Coexistent ARID1A-PIK3CA mutations promote ovarian clear-cell tumorigenesis through pro-tumorigenic inflammatory cytokine signaling

Ovarian clear-cell carcinoma (OCCC) is an aggressive form of ovarian cancer with high ARID1A mutation rates. Here we present a mutant mouse model of OCCC. We find that ARID1A inactivation is not sufficient for tumor formation, but requires concurrent activation of the phosphoinositide 3-kinase catalytic subunit, PIK3CA. Remarkably, the mice develop highly penetrant tumors with OCCC-like histopathology, culminating in hemorrhagic ascites and a median survival period of 7.5 weeks. Therapeutic treatment with the pan-PI3K inhibitor, BKM120, prolongs mouse survival by inhibiting tumor cell growth. Cross-species gene expression comparisons support a role for IL-6 inflammatory cytokine signaling in OCCC pathogenesis. We further show that ARID1A and PIK3CA mutations cooperate to promote tumor growth through sustained IL-6 overproduction. Our findings establish an epistatic relationship between SWI/SNF chromatin remodeling and PI3K pathway mutations in OCCC and demonstrate that these pathways converge on pro-tumorigenic cytokine signaling. We propose that ARID1A protects against inflammation-driven tumorigenesis.


Introduction
Epithelial ovarian cancer (EOC) ranks as the 5 th leading cause of cancer death among women 1 . EOC consists of four major histologic tumor subtypes (serous, clear-cell, endometrioid, or mucinous) and originates from the coelomic epithelium, which is the precursor to the ovarian surface epithelium (OSE) and Müllerian ductal epithelium 2 . The salpinge or fallopian tube of the Müllerian epithelial anlage is the more definitive source of serous EOC 3 . Endometriotic explants originating from more distal sites in the Müllerian epithelial anlage may have some propensity to give rise to clear cell or endometrioid EOC in humans, but this alternate mechanism of pathogenesis has not been thoroughly investigated 4,5 .
In addition to histologic feature, each EOC subtype can be further classified into two distinct groups on the basis of disease progression, metastatic potential, and molecular signature, with low-grade serous, clear-cell, mucinous and endometrioid subtypes being classified as Type I tumors and the high-grade serous subtype as Type II tumors 6 . In general, Type I tumors account for up to 25% of EOCs, are genomically stable and have lower metastatic potential than Type II tumors, which constitute most of the remaining EOC cases 6 . In addition, KRAS, BRAF, PIK3CA, PTEN, and CTNNB1 (β-CATENIN) mutations have historically been associated with Type I tumors at varying frequencies, whereas Type II tumors are largely defined by high frequency TRP53 (p53) mutations and chromosome instability 6,7,8 . Therefore, EOC consists of a heterogeneous group of cancers, each with unique clinical challenges and biology.
Ovarian clear-cell carcinoma (OCCC) is the most common Type I tumor, accounting for 5-25% of all EOC cases, with incidence varying between populations 9, 10, 11 . Endometriosis is recognized as a significant risk factor for OCCC 12,13,14 . Among all EOC subtypes, OCCC has the worst prognosis if diagnosed at an advanced stage of disease because of poor response rates to platinum-based chemotherapy 9,10,15,16 . Consequently, survival rates for women with advanced stage OCCC are low 9,10,15,16 . Although OCCC is the second leading cause of death from ovarian cancer, the etiology and pathogenesis of this devastating disease are poorly understood. together, components of the SWI/SNF complex are mutated in at least 20% of human cancer 19,21,22,23 . ARID1A is a subunit within the SWI/SNF chromatin remodeling complex that facilitates target substrate recognition, however, its role in OCCC tumor initiation and progression has yet to be fully elucidated 24 .
Genetically engineered mouse models offer the opportunity to investigate the contribution of genetic factors to EOC etiology and pathogenesis 25 . In this regard, early research using mouse models of EOC established causative roles for coexistent mutations in PTEN and KRAS or PTEN and WNT/β-CATENIN pathways in endometrioid EOC 7,8 . Recently, Guan et al. reported a mouse model of endometrioid EOC involving coexistent mutations in PTEN and ARID1A 26 . In this study, we describe a new mouse model that genetically and histologically resembles human OCCC and functionally implicates coexistent ARID1A and PIK3CA mutations in cancer causation. We further provide new mechanistic insight into OCCC pathogenesis by establishing a functional link between coexistent ARID1A-PIK3CA mutations and IL-6 signaling.

Generation of a novel Arid1a conditional allele
To establish a functional role for ARID1A mutations in ovarian clear-cell tumorigenesis, we generated a novel Arid1a conditional allele (Arid1a fl ) by inserting loxP sites flanking exons 5 and 6 in ARID1A (Fig. 1A,B). Arid1a-null embryos die around gastrulation, thus they are not amenable to efficient protein extraction 27 . In light of its early embryonic lethality, we monoalleliclly inactivated ARID1A in embryonic tissues (via Sox2Cre 28 ) to verify our Arid1a fl allele in the whole animal. Using this approach, we observed an expected 50% reduction in total protein from whole embryo lysates (Fig. 1C,D). We next sought to inactivate ARID1A in the OSE. To do this, we employed the intrabursal Adenovirus CREmediated (AdCRE) recombination system 7,29 . Efficient CRE-mediated genetic recombination was consistently observed in the OSE and outer epithelial layers of tissues confined to the bursal space ( Fig. 1F,H,I). We performed a series of intrabursal AdCRE injections on two independent cohorts of Arid1a fl/fl mice (n=42) and followed them for approximately one year. The mice remained tumor-free with no significant changes in survival, other than age-related mortality, similar to the phenotypes reported by Guan et al. 26 (Supplementary Fig. 1). The observation that ARID1A loss does not lead to tumor formation supports the notion that ARID1A mutations require additional mutational "hits" in the OSE before tumorigenesis can ensue 26 .

Coexistent ARID1A-PIK3CA mutations initiate ovarian cancer
To explore this further, we focused on the cancer proto-oncogene, PIK3CA (Phosphatidylinositol-4,5-bisphosphate 3-kinase, catalytic subunit alpha or p110α), which also has high OCCC mutation rates 30,31 . Recently, coexistent ARID1A-PIK3CA mutations were reported in gastric cancer 32 . Additionally, it has been observed that OCCC tumors harboring PIK3CA mutations display concomitant loss of ARID1A immunoreactivity, raising the possibility that ARID1A and PIK3CA mutations frequently co-occur in OCCC 33 . To understand the degree of mutational co-occurrence, we compared the incidence of ARID1A and/or PIK3CA mutations in available tumor datasets curated in The Cancer Genome Atlas (TCGA) to OCCC tumor sequencing data 17 . We found that the majority of cancers carry genetic alterations in one of these genes, however the rate of ARID1A-PIK3CA mutational co-occurrence (33%; pval<0.05) is highest in OCCC ( Fig. 2A).
PIK3CA tumor mutations often result in an H1047R 'hotspot' substitution within the kinase domain, generating a catalytically active protein 34 . PIK3CA amplification also frequently occurs in EOC 30 . To investigate the effect of PIK3CA alterations on ovarian tumorigenesis, we used the CRE-inducible (Gt)Rosa26Pik3ca *H1047R allele 35 to drive PIK3CA H1047R expression from the Rosa26 locus. We observed multifocal sites of epithelial hyperplasia in the OSE of AdCRE injected (Gt)Rosa26Pik3ca *H1047R mice, but no tumor formation (Table  1). Intrabursal AdCRE induction of PIK3CA H1047R , under the control of the endogenous promoter, also led to OSE hyperplasia in the mouse 36 , suggesting that constitutive overexpression of PIK3CA H1047R from the Rosa26 locus does not further augment this process.
We next evaluated the combinatorial effects of ARID1A loss and PIK3CA activation by performing AdCRE injections into the ovarian bursa of Arid1a fl/fl ;(Gt)Rosa26Pik3ca *H1047R mice. In stark contrast to the single mutants, Arid1a fl/fl ;(Gt)Rosa26Pik3ca *H1047R mice rapidly developed primary ovarian tumors and showed evidence of abdominal distension, warranting sacrifice with a median latency period of 7.5 weeks post-AdCRE injection ( Fig.  2J-M). Evidence for tumor-specific ARID1A loss and PIK3CA H1047R expression or PI3K pathway activation was detected by PCR amplification of tumor cDNA or by tumor immunohistochemistry (IHC) for ARID1A and phosphorylation of AKT at serine 473 (P-AKT S473) ( Fig. 2B-I). Increased morbidity coincided with the presentation of hemorrhagic ascites (Fig. 2J,L). Peritoneal metastases were detected in approximately 50% of the double mutants, often residing near the contralateral, uninjected ovary (Fig. 2K, see Table 1). Distant metastases beyond the abdominal cavity were not evident upon gross examination. We did not find evidence for genome instability in Arid1a fl/fl ;(Gt)Rosa26Pik3ca *H1047R tumor samples using mouse high-density SNP arrays ( Supplementary Fig. 2).

ARID1A haploinsufficiency does not lead to tumor formation
Several studies point to a gene dose-dependent role for ARID1A in tumor suppression given the high incidence of heterozygous tumor mutations in cancer 19 . To determine if an ARID1A haploinsufficiency coupled with PI3K pathway activation leads to ovarian cancer in our model, we performed AdCRE injections on Arid1a fl/+ ;(Gt)Rosa26Pik3ca *H1047R mice. OSE hyperplasia was observed in 5 of 7 Arid1a fl/+ ;(Gt)Rosa26Pik3ca *H1047R mice, but evidence of tumor formation was not detected after the 11-week observation period, similar to the (Gt)Rosa26Pik3ca *H1047R only cohort (Fig. 2N,O, see Table 1). These data are consistent with the observation that most OCCC tumors carrying heterozygous mutations in ARID1A show loss of ARID1A immunoreactivity and support the notion that additional mechanisms lead to loss of ARID1A protein expression in OCCC 18,37 . Thus, ARID1A likely follows the classical "two-hit" model of tumor suppression in OCCC.

Therapeutic PIK3CA inhibition prolongs animal survival
We next sought to examine the role of continued PIK3CA activation in the progression of ARID1A-deficient ovarian tumors and determine whether therapeutic PIK3CA inhibition would inhibit tumor cell growth and improve animal survival. Primary EOC tumor cells were isolated from the exfoliated tumor cell aggregates present in ascites fluid of tumorburdened Arid1a fl/fl ;(Gt)Rosa26Pik3ca *H1047R mice and subcultured using established methods (Fig. 3A). Isolated tumor cells displayed ARID1A loss and heightened PI3K signaling, as evidence by loss of ARID1A immunoreactivity and increased P-AKT S473 levels compared to normal OSE cells (Fig. 3B). Administration of the pan-class I PI3K inhibitor, BKM120 (Buparlisib) led to a dose-dependent decrease in P-AKT S473 and concomitant reduction in tumor cell viability (IC 50 = 0.96 +/− 0.25 µM) (Fig. 3C,D). We next administered chow-fed BKM120 to Arid1a fl/fl ;(Gt)Rosa26Pik3ca *H1047R mice for 3 weeks, starting at week 4 post-AdCRE injection. The effectiveness of BKM120-dependent inhibition of PI3K signaling in chow-fed mice was accessed by tumor IHC for phosphorylation of the S6 Ribosomal Protein at serine 235/236 (P-S6) (Fig. 3E). BKM120 treatment led to reduced P-S6 levels and extended the median latency period by 3.5 weeks (Fig. 3E,F). These data indicate that therapeutic PI3K inhibition promotes animal survival by inhibiting tumor cell growth, thus providing strong rationale for the use of PI3K inhibitors in OCCC treatment.

Mouse ovarian tumors manifest OCCC-like histopathology
Cardinal histopathological features of human OCCC include cells with clear cytoplasm, stromal hyalinization, solid, papillary, or tubulocystic tumor architectural patterns, and HNF1β immunoreactivity 38 . The ovarian tumors observed in Arid1a fl/fl ; (Gt)Rosa26Pik3ca *H1047R mice were predominately solid in appearance with some papillary areas ( Fig. 4A-D, see Table 1). Tubulocystic patterns were rarely observed (Fig. 4E, see Table 1). Neoplastic cells with clear cytoplasm, prominent nucleoli, and pleomorphic nuclei were observed in the tumors (see high magnification in Fig. 4A'-C'). Elongated spindleshaped cells with finely vacuolated cytoplasm were observed embedded in hyalinized matrix. Solid clear-cell carcinomas were observed directly attached to the ovarian surface ( Fig. 4A). We rarely observed tumor cells organized into bands of glandular epithelium with distinct borders, and, in general, the tumors appeared poorly differentiated and highly disorganized. Primary tumor masses ranged from approximately 100 mg to over 1 g (Fig.  4J). Differences in tumor size, metastatic potential, or morbidity did not correlate with any particular histologic feature.
In human OCCC, neoplastic cells that line luminal spaces often assume a 'hobnail' appearance, which crudely resembles a nucleus standing on the tip of a cytoplasmic stalk. Hobnail cells were often observed on the tumor periphery in regions where tumor cell exfoliation into luminal regions was occurring (Fig. 4D, I, K). Exfoliated, metastatic tumor cell aggregates often consisted of hobnail-shaped cells organized around either clear-cell or hyalinized cores (Fig. 4I). OSE hyperplasia and small clumps of hobnail cells were evident on the ovarian surface as early as one week following AdCRE injection, with exfoliated tumor cell aggregates and clear-cell-like features apparent at 2 weeks ( Supplementary Fig.  3). The tumors were also positive for the human OCCC marker, HNF1β, by IHC (Fig. 5B).

Tumor gene expression supports a role for IL-6 in OCCC
To further evaluate the relevance of our Arid1a fl/fl ; (Gt)Rosa26Pik3ca *H1047R ovarian tumor model to human OCCC, we performed a series of cross-species gene expression comparisons using a well-characterized gene expression dataset containing all four human EOC subtypes 39 . Gene expression profiles of mouse primary tumor samples and matched normal (uninjected) ovaries were generated by microarray. Consistent with our genetic model, ARID1A and PIK3CA expression levels were inversely correlated in comparisons between tumor and normal ovary samples, with PIK3CA upregulation strongly correlated with ARID1A downregulation in the tumor samples ( Supplementary Fig. 4). We verified our microarray gene expression results by RT-PCR validation of KRAS, TFPI2, CDKN2A (P16), CDKN1A (P21), MUC16 (CA125), and VCAN expression ( Supplementary Fig. 5). We next identified all coordinately regulated genes in both mouse tumors and human OCCC, as compared to species-matched normal ovaries. In comparisons of both mouse and human datasets, the control ovary samples showed high levels of gene expression variation, suggesting tissue sampling differences or general interspecies-specific gene expression differences ( Supplementary Fig. 6). Focusing on those genes that remained unchanged in normal mouse and human ovaries, we identified 584 coordinately regulated genes (272 upregulated; 312 downregulated) in mouse and human tumors ( Supplementary Fig. 7). Mouse and human tumor gene expression patterns were enriched for genes involved in Interleukin-6 (IL-6) signaling, immune system function, focal adhesion, and regulation of the actin cytoskeleton, as compared to normal ovaries (Supplementary Figs. 6 and 7). Since many of the upregulated genes we identified in cross-species comparisons with normal ovaries are likely to be found in other EOC subtypes (e.g. CA125) or generally found across most cancers, we next sought to identify an OCCC-specific gene signature for use in subtype-specific gene expression comparisons with the mouse tumors. To do this, we compiled a more refined 'discriminant' gene list for each subtype (e.g. clear-cell vs. all other EOC subtypes). We then compared the normalized raw expression values for each discriminant gene across all of the mouse tumor and human EOC samples. Using this analysis, we found that the gene expression patterns for mouse tumor and human OCCC samples were not statistically different (pval=0.6), suggesting that the OCCC-specific gene expression patterns are conserved between the two species (Fig. 5C). Several well-known markers of OCCC were identified as being highly expressed in the mouse tumors, including IL-6, STAT3, VCAN, and HIF1α, further supporting a role for heightened IL-6 signaling in OCCC tumor pathogenesis 40,41,42 (Fig. 6A).
To further assess similarities between our genetic model and human OCCC, we compared the mouse tumor-specific gene expression profiles for our Arid1a fl/fl ; (Gt)Rosa26Pik3ca *H1047R mouse model to those recently reported for the Arid1a fl/fl ; Pten fl/fl mouse model by Guan et al. 26 , which is genetically similar endometrioid EOC. Both models relied on the intrabursal AdCRE injection method for mutation induction. We applied principal component analysis using the normalized expression values for all genes in the tumor datasets for each mouse model. We found that each tumor sample clustered based on genotype or mutational pattern of the respective genetic model, with the exception of one Arid1a fl/fl ; Pten fl/fl tumor ( Supplementary Fig. 8). These findings are consistent with the marked phenotypic differences observed in our Arid1a fl/fl ;(Gt)Rosa26Pik3ca *H1047R tumors when compared to the phenotypes reported for Arid1a fl/fl ; Pten fl/fl tumors by Guan et al. 26 .
Autocrine IL-6 signaling promotes tumor cell growth IL-6 is an inflammatory cytokine that triggers JAK/STAT3 signaling and has prominent roles in tumor cell growth and differentiation 43,44 . High levels of circulating IL-6 in cancer patients are associated with a poor prognosis 43,44 . In addition to its increased expression in OCCC tumors, IL-6 has been detected in patient serum and tumor-derived cell lines, which has led to its proposed use as a therapeutic target in OCCC 40,41,42 . In support of our crossspecies microarray comparisons, we verified IL-6 expression in primary tumors and peritoneal metastases of Arid1a fl/fl ;(Gt)Rosa26Pik3ca *H1047R mice by RT-PCR (Fig. 6B). We also observed high levels of secreted IL-6 in the body fluids of tumor-burdened mice, with greater than 2000 pg/mL of IL-6 in ascitic fluid aspirates (Fig. 6C,D). Consistent with this, strong IL-6 immunoreactivity was also observed in the mouse tumors by IHC (Fig. 6E).
To understand the role of IL-6 in OCCC pathogenesis, we treated primary Arid1a fl/fl ; (Gt)Rosa26Pik3ca *H1047R mouse ascites-derived tumor cells, isolated as described above, with anti-IL-6 neutralizing antibodies. Anti-IL-6 treatment reduced tumor cell viability and downregulated phosphorylation of STAT3 at tyrosine 705 (P-STAT3 Y705), suggesting that autocrine IL-6-STAT3 signaling promotes tumor cell growth (Fig. 6F,G). Next, we generated primary mouse tumor cell lines that stably express IL-6 short-hairpin RNAs (shRNA) for use in tumor growth assays (Fig. 6H,I). IL-6 shRNA-expressing cells showed reduced growth rates in culture, and media supplementation with recombinant IL-6 restored the normal growth patterns of IL-6 shRNA-expressing cells ( Fig 6J). To further examine tumor cell autonomous roles for IL-6 in ovarian tumor cell growth, we grafted control shRNA-and IL-6 shRNA-expressing mouse tumor cells onto the flanks of immunodeficient nude mice. Remarkably, the ascitic tumor cell grafts were indistinguishable from the ovarian tumors with regards to histopathological features (e.g clear cytoplasm), despite being maintained in tissue culture and further grown under the skin of immunodeficient mice (Fig.  6K,L). The rate of tumorgraft growth for IL-6 shRNA-expressing cells was significantly less than control shRNA-expressing cells, further indicating that autocrine IL-6 signaling contributes to OCCC pathogenesis by promoting tumor cell growth (Fig. 6M).

IL-6 induction requires coexistent ARID1A-PIK3CA mutations
To determine if coexistent ARID1A-PIK3CA mutations are required for IL-6 induction, we utilized an AdCRE-inducible primary OSE cell culture model. IL-6 expression was measured following AdCRE-infection of primary OSE cells isolated from mice carrying all mutant allele combinations. Loss of ARID1A or PIK3CA activation alone led to significant increases in IL-6 expression by RT-PCR and ELISA, and IL-6 induction was further enhanced when ARID1A and PIK3CA were co-mutated together in Arid1a fl/fl ; (Gt)Rosa26Pik3ca *H1047R cells (Fig. 7A,B). Consistent with this, P-STAT3 Y705 levels also correlated with similar increases in IL-6 activity in AdCRE treated OSE cells (Fig. 7C).
ARID1A loss in AdCRE-treated Arid1a fl/fl or Arid1a fl/fl ;(Gt)Rosa26Pik3ca *H1047R OSE cells did not further enhance PI3K pathway activity (Fig. 7C). We found that Arid1a fl/fl ; (Gt)Rosa26Pik3ca *H1047R OSE cell proliferation was suppressed by treatment with anti-IL-6 neutralizing antibodies following mutation induction with AdCRE, and that recombiant IL-6 enhances normal OSE cell proliferation and IL-6-STAT3 signaling, further suggesting that IL-6 is both necessary and sufficient for tumor cell growth (Fig 7D-F). To determine if ARID1A is bound at the IL6 promoter under repressed conditions, we performed a series of ARID1A chromatin immunoprecipitation experiments in primary OSE cells. We used Arid1a fl/fl ;(Gt)Rosa26Pik3ca *H1047R OSE cells for these experiments so that AdCRE-treated or ARID1A-depleted cells could be used as a control for ARID1A immunoprecipitation specificity. In AdControl-treated cells, increased ARID1A occupancy was observed at sites near the IL6 promoter, but not at distal upstream or downstream sites (Fig. 7G). ARID1A occupancy at this site decreased in manner consistent with ARID1A depletion following AdCRE infection. These data indicate that coexistent ARID1A-PIK3CA mutations lead to IL-6 upregulation.

Discussion
Our data support a genetic epistasis model wherein ARID1A and PIK3CA mutations cooperate and ovarian cancer can arise only when these genes are co-mutated in the mouse OSE. The short latency period, lack of copy number variation, and OCCC-like histological and molecular features we observed are also consistent with coexistent ARID1A-PIK3CA mutations being a major driver of OCCC in humans. Our results suggest that high levels of tumor cell-derived IL-6 are promoting tumor growth in OCCC. Indeed, the top pathway we identified in common between species is comprised of genes involved in this pathway, suggesting that IL-6-STAT3 signaling is inherent to OCCC biology. Hence, IL-6 serves as both a molecular marker and a prospective therapeutic target in OCCC. Therefore, anti-IL-6 antibody (Siltuximab) therapy may prove to be a safe and effective treatment strategy for OCCC patients. More broadly, our findings further demonstrate that the continued identification of tumor mutations (i.e. TCGA projects) will allow us to identify potential functional relationships among recurrent mutational patterns or cancer pathways that would have otherwise gone unnoticed.
The OSE-derived tumor phenotypes we observed in our Arid1a fl/fl ; (Gt)Rosa26Pik3ca *H1047R mouse model of OCCC greatly contrasted with those reported by Guan et al. 26 for Arid1a fl/fl ; Pten fl/fl mice. For example, all of the tumor-burdened Arid1a fl/fl ; (Gt)Rosa26Pik3ca *H1047R mice were morbid or succumbed to death within 3 months after AdCRE injection, whereas Guan et al. 26 reported tumors in a subset of Arid1a fl/fl ; Pten fl/fl mice after 6 months, indicating stark differences in tumor latency or survival. Molecular differences were evident in tumor gene expression comparisons between the two mouse models, as each gene expression dataset parsed the majority of tumor samples based on its respective mutational pattern or genotype. These data point to potential functional differences between ARID1A-PIK3CA and ARID1A-PTEN mutational patterns in Type I EOCs and support the idea of a possible PI3K-AKT-mTOR-independent role for PTEN in endometrioid EOC 26 . Alternatively, it remains possible that clear-cell tumorigenesis is driven, in part, by an alternative AKT-independent mechanism downstream of oncogenic PIK3CA mutations 45,46 . Additional pharmacological studies targeting downstream effectors or alternative signaling pathway components will further address potential PI3K pathwayindependent mechanisms of endometrioid versus clear-cell tumorigenesis.
Several theories on the origins of OCCC have been proposed. Our data strongly suggests that the OSE harbors a cancer-prone cell population that gives rise to OCCC-like tumors in the mouse, which is consistent with the classical view of EOC pathogenesis. Nonetheless, our mouse model does not address the role of distal endometriotic explants or other putative precursor lesions in OCCC pathogenesis. Specifically targeting the ARID1A-PIK3CA mutational pattern to other tumor-prone epithelia within the Müllerian duct system would add relevance to our findings and help to further clarify the origins of EOC. An improved set of inducible CRE alleles would facilitate these experiments.
Inflammation of coelomic epithelial cell-derivatives in the female reproductive tract is thought to be a major contributor to malignant transformation in endometriosis-associated EOC, like OCCC 5,47 . Accordingly, inflammatory mediators, aberrant immune modulation, or irregular reproductive hormone levels are thought to play a major role in malignant transformation 5,47 . Inflammatory cytokine signaling pathways, like IL-6, may be part of the natural inflammatory repair process accompanying each ovarian and menstrual cycle that, when left unchecked, promotes unregulated epithelial cell growth. In support of this, our data demonstrate that OSE cells are competent to respond to and upregulate IL-6. Therefore, the incomplete resolution of epithelial cell inflammation during these repair processes may represent a common mechanism underlying OCCC tumorigenesis, irrespective of mutational pattern.
The identification of IL-6 as a physiological target of ARID1A tumor suppressor activity in our OCCC tumor model raises many important questions regarding the role of ARID1Acontaining SWI/SNF complexes in normal coelomic epithelial cell homeostasis. Indeed, ARID1A mutations have been identified in a wide variety of cancers originating from coelomic epithelial-derivatives throughout the gynecologic tract. Moreover, loss of ARID1A immunoreactivity is observed in endometriosis, a disease intimately associated with inflammation 48,49,50 . Endometriosis is also characterized by increased IL-6 expression, but the relationship between ARID1A loss and IL-6 upregulation in endometrial tissue is unknown 51,52 . Thus, ARID1A may regulate coelomic epithelial cell homeostasis and prevent tumorigenic conversion by negatively regulating the inflammatory programs required for normal tissue repair during the female reproductive cycle. It will be interesting to know if reproductive hormones influence ARID1A activity in this regard. Additional work on the propensity of ARID1A mutant epithelial cells to undergo malignant transformation in response to inflammatory "insults" or the role of key inflammatory response pathways in this process will be required to address these hypotheses. Thus, loss of ARID1A-containing SWI/SNF complexes might expose the "darker," pro-tumorigenic side of inflammation.
How do oncogenic PI3K signaling mutations contribute to this process or lead to increased IL-6 expression? Positive regulators of IL-6 include JAK/STAT, PI3K, EGFR/HER2, RAS/RAF/MEK, TNF/TLR/IL1 or RAS/ERK oncogenic pathways and AP-1, SP1, NF-κB, C/EBP, STAT3, and CREB transcription factors 53 . IL-6-mediated JAK/STAT activation can further promote IL-6 transcription 53 . We propose that oncogenic PIK3CA H1047R mutations act like inflammatory "insults" and drive the signaling loop that sustains high levels of IL-6 production in the absence of negative regulation by ARID1A (Fig 7H). Additional studies focusing on the intricacies of this feed-forward signaling loop and the potential crosstalk mechanisms occurring among the various pathways that sustain it will further explain why coexistent ARID1A-PIK3CA mutations are found in a wide variety of human cancers.

Generation of the Arid1a conditional mouse allele
The Arid1a fl conditional mutant mouse allele was engineered using the Bacterial Artificial Chromosome (BAC) recombineering methods developed by Liu et al 54 . Correctly targeted mouse ES cell clones were identified by Southern blotting. The Neo selection cassette used for targeting was removed by FLP-mediated recombination 54 . Targeted ES cells were injected into C57Bl6 blastocysts and implanted into pseudopregnant females. Germline animals were detected by coat color and PCR confirmation of tail DNA samples.

AdCRE-mediated genetic recombination in the ovary
To induce genetic recombination in the ovarian surface epithelium (OSE), we employed a modified version of the ex-vivo Adenovirus-CRE (AdCRE) intrabursal delivery method 7,29 . Briefly, AdCRE particles (obtained from the University of Iowa Gene Transfer Core) were diluted in sterile 1X Dulbeccos's phosphate buffered saline (dPBS) containing 8 µg/mL polybrene. Deeply anesthetized 8-10 week old mice were given a single 5 µL injection of AdCRE particles (2.5×10 7 plaque-forming units or pfu) into the right ovarian bursal cavity of the surgically exposed ovary using a sterile 31-gauge needle. Immediately following AdCRE injection, the surgically excised ovaries were washed with sterile 1X dPBS and placed back into the abdominal cavity. The surgeries were performed on multiple occasions. Age-and genotype-matched, non-littermate animals were randomized prior to AdCRE injection. The surgeon was blinded to the genotypes prior to surgery and AdCRE injection. All surgical procedures were performed in accordance with protocols approved by the University of North Carolina at Chapel Hill Institutional Animal Care and Use Committee.

Mouse husbandry and genotyping
All mice were maintained on an outbred (random) genetic background using CD-1 mice (Charles River). (Gt)ROSA26 Pik3ca*H1047R and Sox2CRE alleles were purchased from The Jackson Laboratory and identified by PCR using published methods 28,35 . Arid1a +/+ ; Arid1a fl/+ ; Arid1a fl/fl alleles were distinguished by PCR using the following primer set: (F) CTAGGTGGAAGGTAGCTGACTGA; (R) TACACGGAGTCAGGCTGAGC. Combining the forward primer (from above) with the following reverse primer amplifies the CREexcised allele: (R) AGAGTAACTAATAACTGCTGGAGGATG. PCR products were resolved by 2%-Tris-Borate-EDTA agarose gel electrophoresis. Endpoints were palpable tumors (>1.5 cm 3 ), visible hemorrhagic ascites or severe abdominal distension, and signs of severe illness, such as dehydration, hunching, ruffled fur, signs of infection, or nonresponsiveness. Sample sizes within each genotype were chosen based on the proportions of animals with tumors and/or survival between each experimental group or a Kaplan-Meyer log rank test for survival differences. Wholemount live animal luminescent imaging was performed using the IVIS Optical System and reagents (Perkin Elmer) according to the manufacturer's instructions. Wholemount images of mice or necropsies were aquired using a Canon PowerShot SD1100 IS digital camera and a tabletop tripod. Wholemount tumor specimens were imaged on a Leica MZ FLIII stereomicroscope using a DS Ri1 digital camera (Nikon). All mice were maintained at the University of North Carolina at Chapel Hill, Animal Facility using standard techniques in accordance with protocols approved by the University of North Carolina at Chapel Hill Institutional Animal Care and Use Committee.

In vivo drug treatments
The pan-PI3K inhibitor, BKM120 (Novartis) was milled by Research Diets using a vehicleor carrier-free formulation similar to standard husbandry diets and dosed at 40 mg kg −1 based on the average daily oral consumption of a mouse 55,56,57 . AdCRE-injected animals were randomized prior to drug treatment. Control or untreated mice were given a standard husbandry diet. All drug treatments were performed in accordance with protocols approved by the University of North Carolina at Chapel Hill Institutional Animal Care and Use Committee.

RT-PCR
Total RNA was extracted from pulverized tumor samples or ovarian epithelial cells using the TRIzol method (Invitrogen), followed by an RNA cleanup step and on-column DNA digestion using the RNAeasy mini prep kit (Qiagen) according to the manufacturers' instructions. To analyze ARID1A and (Gt)Rosa26 Pik3ca*H1047R gene transcript levels, realtime quantitative PCR was performed using Ssofast PCR master mix (Biorad) and a CFX96 thermocycler (Biorad) with the following gene-specific primers: ARID1A (F) CTGAAGGACAAAGGTGACT, ARID1A (R) CACAACTGCGAACTTCTC; R26exon1 (F) CTAGGTAGGGGATCGGGACTCT, PIK3CAcDNA (R) AATTTCTCGATTGAGGATCTTTTCT 35  CTGCAAGTGCATCATCGTTGT. Expression was calculated from three independent experiments using ΔΔC t methods. Expression levels were normalized to βACTIN. Statistical differences were detected using a two-tailed Student's t test. The following H1047 spanning PCR primer pair was used to distinguish endogenous PIK3CA from Rosa26-derived, PIK3CA H1047R gene transcripts: (F) TTCAATGATGCTTGGCTCTG; (R) CTGCTTGATGGTGTGGAAGA. The PCR products were digested with MslI and resolved by 2%-Tris-Borate-EDTA agarose gel electrophoresis.

ELISA
IL-6 ELISAs were performed using the OptEIA mouse IL-6 kit reagents (BD Bioscience) according to the manufacturer's instructions. ELISA measurements were normalized to total cellular protein.

Copy number and TCGA computational analyses
Tumor and matched control DNA samples were processed for hybridization on the high density Mouse Universal Genotyping SNP Array (MegaMUGA, Neogen) using published methods 58 . Copy number analysis was performed using ASCAT 59 . ARID1A and/or PIK3CA TCGA mutation frequency was identified by querying the cBioPortal on 2014-10-21 using the R API 60, 61 .

Microarray and cross-species gene expression comparisons
Total RNA was isolated from 9 primary tumor specimens and 9 matched normal (uninjected) ovaries as described above. Total RNA (250 ng) was used to synthesize fragmented and labeled sense-strand cDNA and hybridize onto Affymetrix Mouse Gene 2.1 ST peg plate arrays. The mouse data were uploaded to the Gene Expression Omnibus (GEO) under the accession number: GSE57380. Affymetrix CEL files were normalized using the Robust Multichip Average normalization method 62 . Expression changes (tumor versus control ovary) were determined using Linear Models for Microarray Data Analysis (LIMMA) 63 or Significance Analysis of Microarray (SAM) analysis 64 . Probes with a false discovery rate (FDR) of 0% were considered statistically significant. For mouse-human cross species comparisons, expression data for human OCCC and normal ovaries were downloaded from GEO under the accession number GSE6008 39 . Heat maps and dendrograms were generated with Java TreeView or R (www.R-project.org). Histological sub-type specific gene signatures were derived by first identifying differentially expressed genes between human normal tissue and cancer tissue in GSE6008. Next, differentially expressed genes were identified comparing each histological subtype to all other histological subtypes. We considered only upregulated genes, and any genes that were upregulated in both the tumor versus normal comparison and in a particular subtype were removed because these are not specific to a histological subtype and may reflect generic pan-cancer gene expression patterns. To compare human and mouse samples from these gene sets we first determined Z-scores for each gene within mouse or human. We combined these Z-scores and Z-scored the combined data using all human and mouse tumor samples in GSE6008 and this study. Pathways analysis was perfomed using the GSEA Molecular Signatures Database (MSigDB) 65,66 or Ingenuity Pathway Analysis (IPA) (Qiagen).

Ascites-derived tumor cell and OSE cell culture
Primary ascites-derived ovarian tumor cells were isolated and grown as previously described 67 . To facilitate complete red blood cell lysis, the cell isolates were incubated and washed in deionized H 2 O prior to re-suspension in growth medium. Mouse OSE cells were isolated and cultured as previously described 68 . To induce CRE-mediated recombination in OSE cells, the cells were infected with AdCRE or AdControl particles (obtained from the University of Iowa Gene Transfer Core) at an MOI of 200 for 2 hours in serum-free media. Infected OSE cells were then placed in normal growth medium and allowed to recover for 3 days prior to further manipulation. BKM120 (Chemitek), low endotoxin rat anti-mouse IL-6 neutralization antibody (R&D systems, MAB406), non-specific rat IgG 1 isotype control antibody (R&D systems, MAB005), or recombinant mouse IL-6 protein (R&D systems, 406-ML-005) were reconstituted according to the manufacturers' instructions and used as described in the text. For all antibody neutralization experiments, the cells were re-fed daily with the appropriate concentrations of antibody. Primary ascites-derived ovarian tumor cells were stably transduced with Mission pLKO.1-puro IL-6 shRNA (Hairpin sequence: CCGGCCAGAGTCCTTCAGAGAGATACTCGAGTATCTCTCTGAAGGACTCTGGTT TTTG) (Sigma, TRCN0000067551) or non-target control shRNA (Sigma, SHC016V) Lentiviral transduction particles according to the manufacturers' instructions. The MTT assay kit (Roche) was used to measure cell viability. Ascites-derived tumor cells and OSE cells were passaged using Accutase (Invitrogen). Half-maximal inhibitory concentrations (IC 50 ) were calculated by curve-fitting using Kaleidagraph software (Synergy).

Tumor grafts
Primary ascites-derived ovarian tumor cells expressing control-or IL-6 shRNAs were harvested by Accutase digestion and washed in 1X Dulbecco's phosphate buffered saline. The cells were then injected subcutaneously into the right flanks of immunodeficient mice at a concentration of 5×10e7 cells/mL in a 1:1 (v/v) solution of 1X Dulbecco's phosphate buffered saline and Cultrex basement membrane extract (3432-005-01, Trevigen). Tumor graft volumes were calculated using bi-weekly caliper-based measurements and growth rates were based on these volumetric measurements. All tumor grafts were performed in accordance with protocols approved by the University of North Carolina at Chapel Hill Institutional Animal Care and Use Committee.

Chromatin immunoprecipitation
Primary OSE cells were fixed in growth media containing 1% methanol-free formaldehyde (Pierce) for 10 min. at room temperature. The cells were washed twice with 1X dPBS and then swelled in cell lysis buffer (5 mM Tris-Cl [pH 8.0]; 85 mM KCl; 1% IGEPAL CA-630) for 10 min. on ice. Swollen cells were homogenized (20 strokes) with a B-type dounce homogenizer on ice to facilitate nuclei release. Isolated nuclei corresponding to 1×10e7 cell equivalents per 500 µL of nuclear lysis buffer (50 mM Tris-Cl [pH 8.0], 10 mM EDTA, and 0.3% SDS) and then sonicated using a water bath-cooled Diagenode Biorupter sonicator and 1.5 mL TPX microtubes (200 µL sample volume per tube) to an average DNA length between 300 and 700 bp. Chromatin concentrations were determined using a BCA protein assay kit (Pierce). Chromatin samples were diluted to 300 µg per 200 µL of nuclear lysis buffer and then cleared by centrifugation at 10,000Xg. Pre-cleared chromatin samples (200 µL each) were added to 800 µL (1:5 dilution) of ChIP dilution buffer (50 mM Tris-Cl [pH 7.5]; 150 mM NaCl; 5 mM EDTA; 0.5% IGEPAL CA-630; 1% TX-100) and incubated with 10 µg of prebound ARID1A (Clone PSG3, 04-080, Millipore) Protein G dynal beads (Invitrogen) overnight at 4°C. The Protein G dynal bead-anti-ARID1A immune conjugates were washed twice with 1 mL ChIP dilution buffer, 3-times with 1 mL ChIP dilution buffer supplemented with 500 mM NaCl, twice with 1 mL ChIP dilution buffer, followed by a final wash in 1 mL low salt TE buffer (10 mM Tris-Cl [pH 8.0]; 1 mM EDTA; 50 mM NaCl). The immunoprecipitated samples were eluted in 100 µL of elution buffer (50 mM NaHCO 3 ; 1% SDS) at 62°C for 20 min. using an Eppendorf tube mixer. The crosslinks were reversed by adding NaCl to a final concentration of 0.2M followed by an overnight incubation at 65°C. The samples were then deproteinated with Proteinase K and purified using the ChIP DNA Clean & Concentrator kit (Zymo Research) according to the manufacturer's instructions. Real-time quantitative PCR using Ssofast PCR master mix (Biorad) and a CFX96 thermocycler (Biorad) was performed using the following gene-specific primer pairs: (IL-6, Site A) GCAACTCTCACAGAGACTAAAGG, GGACAACAGACAGTAATGTTGC 69 ; (IL-6, Site B) CTAGCCTCAAGGATGACTTAAGC, CTATCGTTCTTGGTGGGCTCCAGAGC 70 ; (IL-6, Site C) GCCTTCTTGGGACTGATGCT, GACAGGTCTGTTGGGAGTGG. Data from three independent primary OSE cell isolations were averaged and plotted as a percentage of the total input.

Supplementary Material
Refer to Web version on PubMed Central for supplementary material.     (Gt)Rosa26Pik3ca *H1047R mice were calculated using a two-tailed Student's t test. (E) IL-6 expression in an Arid1a fl/fl ;(Gt)Rosa26Pik3ca *H1047R tumor by IHC. (F) MTT absorbance values plotted with log antibody concentration (µg mL −1 ) for non-specific rat IgG-treated (control) or rat anti-mouse IL-6 treated Arid1a fl/fl ;(Gt)Rosa26Pik3ca *H1047R ascitic tumor cells after 96 hrs. of treatment. Plot represents the average absorbance value ±SD for treatment performed on three independent cell lines. Significant differences between control-and anti-IL-6-treated cells were calculated using a two-tailed Student's t test (*significant p-value <0.05). (G) Primary ascitic tumor cells were treated 10, 1, 0.01, or 0 µg mL −1 rat anti-mouse IL-6 or non-specific Rat IgG-treated (control). (H,I) IL-6 expression in Arid1a fl/fl ;(Gt)Rosa26Pik3ca *H1047R ascitic tumor cells stably expressing control shRNAs or IL-6 shRNAs by ELISA and western blot. Each replicate represents in stable pool of IL-6 shRNA expressing cells from three independently isolated tumor cell lines. Significant differences based on the average protein concentration ±SD between control-versus IL-6 shRNA were calculated using a two-tailed Student's t test. (J) MTT absorbance values plotted over time for control shRNA-, IL-6 shRNA-, or IL-6 shRNA-expressing cells supplemented with 10 ng/mL IL-6. Cells were plated at 2×10e4 cells per mL at hour 0, then MTT measurements were taken every 24 hours for a total of 96 hours. Significant differences between control-and IL-6-shRNA expressing cells were calculated using a twotailed Student's t test (*significant p-value < 0.05). (K,L) Whole-mount images of controland IL-6-shRNA tumorgrafts (arrowheads) on the right flank of nude mice and images of corresponding H&E-stained tumorgraft sections. (M) Plot of growth rates for control-and IL-6-shRNA expressing tumorgrafts over 21 days of measurement. Significant differences between control-and IL-6-shRNA tumorgraft growth rates were calculated using a twotailed Student's t test. Only p-values <0.05 were considered significant.  Table 1 Ovarian tumorigenesis requires concurrent ARID1A loss and PIK3CA activation Only morbid-euthanized mice were scored.