Targeting the IRE1α/XBP1 Endoplasmic Reticulum Stress Response Pathway in ARID1A-Mutant Ovarian Cancers

ARID1A epigenetically regulates gene expression via the SWI/SNF chromatin-remodeling complex by controlling gene accessibility (1, 2). SWI/SNF complexes contribute to both gene activation and repression in a context-dependent manner (3). ARID1A has one of the highest mutation rates across many cancer types (4–6). In fact, on the basis of statistical saturation analyses, ARID1A is among the top 10 most mutated genes and is the most frequently mutated epigenetic regulator across all human cancers (5, 6). Notably, ARID1A is mutated in >50% of ovarian clear cell carcinoma (OCCC; refs. 7–9). Over 90% of the ARID1A mutations observed in epithelial ovarian cancer are frameshift or nonsense mutations that result in the loss of ARID1A protein expression (7, 9, 10). The loss of ARID1A correlates with late-stage disease and predicts early recurrence (11). OCCC is generally refractory to platinum-based chemotherapy, and when diagnosed at advanced stages, carries the worst prognosis among all histosubtypes of ovarian cancer (12). Therefore, there is an even greater need for novel therapeutic approaches that are selective for ARID1A-mutated ovarian cancer.

Upon detecting endoplasmic reticulum (ER) stress, the unfolded protein response (UPR) orchestrates adaptive programs to promote cancer cell survival (13–15). Thus, inhibition of the UPR represents a therapeutic approach for cancers with hyperactive ER stress response (15). The mammalian UPR is governed by three stress transducers that sense ER stress. They include inositol-required enzyme alpha (IRE1α), activating transcription factor 6 (ATF6), and protein kinase RNA-like ER kinase (PERK; ref. 15). IRE1α signaling is the most conserved and well-studied UPR. In response to ER stress, IRE1α signaling involves a conformational change that activates its RNase domain (15). IRE1α processes the mRNA encoding the transcription factor X-box–binding protein 1 (XBP1) by excising a 26-nucleotide intron in the XBP1 mRNA (15). This splicing event shifts the coding reading frame, leading to the translation of a transcription factor termed XBP1s to promote cell survival by resolving ER stress (15). However, the role of SWI/SNF complex in regulating ER stress response has never been explored.

Therapeutic resistance typically enables cancer cells to escape the effects of single-agent treatment. Combinatorial therapeutic strategies offer a solution for this major challenge (16). Histone deacetylase 6 (HDAC6) is a class IIb HDAC isoenzyme (17). Unlike other HDACs, HDAC6 primarily functions in the cytoplasm and targets non-histone proteins (18). Notably, HDAC6 regulates the degradation of misfolded proteins through aggresomes and cells deficient in HDAC6 are hypersensitive to the accumulation of unfolded proteins (19). HDAC6 inhibition is selective against ARID1A inactivation and inhibition of HDAC6 activity using the clinically applicable small-molecule inhibitor ACY1215 reduced the tumor burden of established ARID1A-mutant, but not wild-type, ovarian tumors (20). However, whether HDAC6 inhibition synergizes with ER stress response inhibition in ARID1A-mutated cancers has never been explored. Here, we show that ARID1A transcriptionally represses the IRE1α–XBP1 axis of the ER stress response and pharmacological inhibition of the IRE1α–XBP1 pathway alone or in combination with HDAC6 inhibition represents an urgently needed therapeutic strategy for ARID1A-mutant OCCCs.

The ovarian cancer cell lines OVCA429 (RRID:CVCL_3936), OVTOKO (JCRB Cat# NIHS0301, RRID:CVCL_3117), SKOV3 (JCRB Cat# NIHS0737, RRID:CVCL_4Y20), OVISE (JCRB Cat# JCRB1043, RRID:CVCL_3116), and TOV21G (ATCC Cat# CRL-11730, RRID:CVCL_3613) were cultured in RPMI-1640 supplemented with 10% FBS at 37°C and supplied with 5% CO₂. RMG1 (JCRB Cat# JCRB0172, RRID:CVCL_1662) and KK (RRID:CVCL_F844) cell lines were cultured in 1:1 DMEM/F12 supplemented with 10% FBS at 37°C and supplied with 5% CO₂. Endogenously FLAG-tagged ARID1A and ARID1A knockout RMG1 cells were constructed and cultured as we previously published (21). ARID1B knockout RMG1 cells were generated and cultured as we previously published (22). Primary OCCC cells (VOA4841 and XVOA295) were produced as described previously (20) and cultured in RPMI-1640 supplemented with 10% FBS at 37°C and supplied with 5% CO₂. The protocol for using primary cultures of human ovarian clear cell tumor cells was approved by the University of British Columbia Institutional Review Board (H18-01652). Written informed consent was obtained from human participants. All relevant ethical regulations have been complied with. OVTOKO, SKOV3, OVISE, and RMG1 were purchased from JCRB. TOV21G was purchased from the ATCC. OVCA429 and KK cell lines were obtained from Dr. I.-M. Shih (Johns Hopkins University, Baltimore, MD). Monthly Mycoplasma testing was performed on all cell lines using the LookOut Mycoplasma PCR Detection Kit (Sigma). All cell lines used for experimentation were passaged less than 30 times and authenticated at the Wistar Institute's Genomics Facility using short tandem repeat DNA profiling.

Antibodies against rabbit anti-ARID1A (3 μg ChIP, abcam, 182560), rabbit anti-ARID1A (1:1,000 immunoblot, Cell Signaling, 12354, RRID:AB_2637010), mouse anti-RNA Pol II (2 μg ChIP, Santa Cruz Biotechnology, sc-47701x, RRID:AB_677353), mouse anti-FLAG (3 μg ChIP, 1:1,000 immunoblot, Millipore Sigma, F3165, RRID:AB_259529), mouse anti–β-Actin (1:10,000 immunoblot, Millipore Sigma, A1978, RRID:AB_476692), rabbit anti-ARID1B (1:1,000 immunoblot, abgent, AT1190a, RRID:AB_1551334), rabbit anti-XBP1 (1:1,000 immunoblot, Cell Signaling Technology, 12782, RRID:AB_2687943), rabbit anti-SNF5 (3 μg ChIP, Bethyl, A301-087A, RRID:AB_2191714), rabbit anti-cleaved PARP (1:1,000 immunoblot, Cell Signaling Technology, 5625, RRID:AB_10699459), rabbit anti-c-MYC (1:1,000 immunoblot, Cell Signaling Technology, 5605, RRID:AB_1903938), mouse anti-Ki67 (1:1,000 IHC, Cell Signaling Technology, 9449, RRID:AB_2797703), rabbit anti-cleaved caspase-3 (1:200 IHC, 1:1,000 immunoblot, Cell Signaling Technology, 9661, RRID:AB_2341188), mouse anti-ATF6 (1:1,000 immunoblot, Abcam, 122897, RRID:AB_10899171), rabbit anti-PERK (1:1,000 immunoblot, Cell Signaling Technology, 3192, RRID:AB_2095847), and rabbit anti-ATF4 (1:1,000 immunoblot, Cell Signaling Technology, 11815, RRID:AB_2616025), rabbit anti–acetylated-α-Tubulin (1:1,000 immunoblot, Cell Signaling Technology, 5335, RRID:AB_10544694), and rabbit anti-CHOP (1:1,000 immunoblot, Cell Signaling Technology, 2895, RRID:AB_2089254) were purchased commercially. Rabbit anti–phospho-PERK antibody was kindly provided by Dr. C. Koumenis at the University of Pennsylvania, Philadelphia, PA. Tunicamycin (3516) and B-I09 (6009) were purchased from Tocris Bioscience. Thapsigargin (T9033), GSK2656157 (504651) and MG-132 (C2211) were purchased from Sigma-Aldrich. CeapinA7 (2323027-38-7) was purchased from MedChemExpress. 4μ8c (B1874) was purchased from APExBIO. ACY1215 (S8001) was purchased from Selleck-chem. B-I09 was provided by Dr. C.-C. Andrew Hu and synthesized as previously published (23).

Cells were lysed using RIPA buffer (1% NP-40, 0.5% sodium deoxycholate, 0.1% SDS, 150 mmol/L NaCl, 1 mmol/L EDTA, and 10 mmol/L Tris-HCl, pH 7.4) containing Roche complete protease inhibitors. Protein concentrations were determined by Bradford assays (Bio-Rad). Samples were boiled in SDS-PAGE sample buffer (2% SDS, 200 mmol/L sucrose, 2 mmol/L EDTA, 0.1% bromophenol blue, and 62.5 mmol/L Tris-HCl, pH 6.9) with β-mercaptoethanol and proteins were resolved by SDS-PAGE. Resolved proteins were transferred to nitrocellulose (Bio-Rad) or polyvinylidene fluoride membranes (Millipore) and blocked in 5% w/v nonfat milk in TBS-T. Immunoblotting was performed with the indicated primary antibodies and their appropriate horseradish peroxidase (HRP)–conjugated secondary antibodies. Immunoblots were developed using SuperSignal West Pico PLUS and femto chemiluminescent substrates (Thermo Fisher Scientific).

Cells were cultured until 70%–80% confluency and then 1% formaldehyde crosslinker was added for 10 minutes at room temperature. Formaldehyde reactions were then quenched by 0.125 mol/L glycine for 5 minutes. Next, cells were washed with PBS once and then collected into appropriate conical tubes. The samples were centrifuged at 400 relative centrifugal force (RCF) and PBS was aspirated away from the cell pellet. Then an appropriate volume of PBS was added to the cell pellet to transfer the fixed cells to a 1.5 mL microcentrifuge tube. The sample was centrifuged again at 400 RCF and the PBS was aspirated away from the cell pellet before cell lysis. Fixed cells were lysed using chromatin immunoprecipitation (ChIP) lysis buffer 1 (50 mmol/L HEPES-KOH (pH 7.5), 140 mmol/L NaCl, 1 mmol/L EDTA (pH 8.0), 1% Triton X-100, and 0.1% DOC) on ice for 10 minutes, centrifuged at 1,000 RCF and then supernatant was aspirated. Then, ChIP lysis buffer 2 (10 mmol/L Tris (pH 8.0), 200 mmol/L NaCl, 1 mmol/L EDTA, and 0.5 mmol/L EGTA) was added at room temperature for 10 minutes, centrifuged at 1,000 RCF and then supernatant was aspirated. Chromatin was digested with MNase (Cell Signaling Technology, 10011) in digestion buffer (10 mmol/L Tris 8.0, 1 mmol/L CaCl₂, and 0.2% Triton X-100) at 37°C for 15 minutes. The sample was then centrifuged ≥16,000 RCF at 4°C for 1 minute. The supernatant was then extracted to a new 1.5 mL microcentrifuge tube, NaCl was added to 100 mmol/L, and placed on ice. The remaining nuclear pellet was disrupted by two pulses of the Bioruptor (Diagenode) under high output in 0.4 mL of ChIP lysis buffer 3 (10 mmol/L Tris-HCl (pH 8.0), 100 mmol/L NaCl, 1 mmol/L EDTA, 0.5 mmol/L EGTA, 0.1% ODC, and 0.5% N-lauroylsarcosine). The lysed nuclear pellet sample was then combined with the supernatant from the previous step and centrifuged at ≥16,000 RCF at 4°C for 10 minutes. The supernatant was transferred to a new 1.5 mL microcentrifuge tube and DNA was quantified using the NanoDrop (ThermoFisher Scientific). All samples contained the same amount of DNA and antibody per immunoprecipitation and were compared with IgG isotype controls. 1% input was taken for ChIP-qPCR quantitation. Samples were incubated on a tube rocker over night at 4°C with the appropriate antibody. The following day, protein A/G magnetic Dynabeads (ThermoFisher Scientific, 10002D/10003D) were added to each sample and incubated at 4°C for 1 hour on the tube rocker. Then, all samples were washed twice with ChIP lysis buffer 1, ChIP lysis buffer 1 with 0.65 mol/L NaCl, and wash buffer [10 mmol/L Tris-HCl (pH 8.0), 250 mmol/L LiCl, 0.5% NP-40, 0.5% DOC, and 1 mmol/L EDTA (pH 8.0)]. TE [50 mmol/L Tris-HCl (pH 8.0), 10 mmol/L EDTA (pH 8.0)] was added to the samples and the beads were transferred to a new tube to reduce background signal. Lastly, samples (including 1% input) were eluted with TE supplemented with 1% SDS on a ThermoMixer (Eppendorf) at 65°C at 1,000 RPM for 15 minutes. Crosslinks were then reversed by adding NaCl and proteinase K to final concentrations of 200 mmol/L and 0.3 μg/μL, respectively. Samples were incubated at 65°C for 3 hours, cooled, and DNA was purified using the ChIP DNA Clean and Concentrator Kit (Zymo Research). Purified ChIP DNA was analyzed by qPCR using iTaq Universal SYBR Green Supermix (Bio-Rad).

For ChIP-qPCR the following primers were used:

XBP1 locus forward: 5′-CGACCTCATGTCCGAGTTAAG-3′ and reverse 5′-ACTCTCTCGTTAGAGATGACCA-3′; ERN1 locus forward: 5′-CAGGGCAAGTGGCAGAA-3′ and reverse 5′-GCGCTTCGAATCCTTGTTTG-3′.

Total RNA was isolated using TRizol reagent (Invitrogen). Complimentary DNA was produced using iTaq Universal SYBR Green One-step kit (Bio-Rad). RNA expression was determined using the QuantStudio 3 Real-Time PCR system (ThermoFisher Scientific) with the following sets of primers: hERN1 forward 5′-CCCATCAACCTCTCTTCTGTATC-3′ and reverse 5′-AGGCCGCATAGTCAAAGTAG-3′; hXBP1s forward 5′-CCGCAGCAGGTGCAGG-3′ and reverse 5′-GAGTCAATACCGCCAGAATCCA-3′; hXBP1u forward 5′-GCGCTGTCTTAACTCCTGGT-3′ and reverse 5′-GCCTCTTATGAACTTTCTTCCAG-3′; h18s forward 5′-AACTTTCGATGGTAGTCGCCG-3′ and reverse 5′-CCTTGGATGTGGTAGCCGTTT-3′. 18s expression was used as an internal control. To determine Xbp1 mRNA expression in mice tumor samples the following sets of primers were used: mXbp1 forward 5′-GCAGACTGCTCGAGATAGAAAG-3′ and reverse 5′-AGCTGGAGTTTGTGGTTCTC-3′; mβ-Actin forward 5′-GAGGTATCCTGACCCTGAAGTA-3′ and reverse 5′-CACACGCAGCTCATTGTAGA-3′.

Cells were seeded in 24-well plates according to their growth rates. Fresh culture media containing the appropriate drug concentration were added to each well. Media containing the appropriate drug concentration were changed every 3 days up to 12 days. On the final day, cells were washed once with PBS, fixed, and stained with a 0.05% crystal violet, 10% methanol solution for 10 minutes. To calculate IC₅₀ values, integrated density was determined using NIH ImageJ software (version 1.52v).

Tissue samples were paraffin embedded and sliced onto glass microscopy slides by the Histotechnology Facility at The Wistar Institute. Immunohistochemical staining was performed on consecutive tissue sections. Using the EnVision+ HRP and peroxidase (DAB) systems (DAKO Corporation) protein tissue expression was determined following the manufacturer's instructions. Stained proteins were scored using histological scoring (H score). The H score was calculated upon the intensity and frequency of the stain.

2-mm tips of mice tails were digested by proteinase K and DNA was purified. Purified mouse tail DNA was subjected to PCR using Platinum Hot start PCR Master Mix (2x; ThermoFisher, 13000012). The PCR cycle parameters are as follows for the indicated genes:

loxP-Xbp1: (i) 94°C for 4 minutes, (ii) 94°C for 30 seconds, (iii) 60°C for 30 seconds, (iv) 72°C for 30 seconds, (v) Go to step 2, 34x, (vi) 72°C for 2 minutes, (vii) 12°C infinitely; loxP-Arid1a and wild-type/mutant Pik3ca: (i) 94°C for 4 minutes, (ii) 94°C for 30 seconds, (iii) 55°C for 30 seconds, (vi) 72°C for 1 minute, (v) Go to step 2, 34x, (vi) 72°C for 5 minutes, (vii) 12°C infinitely. The following primer sets were used to amplify the indicated genes: Xbp1: forward 5′-ACTTGCACCAACACTTGCCATTTC-3′ and reverse 5′-CAAGGTGGTTCACTGCCTGTAATG-3′; Arid1a: forward 5′-GTAATGGGAAAGCGACTACTGGAG-3′ and reverse 5′-TGTTCATTTTTGTGGCGGGAG-3′; Pik3ca: forward 5′-AAAGTCGCTCTGAGTTGTTAT-3′ and reverse wild type 5′-GGAGCGGGAGAAATGGATATG-3′ and reverse mutant 5′-GCGAAGAGTTTGTCCTCAACC-3′.

All experimental protocols were approved by the Wistar Institutional Animal Care and Use Committee. All mice were maintained in specific pathogen-free barrier facilities. To establish patient-derived xenograft (PDX) tumors, patient samples were acquired from Christiana Hospital and deidentified as we previously described (22, 24). Samples were cut into small pieces and implanted intrabursally into immunocompromised 6–10-week-old female mice. Tumors were allowed to grow, be removed, and redistributed intrabursally among mice for multiple generations. Once established and validated by genomic sequencing to carry ARID1A frame-shift mutations, which result in loss of expression, tumors were removed and distributed intrabursally amongst 6 mice per group. 6 months after intrabursal distribution the tumors were large enough to be randomized into two groups. Mice were treated with vehicle control (DMSO) or B-I09 (50 mg/kg daily, i.p.) for two weeks.

To establish cell line–derived tumors, 1 × 10⁶ RMG1 or TOV21G cells were unilaterally injected into the ovarian bursa sac of 6–10-week-old female mice (n = 10 per group; ref. 25). Two weeks after intrabursal injection, mice were randomized into two groups and treated with vehicle control (DMSO) or B-I09 (50 mg/kg daily, i.p.) for two weeks. At the end of both experiments, ovarian tissue was surgically removed, and tumor burden was assessed based upon tumor weight.

Transgenic mice with latent mutations in Arid1a and Pik3ca were generated by crossing Arid1a^flox/flox mice (kindly provided by Dr. Wang, University of Michigan and crossed onto a C57BL/6J background for 9 generations) with R26-Pik3ca^H1047R mice carrying inducible Pik3ca mutations (The Jackson Laboratory, Jax#016977) as we and others previously published (20, 26). In these models, adenovirus-Cre intrabursal injection induces OCCC in approximately 45 days. To achieve the Arid1a^flox/flox/Pik3ca^H1047R/Xbp1^flox/flox genotype, Xbp1^flox/flox mice (kindly provided by Dr. C.-C. Andrew Hu; ref. 23) were crossed with the Arid1a^−/−/Pik3ca^H1047R mice. All mice were genotyped before experimentation to validate their genotypes. To induce tumor growth 6–8-week-old female mice were injected intrabursally with adenovirus-Cre. For drug treatment studies, mice were randomized before treatment with B-I09 (50 mg/kg) or vehicle control (DMSO). For synergy studies between B-I09 and ACY1215 mice were treated with 25 mg/kg of each compound. Mice were treated for five-day intervals with two days of rest between treatment intervals up to twenty-one days. Following drug treatment mice were sacrificed, and reproductive tracts were removed. Tumor weights of the injected ovary were weighed and reported.

RNA was extracted with TRIzol (Invitrogen) for all RNA preparations, then subsequently cleaned and DNase treated using RNeasy columns (Qiagen). DNase-treated RNA was subjected to library preparation. Libraries for RNA-sequencing (RNA-seq) were prepared with ScriptSeq complete Gold kit (Epicenter) and subjected to a 75 bp paired-end sequencing run on NextSeq 500, using Illumina's NextSeq 500 high-output sequencing kit following the manufacturer's instructions.

RNA-seq data were aligned using bowtie2 (RRID:SCR_016368; ref. 27) against the hg19 version of the human genome, and RSEM v1.2.12 software (RRID:SCR_013027; ref. 28) was used to estimate raw read counts and Reads Per Kilobase of transcript, per Million (RPKM) using the Ensemble transcriptome. DESeq2 (RRID:SCR_015687; ref. 29) was used to estimate the significance of differential expression between group pairs. Overall gene expression changes were considered significant if they passed FDR thresholds of <5%.

For analysis of previously published ARID1A ChIP-seq (30), genes that had ARID1A peaks within 1,500 base pairs from the transcription start site (TSS) were considered. Significance of overlap was tested using hypergeometric test using 22, 184 Ensemble genes with detected expression in the RNA-seq experiment as a population size. Gene set enrichment analysis of gene sets was done using QIAGEN's Ingenuity Pathway Analysis software (IPA, QIAGEN Redwood City, www.qiagen.com/ingenuity, RRID:SCR_008653) using “Diseases and Functions” options and FDR <5% results were considered significant.

Statistical analyses were performed using GraphPad Prism 8 (GraphPad, RRID:SCR_002798) for Mac OS. Quantitative data are expressed as mean ± SD unless otherwise stated in figure legends. Drug synergy analyses were conducted by Qin Liu within the Biostatistics Unit of the Wistar Institute. ANOVA with Fisher's least significant difference was used to identify significant differences in multiple comparisons. For all statistical analyses, the level of significance was set at 0.05.

To evaluate whether there is an overall significant synergistic effect and at which dose levels the combination could reach significant synergistic effect, we calculated Interaction index with 95% confidence interval (CI) in overall as well as at various doses of each studied drug when it was combined with the other drug using Bliss independence model (31). Interaction index <1 indicates synergistic effect, and the vertical bar below the line of 1.0 indicates significant synergistic effect.

The following previously published datasets were used in the presented studies: ARID1A ChIP-seq: GSE104545 (30); RNA Pol II ChIP-seq: GSE120060 (21) and GSE106665 (32); ATAC-seq: GSE101966 (33), GSE124224 (34) and GSE106665 (32); and gene expression microarray: GSE6008 (35) and GSE29450 (36). The newly generated RNA-seq datasets were deposited into the Gene Expression Omnibus (GEO, RRID:SCR_005012) database can be accessed by the accession number GSE180468.

To determine whether ARID1A regulates ER stress response genes, we profiled changes in gene expression by RNA-seq in control and ARID1A-knockout RMG1 cells treated with or without the ER stress inducer Tunicamycin. In addition, we cross-referenced an ARID1A ChIP-seq dataset in ARID1A wild-type RMG1 cells that we previously published (21) with the RNA-seq datasets (Fig. 1A). Notably, ARID1A knockdown does not affect cell growth (20, 37). The ARID1A-containing SWI/SNF complex either activates or represses gene expression in a context-dependent manner (3). Compared with its function as a transcriptional activator, ARID1A's role in repressing its target genes is less well understood. Thus, we focused our analysis on genes that are upregulated by ARID1A knockout. The analysis revealed that the ER stress response was significantly enriched among ARID1A-regulated direct target genes (Fig. 1A and Supplementary Fig. S1A). Notably, inactivation of another SWI/SNF subunit SMARCB1, also known as SNF5, in malignant rhabdoid tumor activates the UPR response through upregulating the MYC protein (38). However, MYC expression was not affected by ARID1A knockout (Supplementary Fig. S1B). This suggests new mechanisms underlying the observed activation of the ER stress response/UPR in ARID1A knockout cells.

We next sought to validate these findings. We showed that there was an increase in phosphorylated PERK and spliced XBP1 by ARID1A knockout (Supplementary Fig. S1C). Similar observations were also made in ARID1B knockout RMG1 cells compared with controls (Supplementary Fig. S1D). This suggests that the observed changes may be SWI/SNF-complex dependent. In contrast, ARID1A knockout did not increase the expression of either ATF4 or ATF6 (Supplementary Fig. S1C). Given the fact that IRE1α/XBP1 signaling is the most conserved UPR pathway and XBP1 plays a critical role in mediating the UPR response (15), we focused our analyses on XBP1. Upregulation of both spliced and unspliced forms of XBP1 were confirmed at both mRNA and protein levels with or without ER stress inducers such as Tu, MG-132 and Thapsigargin (Fig. 1B–D, Supplementary S1E–S1F). Similar observations were also made in ARID1A wild-type control and the matched isogenic ARID1A knockout OVCA429 OCCC cells (Supplementary Fig. S1G–S1H). This suggests that these findings are not cell line specific. We next directly correlated the ARID1A mutational status with XBP1 mRNA expression in The Cancer Genome Atlas (TCGA) cancer types such as uterine corpus endometrial carcinomas that display high ARID1A mutation frequencies (∼30% cases with ARID1A mutation), as the TCGA does not have OCCC datasets. Compared with ARID1A wild-type tumors, XBP1 mRNA was expressed at significantly higher levels in ARID1A-mutant tumors (Fig. 1E). Notably, ARID1A mutation correlates with an increase in expression of the majority of canonical IRE1α–XBP1s target genes identified in a previous study (Supplementary Table S1; ref. 39). Likewise, in two independently published microarray datasets, compared with normal ovary or ovarian surface epithelium, expression of the majority of identified ARID1A-regulated direct genes implicated in ER stress responses was higher in OCCCs (Supplementary Fig. S1I–S1J; refs. 35, 36).

Our analysis of previously published ChIP-seq datasets (30) revealed that ARID1A directly bound to the promoter regions of the ERN1 and XBP1 genes (Fig. 2A and Supplementary S2A). Consistent with the notion that ARID1A functions as a transcriptional repressor in this context, ARID1A knockout increased the expression and the association of RNA polymerase II (RNA Pol II) with the promoters of both XBP1 and ERN1 genes in our previously published datasets (Figs. 1B and 2B; Supplementary S2B–S2C; ref. 21). We validated the association of ARID1A and SNF5, a core subunit of the SWI/SNF complex, with the promoters by ChIP-qPCR analyses. These associations were reduced by ARID1A knockout, indicating the specificity of our analyses (Fig. 2C). To further validate these findings, we tagged the endogenous ARID1A locus with a FLAG epitope using CRISPR (Fig. 2D). ChIP-qPCR analyses in these cells revealed an association of FLAG-tagged ARID1A with both XBP1 and ERN1 promoters (Fig. 2E and Supplementary S2D). Consistently, data mining of the published assay for transposase-accessible chromatin using sequencing (ATAC-seq) revealed that ARID1A knockout increased the accessibility to the promoters of the XBP1 and ERN1 genes (Supplementary Fig. S2E–S2F; refs. 32–34). Likewise, the associations of RNA Pol II with the promoters of the XBP1 and ERN1 genes were increased by ARID1A knockout (Supplementary Fig. S2G; refs. 21, 32). Together, we conclude that ARID1A represses the IRE1α/XBP1 pathway of the ER stress response/UPR.

Because we show that ARID1A inactivation upregulates the IRE1α/XBP1 pathway, we next sought to determine whether inhibition of the IRE1α/XBP1 pathway is selective against ARID1A-inactivated cells. Toward this goal, we treated control and ARID1A knockout RMG1 cells with a selective IRE1α RNase inhibitor B-I09 (23). We chose B-I09 for our study because of its ability to specifically target IRE1α RNase activity and its strong safety profile in vivo in preclinical studies (23, 40–42). Compared with controls, ARID1A knockout decreased the IC₅₀ value of B-I09 in RMG1 cells (Fig. 3A and B). Consistent with the notion that unresolved ER stress leads to apoptosis, B-I09 induces apoptosis in ARID1A knockout RMG1 cells in a dose-dependent manner as evidenced by upregulation of apoptotic markers such as cleaved PARP p85 and cleaved caspase-3 (Fig. 3C). As a control, B-I09 was not effective in inducing apoptosis in ARID1A wild-type RMG1 controls (Fig. 3C). Similarly, compared with ARID1A wild-type RMG1 cells, ARID1A-mutant TOV21G OCCC cells were sensitive to apoptotic induction by B-I09 (Supplementary Fig. S3A). Likewise, B-I09 is more effective in inducing apoptosis in ARID1A-mutant compared with wild-type primary OCCC cultures (Fig. 3D). In addition, in a panel of OCCC cells, the IC₅₀ value of B-I09 was significantly lower in ARID1A-mutant cells (OVISE, SKOV3, OVTOKO, and TOV21G) compared with ARID1A wild-type cells (RMG1, OVCA429, and KK; Fig. 3E). Similar findings were also made using another IRE1α RNase inhibitor 4μ8c (Supplementary Fig. S3B). In contrast, there was no statistical difference in IC₅₀ value of the PERK inhibitor GSK2656157 (43) or ATF6 inhibitor CeapinA7 (44, 45) between ARID1A-mutant and wild-type cell lines (Supplementary Fig. S3C and S3D). Together, these findings support that ARID1A-inactivated cells are selectively sensitive to inhibition of the IRE1α/XBP1 pathway by inducing apoptosis.

Because we show that inhibition of the IRE1α/XBP1 pathway suppresses the growth of ARID1A-mutant OCCC cells, we sought to determine whether inactivation of the XBP1 pathway is tumor suppressive in ARID1A inactivated OCCC in vivo. Toward this goal, we crossed Arid1a^flox/flox/Pik3ca^H1047R genetic OCCC model (20, 46) with a conditional Xbp1^flox/flox model (23) to generate the Arid1a^flox/flox/Pik3ca^H1047R/Xbp1^flox/flox model (Supplementary Fig. S4A). To inactivate Arid1a and Xbp1, and induce Pik3ca^H1047R expression, adeno-cre was intrabursally administered as previously reported (20). To determine the effects of Xbp1 knockout on tumor burden, we measured the tumor weight 6 weeks after tumor induction (Supplementary Fig. S4B and S4C). We show that Xbp1 knockout significantly decreased the weight of tumors developed in the Arid1a^−/−/Pik3ca^H1047R model (Fig. 4A and B). Consistently, Xbp1 knockout significantly improved the survival of mice bearing the Arid1a/Pik3ca^H1047R tumors (Fig. 4C). Together, we conclude that Xbp1 knockout reduces the tumor burden and improves the survival of mice bearing Arid1a-inactivated OCCCs.

We next sought to determine the therapeutic potential of targeting the IRE1α/XBP1 pathway in ARID1A-mutant tumors. Toward this goal, we used three different mouse models. First, we used orthotopic xenograft models formed by ARID1A-mutant TOV21G OCCC cells. Briefly, the orthotopically transplanted cells were allowed to grow for one week to establish the orthotopic tumors (Supplementary Fig. S5A). Mice were then randomized and treated for two weeks with vehicle control or B-I09 (50 mg/kg, i.p.), the same dose as previously reported (23). We used tumor weight as a surrogate for tumor burden. Our results show that the B-I09 treatment significantly reduced the burden of orthotopic xenografts formed by ARID1A-mutant cells (Fig. 5A and B). Notably, the observed tumor-suppressive effects by B-I09 treatment are ARID1A status dependent. For example, B-I09 did not significantly affect the growth of tumors formed by ARID1A wild-type control RMG1 cells (Supplementary Fig. S5B and S5C). Furthermore, B-I09 significantly reduced the expression of the cell proliferation marker Ki67 in tumors formed by ARID1A-mutant TOV21G but not ARID1A wild-type RMG1 cells (Fig. 5C and D; Supplementary S5D and S5E). In contrast, expression of the apoptosis marker cleaved caspase-3 was induced by B-I09 treatment in tumors formed by ARID1A-mutant TOV21G cells but not ARID1A wild-type RMG1 cells (Fig. 5C and D; Supplementary S5D and S5E).

We next sought to expand these studies into ARID1A-mutant OCCC patient-derived xenografts (Supplementary Fig. S5F and S5G). B-I09 significantly reduced the tumor burden in ARID1A-mutated OCCC PDXs (Fig. 5E and F). Likewise, B-I09 significantly reduced the tumor burden in Arid1a-inactivated OCCCs developed in the genetic Arid1a^−/−/Pik3ca^H1047R models (Fig. 5G and H; Supplementary S5H). The reduction in tumor burden by B-I09 treatment correlated with an improvement of the survival of tumor bearing mice (Fig. 5I). To validate the in vivo on target effects, we examined the protein expression of spliced Xbp1 (Xbp1s) in vehicle control and B-I09–treated mice by immunoblotting. B-I09 treatment decreased Xbp1s protein expression in vivo (Fig. 5J and K). This result confirmed the on-target effects of B-I09. Consistent with previous reports, B-I09 was well tolerated in vivo. For example, B-I09 treatment did not affect the body weight of the treated tumor-bearing mice (Supplementary Fig. S5I). Thus, we conclude that the IRE1α RNase inhibitor B-I09 is effective in treating ARID1A-inactivated OCCCs.

Inhibition of HDAC6 activity is effective against ARID1A-mutant cancers (20). Notably, HDAC6 plays an important role in clearing misfolded proteins and inhibition of HDAC6 activity sensitizes cells to UPR (20, 47, 48). This raises the possibility that IRE1α and HDAC6 inhibitors may synergize with each other in suppressing ARID1A-inactivated cells. Toward testing this possibility, we performed a contour plot analysis using a serial dilution of the IRE1α RNase inhibitor B-I09 and the HDAC6 inhibitor ACY1215. We calculated an overall significant synergistic effect based on interaction index with 95% CI. The analysis indicated an overall significant synergistic effect in combining B-I09 and ACY1215 (Supplementary Fig. S6A). Consistently, B-I09 and ACY1215 combination is synergistic in inducing expression of apoptotic markers such as cleaved caspase-3 and cleaved PARP p85 in ARID1A-mutant TOV21G cells (Fig. 6A). Notably, compared with ARID1A wild-type RMG1 cells, the synergy in inducing expression of these apoptotic markers was enhanced by ARID1A knockout (Fig. 6B). This suggests that the combination is more effective in ARID1A-inactivated cells. Interestingly, ACY1215 increased phospho-PERK expression (Fig. 6C), which is consistent with the notion that HDAC6 inhibition increases ER stress. In addition, ACY1215 decreases XBP1s expression in ARID1A-inactivated cells (Supplementary Fig. S6B). These results are consistent with the notion that ACY1215 simultaneously increases ER stress and suppresses the XBP1s mediated ER-stress response.

We next sought to test the combination in vivo in the genetic Arid1a^−/−/Pik3ca ^H1047R mouse model. Notably, we reduced the dose of both B-I09 and ACY1215 to 25 mg/kg to allow for the observation of synergy in vivo. Under these conditions, although neither B-I09 nor ACY1215 alone demonstrated a significant inhibition on tumor growth compared with vehicle controls, the combination decreased tumor weight in Log10 U by 0.62 more than the sum of single-agent effects (Fig. 6D and E). From interaction effect analyses, the combination effect is significantly better in inhibiting tumor growth than the sum of single-agent effects (P = 0.002), which indicates a synergistic effect. Furthermore, the combination of B-I09 and ACY1215 improved the survival of tumor bearing mice significantly compared with individual treatments alone (Fig. 6F). The combinations at the doses used here were well-tolerated in vivo. For example, the combination treatment did not significantly affect the body weight of the treated tumor-bearing mice (Supplementary Fig. S6C). Consistently, the combination did not visibly change the histological morphology of liver and kidney (Supplementary Fig. S6D). Together, we conclude that the IRE1α inhibitor B-I09 synergizes with the HDAC6 inhibitor ACY1215 in suppressing the growth of ARID1A-inactivated OCCC to improve the survival of tumor-bearing mice.

Here, we show that ARID1A represses the IRE1α/XBP1 pathway of the ER stress response. This is consistent with the notion that ARID1A functions as a tumor suppressor and its inactivation promotes the survival of cancer cells by upregulating the ER stress response to adapt within the harsh tumor microenvironment. We speculate that the increase in the IRE1α–XBP1 signaling is cytoprotective due to its role in dealing with an overall increase in ER stress as evidenced by an increase in both IRE1α–XBP1 and PERK pathways of the ER stress response. Consequently, ARID1A loss creates a dependence on the upregulated IRE1α–XBP1 pathway and inhibition of this pathway is tumor suppressive in ARID1A inactivated tumors by decreasing their ability to adapt to ER stress.

Interestingly, ARID1A represses ERN1 and ARID1A inactivation upregulates ERN1 transcriptionally. In addition, ARID1A directly represses XBP1 transcription. Thus, ARID1A inactivation upregulates the IRE1α/XBP1 pathway by upregulating both the substrate (XBP1) and enzyme (IRE1α) that produces the XBP1s transcription factor. In addition to the IRE1α/XBP1 pathway, the PERK pathway was also activated by ARID1A knockout. However, there is no selectivity against ARID1A mutation among cell lines by PERK inhibition. Regardless, future studies will elucidate the contributions of the PERK pathway in promoting the survival of ARID1A-mutant cancer cells.

Our data show that ARID1A inactivation increases XBP1s expression, which correlates with the observed hypersensitivity to B-I09 treatment. Notably, HDAC6 inhibition reduces XBP1s expression in ARID1A inactivated cells. These results suggest a model whereby HDAC6 and IRE1α inhibition converges on the inhibition of XBP1s as a potential mechanism for the observed synergy. In addition, HDAC6 inhibition increases the expression of phospho-PERK, indicating that HDAC6 inhibition increases ER stress in ARID1A-inactivated cells. Thus, HDAC6 inhibition simultaneously increases ER stress and suppresses ER stress response via downregulating XBP1s.

Epigenetic regulators can control multiple specific pathways simultaneously. For example, ARID1A mutation inactivates the p53 tumor-suppressive pathway through upregulating HDAC6 (20), while increasing the IRE1α/XBP1 pathway as reported here. Indeed, a combination of the HDAC6 inhibition and IRE1α inhibition synergistically suppresses ARID1A-mutant cancers. Thus, epigenetic dysregulations are ideally suited for developing combinatorial therapeutic strategies to prevent and/or overcome therapy resistance. Notably, in addition to tumor intrinsic function, ER stress responses such as the IRE1α/XBP1 pathway are implicated in intratumoral immune cells (13, 14). For example, activation of the IRE1α/XBP1 pathway is known to be immune suppressive in various populations of immune cells (13). Thus, targeting ER stress responses may reinvigorate endogenous antitumor immunity, which could synergize with immunotherapies such as immune checkpoint blockade (13). Because ARID1A represses PD-L1 expression and ARID1A mutation correlates with an increase in PD-L1 expression (49, 50), it would be interesting to examine whether IRE1α inhibition also synergizes with immune checkpoint blockade in ARID1A-mutant cancers.

A limitation of our study is that in addition to regulating the UPR or ER stress response, HDAC6 is implicated in multiple biological processes and pathways, including protein trafficking and degradation, cell shape, and migration (18). Thus, in addition to regulating the ER stress response, other mechanisms regulated by HDAC6 may also contribute to the observed synergy between the HDAC6 inhibitor ACY1215 and the IRE1α inhibitor B-I09. In summary, our findings establish that pharmacological inhibition of the IRE1α/XBP1 pathway of the ER stress response alone or in combination with HDAC6 inhibitor represents a novel therapeutic strategy for ARID1A-mutated cancers.

C.C.A. Hu reports a patent for US10,323,013 issued. R. Zhang reports grants from NIH during the conduct of the study. No disclosures were reported by the other authors.

J.A. Zundell: Conceptualization, data curation, formal analysis, funding acquisition, validation, investigation, methodology, writing–original draft, writing–review and editing. T. Fukumoto: Data curation, formal analysis, investigation, writing–review and editing. J. Lin: Investigation. N. Fatkhutdinov: Investigation, writing–review and editing. T. Nacarelli: Investigation, writing–review and editing. A.V. Kossenkov: Formal analysis, investigation, methodology, writing–review and editing. Q. Liu: Formal analysis, methodology, writing–review and editing. J. Cassel: Data curation, investigation, methodology. C.C.A. Hu: Resources, supervision, writing–review and editing. S. Wu: Data curation, formal analysis, supervision, investigation, writing–original draft, writing–review and editing. R. Zhang: Conceptualization, supervision, funding acquisition, writing–original draft, project administration, writing–review and editing.

We thank X. Hua for assistance with intrabursal injection procedures, Dr. B. Keith for critical reading of the article, and Dr. D. Huntsman for providing the primary cultures of OCCCs. This work was supported by US National Institutes of Health grants (R01CA202919, R01CA239128, R01CA260661, and P50CA228991 to R. Zhang; R01CA163910 and R01CA190860 to C.C.A. Hu; and F31CA247336 to J.A. Zundell), US Department of Defense (OC180109 and OC190181 to R. Zhang), The Honorable Tina Brozman Foundation for Ovarian Cancer Research and The Tina Brozman Ovarian Cancer Research Consortium 2.0 (to R. Zhang), and the Ovarian Cancer Research Alliance [Collaborative Research Development Grant #596552 to R. Zhang and Ann and Sol Schreiber Mentored Investigator Award #649658 to J. Lin]. Support of Core Facilities was provided by Cancer Center Support Grant (CCSG) CA010815 to The Wistar Institute.

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.