalassemia mutation, IVSI-110. This mutation introduces a new, strong and thus preferred 3′ splice site into the intron, which resulted in more efficient splicing and more efficient cleavage compared to the WT minigene, suggesting coupling between splicing and CPA. In S. pombe, reduced all-or-none splicing was observed after a mild inhibition of splicing through mutation of splicing factor prp2 (homolog of U2AF65) and after changing the global gene expression through caffeine treatment¹⁰¹. This led to lower mRNA expression, as unspliced and uncleaved transcripts could be degraded by the RNA exosome in the nucleus¹⁰¹. Altogether, these findings indicate that all-or-none pre-mRNA processing regulates gene expression by altering mRNA levels.

The regulation and mechanisms of all-or-none pre-mRNA processing are not yet clear. It has been proposed that local concentrations of splicing factors contribute to splicing coordination and the coordination of CPA happening at the level of spliceosome assembly¹⁰¹. A likely mechanism could be the prolonged binding of U1 snRNP to unspliced transcripts, which could repress 3′ end cleavage at a distance^103,140,171. The all-spliced situation might be caused by the recruitment of the exon junction complex (EJC) and SR proteins to the newly formed exon-exon junction immediately after splicing, which could promote splicing of the downstream intron¹⁷². A recent structural study visualized an mRNP containing the EJC bound to two megadalton transcription–export complex (TREX) for nuclear export¹⁷³; if the nascent mRNP is topologically similar, the winding of the nascent RNA in a proteinaceous RNP and its coating by TREX could promote or inhibit downstream splicing by influencing the splicing factors that can or cannot associate. SR proteins are also associated with selecting the PAS and stimulating cleavage via the RS domain in Fip1^174,175. Thus, one can speculate that all-or-none pre-mRNA processing might depend on a fast removal of the inhibitory factors, such as U1 snRNP, the recruitment of positive effectors, such as EJC and SR proteins, and/or the co-transcriptional assembly of the nascent RNP (Figure 4a).

Pre-mRNA processing under stress

Pre-mRNA processing is an important regulatory point in the cellular response to stress. Stress encompasses any imbalance in cellular conditions that could damage the cell if left unchecked. A stress response can be triggered by a variety of stimuli, including high salt (hyperosmotic stress), elevated temperature (heat shock), low levels of oxygen (hypoxia), and unbalanced levels of reactive oxygen species (oxidative stress). Interestingly, nutritional stress in budding yeast leads to intron retention in GCR1 mRNA, usage of a cryptic translational start site within the intron, and production of an alternative transcription factor that feeds back to generate a new transcriptional program¹⁷⁶. Alternative splicing (AS) has emerged as an important gene regulatory mechanism in response to temperature in yeast, plants, fruit flies, and mammals^{120,177–183}. Elevated temperature and heat shock disrupt protein folding, depending on the characteristic stability of the folded domain/protein. Proteins and RNAs can register structural changes within a normal range of temperatures. Daily fluctuations in body temperature as low as 1°C have significant effects on pre-mRNA processing in mammals by changing splicing patterns^178–180. These changes are mediated by CDC-like kinase 1 (CLK1), which phosphorylates and activates serine/arginine-rich proteins – known regulators of AS – at lower temperatures¹⁷⁸. The CLK1 ortholog AFC2 in the model plant Arabidopsis thaliana exhibits similar behavior in the context of non-stressful temperature fluctuations¹⁸⁴. In budding yeast and A. thaliana (Figure 4b), inhibitory intron structures melt and allow access to splice sites^181–183. These examples highlight the influence of subtle as well as extreme changes in temperature on pre-mRNA processing, which are not only relevant to human health and disease but also how organisms will respond to the effects of climate change. Below, we focus on recent discoveries on the effects of stress on 3′ end cleavage.

Stress blocks 3′ end cleavage

One of the major phenotypes of the stress response is a failure to cleave nascent RNA 3′ ends, resulting in extensive transcriptional readthrough¹⁸⁵. This produces long non-coding RNAs, or downstream-of-gene RNAs (DoGs), that appear to remain nascent, or uncleaved (Figure 4c)¹⁸⁶. DoGs were independently discovered by three groups in 2015 who observed an increase in read coverage downstream of polyadenylation sites under stress or disease conditions: hyperosmotic stress¹⁸⁶, clear cell renal cell carcinoma (ccRCC)¹⁸⁷, and HSV-1 infection¹⁸⁸. Some DoGs are common to multiple stresses, while others are unique to specific stresses. Several functional roles have been proposed for DoGs. One suggestion is that DoGs may serve as a reservoir of unprocessed transcripts that could be processed post-transcriptionally to aid in cell survival¹⁸⁹. Alternatively, DoGs may repress transcription of neighboring genes or change their surrounding chromatin landscape in a way that promotes the stress response and/or recovery¹⁸⁹. Under stress, read-in genes in mammals exhibit strong intron retention in first introns¹⁹⁰, while read-in genes in yeast are completely unspliced¹⁵¹. Whether DoG regions have intron/exon structure and if splicing of parent genes is affected by stress remains to be elucidated¹⁹¹. LRS approaches that can capture full-length DoG transcripts will allow for validation of both splicing status and connectivity of parent genes, readthrough regions, and read-in genes within the same transcript. So far, this goal has been complicated by the lengths of DoGs, which can reach up to 200 kb in mammalian cells¹⁸⁶. Overall, transcriptional readthrough is now considered a hallmark of the mammalian stress response¹⁹².

In addition to transcriptional readthrough, 3′ end formation is also regulated under stress through APA. A recent study using a genetic screen in Caenorhabditis elegans revealed that mutations to genes coding for CstF64 and Symplekin – components of the cleavage and polyadenylation complex (CPC, see above) – inhibited APA and gene expression required for the hyperosmotic stress response¹⁹³. These results suggest a key role for the CPC and APA in allowing cells to survive high salt conditions. Interestingly, this effect was not applicable to heat stress, suggesting an alternative mechanism of regulating pre-mRNA processing at high temperatures¹⁹³.

Conclusions and perspectives

Aside from 3′ end cleavage, each co-transcriptional RNA processing event can theoretically take place after transcription, i.e. post-transcriptionally, and some splicing events do^22,194. Retained introns have an impact on subcellular localization and polyA tail length^195,196. So, the decision to process RNA co-transcriptionally influences transcript fate. Comparing co-transcriptional gene regulation among prokaryotes and eukaryotes at this point in our respective “fields” is a valuable exercise that inspires new concepts and methodologies. For example, the recent discovery that some bacterial species lack functional coupling of translation and transcription owing to the greater speed of transcription contrasts with textbook knowledge that co-transcriptional translation regulates transcription in prokaryotes. This dichotomy of fundamental behaviors is reflected in the state-of-the-art measurements of co-transcriptional RNA processing events, where both rapid and delayed splicing and/or 3′ end formation are emerging as different regulatory strategies in eukaryotes. An appreciation of co-transcriptional processes and the coupling of RNA processing steps with transcription are most obviously addressed by sequencing nascent RNA (Boxes 2 and 3). The association of Pol II position (3′ end) with transient states of the attached nascent RNA (e.g. fully spliced, fully unspliced, partially spliced, 3′ end cleaved or uncleaved) allows us to infer an order of events. In contrast, metabolic labeling can give answers in units of time, enabling the exploration of true kinetics rather than Pol II position within the gene. However, our current inability to bridge these data and platforms prevents the field from establishing greater generality and potentially defining new open questions. For example, LRS addresses short genes with short introns well but leaves many questions about long intron splicing unaddressed. Metabolic labeling and imaging have provided unique insights that likewise cannot be generalized to the short gene/short intron datasets, because we are limited by a lack of “ground truth” regarding Pol II elongation rates along the lengths of gene bodies, ideally in the absence of drug treatments. Current estimates of Pol II elongation rates along introns and exons differ by about 6-fold, and possible local pausing behaviors are unresolved. Recent breakthroughs highlight the unifying importance of AT- and GC-rich sequences combined with regulation by the U1 snRNP to promote elongation over AT-rich introns¹⁴¹. Yet, new methods for determining regional elongation rates in units of time would vastly improve our ability to integrate approaches to understanding the molecular mechanisms of gene expression in vivo.

The development and improvement of RNA structure probing methods have driven significant discoveries in this area. For example, DMS and SHAPE probing revealed single-stranded regions near the 5′ splice site and branch point favoring the selection of these sites in A. thaliana¹⁹⁷, showed regulation of MAPT (tau) splicing through structure¹⁹⁸, and discovered multiple conformations regulating HIV-1 proviral RNA splicing¹⁹⁹. Despite major advances particularly in the last decade, we lack an overarching understanding of how base-pairing and tertiary intramolecular contacts influence transcription and processing. Part of the difficulty lies in the dynamic nature of transcription, and the fact that transient states are difficult to capture with structure probing methods, which are so far the only tool to reveal base pairing in vivo. In addition, discovery of new regulatory RNA motifs is currently mostly based on structural models built through sequence comparison. With the availability of new genomes and the improvement and addition of RNA structure probing methods that work co-transcriptionally, we are likely to discover new instances of co-transcriptional regulation through RNA structures, including APA and splicing, particularly beyond prokaryotic systems.

In addition to elucidating RNA processing on a single-transcript level, as done in most sequencing-based approaches, single-cell resolution is also needed to resolve cellular variation, as well as stochasticity¹⁰⁷. Most of the principles and findings discussed in this review are based on population measurements and thus reflect the average profile of often millions of individual transcripts and cells (Figure 3c). Single-cell transcriptome analysis promises to reveal novel and often unexpected aspects and variation to co-transcriptional pre-mRNA splicing and RNA processing coordination. Imaging-based methods (Box 3), integrated also with transcriptome studies by sequencing, have taken first steps in this direction, indicating that splicing is as “bursty” as transcription itself and stochastic¹⁰⁷. Further methodological developments are needed to yield a single-cell perspective on coordination between co-transcriptional RNA processing events, such as in all-or-none processing, and the interconnection with transcript starts and ends and pre-mRNA splicing. To this end, single-cell RNA-seq (scRNA-seq) methodology developed for prokaryotes^200–205 could be inspirational, as some of the challenges that delayed scRNA-seq use in prokaryotes compared to eukaryotic cells²⁰⁶ are the same that prevent single-cell analysis for nascent transcripts in eukaryotic cells. Namely, nascent RNAs are rare among the pool of mRNAs, which is reminiscent of prokaryotic cells containing small amounts of mRNAs. They are also not adenylated as are prokaryotic mRNAs and vary substantially in their lengths as do prokaryotic mRNAs that can encode multiple CDSs. Thus, methodological as well as biological principles are bridging scales between prokaryotes and eukaryotes.

Co-transcriptional processes in eukaryotes are restricted to the nucleus, whereas prokaryotic co-transcriptional processes occur within a common plasm. The eukaryotic nucleus is a highly specialized compartment thanks to the selectivity of the nuclear envelope, that is sub-compartmentalized even further by liquid-liquid phase separated (LLPS) organelles (Figure 1b). Even genes themselves can be seen as condensates and many nuclear proteins share the propensity for phase-separation by harboring intrinsically disordered domains^207,208. Yet, our view of compartmentalization in prokaryotic cells is changing. High-resolution imaging of individual components in mRNA regulation has revealed unprecedented cellular organization that may be seen to compensate for the absence of a nucleus separating co- from post-transcriptional RNA processing^209,210 (Figure 1a). Local LLPS has been associated with RNA degradosomes, hubs for degrading prokaryotic RNAs that have often species-specific components and localization, and areas of the nucleoid^211–214. Thus, sub-compartments in the prokaryotic cell could function similarly to biomolecular condensates – nucleoli, Cajal bodies, speckles – in the eukaryotic cell nucleus to render gene expression more efficient²¹⁴. Overall, the power of the nuclear envelope is to add a layer of regulatory potential, e.g. by pre-mRNA splicing, onto fundamental biology that is in principal shared among prokaryotes and eukaryotes with the shared chemistry of the central dogma.

Here we have argued that co-transcriptional processing is more than just temporal and spatial co-occurrence in both pro- and eukaryotes. Multiple instances of interplay have been highlighted in this review connecting RNA processing and transcription through specific features and behaviors like changing Pol II elongation rate, APA and U1 snRNP, all-or-none processing, RNA secondary structures and more. We acknowledge that this is not an exclusive list with many other important regulatory aspects, e.g. RNA editing and modifications adding a dimension to cellular transcriptomes that is beyond the scope of this review. In prokaryotes, we discussed how the absence of the nucleus enables different modes of co-transcriptional translation, but also RNA decay and processing and transcript isoforms can be surprisingly complex, enabling gene expression regulation beyond the logic that one (polycistronic) operon always yields the same number of proteins. This all points to the challenge and opportunity of attaining a holistic understanding of gene expression as an interactive network in which the co-transcriptional level of regulation comprises a unique set of interactions facilitated by RNA polymerases tethering nascent RNAs to chromatin across the tree of life. Crosstalk among researchers working in prokaryotes and eukaryotes might also benefit research and discoveries in any system where RNA regulation in conditions of cellular stress, disease, and/or physiological variation, such as daily temperature fluctuations, are of immense interest.

Glossary

3′ Splice Site (3′SS)

the 3′-most sequence of each intron, which is recognized by U2AF along with the polypyrimidine tract.

5′ Splice Site (5′SS)

the 5′-most sequence of each intron harboring ~6nt complementarity to the 5′ end of U1 snRNA, enabling U1 snRNP binding for intron identification and, later during splicing chemistry, U6 snRNA interactions with some of the same nucleotides.

Alternative Polyadenylation (APA)

usage of distinct polyadenylation sites (PAS) to generate multiple transcript isoforms from the same gene, which can give rise to protein isoforms with distinct C-termini or contribute to transcript regulation through changes to 3′ untranslated regions (UTRs).

Alternative Splicing (AS)

the process of generating multiple mRNA isoforms from a single gene by using different combinations of splice sites (e.g., intron retention, exon skipping, alternative 5′ or 3′ splice site usage).

Branchpoint

Short cis-regulatory element containing an adenine base that is involved in lariat formation during the first step of pre-mRNA splicing and resides within the 3′ portion of the intron usually near and upstream of another cis-regulatory element, the polypyrimidine tract.

Cis-elements for Translation Initiation

For most CDSs these are the Shine Dalgarno sequence, an A-rich sequence upstream of start codon and the start codon itself in bacteria.

Cleavage and Polyadenylation (CPA)

Endonucleolytic cleavage followed by polyadenylation of the transcript by the CPC while Pol II continues transcription until termination downstream.

Cleavage and Polyadenylation Complex (CPC)

Macromolecular protein complex that binds to cis-regulatory elements involved in 3′ end formation of eukaryotic mRNAs and facilitates coupling to Pol II.

Exon Definition

Definition of 3′ and 5′SSs adjacent to short exons that belong to the two surrounding introns.

Exon Junction Complex (EJC)

Macromolecular protein complex that forms at exon-exon junctions after splicing is completed.

Intrinsic Transcription Terminator

RNA stem-loop associated with A- and U-tract in nascent RNA that causes transcription termination of bacterial RNA polymerase.

Intron Retention

an alternative splicing event in which an intron is included in a mature mRNA transcript.

Nascent Elongating Transcript Sequencing (NET-Seq, mNET-Seq)

a method that sequences the 3′ end of all RNAs that co-purify with RNA polymerase, thereby indicating the position of all RNA polymerase holoenzymes that have initiated transcription.

Polyadenylation Site (PAS)

the signal for polyadenylation (AAUAAA in humans) is read by the cleavage and polyadenylation specificity factor (CPSF), recruiting additional proteins to assemble the cleavage and polyadenylation complex (CPC) and release and polyadenylate the nascent transcript.

Polymerase Backtracking

transcriptional pausing through diffusive movement of RNA polymerase along template DNA, maintained by feeding nascent RNA through a channel at the front of the enzyme.

Polypyrimidine Tract (PPT)

a ~15–20-nt-long region of pre-mRNA located near the 3′ end of introns that is rich in pyrimidine bases (i.e., cytosine and uracil) and promotes spliceosome assembly by serving as a binding site for the spliceosome component U2AF.

Polypyrimidine Tract-Binding Protein (PTB)

a protein that binds to polypyrimidine tracts within introns and serves as a negative regulator of pre-mRNA splicing.

Precision Run-On Sequencing (PRO-Seq, similar to GRO-Seq)

a method that determines the density of elongating RNA polymerase holoenzymes, through the addition of a single biotinylated nucleotide (or BrUTP), which can be used to select the nascent RNA, obtain short reads, and map the 3′ end/polymerase position.

Rho

Homohexameric bacterial RNA helicase and chaperone whose binding to nascent RNAs often between not closely trailing ribosomes and RNA polymerase can lead to transcription termination.

Runaway Transcription

Transcription elongation is substantially faster than translation and thus RNA polymerase outpaces ribosomes generating ‘slack’ of nascent RNA in between the two molecular machines; first it was experimentally detected in B. subtilis, it is also likely to occur in other bacteria.

Serine/Arginine-rich (SR) Proteins

a family of proteins whose members contain an RNA-recognition motif and a serine/arginine-rich domain and regulate splicing.

Spliceosome

a multi-megadalton molecular machine composed of the U1, U2, U4, U5, and U6 small nuclear ribonucleoproteins (snRNP) and associated proteins that catalyzes the removal of introns from pre-mRNA.

sRNA

Regulatory bacterial small RNAs that are up to 200nt long and can affect all steps in prokaryotic gene expression, including secondary structure formation, by annealing to mRNAs in coding and non-coding sequence regions, often facilitated by RNA chaperones, such as Hfq.

Tight Co-transcriptional Translation

The first initiating ribosome on nascent RNA is closely trailing RNA polymerase, possibly even physically interacting mediated by transcription factors; this has been the core model for co-transcriptional translation and is based on work in E. coli.

Transcription Polarity

5′ to 3′ ramp of RNA signal across operons caused by transcription termination within operons at varying sites, e.g., in Rho-dependent termination.

Transient Transcriptome Sequencing (TT-Seq)

a method that detects newly synthesized RNA by metabolic incorporation of 4sU in live cells and focusing on the proportion of transcripts that have been synthesized in the labeling period.