Advances in sequencing technologies and bioinformatics, coupled with the identification of the sequence of the human genome, have enabled more than a dozen tumor types to be evaluated for mutations over their entire exomes [Meyerson et al., 2010; Stratton, 2011]. These studies have demonstrated that the landscape of each particular tumor type is defined by a small number of genes mutated at a high frequency, called “gene mountains” and a larger number of gene “hills” that are present in a smaller proportion of cases [Wood et al., 2007].
Members of our group recently used next generation sequencing to evaluate the exomes of ovarian clear cell carcinomas (OCCCs) and identified truncating mutations in ARID1A (MIM# 603024) in 57% of these tumors [Jones et al., 2010]. Independently, Wiegand et al.  discovered a high prevalence of ARID1A mutations in both OCCC (45%) and endometriod carcinoma of the ovary (30%). Combining both studies, two mutations were identified in the same tumor in 30% of the mutated cases, which, taken together with the inactivating nature of the mutations and their remarkable frequency, provided unequivocal evidence that ARID1A is a tumor suppressor gene in these two tumor types. In addition, loss of ARID1A expression was observed in approximately 20% of uterine carcinomas [Wiegand et al., 2011]. In previous studies, chromosomal translocations involving ARID1A were identified in a breast and a lung cancer, though the interpretation of these alterations was challenging [Huang et al., 2007].
The protein encoded by ARID1A is a key component of the highly conserved SWI–SNF (switch/sucrose non-fermentable) chromatin remodeling complex that uses adenosine triphosphate (ATP)-dependent helicase activities to allow access of transcriptional activators and repressors to DNA [Wang et al., 2004; Wilson and Roberts, 2011]. The protein therefore appears to be involved in regulating processes including DNA repair, differentiation, and development [Weissman et al., 2009]. Functional studies by Nagl et al.  have demonstrated that the SWI–SNF complex suppresses proliferation. The ARID1A-encoded protein, BAF250a, is one of two mutually exclusive ARID1 subunits. BAF250a has a DNA-binding domain that specifically binds to AT-rich DNA sequences and is thought to confer specificity to the complex [Wu et al., 2009].
Passenger mutations are best defined as those that do not confer a selective growth advantage to the cells in which they occur, while driver mutations are those which do confer a growth advantage. It is often difficult to distinguish driver mutations from passenger mutations when the mutations occur at low frequency. One of the best examples of this challenge is provided by IDH1 mutations. A single mutation of IDH1, R132H, was discovered in a whole exomic screen of 11 colorectal cancers (CRCs) [Sjöblom et al., 2006]. This mutation was not identified in more than 200 additional CRC samples and was presumed to be a passenger mutation. However, frequent IDH1 mutations at the identical residue were found when brain tumors, such as lower grade astrocytomas and oligodendrogliomas were evaluated [Parsons et al., 2008; Yan et al., 2009]. Thus, the IDH1 mutation in that original CRC in retrospect was undoubtedly a driver.
This example illustrates that once a genetic alteration is identified as a driver in one tumor type, infrequent mutations of the same type in the same gene in other tumors can be more reliably interpreted. Given that, it is now known that ARID1A is a bona fide tumor suppressor gene in OCCC, we applied this principle to the evaluation of ARID1A mutations in other tumor types. As described below, we studied more than 700 different neoplasms of seven different types using Sanger sequencing to determine the contribution of ARID1A alterations to tumorigenesis in general.
Somatic mutations were identified in 43 of the 759 neoplasms studied (6%) (Table 1). Eight neoplasms contained two or three (one case) different mutations, presumably on different alleles, so the total number of mutations was 52. A relatively high frequency of mutations was observed in neoplasms of the colon (10%; 12/119), stomach (10%; 10/100), and pancreas (8%; 10/119). Though only a small number of prostate tumors was available for study, we identified two carcinomas with mutations among the 23 studied. Mutations were observed in three of 125 (2%) medulloblastomas, in four of 114 (4%) breast cancers, and in two of 36 (6%) lung carcinomas (Table 1; Fig. 1). No mutations were observed among 34 glioblastomas or 89 leukemias tested.
|Sample||Tumor type||Nucleotide (genomic)b||Nucleotide (cDNA)c||Amino acid (protein)||Mutation type||MSI status|
|Pa102Ca||Pancreas||g.chr1:26965645G>A||IVS10+1G>A||Splice site||Splice site||MSS|
As expected for inactivating mutations of a tumor suppressor gene, the mutations were distributed throughout the gene and included nonsense variants, out-of-frame and in-frame small insertions and deletions, as well as a small number (three) of missense changes. Mutations were most commonly observed in a seven-base G tract around position g.chr1:26978524 (genomic coordinates refer to hg18) (c.5548), where there were six single base pair deletions and three duplications among gastric, colon, prostate, and pancreas carcinomas. This G tract is the longest mononucleotide repeat in the coding region and the probability of slippage at mononucleotide repeats clearly increases with run length [Eshleman et al., 1996; Markowitz et al., 1995]. Thirty-eight of the 43 samples with somatic mutations were available for microsatellite instability (MSI) testing. Twelve tumors (six colon, five gastric, and one prostate) were shown to be MSI high, and all carried mutations at mononucleotide tracts in the ARID1A gene (Table 1). It is therefore possible that ARID1A, such as TGFβRII or BAX, is associated with MSI and that the homopolymeric repeat frameshifts may result from defects in mismatch repair [Markowitz et al., 1995; Rampino et al., 1997]. Though the interpretation of mutations in mismatch repair-deficient tumors is challenging [Kern, 2002], the fact that approximately 40% of the CRCs with ARID1A mutations did not have MSI leaves little doubt that ARID1A plays a role in this tumor type.
The identification of mutations in ARID1A in several different types of cancer indicates that this gene has a wider role in human tumorigenesis than previously appreciated. These findings are supported by the demonstration of loss of the ARID1A protein, BAF250a, by immunohistochemistry in 14% of gastric and anaplastic thyroid carcinomas [Wiegand et al., 2011] and by the identification of ARID1A point mutations in 3 of 48 pancreatic cancers by Birnbaum et al. . More recently, ARID1A mutations have also been observed in 13% of bladder carcinomas [Gui et al., 2011]. In addition, ARID1A appears to be frequently mutated in gastrointestinal tumors displaying high levels of MSI. Mutations in other members of the SWI–SNF chromatin remodeling complex have also been reported. For example, truncating mutations in SMARCA4/BRG1 were identified in three pancreatic cancers, in a medulloblastoma, and in several lung cancers [Jones et al., 2008; Medina et al., 2008; Parsons et al., 2011]. More recently, 41% of renal cancers have been shown to have truncating mutations in the SWI–SNF chromatin remodeling complex gene, PBRM1 [Varela et al., 2011]. In addition, a pattern of somatic mutation of genes involved more generally in chromatin remodeling is starting to appear. MLL3 appears to be involved in a small number of colon and pancreatic cancers and medulloblastomas [Jones et al., 2008; Parsons et al., 2011; Wood et al., 2007]; MLL2 is mutated in 14% of medulloblastomas and a large fraction of non-Hodgkin's lymphomas [Morin et al., 2011; Parsons et al., 2011] and JARID1C is genetically altered in a small proportion of kidney cancers [Dalgliesh et al., 2010]. These data collectively link genetic alterations to epigenetic changes and pave the way for a better understanding of both.