Hyperosmotic stress alters the RNA Polymerase II interactome and induces readthrough transcription despite widespread transcriptional repression

Summary

Stress-induced readthrough transcription results in the synthesis of downstream-of-gene (DoG) containing transcripts. The mechanisms underlying DoG formation during cellular stress remain unknown. Nascent transcription profiles during DoG induction in human cell lines using TT-TimeLapse-seq revealed widespread transcriptional repression upon hyperosmotic stress. Yet, DoGs are produced regardless of the transcriptional level of their upstream genes. ChIP-seq confirmed that stress-induced redistribution of RNA Polymerase (Pol) II correlates with the transcriptional output of genes. Stress-induced alterations in the Pol II interactome are observed by mass spectrometry. While certain cleavage and polyadenylation factors remained Pol II-associated, Integrator complex subunits dissociate from Pol II under stress leading to a genome-wide loss of Integrator on DNA. Depleting the catalytic subunit of the Integrator using siRNAs induces hundreds of readthrough transcripts, whose parental genes partially overlap those of stress-induced DoGs. Our results provide insights into the mechanisms underlying DoG production and how Integrator activity influences DoG transcription.

Graphical Abstract

In brief

Rosa-Mercado et al. report that hyperosmotic stress causes widespread transcriptional repression in human cells, yet downstream-of-gene transcripts (DoGs) arise regardless of the transcriptional response of their upstream genes. Interactions between Pol II and Integrator are disrupted by hypertonicity and knockdown of the Integrator nuclease leads to DoG production.

Introduction

Conditions of cellular stress, such as hyperosmotic stress, heat shock, oxidative stress, viral infection and renal cancer lead to readthrough transcription (Vilborg et al. 2015; Vilborg et al. 2017; Rutkowski et al. 2015; Bauer et al. 2018; Grosso et al. 2015; Vilborg and Steitz 2017). This phenomenon is defined as transcription extending past the annotated termination sites of genes. This stress-induced transcription produces thousands of long noncoding RNAs (lncRNAs), known as downstream-of-gene (DoG) containing transcripts. DoGs arise minutes after stress from approximately 10% of human protein-coding genes and are continuous with their upstream mRNAs. By definition, DoGs extend over 5 kilobases (kb) past the normal termination site of their gene of origin. They can be over 200 kb long, but have an average length of 14 kb. DoGs are retained in the nucleus and largely remain localized close to their site of transcription (Vilborg et al. 2015; Vilborg et al. 2017; Hennig et al. 2018).

Upon influenza and HSV-1 infection, the induction of readthrough transcripts is orchestrated in part by viral proteins NS1 and ICP27, respectively. These proteins interfere with transcription termination by interacting with host cleavage and polyadenylation (CPA) factors (Nemeroff et al. 1998; Bauer et al. 2018; Wang et al. 2020). Accordingly, knockdown of the catalytic subunit of the CPA complex, CPSF73, partially induces DoGs (Vilborg et al. 2015). However, host-dependent mechanisms underlying DoG production upon environmental stress, such as hyperosmotic stress, remain unknown.

Readthrough transcripts have been previously characterized using nuclear RNA sequencing, 4sU-sequencing, mNET-seq and total RNA sequencing (Vilborg et al. 2015; Rutkowski et al. 2015; Hennig et al. 2018; Bauer et al. 2018; Grosso et al. 2015). While these techniques enable the genome-wide identification of DoG transcripts, most do not provide quantitative information regarding the nascent transcription profiles that coincide with DoG induction. Here we use Transient Transcriptome (TT) sequencing coupled with TimeLapse (TL) chemistry (Schwalb et al. 2016; Schofield et al. 2018) to characterize transcriptional profiles that accompany DoG induction upon hyperosmotic stress. This method enables reliable detection of nascent RNAs by using short pulses of metabolic labeling coupled with transcript enrichment and nucleoside recoding chemistry.

The Integrator complex is important for the 3´ end processing of various noncoding RNAs, such as small nuclear RNAs (snRNAs), enhancer RNAs, viral microRNAs, and lncRNAs, as well as replication-dependent histone pre-mRNAs (Albrecht and Wagner 2012; Lai et al. 2015; Xie et al. 2015; Nojima et al. 2018; Skaar et al. 2015). Integrator also regulates promoter-proximal Pol II pausing at protein-coding genes (Stadelmayer et al. 2014; Gardini et al. 2014; Elrod et al. 2019; Beckedorff et al. 2020). Integrator 11 (Int11) is the catalytic subunit of the complex and forms a heterodimer with Int9 (Baillat et al. 2005; Wu et al 2017). These subunits are analogous to CPSF73 and CPSF100 (Dominski et al. 2005; Baillat and Wagner 2015), respectively, which are essential for the 3´ processing of mRNAs (Mandel et al. 2006; Shi et al. 2009).

Here, we report that hyperosmotic stress causes widespread transcriptional repression, from which a minority of human genes escape. Yet, DoGs are induced from genes that experience different transcriptional responses to hyperosmotic stress. ChIP-seq experiments show that Pol II binding profiles correlate with the transcriptional output of genes after hyperosmotic stress. Furthermore, our proteomics results reveal changes in the Pol II interactome after stress. Specifically, we observe a decreased interaction between Pol II and subunits of the Integrator complex, while the interactions between Pol II and factors known to mediate the termination of protein-coding genes are unchanged. ChIP-seq experiments reveal a genome-wide decrease of Int11 and Int3 occupancy on DNA. Correspondingly, siRNA knockdown of Int11 was sufficient to induce readthrough transcription at hundreds of genes. The identity of genes that produce readthrough transcripts upon depletion of functional Int11 partially overlaps the collection of genes that give rise to DoGs after hyperosmotic stress. Together, our results provide mechanistic insights into the biogenesis of a recently characterized class of noncoding RNAs and reveal a novel role for the Integrator complex in transcriptional regulation.

Results

Hyperosmotic stress causes widespread transcriptional repression

We established the nascent transcriptional profiles accompanying DoG induction by performing TT-TL-seq (Schofield et al. 2018) of untreated HEK-293T cells and cells exposed to hyperosmotic stress (Fig. 1A). Specifically, we exposed cells to 80 mM KCl for 60 minutes, but added the nucleoside analog 4-thiouridine (s⁴U) during the last 5 minutes to label RNAs being actively transcribed (Schwalb et al. 2016). After extracting RNA from HEK-293T cells, RNA from Drosophila S2 cells was added to each sample as a normalization control to ensure accurate differential expression analysis (Lovén et al. 2012; Chen et al. 2015). RNAs containing s⁴U were then biotinylated using MTS chemistry and enriched on streptavidin beads (Duffy et al. 2015). Finally, U-to-C mutations were induced using TL chemistry to assess the nascent nature of the enriched RNAs (Schofield et al. 2018). TT-TL-seq experiments were performed using conditions previously found to induce DoGs (Vilborg et al. 2017). RT-qPCR results from HEK-293T cells exposed to hyperosmotic stress for different times confirmed that DoGs were robustly induced after 1 hour of treatment (Fig. S1A), consistent with previous results from NIH-3T3 cells (Vilborg et al. 2017).

Figure 1:

TT-TL-seq reveals transcriptional profiles that accompany DoG induction after hyperosmotic stress. A) Setup for TT-TimeLapse sequencing (TT-TL-seq) experiments in HEK-293T cells. An arrow indicating directionality marks the beginning of each transcription unit. Exons are shown as rectangles and Pol II molecules are light purple ovals with attached nascent RNAs. Genome browser views of TT-TL-seq data for ENAH provide an example of results from untreated (UT) and KCl-treated (KCl) cells after normalization to the spike-in control. B) Browser image of TT-TL-seq data exemplifying a clean gene (HELLS) and a gene that does not meet the criteria for a clean gene (CYP2C18). The DoG produced from HELLS reads into CYP2C18, making the latter appear to be transcriptionally activated by hyperosmotic stress (log₂ FC=5.39). C) The DoG produced from NUCKS1 is assigned to SLC45A3 because of extensive read-in transcription, which also complicates accurate differential expression analysis for SLC45A3 (log₂ FC=4.13). ~4–5 kb upstream of HELLS and NUCKS1 are shown.

It was previously observed that read-in transcription of DoGs into neighboring genes leads to the mis-characterization of overlapping transcripts as being activated by stress (Rutkowski et al. 2015; Hennig et al. 2018; Cardiello et al. 2018; Roth et al. 2020; Fig. 1B). Moreover, read-in transcription confounds the assignment of DoGs to the corresponding parent gene (Fig. 1C). Our analyses suggest that ~55% of expressed genes experience read-in transcription after hyperosmotic stress (Table S1). Therefore, to ensure accurate differential expression analyses and DoG characterization, we generated a sub-list of genes (referred to as “clean genes” throughout the article). The term “clean genes” describes genes that are expressed, do not overlap with readthrough regions that correspond to neighboring genes on either strand and have higher expression within the gene body than the region 1 kb upstream of the gene’s transcription start site (TSS) (Roth et al. 2020; Figs. 1B & C). We identified 4584 clean genes in HEK-293T cells after hyperosmotic stress and analyzed their transcriptional regulation (Table S1).

Consistent with previous reports analyzing steady-state RNAs in human cells (Amat et al. 2019), we identified changes in transcriptional responses after hyperosmotic stress. Our results reveal predominantly decreases in nascent transcript levels after stress (Figs. 2A–C). Specifically, the number of normalized read counts corresponding to clean genes decreases 3-fold after KCl treatment (Fig. S1D). Yet, we find that a subset of clean genes bypasses this transcriptional repression (Figs. 2B, C & S1E), including GADD45B (Fig. 2D), which is known to be induced by hyperosmotic stress (Mak and Kültz 2004). More than 88% of clean genes were repressed after hyperosmotic stress, while only 3% were upregulated (Figs. 2C & S1F). To validate these observations, we extracted total RNA from untreated and KCl-treated cells and performed RT-qPCR using primers targeting intronic regions of several repressed genes. Results from HEK-293T cells and from SK-N-BE(2)C cells confirm the transcriptional decrease of three representative genes upon hyperosmotic stress (Figs. S1B & C) and demonstrate that mature mRNA levels for these genes remain unaffected, while levels of readthrough transcripts increase after hyperosmotic stress. Together our results reveal that hyperosmotic stress alters the nascent transcription profiles of cells by repressing thousands of genes and activating a small subset of genes.

Figure 2:

Hyperosmotic stress leads to widespread transcriptional repression. A) Whole-genome view of TT-TL-seq normalized reads for forward (F) and reverse (R) strands in UT and KCl samples. B) Minus average plot showing the log₂ fold change for clean genes on the y-axis and the mean of normalized counts on the x-axis. Activated genes are shown in purple, genes retaining comparable expression are gray and repressed genes are blue (n=4584). C) Pie chart illustrating the percentage of clean genes within each of the 3 categories of transcriptional regulation (activated gene, log₂ FC >0.58; comparable gene, log₂ FC < 0.58 but > −0.58; repressed gene, log₂ FC<−0.58). D) Browser shots of TT-TL-seq tracks from HEK-293T cells for VANGL1, which is transcriptionally repressed by hyperosmotic stress (log₂ FC= −4.97), PIM2, which retains comparable expression after KCl treatment (log₂ FC=0.28), and GADD45B, which is activated by hyperosmotic stress (log₂ FC=3.72).

Stress-induced readthrough transcripts arise independent of gene transcription levels

Consistent with widespread transcriptional repression, normalized TT-TL-seq read counts within the bodies of DoG-producing clean genes decreased after hyperosmotic stress, while read counts corresponding to DoG regions increased (Fig. 3A). However, log₂ fold changes in nascent RNAs of DoG-producing clean genes show that DoGs are produced from genes that experience all three types of transcriptional responses (Fig. 3B; Table S1). Specifically, 2.9% of DoGs arise from activated clean genes, 87.8% arise from repressed clean genes and 9.3% arise from clean genes that retain comparable expression in stressed and unstressed HEK-293T cells (Fig. 3B). We then asked whether DoGs preferentially arise from genes that are transcriptionally repressed upon hyperosmotic stress. Interestingly, the percentage of DoG-producing genes within each class of transcriptional regulation is consistent, comprising 12–14% (Fig. 3C). We conclude that DoGs are produced regardless of the transcriptional level of their upstream genes (Fig. 3D).

Figure 3:

DoGs arise regardless of the transcriptional levels of their upstream genes upon hyperosmotic stress. A) Interleaved scatter plot showing the sum of normalized TT-TL-seq read counts of DoG-producing clean genes and corresponding DoG regions (n=590) in untreated and KCl-treated HEK-293T cells for two biological replicates. B) Scatter plot showing clean gene log₂ fold change (FC) for the gene body on the x-axis and the log₂ FC for the DoG region on the y-axis. DoG-producing genes that are transcriptionally activated upon stress are represented in purple, genes retaining comparable levels of expression are gray and genes that are repressed are blue. C) Bar graph showing the percentage of DoG-producing clean genes (black) within each category of transcriptional regulation. D) Browser image showing UT and KCl TT-TL-seq reads for OPA1, a transcriptionally repressed DoG-producing clean gene (gene log₂ FC= −3.22), for SERBP1, which retains comparable expression after stress (gene log₂ FC= −0.45), and for a transcriptionally activated DoG-producing clean gene, SERTAD1 (gene log₂ FC= 1.23). E, F) Bar graphs show gene ontology combined scores for the 10 most significantly enriched biological processes in order of increasing p-value for E) DoG-producing clean genes and for F) non-DoG clean genes. Combined scores are the product of the p-value and the z-score as calculated by Enrichr (Chen et al. 2013).

Clean DoG-producing genes are functionally enriched for transcriptional repression

Previous analyses of DoG-producing genes did not reveal any functional enrichment (Vilborg et al. 2017). We suspected that the challenge of assigning DoGs to the correct gene of origin because of their extension into neighboring genes may have complicated previous efforts (Fig. 1C). Therefore, we revisited the question of whether DoG-producing genes are enriched for certain biological processes using only clean genes. We performed gene ontology analysis of DoG-producing clean genes and clean genes that fail to generate DoGs (non-DoG genes) using Enrichr (Chen et al. 2013; Kuleshov et al. 2016; Table S2). Interestingly, 5 out of the 10 enriched terms with the most significant p-values for DoG-producing genes are related to transcriptional repression (Fig. 3E). The remaining 5 terms are related to transcriptional regulation and protein modifications. Non-DoG genes do not show such a striking enrichment for terms related to transcriptional repression compared to other terms (Fig. 3F). Instead, these genes are strongly enriched for general processes related to RNA processing.

Pol II is redistributed along the genome after hyperosmotic stress

We used anti-Pol II ChIP-seq to investigate how Pol II binding correlates with the transcriptional profiles observed through TT-TL-seq in untreated and KCl-treated HEK-293T cells. Meta-analyses of clean DoG-producing genes confirmed a redistribution of Pol II molecules to the downstream regions in stressed cells (Figs. 4A & S2A), consistent with DoG transcription and previous observations (Cardiello et al. 2018; Heinz et al. 2018). As expected, this redistribution of Pol II molecules was not observed downstream of non-DoG genes (Figs. 4A & S2A). Interestingly, our results reveal a decrease in Pol II occupancy near the TES of non-DoG genes after hyperosmotic stress that was not observed for DoG-producing genes.

Figure 4:

Hyperosmotic stress causes redistribution of Pol II molecules across the genome. A) Meta plots showing the log₂ fold change (FC) of anti-Pol II ChIP-seq data normalized to input across the annotated transcription end sites (TES) of DoG-producing clean genes (n=590) and of clean non-DoG genes (n=3994) from UT and KCl-treated HEK-293T cells. B) Volcano plot showing the log₂ FC of read counts from anti-Pol II ChIP-seq data normalized to input for clean genes on the x-axis and the corresponding log₁₀ adjusted p-values on the y-axis. Genes with increased Pol II binding across the gene body are shown in purple, genes retaining comparable levels of Pol II binding are gray and those with decreased Pol II binding across the gene body are blue. C) Browser images of DoG-producing clean genes showing anti-Pol II ChIP-seq tracks normalized to input: OPA1 is a repressed gene, SERBP1 retains comparable expression and SERTAD1 is activated by hyperosmotic stress according to TT-TL-seq data. D) Meta plots showing Pol II binding on DoG-producing clean genes and clean non-DoG genes that are not repressed (top) or genes that are repressed (bottom). Two biological replicates are shown in each meta plot for UT and KCl samples.

We asked to what extent the stress-induced redistribution of Pol II molecules into the regions downstream of genes was responsible for the widespread transcriptional repression observed by TT-TL-seq. Alternatively, the decrease in transcription upon stress might be caused by lowered Pol II binding to repressed genes, as was previously reported for hyperosmotic stress induced by NaCl (Amat et al. 2019). Our results show a similar number of Pol II binding peaks after KCl treatment (Fig. S2B): a subset of overlapping peaks show increased Pol II binding after stress, while a majority of the identified peaks either retain comparable levels or exhibit reduced Pol II occupancy (Fig. S2C). Consistently, we found that Pol II binding decreases over most clean gene bodies after KCl treatment (Fig. 4B).

We verified Pol II occupancy patterns for representative genes within each category of transcriptional regulation. OPA1, a DoG-producing gene that is transcriptionally repressed, shows decreased binding close to the TSS and a shift of Pol II peaks into the downstream region in stressed cells (Fig. 4C). In contrast, SERTAD1, a DoG-producing gene that is activated by hyperosmotic stress, and SERBP1, which retains comparable expression according to our TT-TL-seq data, show increased Pol II binding close to each TSS that extends downstream of these genes. Finally, we assessed the distribution of Pol II molecules across repressed clean genes and across clean genes that are either activated or retain comparable expression after hyperosmotic stress (not repressed). Consistent with the individual gene examples discussed above, clean DoG-producing genes that are not repressed by hyperosmotic stress show increased Pol II binding close to their TSSs and across the gene body, while genes that are transcriptionally repressed by hyperosmotic stress show decreased levels of Pol II binding across the gene body. Similar effects were also observed for clean non-DoG genes (Fig. 4D). Our results are, thus, in agreement with the nascent transcription profiles derived from TT-TL-seq (Figs. 2 & 3), supporting a model where hyperosmotic stress induces both redistribution of Pol II molecules along the genome and a decrease in Pol II binding to repressed genes.

The Pol II interactome is altered by hyperosmotic stress

To gain insight into the mechanisms that promote DoG formation upon hyperosmotic stress, we performed immunoprecipitation of chromatin-bound Pol II coupled with mass spectrometry analysis (Harlen et al. 2016) in SK-N-BE(2)C cells (Fig. 5A). This human neuroblastoma cell line, where KCl-induced DoGs were originally described (Vilborg et al. 2015), was used because we observed more robust DoG induction in these cells than in HEK-293T cells (Figs. S1B & C). We confirmed isolation of chromatin through western blots (Fig. S3A). Overall, our results reveal distinct changes in the Pol II interactome (Fig. 5B). Peptides corresponding to several CPA factors, as well as to Xrn2 were detected in both untreated and KCl-treated samples (Table S3) at least once across six experiments. The presence of Xrn2 and CPSF1 in the anti-Pol II immunoprecipitates of untreated and KCl-treated cells was validated through western blots (Figs. S3B–D). In contrast, 11 of the 14 subunits of the Integrator complex were not detected among the Pol II interactors in the KCl-treated samples (Table S3). To validate this observation, we performed western blots to detect Pol II in chromatin-bound anti-Integrator subunit 3 (Int3) immunoprecipitates from SK-N-BE(2)C and HEK-293T cells. Decreased binding between Pol II and Int3 in KCl-treated samples compared to untreated samples was confirmed for both cell lines (Figs. 5C & D). The specificity of the anti-Int3 antibody was verified by performing siRNA-mediated knockdowns of this subunit followed by western blots of HEK-293T whole cell lysates (Fig. S3E). To ensure that the decreased detection of Integrator subunits was not due to protein degradation, we also assessed the levels of Int3 and Int11 in whole cell lysates from SK-N-BE(2)C cells. Western blots confirmed that the levels of these Integrator subunits do not decrease in cells after hyperosmotic stress (Fig. S3F).

Figure 5:

The Pol II interactome changes upon hyperosmotic stress. A) Experimental setup for mass spectrometry of anti-Pol II immunoprecipitates obtained from the chromatin fractions of SK-N-BE(2)C cells. B) Overlap between Pol II-associated proteins found in UT and KCl-treated samples (803). Peptides for 145 proteins were found only in the UT samples (blue) and peptides corresponding to 53 proteins were found only in the KCl-treated samples at least once across six replicates (white). C, D) Western blots of anti-Int3 immunoprecipitates obtained from chromatin fractions reveal decreased levels of associated Pol II in KCl-treated samples compared to untreated samples in C) SK-N-BE(2)C (p-value=0.0357) and in D) HEK-293T cells (p-value=0.2407). Chromatin fractions from these cell lines are shown as input. Around 8% of the total input and 11% of the immunoprecipitates were loaded on the gels.

Together, our proteomic results show that hyperosmotic stress induces changes to the Pol II interactome that could contribute to the transcriptional landscape revealed by high throughput sequencing. Specifically, important termination factors interact with Pol II despite cellular stress, while the interaction between Pol II and the Integrator complex decreases.

Hyperosmotic stress decreases the occupancy of Integrator subunits on DNA

Further insight into the decreased binding of Integrator subunits to Pol II after hyperosmotic stress (Fig. 5) was gained by performing anti-Int11 and anti-Int3 ChIP-seq in untreated and KCl-treated HEK-293T cells. The Integrator complex was previously shown to bind close to the TSS of genes, where it regulates promoter proximal Pol II pause release (Stadelmayer et al. 2014; Elrod et al. 2019; Beckedorff et al. 2020). Consistently, we predominantly observe Int3 and Int11 peaks close to the TSS of clean genes that decrease after hyperosmotic stress (Figs. 6A & S4A). Meta-analysis of DoG-producing clean genes and clean non-DoG genes revealed similar decreases in Int11 and Int3 occupancy close to the TSSs. Additionally, read counts for identified Int11 binding sites confirm decreased occupancy of this protein on DNA after hyperosmotic stress (Fig. 6B).

Figure 6:

Hyperosmotic stress leads to decreased occupancy of Integrator subunits on DNA. A) Meta-plots showing the log₂ fold change (FC) of anti-Int11 ChIP-seq normalized to input reveal decreased occupancy of Int11 near the TSSs of clean DoG-producing genes (n=590) and clean non-DoG genes (n=3994). B) Volcano plot of read counts for Int11 binding sites. Log₂ FC of KCl/UT is shown on the x-axis with the corresponding adjusted p-values on the y-axis (n=3881). Sites where Int11 occupancy decreases after hyperosmotic stress are blue, sites where occupancy remains comparable are gray and sites where occupancy increases are purple. C) Browser images of anti-Int3 and anti-Int11 ChIP-seq tracks normalized to input for SERBP1 and SERTAD1, which are DoG-producing clean genes that are not repressed by hyperosmotic stress.

Since our TT-TL-seq and anti-Pol II ChIP-seq experiments reveal that hyperosmotic stress causes transcriptional repression accompanied by a decrease in Pol II binding at downregulated genes, we asked whether the observed decreased occupancy of Integrator subunits on DNA was a consequence of a decrease in Pol II occupancy. The occupancy of Int3 and Int11 decreased near the TSS of genes that are not repressed by hyperosmotic stress, including SERTAD1 and SERBP1 (Figs. 3D, 4C & 6C). Furthermore, we quantified the occupancy of Int11 and Pol II across Int11 binding sites that were identified across two biological replicates and observed that the decrease in Int11 binding is not limited to sites exhibiting a concomitant decrease in Pol II occupancy (Figs. 6C & S4B). We conclude that the decreased binding of Integrator subunits to DNA is not a mere consequence of the transcriptional repression induced by hyperosmotic stress.

Depletion of Integrator endonuclease leads to DoG production

The Integrator complex regulates transcription termination at many noncoding RNA loci and has been shown to bind the 3´ end of certain protein-coding genes (Baillat and Wagner 2015; Gardini et al. 2014). Since the interaction between Pol II and Integrator decreases after hyperosmotic stress (Fig. 5), we investigated whether knocking down the catalytic subunit of the complex, Int11, using siRNAs is sufficient to induce DoGs. We transfected HEK-293T cells with an siRNA against Int11 (siInt11) or with a non-targeting siRNA control (siC). HEK-293T cells stably expressing FLAG-tagged, siRNA-resistant wild-type (WT) or catalytically inactive Int11 (E203Q) were also transfected with siInt11 (Baillat et al. 2005; Xie et al. 2015) for 72 hours. DoG levels were then measured by RT-qPCR, confirming that knockdown of endogenous Int11 is sufficient to induce DoGs. Moreover, expression of exogenous WT Int11 reduces DoG induction compared to samples expressing the E203Q mutant or to samples expressing no rescue (Fig. 7A).

Figure 7:

Depletion of Integrator nuclease subunit leads to readthrough transcription. A) RT-qPCR analysis of DoG levels in HEK-293T cells transfected with a scrambled siRNA control (siC) or an siRNA targeting endogenous Int11 (siInt11). The siInt11-transfected cells expressed either exogenous wild-type Int11 (WT), catalytically inactive Int11 (E203Q) or no exogenous Int11 (None). Adjusted p-values compared to siC: None: doSERBP1=0.004; doSNRBP=0.032; doInt6=0.077; doSERTAD1=0.097; WT: doSERBP1=0.011; doSNRBP=0.091; doINT6=0.091; doSERTAD1=0.654; E203Q: doSERBP1=<0.001; doSNRBP=<0.001; doINT6=<0.001; doSERTAD1=0.337. B) Bar graph showing the number of DoGs induced after siRNA knockdown of endogenous Int11. C) Browser image of TT-TL-seq data for two genes that produce DoGs upon depletion of functional Int11. D) Scatter plot showing gene expression log₂ fold change (FC) for siInt11+E203Q HEK-293T cells on the x-axis and the log₂ FC of the corresponding readthrough transcripts on the y-axis. Genes that are activated in siInt11+E203Q cells compared to siC-transfected cells are purple, unaffected genes are gray and repressed genes are blue. All readthrough sites identified in the siInt11+E203Q sample are represented in this plot (n=840). E) Venn diagram displaying overlap between the identities of clean genes that produce readthrough transcripts in siInt11+E203Q cells (dark blue) and those of DoG-producing clean genes in KCl-treated samples (white).

Endogenous Int11 protein levels were reduced by ~90% in siInt11-transfected samples (Fig. S5A). The efficiencies of exogenous WT and E203Q mutant Int11 expression were verified by measuring the levels of unprocessed U2 through RT-qPCR (Skaar et al. 2015; Fig. S5B): unprocessed U2 accumulates in cells without rescue (None) or expressing the E203Q mutant Int11, but not in cells transfected with siC or cells expressing exogenous WT Int11. Northern blots using probes against snRNA U2 showed that levels of mature U2 do not decrease after knockdown (Fig. S5C), consistent with previous reports (Tatomer et al. 2019). Thus, it is unlikely that the effects of Int11 knockdown on DoG production are because of the role of Integrator in snRNA processing.

We assessed the extent to which knockdown of endogenous Int11 induces DoGs genome-wide. To increase cell viability, we performed TT-TL-seq on HEK-293T cells transfected with an siRNA against Int11 for 48 hours (Fig. S5D). Results obtained from cells lacking functional Int11 reveal hundreds of readthrough sites across the genome that are induced by more than 1.5-fold compared to the siC-transfected sample (Fig. 7B; Table S4). According to TT-TL-seq data, induction of readthrough transcription after depletion of endogenous Int11 was most evident upon expression of the E203Q mutant (Figs. 7B & C). Yet, readthrough transcripts observed in siInt11-transfected cells expressing no rescue and in cells expressing the E203Q mutant Int11 were highly correlated (Fig. S5E). As expected, the most highly induced sites of readthrough transcription (Table S4) corresponded to snRNA genes (Baillat et al. 2005; Albrecht and Wagner 2012). We also detected readthrough downstream of lncRNA and histone genes as previously described (Nojima et al. 2018; Skaar et al. 2015). However, most identified readthrough sites were downstream of protein-coding genes (Table S4). Of the 840 readthrough transcripts induced by knockdown of functional Int11 (E203Q sample), 489 exhibited greater than 80% read coverage in the region 5 kb downstream of the annotated termination site of the gene of origin and, therefore, met all criteria to be classified as DoG RNAs (Figs. 7B & C; Table S4).

Depletion of Integrator subunits has been shown to alter the transcriptional levels of certain genes (Gardini et al. 2014; Elrod et al. 2019; Tatomer et al. 2019; Beckedorff et al. 2020). Consistently, we found that depletion of functional Int11 in HEK-293T cells differentially affects more than a thousand genes (Fig. S5F). Examination of the expression levels of the parent genes revealed that readthrough transcripts predominantly arise from upregulated genes or from genes that retain comparable expression after knockdown, while very few arise from genes that are transcriptionally repressed (Fig. 7D).

Given our observation that the interaction between Integrator subunits and Pol II is disrupted by hyperosmotic stress, we asked how the identities of genes producing readthrough transcripts upon depletion of functional Int11 compare to genes that produce DoGs after hyperosmotic stress. We identified 232 clean genes producing readthrough transcripts in siInt11-transfected cells and 351 clean genes producing readthrough transcripts in the siInt11+E203Q mutant sample. Comparison with the 590 DoG-producing clean genes identified in stressed cells showed that up to 25% of KCl-induced DoGs are detected at loci that also produce readthrough transcripts after depletion of functional Int11 (Figs. 7E & S5G). These readthrough transcripts are generally more robustly induced after KCl treatment than upon depletion of functional Int11 (Fig. S5H), suggesting that, although Int11 knockdown is sufficient to produce readthrough transcription, decreased interactions between Integrator and Pol II are not solely responsible for DoG induction upon hyperosmotic stress.

Discussion

We used TT-TL-seq to evaluate nascent transcription profiles that accompany DoG induction under conditions of hyperosmotic stress (Fig. 1) and observed that hyperosmotic stress triggers transcriptional repression that is widespread and of greater magnitude than anticipated (Robbins et al. 1970; Amat et al. 2019). After an hour-long exposure to mild conditions of hyperosmotic stress, only ~12% of clean genes escaped transcriptional repression (Fig. 2). Interestingly, DoGs are produced independent of the transcriptional levels of their upstream genes, and DoG-producing genes are functionally enriched for transcriptional repression according to gene ontology analyses (Fig. 3). Pol II binding profiles demonstrate a correlation with nascent transcription profiles across DoG-producing clean genes and non-DoG genes (Fig. 4). Proteomic analysis of the Pol II interactome revealed decreased binding between the Integrator complex and Pol II upon exposure to hyperosmotic stress (Fig. 5), accompanied by a genome-wide decrease in the occupancy of Integrator subunits on DNA (Fig. 6). Depletion of the catalytic subunit of the Integrator complex, Int11, is sufficient to induce readthrough transcription downstream of hundreds of genes. Loci that produce readthrough transcripts upon depletion of functional Int11 and those that produce DoGs upon hyperosmotic stress partially overlap (Fig. 7).

Our analysis of clean genes (expressed genes lacking overlap with readthrough regions) has enabled us to accurately characterize the relationship between DoGs and their upstream transcripts (Figs. 1 & 3). Consistent with previous observations in heat-shocked cells (Cardiello et al. 2018), we find that DoGs are produced from genes that experience different transcriptional responses to hyperosmotic stress. Importantly, a similar percentage of clean genes that are activated, repressed or unchanged produce DoGs after hyperosmotic stress, revealing that transcriptional levels and DoG production are not mechanistically linked. Our analyses further suggested that ~55% of genes that are expressed in HEK-293T cells might be affected by read-in transcription upon exposure to hyperosmotic stress (Table S1). Intriguingly, other groups have previously reported that read-in transcription from upstream genes negatively affects the splicing of downstream genes upon HSV-1 infection (Rutkowski et al. 2015). Additionally, there is evidence that heat shock and hyperosmotic stress disrupt splicing (Shalgi et al. 2014; Hennig et al. 2018). These findings suggest that altered splicing patterns may be a consequence of DoG production and highlight the extensive impact of DoG production on the transcriptome of stressed cells. Investigating the diversity of unprocessed transcripts that accumulate upon stress will provide a better understanding of how transcriptional and processing dynamics are impacted by stress.

Previously, we were unable to find any functional enrichment for DoG-producing genes in the context of hyperosmotic stress (Vilborg et al. 2017). Upon revisiting this question using only clean genes, we found that many clean DoG-producing genes encode transcriptional repressors (Fig. 3; Table S2). Transcriptional repression has also been reported under other DoG-inducing conditions, including heat shock and viral infection (Mahat et al. 2016; Rutkowski et al. 2015; Bauer et al. 2018). Thus, it is tempting to speculate that DoGs could serve as a store of unprocessed transcripts to facilitate cell survival upon prolonged stress by providing a source of important mRNAs (requiring processing, but not active transcription), thereby reducing the cell’s energetic costs. Other terms of biological enrichment for DoG-producing genes are related to protein modifications. A repository of transcripts that encode such proteins might facilitate recovery from stress. Nuclear-retained, intron-containing transcripts have been shown to serve as a source of gene regulation in other contexts (Boutz et al. 2015). Therefore, there is precedent for unprocessed transcripts, such as DoGs, functioning to regulate gene expression levels via post-transcriptional processing. This idea is also supported by the fact that DoGs have relatively long half-lives (Vilborg et al. 2015). DoG induction might further promote transcriptional repression during hyperosmotic stress by causing transcriptional interference at neighboring genes (Mazo et al. 2007; Muniz et al. 2017). Alternatively, given the transcriptional changes observed upon stress, DoGs may serve to alter the chromatin landscape surrounding their parent and neighboring genes to facilitate prolonged responses throughout stress and recovery from stress.

In agreement with previous observations in heat shock (Cardiello et al. 2018), anti-Pol II ChIP-seq data from KCl-treated cells demonstrate a decrease in Pol II binding across the bodies of repressed clean genes accompanied by a shift of elongating Pol II molecules into downstream regions (Fig. 4). At loci that are not repressed by stress, we observe an increase in Pol II binding along the gene bodies, extending into their downstream regions. These observations are consistent with our TT-TL-seq results (Figs. 2 & 3). The differing levels of Pol II occupancy observed across parent genes suggest an elegant and tightly regulated mechanism of transcriptional termination at DoG-producing genes. Furthermore, our anti-Pol II ChIP-seq data revealed decreased occupancy near the TES of non-DoG genes after hyperosmotic stress, which was not observed for DoG-producing genes (Figs. 4A & S2A). This observation suggests that one factor distinguishing DoG-producing genes from genes that do not produce DoGs is their dependence on Pol II pausing near their TESs for efficient transcription termination.

Surprisingly, rather than observing decreased interactions between Pol II and CPA factors, our proteomic analyses revealed that the binding between Integrator subunits and Pol II decreases after stress (Fig. 5). This was shown by performing anti-Pol II immunoprecipitations followed by mass spectrometry (Table S3) and anti-Int3 immunoprecipitations followed by western blots using an antibody against Pol II (Fig. 5). Furthermore, we found that hyperosmotic stress decreases the occupancy of Integrator subunits 11 and 3 on DNA. Therefore, we asked whether siRNA depletion of Int11, the catalytic subunit of the complex, is sufficient to induce DoGs. TT-TL-seq of siRNA-treated samples revealed that hundreds of DoGs are induced after functional Int11 is depleted from cells (Fig. 7). Moreover, our results reveal that expression of wild-type Int11 reduces the number of DoGs induced in cells where endogenous Int11 has been knocked down compared to no rescue. Additionally, we observe an increase in the number of DoGs induced upon expression of the E203Q mutant Int11, arguing that the nuclease activity of Int11 is important for repressing DoG production under homeostasis. These results highlight the importance of the Integrator complex in the regulation of mRNA production beyond the promoter proximal pause site (Elrod et al. 2019). Perhaps, cells rely on Integrator to mediate a mechanism of transcription termination at a subset of protein-coding genes that is alternative to the canonical mechanism orchestrated by the CPA complex. This might explain why readthrough transcription is not observed downstream of all protein-coding genes. However, our ChIP-seq experiments did not reveal preferential loss of Integrator subunits on DoG-producing genes (Figs. 6 & S4). Therefore, it is possible that the genome-wide loss of Integrator upon hyperosmotic stress plays an indirect role in regulating DoG biogenesis. More work investigating the role of Integrator binding to mRNA 3énds is imperative.

Our results reveal that as many as 25% of genes that produce DoGs upon KCl-treatment also produce readthrough transcripts upon depletion of functional Int11 (Fig. 7). This partial overlap between Int11-depletion-dependent readthrough transcripts and KCl-induced DoGs suggests that the decreased binding of Integrator subunits to Pol II during the stress response contributes to DoG biogenesis. We asked whether the effects of Int11 knockdown and hyperosmotic stress on DoG induction are additive by exposing HEK-293T cells to KCl after 48 or 72 hours of siInt11 transfection, but did not observe such effects on DoG production upon combined treatment (data not shown). However, our results were inconclusive due to the different extents of DoG induction observed in each context (Fig. S5H) as well as by the temporal differences inherent to the experimental protocols. Because of the changes in the transcriptional landscape upon hyperosmotic stress (Fig. 2), it is unlikely that depletion of a single protein is enough to recapitulate the complexity of this response. Hyperosmotic stress induces widespread transcriptional repression to an extent that is not recapitulated by depletion of functional Int11. Additionally, our results reveal changes in Pol II binding patterns to DNA (Fig. 4) and to other proteins (Fig. 5; Table S3) after hyperosmotic stress. The full extent of DoG induction upon cellular stress might also reflect a stress-induced increase in the processivity of elongating Pol II molecules, which would result in readthrough transcription (Fong et al. 2015).

In agreement with the important roles of Xrn2 and CPSF73 in mediating the termination of protein-coding gene transcripts (Proudfoot. 2016), readthrough transcription is observed upon depletion of these proteins (Fong et al. 2015; Eaton et al. 2018; Eaton et al. 2020). However, unlike stress-induced readthrough transcription, depletion of these proteins was not limited to a subset of genes. Using anti-Pol II immunoprecipitation of chromatin fragments coupled with mass spectrometry and western blots, we observed that the overall binding of Pol II and important termination factors, including Xrn2 and CPA factors, is not affected by hyperosmotic stress. Our data, however, do not exclude the possibility that important termination factors are redistributed along the genome after hyperosmotic stress. Anti-CPSF6 ChIP-seq from cells treated with NaCl to induce hyperosmotic stress revealed decreased binding of this subunit at certain genes (Jalihal et al. 2020); yet it is unclear whether this was accompanied by a decrease in Pol II occupancy at these loci. Moreover, anti-CPSF73 ChIP-qPCR data from heat-shocked cells suggest a decrease of this factor at activated DoG-producing genes (Cardiello et al. 2018). It is possible that at genes activated by stress, disruption of the stoichiometry between Pol II and termination factors contributes to DoG production. However, such a model would not explain readthrough transcription at genes that retain comparable expression levels or are transcriptionally repressed, which comprise the majority of DoG-producing genes upon hyperosmotic stress (Fig. 3B). These observations emphasize the need to further explore the features of DoG-producing genes that might make them susceptible to failed termination despite the presence of important termination factors on chromatin.

Limitations:

Our work provides insight into the transcriptional landscape that accompanies DoG biogenesis after hyperosmotic stress and reveals a role for the Integrator complex in DoG production. However, the Integrator complex does not seem to be the only agent involved in the biogenesis of readthrough transcripts upon stress. Additional work assessing the distribution of other important termination factors on DNA after stress should further elucidate how cells selectively induce DoGs from a subset of genes. Furthermore, our data do not provide information about the exact mechanism by which Integrator regulates DoG biogenesis. More work is needed to determine whether the effect of the Integrator complex on readthrough transcripts produced from protein-coding genes is direct or indirect.

STAR Methods

Lead Contact

Additional information and reagent request should be directed to and will be fulfilled by Joan A. Steitz (joan.steitz@yale.edu).

Material Availability

This study did not generate unique reagents.

Data and Code Availability

All high throughput sequencing datasets generated in this study have been deposited to the Gene Expression Omnibus database and can be found as a super series under the following accession number: GSE152063.

Experimental Model and Subject Details

HEK-293T cells were cultured in DMEM supplemented with 10% fetal bovine serum, 1% penicillin and streptomycin, and 1% L-glutamine. SK-N-BE(2)C cells were cultured in a 1:1 mix of DMEM and F-12K media supplemented with 10% fetal bovine serum, 1% penicillin and streptomycin, and 1% L-glutamine. Stable HEK-293T cell lines expressing WT or E203Q mutant Int11 were supplemented using 2 µg/mL of puromycin, as previously described (Xie et al. 2015). Hyperosmotic stress inductions were performed using 80 mM KCl for 1 hour, as described in Vilborg et al. 2015.

Method Details

RNA preparations and RT-qPCRs:

RNA extractions were performed using TRIzol according to the manufacturer’s instructions. After treating samples with RQ1 DNase (Promega), RNA was recovered through PCA extraction followed by ethanol precipitation. cDNA was made from 2 µg RNA using Super Script III and random primers (Invitrogen). Samples were then diluted 1:10 and 2 µL of diluted cDNA were used for RT-qPCR along with 5 µL of iTaq universal SYBR mix (Bio-Rad), 1.5 µL of primer mix (1.5 µM of forward and 1.5 µM of reverse) and 1.5 µL of water. Plates were run on a Bio-Rad CFX384 machine. Results were analyzed using the comparative C_T method (Schmittgen and Livak. 2008). Primers designed for this study were subjected to a primer efficiency test and amplicons were run on agarose gels to ensure that a single band was observed for each pair. Primers used in this study are listed in Table S5.

High throughput sequencing:

TT-TimeLapse:

Samples were fed 1 mM s⁴U (Sigma) during the final 5 minutes of incubation with either KCl or siRNAs. After extracting RNA from HEK-293T cells, samples were spiked with 4% s⁴U-labeled RNA from Drosophila S2 cells. TT-TL-seq was performed as described in Schofield et al. 2018. Sequencing was done by the Yale Center for Genomic Analysis (YCGA).

ChIP-seq:

2.5×10⁷ HEK-293T cells were plated in 15 cm dishes (5×10⁶ per dish) and incubated at 37°C overnight. For anti-Int3 ChIP-seq experiments, 3×10⁷ HEK-293T cells were used. Cells were crosslinked during the last 10 minutes of stress induction using 1% formaldehyde and then washed twice with 1x PBS. Chromatin immunoprecipitation was performed as described in Bieberstein et al. 2014 using 30 µL of anti-Pol II antibody (Cell Signaling D8L4Y), 25 µL of anti-Int11 antibody (Abcam ab75276) or 30 µL of anti-Int3 antibody (Bethyl A302–050A) and 180 µL of Protein A beads. The YCGA prepared libraries and sequenced the samples on the NovaSeq platform. We note that the generation of Int11 ChIP-seq datasets was particularly challenging. We suspect that this was due to the biological properties of this protein as well as the reduced interactions between Integrator subunits and Pol II upon hyperosmotic stress. Subsequently, we performed anti-Int3 ChIP-seq experiments and obtained agreeing results.

Bioinformatics:

TT-TL-seq:

TT-TL-seq reads from HEK-293T cells containing spiked-in RNA from Drosophila S2 cells were filtered with FastUniq (Xu et al. 2012) to remove duplicate sequences and trimmed with Cutadapt v1.16 (Martin 2011) to remove adapter sequences. Unfiltered, trimmed reads were mapped to a combined hg38 and dm6 genome using hisat2 version 2.1.0 with default parameters and --mp 4,2 (Kim et al. 2015). Bam files for uniquely mapped reads (SAM flag 83/163 or 99/147) were created, sorted and indexed using samtools version 1.9 (Li et al. 2009). U-to-C mutation calls were performed as previously described (Schofield et al. 2018). To calculate normalization factors, reads aligning to the dm6 genome were filtered and counted with HTseq htseq-count (Anders et al. 2015). Normalization factors for each sample were calculated with edgeR (Robinson et al. 2010) using read counts mapped to the dm6 genome (calcNormFactors with method = ‘upperquartile’).

Sample	Spike-in normalization factor
siC_rep1	2.65151515
KCl_rep1	0.71138211
siInt11_rep1	2.08333333
siC_rep2	1.94444444
KCl_rep2	0.63405797
siInt11_rep2	1.26811594
WT	2.65151515
mut	1.82291667

Gene counts for reads aligning to the hg38 genome were generated using bedtools multicov version 2.26.0 (Quinlan et al. 2010). Read-in values used to define clean genes were generated using ARTDeco (Roth et al. 2020). DoGs were identified using DoGFinder (Wiesel et al. 2018). Differential expression analysis of normalized read counts in gene bodies and in DoG regions was done using DESeq2 (Love et al. 2014). The list of clean genes is comprised by genes that were expressed by more than 100 read counts after normalization to spike-in, had a read in value ≤ −1 and did not overlap with readthrough regions from neighboring genes on either strand according to bedtools intersect analysis (Quinlan et al. 2010). EnrichR was used for gene ontology analysis of biological enrichments (Chen et al. 2013; Kuleshov et al. 2016). Visualization of the generated data was achieved using ggplot2 (Wickham 2016) and graphPad prism. Eulerr was used to generate proportional Venn diagrams. Normalized tracks were visualized using IGV.

ChIP-seq:

ChIP-seq samples were mapped to the hg38 human genome using bowtie2 (Langmead et al. 2009). Bam files and gene counts were generated as described above. Counts were normalized to input using edgeR normalizeChIPtoInput function (Robinson et al. 2010). MACS2 version 2.1.1 (Zhang et al. 2008) was used to call peaks. Volcano plots were generated using ggplot2 on log₂ fold changes and adjusted p-values calculated with DESeq2. Normalized bigwig files and meta plots were created using deepTools version 3.3.0 (Ramírez et al. 2016). Tracks were visualized using IGV.

Cell fractionation and immunoprecipitations:

For cellular fractionation, 4.8×10⁶ cells were plated in three 15 cm dishes and incubated overnight at 37°C. After KCl treatment, cells were washed twice with 1x PBS, scraped on ice and pelleted by centrifugation (1400rpm at 4°C for 5 minutes). Pellets were resuspended using hypotonic lysis buffer for 5 minutes (Nojima et al. 2016). Nuclei were collected by centrifugation (1400rpm at 4°C for 5 minutes) and the cytoplasmic fraction (supernatant) was discarded. Nuclei were incubated in NUN1 and NUN2 (Wuarin and Schibler 1994) on ice for 15 minutes with intermittent vortexing. Samples were then spun down to pellet the chromatin fraction. Chromatin fragments were generated by incubating the chromatin pellets with 2 µL of micrococcal nuclease for 5 minutes at 37°C, at 1400 rpm in a thermomixer (MNase-digested samples) or by incubating the pellets with 25U of benzonase for 45 minutes (Benzonase-digested samples). After digestion, the samples were clarified (16,000 g at 4°C for 5 minutes) and the supernatant was transferred into a new 1.5 mL tube.

For immunoprecipitation, 10 µg of anti-Pol II antibody (MABI0601) or 3 µg of anti-Int3 (Bethyl A302–050A) antibody were incubated with 25 µL of magnetic anti-mouse beads (NEB) or Protein A beads, respectively, overnight. The following day, the antibody-conjugated beads were washed twice with 1 mL of NET-2 buffer and resuspended in 25 µL of NET-2 buffer. Chromatin fragments were pre-cleared by incubating them with the IgG-conjugated beads for 3 hours in a rotator at 4°C. The beads were then collected using a magnetic rack and the supernatant was transferred to a 1.5 mL tube containing the anti-Pol II conjugated beads. The samples were incubated for an additional 2 hours in the rotator at 4°C. After collecting the beads using a magnetic rack, the supernatant was discarded and the beads were washed four times using 1 mL of NET-2 buffer. The final wash was done using 1x PBS and the samples were resuspended using 8 µL of 10 mM Tris-HCL pH 8.5.

Proteomics:

Digestion.

The anti-Pol II immunoprecipitates were solubilized with a 2.5% solution of ALS-110 (Protea) in 50 mM Tris-HCL pH=8.5 with 5 mM EDTA, and 50 mM DTT directly on the beads. Samples were incubated at 95°C for 10 minutes to reduce cystines. Reactions were cooled on ice for 1 minute, 2 µL of 1 M Tris pH 8.5 were added and cysteines were alkylated by adding 4.7 µL of 100 mM iodoacetamide for 30 minutes in the dark. The reactions were then quenched with the addition of 0.7 µL of 200 mM DTT. Afterwards, 108 µL water, 0.67 µL CaCl₂, 1.33 µL 1 M Tris-HCL pH 8.5 and 7.8 µL 0.5 mg/mL trypsin (Promega) were added and proteins left to digest for 16 hours at 37°C. Surfactant was cleaved with the addition of 20% trifluoroacetic acid to pH < 3 and samples were incubated at 25°C for 15 minutes. Peptides were then separated from beads using a magnetic rack and desalted using C18 silica MicroSpin Columns (The Nest Group). Column elution was performed in 300 µL 80% acetonitrile 0.1% trifluoroacetic acid and peptides were dried by centrifugal vacuum.

Mass spectrometry:

LC-MS/MS was performed using an ACQUITY UPLC M-Class (Waters) paired with a Q Exactive Plus (Thermo). Peptides were separated on a 65-cm-long, 75-µm-internal-diameter PicoFrit column (New Objective) packed in-house to a length of 50 cm with 1.9 µm ReproSil-Pur 120 Å C18-AQ (Dr. Maisch) using methanol as the packing solvent. Peptide separation was achieved using a non-linear 90-min gradient from 1% ACN 0.1% formic acid to 90% ACN 0.1% formic acid with a flowrate of 250 nl/min. Approximately 3–4 µg of peptides were run for each of the six biological replicates with at least one blank run between samples to avoid peptide carryover.

Bioinformatics:

Mass spectra were searched using Maxquant version 1.5.1.2 (Cox et al. 2008) with Carbamidomethyl (C) as a fixed modification, Acetyl (N-Term), Deamidation (NQ), Oxidation (M) as variable modifications and with up to three missed trypsin cleavages, a 7 amino acid minimum length, and 1% false discovery rate (FDR) against the Uniprot Human database (downloaded Dec-2017).

siRNA knockdowns:

For knockdown experiments, 7.5×10⁴ HEK-293T cells were seeded in six well plates and incubated at 37°C overnight. After 24 hours, the cells were transfected with either 50 nM of scrambled siRNA control (siC), siRNA against Int11 (siInt11) (Xie et al. 2015) or siRNA against Int3 (siInt3) (Skaar et al. 2015) using Lipofectamine RNAimax according to manufacturer’s instructions. Cells were incubated at 37°C for 72 hours after which they were washed with 1x PBS, scraped and pelleted by centrifugation (1400 rpm at 4°C for 5 minutes). After discarding the supernatant, cells were resuspended in 150 µL of NET-2 buffer. 50 µL were aliquoted to verify knockdown efficiency through western blots and the rest of the sample was lysed using TRIzol. siRNA experiments for TT-TL experiments were performed by transfecting 1.5×10⁵ cells for 48 hours instead of 72 hours.

Western blots:

For western blots of whole cell lysates, samples were sonicated using 10 cycles of 30 seconds on and 30 seconds off in a Diagenode Bioruptor Pico sonicator (Withers et al. 2018) and digested with micrococcal nuclease for 45 minutes at 37°C. Samples were run at an increasing voltage (up to 120 V) in NuPAGE 4–12% BisTris gels in 1x MOPS buffer (Invitrogen). Proteins were transferred overnight to a 0.45 µm nitrocellulose membrane at 30 V. Membranes were blocked for 1 hour in 5% milk in 1x PBST. Detection of total Pol II and Tyr1-P containing Pol II molecules was achieved using antibodies from MBL (MABI0601, discontinued) and Active motif (61383) respectively, at a 1:1000 dilution. Antibodies against Ints3 (Bethyl A302–050A), Ints11 (Abcam ab75276), Xrn2 (A301–101) and CPSF160 (A301–580A) were used at a 1:800 dilution. Anti-GAPDH antibody (Cell Signaling 14C10) was used at 1:2000, while the antibody against U1–70K (Kastner et al. 1992; Tarn and Steitz. 1994) was used at a 1:500 dilution. All secondary antibodies were used at a 1:2000 dilution of 1x PBST, 3% milk.

Quantification and Statistical Analysis

One sample t-tests and paired t-tests were performed on the fold changes of western blot band intensities and on RT-qPCR data, respectively, using graphPad prism. Differential expression analysis for TT-TL-seq and ChIP-seq datasets were obtained using DESeq2.

Supplementary Material

2

Table S1: Differential expression analysis of clean genes and DoG-producing clean genes. Related to Figures 1–3.

3

Table S2: Gene ontology analysis for DoG-producing clean genes and clean non-DoG genes. Related to Figure 3.

4

Table S3: Proteins identified by mass spectrometry in untreated and KCl-treated samples. Related to Figure 5.

5

Table S5: List of oligos used in this study. Related to Figures S1, S3, 7 and S5.

6

Table S4: Readthrough transcripts induced upon depletion of functional Int11. Related to Figure 7.

Key Resources Table

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
CPSF1	Bethyl Laboratories	A301-580A
GAPDH	Cell Signaling	14C10
H3	Abcam	ab1791
Int3	Bethyl Laboratories	A302-050A
Int11	Abcam	ab75276
RNA Pol II CTD	MBL	MABI0601
RNA Pol II (N-terminus)	Cell Signaling	D8L4Y
RNA Pol II Tyr1-P CTD	Active Motif	61383
U1-70K	Kastner et al. 1992; Tarn and Steitz. 1994	N/A
Xrn2	Bethyl Laboratories	A301-101
Chemicals, Peptides, and Recombinant Proteins
Dynabeads Protein A	Invitrogen	1001D
4-thiouridine	Sigma	T4509
Goat anti-mouse beads	NEB	S1431S
Lipofectamine RNAimax	Invitrogen	13778-150
Micrococcal nuclease	NEB	M0247S
RQ1 DNase	Promega	M610A
Superscript III	Invitrogen	56575
Deposited Data
Raw and analyzed TT-TL-seq data	This paper	GSE152059
Raw and analyzed anti-Pol II ChIP-seq data	This paper	GSE152062
Raw and analyzed anti-Int3 and anti-Int11 data	This paper	GSE159190
Experimental Models: Cell Lines
HEK-293T	Steitz lab	N/A
HEK-293T 2x FLAG-Int11 WT	Xie et al. 2015	N/A
HEK-293T 2x FLAG-Int11 E203Q	Xie et al. 2015	N/A
Drosophila S2 cells	Simon lab	N/A
SK-N-BE(2)C	Steitz lab	N/A
Oligonucleotides
RT-qPCR primers	Table S5	N/A
siRNAs	Table S5	N/A
Northern blot probes	Table S5	N/A
Random primers	Invitrogen	58875
Software and Algorithms
ARTDeco	Roth et al. 2020	N/A
bedtools	Quinlan et al. 2010	N/A
Bowtie2	Langmead et al. 2009	N/A
Cutadapt	Martin 2011	N/A
DESeq2	Love et al. 2014	N/A
DoGFinder	Wiesel et al. 2018	N/A
edger	Robinson et al. 2010	N/A
enrichr	Chen et al. 2013; Kuleshov et al. 2016	N/A
eulerr	Larsson J (2020)	https://github.com/jolars/eulerr
FastUniq	Xu et al. 2012	N/A
ggplot2	Wickham. 2016	N/A
Hisat2	Kim et al. 2015	N/A
HTSeq	Anders et al. 2015	N/A
MACS2	Zhang et al. 2008	N/A
Prism	GraphPad	N/A
samtools	Li et al. 2009	N/A

Highlights.

• Hyperosmotic stress triggers transcriptional repression of many genes.

• DoG RNAs arise independent of the transcriptional level of their upstream gene.

• Interactions between Pol II and Integrator subunits decrease after salt stress.

• Depletion of the Int11 nuclease subunit induces the production of hundreds of DoGs.