ARID1A mutations in cancer: another epigenetic tumor suppressor?

Epigenetic regulators impose upon the genetic code a chromatin structure characterized by chromatin accessibility, nucleosome position, and histone modifications. These features of chromatin structure modulate gene expression, thereby affecting both the identity and function of cells. Sequencing of cancer genomes has revealed frequent mutations in epigenetic regulators, particularly chromatin remodelers and histone modifiers, and disordered chromatin regulation has emerged as a distinct mechanism contributing to tumor development. The canonical example of the links between chromatin remodeling and cancer may well be SMARCB1/SNF5/INI1/BAF47, a core component of the SWI/SNF chromatin remodeling complexes, of which ARID1A is also a member. In the late 1990s, specific biallelic inactivating mutations of SMARCB1 were identified in the vast majority of rhabdoid tumors, highly aggressive cancers arising in kidney, brain, and soft tissues in young children ( 1 , 2 ). Analogous to retinoblastoma and the Rb gene product, children having germline heterozygous inactivation of SMARCB1 are cancer predisposed, having a high risk of developing malignant rhabdoid tumors at an especially young age. Subsequent to these discoveries, implementation of next-generation sequencing technologies have revealed frequent and recurrent mutations in a wide variety of epigenetic modulators, including mediators of DNA methylation (e.g. DNMT3a) and covalent modifiers of histones (e.g. MLL, MLL3, KDM5A, KDM5C, KDM6A, etc.). Furthermore, genes encoding subunits of ATP-dependent chromatin remodelers, especially subunits of the SWI/SNF complex, have emerged as particularly frequently and broadly mutated in cancer ( 3 ). Recently, recurrent inactivating mutations in ARID1A have been identified in a wide variety of cancers, suggesting that it too functions as a tumor suppressor in many different cell types. Here we review the spectrum of ARID1A mutations in cancer and what is known about ARID1A structure and function, as well as discuss potential mechanisms of tumor suppression and clinical implications.

ARID1A has been implicated in numerous protein-protein interactions ( Figure 1B ). Among these interactions, the most widely known and studied are those which make ARID1A a part of SWI/SNF chromatin remodeling complexes. SWI/SNF complexes are multi-subunit protein complexes that utilize the energy of ATP hydrolysis to remodel chromatin structure and are capable of sliding nucleosomes along a DNA template in vitro ( 31 ). The nucleosome remodeling activity is derived from the catalytic ATPase subunit (either SMARCA4/BRG1 or SMARCA2/BRM) and is enhanced by the non-catalytic subunits SMARCB1/SNF5, BAF155, and BAF170 ( 32 ). SWI/SNF complexes have been described as having two main variants, BAF (BRG1 associated factors) and PBAF (Polybromo BRG1 associated factors) ( 33 ). However, many SWI/SNF subunits have multiple isoforms and/or belong to highly homologous multi-gene families ( 34 ). For example, in addition to the two variant ATPase subunits, four different genes encode the BAF45 subunit, three different genes encode the BAF60 subunit, and two different genes encode the BAF53 subunit. Consequently, it has been proposed that several hundred variants of the complex may exist ( 35 ), and that different complex variants may play distinct lineage-, transcription factor-, and chromatin state-specific roles. Several activities have been ascribed to the complex that may contribute to transcriptional regulation, including modulation of chromatin accessibility and transcription factor binding, recruitment of coactivator/corepressor complexes that have histone-modifying activity, and facilitation of the chromatin looping required to approximate promoters with distal enhancers ( Figure 2 ) ( 3 ). As a member of SWI/SNF complexes, ARID1A is thought to contribute to specific recruitment of its chromatin remodeling activity by binding transcription factors and transcriptional coactivator/corepressor complexes ( 36 - 38 ). Several studies have linked SWI/SNF and ARID1A to transcriptional regulation, particularly nuclear hormone-induced transcription and expression of cell cycle regulators ( 21 , 37 - 43 ). Re-expression of ARID1A in an ARID1A-deficient breast cancer cell line augmented transcriptional activation of reporter plasmids by glucocorticoid, estrogen, and androgen receptors ( 40 ). Interaction with these nuclear hormone receptors, particularly the glucocorticoid receptor, has been shown to depend on a broadly defined C-terminal region that contains several putative nuclear hormone receptor binding sites (LXXLL motifs) ( 37 ). It is unknown whether roles for ARID1A in the regulation of hormone signaling contribute to tumor suppression, but it is noteworthy that ARID1A mutations are frequently seen in cancers occurring in hormone-responsive tissues (e.g. breast and ovarian).

The vast majority of cancer-associated mutations in ARID1A (>97%) were inactivating, with nonsense or frameshift (rather than silent or missense) mutations detected throughout the gene. However, in only 30% of the OCCs with ARID1A mutations were both alleles affected ( 4 , 5 ). By immunohistochemistry, 73% of the ARID1A heterozygous tumors lacked protein expression, as did 5% of tumors not found to have coding mutations ( 5 ). Several mechanisms may be postulated to explain this interesting finding - loss of ARID1A protein expression in the setting of heterozygous mutation without loss of heterozygosity. Mutations affecting ARID1A expression may occur in non-coding regions of the genome not assayed by exome sequencing techniques. Alternatively, epigenetic silencing might contribute. One group has postulated that post-transcriptional and/or post-translational mechanisms account for loss of ARID1A protein in OCCs harboring heterozygous mutations, based upon the finding that RNA sequencing detects both wild-type and mutant alleles in a small number of cases ( 5 ). Conversely, 27% of the ARID1A heterozygous OCCs retain detectable protein expression. A similar situation occurs in gastric cancer, in which ARID1A mutations were biallelic in only 44% of ARID1A-mutant samples. Again, 25% of samples harboring heterozygous mutations retained ARID1A expression by IHC ( 9 ). Array CGH analysis of primary breast tumors demonstrated loss of chromosomal material encoding one copy of ARID1A in 13% of samples, without identifying any ARID1A coding mutations in the remaining allele ( 14 ). At the protein level, two surveys of ARID1A found complete loss of expression characterized only 1-3% of breast cancers ( 6 , 7 ). Finally, in hepatocellular carcinomas, nearly all of the ARID1A mutations were found to be heterozygous ( 11 ) and, in a separate study, protein expression was detected in all samples ( 7 ). Collectively, observations that ARID1A is recurrently and specifically mutated on one allele but expressed from the other allele, have raised the possibility that reduced levels of ARID1A may mediate a haploinsufficient effect in promoting cancer. In vitro studies provide support for a haploinsufficient tumor suppressor role for ARID1A. Knockdown studies in a variety of cell types, having only partial loss of ARID1A, showed increased cell proliferation and colony formation ( 9 , 13 , 14 , 21 ), impaired differentiation ( 21 , 22 ), as well as decreased apoptosis ( 23 ). Heterozygosity for Arid1a in mice results in embryonic-lethality, suggesting that biologically relevant haploinsufficient effects are caused by loss of a single allele ( 22 ). Furthermore, two studies have found that haploinsufficiency and decreased transcript levels of ARID1A are associated with high-risk, poor-prognosis breast cancers ( 14 , 24 ). Collectively, these findings suggest that, much like p53 and PTEN, haploinsufficiency for ARID1A is capable of promoting tumor formation in some cancers. Given the diversity of the lineages involved, it is also likely that the effects of ARID1A mutation may vary by cell type such that haploinsufficiency may promote transformation in some lineages whereas homozygous inactivation is required in others ( 25 , 26 ). Close attention to tissue-specific effects of ARID1A mutations and incorporation of mouse models of disease will be critical to testing these hypotheses.

Several studies have attempted to analyze the prognostic significance of ARID1A mutations, transcript levels, or protein loss in a variety of cancer subtypes – OCCs, gastric cancer, breast cancer, and bladder cancer ( 8 , 10 , 14 , 18 , 24 , 53 , 56 - 59 ). Even within a single cancer subtype, no consistent relationship has emerged between ARID1A mutation or expression and prognosis. Ultimately, larger prospective studies, ideally assessing not only ARID1A sequence but also loss of heterozygosity and protein expression, will be required to adequately address the prognostic significance of ARID1A mutations. A handful of candidate therapeutic targets, having striking concordance with those identified for SMARCB1-deficient cancers, have emerged from functional studies, including cyclins A, B2, and C ( 41 ); myc ( 42 ); and the Polycomb complexes ( 16 ). A key question going forward is the degree to which therapeutic dependencies will be similar, or distinct, among cancers characterized by mutations of different SWI/SNF subunits. Furthermore, it will be of interest to determine whether inactivation of chromatin remodelers such as ARID1A can be therapeutically exploited by targeting downstream and potentially reversible epigenetic consequences of remodeler mutation ( 60 ).

Within the context of functional models of cancer development ( 44 ), how might the consequences of ARID1A mutation contribute? Studies have suggested roles for ARID1A in three processes relevant to tumor suppression – proliferation, differentiation, and apoptosis – with mixed results. For both breast and gastric cancer cell lines, knockdown of wild-type ARID1A enhanced cellular proliferation, while re-expression of ARID1A in mutant cell lines dampened cell proliferation ( 9 ). Knockdown of ARID1A also enhanced proliferation of normal ovarian surface epithelial cells ( 45 ) and inhibited cell cycle arrest in murine pre-osteoblasts ( 21 , 41 ). In contrast, Arid1a knockout in ES cells resulted in loss of self-renewal properties. With respect to differentiation, Arid1a knockdown disrupted differentiation of cultured osteoblasts, whereas knockout forced ES cell differentiation into primitive endoderm and permitted in vitro development of neurons and skeletal muscle, while preventing that of cardiomyocytes and adipocytes ( 22 ). A single study has examined apoptosis following ARID1A knockdown, finding that Fas-mediated cell death is inhibited in Jurkat leukemia cells ( 23 ). Taken together, these results raise the possibility that ARID1A loss affects three canonical tumor suppressor functions, with tumor promoting effects particularly seen in lineages associated with ARID1A cancer mutations. Given the large number of SWI/SNF subunits that are mutated in cancer, it seems highly likely that the tumor-suppressing activities of ARID1A are related to its role within this complex. In addition to ARID1A and SMARCB1, at least 5 other SWI/SNF subunits have been found to be recurrently mutated in cancer, including: PBRM1 (40% of renal carcinomas) ( 46 ), ARID2 (18% of hepatitis C-associated liver cancer) ( 47 ), ARID1B (liver, breast and pancreatic cancers) ( 12 , 16 , 48 ), BRD7 (breast cancer) ( 49 ), and SMARCA4 (lung cancer, medulloblastoma) ( 50 , 51 ). Also worthy of note is that mutations in individual subunits of the SWI/SNF complex have largely distinct cancer spectra ( 3 ). Consequently, loss of individual subunits may have distinct effects upon SWI/SNF function and its tumor suppressor roles. Loss of ARID1A may have many effects on SWI/SNF complexes that lead to transcriptional dysfunction, including disruption of nucleosome sliding activity, assembly of variant SWI/SNF complexes, targeting to specific genomic loci, and/or recruitment of coactivator/corepressor activites. With respect to chromatin remodeling, ARID1A is thought to be dispensable for the in vitro nucleosome remodeling activity of SWI/SNF, as measured by DNase hypersensitivity patterns of reconstituted nucleosomal arrays ( 32 ). However, contributions of SWI/SNF to the complex states of chromatin that exist in vivo are poorly understood and thus contributions from ARID1A to such activities cannot be excluded. Loss of ARID1A may affect expression or stability of other SWI/SNF subunits. While one study has demonstrated that several components of SWI/SNF (including Baf155, Baf170, Snf5, and Brg1) remain normally expressed and co-associated upon knockdown of ARID1A ( 21 ), another has shown that knockout of one or both alleles of ARID1A in mouse ES cells alters protein levels of Snf5, Baf155, Baf170, and Baf57 ( 22 ). This latter observation suggests ARID1A mutation may affect the composition or assembly of several SWI/SNF complex variants, not just those that usually incorporate this subunit. Targeting of SWI/SNF to chromatin, as measured by binding of Snf5, Baf155, Baf170 and Brg1 to the Myc promoter and a small number of E2F targets, is not affected by knockdown of ARID1A ( 42 ). However, as the SWI/SNF complex binds to thousands of loci, genome-wide analyses will be necessary to determine global effects of ARID1A on SWI/SNF targeting. Finally, ARID1A-containing complexes have been shown to associate with the Sin3 corepressor complex (including HDAC1 and HDAC2), suggesting that SWI/SNF recruitment of additional transcription modulators and histone modifiers may be impaired by ARID1A loss ( 21 ). Another question of interest is the extent to which ARID1A and ARID1B have complementary or distinct roles in cell function. ARID1A and ARID1B have been characterized as mutually exclusive members of BAF variant SWI/SNF complexes, meaning the two proteins do not coimmunoprecipitate ( 31 ). While both are expressed in many tissues, one or the other is present in any single instance of the SWI/SNF complex. Following induction of differentiation in a non-transformed osteoblast model, ARID1A is required for the induction of cell cycle arrest, while knockdown of ARID1B had no effect ( 41 ). Similarly, ARID1A and ARID1B have opposing effects on cell cycle arrest caused by serum deprivation - knockdown of ARID1A delayed arrest while ARID1B knockdown had no effect. Conversely, cell cycle re-entry following deprivation was modestly accelerated by ARID1A knockdown but markedly delayed by ARID1B knockdown. As noted above, ARID1A associates with HDAC1/2-containing complexes, whereas ARID1B co-immunoprecipitates with HDAC3 and is associated with both histone deacetylase and acetylase (HAT) activity ( 21 ). Furthermore, ARID1A and ARID1B were found to have differential interactions with E2F family members ( 21 ). Collectively, such findings could suggest a tumor suppressor model by which the unopposed actions of ARID1B-containing SWI/SNF complexes disrupt cell cycle control and predispose to transformation. However, it is also worthy of note that another variant of SWI/SNF incorporates a third ARID protein, known as ARID2. Interestingly, recurrent mutations of both ARID1B and ARID2 have been reported in some of the same types of cancers that contain frequent ARID1A mutations, including hepatocellular carcinomas ( 12 , 13 , 47 ), and recurrent ARID1B mutations are seen in breast cancers ( 16 ), pancreatic cancers ( 16 ), and gastric cancers ( 8 ). This raises the possibility that ARID1B and ARID2 may also have tumor suppressor activity. Whether these ARID gene mutations are mutually exclusive or have some degree of overlap has not been systematically reported. Consequently, it remains unclear the extent to which complementary or opposing functions of ARID-family proteins contribute to cancer. A further intriguing question is whether ARID1A may contribute to enzymatic activity distinct from ATP-dependent chromatin remodeling. Studies performed on ARID1B demonstrate that it has E3 ubiquitin ligase activity, thought likely to extend to ARID1A based upon structural homology ( 52 ). ARID1B has been shown to facilitate monoubiquitination of lysine-120 of histone H2B – a modification that is thought to be a precursor to acquisition of trimethylated H3K4 at promoters and enhancers of actively transcribed genes. The E3 ubiquitin ligase function has been mapped to a putative C-terminal B/C box. However, it remains to be determined whether this ubiquitylation activity is associated with SWI/SNF complexes and whether it relates to the putative tumor suppressor role of ARID1A.

Ultimately, several lines of evidence support classification of ARID1A as a bona fide tumor suppressor gene. Somatic mutations are found in significant subsets of several cancer types and are not detected in other specific classes of cancer. Studies demonstrate a characteristic pattern of inactivating mutations occurring throughout the gene body. Finally, early functional studies provide evidence that ARID1A affects several canonical tumor suppressor pathways. The case of ARID1A exemplifies the challenges present in assigning mechanistic import to the many mutations being identified through cancer genome sequencing. Haploinsufficient tumor suppressor effects have ample precedent, and their identification and interpretation require synthesis of human sequencing data, as well as cell culture and animal modeling systems. Different mutation patterns among cancer subtypes imply tissue-specific mutational effects, which may limit the extent to which observations in one cancer type or cell line may be applied to other model systems. Chromatin remodelers, and ARID1A in particular, are emerging as a novel class of genes associated with a variety of cancers. While functional pathways that promote transformation are beginning to be identified, much remains to be elucidated regarding the mechanistic basis by which ARID1A alters chromatin structure, contributes to SWI/SNF activity, modulates transcription, and ultimately suppresses cancer formation.

In addition to identifying frequent ARID1A mutations, the studies above have demonstrated several associations with other oncogene and tumor suppressor pathways. In OCC, one study identified a significant correlation between ARID1A loss and the presence of activating mutations in PIK3CA as 46% of ARID1A deficient tumors contained PIK3CA mutations while only 17% of ARID1A expressing tumors contained PIK3CA mutations ( 53 ). There was also a strong concordance between PIK3CA and ARID1A status in gastric cancer, collectively suggesting potential cooperating effects of these mutations in oncogenesis ( 9 ). In breast, gastric, and ovarian cancers, mutation or loss of ARID1A is significantly more common in tumors having wild-type p53 ( 8 , 9 , 24 , 45 ), and p53 has been shown to directly associate with the C-terminus of ARID1A in co-precipitation experiments ( 45 ). In gastric cancer, two additional correlations were noted: rates of ARID1A mutation or loss were significantly higher in tumors having microsatellite instability (ie mismatch repair defects) and those associated with EBV infection ( 9 ). In HCC, ARID1A mutations were correlated with mutations of β-catenin ( 11 ). The SWI/SNF complex has also been reported to physically associate and functionally cooperate with the RB protein and pathway, respectively ( 54 , 55 ). However, no relationship between SWI/SNF mutation and RB mutation has yet been reported in the cancer genome sequencing studies described above. Ultimately, these data reveal potential cooperating interactions between ARID1A mutation and other tumor-promoting pathways.