Dual blockade of BRD4 and ATR/WEE1 pathways exploits ARID1A loss in clear cell ovarian cancer

ARID1A, an epigenetic tumor suppressor, is the most common gene mutation in clear-cell ovarian cancers (CCOCs). CCOCs are often resistant to standard chemotherapy and lack effective therapies. We hypothesized that ARID1A loss would increase CCOC cell dependency on chromatin remodeling and DNA repair pathways for survival. We demonstrate that combining BRD4 inhibitor (BRD4i) with DNA damage response inhibitors (ATR or WEE1 inhibitors; e.g. BRD4i-ATRi) was synergistic at low doses leading to decreased survival, and colony formation in CCOC in an ARID1A dependent manner. BRD4i-ATRi caused significant tumor regression and increased overall survival in ARID1AMUT but not ARID1AWT patient-derived xenografts. Combination BRD4i-ATRi significantly increased γH2AX, and decreased RAD51 foci and BRCA1 expression, suggesting decreased ability to repair DNA double-strand-breaks (DSBs) by homologous-recombination in ARID1AMUT cells, and these effects were greater than monotherapies. These studies demonstrate BRD4i-ATRi is an effective treatment strategy that capitalizes on synthetic lethality with ARID1A loss in CCOC.


Introduction
Ovarian cancer is a heterogeneous disease with multiple histological subtypes 1 . There is a global consensus on the necessity of defining treatment strategies for ovarian cancer by histologic subtype. Even though less common in the US, clear cell ovarian cancer (CCOC) represents more than 25% of all ovarian cancers in Asia including Japan, Taiwan, and Singapore 2 . CCOC is one of the most challenging subtypes to treat as it is relatively insensitive to standard of care chemotherapies and is thus associated with a worse prognosis than more common subtypes, such as high-grade serous ovarian cancer (HGSOC) [3][4][5] . Response rates to standard front-line chemotherapy for patients with measurable residual tumor after debulking surgery with advanced disease are only 11.1% for CCOC compared to 72.5% for HGSOC 5 . Even more challenging is the lack of effective therapies for platinum-resistant recurrent CCOC with response rates of about 1% with various chemotherapy regimens 6 . Strategies to identify more effective therapeutic options for CCOC are clearly a clinical unmet need.
CCOCs demonstrate unique genetic alterations (e.g., ARID1A, PIK3CA, PPP2R1A, KRAS, PTEN) which may be exploited by targeted therapies 7,8 . Despite distinct genomic/molecular alterations in CCOC, they are currently treated similarly to HGSOC with standard platinum-based chemotherapy. Not surprisingly, treatment results are poor as described above [3][4][5][6]9 . Unlike HGSOC which is thought to originate from the distal fallopian tube 10 , CCOC is thought to arise from displaced endometriosis, further supporting a unique molecular landscape 11 . Additionally, CCOCs represent a heterogeneous disease at the genomic level despite having similar histological features 12 . Based on the transcriptome profiling of 212 primary tumors, Bolton et al. identified two distinct molecular subgroups for CCOCs, the first driven by ARID1A mutations (included those with P1K3CA, TERT mutations) and the second, with TP53 mutations (including mesenchymal differentiation and immune -related pathways). These subgroups each have distinct clinical outcomes, with the TP53 subgroup fairing worse, and exhibiting potentially unique therapy responsiveness 13,14 . Recent gene interaction analysis and functional assessment in CCOCs revealed that mutated genes were clustered into groups related to chromatin remodeling, DNA repair, cell cycle checkpoint, and cytoskeletal organization 15 . Integrated analyses uncovered that frequently mutated or amplified/deleted genes were involved in the KRAS/PI3K (82%) and MYC/retinoblastoma (RB) (75%) pathways as well as the critical chromatin remodeling SWI/SNF complex (85%) 15 . ARID1A is the most prevalent mutation in CCOC, with more than 50% of all CCOC tumors harboring this mutation 16 . ARID1A is a member of the SWI/SNF complex, with family members having helicase and ATPase activities which also regulate transcription of a subset of genes by altering the chromatin structure around those genes. This complex also has a major role in the repair of DNA lesions by directly facilitating DNA accessibility on the chromatin or indirectly by facilitating the functions of DNA repair proteins (e.g. p53, BRCA1, ATR, and Fanconi anemia proteins) 17, 18 . In addition, ARID1A maintains genome stability by protecting telomere cohesion and mutations or inactivation of ARID1A causes DNA damage 19 . The SWI/SNF complex is required for dozens of processes that are critical for cell cycle checkpoint control and differentiation 17 . Mutations, translocations and deletions involving various subunits of the SWI/SNF complex were found in ~20% of all human cancers, with ARID1A being the most frequently mutated member 20,21 . Recent studies show that SWI/SNF-mutant cancers depend on residual SWI/SNF complexes for their aberrant growth, thus revealing synthetic lethal interactions that could be exploited for therapeutic purposes 22,23 . Finally, while CCOCs exhibit unique genomic alterations as well as activation of various signaling pathways, including PI3K/AKT/mTOR, VEGF, HNF-1β, and IL-6/STAT3, no novel molecular-targeted therapies have yet been developed for this refractory subtype of ovarian cancer 7 .
Recently, several selective small molecule inhibitors influencing chromatin-modifying proteins have been developed as first-in-class cancer therapies 24,25 . Among them, Bromodomain and extraterminal domain (BET)-family protein inhibitors (BETi; JQ-1, iBET-762, AZD5153, and ZEN3694; clinicaltrials.gov) are being pursued in early clinical trials showing tolerability and antitumor activity for various types of cancers including HGSOC [25][26][27][28][29] . BETi bind to the bromodomain of BET proteins (predominantly BRD4) and prevent their interactions with acetylated histones thereby inhibiting the transcription of genes which are very important in tumorigenesis such as MYC, Receptor Tyrosine Kinases and downstream effectors such as MTOR 25,30 . CCOC and endometrioid ovarian cancer have increased expression of c-MYC by immunohistochemistry (IHC) 31 . BETi exhibit selectivity for tumor cells by preferentially binding to super-enhancers, noncoding regions of DNA critical for the transcription of genes that determine a cell's identity 21,24 .
BRD4 also plays an important role in regulating the expression of genes required for M to early G1 phase transition. BRD4 recruits P-TEFb to chromosomes at late mitosis to promote G1 gene expression and cell cycle progression. Knocking down BRD4 leads to G1 cell cycle arrest and apoptosis 32 . Loss of ARID1A sensitizes breast cancer cells to BET inhibition 33 , and most ARID1A mutated (ARID1A MUT ) ovarian clear cell carcinomas showed relative sensitivity to BETi compared with ARID1A wild type (ARID1A WT ) in vitro and inhibited tumor growth in vivo as monotherapy 34 .
Although BETi have strong and rational indication for ARID1A MUT CCOC, concerns remain about the limited single-agent efficacy in the clinic 24 . Given cancers ultimately develop resistance to most monotherapy approaches, combination therapy is a strategy to potentially prevent emergence of resistance, and increase durability of responses 35 .
ARID1A-depleted cells have impaired G2/M checkpoint initiation and maintenance 36 . When DNA double strand breaks (DSB) are induced, ARID1A is recruited via its interaction with DNA damage checkpoint kinase ataxia telangiectasia and Rad3-related (ATR). It is required for DSBinduced ATR activation and promotes DSB end resection, leading to the sensitivity to PARP inhibitors (PARPi) in ARID1A deficient cancer cells in vitro and in vivo 36 . ARID1A deficiency also results in topoisomerase 2A and cell cycle defects, which cause an increased reliance on ATR checkpoint activity, and ATR inhibitors (ATRi) are synthetic lethal and active in ARID1A mutant tumors preclinically 37 . WEE1 tyrosine kinase is another key G2 cell cycle checkpoint regulator that arrests cells in G2, allowing for DNA repair by homologous recombination 38,39 . WEE1 inhibition abrogates the G2 checkpoint, and triggers premature cell cycle entry into mitosis with resulting mitotic catastrophe leading to cell death 38 . Given that ARID1A MUT cancers are known to have defects in S and G2/M, WEE1 inhibitors (WEE1i) are also a rationale therapeutic approach to exploit ARID1A MUT CCOC 36,37 . ATR and WEE1 inhibitors are currently being evaluated in phase I/II clinical trials as monotherapy and in combination with radiation or cytotoxic chemotherapy [40][41][42][43][44] . Considering effects of ATR/WEE1 inhibitors combined with vulnerabilities of ARID1A MUT tumors induced by aberrations in DNA damage repair (DDR), ATRi and WEE1i are rational candidates for combination with BETi 37,45 .
In this study, we found that BETi, ATRi and WEE1i were the most active monotherapies in ARID1A MUT cells compared to ARID1A WT cells in a drug screen. We found that low dose BRD4 inhibition (BRD4i) in combination with DNA damage response inhibitors (DDRi, BRD4i-ATRi or BRD4i-WEE1i) were synergistic in decreasing cell viability and colony formation in ARID1A MUT cells compared to ARID1A WT cells. Combination of BRD4i-ATRi led to significant tumor regression and increased overall survival comparing to standard chemotherapy or monotherapies in ARID1A MUT CCOC patient-derived xenograft (PDX) models with less effect in ARID1A WT PDXs. The BRD4i-ATRi combination induced G1 arrest, decreased homologous recombination (HR) regulators such as RAD51 and BRCA1 expression, leading to increased DNA DSB and cell apoptosis in ARID1A MUT cells compared to wild type cells. Our studies identify a novel drug combination capitalizing on ARID1A mutations common in CCOC that is highly effective and tolerable, warranting evaluation in the clinic.  1A-B) 6 . At clinically comparable doses, BET inhibitors, BRD4i (AZD5153, 0.6 µM) and JQ1 (0.8 µM) followed by DNA damage response (DDR) inhibitors, WEE1i (AZD1775) and ATRi (AZD6738) where the most active in decreasing viability in ARID1A mutant compared to ARID1A wild type CCOCs. Chemotherapy (carboplatin and paclitaxel), PARPi (olaparib), DNA methyl transferase inhibitors (Decitabine), PI3K inhibitors (BKM120), EZH2 inhibitors (GSK126, EPZ6438) were less effective as measured by MTT assay (Supplementary Fig. 1A-B). Although DDR inhibitors (DDRi) and BET inhibitors are more effective in decreasing viability than chemotherapy and other agents (including inhibitors to PI3K, PARP, EZH2, DNMT), their monotherapy efficacy could potentially be further optimized.
BRD4i-ATRi combination therapy is more effective than monotherapy in ARID1A MUT compared to ARID1A WT CCOC PDX models There has previously been a lack of well-characterized in vivo experimental models for CCOC.
We and other groups have shown that patient-derived xenograft (PDX) models represent the architecture and genomics of the original patient tumor, demonstrating the natural progression of ovarian cancer, and mimicking the drug response of the patient [47][48][49][50][51][52][53] . In this study, we developed 8 PDX models from CCOC patients using an orthotopic transplant technique (suture tumor chunk to the ovary) in NSG mice which results in primary ovarian tumors with metastasis to the peritoneal cavity similar to what is observed in patients with advanced disease. We characterized these 7 CCOC PDX models and genomic profiles are shown (  BRD4i-ATRi vs. control for ARID1A MUT vs. wildtype; Fig. 5A). In further analysis of cell cycle phase related genes, although the ATRi treatment impact on genes expressed in the S-phase of cell cycle were subtle, BRD4i treatment resulted in a significant downregulation, and the effect was further enhanced by combination of BRD4i-ATRi treatment (Fig. 5B). Furthermore, the repression was more evident in HCT116 cells carrying ARID1A KO mutation, suggesting that ARID1A plays a key role in attenuating efficacy of the combination. In addition to repression of genes expressed in S-phase, several genes involved in G1/S progression were also affected. Specifically, CDK2 and CDC25A were down-regulated while CDKN1B, CDKN2B, and RB1, negative regulators of G1/S progression, were upregulated ( Fig 5C). This pattern was more pronounced with BRD4i-ATRi combination treatment than with BRD4i monotherapy, and ATRi alone only had subtle impact, potentially explaining the synergistic effect of the combination therapy. Further, RPPA analyses showed that p-Rb protein level was decreased in HCT116 ARID1A KO compared to that of ARID1A WT after treatment with BRD4i-ATRi combination ( Further, Western blot analysis showed that the protein levels of p-Rb in TOV21G (ARID1A MUT ) and HCT116 ARID1A KO were significantly decreased compared to that in OVKATE (ARID1A WT ) and HCT116 ARID1A WT cells, suggesting G1 arrest in ARID1A deficient cells (Fig. 5F). Thus, although DDRi and BRD4i treatments have strong effects on cell cycle progression, ARID1A-loss led to higher susceptibility to BRD4i-associated G1 cell cycle arrest. We also examined levels of protein expression of additional cell cycle regulation factors after treatment with ATRi and BRD4i.
ATR and BRD4 downstream proteins, pChk1, CtIP, and c-Myc, were downregulated by BRD4i-ATRi independent of ARID1A protein level ( Supplementary Fig. 5). The protein level of P27 (CDKN1B), an inhibitor of cell cycle progression, was increased after drug treatment (Fig. 5F), similar to its RNA level change observed with RNAseq. Cdc6, cell division cycle 6, is a critical regulator of DNA replication and required for recruiting minichromosome maintenance (MCM) protein complexes to DNA. Cdc6, a key regulator of cell cycle, also works as a cycle checkpoint maintenance which orchestrates S phase and mitosis 57 . BRD4i-ATRi treatment led to significant decreases of p-Cdc6 and total Cdc6 in cell lines carrying ARID1A MUT or ARID1A KO (Fig. 5F), again suggesting inhibition of cell cycle progression was more pronounced in the setting of ARID1A loss.  6C, Supplementary Fig. 7). We then further tested levels of apoptotic protein markers (cleaved caspase-3, cleaved caspase-7, and cleaved PARP) in TOV21G and HCT116 ARID1A deficient cells, and OVKATE and HCT116 (ARID1A WT ) cells after treating with ATRi, BRD4i, or the combination. Apoptosis marker proteins were increased more by BRD4i-ATRi in cells with ARID1A loss compared to ARID1A WT cells (Fig. 6D).

BRD4i-ATRi combination decreases homologous recombination and induces cell apoptosis in CCOC with ARID1A loss
Taken together, DDRi-BRD4i combination treatment significantly activated DNA damage and apoptosis pathways more so in cells with ARID1A MUT or ARID1A KO compared to ARID1A WT . In CCOC with ARID1A loss, inhibition of BRD4 decreases BRCA1expression levels and prevents RAD51 loading, thus leading to decreased homologous recombination and increased DNA double strand (DS) DNA breaks. ATRi inhibits CHK1 and activates CDK1 activation, resulting in loss of the G2 checkpoint, at the same time, it also increases replication stress and DS DNA breaks (Fig   7). Combination of BRD4i-ATRi significantly induced DNA damage and cell apoptosis (Fig. 7).
However, in ARID1A wild type CCOC cells, BRD4i has less effect in decreasing BRCA1 expression and RAD51 loading, and leads to minimal induction of DNA damage and cell apoptosis when in combination with ATRi (Fig. 7).

Discussion
Clear cell ovarian cancer (CCOC) is one of the most challenging subtypes of ovarian cancers to treat. They are intrinsically resistant to standard chemotherapy and thus, developing effective treatment for this subtype is critical and an unmet medical need [3][4][5] . The ARID1A gene, a member of the SWI/SNF family, is the most prevalently mutated gene in CCOC 7  To identify effective drugs for CCOC treatment, either as a monotherapy or combination therapy, we performed a drug screen with various targeted drug candidates including standard chemotherapy, epigenetic regulators, and inhibitors of tyrosine kinase signaling pathways and the DNA damage response. We identified BRD4i followed by DDR inhibitors (ATRi, WEE1i) as the most effective drugs in CCOC cells and cells with ARID1A mutations were especially sensitive to Given the rapid emergence of resistance to monotherapy in cancer, we investigated whether combination BRD4i with WEE1 or ATR inhibition would be a strategy to exploit ARID1A loss by targeting two critical pathways critical for survival. To validate drug screening results, a large library of ARID1A mutant and wild type CCOC cells were tested with BRD4i in combination with ATRi or with WEE1i. We found both combinations were synergistic by coefficient of drug interaction (CDI) scores (CDI <1=synergism) 46 in decreasing viability and colony formation more so in the ARID1A MUT lines compared to ARID1A WT at in vitro doses lower or comparable to doses used in the clinic (BRD4 10-100nM, WEE1i 100-250nM, ATRi 100-1000nM; Fig. 1). We validated that response to BRD4i with ATRi or BRD4i with WEE1i was dependent on ARID1A loss status by ARID1A knockout or knock down increasing response to the combination; and overexpression of ARID1A, decreasing the response to the BRD4i-DDRi combination. Others have shown that combination inhibition of BRD4 and ATR demonstrates synergistic cytotoxic activity in other cancers such as lymphomas, melanoma and high-grade serous ovarian cancer [58][59][60] , however ARID1A loss as a biomarker of sensitivity was not explored. We further show that drug sensitivity was impaired by increased cellular levels of ARID1A protein not only in CCOC, but in endometrial, colon, and HGSOC cells (Fig. 2) warranting further investigation of this combination in other cancer types with ARID1A loss. Eric: Collectively, these results support the concept that ARID1A pathogenic variants sensitize cancer cells to the BRD4i-ATRi combination and may serve as biomarkers to select patients for this efficacious treatment.
Advancements in CCOC treatments have been hampered by lack of preclinical models.
PDX models developed represented similar genomic (e.g., ARID1A, PIK3CA, KRAS etc.) and protein (PAX8, ARID1A, napsin A, racemase) profiles to the native patient tumor from which they were derived supporting use of these models as surrogates of the patient tumor. Using the CCOC PDX platform, we compared BRD4i-ATRi and BRD4i-WEE1i vs. monotherapy and carboplatin in ARID1A MUT (WO-38, WO-24, and WO-93) and wild type models (WO-30, WO-120; Fig. 4 and Supplementary Fig. 1). Although some tumor growth suppression was observed with monotherapies (e.g carboplatin, ATRi, WEE1i and BRD4i), BRD4i-ATRi combination was tolerable and led to significant robust decreases in tumor volume and increased overall survival relative to single-agent therapies that were maintained for >15-35 weeks in ARID1A MUT in contrast to modest effects in the ARID1A WT PDX models (Fig. 4). Indeed, in the ARID1A WT (WO-30, WO-120) PDXs, BRD4i-DDRi did not demonstrate significant antitumor activity (Fig. 4D, E). BRD4i-DDRi combinations were tolerated at clinically relevant doses as shown by stable body weight.
Interestingly, the BRD4i-ATRi was more active than the BRD4i-WEE1i likely because ATR has distinct roles in protecting replication fork stability in S phase in addition to G2/M regulation, a key function of the downstream WEE1 kinase. Alternatively, the dual effects of WEE1i on G1/S and G2/M progression may compromise its cooperation with BRD4i. In summary, more relevant to clinical applications, BRD4i-ATRi treatment causes complete tumor regression in chemotherapy-resistant CCOC PDX tumors more so than monotherapy in an ARID1A leveldependent manner.
Mechanistically, RNA-seq and RPPA drug response studies revealed that cell cycle regulators of G1/S including p27 and pRB, were significantly reduced by BRD4i-ATRi combination in an ARID1A mutation-dependent manner. p27 Kip1 arrests cells at G1 by inhibiting the activity of cyclin E-CDK2 complexes 61 , and phosphorylation of RB1 and CDC6 mark passage from G1 into S phase. 62 . We found combination BRD4i-ATRi increased G1 arrest and decreased S phase, which indeed correlates with decreased phosphorylation of RB and CDC6 more so than monotherapy in a manner dependent on ARID1A mutation (Fig 5.). Similarly, others have shown BRD4 inhibition deregulates CDC6 activity and results in aberrant DNA replication re-initiation and sensitization to replication stress-inducing agents 58 . Since ARID1A-null or KO is defective in the G2M checkpoint 36 , it is possible that the combination of replication fork collapse and G/M checking inhibition with ARID1A mutation allows damaged cells to pass through M phase and arrest in G1 through an alternative checkpoint mechanism (e.g. ATM), ultimately leading to apoptosis. Similarly, ATR-mediated phosphorylation of CHK1 is higher in ARID1A-mutant or deleted cells (Supplementary Fig. 3), thus making these cells more reliant on ATR function to stabilize stalled replication forks 37 .
Given RNA seq analysis identified DNA Damage Repair pathways as significantly more effected in ARID1A MUT compared to wildtype cells, we evaluated drug effects on homologous recombination (HR), the primary mechanism to repair DNA DSBs 63 . BRCA1 regulates RAD51 deposition in response to DNA damage by recruiting PALB2 and BRCA2 64 . BRCA2 loads RAD51 onto resected DSBs for HR 65,66 . RAD51 foci formation has been used as a functional biomarker of HR 67 which is critical for DNA repair. RNA-seq revealed that BRD4i decreased BRCA1 expression, with a further decrease observed following BRD4i-ATRi treatment in the ARID1A mutant cells (Fig. 5).
ARID1A loss and BRD4i increases reliance on ATR function 37,60 . Thus, we examined the effect of ATRi added to BRD4i on ɣH2AX, a DNA double stand break (DSB) marker 68 and cleaved caspase 3, a marker for apoptosis 69 . Treatment of ARID1A MUT PDXs indeed exhibited significantly increased ɣH2AX and CC3-positive cells compared with ARID1A WT (Fig. 4 and Supplementary Fig. 1) suggesting ARID1A has a role in maintaining genomic stability. However, ATRi and BRD4i caused more DSBs in combination combined to monotherapies, and these breaks are further increased by ARID1A loss. Therefore, we observed that BRD4i-ATRi combination significantly induced more γH2AX in ARID1A deficient CCOC lines compared to wild type models.
Regarding the mechanism by which BRD4i-ATRi increase breakage, we also found that monotherapy each decreased RAD51 foci in ARID1A KO cells, which is consistent with previous findings in other genetic contexts 49,70 . However, combination of BRD4i-ATRi decreased RAD51 foci more so than monotherapies, suggesting that this combination further inhibits HR-mediated repair of DNA double strand breaks, leading to apoptosis (Fig. 6). The mechanisms underlying these effects varied and are likely cumulative. For example, ARID1A loss causes decreased transcription of HR genes 18 . BETi may cooperate with ARID1A loss in producing these changes 71 .
ARID1A loss leads to the ATR-mediated phosphorylation of CHK1, which implies replication fork stalling (Supplemental Fig. 5). In addition, inhibition of the ATR-CHK1 pathway has been shown to suppress the accumulation of RAD51 on DSBs 72 . Although each drug demonstrates distinct mechanisms of action, the combination of BRD4i and ATRi synergizes with ARID1A MUT to generate DSBs and permit aberrant progression though M phase into G1, where alternative checkpoint mechanisms are triggered (Fig. 7).
In summary, our study identified a novel drug combination, BRD4i-ATRi that exploits the most common genetic vulnerability, ARID1A loss in CCOCs, which are in dire need of new treatment options. Combination low dose BRD4i-ATRi was tolerable and highly effective in aggressive platinum resistant CCOC models justifying further evaluation of this combination in the clinic.

In vitro proliferation assay
Two thousand cells were seeded in 96-well plate at least in triplicate on Day 0. The cell numbers were measured by CyQUANT Cell Proliferation Assay kit (C7026, Thermo Fisher Scientific) after 24, 48, 72, 96, or 120 hours incubation following the manufacture's protocol. Doubling times were calculated by http://doubling-time.com/.

Western blotting
Cells were seeded into 6-well plates at 3× 10 5

Patient-derived xenograft (PDX) models
The WO-38, WO-24, WO-93, WO-30, WO-120, WO-28, and WO-36 models were developed by orthotopic transplantation of patient tumor to the ovaries of mice using methods previously described 47  These dosages and schedules were optimized as maximum tolerable dose by prior dose-deescalating preliminary studies. The body weights and condition scores of mice were monitored weekly. In all the models, percentage change in body weight during treatment was used as a marker for toxicity and dose level adjustments. Significant treatment toxicity was defined as a 15% drop in body weight and the mice require treatment reduction at 25% dose and supportive supplements care (i.e. gel pack supplement and subcutaneous fluid injection if necessary). For mice with 20% drop in body weight, treatment was stopped, and supportive measures were provided. Body weight was rechecked every 3-4 days. Once improved, treatment was restarted at 25% reduced dose. If body weight was not regained after one week, PDX was euthanized in accordance with IACUC protocols. Trial endpoints were defined as tumor volume > 1000 mm3 or poor condition score (defined as score of 1 on a 1-5-point scale) 75 . Tumors were collected and snap frozen for protein and genomic analysis and fixed in formalin for IHC.