Oncogene (2012) 31, 3939–3948 & 2012 Macmillan Publishers Limited All rights reserved 0950-9232/12 www.nature.com/onc ORIGINAL ARTICLE Identification of novel CHD1-associated collaborative alterations of genomic structure and functional assessment of CHD1 in prostate cancer

W Liu1,2, J Lindberg3, G Sui4, J Luo5, L Egevad6,TLi1,2, C Xie1,2, M Wan4, S-T Kim1,2, Z Wang1,2, AR Turner1,2, Z Zhang1,2, J Feng1,2, Y Yan7, J Sun1,2, GS Bova8, CM Ewing5, G Yan5, M Gielzak5, SD Cramer4, RL Vessella9, SL Zheng1,2, H Gro¨nberg3, WB Isaacs5 and J Xu1,2,7

1Center for Cancer Genomics, Wake Forest University School of Medicine, Winston-Salem, NC, USA; 2Center for Human Genomics and Personalized Medicine Research, Wake Forest University School of Medicine, Winston-Salem, NC, USA; 3Department of Medical Epidemiology and Biostatistics, Karolinska Institute, Stockholm, Sweden; 4Department of Cancer Biology, Wake Forest University School of Medicine, Winston-Salem, NC, USA; 5Brady Urological Institute, Johns Hopkins Medical Institutions, Baltimore, MD, USA; 6Department of Pathology and Cytology, Karolinska University Hospital, Stockholm, Sweden; 7Center for Genetic Epidemiology and Prevention, Van Andel Research Institute, Grand Rapids, MI, USA; 8Department of Pathology, Johns Hopkins Medical Institutions, Baltimore, MD, USA and 9Department of Urology, University of Washington, Seattle, WA, USA

A clearer definition of the molecular determinants that drive Introduction the development and progression of prostate cancer (PCa) is urgently needed. Efforts to map recurrent somatic deletions Although many prostate cancer (PCa) tumors are in the tumor genome, especially homozygous deletions indolent and pose little health hazard, an important (HODs), have provided important positional information in subset are aggressive and progress to disseminated the search for cancer-causing . Analyzing HODs in the disease, resulting in B34 000 deaths in the United States tumors of 244 patients from two independent cohorts and 22 every year (Siegel et al., 2011). Recent studies have PCa xenografts using high-resolution single-nucleotide highlighted the collaborative nature of multiple genomic polymorphism arrays, herein we report the identification of alterations underlying the critical process of PCa CHD1, a chromatin remodeler, as one of the most frequently progression (Carver et al., 2009; King et al., 2009; Ding homozygously deleted genes in PCa, second only to PTEN et al., 2011). in this regard. The HODs observed in CHD1, including Indeed, recent deep sequencing of the exome and deletions affecting only internal exons of CHD1, were found the whole genome of tumor cells has revealed the to completely extinguish the expression of mRNA of this extraordinary complexity of genomic alterations that in PCa xenografts. Loss of this chromatin remodeler in characterize human cancers (Berger et al., 2011; clinical specimens is significantly associated with an Robbins et al., 2011). This complexity consists of increased number of additional chromosomal deletions, both various combinations of base substitution, transloca- hemi- and homozygous, especially on 2q, 5q and 6q. tions, gene fusion and copy-number alterations (CNAs). Together with the deletions observed in HEK293 cells stably It is becoming increasingly clear that CNAs are a major transfected with CHD1 small hairpin RNA, these data component of the landscape of the PCa tumor genome suggest a causal relationship. Downregulation of Chd1 in (Taylor et al., 2010; Robbins et al., 2011; Kan et al., mouse prostate epithelial cells caused dramatic morphologi- 2010). To assess the significance of these alterations in cal changes indicative of increased invasiveness, but did not the development of PCa, it is necessary to distinguish result in transformation. Indicating a new role of CHD1, alterations driving proliferation of cancer cells versus these findings collectively suggest that distinct CHD1- random changes or passengers. Beroukhim et al. (2007) associated alterations of genomic structure evolve during have developed an important tool (Genomic Identifica- and are required for the development of PCa. tion of Significant Targets in Cancer (GISTIC)) that has Oncogene (2012) 31, 3939–3948; doi:10.1038/onc.2011.554; been used in the identification of a number of cancer published online 5 December 2011 genes (Beroukhim et al., 2010; Taylor et al., 2010). After candidate regions of CNAs are identified, a Keywords: CHD1; homozygous deletion; prostate cancer major challenge is how to then further identify candidate cancer genes, as the size of deletions and amplifications are usually very large, covering many genes, especially in high-grade and metastatic PCa. Correspondence: Dr WB Isaacs, Brady Urological Institute, Johns Hopkins Medical Institutions, Marburg 115, 600 North Wolfe Street, Using GISTIC with stringent criteria can help to narrow Baltimore, MD 21287, USA. the search to a target region with fewer genes. Even so, a E-mail: [email protected] or Dr J Xu, Center for Cancer Genomics, majority of such regions in the tumor genome are Wake Forest University School of Medicine, Medical Center Boulevard, hemizygous, with their in vivo biological effect on the Winston-Salem, NC 27157, USA. growth advantage of cancer cells being difficult to infer E-mail: [email protected] Received 5 July 2011; revised 14 September 2011; accepted 9 October because a second, unaltered allele is still present, 2011; published online 5 December 2011 notwithstanding established haploinsufficient or CHD1 and genomic alterations in prostate cancer WLiuet al 3940 dominant cancer genes. The unequivocal loss-of-func- Table 1 Screening for HOD in significant CNAs regions of deletion tion associated with homozygous gene deletions (HODs) identified by GISTICa can simplify this process. As a result, identifying and Representative JHH Sweden Combined characterizing HODs have led to the discovery of gene and CNA-region multiple recessive human cancer genes with PTEN (cytoband) being an important example in PCa. Number % Number % Number % Herein we report the assessment of HOD in the tumor PTEN(10q23.31) 18 12.77 16 15.53 34 13.93 genomes of 244 patients with primary PCa from two CHD1(5q21.1) 10 7.09 11 10.68 21 8.61 independent cohorts and 22 xenografts via genome-wide BNIP3L(8p21.2) 3 2.13 2 1.94 5 2.05 analysis of CNAs. While uncovering a number of new LRP1B(2q22.1) 2 1.42 3 2.91 5 2.05 RB1(13q14.2) 3 2.13 1 0.97 4 1.64 recurrent HODs using allele-specific analysis, we dis- USP10(16q24.1) 1 0.71 2 1.94 3 1.23 cover that CHD1 is the second, only to PTEN, most TMPRSS2- 3 2.13 0 0.00 3 1.23 frequent homozygously deleted gene in PCa. We find ERG(21q22) that loss of both the alleles of CHD1 is significantly HTR3A(11q23.2) 2 1.42 NA NA 2 0.82 associated with additional, potentially targeted HOD in RYPB(3p13) 2 1.42 0 0.00 2 0.82 MAP3K7(6q15) 1 0.71 1 0.97 2 0.82 the tumor genome. Accordingly, chromosomal deletions TP53(17p13.1) 0 0.00 1 0.97 1 0.41 associated with experimental knockdown of CHD1 CDKN1B(12p13.1) 1 0.71 0 0.00 1 0.41 provide evidence that CHD1 may have a causal role in SERPINB5(18q21.33- 0 0.00 0 0.00 0 0.00 prevention of deletion events. These data revealed a 22.1) PDE4D(5q11.2) 0 0.00 0 0.00 0 0.00 novel CHD1-associated coordinative network of altera- DISC1(1q42.2) 0 0.00 0 0.00 0 0.00 tions, and suggest a new role for this chromatin remodeler in the development of PCa. Abbreviations: CNAs, copy-number alterations; GISTIC, Genomic Identification of Significant Targets in Cancer; HOD, homozygous deletions; JHH, the Johns Hopkins Hospital. Results NA, not tested due to in significant in GISTIC analysis. aGISTIC was used to distinguish the regions of CNAs that drive cancer growth from numerous random CNAs that accumulate during cancer CHD1 is the second most frequent homozygously deleted development. gene in PCa Recurrent somatic deletions in the tumor genome, especially HODs, have been informative targets in the (Supplementary Figures 2–5) with informative samples search for tumor-suppressor genes. To uncover the full harboring deletions in these regions and confirmed our spectrum of HODs in primary PCa tumors, we findings from the GISTIC analysis as described above. performed a comprehensive analysis of DNA CNAs in Importantly, genome-wide allele-specific analysis re- the tumor genomes of surgical specimens of cancer vealed multiple occurrences of CHD1 HOD in both of tissues from 244 PCa patients using Affymetrix high- the cohorts from JHH and Sweden, with frequencies of resolution single-nucleotide polymorphism arrays 7.1% and 10.7%, respectively (Table 1 and Supplemen- (Affymetrix, Santa Clara, CA, USA). To identify CNAs tary Figures 6 and 7). In comparison, HOD frequency at that likely drive cancer growth, we first used GISTIC PTEN was observed in B13% and 16%, respectively, of with a q ¼ 0.01 (false discovery rate (FDR)) and a join- the sample in these two cohorts. Thus, at this resolution, segment-size of 80 probes to identify the significant CHD1 is second only to PTEN as the most frequent regions of deletions. The data revealed 20 significant homozygously deleted gene in the tumor genome of CNA regions, including 15 and 13 deletions (Supple- PCa. mentary Figure 1) in the Johns Hopkins Hospital (JHH) In primary PCa tumors, the size of HODs affecting and Swedish cohorts, respectively, as well as 5 and 7 CHD1 ranged from B138 kb to 2898 kb, in some amplifications. Although distinct CNAs unique to a instances covering RGMG, FAM174A and ST8SIA4, specific cohort were observed, the overall patterns were in addition to CHD1 (Supplementary Table 1). Three of remarkably similar. For example, all the 13 regions of the HODs eliminated the 50 region of CHD1 (Figure 1a), deletion identified in the Swedish cohort overlapped while the majority removed the whole gene (Figure 1b). with the regions of deletion identified in the JHH cohort Three of the HODs removed the 30-coding region of with a positive agreement score of B93% that is not CHD1. We further analyzed CNAs among an additional statistically different from 100% agreement 22 prostate tumor xenografts and identified four (P ¼ 0.0781), although seven of the peak regions were additional HODs at 5q21.1 that affected CHD1 different. Two distinct regions of deletion with peaks at (Supplementary Figure 7). One of the HODs removed 1q42.2 and 11q23.2 were observed in the JHH cohort at least 10 internal exons of CHD1 (Figure 1c). We but did not reach to the level of significance at q ¼ 0.01 confirmed the complete loss of CHD1 mRNA expres- in the Swedish cohort. Among regions having the same sion in all the four xenograft samples with CHD1 HOD peak between these two cohorts, four of them including (Figures 1d and e), but only a moderate loss of CHD1 2q22.1, 3q13, 5q21.1 and 10q23.31 contain only one mRNA in the samples with hemizygous deletion significant (q ¼ 0.01) gene each: LRP1B, RYBP1, CHD1 (Supplementary Figures 8A and B). Although one and PTEN, respectively, as determined by GISTIC. We HOD (in Lu81) caused a contiguous loss of only an also carried out a minimum-overlap-region analysis internal portion of CHD1 (Figure 1c), loss of this partial

Oncogene CHD1 and genomic alterations in prostate cancer WLiuet al 3941

Figure 1 CHD1 HODs and complete loss of the expression of CHD1 transcript in the tumors harboring CHD1 HOD. (a–c) Allele-specific analysis showing examples of HODs affecting the 50 of CHD1, multiple genes including the entire CHD1 gene, and only internal exons, respectively. Blue and red dot lines represent different alleles. (a) Mapping of a HOD that resulted in complete loss of only the 50 of the CHD1 gene from 98 265 502 bp to 98 663 965 bp (hg18) in a primary tumor. (b) Allele-specific analysis revealed a large HOD affecting multiple genes including the entire CHD1 and RGMB genes in a primary tumor. (c) Fine mapping a HOD that affects only internal exons in a xenograft tumor. (d) Relative expression of CHD1 mRNA in PCa xenografts with CHD1 HODs (Lu69, Lu81, Lu78 and Lu141) and without deletion of CHD1 (Lu49, Lu70, Lu92.1, Lu147, Lu153, Lu77, Lu145.1 and Lu146) detected by the Agilent Whole Human Microarray. (e)TheaverageofCHD1 relative expressions in xenografts with HODs and without deletion of CHD1.

Oncogene CHD1 and genomic alterations in prostate cancer WLiuet al 3942 gene segment resulted in complete loss of the expression of all exons including the intact exons, as revealed by exon expression analysis using the Affymetrix human Exon 1.0 ST array (Supplementary Figure 8C). These results suggest that HODs affecting CHD1 typically result in complete absence of CHD1 expression, as opposed to hemizygous deletions of CHD1, which resulted in decreased but not absent CHD1 mRNA expression (Supplementary Figure 8A).

Complete loss of CHD1 is associated with a larger number of HODs at other locations in the tumor genome To explore the possible genomic and functional effects of CHD1 HOD in PCa, we began by characterizing the genome-wide DNA CNA profiles in tumor genomes with CHD1 HODs, initially comparing such cases with those harboring PTEN HODs and then comparing the cases harboring CHD1 or PTEN HODs with those harboring no CNAs at these two genes. As the frequency of HODs may be associated with general genomic instability as tumor cells evolve during cancer development and PTEN is the most frequent homo- zygously deleted gene in PCa, we therefore used tumors with PTEN HODs as ‘positive controls’ and tumors without deletions at these two genes as ‘negative controls’. In these comparisons, we observed a substan- tially higher frequency of HODs in other genomic regions among samples with CHD1 HOD. Cases harboring CHD1 HODs contained an average of 4.5 additional HODs per genome compared with an average of 0.89 per genome in cases harboring PTEN HODs Figure 2 Complete loss of CHD1 is associated with increased (P ¼ 0.0003, Wilcoxon two-sample test; Figure 2a) in the number of additional HODs in the tumor genome of PCa. Neither JHH cohort. We validated our findings using the (green bars; N ¼ 30, 43 and 73 in the JHH, Swedish and combined Swedish cohort, where we observed B4.64 additional cohorts, respectively) in the legend entry represents tumors HODs per genome among tumors harboring the CHD1 harboring no CNAs at CHD1 and/or PTEN revealed by allele- specific analysis. Blue bars represent CHD1 HOD with N ¼ 10, 11 HOD, in contrast to 0.53 additional HODs per genome and 21 in the JHH, Swedish and combined cohorts, respectively. among tumors harboring the PTEN HOD (P ¼ 0.0001). Red bars represent PTEN HOD with N ¼ 18, 15 and 33 in the Combining these two cohorts we identified a total of 21 JHH, Swedish and combined cohorts, respectively. (a) Compar- patients harboring CHD1 HODs with an average of 4.57 isons of the average numbers of additional HODs among tumors harboring CHD1 HOD, PTEN HOD or no CNAs at CHD1 and/or additional HODs per genome in the patients harboring PTEN in two PCa cohorts from JHH and Sweden. P-values were CHD1 HOD, which is significantly higher than 0.71 calculated using the Wilcoxon two-sample test. (b) Comparisons of additional HODs per genome in the patients harboring the average numbers of cytobands affected by these additional PTEN HODs (P ¼ 1.15 Â 10À7). In contrast to tumors HODs among tumors harboring CHD1 HOD, PTEN HOD or no CNAs at CHD1 and/or PTEN in two PCa cohorts from JHH harboring no CNAs at CHD1, based on allele-specific and Sweden. P-values were calculated using the Wilcoxon analysis, tumors harboring CHD1 HOD contained two-sample test. significantly more additional HODs in both the JHH and the Swedish cohorts, with P ¼ 4.67 Â 10À7 and P ¼ 9.95 Â 10À8, respectively (Figure 2a). by these additional HODs in the genomes with complete Analyzing these HODs, we noticed multiple small- loss of CHD1 than by those in the genomes containing sized HODs that clustered in the vicinity of particular either complete loss of PTEN or no CNAs at CHD1 regions on the same (Figures 1b and c). and/or PTEN (Figure 2b). These clustered HODs may be derived from a single event that led to the initial genomic deletion/alteration Loss of CHD1 is associated with deletions on 2q22, (Stephens et al., 2011). To minimize any possible effects 5q11.2. and 6q15 in the tumor genome of these multiple, clustered HODs on calculations of the We next analyzed the distribution of these additional number of HOD events, we compared the number of HODs in tumors with complete loss of CHD1. As shown cytological chromosomal bands (cytobands) affected by in Figure 3, in a comparison among these tumors from additional HODs in the tumors harboring CHD1 HODs both the JHH and the Swedish cohorts, most of these versus tumors with complete loss of PTEN and with no additional HODs were not randomly distributed, but CNAs at CHD1 and/or PTEN. The data revealed that a rather appeared preferentially located on significantly larger number of cytobands were affected 2, 5 and 6, being absent from chromosomes 9, 11 and 14

Oncogene CHD1 and genomic alterations in prostate cancer WLiuet al 3943

Figure 4 Significant correlations between CHD1 deletion and CNAs at other genes (loci). y-Axis, correlation coefficient;. x-axis, genes representing significant CNAs loci identified by GISTIC. TMPRSS2–ERG represents the region where CNA affected both 30 of TMPRSS2 and 50 of ERG to potentially create the fusion gene. JHH, the JHH cohort. SWD, the Swedish cohort. Fisher’s exact test was used to assess the associations between loss of CHD1 and other CNAs. Only significant associations identified in both of these cohorts are presented. All other correlations and significant tests are presented in Supplementary Table 3.

representative genes identified by GISTIC (Table 1). As shown in Figure 4, deletions of CHD1 was positively associated with deletions of LRP1B at 2q22.1, PDE4D at 5q11.2, MAP3K7 at 6q15 and gain of COL1A2 at 7q21.3 in both of the cohorts (P ¼ 0.0079, Supplemen- tary Table 3). We next compared the GISTIC signatures with or without the tumors harboring CHD1 deletions. Figure 3 The chromosomal cytobands affected by additional As shown in the Supplementary Figure 11, removing the HODs in tumors with complete loss of CHD1 are significantly tumors harboring CHD1 deletions (right panels in A, B, located on chromosomes 2, 5 and 6. (a) Distribution of the cytobands affected by these additional HODs in the tumors with C and D) substantially reduced the significance levels of complete loss of CHD1 in the JHH cohort. (b) Distribution of the signature peaks associated with deletions of LRP1B at cytobands affected by these additional HODs in the tumors with 2q22.1, PDE4D at 5q11.2, MAP3K7 at 6q15 and gain of complete loss of CHD1 in the Swedish cohort. y-Axis, the number COL1A2 at 7q21.3 (marked by light blue ovals), in of affected cytobands; x-axis, chromosome; z-axis, tumors harbor- comparison to the significance levels of these signature ing CHD1 HOD. The binomial proportion test was used to assess the significance in the distribution of cytobands affected by these peaks derived from all tumors including the ones additional HODs in the tumors harbored complete loss of CHD1. harbored CHD1 deletions (left panels in A, B, C and Specific P-values obtained from the binomial proportion test mark D). These significant concurrences of CNAs suggest a chromosomes that were significantly affected by these additional novel collaborative CNA-network among these loci in HODs. the evolution of the PCa tumor genome. In addition, we observed a significantly negative correlation between CNAs at CHD1 and TMPRSS2– through 22 among the tumor genomes with complete ERG on 21q22 (Figure 4) in both the JHH and the loss of CHD1. One exception was chromosome 4 that Swedish cohorts (P ¼ 0.0057 for the direction of harbored HODs in the Swedish cohort (P ¼ 9.91 Â 10À10, association). Indeed, none of the 21 patients with Figure 3b), but not in the JHH cohort (Figure 3a). Some CHD1 HOD harbored the deletion from 30 of recurrent additional HODs targeted the same region or TMPRSS2 to 50 of ERG, which is statistically significant affected the same genes (Supplementary Figures 2, 6, 9, (P ¼ 7.34 Â 10À4, Fisher’s exact test). These findings and Supplementary Table 2, respectively), while other indicate that mutually exclusive selection of these two additional HODs occurred only once in tumors harbor- CNAs might occur during proliferation of PCa cells, ing CHD1 HOD (Supplementary Figures 2, 3, 6, 7, 9 resulting in two distinct subgroups of PCa. and 10). Furthermore, we explored the association between To further evaluate the effects of loss of CHD1 on the loss of CHD1 and PCa aggressiveness. As the total changes in CNA signature among the tumor genomes in number of CHD1 HOD is too small to perform a the JHH and Swedish cohorts, we first tested the meaningful statistical analysis, we combined tumors association between any deletion (hemizygous and with either hemizygous or HODs to explore the homozygous) at CHD1 and the other CNAs with their relationship between the loss of CHD1 and the Gleason

Oncogene CHD1 and genomic alterations in prostate cancer WLiuet al 3944

Figure 5 Downregulation of Chd1 in MPECs alters morphological characteristics in the growth of the cells. (a) Clonogenic assay shows more colonies formed from the cells with reduced level of Chd1 (U6/CHD1). (b) Quantitative data of clonogenic assay demonstrate that cells with knockdown expression of Chd1 displayed a significantly larger number of pixels from formed colonies. Western blotting shows a reduction of Chd1 protein in MPECs transfected by CHD1 shRNA (U6/CHD1) in comparison to those transfected by control shRNA (U6/control). (c) Cell spheroids formed by MPECs transfected with control shRNA in a 3D culture system. (d) Multiple branching structures (marked by red arrows) were generated by MPECs transfected with CHD1 shRNA in a 3D culture system. (e) Comparison of the length of branches formed by the cells transfected with U6/control or U6/CHD1 shRNA. Four representative views from each of the treatments were taken and five of the longest branches within each of the views were measured. The average and standard deviation are presented. Significantly longer branches formed by the cells transfected with U6/CHD1 shRNA in this 3D culture indicate an invasive property of cells with downregulated Chd1. All of the experiments were carried out in triplicates.

score. We found that loss of CHD1 was significantly without transfection or stably transfected with control associated with tumors having a higher Gleason score shRNA. We also observed one HOD each on chromo- 47 in the JHH cohort (P ¼ 0.032) but not in the somes 3 and X, as well as hemizygous deletions on Swedish cohort. These findings warrant further investi- chromosomes 16 and 18 (Supplementary Figure 12) in gation using additional larger cohorts. the cell lines stably transfected with CHD1 shRNA but not in these two types of control cells. CNAs and morphologic changes after knockdown To assess possible phenotypic alterations associated expression of CHD1 with CHD1 downregulation, we chose to use our mouse The data presented above suggest that loss of CHD1 prostate epithelial cells (MPECs) model because the cells may predispose cells to genomic deletions, potentially at maintain progenitor cell characteristics over long-term specific loci. To begin to address this question, we chose culture and reflect the true nature of prostate epithelial a well-established human cell transfection model, stem cells in vivo (Barclay et al., 2008). We infected the HEK293, for manipulating CHD1 expression and MPECs using lentiviruses expressing CHD1 and control analyzed CNAs in cells with stable knockdown of shRNAs. As shown in Figure 5b, CHD1 shRNA Chd1 protein expression. We identified a deletion that effectively reduced expression of Chd1 protein. In affected LRP1B in a cell line stably transfected with clonogenic analysis, we observed that MPECs expres- CHD1 small hairpin RNA (shRNA; Supplementary sing CHD1 shRNA formed more colonies than the cells Figure 12), which is consistent with our findings in the expressing control shRNA (Figure 5a). Quantitative primary tumors, and suggests a causal relationship. The analysis of the clonogenic data revealed that cells deletion at LRP1B was not observed in the cells either with knockdown expression of CHD1 displayed a

Oncogene CHD1 and genomic alterations in prostate cancer WLiuet al 3945 consistently higher (P ¼ 0.0489) survival and prolifera- genes in the development of PCa are subject to further tion rate (Figure 5b). This result suggests that silencing investigation. However, significant correlation between endogenous CHD1 expression might enhance cell the loss of CHD1 and deletions of LRP1B and PDE4D clonogenicity or survivability. We further analyzed the rather than genes located at other fragile sites, suggests morphological characteristics of MPECs growth, using their concurrent losses provide advantages for selection silenced expression of endogenous CHD1 in a three- and proliferation of cancer cells. Inactivation of LRP1B dimensional (3D) culture system. As expected, cell by mutation, deletion and methylation has been spheroids were formed by MPECs transfected with previously reported in the lung cancer (Ding et al., control shRNA (Barclay and Cramer, 2005; Figure 5c). 2008; Kohno et al., 2010), glioblastoma multiforme (Yin Surprisingly, the MPECs transfected with CHD1 et al., 2009), oral cancer (Cengiz et al., 2007) and gastric shRNA formed branch structures growing into the cancer (Lu et al., 2010). PDE4D, encoding cyclic collagen matrix, with significantly (Po0.0001) lengthen- adenosine monophosphate-specific phosphodiesterase ing of branches and side branches (Figures 5d and e) 4D, has been previously identified as a potential that were not observed in cells transfected with control tumor-suppressor gene in esophageal adenocarcinoma shRNA. These characteristics suggest that silenced (Nancarrow et al., 2008). On the other hand, Rahrmann CHD1 expression could significantly enhance cell et al. (2009) identified PDE4D as a proliferation- invasiveness and/or developmental changes. However, promoting factor in PCa. renal-grafting these cells in mice did not result in CHD1 encodes a DNA- tumorigenesis in our in vivo study (data not shown). binding protein known to have important roles in These findings indicate that alterations of other genes, regulating gene expression in mammalian cells (Sims such as those were associated with CHD1 deletion in the et al., 2007), gametogenesis in Drosophila melanogaster primary tumors described above, may be needed to (McDaniel et al., 2008), and pluripotency of stem cells in initiate the development of PCa. mice (Gaspar-Maia et al., 2009). Multiple chromatin- remodeling genes have recently been implicated through mutational analyses of various cancer types, such as Discussion renal cell (Dalgliesh et al., 2010), ovarian (Jones et al., 2010), lung (Medina et al., 2008) and others (Weissman Genetic alterations observed in a specific tumor genome and Knudsen, 2009), emphasizing the importance of this at a particular stage are cumulative products of epigenetic function in human carcinogenesis. successive clonal expansion and mutational evolution To explore the function of CHD 1 in the tumor (Liu et al., 2009; Anderson et al., 2011). However, HODs genome of PCa, we compared the CNAs in tumors that do not affect cell survival could be maintained as harboring HODs of CHD1 to either those harboring random passenger mutations. When a HOD affects HODs of PTEN or those with no CNAs at CHD1 and/ genes whose absence provides growth advantages, this or PTEN. To our surprise, the tumors harboring CHD1 HOD would be selected during cell proliferation and HOD had a significantly larger number of additional could be observed in a significant larger number of the HODs in other locations of the genome. Most of tumor genomes in a large population of PCa patients. these additional HODs are significantly located on Analysis of HODs may reveal not only recessive cancer chromosomal regions 2q, 5q and 6q where hemizygous genes but can also uncover regions of genetic fragility deletions occurred and were associated with the loss of (Bignell et al., 2010) sites predisposed to undergo DNA CHD1. rearrangements in the tumor genome. Using an unbiased The consistent associations of CNAs found in clinical genome-wide analysis of tumor DNA, we uncovered samples from two independent cohorts might be also CHD1, the chromatin remodeler, as one of the two most due to selection for growth and proliferation advantage frequent homozygously deleted genes in the tumor of the tumor cells during cancer development in the genome of PCa. Importantly, inactivating mutations in prostate, in addition to a possibly causal relationship. the chromatin remodelers, CHD1 and CHD5, have been The cells transfected with shRNA against CHD1 also reported recently in PCa (Berger et al., 2011; obviously did not evolve as the tumor cells did in the Robbins et al., 2011). clinical specimens. Multiple deletions, including those Among the three genes concurrently deleted with observed in clinical specimens, in the cells with knock- CHD1, LRP1B and PDE4D are apparently located at down expression of CHD1 provide evidence supporting fragile sites on chromosomes of 2 and 5, respectively the protective role of CHD1 and provide a basis for (Bignell et al., 2010). However, MAP3K7 seems not to CHD1-associated CNAs if these cells could evolve and be associated with inherent fragility and rather may proliferate in the same way as those in clinical samples. function as a tumor-suppressor gene. Encoding TGF-b The fact that experimental knockdown expression of activated kinase-1, MAP3K7, has been reported to be CHD1 itself did not result in transformation and frequently deleted in PCa, the occurrence of which is deletions on chromosomes 5 and 6 suggests that highly associated with high-grade disease (Liu et al., deletions on chromosomes 5q and 6q might be caused 2007). Recent work has demonstrated a tumor-suppres- by (1) further genomic evolution of the cells with loss sive role of MAP3K7 in liver cancer (Bettermann et al., of CHD1 induced by tumor microenvironment, (2) a 2010). As LRP1B and PDE4D are located within fragile different mechanism other than loss of CHD1. sites of the genome, their functions as tumor-suppressor From this point of view, the data from experimental

Oncogene CHD1 and genomic alterations in prostate cancer WLiuet al 3946 knockdown expression of CHD1 is consistent with the tumorigenesis in our in vivo study indicates that CHD1 findings in clinical samples from the two cohorts. itself may not be sufficient for tumor suppression More suitable in vitro and in vivo models with human although CHD1 HODs were frequently observed in prostate epithelial cells and tissue recombination could PCa. It has been reported that transgenic mice expres- shed more light on mechanism of CHD1-associated sing the TMPRSS2–ERG fusion gene also failed to CNAs. develop prostatic intraepithelial neoplasia or prostate Together with the results from HEK293 cells stably tumor (King et al., 2009). However, when these ERG transfected with CHD1 shRNA, these data indicate that overexpressing transgenic mice were crossed with CHD1 may either have a role in protecting the genome PTEN-deficient mice, PCa was observed in their off- from loss of DNA, or it may, in conjunction with the spring (Carver et al., 2009; King et al., 2009). These other associated CNAs, provide selection and/or growth studies demonstrated that alterations of multiple genes advantages for tumor cells during the development of are required for a particular phenotype in the develop- PCa. It is plausible that CHD1, a chromatin remodeler, ment of PCa. The genes concurrently altered with may indirectly protect DNA because it facilitates deletion of CHD1 that were identified in this study deposition of H3 into chromatin (Konev et al., may represent a collaborative network that is acquired 2007; Sims and Reinberg, 2009). In addition, direct in cancer cells and cumulatively drives the development binding of CHD1 to DNA (Stokes and Perry, 1995) may of PCa. This may explain why knockdown CHD1 in provide an alternative mechanism of protection, how- isolation failed to result in tumorigenesis experimentally. ever, it is unknown whether CHD1 is directly involved in Future studies should reveal more details of mechanisms DNA repair. Post-translational modifications to his- by which these genes (loci) interact in promoting the tones have been reported to influence DNA repair development of PCa. (Avvakumov et al., 2011). On the other hand, deletion at The observation that loss of CHD1 is negatively LRP1B in cells with knockdown expression of CHD1 correlated with the deletion at 21q22 that creates the but not in the controls could be due to chance, and its fusion of TMPRSS2–ERG indicates tumors harboring association with loss of CHD1 in clinical cohorts might these alterations may represent different subtypes of be caused by selective advantage for cell proliferation. PCa. Berger et al. (2011) reported that the locations of Extensive studies using a more suitable model with DNA rearrangement breakpoints in tumors harboring human prostate epithelial cells are needed to fully TMPRSS2–ERG fusion inversely correlated with break- evaluate the effects of Chd1 knockdown on genomic points in tumor lacking the ETS fusion. Their findings alterations and tumorigenesis. suggest a link between and To further evaluate the effect of CHD1 loss in the genesis of genomic alterations. Our data provide further development of PCa, we knocked down Chd1 expres- support for the role of a particular chromatin remodeler, sion in MPECs. These mouse prostate progenitor/stem CHD1, in the genesis of CNAs. cells maintain progenitor characteristics in long-term In summary, we have identified CHD1, as second only culture in vitro and are capable of fully differentiating to PTEN, the most frequently homozygously deleted into prostatic structures in vivo (Barclay et al., 2008). gene in PCa. We demonstrate that (1) complete loss of CHD1 is known for its role in development and in CHD1 was associated with increased number of addi- maintaining the pluripotency of embryonic stem cells. tional HODs on 2q, 5q and 6q, and (2) loss of CHD1 We speculate that loss of Chd1 in these stem cells may was positively correlated with CNAs at LRP1B, disrupt normal development of prostatic structures and PDE4D, MAP3K7 and COL1A2, but negatively asso- lead to prostatic abnormality or eventually tumorigen- ciated with deletion at 21q22 that creates TMPRSS2– esis. It is therefore better to use a developmental model ERG fusion. Assessing the function of CHD1, we found with prostate progenitor/stem cells as the basis upon that downregulation of Chd1 protein resulted in which to test our hypothesis. As prostate and tumor morphological changes in the growth of MPECs, but stem cells possess similar capabilities such as self- did not produce in vivo tumorigenesis. Although these renewal and androgen independence, it has been findings suggest that collaborative alterations were speculated that tumor stem cells are derived from acquired in a cumulative manner in the tumor genome aberrant stem cells. Therefore, we believe that using and their interaction may drive the development this developmental model with prostate progenitor/stem of PCa, further study is warranted to illustrate how cells to evaluate the role of CHD1 can provide critical CHD1 interacts with other genes, including LRP1B, insight into the mechanisms that drive the development PDE4D and MAP3K7, to affect tumorigenesis and the of PCa. Although the MPECs expressing CHD1 shRNA development of PCa. It will also be important to generated only slightly more colonies than the cells evaluate whether loss of this chromatin remodeler expressing the control shRNA in the clonogenic assay, directly causes aberrations of genomic structure such they produced significantly longer branching structures as deletions and amplifications, or indirectly lead to that extended into the collagen matrix in a 3D culture concurrent accumulation of these aberrations, especially system. These morphological characteristics suggest at/near the specific fragile sites at 2q22.1 and 5q11.2. invasive growth, as well as developmental changes Although further study is necessary, these findings (Barclay and Cramer, 2005) of the cells with down- strongly implicate CHD1 as a critical factor in the regulation of Chd1. The fact that renal-grafting these dramatic genomic reorganization that accompanies cells with downregulated Chd1 in mice did not result in prostate carcinogenesis.

Oncogene CHD1 and genomic alterations in prostate cancer WLiuet al 3947 Materials and methods containing a puromycin-resistant gene that we previously described (Deng et al., 2009), in addition to the shRNAs against Study subjects and DNA copy-number analysis CHD1 and non-specific target from OriGene. The shRNA- Somatic tumor DNA from a total of 22 xenografts and 244 PCa containing lentiviruses were packaged following a standard patients undergoing radical prostatectomy for the treatment of protocol (Deng et al., 2007) and used to infect MPECs (Barclay clinically localized disease at two centers, one in the United et al., 2008). Two days post infection, 1.5 mgmlÀ1 of puromycin States, (JHH) and the other in Sweden (Karolinska Institute) was added to select the transfected cells for at least 3 days and from 1988 to 2006 was used in this study. They were selected then used for downstream assays. We performed immunostaining based on the availability of genomic DNA of sufficient quantity and western blotting according to a previous protocol (Sui et al., (45 mg) and purity (470% cancer cells for cancer specimens, no 2002) with modifications (Supplementary Method) using a CHD1 detectable cancer cells for normal samples). Tissue samples were antibody (Bethyl Laboratories, Inc., Montgomery, TX, USA). obtained by macro-dissection of matched non-malignant (nor- mal) and cancer-containing areas of prostate tissue as deter- Clonogenic assay mined by histological evaluation of hematoxylin and eosin- Clonogenic analysis was performed as described previously stained frozen sections of snap-frozen radical prostatectomy (Cao et al., 2010). Briefly, cells infected by the lentivirus specimens. Amongst these 244 patients, 193 had normal control expressing the control shRNA, and CHD1shRNA were DNA, whereas 51 of them had no matched normal DNA individually plated at different densities (125, 250, 500, 1000 available at the time of DNA analysis. Most of the 141 patients and 2000 per dish) in 6-cm cell culture dishes. After 7–10 days, in the JHH cohort had a more aggressive form of PCa; 31%, the cells were fixed in 10% formalin and stained by 0.1% 30%, and 46% of patients had pathologic Gleason score X8, crystal violet. Photoshop software was used to quantify the pathologic stage XT3b and pretreatment serum PSA X10 ng pixels of the stained cells. mlÀ1, respectively. Most of the 103 patients in the Swedish cohort had a less aggressive form of PCa; B51 and B11% of these 3D culture of the MPECs patients had a preoperative Gleason score p6 and a patholo- gical Gleason score X8, respectively. Experimental procedures According to a previously published method of 3D culture for assay of single-nucleotide polymorphism array, DNA copy- (Barclay and Cramer, 2005), MPECs infected by the control shRNA and CHD1 shRNA were individually trypsinized and number analysis and identification of target genes are presented 4 in the Supplementary Methods. resuspended in the collagen matrix at a density of 8 Â 10 cells per ml. A 0.5 ml aliquot of this solution was dispensed into 4 Expression microarrays each well of a 24-well plate (4 Â 10 cells per well). After the gel We isolated the total RNA from fresh frozen tissues of xenograft solidified at room temperature, 1.0 ml of culture medium was tumors using a TRIzol Reagent kit (Invitrogen, Grand Island, added to the top of the gel in each well. The cells were NY, USA). The amount and quality of total RNA were assessed maintained in the 3-D culture for 8–14 days in a cell culture using the Agilent 2100 Bioanalyzer (Agilent Technologies, incubator with media changes every other day. At the end of Wilmington, DE, USA) before expression analysis. Expression the incubation period, the branch structures for pseudoductal microarray analysis was carried out according to the instructions morphogenesis were imaged using digital photomicrography, in the Agilent 4 Â 44 Whole Genome Expression Microarray and the amount of outgrowth from spheroids over time was System (Agilent Technologies) using an input of 500 ng of total determined using Photoshop and digital photomicrography. RNA, and a 2-color design involving cohybrization of Cye-5- labeled test samples with a Cye-3-labeled common reference Statistical methods sample from benign prostate cells. The expression profile for The significance in the number of additional HODs in the tumor each sample was represented as a normalized ratio of sample/ genomes with complete loss of CHD1 in comparison to those in reference for all entities represented on the array. Statistical the tumor genomes either with PTEN HOD or without CNAs at analyses of the differentially expressed genes were performed PTEN and/or CHD1 was assessed using Wilcoxon’s two-sample using GeneSpring software (Agilent Technologies). In addition, test. We used Fisher’s exact test to test the associations between these RNA samples were also analyzed for exon expression using loss of CHD1 and other significant CNAs identified by the the Affymetrix Human 1.0 ST array, according to the GISTIC across the whole genome. We used the binomial manufacturer’s instructions. We analyzed the hybridization proportion test to assess the significance in the distribution of intensity and exon expression levels using the Partek Genomic additional HODs in the tumors that harbored complete loss of Suite 6.4 (Partek Inc., St Louis, MO, USA). CHD1. All the statistical analyses were performed using SAS software version 9.2 (SAS Institute Inc., Cary, NC, USA). RNA interference and western blotting. The CHD1 shRNAs, were purchased from OriGene (Rock- ville, MD, USA) and used for transfection of human cell lines Conflict of interest HEK293 following the manufacturer’s instructions. These shRNAs include TI355777 (50-ATG ATG GAG CTA AAG The authors declare no conflict of interest. AAA TGT TGT AAC CA), TI355778 (CAC AAG GAG CTT GAG CCA TTT CTG TTA CG), TI355779 (AGT GTC AGA TGC TCC AGT TCA TAT CAC GG), and TI355780 (AAT Acknowledgements GGA CAC AGT GAT GAA GAA AGT GTT AG-30). Additional shRNAs were designed and constructed according The study is partially supported by the National Institutes of to our previously published method (Sui et al., 2002; Sui and Health Grants CA135008 and CA133066 (to W Liu and WB Shi, 2005) for lentiviral infection of the MPECs. We directly Isaacs), CA119069 (to J Xu), CA131338 and 133009 (to SL subcloned a control shRNA with a scrambled sequence of Zheng and WB Isaacs). We thank Tamara Adams for editing ‘50-GGG CCA TGG CAC GTA CGG CAA G-30’ and a the manuscript. The support of William T Gerrard, Mario CHD1 shRNA with a target sequence of ‘50-GAA GAT GTG Duhon, Jennifer and John Chalsty and P Kevin Jaffe (to WBI) GAA TAT TAT AAT T-30’ into a lentiviral vector pLU is gratefully acknowledged.

Oncogene CHD1 and genomic alterations in prostate cancer WLiuet al 3948 References

Anderson K, Lutz C, van Delft FW, Bateman CM, Guo Y, Kohno T, Otsuka A, Girard L, Sato M, Iwakawa R, Ogiwara H et al. Colman SM et al. (2011). Genetic variegation of clonal architecture (2010). A catalog of genes homozygously deleted in human lung and propagating cells in leukaemia. Nature 469: 356–361. cancer and the candidacy of PTPRD as a tumor suppressor gene. Avvakumov N, Nourani A, Coˆte´J. (2011). Histone chaperones: Genes Chromosomes Cancer 49: 342–352. modulators of chromatin marks. Mol Cell 41: 502–514. Konev AY, Tribus M, Park SY, Podhraski V, Lim CY, Emelyanov Barclay WW, Axanova LS, Chen W, Romero L, Maund SL, Soker S AV et al. (2007). CHD1 motor protein is required for deposition of et al. (2008). Characterization of adult prostatic progenitor/stem histone variant H3.3 into chromatin in vivo. Science 317: 1087–1090. cells exhibiting self-renewal and multilineage differentiation. Stem Liu W, Chang BL, Cramer S, Koty PP, Li T, Sun J et al. (2007). Cells 26: 600–610. Deletion of a small consensus region at 6q15, including the Barclay WW, Cramer SD. (2005). Culture of mouse prostatic epithelial MAP3K7 gene, is significantly associated with high-grade prostate cells from genetically engineered mice. Prostate 63: 291–298. cancers. Clin Cancer Res 13: 5028–5033. Berger MF, Lawrence MS, Demichelis F, Drier Y, Cibulskis K, Liu W, Laitinen S, Khan S, Vihinen M, Kowalski J, Yu G et al. (2009). Sivachenko AY et al. (2011). The genomic complexity of primary Copy number analysis indicates monoclonal origin of lethal human prostate cancer. Nature 470: 214–220. metastatic prostate cancer. Nat Med 15: 559–565. Beroukhim R, Getz G, Nghiemphu L, Barretina J, Hsueh T, Linhart D Lu YJ, Wu CS, Li HP, Liu HP, Lu CY, Leu YW et al. (2010). et al. (2007). Assessing the significance of chromosomal aberrations Aberrant methylation impairs low density lipoprotein receptor- in cancer: methodology and application to glioma. Proc Natl Acad related protein 1B tumor suppressor function in gastric cancer. Sci USA 104: 20007–20012. Genes Chromosomes Cancer 49: 412–424. Beroukhim R, Mermel CH, Porter D, Wei G, Raychaudhuri S, McDaniel IE, Lee JM, Berger MS, Hanagami CK, Armstrong JA. Donovan J et al. (2010). The landscape of somatic copy-number (2008). Investigations of CHD1 function in transcription and alteration across human cancers. Nature 463: 899–905. development of Drosophila melanogaster. Genetics 178: 583–587. Bettermann K, Vucur M, Haybaeck J, Koppe C, Janssen J, Medina PP, Romero OA, Kohno T, Montuenga LM, Pio R, Yokota J Heymann F et al. (2010). TAK1 suppresses a NEMO-dependent et al. (2008). Frequent BRG1/SMARCA4-inactivating mutations in but NF-kappaB-independent pathway to liver cancer. Cancer Cell human lung cancer cell lines. Hum Mutat 29: 617–622. 17: 481–496. Nancarrow DJ, Handoko HY, Smithers BM, Gotley DC, Drew PA, Bignell GR, Greenman CD, Davies H, Butler AP, Edkins S, Andrews Watson DI et al. (2008). Genome-wide copy number analysis in JM et al. (2010). Signatures of mutation and selection in the cancer esophageal adenocarcinoma using high-density single-nucleotide genome. Nature 463: 893–898. polymorphism arrays. Cancer Res 68: 4163–4172. Cao P, Deng Z, Wan M, Huang W, Cramer SD, Xu J et al. (2010). Rahrmann EP, Collier LS, Knutson TP, Doyal ME, Kuslak SL, Green MicroRNA-101 negatively regulates Ezh2 and its expression is LE et al. (2009). Identification of PDE4D as a proliferation modulated by androgen receptor and HIF-1alpha/HIF-1beta. Mol promoting factor in prostate cancer using a Sleeping Beauty Cancer 9: 108. transposon-based somatic mutagenesis screen. Cancer Res 69: Carver BS, Tran J, Gopalan A, Chen Z, Shaikh S, Carracedo A et al. 4388–4397. (2009). Aberrant ERG expression cooperates with loss of PTEN to Robbins CM, Tembe WA, Baker A, Sinari S, Moses TY, Beckstrom- promote cancer progression in the prostate. Nat Genet 41: 619–624. Sternberg S et al. (2011). Copy number and targeted mutational Cengiz B, Gunduz M, Nagatsuka H, Beder L, Gunduz E, Tamamura analysis reveals novel somatic events in metastatic prostate tumors. R et al. (2007). Fine deletion mapping of chromosome 2q21-37 Genome Res 21: 47–55. shows three preferentially deleted regions in oral cancer. Oral Oncol Siegel R, Ward E, Brawley O, Jemal A. (2011). Cancer statistics, 2011: 43: 241–247. the impact of eliminating socioeconomic and racial disparities on Dalgliesh GL, Furge K, Greenman C, Chen L, Bignell G, Butler A premature cancer deaths. CA Cancer J Clin 61: 212–236. et al. (2010). Systematic sequencing of renal carcinoma reveals Sims 3rd RJ, Millhouse S, Chen CF, Lewis BA, Erdjument-Bromage inactivation of histone modifying genes. Nature 463: 360–363. H, Tempst P et al. (2007). Recognition of trimethylated histone H3 Deng Z, Wan M, Cao P, Rao A, Cramer SD, Sui G. (2009). Yin Yang lysine 4 facilitates the recruitment of transcription postinitiation 1 regulates the transcriptional activity of androgen receptor. factors and pre-mRNA splicing. Mol Cell 28: 665–676. Oncogene 28: 3746–3757. Sims 3rd RJ, Reinberg D. (2009). Stem cells: escaping fates with open Deng Z, Wan M, Sui G. (2007). PIASy-mediated sumoylation of Yin states. Nature 460: 802–803. Yang 1 depends on their interaction but not the RING finger. Mol Stephens PJ, Greenman CD, Fu B, Yang F, Bignell GR, Mudie LJ Cell Biol 27: 3780–3792. et al. (2011). Massive genomic rearrangement acquired in a single Ding L, Getz G, Wheeler DA, Mardis ER, McLellan MD, Cibulskis K catastrophic event during cancer development. Cell 144: 27–40. et al. (2008). Somatic mutations affect key pathways in lung Stokes DG, Perry RP. (1995). DNA-binding and chromatin localiza- adenocarcinoma. Nature 455: 1069–1075. tion properties of CHD1. Mol Cell Biol 15: 2745–2753. Ding Z, Wu CJ, Chu GC, Xiao Y, Ho D, Zhang J et al. (2011). Sui G, Shi Y. (2005). Gene silencing by a DNA vector-based RNAi SMAD4-dependent barrier constrains prostate cancer growth and technology. Methods Mol Biol 309: 205–218. metastatic progression. Nature 466: 869–873. Sui G, Soohoo C, Affar el B, Gay F, Shi Y, Forrester WC et al. (2002). Gaspar-Maia A, Alajem A, Polesso F, Sridharan R, Mason MJ, A DNA vector-based RNAi technology to suppress gene expression Heidersbach A et al. (2009). Chd1 regulates open chromatin and in mammalian cells. Proc Natl Acad Sci USA 99: 5515–5520. pluripotency of embryonic stem cells. Nature 460: 863–868. Taylor BS, Schultz N, Hieronymus H, Gopalan A, Xiao Y, Carver BS Jones S, Wang TL, Shih IeM, Mao TL, Nakayama K, Roden R et al. et al. (2010). Integrative genomic profiling of human prostate (2010). Frequent mutations of chromatin remodeling gene ARID1A cancer. Cancer Cell 18: 11–22. in ovarian clear cell carcinoma. Science 330: 228–231. Weissman B, Knudsen KE. (2009). Hijacking the chromatin remodel- Kan Z, Jaiswal BS, Stinson J, Janakiraman V, Bhatt D, Stern HM ing machinery: impact of SWI/SNF perturbations in cancer. Cancer et al. (2010). Diverse somatic mutation patterns and pathway Res 69: 8223–8230 Review. alterations in human cancers. Nature 466: 869–873. Yin D, Ogawa S, Kawamata N, Tunici P, Finocchiaro G, Eoli M et al. King JC, Xu J, Wongvipat J, Hieronymus H, Carver BS, Leung DH (2009). High-resolution genomic copy number profiling of glioblas- et al. (2009). Cooperativity of TMPRSS2-ERG with PI3-kinase toma multiforme by single nucleotide polymorphism DNA pathway activation in prostate oncogenesis. Nat Genet 41: 524–526. microarray. Mol Cancer Res 7: 665–677.

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

Oncogene