The Promoter Proximal Pause: a decision point governing RNA polymerase II fate

Abstract

A signature feature of transcription on most genes in multicellular animals is that RNA polymerase II (RNAPII) piles up about 50 bases downstream of the start site at the promoter-proximal pause (PPP). Promoter proximal pausing is controlled by positive and negative elongation factors that associate with RNAPII. There are two major outcomes for promoter-proximal paused RNAPII complexes: release into the gene body and premature termination. Here we discuss how RNAPII dynamics at the PPP function in a quality control checkpoint and in regulation of RNAPII flux through genes. We propose a Pause Release Attenuation model to describe RNAPII dynamics at the PPP.

Introduction

The transcription cycle of initiation, elongation, and termination by RNA polymerase II (RNAPII) begins with recruitment of the polymerase to the promoter together with general transcription factors (GTFs) and mediator to form a large pre-initiation complex (PIC). RNAPII recruitment and PIC assembly are controlled by chromatin remodelers (1) and sequence-specific transcription factors that contact mediator and other PIC components (2–5). After PIC assembly, promoter DNA is melted by the TFIIH-associated XPB helicase to expose the template strand, and transcription is initiated followed by RNAPII escape from the promoter when an RNA-DNA hybrid of ~10 base pairs completes the transcription bubble (6). At most metazoan genes, RNAPII stalls at the promoter proximal pause (PPP) after transcribing 40–50 bases and before encountering the first nucleosome downstream of the transcription start site (TSS) (7–10). The PPP marks a decision point where a choice is made between two “fates”: (i) pause release and a resumption of elongation which may take many hours for the longest human genes or (ii) premature termination that evicts the polymerase from the template (Figure 1a). Here, we review how the PPP is established, the two alternative “fates” of paused RNAPII, and how the decision between these fates may be regulated to control gene expression and impose quality control over elongation complexes (ECs) entering the gene body. Other excellent recent reviews of this topic include (11–14)

Figure 1. The Pause Release Attenuation model of RNAPII dynamics at the PPP.

(a) Paused RNAPII has alternative and competing fates: pause release and premature termination. (b) Idealized RNAPII density plots before (red) and after gene activation (green) based on anti-RNAPII ChIP-seq.

Discovery of promoter-proximal pausing and premature termination

The first evidence of promoter proximal pausing came from studies of DNA tumor viruses. In vivo pulse labelling and in vitro nuclear run-on assays of infected cells revealed an unexpected bias with much higher levels of transcription at the 5’ ends than at the 3’ ends of viral genes, consistent with a slow-down or pause resulting in a pile-up of polymerases in the 5’ region (15–19). Pausing near the 5’ ends of the SV40 late, Adenovirus major late, and HIV-1 transcription units can be followed by premature termination to produce discrete short transcripts (15; 18; 20–24). Notably, the HIV-1 transactivator Tat suppresses premature termination to induce transcription of the provirus (23). Pausing and premature termination at a run of T residues about 90 bases downstream of the Adenovirus major late promoter can be recapitulated in HeLa nuclear extract (NE) (25). Seminal observations of the Adenovirus Major late transcription unit revealed that the antiviral ATP analogue DRB (5,6-Dichloro-1-β-D-ribofuranosylbenzimidazole) limited transcription to a few hundred bases arguing that it inhibits elongation, and that sensitivity to this inhibitor was lost in a partially purified transcription system (18; 20; 26). Together, these studies suggested that viral transcript production might be regulated by a decision between productive elongation and premature termination (17–19; 23).

Subsequently, nuclear run-on measurements of RNAPII occupancy on cellular genes, including chicken β-globin (27), MYC and Hsp70 by identified a similar 5’ bias consistent with promoter proximal pausing. Notably a 5’ peak of RNAPII was detected on the MYC and heat shock genes even when these genes were not actively expressed and polymerases failed to enter the gene body (28–32). Stalled RNAPII complexes were resistant to Sarkosyl and associated with a transcription bubble, demonstrating that they were fully engaged EC’s (32; 33). Groudine and colleagues suggested that the PPP was a “general rate-limiting step” in the RNAPII transcription cycle (34) and this prediction was subsequently validated by genome-wide RNAPII profiling in Drosophila and human cells (7–9; 35; 36). Thus, the PPP is not a specialized adaptation for a few highly inducible genes, but rather is a hallmark of most genes regardless of expression level.

Models of RNAPII dynamics at the PPP

Early work showed that when highly inducible genes like Hsp70 and MYC, and the HIV-1 provirus are activated, RNAPII density within the gene body increases relative to the peak at the 5’ end (23; 28; 29; 31) (Figure 1b). These findings suggested that control could be exerted post-initiation by regulating the elongation phase of the transcription cycle echoing attenuation in bacterial operons which also depends on polymerase pausing (37). The most popular interpretation of the altered RNAPII density profile following gene activation is the “Pause Release model” that invokes a mechanism similar to a “valve” that controls the fraction of polymerases at the PPP which are released into the gene body to carry out productive elongation (14; 34; 38; 39). Pause release differs in an important way from bacterial attenuation where transcription is limited by premature termination. Recent work has revealed that in addition to release into productive elongation, premature termination is a major alternative fate of RNAPII at the PPP that compete with one another (Figure 1a). Pausing is a universal pre-requisite for termination, and the longer a pause persists, the greater the window of opportunity for premature termination to occur. Conversely, pause release precludes termination. Hence, pausing and premature termination are not independent of one another; instead they are both integral to RNAPII dynamics at the 5’ ends of genes. The original pause-release model does not account for the active RNAPII turnover now known to occur at the PPP. We propose a revised hybrid Pause Release Attenuation model (Figure 1a) to describe RNAPII trafficking at the PPP.

Pausing and premature termination factors

Pausing is an intrinsic property of elongation by all RNA polymerases. Elongation is not continuous but is interrupted frequently by RNAPII stopping for a period before adding the next nucleotide. Pauses may persist anywhere from a few seconds to many minutes in vitro depending on NTP concentration and DNA and RNA sequence context (40). Paused RNAPII can have multiple alternative fates (Figure 1a): it can resume productive elongation through pause release, it can be evicted from the template through premature termination, or it can backtrack with displacement of the 3’ end of the transcript from the active site. Many factors influence entry and exit from the paused state.

• RNAPII CTD: The C-terminal domain (CTD) of the large RPB1 subunit is a unique signature feature of RNAPII that is composed of 52 conserved heptad repeats in mammals (YSPTSPS). Dynamic phosphorylation and dephosphorylation of the CTD, primarily at S2, S5, and S7, is coordinated with the transcription cycle and is associated with changes in elongation, pausing, and premature termination (41–46).

• CDK7: The CDK7 kinase is integral to the GTF TFIIH and it phosphorylates the RNAPII CTD at Ser5 and Ser7 residues when transcription initiates to promote release of the polymerase from the mediator and dissociation of the PIC (47). CDK7 activity is critical for the exchange of initiation factors for elongation factors (e.g., DSIF) (48; 49), and inhibition impairs the transition to productive elongation (48; 50; 51).

• DSIF: The DSIF (DRB-sensitivity-inducing factor) complex (comprising Spt4 and Spt5) was discovered as one of the two factors (along with NELF) that confers sensitivity to inhibition of elongation in vitro by DRB (52). DSIF has both positive and negative effects on pausing that are toggled by phosphorylation of different regions of Spt5 (53–56). DSIF “travels” with RNAPII along genes (57) and is localized at promoters, within gene bodies and at 3’ ends of active genes (36; 58).

• NELF: Negative Elongation Factor (NELF) is a 4-subunit complex that, like DSIF, is required for inhibition of elongation in vitro by DRB (59). NELF binds to RNAPII and RNA and inhibits elongation by covering the funnel through which NTPs enter the RNAPII secondary channel (60) that leads to the active site. NELF binding also competes with the positive elongation factors TFIIS and PAF1C (61). NELF is localized with RNAPII at the PPP and is normally released prior to entry into the gene body (36; 62).

• P-TEFb: Positive-Transcription Elongation Factor-b (P-TEFb) comprises the cyclin dependent kinase CDK9/CyclinT1 which is inhibited by DRB (63–65). P-TEFb phosphorylates numerous substrates (66) including the RNAPII CTD (67), DSIF and NELF. P-TEFb can be sequestered in an inactive form with the 7SK RNP (68) and its major active form is within the super elongation complex (SEC) (69). Phosphorylation of DSIF and NELF by P-TEFb is sufficient to alleviate the negative effects of these factors and permit entry into productive elongation in vitro (70; 71). P-TEFb is primarily localized at 5’ ends of genes (72; 73).

• PAF1C: The Polymerase-Associated Factor 1 Complex (PAF1C) (74) is a 6 subunit complex that makes multiple contacts with RNAPII (75) including the phosphorylated CTD (76) and is a major potentiator of transcription elongation (77). It stimulates elongation by stabilizing RNAPII complexes with the SEC (73) and through direct of its Rtf1 subunit with phosphorylated Spt5 (78). PAF1C also stimulates elongation by recruitment and allosteric activation of the major CTD Ser2P kinase CDK12/CyclinK (43; 73). PAF1C appears to travel with RNAPII and localizes throughout the length of genes (62).

• CDK12/CyclinK: The closely related CDK12/CDK13 cyclin dependent kinases are allosterically activated by the Cdc73 subunit of PAF1C to preferentially deposit a dual phosphorylation mark at Ser2 and Ser5 of the CTD heptad repeats which can be processed by a phosphatase to generate a single Ser2P modification (43) characteristic of elongating RNAPII (79). Simultaneous inhibition of CDK12/CDK13 reduces the speed and processivity of transcription arguing that these kinases prevent early termination within the gene body (80; 81). CDK12 co-localizes with RNAPII in a peak at the PPP (73).

• TFIID: TFIID is a GTF that nucleates PIC assembly by binding to promoter DNA elements upstream and downstream of the TSS including the TATA box and the downstream promoter element/pause button (82). Binding of TFIID to downstream promoter DNA likely contributes to its stabilization of paused RNAPII (83).

• TFIIS: TFIIS binds to RNAPII by insertion of its N-terminal Zn ribbon domain into the secondary channel to contact the active site and allosterically activate the endogenous RNA endonuclease activity of the polymerase (84). RNA cleavage creates a new 3’ OH end that can be re-positioned in the active site to allow a backtracked polymerase to recover and resume elongation (85; 86). TFIIS is localized with RNAPII along the entire length of active genes. It stimulates exit from the PPP (35; 87) and maintains rapid elongation throughout the gene body (88).

• TFIIF: TFIIF is a GTF required for initiation that also stimulates early elongation (89) by competing with the negative effector NELF for a common binding site (61; 90) and by cooperating with TFIIS to stimulate elongation (91). TFIIF is displaced when the paused EC is established, likely by PAF1C and SEC (6). Gdown1 is a negative factor that stabilizes the PPP by competing with TFIIF binding (92; 93).

• SPT6: SPT6 is a histone chaperone that binds RNAPII in a manner that depends on phosphorylation of the linker between the RNAPII body and the CTD (94). SPT6 facilitates elongation through nucleosomes that present a barrier to polymerase causing it to pause at specific histone-DNA contact points (95). Acute depletion of SPT6 leads to RNAPII accumulation at the +1 nucleosome, suggesting that it probably facilitates nucleosome transfer from the front to the back of the polymerase (96; 97). SPT6 also facilitates elongation by promoting recruitment of PAF1C and release of NELF from paused complexes (62). SPT6 probably travels with RNAPII and is located throughout gene bodies and at 3’ ends (98).

• FACT: The Facilitator of chromatin transcription (FACT) complex is a histone chaperone composed of SPT16 and SSRP1 (99) that transfers nucleosomes from the front to the rear of transcribing polymerase (97; 100; 101) thereby alleviating the nucleosomal barrier to elongation.

• CRL3^ARMC5: ARMC5 and Cullin3 (CRL3) are components of an E3 Ubiquitin ligase that contacts RPB1 directly through ARMC5 to direct its ubiquitylation and degradation (102). This degradation targets Ser5P phosphorylated RNAPII at the PPP resulting in premature termination of defective ECs and thereby serves a transcriptional quality control checkpoint function (103; 104).

• Integrator: This 17-subunit complex is responsible for termination of transcription at the 3’ ends of snRNA genes (105) and for premature termination of RNAPII complexes at the PPP (106). Premature termination by integrator involves the concerted action of its RNA endonuclease that cuts the nascent transcript about 20 bases upstream of the 3’ end (107; 108) and its PP2A phosphatase that dephosphorylates the CTD and Spt5 (109–112). Integrator can localize throughout the length of genes (113) but is enriched at 5’ ends (114).

• Xrn2: Xrn2 is a nuclear 5’-3’ exoribonuclease that digests nascent RNA with an exposed 5’ phosphate and functions as a “torpedo” that is essential for termination coupled to polyA site cleavage at 3’ ends (115; 116). In fission yeast at least, Xrn2’s “torpedo” function is stimulated by contact with Spt5 (117). Xrn2 also co-localizes with RNAPII within gene bodies and at the PPP where it is implicated in premature termination (117–119).

• PP1/PNUTS: The PP1 phosphatase and its partner PP1 Nuclear Targeting Subunit (PNUTS) control deceleration of RNAPII and release of Paf1C after transcription of the polyA site in preparation for termination (58; 120; 121). In a potentially related function, PP1/PNUTS suppresses expression of Kaposi’s Sarcoma Herpes Virus (KSHV) genes possibly by promoting premature termination (122). On the other hand, PP1/PNUTS was also found to enhance promoter-proximal pause release and expression of cellular genes (123; 124).

• Restrictor: Restrictor is a termination factor comprising the RNA binding Zinc finger protein ZC3H4 and the WD40 domain protein WDR82 (125–127). Termination requires the PP1-PNUTS phosphatase which is contacted by the WDR82 subunit (128–130). ZC3H4 recruits the exosome via contacts with ARS2 to degrade terminated transcripts (130; 131). Restrictor operates predominantly on divergent antisense transcripts from bidirectional promoters, but it also terminates some sense transcription in a region downstream of the PPP (132).

• Spliceosomal snRNPs: Much pre-mRNA splicing is performed co-transcriptionally on the nascent transcript by the spliceosome acting near RNAPII (133). How splicing might be coordinated with transcriptional pausing is an active area of investigation. The U1 snRNP which recognizes 5’ splice sites makes protein-protein contacts with elongating RNAPII and SPT6 (134; 135) and could affect pausing by competing with the elongation factor Rtf1 (a subunit of PAF1C) (136). Furthermore, Pladienolide B inhibition of the U2 snRNP, which recognizes branch points within introns, increases RNAPII accumulation at the PPP though the underlying mechanism remains unclear (137).

The Functions of pausing

Limiting transcriptional output

The PPP delays RNAPII entry into the gene body and if the pause is prolonged, it could significantly limit the flux of polymerases through the gene body and hence transcriptional output. This is the basis of the Pause Release model of regulated elongation. A second way that a paused RNAPII could limit transcriptional output is by occluding the promoter to prevent new initiation events (Figure 2a). While this idea has not been tested directly, for example by monitoring formation of newly capped 5’ ends (138), inhibition of pause release modestly diminished RNAPII occupancy in a region upstream of the TSS consistent with reduced initiation (139). At promoters where the paused polymerase is highly dynamic however (see below), steric hindrance of initiation is expected to be less important.

Figure 2. Functions of RNAPII pausing.

(a) Pausing limits transcription by two potential mechanisms: (i) by blocking re-initiation and (ii) increasing the time spent in the promoter to delay entry into the gene body and promote premature termination. (b) Pausing primes genes for RNA synthesis in later stages of development or for rapid activation in response to environmental stimuli. (c) Pausing provides time to carry out a quality control “checkpoint” that eliminates “incompetent” EC’s including those where capping has failed.

Potentiating transcriptional activation

In contrast to its potential negative regulatory functions, pausing could play a positive role by “poising” promoters for activation (Figure 2b). Pre-loading a paused and transcriptionally engaged polymerase could accelerate activation because potentially rate limiting establishment of an open chromatin conformation and RNAPII recruitment have been completed in advance. Consistent with this idea, a peak of paused RNAPII without productive transcription into gene bodies often anticipates future activation of developmentally regulated (140–142) and highly inducible (32; 143) genes (Figure 2b). In agreement with the ‘poising” model, genes in the Drosophila embryo with paused RNAPII were activated synchronously throughout a tissue whereas those without paused RNAPII were activated more stochastically (144; 145). On the other hand, the prominent PPP is not peculiar to inducible genes but is instead common to almost all genes in metazoans including housekeeping genes, begging the question of whether it can serve a function other than priming gene activation. Additional potential positive functions of pausing are: (i) paused RNAPII may exclude nucleosomes and provide a “bookmark” in the chromatin that can be easily accessed by the transcriptional machinery (146; 147) and (ii) paused RNAPII may stabilize loops between promoters and enhancers that are thought to prime genes for developmentally regulated expression (148). Most genes are transcribed in bursts separated by periods of inactivity (149; 150), and it is also possible that in the intervals between bursts, paused polymerases maintain promoters in a permissive state for the next burst (151). While many studies show a correlation between gene activation and an increased ratio of promoter distal:proximal RNAPII density consistent with pause release (31; 152–155), it remains to be unequivocally established that a pause is actually necessary for gene activation in most cases.

Quality Control

Treatments that disrupt assembly of a competent EC destabilize RNAPII at the PPP which is phosphorylated on CTD Ser5 residues. These treatments include low level triptolide (Trp), the inhibitor of initiation that covalently inactivates the XPB helicase (156), inhibitors of CDK7 and CDK9 kinases, and knockdown of capping enzyme, Spt5 or integrator subunits (103; 104; 157; 158). Two partially redundant mechanisms are responsible for the removal of promoter proximal polymerase under these conditions; CRL3^ARMC5 dependent ubiquitylation/degradation and Integrator dependent premature termination (103; 104; 157; 158). These results strongly suggest that a checkpoint at the PPP eliminates complexes that fail to qualify as competent for release into the gene body (159) (Figure 2c) although the details of how this quality control test works remain to be explored. Loss of the ARMC5 component of the checkpoint apparatus, which occurs in patients with adrenal hyperplasia, results in over two-fold elevation of the cellular pool of RNAPII (160). Hence, the checkpoint appears to target a remarkably large fraction of RNAPII for degradation, suggesting that the failure rate of ECs at the PPP is very high, in agreement with the notion that premature termination is the predominant fate of paused RNAPII (161–164).

Detection of pauses

Pausing is detected in vitro by following RNA chain growth over time using pulse-chase protocols where the 5’ region of the transcript is labelled with limiting NTPs that cause RNAPII to arrest (165) or in kinetically synchronized reactions with limiting NTPs (70). Transcription in NE with physiological NTP concentrations does not normally produce many paused transcripts (166) but withholding nucleotides causes pausing that can be reversed by re-addition of NTPs (165). This approach has been invaluable to studies of pause release and the factors that control it. Pausing and the resumption of elongation by individual RNAPII molecules can also be directly observed using single molecule methods even when NTPs are not limiting (40; 167).

Most studies of pausing in vivo measure it indirectly by monitoring relative RNAPII density at different positions within genes. The underlying principle is that polymerases will pile up at a pause site and following release they will spread out and density will decline analogous to cars on the road before and after a stop light. RNAPII density is measured by nuclear run-on, anti-RNAPII chromatin immunoprecipitation (ChIP), and anti-RNAPII CUT&RUN, all of which can be applied genome-wide. These methods provide average values of RNAPII occupancy over a population of cells, but it can also be measured on individual DNA molecules by DNA methyltransferase footprinting (168). In addition, imaging of tagged RNAPII by Fluorescence Recovery After Photobleaching (FRAP) is a powerful way to infer RNAPII residence times in live cells (161; 162).

Anti RNAPII ChIP-Seq and CUT&RUN reveal RNAPII occupancy at a resolution of ~20–100 bases which can be improved by employing exonuclease digestion (139). A disadvantage of these methods is that they cannot distinguish transcriptionally engaged polymerases from others in proximity to DNA nor do they distinguish the strand being transcribed. The global nuclear run-on methods, PRO-Seq (10) and GRO-Seq (8), which both provide strand information, utilize elongation in isolated nuclei or permeabilized cells to incorporate labeled nucleotides and PRO-Seq has single nucleotide resolution. These assays are performed in the presence of Sarkosyl, which inhibits initiation but permits ongoing elongation, thereby capturing engaged RNAPII complexes. The disadvantage of the run-on approach is that any polymerases which fail to resume elongation in the reaction will not be detected. The most powerful method for pause detection in vivo is arguably Nascent Elongating Transcript sequencing (NET-Seq) introduced by Churchman and Weissman (169) and improved by Proudfoot and colleagues (170). NET-Seq maps the 3’ ends of RNA fragments which are immunoprecipitated with RNAPII thereby precisely localizing the active site of transcriptionally engaged polymerases. A limitation of the method is that the 3’-OH of a RNAPII-associated RNA will not be at the active site if the polymerase has backtracked. Churchman and Weissman (169) introduced a useful operational definition of a pause site as a nucleotide position where polymerase density is >3 standard deviations above the mean for the window surrounding it. NET-Seq identified a consensus sequence for pause sites in human cells at G residues followed by a pyrimidine with a modest preference for elevated G content in the transcript upstream of the pause (54; 171). The same consensus sequence is enriched at the promoter proximal pause and at pauses within the gene body (54).

P-TEFb, DSIF and NELF; key regulators of the PPP

Marshall and Price observed that during a short pulse of transcription with limiting NTPs, paused transcripts ~20–100 bases long were rapidly produced, but in a subsequent chase these paused complexes were slow to resume elongation and frequently terminated prematurely (64). They called this inefficient mode of transcription abortive elongation. There is a slow transition from abortive elongation to productive elongation which was stimulated by high salt, Sarkosyl, or heparin (64). These treatments, which strip factors off ECs, greatly increase processivity, that is the efficiency of full-length transcript synthesis, presumably by suppressing premature termination. In summary, this early in vitro work revealed two possible fates of paused polymerases: premature termination and productive elongation (Figure 1a). In a pivotal experiment, it was discovered that the addition of NE to isolated pre-initiation complexes stimulated the production of long transcripts and that this activity, designated Positive Transcription Elongation Factor-b (P-TEFb), was inhibited by DRB (64). P-TEFb was purified and identified as the cyclin dependent kinase CDK9/Cyclin T1 (63) which is the same kinase required for Tat activation of HIV transcription (172; 173). Immunodepletion of P-TEFb from NE inhibited production of long transcripts and rendered transcription DRB-resistant, suggesting that this kinase is the target responsible for DRB inhibition of elongation (173).

While transcription in unfractionated NE is inhibited by DRB, this is not the case for a partially purified system (26). In seminal studies, Handa and colleagues purified two factors, DSIF and NELF, that restored sensitivity to DRB when added to a partially purified transcription system (52; 59). Conversely, immunodepletion of either factor from NE, rendered transcription DRB resistant (52; 59). DRB sensitivity inducing factor (DSIF) is composed of Spt5 and Spt4 that correspond to genes identified in budding yeast as suppressors of transcription defects caused by transposable element insertions. Spt5 is a homologue of the bacterial and archaeal elongation factors NusG (52; 174), making it the only TF conserved in all 3 kingdoms of life. NELF comprises 4 subunits (NELF-A, B, C/D, E) and, unlike DSIF, is unique to a subset of multicellular animals (absent from C. elegans) where transcription is characterized by a prominent PPP. Both DSIF and NELF bind directly to RNAPII and to RNA, and when added to a partially purified transcription system, they cooperatively inhibit elongation by prolonging dwell time at promoter proximal and distal pauses (59; 90) and this activity is augmented by DRB. On the other hand, DSIF also accelerates elongation in vitro when NTP’s are limiting (52). Hence DSIF is both a positive and a negative regulation of transcription elongation. While immunodepletion of DSIF or NELF from extracts is sufficient to make them resistant to DRB, it does not greatly affect the efficiency of transcription without DRB (52; 59). On the other hand, add back of recombinant DSIF or NELF to depleted extracts inhibits production of full-length transcripts (59; 70). The inhibition of elongation by NELF and DSIF is abolished if they are phosphorylated by P-TEFb (90). While P-TEFb greatly stimulates transcription in the presence of DSIF and NELF, it has no effect in their absence (70), showing that the positive function of this kinase in vitro is to counteract the negative effects of DSIF and NELF.

Precise mapping of pause sites by NET-Seq showed that high frequency pausing occurs throughout a pausing zone that can extend hundreds of bases from the TSS depending on the gene (54) and is not always limited to a small region upstream of the first nucleosome. A long pausing zone is consistent with transcripts extending several hundred bases even when elongation is inhibited by DRB (20; 175; 176). When Spt5 is depleted gradually, RNAPII levels are maintained and the PPP is relatively unaffected, but pausing zones extend much further into genes consistent with the model that DSIF facilitates the “maturation” of ECs from a high pausing to a low pausing state competent to transcribe long distances (53; 54). Degron depletion of Spt5 and add back of mutants has revealed insights into dual effects of DSIF on pausing and elongation. The switch between the positive and negative effector functions of DSIF is controlled by P-TEFb phosphorylation (53; 75) and PP1/PP2A dephosphorylation (58; 109; 120) of multiple sites on Spt5. The hypophosphorylated C-terminal repeat region 1 (CTR1) of Spt5 enhances pausing, and phosphorylation reverses this effect converting Spt5 into a suppressor of pausing throughout the length of genes (53; 54). Phosphorylation of Thr4 residues in the CTR1 repeats (GSR/QTP) enhances elongation by recruiting the Rtf1 subunit of PAF1C (78) that acts as an allosteric activator of RNAPII translocation (60). In contrast, the CTR2 region of Spt5 acts as a “brake” that limits elongation possibly by inhibiting Rtf1 recruitment (55). Dephosphorylation of the RNA binding KOW4-5 linker region of Spt5 switches the protein into a potent enhancer of pausing (54; 56). Indeed inhibition of Spt5 KOW4-5 linker phosphorylation by DRB is likely to explain much of this drug’s inhibitory effect on elongation (54; 109).

How DSIF and NELF cause RNAPII to pause has been elucidated by critical cryo-EM studies reviewed in detail elsewhere (12) and only briefly summarized here. DSIF stabilizes the EC by contacting RNAPII itself, as well as the non-template strand of the transcription bubble via its DNA clamp, and the nascent RNA via its RNA clamp (177). A transcript at least 18 bases long is required for DSIF recruitment (178). Spt5 provides a docking site for NELF, and the RNAPII-DSIF-NELF complex assumes a “tilted” RNA-DNA hybrid conformation within the active site, which is also found in paused bacterial RNAP (61; 179). This conformation hampers polymerase progress because there is no unpaired base in the template strand available for base pairing with an incoming NTP. In addition, NELF impedes forward progress by occluding the secondary channel which prevents entry of NTPs and blocks TFIIS binding. NELF also interferes with trigger loop movement that is critical for translocation (61). In a second so-called “poised” conformation of the RNAPII-DSIF-NELF complex, the RNA-DNA hybrid is not tilted and NELF does not interfere with TFIIS binding (180). It may represent an intermediate between paused and productively elongating states.

In addition to the canonical elongation regulators DSIF and NELF, the GTF TFIID also facilitates pausing in vitro in a purified system (83). This activity is consistent with the fact that at least in Drosophila, TFIID recognizes two sequence motifs that may lie downstream of the TSS: “pause button” and the “initiator” which are both enriched at genes with a prominent PPP (82; 181). The TATA box, which is bound by TFIID upstream of the TSS, is also implicated in regulation of pausing at NF-kB target genes by DSIF and P-TEFb (182).

PPP and gene regulation

The importance of pausing for gene regulation has been tested by perturbing the canonical pausing factors NELF and DSIF using mutants or depletion with RNAi knock down (KD) or CRISPR-engineered degrons. Depletion of NELF or Spt5 does not suppress the PPP at most genes, but shifts the RNAPII peak a few bases downstream to a position that abuts the +1 nucleosome (54; 157; 183; 184). Following RNAi KD of NELF in Drosophila S2 cells or mouse embryonic stem cells (mESCs), only a small subset of genes was upregulated, as would be expected if there was uncontrolled pause release (147; 185). NELF depletion also has relatively minor effects on developmentally regulated gene expression programs (140; 186), but it does disrupt FGF signaling and NANOG activation in mESCs (185; 187) and regulation of genes required for energy homeostasis (188). On the other hand, NELF is dispensable for rapid activation of heat shock genes (183; 189), indicating that a NELF stabilized pause is not essential for “priming” this response. Mutation of Spt5 or inhibition with a drug disrupts specific regulatory pathways including EGF (53) and NF-KB signaling (190) as well as differentiation but not self-renewal of mESCs (191). In summary, contrary to what might have been expected based on their in vitro properties, NELF and DSIF are not generally required to establish the PPP in vivo and their depletion does not cause widespread release of paused ECs, but specific gene regulatory programs are disrupted. Somewhat paradoxically, NELF KD downregulated more genes than it upregulated, suggesting that in most cases this factor functions as a positive rather than a negative regulator of gene expression. Such a positive effect is likely mediated through “bookmarking” of open chromatin regions resulting from nucleosome exclusion by paused RNAPII (146; 147; 185; 192).

The apparent differences between NELF and DSIF effects on PPP in vivo versus in vitro, where the template is naked DNA, likely reflect the influence of chromatin. As it invades a nucleosome, RNAPII pauses at positions where histones contact DNA (95; 169; 193). The +1 nucleosome downstream of the TSS is the strongest barrier to RNAPII, which is weakened by incorporation of the histone variant H2A.Z (194). Acute depletion of the histone chaperone FACT delocalizes the +1 nucleosome and re-positions the PPP highlighting the importance of this nucleosome for pausing (101). It is unclear whether with respect to gene regulation, there is any functional redundancy between “intrinsic” NELF/DSIF dependent pausing and pausing that results from collision with the +1 nucleosome. Histone N-terminal tail modifications can also influence pausing within nucleosomes (195). The Set1 containing COMPASS complex methylates K4 residues in the Histone H3N-terminal tail, a modification enriched in promoter-proximal regions. Degron depletion of COMPASS reduced H3K4me3 and increased the stability of paused RNAPII while reducing transcription within gene bodies, suggesting that this modification stimulates pause release. However, it is also possible that reduced integrator recruitment when H3K4me3 is suppressed could stabilize paused RNAPII by preventing premature termination (196).

How stable is the promoter-proximal pause?

A critical question about the PPP is how long RNAPII remains there as this determines the windows of opportunity for the competing processes of pause release and premature termination. This question has been addressed by inhibiting transcription initiation with Triptolide (Trp) and monitoring the decay in RNAPII occupancy at the 5’ peak by ChIP (139; 197; 198), global run-on (199; 200), or single-molecule methylation protection (168). This approach has led to estimates of the pause half-life that vary between genes from 1–2 min to over 30 minutes depending in part on the Trp concentration used. Some of this variation is related to the limitation that Trp acts quite slowly (184). Furthermore, not all initiation events are necessarily dependent on XPB (201) and not all RNAPII molecules detected by ChIP-Seq are transcriptionally engaged as they include PICs (197; 198). Hence, estimates of residence time using Trp likely represent an upper limit. Another strategy is to rapidly inhibit PIC formation by osmotic shock (202) and then monitor the decay of RNAPII at the PPP. This approach yielded t_1/2 estimates for paused ECs of <2 minutes even when elongation was inhibited by DRB, so that presumably the only possible fate is termination (198).

Paused RNAPII dwell time can also be measured in live cells by imaging fluorescently tagged RNAPII. At the uninduced Drosophila Hsp70 gene, paused RNAPII is relatively stable (t_1/2 ~5 min) and was further stabilized when elongation was blocked by inhibiting P-TEFb (203). On the other hand, FRAP analysis of endogenous GFP tagged RNAPII averaged over many loci in human fibroblasts demonstrated an average residence time of only ~40 seconds at the PPP (162; 204). Dynamic clusters of RNAPII molecules detected by super-resolution microscopy are abundant in live cells with half-lives of a few seconds. Interestingly, these clusters are stabilized by DRB suggesting a possible relation to paused EC’s (205). Another FRAP study of an integrated reporter gene identified a population of paused polymerases that persisted in the presence of DRB with a t_1/2 ~50 seconds (161). Remarkably, at this reporter gene only about 10% of RNAPII encounters with the promoter resulted in initiation and only about 1% produced a full length transcript (161) suggesting that the process is very inefficient.

Paused polymerases are associated with short capped RNAs (scRNAs) (35; 206) and the turnover of these transcripts can be estimated by measuring the extent of 4-thiouracil incorporated after different pulse labelling times using a method called Start-TimeLapse-seq (STL-Seq) (163). This approach found the median t_1/2 of scRNAs is ~ 5 minutes, but it is possible that this differs from the t_1/2 of RNAPII itself. This study also revealed that premature termination removes ~80% of paused polymerases and that termination occurs more rapidly than pause release. Modeling of GRO-seq data following Trp inhibition was used to estimate rates of initiation, pause release, premature termination and elongation from the decay in RNAPII occupancy at the pause site and the gene body (164). This study, which assumed a constant rate of elongation across gene bodies and that pausing does not widely interfere with initiation, concluded that the average turnover time for paused polymerases is less than a minute, in good agreement with FRAP studies (162). At most genes the premature termination rate was estimated to far exceed the pause release rate in agreement with the STL-Seq results (163). Indeed the rate of premature termination must exceed that of pause release if regulation of release is to exert a meaningful effect on RNA output (164). If, on the contrary, pause release were faster than premature termination, then RNA output would be predominantly determined by the rate of initiation rather than pause release. In summary, while there is considerable variation in published estimates of the PPP half life depending on the method and biological system employed, it appears in many cases to be only a few seconds. Such rapid RNAPII turnover at the PPP has the potential advantage that many promoters could be sampled by the pool of RNAPII in a short period to maintain a state of readiness for gene activation. If in contrast a sizeable fraction of the RNAPII pool were stably sequestered at pause sites, then the system is expected to be less responsive (204).

Pause Release

Pause release is the conversion of promoter-proximal ECs with high pausing propensity into elongation-competent complexes with low pausing propensity that translocate rapidly through the gene body. Because it is not possible to follow labelled ECs over time from initiation through productive elongation in vivo as one can in vitro, pause release in vivo must be inferred from indirect evidence. This is usually done by determining the level of RNAPII occupancy at the promoter proximal peak divided by occupancy downstream in the gene body. A higher proximal:distal density ratio is often interpreted as evidence of more pausing and less release. Indeed, the proximal:distal density ratio is often called the “pausing index” but this designation can be misleading because many parameters other than pausing can affect it, including rates of initiation, premature termination, and elongation within the gene body. We advocate that relative RNAPII density ratios be interpreted as a measure of “entry into the gene body” to remain agnostic about mechanism unless additional information is available.

P-TEFb was identified as a DRB-sensitive driver of pause release in vitro (63). Inhibition or degradation of P-TEFb in vivo suppresses productive transcription and depletes RNAPII from gene bodies while increasing accumulation at the PPP. The response to P-TEFb inhibition strongly suggests that it stimulates pause release in vivo as it does in vitro. CDK9 has hundreds of candidate target substrates (66) but the most relevant ones for pause release are NELF, DSIF, and the RNAPII CTD. Phosphorylation of NELF-A and –E triggers eviction of NELF from the paused EC (59; 71), and its replacement on the RNAPII surface by the positive elongation factor PAF1C (75). Spt5 phosphorylation in the KOWx-4/5 linker and CTR1 domains converts DSIF from a negative to a positive effector of elongation (53). Pol II CTD phosphorylation by P-TEFb in vitro occurs principally on Ser5 residues of the heptad repeats (207) but Ser2 phosphorylation can occur if the substrate is primed by Tyr1 phosphorylation (208). CTD Ser5 phosphorylation is associated with reduced pausing (129) and resistance to the negative effects of DSIF and NELF (59). CTD Ser2 phosphorylation in vivo correlates with P-TEFb activity (45), but whether this phosphorylation is direct and/or indirect is not entirely clear. P-TEFb likely stimulates Ser2 phosphorylation indirectly through recruitment of PAF1C (73) whose CDC73 subunit allosterically activates the Ser2 kinases Cdk12/13 (43).

P-TEFb associates with other factors to form catalytically active and inactive complexes. The major active form is a complex with AFF1/4, ELL/ELL2, Eaf1/2, and AF9 that constitutes the super-elongation complex (SEC) (69; 209; 210). The SEC makes direct contact with RNAPII subunits RPB2 and RPB9 via the ELL-EAF subunits. Association of the SEC with RNAPII stimulates elongation in several ways in addition to P-TEFb recruitment: by allosteric activation of the polymerase (211) and by contacting PAF1C (212). Thus, the SEC and PAF1 complexes cooperate with one another soon after initiation to convert the EC from a high pausing to a low pausing state.

Multiple factors are implicated in recruitment of P-TEFb to activate target genes by stimulating pause release, including sequence specific transcription factors like MYC, NF-KB and HIF (36; 213–215), mediator (216; 217), and the co-activators BRD4 (218) and KAP1/TRIM28 (219; 220). HIV-1 Tat is a potent activator of transcription elongation that binds the TAR RNA hairpin loop at the 5’ end of the HIV-1 nascent transcript and recruits the SEC to the EC via contacts with CyclinT1 (221) and AFF4 (222). Indeed, the SEC was initially discovered as a complex with Tat (209; 210). BRD4 and KAP1/TRIM28 contact chromatin in promoter regions (223; 224) and are thought to recruit P-TEFb (218; 219; 225), but unexpectedly, degron depletion of BRD4 or KAP1/TRIM28 caused widespread elongation defects without markedly affecting CDK9 recruitment (226; 227). More work is required to clarify the functional links between BRD4, transactivators and P-TEFb phosphorylation of the paused EC.

The PP1-PNUTS phosphatase is also implicated as a positive regulator of pause release since degron-mediated depletion of PNUTS reduced transcription and RNAPII occupancy across gene bodies, while maintaining the peak of RNAPII at the PPP (123; 124). Potentially relevant substrates include SPT6, PAF1, CDK9, GDOWN1, and AFF4 (123; 124), but how their dephosphorylation might enhance pause release remains to be elucidated.

Premature termination

It is frequently assumed that polymerases which accumulate at the PPP are poised to resume elongation like those that can be chased into long products by addition of NTPs in vitro, however there is an alternate fate for paused RNAPII which is premature termination (228; 229). Recent work has defined two major termination mechanisms that remove a large fraction of polymerases at the PPP: i) endonucleolytic cleavage of the nascent transcript and dephosphorylation of RNAPII by the Integrator complex and ii) ubiquitylation and proteasomal degradation of the polymerase by CRL3^ARMC5. Some premature termination by the restrictor also occurs at positions downstream of the PPP (132). It is estimated that as much as 90% of RNAPII molecules that initiate, terminate prematurely rather than undergo pause release (161; 162; 164; 204).

Integrator

Integrator is a 17-subunit complex first identified as a RNAPII CTD-associated factor that terminates non-adenylated U snRNA transcripts (105). The cleavage module includes the RNA endonuclease INTS11 and its binding partner INTS9 and the phosphatase module includes the catalytic subunit PP2A-C, the scaffolding subunit PP2A-A/PR65, and INTS6 (106; 111). Nascent RNA profiling reveals that Integrator preferentially mediates transcription termination within the first 2 kilobases of almost all human genes (114; 230; 231). This 5′ region, where Integrator-dependent termination is most frequent, roughly coincides with the zone of high-frequency pausing (54).

Integrator specifically binds to paused ECs through contacts between INTS6 and NELF-B (107; 232–234). The KOWx-KOW4 domain of Spt5 directly contacts INTS11, opening its RNA binding channel and activating the endonuclease. Both enzymatic activities of Integrator act on paused RNAPII to facilitate termination. The cleavage module cuts the RNA about 23 bases from its 3′ end, leaving a 5′ phosphate (5′ PO₄) end on the RNA attached to RNAPII (107; 233). This renders the RNA susceptible to 5′–3′ exonucleases including INTS11 itself (232) and the torpedo exonuclease Xrn2 (118; 119). The yeast Xrn2 homologue Rat1 can terminate transcription by pulling on the nascent transcript and/or pushing on the polymerase to disrupt the RNA-DNA hybrid to dislodge the polymerase (235). Integrator-associated PP2A also antagonizes pause release by dephosphorylating the KOWx-4 RNA-binding domain of Spt5 and the RNAPII CTD. Notably, compromising the PP2A phosphatase module by deleting INTS6, a tumor suppressor, confers resistance to P-TEFb inhibitors, and conversely PP2A agonists synergize with CDK9 inhibition to promote cell death, highlighting the antagonism between these two regulators of Spt5 phosphorylation (72). Dephosphorylation of the CTD and Spt5 may promote termination by prolonging the promoter proximal pause and/or slowing the speed of elongation. The same two substrates are dephosphorylated downstream of polyA sites (58; 120; 236), suggesting that dephosphorylation of the EC facilitates termination at both 5’ and 3’ ends of genes.

Ubiquitylation/degradation

An alternative “extreme” mechanism of premature termination involves not just eviction from the template but also degradation of the polymerase by a ubiquitin-dependent (Ub/Deg) mechanism. Ub/Deg was originally shown to remove RNAPII stalled at sites of DNA damage (237–239) and more recently found to prematurely terminate paused ECs that fail a quality control checkpoint (Figure 2c). This premature termination pathway targets RNAPII at the PPP (103; 104; 240; 241) and involves the ATP dependent segregase Valosin-Containing Protein (VCP/p97/CDC48), which disassembles multisubunit complexes, and the CRL3-RING ubiquitin ligase complex (103; 104). CRL3 binds the substrate adaptor ARMC5, which specifically recognizes the RPB1 subunit of RNAPII (160) in a manner that is facilitated by CTD Ser5 phosphorylation (103; 157; 158). ARMC5 loss in patients with adrenal hyperplasia is associated with reduced K48 linked ubiquitylation of RPB1 and elevated cellular pools of RNAPII (160).

Disrupting assembly of a competent EC by depleting Spt5, capping enzymes or INTS6 causes CRL3^ARMC5 dependent ubiquitylation and degradation of RPB1 (56; 103; 157). Premature termination by the integrator and CRL3^ARMC5 are partially redundant. Depleting components of either mechanism alone has little effect on cells, but when both are compromised, cells are inviable and “incompetent” RNAPII complexes with high pausing propensity “leak” into gene bodies (103; 104).

Whereas control of pause release via regulated P-TEFb recruitment and activation is relatively well-established, it is much less clear whether premature termination is also modulated to control transcriptional output. Some studies suggest that premature termination occurs constitutively regardless of the state of gene activation and therefore does not serve a regulatory function (163; 203). However, other estimates of premature termination rates during macrophage differentiation (242) based on RNAPII turnover following Trp treatment showed decreased termination at a subset of activated genes and increased termination at a subset of repressed genes, suggesting that termination is regulated to control mRNA synthesis. It has also been suggested that premature termination is specifically down-regulated in response to oxidative stress resulting in rapid accumulation of RNAPII at pause sites which could serve to stabilize paused ECs that ‘bookmark” regions of open chromatin for rapid recovery of transcription post-stress (184).

Perspective

Promoter-proximal pausing is a hallmark of metazoan transcription, marked by a pile-up of RNAPII ~50 bases downstream of the start site with much lower occupancy across the gene body. This density profile reflects a dynamic balance between pause release into productive elongation and premature termination. Gene activation increases RNAPII density within gene bodies relative to the 5′ peak, and this phenomenon inspired the Pause Release model which proposes that the flux of polymerases that exit the PPP and enter productive elongation is controlled by a “valve”, operated by P-TEFb, governing the frequency that paused polymerases resume elongation. However, recent work suggests that the PPP is also integral to multiple premature termination mechanisms that turn over a large fraction of early ECs. Pausing and premature termination are tightly linked: prolonged pausing increases the chance of termination, whereas release precludes it. Thus, RNAPII turnover at promoters reflects a competition between attenuation and elongation. More work is required to resolve whether active turnover of paused RNAPII serves exclusively as a quality control mechanism that weeds out defective elongation complexes, or if it also serves to regulate transcriptional output, but these two scenarios are not mutually exclusive. We propose that a hybrid Pause Release–Attenuation model (Figure 1a) is a useful framework to describe RNAPII dynamics at the PPP.