Nascent RNA Analyses: Tracking Transcription and Its Regulation

Abstract

The genome encodes information to program an organism’s development and maintenance, and its decoding begins with regulated transcription of genomic DNA into RNA. Transcription and its control can be tracked indirectly by measuring stable RNAs, or directly by measuring nascent RNAs. The latter reveals the immediate regulatory changes in response to developmental, environmental, disease, and metabolic signals. Multiple complementary methods have been developed to quantitatively track nascent transcription genome-wide at nucleotide-resolution, providing novel insights to mechanisms of gene regulation and transcription-coupled RNA processing. Here, we critically evaluate the array of strategies used for investigating nascent transcription and discuss recent conceptual advances they provide.

Introduction

Transcription is a process whereby RNA polymerases (Pol) synthesize RNA molecules from a DNA template. In eukaryotes, Pol I and III synthesize ribosomal RNAs and small structural RNAs, respectively; whereas Pol II produces protein-coding mRNAs, long non-coding RNAs, primary microRNAs, and enhancer RNAs (eRNAs) [G]. While every cell in an individual organism contains an identical genome, regulation of gene expression defines which RNAs and proteins are synthesized, and the level to which they are produced (reviewed in Ref. 1). Consequently, genome-wide coordination of transcription underlies cellular differentiation, responses to internal and extracellular signals, and organismal functions.

Transcription of genes is controlled via proximal and distal regulatory elements, termed promoters and enhancers, respectively (reviewed in Refs. 2 and 3). These regulatory elements contain binding sites for promoter - and enhancer-specific transcription factors (TFs) that define when a gene is active, and the frequency with which it is transcribed (Figure 1). In mammals, promoters and enhancers contain two core initiation regions^4,5 [G]. These regions are bound by general transcription factors (GTFs), that together with Pol II comprise the pre-initiation complex (PIC). Both promoters and enhancers have a very similar chromatin architecture, including the constellation of GTFs, TFs and Pol II complexes, and drive divergent transcription from their core initiation regions⁴ (Figure 1). Enhancer transcripts, however, are generally short and unstable, whereas the coding strand of a gene produces transcripts (pre-mRNA) that are predominantly long and stable (Figure 1)^4,6,7. The mechanisms of enhancer-promoter communication remain poorly understood, but recent reports suggest eRNA production correlates with its functional enhancer capacity^2,3 (reviewed in Ref. 8).

Figure 1 |. The transcription cycle.

a | Typical gene architecture depicting DNA elements that affect transcription and transcript stability. At promoters, binding sites for gene-specific transcription factors are found between two core promoters (i.e., core initiation regions) that drive divergent transcription. The coding strand (right) produces mRNA that is stabilized by the presence of splice sites, and the anti-sense strand (left) produces an unstable upstream antisense RNA. b | The transcription cycle consists of the following steps. 1) A pioneer transcription factor binds a specific sequence motif and increases chromatin accessibility. 2) Additional sequence-specific TFs bind near the pioneer factor. Core promoters recruit GTFs and Pol II to form the PIC. 3) The GTF TFIIH unwinds DNA, and Pol II initiates transcription. 4) Pol II undergoes promoter-proximal pausing after transcribing 20–60 nt, and the pause is stabilized by binding of DSIF and NELF. Before or during pausing, Pol II’s CTD is phosphorylated at Ser5 and Ser7, and the RNA undergoes 5’ capping. 5) Pol II escapes promoter-proximal pausing and enters productive elongation, largely due to CDK9 phosphorylating multiple targets: NELF (ending its interaction with Pol II), DSIF (converting it to an elongation factor), and Pol II Ser2 (which interacts with RNA processing factors). Additional elongation factors, such as PAF1, promote this escape. 6) During productive elongation, co-transcriptional processing, including splicing, RNA methylation, and RNA editing occur. Nucleosomes are removed in highly transcribed genes, while chromatin accessibility increases at moderately transcribed genes^207,208. 7) The RNA is cleaved and polyadenylated. After cleavage, Pol II continues elongating, but the nascent RNA lacks a cap and subject to XRN2-mediated degradation, which destabilizes Pol II and contributes to termination. After termination, Pol II can be recycled to new initiation, repeating the transcriptional cycle.

Beyond binding to genomic sequences, certain transcription regulators directly act at the transcription machinery. One of the major signaling platforms for transcription regulation is the C-terminal domain (CTD) of Pol II’s Rbp1 subunit [G] (reviewed in Refs. 9 and 10). This CTD consists of multiple heptad repeats of a consensus Tyr-Ser-Pro-Thr-Ser-Pro-Ser sequence, and undergoes regulated post-translational modifications (PTMs) before and during transcription, leading to dramatic changes in the entourage of factors associated with Pol II as it progresses through the transcription cycle (reviewed in Ref. 10).

Detailed mechanistic studies at genes have revealed that transcription consists of multiple regulated steps (Figure 1). Transcription begins (step 1) when a pioneer factor binds closed chromatin and increases chromatin accessiblility by recruiting nucleosome remodelers and histone acetyltransferases (reviewed in Refs. 11 and 12). As chromatin opens, DNA elements become accessible to additional TFs and formation of PIC (step 2)¹³ (reviewed in Ref. 14). Once the PIC is formed, its TFIIH subunit unwinds double-stranded DNA, allowing Pol II to initiate transcription (step 3)^15,16. Engaged Pol II initially transcribes 20 to 60 nucleotides (nts), then undergoes promoter-proximal pausing, which in mammals and many other metazoans is a major rate-limiting step in gene expression, and a regulatory check-point for execution of transcription programs (step 4)^17–23 (reviewed in Ref. 24). During this early transcription, the nascent RNA is protected by 5′-capping²⁵. Pausing of Pol II is stabilized by NELF and DSIF^26–28, whereas release from the pause (step 5) requires P-TEFb, whose CDK9 subunit phosphorylates NELF-E, DSIF, and Ser2 residue of the Pol II CTD (reviewed in Ref. 29). During productive elongation (step 6), multiple elongation factors enhance processivity of Pol II^30,31 and couple nascent transcription to processes such as splicing (reviewed in Ref. 32). Finally, at the 3′-end of the gene, the transcript is cleaved and the pre-mRNA polyadenylated³ (reviewed in Ref. 33). This cleavage releases pre-mRNA from Pol II and leaves uncapped nascent RNA unshielded from XRN2-mediated degradation, which destabilizes Pol II and contributes to termination (step 7) (reviewed in Ref. 34). The terminating Pol II dissociates from the DNA and is recycled to a new round of transcription³⁵.

The current mechanistic view of transcription regulation (Figure 1) combines data from a variety of techniques that have mapped the chromatin composition and interrogated nascent RNA synthesis at each step. Over the past 50 years of tracking RNA synthesis^36–38, our ability to investigate transcription has evolved greatly, expanding single-locus analyses to probing transcriptional mechanisms across genomes. Moreover, advances in electron microscopy has revealed structures of transcription factors and transcription complexes, including Pol II in its pause- and elongation complex^27,28,39,40. Particularly, RNA sequencing methods have been crucial in quantifying RNA molecules in the cell, and more recently, allowed robust measures of genome-wide changes in nascent RNA production. Techniques that directly measure RNA synthesis differ in their biochemical approaches to capture transcripts from the total pool of RNA, and therefore, have different abilities to track distinct RNA species, such as pre-mRNAs, divergent transcripts [G], or eRNAs. In this Review, we compare and contrast some of these key strategies, discussing their strengths and limitations, and how their usage in combination can provide a holistical view of regulation throughout the transcription cycle. We begin with a description of the conceptual differences between nascent RNA sequencing methods, and outline complementary imaging techniques. We then discuss the technologies best suited to addressing particular biological and mechanistic questions while highlighting the discoveries made with these methods.

Methods to track nascent RNA synthesis

Methods that investigate RNA synthesis biochemically enrich for nascent or newly synthesized RNA species from the total pool of cellular RNA. In all of these techniques, isolated transcripts are reverse transcribed, ligated to adapters, deep sequenced, and after initial quality filtering and trimming, mapped against the reference genome. As a result, genome-wide sequencing of nascent or newly synthesized transcripts provides rich datasets, including density profiles of transcribing Pol II molecules across the genome, coordinates of transcription start sites or Pol II active sites at nucleotide resolution, and measures of gene and enhancer transcription. Albeit many of the library preparation and data analysis steps are shared between RNA-seq methodologies, distinct approaches differ remarkably in their ability to enrich for the nascent transcripts. Consequently, the method of choice defines which RNA species or steps of the transcription cycle can be analysed and affects the stringency and resolution of the generated data.

The various strategies to enrich for nascent RNA include isolation of chromatin-associated RNA (caRNA)^41–44 , Pol II–associated RNA^45,46, small capped RNA^47,48, recently synthesized RNA^49,50, or RNA from elongation-competent Pol II complexes^48,51,52 (Figure 2). Most of these methods can reliably discern changes in gene expression; however, they differ considerably in their sensitivity to detect different steps of transcription, spatiotemporal resolution, and abilities to identify distinct RNA species, such as eRNAs, divergent transcripts, un-spliced intermediates, and other unstable non-coding RNAs. While genome-wide nascent RNA sequencing methods can precisely map molecular-level regulation of Pol II at genes and enhancers^{17,18,53–56}, they do not yet distinguish cell-to-cell variation or provide spatial maps of trancription within the nucleus. By contrast, imaging-based methods have limited genomic resolution and ability to discern steps of transcripton, but they can quantify transcript production in real-time in the 3D space of the nucleus, tissue or organism⁵⁷. Below, we summarize distinct sequencing and imaging methods for tracking RNA synthesis and emphasize key steps that determine the sensitivity, specificity, and resolution of each method (Figure 2).

Figure 2 |. Comparison of nascent RNA enrichment and sequencing assays.

Transcription profiles of a highly expressed gene (β-actin), a highly paused gene (MED17), and a gene with a nearby eRNA (IPO7 for human cells, Klf4 for mouse cells), generated with distinct RNA-seq methodologies. a | In chromatin-associated RNA (caRNA) methods, high salt washes are used to isolate chromatin-bound RNAs. In traditional caRNA-seq (lower), that material is directly sequenced. Start-seq (upper) further enriches for capped caRNAs and uses size selection to capture initiation and pause sites of individual transcripts. Data for caRNA originates from Ref. 46 and Start-seq from Ref. 23. b | Immunoprecipitation of Pol II complexes enriches for RNAs that associate with Pol II in mNET-seq. Antibody-mediated isolation of Pol II removes most chromatin-bound RNAs. mNET-seq data that used an antibody targeting total Pol II was obtained from Ref. 61. . c | Run-on techniques mark nascent RNAs with labeled nucleotides. The use of the anionic detergent sarkosyl in the run-on reaction releases paused polymerases but not back-tracked or terminated Pol II. The original genome-wide nuclear run-on assay, GRO-seq, has been adapted to provide single-nucleotide resolution of the position of engaged Pol II on nascent RNA (PRO-seq). Using cap selection and sequencing from the 5’ end of the labeled RNA reports the initiating base (PRO-cap, upper). Data for PRO-cap from Ref. 209 (GSE110638), for PRO-seq from Ref. 4. d | Metabolic RNA labeling methods feed living cells modified ribonucleotides that will be incorporated into nascent RNAs. After labeling, nascent RNA can be enriched from the total RNA pool with immuno-purification (TT-seq, upper). Additionally, 4-thiouridine allows U to C base conversion for mutation-based identification of nascent transcripts after sequencing (TT-TimeLapse-seq, lower). Data for TT-seq from Ref. 210, for TimeLapse-seq from Ref. 76.

Isolating chromatin-associated RNAs.

The simplest approach for isolating nascent RNA uses strong washes to separate RNA present on chromatin from other RNAs in the cells, thus relying on the stability of the association between nascent transcripts and polymerases during salt fractionation of chromatin (Figure 2a)^41–44,58. Because cytoplasmic and nucleoplasmic RNAs are more abundant and stable than nascent RNAs, enriching for caRNAs significantly improves the dynamic range for detecting changes in gene and enhancer synthesis. Certain mature RNAs, however, are also stably associated with chromatin, and therefore are captured in caRNA-seq. These mature RNAs include long non-coding RNAs such as Xist, which coats and inactivates one of the two X chromosomes in female mammals, as well as small nuclear RNAs (snRNAs), including spliceosomal U1, U2, U4, U5, and U6. To further increase sensitivity and specificity for nascent RNAs associated with Pol II, Nascent Elongating Transcript Sequencing (NET-seq)⁴⁵ was developed in yeast and uses immunoprecipitation to capture Pol II-associated RNAs (Figure 2b). NET-seq shares many biochemical and data-analytical approaches with chromatin immunoprecipitation (ChIP-seq) [G] and RNA immunoprecipitation RIP-seq [G]⁵⁹, which has contributed to it being readily adopted by the transcription field. NET-seq was subsequently adapted for mammalian cells (mNET-seq); in this approach, Pol II immunoprecipitation was performed with antibodies against the post-translationally modified CTD of Pol II to identify regulatory status of the transcription complex at distinct regions of genes^46,60. The originalNET-seq enriched for nascent RNAs by removing most non-nascent RNAs, as well as Pol I and Pol III associated transcripts. However, certain non-nascent RNAs that stably associated with Pol II remained, including the notable example of snRNAs⁶⁰. Removal of non-nascent RNAs was improved by use of the detergent empigen, which dissociated splicing intermediates that bound the spliceosome, as well as miRNAs that associated with Microprocessor ⁶¹. Comparing these two versions of the mNET-seq protocol provided valuable information about co-transcriptional splicing^60,61.

To specifically identify transcription start sites (TSSs), Start-seq enriches for chromatin-associated capped RNA-species⁴⁷. The selection of capped RNA is conducted by enzymatic degradation of caRNAs that are not protected by the 5′-cap, such as rRNA or unshielded mRNA fragments generated during preparation of the sequencing library. Retaining capped caRNAs that are smaller than 80 nt in length, further selects for transcripts that are undergoing initiation, pausing, or early elongation. With paired-end sequencing [G], these short, capped RNAs provide high-resolution information both on TSSs (5′-end of each read), and active sites of transcription (3′-end of each read) at genes and enhancers⁴⁷.

Isolating RNAs from transcriptionally-competent Pol II.

Unlike caRNA-seq and NET-seq, run-on assays specifically capture RNAs that are undergoing synthesis. The run-on experiment starts by placing the cells on ice, which stops Pol II from transcribing. Subsequently, the cells are permeabilized, or nuclei or chromatin isolated, to remove endogenous nucleotides and enable labelled NTPs to reach Pol II complexes at the chromatin. Traditional ‘run-on’ reactions used radio-labelled NTPs⁶², which were detected at transcripts by hybridization to complementary DNA sequences^62,63. With current high-throughput sequencing techniques, RNAs with incorporated labelled nucleotides can be mapped across the genome⁵¹, and the initiating base and active site identified at nucleotide resolution^4,48,52.

Like caRNA-seq, run-on reactions rely on the strong interaction between transcriptionally-engaged Pol II and DNA. During the run-on reaction, anionic detergent sarkosyl is used to remove proteins, including pausing factors, from chromatin, enabling transcription-competent Pol II to proceed along the genome in vitro. The addition of sarkosyl is particularly important to reactivate paused Pol II complexes, and therefore, to detect nascent RNAs in the pause complexes⁶⁴. Since run-on techniques rely on incorporation of nucleotides, only transcription-competent polymerases are detected. Therefore, Pol II in initiation complex remains undetectable with nuclear run-on techniques⁶⁴. Moreover, sarkosyl cannot rescue Pol II from a backtracked state (reviewed in Ref. 65), which occurs when Pol II moves back in the DNA template, for example as a result of misincorporation of a nucleotide, displacing the 3′-end of the nascent transcript. The first genome-wide adaptation of the run-on reaction was Global Run-On sequencing (GRO-seq), which labeled nascent transcripts with 5-bromouridine 5′ triphosphate (BrUTP), and immunopurified the transcripts using an antibody against BrUTP⁵¹. Precision Run-On sequencing (PRO-seq) refined this approach to generate nucleotide-resolution maps of active transcription. In PRO-seq, biotin-11-NTPs are incorporated into a competent Pol II’s active site, and due to their bulkiness, only a single nucleotide is added to the transcript⁴⁸. When using all four biotinylated nucleotides, which have similar rates of incorporation⁴⁸, all transcripts can be captured without biasing for presence for certain nucleotide sequences. After the run-on, biotinylated nascent transcripts are isolated using streptavidin-coated magnetic beads, providing nucleotide-resolution profile of the active sites of transcription (Figure 2c)⁴⁸. The sensitivity of run-on methodologies is provided by the polymerase-catalyzed incorporation of a single biotinylated nucleotide to the active site of each nascent transcript, followed by and three biotin-affinity purifications to isolate nascent transcripts during library preparation. This results in very low background, and a nearly a million-fold dynamic range estimated from various data-sets^48,66. While the in vitro aspect of run-on reactions may cause some concern that the read-out is distorted, run-on methods show agreement with permanganate IP-seq^48,67 (PIP-seq) [G] ChIP-seq⁶⁴, ChIP-exo⁶⁸, mNET-seq⁴⁶, and Start-seq⁶⁹ in that promoter-proximal pausing co-localizes genomewide⁴⁶ and occurs predominantly at the nucleotide preceding cytosine in run-on^52,70.

By using chromatin as a starting material for run-on reactions, a variant method, Chromatin Run-on sequencing (ChRO-seq), captures nascent RNAs from tissue samples that are not readily amenable to nuclei isolation protocols or have experienced RNA degradation during handling or storage⁷¹. Both GRO-seq and PRO-seq have been adapted for nucleotide-resolution identification of TSSs by enzymatic cap selection, similar to that used in Start-seq. These variants are known as GRO-cap and PRO-cap^4,48. The PRO-cap protocol has recently been further modified to coordinate analysis of the initiating nucleotide, capping status, and length of RNA molecules (CoPRO)⁵². This simultaneous identification of the initiating base and active site of each nascent transcript allows systematic investigation of connections between transcriptional initiation and elongation, whereas capping status provides mechanistic insight into early co-transcriptional processes. We anticipate that improvement of long-read sequencing techniques, which would generate sequencing reads that span multiple exons, will allow development of similar strategies to reach beyond the early coding sequences to map other co-transcriptional processes, such as termination and splicing.

Isolating metabolically labeled RNAs

In metabolic labeling, cells are grown in medium supplemented with modified cell-permeable nucleosides, such as 4-thiouridine^50,72–79 (4sU), bromouridine⁸⁰ (BrU), or 5-ethenyluridine^81,82 (EU), which are incorporated into nascent RNAs by active polymerases. Importantly, these nucleotides lack 5′ triphosphates, and are thus processed by the nucleoside salvage pathway, whereby U is converted to UTP in a series of enzymatic reactions⁸³, which explains the need for extended labeling times in cells (generally 10–60 minutes). In comparison, genome-wide run-on reactions start with isolated nuclei, permeabilized cells or chromatin, which allows NTPs to reach the active site of Pol II from the onset of the run-on reaction^48,51. Metabolic labeling does not allow for tracking RNA synthesis at nucleotide resolution, but it can measure RNA synthesis in living cells, which allows following lifetime and turnover of RNAs (Figure 2d). Recently, advances have been made in shortening labeling with 4sU to 5 minutes in Transient Transcriptome sequencing (TT-seq)⁵⁰ and TimeLapse-seq⁷⁹, which has increased detection of newly synthesized RNAs. In TT-seq, metabolically labeled RNAs are enriched by using sonication to separate fragment RNA followed by immuno- or affinity-purification, then compared to unlabeled or total RNA⁵⁰. Alternatively, RNAs can be chemically treated to convert incorporated 4sU to cytosine, creating single-nucleotide mutations that are quantified after deep sequencing^76–79. Chemical conversion does not require isolation of labeled RNA, and compares nascent and non-nascent RNA in the same sample. This direct analysis of the levels of nascent or newly synthesized RNA (converted nucleotides) to unlabeled RNA allows estimates of average RNA production and decay rates genome-wide^76,78,79, as long as background rates of 4sU incorporation and sequencing errors are taken into account⁷⁷. Besides allowing incorporation of labeled nucleotides in cultured cells, metabolic labeling of RNA has also been demonstrated in living animals^81,82.

Imaging-based methodologies.

Imaging-based methods have limited genomic resolution and capacity to identify regulatory steps, but can quantify transcript production in real-time in the 3D space of the nucleus, tissue or organism (reviewed in Ref. 84). Nascent transcripts can be imaged with techniques such as fluorescent in situ hybridization (FISH), in which labeled oligonucleotides are hybridized with transcripts (Fig 3A)^85–88. Using these methods in mouse cells from parents of two different strains can provide information about the effects of genome variation in bursting within the same cell⁸⁹. Alternatively, transcripts can be engineered to encode hairpin structures that are recognized in vivo by tagged cognate binding proteins, such as GFP-coupled bacterial coat protein MS2 (Fig 3B)^90,91. In these techniques, the site of nascent RNA synthesis in the nucleus is identified as the brightest spot of transcript signal^90,92,93, by labeling the locus producing the transcript⁹⁴, or by targeting fluorophores specifically to introns that are present in short-lived pre-mRNA species^57,95,96. To visualize single unspliced pre-mRNA molecules, several fluorophores must hybridize at each molecule of a transcript’s intron. Monitoring the synthesis of pre-mRNAs, one molecule at the time, can then be used as a proxy for elongating Pol II molecules at the gene. This transcript-by-transcript approach has revealed Pol II dynamics at single model loci. Together with analyses of factor dynamics at regulatory site^97,98 (reviewed in Ref. 99), transcription-driven mobility of chromatin^92,100, and enhancer-mediated gene activation^101–103, single molecule imaging of nascent RNA has elucidated kinetics of TF binding and chromatin regulation. Moreover, novel targeting and multiplexing strategies have begun to reveal chromatin features and RNA synthesis simultaneously at multiple endogenous loci^57,100, setting the stage for visualizing transcriptional programs and genome regulation within individual cells.

Figure 3 |. Imaging nascent RNA.

a | Nascent RNAs can be detected via fluorescence in situ hybridization. Dye-labeled probes that are complementary to RNA hybridize to intronic sequences, permitting detection of endogenous nascent transcripts. b | Hairpin-forming sequences that bind the GFP-tagged MS2 protein are engineered into an intron and can be imaged in vivo.

Different methodologies provide complementary information.

Each different approach that measures nascent RNA has unique strengths and limitations (Table 1). While sequencing mature mRNAs reliably measures steady-state levels, only nascent RNA-seq techniques enable robust quantification of changes in RNA synthesis and detect unstable RNA species. These distinct nascent techniques were designed to answer different questions about transcriptional regulation and co-transcriptional processses, and so comparing their findings provides a detailed look at many stages of gene regulation. Of the various nascent RNA-seq methods, PRO-seq⁴⁸ provides a direct measurement of positions of transcription complexes across the genome, enabling mechanistic studies of regulation of Pol II. On the other hand, metabolic labeling, such as TT-seq⁵⁰ and SLAM/TimeLapse-seq^76,79, measures the RNAs synthesized per unit of time. Generally, polymerase densities detected with mNET-seq and in vitro Run-On seq (RO-seq) techniques quantitatively agree across genes. However, mNET-seq can detect all chromatin-associated Pol II molecules, whereas RO-seq methods only detect active RNA in transcriptionally-competent complexes. The requirement in RO-seq for adding nucleotides in vitro excludes detection of Pol II in pre-initiation complexes⁶⁴, and is insensitive for stalled complexes, such as backtracked Pol II^19,104. Reduced detection of initiation complexes is present also in caRNA and mNET-seq techniques due to mappability limit of short (<20 nt) reads. Although RNAs associated with newly initiated elongation complexes are efficiently captured biochemically in caRNA, NET-seq and RO-seq techniques, due to mappability limit of short (<20 nt) read, they are less likely to uniquely align to the reference genome. However, GRO-seq⁵¹, and, length extended run-on sequencing (leChro-seq)⁶⁸ allow generation of longer transcripts during the in vitro transcription, which improves mappability of short transcripts and those that arise from repetitive genomic regions.

Table 1 |.

Strengths and limitations of nascent RNA methodologies

Method	Advantages	Considerations
Chromatin-associated RNA sequencing	• Can be used to isolate all chromatin-associated RNA species • Can be combined with methods that assay co-transcriptional processes, including RNA methylation and editing	• Sequences non-nascent RNAs that stably associate with chromatin also
Start-seq	• Simultaneously identifies initiation and pausing sites • Detects eRNAs	• Does not report transcription beyond the first ~100 nucleotides.
NET-seq	• Is Pol II-specific (antibody enrichment)	• Is limited to cells with epitope-tagged Pol II
mNET-seq	• Is Pol II-specific (antibody enrichment) • Can isolate Pol II with different modifications	• Includes RNAs that are stably associated with Pol II • Does not currently include RNA <30 nt in length
PRO-cap	• Identifies transcription initiation sites • Detects eRNAs	• Does not report transcription beyond the first ~100 nucleotides
PRO-seq	• Captures RNAs from transcriptionally-competent polymerases • Identifies positions of active transcription at nucleotide-resolution genome-wide • Detects eRNAs	• Does not measure polymerase backtracking • Captures RNAs being transcribed from Pol I and Pol III also
CoPRO	• Simultaneously identifies initiation and pausing sites • Measures RNA capping status	• Does not measure transcription beyond promoter-proximal pause site
SMIT-seq	• Measures splicing status during transcription	• Limited to species with short introns
TT-seq	• Captures RNAs from actively-transcribing polymerases • Identifies transcription termination sites	• Does not detect Pol II pausing
SLAM-seq and TimeLapse-seq	• Captures RNAs from actively-transcribing polymerases • Can be used to determine RNA stability	• Requires deep sequencing to measure chemical conversion rate • Long labeling times do not capture newly synthesized RNA
Intron seq-FISH	• Detects transcription of 1000s of genes in single cells • Contains positional information of transcribed genes in the 3D space of the nucleus.	• Does not report chromosomal positions of active Pol II complexes • Does not distinguish different steps of transcription • Requires a library of intron-targeting probes and series of hybridizations

Metabolic labeling also tracks nascent or newly synthesized transcripts, but long labeling times used in traditional experiments causes poor resolution for measuring nascent transcription. Fragmentation of RNA and chemical conversion of labeled nucleotides have improved the positional resolution of the active site of RNA synthesis in metabolic labeling techniques by allowing selection of labeled fragments of transcripts after as little as five minutes of nucleotide incubation^50,79. Moreover, metabolic labeling does not efficiently label nascent RNAs in stationary or slow moving Pol II, and thus is far less sensitive in detecting promoter-proximal pausing than PRO-seq and NET-seq (Figure 2). Isolation of RNA with antibodies to Pol II provides additional information about the status of Pol II. For example, mNET-seq has provided maps of Pol II’s post-translational modifications genome-wide⁴⁶. While caRNA-seq is the least sensitive method for capturing nascent transcripts, it is straightforward to combine with methods that investigate co-transcriptional processes^42,105–107.

Nascent RNA sequencing methods have been adapted for genome-wide identification of enhancers and for analyses of promoter and enhancer architecture. Particularly, high-resolution and high-sensitivity maps of transcription initiation at putative enhancers have been obtained with GRO-cap, PRO-cap, and Start-seq, which identify far more TSSs than prior studies of capped RNA 5′ ends^{4,18,68,108–110}. The distinct cap-selection methods use slightly different strategies to degrade un-capped and to enrich short capped RNAs for library preparation. Although these cap selection strategies highly correlate⁴⁸, the usage of different enzymes could generate differences in cap-selection datasets. Moreover, mis-annealing of reverse-transcript primer has been suggested to cause occasional mis-calling of pause-sites in any data that relies on reverse-transcription from the 3’-adaptor¹⁰⁸. This mis-calling of inidividual pause sites is minor in high-quality datasets¹¹¹. Using several nascent RNA methods together can therefore provide an integrated view of active transcription, co-transcriptional processes, and the half-lives of RNAs. Furthermore, integrating analyses of nascent transcription with mapping of chromatin states and TF binding reveals mechanistic regulation at different steps, such as: promoter or enhancer remodeling, assembly of the PIC, initiation of transcription, Pol II pausing and entry into productive elongation, and termination (Figure 1). Furthermore, biochemical and computational steps have been adjusted to analyze transcripts at early coding sequences, splice-junctions, and polyadenylation and cleavage sites to measure co-transcriptional processes, such as RNA capping, splicing, and cleavage.

Regulating the transcription cycle

Mechanistic regulation of Pol II at distinct steps of transcription underlies expression of individual genes, and coordinates transcription programs. In this section, we will describe different stages of the transcription cycle (Figure 1) and how nascent RNA methods have enhanced our understanding of the mechanistic regulation of the transcription steps and the co-occurring RNA processing (Table 2).

Table 2 |.

Methods used to investigate different steps of transcription

Transcription step	Methods used
Transcription initiation site*	GRO-cap4, PRO-cap48, 5′ Start-seq47, CoPRO52
RNA capping	CoPRO52
Promoter-proximal pausing	mNET-seq46, GRO-seq51, PRO-seq48, CoPRO52, Start-seq47
Co-transcriptional RNA processing	Chromatin-associated RNA-sequencing42,105–107, mNET-seq46,50,60, SMIT-seq163, metabolic labelling75,169, CoPRO52
Transcription termination	TT-seq50, mNET-seq46, PRO-seq48
Pol II CTD modification status	mNET-seq46,60
Transcription bursting	Intron seq-FISH57

^*While the listed methods were designed specifically to detect TSSs, GRO-seq can infer TSSs, as can PRO-seq and NET-seq with long enough reads.

Gene activation is defined by initiation and promoter-proximal pausing.

Early experiments in bacteria and yeast suggested that the main rate-limiting step of transcription is recruitment of Pol II to promoters. In this model, gene activity is primarily controlled by chromatin state and PIC assembly. Promoter-proximal pausing in eukaryotes was characterized at Drosophila major heat shock genes^19,20,112, and similar analyses revealed human c-myc and c-fos genes appeared to have a similar promoter-proximal pausing^113–115. This mode of regulation was initially considered by the field to be a regulatory step that was specific to a few highly-inducible genes; however, studies tracking nascent RNA synthesis, together with other methods, including genome-wide ChIP- and PIP-seq⁶⁷, demonstrated that promoter-proximal pausing is widespread throughout metazoan genomes^{17,18,22,23,44,46,51,55,67,116–118} (reviewed in Refs. 24 and 65). Furthermore, many transcriptional programs, such as tissue morphogenesis in Drosophila and heat shock response in mammals, are coordinated by pause-release^{17,18,21,55,119–124}. However, some programs, including response to estrogen signaling, do primarily drive Pol II recruitment^53,118,125. Perturbation of Pol II recruitment with triptolide and inhibition of promoter-proximal pause-release with flavopiridol confirmed that these two processes are the main rate-limiting steps for gene transcription. Triptolide interferes with transcription initiation by preventing TFIIH helicase from melting the DNA strands, which causes a genome-wide block in transcription initiation^126,127. Flavopiridol, in turn, inhibits CDK9, preventing the release of Pol II from the promoter-proximal paused¹²⁸ (Figure 1). A GRO-seq time-course after flavopiridol treatment demonstrated that 95% of mouse genes require CDK9-dependent pause-release, even if pausing is not apparent in steady-state¹²⁹.

Polymerase recruitment and pausing are inextricably linked. Structural modeling of DNA-bound PIC, Mediator, and promoter-proximal paused Pol II indicated that pausing within 50 nucleotides of the TSS blocks new initiation owing to steric hindrance⁷⁰ (Figure 4A). A recent CoPRO study demonstrated the endogenous relevance of this work by measuring the distance of the pause site from the inititating base genome-wide, finding that pausing nearly always occurs within 50 nt of the TSS in human cells (Figure 4B)⁵². Combined usage of mNET-seq and TT-seq demonstrated that new initiation requires pause release of Pol II⁷⁰. These results together strongly suggest that promoter-proximal pausing precludes a new round of initiation (Figure 4C).

Figure 4 |. Promoter-proximal pausing interferes with transcription initiation.

a | Structural modeling shows that Pol II pausing within 50 nt of the initiation site precludes PIC binding due to steric hindrance. Reprinted from Ref. 70. b | Analysis of coPRO data demonstrates that promoter-proximal pausing occurs within 60 nt genome-wide. Pausing predominantly occurs one base upstream from a cytosine in a GC-rich region. Adapted from Ref. 52. c | Updated transcription cycle model demonstrating that paused Pol II blocks PIC formation.

Molecular events that trigger Pol II to switch from a paused state to an elongating complex have been biochemically characterized (reviewed in Ref. 29) and structurally resolved in context of the Pol II complex by cryo-EM [G]^27,28. In the Pol II pause complex, NELF binding appears to facilitate the RNA-DNA hybrid adopting a tilted conformation that prevents NTPs accessing the Pol II active site²⁸. The structure further suggests that during the switch to productive elongation, PAF binding displaces NELF, thus allowing NTPs to enter²⁷. PRO-seq and mNET-seq measurements of endogenous Pol II pause sites have demonstrated Pol II pausing most likely occurs immediately prior to incorporation of a cytosine, which is the least abundant nucleotide, thus potentially slowing Pol II to permit NELF binding^52,70. Interestingly, GC-rich sequences are prevalent at active promoter-proximal regions with pause sites located >36 nt from the start nucleotide. In comparison, Pol II pausing more proximal to the TSS (<35 nt from the TSS) is generally found at non-protein coding regulatory elements, such as enhancers, and contain less Pol II pausing signal⁵².

Initiation and pausing are regulated at enhancers.

Like promoters, enhancers undergo widespread transcription, as first seen at the β-globin locus ^130,131, and since demonstrated genome-wide in many cell types^{4,18,53,54,109,110,125} (reviewed in Ref. 132). Nascent transcription assays have identified Pol II pausing also at enhancers, in fact the similar divergent pattern of transcription serves as a robust marker for de novo identification of transcribed enhancers genome-wide^{4,18,68,69,109}. The mechanisms and importance of Pol II pausing at enhancers, however, remain largely uncharacterized. At enhancers, Pol II has been shown to exhibit short-lived pausing that is regulated by DSIF, but unlike at promoters, this pausing is not sensitive to NELF knockdown, as seen by Start- and PRO-seq⁶⁹. However, some enhancers have been shown to bind NELF upon gene activation, suggesting that eRNAs may ‘sponge’ NELF from promoters¹³³. Similar steps of transcription regulation at genes and enhancers are further supported by the rapid transcriptional response to heat shock, which causes a genome-wide increase in Pol II density at promoter-proximal as well as enhancer-proximal pause sites^18,52. The coordination of regulatory events at genes and their functionally connected enhancers remain to be characterized, however, release of paused Pol II from promoter-proximal regions can be stimulated by TF-binding to enhancers^134,135, which suggests that CDK9 may reach gene promoters via distal regulatory regions.

Bursting requires initiation and pause-release.

Imaging of nascent RNA production in individual cells has provided evidence that transcription occurs in discontinuous bursts. These bursts were first observed in EM-images of chromosomal spreads¹³⁶, then demonstrated in vivo in a variety of organisms from bacteria to human^{21,83,137,138}. The exact mechanisms that control bursting remain elusive, and the occurrence and kinetics of bursting likely depend on the organism and the regulatory state of the gene¹³⁹. Single molecule imaging of pre-mRNAs from yeast to humans has estimated that a burst is comprised of ~2–100 transcribing Pol II molecules^96,139,140. The probability of a burst occurring (burst frequency) positively correlates with accessibility and priming of the promoter¹⁰¹, and promoters with TATA and Initiator elements tend to have increased burst size⁸⁹. Bursting frequency can be increased by forced enhancer connection¹⁰¹ and inducing histone acetylation at enhancers via targeted recruitment of dCas9-p300¹⁴¹. TFs can influence burst frequency by modulating the chromatin state of promoters and enhancers, but they can also control burst duration, likely by directly activating Pol II machinery at gene promoters¹⁰¹.

Many models of transcriptional bursting suggest that the rapid switching between on and off states is defined at a single regulatory step^85,96 (reviewed in Ref. 142). However, imaging-based methodologies rely on labeling introns in pre-mRNAs and therefore cannot distinguish between Pol II recruitment and initiation and the release of promoter-proximal paused Pol II to productive elongation (Figure 3). To send convoys of Pol II into productive elongation, the ‘single step’ of activation in metazoan species must constitute rapid rounds of coupled Pol II recruitment, initiation, and pause-release⁷⁰. Recent computational modeling of bursting that used intron fluorescence in conjunction with Pol II ChIP-seq suggests that Pol II pause release, not recruitment, causes a burst to initiate¹⁴³.

Stability of promoter-proximal pausing.

The interplay between Pol II pausing, premature termination, and release into productive elongation has been a point of ongoing debate and investigation. Nascent RNA sequencing measurements^{22,70,129,144} show highly varied pause duration, likely owing to distinct promoter sequences, chromatin accessibility, and rate of pause escape. For example, studiesthat inhibited transcription inititation with triptolide used a GRO-seq time-course to track Pol II as it escaped from promoter-proximal pausing, and found that the pause stability ranges genome-wide from 2.5 to 20 minutes (median = 6.9 minutes)^22,129,144. Another study, which modeled Pol II pausing, termination, and release into productive elongation by compaing mNET-seq and TT-seq, found similar pause durations⁷⁰. These datasets agree well with an imaging and biochemical study that so far provides the only direct measure of the promoter-proximal Pol II half-life, quantifying Pol II dynamics at the Hsp70 gene in polytene chromosomes of Drosophila. In its non-induced state, the Hsp70 gene had a 5 minute Pol II half-life at the promoter, but upon heat-activation, the release of paused Pol II into productive transcription increased ~100 fold¹⁴⁵, corresponding to an estimated three second pause half-life in active state. Of note is that premature termination did not decrease upon heat-shock¹⁴⁵, demonstrating that the heat shock induced increase in transcription is regulated by an increase in the rate of released of paused Pol II into productive elongation rather than by an anti-termination mechanism.

Recently, other methodologies have estimated pause durations to be shorter than reported by nascent RNA sequencing studies. In one study, the kinetics of GFP-tagged Pol II were measured en masse: By photobleaching regions of nuclei and measuring the rates at which unbleached GFP-Pol II recovered the fluorescence, the Marteijn group deduced rates at which Pol II can be freed from chromatin¹⁴⁶. This bulk visualization of Pol II, however, cannot discern whether the measured Pol II population came from promoters, gene bodies, enhancers, or even if it was bound to DNA. To overcome these limitations, the study used different drugs and computational modeling, arriving to a conclusion that 10% of Pol II molecules have a half-life of 2.4 seconds, 23% have a half-life of 42 seconds, and the rest have longer residence times in the genome, likely because they are engaged in pausing or gene body transcription¹⁴⁶. Similarly, a single-molecule foot-printing experiment concluded that 68% of genes have pause half-lives up to 5 minutes¹⁴⁷, which is in rough agreement with triptolide inhibition and sequencing measurements¹²⁹. This footprinting assay used methylation to track if DNA sequences were protected by proteins and used modeling to determine if they consituted the PIC or paused Pol II. However, methylation in this assay was performed with ~30-minute enzymatic treatments at physiological temperatures¹⁴⁷, during which time Pol II could escape pausing. Based on various sequencing and imaging studies conducted in distinct laboratories, we reiterate that the half-life of paused Pol II may range from seconds to many minutes, and that the kinetics of Pol II at the pause site are influenced by DNA and RNA sequence, chromatin environment and level of gene activity. In any case, the reported pause durations are long relative to the normal average elongation rate of Pol II (0.024 sec per nucleotide)^49,118,129.

Co-transcriptional processes

Releasing Pol II into elongation is only the beginning: co-transcriptional processes such as 5′ capping, intron splicing, and polyadenylation are essential for productive transcription, as well as for RNA stability and trafficking (Figure 1). Beyond being coupled to Pol II elongation, these co-transcriptional processes influence the efficiency of translation, and can create different transcript isoforms by alternative splicing or polyadenylation (reviewed in Refs. 148 and 149). Recent studies have begun to identify mechanisms of RNA processing, as well as elucidate roles for RNA epi-transcriptomic marks. RNA editing alters individual nucleotides, which can cause changes in splicing, stability, and the protein isoform production (reviewed in Ref. 150). Epi-transcriptomic marks, which include RNA methylation, can alter RNA stability and translation efficiency (reviewed in Ref. 151). Given the importance of such modifications to proper RNA function, understanding where and when these processes occur relative to transcription can offer insights into whether these processes are regulated together, or as independent steps.

5′ capping of RNA occurs prior to or during promoter-proximal Pol II pause.

Shortly after transcription initiation, a 5′ inverted methyl-guanosine cap is added to the 5′-most nucleotide, which provides protection from exonucleases. Early run-on studies using chain-terminating biotin-dNTPs identified that capping occurs shortly after transcription initiation at two Drosophila melanogaster heat shock genes²⁰. These run-on reactions were paired with de-capping assays that demonstrated that 5’-capping predominantly occurred at or before promoter-proximal pause sites²⁰. Additional studies showed capping enzymes bind Ser5-phosphorylated Pol II CTD¹⁵² and interact with DSIF^153,154. Recently, CoPRO (Figure 2C) was used to investigate genome-wide connections between initiation, pausing, and capping in human cells⁵². In this study, the initiating base, active site, and capping status of individual transcripts as short as 18 nt were measured, demonstrating that capping begins when the 5’-end of the RNA emerges from Pol II exit channel, which corroborated earlier work on individual genes²⁰. Capping of RNAs was further compared at pause sites that have either short (20–32 nt) or long (33–60 nt) distance to the TSS, demonstrating a connection between pausing and capping, and suggesting a regulatory role for sequences that influence the pause distance⁵². Promoter-proximal pausing acts as a quality control checkpoint²² and as a mechanism for synchronous activation of genes during development or in response to signaling^{17,18,21,24,55,124}. The distance between initiation and pausing may, therefore, add another dimension for coordinated gene-regulation, providing different opportunities for early RNA processing and Pol II CTD modifications, resulting in preferential recruitment, maturation, and release of Pol II, or folding and modification of the RNA.

Mechanisms of co-transcriptional splicing in yeasts and mammals.

During splicing, non-coding introns are removed from exons by the spliceosome, and these events largely occur co-transcriptionally (reviewed in Refs. 155 and 156). Splicing components have been visualized on nascent RNAs by electron microscopy¹⁵⁷ and ChIP studies have shown them to be associated with chromatin^158–160. One of the most fundamental questions in the splicing field is how splice sites are recognized with a nucleotide precision from the vast variety of pre-mRNA sequences. It has been hypothesized that yeast, which have strong consensus splicing motifs in short introns, use intron definition [G] to demarcate splice sites (reviewed in Ref. 161). Mammals, however, have more degenerate motifs, and introns are generally multiple kilobases in length. These introns are therefore hypothesized to be recognized via exon definition [G] (reviewed in Ref. 161). Sequencing of nascent RNA has allowed more precise measurements of when and where splicing occurs in comparison to transcribing Pol II (reviewed in Ref. 162).

The most direct study of co-transcriptional splicing was performed in yeast using Single-Molecule Intron Tracking sequencing (SMIT-seq). S. cerevisiae yeast have only 250 intron-containing genes, most of which only have one intron. In SMIT-seq, total caRNA is isolated, then intron-containing genes are enriched by using primers specific to their first exons. By using paired-end sequencing, the position of the polymerase is detected with the 3′ read, and the splicing status is measured with the 5′ read¹⁶³. In S. cerevisiae, completed splicing was observed once Pol II had traveled 26 nt downstream of the 3′ splice site, which is the distance required for the spliceosome to access the site, providing strong evidence for intron definition¹⁶³. SMIT-seq in S. pombe, which has more introns and more multi-intronic genes¹⁶⁴, demonstrated that transcripts that failed to splice co-transcriptionally were not cleaved and polyadenylated¹⁶⁵. The coupled transcription, splicing and polyadenylation suggests connections between multiple co-transcriptional events.

SMIT-seq is not readily adapted to most metazoan introns because of their far greater length. Instead, metazoan co-transcriptional splicing has mostly been investigated with indirect methods such as metabolic labeling^75,166, caRNA-seq^42,105, and mNET-seq^46,60. One form of evidence for exon definition would be if splicing was only detected once the subsequent exon had been fully transcribed. mNET-seq has provided some support for the exon definition model in mammals, because spliced transcripts were not detected, even in the reads where Pol II had transcribed into an exon⁶⁰. However, long-read sequencing is needed to interrogate the Pol II position with respect to completed splicing in mammals. Results from NET-seq and GRO-seq demonstrate that Pol II accumulates at exons due to reduced elongation rate^44,129, suggesting that sequence and chromatin features of exons impact splicing. Interestingly, caRNA-seq demonstrated that transcripts largely remain chromatin-associated until splicing completes, even when those transcripts have already been cleaved and polyadenylated^42,105. Although spliceosome components are present in non-chromatin regions of nuclei, these results suggest that proximity to chromatin promotes splicing efficiency^42,167. Further, sequencing RNAs present on the chromatin and in the nucleoplasm demonstrated that constitutive introns were spliced almost entirely co-transcriptionally, whereas alternative introns were often spliced post-transcriptionally¹⁰⁵. Exon definition for constitutive exons, which have stronger splice sites than alternative exons, therefore occurs efficiently during transcription, whereas additional, non-chromatin factors may be required for recognizing alternative exons, as weak splice sites may take more time to be recognized than strong ones.

RNA editing and methylation can occur co-transcriptionally.

RNA can be modified during and after synthesis, including “editing” of individual nucleotides (reviewed in Ref. 150) and modifications, the most common of which is adenosine methylation at the N-6 position (m6A) (reviewed in Ref. 151). Two groups combined nascent RNA sequencing techniques with immunoprecipitation of m6A and found that this methylation occurs rapidly during transcription^106,168. Both groups also measured the effect on splicing in cells with reduced levels of the m6A methylase METTL3. While the Darnell group found that alternative splicing is not affected by m6A in caRNA from mouse embryonic cells¹⁰⁶, the Ørom group did detect an m6A-dependence on splicing rate as measured in HEK293 cells by metabolic labeling¹⁶⁸.

During RNA editing, one nucleotide is chemically converted into another (reviewed in Refs. 150 and 169). Both GRO-seq and PRO-seq show that this editing can occur very rapidly as differences between RNA sequence and the underlying DNA sequence can be detected within 35 nt of the Pol II exit channel, likely occurring in R-loops [G]¹⁷⁰. The most investigated form of editing, adenosine-to-inosine, has been studied with caRNA-seq and shown to affect splicing efficiency in both Drosophila and humans^107,171. More work is needed to determine the timing of competing or synergizing co-transcriptional events.

RNA cleavage precedes transcription termination.

Directly measuring cleaved and polyadenylated RNAs with nascent RNA methodologies is challenging because the polyadenylated transcript is no longer associated with Pol II. However, increased Pol II density, caused by pausing or slowing down at cleavage and polyadenylation sites, is apparent in GRO-seq, PRO-seq, and NET-seq data^{44,46,48,51,172}. Knockdown of the CPSF and CstF cleavage factors results in decreased accumulation of Pol II at the cleavage and polyadenylation site (CPS), which indicates a functional connection between Pol II and RNA-processing factors during co-transcriptional RNA cleavage (Figure 5)⁴⁶. Importantly, knocking down cleavage factors and XRN2 also increases Pol II accumulation near the TSS⁴⁶, suggesting that many transcripts are cleaved at an early stage of transcription, likely as a part of premature termination. These prematurely cleaved transcripts often use cryptic cleavage sites in the first intron and are targeted for degradation by the nuclear exosome¹⁷³. Alternative cleavage and polyadenylation is widespread in mammals (reviewed in Ref. 174); however, we do not yet have robust methods for studying the co-transcriptional dynamics of this process.

Figure 5 |. Observing cleavage and polyadenylation in nascent RNA datasets.

The 3’ end of a gene is demarcated by a polyadenylation signal (PAS) that is slightly upstream of the cleavage and polyadenylation site (CPS). In typical cells, Pol II accumulates at the CPS, which is indicative of a reduced elongation rate. After cleavage, Pol II continues transcribing, but its levels decrease as termination occurs in a termination window. During cleavage factor knock down, Pol II accumulates far less at the PAS and terminates later. Adapted from Ref. 46. When the XRN2 protein is rapidly degraded, Pol II has slightly less accumulation at the CPS but very delayed termination. Adapted from Ref. 172. These results support the torpedo model, whereby XRN2 chases down unshielded nascent RNA after the cleavage, destabilizing Pol II.

After cleavage and polyadenylation, Pol II continues transcribing, on average 8 kb, but it eventually dissociates from DNA and terminates transcription^34,46,50,51. Two models for the mechanism of termination have been proposed: allosteric hindrance, in which Pol II interacts with other factors that destabilize it; and the torpedo model, in which an exonuclease ‘torpedo’ chases down Pol II to trigger termination (reviewed in Ref. 34). Termination is slowed upon degradation of the XRN2 exonuclease, which strongly supports the torpedo model (Figure 5)¹⁷². TT-seq data has been used to identify termination sites by capturing the uncapped RNAs that associate with Pol II after cleavage⁵⁰. On average, genes have four transcription termination sites and the median termination window spans 3 kb. These termination sites are GC-rich, and they overlap with positions where Pol II pauses or slows, suggesting that pausing can aid termination⁵⁰. Conversely, termination may be impaired during cellular stress, resulting in transcriptional readthrough^175,176.

Enhancer RNAs and microRNAs do not appear to use the same cleavage factors as genes. For example, a GRO-seq study showed that eRNAs are cleaved by Integrator¹⁷⁷, which also processes non-polyadenylated snoRNAs¹⁷⁸. Primary microRNAs form a hairpin as they are transcribed, which is recognized and cleaved by the Microprocessor complex in the nucleus, leaving a characteristic 3’ overhang (reviewed in Ref. 179). Studies using caRNA-seq and mNET-seq found that this processing occurs co-transcriptionally^46,180.

Posttranslational modifications of Pol II and chromatin

Regulation of transcription complexes by the C-terminal domain of Pol II.

Pol II is a multisubunit protein that is regulated by GTFs, positive and negative elongation factors, chromatin environment, and sequence-specific TFs (reviewed in Refs. 9, 29, 181, and 182). ChIP-seq and mNET-seq data largely agree on the distribution of Pol II CTD phosphorylated Ser2, the density of which increases towards the 3′ ends of genes and is greatest at the regions where 3′-end processing and termination occur^46,183,184. However, unlike ChIP-seq, mNET-seq does not detect significant accumulation of phosphorylated Ser5 CTDs at promoter-proximal regions of genes after normalizing CTD modification levels to the total amount of Pol II⁴⁶. Moreover, mNET-seq reported high enrichment of Ser5 CTD phosphorylation at exons^46,60, a finding that has gained support from mass spectrometry analyses in yeast¹⁸⁵, but has not been detected by ChIP-seq (reviewed in Ref. 10). The upstream splicing RNA intermediates are likely closely tracking with phosphorylated Ser5 Pol II after the 5’splice site cleavage, thereby enriching mNET-seq libraries with RNAs at those positions^37,50. The discrepancies between ChIP-seq and mNET-seq could be caused by a number of methodological differences. In essence, ChIP-seq detects DNA fragments that are occupied by Pol II after formaldehyde-mediated stabilization of protein-protein and protein-DNA interactions, whereas mNET-seq captures the RNAs that associate with Pol II-chromatin complex without cross-linking, and in the original protocol, without stringent washing⁴⁶. Beyond enriching for distinct chromatin complexes, current NET-seq datasets under-report Pol II that occupies the first ~35 nt from the TSS, likely a consequence of size selection during library, reducing the efficiency of reporting paused Pol II complexes at genes and enhancers⁴⁶. The differences in Pol II CTD phosphorylation obtained by ChIP-seq and mNET-seq need to be clarified to better understand Pol II regulation at initiation, pause-release and splicing.

Interplay of Pol II with local chromatin architecture.

Similar to the CTD code of Pol II, different chromatin modifications also are coupled to transcription. These modifications are suggested to both influence transcription, and to be affected themselves by the progression of Pol II and co-transcriptional processes (reviewed in Ref. 186). For example, histone H3 lysine 4 trimethylation (H3K4me3) is thought to increase transcriptional activity by maintaining an active chromatin state^187,188 and recruiting chromodomain helicase DNA binding protein 1 (CHD1). CHD1 is an ATP-dependent chromatin remodeler that, in turn, maintains trimethylation of H3K36 and H3K4 at transcribed genes, and promotes association of splicing factors and transcription elongation¹⁸⁹. Intriguingly, co-transcriptional splicing of the first exon increases H3K4me3 at the promoter¹⁹⁰. Furthermore, ChIP studies indicate that promoter levels of H3K4me3 inversely correlate with the distance to the first exon, and H3K4me3 at promoters decreases upon inhibition of splicing¹⁹⁰. In a GRO-seq study, the rate of Pol II elongation and its degree of acceleration correlated with the length of the first intron¹²⁹. Introns contain histone H3 lysine 79 dimethylation (H3K79me2)¹⁹¹ and are less nucleosome-dense than exons¹⁹², and long introns may therefore provide Pol II the opportunity to gain full speed¹²⁹ (reviewed in Ref. 181). H3K36me3 is deposited co-transcriptionally at genes and is enriched in exons^193–196 (reviewed in Ref. 197), where ChIP studies show that it blocks cryptic transcriptional initiation in yeast¹⁹⁶ (reviewed in Ref. 181).

The characteristic patterns of histone modifications across the genome have been used to predict functions of specific regions¹⁹⁸. In particular, histone modifications and DNA accessibility have been widely used to distinguish between active enhancers and promoters; the presence of histone H3 lysine 27 acetylation (H3K27ac) and the relative enrichment of mono-methylated H3K4 (H3K4me1) over H3K4me3 are commonly taken to be indicative of enhancer function¹⁹⁹. However, recent studies have questioned whether histone modifications can be used to discern the classes of regulatory elements. A study from our lab identified transcription initiation sites using GRO-cap and categorized transcripts produced from these sites to stable RNAs (i.e., mRNAs) and unstable RNAs (i.e., uaRNA or eRNA) based on cap analysis gene expression (CAGE) [G] data⁴. The results showed that H3K4me1 and H3K4me3 are strongly correlated with polymerase occupancy at both promoters and distal enhancers. H3K4me1 was anti-correlated with polymerase density, whereas marks traditionally associated with promoters, such as acetylation of H3K9 (H3K9ac) and occupancy of GTFs, positively correlated with transcription at the locus, whether it was predicted to be an enhancer or promoter (Figure 6)⁴. Recent work from the Adelman lab extended these findings by showing that genomic sequences that have episomal enhancer activity in self-transcribing active regulatory region sequencing (STARR-seq)²⁰⁰ [G] have prominent levels of H3K4me3 and transcription initiation in vivo as measured by ChIP-seq and Start-seq⁶⁹. Similarly, using PRO-cap, the Furlong group found that bidirectional promoters were more likely to act as enhancers, and that certain enhancers could function as promoters²⁰¹. Furthermore, blocking transcription using flavopiridol decreases H3K4me3 at enhancers⁵⁴. These results demonstrate that H3K4me3 reflects transcription activity but may not distinguish promoters from enhancers. Taken together, the similarities in chromatin state and transcription factor constellation at promoters and enhancers suggest that regulatory elements may exist along a functional spectrum of promoter-ness and enhancer-ness where the histone modifications reflect transcription activity rather than functionally categorize regulatory elements^202,203 (reviewed in Ref. 132).

Figure 6 |. Histone modifications correlate with transcription level.

Previous work suggests that H3K4me1 demarcates enhancers and H3K4me3 demarcates promoters. Nascent RNA experiments demonstrate that the degree of H3K4 methylation instead correlates with transcription level. Enhancers are commonly enriched for H3K4me1 because they have lower transcription than promoters.

Conclusions and future perspectives

Control of RNA synthesis is critical in defining cell types, cellular responses, and organismal functions. Current methodologies can quantify transcribing Pol II complexes at nucleotide resolution with high sensitivity across the genome. The resulting genome-wide profiles of nascent RNA synthesis, coupled with high-resolution maps of factor binding, have uncovered mechanisms that drive gene transcription via regulatory steps, including Pol II recruitment and initiation, promoter-proximal pause-release, productive elongation, and termination (Figure 1). Despite recent advances, several seminal questions remain unanswered. Specific factors have been implicated in regulating certain steps of transcription, yet their detailed mechanistic roles, as well as their interplay with co-factors and the transcription machinery, need to be uncovered in detail. Moreover, RNA processing steps occur co-transcriptionally, but the timing and interconnections between splicing, RNA modifications, and Pol II elongation have not been disentangled. Further, understanding gene regulation requires expanding the mechanistic studies of transcription from genes to transcribed distal regulatory elements, such as enhancers. How genes and enhancers produce fundamentally different transcripts from very similar chromatin architectures, as well as how a promoter establishes directionality to encode stable transcripts in only one direction, remain incompletely understood. It is also unclear how promoters and enhancers each contribute to different steps of transcription, such as Pol II recruitment, initiation, and release from the promoter-proximal pause. Consequently, how transcription is orchestrated in the networks of genes and their distal regulatory elements remain to be fully elucidated, and requires integrative techniques to track nascent RNA synthesis, chromatin connections and factor binding.

Addressing the above questions requires continued and improved use of approaches that monitor the mechanisms by which regulatory signals affect RNA synthesis, binding dynamics of regulatory factors, and chromatin architecture. Both existing and new tools are needed to rapidly perturb specific functions of targeted transcription factors and features of chromatin in living cells. These perturbation approaches include rapid degradation (degron, reviewed in Ref. 204 [G]) of TFs or their inhibition with high-affinity and high-specificity drugs or macromolecules, such as peptides or RNA aptamers (reviewed in Ref. 205) [G]. The perturbation-triggered changes in Pol II distribution can be instantly assessed by nascent RNA methods, providing insights to the mechanistic role of each factor before secondary effects confound the interpretation of results. To address coordinated execution of transcription and of co-transcriptional RNA processing, techniques that use long-read sequencing and require less starting material will open new avenues. Use of less starting material will expand analyses of transcription regulation to individual cells and cell types, including patient-derived samples, specific developmental stages, and comparison of cells from diseased versus healthy tissue. Particularly, the ability to interrogate nascent RNA synthesis in single cells is a prerequisite for understanding cellular heterogeneity, stochastic gene expression, and regulation of cells’ responses in multicellular model systems, such as tissues and organisms. All of these in vivo analyses should be complemented by detailed atomic resolution structures of the large machines like Pol II complexes, chromatin remodelers and enhanceosomes that perform or regulate transcription. In this regard, cryo-EM is already providing precise structures of Pol II and other large regulatory machineries. In the future, applying cryo-EM to complexes in their native state will provide critical insights into interactions controlling transcription. Finally the dynamics of transcription need to be tracked in living cells, ideally with super-resolution [G] methods (reviewed in Ref. 205) to understand the interplay of multiple TFs and elongation factors at enhancers and promoters, and to clarify how transcription is orchestrated in vivo.

Glossary

CORE INITIATION REGION

Short (~ 60 nucleotide) region at promoter and enhancer TSSs that provides a binding platform for GTFs. At promoters, core promoter is used as a synonym for core initiation region

C-TERMINAL DOMAIN OF POL II RBP1

Contains multiple repeats of seven amino acids, and serves as a flexible binding scaffold for transcriptional regulators. The post-translational modifications of the heptad-repeat greatly influence the regulatory interactions, and therefore, transcriptional processes throughout the cycle

CRYO-EM

Electron microscopy technique that visualizes molecules in cryogenic (-200°C) temperatures. Allows near atomic resolution (<4Å) imaging of complex molecules and molecule complexes in their native conformation without crystallization or embedding of the sample

SUPER-RESOLUTION MICROSCOPY

A collective term for light microscopy techniques that provide higher resolution (< 200 nm) than imposed by the diffraction limit of visible light

ENHANCER RNAS

Short (50 – 2000 nt) non-coding RNAs that are produced by Pol II from enhancers. The production of eRNAs moderately correlate with the functional activity of the enhancer

DIVERGENT TRANSCRIPTION

In metazoan, genes and enhancers drive transcription from two core initiation regions from both strands in opposing directions. On genes, the mRNA-coding strand is termed sense, and the anti-sense strand produces the divergent transcript

ChIP-SEQ

Method in which protein-DNA interactions are stabilized, chromatin is sheared, and fragments with a protein of interest enriched using an antibody. Purified DNA from the enriched fragments is sequenced, providing genome-wide maps of protein localization

RIP-seq

A technique by which RNAs that interact with a given protein can be identified. Utilizes antibody-mediated enrichment of a protein, after which, its interacting RNAs can be isolated, reverse-transcribed, and sequenced

PAIRED-END SEQUENCING

High-throughput sequencing of DNA-molecules from both ends, which provides information from 3’- and 5’-ends of each DNA fragment, and allows more accurate mapping of the reads to the reference genome

PIP-SEQ

Permanganate footprinting detects unpaired thymine in the DNA, providing the exact locations of open transcription complexes (transcription bubble) across the genome

INTRON DEFINITION

Splicing model in which specific sequences that demarcate introns are sufficient for spliceosomes to recognize intron boundaries

EXON DEFINITION

A model by which proteins that bind to exons are required for the spliceosome to to recognize sequences that demarcate introns

R-LOOPS

Three-stranded DNA:RNA hybrid structures formed for example by template DNA, the complement nascent RNA and the non-template single-stranded DNA

CAGE

Cap analyses gene expression technique that measures RNA expression and maps TSS of gene promoters. Provides precise maps of TSSs of genes that produce long-lived transcripts

STARR-SEQ

Method that assays enhancer activity for millions of candidate sequences by cloning them downstream of a reporter gene and upstream of a CPS. Functionally active enhancers drive expression of RNA molecules that contains the candidate sequence

DEGRON

Tool in which genome editing adds a tag that is encodes a protein recognized by the E3 ubiquitin-ligase complex. After addition of a small molecule, the edited factor is inducibly degraded

RNA APTAMER

Structured RNA molecule selected for binding to a factor of interest to disrupt its functions or interactions