Evolution of promoter-proximal pausing enabled a new layer of transcription control

Promoter-proximal pausing of RNA polymerase II (Pol II) is a key regulatory step during transcription. Despite the central role of pausing in gene regulation, we do not understand the evolutionary processes that led to the emergence of Pol II pausing or its transition to a rate-limiting step actively controlled by transcription factors. Here we analyzed transcription in species across the tree of life. We found that unicellular eukaryotes display a slow acceleration of Pol II near transcription start sites. This proto-paused-like state transitioned to a longer, focused pause in derived metazoans which coincided with the evolution of new subunits in the NELF and 7SK complexes. Depletion of NELF reverts the mammalian focal pause to a proto-pause-like state and compromises transcriptional activation for a set of heat shock genes. Collectively, this work details the evolutionary history of Pol II pausing and sheds light on how new transcriptional regulatory mechanisms evolve.


Introduction
The evolution of complex transcriptional regulatory programs is one of the defining characteristics of metazoans which enables the organismal complexity required for animal development. "Pausing" is one of the regulatory stages during transcription by RNA Polymerase II (Pol II). Pol II transiently "pauses" 20-60 bases downstream of the transcription start site (TSS) at all genes in Drosophila and mammals, disrupting the continuous flow of transcription (Fig. 1A) [1][2][3][4] . The rate at which polymerases are "released" from a paused state into productive elongation is actively regulated by transcription factors 5 , and is therefore essential for proper development in most animal species [6][7][8] . However, unicellular model organisms, including yeast 9,10 , do not have a promoterproximal pause. To date, no study has characterized the distribution of Pol II outside of a few key model organisms, leaving when and how the pause evolved as an open question.

NELF subunit evolution increased the residence time of Pol II in a pause state
We used Precision Run-On and Sequencing (PRO-seq) 11 to study transcription in 20 extant organisms that represent two billion years of evolution (Fig. 1B), including multiple species near the base of the Metazoa phylogeny and the transition between plants and animals. Our atlas of organisms adds new data representing two prokaryotic organisms (Escherichia coli and Haloferax mediterranei, representing the bacteria and archaea domains, respectively), and single-celled eukaryotes including the social amoeba (Dictyostelium discoideum), two ichthyosporeans (Creolimax fragrantissima, and Sphaeroforma arctica), and a filasterean (Capsaspora owczarzaki). We also included a number of metazoan organisms representing major taxa, including the cnidarian (Nematostella vectensis), the sea urchin (Strongylocentrotus purpuratus), the water flea (Daphnia pulex), the butterfly (Dryas iulia), and the cyclostome (Petromyzon marinus). Finally, we have augmented our PRO-seq atlas by integrating published data from a fly (Drosophyla melanogaster 11 ), a nematode (Caenorhabditis elegans 12 ), yeast (Saccharomyces cerevisiae and Schizosaccharomyces pombe 9,10 ), model plants (Arabidopsis thaliana, Oryza sativa, and Zea mays [13][14][15], and mammals (Homo sapiens and Mus musculus 7,16 ). These species occupy key positions along the phylogenetic tree, allowing us to investigate most major transitions in the animal lineage.
As expected, most metazoan organisms exhibited a pileup of RNA polymerase 30-100 base pairs downstream of the TSS indicative of Pol II promoter-proximal pausing (Fig. 1C). Conversely, prokaryotic organisms lacked a prominent Pol II peak in our data. Plants and unicellular eukaryotes exhibited more diverse pause variation: the plant Z. mays showed a focused peak, while the plant O. sativa and the yeast S. pombe displayed a more dispersed peak downstream of the TSS. The plant A. thaliana and yeast S. cerevisiae showed no evidence of pausing. To reveal more subtle differences in the dynamics and gene-by-gene variation at the pause site than observed in meta profiles, we computed pausing indexes which quantify the duration Pol II spends in a promoterproximal paused state 3,17 . Pausing indexes revealed that the residence time of Pol II at the pause site is, on average, 1-2 orders of magnitude higher in metazoans than unicellular eukaryotes or plants ( fig. S1). Consistent with the meta plots, we also noted wide variation in pausing indices in unicellular eukaryotes and plants. Taken together, our results suggest that an ancestral slowdown in Pol II transcription near the TSS may have arisen in unicellular eukaryotes and became longer in duration and more focused during the early evolution of metazoans.
Pausing is mediated by interactions between Pol II, DRB sensitivity inducing factor (DSIF), and the negative elongation factor (NELF) complex 18,19 . Of these proteins, the NELF complex can establish pausing both in vitro and in vivo 20 . The NELF complex consists of multiple subunits, including NELF-A, -B, -C (or its isoform -D), and -E. CryoEM studies revealed that NELF-B and -E form a sub-complex, while NELF-C/-D and -A form a separate subcomplex. NELF-B and -C/-D interact with one another forming a core structure that holds the entire NELF complex together (Fig. 1D) [21][22][23] .
We hypothesized that the evolution of NELF proteins was associated with the gain of pausing in eukaryotes. To test this hypothesis, we used BLASTp to identify potential orthologs of the human NELF subunits among a group of 30 organisms representative of key eukaryotic taxa ( Fig. 1E; fig. S2; fig. S3). We found that NELF-B and -C/-D are widely distributed in eukaryotes, suggesting that the core NELF subunits were present in a shared common ancestor of all eukaryotes. Both subunits were secondarily lost in yeast, S. pombe and S. cerevisiae, as well as in the nematode C. elegans and in land plants. NELF-A is present in some unicellular eukaryotes; a strong match to the metazoan protein first appeared in a common ancestor of Ichthyosporea and metazoans. We only found strong evidence for NELF-E in metazoans, and the most parsimonious model is that NELF-E evolved early in Metazoa or just before the transition to multicellularity. Collectively, these findings demonstrate a strong association between the evolution of NELF proteins and paused polymerase near TSSs.
To test whether other factors besides NELF were associated with the evolution of paused Pol II, we examined the evolutionary conservation of other proteins linked to pausing. Most other proteins implicated in the early steps of transcription elongation, including PAF1, DSIF (SPT4, and SPT5), the positive transcription elongation factor (P-TEFb; CDK9, cyclins [human Cyclin-T1, Cyclin-T2]), and 7SK (MEPCE, LARP7) are deeply conserved among eukaryotes ( Fig. S2; Fig. S3), with some structural conservation extending back to archaea 24 . Thus, proteins responsible for the release from pause in metazoans (especially P-TEFb and PAF1) were part of the ancestral eukaryotic transcription complex, and evolved before the high pausing indices found in metazoans. The sole exceptions were the HEXIM proteins (HEXIM1 and HEXIM2), which are part of the 7SK complex that works in preventing P-TEFb-mediated pause release 25 . The most parsimonious model is that HEXIM proteins evolved in a common ancestor of Metazoa, perhaps coincident with NELF-E, as an additional checkpoint to increase the residence time of paused Pol II or to regulate pausing in metazoans.
Our finding that NELF and HEXIM protein evolution were uniquely associated with polymerase pausing led us to suspect that gains of specific NELF and HEXIM protein subunits may have resulted in incremental alteration of the residence time of paused Pol II along the animal stem lineage. To determine how the addition of multiple NELF subunits affected the strength of pausing, we compared pausing indexes between species with a different complement of NELF or HEXIM subunits. Species containing NELF-B and -C/-D (which we refer to as the "core" NELF complex) have higher pausing indexes than species without any NELF subunits ( Fig. 1F; Fig S4). The addition of NELF-A increased pausing indexes to the same order of magnitude observed in metazoan model organisms (files and mammals). Thus, NELF-B and -C/-D are sufficient for pausing, but the addition of NELF-A, NELF-E, and HEXIM proteins correlates with higher pausing indexes and suggests the derived proteins may act together with the core NELF complex to fine-tune the function of paused RNA polymerase (Fig. S4D).

Organisms without NELF show different types of pausing behavior
In some unicellular organisms, which do not have all four of the NELF subunits found in metazoans, we observed that Pol II moved slowly through the first 30-100 bp after the TSS. We hypothesized that this "proto-pause" may serve as an ancestral substrate pre-dating the highly focused, long-duration Pol II pausing observed in extant metazoans. In some cases, we observed examples of extreme phenotypes in which Pol II moved slowly despite having a complete absence of the NELF core complex. For instance, Z. mays, S. pombe, and O. sativa all display an accumulation of Pol II near the TSS, despite having lost both NELF-B and NELF-C/D. This extreme example of a protopause appears in a similar location as a canonical pause (or just downstream), but does not have the same magnitude of pausing index (Fig. 1C). To explore why some extant species have a proto-pause, despite not containing any of the NELF subunits, we examined the DNA sequence under the quartile of genes with the strongest positioned proto-pause in each organism ( Fig. 2A). Consistent with previous work 11, [26][27][28][29][30] , metazoan organisms show a well-defined pause motif, which is also present in three organisms that show a proto-pause, including Z. mays, S. pombe, and O. sativa (Fig. 2B). These observations may suggest that a pause DNA sequence motif contributes to a transient slowdown of Pol II at this position in organisms that have lost the core NELF subunits, NELF-B and NELF-C/D.
To determine whether the pause motif was associated with pausing index variation across all 20 species, we examined the enrichment of the human pause motif near the pause position (Fig. 2C). Despite the pause motif we used being derived from humans 28 , we nevertheless found that it explained variation in the pausing index across all organisms surprisingly well (R 2 = 0.306, p = 0.011; Fig. 2D; fig. S5A-F). Conversely, the DNA sequence motif of the TATA box and Initiator were not correlated with pausing index ( fig. S5G-H). Altogether, our data support the idea that a pause sequence motif, featuring a C (or possibly G) in the Pol II active site at the pause position, serves as an ancestral step limiting the rate of transcription after initiation and can be linked to the formation of a proto-pause. This pause-associated DNA sequence alongside other chromatin factors, such as the position of the +1 nucleosome and the rate at which the P-TEFb subunit CDK9 phosphorylates the early elongating Pol II complex, may then be sufficient to explain much of proto-pause formation in species such as S. pombe 9 .

Loss of core NELF-B impacts chromatin localization of NELF-E and alters Pol II pausing
Our analysis of NELF evolution shows that the core NELF subunits, NELF-B and -C/D, evolved earlier than the ancillary subunits, NELF-A and -E. To test the functional impact of core and ancillary subunits in mammalian cells, we generated FKBP12homozygously tagged mouse embryonic stem cell (mESC) lines that rapidly degrade either NELF-B (13) or NELF-E after treatment with the small molecule dTAG-13 ( Fig. 3A-B 7,31 ). The NELF-B dTAG was reported and validated in a recent paper 7 , while the NELF-E dTAG cell line is novel here. We verified that the FKBP12-tagged NELF-E protein was properly localized and that NELF-E was nearly undetectable within 30 min after the addition of 500 nM dTAG ( fig. S6; fig. S7). We also verified that the rapid depletion of both NELF subunits decreased Pol II levels at the pause site following 30-60 min of dTAG-13 treatment, as measured by PRO-seq ( Fig. 3E-F; fig. S8A). We hypothesized that loss of NELF-B would have a greater impact on NELF complex assembly on chromatin than loss of NELF-E due to the central role of NELF-B in the complex (Fig. 3A 21 ). Consistent with our hypothesis, we observed that loss of NELF-B led to a decrease of the entire NELF-B/E sub-complex from chromatin, while a loss of NELF-E resulted in only a moderate reduction of 40% in NELF-B protein levels ( Fig. 3B-C left panel; fig. S7 fig.  S6A-B). These findings confirm that the functions of NELF-B and -E in mESCs mirror the structure and evolutionary history of these NELF subunits.

Pol II recovery after prolonged NELF-B degradation mirrors a proto-paused-like state
Our PRO-seq data in the NELF-B cell line showed that many genes partially recovered Pol II at the pause site following 60 min of treatment ( fig. S8B-C). To investigate the observed Pol II signal recovery, we first clustered genes based on their changes in Pol II loading between 30 and 60 min of dTAG treatment (Fig. 3E, clusters 1, 2, and 3). Cluster 1 showed a localized recovery of PRO-seq signal near the position of the canonical pause. Cluster 2 showed no indication of recovery and, relative to the other clusters, it was enriched in transcribed enhancer sequences ( fig. S9A). And, cluster 3 exhibited a recovery of Pol II further into the gene body in a similar position as the slowdown of Pol II observed in S. pombe and O. sativa, potentially near the location of the +1 nucleosome, as reported by a previous study 32 .
We hypothesized that after the depletion of NELF, DNA sequences associated with the proto-pause in organisms without NELF-B may be sufficient to re-establish some paused Pol II. We looked for enrichment of the DNA proto-pause motif at loci that exhibited recovery of the paused state after NELF depletion. We found both higher enrichment of the pause motif and better positioning of the +1 nucleosome in clusters 1 and 3 when compared to cluster 2 ( Fig. 3F-G; fig. 9B). Interestingly, the main difference between clusters 1 and 3 was that genes in cluster 3 had higher initiation rates, as determined by both TT-seq 33 and a computational modeling approach analyzing steadystate PRO-seq data 17 (Fig. 3H-I; fig. S9C). Genes in cluster 3 also had much higher binding of some components of the pre-initiation complex and more clearly defined DNA sequence motifs that specify transcription initiation 34 ( fig. S9D-E), potentially consistent with higher initiation rates. Based on these results, we propose that clusters 1 and 3 partially recover Pol II near the pause due to a combination of DNA sequence and interactions with well-positioned nucleosomes. We also speculate that genes in cluster 3 recover in a more downstream position as a result of a higher rate of initiation. Greater initiation rates at these genes may lead to an accumulation of Pol II at the start of the gene that causes polymerases to be pushed downstream due to interactions between newly incoming Pol II. In sum, we found that after NELF depletion, Pol II signal resembles the pattern found in proto-paused organisms that have lost the core NELF subunits. Furthermore, this proto-paused-like state is associated with the same DNA sequence features and the presence of strongly positioned nucleosomes.

Pol II pausing allows transcription factors to regulate pause release
We speculate that the evolution of a focal pause was required for the evolution of a system that could control gene expression by releasing paused Pol II. In metazoans, sequence-specific transcription factors can modulate pause release and thereby tune the level of gene expression 5,6,[35][36][37][38] . The factors that are responsible for pause release (e.g., p-TEFb) are conserved in all eukaryotes ( fig. S2; S3), pointing to the critical role of release (or an analogous step in early elongation) in eukaryotic organisms 10 . After the depletion of NELF-B, we observed paused Pol II "creeping" across the first couple of kilobases of the gene body ( fig. S10), similar to observations made in S. pombe, which has no NELF 7,10 . As a result, Pol II which needs to be released from pause by p-TEFb is no longer in a fixed location, in proximity to promoter-bound transcription factors.
We hypothesized that the downstream redistribution of Pol II after NELF-B depletion would prevent the targeted regulation of gene expression by transcription factors acting to release paused Pol II into productive elongation. To test this hypothesis, we turned to the well-studied heat shock system, where the transcription factor heat shock factor 1 (HSF1) activates transcription of a core group of a few hundred genes following heat stress by the release of paused Pol II 36,39,40 . We asked whether HSF1 could release paused genes as efficiently following the depletion of NELF-B and -E in mESCs (Fig. 4A). We first identified genes that were up-and down-regulated using a regular heat shock experiment in mESCs. Our analysis confirmed the induction of a core group of heat shockresponsive genes 36,39,40 , despite some differences in basal gene expression between NELF-B and NELF-E cell lines ( fig. S11; fig. S12A-D). Although many classical upregulated genes were properly up-regulated following the depletion of NELF-B and NELF-E, heat shock (HS)-dependent genes on average had a lower induced fold-change following NELF-B depletion ( fig. S12; fig. S13E). Thus, Pol II redistribution after NELF-B depletion does prevent HSF1 from acting efficiently as a transcriptional activator.
To rule out the possibility that our observed differences in HSF1-dependent gene activation were driven by changes in gene expression following dTAG-13 7 , including the accumulation of Pol II trickling into the gene body ( fig. S10), we focused our analysis on the gene body downstream from NELF-induced Pol II trickling regions. We also excluded genes with altered gene body density following dTAG-13 treatment in either cell line (fig. S13F; see Methods). For the remaining genes, we noted a clear defect in the HSinduction of up-regulated genes, but not in HS-repression at down-regulated genes, consistent with a model in which HSF1 failed to adequately release Pol II after NELF depletion (Fig. 4B). The up-regulation defect was more prominent following the depletion of NELF-B than NELF-E (unpaired Mann-Whitney, p-value = 2.8e-4) (Fig. 4B), potentially consistent with a more direct role for NELF-B in the formation of a focal pause.
Interestingly, many of the most highly HS-induced genes did not show a large defect in up-regulation as seen here (e.g. Hspa1b, Hsp1h1; fig. S12) and in a previous study 32 . The high rate of firing at these genes may be associated with a high concentration of p-TEFb, resulting in a higher probability of releasing Pol II in the right location before it trickles away from the promoter. In contrast, the more moderately induced and highly paused HS genes are firing less frequently and the trickling of paused Pol II to more downstream locations may prevent their proper activation ( fig. S13). Altogether, these findings support our model in which the evolution of pausing facilitated the ability for transcription factors to act on pause-release, providing an additional step to more tightly control gene expression.

Discussion
Our work offers mechanistic insights into how new regulatory complexity evolved by enabling targeted regulation of a preexisting step in the transcription cycle through the evolution of a focal promoter-proximal pause. We propose that the recruitment of P-TEFb and PAF1, which cause pause-release in metazoans, actually serve a more general role that is necessary at all genes in all eukaryotic organisms, regardless of whether the organism has a long-lived focal pause. The evolution of a "focal" pause collapsed the substrate for this step in transcription to a single location at each gene and increased the pause residence time. The degree to which Pol II slows down at the pause position, which progressively increased in the branches leading to metazoans, appears to be affected by the evolution of NELF-E, NELF-A, and HEXIM proteins. Together these evolutionary innovations collapsed a rate-limiting step in all eukaryotes into a single position in metazoans. A centralized location for paused Pol II allows transcription factors to catalyze the release of Pol II into productive elongation, by providing a focused and promoterproximal target adjacent to transcription factor binding sites, as shown here in the case of HSF1. This innovation provided a new rate-limiting step in transcription that could be targeted for gene-specific regulation. The evolution of additional regulatory complexity may have helped to enable the evolution of complex, multicellular metazoan organisms.      A phylogenetic tree cluster of all species analyzed in this study is depicted on the left. On the right, percentage identities from running BLASTp using the indicated human protein are reported.The criteria for being marked "present" is that the human protein sequence needs to identify an ortholog in the indicated species, and that the ortholog in the indicated species needs to reciprocally identify the indicated subunit in human using BLASTp. We required that both the initial and reciprocal BLAST searches identified the indicated protein with an E-value less than 1e-06. NA indicates that no value was output by BLASTp. We note that more sensitive approaches have identified conservation of Spt4 and Spt5 extending back to archaea 24 , but this conservation was not evident in the protein sequence similarity analyzed here. We also note that several proteins have undergone recent duplication/ divergence events which are not reflected in the figure (CyclinT1 and Cyclin T2; HEXIM1 and HEXIM2).

Figure S3: E-values for NELF and other transcription-associated proteins.
A phylogenetic tree cluster of all species analyzed in this study is depicted on the left. The table shows E-values output by BLASTp for the indicated human transcription-associated protein. The criteria for being marked "present" is that the human protein sequence needs to identify an ortholog in the indicated species, and that ortholog needs to reciprocally identify the indicated subunit in human using BLASTp. We required that both the initial and reciprocal BLAST searches identified the indicated protein with an E-value less than 1e-06. The E-value shown in the table reflects the first BLAST search (i.e., human protein to the indicated species). NA indicates that no value was output by BLASTp. We note that more sensitive approaches have identified conservation of Spt4 and Spt5 extending back to archaea 24 , but this conservation was not evident in the protein sequence similarity analyzed here. We also note that several proteins have undergone recent duplication/ divergence events which are not reflected in the figure (CyclinT1 and Cyclin T2; HEXIM1 and HEXIM2).

Figure S10: Pol II trickles into gene bodies effect after NELF-B degradation.
Heatmaps of log2 fold changes in PRO-seq signal in NELF-B tagged cell lines at all TSSs, Clusters 1, 2, and 3 (in this order from top to bottom rows). The heatmaps depict log2 fold change relative for the following comparisons (from left to right columns): untreated PRO-seq signal, log2 fold change for 30min/0min, 60min/0min, and 60min/30min of dTAG-13 treatment.

Figure S14: Evaluation of PRO-seq library quality.
Profiles show the number of PRO-seq reads per species is reported as a function of insert size. A color gradient from orange (depicting highly degraded RNA) to white (depicting lowly degraded RNA) marks the quality of each sample. A degradation ratio score is also reported at the top of each plot. Degradation ratios for C.elegans and A.thaliana were computed manually using the scripts in 43 .

Acknowledgments
We thank members of the Danko and Lis labs for valuable discussions and suggestions throughout the life of this project, and Meritxell Antó Subirats for preparing samples from C. owczarzaki, C. fragrantissima and S. arctica. Work in this publication was primarily supported by a grant from the NASA exobiology program (17-EXO-17-2-0112). Additional funding was also available from NHGRI (R01-HG010346 and R01-HG009309) to CGD, from the NIGMS (R01 GM147731) to ILB and CGD, and from NIH (RM1-GM139738) to JTL. AA was supported by the NIH (T32GM007739, F30HD103398). MML was supported by an Ayuda Juan de la Cierva

Competing Interests Statement
The authors declare no competing interests.

Data Availability
Tables in CSV format can be downloaded from: https://github.com/alexachivu/PauseEvolution_prj Data generated in this study can be found in Gene Expression Omnibus at: GSE223913.

Code Availability
Custom code for analyzing sequencing data can be found on GitHub under: https://github.com/alexachivu/PauseEvolution_prj/

Experimental methods
Sample collection: E. coli: An overnight culture of E. coli MG1655 was subcultured in 50 mL LB and grown at 37°C to OD 600 = 0.95. 5 mL aliquots were pelleted by centrifugation at 3000 × g. Pellets were permeabilized, washed, and flash-frozen as described in 44 .
H. mediterranei: ATCC 33500 was grown for 48 hours at 35 °C in ATCC Medium 1176. 12.5 mL culture was centrifuged, and the cell pellet was resuspended in 3 mL cold non-yeast permeabilization buffer. To increase permeabilization of archaeal cells, the cell suspension was split into 3 × 1 mL aliquots in screw-cap tubes and combined with 400 µL sterile 0.5 mm glass beads. Cells were subject to bead-beating for 3 cycles of 2 minutes vortexing, 2 minutes on ice. Supernatants were transferred to 1.5 mL tubes, centrifuged to collect cell contents, and washed twice by resuspension in 500 µL storage buffer. Cells were resuspended in a final volume of 50 µL storage buffer and snap-frozen. The permeabilization and storage buffers were the same as reported previously 44 , and include: ATCC Medium 1176 recipe (1 L), 156 g NaCl, 13 g MgCl 2 × 6H 2 O, 20 g MgSO 4 × 7H 2 O, 1 g CaCl 2 × 2H 2 O, 4 g KCl, 0.2 g NaHCO 3, 0.5 g NaBr, 5 g yeast extract, 1 g glucose. After mixing components, the pH was adjusted to 7.0 and the buffer was autoclaved. D. iulia: Wing tissues were sampled from Day 3 pupae derived from Costa Rican stock following standard protocols (e.g. 46 and 16 ). Wing tissues were dissected from pupae in cold PBS, after which nuclei were extracted in cold PBS using a dounce homogenizer. Nuclei were spun down and resuspended in nuclei storage buffer before flash freezing.
Creolimax fragrantissima and Sphaeroforma arctica were cultured axenically in BD Difco™ Marine Broth 2216, in tissue culture-treated flat-bottomed polyethylene tissue culture flasks at 12°C; confluent cells were harvested by centrifugation, and the pellet flash-frozen and stored at -80°C.
Dictyostelium discoideum AX3 wildtype cells were cultured axenically in HL-5 (Formedium) on untreated polystyrene petri dishes at 22°C. Confluent cells were resuspended in fresh media and centrifuged at 300xg for 5 min. The pellet was flash frozen and stored at -80°C.
Nematostella vectensis: Adult Nematostella were reared in 1/3 strength artificial seawater at 18°C in dark conditions. Spawning was induced using the protocol described in 47 . Adult males and females were induced to spawn in small glass bowls, and fertilized egg masses were removed and cultured in small glass bowls at 25°C. Swimming gastrula/early polyp stage animals were harvested for nuclei isolation.
Generation of NELFB and NELFE mESCs: both cell lines were generated using an identical approach to endogenously and homozygously tag the C-terminus of each protein with FKBPF36V tag. The NELFB line has been previously described, and the NELFE line was generated for this study. The methods below describe the NELFE line generation, for more details of NELB line, please refer to 7 .
dTAG drug treatment: The dTAG-13 reagent (Bio-Techne: https://www.bio-techne.com/p/small-molecules-peptides/dtag-13_6605) was reconstituted in DMSO (Sigma) to a final concentration of 5 mM. The dTAG-13 solution was diluted in culture medium to 500 nM and added to cells for the indicated time period during medium changes.
Immunofluorescence: Cells plated on u-Slide eight-well plates (Ibidi) were washed with PBS+/+ and fixed in 4% PFA (Electron Microscopy Sciences) in PBS+/+ for 10 min at room temperature. Cells were subsequently washed twice with PBS+/+, followed by wash buffer and 0.1% Triton X-100 (Sigma) in PBS+/+, and permeabilized in 0.5% Triton X-100 (Sigma) in PBS+/+ for 10 min. Then blocked with 3% donkey serum (Sigma) and 1% BSA (Sigma) for 1 h at room temperature. Cells were incubated with primary antibodies in blocking buffer overnight at 4°C (antibodies and concentrations are listed in Supplemental Table S1). Then, they were washed three times in wash buffer and incubated with suitable donkey Alexa Fluor (1:500; Invitrogen) for 1 h at room temperature. Finally, cells were washed three times with wash buffer, with the final wash containing 5 μg/mL Hoechst 33342 (Invitrogen), and imaged.
Imaging: Fixed immunostained samples were imaged using a Zeiss LSM880 laser scanning confocal microscope. An air plan-apochromat 20×/NA 0.75 objective was used. Images represent a 2D plane correlating to the monolayer of cells in culture. No further image processing was performed.
The harvested cells were incubated on ice for 5 min, scraped and collected then sonicated for 15 sec to complete lysis and then spun down at 12,000g for 10 min at 4°C. The supernatant was collected, and protein concentration was measured using Pierce BCA protein assay kit (Thermo). Samples were prepared by mixing 10 to 20 µg of protein with Blue loading buffer (Cell Signaling) and 40 mM DTT (Cell Signaling), followed by boiling for 5 min at 95°C for denaturation. Cellular compartment fractions were prepared using subcellular protein fractionation kit (Thermo) following the manufacturer's instructions. The samples were run on a Bio-Rad Protean system and transferred to a nitrocellulose membrane (Cell Signaling) using transblot semidry transfer cells (Bio-Rad) following the manufacturer's instructions and reagents. The nitrocellulose membrane was briefly washed with ddH2O, stained with Ponceau S (Sigma) for 1 min, and washed three times with TBST (0.1% Tween 20 [Fisher] in TBS) to check for transfer quality and serve as a loading control. Then it got blocked with 4% BSA in TBST for 1 h at room temperature and incubated with primary antibodies diluted in blocking buffer overnight at 4°C. The membrane was then washed three times with TBST, incubated with secondary antibodies in blocking buffer for 1 h, and washed three times with TBST. Last, the nitrocellulose membrane was incubated with ECL reagent SignalFire for 1-2 min and imaged using a ChemiDoc (Bio-Rad).
The following antibodies were used in this paper:

Antibody
Source Identifier Application Conc. Heat shock experiments on mESCs: Heat shock was administered as described in recent work from the Lis lab 36,40 . We started the heat stress after 30 min of dTAG-13 treatment, which corresponds to the maximal depletion of paused Pol II based on PRO-seq data.
We performed the analysis of the HS data in two different ways: -On a first analysis, by calling gene expression changes using DEseq (log2foldChange > or < 0, and padj < 0.05) between a regular HS and NHS experiment. Then, we plotted log2 fold changes of (HS+dTAG)/(NH) and (HS+dTAG)/(NHS+dTAG) at these pre-defined HS up-regulated or down-regulated genes coordinates. -For a second approach, we considered the effect that NELF-B depletion has on Pol II trickling into gene bodies. To eliminate any biases from increased PRO-seq signal downstream from the TSS due to the dTAG-13 treatment alone, we re-analyzed our data after removing the first 3 kb downstream from the TSS and we focused on genes that remain unchanged following dTAG-13 treatment, but are either up or down-regulated after HS ( fig. S13F). We confirmed a slight up-regulation of genes in the vicinity of the TSS after dTAG-13 treatment by plotting the correlation between the log2 fold change of dTAG-13 treatment after NELF-B degradation ( fig. S10). We used deeptools to compute log2 fold change bigwigs and plot heat maps.
Both of these analyses confirmed a defect in up-regulation across many HS-dependent genes. In the second analysis, we observed that a significant number of genes that were meant to show HS-dependent upregulation failed to reach their full transcription potential in the absence of NELF-B (Fig. 4C left; fig. S13A). The same effect was also observed, though in far fewer genes in the absence of NELF-E (Fig. 4C right; fig. S13B). We noted no defect in down-regulated genes using this analysis approach.
PRO-seq library prep: PRO-seq or ChRO-seq 11,48 libraries were prepared from snap-frozen cell pellets following the protocol described in 44 . All PRO-seq libraries were evaluated for data quality and sequencing depth using PEPPRO 49 . Data and data quality are shown in ( fig. S14; Table 1).

Computational analyses
*In this paper we refer to PRO-seq, GRO-seq, and ChRO-seq as PRO-seq.
Mapping and processing PRO-seq data: Single and paired-end reads of PRO-seq data were aligned to its reference genome using the proseq2.0 pipeline from the Danko lab (https://github.com/Danko-Lab/proseq2.0) using the following parameters: -RNA5=R1_5prime --RNA3=R2_5prime --ADAPT1=GATCGTCGGACTGTAGAACTCTGAACG --ADAPT2=AGATCGGAAGAGCACACGTCTGAACTC --UMI1=4 --UMI2=4 --ADD_B1=6 --ADD_B2=0 --thread=8 --map5=FALSE. Library processing included adapter trimming using cutadapt, PCR deduplication (where UMIs are present) using printseq-lite.pl, followed by mapping to the reference genome using BWA. Mapped BAM files were then trimmed either to the 3'-end of the RNA (to map the location of RNA Pol II) or the 5'-end (to map the beginning of the RNA) and the 1bp position was converted to bedGraphs and BigWigs. PRO-seq libraries were also RPM normalized to account for differences in sequencing depth.

Reannotation of transcription start sites:
We took PRO-seq mapped BAM files and ran it through the RunOnBamToBigWig tool developed in the Danko lab (https://github.com/Danko-Lab/RunOnBamToBigWig) to compute 5'prime mapped BigWigs (parameter for paired end data: --RNA5=R1_5prime; parameter for single end data: --SE_READ=RNA_5prime). Then, we used published gene annotations in each species, resized them to a 1kb window centered on the gene annotation start site, and computed the total number of 5'-prime mapped PRO-seq reads that fall within this interval using 10bp sliding windows. Last, to reannotate gene start sites, we took the start position of the 10bp window with the maximum number of 5'-prime PRO-seq reads. We used these annotated TSSs for all further analyses.

Computing Pausing indexes:
Pausing indexes were computed as the ratio between Pol II density in the pause region and gene body region. We defined the pause region as the interval between [ TSS-150, TSS+150 ]bp and the gene bodies as [ TSS+300, TES-300 ]bp (where TES = transcription end site). Genes shorter than 300bp were removed from the analysis.

Heatmaps and meta profiles:
We use DeepTools to functions (bigwigCompare, compute matrix, plotHeatmap, and plotProfile) to compute heatmaps and meta profiles of PRO-seq data. We also used DeepTools to cluster and compute correlations between the heat shock PRO-seq libraries as BAM files (bamCompare, and multiBamSummary, plotPCA, and plotCorrelation) 50 .
Before running bigwigCompare, we generated combined BigWigs of the plus and minus PROseq data for each sample.

Reciprocal BLASTp to compare transcription proteins across species:
To determine if human orthologs of key transcription machinery proteins (NELF complex; DSIF; PAF1; 7SK proteins) were present in other species, we performed a reciprocal BLAST using the rBLAST library (version 0.99.2). First, given a human protein, H, a BLASTp was performed against the protein database of a different organism (X). Sequences in species X that produced a BLASTp E-value lower than 1e-6, were considered candidates for a second BLASTp run. On the second BLASTp run, we performed a reciprocal BLAST search using the candidates in species X relative to the human proteome. If this reciprocal search yielded any valid hits passing the scoring same threshold of E-value < 1e-6, the protein was considered present in species X. This approach was repeated for all protein sequences in fig. S2 and S3. For this analysis, we downloaded human protein sequences from UniProt, along with complete proteomes (.pep.all.fa files) of all species analyzed in this study from NCBI or Ensembl (only for C. fragrantissima and D. iulia). Then, we used the following script to compare (https://github.com/alexachivu/PauseEvolution_prj/blob/main/Protein%20BLAST%20(BLASTp)) the human homologs of those proteins to the entire proteome of all species in this study. We provide the human protein sequences used in this reciprocal BLAST search here: https://github.com/alexachivu/PauseEvolution_prj/blob/main/Human.orthologs_sequences.fa Defining DNA sequence motif under the pause: The position of the pause site in each species was defined as follows. First, we utilized our reannotated TSS positions and created a 100bp window starting from the TSS: [ TSS, TSS+100 ]bp. Next, we designated the base with the maximum PRO-seq counts within this 100bp window as the pause site. Finally, we created either a 1kb or 20bp window centered around the identified pause site for further motif analyses.

#3.
For the motif enrichment analysis, we used a different script (https://github.com/alexachivu/PauseEvolution_prj/blob/main/MotifEnrichment) to compare the DNA sequences in each species with a human pause motif described in 28 .

Differential analysis:
We performed differential analysis to quantify changes after heat shock, dTAG-13, and the dual treatment of heat shock and dTAG-13. To accomplish this, we run DEseq2. We used the total number of dm3 spike-in reads (divided by the mean of the spike-ins) as scaling factors.

Figure design:
We used BioRender to draw all of the illustrations and cartoons in this paper, with the exception of the schematic representing the relationships between species. The latter was prepared using Interactive Tree of Life (iTOL) v.6.7 51 based on relationships depicted in 52 and edited in InkScape and Illustrator.