Multimodal regulatory elements within a hormone-specific super enhancer control a heterogenous transcriptional response

Summary:

The hormone stimulated Glucocorticoid Receptor (GR) modulates transcription by interacting with thousands of enhancers and GR binding sites (GBSs) throughout the genome. Here, we examined the effects of GR binding on enhancer dynamics and interrogated the contributions of individual GBSs to the hormone response. Hormone treatment resulted in genome-wide reorganization of the enhancer landscape in breast cancer cells. Upstream of the DDIT4 oncogene, GR bound to four sites constituting a hormone-dependent super enhancer. Three GBSs were required as hormone-dependent enhancers that differentially promoted histone acetylation and transcription frequency and burst size. Conversely, the fourth site suppressed transcription and hormone treatment alleviated this suppression. GR binding within the super enhancer promoted a loop-switching mechanism that allowed interaction of the DDIT4 TSS with the active GBSs. The unique functions of each GR binding site contributed to hormone-induced transcriptional heterogeneity and demonstrate the potential for targeted modulation of oncogene expression.

Graphical Abstract

eTOC blurb:

Hoffman et al. show that the Dex-responsive gene DDIT4 is regulated by four GR binding sites (GBSs) within a Dex-specific super enhancer. GR promotes DDIT4 transcription via a loop-switching mechanism that disrupts a suppressive interaction between the TSS and GBS4 in favor of TSS interaction with the other enhancer-like GBSs.

Introduction:

Synthetic glucocorticoids such as Dexamethasone (Dex) are widely prescribed to treat a variety of human diseases including auto-immune disorders, asthma, cancer, and COVID-19. In human cells, glucocorticoid exposure elicits a rapid and robust transcriptional response mediated by the Glucocorticoid Receptor (GR)(Dietrich et al., 2020, Miranda et al., 2013, Weikum et al., 2017). For instance, in a variety of cancer cell lines, Dex treatment triggers GR binding at tens of thousands of GR binding sites (GBSs) throughout the genome and alters the expression of hundreds to thousands of genes (Hoffman et al., 2020, Hoffman et al., 2018, Mcdowell et al., 2018, Thomas-Chollier et al., 2013). Many of these GBSs exhibit characteristics of transcriptional enhancers such as chromatin accessibility and acetylation of Histone H3 Lysine 27 residue (H3K27ac) (Hoffman et al., 2018, Vockley et al., 2016). Furthermore, some GBSs have the capacity to activate transcription in reporter assays (Vockley et al., 2016). Thus, the transcriptional response to glucocorticoids is likely driven by the modulation of enhancer activity in response to GR binding.

Super enhancers (SEs) are regions of the genome that contain clusters of multiple enhancers and are significantly enriched for enhancer characteristics (Hnisz et al., 2013, Whyte et al., 2013). SEs can be cell-type and/or disease specific and are often associated with genes that are critical drivers of cell identity and disease progression (Hnisz et al., 2013, Whyte et al., 2013). The individual constituent enhancers within SEs can be functionally distinct and constituent enhancers critical for SE function are referred to as hub enhancers (Huang et al., 2018). GR and estrogen receptor motifs are enriched in super enhancers in various cell lineages including breast cancer cells (Bojcsuk et al., 2017, Shin et al., 2016) and SEs have been shown to control the estrogen response in the mouse uterus (Hewitt et al., 2020). Intriguingly, GR often binds to clusters of GBSs within and around target genes (Grange et al., 2001, Vockley et al., 2016, Chandler et al., 1983, Miranda et al., 2013). For instance, 11 GBSs were identified around the GR target gene TSC22D3 (GILZ) in U2OS osteosarcoma cells and deletions of large regions of the genome containing 8 or more of these GBSs were required to ablate GILZ glucocorticoid responsiveness (Thormann et al., 2018). Thus, the transcriptional response to glucocorticoids may involve the binding and regulation of SEs by GR.

Recent work has demonstrated that the transcriptional response to steroid hormones is heterogeneous. For instance, TFF1 exhibits highly variable transcription dynamics in estrogen-treated breast cancer cells, with some cells showing a rapid and robust response while others remain inactive for hours or days (Rodriguez et al., 2019). Single-cell RNA-seq in breast cancer cells revealed that while every Dex-treated cell exhibited a transcriptional response, there was a large degree of variability in the response in individual cells, with on average only ~30% of GR target genes showing a response in each cell (Hoffman et al., 2020). However, the mechanisms by which GR-modulated enhancer activity contributes to transcriptional heterogeneity remain unknown. As the hormone response is so frequently dysregulated and/or pharmacologically targeted in human disease, understanding the molecular mechanisms underlying the heterogeneity in the transcriptional response to Dex and other steroid hormones is critical.

Here we show that Dex treatment causes genome-wide changes in H3K27ac that result in a reorganization of the enhancer landscape in breast cancer cells. We also identify a subset of Dex-specific super enhancers. Among these is an SE that encompasses DDIT4, a GR target gene that is implicated in a variety of cancers and exhibits a highly heterogenous Dex response. We utilize this super enhancer as a model to characterize and dissect the mechanisms of transcriptional regulation by GR. We identify four GBSs within the DDIT4 super enhancer that have distinct chromatin characteristics. CRISPR based knockouts of these GBSs reveal that each makes unique contributions to the Dex-dependent induction of DDIT4 transcription. We demonstrate that three of the GBSs act as enhancers of DDIT4 transcription and the fourth acts as a suppressor of transcriptional activity within the super enhancer. The GBSs control DDIT4 through mechanisms regulating histone acetylation, transcription burst size and frequency, and dynamic chromatin interactions. The unique characteristics and functions of each GBS suggest that DDIT4 transcriptional heterogeneity arises from differential utilization of constituent GR-bound regulatory elements within a hormone-specific super enhancer.

Results:

Glucocorticoid treatment alters the enhancer landscape in breast cancer cells

We define GR binding sites (GBSs) as regions of Dex-induced GR enrichment identified as peaks in ChIP-seq datasets. Our previous classification of GBSs in T47D A1-2 human breast cancer cells (Archer et al., 1994) revealed subsets of GBSs that either gained or were pre-patterned with enhancer characteristics such as eRNA transcription, ATAC accessibility, and acetylation of Histone H3 Lysine 27 (H3K27ac) (Hoffman et al., 2018). However, this analysis did not specifically identify hormone-regulated enhancers. To do so, we performed H3K27ac ChIP-seq treated with Dexamethasone (Dex) or ethanol vehicle (Veh) as a control for 1 hour. Peak calling identified ~30,000 H3K27ac peaks in each sample. 3763 H3K27ac peaks were specific to Veh-treated cells and 8329 H3K27ac peaks were specific to Dex-treated cells (Figure 1A). Thus, Dex treatment substantially increased and altered the distribution of H3K27ac in the genome of A1-2 cells.

Figure 1: Glucocorticoid treatment alters the super enhancer landscape.

A) Comparison of H3K27ac peak sets between Veh-treated and 1 hour Dex-treated A1-2 cells. B-D) Scatter plots comparing the ROSE enhancer rankings of super enhancers called from H3K27ac ChIP-seq experiments.

We next used the ROSE rank ordering of super-enhancers code (Lovén et al., 2013, Whyte et al., 2013) to determine whether Dex-treatment altered the categorization of super enhancers (SEs). In each sample, approximately 600 regions were ranked as SEs. Comparison of the enhancer ranks of the SEs called in either two Veh-treated or two Dex-treated samples revealed that there was little variability in ranking of SEs across biological replicates (Figure 1B,C). However, comparison of SE rankings between Veh- and Dex-treated cells revealed substantial differences (Figure 1D). Indeed, this comparison revealed 105 Dex-gained SEs and 101 Dex-lost SEs (Figure 1D). Similar results were obtained using RNA Polymerase II ChIP-seq signal to rank SEs (Figure S1A). As expected, examination of the constituent enhancers within these treatment-specific SEs revealed that upon Dex treatment, H3K27ac signal was reduced over Dex-lost SEs and increased over Dex-gained SEs (Figure S1B). Intriguingly, GR coverage was poorly organized over these constituent enhancers (Figure S1C), suggesting that they did not fully overlap with GBSs. However, GBSs were present within SEs, and Dex-gained SEs were enriched for GR binding compared to Dex-lost SEs (Figure S1D). Furthermore, GBSs within Dex-gained SEs exhibited stronger GR enrichment than GBSs within Dex-lost SEs (Figure S1E). Thus, Dex treatment altered the enhancer and SE landscape.

DDIT4 hormone induction and transcriptional heterogeneity

We previously characterized the transcriptional response to stress hormone exposure in the A1-2 cells using bulk RNAseq and single-cell RNAseq (scRNAseq) over a time course of Dex treatments (Hoffman et al., 2020). In these datasets, we found a striking degree of cell-to-cell heterogeneity in both the number of Dex responsive genes and the degree to which individual genes responded to Dex.

To determine whether Dex-dependent enhancer dynamics drove Dex-induced transcriptional heterogeneity, we chose a representative Dex-gained SE for detailed analysis. DNA Damage Inducible Transcript 4 (DDIT4) is a Dex-induced gene located within a ~40kb Dex-gained SE hereafter referred to as the DDIT4 SE (Figure 1D). DDIT4 made a salient target for further study as it exhibited heterogeneity in both the level of induction and in the number of expressing cells in response to Dex. In scRNAseq, DDIT4 was not detected in every cell (Figure 2A). DDIT4 was detected in less than half of Veh-treated cells and the detection rate increased to nearly 90% during Dex treatment (Figure 2A). Furthermore, Dex-treated cells displayed a wide range of DDIT4 responses, with some cells inducing DDIT4 greater than 10-fold above vehicle while the majority of cells induced DDIT4 to intermediate levels (Figure 2A).

Figure 2: Heterogeneity in Dex-induced DDIT4 transcription.

A) Scaled Log10 DDIT4 expression (box and whiskers with overlayed dots representing individual cells, left axis) and detection rate (line plot, right axis) in scRNAseq (n = 400 cells per timepoint). B) Fluorescent images of DDIT4 mRNA smFISH. Green spots mark DDIT4 mRNA; DAPI nuclear stain in blue. Scale bar = 10 um. C) Box and whisker plot depicting average smFISH Log10 mRNA per cell across 3 biological replicates with ~1000 cells per replicate. D) Smoothed density histogram depicting the distribution of DDIT4 mRNA/cell counts in smFISH. E) Smoothed density histogram depicting the distribution of raw DDIT4 mRNA/cell counts in scRNAseq. F) Table depicting the number of cells analyzed and the percent responding cells in each method.

As scRNAseq is subject to technical limitations and biased toward the detection of highly expressed genes, we sought to independently validate our findings to confirm that DDIT4 exhibited a heterogeneous transcriptional response to Dex. To do so, we quantified DDIT4 mRNA expression using single-molecule fluorescent in situ hybridization (smFISH). 30 singly labeled mRNA probes were designed to DDIT4 exons, and smFISH was performed over a time course of Dex treatment (Figure 2B). At least one DDIT4 mRNA was detected in 98% of vehicle-treated cells (Figure 2B,C). The pattern of DDIT4 induction in smFISH was very similar to that observed in scRNAseq, with DDIT4 being upregulated at 1 hour of Dex treatment and peaking at 4 hours (Figure 2D). Like with scRNAseq, upregulation of DDIT4 mRNA in individual cells was highly heterogeneous in smFISH, with the number of mRNA/cell in Dex-treated cells ranging from 0 to over 200 (Figure 2C,D). Thus, while smFISH was far more sensitive than scRNAseq, both assays provided similar profiles of the relative amounts of DDIT4 RNA/cell across a population of cells (Figure 2D,E). To determine the percentage of cells that activated DDIT4 in response to Dex in each experiment, we set expression thresholds using the mean plus one standard deviation of DDIT4 RNA/cell in Veh-treated control cells. In both assays, 65% of cells treated with Dex for 1-8 hours exhibited DDIT4 upregulation in response to Dex (Figure 2F). Thus, scRNAseq and smFISH independently demonstrated that DDIT4 exhibits a heterogeneous transcriptional response to Dex. As such, DDIT4 was an attractive target to study the role of GR-regulated enhancer dynamics in the transcriptional response to Dex.

Dex-induced activation of the DDIT4 super enhancer

As the transcriptional response to Dex is mediated by GR chromatin interactions at GBSs throughout the genome, we searched our previously published GR ChIP-seq data (Hoffman et al., 2018) for GBSs within the DDIT4 SE. We identified four GR ChIP-seq peaks at sites 19kb, 20kb, 25kb, and 30kb upstream of the DDIT4 TSS (Figure S2A, Figure 3A). These peaks were named GBS1 through GBS4 in order of their proximity to the DDIT4 TSS (Figure S2A, Figure 3A). To examine the dynamics of GR binding to these GBSs, we performed GR ChIP-seq over a time course of Dex treatments from 15 minutes to 8 hours. GR bound to all 4 GBSs within the first 15 minutes of Dex treatment (Figure 3A). GR enrichment was reduced at all GBSs by 8hrs, with enrichment at GBS1, GBS2, and GBS3 reduced to approximately 50% maximum and enrichment at GBS4 reduced to 19% maximum (Figure 3A). Thus, we reasoned that Dex induction of DDIT4 transcription was likely mediated by GR binding to this upstream cluster of 4 GBSs.

Figure 3: Identification of GR bound enhancers within the DDIT4 super enhancer:

A) GR ChIP-seq coverage time course over the DDIT4 super enhancer. Percentage of maximum enrichment at 8hrs are overlayed on 8hr track. Y-axis scale for each track = 0-279. B) H3K27ac and PolII ChIP-seq coverage over the DDIT4 gene. H3K27ac Y-axis scale = 0-86, PolII Y-axis scale = 0-395. C) H3K27ac and PolII ChIP-seq coverage over the DDIT4 GBSs. H3K27ac Y-axis scale = 0-77, PolII Y-axis scale = 0-50. D-F) Luciferase reporter assays using fragments of the DDIT4 GBSs inserted upstream of the Luciferase gene alone (D) and either upstream (E) or downstream (F) of the Luciferase gene with a tk promoter. Error bars represent standard deviation of biological triplicates. G) RT-PCR for transcripts generated within the DDIT4 super enhancer. Error bars represent standard deviation of biological triplicates

To determine whether these GBSs represented constituent enhancers within the Dex-gained DDIT4 SE, we examined H3K27ac ChIP-seq coverage over the region. H3K27ac was strongly enriched around the DDIT4 TSS and was largely unaffected by Dex treatment (Figure 3B). In vehicle-treated cells, lower levels of H3K27ac enrichment were observed over GBS2, GBS3, and another region approximately 13kb upstream (−13kb) of the DDIT4 TSS (Figure 3B,C). After 1 hour Dex treatment, there was a broad induction of H3K27ac enrichment between the −13kb region and GBS4, with peaks in coverage over GBS1, 2, and 3 (Figure 3B,C). Upstream of GBS3, the Dex-induced H3K27ac gradually tapered off and relatively little enrichment was detected beyond GBS4 (Figure 3B,C). Thus, the chromatin surrounding the GBSs represented the Dex-induced region within the DDIT4 SE.

We next examined whether the constituent GBSs of the DDIT4 SE could individually function as active regulatory elements. We cloned ~100bp fragments from the center of each of the GBSs into pGL3B-derived expression vectors and assayed for Dex induction of luciferase activity. When inserted directly upstream of the Luc coding sequence, each of the GBS fragments conferred some degree of Dex-responsiveness (Figure 3D). This Dex-responsiveness was increased by the addition of a minimal Thymidine Kinase (TK) promoter sequence between the GBS fragment and the Luc gene (Figure 3E). Furthermore, insertion of each of the GBS fragments downstream of the Luc coding sequence with a 5’ TK promoter also resulted in Dex-induced luciferase activity (Figure 3F). Thus, each constituent DDIT4 GBS had the capacity to individually activate transcription and act as transcriptional enhancers in response to Dex.

Prior to Dex treatment, GBSs in A1-2 cells are often enriched for chromatin accessibility, presence of the BRG1 chromatin remodeler, and pioneer factor binding (Burd et al., 2012, Hoffman et al., 2018). We utilized these previously published datasets to determine whether these features were present within the DDIT4 SE. In Veh-treated cells, chromatin accessibility as measured by ATAC-seq was strongly enriched at the DDIT4 TSS and low levels of enrichment were detected at the 13kb upstream region and at GBSs 2, 3, and 4 (Figure S2A,B). Upon Dex treatment, ATAC coverage was increased around each of the GBSs, indicating that Dex treatment promoted chromatin accessibility (Figure S2B). Along with this gain of accessibility, Dex-dependent enrichment of BRG1 was observed at GBSs 1, 2, and 3 (Figure S2A,B). However, the pioneer factors FOXA1 and GATA3 were not enriched within the DDIT4 SE (Figure S2C). These data indicated that while the DDIT4 GBSs exhibited some accessibility prior to hormone treatment, they were not pre-patterned by BRG1, FOXA1, or GATA3. Instead, GR binding in response to Dex led to the recruitment of the BRG1 chromatin remodeler and opening of the chromatin.

RNA Polymerase II (PolII) binding and eRNA transcription are additional hallmarks of active enhancers. In Veh-treated cells, PolII was enriched over the DDIT4 TSS and gene body and a low level of enrichment was detected over the −13kb region (Figure 3B,C). After 1hr Dex treatment, PolII enrichment was markedly increased over DDIT4 as well as GBS2, GBS3, and the −13kb region (Figure 3B,C). To determine whether PolII was actively transcribing RNA from these sites, we examined our previously published total RNA-seq coverage (Hoffman et al., 2020). In Veh-treated cells, RNA was produced by the −13kb region, but not GBS2 or GBS3 (Figure S2D). After Dex treatment, transcription of potential eRNAs was induced at both GBS2 and GBS3 and increased at the −13kb region (Figure S2D). RT-PCR over a time course of short Dex treatments revealed that induction of transcription at GBS2 and GBS3 preceded induction of DDIT4 transcription. Nascent DDIT4 transcription (detected by RT-PCR for exon-intron junctions in the unspliced mRNA) was significantly induced at 30 minutes of Dex treatment, whereas transcription at GBS2 and GBS3 was significantly induced at 10 and 15 minutes, respectively (Figure 3G). Conversely, induction of transcription at the −13kb region was not significantly different than DDIT4 transcription (Figure S2E). Thus, we hypothesized that GBS2 and GBS3 were GR-dependent enhancers that mediated the induction of DDIT4 expression in response to Dex.

Generation of DDIT4 enhancer knockout cell lines

To examine the requirement for the constituent GBSs within the DDIT4 SE, we used CRISPR to individually delete each GBS. We designed CRISPR guides flanking each GBS and used a dual-guide approach to delete a 300-500bp region around each GBS to ensure the entire binding site was deleted (Figure S3A). Following selection, colonies were selected and expanded as independent potential knockout lines. Genomic DNA from these expanded lines was used for a PCR genotyping screen to identify cell lines with complete genomic deletion of each GBS (as judged by complete loss of a WT band)(Figure S3B). Using this approach, we developed enhancer knockout (eKO) cell lines for each GBS, heretofore named GBS1-eKO, GBS2-eKO, GBS3-eKO, and GBS4-eKO. Despite screening over 200 potential GBS1-eKO clones, we were only able to isolate a single complete GBS1-eKO cell line. For the GBS2, 3, and 4 eKOs, we isolated at least three independent complete knockout clones and selected a single clone at random to proceed with further analysis. We validated GBS deletion using GR ChIP-QPCR in each line. Each individual eKO resulted in a complete loss of GR ChIP enrichment at the corresponding GBS (Figure S3D). Each eKO maintained GR binding at the remaining GBSs, suggesting that GR binding to the four GBSs within the DDIT4 SE was largely independent (Figure S3D). To ensure that the deletions did not result in spurious GR binding, we used the JASPAR2020 database of transcription factor binding profiles (Fornes et al., 2020) to identify GR motifs (GR regulatory elements or GREs) within the DDIT4 SE. GREs were identified within each of the GBSs, as well as at four other locations with the DDIT4 SE (Figure S3C). GR ChIP enrichment was not detected at any of these additional GREs in parental A1-2 cells or any of the eKOs (Figure S3D). As such, each eKO demonstrated a complete ablation of GR binding to the appropriate GBS without any evidence of spurious GR binding. These eKO lines were utilized for subsequent experiments to interrogate the role of each GBS in the DDIT4 transcriptional response to Dex.

Differential requirements for constituent GBSs within the DDIT4 super enhancer

To determine the effects of each eKO on DDIT4 transcription, we treated each cell line with Dex for 2, 4, and 8 hour timepoints and compared to Veh-treated controls. For the GBS2, 3, and 4 eKOs, the pattern DDIT4 induction was validated in three independent eKO clones (Figure S3E). In this experiment, the DDIT4 mature and nascent transcripts were respectively upregulated 3.75-fold and 3.3-fold at 2 hours Dex and remained at similar levels at 4 and 8 hours Dex (Figure 4A). The GBS1-eKO and GBS3-eKO each resulted in delayed and reduced induction of DDIT4 expression (Figure 4, yellow and orange lines). In both eKOs, fold-induction of the DDIT4 mature transcript was significantly reduced at 2 hours Dex and failed to reach the maximum level of induction observed in A1-2 cells within the time course (Figure 4A). Similarly, induction of the DDIT4 nascent transcript was also delayed in the GBS1-eKO and GBS3-eKO cell lines (Figure 4B). However, the nascent transcript was slightly upregulated in these lines in Veh-treated cells and at the 8 hour Dex timepoint where it reached the levels seen in A1-2 cells at 2 hours Dex (Figure 4B). Thus, GBS1 and GBS3 were individually required for the timing and extent of DDIT4 transcriptional induction in response to Dex.

Figure 4: GBS-eKOs differentially alter DDIT4 gene and eRNA expression:

A-E) line plots depicting mRNA and eRNA expression in EtOH-treated cells and cells treated with Dex for 2, 4, or 8 hours. Error bars represent standard deviation of biological triplicates. Asterisks (*) indicate data points significantly above the matched A1-2 timepoint (p <0.05). Number (#) signs indicate data points significantly below the matched A1-2 timepoint (p < 0.05).

Deletion of GBS2 had more dramatic effects on DDIT4 expression (Figure 4, blue lines). In GBS2-eKO cells, the mature DDIT4 transcript was not up-regulated at the 2 or 4 hour Dex timepoints and was only upregulated 1.5-fold at the 8 hour timepoint (Figure 4A). Similarly, the nascent DDIT4 transcript was not induced at 2 hours Dex and induction fell significantly short of A1-2 cells at 4 and 8 hours Dex (Figure 4B). At all Dex timepoints, the mature and nascent transcripts were significantly reduced in the GBS2-eKO relative to the corresponding timepoints in the A1-2 cells (Figure 4A,B). These data demonstrated that GBS2 was required for full induction of DDIT4 transcription in response to Dex.

Surprisingly, deletion of GBS4 resulted in enhanced induction of DDIT4 transcription (Figure 4, green lines). Basal levels of the mature and nascent DDIT4 transcripts were significantly upregulated in the GBS4-eKO (Figure 4A,B). Upon Dex treatment, induction of the DDIT4 mature transcript in the GBS4-eKO cells was significantly upregulated at 4 and 8 hours of Dex treatment (Figure 4A). Induction of the nascent DDIT4 transcript was significantly upregulated at all Dex timepoints (Figure 4B). Thus, GBS4 acted as a suppressor of DDIT4 expression and limited the levels of DDIT4 transcriptional induction in response to Dex.

We also examined the effects of each eKO on eRNA transcription within the DDIT4 SE. In wild-type A1-2 cells, the eRNAs transcribed from GBS2, GBS3, and the −13kb region were respectively induced 19-fold, 16-fold, and 5-fold at 2 hours Dex and remained at similar levels at 4 and 8 hours (Figure 4C-E). As expected, the GBS2-eKO and GBS3-eKO completely silenced transcription of their corresponding eRNAs (Figure 4C,D). While the GBS3-eKO had little effect on the transcription of the other eRNAs, each of the other eKOs had distinct effects on eRNA expression. Unexpectedly, in GBS1-eKO cells, induction of the GBS2 and −13kb eRNAs was significantly increased at 4 and 8 hours of Dex treatment (Figure 4D,E). In the GBS2-eKO, induction of the GBS3 eRNA was increased at all timepoints whereas Dex induction of the −13kb eRNA was blocked (Figure 4C,E). Finally, the GBS4-eKO resulted in massive upregulation of all three eRNAs in Veh- and Dex-treated cells (Figure 4C-E). Taken together, these data indicated that within the DDIT4 SE, each of the constituent GBSs was uniquely required to regulate the DDIT4 and eRNA transcription. GBS1, GBS2, and GBS3 each acted as enhancers that were required for proper timing and levels of DDIT4 and eRNA induction in response to Dex. Conversely, GBS4 suppressed DDIT4 transcription and the activity of the other constituent GBSs .

DDIT4 enhancer knockouts alter hormone-dependent histone acetylation

Dex treatment induced a broad region of H3K27ac accumulation within the DDIT4 super enhancer (Figure 3C). To determine the effects of the eKOs on the induction of histone acetylation, we performed H3K27ac ChIP-seq in all four lines. At the DDIT4 TSS, the pattern of H3K27ac was largely unaffected in each of the eKOs (Figure S4). However, each knockout resulted in unique effects on the patterns of H3K27ac across the GBSs. In comparison to A1-2 cells, Dex-induced H3K27ac in the GBS1-eKO was diminished over a region starting at GBS1 and extending 1.5kb downstream (Figure 5A). Similarly, in the GBS3-eKO, Dex-induced K27ac was markedly reduced over a 5kb region centered on GBS3 (Figure 5C). In the GBS2-eKO, the loss of Dex-dependent H3K27ac was more widespread, with a reduction of coverage spanning 20kb from GBS4 to the −13kb region (Figure 5B). Finally, in the GBS4-eKO, there was an increase in Dex-induced H3K27ac over a 1kb region flanking the downstream side of GBS4 with a sharp drop in H3K27ac coverage at the site of the GBS4 deletion (Figure 5D). In summary, GBSs 1, 2, and 3 were each required for regions of Dex-dependent histone acetylation within the DDIT4 super enhancer, whereas there was slightly more H3K27ac in the GBS4-eKO.

Figure 5: GBS deletions disrupt Dex-induction of H3K27 acetylation.

A-D) representative browser tracks depicting H3K27ac ChIP-seq coverage over the DDIT4 GBSs in cells treated with Dex for 1 hour. Shaded boxes indicate regions with differential patterns in the eKO lines relative to parental A1-2 cells. Y-axis scales for all tracks = 0-93. E) Table indicating whether the region encompassing DDIT4 and the GBSs ranks as a super enhancer in EtOH or Dex treatment.

Intriguingly, the loss of acetylation in the GBS1, 2, and 3 eKOs was sufficient to downgrade the ROSE rank-ordering of the DDIT4 region (Figure 5E). As such, the region was no longer called as a Dex-gained SE in these eKO lines, indicating that each of the constituent active GBSs were required for SE status. Importantly, we did not observe any spurious acetylation of H3K27 in the eKOs, indicating that new enhancers did not arise in the eKOs. All together, these data demonstrated that the constituent GBSs were individually required for regulation of both transcription and histone acetylation within the DDIT4 SE.

Dynamic chromatin looping between the DDIT4 TSS and GBSs

We next sought to examine whether differential binding of other transcription factors played a role in the unique suppressive function of GBS4. We again used the JASPAR2020 database of transcription factor binding profiles (Fornes et al., 2020) to identify transcription factor motifs within the regions deleted for each eKO. To identify factors likely to bind within the GBSs, we filtered potential motifs by gene expression to only consider transcription factors expressed at a similar or higher level than GR (GR is in the top 15% of expressed genes; we included all transcription factors within the top 25% of expressed genes). As expected, motifs for known GR cofactors such as STAT1/3, SREBF1/2, and SP1/3 were found within all four GBSs (Figure S5A). Only E2F1 and E2F2 motifs were unique to GBS4. While E2F1 and E2F2 are typically thought of as transcriptional activators, we sought to determine whether they indeed bound to GBS4. Publicly available ENCODE E2F1 ChIP-seq data in MCF7 breast cancer cells revealed E2F1 was enriched at the DDIT4 TSS and in the −13kb region, but was not observed at any of the GBSs (Figure S5B). A similar binding pattern was also observed in ENCODE data from HeLa cells (Figure S5B). To examine whether E2F1 bound to GBS4 in A1-2 cells, we performed ChIP-PCR with two independent antibodies against E2F1. Both gave similar patterns of enrichment, with hormone-independent E2F1 binding detected at the TSS and at GBS2, but not at GBS4 (Figure S5C,D). As such, we could not conclude that the suppressive function of GBS4 was driven by E2F1 or other transcription factors.

Distal enhancer elements have been demonstrated to regulate the activity of TSSs through the formation of chromatin loops. Sites involved in forming these chromatin loops are often bound by CTCF and the Cohesin complex. Indeed, CTCF motifs were found under GBS1 and GBS2 (Figure S5A). To determine whether chromatin looping occurred in the DDIT4 SE, we first performed ChIP-QPCR for CTCF and the Cohesin subunits RAD21 and SMC3 in wild-type A1-2 cells. Relative to IgG negative controls, there was enrichment for both RAD21 and SMC3 at the DDIT4 TSS and each of the GBSs (Figure S6B,C). However, CTCF was not markedly enriched at any of these sites (Figure S6D). We searched the entire 40kb DDIT4 SE using JASPAR2020 and found 10 potential CTCF binding sites (Figure S6A). Among these, only site “CTCF_i” within the DDIT4 gene body showed a strong level of Dex-independent enrichment (Figure S6D). This site also showed strong enrichment for RAD21 and SMC3 (Figure S6B,C). These patterns of Cohesin and CTCF binding suggested that chromatin looping may be occurring within the DDIT4 SE.

To examine the possibility of chromatin looping within the DDIT4 SE in greater detail, we performed chromosome confirmation capture with sequencing (4C-seq) in biological triplicates. First, we used the DDIT4 TSS as a capture viewpoint to identify chromatin loops that might be contributing to DDIT4 transcription. Nearly all detected TSS interactions occurred proximally to DDIT4, with some above-background interactions detected in the surrounding ~50mb of chromosome 10 (Figure S6E,F). Additional above-background interactions were detected on chromosome 20, likely the result of the translocations between chromosomes 10 and 20 observed in T47D cells (Rondón-Lagos et al., 2014). At the chromosome level, there were no apparent differences in the interactions detected in vehicle compared to Dex (Figure S6E,F). However, focusing on the DDIT4 SE region revealed a consistent Dex-dependent gain in chromatin interactions over the GBSs (Figure 6A, blue line, Figure S6G). The strongest Dex enrichment occurred over GBS3 and Dex-enriched interactions stopped at GBS4 (Figure 6A, blue line, Figure S6G). Thus, Dex treatment promoted the interaction of the DDIT4 TSS with the GBSs that enhanced DDIT4 transcription.

Figure 6: Dynamic chromatin looping within the DDIT4 SE.

A) Ribbon plots depicting Veh-subtracted 4C coverage (Dex signal minus Veh signal) from the DDIT4 TSS (blue) and GBS4 (red) viewpoints. Bold lines represent the mean Veh-subtracted coverage and ribbons represent standard deviation between biological triplicates. Black bars represent GBSs and DDIT4 gene. B) Ribbon plots depicting Veh-subtracted 4C coverage using the DDIT4 TSS as the interaction viewpoint. In each plot, the black line represents the mean subtracted 4C signal in A1-2 cells and the colored line represents the mean subtracted 4C signal in the indicated GBS-eKO. Ribbons on both lines represent the standard deviation between biological triplicates. Black dots above each plot indicate 4C fragments significantly lost in the eKOs with t-test p-value <0.05. C) Browser track with bars depicting the regions defined for comparison in D and E. D) Box and dot plots depicting the Veh-subtracted 4C coverage for individual restriction fragments within the regions defined in C using the TSS (blue) and GBS4 (red) viewpoints. E) Slope plots comparing 4C coverage in Veh- and Dex-treated using a viewpoint within the 22k region.

We performed 4C-seq using the DDIT4 TSS viewpoint in each of the eKO cell lines in biological triplicate to determine whether any of the GBSs were required for these interactions. Surprisingly, all four of the GBS-eKOs resulted in reduced Dex-dependent interaction between the TSS and the region encompassing the GBSs (Figure 6B). The greatest disruption in TSS interactions occurred in the GBS2-eKO, where Dex-dependent interactions were diminished or lost over the 10kb surrounding GBSs 1, 2, and 3 (Figure 6B). TSS interactions over GBS1 and GBS3 were also diminished/lost in the GBS1-eKO and diminished in the GBS4-eKO (Figure 6B). The loss of interaction in the GBS3-eKO was largely specific to the region surrounding GBS3 (Figure 6B). Thus, all four GBSs were required for Dex-dependent chromatin interactions between the TSS and the GBSs required for DDIT4 induction.

As Dex-induced TSS interactions within the DDIT4 SE appeared to stop at GBS4, we next performed 4C-seq with GBS4 as the viewpoint to determine whether it was involved in other chromatin interactions. At the chromosome level, GBS4 interactions were largely similar to the TSS viewpoint, with no apparent hormone effect and nearly all interactions occurring proximally to GBS4 (Figure S6H,I). Within the DDIT4 SE, GBS4 interactions with the other GBSs were reduced by Dex treatment (Figure6a, red line, Figure S6J). To further compare the TSS and GBS4 interactions, we defined regions surrounding the GBSs and other points of interest within the DDIT4 SE (Figure 6C) and generated boxplots of the Veh-subtracted 4C signal (Figure 6D). This revealed that upon Dex treatment, the region encompassing GBS1, 2, and 3 lost interactions with GBS4 and gained interactions with the DDIT4 TSS (Figure 6D). The region between GBS2 and GBS3 (labeled as “22k”) exhibited the greatest difference in interactions. To confirm these dynamics, we performed 4C-seq using a restriction fragment within the 22k region as the viewpoint. Interactions between the 22k viewpoint and fragments surrounding GBS4 were uniformly reduced following Dex treatment, and fragments surrounding the DDIT4 TSS uniformly gained interactions (Figure 6E). Thus, the region surrounding GBS1, 2, and 3 was involved in bi-directional interactions, with Dex treatment promoting a loss of interaction with GBS4 in favor of gained interactions with the DDIT4 TSS.

Transcriptional mechanisms of GBS enhancer function

To explore potential mechanisms by which the GBSs regulated DDIT4 transcription, we performed smFISH for DDIT4 mature transcripts in each of the eKOs (Figure 7A,B). In A1-2 cells, the average number of DDIT4 mRNA per cell increased from ~15 in Veh to ~40 after 4 hours Dex treatment. None of the eKOs had significant effects on the level of DDIT4 mature transcripts in Veh-treated cells (Figure 7C). However, after 4 hours Dex treatment, GBS1-eKO, GBS2-eKO, and GBS3-eKO cells had significantly reduced numbers of DDIT4 mature transcripts (Figure 7C). The GBS1-eKO and GBS2-eKO exhibited the most severe reduction in DDIT4 induction (Figure 7C). Conversely, the number of mature transcripts per cell in Dex-treated GBS4-eKO cells was significantly increased relative to Dex-treated A1-2 cells (Figure 7C).

Figure 7: GBS deletions alter DDIT4 transcription dynamics.

A) Model of two-color sequential HCR and Stellaris smFISH probe design. B) Representative smFISH images in each cell line/treatment pair. Active transcription sites are identified by co-localization of green (exon probes) and red (intron probes) spots. Scale bar = 10 um. C-E) Dot and whisker plots depicting mean DDIT4 RNA/cell (C), the percent of cells actively transcribing DDIT4 (D) and burst size (E). Whiskers represent standard error of 6 biological replicates with >750 cells per replicate; asterisks indicate t-test p-values <0.0125 for eKO lines versus parental A1-2 cells. F) Box and dot plot depicting DDIT4 transcriptional response rate. Black dots represent 6 biological replicates; asterisks indicate t-test p-values <0.01.

We next sought to interrogate the dynamics of DDIT4 transcription in the eKO cells lines. To determine how frequently cells were actively transcribing DDIT4, we designed probes to detect the introns present in nascent DDIT4 transcripts to allow identification of transcription sites (TSs) in the nuclei (Figure 7A). Given that the DDIT4 introns were too small for standard smFISH probe design, we designed hybridization chain reaction (HCR) probes to both introns of DDIT4 (supplementary table). We developed a two-color sequential smFISH protocol by which active TSs could be identified by colocalization of Stellaris mature transcript spots and HCR intron/nascent transcript spots (Figure 7A). On average, 13.8% of Veh-treated A1-2 cells exhibited at least one active TS (Figure 7D). After 4 hours Dex treatment, the fraction of cells with an active DDIT4 TS increased to 41% (Figure 7D). In Dex-treated GBS1-eKO and GBS2-eKO cells, there was a significant reduction in the ratio of cells with active TSs (Figure 7D). Conversely, Dex-treated GBS4-eKO cells had a significant increase in cells with active TSs (Figure 7D). Thus, the heterogeneity of DDIT4 transcriptional activation was governed by the activity of the GBSs within the DDIT4 super enhancer.

Transcriptional heterogeneity could arise from variability in either the frequency or the amplitude of activation. While the frequency of TS activation appeared to be altered in the eKOs, this could be caused by changes in the amplitude, or burst size, of each TS activation. To determine the burst size, we quantified the intensity of mature DDIT4 transcript spots at TSs and normalized this to the median intensity of mature transcript spots in the cytoplasm (assumed to be single transcripts) (Rodriguez et al., 2019). In A1-2 cells, the median burst size in Veh-treated cells was 4.6 RNA per TS, and Dex treatment increased the burst size to 9.3 RNA per TS (Figure 7E). None of the eKOs altered the burst size in Veh-treated conditions. However, the Dex-induced increase in burst size was severely reduced in GBS1-eKO and GBS2-eKO cells, and moderately reduced in GBS3-eKO cells (Figure 7E). Therefore, GR modulation of burst size was an important component of the overall DDIT4 Dex response. Intriguingly, the GBS4-eKO had no effect on burst size (Figure 7E), indicating that GBS4 modulated the frequency of TS activation.

Finally, we calculated the DDIT4 transcriptional response rate in each replicate as in Figure 2G. In A1-2 cells, ~56% of Dex-treated cells had mRNA/cell counts more than one standard deviation above the mean level in Veh-treated A1-2 cells (Figure 7F). The response rate was significantly reduced in the GBS1-eKO, GBS2-eKO, and GBS3-eKO cells, with the GBS1-eKO and GBS2-eKO showing the greatest reduction (Figure 7F). On average, the response rate was increased in the GBS4-eKO, however greater variability between replicates prevented statistical significance. Taken all together, these experiments demonstrated that the constituent GBSs of the DDIT4 super enhancer regulated DDIT4 transcription through distinct and sometime opposing mechanisms involving histone acetylation, chromatin looping, and modulation of the frequency and amplitude of transcriptional bursts.

Discussion:

In A1-2 human breast cancer cells, Dex treatment resulted in reorganization of H3K27ac marked enhancers throughout the genome. This reorganization further resulted in differential SE rankings and the identification of 105 Dex-gained SEs. The GR target gene DDIT4 resides within a Dex-gained super enhancer and responds rapidly and heterogeneously to Dex treatment. We identified four GBSs in a region within the SE and 18-30kb upstream of the DDIT4 TSS. Fragments of each of these GBSs were able to confer Dex-responsiveness as promoters and enhancers in vector-based luciferase assays. The GBSs were also marked by Dex-induced H3K27ac. Utilizing CRISPR, we individually deleted each of the GBSs to determine their contribution to Dex induction of DDIT4 transcription and their function as constituents of the DDIT4 super enhancer.

GBS1, GBS2, and GBS3 were required as enhancers of DDIT4 transcription in response to Dex. Loss of either GBS resulted in delayed induction of DDIT4 and a reduction in local Dex-induced H3K27ac and chromatin looping with the DDIT4 TSS. However, they each exhibited differential characteristics and mechanisms of action. GBS3 produced an eRNA in response to Dex and there was a broad loss of Dex-induced H3K27ac in the GBS3-eKO. The GBS1-eKO had greater effects on transcription burst size and the number of DDIT4 mRNA/cell, and also reduced the ratio of cells with active DDIT4 TSs. GBS2 was required as the major driver of DDIT4 mRNA levels within the DDIT super enhancer. GBS2 also produced an eRNA in response to Dex and exhibited the greatest level of H3K27ac among the GBSs. Deletion of GBS2 almost completely blocked the DDIT4 transcriptional response. Concomitant with this, the GBS2-eKO exhibited reduction of H3K27ac accumulation throughout the DDIT4 SE. GBS2-eKO cells also had reduced mRNA/cell, percent cells transcribing, and burst size following Dex treatment. Finally, Dex treatment promoted the interaction of the DDIT4 TSS with the region surrounding these GBSs. As such, GBSs 1, 2, and 3 acted as Dex-dependent enhancers with distinct mechanisms of action that cooperatively interacted with the DDIT4 TSS to promote transcription.

GBS4 acted as a silencer or suppressor of transcriptional activity within the DDIT4 super enhancer. Negative regulatory elements or silencers play an opposing role to enhancers by repressing the transcription of active TSSs (Ogbourne and Antalis, 1998). Like enhancers, silencers can act on TSSs from distance through long range chromatin interactions (Gisselbrecht et al., 2020). Intriguingly, some silencers are bifunctional and can act as enhancers in other cellular contexts or specific tissue/cell types (Gisselbrecht et al., 2020). In the GBS4-eKO, the frequency of TS activation in Dex was upregulated, Dex-induction of DDIT4 mRNA doubled, and eRNA expression within the DDIT4 super enhancer was massively upregulated. However, compared to the other eKOs, the observed changes in H3K27ac were minor. Unlike GBS1, 2, and 3, BRG1 was not recruited to GBS4 and Dex treatment did not result in enhanced chromatin accessibility, indicating that GR-recruited chromatin remodeling was not responsible for Dex-mediated alleviation of the suppression. Additionally, DDIT4 and the eRNAs were also upregulated in Veh-treated GBS4-eKO cells, indicating that suppression of the DDIT4 super enhancer by GBS4 was hormone-independent. Finally, in Veh-treated cells, GBS4 formed chromatin interactions with GBSs 1,2, and 3 and Dex treatment reduced the frequency of these interactions. Taken together, these data indicated that GBS4 suppressed DDIT4 transcription through interaction with the other GBSs that blocked their activity and interaction with the TSS.

Based on these findings, we can organize the results to propose the following model of DDIT4 transcriptional regulation (Figure S7). In uninduced cells, DDIT4 is expressed at a basal level, low levels of H3K27ac are present around the GBSs, and a chromatin loop is formed between GBS4 and the other GBSs. In these conditions, GBS4 suppresses transcription of GBS2 and GBS3 eRNAs and blocks interaction of GBSs 1, 2, and 3 with the DDIT4 TSS. Upon Dex treatment, GR enters the nucleus and interacts with chromatin at the 4 GBSs. This inhibits the chromatin interactions between GBS4 and the other GBSs and promotes the interaction of GBSs 1, 2, and 3 with the TSS. Dex treatment also promotes histone acetylation, BRG1 recruitment, and eRNA transcription at GBSs 1,2, and 3. Thus, GR binding to the four GBSs promotes a loop-switching mechanism that drives the induction of DDIT4 transcription in response to Dex.

Whereas it may be expected that loss of GBS4 would increase TSS interaction with GBSs 1, 2, and 3, these interactions were reduced in all four eKOs. We hypothesize that the greatly increased level of active transcription within the DDIT4 SE in the GBS4-eKO may result in reduced detection of TSS chromatin interactions. As transcription of the GBS2 and GBS3 eRNAs was upregulated ~100-fold, it is possible that greater accumulation of transcriptional machinery with the DDIT4 SE either reduced the ability to detect chromatin interactions or obviated the requirement for chromatin interactions for DDIT4 induction. Alternatively, this massive upregulation of eRNA transcription in the GBS4-eKO could suggest a functional role for the GBS2 and GBS3 eRNAs in promoting DDIT4 transcription. In the Estrogen and Androgen responses, hormone-induced eRNAs have been shown to stabilize chromatin looping between receptor-bound enhancers and target promoters (Hah et al., 2013, Hsieh et al., 2014, Li et al., 2013). However, the reduced interactions in the GBS4 suggested that eRNA function within the DDIT4 SE occurred through a mechanism independent of chromatin looping. eRNAs have been shown to promote histone acetylation and active chromatin environments through interactions with CBP and BRD4 (Bose et al., 2017, Rahnamoun et al., 2018). eRNAs have also been demonstrated to act as decoys for NELF complex, promoting release of paused PolII into elongation and increased transcription (Schaukowitch et al., 2014). As such, increased basal and Dex-induced DDIT4 expression in the GBS4-eKO could be the result of enhanced pause release or histone acetylation triggered by the upregulation of the GBS2 and GBS3 eRNAs.

The transcriptional heterogeneity observed in the DDIT4 Dex response could arise from stochasticity within the DDIT4 SE. GR chromatin interactions are short-lived (Paakinaho et al., 2017) and if GR binding to the four GBSs is stochastic, then the output of DDIT4 transcription would be determined by the frequency of usage of each GBS across multiple DDIT4 alleles. Indeed, our ChIP-seq revealed that the dynamics of GR binding differed between the 4 GBSs, demonstrating that GR binding within the DDIT4 SE was not uniform. As GR often binds to accessible regions of chromatin (Burd et al., 2012, Hoffman et al., 2018), GR binding could also be influenced by heterogeneity in the underlying chromatin landscape or by differential chromosome topologies between genomic copies of the DDIT4 locus. Persistent transcriptionally repressive gene states in subsets of cells contributed to heterogeneity in the induction of the Estrogen-responsive gene TFF1 (Rodriguez et al., 2019). Here, we observed that Dex treatment shifted the DDIT4 SE it to a more “active” chromatin state with Dex-dependent enrichment of H3K27ac, chromatin accessibility, and transcriptional activity. Furthermore, the chromatin interactions between GBS4 and GBSs 1, 2, and 3 demonstrated that chromatin interactions limited the activity of the DDIT4 SE in the absence of hormone. If the chromatin landscape and looping interactions within the DDIT4 SE are stochastic, then differential usage of the GBSs would result in differential alleviation of the repressive chromatin states. Thus, we hypothesize that variability in the underlying chromatin state and GR binding are intertwined and together give rise to the heterogeneous DDIT4 transcriptional response.

Dysregulation of DDIT4 transcription has been found in a variety of cancer types. DDIT4 is upregulated and implicated as an oncogene in gastric cancer and ovarian carcinoma (Chang et al., 2018, Du et al., 2018). In acute myeloid leukemia, glioblastoma multiforme, and breast, colon, skin, and lung cancer, high levels of DDIT4 were significantly associated with worse prognoses (Pinto et al., 2017). The oncogenic mechanism of DDIT4 is unknown, but it has been implicated in promoting resistance to Dex-induced cell death in lymphocytes (Molitoris et al., 2011). Conversely, suppression of DDIT4 expression by miR-495 promoted proliferation of breast cancer stem cells in hypoxic conditions, suggesting DDIT4 could also function as a tumor suppressor. Here we demonstrate that levels of DDIT4 can be modulated by targeting individual regulatory elements within the hormone-responsive DDIT4 SE. The transcriptional response to hormones has been pharmacologically targeted with great success in a variety of hormone-dependent cancers (Ulm et al., 2019) and the dysregulation of cellular transcriptional programs in cancer has arisen as an attractive therapeutic target (Bradner et al., 2017). As cancer treatments become increasingly targeted based on patient-specific genetics, understanding the mechanistic contributions of individual regulatory elements to oncogene expression is becoming critical for the development of prospective therapeutics.

Limitations of the Study:

In this study, we utilize an in vitro model of cultured human breast cancer cells. As we do not observe any significant difference in growth rate, wound healing, or other biological phenotypes in the eKO cell lines, our ability to assess the biological impact of glucocorticoid induction of DDIT4 expression is limited. This limitation may be due to the heterogeneity in the DDIT4 response to hormone in A1-2 cells. Because only a subset of the cells exhibit high DDIT4 inducibility, it may be difficult to detect gross biological defects. To gain further insight into the physiological role of the DDIT4 hormone response, we hope to extend our study with future work investigating the role of the DDIT4 GBSs in more clinically relevant model systems of tumorigenesis and stress response. This study is also limited in scope to the mechanisms of regulation of a single gene and lacks further genome-wide investigation of the effects caused by either the genetic deletions or the disrupted DDIT4 transcription response. While RNA-seq and further ChIP-seq experiments were outside the scope of this manuscript, further investigation into transcriptomic effects of the eKOs could provide additional insight into the requirement for hormone induction of DDIT4.

STAR Methods:

Resource Availability:

Lead Contact:

The lead contact is Trevor K. Archer. Please address correspondence to archer1@niehs.nih.gov

Materials Availability:

Materials generated for or used within this study are available upon correspondence.

Data and Code Availability

• All sequencing data generated for this study has been deposited at GEO GSE173798. Previously published data that was used in this study are available at GEO GSE112491 and GEO GSE141834.

• This paper does not report original code.

• Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request.

Experimental Model and Subject Details:

T47D A1-2 cells (Archer et al., 1994) were grown in MEM (Gibco 10370) supplemented with 10% FBS, GlutaMAX (Gibco 3505), HEPES (Sigma H0887), Penicillin/Streptomycin (source), and G418 (Source). 24 hours prior to hormone treatment, cells were switched to phenol red-free MEM (GIBCO 51200) supplemented with 5% Charcoal/Dextran treated FBS (R&D Systems S11650) with the same additional supplements. For hormone treatments, Dexamethasone (100nM) or vehicle (equal volume 95% ethanol) were added to media for the indicated times.

Method Details:

RNA isolation and RT-PCR

RNA was isolated using Qiagen RNeasy kits with on-column DNase treatment. cDNA was synthesized using Superscript III kits with random hexamers (Thermo Scientific). qPCR was run using BioRad ssoAdvanced Universal SYBR Green Supermix and gene/transcript specific primers (see supplemental table). All experiments were performed in biological triplicate. qPCR data was normalized to the geometric mean of ACTB, GAPDH, and TUBA1B expression.

ChIP-seq

Cells were fixed in PBS with 1% methanol-free formaldehyde (Thermoscientific 28906) at 37C for 10 min and quenched with glycine. Cell pellets were resuspended in MNase swelling buffer (25 mM HEPES pH 7.9, 1.5 mM MgCl2, 10 mM KCl, 0.1% NP-40, 5 mM Sodium Butyrate, 0.5 mM PMSF, and protease inhibitor cocktail (PIC) (ThermoFisher Scientific, 78429)) for enzymatic fragmentation or into hypotonic buffer (10 mM HEPES-NaOH pH 7.9, 10 mM KCl, 1.5 mM MgCl2, 340 mM sucrose, 10% glycerol, 0.1% Triton X-100, 5 mM Sodium Butyrate, 0.5 mM PMSF, and PIC) for mechanical sonication. Cells were incubated on ice for 10 min then subjected to Dounce homogenization using 20-strokes (Duran Wheaton Kimble, 357542).

For enzymatic fragmentation with MNase, nuclei were sedimented through digestion buffer (15mM Hepes pH 7.9, 60mM KCl, 15mM NaCl, 0.32M Sucrose, 5 mM Sodium Butyrate, 0.5 mM PMSF, and PIC) by centrifugation for 7 min at 930g (4C). Nuclei pellets were fully resuspended in digestion buffer containing 3.3 mM CaCl2 and digested with MNase (0.7U/1x10^6 cells, Worthington, cat#) at 37C for 15min. MNase fragmentation was stopped by addition of 10 mM EGTA and incubation for 5 min on ice. Digested nuclei were lysed in LB3 buffer (10 mM Tris-HCl pH 8.0, 100 mM NaCl, 1 mM EDTA, 0.5 mM EGTA, 0.1% Na-deoxycholate, 0.5% N- Lauroylsarcosine, 1% Triton X-100, 5 mM Sodium Butyrate, 0.5 mM PMSF, and PIC) using 5-passages through a 25g ( 0.508 mm x 1.6 cm) needle and brief sonication using 4 cycles (30 sec on/off, power high, 4C) with a Bioruptor sonicator (source). MNase digested chromatin lysate was recovered after 15 min centrifugation at 13,000g (4C).

For mechanical sonication, nuclei pellets were resuspended in Shearing buffer (10 mM Tris-HCl pH 8.0, 1 mM EDTA, 0.5 mM EGTA, 0.5 mM PMSF, 5 mM Sodium Butyrate, 0.1% SDS, and PIC) and chromatin was fragmented by sonication using the Covaris S220 (Peak Power, 175; Duty Factor, 10; Cycles/Burst, 200; Total Time, 600 sec). Sonicated chromatin was recovered by centrifugation at 20,000g (4C) for 10 min.

The concentration of prepared fragmentated chromatin lysates determined by Bradford assay. Fragmentated chromatin was diluted two-fold in 2xIP buffer (20 mM Tris-HCl pH 8.0, 300 mM NaCl, 2 mM EDTA, 20% Glycerol, 1% Triton X-100, 0.5 mM PMSF, 5 mM Sodium Butyrate and PIC) and immunoprecipitation performed with the indicated antibodies at a ratio of 1 ug antibody per 200 ug chromatin. Immune complexes were captured using protein A and G dynabeads (source), washed once each with low salt buffer (20 mM Tris-HCl pH 8.0, 150 mM NaCl, 2 mM EDTA, 1% Triton X-100, 0.1% SDS), high salt buffer (same as low salt buffer, except 500 mM NaCl), and LiCl buffer (Tris-HCl pH 8.0, 250 mM LiCl, 2 mM EDTA, 1 % NP-40, 1% (wt/vol) sodium deoxycholate), and twice with TE. Eluted DNA was RNaseA (source) and Proteinase K (source) treated and purified using Qiagen PCR purification columns. ChIP-seq libraries were generated using the ACCEL-NGS^® 2S PLUS DNA library kit according to manufacturer’s protocol and sequenced on the Illumina NextSeq platform. ChIP-seq data analysis was performed as previously described (Hoffman et al., 2018) and ChIP-seq coverage was visualized with the IGV genome browser (Robinson et al., 2011).

Super enhancer analysis

To rank enhancers and identify super enhancers, we utilized the ROSE (Lovén et al., 2013, Whyte et al., 2013) software according to the recommended protocol. Briefly, we generated a union set of H3K27ac peaks from the H3K27ac ChIP-seq data sets generated for this manuscript. Enhancers were stitched together and enhancer/SE rankings were calculated for every H3K27ac ChIP-seq replicate in A1-2 and GBS-eKO cell lines. Constituent enhancers and GR peaks within SEs were identified using Bedtools (Quinlan and Hall, 2010).

Sequential smFISH

The DDIT4 mRNA probe set (30 probes) was designed to all DDIT4 exons using Stellaris Probe Designer (https://www.biosearchtech.com/stellaris-designer) with the following parameters: masking level of 5, oligo length of 20, minimum spacing of 2 nucleotides. The probes were ordered from Biosearch Stellaris in Quasar 570. Because of the intron length, a hybridization chain reaction (HCR) probe set was designed to both DDIT4 introns (Choi et al., 2018). Three pairs of HCR initiator probe sequences were detected using Cy5 labeled DNA HCR hairpins. Sequential smFISH was carried out by performing HCR FISH first, followed immediately by performing Stellaris FISH according to the protocols outlined by (Choi et al., 2018) and Biosearch Stellaris. These protocols were performed with minor modifications. Briefly, cells were grown in 12 well plates on 18mm No. 1.5 coverslips. Cells were rinsed with PBS before fixation with 4% PFA in PBS. Post fixation cells were washed with PBS and stored overnight in 70% ethanol at 4°C. Cells were then washed with 2x SSC prior to a 30 min pre-hybridization at 37°C in 30% probe wash buffer without citric acid. HCR initiator probes were hybridized overnight at 37°C in 30% probe hybridization buffer without citric acid. After hybridization the cells were washed 4 times with 30% probe wash buffer and twice with 5x SSCT. Signal was then amplified using 90 minute incubation at room temperature in amplification buffer containing both Cy5 labeled DNA HCR hairpins. Excess hairpins were removed with 5 washes in 5x SSCT. After completion of HCR smFISH protocol, the adherent mammalian cell Stellaris RNA FISH Protocol was followed. Hybridized samples were mounted in Prolong Gold with DAPI and allowed to fully dry before imaging.

Microscopy

smFISH imaging was performed on a custom-built microscope. This microscope consisted of an ASI (www.asiimaging.com) Rapid Automated Modular Microscope System (RAMM) base, a Hamamatsu ORCA-Flash4 V3 CMOS camera (https://www.hamamatsu.com/, C13440-20CU), Lumencore SpectraX (https://lumencor.com/), an ASI High Speed Filter Wheel (FW-1000), ASI MS-2000 Small XY stage, and a Zeiss C-Apochromat 40x / 1.20 NA UV-VIS-IR objective. DAPI, Quasar 570, and Cy5 were excited using the SpectraX violet, green, and red filters, respectively. Images were acquired using Micro-Manager (Edelstein et al., 2010). Twenty-five fields were imaged using 10 micron Z stacks at 0.5 micron intervals. The maximum intensity projections were taken and used for analysis. The number of cells used per sample used for the analysis ranged from 750-3400 cells.

Image analysis

smFISH spot segmentation was performed on the maximum intensity projections. RNA smFISH spots were identified by fitting to a 2D Gaussian mask and performing local background subtraction using custom python code adapted from (Donovan et al., 2019) and (Gowthaman et al., 2021). Cytoplasmic and nuclear masks were generated using Cellprofiler (Carpenter et al., 2006). Transcription sites were identified by mRNA spots that were colocalized with intron spots within the same nuclear masks. Spots with a Euclidean distances less than or equal to 5 pixels were considered colocalized. Burst size was determined by dividing the mRNA intensity at the transcription site by the median intensity of cytoplasmic mRNA in the same cell. Only cells with at least 5 cytoplasmic RNA were considered for transcription site normalization.

4C-seq

4C-seq was performed according to the published protocol (Krijger et al., 2020). Briefly, batches of 5-10 million cells were fixed in 2% Formaldehyde solution and flash frozen in liquid nitrogen. A single pellet of 5-10 million cells was used for each 4C replicate. Chromatin was digested with the primary restriction enzyme, ligated with T4 DNA Ligase to generate the 3C template, and purified with NucleoMag PCR beads. The purified 3C template was subsequently digested with the secondary restriction enzyme, ligated with T4 DNA Ligase to generate the 4C template, and again purified with NucleoMag PCR beads. 4C-seq libraries were generated by two-step PCR using Expand Long Template PCR kit. The first step PCR reaction used reading and non-reading primers to amplify 4C fragments and was purified with a 0.8X AMPure XP purification. The second step PCR reaction added sequencing adapters and indexes and was purified using Qiagen PCR purification kits. Libraries were sequenced on an Illumina MiSeq with a target of 1-2 million single-end, 100bp reads per library. Data was processed using Pipe4C (Krijger et al., 2020) and run via Rscript for R version 3.6.1 with the following parameters: trimLength=0, mismatchMax=0, and minAmountReads=10. Data QC involved reviewing output reports to verify that the overall alignment rate was typically >85% and resulted in at least 1 million reads aligned to the fragment index. Results were visualized using ggplot2 (Wickham, 2016).

Generation of GBS-eKO cell lines

CRISPR guide pairs flanking each DDIT4 GBS were designed using the now discontinued web tool hosted by the Zhang Lab. Complementary guides were annealed and ligated into px330. For each GBS, A1-2 cells were co-transfected with both guide plasmids and with an empty vector containing a hygromycin resistance cassette. Transfected cells were selected with hygromycin for 1-2 weeks, individual colonies were picked and expanded, and GBS deletion was verified by genotyping PCR using primers flanking the expected deletion sites.

Quantification and Statistical Analysis:

Statistical details of experiments can be found in the figure legends, method details, and results sections. For RT-PCR data, t-tests were performed in Microsoft Excel using data from biological triplicates. For 4C-seq data, t-tests were performed in R using data from biological triplicates. For smFISH, t-tests were performed in R using data from six biological replicates for each condition.

Supplementary Material

3

Supplemental Table 1: Oligonucleotide Sequences used in this study, Related to STAR Methods section.

Key Resources Table

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
GR	Novus	NBP2-42221, RRID:AB_2894721
H3K27ac	Abcam	ab4729, RRID:AB_2118291
RNAP2	Active Motif	102660, RRID:AB_2732926
RAD21	Abcam	ab992, RRID:AB_2176601
SMC3	Abcam	ab9263, RRID:AB_307122
CTCF	Abcam	ab70303, RRID:AB_1209546
E2F1 (C20)	Santa Cruz	sc-193, RRID:AB_631394
E2F1	Millipore Sigma	05-379, RRID:AB_2096772
Chemicals, peptides, and recombinant proteins
16% Formaldehyde Solution	ThermoFisher Scientific	28906
CutSmart Buffer	NEB	B7204S
NlaIII	NEB	R0125L
Csp6I	ThermoFisher Scientific	ER0211
Ligation Buffer	NEB	B0202S
T4 DNA Ligase	NEB	M0202L
Proteinase K	ThermoFisher Scientific	25530-015
DpnII Buffer	NEB	B7006S
DpnII	NEB	R0543L
HALT Protease Inhibitors	ThermoFisher Scientific	78430
Dexamethasone	Millipore Sigma	D4902
Formamide (99.5%)	ThermoFisher Scientific	BP228-100
20x SSC	Quality Biological	351-003-131
TWEEN 20	Millipore Sigma	P7949
Heparin sodium salt from porcine intestinal mucosa	Millipore Sigma	H3149
Denhardt’s Solution 50x	Millipore Sigma	D2532
Dextran sulfate sodium salt from Leuconostoc spp.	Millipore Sigma	D8906
ProLong Gold Antifade Mountant with DAPI	ThermoFisher Scientific	P36935
18 mm #1.5 Coverslip	Electron Microscopy Sciences	72222-01
MEM	ThermoFisher Scientific	10370
Phenol Red-Free MEM	ThermoFisher Scientific	51200
GlutaMAX	ThermoFisher Scientific	3505
HEPES	Millipore Sigma	H0887
Penicillin/Streptomycin	Millipore Sigma	P0781
G418	ThermoFisher Scientific	10131-035
Charcoal/Dextran Treated FBS	R&D Systems	S11650
Protease Inhibitor Cocktail	ThermoScientific	78429
Micrococcal Nuclease (Mnase)	Worthington	LS004798
RNase A	ThermoFisher Scientific	12091-021
Critical commercial assays
NucleoMag PCR	Takara	744100
Expand Long Template PCR	Millipore Sigma	11681842001
AMPure XP beads	Beckman Coulter	A63881
SuperScript III First-strand kit	ThermoFisher Scientific	18080-051
ssoAdvanced Universal SYBR Green Supermix	Bio-Rad	172-5274
RNeasy Mini Kit	Qiagen	74104
QiaQuick PCR Purification Kit	Qiagen	28104
Dounce Homogenizer	Duran Wheaton Kimble	357542
Protein A/G Dynabeads	ThermoFisher Scientific	88802
Deposited data
Previously published ChIP-seq Data	(Hoffman et al., 2018)	GEO GSE112491
Previously published RNA-seq data	(Hoffman et al., 2020)	GEO GSE141834
ChIP-seq and 4C-seq	this study	GEO GSE173798
MCF7 E2F1 ChIP-seq	ENCODE	ENCSR000EWX
HeLa E2F1 ChIP	ENCODE	ENCSR000EVM
Experimental models: cell lines
A1-2	(Archer et al., 1994)	RRID:CVCL_0I95
A1-2 GBS1 eKO	this study	N/A
A1-2 GBS2 eKO	this study	N/A
A1-2 GBS3 eKO	this study	N/A
A1-2 GBS4 eKO	this study	N/A
Oligonucleotides
See supplemental table
Recombinant DNA
px330_GBS1_5p	this study	N/A
px330_GBS1_3p	this study	N/A
px330_GBS2_5p	this study	N/A
px330_GBS2_3p	this study	N/A
px330_GBS3_5p	this study	N/A
px330_GBS3_3p	this study	N/A
px330_GBS4_5p	this study	N/A
px330_GBS4_3p	this study	N/A
pGL3B	Promega	E1751
pGL3B_GBS1	this study	N/A
pGL3B_GBS2	this study	N/A
pGL3B_GBS3	this study	N/A
pGL3B_GBS4	this study	N/A
pGL3B_tk	this study	N/A
pGL3B_tk_GBS1	this study	N/A
pGL3B_tk_GBS2	this study	N/A
pGL3B_tk_GBS3	this study	N/A
pGL3B_tk_GBS4	this study	N/A
pGL3B_tk_3pGBS1	this study	N/A
pGL3B_tk_3pGBS2	this study	N/A
pGL3B_tk_3pGBS3	this study	N/A
pGL3B_tk_3pGBS4	this study	N/A
Software and algorithms
Cutadapt	(Martin, 2011)	RRID:SCR_011841
Bowtie2	(Langmead and Salzberg, 2012)	RRID:SCR_016368
Samtools	(Li et al., 2009)	RRID:SCR_002105
MACS2	(Zhang et al., 2008)	RRID:SCR_013291
Bedtools	(Quinlan and Hall, 2010)	RRID:SCR_006646
Deeptools	(Ramírez et al., 2016)	RRID:SCR_016366
ggplot2	(Wickham, 2016)	RRID:SCR_021139
IGV	(Robinson et al., 2011)	RRID:SCR_011793
Micro-Manager	(Edelstein et al., 2010)	RRID:SCR_000415
ROSE	(Lovén et al., 2013, Whyte et al., 2013)	RRID:SCR_017390
pipe4C	(Krijger et al., 2020)	N/A
CellProfiler	(Carpenter et al., 2006)	RRID:SCR_007358

Highlights:

• Dex treatment reorganizes the enhancer landscape in breast cancer cells

• A Dex-specific super enhancer encompasses DDIT4 and four GR binding sites

• Each binding site is uniquely required to promote or suppress DDIT4 expression

• Dex treatment triggers a loop-switching mechanism to induce DDIT4 transcription