Genome-Wide DNA Methylation Events in TMPRSS2–ERG Fusion-Negative Prostate Cancers Implicate an EZH2-Dependent Mechanism with Mir-26A Hypermethylation

Published OnlineFirst August 28, 2012; DOI: 10.1158/2159-8290.CD-12-0041

RESEARCH ARTICLE

Genome-wide DNA Methylation Events in TMPRSS2–ERG Fusion-Negative Prostate Cancers Implicate an EZH2-Dependent Mechanism with miR-26a Hypermethylation

Stefan T. Börno 1 , 3, Axel Fischer 1 , Martin Kerick 1 , Maria Fälth 4 , 6, Mark Laible 4, Jan C. Brase 4 , 7, Ruprecht Kuner 4 , Andreas Dahl 8 , Christina Grimm 1 , Behnam Sayanjali 1 , Melanie Isau 1 , 3, Christina Röhr 1 , 3, Andrea Wunderlich 1 , 3, Bernd Timmermann 2 , Rainer Claus5 , Christoph Plass 5 , Markus Graefen 9 , Ronald Simon 10 , Francesca Demichelis 11 , 13, Mark A. Rubin 12 , Guido Sauter 10 , Thorsten Schlomm 9 , Holger Sültmann 4 , Hans Lehrach 1 , and Michal R. Schweiger 1

ABSTRACT Prostate cancer is the second most common cancer among men worldwide. Altera- tions in the DNA methylation pattern can be one of the leading causes for prostate cancer formation. This study is the fi rst high-throughput sequencing study investigating genome-wide DNA methylation patterns in a large cohort of 51 tumor and 53 benign prostate samples using methylated DNA immunoprecipitation sequencing. Comparative analyses identifi ed more than 147,000 cancer-associated epigenetic alterations. In addition, global methylation patterns show signifi cant differences based on the TMPRSS2–ERG rearrangement status. We propose the hypermethylation of miR-26a as an alternative pathway of ERG rearrangement-independent EZH2 activation. The observed increase in differential methylation events in fusion–negative tumors can explain the tumorigenic process in the absence of genomic rearrangements.

SIGNIFICANCE: In contrast to TMPRSS2–ERG-rearranged tumors, the pathomechanism for gene fusion– negative tumors is completely unclear. Using a sequencing-based approach, our work uncovers signiﬁ cant global epigenetic alterations in TMPRSS2–ERG gene fusion–negative tumors and provides a mechanistic explanation for the tumor formation process. Cancer Discov; 2(11); 1–12. ©2012 AACR.

INTRODUCTION transmembrane protease serine 2 ( TMPRSS2) gene (4 ), most com- monly involving the v-ets erythroblastosis virus E26 oncogene homolog More than 900,000 men are diagnosed with prostate cancer (ERG) that is observed in approximately 50% of all prostate each year, making it the second most common cancer among cancer cases (5 ). The overexpression of ERG is thought to be men worldwide ( 1 ). The clinical course of prostate cancer is het- suffi cient for the initiation of prostate intraepithelial neoplasia erogeneous, ranging from indolent tumors requiring no therapy lesions, a precursor of prostate cancer ( 6 ). Other rearrangements during the patient’s lifetime to highly aggressive prostate cancer are less frequent and, interestingly, tend to be present in prostate developing into a metastatic disease. Despite its high prevalence, cancers harboring the TMPRSS2–ERG gene fusion (FUS+ ; refs. the clinical management of prostate cancer is limited by the low 7, 8 ). Recent work has uncovered frequent somatic deletions at specifi city of the existing diagnostic and prognostic tools and 5q21 and 6q21 including CHD1 and FOXO3 as well as recurrent the lack of effective systemic therapeutic strategies. SPOP mutations within ETS gene fusion–negative prostate can- Recent years have brought about a marked extension of cers, suggesting distinct subclasses within gene fusion–negative our understanding of the somatic basis of prostate cancer. tumors ( 9, 10 ). With 0.33 somatic protein altering mutations per megabase During prostate cancer progression, one of the most sig- (Mb), the mutation frequency in prostate cancer is signifi cantly nifi cantly upregulated genes is enhancer of zeste homolog 2 lower than in breast (∼1/Mb) or lung cancer (3.8/Mb) and lies ( EZH2) , which is also associated with invasiveness and high within the lowest range of cancer-associated mutations (2, 3). A malignancy of several other tumor entities ( 11, 12 ). The large proportion of prostate cancers harbor gene fusions involv- polycomb group protein EZH2 is responsible for silencing of ing members of the ETS family and the androgen-regulated homeobox (HOX) genes through H3K27 methylation during tissue development ( 13–15 ), and it links histone modifi cations to DNA methylation as EZH2 target genes may consec- Authors’ Affi liations: 1 Department of Vertebrate Genomics and 2 Next Generation Sequencing Group, Max Planck Institute for Molecular Genet- utively become hypermethylated (16, 17). In prostate cancer, ics; 3 Department of Biology, Chemistry, and Pharmacy, Free University, aberrant DNA methylation is found to be associated with Berlin; 4Cancer Genome Research and 5 Division of Epigenomics and Can- altered EZH2 expression, correlates with tumor progression, cer Risk Factors, German Cancer Research Center (DKFZ) and National and is proposed to be one of the earliest events in oncogenesis Center for Tumor Diseases, Heidelberg; 6 Cellzome AG, Heidelberg; 7 Sividon Diagnostics GmbH, Cologne; 8 Biotechnology Center, Technical University (12, 13, 15 , 18–20 ). Hypermethylation of the glutathione Dresden, Dresden; 9 Martini Clinic, Prostate Cancer Center and 10 Institute S-transferase pi 1 ( GSTP1) is considered to be a cardinal gate- of Pathology, University Medical Centre Hamburg-Eppendorf, Hamburg, keeper event in early prostate cancer development ( 21 ). Germany; 11Institute for Computational Biomedicine and 12 Department Most epigenetic studies conducted so far are focused on of Pathology and Laboratory Medicine, Weill Cornell Medical College, specifi c gene regions and interrogate cancer-associated differen- New York, New York; and 13 Centre for Integrative Biology, University of Trento, Trento, Italy tial methylation events in general without asking if epigenetic Note: Supplementary data for this article are available at Cancer Discovery alterations might differ in specifi c subgroups of prostate can- Online (http://cancerdiscovery.aacrjournals.org/). cer. This might be particularly important for TMPRSS2–ERG- Corresponding Author: Michal-Ruth Schweiger, Max Planck Institute for negative tumors, in which the pathomechanism of oncogenesis Molecular Genetics, 14195 Berlin, Germany. Phone: 49-30-84131339; is so far unclear. Recent identifi cations of different patterns Fax: 49-30-84131380; E-mail: [email protected] of DNA methylation in promoters of benign, cancerous, and doi: 10.1158/2159-8290.CD-12-0041 metastatic prostate tissues as well as decreased levels of LINE-1 © 2012 American Association for Cancer Research. methylation in fusion-negative prostate cancer revealed a more

NOVEMBER 2012CANCER DISCOVERY | OF2

RESEARCH ARTICLE Börno et al.

detailed picture of epigenetic alterations in prostate cancer samples and might be used as prostate cancer biomarkers ( 22 ). However, the epigenetic landscapes of prostate cancer (Supplementary Fig. S2C and S2D). subgroups are still not suffi ciently investigated to draw any con- Hypermethylated regions were more homogeneous among clusions. To understand alterations in the ERG fusion–negative the tumor samples (lower P values) than hypomethylated class of prostate cancer, we used a deep sequencing readout of regions. This might indicate a specifi c and site-directed meth- methylated DNA immunoprecipitation sequencing (MeDIP- ylation process as compared with a more unspecifi c global Seq) to screen 51 tumor and 53 benign prostate tissues (23–26 ). loss of methylation. We integrated the results with gene and microRNA (miRNA) Although most hypomethylated regions were detected in expression analyses and proposed a model for the development intergenic regions, only a quarter of the hypermethylated bins of aberrant DNA methylation patterns in ERG fusion–negative were located outside of genes and promoters (Supplementary (FUS− ) prostate cancers. Fig. S2E). We found a peak of hypermethylation within ±2 kb windows around the transcription start sites (TSS). In addi- RESULTS tion, we detected that hypermethylation is enriched within conserved [OR = 1.47, confi dence interval (95% CI ) = 1.43–1.50, Catalogs of Tumor-Specifi c P = 1.08 × 10 −209 ] and miRNA (OR = 2.05, CI = 1.83–2.29, P = Epigenetic Alterations 8.79 × 10 −32) regions, whereas hypomethylation of these sites = = = For the analysis of genome-wide methylation patterns in is lower than expected (ORcons 0.42, CI 0.4–0.44, P 0 and = = = × −6 51 prostate cancers and 53 normal prostate tissues, we used ORmiRNA 0.63, CI 0.51–0.77, P 4.37 10 ; Supplementary MeDIP followed by high-throughput sequencing by oligo- Fig. S2F). Pathway analyses revealed the developmental gene nucleotide ligation and detection (SOLiD) ( 26 ). All tumors group to be the most signifi cantly differentially methylated. Of selected for this study were staged pT2–pT4 and had Gleason 231 homeobox genes ( 28 ), we identifi ed 175 with a signifi cant scores ranging from 6 to 9. The TMPRSS2–ERG fusion TSS-associated hypermethylation. The differential methylation was present in 17 tumors; 20 tumors were TMPRSS2–ERG - of homeobox genes also resulted in a more stringent impact negative as determined by PCR (Supplementary Table S1). on transcription: In the complete data set, although we found For a validation of the genome-wide MeDIP-Seq data we signifi cant negative and positive correlations for 1,143 and 839 used a focused, bisulfi te-based, mass spectrometry approach genes, respectively, we found 53 negative and 8 positive cor- [BS-MS (Epityper)] in 79 heterogeneous regions covering pro- relations for homeobox genes. Consistent with the literature, moters, high and low CpG content areas, and endo- and exog- we found hypomethylations predominantly localized within enous regions (Supplementary Table S2 and Supplementary repeat regions mainly within long interspersed nuclear ele- Fig. S1A and S1B). Pearson correlation coeffi cients greater ments (LINE) and long terminal repeats (LTR) ( 20 , 24 ). than 0.8 were achieved. To assess whether the identifi ed loci Methylation Patterns Differentiate of differential methylations are tumor cell–specifi c alterations, and are not due to different cellular compositions of TMPRSS2–ERG FUS and FUS Tumors tumor and normal samples, we also applied BS-MS analyses Although methylation patterns clearly distinguished normal (EpiTyper) to investigate 36 differentially methylated regions from tumor tissues, we also found FUS− and FUS+ samples on matched micro- and macrodissected samples from two separated in a PCA ( Fig. 1A , Supplementary Fig. S3A). In line additional patients and achieved correlation values greater with previous reports on gene expression ( 29, 30 ), the PCA than 0.91 (Supplementary Fig. S1B and S1C). analyses of the MeDIP-Seq values suggested that FUS + samples Principal component analyses (PCA) of global methyla- exhibit a higher homogeneity than FUS− samples. In addition, tion revealed comprehensive differences between tumor and the FUS+ class seems more similar to normal tissues than to normal samples (Supplementary Fig. S2A). Similar results FUS− samples, implying that the DNA methylation patterns in were obtained after restricting the analyses to regions outside FUS− prostate cancer are considerably more altered. In testing of somatic copy number alterations (sCNA; Supplementary the extend of the global methylation profi les in normal, FUS+ , Fig. S2A). Within the genome-wide MeDIP-Seq data sets, and FUS− samples, we found a signifi cantly higher number of we identifi ed approximately 147,000 cancer-specifi c differen- methylation events in FUS − compared with normal and FUS+ tially methylated regions (cDMR): 85,406 hypermethylated tissues (P < 0.011 and P < 0.021, respectively; Fig. 1B ), with FUS+ and 61,308 hypomethylated [Benjamini–Hochberg (BH) cor- samples showing similar counts as normal samples. These rected Mann–Whitney test P < 0.05; Supplementary Table results were also valid after excluding regions with somatic S3A], including the well-characterized differentially methyl- copy number alterations (Supplementary Fig. S3B). We calcu- ated region in GSTP1 (Supplementary Fig. S2B). We identifi ed lated the patient-wise number of differential methylations in additional biomarkers that were equally, or even more, sig- the 147,000 cDMR regions identifi ed previously in the tumor nifi cantly differentially methylated than previously described to normal comparison and found FUS− samples enriched for ones. From a recently published list by Kobayashi and col- differential methylation (Supplementary Fig. S3C). leagues (27 ) containing 87 potential biomarker regions, we FUS − and FUS+ samples signifi cantly differ from one confi rmed 84 regions with our high-throughput sequencing another at 27,500 regions encompassing cancer census genes approach. Notably, within our list of signifi cantly differen- ( 31 ), homeobox ( 28 ), and miRNA genes [OR = 1.99 (P = 7.16 tially methylated regions, their top candidate was ranked × 10 −12, 95% CI = 1.65–2.38), OR = 1.56 ( P = 0.004, 95% CI = 36th and GSTP1 108th. Of the 110 most signifi cant differ- 1.14−2.09), and OR = 1.82 (P = 4.87 × 10−7 , 95% CI = 1.45– entially methylated regions, 887 combinations of 2 regions 2.26), respectively; Fig. 1C and Supplementary Table S3)]. We already allowed a perfect separation of tumor and normal also detected more hypomethylated regions in FUS− samples

OF3 | CANCER DISCOVERYNOVEMBER 2012 www.aacrjournals.org

Deregulated Methylomes in Prostate Cancer RESEARCH ARTICLE

ABMethylated bins

FUS+ FUS– Undet. 100 * 50 4e + 05 *

PC3 0 Bin count

2e + 05 –100

–100 0 100 200 300 NORM FUS+ FUS– PC1

CED FUS–/FUS+ LINE L1 160,000 3

2 140,000

1 TMP2 OR 2 rpm

log 120,000 0 * Cons Rmsk miR2k

–1 HOX2k ** 100,000 TMP1 CGI+shores –2 – +

– + FUS FUS FUS FUS NORM

Figure 1. Global methylation patterns of TMPRSS:ERG FUS− tumors are signifi cantly different from FUS + tumors. A, PCA of all methylated genomic regions for 51 prostate tumors. For each region (bin) the reads per million (rpm) values are used as input. Orange circles: TMPRSS2–ERG FUS + , red circles: FUS− prostate cancer samples. Gray circles: tumor samples with unknown fusion status. Undet., undetermined. B, genome-wide methylation in FUS +, FUS− , and normal (NORM) prostate tissues. Depicted are the counts of genomic regions (500 bp) found signifi cantly methylated (Binomial test: P < 0.001). Signifi cant differences are highlighted by asterisks ( t test: *, P < 0.05; **, P < 0.01). t test P values are normal versus FUS + P < 0.823; normal versus FUS− P < 0.011; FUS + versus FUS− P < 0.021. C, enrichments of hypermethylated regions in genomic areas in FUS + versus FUS− . Depicted are + log2 (ORs) of regions associated with homeobox genes (HOX2k), CpG islands and shores (CGI shores), miRNAs (miR2k), conserved regions (Cons), and repeat regions (Rmsk). D, methylation levels of LINE L1 elements in FUS + , FUS − , and normal tissues. Depicted are the amounts of reads in bins with LINE L1 elements in rpm. Signifi cant differences are highlighted by asterisks (t test: *, P < 0.05; **, P < 0.01). t test P values are normal versus FUS + < 0.254; normal versus FUS − <0.001; FUS + versus FUS − < 0.004. E, heatmap of 2 differentially methylated genomic regions for FUS− and FUS + samples (TMP1 = chr1:149033001–149033500; TMP2 = chr16:46414001–46414500). Given below is the TMPRSS2–ERG fusion status: FUS− (brown) and FUS + (orange). Methylation levels are color coded (red = low, yellow = intermediate, white = high).

than in FUS + and normal samples. In particular, in LINE L1 ferases ( DNMT1, DNMT3A, DNMT3B, EZH1, EZH2, Fig. 2A). elements, we found signifi cant hypomethylations ( Fig. 1D ), In our tumor samples, DNMT1, DNMT3A, and EZH2 levels consistent with Kim and colleagues ( 22 ). From our data set, the were increased. In addition, for EZH2 we could observe a − top 2 marker regions (TMP1: chr1:149033001–149033500 and signifi cant increase in the expression level in FUS in compar- + TMP2: chr16:46413501–46414500) were suffi cient to distin- ison with FUS tissues (Mann–Whitney P = 0.013). Upregula- + guish between FUS + and FUS− samples with 2 samples not cor- tion of EZH2 in FUS tissues can be explained by increased rectly assigned ( Fig. 1E ). This fi nding was validated with BS-MS ERG levels in this tumor subgroup ( Fig. 2B ; ref. 32 ). using 46 of the tumor samples already investigated by MeDIP-Seq EZH2 is a polycomb group protein with H3K27 methyl- (Supplementary Fig. S3D and Supplementary Table S2D). transferase activity that functions as a transcriptional repres- sor in prostate cells ( 12 , 32 ). Its function depends on an intact EZH2 Overexpression Correlates with Altered SET domain as well as an endogenous HDAC activity and is Methylation Patterns thought to induce a DNA hypermethylation of its target genes To identify genes that might be responsible for the during cancer development ( 12 , 16 ). We confi rmed a positive increased methylations in FUS− tumors, we investigated the correlation between the expression of EZH2 and promoter gene expression levels of DNA and histone methyltrans- methylation for 77 of 83 polycomb group target genes that

NOVEMBER 2012CANCER DISCOVERY | OF4

RESEARCH ARTICLE Börno et al.

ABERG miR26a miR26a DNMT1 DNMT3A DNMT3B EZH1 EZH2 expression expression methylation 6.5 7.6 ** ** ** 10 * ** ** ** ** * 5.8 7.2 7.0 7.4 ** ** ** * ** 11.0 * 6 ** 6.0 5.7 9 7.2 7.0 10.5 6.5 4 5.6 5.5 8 7.0 6.8 10.0 (expression) (expression) (expression) 5.5 3 2 2 2 6.8 6.0 7 MeDIP value log log log 5.0 9.5 6.6 5.4 6.6 2 6 9.0 5.3 5.5 4.5 6.4 6.4 1 – – – – – – – – + + + + + + + + FUS FUS FUS FUS FUS FUS FUS FUS FUS FUS FUS FUS FUS FUS FUS FUS NORM NORM NORM NORM NORM NORM NORM NORM

Figure 2. EZH2 expression is signiﬁ cantly increased in FUS − prostate cancers due to decreased miR-26a levels. A, gene expression levels of DNMT1, + − DNMT3A, DNMT3B, EZH1, and EZH2 in normal, FUS , and FUS prostate tissues are given as log2 values of the array signal. B, ERG (left) and miR-26a (mid- dle) expression in normal, FUS +, and FUS − tissue samples. Right: DNA methylation in the miR-26a region (chr12:58217001–58219000) in normal, FUS +, and FUS − samples. Depicted are the MeDIP-seq values (rpm). Signiﬁ cances were tested with the Mann–Whitney Rank Sum Test (*, P < 0.05; **, P < 0.01).

are associated with metastatic prostate cancer ( 33 ). Thirty-fi ve than in FUS + samples (Fig. 2A). We therefore looked at EZH2 of these also showed a signifi cant correlation between gene promoter methylation, but found no association of enhanced expression and EZH2 expression (Supplementary Fig. S4 and expression of EZH2 with promoter methylation in FUS− sam- Supplementary Table S4). An analysis of 204 HOX genes ( 28 ), ples; rather, the methylation levels were found to be comparable the primary targets of EZH2-mediated cellular effects, revealed in both tumor subsets (minimal P value after BH correction = a positive correlation for 189 (92%) of them between EZH2 0.29). We next turned our attention to the expression levels of expression and HOX gene promoter methylation ( Fig. 3A and miRNAs, as we had found miRNA genes to be preferred sites B and Supplementary Table S4). Of these, 117 (62%) were of differential methylation in FUS− samples. In particular, we signifi cantly downregulated. Interestingly, for HOXC6 , we fi nd investigated the expression levels of 6 miRNAs (miR-26a , -101, a hypermethylation and upregulation of the short transcript -138 , -124 , -214, and let-7b ), suggested regulators of EZH2 ( 11 , forms, whereas for the long isoforms, no differential methyla- 36 ). For the quantitative PCR (qPCR) analyses, we used the same tion and no change in the expression level could be detected. set of primary prostate tumors that had previously been used for the methylation analyses. For miRNA-124 , -214 and -138 , we EZH2 Expression Is Regulated by ERG found signifi cant differences between normal and tumor tissues + - and miR-26a in FUS and FUS Prostate ( Fig. 5A ). However, these miRNAs showed no differential expres- Cancers, Respectively sion between FUS − and FUS + samples. In contrast, for miR-26a , a EZH2 and MYC are known targets of the ERG transcription miRNA whose downregulation is associated with EZH2 upregu- factor and are signifi cantly overexpressed in TMPRSS2 –ERG lation in nasopharyngeal carcinoma ( 11 ), breast cancer ( 37 ), as FUS + samples (34, 35). As expected, we found increased EZH2 well as in a murine lymphoma model ( 38 ), we found signifi cant and MYC expression in FUS+ samples and observed ERG/ expression differences between FUS + and FUS − samples. We cor- EZH2 and ERG/MYC expression correlation in the FUS + sam- roborated this fi nding in prostate cancer cell lines ( Fig. 5B ). In ple set (Spearman correlation = 0.52, P < 1 × 10−5 and correla- line with the in vivo data, the FUS − cells DU145 and PC3 express tion = 0.79, P < 2.2 × 10 −6, respectively). Consistent with this lower levels of miR-26a and more EZH2 than the FUS + cells VCaP fi nding, a knockdown of ERG in VCaP prostate cancer cells and NCI-H660. We also transfected DU145 cells with miR-26a revealed a downregulation of EZH2 and MYC (Fig. 4A–C). mimics and found a more than 2-fold decrease of EZH2 gene In contrast, in FUS − samples the ERG expression level is low, expression level ( Fig. 5C ). but MYC and EZH2 are still increased. MYC is signifi cantly Hypermethylation of a region of 2 kb in FUS − samples increased in tumor samples (P < 1.33 × 10 −21 ) with no signifi cant (chr12:58217001–58219000) in the vicinity of miR-26a was difference between FUS− and FUS+ tumors (P = 0.267). The found to be negatively correlated with miR-26a expression increase in MYC might be due to decreased miRNA-34c levels in in FUS − (correlation = −0.47; P < 4.8 × 10 −5 ), but this region our samples, a predicted regulator of MYC. In addition, we found was not found to be hypermethylated in FUS + samples (Fig. hypermethylation of the miRNA-34 region, suggesting a suppres- 2B and Supplementary Fig. S5A). Furthermore, expression sion of the miRNA transcription. For EZH2 , the situation is even of miR-26a was negatively correlated to EZH2 expression in more pronounced because the expression of EZH2 is even higher FUS − samples but not in FUS + samples (correlation = −0.37,

OF5 | CANCER DISCOVERYNOVEMBER 2012 www.aacrjournals.org

Deregulated Methylomes in Prostate Cancer RESEARCH ARTICLE

A 100 80 60 40 20 0 5 10 20 2515 0 Fusion status

Expression in tumor Downregulated Upregulated

Fusion status FUS– FUS+

0102030 Patient B

FUS–/NORM FUS+/NORM Figure 3. Expression and methylation of homeobox genes in TMPRSS2–ERG FUS + and FUS − tumors. A, patientwise patterns of promoter hypermethylation and corresponding gene expression of homeobox genes. Homeobox genes had to be differentially methylated in at = − = + 98 451 least 5 patients. Patients are ordered according to fusion state (brown FUS , orange FUS ) and sum of affected genes. Each visible square in the matrix marks a hypermethylated promoter. Color coding of squares depicts expression changes (red: upregulated in tumor, green: downregulated in tumor, gray: no differential expression). The histograms on both sides summarize the data patientwise (top) and genewise (side). B, number of homeobox genes that 51 show a correlation of promoter methylation and EZH2 expression of below −0.4 differenti- ated by TMPRSS2–ERG fusion state (FUS − and FUS + ).

P < 3 × 10−3 and correlation = 0.09, P = 0.48, respectively). A in PC3 cells, with less strong effects in DU145 cells (Fig. 5F luciferase reporter assay with the region of the differentially and data not shown). methylated region (DMR) at either the miR-26a locus or the positive control dual oxidase 1 (DUOX1) promoter region upstream of the luciferase gene resulted in signifi cantly lower DISCUSSION luciferase signals in DUOX1 and the miR-26a system after in Our analyses provide a molecular mechanism for the devel- vitro methylation (Fig. 5D). The miR-26a region is methylated opment of fusion gene–negative prostate cancer. Comparing in FUS − DU145 and PC3 cell lines, with less methylation in FUS − and FUS + tumor samples on a genome-wide scale, we DU145 cells ( Fig. 5E and Supplementary Fig. S5A). We vali- found signifi cantly more differentially methylated regions in dated the methylation in the miR-26a region with a bisulfi te FUS − than in FUS + tumors and normal samples. Among the sequencing approach and found the region in DU145 less hypermethylated regions, we found a signifi cant enrichment methylated than in FUS + VCaP cells (Supplementary Fig. of homeobox gene promoters (28 ). Homeobox genes are S5B). We thus tested if a treatment of fusion gene–negative involved in embryonic development and cell lineage determi- cells with 5-aza-2′-deoxycytidine results in an increased miR- nation and a deregulation of these genes might partly explain 26a level. Indeed, we determined an increased expression of the tumor dedifferentiation phenotype. A candidate for miR-26a and accompanying decrease in EZH2 after treatment the deregulation of homeobox genes, EZH2 , is signifi cantly

NOVEMBER 2012CANCER DISCOVERY | OF6

RESEARCH ARTICLE Börno et al.

tumors. The study from Kim and colleagues is based on a ABCERG EZH2 MYC 1.2 1.2 1.2 MethylPlex-next generation sequencing (M-NGS) approach concentrating on high guanine and cytosine (GC) content 1.0 1.0 1.0 regions. A comparison of both technologies resulted in con- 0.8 0.8 0.8 cordance rates of 89% in CpG island regions and 62% outside 0.6 0.6 0.6 of CpG islands, with more than 3 times more DMRs detected 0.4 0.4 0.4 in MeDIP-Seq experiments (3,928 in MeDIP-Seq and 1,274 Target/GAPDH 0.2 0.2 0.2 in M-NGS in CpG containing promoter regions; ref. 22 ). This low number of detected DMRs within M-NGS might 0.0 0.0 0.0 nt nt

nt be due to a speciﬁ c GC enrichment step within the M-NGS protocol, which resulted in a loss of low CpG density DMRs. Untrans. Untrans. Untrans.

ERG.KO1 ERG.KO2 ERG.KO1 ERG.KO2 Thus, on an extended global scale, we fi nd a higher degree of ERG.KO1 ERG.KO2 deregulated methylations in FUS − tumors with higher num- 10 7.5 bers of differential methylation events. 7.0 10 Our fi nding of extensive differences between the methyla- 9 + − 6.5 tion patterns of FUS and FUS samples suggests 2 patho- 9 8 mechanistic models (Fig. 6). 6.0 + 8 EZH2 overexpression in TMPRSS2–ERG FUS samples can 7 5.5 gene expression be explained by ERG overexpression, as EZH2 is a target 2 5.0 7 gene of the ERG transcription factor (34 ). In contrast, EZH2 Log 6 − 4.5 overexpression in FUS tumors seems to be regulated in a 6 – – – + + + different manner, as ERG is not increased in these tumors, but EZH2 levels are even higher. Increased MYC expression FUS FUS FUS FUS FUS FUS NORM NORM NORM in prostate cancer alone does not suffi ce as an explanation for the regulation of EZH2 because MYC expression levels are comparable between FUS + and FUS − tumors and thus do + Figure 4. ERG is overexpressed in FUS tumors and cell lines, resulting EZH2 in increased EZH2 and MYC levels. Expression of (A) ERG, ( B) EZH2 , and not explain the increased and differential methylation − + (C) MYC in PC-3 cell lines (top) and tissues (bottom). Shown are real-time levels in FUS cancers. Interestingly, by comparing FUS and PCR (RT-PCR) results of VCaP cells without transfection (untrans.), trans- FUS − tumors, we found signifi cantly more DMRs in FUS − fected with a nontargeting siRNA control (nt) or siRNAs against ERG (ERG. than in FUS + and normal samples, suggesting an epigenetic KO1, ERG.KO2). For normal and prostate cancer tissue samples, microarray − expression values for normal, FUS − , and FUS + samples are plotted. mechanism to play a predominant role in FUS tumors. In the course of our study, we noticed miRNA regions to be signifi cantly hypermethylated. We have shown here that overexpressed in our tumor datasets and even more highly one of the miRNAs targeting EZH2 , miR-26a , is suppressed expressed in FUS − tumors compared with FUS + ; cell line in FUS − prostate cancer. This fi nding is supported by reports data support these fi ndings. An increase in EZH2 expression describing miR-26a to be suppressed in nasopharyngeal car- induces epigenetic alterations and contributes to an errone- cinoma (13 , 41 ), lymphomas (38 ), and breast cancer (37 ). We ous expression of developmental genes, further promoting show here that the hypermethylation of the miR-26a locus is tumor formation (12 , 16 , 33 ). EZH2 has been associated with associated with high EZH2 levels in primary FUS − prostate high-grade prostate cancer ( 15 , 33 , 39, 40 ) and is also found cancers as well as in PC cell lines, whereas miR-26a mimics led to be overexpressed in many other tumor types ( 11 , 37 ). to a reduction of EZH2 . Thus, we suggest that in FUS − sam- Hypomethylations are predominating within repeat ples, a suppression of miR-26a caused by a hypermethylation regions in FUS − tumors (Fig. 1C and D). This is in line with leads to a decreased inhibition of EZH2 , thereby adding fur- a recent study by Kim and colleagues ( 22 ) showing that ther perturbations to the global DNA methylation profi le. fusion gene-negative prostate cancers in comparison with Recurrent gene fusions are also found in other tumors fusion gene-positive and normal harbor decreased levels including leukemia, lymphoma, sarcoma, non–small cell of methylations within LINE L1 regions. Notably, they not lung cancer, and other less common tumor entities ( 42 ). only showed lower numbers of methylations within LINE L1 Thus, the pathomechanistic model of a deregulated epi- regions in ETS − tumors, but their data also implicate lower genome in tumors without gene fusions might also be overall numbers of DMRs in ETS − tumors compared with applicable to these. First evidence may be seen in studies on FUS + tumors and adjacent benign tissues. This is in contrast acute myeloid leukemia (AML) where mutations in DNMT3A to our results because we found FUS + tumors containing appear exclusively in AML samples without translocations increased genome-wide numbers of DMRs. The discrep- (43 ) and where AML tumors with somatic mutations of ancy is most likely due to the different technologies used. isocitrate dehydrogenase 1 or 2 (IDH1 or 2 ) reveal different Even though the MeDIP-Seq technology is an affi nity-based DNA methylation patterns than patients with mixed line- approach, and as such high CpG regions are preferentially age leukemia (MLL) translocations, suggesting subgroup- enriched, we do not fi nd this to be an explanation for the specifi c methylation events ( 44 ). discrepancy between our study and the one from Kim and Gaining insight into tumor formation processes as colleagues ( 22 ). When we restrict our data sets to similar described in this study opens the way for personalized targeted regions, we still fi nd more cDMRs in FUS − than in FUS + therapeutics and for novel drug developments. In this regard,

OF7 | CANCER DISCOVERYNOVEMBER 2012 www.aacrjournals.org

Deregulated Methylomes in Prostate Cancer RESEARCH ARTICLE

A miR-124 miR-26a miR-101 miR-214 miR-138 let-7b 4.5 ** ** ** 4 ** 0 8.0 12.0 ** 11.0 * 4.0 ** ** 7.5 3 11.5 10.5 3.5 –5 7.0 2 11.0 3.0 10.0 6.5 10.5 2.5 1

gene (expression) –10 9.5 6.0 10.0 2 2.0 0 log 5.5 9.5 9.0 1.5 –15 –1 – – – – – – + + + + + + FUS FUS FUS FUS FUS FUS FUS FUS FUS FUS FUS FUS NORM NORM NORM NORM NORM NORM

BCMiR26a mimics miR26a EZH2 ERG in DU145 cells 3.5 1.0

3.0 0.8 EZH2 2.5

0.6 2.0

1.5

Target/TBP 0.4 Relative expression Relative 1.0 0.2 0.5

0.0 0.0 96 h 48 h 72 h 24 h PC3 PC3 PC3 120 h VCaP VCaP VCaP DU145 DU145 DU145 NCI-H660 NCI-H660 NCI-H660 DE F5-aza treatment Luciferase methylation assay miR26a of PC3 cells 2.5 1.0 miR26a EZH2 2.0 1.0 0.6 1.5

0.6 1.0 0.2 Relative expression Relative

Differential MeDIP value Differential 0.5 Normalized luciferase signal Normalized luciferase 0.2 –0.2 0.0 U M U M U M 58215000 58219000 DUOX1 miR26 pGI3 Aza 2 Aza 2 DMSO DMSO Aza 0.5 Aza 0.5

Figure 5. Methylation and expression of miR-26a in relation to EZH2 expression. A, expression levels of EZH2 targeting miRNAs in normal, FUS + , + and FUS prostate tissues are depicted as negative normalized C t values. Signiﬁ cant differences are highlighted by asterisks (Mann–Whitney test: *, P < 0.05; **, P < 0.01). B, RT-PCR determination of miR-26a, EZH2, and ERG expression in FUS − (PC3, DU145) and FUS + (VCaP, NCI-H660) cell lines. Normalizations were done in relation to TBP . C, time course of EZH2 expression in DU145 cells after treatment with miR-26a mimics. D, luciferase reporter assays with the DUOX1 or the miR-26a region cloned upstream of the luciferase reporter gene. pGl3 indicates empty vector controls, without (U) or with (M) in vitro methylation of the plasmid. E, differential methylation in DU145 (solid black line), PC3 cells (solid gray line), and tumor samples (dashed black line) compared with normal samples in the cDMR of miR-26a (black). F, relative expression of miR-26a and EZH2 after treatment with 5-aza-2′-deoxycytidine (0.5 or 2 μmol/L).

NOVEMBER 2012CANCER DISCOVERY | OF8

RESEARCH ARTICLE Börno et al.

who underwent radical prostatectomy for prostate cancer. Only sec- Normal + FUS– FUS tions containing exclusively normal tissue material with epithelial cell TMPRSS2–ERG content between 20% and 40% were included in the study. For Laser miR26a miR26a miR26a Capture Microdissection (LCM; Zeiss) of epithelial cells, 16-μm tissue sections were mounted on special LCM slides and brieﬂ y stained with TMPRSS2 TMPRSS2 hematoxylin and eosin to facilitate the localization of epithelial cells. Me ERG ERG Epithelial cells were collected by LCM from 10 tissue sections each. Me DNA was isolated using the DNA Mini Kit (Qiagen) according to the manufacturer’s instructions. The TMPRSS2–ERG fusion status was determined with real-time PCR (RT-PCR) following protocols from ++ +++ ++ Jhavar and colleagues (50 ) and Mertz and colleagues (51 ).

Methylation Profi ling by MeDIP-Seq EZH2 EZH2 – – – – EZH2 – In brief, 2.5 μg of genomic DNA were fragmented to 100 to 200 bp using the Covaris S2 system and end repaired with End Repair mix (Enzymatics) followed by a purifi cation step (Qiagen DNA Purifi ca- + ++ +++ tion Kit). Barcoded sequencing adapters (Supplementary Table S2) were ligated followed by nick translation with DNA polymerase I HOX (NEB, 10 U). genes Me Me Me For the enrichment step of the methylated DNA immunoprecipi- μ H3K27me3 Promoter tation (MeDIP), 5 g of an anti-5-methyl cytosine antibody (Eurogen- hypermethylation tec) coupled to magnetic beads was used. Coupling was conducted by incubation overnight in 1 × PBS + 0.5% bovine serum albumin (BSA). Sequencing libraries were generated before the enrichment and incubated with the beads for 4 hours in IP Buffer (10 mmol/L Figure 6. Model for the homeobox gene promoter methylation for sodium phosphate buffer pH 7, 140 mmol/L NaCl, 0.25% Triton normal, FUS −, and FUS + prostate cancers. In normal samples EZH2 expression is controlled by miR-26a . FUS + samples show an overexpres- X100). Beads were washed 3 times with IP buffer and DNA was sion of ERG targeting the EZH2 promoter and thus leading to EZH2 over- eluted in elution buffer (50 mmol/L Tris-HCl pH 7.5, 10 mmol/L expression. In FUS − samples, on the other hand, a region near miR-26a-2 EDTA, 1% SDS) by incubation for 15 minutes at 65°C. After 2 hours is hypermethylated and miR-26a expression is lowered, resulting in over- of incubation with proteinase K, the DNA was phenol/chloroform expression of EZH2 . EZH2 in turn suppresses homeobox gene expression extracted and ammonium acetate/ethanol precipitated. Enrichment by methylation of H3K27 and fi nally even manifests this silencing by DNA controls were conducted with RT-PCR targeting methylated as well hypermethylation of homeobox gene promoters. This effect is stronger in FUS − samples as these show a higher expression of EZH2 and a stronger as unmethylated regions. methylation of homeobox genes. SOLiD sequencing libraries were prepared following the SOLiD V3 fragment multiplex library preparation protocol (Life Technologies) with slight modifi cations. Following MeDIP, enrichment libraries were DNA methyltransferase inhibitors, like the well-known but amplifi ed with multiplex library PCR primers 1 and 2 and size-selected ′ toxic nucleoside DNMT inhibitor 5-aza-2 -deoxycytidine (45 ) and quantifi ed using qPCR with library PCR primer 1 and 2 (Sup- or the direct DNMT1 inhibitor procaine (46 ), might preferen- plementary Table S2). Dilutions of a prequantifi ed SOLiD fragment − tially be useful as treatment for patients with FUS prostate library control (DH10B) were used to create the standard. Samples tumors (47 ). In contrast, FUS + patients might benefi t from were diluted to 100 pg/μL using 1x Low TE buffer (ABI) and qPCR was PARP1 inhibitors, as suggested by Brenner and colleagues repeated. Identical amounts of up to 8 barcoded libraries were pooled. ( 48 ) who showed that PARP1 is essential for TMPRSS2–ERG Libraries were fi xed to sequencing beads by emulsion PCR following target gene transcription and directly interacts with ETS the templated bead preparation protocol for SOLiD V3. For each pool genes. Given the central role of EZH2 in prostate cancer, a of 8 samples, approximately 4 to 5 emulsion PCRs were needed to fi ll a full slide with up to 600 million beads. Sequencing was conducted on combination therapy with 3-deazaneplanocin A ( 49 ) might a SOLiD 3+ using barcode sequencing chemistry (5 + 35 bp; Lifetech). be considered. However, additional studies are required to The sequencing statistics are listed in Supplementary Table S1. further elucidate the ideal treatment strategies within the different prostate cancer subtypes. Alignment and Peak Detection Reads were aligned to HG19 using Applied Biosystem’s Bioscope METHODS Alignment module in seed and extend mode taking the fi rst 25 bp of the reads as seeds allowing 2 mismatches and a mismatch penalty Biological Samples score of −2 for extension. The aligned reads were elongated to 200 Prostate tissue samples were obtained from the University Medical bp in a strand-oriented manner. Redundant reads and reads with Centre Hamburg-Eppendorf (Hamburg, Germany). Approval for the no CpGs in the elongated sequence were excluded from further study was obtained from the local ethics committee and all patients analyses. Next, the HG19 reference genome was split into adjacent agreed to provide additional tissue sampling for scientifi c purposes. 500-bp bins, and the number of reads per bin was counted. Reads Tissue samples from 51 prostate cancer and 53 normal prostate tis- were assigned to a bin when their center was located within the bin. sues were included (Supplementary Table S1). None of the patients For sample-wise normalization, binwise read counts were related to had been treated with neoadjuvant radiation, cytotoxic, or endocrine the total read count of each sample (reads per million = rpm). Sta- therapy. To confi rm the presence of tumor, all punches were sec- tistical analyses were conducted using R (version 2.9.2). A binomial tioned, and tumor cell content was determined in every 10th section. distribution of the reads (null hypothesis) was assumed and, thus, a Only sections containing at least 70% tumor cells were included in the probability value for each read count was assigned. Differential meth- study. Normal prostate tissue samples were obtained from 53 patients ylations between normal and tumor tissues were calculated with the

OF9 | CANCER DISCOVERYNOVEMBER 2012 www.aacrjournals.org

Deregulated Methylomes in Prostate Cancer RESEARCH ARTICLE

Mann–Whitney test with corrections for multiple testing using the Affymetrix data was MIAMI-compliant submitted to GEO database Benjamini–Hochberg approach. Bins with corrected P values of <0.05 (NCBI, GEO, GSE29079). Expression values of the HOX (28 ) and are called differentially methylated. EZH2 target ( 33 ) genes were correlated with MeDIP-Seq data of all Primary sequencing data (read sequence fi les *.csfasta and read corresponding bins in a region of 2 kb around the TSS. quality fi les *.qual) and secondary analysis data (bed fi les) for all prostate cancers are available under the GEO accession number Determination of Copy Number Variations GSE35342. Bed fi les contain only extended, 0-CpG, and duplicate DNA from 41 tumor samples was hybridized on Affymetrix depleted reads. Genome-Wide SNP Arrays 6.0. The raw data were preprocessed and copy number number alterations were estimated with CRMA v2 ( 52 ). PCA and Hierarchical Clustering The median of all tumor samples was used as reference sample. PCA were conducted with the prcomp -function in R using all bins methylated in normal or tumor samples. Hierarchical clusterings Statistical Analyses were conducted with the heatmap.2-function included in the R pack- Statistical analyses were conducted using R (version 2.9.2). For deter- age “gplots” using the Euclidian distances and the agglomerative mining methylated bins, a binomial distribution of the reads (null algorithms “complete” and “ward.” hypothesis) was assumed, and thus a sample-wise probability value for each read count was assigned. Bins with P values <0.001 were called Annotations “signifi cantly methylated.” PCA were conducted with the prcomp - Additional information of the following databases was annotated function in R using all bins methylated in normal or tumor samples. to the 500-bp bins for functional and topographic analyses: CpG- For determining DMRs between tumor (n = 51) and normal sam- island (cpgIslandExt) and repeat masker data (rmsk) were extracted ples (n = 54), or FUS + (n = 17) and FUS − ( n = 20) samples, multiple from the UCSC database version GRCh37/HG19. Bins containing at 2-tailed Mann–Whitney tests were used and the P values corrected for least one base of an annotated CpG island were annotated as “CGI” multiple testing with the Benjamini–Hochberg approach. Corrected and the ±2 kb surrounding bins as CGI-shore. P values below 0.05 were called “signifi cant.” Analyses of differen- Genic, exonic, and transcription start sites were used from Biomart tially expressed genes and miRNAs were conducted similarly. Ensembl59, miRNA regions from miRBASE v16, homeobox genes For correlation analyses, only samples with gene expression and from Holland and colleagues (28 ). The cancer gene census list was methylation data were considered (gene expression data: n = 48, ± normal obtained from Futreal and colleagues ( 31 ) . Regions of 2 kb around = + = − = = n tumor 47, n FUS 17, n FUS 20; miRNA expression data: n normal 48, a TSS were annotated as “promoter region.” Conserved regions were n = 51, n + = 17, n − = 20) and Spearman correlation analyses annotated as follows: the nucleotide-wise conservation scores of a set tumor FUS FUS were conducted. Correlations with Benjamini–Hochberg corrected P of 46 species were extracted from UCSC (50 ) and the sums of these values of <0.05 were called “signifi cant.” scores for 100-bp sized consecutive bins were calculated. The highest scoring 1% quantile consisting of 281,450 100-bp bins were assigned miR-26a and EZH2, MYC, and ERG RT-PCR Quantifi cation to 176,060 different genomic 500 bp bins that are annotated as

“conserved.” For enrichment analyses, the log 2-ratios were calculated Total RNA, including miRNA, was extracted from tissue sections from the ORs using a Benjamini–Hochberg corrected P value cut-off as described by Brase and colleagues (53 ). Quantitative RT-PCR of 0.05 for differential methylation. amplifi cation of miRNAs was conducted using low-density TaqMan arrays v2.0, the TaqMan MicroRNA Reverse Transcription Kit, and Bisulfi te Mass Spectrometry TaqMan MicroRNA Assays (Applied Biosystems; ref. 53 ). For RT-PCR measurements of EZH2, MYC, ERG, and TBP, the following primers μ For bisulfi te analyses, 1 g of DNA was bisulfi te-converted using were used: EZH2-forward: 5′-tgt gga tac tcc tcc aag gaa, EZH2-reverse: Epitect bisulfi te Conversion Kit (Qiagen) and subsequently ampli- 5′-gag gag ccg tcc ttt ttc a, ERG-forward: 5′-aag tag ccg cct tgc aaa t, fi ed with specifi c primer pairs (Supplementary Table S2 ) carry- ERG-reverse: 5′-cag ctg gag ttg gag ctg t, TBP-forward: 5′-cgg ctg ttt ing a T7 promoter that was designed using the Epidesigner tool aac ttc gct tc, and TBP-reverse: 5′-cac acg cca aga aac agt ga. (51 ) with standard criteria (amplicon length: 400–600 bp). In vitro transcription was conducted, and the transcripts were cleaved and Cell Lines and Knockdown of ERG in VCaP subsequently analyzed using MALDI-TOF mass spectrometry on a Cells, Transfection of miR-26a Mimics, and MassARRAY Analyser 4. 5-aza-2′-deoxycytidine Treatment Bisulfi te Illumina Sequencing of the miR-26a Region VCaP, PC-3, DU145, and NCI-H660 cells were obtained from Ameri- can Type Culture Collection (ATCC). All cells were maintained and prop- Enrichment and sequencing of the miR-26a region was conducted agated according to the recommendations of ATCC; PC-3 cells were, in according to Agilent’s SureSelect Methyl-Seq Target Enrichment Proto- addition to ATCC, authenticated at the German Cancer Research Center col. Mapping of 100-bp paired end reads generated on a Illumina HiSeq (DKFZ). For the knockdown of ERG , VCaP cells were transfected with device was conducted with BSMAP with default parameters except the 50 nmol/L of ON-TARGETplus Non-Targeting Pool or ERG custom mismatch number (-v 10) and the quality trimming threshold (-p 2). siRNA ( 35 ; CGACAUCCUUCUCUCACA duplexes with UU overhang; Duplicates were removed with Picard Tools’ MarkDuplicate module both from Dharmacon) using Lipofectamine RNAiMAX (Invitrogen) as using default parameters. Methylation status of each CpG was calcu- recommended by the supplier. Cells were harvested for DNA and RNA lated by the ratio mC/(C+mC) and visualized with R. extraction 96 hours after transfection. DU145 cells were transfected with miR-26a mimics constructs with Gene Expression Analyses Lipofectamine 2000. After 24, 48, 72, 96, and 120 hours, incubation The Affymetrix GeneChip Whole Transcript Sense Target Labe- RNA was isolated and EZH2 levels were determined by qPCR. For the ling Assay was used to generate amplifi ed and labeled sense DNA. 5-aza-2′-deoxycytidine (Sigma) treatment, PC-3 cells were incubated Briefl y, 1 μg of total RNA was initially used for rRNA reduction. for 96 hours with 0.5 μmol/L, 2 μmol/L, or a corresponding amount Following the manufacturer’s instructions, cDNA was hybridized of dimethyl sulfoxide (DMSO; vehicle control). The medium was to the Affymetrix 1.0 Human Exon ST arrays. The raw data fi les changed every 24 hours. Cells were then harvested for DNA and RNA were preprocessed and normalized using Affymetrix powertools. The extraction.

NOVEMBER 2012CANCER DISCOVERY | OF10

RESEARCH ARTICLE Börno et al.

In Vitro Methylation and Luciferase Reporter Assay fi er,” and “Predict” (BMBF 01GS0890 to H. Sültmann, M. Fälth, and Target regions were cloned in the pGL3_basic vector (Promega) R. Kuner; 01GS0891 to S.T. Börno and A. Wunderlich; 01GS089105 using the following primer sequences: miR26a_Forward: 5′-ggg tga cag to M. Graefen; 01GS08111 to C. Grimm and M.R. Schweiger; Predict gag agg aga ca, miR26a_Reverse: 5′-tgg tca ttg agg gga aaa ag, DUOX1- to M. Isau). short-Forward: 5′-gca ccg acg gaa cat ctc ta, and DUOX1-short-Reverse: 5′-ctc tcg tcc ggt gcc tct. In vitro plasmid methylation was conducted Received February 8, 2012; revised July 19, 2012; accepted July 20, with SssI-CpG-Methyltransferase (NEB) following the manufactur- 2012; published OnlineFirst August 28, 2012. er’s instruction. Methylation was confi rmed by methylation-sensitive restriction enzyme digest (ApaL1, NEB) and sequencing. Methylated and unmethylated plasmids were transfected into HEK293 cells using REFERENCES PEI (polyethyleneimine) transfection reagents. After 24 hours, Dual- 1. Jemal A , Bray F , Center MM , Ferlay J , Ward E , Forman D . Global Luciferase Reporter assays (Promega) were conducted according to the cancer statistics. CA Cancer J Clin 2011 ; 61:69 – 90 . manufacturer’s instructions, and luciferase signals were detected using 2. Berger MF , Lawrence MS , Demichelis F , Drier Y , Cibulskis K , the Promega Multi Detection System. Sivachenko AY , et al. The genomic complexity of primary human prostate cancer. Nature 2011 ; 470:214 – 20 . Disclosure of Potential Confl icts of Interest 3. Kan Z , Jaiswal BS , Stinson J , Janakiraman V , Bhatt D , Stern HM , et al. Diverse somatic mutation patterns and pathway alterations in S.T. Börno, T. Schlomm, and M.-R. Schweiger have ownership human cancers. Nature 2010 ; 466:869 – 73 . interest (including patents) in pending patent. No potential confl icts 4. Rubin MA , Maher CA , Chinnaiyan AM . Common gene rearrange- of interest were disclosed by the other authors. ments in prostate cancer. J Clin Oncol 2011 ; 29:3659 – 68 . 5. Tomlins SA , Rhodes DR , Perner S , Dhanasekaran SM , Mehra R , Authors’ Contributions Sun XW , et al. Recurrent fusion of TMPRSS2 and ETS transcription Conception and design: S.T. Börno, T. Schlomm, H. Sültmann, factor genes in prostate cancer. Science 2005 ; 310:644 – 8 . H. Lehrach, M.-R. Schweiger 6. Klezovitch O , Risk M , Coleman I , Lucas JM , Null M , True LD , et al. A Development of methodology: S.T. Börno, A. Dahl, M.-R. Schweiger causal role for ERG in neoplastic transformation of prostate epithe- Acquisition of data (provided animals, acquired and managed lium. Proc Natl Acad Sci U S A 2008 ; 105:2105 – 10 . patients, provided facilities, etc.): S.T. Börno, M. Laible, J.C. Brase, 7. Palanisamy N , Ateeq B , Kalyana-Sundaram S , Pfl ueger D , Ramnaray- R. Kuner, A. Dahl, B. Timmermann, R. Claus, M. Graefen, R. Simon, anan K , Shankar S , et al. Rearrangements of the RAF kinase pathway in prostate cancer, gastric cancer and melanoma . Nat Med 2010 ; 16:793 – 8 . M.A. Rubin, G. Sauter, T. Schlomm, H. Sültmann, M.-R. Schweiger 8 . P fl ueger D , Terry S , Sboner A , Habegger L , Esgueva R , Lin PC , et al. Analysis and interpretation of data (e.g., statistical analysis, biosta- Discovery of non-ETS gene fusions in human prostate cancer using tistics, computational analysis): S.T. Börno, A. Fischer, M. Kerick, next-generation RNA sequencing. Genome Res 2011 ; 21:56 – 67 . M. Fälth, M. Laible, J.C. Brase, R. Kuner, C. Grimm, B. Sayanjali, M. Isau, 9. Barbieri CE , Baca SC , Lawrence MS , Demichelis F , Blattner M , C. Röhr, A. Wunderlich, B. Timmermann, R. Claus, C. Plass, M. Graefen, Theurillat JP , et al. Exome sequencing identifi es recurrent SPOP, FOXA1 F. Demichelis, M.A. Rubin, G. Sauter, H. Sültmann, M.-R. Schweiger and MED12 mutations in prostate cancer . Nat Genet 2012;44:685–9 . Writing, review, and/or revision of the manuscript: S.T. Börno, 10. Grasso CS , Wu YM , Robinson DR , Cao X , Dhanasekaran SM , A. Fischer, M. Kerick, M. Fälth, M. Laible, R. Kuner, C. Grimm, Khan AP , et al. The mutational landscape of lethal castration-resist- M. Isau, C. Röhr, A. Wunderlich, B. Timmermann, R. Claus, ant prostate cancer. Nature 2012;487:239–43. C. Plass, M. Graefen, R. Simon, F. Demichelis, M.A. Rubin, G. Sauter, 11. Alajez NM , Shi W , Hui AB , Bruce J , Lenarduzzi M , Ito E , et al. T. Schlomm, H. Sültmann, M.-R. Schweiger Enhancer of Zeste homolog 2 (EZH2) is overexpressed in recurrent Administrative, technical, or material support (i.e., reporting or nasopharyngeal carcinoma and is regulated by miR-26a, miR-101, organizing data, constructing databases): S.T. Börno, A. Fischer, and miR-98. Cell Death Dis 2010 ; 1:e85 . A. Dahl, T. Schlomm, H. Sültmann, M.-R. Schweiger 12. Varambally S , Dhanasekaran SM , Zhou M , Barrette TR , Kumar-Sinha Study supervision: T. Schlomm, H. Sültmann, H. Lehrach, M.-R. Schweiger C , Sanda MG , et al. The polycomb group protein EZH2 is involved in Performed MeDIP experiments: S.T. Börno progression of prostate cancer. Nature 2002 ; 419:624 – 9 . Provided bioinformatic analyses: S.T. Börno, A. Fischer, M. Kerick, 13. Hoffmann MJ , Engers R , Florl AR , Otte AP , Muller M , Schulz W A. M. Fälth Expression changes in EZH2, but not in BMI-1, SIRT1, DNMT1 or Provided RNA and microRNA data and bioinformatics analyses: DNMT3B are associated with DNA methylation changes in prostate cancer. Cancer Biol Ther 2007 ; 6:1403 – 12 . M. Fälth, J.C. Brase, R. Kuner, H. Sültmann 14. Pirrotta V . Polycombing the genome: PcG, trxG, and chromatin Generated sequence data: S.T. Börno, A. Dahl, B. Timmermann silencing. Cell 1998 ; 93:333 – 6 . Contributed to the procurement of tumor tissue, preparation of 15. Yu J , Mani RS , Cao Q , Brenner CJ , Cao X , Wang X , et al. An integrated DNA and RNA, clinical data, copy number, TMPRSS2–ERG sta- network of androgen receptor, polycomb, and TMPRSS2-ERG gene tus, and gene expression analyses: M. Laible, R. Kuner, M. Graefen, fusions in prostate cancer progression. Cancer Cell 2010 ; 17:443 – 54 . R Simon, G. Sauter, T. Schlomm, H. Sültmann 16. Vire E , Brenner C , Deplus R , Blanchon L , Fraga M , Didelot C, et al. Performed cell culture experiments: S.T. Börno, M. Laible, B. Sayanjali The Polycomb group protein EZH2 directly controls DNA methyla- Performed BS-MS experiments: S.T. Börno, R. Claus, C. Plass tion. Nature 2006 ; 439:871 – 4 . 17. Widschwendter M , Fiegl H , Egle D , Mueller-Holzner E , Spizzo G , Acknowledgments Marth C , et al. Epigenetic stem cell signature in cancer. Nat Genet This work is dedicated to Professor Manfred Schweiger who was a 2007 ; 39:157 – 8 . passionate scientist. The authors thank Anna Kosiura, Nada Kumer, 18. Feinberg AP , Vogelstein B . Hypomethylation distinguishes genes of some Uta Marchfelder, and Michael Zinke for excellent technical assist- human cancers from their normal counterparts . Nature 1983 ; 301:89 – 92 . 19. Schulz WA , Hoffmann MJ. Epigenetic mechanisms in the biology of ance, and Julia Liep for providing RNA samples. prostate cancer . Semin Cancer Biol 2009 ; 19:172 – 80 . Grant Support 20. Yegnasubramanian S , Haffner MC , Zhang Y , Gurel B , Cornish TC , Wu Z , et al. DNA hypomethylation arises later in prostate cancer This work has been supported by grants from the Federal Ministry progression than CpG island hypermethylation and contributes to of Education and Research “Proceed,” “Mutanom,” “Intestinal Modi- metastatic tumor heterogeneity. Cancer Res 2008 ; 68:8954 – 67 .

OF11 | CANCER DISCOVERYNOVEMBER 2012 www.aacrjournals.org

Deregulated Methylomes in Prostate Cancer RESEARCH ARTICLE

21. Lee WH , Morton RA , Epstein JI , Brooks JD , Campbell PA , Bova GS , 38. Sander S , Bullinger L , Klapproth K , Fiedler K , Kestler HA , Barth TF , et al. Cytidine methylation of regulatory sequences near the pi-class et al. MYC stimulates EZH2 expression by repression of its negative glutathione S-transferase gene accompanies human prostatic car- regulator miR-26a. Blood 2008 ; 112:4202 – 12 . cinogenesis. Proc Natl Acad Sci U S A 1994 ; 91:11733 – 7 . 39. Karanikolas BD , Figueiredo ML , Wu L . Comprehensive evaluation of 22. Kim JH , Dhanasekaran SM , Prensner JR , Cao X , Robinson D , the role of EZH2 in the growth, invasion, and aggression of a panel of Kalyana-Sundaram S , et al. Deep sequencing reveals distinct patterns prostate cancer cell lines. Prostate 2010 ; 70:675 – 88 . of DNA methylation in prostate cancer . Genome Res 2011 ; 21:1028 – 41 . 40. van Leenders GJ , Dukers D , Hessels D , van den Kieboom SW , 23. Chavez L , Jozefczuk J , Grimm C , Dietrich J , Timmermann B , Lehrach H , Hulsbergen CA , Witjes JA , et al. Polycomb-group oncogenes EZH2, et al. Computational analysis of genome-wide DNA methylation BMI1, and RING1 are overexpressed in prostate cancer with adverse during the differentiation of human embryonic stem cells along the pathologic and clinical features. Eur Urol 2007 ; 52:455 – 63 . endodermal lineage. Genome Res 2010 ; 20:1441 – 50 . 41. Lu J , He ML , Wang L , Chen Y , Liu X , Dong Q , et al. MiR-26a inhibits 24. Feber A , Wilson G , Zhang L , Presneau N , Idowu B , Down TA , et al. cell growth and tumorigenesis of nasopharyngeal carcinoma through Comparative methylome analysis of benign and malignant peripheral repression of EZH2 . Cancer Res 2011 ; 71:225 – 33 . nerve sheath tumours. Genome Res 2011;21:515–24. 42. Edwards PA . Fusion genes and chromosome translocations in the 25. Laird PW . Principles and challenges of genomewide DNA methyla- common epithelial cancers. J Pathol 2010 ; 220:244 – 54 . tion analysis. Nat Rev Genet 2010 ; 11:191 – 203 . 43. Ley TJ , Ding L , Walter MJ , McLellan MD , Lamprecht T , Larson DE , 26. Weber M , Davies JJ , Wittig D , Oakeley EJ , Haase M , Lam WL , et al. et al. DNMT3A mutations in acute myeloid leukemia. N Engl J Med Chromosome-wide and promoter-specifi c analyses identify sites of 2010 ; 363:2424 – 33 . differential DNA methylation in normal and transformed human 44. Akalin A , Garrett-Bakelman FE , Kormaksson M , Busuttil J , Zhang L, cells. Nat Genet 2005 ; 37:853 – 62 . Khrebtukova I , et al. Base-pair resolution DNA methylation sequenc- 27. Kobayashi Y , Absher DM , Gulzar ZG , Young SR , McKenney JK , ing reveals profoundly divergent epigenetic landscapes in acute mye- Peehl DM , et al. DNA methylation profi ling reveals novel biomarkers loid leukemia . PLoS Genet 2012 ; 8:e1002781 . and important roles for DNA methyltransferases in prostate cancer . 45. Samlowski WE , Leachman SA , Wade M , Cassidy P , Porter-Gill P , Genome Res 2011 ; 21:1017 – 27 . Busby L , et al. Evaluation of a 7-day continuous intravenous infusion 28. Holland PW , Booth HA , Bruford EA . Classifi cation and nomencla- of decitabine: inhibition of promoter-specifi c and global genomic ture of all human homeobox genes. BMC Biol 2007 ; 5:47 . DNA methylation. J Clin Oncol 2005 ; 23:3897 – 905 . 29. Sboner A , Demichelis F , Calza S , Pawitan Y , Setlur SR , Hoshid a Y , 46. Villar-Garea A , Fraga MF , Espada J , Esteller M . Procaine is a DNA- et al. Molecular sampling of prostate cancer: a dilemma for predict- demethylating agent with growth-inhibitory effects in human cancer ing disease progression. BMC Med Genomics 2010 ; 3:8 . cells. Cancer Res 2003 ; 63:4984 – 9 . 30. Setlur SR , Mertz KD , Hoshida Y , Demichelis F , Lupien M , Perner S , 47. Perry AS , Watson RWG , Lawler M , Hollywood D. The epigenome as a et al. Estrogen-dependent signaling in a molecularly distinct subclass therapeutic target in prostate cancer. Nat Rev Urol 2010 ; 7:668 – 80 . of aggressive prostate cancer. J Natl Cancer Inst 2008 ; 100:815 – 25 . 48. Brenner JC , Ateeq B , Li Y , Yocum AK , Cao Q , Asangani IA , et al. 31. Futreal PA , Coin L , Marshall M , Down T , Hubbard T , Wooster R , et al. Mechanistic rationale for inhibition of poly(ADP-ribose) polymer- A census of human cancer genes. Nat Rev Cancer 2004 ; 4:177 – 83 . ase in ETS gene fusion-positive prostate cancer. Cancer Cell 2011 ; 32. Cao R , Wang L , Wang H , Xia L , Erdjument-Bromage H , Tempst P , 19:664 – 78 . et al. Role of histone H3 lysine 27 methylation in Polycomb-group 49. Chiba T , Suzuki E , Negishi M , Saraya A , Miyagi S , Konuma T, et al. silencing. Science 2002 ; 298:1039 – 43 . 3-deazaneplanocin A is a promising therapeutic agent for the eradica- 33. Yu J , Rhodes DR , Tomlins SA , Cao X , Chen G , Mehra R , et al. A tion of tumor-initiating hepatocellular carcinoma cells . Int J Cancer polycomb repression signature in metastatic prostate cancer predicts 2012;130:2557–67 . cancer outcome. Cancer Res 2007 ; 67:10657 – 63 . 50. Jhavar S , Reid A , Clark J , Kote-Jarai Z , Christmas T , Thompson A , 34. Kunderfranco P , Mello-Grand M , Cangemi R , Pellini S , Mensah A , et al. Detection of TMPRSS2-ERG translocations in human prostate Albertini V , et al. ETS transcription factors control transcription of cancer by expression profi ling using GeneChip Human Exon 1.0 ST EZH2 and epigenetic silencing of the tumor suppressor gene Nkx3.1 arrays . J Mol Diagn 2008 ; 10:50 – 7 . in prostate cancer. PLoS One 2010 ; 5:10547 . 51. Mertz KD , Setlur SR , Dhanasekaran SM , Demichelis F , Perner S , 35. Sun C , Dobi A , Mohamed A , Li H , Thangapazham RL , Furusato Tomlins S , et al. Molecular characterization of TMPRSS2-ERG gene B , et al. TMPRSS2-ERG fusion, a common genomic alteration in fusion in the NCI-H660 prostate cancer cell line: a new perspective for prostate cancer activates C-MYC and abrogates prostate epithelial an old model. Neoplasia 2007 ; 9:200 – 6 . differentiation. Oncogene 2008 ; 27:5348 – 53 . 52. Bengtsson H , Wirapati P , Speed TP. A single-array preprocessing 36. Varambally S , Cao Q , Mani RS , Shankar S , Wang X , Ateeq B, et al. method for estimating full-resolution raw copy numbers from all Genomic loss of microRNA-101 leads to overexpression of histone Affymetrix genotyping arrays including GenomeWideSNP 5 & 6 . Bio- methyltransferase EZH2 in cancer. Science 2008 ; 322:1695 – 9 . informatics 2009 ; 25:2149 – 56 . 37. Zhang B , Liu XX , He JR , Zhou CX , Guo M , He M , et al. Pathologically 53. Brase JC , Johannes M , Mannsperger H , Falth M , Metzger J , decreased miR-26a antagonizes apoptosis and facilitates carcinogen- Kacprzyk LA , et al. TMPRSS2-ERG- specifi c transcriptional modu- esis by targeting MTDH and EZH2 in breast cancer. Carcinogenesis lation is associated with prostate cancer biomarkers and TGF-beta 2011 ; 32:2 – 9 . signaling. BMC Cancer 2011;11:507 .

NOVEMBER 2012CANCER DISCOVERY | OF12

Genome-wide DNA Methylation Events in TMPRSS2−ERG Fusion-Negative Prostate Cancers Implicate an EZH2-Dependent Mechanism with miR-26a Hypermethylation

Stefan T. Börno, Axel Fischer, Martin Kerick, et al.

Cancer Discovery Published OnlineFirst August 28, 2012.

Updated version Access the most recent version of this article at: doi:10.1158/2159-8290.CD-12-0041

Supplementary Access the most recent supplemental material at: Material http://cancerdiscovery.aacrjournals.org/content/suppl/2012/07/30/2159-8290.CD-12-0041.DC1

E-mail alerts Sign up to receive free email-alerts related to this article or journal.

Reprints and To order reprints of this article or to subscribe to the journal, contact the AACR Publications Department at Subscriptions [email protected].

Permissions To request permission to re-use all or part of this article, use this link http://cancerdiscovery.aacrjournals.org/content/early/2012/10/23/2159-8290.CD-12-0041. Click on "Request Permissions" which will take you to the Copyright Clearance Center's (CCC) Rightslink site.