An Integrated Genomic Approach Identifies ARID1A As a Candidate

Oncogene (2012) 31, 2090–2100 & 2012 Macmillan Publishers Limited All rights reserved 0950-9232/12 www.nature.com/onc ORIGINAL ARTICLE An integrated genomic approach identiﬁes ARID1A as a candidate tumor-suppressor gene in breast cancer

A Mamo1,9, L Cavallone2,9, S Tuzmen3, C Chabot1, C Ferrario1, S Hassan1, H Edgren4,5, O Kallioniemi4,5, O Aleynikova6, E Przybytkowski1, K Malcolm1, S Mousses3, PN Tonin2,7,8 and M Basik1

1Department of Oncology, Lady Davis Institute, Sir Mortimer B Davis Jewish General Hospital, McGill University, Montreal, Que´bec, Canada; 2Department of Human Genetics, McGill University, Montreal, Que´bec, Canada; 3Pharmaceutical Genomics Division, The Translational Genomics Research Institute, Scottsdale, AZ, USA; 4University of Helsinki, Institute for Molecular Medicine (FIMM), Helsinki, Finland; 5VTT Technical Research Centre of Finland and University of Turku, Medical Biotechnology, Turku, Finland; 6Department of Pathology, Jewish General Hospital, Montreal, Que´bec, Canada; 7Department of Medicine, McGill University, Montreal, Que´bec, Canada and 8The Research Institute of the McGill University Health Centre, Montreal, Que´bec, Canada

Tumor-suppressor genes (TSGs) have been classically defined Introduction as genes whose loss of function in tumor cells contributes to the formation and/or maintenance of the tumor phenotype. Cancers occur as a result of dysregulation in the TSGs containing nonsense mutations may not be expressed function of oncogenes and tumor-suppressor genes because of nonsense-mediated RNA decay (NMD). We (TSGs). TSGs have been classically defined as genes combined inhibition of the NMD process, which clears whose loss of function in tumor cells contributes to the transcripts that contain nonsense mutations, with the applica- formation and/or maintenance of the tumor phenotype tion of high-density single-nucleotide polymorphism arrays (Presneau et al., 2003). Various mechanisms of inactiva- analysis to discriminate allelic content in order to identify tion of TSGs have been reported and can include candidate TSGs in five breast cancer cell lines. We identified deletion of one allele through chromosomal non- ARID1A as a target of NMD in the T47D breast cancer cell dysjunction and/or intra-chromosomal rearrangement, line, likely as a consequence of a mutation in exon-9, which and mutational inactivation of the remaining allele introduces a premature stop codon at position Q944. (Presneau et al., 2003). TSGs have been identified in ARID1A encodes a human homolog of yeast SWI1, which cancer both as a result of a search for germline is an integral member of the hSWI/SNF complex, an ATP- mutations in hereditary cancers and the molecular dependent, chromatin-remodeling, multiple-subunit enzyme. characterization of somatic deletions and mutations in Although we did not find any somatic mutations in 11 breast tumor tissue RNA and DNA (Vogelstein and Kinzler, tumors, which show DNA copy-number loss at the 1p36 locus 2004). However, TSGs containing nonsense mutations adjacent to ARID1A, we show that low ARID1A RNA or may not be detectable using RNA-based next-genera- nuclear protein expression is associated with more aggressive tion re-sequencing because RNA containing such breast cancer phenotypes, such as high tumor grade, in two aberrant transcripts is eliminated by the nonsense- independent cohorts of over 200 human breast cancer cases mediated RNA decay process (NMD) (Noensie and each. We also found that low ARID1A nuclear expression Dietz, 2001). Indeed, an interesting strategy to discover becomes more prevalent during the later stages of breast potential TSG candidates was elaborated by Noensie tumor progression. Finally, we found that ARID1A re- and Dietz (2001) and adapted by Huusko et al. (2004). expression in the T47D cell line results in significant inhibition Noensie and Dietz (2001) proposed that it would be of colony formation in soft agar. These results suggest that possible to discover nonsense mutation containing ARID1A may be a candidate TSG in breast cancer. mRNA by inhibiting the NMD process. In this way, Oncogene (2012) 31, 2090–2100; doi:10.1038/onc.2011.386; transcripts containing nonsense mutations are stabilized published online 5 September 2011 and selectively increase in quantity relative to non- mutated transcripts. Huusko et al. (2004) then combined Keywords: ARID1A; nonsense-mediated mRNA decay; NMD inhibition with array comparative genomic tumor-suppressor gene; breast cancer hybridization (aCGH) in order to enrich for the selection of candidates mapping to deleted regions in prostate cancer cell lines, thereby identifying EPHB2 as Correspondence: Dr M Basik, Department of Oncology, Lady Davis a novel candidate TSG in prostate cancer. Institute, Sir Mortimer B Davis Jewish General Hospital, McGill We have adapted the approach of Huusko et al. by University, 3755 Cote Ste-Catherine, Montreál, Que´bec, Canada integrating the allelic content inferred from high-density H3T 1E2. genotyping single-nucleotide polymorphism (SNP) E-mail: [email protected] arrays to identify candidates TSGs in five breast cancer 9These authors contributed equally to this work as first co-authors. Received 16 February 2011; revised 6 July 2011; accepted 26 July 2011; cell lines. Here we report the discovery of ARID1A as a published online 5 September 2011 candidate TSG in the T47D breast cancer cell line by Inactivation of ARID1A in breast cancer A Mamo et al 2091 using this approach. ARID1A encodes a human were identified. The percentage of total transcripts homolog of yeast SWI1, which contains a DNA-binding mapping to LOH regions with an NMD ratio >2 motif (AT-rich interactive domain, ARID) and is an ranged from 1.4 to 4% of all transcripts (Supplementary integral member of the hSWI/SNF complex, an ATP- Figure 2). Reasoning that under-expression is a pheno- dependent, chromatin-remodeling, multiple-subunit en- type of classical TSGs, we then selected candidates zyme (Takeuchi et al., 1997, 1998; Dallas et al., 1998, showing levels of gene expression below the overall 2000; Wang et al., 2004; Wilsker et al., 2004; Patsialou median value of expression for each cell line. Using this et al., 2005). An increasing body of evidence has strategy, the number of candidates decreased to 0.6– demonstrated that SWI/SNF complexes have important 1.9% of total transcripts depending on the cell line roles in gene regulation (Winston and Carlson, 1992; (Supplementary Figure 2). To prioritize candidates for Carlson and Laurent, 1994; Hirschhorn et al., 1995; further analysis, we selected those genes, which showed Kennison, 1995; Sudarsanam et al., 1999, 2000), cell an NMD ratio >2 uniquely in one of the five breast proliferation (Nagl et al., 2005, 2006, 2007) and cancer cell lines. The rationale was based on the development (Carlson and Laurent, 1994; Kennison, assumption that if one gene harbors a nonsense 1995). Several genes encoding hSWI/SNF components mutation in a given cell line, it would be very unlikely have been associated with tumorigenesis (Medina and that the same gene would have a nonsense mutation in Sanchez-Cespedes, 2008; Roberts and Biegel, 2009; another cell line of a small sample set because of the very Rodriguez-Nieto and Sanchez-Cespedes, 2009). More- low frequency of nonsense mutations in tumor cells over, Huang et al. (2007) reported that an antisense (Sjoblom et al., 2006). Applying this selection strategy to cDNA resulting from a genomic rearrangement invol- all five breast cancer cell lines led to the selection of 53 ving the ARID1A gene in a primary breast carcinoma candidate TSGs, or 0.03% of the total number of could transform NIH3T3 cells. During the course of our transcripts on this microarray platform (Supplementary investigation, mutations in the ARID1A were reported Figure 1 and Table 1). to occur in up to 50% of ovarian clear cell carcinomas Sequencing of the 53 candidate genes was begun in the (Jones et al., 2010; Wiegand et al., 2010). Thus ARID1A respective cell lines used for their selection. We began appears to show all of the hallmarks of a classical validating our list starting with genes that have been TSG (Presneau et al., 2003). reported to possess potential tumor-suppressive function We also report on ARID1A expression in human or to be involved in breast cancer. A total of 15 genes breast cancers. We found that low ARID1A RNA and/ were sequenced before we found a nonsense mutation in or nuclear protein expression is associated with more CDH1 in the MDA-MB-453 cell line. CDH1 encodes aggressive breast cancer phenotypes. Finally, we found cadherin-1 and is known to be mutated in the MDA-MB- that ARID1A re-expression in the T47D breast cancer 453 cell line (Forbes et al., 2010), a mutation, which was cell line results in significant inhibition of colony verified in our laboratory by DNA sequencing. CDH1 or formation in soft agar. These results suggest that the cadherin is a well-known TSG, responsible for a ARID1A may be a TSG in breast cancer, and that it hereditary cancer syndrome (Campeau et al., 2008). warrants further investigation as a potential diagnostic As Huang et al. (2007) reported that the ARID1A and therapeutic marker in breast cancer. gene was genomically rearranged in a primary breast carcinoma, we also focused on the ARID1A gene in our list. We identified a nonsense mutation, c.944C>T, in exon-9 in the ARID1A gene, which introduces a Results premature stop codon at amino-acid position 944 (Figure 1a). We confirmed that the RNA levels of NMD inhibition combined with genomic analysis identifies ARID1A were very low and increased when the T47D ARID1A as a potential candidate TSG in breast cancer cell line was treated with emetine, the NMD inhibitor A modification of a strategy described by Huusko et al. (Figure 1b). ARID1A is located at chromosomal band (2004) was used to identify transcripts with nonsense 1p36.11, mapping within the LOH region shown by the mutations. We inhibited NMD using emetine followed entire 1p arm in the T47D cell line (Figure 1c). ARID1A by inhibition of de novo transcription using actinomycin- encodes a human homolog of yeast SWI1, which D (Supplementary Figures 1 and 2), in five breast cancer contains a DNA-binding motif (AT-rich interactive cell lines (MCF-7, MDA-MB-231, MDA-MB-361, domain, ARID). As this TSG candidate has not been MDA-MB-453 and T47D). Gene expression analysis extensively characterized, we further investigated its role comparing emetine-treated to untreated cells showed as a putative TSG in breast cancer cells and tumors. that the percentage of transcripts stabilized (that is, upregulated by two-fold) after drug treatment ranged from 8.5 to 10% of total transcripts in each of the five Genomic alterations involving ARID1A in primary breast treated breast cancer cell lines. cancers To enrich for classical TSG candidates, genes showing To determine whether the chromosomal region contain- an NMD ratio of >2 in emetine-treated versus ing ARID1A is the target of genomic alterations in untreated cell lines and mapping to regions of loss of primary breast cancers, we analyzed 82 consecutive heterozygosity (LOH) or homozygous deletions inferred primary breast tumors by high-resolution aCGH and by high-density, whole-genome SNP BeadChip analysis found that 11 (13%) of these tumor samples showed a

Oncogene Inactivation of ARID1A in breast cancer A Mamo et al 2092 Table 1 Candidate tumor suppressor genes mapping to LOH regions determined by SNP array analyses Cell line Gene symbol Chromosomal position Cytoband NMD cDNA microarray location ratio gene expression

MCF7 CASP5 chr11:104,864,967–104,893,895 11q22.3 2.7 0.1 MCF7 COL4A2 chr13:110,959,631–111,165,371 13q34 2.0 0.0 MCF7 CXADR chr21:18885330–18937929 21q21.1 3.1 0.2 MCF7 DNAH1 chr3:52350335–52434512 3p21.1 2.2 1.0 MCF7 ETV1 chr7:13930858–14031050 7p21.2 2.1 0.1 MCF7 POU2F3 chr11:120110951–120190653 11q23.3 16.1 0.2 MCF7 SLITRK2 chrX:144899347–144907358 Xq27.3 2.8 0.7 MCF7 TMEM56 chr1:95583479–95710916 1p21.3 2.0 0.8 MCF7 ZNF192 chr6:28109716–28125236 6p22.1 2.5 1.0 MCF7 RNF160 (LTN1) chr21:30300466–30365277 21q21.3 2.5 0.6

MDA-MB-231 ChGn chr8:19297341–19460056 8p21.3 2.6 0.5 MDA-MB-231 CPNE4 chr3:131,253,584–131,754,286 3q22.1 2.1 0.4 MDA-MB-231 ITGA11 chr15:68,594,050–68,724,492 15q23 3.8 0.6 MDA-MB-231 LDOC1L chr22:44,888,450–44,894,005 22q13.31 2.2 1.0 MDA-MB-231 NXF5 chrX:101087085–101112549 Xq22.1 2.1 0.5 MDA-MB-231 OR1D2 chr17:2,995,353–2,996,290 17p13.3 2.5 0.8 MDA-MB-231 PRO1768 chr14:90,042,560–90,043,818 14q32.11 2.1 1.0 MDA-MB-231 SERPINA1 chr14:94,843,085–94,857,029 14q32.13 3.0 0.6 MDA-MB-231 SMPD2 chr6:109,761,931–109,765,121 6q21 2.7 0.6 MDA-MB-231 SNAI2 chr8:49,830,239–49,833,988 8q11.21 3.0 0.6 MDA-MB-231 UBN1 chr16:4,897,912–4,932,357 16p13.3 2.8 0.8 MDA-MB-231 WDR59 chr16:74,907,471–75,019,017 16q23.1 2.2 0.6

MDA-MB-361 BNC2 chr9:16,409,502–16,870,786 9p22.3-p22.2 2.1 0.6 MDA-MB-361 DCT chr13:95,091,843–95,131,936 13q32.1 2.2 0.6 MDA-MB-361 FGF14 chr13:102,373,205–103,054,124 13q33.1 2.4 0.5 MDA-MB-361 GAS7 chr17:9,813,926–10,101,868 17p13.1 2.3 0.5 MDA-MB-361 IFNA5 chr9:21,304,687–21,305,255 9p21.3 2.2 0.7 MDA-MB-361 ITIH3 chr3:52,828,784–52,843,025 3p21.1 2.1 0.8 MDA-MB-361 KCNH4 chr17:40,308,910–40,333,296 17q21.2 2.0 0.7 MDA-MB-361 KCNJ12 chr17:21,279,699–21,323,179 17p11.2 2.7 0.0 MDA-MB-361 PLD2 chr17:4,710,421–4,726,727 17p13.2 2.2 0.6 MDA-MB-361 RRP9 chr3:51967446–51975922 3p21.1 2.1 0.7 MDA-MB-361 SPAG11 chr8:7705402–7721318 8p23.1 2.2 0.7 MDA-MB-361 XKRX chrX:100,168,431–100,183,898 Xq22.1 2.2 0.9

MDA-MB-453 CDH1 chr16:68,771,195–68,869,444 16q22.1 6.4 0.2 MDA-MB-453 CLDN23 chr8:8,559,666–8,561,616 8p23.1 2.1 0.7 MDA-MB-453 GRP183 chr13:99946790–99959749 13q32.3 2.3 0.5 MDA-MB-453 HOMER2 chr15:83,517,738–83,621,473 15q25.2 4.2 0.2 MDA-MB-453 LOXL1 chr15:74,218,789–74,244,478 15q24.1 2.2 0.1 MDA-MB-453 OBP2A chr9:138,437,985–138,441,814 9q34.3 2.1 0.8 MDA-MB-453 PTCH chr9:98231033–98242870 9q22.32 2.6 0.2 MDA-MB-453 SERPINF2 chr17:1,646,130–1,658,559 17p13.3 26.2 0.6 MDA-MB-453 TNFSF12 chr17:7452375–7461206 17p13.1 3.3 0.1

T47D ADCY7 chr16:50,321,823–50,352,043 16q12.1 2.2 0.2 T47D ARID1A chr1:27,022,522–27,108,601 1p36.11 2.6 0.4 T47D CREBL2 chr12:12,764,831–12,798,041 12p13.1 2.4 0.5 T47D FAM171B chr2:187,558,789–187,628,510 2q32.1 2.3 0.4 T47D HBE1 chr11:5,289,580–5,291,386 11p15.4 2.3 0.7 T47D MAMLD1 chrX:149,531,686–149,682,446 Xq28 3.1 0.2 T47D PPIL6 chr6:109,711,419–109,761,847 6q21 3.0 0.3 T47D SPEN chr1:16,174,359–16,266,950 1p36.21- 2.8 0.5 p36.13 T47D STK32C chr10:134,020,996–134,121,477 10q26.3 2.1 0.9 T47D WSCD1 17p13.2 2.5 1.0

Abbreviations: NMD, nonsense-mediated RNA decay; SNP, single-nucleotide polymorphism. Gene symbol, chromosomal position and cytoband location based on Human Genome Browser Gateway (NCBI36/hg 18) assembly.

DNA copy-number loss in the region overlapping the ARID1A RNA expression in primary breast tumor ARID1A locus on chromosome 1p36.11 (Figure 2b). samples Sequencing of the protein-encoding exons of ARID1A To better understand the role of ARID1A expression in in breast tumor samples showing DNA copy-number breast cancer, we studied the clinico-pathological factors loss revealed no evidence of sequence variations. associated with its gene expression in a well-annotated

Oncogene Inactivation of ARID1A in breast cancer A Mamo et al 2093

Emetine - + - + (T= 8h) Reference ARID1A

Exon 9 ARID1A ARID1A in T47D

GAPDH

B allele frequency

Log R ratio

DS2885 ARID1A D1S247 Figure 1 The ARID1A gene is mutated in the T47D breast cancer cell line. (a) Sequence analysis identified a C-to-T substitution resulting in a TAG stop codon (the plus strand is shown) in exon-9 of ARID1A in T47D breast cancer cell line. (b) Reverse transcription–-PCR analysis of ARID1A mRNA levels in both emetine-treated and untreated breast cancer cell lines, showing an emetine-induced increase in ARID1A mRNA levels in T47 cells and not in MDA-MB-453. GAPDH is used as the control gene. (c) SNP array imaging results for chromosome-1 of the T47D cell line. The top plot of the figure shows the B-allele frequency for each SNP marker aligned to its chromosomal position, and, as shown, indicates LOH for the entire chromosome-1p arm. The bottom plot of the figure contains the log R ratio providing an indication of the copy number for each SNP marker aligned to its chromosomal position, indicating a lower copy number for the entire 1p arm relative to the 1q arm. The location of the ARID1A gene on the chromosomal map, as well as of the polymorphic microsatellite repeat markers (D1S247 and D1S2885) used to assess LOH in tumors, are indicated below left. database containing 251 independently ascertained in the p53-‘deficient’ tumors than in the p53 ‘wild-type’ breast tumors (Miller et al., 2005). ARID1A RNA tumors (Figure 3). expression values were significantly lower in grade-III tumors (versus grade-I/II) (Po0.0001), PRÀ (Po0.05), high Ki67 RNA expression (Po0.05) and high ERBB2 ARID1A protein expression in primary breast RNA expression values (Po0.05) (Figure 3). These data tumor samples suggest that ARID1A RNA expression is lower in high- As ARID1A functions as part of the SWI/SNF grade, hormone-independent, HER2 þ and highly transcription-regulating complex, we expected that its proliferative breast tumors, all of which have been expression in the nucleus would be important for TSG shown to be indicators of more aggressive breast cancer activity and that such expression would be higher in phenotypes. normal breast as compared with breast tumor tissues. ARID1A RNA expression was also compared with We performed immunohistochemical analysis of the expression of a p53 signature gene set in the same ARID1A protein by using a tissue microarray (TMA) data set. This gene set distinguishes tumors with a containing samples from 236 breast tumors, and transcriptional fingerprint of p53 deficiency or low matching normal breast tissues and adjacent pre- expression, thus potentially identifying not only tumors invasive breast lesions from many of the patients. with p53 mutations, but also those harboring other Nuclear staining was categorized into low (0 or forms of inactivation of p53 expression (for example, 1 þ staining) or moderate/high (>1 þ staining). We p53 methylation, MDM2 amplification). We found that observed a progressive decrease in the relative propor- ARID1A expression was significantly lower (Po0.001) tion of moderate/high nuclear staining as compared with

Oncogene Inactivation of ARID1A in breast cancer A Mamo et al 2094

Figure 2 ARID1A genomic alterations in breast tumors. Examples of breast tumor samples (numbers 1, 30 and 231) depicting copy- number anomalies of chromosome-1 loci using the 244K aCGH microarray. Individual probe values are represented as normalized log2 ratios. The red dots represent probes with increased copy-number values (>log2 ¼ 1) and the green dots represent probes with decreased copy-number values (olog2 ¼À1). DNA copy-number losses are apparent at the telomeric end of the 1p arm, including the locus of the genomic region containing ARID1A, as indicated by the blue box inset.

ARID1A ––––* *** *** * *

9.4

9.2

9.0

8.8 expression (log2) expression

8.6

8.4 G1 G2 G3 LN– LN+ ER– ER+ PgR– Basal PgR+ LumA LumB p53_wt Normal ERBB2 p53_mut tsize >40 tsize Ki67_low Ki67_high tsize <=20 tsize survival <3 survival >8 PCNA_low tsize 21–40 tsize PCNA_high ERBB2_low survival 3–8 ERBB2_high death_cancer age at diag <50 age at diag >65 age at diag 50-65 alive_or_censored Figure 3 Clinico-pathological correlations of ARID1A RNA expression. Plot of ARID1A RNA expression in groups of breast tumors classified according to different clinico-pathological variables, including RNA expression of ER, PR, PCNA, Ki67 and ERBB2 genes, as well as grade (G1, G2, G3), tumor size (tsize), age at diagnosis, lymph node status (LN), overall survival, cancer-related mortality and molecular subtype of breast cancer in an independent gene expression database of breast cancers. The asterisks indicate statistical significance for categories with two variables. One asterisk, Po0.05; three asterisks, Po0.001. (À) indicates no statistical significance.

Oncogene Inactivation of ARID1A in breast cancer A Mamo et al 2095 low nuclear ARID1A protein staining as one moved and empty vector-transfected T47D cells in culture from samples of normal breast tissue adjacent to (Supplementary Figure 3). Importantly, ARID1A- tumors, through progressively more advanced stages of transfected T47D cells were able to form significantly breast tumor progression (hyperplasia without atypia, fewer colonies in soft agar than cells transfected with the atypical ductal hyperplasia, ductal carcinoma in situ)to empty vector (Figure 5). samples of invasive breast carcinoma and metastatic Conversely, transfection of ARID1A small interfering lymph nodes (Figure 4a). Indeed, only 37% of normal RNA (siRNA) into the MCF-7 breast cancer cell breast epithelial cells show low nuclear staining com- line that expresses high levels of ARID1A resulted in a pared with 57% of ductal carcinoma in situ, 64% of significant increase in proliferation in these cells invasive breast tumor cells and 80% of metastatic lymph (Supplementary Figure 4). Taken together, these find- nodes. This trend of progressively lower ARID1A protein ings support a potential TSG function for ARID1A in nuclear expression corresponding to the various phases of breast cancer cells. breast tumor progression was significant using Spear- man’s rank correlation (r-value of À0.28, Po0.0001). By contrast, cytoplasmic staining of ARID1A protein was less intense across all lesions (data not shown). Thus Discussion ARID1A protein expression may be downregulated during breast tumor progression, or a selection of breast We describe the application of an integrated genomic epithelial cells expressing lower nuclear protein levels may approach aiming at identifying candidate TSGs in take place during tumor progression. breast cancer, which resulted in the discovery of an We also found that nuclear ARID1A protein expres- ARID1A gene mutation in the T47D breast cancer cell sion correlated with estrogen receptor (ER) and line. This approach is based on the manipulation of a progesterone receptor (PR) positivity, and correlated homeostatic cellular process, NMD, whose function is inversely with histological grade and ERÀ/PRÀ/Her2À to eliminate transcripts containing nonsense mutations. (triple negative) status. Thus, low-grade or hormone Such a manipulation enables the analysis of transcripts, receptor-expressing tumors were more likely to show which would otherwise not be present in tumor RNA, moderate/high nuclear protein expression of ARID1A and therefore not be detectable using RNA-based tumor than high-grade and triple-negative tumors (Figures 4b re-sequencing approaches. Once candidate NMD gene and c, and Table 2). These findings suggest that low/ targets are uncovered, we performed whole-genome absent nuclear protein levels are associated with more genotyping using SNP arrays in order to identify aggressive clinico-pathological features of breast NMD target candidates located in regions of LOH or tumors, and are consistent with our observations for deletions. These criteria allowed us to enrich for 0.03– ARID1A RNA expression. 0.04% of all genes as candidate TSGs in five breast To determine the prognostic significance of ARID1A cancer cell lines. We obtained 9–13 top candidates per protein expression, survival analysis was performed by cell line, including CDH1, a gene known to contain a using the nuclear staining score as above. In univariate nonsense mutation in the MDA-MB-453 cell line in the analysis, patients with low nuclear protein staining had Catalogue of Somatic Mutations in Cancer (COSMIC) a trend with a two-fold higher risk of breast cancer- database (Forbes et al., 2010), validating our method of related mortality (hazards ratio, 1.98; 95% confidence enriching for candidate TSGs. CDH1 is a known TSG, interval, 0.86-6.3; P ¼ 0.088). implicated in rare cases of hereditary diffuse gastric cancer Finally, many of the samples for which we had syndrome, which also features lobular breast carcinomas performed aCGH had tissue cores represented on this (Campeau et al., 2008). Somatic mutations in CDH1 have TMA. Nine of the 11 clinical samples with 1p36.11 been reported in breast carcinomas (Berx and van Roy, DNA copy-number loss were represented on the 2009) and loss of function is thought to contribute to TMA used for ARID1A protein expression analysis. progression in cancer by increasing proliferation, invasion We found that eight of these nine samples with DNA and/or metastasis (Strumane et al., 2004). copy-number loss at 1p36.11 also had low levels A previously unreported nonsense mutation was of ARID1A nuclear protein expression, suggestive of a identified in ARID1A in the T47D breast cancer cell concordance between DNA copy-number loss and line. Although somatic mutations were not identified in ARID1A protein expression. breast tumors harboring LOH of the ARID1A locus, mutation screening was limited to only 11 samples. However, we did find that both ARID1A gene and ARID1A inhibits the growth of T47D breast cancer protein expression levels are related to clinico-patholo- cells in soft agar gical features of breast cancer, including hormone To assess the tumor-suppressive properties of ARID1A receptor status, tumor grade and p53 activation status. in vitro, we introduced a full-length ARID1A-expressing We showed that low ARID1A expression is associated vector into T47D breast cancer cells. T47D breast with high tumor grade, p53 inactivation, and high cancer cells were stably transfected with the empty Ki67 and ERBB2 RNA expression. We also showed vector alone or with a full-length, sequence-verified that ARID1A expression significantly inhibits the ability ARID1A cDNA (Figure 5). There was a significant of T47D breast cancer cells to form colonies in soft agar, difference in proliferation between ARID1A-transfected and that its silencing by siRNA in highly expressing

Oncogene Inactivation of ARID1A in breast cancer A Mamo et al 2096

100%

90%

80% Low ARID1A 70%

60% Med-High 50% ARID1A 40%

30%

20%

10%

0% N H ADH DCIS T LN n=134 n=20 n=7 n=99 n=236 n=41

100% 100% 90% 90% 80% 80% 70% 70% 60% 60%

50% ARID1A 50% mod/high 40% 40% 30% ARID1A 30% low 20% 20% 10% 10% 0% 0% ER positive ER negative gr I gr II gr III

Normal breast Breast tumor Figure 4 Clinico-pathological correlations of ARID1A nuclear protein expression. (a) Bar graph of relative proportion of absent/low (0 or 1 þ ; white) versus moderate/high (2 þ or 3 þ ; black) nuclear staining intensity for the different stages of breast tumor progression (normal adjacent breast tissue (N), hyperplasia without atypia (H), atypical ductal hyperplasia (AH), ductal carcinoma in situ (DCIS), invasive adenocarcinoma (T) and metastatic lymph nodes (LN)) on a TMA. Note the progressive relative increase in absent/low nuclear staining as one moves to more invasive tumors. The value on the y-axis represents the percentage of samples within each category showing different ARID1A staining. The total number of samples in each category is shown below on the x-axis. (b) Bar graph showing relative proportion of ARID1A absent/low (0 or 1 þ ; black) and ARID1A moderate/high (2 þ or 3 þ ; stripes) expression as ER þ and ERÀ tumors. ER status was deﬁned at clinical pathological examination of the tumors. (c) Bar graph showing relative proportion of ARID1A absent/low (0 or 1 þ ; black) and ARID1A moderate/high (2 þ or 3 þ ; stripes) expression as a function of grade-I, grade-II and grade-III breast tumors on a breast tumor TMA. Assessment of histological grade was performed at clinical pathological examination of the tumors. (d) Photomicrographs of ARID1A staining of matched tumor and adjacent normal tissue from the same patient showing nuclear staining in the normal breast epithelium and not in tumor cells.

MCF-7 cells increases proliferation rates. Taken to- ARID1A encodes for AT-Rich interactive domain- gether, our data further support the notion of a tumor- containing protein-1A and is a component of the SWI/ suppressive function for ARID1A in breast cancer. SNF chromatin-remodeling complex, which functions

Oncogene Inactivation of ARID1A in breast cancer A Mamo et al 2097 Table 2 Clinico-pathological parameters of ARID1A nuclear expres- cancers and skin cancers, but none in a variety of cancer sion in breast cancer types including 11 breast cancers (Forbes et al., 2010). DNA copy-number changes of the 1p36 locus contain- ARID1ANuc ing ARID1A were observed in 13% of breast tumors. Variable r P-value Although associated with low ARID1A protein expression, these changes cannot account for all of the breast ER-positive 0.24 0.0002 PR-positive 0.19 0.0028 tumors showing low or absent ARID1A expression. ERÀ/PRÀ/Her2ÀÀ0.23 0.0013 Epigenetic inactivation of ARID1A expression may HER2-positive À0.096 0.18 account for the apparent decreased expression of Lymph nodes À0.048 0.49 ARID1A RNA. Methylation silencing of the predicted Stage À0.048 0.49 Grade À0.24 0.0006 CpG island overlapping the ﬁrst exon of ARID1A as a Tumor size À0.046 0.49 possible mechanism of inactivation of gene expression requires further investigation (Dworkin et al., 2009; Orlando and Brown, 2009). Although microRNAs 0 1400 targeting the 3 -untranslated region of ARID1A have 1200 been predicted by TargetScanS bioinformatic analysis 1000 (http://genome.ucsc.edu), to our knowledge none * p< 0.01 have been investigated as a possible mechanism for the 800 under-expression of ARID1A. 600 There is an increasing body of evidence that several 400 members of the SNF/SWI are implicated in different

Number of colonies 200 types of cancers (Rozenblatt-Rosen et al., 1998; Adler 0 pCMV6-XL4 pCMV6-XL4-ARID1A et al., 1999; Grand et al., 1999; Yuge et al., 2000; Biegel and Pollack, 2004; Medina and Sanchez-Cespedes, 2008; Roberts and Biegel, 2009), which suggests a role for chromatin remodeling in the pathogenesis of many cancers. Other members of the SWI/SNF complex besides ARID1A may affect ER-mediated transcription (Belandia et al., 2002), as well as sensitivity to anti- estrogens (Zhang et al., 2007). The presence of a pCMV6-XL4 pCMV6-XL4-ARID1A recognized DNA-binding domain in ARID1A and its demonstrated ability to bind to linear duplex DNA suggest that ARID1A contributes to the DNA binding activity in the SWI/SNF complex, although DNA binding through ARID regions is not restricted to pCMV6-XL4 pCMV6-XL4- ARID1A AT-rich sequences, consistent with a wider range of ARID1A DNA interactions for these proteins (Dallas et al., 2000). The two ARID-containing proteins (ARID1A and Lamin B ARID1B) are mutually exclusive subunits of the SWI/ SNF complexes, such that SWI/SNF complexes Figure 5 Overexpression of ARID1A decreases anchorage-inde- endowed with anti-proliferative properties contain pendent growth of the T47D breast cancer cell line. (a) Anchorage ARID1A, whereas those endowed with pro-proliferative independence was determined by colony formation assay in soft agar. The means and standard errors were obtained from two properties contain ARID1B (Nagl et al., 2007). The independent experiments each performed in triplicates. Difference anti-proliferative activity of ARID1A is underscored by was analyzed by t-test; *Po0.01. (b) Photomicrographs of tissue the ability of ARID1A-containing SWI/SNF complexes culture plates showing results of soft agar growth of T47D cells to repress E2F1 (Van Rechem et al., 2009), Myc, cyclin- transfected with ARID1A (right well) or empty vector (left well). (c) Nuclear extracts from T47D stably transfected with the B2 and Cdc2 function (Nagl et al., 2006). This is pCMV6-XL4 empty vector or the ARID1A expression vector consistent with our finding that ARID1A RNA levels (pCMV6-XL4-ARID1A) were analyzed by immunoblotting using are significantly lower in high-Ki67-expressing breast the indicated antibodies. tumors (Figure 2). More recent evidence also implicates ARID1A in the regulation of histone modifications, to regulate gene transcription as well as chromatin specifically in the targeting of histone H2B for ubiqui- dynamics (Winston and Carlson, 1992; Carlson and tination (Li et al., 2010), with consequent effects on E2F, Laurent, 1994; Hirschhorn et al., 1995; Kennison, 1995; histone deacetylase and Myc-dependent transcription. Sudarsanam et al., 1999, 2000). During the course of this Our study confirms that the ARID1A protein has the investigation, there were two independent reports of potential to play a tumor-suppressive role in the ARID1A mutations in 50% of endometroid ovarian formation of breast cancer, and that its low expression carcinomas and 57% of clear cell ovarian carcinomas is associated with more aggressive phenotypes of breast (Jones et al., 2010; Wiegand et al., 2010). The COSMIC cancer such as high tumor grade, and with colony reports only four mutations in ARID1A: one each in formation capacity in vitro in breast cancer cells carrying central nervous system cancers, kidney cancers, lung a nonsense mutation in this gene. These data add to the

Oncogene Inactivation of ARID1A in breast cancer A Mamo et al 2098 growing list of evidence supporting a critical role for generated by using the Primer3 software (http://frodo.wi.mit. SWI/SNF complex members in solid tumors such as edu/cgi-bin/primer3/primer3_www.cgi). The primer pairs for breast cancer. ARID1A are provided in Supplementary Table 1. Forward and reverse PCR primers were required to be located no closer than 50 bp to the target exon boundaries. Exons larger than 450 bp were analyzed as multiple overlapping amplicons. PCR Materials and methods products were designed to range in size from 300 to 700 bp, which was considered optimal for amplification, purification Cell lines and breast tumors and sequencing. To minimize amplification of homologous Five breast cancer cell lines (MCF7, MDA-MB-231, MDA- genomic sequences, primer pairs were filtered by using the MB-361, MDA-MB-453 and T47D) were obtained from the UCSC In Silico PCR software, and only pairs yielding a single American Type Culture Collection (ATCC, Manassas, VA, product were used. PCR products were purified as recom- USA) and tested for mycoplasma at reception and every 6 mended by the manufacturer (Qiagen) and DNA sequencing of months thereafter using the Mycoplasma Detection kit products was performed at the McGill University and Genome (Invitrogen, Burlington, ON, Canada). Cells were maintained Quebec Innovation Centre as well as at the TGEN DNA in RPMI supplemented with 10% fetal calf serum (Invitrogen) sequencing facility. Sequence chromatograms were aligned and at 37 1C in a humidified incubator in 5% CO2. DNA was analyzed with the Staden package (Staden et al., 2000) and extracted from these cell lines by using the Qiagen DNAmp the Mutation surveyor software Version 3.24 by using the extraction kit (Qiagen, Germantown, MD, USA) according reference sequence NM_006015.4. to the manufacturer’s instructions. Three hundred consecutive breast tumors were collected as part of a tumor banking effort funded by the Fonds de la Reverse transcription reactions of ARID1A recherche en sante´du Que´bec tumor bank at the Centre For semi-quantitative reverse transcription–PCR, cDNA was Hospitalier de l’Universite´de Montreál (CHUM) from 2000 to generated from 10 mg of total RNA extracted from breast 2003. Patients had signed informed consent for breast tumor cancer cell lines by using oligodT and SuperScript II reverse transcriptase (Invitrogen). The primers used for exon-9 of the banking. One hundred of these tumor samples were used for 0 DNA extraction. Tissue sections for each tumor showed ARID1A gene were 5 -TATATGCAGAGGAACCCCCAG AT-30 (forward reaction) and 50-CATTGGGCAAGGCATTA >70% tumor cells as determined by a hematoxylin and eosin 0 staining for each sample. TAGTT-3 (reverse reaction). PCR was performed at an annealing temperature of 60 1C and for 26 cycles. Glyceralde- hyde-3-phosphate dehydrogenase (GAPDH) was amplified NMD assay of breast cancer cell lines simultaneously as a control for the efficiency of cDNA synthesis. NMD inhibition was performed as described previously (Huusko et al., 2004) and as shown in supplementary Figure 1. Cell pellets were frozen and total RNA was extracted aCGH of breast tumors by using the RNeasy kit (Qiagen) in accordance with the DNA was extracted from 14 frozen breast tumors by using the manufacturer’s instructions. Gene expression profiling was DNAamp kit (Qiagen), a subset of the 100 consecutively banked performed as reported previously (Hosein et al., 2010). The breast tumors that showed LOH as above. DNA quality was ratio of individual transcript expression in emetine-treated assessed on a 2100 bioanalyzer using a DNA 12000LabChip versus untreated cells was determined at different time points. kit (Agilent Technologies, Palo Alto, CA, USA). aCGH was The NMD ratio used for analysis and selection of TSG performed as reported previously (Hosein et al., 2010). candidates is the average of the ratios obtained per each of the four time points. Gene expression database analysis A published data set of 251 breast tumors profiled on both High-density SNP arrays of cell lines Affymetrix U133A and U133B arrays (Miller et al., 2005) was Genome-wide chromosomal abnormalities were inferred by reanalyzed for this study. Briefly, microarray data were the Infinium genotyping technology using Sentrix Human-1 preprocessed by using the open source R statistical program- BeadChip (Illumina, San Diego, CA, USA; http://www. ming language (R development core team) and the RMA illumina.com), performed at the McGill University and method as implemented in the Bioconductor affy Package Genome Quebec Innovation Centre (Montreal, QC; gqinno- (www.bioconductor.org). Data were preprocessed directly to vationcenter.com). A 750-ng weight of DNA was used and the Ensemble gene IDs (Dai et al., 2005). For genes appearing on BeadChips were scanned by using the BeadArray Reader. both array types, data were combined by calculating their Genotyping analysis was performed by using the BeadStudio median expression values from both arrays. ERBB2, Ki67 and software version 2.1, which provides an estimate of the allele PCNA status was estimated from the expression data itself. frequency and copy number (log R ratio). LOH is inferred Tumor molecular subtype classification was performed based by (B) allele frequency where values deviating from 0.5 (less on the parameters described by Parker et al. (2009). Mann– than 0.2 and greater than 0.6) indicate allelic imbalance or Whitney U-test was used to test the significance of the LOH. SNP information was based on NCBI Build 36.1 difference between the medians of two category phenotypes. (www.ncbi.nlm.nih.gov/genome). Immunohistochemistry Mutation analysis of candidate TSGs Immunohistochemical staining of ARID1A was performed The genomic sequences of the entire coding region of each by using a TMA slide set containing duplicate or triplicate exon of each candidate TSG were obtained from the UCSC 0.6-mm cores from 237 breast adenocarcinoma samples as well Santa Cruz Genome Bioinformatics Site (http://genome. as pre-invasive lesions from the same surgical resection ucsc.edu). All genomic positions correspond to UCSC Santa specimens, as described by Hassan et al. (2009). The samples Cruz hg18 build 36 human genome sequence. Primer pairs for were from patients with a 3.3-year median follow-up. Nine PCR amplification and sequencing of each coding exon were percent were Her2 þ by immunohistochemistry (either 2 þ or

Oncogene Inactivation of ARID1A in breast cancer A Mamo et al 2099 À À À À À À 3 þ ), and 18% were ER ,PR and Her2 (ER /PR /Her2 ), tase inhibitors Na3VO4 (1 mM) and NaF (5 mM) were added to or ‘triple negative’, as determined by the clinical pathology both homogenization and extraction solutions. siRNA-treated report. TMA staining was performed as reported previously MCF-7 cells were collected 72 h after siRNA transfection, (Hassan et al., 2009). We used the mouse monoclonal anti- washed in cold phosphate-buffered saline, resuspended ARID1A antibody (sc-81193; Santa Cruz, Santa Cruz, CA, in Kinexus lysis buffer (Kinexus, Victoria, BC, Canada) and USA) diluted 1/30 in blocking solution. Bound antibody was incubated on ice for 15 min. Cell lysates were then centrifuged detected with goat anti-mouse biotin-conjugated secondary at 13 000 r.p.m. for 15 and the supernatants were collected. antibodies (2 mg/ml; Jackson ImmunoResearch Laboratories, Protein concentration was determined by the Bradford protein West Grove, PA, USA). Two independent observers read the assay (Coomassie Plus-The better Bradford assay reagent; Pierce, slides in a blinded manner (MB and OA). Only epithelial cells Rockford, IL, USA). Proteins from nuclear extracts (80 mg) were were evaluated and the result for each core was recorded resolved by 8% sodium dodecyl sulfate–PAGE and transferred separately. The average maximal staining intensity (no staining to nitrocellulose Hybond-C Extra membranes (Amersham (0), low (1 þ ), moderate (2 þ ) or high (3 þ ) for each of two Biosciences/GE Healthcare, Quebec, Canada). Western blotting cores per sample) was recorded for each sample. Statistical and enhanced chemiluminescence detection were then performed analysis was performed according to Hassan et al. (2008). according to the manufacturer’s recommendations (Amersham Biosciences/GE Healthcare). Generation of stable ARID1A-transfected T47D cell lines and siRNA transient transfection of MCF-7 cells In vitro assays The pCMV6-XL4 and pCMV6-XL4-ARID1A (catalog no. T47D cells stably transfected with pCMV6-XL4-ARID1A or SC303719) vectors were purchased from OriGene Technolo- the empty vector pCMV6-XL4 were seeded in a 96-well plate gies (Rockville, MD, USA). The pIRES2-DsRed2 vector was (5000 cells per well) in 200 ml of RPMI-1640 medium contain- provided by Dr Raquel Aloyz at the Segal Cancer Center. ing 10% fetal bovine serum. For the siRNA-treated MCF-7 Antibodies against ARID1A (BAF250a; clone PSG3) cells, these cells were seeded in a 96-well plate (2500 cells per and lamin-B (clone C-20) were purchased from Santa Cruz well) 72 h after siRNA transfection in 200 ml of Dulbecco’s Biotechnology. Antibody against a-tubulin (DM1A) was modified Eagle’s medium high-glucose medium containing obtained from Abcam (Cambridge, UK). Generation of the 10% fetal bovine serum. Twenty-four hours after seeding, the T47D polyclonal cell line stably overexpressing ARID1A was medium was changed and cell number was determined 5 days accomplished by co-transfection of the pCMV6-XL4-ARI- later by the Alamar Blue assay (Invitrogen) according to the D1A expression plasmid and the pIRES2-DsRed2 plasmid in a manufacturer’s instructions. For the soft agar growth assay, ratio of 10:1 with the Lipofectamine LTX and PLUS Reagents T47D cells (0.5 Â 104) stably transfected with pCMV6-XL4- according to the manufacturer’s instructions (Invitrogen). ARID1A or the control empty vector were suspended in 1 ml of Forty-eight hours after transfection, transfected cells were soft agar (0.30% Bactoagar in RPMI with 10% fetal bovine selected by treatment with 400 mg/ml geneticin (G418; Wisent, serum), plated onto 2 ml of solidified agar (0.70% Bactoagar in Quebec, Canada) and stably transfected cells were maintained RPMI with 10% fetal bovine serum) in six-well plates and in selection for subsequent passages at the same concentration cultured for 1 month. Colonies were scored electronically using of G418. Knockdown of ARID1A in MCF-7 cells was an automated cell colony counter (GelCount; Oxford Optronix, performed by transient transfection of 2.5 Â 105 cells in a 60- Oxford, UK). mm tissue culture dish with 12.5 ml (20 nM) of the control (CTL) or the human ARID1A siRNA (Hs_ARID1A_5), and 5 ml of the Attractene reagent according to the manufacturer’s recommendations (Qiagen). Reduction efficiency was mon- Conflict of interest itored by immunoblotting 72 h after siRNA transfection. The authors declare no conflict of interest. Protein extraction and immunoblotting Stably T47D transfected cells were washed once in phosphate- buffered saline and nuclear extracts were obtained as described Acknowledgements by Vallone et al. (1997). Briefly, cells were resuspended in 10 volumes of homogenization solution and forced through a This study was supported by a grant from the Quebec Breast 26-gauge needle. The nuclei were then collected by centrifuga- Cancer Foundation to Mark Basik and Patricia N Tonin. EP is tion, resuspended and incubated for 12 h at 4 1C, centrifuged supported by the McGill Integrated Cancer Research Training for 15 min at 12 000 r.p.m. and the supernatant was retained. Program. The tumor bank was supported by the Fonds de The protease inhibitors aprotinin (10 mg/ml), leupeptin (10 mg/ Recherche en Sante´du Quebec (FRSQ) through the Re´seau de ml), phenylmethylsulfonyl fluoride (1 mM), and the phospha- Cancer—Axe cancer du sein et de l’ovaire to MB.

References

Adler HT, Chinery R, Wu DY, Kussick SJ, Payne JM, Fornace Jr AJ Biegel JA, Pollack IF. (2004). Molecular analysis of pediatric et al. (1999). Leukemic HRX fusion proteins inhibit GADD34- brain tumors. Curr Oncol Rep 6: 445–452. induced apoptosis and associate with the GADD34 and hSNF5/ Campeau PM, Foulkes WD, Tischkowitz MD. (2008). Hereditary INI1 proteins. Mol Cell Biol 19: 7050–7060. breast cancer: new genetic developments, new therapeutic avenues. Belandia B, Orford RL, Hurst HC, Parker MG. (2002). Targeting of Hum Genet 124: 31–42. SWI/SNF chromatin remodelling complexes to estrogen-responsive Carlson M, Laurent BC. (1994). The SNF/SWI family of global genes. EMBO J 21: 4094–4103. transcriptional activators. Curr Opin Cell Biol 6: 396–402. Berx G, van Roy F. (2009). Involvement of members of the Dai M, Wang P, Boyd AD, Kostov G, Athey B, Jones EG et al. (2005). cadherin superfamily in cancer. Cold Spring Harb Perspect Biol 1: Evolving gene/transcript deﬁnitions signiﬁcantly alter the inter- a003129. pretation of GeneChip data. Nucleic Acids Res 33: e175.

Oncogene Inactivation of ARID1A in breast cancer A Mamo et al 2100 Dallas PB, Cheney IW, Liao DW, Bowrin V, Byam W, Pacchione S et al. Orlando FA, Brown KD. (2009). Unraveling breast cancer hetero- (1998). p300/CREB binding protein-related protein p270 is a compo- geneity through transcriptomic and epigenomic analysis. Ann Surg nent of mammalian SWI/SNF complexes. MolCellBiol18: 3596–3603. Oncol 16: 2270–2279. Dallas PB, Pacchione S, Wilsker D, Bowrin V, Kobayashi R, Parker JS, Mullins M, Cheang MC, Leung S, Voduc D, Vickery T Moran E. (2000). The human SWI–SNF complex protein p270 is et al. (2009). Supervised risk predictor of breast cancer based on an ARID family member with non-sequence-specific DNA binding intrinsic subtypes. J Clin Oncol 27: 1160–1167. activity. Mol Cell Biol 20: 3137–3146. Patsialou A, Wilsker D, Moran E. (2005). DNA-binding properties of Dworkin AM, Huang TH, Toland AE. (2009). Epigenetic alterations ARID family proteins. Nucleic Acids Res 33: 66–80. in the breast: implications for breast cancer detection, prognosis and Presneau N, Manderson EN, Tonin PN. (2003). The quest for a tumor treatment. Semin Cancer Biol 19: 165–171. suppressor gene phenotype. Curr Mol Med 3: 605–629. Forbes SA, Tang G, Bindal N, Bamford S, Dawson E, Cole C et al. Roberts CW, Biegel JA. (2009). The role of SMARCB1/INI1 in (2010). COSMIC (the Catalogue of Somatic Mutations in Cancer): development of rhabdoid tumor. Cancer Biol Ther 8: 412–416. a resource to investigate acquired mutations in human cancer. Rodriguez-Nieto S, Sanchez-Cespedes M. (2009). BRG1 and LKB1: Nucleic Acids Res 38: D652–D657. tales of two tumor suppressor genes on chromosome 19p and lung Grand F, Kulkarni S, Chase A, Goldman JM, Gordon M, Cross NC. cancer. Carcinogenesis 30: 547–554. (1999). Frequent deletion of hSNF5/INI1, a component of the SWI/ Rozenblatt-Rosen O, Rozovskaia T, Burakov D, Sedkov Y, Tillib S, SNF complex, in chronic myeloid leukemia. Cancer Res 59: 3870–3874. Blechman J et al. (1998). The C-terminal SET domains of ALL-1 and Hassan S, Baccarelli A, Salvucci O, Basik M. (2008). Plasma stromal TRITHORAX interact with the INI1 and SNR1 proteins, components cell-derived factor-1: host derived marker predictive of distant of the SWI/SNF complex. Proc Natl Acad Sci USA 95: 4152–4157. metastasis in breast cancer. Clin Cancer Res 14: 446–454. Sjoblom T, Jones S, Wood LD, Parsons DW, Lin J, Barber TD et al. Hassan S, Ferrario C, Saragovi U, Quenneville L, Gaboury L, Baccarelli (2006). The consensus coding sequences of human breast and A et al. (2009). The influence of tumor-host interactions in the stromal colorectal cancers. Science 314: 268–274. cell-derived factor-1/CXCR4 ligand/receptor axis in determining Staden R, Beal KF, Bonfield JK. (2000). The Staden package, 1998. metastatic risk in breast cancer. Am J Pathol 175: 66–73. Methods Mol Biol 132: 115–130. Hirschhorn JN, Bortvin AL, Ricupero-Hovasse SL, Winston F. (1995). A Strumane K, Berx G, Van Roy F. (2004). Cadherins in cancer. Handb new class of histone H2A mutations in Saccharomyces cerevisiae causes Exp Pharmacol 165: 69–103. specific transcriptional defects in vivo. MolCellBiol15: 1999–2009. Sudarsanam P, Cao Y, Wu L, Laurent BC, Winston F. (1999). Hosein AN, Wu M, Arcand SL, Lavallee S, Hebert J, Tonin PN et al. The nucleosome remodeling complex, Snf/Swi, is required for the (2010). Breast carcinoma-associated fibroblasts rarely contain p53 maintenance of transcription in vivo and is partially redundant with mutations or chromosomal aberrations. Cancer Res 70: 5770–5777. the histone acetyltransferase, Gcn5. EMBO J 18: 3101–3106. Huang J, Zhao YL, Li Y, Fletcher JA, Xiao S. (2007). Genomic and Sudarsanam P, Iyer VR, Brown PO, Winston F. (2000). Whole- functional evidence for an ARID1A tumor suppressor role. Genes genome expression analysis of snf/swi mutants of Saccharomyces Chromosomes Cancer 46: 745–750. cerevisiae. Proc Natl Acad Sci USA 97: 3364–3369. Huusko P, Ponciano-Jackson D, Wolf M, Kiefer JA, Azorsa DO, Takeuchi T, Chen BK, Qiu Y, Sonobe H, Ohtsuki Y. (1997). Tuzmen S et al. (2004). Nonsense-mediated decay microarray Molecular cloning and expression of a novel human cDNA analysis identifies mutations of EPHB2 in human prostate cancer. containing CAG repeats. Gene 204: 71–77. Nat Genet 36: 979–983. Takeuchi T, Furihata M, Heng HH, Sonobe H, Ohtsuki Y. (1998). Jones S, Wang TL, Shih Ie M, Mao TL, Nakayama K, Roden R et al. Chromosomal mapping and expression of the human B120 gene. (2010). Frequent mutations of chromatin remodeling gene ARID1A Gene 213: 189–193. in ovarian clear cell carcinoma. Science 330: 228–231. Vallone D, Battista S, Pierantoni GM, Fedele M, Casalino L, Santoro Kennison JA. (1995). The Polycomb and trithorax group proteins of M et al. (1997). Neoplastic transformation of rat thyroid cells Drosophila: trans-regulators of homeotic gene function. Annu Rev requires the junB and fra-1 gene induction which is dependent on Genet 29: 289–303. the HMGI-C gene product. EMBO J 16: 5310–5321. Li XS, Trojer P, Matsumura T, Treisman JE, Tanese N. (2010). Van Rechem C, Boulay G, Leprince D. (2009). HIC1 interacts with a Mammalian SWI/SNF—a subunit BAF250/ARID1 is an E3 specific subunit of SWI/SNF complexes, ARID1A/BAF250A. ubiquitin ligase that targets histone H2B. Mol Cell Biol 30: 1673–1688. Biochem Biophys Res Commun 385: 586–590. Medina PP, Sanchez-Cespedes M. (2008). Involvement of the Vogelstein B, Kinzler KW. (2004). Cancer genes and the pathways they chromatin-remodeling factor BRG1/SMARCA4 in human cancer. control. Nat Med 10: 789–799. Epigenetics 3: 64–68. Wang X, Nagl NG, Wilsker D, Van Scoy M, Pacchione S, Yaciuk P Miller LD, Smeds J, George J, Vega VB, Vergara L, Ploner A et al. et al. (2004). Two related ARID family proteins are alternative (2005). An expression signature for p53 status in human breast subunits of human SWI/SNF complexes. Biochem J 383: 319–325. cancer predicts mutation status, transcriptional effects, and patient Wiegand KC, Shah SP, Al-Agha OM, Zhao Y, Tse K, Zeng T et al. survival. Proc Natl Acad Sci USA 102: 13550–13555. (2010). ARID1A mutations in endometriosis-associated ovarian Nagl Jr NG, Patsialou A, Haines DS, Dallas PB, Beck Jr GR, Moran carcinomas. N Engl J Med 363: 1532–1543. E. (2005). The p270 (ARID1A/SMARCF1) subunit of mammalian Wilsker D, Patsialou A, Zumbrun SD, Kim S, Chen Y, Dallas PB et al. SWI/SNF-related complexes is essential for normal cell cycle arrest. (2004). The DNA-binding properties of the ARID-containing Cancer Res 65: 9236–9244. subunits of yeast and mammalian SWI/SNF complexes. Nucleic Nagl Jr NG, Wang X, Patsialou A, Van Scoy M, Moran E. (2007). Acids Res 32: 1345–1353. Distinct mammalian SWI/SNF chromatin remodeling complexes Winston F, Carlson M. (1992). Yeast SNF/SWI transcriptional activators with opposing roles in cell-cycle control. EMBO J 26: 752–763. and the SPT/SIN chromatin connection. Trends Genet 8: 387–391. Nagl Jr NG, Zweitzig DR, Thimmapaya B, Beck Jr GR, Moran E. Yuge M, Nagai H, Uchida T, Murate T, Hayashi Y, Hotta T et al. (2006). The c-myc gene is a direct target of mammalian SWI/SNF- (2000). HSNF5/INI1 gene mutations in lymphoid malignancy. related complexes during differentiation-associated cell cycle arrest. Cancer Genet Cytogenet 122: 37–42. Cancer Res 66: 1289–1293. Zhang B, Chambers KJ, Faller DV, Wang S. (2007). Reprogramming Noensie EN, Dietz HC. (2001). A strategy for disease gene of the SWI/SNF complex for co-activation or co-repression in identification through nonsense-mediated mRNA decay inhibition. prohibitin-mediated estrogen receptor regulation. Oncogene 26: Nat Biotechnol 19: 434–439. 7153–7157.

Supplementary Information accompanies the paper on the Oncogene website (http://www.nature.com/onc)

Oncogene