ARID1A Deficiency Impairs the DNA Damage Checkpoint and Sensitizes Cells to PARP Inhibitors

ARID1A (AT-rich interactive domain 1A) has been identified as one of the most frequently mutated genes in human cancers by multiple next-generation genomic sequencing studies (1–3). ARID1A mutation rates ranging from 10% to 57% have been identified across multiple tumor lineages, including ovarian clear cell carcinoma, uterine endometrioid carcinoma, gastric cancer, hepatocellular carcinoma, esophageal adenocarcinoma, breast cancer, pancreatic cancer, transitional cell carcinoma of the bladder, renal cancer, Waldenström macroglobulinemia, pediatric Burkitt lymphoma, and cholangiocarcinoma (1–3). ARID1A, also known as BAF250A, is a subunit of the evolutionarily conserved SWI/SNF chromatin remodeling complex (4, 5). The SWI/SNF complex repositions, ejects, or exchanges nucleosomes, which serves to modulate DNA accessibility to cellular processes involved in chromatin structure, such as transcription, DNA replication, and DNA repair (6–8). However, how ARID1A deficiency contributes to cancer development and approaches to exploit ARID1A deficiency therapeutically are not known.

Ataxia telangiectasia and RAD3-related protein (ATR) is a member of the phosphatidylinositol 3-kinase–like kinase family. Along with another kinase, ataxia telangiectasia-mutated (ATM), ATR functions as a central regulator controlling cellular responses to DNA damage (9–11). In general, ATM is activated by double-strand DNA breaks (DSB), whereas ATR responds to single-strand DNA breaks (SSB; ref. 12). However, the ATM- and ATR-activating DNA lesions are interconvertible: DSBs activate ATM but can also activate ATR as a consequence of DSB end resection, which generates a single-stranded region (13–15). Unlike ATM, ATR is essential for cell survival (16), supporting the functional importance of ATR for genome maintenance programs. For example, in S phase, ATR regulates replication initiation, replisome stability, and replication fork restart (17). In G₂ phase, ATR prevents premature mitotic entry in the presence of damaged DNA via the G₂ checkpoint (18, 19). Thus, a key question remains unanswered: how is ATR signaling regulated, allowing it to perform versatile roles in the DNA damage response (DDR)? One possibility is that ATR-interacting proteins fine-tune the temporal and spatial functions of ATR in DDR. Therefore, we conducted a proteomic analysis to systematically identify ATR-interacting proteins. In addition to many known ATR-binding proteins, such as ATRIP, we identified ARID1A as an unexpected interacting partner of ATR. Human cancers result in large part from the accumulation of multiple genetic alterations, including mutations, deletions, translocations, and amplifications (20). Thus, our proteomic result raised the intriguing question of whether ARID1A, through its interaction with ATR, plays a role in maintaining genomic integrity that could be exploited as a therapeutic liability.

In this study, we found that ARID1A is recruited to DSBs via its interaction with ATR. In response to DNA damage, ARID1A facilitates DNA DSB end processing to generate Replication Protein A (RPA)–coated single-strand DNA (ssDNA) and sustains ATR activation in response to DSBs. Loss of ARID1A leads to impaired checkpoint activation and repair of DNA DSBs, which sensitizes cells to DSB-inducing treatments, such as radiation and PARP inhibitors. Thus, our results provide biologic insights into the function of ARID1A as a tumor suppressor in human cancers and a mechanistic basis for targeting ARID1A-deficient tumors.

To explore the mechanisms regulating the functions of ATR in DDR, we conducted an immunoprecipitation (IP) assay to enrich ATR-associated protein complexes which were then subjected to silver staining and mass spectrometry (Fig. 1A). In addition to known ATR-binding proteins, such as ATRIP, we identified ARID1A as a binding partner of ATR (Fig. 1A and Supplementary Fig. S1). Notably, in addition to ARID1A, multiple subunits of the SWI/SNF complex, including BRG1, BAF57, BAF60, BAF170, and SNF5, were also identified by the mass spectrometry analysis, suggesting that ATR interacts broadly with the SWI/SNF complex. To confirm the interaction between ARID1A and ATR, we performed reciprocal IP with V5-tagged ARID1A (Fig. 1B) and endogenous IP analyses (Fig. 1C and Supplementary Fig. S2), which confirmed that ARID1A interacts with ATR. Given the important role of ATR in DDR, we next tested whether ARID1A is recruited to DNA breaks. We used chromatin IP (ChIP) assay to examine whether ARID1A was recruited to the proximity of a single site-specific I-SceI–induced DSB (Fig. 1D and Supplementary Fig. S3A), as previously described (21, 22). Interestingly, we found that ARID1A was enriched at the chromatin region close to this DSB (Fig. 1D and Supplementary Fig. S3B and S3C). To facilitate visualization of the recruitment of ARID1A to DNA lesions, we used a light activation system (KillerRed System; ref. 23) to determine whether ARID1A localized at sites of DNA damage. Briefly, KillerRed (KR) is a light-stimulated reactive oxygen species inducer fused to a tet-repressor (tetR) or transcription activator (TA; tetR+VP16), which binds to a TRE cassette (∼90 kb) integrated at a defined genomic locus in U2OS cells (U2OS TRE cell line; ref. 24). KR facilitates the formation of oxygen radicals and superoxide through the excited chromophore (25, 26) to induce DNA damage, including SSBs and DSBs. Targeting the expression of KR to one specific genome site allows visualization of the recruitment of specific proteins to a site of DNA damage. As shown in Fig. 1E and Supplementary Fig. S4, EGFP-tagged ARID1A predominantly localized in the nucleus. Upon activation of KillerRed, ARID1A showed a specific enrichment at the DNA damage site, which colocalized with TA-KR focus but not with TA-mcherry, indicating that the recruitment of ARID1A at TA-KR is specific to DNA damage. These findings revealed that ARID1A interacts with ATR and is recruited to DSBs.

We next set out to determine whether the recruitment of ARID1A to DSBs is dependent on its interaction with ATR. First, we transiently knocked down ATR and examined the recruitment of ARID1A to DSBs via I-SceI–based ChIP. As shown in Fig. 1F, in ATR-knockdown cells, recruitment of ARID1A to DSBs was significantly reduced. This suggested that ATR is required for recruiting ARID1A to DNA lesions. Second, we examined whether the DNA damage signaling induced by ATM and ATR is required for recruitment of ARID1A to DSBs. We treated cells with chemical inhibitors of ATM or ATR. These inhibitors effectively blocked DNA damage signaling, as shown by their impact on phosphorylation of ATM and CHK1 (Fig. 1G). The ATM inhibitor markedly decreased the recruitment of ARID1A to DSBs (Fig. 1G). The ATR inhibitor also decreased recruitment of ARID1A to DSBs, but to a lesser extent, whereas ATR knockdown significantly reduced the recruitment of ARID1A to DSBs (Fig. 1F and G). Consistent with previous findings that ATR recruitment to DSBs requires ATM and not the converse (13–15), these data suggested that interaction between ARID1A and ATR, with ATR acting as a scaffold, and DNA damage signaling initiated by ATM are required for efficient ARID1A recruitment to DSBs.

To understand the molecular details of the ARID1A–ATR interaction and the recruitment of ARID1A to DSBs, we first used two deletion constructs to test whether ARID1A binds to ATR through its N-terminal or C-terminal half. As shown in Fig. 1H, binding of ATR to the C-terminal of ARID1A was readily detectable, whereas N-terminal ARID1A did not pull down ATR even though the N-terminal ARID1A expressed at levels similar to the full-length protein. This result suggested that the domains required for ATR interaction are located in the C-terminal half of ARID1A. Next, we generated deletion mutants of ARID1A to further map the regions mediating its interaction with ATR (Fig. 1I–K). Using these constructs, we found that regions from 1800 to 1900 amino acids (aa) and 2100 to 2200 aa at the C-terminal half of ARID1A were essential for its interaction with ATR (Fig. 1I and J). We further examined the recruitment of deletion mutants to DSBs. Indeed, we found that deletion of ARID1A-interacting domain suppressed the recruitment of ARID1A to DSBs (Fig. 1L and Supplementary Fig. S5).

Collectively, these data showed that ARID1A interacts with ATR via its C-terminal region, which mediates its recruitment to DSBs.

Next, we tested whether ARID1A deficiency impairs the cellular response to DNA damage. As our model system, we used isogenic HCT116 cell lines with wild-type ARID1A and a knock-in mutant ARID1A (Q456*/Q456*) that abolishes ARID1A expression because of an early stop codon. We first examined cell cycle distribution at different time points after exposure to ionizing radiation (IR). As shown in Fig. 2A, 1 hour after irradiation, control cells started to accumulate at the G₂–M checkpoint. At 4 and 8 hours after irradiation, there was a significant increase in cells at the G₂–M phase, suggesting that ARID1A-depleted cells had weakened G₂–M checkpoint activation. Sixteen hours after irradiation, control cells still had a large proportion of cells arrested at the G₂–M checkpoint, whereas ARID1A-depleted cells exhibited a markedly reduced proportion of cells at the G₂–M checkpoint and a significant increase of cells in G₁ phase (Fig. 2A). These results indicated that ARID1A deficiency leads to impaired G₂–M checkpoint initiation and maintenance. To confirm these results, we used phospho-Histone H3 staining to measure the fraction of mitotic cells after exposure to IR in ARID1A-depleted cells. Without IR treatment, there was no apparent difference in the percentage of mitotic cells between control and ARID1A-depleted cells (Fig. 2B). After exposure to IR, in control cells the percentage of mitotic cells was significantly reduced and started to recover 16 hours after IR (Fig. 2B). In contrast, ARID1A-deficient cells showed a slower decrease in numbers of mitotic cells at early time points after IR exposure and a significantly increased percentage of cells reentering mitosis at 16 hours after IR, suggesting defective G₂–M checkpoint initiation and maintenance (Fig. 2B). In accord with this finding, as shown in Fig. 2C, ARID1A depletion led to a marked increase in cumulative mitotic reentry after IR exposure revealed by paclitaxel treatment, which blocks mitotic exit. Defective G₂–M checkpoint maintenance was not due to a differential response to paclitaxel between control and ARID1A-null cells, because we observed a similar block in mitotic accumulation in cells not treated with IR (Fig. 2C). ARID1A expression was effectively depleted in ARID1A-null cells (Fig. 2D). These data are consistent with ARID1A deficiency significantly impairing G₂–M checkpoint initiation and maintenance.

As ARID1A is a subunit of the SWI/SNF complex, we next asked whether the chromatin remodeling activity of SWI/SNF is required for G₂–M checkpoint response. We knocked down the core catalytic subunits BRG1 or BRM in U2OS cells (Supplementary Fig. S6) and found that BRG1 deficiency led to an increase of mitotic cells 16 hours after exposure to IR, similar to the increase observed in ARID1A-null cells (Fig. 2E). This result suggested that ARID1A-associated BRG1-containing SWI/SNF complexes are required for maintaining G₂–M cell-cycle arrest after induction of DSBs.

Having observed a defective DNA damage checkpoint in ARID1A-deficient cells, we further examined whether ARID1A deficiency impairs the DNA damage checkpoint signaling pathway. We treated cells with IR and examined the activation of CHK1, a key G₂–M checkpoint regulator. In ARID1A-depleted cells, we found reduced CHK1 (S317) phosphorylation in response to IR, particularly at 8 hours after IR (Fig. 3A). In response to DSBs, CHK1 (S317) can be phosphorylated by either ATM or ATR. Thus, we examined whether ARID1A deficiency affected ATM and/or ATR activation. In response to DNA damage, ATM (S1981) and ATR (T1989) undergo autophosphorylation, which provides a surrogate marker for kinase activation (13, 27–29). ARID1A depletion remarkably reduced ATR activation [phosphorylation of ATR (T1989)] in response to IR (Fig. 3B), but did not alter ATM activation [phosphorylation of ATM (S1981)] or recruitment of ATM to DNA damage sites (Fig. 3C and Supplementary Fig. S7). Previous studies showed that the G₂–M checkpoint is impaired in the absence of ATR (16). Thus, our data are consistent with ARID1A depletion impairing ATR activation in response to DSBs and thereby altering checkpoint signaling.

In general, DSB ends are the preferred substrate for ATM binding, which activates first ATM and then ATR to sustain ATM-initiated signaling (11). It has been shown that phosphorylation of H2AX in response to IR is mediated by both ATM and ATR (30, 31). Therefore, we examined the effect of ARID1A deficiency on the dynamics of H2AX phosphorylation. As we expected, chromatin binding of γ-H2AX was significantly reduced in ARID1A-deficient cells at 8 hours compared with 4 hours after IR, indicating impairment of sustained γ-H2AX foci formation (Fig. 3D). To confirm this result, we tested whether ARID1A deficiency reduced γ-H2AX foci formation, which directly reflects the accumulation of γ-H2AX at DSBs. As shown in Fig. 3E, γ-H2AX foci formation was significantly reduced at later time points after IR. In addition, we examined the foci formation by DNA damage–responsive protein 53BP1, a key adaptor protein in checkpoint response whose recruitment to DSBs is dependent on the protein platform assembled by γ-H2AX formation (32). Consistent with the reduced γ-H2AX foci formation at 8 hours after exposure to IR, 53BP1 foci formation was remarkably reduced in ARID1A-deficient cells (Fig. 3F). As shown in Supplementary Fig. S8A, we used a comet assay to determine the presence of DSBs. Control and ARID1A-depleted cells had similar levels of DSB formation 8 hours after exposure to IR, suggesting that reduced γ-H2AX and 53BP1 foci formation was not due to a reduced level of DSBs. Collectively, these results indicated that ARID1A deficiency impairs ATR-mediated signaling in response to DSBs, which is required for sustaining DSB-induced DNA damage signaling.

Next, we sought to determine the molecular mechanism underlying the defective DSB-induced ATR activation in ARID1A-deficient cells. In response to DSBs, ATM is directly activated by the MRE11–RAD50–NBS1 (MRN) complex, which is required to recruit ATM to DSBs (33–35), whereas ATR activation and recruitment to DSBs require the formation of RPA-coated SSBs, which arise from 5′-3′ resection of DSB ends (18, 36). Therefore, we asked whether ARID1A depletion affects the process of DSB end resection, leading to reduced efficiency of ATR activation.

First, we examined phosphorylation of the ssDNA-binding protein RPA in ARID1A-depleted cells as an indicator of DSB resection efficiency (37). As shown by Western blotting (Fig. 4A), IR-induced phosphorylation of the RPA2 subunit (Ser4 and Ser8) was significantly reduced after ARID1A depletion. In contrast, ARID1A deficiency did not affect RPA phosphorylation in response to replication stress stimuli hydroxyuridine (HU) and ultraviolet light (UV) even though both HU and UV induce much stronger RPA phosphorylation than IR (Fig. 4B). This result suggested that ARID1A specifically affects DSB-induced formation of ssDNA. To confirm this observation, we used immunofluorescent staining to detect RPA phosphorylation (p) at DNA damage sites. Notably, ARID1A depletion markedly reduced pRPA (Ser4/Ser8) foci formation, consistent with decreased RPA accumulation and impaired ssDNA formation at DSBs (Fig. 4C and Supplementary Fig. S8B). We also tested the effect of ARID1A loss on the chromatin environment around DSBs by examining histone H3 occupancy at I-SceI–induced DSBs by ChIP assay. We found that H3 deposition was not altered in ARID1A-depleted cells before DNA damage (Fig. 4D). However, H3 occupancy was much higher in ARID1A-depleted than control cells after induction of DSBs by I-SceI. These data supported impaired DSB end resection due to loss of ARID1A, suggesting that recruitment of ARID1A to DSB is required to create a favorable chromatin environment for efficient DSB end resection. Next, we determined whether ARID1A depletion affects DSB repair via homologous recombination (HR) or single-strand annealing (SSA), which are the repair mechanisms requiring DSB end resection (Supplementary Fig. S9). In line with our finding that ARID1A is required for efficient DSB end resection, we found that ARID1A knockout impaired both HR repair and SSA repair efficiency (Fig. 4E and F). In addition, to exclude an indirect effect of ARID1A depletion on gene transcription regulation, we examined the key molecules involved in DDR and DSB end resection in ARID1A-depleted cells and observed no apparent reduction in protein levels (Supplementary Figs. S9 and S10). Together, these data suggested that initial DSB end resection resulting from ATM–MRN complex–dependent signaling recruits ATR in complex with ARID1A to promote DSB end resection and ATR activation and thereby augment DSB-induced DNA damage signaling.

Inhibitors of PARP1, an enzyme involved in repairing DNA SSBs (38, 39), are now in clinical trials and showing promising activity. PARP inhibitor treatment causes failure of SSB repair, which can lead to DSBs when DNA replication forks stall and collapse at persistent SSB lesions (38, 39). Furthermore, PARP inhibitors trap PARP on DNA, eventually leading to DSBs. Therefore, PARP inhibitors are selectively lethal to cells lacking BRCA1 or BRCA2, two proteins involved in repairing DSBs, or lacking other components of the DSB break repair pathway, but exhibit minimal toxic effects on normal cells and less activity in cancer cells without DSB repair deficiency (38, 39). We found that ARID1A-depleted cells exhibited a significant G₂–M checkpoint defect in response to DSBs, which may lead to insufficient cell-cycle arrest to allow efficient DSB repair (Fig. 2). In addition, ARID1A deficiency impaired DSB repair through both HR and SSA mechanisms (Fig. 4). On the basis of these observations, we reasoned that ARID1A deficiency may render cells vulnerable to DSBs induced by PARP inhibitors. We tested this hypothesis in a variety of isogenic models with multiple PARP inhibitors that are currently used in clinical trials.

First, we knocked down ARID1A in two nontransformed breast epithelial cell lines, MCF10A and HMEC, and then treated the cells with the PARP inhibitors olaparib, rucaparib, and veliparib. As shown in Fig. 5A and B, PARP inhibitors selectively decreased the survival of cells with decreased ARID1A expression. Next, we tested whether ARID1A depletion sensitized HCT116 colon cancer cells and MDA-MB-231 breast cancer cells to PARP inhibitors. We treated the cells with olaparib and BMN673, a potent PARP inhibitor (40, 41). Strikingly, ARID1A-knockdown cells showed remarkably reduced colony formation in the presence of PARP inhibitors (Fig. 5C and Supplementary Fig. S11A). As we observed the reduced colony formation in ARID1A-deficient HCT116 cells compared with control cells, we further examined the growth rate of these cells. As shown in Supplementary Fig. S11B, ARID1A knockdown reduced cell growth. However, we did not observe cytotoxic effects of ARID1A knockdown alone. It is noteworthy that not only the number of colonies but also the average colony size was lower with BMN673 than with olaparib in ARID1A-knockout cells, suggesting that BMN673 strongly inhibits cancer cell survival. In line with these findings, significantly enhanced apoptosis after exposure to PARP inhibitors was observed in ARID1A-depleted MDA-MB-231 cells (Fig. 5D and E), HCT116 cells (Fig. 5F), and HOC8 ovarian cancer cells (Supplementary Fig. S12). To confirm that ARID1A deficiency sensitizes cells to PARP inhibitors, we conducted four sets of experiments. First, we examined the correlation between ARID1A expression and PARP inhibitor sensitivity in a collection of ovarian cancer cells. As shown in Supplementary Fig. S13A–S13D, sensitivity to BMN673 was associated with ARID1A protein levels. Second, we reconstituted ARID1A-knockout HCT116 cells and ARID1A-knockdown U2OS cells with wild-type ARID1A. As shown in Supplementary Fig. S14A and S14B, wild-type ARID1A significantly reduced PARP-inhibitor–induced apoptosis. In the third experiment, we reconstituted ARID1A-knockout HCT116 cells with deletion mutants, which lack the ATR interaction domain and are not recruited to DSBs. In contrast to wild-type ARID1A, these deletion mutants did not rescue the apoptosis induced by PARP inhibitors in ARID1A-depleted cells (Fig. 5G and Supplementary Fig. S14C). This result suggested that the role of ARID1A in DSB repair contributes to cellular sensitivity to PARP inhibitor treatment. In the fourth experiment, we reconstituted patient-derived mutations of ARID1A in HCT116-knockout cells (Supplementary Fig. S14D). We found that two frameshift mutations (c.5715delA and c.5548delG) reduced protein expression and failed to rescue PARP inhibitor–induced apoptosis. In addition, a missense mutation (c.A5337G), which is localized in the ARID1A–ATR interaction region, was unable to rescue apoptosis induced by BMN673. In contrast, another missense mutation (c.C6693T), which is localized outside the interaction region, partially rescued apoptosis. Furthermore, we determined whether ARID1A-deficient cells are sensitive to PARP inhibitors because of defective ATR signaling. To this end, we treated wild-type cells and ARID1A-knockdown cells with a selective ATR inhibitor (Supplementary Fig. S15). We found that ATR inhibitor treatment significantly increased BMN673-induced apoptosis in wild-type cells, which showed an even stronger effect than ARID1A knockdown, suggesting that ARID1A deficiency only partially inhibits the function of ATR. Moreover, ARID1A knockdown did not significantly increase PARP inhibitor sensitivity in ATR inhibitor–treated cells, suggesting that the major effect of ARID1A loss is mediated by defective ATR function. These results showed that ATR inhibition of wild-type cells confers sensitivity to PARP inhibitors, which is epistatic with ARID1A deficiency. Collectively, these data further demonstrated that ARID1A deficiency sensitizes cancer cells to PARP inhibition, potentially through inhibition of ATR function in DSB repair.

Compared with olaparib and veliparib, rucaparib and BMN673 at the same concentration showed stronger effects on cell survival and apoptosis in ARID1A-deficient cells (Fig. 5E and F). Whether this represents differential activity on PARP inhibition or trapping or EC₅₀ values for the enzyme in intact cells remains to be determined. Based on the increased activity, we examined the dose dependence of rucaparib- and BMN673-induced apoptosis in ARID1A-depleted cells. As shown in Fig. 5H, both drugs induced a stronger apoptotic effect in ARID1A-depleted cells than in control cells. Consistent with the EC₅₀ for PARP inhibition in intact cells (40, 41), BMN673 induced apoptosis at much lower concentrations than rucaparib in ARID1A-deficient cells. Therefore, in the next step, we tested the antitumor effects of BMN673 against ARID1A-deficient cancer cells in vivo.

We treated nude mice bearing ARID1A-deficient and parental MDA-MB-231 breast cancer tumors and ARID1A-depleted and parental HCT116 colon cancer tumors in the opposite flanks with and without oral 0.33 mg/kg BMN673 daily. Because of the significant tumor burden in the untreated group, HCT116 xenograft mice were only treated with BMN673 for 16 days. After 1 week of treatment, a selective antitumor efficacy of BMN673 was observed in ARID1A-depleted HCT116 cells, and this antitumor effect was more marked at treatment day 16 (Fig. 6A and B). In contrast, BMN673 was without effect in mice bearing parental HCT116 cells (Fig. 6C). MDA-MB-231 xenografts, which grew more slowly, were treated with BMN673 for 30 days. Once again BMN673 significantly inhibited tumor growth in ARID1A-deficient cancer cells but not wild-type cells (Fig. 6A, D, and E). Thus in both xenograft models, growth of ARID1A-deficient xenografts was suppressed by BMN673 compared with vehicle (Fig. 6F), but growth of ARID1A-wild-type xenografts was not (Fig. 6F).

We analyzed the expression of the DDR marker phosphorylated CHK1 and the apoptosis marker cleaved caspase 3 in xenograft tumor tissues. As expected, in response to BMN673, ARID1A-deficient cancer cells exhibited much lower expression of pCHK1 than did control cells with wild-type ARID1A (Fig. 6G), consistent with loss of ARID1A decreasing the DDR induced by BMN673. Furthermore, ARID1A-deficient tumor cells exhibited enhanced apoptosis after BMN673 treatment (Fig. 6G). Taken together, our data suggest that targeting the defective DDR that occurs in ARID1A-deficient cells with PARP inhibitors could be beneficial for cancer patients with ARID1A-deficient tumors.

As shown in the proposed model (Supplementary Fig. S16), our study indicates that ARID1A interacts with ATR, is recruited to sites of DNA damage in an ATR-dependent manner, and facilitates and/or accelerates effective DNA DSB end resection. This ARID1A-mediated chromatin remodeling, by promoting efficient DSB end resection, is required for sustaining ATR-dependent signaling from DSBs and repair of DSBs through HR pathways. Our in vitro and in vivo data further show that PARP inhibitors demonstrate selective activity against ARID1A-deficient cells. Collectively, our results provide mechanistic insights into how ARID1A suppresses tumorigenesis and how therapeutic liabilities engendered by ARID1A deficiency could be exploited clinically.

In the DNA damage signaling network, similar regulatory mechanisms are used for promoting/sustaining cellular responses to DNA damage. For example, when DSBs occur, ATM and/or ATR phosphorylate the histone H2A variant H2AX, which can spread thousands of base pairs around DSB sites (42, 43). The presence of phosphorylated H2AX (γ-H2AX) provides docking sites to recruit DNA damage–responsive sensors, such as NBS1 and MDC1, and recruitment of these sensor proteins further activates or maintains ATM kinase activity (phosphorylation of H2AX) and amplifies ATM signaling (44–46). Processing of DSB ends to single-strand ends is required for ATR activation. The initial activation of ATR, which is dependent on MRN complex–ATM signaling, may not require ARID1A. Consistent with this recruitment–amplification/sustaining model, our data showed that ARID1A is recruited to DSBs via its interaction with ATR and thereby helps sustain ATR activation via promoting resection.

It is worth noting that our data showed that ARID1A deficiency reduced ATR activation in response to IR-induced DSBs. However, no significant change of ATR activation in response to HU was observed in ARID1A-deficient cells as examined in our current experimental condition. There are two possible explanations for this observation. First, the different structures of DSBs caused by IR and HU may be associated with distinct mechanisms of ATR activation. Whereas IR induces DSBs with blunt ends, HU induces replication stress–associated DSBs in S phase (47). These types of DSBs occur at stalled/collapsed replication forks, which normally are one-ended and associated with replication fork structure. We speculate that in response to IR-induced DSBs, chromatin remodeling activity mediated by ARID1A is required to promote efficient DSB end resection to sustain ATR signaling. However, in response to HU-induced (replication stress–associated, one-end) DSBs, ARID1A-mediated chromatin remodeling may not be required to promote ATR activation given the relatively open chromatin structure of replication forks during DNA replication in S phase. If the chromatin remodeling activity is required, it is likely that a different chromatin remodeling mechanism independent of ARID1A is involved in HU-induced one-end DSBs. Second, although HU induces replication stress–associated DSBs, HU primarily induces extensive formation of single-strand DNA coated with RPA. This DNA structure provides a molecular basis for activating ATR through ATRIP and TOPBP1 in S phase, which does not require DNA end resection. Thus, on the basis of our data, we speculate that ARID1A does not play a determinative role in regulating HU-induced ATR activation.

Multiple studies have indicated tumor-suppressor roles for ARID1A in transcriptionally regulating cell-cycle progression (48–50). Interestingly, our study shows that ARID1A directly regulates the DNA damage checkpoint independent of its roles in transcriptionally regulating cell-cycle progression. Consistent with this caretaker role of ARID1A, previous studies found that knockdown of SWI/SNF core component BRG1 or BRM could affect DNA repair pathways and proper mitosis segregation (1, 51). Consistent with this contention, a recent study reported that SWI/SNF factors are required for cellular resistance to DNA damage (52). To understand detailed mechanisms of how ARID1A regulates ATR activation in response to DSBs, two additional questions are of critical importance. First, is the role of ARID1A in regulating the function of the SWI/SNF complex related to its effects on ATR function? Our data suggest that the SWI/SNF complex is required for maintaining checkpoint activation. It is likely that ARID1A is required for recruiting or stabilizing the SWI/SNF complex at DSBs. Whether there are functionally restricted ARID1-containing SWI/SNF complexes independently involved in DNA damage repair or transcriptional regulation or whether the same protein complexes can mediate both processes depending on the cellular context will be important to ascertain. Second, how do patient-derived mutations of ARID1A affect its function in the DNA damage response? A variety of mutations are found in ARID1A in human cancers. Many of the mutations result in mutation-mediated RNA decay and unstable protein, but a number of the mutant ARID1A proteins are expressed. Thus to functionalize patient-derived mutations is not only important for understanding the molecular mechanisms of ARID1A in the DNA damage response, but also important for stratifying patients for clinical trials of targeted therapy with PARP inhibitors.

Our results showed that ARID1A deficiency impairs DSB repair, which normally should delay the progression of the cell cycle in order to allow resolution of the DSBs. However, we observed that ARID1A-deficient cells progress more quickly through the cell cycle, suggesting that in the ATR-mediated DNA damage checkpoint, ARID1A plays a dominant role in determining cell-cycle progression after DNA damage. It is likely that impaired DSB repair and cell-cycle checkpoint activity may allow ARID1A-deficient cells to transmit damaged (unrepaired) DNA to the next cell cycle, which could potentially lead to genomic instability. As suggested by a recent study in gastric cancer, ARID1A mutations are associated with increased microsatellite instability (MSI), and it is of interest to investigate whether ARID1A deficiency promotes the mutational process in tumorigenesis or whether ARID1A is a target for MSI.

PARP inhibitor monotherapy is well tolerated in patients but has minimal therapeutic effects in the absence of specific genetic defects, such as BRCA1/BRCA2 mutations (53). Thus, our study provides a mechanistic rationale for testing the efficacy of PARP inhibitors in ARID1A-deficient tumors either alone or in combination. However, as noted above, not all ARID1A mutations are demonstrated to compromise the DNA damage response. Therefore, it is imperative to select ARID1A-mutant tumors for PARP inhibitor treatment on the basis of the biologic significance of their specific mutations.

Specific inactivating mutations in several other SWI/SNF subunits, including PBRM1, ARID2, ARID1B, BRG1, SNF5, and BRD7, have also been frequently found in human cancers (1). However, mutations in different SWI/SNF subunits lead to distinct cancer spectrums (1). In addition, mouse models with genetic knockout of different SWI/SNF subunits exhibit distinct phenotypes (1, 54). These findings suggest that individual subunits may have distinct tumor-suppressor roles in specific tissue contexts. Given the diverse combinations and interactions of subunits in a particular SWI/SNF complex, it is possible that ARID1A has distinct effects on tumorigenesis in different tissues. Therefore, the mechanisms by which ARID1A exerts its tumor-suppressor function may be tissue dependent. The answer to this question will be instrumental to guide therapy with DNA damage–inducing agents such as PARP inhibitors. In the current study, we used well-established cancer cell models such as HOC8, HCT116, and MDA-MB-231 and normal diploid epithelial cell models such as MCF10A and HMEC as representative model systems to test our hypothesis. Thus, it is necessary to confirm our findings in additional cancer models, such as ovarian clear-cell carcinoma models, which exhibit the highest ARID1A mutation rate among all cancer types. It has been found that ARID1A mutations cooperate with activation of the PI3K–AKT pathway in promoting tumorigenesis (2, 3). A recent study showed that depletion of ARID1B reduces survival of ARID1A-mutant cancer cells (55). Thus, PARP inhibitors in combination with PI3K–AKT inhibitors or strategies targeting ARID1B may provide new therapeutic avenues for patients with ARID1A-mutant tumors.

HCT116 parental and ARID1A-knockout (Q456*/Q456*) cell lines were purchased from Horizon Discovery Ltd. and maintained according to the manufacturer's instructions. U2OS, MCF10A, and HMEC cells and breast cancer cell line MDA-MB-231 were purchased from the American Type Culture Collection (ATCC). Cell lines were validated by short tandem repeat (STR) DNA fingerprinting using the AmpF_STR Identifiler Kit according to the manufacturer's instructions (Applied Biosystems; cat. 4322288). The STR profiles were compared to known ATCC fingerprints, to the Cell Line Integrated Molecular Authentication database (CLIMA) version 0.1.200808 (Nucleic Acids Research 37:D925-D932 PMCID: PMC2686526), and to the MD Anderson fingerprint database. The STR profiles matched known DNA fingerprints or were unique. The U2OS cell lines were obtained from the ATCC in 2013. The authentication of the breast cancer cell lines and ovarian cancer cell lines was performed in the MD Anderson Cancer Center (MDACC) Characterized Cell Line Core facility in 2012 and 2013. HCT116 and ARID1A-knockout cell lines were obtained from Horizon Discovery, Inc., at the end of 2012 and automatically replaced with a new batch of the cell lines every year since then, through a contract with the company.

U2OS cells were maintained in McCoy's 5A medium (Cellgro) supplemented with 10% FBS with glutamine, penicillin, and streptomycin. MDA-MB-231 breast cancer cells were grown in RPMI 1640 medium supplemented with 10% FBS. HMEC cells were grown in HuMEC medium with growth supplement. MCF10A cells were maintained in mammary epithelial growth medium (Clonetics), a proprietary serum-free medium containing insulin, hydrocortisone, epidermal growth factor, and bovine pituitary extract. Cells were incubated at 37°C in a humidified incubator with 5% CO₂.

The pCDNA6-V5-ARID1A and its deletion forms 1-1758aa and 1759-2285aa were kindly provided by Dr. Ie-Ming Shih (Johns Hopkins University School of Medicine). The EGFP-ARID1A and its mutant forms were generated by Custom DNA Constructs. Other mutations were generated by the QuickChange II Site-Directed Mutagenesis Kit (Stratagene). The identity of all plasmids was confirmed by sequencing at the MDACC Sequencing and Microarray Facility.

Anti-ARID1A (1:500) and anti-RPA (1:1,000) antibodies were purchased from Bethyl Laboratories. Anti-V5 (1:1,000) was purchased from Life Technology. Anti–γ-H2AX (1:1,000) was purchased from Millipore. Anti-53BP1 (1:2,000) was purchased from Novus Biologicals. Anti-ATR (1:500) was purchased from Santa Cruz Biotechnology. Anti-ATRIP (1:1,000), anti-pATM (1:500), anti-pCHK1 (1:500), anti-pCHK2 (1:500), anti-GFP (1:1,000), and anti-pH3 (1:500) were purchased from Cell Signaling Technology. Anti-pATR antibody (1:500) was kindly provided by Dr. Lee Zou (Massachusetts General Hospital Cancer Center, Harvard Medical School). Anti-Tubulin (1:2,000) and anti–β-actin (1:2,000) were purchased from Sigma. An apoptosis detection kit was purchased from BD Biosciences. The PARP inhibitors olaparib, veliparib, rucaparib, and BMN673 were purchased from Selleckchem. The silver staining kit was purchased from Thermo Scientific.

ARID1A knockdown was achieved by RNA interference using a lentiviral vector–based MISSION shRNA or siRNA (Sigma). Lentiviral particles corresponding to the MISSION shRNA ARID1A- NM_006015 target set were used, as well as a MISSION nontarget shRNA control. The shRNA sequences were as follows: V3LHS 410041 (#1), TAAATAGCTGTGTCTCGCT; V2LHS_71866 (#2), TCTTGAGATAGCTCCTGCG. Specificity and efficacy of the shRNA ARID1A procedure were evaluated by Western blotting after transduction and puromycin selection in cells. siRNA transfection was conducted using oligofectamine (Life Technology) according to the manufacturer's instructions.

Cells were washed in PBS, and cellular proteins were extracted in 8 mol/L urea lysis buffer plus protease and phosphatase inhibitors (GenDEPOT) for 30 minutes at 4°C. Lysates were cleared by centrifugation, and proteins were separated by gel electrophoresis. Membranes were blocked in PBS 0.1% Tween 20 (PBS-T)/5% (w/v) milk for 1 hour at room temperature. Membranes were then incubated with primary antibodies diluted in PBS-T/5% (w/v) milk at 4°C overnight. Subsequently, membranes were washed with PBS-T and incubated with horseradish peroxidase secondary antibody (1:2,000; Jackson ImmunoResearch) diluted in PBS-T/5% skim milk. Membranes were washed in PBS-T, and bound antibody was detected by enhanced chemiluminescence (GE Healthcare). IP was performed by incubating lysates from 6 × 10⁶ cells with 1 μg of antibody at 4°C overnight, followed by addition of 20 μL of protein A/G-conjugated agarose beads (GE Healthcare). The precipitates were washed four times with ice-cold PBS, resuspended in 6× Laemmli buffer, and resolved by SDS-PAGE followed by immunoblotting. The preparation of chromatin fractions and Western blot analyses, including the conditions for RPA analysis, were performed as described previously (21, 56).

SSA assay cells were kindly provided by Dr. Jeremy Stark (Beckman Research Institute of the City of Hope). The HR and SSA repair assays were performed as described previously (21, 57–60).

For detection of DNA damage–induced foci of γ-H2AX, 53BP1, and pRPA32, immunofluorescent staining was performed essentially as described previously (21, 56). After treatment, cells were first subjected to cytoskeleton extraction/stripping (cytoskeleton buffer: 10 mmol/L PIPES, pH 6.8; 100 mmol/L NaCl; 300 mmol/L sucrose; 3 mmol/L MgCl2; 1 mol/L EGTA; 0.5% Triton-X 100; stripping buffer: 10 mmol/L Tris-HCl, pH 7.4; 10 mmol/L NaCl; 3 mmol/L MgCl2; 1% Tween 20; 0.25% sodium deoxychalate) to remove the unbound proteins and, second, fixed in PBS-buffered 4% paraformaldehyde. Primary antibodies were incubated at 4°C overnight, and secondary antibody Alexa 488–conjugated goat anti-rabbit IgG was incubated for 1 hour at room temperature. Slides were mounted in medium containing DAPI (Vector Laboratories) and analyzed under a fluorescence microscope. At least 50 cells per sample were scored, and foci were counted. Although currently available ARID1A antibodies are suitable for Western blot analysis and general fluorescent staining, they failed to detect formation of IR-induced foci of ARID1A at DNA breaks.

DSBs were induced in cells transfected with control siRNA or ATR siRNA by I-SceI expression. At indicated time points, cells were cross-linked with formaldehyde, and ChIP assays were performed with an EZ ChIP kit (Upstate) according to the manufacturer's instructions. Cellular lysates were subjected to five sets of sonication on wet ice with a 60 Sonic Dismembrator (Fisher Scientific). Each set consisted of 8 seconds of sonication separated by 1-minute intervals on ice. ARID1A and V5 antibodies (5 μL/reaction) were used for IP. The ChIP primers used to analyze proteins binding at DSBs were primer 1, 5′-TACGGCAAGCTGACCCTGAA-3′ (sense) and 5′-GCCCATATATGGAGTTCCGC-3′ (antisense); primer 2, 5′-GCCCATATATGGAGTTCCGC-3′ (sense) and 5′-GCCCATATATGGAGTTCCGC-3′ (antisense); primer 3, 5′-CAGGCAACTCTGCTCACTCA-3′ (sense) and 5′-GGCTATTGCTATAGGGTCCG-3′ (antisense); primer 4, 5′-GAGGCACCTGGAGCTGAG-3′ (sense) and 5′-GCTGAACTTGTGGCCGTTTA-3′ (antisense); and primer 5, 5′-GAGAAAGGTGCCGAAGAAGC-3′ (sense) and 5′-GTAGCAGATGGGGCAGGTG-3′ (antisense).

Male athymic nu/nu mice (6–8 weeks old) were used for all in vivo xenograft studies. Mice were quarantined for at least 1 week before experiments. All animal studies were conducted in compliance with protocols approved by the MDACC Institutional Animal Care and Use Committee. Exponentially growing MDA-MB-231 (1 × 10⁶) or HCT116 (2 × 10⁶) cells were implanted subcutaneously at the flank of nude mice (left: control cells; right: ARID1A-deficient cells). Mice were treated with vehicle or BMN673 (0.33 mg/kg) once daily by oral gavage. Tumors were measured every 2 days by caliper to determine tumor volume using the formula [length/2] × [width²]. Each cell line was tested in six different animals. For immunohistochemistry, tumor tissue samples were fixed in 4% buffered paraformaldehyde and processed for histopathologic evaluation by paraffin embedding and antibody staining.

Methods for the following assays are available in the supplementary information: KillerRed System, Flow Cytometric Analyses, Comet assay, Colony-forming assay, and Tet-On expression of ARID1A.

All statistical analyses were performed with a two-tailed Student t test.

G.B. Mills has received PI sponsored research clinical trial support from AstraZeneca and clinical trial support from BioMarin, has licensed a patent to Myriad Genetics, and is an advisory board member for AstraZeneca. No potential conflicts of interest were disclosed by the other authors.

Conception and design: J. Shen, W. Zhang, P. Kapoor, I.-M. Shih, X. Wu, X. Shen, G. Peng

Development of methodology: J. Shen, W. Zhang, I.-M. Shih, X. Wu

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): J. Shen, Y. Peng, L. Wei, L. Yang, L. Lan, I.P. Uray, X. Wu

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): J. Shen, L. Wei, L. Yang, Z. Ju, Q. Mo, G. Peng

Writing, review, and/or revision of the manuscript: J. Shen, L. Lan, P.H. Brown, X. Shen, G. Peng

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): J. Shen, G. Peng

Study supervision: G.B. Mills, G. Peng

The authors thank Dr. Shiaw-Yih Lin (MD Anderson Cancer Center) and Dr. Grzegorz Ira (Baylor College of Medicine) for scientific discussions; Dr. Jeremy Stark (Beckman Research Institute of the City of Hope) and Dr. Lee Zou (Massachusetts General Hospital Cancer Center, Harvard Medical School) for reagents; and S. Deming for proofreading the manuscript.

This work was supported in part by Cancer Center Support Grant CA016672 from the National Cancer Institute (NCI); the NCI Ovarian SPORE (MDACC) Career Development Award, P50 CA83639 (to G. Peng); NCI grant R00 CA149186 (to G. Peng); P50 CA098258 and P50 CA083639 (to G.B. Mills); and the Adelson Medical Research Foundation (00000205 and 00000196; to G.B. Mills).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.