Activation of Oncogenic Super-Enhancers Is Coupled with DNA Repair by RAD51

Summary

DNA double-strand breaks (DSBs) are deleterious and tumorigenic but could also be essential for DNA-based processes. Yet the landscape of physiological DSBs and their role and repair are still elusive. Here, we mapped DSBs at high resolution in cancer and non-tumorigenic cells and found a transcription-coupled repair mechanism at oncogenic super-enhancers. At these super-enhancers the transcription factor TEAD4, together with various transcription factors and co-factors, co-localizes with the repair factor RAD51 of the homologous recombination pathway. Depletion of TEAD4 or RAD51 increases DSBs at RAD51/TEAD4 common binding sites within super-enhancers and decreases expression of related genes, which are mostly oncogenes. Co-localization of RAD51 with transcription factors at super-enhancers occurs in various cell types, suggesting a broad phenomenon. Together, our findings uncover a coupling between transcription and repair mechanisms at oncogenic super-enhancers, to control the hyper-transcription of multiple cancer drivers.

Graphical Abstract

Highlights

• •Physiological DSBs are enriched at highly active oncogenic super-enhancers (SEs)
• •RAD51 co-localizes with transcription factors at SE in various cells
• •TOP1 mediates DSBs at SEs that are repaired by a RAD51-dependent mechanism
• •Depletion of RAD51 increases DSBs at SEs and decreases expression of related oncogenes.

By mapping physiological double-strand breaks (DSBs) of tumorigenic and non-tumorigenic cells Hazan et al. uncover a coupled transcription-repair mechanism at oncogenic super-enhancers in which RAD51 of the homologous recombination pathway plays a key role supporting the hyper-transcription of related oncogenes.

Introduction

Emerging evidence suggests that DNA double-strand breaks (DSBs), although considered the most deleterious type of DNA damage and the main source of cancer-associated lesions, are essential for DNA-based processes such as transcription, replication, and genome organization (Barlow et al., 2013, Canela et al., 2017, Ju et al., 2006, Madabhushi et al., 2015, Wei et al., 2016). However, the genomic landscape of physiological DSBs and their repair mechanisms remain unknown for two main reasons. First, the methods used to identify DSBs were indirect or insufficient to detect all DSBs, especially naturally occurring DSBs. Second, the well-characterized DNA damage response (DDR) signaling for DSBs is based mainly on studies involving external artificially induced DSBs (typically by ionizing radiation, chemotherapeutic drugs, or restriction enzymes) or internal DSBs induced by reactive oxygen species (reviewed by Jackson and Bartek, 2009). More recently, signal-induced DSBs, occurring upon transcription of early-response genes (following neuronal activation [Madabhushi et al., 2015], estrogen treatment [Ju et al., 2006], androgen induction [Haffner et al., 2010], serum induction [Bunch et al., 2015], and heat shock [Bunch et al., 2015]), have been revealed, adding to the complexity of mapping DSBs. Physiological DSBs are required for releasing the torsional tension that develops during transcription and replication, and at loop anchors involved in three-dimensional (3D) genome folding (Barlow et al., 2013, Canela et al., 2017, Canela et al., 2019, Gothe et al., 2019, Ju et al., 2006, Madabhushi et al., 2015, Wei et al., 2016). In particular, DSBs and subsequent DDR signaling were shown to promote effective RNA polymerase II (Pol II) pause release and transcriptional elongation for induction of early-response genes (Bunch et al., 2015). However, although DSBs were shown in signal-inducible models, the patterns, roles, and repair of internal homeostatic DSBs (termed physiological DSBs) in normal and cancer cells are largely unknown.

Here, we used the recently developed method BLISS (breaks labeling in situ and sequencing), which allows direct and quantitative profiling of DSBs in a genome-wide manner (Yan et al., 2017). BLISS directly detects DSBs at a resolution of a single nucleotide and has sensitivity that allows identifying not only artificially induced but also naturally occurring physiological DSBs (Yan et al., 2017). BLISS was shown to be an accurate and sensitive quantitative method, in particular because it quantifies DSBs through unique molecular identifiers (Yan et al., 2017, Iannelli et al., 2017). Yan et al. (2017) were able to assess the genome-wide off-target activity of two CRISPR-associated RNA-guided endonucleases, Cas9 and Cpf1, demonstrating that Cpf1 has higher specificity than Cas9. More recently, BLISS was used to map DSBs at sites involved in recurrent genome rearrangements and chromosomal translocations in cancer cells (Dellino et al., 2019, Gothe et al., 2019). Nevertheless, the landscape of DSBs across the genome and their repair in cancer cells is poorly characterized. We therefore set out to unbiasedly map naturally occurring physiological DSBs which we define as the “breakome,” using BLISS and to characterize its significance. Our analysis revealed that the breakome is cell-type specific. Further characterization of these DSBs uncovered their enrichment around regulatory elements, including promoters and super-enhancers, with the latter defined by extensive acetylation of histone H3 lysine 27 (H3K27ac; Whyte et al., 2013). Surprisingly, the sequences around the identified DSBs are highly enriched for TEAD transcription factor binding motifs. Sites bound by, for example, TEAD4, as revealed by chromatin immunoprecipitation assays, lack DSBs, suggesting efficient DNA repair at these sites. Indeed, TEAD4 binding overlaps with that of the repair factor RAD51 of the homologous recombination (HR) pathway mainly at super-enhancers. Depletion of either of these factors by small interfering RNA (siRNA) increases DSBs at RAD51/TEAD4 co-binding sites at super-enhancers and decreases the expression of related oncogenes. Together, our findings suggest an unforeseen coupling of RAD51 with the transcriptional machinery that is required for regulating the expression of oncogenic super-enhancers.

Results

The Breakome Is Cell-Type Specific

The extensively studied field of DSB repair mainly focuses on artificial or signal-induced DSBs while naturally occurring DSBs have remained relatively unidentified. To characterize these physiological DSBs, we set out to map the “breakome” in several cell types, in the absence of experimentally induced DNA damage, using the recently established method BLISS (Yan et al., 2017). BLISS was performed in duplicates in various human cell lines of different origins: MCF7 and MCF10A (epithelial breast cancer and pre-cancerous, respectively), BJ (fibroblast), and EndoC β-cells (endocrine). The genome-wide distribution of DSBs shows high similarities within each cell line, while dissimilarities among the different cell types are evident (Figure 1A). To verify these observations, we calculated cross correlations between these BLISS samples using bins of 100 kb (Figure S1A) and found highly significant (p < 1 × 10⁻¹⁰⁰) correlations within each cell type, further indicating that the pattern of DSBs along the genome is cell-type specific.

Figure 1.

Landscape and Characterization of the “Breakome”

(A) Genome-wide distribution of DSBs in cells from different lineages shows cell type-specific patterns. Two BLISS samples from each cell type are shown: MCF7 (epithelial breast cancer, in blue), MCF10A (non-tumorigenic breast epithelial, in light blue), BJ (fibroblast, in purple), and EndoC (endocrine, in orange).

(B) The distribution of DSBs in breast cancer MCF7 and pre-cancerous MCF10A cells along ChromHMM-defined chromatin states of HMEC. The bar height of each regulatory element is calculated as the ratio between observed and expected DSB counts. The figure shows DSB enrichment at insulators, strong enhancers, and active promoters. The dashed line marks a ratio of 1, and p values are indicated at the top of each bar.

(C) Enrichment of DSBs at MCF7-specific enhancers classified into super, clustered, and single enhancers. p values are indicated at the top of each bar.

(D) Treatment with Pol II inhibitor DRB for 30 min decreases DSBs around highly expressed genes (blue), while the effect on low-expressed genes is mild (green).

(E) E2-induced transcription at enhancers is associated with DSB induction. Gro-seq data from Hah et al. (2013) demonstrate increase in eRNA at E2-induced enhancers. Similarly, BLISS of cells grown on E2-free media for 2 days (−E2) and treated with 100 nM E2 for 1 h (+E2) show increase of DSBs upon E2 treatment relative to the genomic background.

MCF7 Breakome Reveals Clusters of Recurrent DSBs

Analysis of the MCF7 breakome revealed the existence of recurrent clusters of DSBs on chromosomes 3, 17, and 20 (Figure 1A). These clusters appear in four non-treated replicates and also after treatments such as ionizing radiation (IR) and estradiol (E2) treatment (Figures S1B and S1C). Notably, although the first two BLISS replicates were sequenced in higher coverage reads than all other BLISS samples (∼6 × 10⁷ and ∼4 × 10⁷ reads, respectively) the clusters exhibit similar patterns, implying that the regular coverage is close to saturation. The genomic regions of these clusters are amplified in MCF7 compared with the hg38 reference genome and are highly accessible as detected by the DNaseI hypersensitivity (DHS) assay (Figure S1B). In addition, the clusters overlap with regions of high contact rates, as determined from Hi-C chromosome contact maps (from Wang et al., 2013) (Figure S1B), which may lead to genomic instability (Kaiser and Semple, 2018). Of note, these clusters do not contain more repetitive and low-complexity sequences compared with random genomic regions. The presence of genes with various oncogenic activities within these clusters (BRIP1, RAD51C, TBX4, MIR21, NCOA3, PPM1E, TUBD1, and CYP24A1) suggests that their amplifications might have been positively selected during cancer development. These results suggest that BLISS can uncover recurrent DSBs that might be associated with chromosomal interactions and genomic instability.

DSBs Are Enriched at Regulatory Elements

To further characterize these physiological DSBs, we examined their distribution in non-tumorigenic MCF10A and tumorigenic MCF7 cells across the different chromatin states annotated by ChromHMM, which considers epigenomic information of histone modifications in a combinatorial manner (Ernst et al., 2011). We used the chromatin states as identified in breast epithelial HMEC cells (GSE38163) and found that the identified DSBs are spread along the genome, but their distribution relative to regulatory elements such as insulators, enhancers, and promoters is significantly higher than expected in both MCF10A and MCF7 cells (Figure 1B). Next, we examined the distribution of DSBs around MCF7-specific enhancers separated into three categories: (1) super-enhancers (n = 117, identified in dbSUPER; Khan and Zhang, 2016); (2) clustered enhancers (n = 716, identified as five or more consecutive enhancers within a 10 kb band); and (3) single enhancers (n = 28,871, identified as all other Encode-predicted enhancers). The ratio of observed versus expected DSB density is even higher for these MCF7-specific enhancers compared with the HMEC enhancers (Figure 1C) suggesting the formation of DSBs at regions with strong enhancer activity.

In order to verify that BLISS detects transcription-mediated DSBs, we treated the cells with the Pol II inhibitor DRB, which prevents entry into transcription elongation. Upon transcriptional inhibition with DRB, levels of DSBs decrease around highly expressed promoters, while lowly expressed promoters are less affected (Figure 1D, blue versus green, respectively). In addition, we used the well-characterized E2 induction system in which cells were grown on E2-free media for 2 days followed by 1 h treatment with 100 nM E2. Upon E2 treatment, enhancer RNAs (eRNAs) are transcribed from the E2-induced enhancers, in accordance with published Gro-seq data (Hah et al., 2013), and they exhibit increased DSB frequency at the same genomic locations as detected by BLISS (Figure 1E).

Together these results are consistent with previous studies showing enrichment of transcription-mediated DSBs (TM-DSBs) around promoters, enhancers (Yan et al., 2017, Iannelli et al., 2017, Madabhushi et al., 2015, Bunch et al., 2015, Wei et al., 2016) and the recently reported loop anchors (insulators) (Canela et al., 2017, Canela et al., 2019, Gothe et al., 2019, Yan et al., 2017).

TEAD Motif Is Highly Enriched around DSBs

The enrichment of DSBs around regulatory elements may suggest involvement of transcription factors in genome fragility. To explore this, we computationally searched the sequence around the identified DSBs for consensus patterns of known DNA-binding factors or de novo motifs (as defined using HOMER; Heinz et al., 2010). Our analysis revealed that the most abundant motif in the MCF7 breakome is for TEAD and its derivatives TEAD4 and TEAD2 (ATTCC/GGAAT), which are known to be involved in the Hippo signaling pathway that is involved in regulation of organ size and tumorigenesis (reviewed by Saucedo and Edgar, 2007) (Figure 2A). The TEAD motif is also consistently more abundant in the breakome of other cell types, including epithelial (MCF10A, EndoC), and mesenchymal (BJ), as well as in primary mouse embryonic stem cells and mouse neuronal stem cells (Figure 2A).

Figure 2.

The Breakome Is Enriched in TEAD Motifs, While Enhancers Are Enriched for AP-1 Motifs

(A) Top de novo and known motifs enriched in the proximity of DSBs across the whole genome of MCF7, MCF10A, BJ, and EndoC as well as in primary mouse embryonic and neuronal stem cells. p values and percentage of target sequences with the motif are presented.

(B) Motifs enriched in BLISS sequences at enhancers of MCF7 and MCF10A. p values were obtained using HOMER (Heinz et al., 2010).

In contrast to the dominant enrichment of the TEAD motif in proximity to DSBs within the whole genome, the BLISS breaks that mapped to enhancers of MCF7 are more frequently close to regions containing the palindromic motif (TGA G/C TCA) of FOS and JUN (Figure 2B). These factors belong to activator protein 1 (AP-1) complex and bind as heterodimers to regulate gene expression in response to a variety of stimuli. Within MCF7 enhancers, 20% of the BLISS breaks display proximity of TEAD and AP-1 motifs (Figure S2A), and in most of them the distance between the two motifs is <200 bases (Figure S2B), suggesting association between TEAD and AP-1 factors at MCF7 enhancers.

Similar de novo motif analysis of DSBs within enhancers in pre-cancer MCF10A cells shows no enrichment of the AP-1 motif, and among the known motifs, the AP-1 motif is lower than that of TEAD (Figure 2B), suggesting involvement of AP-1 factors in activation of enhancers that are associated with tumorigenic traits in cancer cells (MCF7), termed oncogenic enhancers.

Together, these results indicate that the tendency of DSBs to occur around the TEAD motif is a common phenomenon in many cell types that is conserved from mice to human. In contrast, DSBs at oncogenic enhancers tend to occur around AP-1 and TEAD motifs within close proximity, suggesting cooperation between TEAD and AP-1 factors at these sites.

TEAD4 Protects against DSBs at Enhancers

The enrichment of TEAD motifs around BLISS peaks suggests an involvement of TEAD transcription factor(s) in genome fragility. To further explore this, we compared the genome-wide distribution of TEAD binding (published chromatin immunoprecipitation sequencing [ChIP-seq] of TEAD4; ENCSR000BUO) to the MCF7 breakome. Separating the top 8,900 TEAD4 binding sites according to the presence of DSBs, as detected by BLISS, revealed that most of the TEAD4 binding sites are devoid of DSBs (Figure 3A; 4,900 TEAD4 binding sites without DSBs versus 4,000 with DSBs). Reciprocal sorting of DSBs by TEAD4 enrichment level showed that the majority of DSBs are not enriched for TEAD4 binding (Figure S3A; 10,000 TEAD4-depleted DSBs [hereafter −DSB] versus 6,000 enriched sites [+DSB]). To determine the role of TEAD4, we depleted its expression using siTEAD4 and compared the pattern of DSBs with a control sample (siCtrl, universal negative controls of independent RNAi pools that rule out off-targets), along TEAD4 binding sites separated into two groups of −/+ DSBs from the siCtrl sample. Interestingly, depletion of TEAD4 has no effect in the +DSBs group (Figure 3A, upper part, blue), while DSBs increase within the −DSBs group (Figure 3A, lower part, red). The localization of TEAD4 binding sites (−/+DSBs) across the 15-chromatin states (as defined by Ernst et al., 2011) indicated that the non-affected group (+DSBs) is enriched around active promoters, whereas the affected group (−DSBs) is enriched at enhancers (Figure 3B), implying that TEAD4 protects against DSBs specifically at the most active enhancers but not at promoters. Furthermore, high overlap between TEAD4 and AP-1 factors such as FOSL2, FOS, JUN, and JUND is observed and supports their mutual role of inducing oncogenic enhancers (Figure 3A). Together, these findings support the mutual role of TEAD and AP-1 factors, which together with YAP/TAZ were previously shown to induce enhancer activity driving oncogenic growth (Zanconato et al., 2015).

Figure 3.

TEAD4 Protects against DSBs Mainly at Enhancers

(A) TEAD4 binding sites, plotted together with AP-1 factors JUN, JUND, FOS, and FOSL2, were classified into two groups: a DSB-enriched group (+DSB) and a DSB-deprived group (−DSBs). Data from siCtrl and siTEAD4 on the right-hand side show that TEAD4 depletion by siRNA does not affect the fragility of the +DSB group (blue), while breaks are increased in the −DSB group (red).

(B) The distribution of TEAD4 binding sites from −/+DSB groups across the chromatin states demonstrating that the −DSB group (red bars) is mainly enriched at strong enhancers, while the +DSB group (blue bars) is largely enriched at active promoters. The siTEAD4 experiment was repeated twice, and the analysis showed similar results.

TEAD4 Co-localizes with the Repair Factor RAD51, Especially at Super-enhancers

Protection against DSBs at TEAD4 binding sites within strong enhancers suggests a collaboration with repair machinery. Previous observations demonstrated that artificially induced DSBs at transcriptionally active regions are preferentially repaired through RAD51-dependent mechanisms (Aymard et al., 2014). In order to check involvement of such mechanisms, we examined ChIP-seq data of RAD51 (GSE105597). The distribution of the top RAD51 binding sites across the chromatin states demonstrates significant enrichment around strong enhancers, active promoters, and insulators (Figure S4A). To reveal the role of RAD51 at these three regulatory element types, we depleted its expression using siRNA (two independent pools) and compared the pattern of DSBs around RAD51 binding sites with siCtrl. Interestingly, at strong enhancers and insulators normally bound by RAD51, DSBs increase upon depletion of RAD51, implying a loss of repair activity, whereas at promoters there is a depletion of DSBs around RAD51 binding sites, in both siRAD51 and siCtrl, despite the stronger enrichment of RAD51 in untreated cells (mean scale = 25 compared with 3 at the other two elements) (Figure S4B).

Next, we examined the genome-wide distribution of RAD51 relative to TEAD4. Although only 7.6% of RAD51 binding sites overlap with TEAD4 (Figure 4A), the distribution of these RAD51/TEAD4 common sites across chromatin states shows high enrichment at strong enhancers (Figure 4B; observed/expected ratio = 43.4), and further classification of MCF7 enhancers showed that these common sites appear 30 times more than expected at super-enhancers (Figure 4C). These RAD51/TEAD4 common sites overlap with AP-1 factors and co-factors such as P300, BRD4, and MED1 within super-enhancers that are regulating the expression of oncogenes such as GATA3, XBP1, TFF1, and GREB1 (Figure 4D; Figure S3B). Furthermore, these enhancers are specific for MCF7 and absent in pre-cancerous HMEC cells, suggesting that they were acquired de novo during cellular transformation. Interestingly, while MCF7 has only 117 super-enhancers, relative to 1,098 super-enhancers in normal mammary epithelial HMEC cells, their effect on transcription of related genes, assessed by RPKM (reads per kilobase of transcript per million mapped reads) counts, is much more prominent (Figure S5A), suggesting that the oncogenic super-enhancers have a stronger effect on the expression of their associated genes compared with their normal or pre-malignant counterparts. Together, the overlap between RAD51 and TEAD4 at oncogenic super-enhancers suggests that collaboration between repair and transcription induction is required to support this high transcriptional activity of the related oncogenes.

Figure 4.

TEAD4/RAD51 Co-localize at Super-enhancers of Oncogenes

(A) Venn diagram of RAD51 and TEAD4 binding sites showing some overlap. Peaks were obtained from RAD51 and TEAD4 ChIP-seq experiments (from ENCODE: ENCSR442VBJ and ENCSR000BUO, respectively).

(B) The distribution of TEAD4/RAD51 common sites across chromatin states of HMEC shows enrichment mainly around strong enhancers. p values are marked at the top of each bar.

(C) The enrichment of TEAD4/RAD51 common sites at the three types of MCF7-specific enhancers showing high enrichment at super-enhancers.

(D) Several examples for oncogenes regulated by super-enhancers containing overlapped sites of TEAD4, RAD51, JUN, JUND, FOS, and FOSL2. Also shown are the marker of active enhancers, H3K27Ac, Groseq data of both strands, and super-enhancer regions in MCF7 (MCF7 SE trace). Scale bars: gray, 10 kb; black, 50 kb.

RAD51 and TEAD4 Are Involved in Repair and Transcription at Super-enhancers

Given that depletion of TEAD4 or RAD51 increased DSBs around their binding sites within strong enhancers (Figures 3 and S4, respectively), we next asked how depletion of these factors would affect DSBs at RAD51/TEAD4 common sites compared with their unique sites within enhancers. Separating common and unique RAD51 binding sites at single, clustered, and super-enhancers (Figure 5A) revealed that depletion of RAD51 increases DSBs at RAD51/TEAD4 common sites within super-enhancers, whereas clustered and single enhancers are less affected (Figure 5B). Using additional siRNA of RAD51 (SMARTpool, which is a pool of siRNAs designed and modified for greater specificity and reduced off-targets) resulted in similar increase of DSBs at RAD51/TEAD4 common sites within super-enhancers (Figure S6A).

Figure 5.

TEAD4 and RAD51 at Super-enhancers Co-regulate Induction and Repair of DSBs

(A) RAD51 binding sites at super, clustered, and single enhancers grouped into two clusters: RAD51/TEAD4 common sites and RAD51 unique binding sites. The graphs in the upper panel show the accumulation of RAD51 and TEAD4 in the common cluster (blue trace) and in the unique cluster (light green trace).

(B and C) The distribution of DSBs from siCtrl, siRAD51, and siXRCC4 BLISS samples (B) and from siCtrl and siTEAD4 BLISS samples (C) across the RAD51/TEAD4 common sites cluster (blue trace) and RAD51 unique cluster (light green trace) for each of the three types of enhancers. The red arrows mark the increase of DSBs in siRAD51 and siTEAD4 specifically at RAD51/TEAD4 common sites within super-enhancers. To verify these results, we repeated depletion of RAD51 and TEAD4 twice, and the analysis showed similar increase in DSBs in RAD51/TEAD4 common sites within super-enhancers.

(D and E) qRT-PCR demonstrates downregulation of super-enhancer-regulated oncogenes upon depletion of RAD51 (D) and TEAD4 (E) relative to siCtrl samples.

(F and G) Inhibition of RAD51 or TEAD4 by B02 or super-TDU, respectively, shows similar decrease of gene expression (F) as well as their eRNAs (G). The qRT-PCR values in (D)–(G) represent mean ± SD from three independent biological samples with technical triplicates.

(H) Total RNA-seq around RAD51/TEAD4 common (blue) and RAD51 unique (green) sites within super-enhancers demonstrates downregulation of eRNAs around the common sites upon B02 treatment.

To exclude the possibility that the repair at RAD51/TEAD4 common sites within super-enhancers is through non-homologous end joining (NHEJ), we depleted one of its key factors, XRCC4. Indeed, knockdown of XRCC4 does not change the pattern of DSBs at these sites (Figure 5B; Figure S6B). Furthermore, depletion of BRCA1, another factor of the HR pathway, has a similar effect to RAD51, showing more DSBs at RAD51/TEAD4 common sites within super-enhancers (Figure S6C). Together, these results support repair of physiological DSBs at super-enhancers through a RAD51-dependent mechanism.

In addition, further analysis of the BLISS data from siRAD51 and siCtrl samples on the basis of levels of the enhancer marker H3K27ac, rather than using previously defined super-enhancers, revealed a similar increase of DSBs around TEAD4/RAD51 common sites at strong enhancers (Figure S6B).

Interestingly, although RAD51 binding is significantly higher in the unique compared with the common sites (p = 3.9 × 10⁻²; Figure 5A, light green versus blue graphs), the pattern of DSBs around these unique sites does not change in the siRAD51-treated cells, indicating that RAD51-mediated repair is not correlated with the level of RAD51 binding but rather occurs at sites shared with TEAD4. Indeed, depletion of TEAD4 increases DSBs at TEAD4/RAD51 common sites at enhancers (Figure 5C).

In order to validate the association between super-enhancer activity and DSBs, we suppressed their activity by using the BRD4 inhibitor JQ1 (Lovén et al., 2013). Treating the cells with 1 μM JQ1 for 4 h caused a 15% decrease of DSBs at super-enhancers compared with mock-treated cells (Figure S5B). The specificity for super-enhancers, as calculated by comparing the 117 super-enhancers with 100 random sets of regions with the same widths, is significant (p = 1.1 × 10⁻¹²). Thus, the regulatory activity at super-enhancers leads to enrichment of DSBs that are repaired through a RAD51-dependent mechanism.

To examine how RAD51 and TEAD4 affect expression of super-enhancer regulated genes, we performed qRT-PCR on RAD51- and TEAD4-depleted cells and found downregulation of target genes and eRNAs (Figures 5D and 5E; Figure S7). Similarly, pharmacological inhibition of RAD51 and TEAD4 (by B02 and super-TDU, respectively) downregulates super-enhancer genes and their eRNAs, although the effect of RAD51 inhibition is more significant (Figures 5F and 5G). To verify the effect of RAD51 inhibition, we performed total RNA sequencing (RNA-seq), which detects coding as well as multiple forms of noncoding RNA, including eRNAs. Upon RAD51 inhibition, eRNAs around RAD51/TEAD4 sites within super-enhancers are significantly downregulated (Figure 5H), suggesting reduced activity. Indeed, 17 super-enhancer genes are significantly downregulated compared with 100 random sets of genes (Figure S7D).

These findings demonstrate that RAD51 has a role in supporting the activity of super-enhancers. Given that at these sites, RAD51 is involved in repair of DSBs, we propose that the coupling between transcription and repair at super-enhancers is required for the high expression levels of related oncogenes.

TOP1 Is Involved in RAD51-Associated DSBs

The topoisomerases TOP1 and TOP2 cooperate to optimize transcription through induction of single-strand breaks and double-strand breaks, respectively (Pommier et al., 2016, Teves and Henikoff, 2014, King et al., 2013, Husain et al., 2016). Several reports have shown that TOP2 isoforms are associated with DSBs and transcriptional activity (Ju et al., 2006, Madabhushi et al., 2015, Bunch et al., 2015, Canela et al., 2019). To examine whether TOP2 mediates the DSBs at RAD51-associated super-enhancers, we initially examined TOP2 binding (ChIP-seq from Manville et al., 2015) and found, as expected, enrichment at promoter regions, whereas no overlapping with that of RAD51 at strong enhancers or insulators was observed (Figure 6A). We next performed BLISS following treatment with the TOP2 inhibitor etoposide (ETO). ETO inhibits the ligation step catalyzed by TOP2 and has been shown to modulate transcription (reviewed by Bouwman and Crosetto, 2018). Treatment with ETO showed involvement at promoters, while the pattern of DSBs does not change around RAD51 sites in strong enhancer regions or insulators (Figure 6A). Altogether, these findings suggest that TOP2 activity is less involved in RAD51-associated DSBs at strong enhancers.

Figure 6.

TOP1, but Not TOP2, Is Involved in RAD51-Associated DSBs

(A) TOP2 binding (from Manville et al., 2015) and DSBs after ETO treatment (5 μM for 4 h) show no correlation with RAD51 binding sites at strong enhancers.

(B) TOP1 ChIP-seq compared with RAD51 binding sites at strong enhancers, insulators, and promoters showing high overlap at strong enhancers.

(C) TOP1 is enriched at DSBs in enhancers. DSBs are compared between 1,000 enhancers enriched with TOP1 and 1,000 enhancers exhibiting low TOP1 coverage (blue and green graphs, respectively). Rich TOP1 regions are correlated with more DSBs.

(D) qPCR on genomic DNA isolated from cells treated with DMSO (mock), 5 μM B02, and 10 μM CPT for 4 h using primers designed for the sites shown in Figure S6D. Decrease in PCR products indicates increase of DSBs, which prevents amplification. Site with no recurrent DSBs was used for normalization. Values represent mean ± SD from three independent biological samples with technical triplicates.

Next, we set to determine whether TOP1 might be involved in generation of these DSBs. TOP1 was shown to be involved in activation of mTEC super-enhancers and recruitment of DSB repair factors, suggesting that TOP1-induced single-strand breaks (SSBs) may develop into DSBs (Bansal et al., 2017, Abramson et al., 2010). To examine whether TOP1 is involved in RAD51-associated DSBs, we first determined its genome-wide distribution using ChIP-seq. Comparing TOP1 and RAD51 binding sites at strong enhancers and promoters shows high overlap in MCF7 cells (Figure 6B). We next compared DSBs on enhancers rich with TOP1 versus enhancers deprived of TOP1 and found that regions enriched with TOP1 binding display more BLISS signal (Figure 6C). Furthermore, pharmacological inhibition of RAD51 (B02) or TOP1 (camptothecin [CPT], which prevents DNA re-ligation) increases DSBs as detected by qPCR on genomic DNA isolated from cells treated with DMSO (mock), B02, or CPT (Figure 6D). The primers were designed for regions with more recurrent DSBs in siRAD51 compared with the siCtrl (Figure S6D). A decrease in PCR products upon treatment with RAD51 or TOP1 inhibitors suggests an increase of DSBs, which prevents amplification (Figure 6D). Together, these findings imply that TOP2 is unlikely to be involved in RAD51-associated DSBs and that these DSBs are likely developed from SSBs induced by TOP1.

A Signature of Luminal-Specific Super-enhancers Is Regulated by RAD51/TEAD4

Super-enhancers differ from typical enhancers in size, transcription factor density, and ability to induce transcription (Hnisz et al., 2013). However, it is controversial whether they also dictate cellular identity through regulation of key lineage-specific genes (Pott and Lieb, 2015). To address this question regarding the super-enhancers in MCF7 cells, which represent the luminal A breast cancer subtype, we compared expression profiles from a cohort of 51 breast cancer patients (Gruosso et al., 2016). We defined a subset of 21 genes regulated by super-enhancers with high levels of TEAD4/RAD51 and Gro-seq (GSE94479 from Liu et al., 2017; Table S1). We found that this subset of genes is expressed at higher levels in subtypes luminal A and B compared with normal, HER2 and basal subtypes (Figure S8A; p = 6.7 × 10⁻⁷). As a control, the expression levels of housekeeping genes were relatively constant across the different subtypes (Figure S8B). Moreover, lower expression of housekeeping genes compared with super-enhancer-regulated genes demonstrates the exceptionally high transcription levels of these super-enhancer-regulated genes, which is presumably established through, among others, the collaboration between TEAD4 and RAD51.

To further characterize these 21 genes, we used a recently published systematic screening of driver genes required for tumor survival in ∼500 cancer cell lines, including MCF7 (Tsherniak et al., 2017). According to this screening, 8 of the 21 genes are MCF7-dependent oncogenes (dependency score in minus direction; Table S1). In addition, many of the other genes were previously demonstrated to have oncogenic functions, including FOS, XBP1, TRIM37, SULF2, PARD6B, and ZDHHC7, or to be associated with cancer: BCAS4 and MLPH (Table S1 and references therein). It should be noted that 16 of these 21 super-enhancers of MCF7 are not active in pre-cancerous HMEC cells, suggesting that these oncogenic super-enhancers were generated during tumorigenesis and likely supported that process. Our results suggest that this signature of 21 genes, regulated by RAD51/TEAD4 at the super-enhancers of luminal MCF7, may define key luminal oncogenes in human specimens.

Given that many super-enhancer-regulated genes are oncogenes, we examined whether inhibition of RAD51 alters tumorigenesis. First, we performed a colony formation assay and found that RAD51 inhibition dramatically decreases both the number and size of the colonies (Figure S9A). Next, we checked for apoptosis and found that the tumorigenic MCF7 are more sensitive to RAD51 inhibition compared with non-tumorigenic MCF10A. Thus, inhibiting RAD51 induces loss of tumorigenesis.

RAD51 Overlaps with Transcription Factors at Super-enhancers of Various Cell Types

To examine whether the overlap between RAD51 and TEAD4 (or other TFs) at super-enhancers occurs also in other cell types, we compared RAD51 and TEAD4 distribution in K562 (chronic myelogenous leukemia cells) and HepG2 (hepatocellular carcinoma cells). Whereas in MCF7 and K562 cells, the vast majority of super-enhancers have TEAD4/RAD51 common peaks (89% and 79%, respectively), in HepG2 cells they have only 19% common peaks (Figure 7A). In addition, RAD51 but not TEAD4 ChIP-seq data were available for B lymphocytes GM12878. Therefore, we compared RAD51 binding sites with another transcription factor, JUNB, of the AP-1 complex and found that they co-localize at 94% of the super-enhancers (Figure 7B). Interestingly, the super-enhancer-regulated genes express higher in the three cell types with RAD51 in the majority of the super-enhancers compared with HepG2 (Figure 7C), further suggesting that RAD51 supports the high transcription levels. Together, these findings suggest that the collaboration between RAD51 and transcription factors at super-enhancers is not exclusive for MCF7 breast cancer but rather is a broad phenomenon.

Figure 7.

RAD51/TEAD4 Coupling Regulates Expression of Oncogenes

(A) Pie charts of RAD51 and TEAD4 unique and common binding sites within super-enhancers of tumorigenic MCF7 breast cancer, K562 chronic myelogenous leukemia, and HepG2 hepatocellular carcinoma.

(B) RAD51 co-localizes with JUNB in most super-enhancers of GM12878 B-lymphocytes.

(C) The expression of super-enhancer-regulated genes in MCF7, K562, and GM12878 is significantly higher compared with HepG2, while the overall expression level is similar.

(D) Model: at super-enhancers, TEAD4 together with YAP1 and JUN/FOS (AP1 complex) induce transcription of related oncogenes. The high transcriptional activity is accompanied by DSBs, which are continuously being repaired in a RAD51-dependent mechanism. Depletion of RAD51 or TEAD4 results in an increase of DSBs specifically at TEAD4/RAD51 common sites and consequently a decrease in expression of related oncogenes.

Discussion

Using genome-wide mapping of physiological DSBs in cancer and normal cells, we uncovered a mechanism of DNA repair coupled with transcription at oncogenic super-enhancers. This mechanism involves high transcriptional activity mediated by TEAD and AP-1 factors P300, BRD4, and MED1 combined with repair through RAD51. Upon depletion of RAD51, DSBs increase around TEAD4/RAD51 sites at super-enhancers, leading to downregulation of the associated oncogenes (Figure 7D). Therefore, we propose a model in which the extensive transcription and regulatory activity at oncogenic super-enhancers is accompanied by DSBs, likely mediated by TOP1, that are constantly challenging genomic integrity. The coupling with RAD51 provides an efficient and accurate repair leading to tumor propagation advantage.

Induction and Repair of Transcriptional-Mediated DSBs (TM-DSBs)

Our findings that DSBs associate with transcription are consistent with previously published data showing essential roles for TM-DSBs in transcriptional activation (Canela et al., 2017, Ju et al., 2006, Madabhushi et al., 2015, Wei et al., 2016). Both TOP1 and TOP2 were shown to be important for gene transcription through cleavage and resealing of one or two strands (reviewed by Pommier et al., 2016). Previously, TOP1 was shown to be involved in eRNA transcription and nucleosome depletion at enhancers, and TOP1-induced breaks can mobilize the DDR (Puc et al., 2015). Bansal, Mathis, and colleagues showed that TOP1 is involved in activation of super-enhancers of mTEC cells, where it associates with the DSB marker γH2AX and DSB repair factors. The authors proposed that TOP1-induced SSBs might develop into DSBs that promote DDR signaling (Abramson et al., 2010, Bansal et al., 2017). Our findings that enhancers enriched with TOP1 binding display more BLISS signal and that TOP1 and RAD51 co-localize strongly support that TOP1 is involved with RAD51-associated DSBs at super-enhancers, while TOP2 is less involved at these sites (Figure 6). TOP1 may release the torsional tension of eRNA transcription by inducing SSBs, and when these are developed into DSBs, RAD51 participates in their repair. In addition, TOP1 may resolve R-loops generated during eRNA transcription, which was previously shown to induce DSBs (Aguilera and García-Muse, 2012). Detailed analysis and mechanistic insights into the precise action of TOP1/TOP2 in regulating super-enhancers’ associated breaks will be further investigated in future work.

The repair of these TM-DSBs in normal adult cells, which are primarily post-mitotic, is likely through NHEJ, because HR requires transition through S phase in order to use the other chromatid as a template. In support, TM-DSBs in early-response genes of post-mitotic neuronal cells were shown to be repaired through NHEJ (Madabhushi et al., 2015). Likewise, TM-DSBs at super-enhancers of mTECs were shown to recruit NHEJ factors such as DNA-PKs, Ku80, and PARP-1 (Bansal et al., 2017). In contrast, proliferating cancer cells could use the HR machinery to ensure error-free repair. Indeed, re-analysis of Gro-seq data (GSE94479 from Liu et al., 2017) demonstrates that super-enhancer-regulated genes are expressed throughout the cell-cycle phases (data not shown).

Our study mapping physiological DSBs of proliferating cancer cells demonstrates that depletion of RAD51 and BRCA1, but not XRCC4, increases DSBs at RAD51/TEAD4 common sites within super-enhancers, suggesting HR-mediated repair at these sites. Notably, these findings do not exclude the possibility of repair through XRCC4/NHEJ at other regions. Therefore, although RAD51/HR are involved in accurate repair and thus protect the genome against mutations and alterations, increased levels of RAD51 and its repair activity might overcome cancer-related DNA damage, associated with uncontrolled proliferation, transcription, and metabolism, or in resistance to chemotherapy and other DNA damage-inducing drugs. Indeed, RAD51 levels were shown to increase in many types of cancer, and its overexpression was implicated in drug resistance (Raderschall et al., 2002, Richardson, 2005, Henning and Stürzbecher, 2003, Klein, 2008).

Normal versus Oncogenic Super-enhancers

The definition of super-enhancers as long chromatin stretches overloaded with H3K27ac and H3K4me1, proposed to act as platforms hosting a high density of tissue-specific and common transcription factors (Hnisz et al., 2013, Whyte et al., 2013), associates them with hyper-transcription. In normal cells, super-enhancers regulate cell identity and control the activity of post-mitotic differentiated cells following environmental or physiological stimulation. In contrast, in cancer cells, which have an altered pattern of gene expression, the super-enhancers regulate genes with oncogenic functions (Lovén et al., 2013, Parker et al., 2013, Hnisz et al., 2013). From a clinical perspective, expression data from breast cancer patients show that a signature set of luminal-specific oncogenes is controlled by super-enhancers, supporting the concept that super-enhancers play a role in defining tumorigenic cell identity (Figure S8). Super-enhancers are acquired during tumorigenesis through mutations (Mansour et al., 2014), amplifications (Zhang et al., 2016), or translocations (Drier et al., 2016, Hnisz et al., 2013). A recent study has indeed demonstrated that stratifying tumors by their enhancer landscape can reveal new epigenetic subtypes with distinct clinical course and outcome (Cejas et al., 2019). Our finding that luminal super-enhancers are regulated by RAD51/TEAD4 implies that during transformation, tight cooperation between transcription and repair machinery supports activation of oncogenic super-enhancers. Of note, many of the genes regulated by these luminal super-enhancers are common for various cancer types (e.g., MYC, XBP1, GATA3, FOS, HES1), and therefore depletion of RAD51 and TEAD4 is expected to affect these targets not only in luminal cells. Future work will assess cell type-specific signatures of other breast cancer subtypes and other cancer types.

The effect of the 117 super-enhancers of MCF7 on transcription is potent compared with the 1,098 super-enhancers of pre-cancerous HMEC cells (Figure S5A), which suggests that the higher transcriptional activity may be supported by repair through RAD51. Most important, the correlation between the presence of RAD51 and high transcription of their genes (Figure 7C) suggests that RAD51 supports the hyper-transcription in various cell types.

Collaboration of TEAD/YAP/TAZ with AP-1 Complex Drives Oncogenic Enhancers

The observation that TEAD4, and likely other transcription factors, regulates the expression and integrity of super-enhancers and associated genes is another interesting finding of our study. Unlike RAD51, TEAD4 depletion increase DSBs at various types of enhancers, making it less specific for super-enhancers. At present, the detailed mechanism of TEAD4 involvement and whether it is upstream or downstream of TOP1 is not yet known and will be the subject of future studies.

The family of TEAD transcription factors (1–4) together with YAP1 belongs to the Hippo signaling pathway, which controls organ size through restraining proliferation and promoting apoptosis (Salah and Aqeilan, 2011). Tissue overgrowth and tumor formation are suppressed by limiting the oncogenic activity of YAP (reviewed by Saucedo and Edgar, 2007). The YAP/TEAD complex together with the AP1 heterodimer, which contains JUN/FOS proteins, was shown to cooperate on cis-regulatory regions to regulate migration and invasion in a diverse range of cancer cells (Lin et al., 2017, Liu et al., 2016, Wu et al., 2008). Galli et al. (2015) demonstrated that binding of YAP1 to TEAD4 at super-enhancers of liver cancer releases paused Pol II from promoters toward elongation (Galli et al., 2015). In breast cancer models, the collaboration between TEAD/YAP/TAZ and AP-1 factors such as FOS/JUN was shown to drive oncogenic enhancers and inhibition of TEAD4/YAP/TAZ or AP-1 suppressed oncogenic features (Zanconato et al., 2015, Lu et al., 2005).

Importantly, our findings showing increased density of DSBs and decreased expression of eRNAs at RAD51/TEAD4 common sites in RAD51-deficient cells indicate that RAD51 supports super-enhancer activity. Overall, the coupling between induction and repair of DSBs is required for the high expression of related oncogenes.

Given that inhibition of RAD51 or TEAD restricts tumor growth, it is plausible to assume that inhibiting both will impair transcription and repair at super-enhancers. Development of specific inhibitors targeting TEAD4/RAD51 collaboration might be a promising strategy to abolish the activity of super-enhancers in cancer cells.

STAR★Methods

Key Resources Table

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies, Chemicals, Peptides, and Recombinant Proteins
TOP1 Ab	Abcam	Cat# ab3825
TOP1 Ab	Abcam	Cat# ab85038
HaeIII	New England Biolabs	Cat# R0108S
T4 DNA Ligase	New England Biolabs	Cat# M0202M
Truncated RNA Ligase	New England Biolabs	Cat# M0242S
T4 Polynucleotide Kinase	New England Biolabs	Cat# M0201S
N-TERTM Nanoparticle siRNA Transfection	Sigma	Cat# N2913
Lipofectamine 2000 Reagent	Thermo Fisher Scientific	Cat# 11668-027
RAD51 inhibitor B02	Sigma	Cat# SML0364
TOP1 inhibitor Camptothecin	Sigma	Cat# C9911
TEAD inhibitor Super-TDU	Selleckchem	Cat# S8554
RNApolII inhibitor DRB	Sigma	Cat# D1916
Dynabeads Protein G	ThermoFisher	Cat# 10004D
Formaldehyde	ThermoFisher	Cat# 28906
Critical Commercial Assays
NEBNext PCR Master Mix	New England Biolabs	Cat# M0541S
Quick Blunting Kit	New England Biolabs	Cat# M1201A
CutSmatr buffer	New England Biolabs	Cat# B7204
RNeasy Minelute	Qiagen	Cat# 74204
MegaScript T7 Transcription Kit	ThermoFisher scientific	Cat# AM1333
Agencourt AMPure XP Beads	Beckman coulter life sciences	Cat# A63880
SuperScript III First-Strand Synthesis	Thermo Fisher Scientific	Cat# 18080051
Quick-RNA miniprep	Zymo Research	Cat# R1057
Cell lines
MCF7	ATCC	HTB-22
MCF10A	ATCC	CRL-10317
siRNA
siRAD51	Sigma	Cat# EHU045521
siRAD51	Dharmacon	Cat# L-003530-00-0005
siXRCC4	Sigma	Cat# EHU091461
siTEAD4	Sigma	NA
siBRCA1	Dharmacon	Cat# L-003461-00-0005
siCtrl (universal negative control)	Sigma	Cat# SIC001
Deposited Data
RNAseq of MCF7	(Jin et al., 2013)	GEO: GSM1154039
RNAseq of MCF10A	(Kang et al., 2013)	GEO: GSE45258
Gro-seq of MCF7 -/+E2	(Hah et al., 2013)	GEO: GSE43836
Gro-seq of MCF7 G0/G1	(Liu et al., 2017)	GEO: GSE94479
Exp profiles BC Samples	(Gruosso et al., 2016)	GEO: GSE45827
Chromatin states of HMEC	Bernstein lab, Broad	GEO: GSM936084
DNaseI hypersensitivity assay in MCF7	Crawford lab, Duke	GEO: GSM816627
Hi-C in MCF7	(Wang et al., 2013)	GEO: GSM1154027
POLR2A ChIP-seq in MCF7	Lyer lab, UTA	GEO:GSM822295
TEAD4 ChIP-seq in MCF7	Myers lab, HAIB	GEO:GSM1010860
RAD51 ChIP-seq in MCF7	Snyder lab, Stanford	GEO:GSE105597
CTCF ChIP-seq in MCF7	Bernstein lab, Broad	ENCSR560BUE
H3K27ac ChIP-seq in MCF7	Bernstein lab, Broad	GEO:GSE96352
H3K4me1 ChIP-seq in MCF7	Bernstein lab, Broad	GEO:GSE86714
JUN ChIP-seq in MCF7	Snyder lab, Stanford	GEO:GSE91550
JUND ChIP-seq in MCF7	Myers lab, HAIB	GEO:GSM1010892
FOSL2 ChIP-seq in MCF7	Myers lab, HAIB	GEO:GSM1010768
FOS ChIP-seq in MCF7	Snyder lab, Stanford	GEO:GSE105734
BRD4 ChIP-seq in MCF7	(Nagarajan et al., 2014)	GEO:GSE55921
MED1 ChIP-seq in MCF7	(Liu et al., 2014)	GEO:GSM1469999
EP300 ChIP-seq in MCF7	Myers lab, HAIB	GEO:GSM1010800
TOP2 ChIPseq in MCF7	(Manville et al., 2015)	GEO:GSE66753
DRIP-seq in MCF7	(Stork et al., 2016)	GEO:GSE81851
TEAD4 Chip-seq in K562	Myers lab, HAIB	GEO:GSM1010895
RAD51 Chip-seq in K562	Snyder lab, Stanford	GEO:GSE91838
TEAD4 ChIP-seq in HepG2	Myers lab, HAIB	GEO:GSM1010875
RAD51 ChIP-seq in HepG2	Snyder lab, Stanford	GEO:GSE91460
RAD51 ChIP-seq in GM12878	Snyder lab, Stanford	GEO:GSE105699
JUNB ChIP-seq in GM12878	Myers lab, HAIB	GEO:GSE96455
RNA-seq in GM12878	Graveley lab, Uconn	GEO:GSE78552
Software and Algorithms
ImageJ	(Schneider et al., 2012)	https://imagej.nih.gov/ij/
Prism 6	GraphPad	https://www.graphpad.com/scientific-software/prism/
R version ver. 3.4.3	(R Core Team, 2017)	https://www.R-project.org/
Python Language Reference, ver. 2.7	Python Software Foundation.	http://www.python.org/
Bowtie2 ver. 2.3.4.2	Johns Hopkins University	http://bowtie-bio.sourceforge.net/bowtie2/
Homer findMotifs and annotatePeaks	(Heinz et al., 2010)	http://homer.ucsd.edu/homer/
Deeptools computeMatrix, plotHeatmap and plotProfile ver 3.1.3	(Ramírez et al., 2016)	https://deeptools.readthedocs.io/en/develop/
MACS2 1.4.3	(Zhang et al., 2008)	https://github.com/taoliu/MACS/wiki
samtools	(Li et al., 2009)	http://www.htslib.org/download/
Bedtools V2.26.0	(Quinlan and Hall, 2010)	https://bedtools.readthedocs.io/en/latest/
IGV2.3	(Robinson et al., 2011)	http://software.broadinstitute.org/software/igv/

Lead Contact and Materials Availability

Further information and requests for resources should be directed to and will be fulfilled by the Lead Contact, Dr. Rami Aqeilan (ramiaq@mail.huji.ac.il).

This study did not generate new unique reagents.

Experimental Model and Subject Details

Human Cell Lines

MCF7 were grown in RPMI supplemented with 10% (vol/vol) FBS (GIBCO), glutamine, and penicillin/streptomycin (Beit-Haemek) whereas MCF10A were grown on DMEM/F12 supplemented with 5% Horse serum, 20 ng/ml EGF, 0.5 mg/ml Hydrocortisone, 100 ng/ml Cholera toxin, 10 μg/ml Insulin and Pen/Strep. Cells were grown at 37°c under a humidified atmosphere with 5% CO₂. Cells were routinely authenticated by STR profiling, tested for mycoplasma, and cell aliquots from early passages were used.

Method Details

Transient Transfection

Transient transfections of siCtrl, siRAD51, siXRCC4 and siTEAD4 were achieved using N-TER™ Nanoparticle siRNA Transfection System (Sigma, N2913) while siBRCA1 (from Dharmacon; L-003461-00-0005) was transfected with lipofectamin (from ThermoFisher) according to the manufacturers.

RNA Extraction, RT-PCR, and Real-Time PCR

Total RNA was prepared using TRIreagent (Sigma Aldrich) as described by the manufacturer. One microgram RNA was used for cDNA synthesis using the First-Strand cDNA Synthesis Kit (Bio-Rad). Quantitative real-time PCR was performed using Power SYBR Green PCR Master Mix (Applied Biosystems). All measurements were performed in triplicate and standardized to the levels of HPRT.

Validation of DSBs using PCR

Quantitative PCR on genomic DNA isolated from cells treated with DMSO (mock), RAD51 inhibitor (B02; 5 μM) and TOP1 inhibitor (CPT; 10 μM) was performed using primers designed for selected sites [TRIM37, BRIP1, Chr20 HS #1, Chr20 HS #2]. Decrease in PCR products suggest increase of DSBs. Sites with no recurrent DSBs were used for normalization.

Total RNA-seq

Total RNA was isolated using Quick-RNA miniprep (Zymo research) according to manufacturer instructions. The RNA-seq libraries were prepared using Ribo-Zero rRNA Removal Kit (Illumina).

Chromatin Immunoprecipitation (ChIP) and ChIP-seq

MCF7 cells (∼10⁶) were crosslinked with 1% formaldehyde for 10 min at room temperature and quenched with glycine, 125 mM final concentration. Fixed cells were washed twice in PBS and incubated in sonication buffer (0.5M NaCl, 0.5%SDS RIPA buffer containing PMSF, protease and phosphatase inhibitors) for 30min on ice. Cells were sonicated using Covaris M220 for 10min to produce chromatin fragments of ∼200-300 bp. The sheared chromatin was centrifuged 10min at maximum speed. From the supernatant, 50 μL were saved as input DNA and the rest was diluted in RIPA buffer without NaCl and SDS. The chromatin was immunoprecipitated by incubation with 5 μg of TOP1 antibodies (an3825, ab85038) pre mixed with 100 μL Dynabeads Protein G and incubated over-night at 4°C with rotation. Immunoprecipitates were washed twice with RIPA buffer + 150mM NaCl, twice with RIPA buffer + 300 mM NaCl, and twice with TE buffer. The chromatin was eluted from the beads with 200 μL of direct elution buffer (10mM Tris pH 8, 0.3M NaCl, 5mM EDTA, 0.5% SDS) and incubated overnight at 65°C to reverse the cross-linking. Samples were treated with RNase for 1h at 37°C and with proteinase K at 55°C for 2 h. DNA was cleaned up by QIAquick PCR purification column (QIAGEN), according to the manufacturer’s instructions, and eluted in 30 μl of elution buffer. The ChIPed and the Input DNA were used for real-time PCR or to prepare libraries by Truseq (Ilumina) and sequenced in Nextseq (illumine).

BLISS

BLISS was performed as described in detail in Yan et al. (2017) (Mirzazadeh et al., 2018). Briefly, :300,000 cells were grown on 12mm coverslips in 24-well plates. After treatment (or mock), cells were washed twice with PBS and fixed in 4% paraformaldehyde for 10 min at room temperature. After washing, cells were lysed and submitted to in situ DNA ends blunting (with Quick Blunting kit, NEB), followed by in situ DNA ends ligation with BLISS linker. After washing, genomic DNA was digested by HaeIII (NEB) for 3hr at 37°C and the fragmented DNA was extracted. The fragmented DNA was in vitro transcribed using MessageAmpII kit (Ambion) for 14 h at 37°C. After RNA purification and ligation of the 3′-Illumina adaptor, the RNA was reverse transcribed. The final step of library indexing and amplification was performed using the Illumina TruSeq Small RNA Library Prep Kit.

Suspension BLISS (sBLISS)

sBLISS was performed as described in detail in Gothe et al. Molecular Cell, 2019. Briefly, 10⁶ cells in suspension were fixed in 4% paraformaldehyde in PBS/10%FCS for 10min at RT. Fixation was quenched with 125 mM glycine for 5min at RT followed by two washes with ice-cold PBS. The cells were lysed for 60min on ice and the nuclei were permeabilized for 60min at 37°c. Then, nuclei were washed twice with CutSmart Buffer supplemented with 0.1% Triton X-100 (CS/TX100), and DSB ends were in situ blunted with NEB’s Quick Blunting Kit for 60min at RT. Blunted nuclei were washed twice with 1x CS/TX100 before proceeding with in situ ligation of sBLISS adapters to the blunted DSB ends. Adaptor ligation was performed with T4 DNA Ligase for 20-24h at 16°C and supplemented with BSA and ATP. After ligation, nuclei were washed twice with 1x CS/TX100 and genomic DNA was extracted with Proteinase K at 55°c for 14-18h while shaking at 800rpm. Afterward, Proteinase K was heat-inactivated for 10 min at 95°c, followed by extraction using Phenol:Chloroform:Isoamyl Alcohol, Chloroform, and ethanol precipitation. The purified DNA was sonicated in 100 μL TE using a BioRuptor Plus (Diagenode) with the following settings: 30 s ON, 60 s OFF, HIGH intensity, 40 cycles. Sonicated samples were concentrated with Agencourt AMPure XP beads (Beckman Coulter) and fragment sizes were assessed using a BioAnalyzer 2100 (Agilent Technologies) to range from 300bp to 800bp, with a peak around 400-600bp. The sonicated DNA was in vitro transcribed using MessageAmpII kit (Ambion) for 14 h at 37°C. After RNA purification and ligation of the 3′-Illumina adaptor, the RNA was reverse transcribed. The final step of library indexing and amplification was performed using the Illumina TruSeq Small RNA Library Prep Kit.

Quantification and Statistical Analysis

ChIP-seq processing

The sequenced files, containing 84bp single-ended reads, were trimmed for Illumina adaptor and for base quality, keeping Phred score above 20, with trim galore utility of Babraham Bioinformatics. Reads were then aligned to the human genome (GRCh38) with Bowtie2 (v2.2.9) (Langmead and Salzberg, 2012) keeping the default end-to-end scoring coefficients. Coverage files, both bigwig and bedgraph, were created with the bamCompare utility of deeptools (Ramírez et al., 2016) by comparing the treated BAM file with the control input file while ignoring duplicates. We have also created our own black list of regions, mainly consisting of chromosomal centromeres, and filtered the results accordingly.

Peak calling was performed with the MACS2 tool (Zhang et al., 2008). Prior to applying peak calling, filtered bed files were created from the aligned SAM files (for both treated and control) using a custom script. The filter kept reads exhibiting high mapping quality of 42 (highest figure for for Bowtie2) and an average base quality of at least 32. For peak calling, the treated file was compared the control file while ignoring duplicates, setting fold limits at their default of 5-50 but lowering q-value to a stricter figure of 0.01.

BLISS data processing

BLISS fastq files are initially scanned and demultiplexed according to the sample barcode and UMIs (Unique molecular identifiers). Quality control is applied with the trim galore tool, trimming to base quality of 20 and removing adapters. The preprocessed fastq files are then aligned to hg38 assembly with Bowtie2, creating SAM files. A custom Python script is used to extract detailed statistical information from these SAM files, including distribution of: mapping qualities, alignment scores, minimum, and read-averaged quality scores. Filtering by this information ensures very high-quality reads while reads within a predefined black list of genomic regions (mainly centromeres) are discarded. Next, unique reads are kept by removing ‘pure’ duplicates (containing the same UMI and genomic start position) or ‘candidate’ duplicates. The criteria for candidate duplicates are: (I) share the same genomic start position and the Hamming distance between UMIs is less than two; (II) share the exact same UMI and their genomic start position is less than 3 bases and (III) if there is a fold change in their counts larger than two.

The list of unique reads is converted to BED files for further analysis such as motif search and break enrichment.

Algorithms used for plotting the data

Some Heatmaps were generated by using “ComplexHeatmap,” R package and “matplotlib,” Python package (from Hunter, 2007). Other heatmaps were generated with deeptools (Ramírez et al., 2016), specifically computeMatrix, plotHeatmap and plotProfile tools.

Expression Boxplots were created by compering two bed files with genomic regions of super enhancers (in MCF7/HMEC cell lines from dbSUPER (http://bioinfo.au.tsinghua.edu.cn/dbsuper/) using Bioconductor’s DiffBind package (v2.4.8) in order to find common and unique super enhancers. The outputs were annotated using HOMER to detect the closet genes. Two lists of genes and their expression values, from two different RNA-Seq experiments of MCF7 and MCF10 were searched for those genes. Their expression values were the basis of the boxplot, which was generated using gplot2 (v2.2.1).

Plotting Data: R package version 3.0.1. “gplots” (https://CRAN.R-project.org/package=gplots).

Comparison between TEAD4 and RAD51 ChIP-seq: Peaks from the two experiments were compared using Bioconductor’s DiffBind package (v2.4.8). The graph was created using the VennDiagram package (v1.6.17).

Enhancer to Gene mapping- Several databases suggest possible enhancer to gene matching. In this research, we adopt Homer (Heinz et al., 2010) utility for peak annotation, annotatePeaks.pl, associating peaks with their nearby genes.

Super enhancers of MCF7 and HMEC were downloaded from dbSUPER (http://bioinfo.au.tsinghua.edu.cn/dbsuper).

Motif analysis was performed with Homer utilities (Heinz et al., 2010). Motif search is performed with utility: findMotifsGenome.pl, which searches for enriched motifs within predefined genomic regions. A proper background of regions for comparison is also selected by the utility. In most cases, the search was repeated for several different aperture widths around the BLISS sites or around other annotated regions like enhancers (typical apertures between 50 and 2000 bp). In some cases, the search was also repeated for varying motif sizes (8-20). All variations had minor effect on the resulting dominance of the TEAD motif. Homer utility also provides p values for the motif occurrences in comparison with the background. The utility reports known motifs and also de-novo motifs. De-novo motifs are also assigned with the best-matched know motif as presented in the figures.

Exact genomic location of motifs is found with Homer utility annotatePeaks.pl (appending the -m tag). Genomic distance between motifs is calculated with Python custom scripts.

Data and Code Availability

The BLISS datasets generated during this study are available at GEO: GSE136365.

Published: October 8, 2019