The BET inhibitor JQ1 selectively impairs tumour response to hypoxia and downregulates CA9 and angiogenesis in triple negative breast cancer

The availability of bromodomain and extra-terminal inhibitors (BETi) has enabled translational epigenetic studies in cancer. BET proteins regulate transcription by selectively recognizing acetylated lysine residues on chromatin. BETi compete with this process leading to both downregulation and upregulation of gene expression. Hypoxia enables progression of triple negative breast cancer (TNBC), the most aggressive form of breast cancer, partly by driving metabolic adaptation, angiogenesis and metastasis through upregulation of hypoxia-regulated genes (for example, carbonic anhydrase 9 (CA9) and vascular endothelial growth factor A (VEGF-A). Responses to hypoxia can be mediated epigenetically, thus we investigated whether BETi JQ1 could impair the TNBC response induced by hypoxia and exert anti-tumour effects. JQ1 significantly modulated 44% of hypoxia-induced genes, of which two-thirds were downregulated including CA9 and VEGF-A. JQ1 prevented HIF binding to the hypoxia response element in CA9 promoter, but did not alter HIF expression or activity, suggesting some HIF targets are BET-dependent. JQ1 reduced TNBC growth in vitro and in vivo and inhibited xenograft vascularization. These findings identify that BETi dually targets angiogenesis and the hypoxic response, an effective combination at reducing tumour growth in preclinical studies.


INTRODUCTION
Epigenetic regulators are promising targets in cancer as transcriptional dysregulation and mutations in chromatin modulators and transcription factors (TF) are common in many malignancies. The bromodomain and extra-terminal (BET) proteins are lysine acetylation readers that mediate gene expression, including oncogenes. 1 BET inhibitors (BETi) demonstrate many anticancer effects by downregulating gene expression of oncogenic factors. 2 Breast cancer is the most common female cancer and triple negative breast cancer (TNBC) is its most aggressive subtype. Low oxygen (hypoxia) can drive TNBC progression, 3 promoting adaptation through genes within the major hallmarks of cancer. 4 Hypoxia can control gene expression recruiting chromatin remodelling complex 5 and histone deacetylases. 6 Thus, we investigated whether BETi JQ1 could impair the hypoxia response in TNBC and exert therapeutic effects.
Hypoxia is found in 450% of breast tumours and arises from high metabolic and proliferative rates and aberrant tumour vascularization. Clinically, hypoxia is associated with chemoradiotherapy resistance, metastasis and poor survival, 7 being a key area for targeted therapeutic development. 3,8 Most hypoxic responses are mediated by the hypoxia-inducible factors 1α and 2α (HIF-1α and HIF-2α), which in the absence of O 2 stabilize and heterodimerise with HIF-1β. 7,9 This heterodimer binds to the hypoxia response element in gene promoters and induces transcription of genes, which drive molecular adaptation through many pathways, including pH regulation (CA9), angiogenesis (VEGF-A), metabolism (LDHA) and metastasis (LOX). 7,8 Other pathways including the unfolded protein response, XBP1 and ATF4 are also important. 8,10 TNBC shows overexpression of HIF target genes and is the breast cancer subtype most frequently associated with hypoxia. 11,12 Targeting HIF directly is a major challenge, while targeting HIF downstream targets such as VEGF has proven more feasible, although targeting just one protein has had less effect on overall survival than expected. [13][14][15] The BET proteins (BRD2-4 and BRDT) regulate transcription by 'reading' acetylated histones and recruiting TFs and epigenetic regulators. 1,2 JQ1 is a BETi 16 that showed effects on tumour growth and survival, 17 cell cycle arrest, and differentiation. 16,[18][19][20] Although many attributed JQ1 effect to its MYC downregulation, 18,21 it is unlikely that this is the sole mechanism 17 and some studies do not corroborate this, 22 as MYC downregulation is not always sufficient to inhibit cell growth 23 and JQ1 effects are observable without MYC alteration. 24 BET proteins can associate with many TFs 21,25 and other genes are regulated by BETi, such as p21, BCL-xl, BCL2, AKT, FOSL1 and RUNX2. [20][21][22]24 Although the oncogenic driver varies, tumour addiction to BET activation seems common to many malignancies. These results led to clinical evaluation and there are 13 BETi clinical trials currently underway (www.clinicaltrials.gov).
Given the clinical investigation of BETi, we assessed whether the BETi JQ1 could alter the hypoxia response, exerting an anti-tumour effect. JQ1 modulated 44% of hypoxia-responsive genes, of which two-thirds were downregulated including CA9 and VEGF-A. JQ1 reduced TNBC growth in monolayer and spheroid culture. Furthermore, JQ1 prevented HIF binding to the CA9 promoter. Finally, JQ1 downregulated CA9 and VEGF-A expression and reduced growth and vascularization (CD31 positivity) in a TNBC xenograft model. These data show that JQ1 impairs tumour response to hypoxia.  Figure S1A). Interestingly, JQ1 had a greater impact on hypoxia-sensitive genes than those that were not hypoxia-sensitive (P = 0.046; median logFC = 1.22 × 1.02) (Supplementary Figure S1C).

RESULTS
To further investigate the effect of JQ1 on tumour response to hypoxia, we defined which pathways were hypoxia-regulated and evaluated their expression under JQ1 treatment. Then, we developed a Hypoxia Network (HyN) (Figure 1b) containing the Hypoxia Signature 27 and a list of genes for each pathway found to be hypoxia-regulated (obtained from KEGG, Supplementary Table S1). Hypoxia upregulated most of the HyN clusters in both cell lines (Figure 1c and Supplementary Figure S1B). Gene Set Enrichment Analysis showed that hypoxia upregulates angiogenesis, glycolysis, oxidative phosphorylation and pentose phosphate pathway in MDA-MB-231 (Figure 1c and Supplementary Table S2). The Hypoxia Signature 27 set of genes was upregulated in hypoxia as expected (MDA-MB-231: ES = − 0.79, normalized enrichment score = − 2.68, Po0.001, false discovery rate q-valueo0.001; MCF-7: ES = − 0.83, normalized enrichment score = − 2.63, Po0.001, false discovery rate q-valueo0.001, n = 3). JQ1 treatment prevented hypoxic upregulation of the hypoxia signature, angiogenesis, oxidative phosphorylation and pentose phosphate pathway gene data sets, but did not alter glycolysis or MYC expression. In addition, JQ1 treatment downregulated the cell cycle and TCA sets of genes in hypoxia (Figure 1c and Supplementary Table S2). MCF-7 results confirmed most of these findings (Supplementary Figure S1B and Supplementary Table S3 Figure S4).
Additional BETi were tested (I-BET151 and I-BET762) and confirmed these findings, also downregulating CA9, VEGF, CXCR7, TMEM45A and LOX in hypoxia (Supplementary Figure S5). In contrast, I-BET151 in hypoxia induced VEGF-A in HCC1806 cell line and PFKFB3 in both cell lines tested. No effect was observed on HIF-1α expression, but there was a significant upregulation of HIF-2α at the RNA level.
Immunoblot analysis further confirmed that JQ1 induced a significant reduction of CA9 protein induction in hypoxia in both TNBC cell lines (P o 0.01, n = 3) (Figure 4b). JQ1 did not significantly alter HIF-1α and HIF-2α protein expression ( Figure 4b).
JQ1 reduces HIF binding to the CA9 promoter To investigate how JQ1 prevents hypoxia-responsive gene expression, we evaluated chromatin immunoprecipitation (ChIP) Hypoxia Network (HyN) of HIF-1β to the hypoxia response element of the CA9 promoter, as HIF-1α dimerizes with HIF-1β prior to transcription induction. As expected, hypoxia increased HIF binding in both the MDA-MB-231 (P = 0.002, n = 3) and HCC1806 (P = 0.02, n = 3) (Figure 5a). JQ1 treatment in hypoxia reduced HIF binding to the CA9 promoter to normoxic levels in both MDA-MB-231 (P o 0.01, n = 3) and HCC1806 (P o0.05, n = 3) (Figure 5a). This suggests HIF-1 is BETdependent for binding/recruitment in some of its downstream targets, explaining how BET inhibition reduces the expression of hypoxia-induced genes.
In addition to this, expressions of BRD2-4 are increased in hypoxia (Supplementary Figure S6A), suggesting a role for BET proteins in hypoxia. Although JQ1 is a potent inhibitor for all the BET proteins, 16 some of its effects are attributed to a specific isoform. We performed siRNA knockdown to investigate which isoform was responsible for these effects (Supplementary Figure S6B). BRDT was excluded as its expression was below the limit of detection. The BRDT siRNA did not change CA9, VEGF or  (Figure 5b). Therefore, some HIF targets were shown to be BET-dependent (CA9 and VEGF-A), while others, such as LDHA are BET-independent.
To better comprehend how JQ1 can have these effects, BRD4 binding and acetylation of H3K27 and H4 in hypoxia at the promoters of VEGF and CA9 in HCC1806 and MDA-MB-231 were investigated by ChIP qPCR (Figures 5c and d  One-way analysis of variance, n = 3, *P o0.05, **P o0.01, ***P o0.001. We investigated available BRD4 ChIP-Seq data for MCF-7 in normoxia 31 and compared these with those genes that were downregulated by JQ1 in hypoxia and found that: 61% had BRD4 binding by ChIP-Seq in normoxia (Supplementary  Tables S5 and S6). We investigated the BRD4 binding to genes that were downregulated by JQ1 in hypoxia and also upregulated in response to hypoxia; 39% of these genes showed direct binding by BRD4 (Supplementary Tables S5  and S6). Finally, we investigated the overlap between BRD4 binding in normoxia in MCF-7 31 and HIF-1α or HIF-2α binding in hypoxia 26 in MCF-7 published data sets. These data confirmed BRD4 and HIF-1 or HIF-2 binding in 14% of the genes that we identified are increased in hypoxia and downregulated by JQ1.

DISCUSSION
Hypoxia represents a key target for the development of therapies in cancer. 3,32 Hypoxia induces a transcriptomic shift largely dependent on HIF, 33 and there is evidence for HIF dependence upon epigenetic regulation in response to hypoxia. 5, 6 We demonstrate an epigenetic approach to modulate the tumour response to hypoxia and reduce growth in TNBC. JQ1 modulated the expression of 44% of hypoxiaresponsive genes in MDA-MB-231 TNBC cell lines, of which two-thirds were downregulated. More specifically, JQ1 downregulated the expression of the major regulators of hypoxic pH regulation and angiogenesis, CA9 and VEGF-A, in TNBC cell lines and xenografts. We observed that in hypoxic conditions, there was an increased histone acetylation at, and BRD4 binding to, the CA9 and VEGF promoters, suggesting an explanation for JQ1 effectiveness in this context. It is possible that BRD2 and/or BRD3 may also be important in the regulation of CA9 and VEGF. Analysis of published data sets 26,31 identified that many of the hypoxic JQ1-regulated genes have BRD4 binding and that 14% of the genes simultaneously upregulated by hypoxia, downregulated by JQ1 and bound by BRD4 are direct targets of either HIF-1α or HIF-2α; these included VEGFA and CA9. Finally, as JQ1 prevents HIF binding to the CA9 promoter but not all HIFregulated transcription, the data suggest that some HIF targets are BET-dependent.
JQ1 consistently downregulated CA9 in in vitro and in vivo models. Hypoxic tumours develop in an acidic microenvironment, owing to increased production of metabolic acids and poor vascularization. 34 CA9 is highly induced in hypoxia, where it allows adaptation to this environment maintaining a more neutral intracellular pH. 35,36 Increased CA9 expression is a marker of poor prognosis in breast cancer and is more common in TNBC than other breast cancer subtypes. 36 CA9 inhibition reduces tumour growth and metastasis. [35][36][37]  JQ1 also consistently demonstrated an anti-angiogenic effect, as it reduced the expression of the angiogenic pathway, the key angiogenic inducer VEGF-A and blood vessel count. JQ1-treated xenografts showed lower levels of Tie2 and NRP, involved in vascular stabilization and branching and promotion of arterial growth. 38 Conversely, there was a higher expression of EFNB2/ ephrinB2 (Figure 6b), described as a regulator of arterial/venous specialization and vessel branching. 38 Collectively, this indicates that JQ1 could impair the early steps of angiogenesis, a major hallmark of cancer.
Many studies showed an anti-tumoural effect of BETi, 19,21,22,24 and recently it was found that JQ1 shows a typical behaviour of anti-angiogenic agents and, in fact, JQ1 can reduce tumour vascularization by suppressing VEGF stimulation. 39 Angiogenesis is upregulated by hypoxia and supports tumour progression. 7 Anti-angiogenic therapy is a major cancer treatment strategy used to treat eight solid tumour types. However, this strategy was found to induce hypoxia in around 50% of patients. 7 Hypoxic adaptation enables resistance to anti-angiogenic therapy and may in part explain why the promise of anti-angiogenic therapy in *p<0.05, **p<0.01, ***p<0.001 compared to untreated normoxia # p<0.05, ## p<0.01, ### p<0.001 compared to untreated hypoxia   breast cancer has not been fulfilled. We have shown that combined inhibition of VEGF and CA9 act at least additively and in some examples synergistically to reduce tumour growth. 35 Anti-angiogenic therapies can lead to metabolic adaptation. 7,35 JQ1 treatment increased the expression of LDHA and PFKFB4 in xenografts, but not in cell cultures. LDHA was reported to be downregulated by JQ1 in ovarian cancer, 40 may be because of its regulation by MYC. 41 Whereas JQ1 reduced the expression of oxidative phosphorylation, pentose phosphate pathway, TCA gene data sets but not glycolysis in TNBC cell line MDA-MB-231, in the ER+ cell line MCF-7, JQ1 increased TCA. This may be due to differences in the metabolic requirements of these subtypes of breast cancer. Thus, we might expect that a co-treatment with an anti-glycolytic or pro-OXSPHOS drug (such as metformin) could lead to a synergistic effect and be a promising therapy, especially in TNBC.
We also highlight the impact of BET inhibition on wider hypoxic gene expression. The hypoxic regions of tumours are resistant to other therapies, therefore we propose that utilizing BET inhibitors to target the hypoxic tumour cells in combination with additional chemotherapy or radiotherapy may provide better responses. Combining hypoxia targeting with radiotherapy or chemotherapy has been shown previously to provide a greater therapeutic response. For example, targeting hypoxia-regulated genes including CA9, one of the JQ1-regulated genes, increases sensitivity to radiotherapy and chemotherapy. 37,42 Stem cell characteristics comprise another important hallmark of cancer and epigenetic regulation has an important role in this. BRD4 has been proposed as a marker for self-renewal 43 and JQ1 can downregulate genes involved in this process in human cordderived mesenchymal stem cells. 44 Stem cells are maintained in an undifferentiated state through expression of the core transcriptional factors Nanog, Oct4 and Sox2. BRD4 is required for Nanog expression and JQ1 inhibits this inducing rapid differentiation of murine embryonic stem cells 45,46 as well as significantly downregulating Oct4 and SOX2. 46 BET inhibition or BRD4 depletion reduces the expression of pluripotent genes and shifts cellular fate. 45,46 Collectively, these data show that BRD4 is critical for the maintenance of pluripotency and maintaining stem cell fate, while inhibition of BET proteins enhances differentiation.
Initially, studies described JQ1 effects as MYC-dependent. 16,18,47 Although some studies reported MYC expression predicts JQ1 sensitivity, our results indicate other mechanisms are relevant; as JQ1 reduced tumour cell growth both in MYC-amplified (MCF-7 and HCC1806) and MYC-non-amplified cell lines (MDA-MB-231, Cal51 and SUM159). Our data provide further evidence for the context dependence of BETi.
Several studies show that BETi have broader MYC-independent effects. 19,48,49 JQ1 impairs the recruitment of multiple TFs to their targets by physical disruption, for example, between the BRD4 and the N-terminal domain of the androgen receptor. 19,21 Thus, JQ1 acts by blocking BET protein ability to bind to chromatin, which in turn prevents TF recruitment, possibly including HIF. This is in agreement with our observation of reduced HIF binding to the CA9 promoter region in response to JQ1. Other important hypoxia-regulated genes demonstrated a similar pattern of downregulation by JQ1 in this study. These include CXCR7 and LOX. CXCR7 is a G protein-coupled receptor upregulated in breast cancer associated with worst outcome that mediates angiogenesis and metastasis. 50 LOX is also upregulated in breast cancer and confers a poor prognosis, where it enables angiogenesis 51 and disrupts bone homeostasis providing a favourable environment for metastatic cells from hypoxic ER − breast cancer. 52 Taken together, this led us to the original suggestion that HIF targets can be divided into BET-dependent and BET-independent.
The SWI/SNF chromatin remodelling complex was the first epigenetic factor demonstrated to regulate the response to hypoxia. 5,53 This complex makes DNA accessible to other factors, especially through its ATPase subunits BRM and BRG1. 54 In breast cancer, BRG1 and BRM are overexpressed in most primary breast cancers and are needed for in vivo tumour formation and TNBC cell line proliferation. 55 SWI/SNF can either directly regulate the expression of HIF-1α and HIF-2α or regulate the expression of hypoxia-responsive genes, including CA9. 5 The CA9 promoter nucleosome is BRG1-depedently remodelled in response to hypoxia. 5 JQ1 does not bind to BRM or BRG1, 16 and there is no current knowledge regarding interactions between BET proteins and SWI/SNF complex. While JQ1 prevents acetylated histones from being 'read', SWI-SNF can promote deacetylation. 54 Both interact with MYC. 16,18,47,54 Finally, just as CA9 was modulated by both factors, LDHA was not. Future studies should address to what extent the set of genes affected by these factors overlap. It might be the case that BET proteins and SWI/SNF complex interact at some level forming an enhanceosome and only some HIF targets are epigenetically regulated, rather than being BET-or SWI/SNF-dependent.
In conclusion, we showed that BETi impairs tumour response to hypoxia, targeting multiple pathways such as angiogenesis and pH control. Our findings alter the understanding of tumour response to hypoxia and identify a new avenue for epigenetic therapy to target the hypoxic tumour microenvironment. Furthermore, these results have a clear impact on the interpretation of the results from current clinical trials and future clinical use of drugs that inhibit the BET proteins in solid tumours.
Functional analysis was carried out on DEG to identify statistically overrepresented ontologies using Database for Annotation, Visualization and Integrated Discovery (https://david.ncifcrf.gov/). DEG fulfilled the criteria: FC log2 ⩾ 1 and Po0.05.

Real-time PCR (qPCR)
qPCR was performed (n = 3) as described previously. 35 Primers sequences are available in Supplementary Table S4. Immunoblotting Immunoblotting was performed as described previously (n = 3) 35 with primary antibodies listed in Supplementary Table S7. Bands were quantified using ImageJ.
Tumour growth was measured with calipers by experienced technicians blinded to the experimental hypothesis and after three out of five animals reached 150 mm 3 , animals received JQ1 or vehicle (10% DMSO, 10% hydroxypropyl beta cyclodextrin) intraperitoneally at 50 mg/kg daily. Animals were randomly grouped at injection and one group was treated with JQ1 whilst the other was untreated (mean tumour sizes at the treatment starts were; shCTL, 120 mm 3 and JQ1-treated, 137 mm 3 ). When tumours reached 1.44 cm 3 , the mice were killed by cervical dislocation.

ChIP assay
ChIP assay was performed for antibodies listed in Supplementary Table S7 using the EZ-ChIP Chromatin Immunoprecipitation Kit (#17-371, Millipore, Billerica, MA, USA) or ChIP-IT Express Enzymatic (Active Motif) according to the manufacturer's instructions. DNA isolated from ChIP was quantified by qPCR using CA9 or VEGF promoter primers (Supplementary Table S4) (n = 3).

Statistical analysis
Statistical analysis and graphs were performed using GraphPad Prism v6.0 (GraphPad, La Jolla, CA, USA). Results are plotted as mean values with standard deviation. Statistical tests and the number of repeats are described in the figure legends. Student's t-test was used for two sample analyses and normal distributions were assumed, otherwise the nonparametric Mann-Whitney test was used. Analysis of variance was used for 42 sample analyses. No samples or experimental repeats were excluded from analyses. For the in vivo experiment to detect a FC of 2 at alpha (P value) = 0.05, five animals in each group gave a 75% power of detection. No statistical methods were used for the samples size selection of other experiments.