Journal of Pathology J Pathol (2012) Published online in Wiley Online Library (wileyonlinelibrary.com) DOI: 10.1002/path.4027 ORIGINAL PAPER Chemical castration and anti-androgens induce differential gene expression in prostate cancer Saara Lehmusvaara,1 Timo Erkkilä,2 Alfonso Urbanucci,1 Kati Waltering,1,2 Janne Seppälä,2 Antti Larjo,2 Vilppu J Tuominen,1 Jorma Isola,1 Paula Kujala,3 Harri Lähdesmäki,4 Antti Kaipia,5 Teuvo LJ Tammela5 and Tapio Visakorpi1 * 1 2 3 4 5 Institute of Biomedical Technology and BioMediTech, University of Tampere and Tampere University Hospital, Finland Department of Signal Processing and BioMediTech, Tampere University of Technology, Finland Department of Pathology, Tampere University Hospital, Finland Department of Information and Computer Science, Aalto University School of Science, Helsinki, Finland Department of Urology, University of Tampere and Tampere University Hospital, Finland *Correspondence to: Tapio Visakorpi, Institute of Medical Technology, FIN-33014 University of Tampere, Tampere, Finland. e-mail: tapio.visakorpi@uta.fi Abstract Endocrine therapy by castration or anti-androgens is the gold standard treatment for advanced prostate cancer. Although it has been used for decades, the molecular consequences of androgen deprivation are incompletely known and biomarkers of its resistance are lacking. In this study, we studied the molecular mechanisms of hormonal therapy by comparing the effect of bicalutamide (anti-androgen), goserelin (GnRH agonist) and no therapy, followed by radical prostatectomy. For this purpose, 28 men were randomly assigned to treatment groups. Freshly frozen specimens were used for gene expression profiling for all known protein-coding genes. An in silico Bayesian modelling tool was used to assess cancer-specific gene expression from heterogeneous tissue specimens. The expression of 128 genes was > two-fold reduced by the treatments. Only 16% of the altered genes were common in both treatment groups. Of the 128 genes, only 24 were directly androgen-regulated genes, according to re-analysis of previous data on gene expression, androgen receptor-binding sites and histone modifications in prostate cancer cell line models. The tumours containing TMPRSS2–ERG fusion showed higher gene expression of genes related to proliferation compared to the fusion-negative tumours in untreated cases. Interestingly, endocrine therapy reduced the expression of one-half of these genes and thus diminished the differences between the fusion-positive and -negative samples. This study reports the significantly different effects of an anti-androgen and a GnRH agonist on gene expression in prostate cancer cells. TMPRSS2-ERG fusion seems to bring many proliferation-related genes under androgen regulation. Copyright  2012 Pathological Society of Great Britain and Ireland. Published by John Wiley & Sons, Ltd. Keywords: neoplasia; prostatic; neoadjuvant; endocrine therapy; LHRH; TMPRSS2–ERG Received 8 December 2011; Revised 4 February 2012; Accepted 9 March 2012 No conflicts of interest were declared. Introduction Prostate cancer is a highly hormone-sensitive malignancy [1]. More than 70 years ago, Huggins and Hodges [2] observed the relationship between testosterone and prostate cancer progression; they also documented the clinical benefits of castration or oestrogen injections in patients with advanced prostate cancer. Thirty years later, the discovery of testosterone reduction in male rats with gonadotropin-releasing hormone (GnRH) treatment [3] quickly led to the development of GnRH agonist therapy for prostate cancer [4]. Several randomized trials have demonstrated the equivalency of chemical and surgical castration on treatment response of prostate cancer [5,6]. Anti-androgen treatment was originally designed to be used in combination with castration. Later, Copyright  2012 Pathological Society of Great Britain and Ireland. Published by John Wiley & Sons, Ltd. www.pathsoc.org.uk similar survival rates for non-steroidal anti-androgen monotherapy and castration have been reported in patients with locally advanced cancer [7,8]. In metastasized cancers, however, castration is superior to anti-androgen treatment [7]. Therefore, non-steroidal anti-androgen treatment can be used as an alternative to castration for patients with locally advanced prostate cancer [9,10]. Although 95% of patients initially respond to endocrine treatment [11], practically all cancers eventually become resistant to it. Mechanisms such as amplification and over-expression of the androgen receptor (AR) gene, mutations of the AR leading to promiscuous ligand usage, altered expression of the AR coregulators, truncated AR splice variants and the expression of steroidogenic enzymes enabling intracrine testosterone J Pathol (2012) www.thejournalofpathology.com S Lehmusvaara et al. production have been suggested to mediate the development of castration-resistant prostate cancer (CRPC) [12]. The AR is a transcription factor that regulates the expression of hundreds of genes. However, only one bona fide common AR target gene important in prostate cancer has been identified. The TMPRSS2–ERG fusion gene has been identified in up to one-half of all prostate cancers [13]. Despite several studies, the functional mechanism of ERG in prostate cancer progression remains unclear. Recently, Yu et al [14] reported that chromatin binding of the ERG and AR transcription factors overlaps and that ERG disrupts AR signalling. The treatment of CRPC remains a major clinical problem. The mean overall survival after disease progression is only approximately 20 months [15]. However, new potential treatments for CRPC have recently emerged. In addition to docetaxel and cabazitaxel, a novel anti-androgen, MDV3100, and a CYP17 inhibitor, abiraterone, have also demonstrated efficacy for the treatment of CRPC [16,17]. Therefore, the ability to predict treatment response to initial endocrine therapy is important to assess whether additional treatments should be initiated. A decline in prostate-specific antigen (PSA) after endocrine therapy has been reported to predict treatment response [11]. Still, improved biomarkers that can be evaluated at the time of diagnosis are needed. Despite the wide clinical use of GnRH agonists and anti-androgens, their effects at the transcriptome level are poorly studied. To the best of our knowledge, direct comparisons of the effects of these two treatment modalities on gene expression have not been published. In this study, we utilized rare clinical specimens from neoadjuvant endocrine-treated patients and examined the gene expression patterns induced by the GnRH agonist, goserelin, and a non-steroidal anti-androgen, bicalutamide. In addition, we evaluated the effects of the TMPRSS2–ERG fusion on the expression profiles induced by the endocrine treatments.  from prostatectomies were embedded in Tissue-Tek (Sakura, Alphen aan den Rijn, The Netherlands) and frozen in liquid nitrogen. Total RNA was extracted with Trizol (Invitrogen, Carlsbad, CA, USA) according to the manufacturer’s instructions. Adjacent sections before and after RNA extraction site were also cut and stained with haematoxylin and eosin (H&E). The stained slides were scanned and visualized with a virtual microscope system [18] and the amounts of cancer, benign epithelium and stroma in the specimens were assessed. A tissue microarray (TMA) was constructed from formalin-fixed, paraffin-embedded prostatectomy specimens. The second set of samples (see Supporting information, Table S1) was obtained from the Tampere University Hospital (Tampere, Finland). The specimens were confirmed to contain > 70% of malignant or nonmalignant epithelial cells using H&E-stained slides. Total RNA was extracted from the frozen sections with Trizol (Invitrogen), and first-strand cDNA synthesis was performed using SuperScript III reverse transcriptase (Invitrogen) and random primers (Fermentas, Glen Burnie, MA, USA). The use of clinical material in this study was approved by the Ethical Committee of the Tampere University Hospital. Written informed consent was obtained from the patients. Expression profiling Microarray hybridization was performed in the Finnish Microarray Centre at the Turku Centre for Biotechnology, Turku, Finland. First, 300 ng RNA was amplified using the Illumina RNA TotalPrep Amplification kit (Ambion, Austin, TX, USA), followed by cRNA hybridization with Illumina’s Human HT-12 Expression BeadChip, v 3 (targeting > 25 000 annotated genes), according to the manufacturer’s instructions. Finally, microarray chips were scanned with the Illumina BeadArray Reader, BeadScan software v 3.5. (submitted to Array Express, Ref ID E-MEXP-3081). Methods In silico data analysis Clinical samples A randomized clinical trial comparing the neoadjuvant GnRH analogue and an anti-androgen was conducted at the Tampere University Hospital in Finland between 2004 and 2006. Twenty-eight men with localized prostate cancer were randomized into three groups: no treatment (11 men); anti-androgen (bicalutamide, 150 mg/day administered orally for 12 weeks, 9 men); or GnRH agonist (goserelin acetate, 3.6 mg administered by subcutaneous injection every 4 weeks for 12 weeks, 8 men) (see Supporting information, Table S1; clinicopathological characteristics of the cases are found in Table S2). After neoadjuvant endocrine treatment (or no treatment), patients underwent a radical prostatectomy. Fresh specimens Copyright  2012 Pathological Society of Great Britain and Ireland. Published by John Wiley & Sons, Ltd. www.pathsoc.org.uk Because the amount of cancer in the prostatectomy specimens varied (0–85%), an in silico Bayesian modelling tool was used to predict the gene expression in different tissue compartments [19]. Briefly, using the percentages of different cell types for each sample, DSection analysis calculated the expression values for each probe in benign epithelia, stroma and malignant cells. Before DSection analysis, the samples were normalized. First, the probes with a value of < 100 in all of the samples were excluded. Second, the average and standard deviation (SD) of the probes in every sample were calculated separately and defined to 0 and 1, respectively. DSection analysis calculated the average expression values of the individual genes for each treatment group and each tissue compartment. Therefore, J Pathol (2012) www.thejournalofpathology.com Effect of endocrine therapy in prostate cancer rather than assessing the differences between individuals, the expression differences between treatment groups were quantified. Microdissection Freshly frozen slides from the prostatectomy specimens were stained with HistoGene Staining Solution (Arcturus Bioscience, CA, USA). Both the cancerous and stromal tissue compartments were obtained by laser capture microdissection (Arcturus, Veritas). RNA from the microdissected samples was extracted with the PicoPure RNA Isolation Kit (Arcturus Bioscience), according to the manufacturer’s instructions, and followed by first-strand cDNA synthesis as described above. qRT–PCR For qRT–PCR analysis, the Maxima SYBR Green/Rox qPCR Master Mix (Fermentas) and CFX96 RealTime System apparatus (Bio-Rad) were used. Primer sequences are listed in Table S3 (see Supporting information). All annealing steps occurred at 60 ◦ C, and β-actin was used as a reference gene. Immunohistochemistry Immunostainings were performed using polyclonal rabbit antibodies (Sigma-Aldrich, St. Louis, MO, USA) against TMEFF2 (HPA026553, diluted 1 : 200), TPD52 (HPA028427, diluted 1 : 8000) and NEDD4L (HPA024618, diluted 1 : 20). Polyclonal rabbit antibody against ERG (EPR3864, diluted 1 : 100; Epitomics, Burlingame, CA, USA) was used. The TMA sections were deparaffinized, followed by antigen retrieval in 5 mM Tris–HCl:1 mM EDTA, pH 9, at 121 ◦ C for 2 min in an autoclave. Bound antibody was visualized with the Power-Vision+ Poly-HRP IHC Detection Kit (ImmunoVision Technologies, Brisbane, CA, USA). The sections were counterstained with haematoxylin and the staining was scored on a scale of 0–3 (0, no staining; 1, weak staining; 2, moderate staining; and 3, high-intensity staining) for TMEFF2, TPD42 and NEDD4L. ERG staining was scored as positive or negative. Staining intensity was measured only from the cancerous areas in a blinded fashion by one of the authors (SL). Cancerous areas were confirmed with a mixture of two mouse monoclonal antibodies against the basal cell layer (p63, diluted 1 : 200; and HMW keratin, clone 34β12, diluted 1 : 100; both from LabVision, Fremont, CA, USA) and a rabbit monoclonal antibody against AMACR (clone 13H4; Dako, Copenhagen, Denmark), as described [20]. Fluorescence in situ hybridization Three-colour fluorescence in situ hybridization (FISH) was carried out on the TMA slides as previously described [21]. Probes for ERG (RP11-164E1), TMPRSS2 (RP11-814F13) and the region in between (RP11-367P1) were used. Copyright  2012 Pathological Society of Great Britain and Ireland. Published by John Wiley & Sons, Ltd. www.pathsoc.org.uk Ontology analysis The web-based integrated data-mining system WebGestalt [22] was used to determine the gene ontologies. Statistical analyses The Benjamini–Hochberg method, one-way ANOVAs with Bonferroni’s multiple comparison, χ2 and Fishers’ exact tests were used for statistical analyses. Results To assess gene expression specifically in cancer cells from heterogeneous tissue samples, we used an in silico Bayesian modelling tool, DSection [19], to analyse microarray data. This tool allows the simultaneous estimation of gene expression in three tissue compartments (cancer, epithelium and stroma) and in the three experimental groups (control, goserelin and bicalutamide). To validate the modelling tool, we first assessed the differential expression in cancer tissue and between nontreated control and treatment groups with and without DSection. Without DSection, the samples were first categorized into three groups, depending on their cancer tissue content: low (0–9%), moderate (23–49%) and high (74–85%). We then compared the results with the DSection model. In all the groups, the genes were ranked based on their fold change (FC) between the treatment and control. We observed clear dissimilarity between the ranked lists of the lowest purity group (0–9%) and the DSection model (9% of genes in common), whereas the group with 74–85% purity were 71% similar with the DSection model. Therefore, DSection seems to reliably estimate cancer tissuespecific expression in heterogeneous tissue specimens. To further validate DSection, we microdissected tissue specimens and extracted RNA from both the cancer and stromal compartments of 11 samples (three or four from each group) and measured the expression of 10 genes by qRT–PCR. Only one of 10 genes (MAOA) showed differences between the DSection prediction and qRT–PCR (Figure 1). To determine similarities and differences between the bicalutamide and goserelin treatment groups, we compared gene expression profiles induced by the two treatments in cancer tissues. Altogether, 128 genes had > two-fold reduced expression [false discovery rate (FDR) < 0.01], among which 33 genes were common to both treatment groups (Figure 2a; see also Supporting information, Table S4). Among 86 genes that showed increased expression, two genes were common (Figure 2b; see also Supporting information, Table S5). Overall, only 16% of the most differently expressed genes were common to both treatments. Given that both bicalutamide and goserelin treatments aim to inhibit AR signalling, we focused on the genes with reduced expression after the treatments. To J Pathol (2012) www.thejournalofpathology.com S Lehmusvaara et al. In silico predicted In silico predicted qPCR validated 1.0 1.0 0.5 0.5 0.0 0.0 ctrl bic. ctrl gos. bic. Cancer Stroma 1.0 1.0 0.5 0.5 0.0 ctrl bic. gos. 1.0 0.5 0.5 0.0 0.0 ctrl bic. gos. ctrl bic. gos. 1.5 of MAOA 1.0 Cancer Stroma Relative expression 1.5 1.5 0.0 ctrl bic. gos. 1.5 1.0 0.5 0.0 Cancer Stroma 1.0 0.5 ctrl bic. gos. 0.0 ctrl bic. gos. 1.5 1.0 1.0 0.5 0.5 0.0 0.0 ctrl bic. Cancer Stroma ctrl gos. bic. gos. 1.5 1.0 0.5 0.0 Cancer Stroma 1.0 0.5 ctrl bic. gos. 0.0 ctrl bic. gos. 1.5 1.0 1.0 0.5 0.5 0.0 ctrl bic. gos. Cancer Stroma 0.0 ctrl bic. gos. of NEDD4L I 1.5 Relative expression of TMEFF2 D 1.5 of MBOAT2 1.5 Relative expression H of DHCR24 2.5 2.0 1.5 1.0 0.5 0.0 ctrl bic. gos. 2.5 2.0 1.5 1.0 0.5 0.0 Cancer Stroma ctrl bic. gos. E 3 1.5 2 1.0 1 0.5 0 0.0 ctrl bic. gos. Cancer Stroma ctrl bic. gos. 1.5 of TMPRSS2 of TPD52 E Relative expression Relative expression 1.5 1.5 gos. C Relative expression 2.0 G of NPY Relative expression Cancer Stroma of GSTT1 1.5 Relative expression 1.5 B Relative expression qPCR validated F of PSA Relative expression A 1.5 1.0 0.5 0.0 Cancer Stroma 1.0 0.5 ctrl bic. gos. 0.0 ctrl bic. gos. Figure 1. Validation of DSection in silico Bayesian model prediction. In silico-predicted and qRT–PCR-assayed gene expression differences between non-treated control, bicalutamide and goserelin treatments of 10 genes: PSA (A); NPY (B); DHCR24 (C); TMEFF2 (D); TPD52 (E); GSTT1 (F); MAOA (G); MBOAT2 (H); NEDD4L (I); and TMPRSS2 (J). The expression of cancer tissue in control sample was normalized to 1. Ctrl, control; bic, bicalutamide; gos, goserelin. A Reduced expression 65 Bicalutamide 33 30 Goserelin B Increased expression 44 2 Bicalutamide 40 Goserelin Figure 2. Number of differentially expressed genes after bicalutamide or goserelin treatments (FC > 2; FDR < 0.01). (A) Genes with reduced expression compared to non-treated control group. (B) Genes with increased expression compared to non-treated control group. examine which of the genes with reduced expression are direct AR targets, we first utilized our previously published data on DHT-stimulated genes in an AR over-expressing LNCaP-based cell line model and in a VCaP cell line [23]. From 128 genes, 45 (35%) were > two-fold up-reguated in cell lines (see Supporting information Table S4). In addition, we used two Copyright  2012 Pathological Society of Great Britain and Ireland. Published by John Wiley & Sons, Ltd. www.pathsoc.org.uk independent chromatin immunoprecipitation (ChIP)sequencing (seq)-derived AR binding sites (ARBSs) data in an LNCaP-based cell line model [24,25]. ARBS data overlapped > 60% between the two studies (see Supporting information, Table S4). To confirm the active ARBSs, we utilized histone methylation data from two independent datasets [14,26]. The other dataset contained ChIP-seq data of the monomethylated H4K3, and the other dataset utilized the changes in H3K4me2 signal resulting nucleosome stabilization–destabilization (NSD) score for the nucleosomes. Both histone modifications have been shown to be associated with active enhancer areas [26,27]. Approximately half of the 128 genes contained a potentially active enhancer area with ARBS (see Supporting information, Table S4). Taken together, from the 128 genes that had > two-fold reduced expression after the endocrine treatment, 24 genes were induced by DHT stimulation in the cell line models, showed ARBSs close to the TSS according to the two independent studies or, alternatively, showed ARBS close to the TSS according to one of the studies and contained J Pathol (2012) www.thejournalofpathology.com Effect of endocrine therapy in prostate cancer Table 1. Genes whose expression was > two-fold reduced after bicalutamide and/or goserelin treatments, were > two-fold induced by DHT treatment, in cell culture models, had ARBS, and potentially active enhancer Gene name FKBP5 TMEFF2 FAM110B NEDD4L MME TMPRSS2 MBOAT2 CLDN8 TBC1D8 HOMER2 DHCR24 KLK4 RAB3B BRP44 C1orf116 TPD52 RDH10 KCNN2 LCP1 PMEPA1 KHDRBS3 LAMA3 ATAD2 FC ctrl vs bic FDR ctrl vs bic FC ctrl vs gos FDR ctrl vs gos 4.0 3.8 3.5 3.2 3.1 3.0 2.9 2.7 2.6 2.5 2.4 2.4 2.3 2.2 2.1 2.1 2.0 2.0 2.0 2.0 2.0 1.8 1.7 1.9E-05 0.008 0.0003 8.5E-05 1.9E-05 0.0007 5.8E-07 0.003 5.6E-07 5.7E-05 0.0003 0.006 0.0007 0.002 0.003 0.0001 0.006 0.001 0.0001 0.001 0.001 0.0009 0.002 2.8 7.2 2.1 0.5 4.8 0.8 1.7 1.7 1.4 1.3 1.8 1.2 1.2 1.4 1.5 0.4 2.2 2.4 1.1 0.8 2.6 2.5 2.1 0.0004 0.001 0.02 0.001 1.9E-06 0.2 0.0005 0.1 0.005 0.2 0.01 0.5 0.3 0.1 0.1 2.9E-05 0.007 0.0006 0.5 0.1 0.0002 5.5E-05 0.0003 For more details, see main text and Table S4 (see Supporting information). FC, fold change; FDR, false discovery rate; bic, bicalutamide; gos, goserelin. Genes that are indicated in bold type are > two-fold reduced after both endocrine treatments. potentially active enhancer (Table 1; see also Supporting information, Table S4). The differential expression of six of the 24 genes was confirmed with qRT–PCR (Figure 1). Next, we studied the expression of three of these directly androgen-regulated genes at the protein level. We immunostained the trial specimens with antibodies against TMEFF2, TPD52 and NEDD4L and observed abundant staining in all of the samples with the TPD52 and NEDD4L antibodies, which clearly illustrates high expression at the protein level in prostate cancer. However, significant differences between the treatment groups could not be assessed (Figure 3a, b). Instead, TMEFF2 staining was significantly weaker in both the bicalutamide- and goserelin-treated samples compared to control samples (p < 0.0001, χ2 test; Figure 3c, d). Further, we determined whether the genes with reduced expression after the endocrine treatment would be reactivated in CRPC. We measured the expression levels of four selected genes, TMEFF2, DHCR24, TPD52 and NEDD4L, in an independent set of benign prostate hyperplasia (BPH), previously untreated prostate cancers (prostatectomy specimens) and CRPC samples, using qRT–PCR (Figure 3e–h). We found that DHCR24, TPD52 and NEDD4L were significantly over-expressed in the prostate cancer samples when compared to the BPH samples and that TMEFF2 had a trend towards significance (p < 0.0001). The expression of TPD52 and NEDD4L was also significantly increased in the CRPC samples compared to the BPH samples (p < 0.001 and < 0.05, respectively), and DHCR24 and TMEFF2 had a trend towards significance. Copyright  2012 Pathological Society of Great Britain and Ireland. Published by John Wiley & Sons, Ltd. www.pathsoc.org.uk Finally, we assayed TMPRSS2-ERG fusion in the trial samples by using FISH and IHC (see Supporting information, Figure S1). Of the 25 samples containing enough cancerous area for the assay, 15 samples (60%) were positive for the fusion gene (see Supporting information, Table S6). Due to the small number of cases, we combined the bicalutamide and goserelin groups into one endocrine-treated sample group. First, we evaluated role of ERG and AR in the control and endocrine treated groups. In the control group, we detected substantially higher expression levels of 869 genes in the fusion-positive (F+ ) cases compared to the fusion-negative (F− ) cases (Figure 4a; see also Supporting information, Table S7). In contrast, the treatment reduced the expression of 601 genes in the F+ cases but only 69 genes in the F− cases (p < 0.0001, χ2 test, Figure 4b; see also Supporting information, Table S8). Interestingly, one-half (430) of the genes that were up-regulated in the F+ cases versus F – cases in the control group were common to those that were down-regulated after the treatment in the F+ cases (Figure 4c). To assess whether the differentially expressed genes are direct targets of ERG and AR, we utilized ChIP-seq data from the TMPRSS2–ERG fusion-positive VCaP cell line published by Yu et al [14]. We re-analysed the data and determined the genes closest to the ERG binding sites (ERGBSs) and ARBSs. On average, 68% and 25% of all genes in the genome possess an ERGBS and ARBS, respectively, and, as previously shown [14], > 90% of the genes with an ARBS are also targets of ERG. The most significant enrichment in ERGBSs occurred in a group of 869 genes that J Pathol (2012) www.thejournalofpathology.com S Lehmusvaara et al. 20 ct rl s go bi c ct rl 3 F G 60 TPD52/b-actin 40 20 30 20 10 PC C PC H PC R C * *** 60 PC H BP PC R C PC H BP H 40 0 0 0 *** *** BP DHCR24/b-actin 100 50 *** R s 2 200 0-1 2 3 s 40 100 80 60 40 20 0 go 60 go bi c 0-1 300 TMEFF2/b-actin 80 0 E NEDD4L/b-actin TMEFF2 % of samples % of samples 100 80 60 40 20 0 D C NEDD4L 100 bi c B TPD52 ct rl % of samples A 40 20 C R PC PC BP H 0 Figure 3. Immunohistochemical staining of the trial material with antibody against TPD52 (A), NEDD4L (B) and TMEFF2 (C); bic, bicalutamide; gos, goserelin. (D) Representative images of TMEFF2 staining; 0–1, no or low staining; 2, moderate staining; 3, high-intensity staining. Staining intensity was measured only from cancerous areas, which were confirmed with AMACR, p63 and keratin HMW triple staining. Differences between treatment groups (non-treated control, bicalutamide and goserelin) were significant with TMEFF2 antibody (p < 0.0001, χ2 test) but not with TPD52 of NEDD4L. (E–I) Gene expression levels of four candidate genes, TMEFF2 (E), DHCR24 (F), TPD52 (G) and NEDD4L (H), in benign prostate hyperplasia (BPH), hormone-naive prostate cancer (PC) and castration-resistant prostate cancer (CRPC), measured with qRT–PCR; ∗ p < 0.05, ∗∗∗ p < 0.001, one-way ANOVA with Bonferroni’s multiple comparison test. showed increased gene expression in the F+ cases compared to the F− cases in the control group (85% harboured an ERGBS; p = 3.3e-25, Fishers’ exact test; Figure 4a). Similarly, in a group of the 601 genes that showed decreased expression in the F+ cases after the treatment 86% harboured an ERGBS (p = 4.4e-21; Figure 4b). In addition, the same groups of genes had also significant ARBSs enrichment: 36% of the 869 genes and 40% of the 601 genes harboured an ARBS, (p = 1.8e-25 and p = 6.2e-27, respectively; Figure 4a, b). Copyright  2012 Pathological Society of Great Britain and Ireland. Published by John Wiley & Sons, Ltd. www.pathsoc.org.uk Finally, we performed an ontology analysis for the most interesting groups of genes. Those were the group of 869 genes with increased expression in the F+ cases compared to the F− cases in a control group and the group of 601 genes with decreased expression in the F+ cases after the endocrine treatment (Figure 4a, b). Same groups also had the most significant enrichment of ERGBS and ARBS, as mentioned above. Interestingly, we detected that the significantly enriched ontologies in both groups were strongly related to proliferation, eg ‘cell cycle’, ‘M phase’ and ‘mitosis’ J Pathol (2012) www.thejournalofpathology.com Effect of endocrine therapy in prostate cancer F+ vs F− A Ctrl D treat 869 genes (F+ vs F− in ctrl) 869 116 76% ERGBS 38% ARBS*** 85% ERGBS*** 36% ARBS*** 67 239 68% ERGBS 32% ARBS* 76% ERGBS** 38% ARBS Treat vs ctrl B cell cycle cell cycle process cell cycle phase mitotic cell cycle M phase M phase of mitotic cell cycle mitosis nuclear division organelle fission Organelle organization chromosome segregation cell division 10−0 10−2 E 10−4 10−6 10−8 Adj.P-val F− F+ 601 genes (treat vs ctrl in F+) 66 260 67% ERGBS 38% ARBS*** 72% ERGBS 36% ARBS*** 66 601 68% ERGBS 34% ARBS** 86% ERGBS*** 40% ARBS*** cell cycle cell cycle process cell cycle phase mitotic cell cycle M phase M phase of mitotic cell cycle mitosis nuclear division organelle fission chromosome segregation 10−0 10−2 10−4 10−6 10−8 Adj.P-val C F+ vs F− ctrl Treat vs ctrl F+ 439 85% ERGBS*** 36% ARBS*** 430 171 86% ERGBS*** 40% ARBS*** Figure 4. Number of differentially expressed genes and their ontologies according to TMPRSS2–ERG fusion and endocrine treatment status. (A) Genes with increased and decreased expression in the fusion-positive (F+ ) versus fusion-negative (F− ) cases in control (crtl) and endocrine-treated (treat) groups. (B) Genes with increased and decreased expression after the endocrine treatment in the F+ and F− cases. (C) Half (430) of those genes with increased expression in the F+ cases in the control group (left circle, altogether 869 genes) had reduced expression levels after the endocrine treatment (right circle, altogether 601 genes). Fold change > 1.6. Asterisks represent the statistical significance of AR and ERG binding site enrichment: ∗ p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001, Fisher’s exact test. Binding frequency in the whole genome was 68% for ERG and 25% for AR. Binding sites were retrieved from the publicly available VCaP cell line ChIP-seq datasets [14]. (D–E) Significantly enriched gene ontologies of biological processes. Ontologies were determined from the group of 869 genes with increased expression in F+ cases in the control group (D) and from the group of 601 genes with reduced expression after endocrine treatments in F+ cases (E), with the limits: > seven genes/ontology group and adjusted p value (adj.P-val.) < 0.0001; calculated with Fisher’s exact test and Benjamini–Hochberg multiple test. (Figure 4d, e; cut-off > 7 genes and an adjusted p < 0.0001). Discussion In this study, we took advantage of rare clinical specimens from neoadjuvant endocrine-treated patients and studied the differences between the two most commonly used endocrine treatments, the GnRH agonist goserelin and the anti-androgen bicalutamide. Surprisingly, the two endocrine treatments appeared to regulate different genes. A chronic GnRH agonist administration leads to down-regulation of its receptor Copyright  2012 Pathological Society of Great Britain and Ireland. Published by John Wiley & Sons, Ltd. www.pathsoc.org.uk and subsequently to reduction of androgen secretion to castrate levels. The anti-androgens, in contrast, bind directly to the AR ligand binding domain and compete with DHT or testosterone binding. However, both treatments aim to inactivate the AR signalling pathway. Despite similar clinical outcome, the findings here indicate that the molecular responses induced by GnRH agonists and anti-androgens are at least partially different. Mostaghel et al [28] previously measured gene expression levels after medical castration, using the GnRH antagonist acyline in healthy volunteers. In addition, Holzbeierlein et al [29] characterized the gene J Pathol (2012) www.thejournalofpathology.com S Lehmusvaara et al. expression changes after combined androgen blockade with goserelin and flutamide for 3 months in patients with localized prostate cancer. When we compared the most differentially expressed genes, few genes were common to this study and Mostaghel’s and Holzbeierlein’s studies. The lack of common differentially expressed genes could be explained by the use of different drugs or combinations of drugs in the three studies. In addition, different microarray platforms, normalization methods, choices of arbitrary threshold values and statistical analyses may contribute to the differences observed in these studies. From 128 genes with > two-fold reduced expression, only 24 had ARBS closest to their TSS and were induced by DHT > two-fold in AR-dependent cell line models. These 24 genes are most probably directly androgen-regulated, while the reduced expression of other 104 genes may be due to the secondary effects. It is noteworthy that the samples from the trial have been collected after 3 months of therapy, which may explain a large proportion of secondary effects. This time point also excludes early androgen-responsive genes and reveals only long-term AR targets. Furthermore, the ARBSs, H3K4 methylation data and gene expression data following DHT stimulation are derived from prostate cancer cell line models. Thus, it is possible that some of these 104 genes could be direct AR targets in prostate tissue. H3K4 momomethylation has been shown to be associated with transcription binding at enhancers and demethylation with both TSS and enhancers [27,30]. In this study, we utilized the histone methylation data of two independent studies to identify potential enhancer areas [14,26]. Although potential enhancer areas were found, they do not necessarily regulate the expression of the closest gene, and therefore the possibility of false positives and false negatives is present. Only reporter and chromatin conformation assays could reliably identify the correct target genes for enhancer areas. Thus, to reliably detect AR target genes, we combined the methylation data with ARBSand DHT-induced expression data. Because endocrine treatment is not curative and the disease eventually relapses, we investigated the genes that are down-regulated after the endocrine treatment and reactivated in the castration-resistant stage. Such genes could be potential biomarkers for response to hormonal therapy. We detected a trend of increased TMEFF2 and DHCR24 expression levels and a statistically significant over-expression of TPD52 and NEDD4L in CRPC cases compared to cases of BPH. The expression levels of TPD52, DHCR24 and TMEFF2 have previously been shown to be androgen-regulated [31–33]. In addition, NEDD4L has been implicated in AR signalling [34,35]. Moreover, the expression levels of TPD52 and TMEFF2 have been demonstrated to be increased in prostate cancer, especially in CRPC [32,36,37]. We also assessed the influence of TMPRSS2–ERG fusion on gene expression in both untreated and Copyright  2012 Pathological Society of Great Britain and Ireland. Published by John Wiley & Sons, Ltd. www.pathsoc.org.uk endocrine-treated cases. It appears that ERG sensitizes cells to the endocrine therapy, because 8.7-times more genes were down-regulated after endocrine treatment in the F+ cases compared to the F− cases (601 versus 69 genes). Both ERGBSs and ARBSs were significantly enriched in the regions near the down-regulated genes. Notably, the majority of the down-regulated genes are the same genes that showed increased expression in the F+ cases compared to the F− cases in the control group. Therefore, the endocrine treatment mainly affected the genes that were highly expressed in the F+ cases and diminished the differences between the F+ and F− tumours. It has been shown that ERG expression is increased by androgens in the VCaP cell line [14]. This can also be seen in our data, with increased expression of putative ERG target genes in the F+ cases of the hormone-naı̈ve control group. In our data, endocrine treatment reduced the expression of 601 genes in the F+ cases but only 69 genes in the F− cases. These 601 genes are most probably direct targets of ERG, because the androgen-dependent nature of TMPRSS2 expression renders the genes under control of fused ERG. Indeed, 86% of these genes had ERGBS closest to their TSS. Interestingly, the ontology analysis revealed that these genes are strongly involved in cell cycle and mitosis. Thus, it seems that the fusion brings many proliferation-associated genes under androgen regulation. Previously it has been shown that ETS transcription factor binding sites were often close to ARBSs [38]. Recently, Yu et al [14] also reported that AR and ERG chromatin binding profiles overlap. Additionally, Yu et al reported that ERG may reduce AR activity and, in the presence of androgens, the knock-down of ERG enhances the expression of AR-regulated genes. However, our data suggest more synergistic roles for AR and ERG; co-activity of these transcription factors enhances the expression of their target genes, as in the F+ cases in the control group, and reduction of androgens in the treatment group reduces the expression of ERG target genes in the F+ cases. Several other studies have also identified differences in the gene expression between TMPRSS2–ERG fusion-positive and -negative prostate cancers [39–41]. Similar to our study, they have identified more genes with increased expression than reduced expression in the fusion-positive compared to the fusionnegative cases [39,40]. Approximately 30–50% of genes with increased expression in the fusion-positive cases reported in these studies were also increased > 1.3-fold in this study. Sample heterogeneity is a major problem in microarray studies, especially in diseases such as prostate cancer, where the cancerous areas are small and often surrounded by normal cells. In this study, we used the in silico probabilistic analysis tool, DSection, to artificially isolate the cancerous areas. We detected a 71% overlap when we compared the gene expression profile of the DSection model to the gene expression J Pathol (2012) www.thejournalofpathology.com Effect of endocrine therapy in prostate cancer profile of samples with a high percentage of cancerous area (high-cancer). Interestingly, TPD52, one of the most differentially expressed genes, was part of the group of genes that DSection predicted to be the most differentially expressed. However, TPD52 expression was not different, according to the expression profile of the samples with a high cancer compartment. The qRT–PCR from the microdissected samples confirmed the differential expression of TPD52. This example shows that the DSection prediction model can produce reliable expression data that overcome the heterogeneity of the prostate cancer tissue specimens. In conclusion, we utilized this rare clinical material from neoadjuvant-treated PC patients and found a clear difference in the gene expression levels induced by an anti-androgen and a GnRH agonist. This indicates different cellular consequences of these two forms of androgen deprivation. In addition, we showed that the endocrine treatments induce different gene expression changes in PC, depending on TMPRSS2–ERG fusion. Many of the treatment-responsive genes in the fusion-positive cases were related to proliferation. The weakness of the study was the low number of cases. Thus, it is vital that the findings should be validated in larger samples, although unlikely in similar trial settings. Acknowledgment We wish to thank Ms Mariitta Vakkuri, Ms Päivi Martikainen, Ms Anni Järvinen, Ms Salla Kolmihaara and Ms Anne Luuri for their skilled technical assistance. Grant support was received from the Academy of Finland, the Cancer Society of Finland, the Reino Lahtikari Foundation, the Sigrid Juselius Foundation, competitive research funding from Pirkanmaa Hospital District and a non-restricted grant from AstraZeneca. Author contributions AK, PM and TT guided and carried out the clinical trial; JI established dual a-colour three-antibody immunostaining; VT developed the virtual microscopy system to facilitate IHC analysis; AU and KW carried out the cell line experiments; TE and HL developed the DSection analysis and carried out bioinformatic analysis; JS analysed AR and ERG binding site data; AL analysed the histone methylation data; TV planned the experiments and guided the work; and SL analysed data and carried out experiments. The manuscript was written by SL and TV with the assistance of all other authors. References 1. Labrie F, Belanger A, Luu-The V, et al . Gonadotropin-releasing hormone agonists in the treatment of prostate cancer. Endocr Rev 2005; 26: 361–379. Copyright  2012 Pathological Society of Great Britain and Ireland. Published by John Wiley & Sons, Ltd. www.pathsoc.org.uk 2. Huggins C, Hodges CV Studies on prostatic cancer. I. The effect of castration, of estrogen and of androgen injection on serum phosphatases in metastatic carcinoma of the prostate. Cancer Res 1941; 1: 293–297. 3. Auclair C, Kelly PA, Labrie F, et al . Inhibition of testicular luteinizing hormone receptor level by treatment with a potent luteinizing hormone-releasing hormone agonist of human chorionic gonadotropin. Biochem Biophys Res Commun 1977; 76: 855–862. 4. Seidenfeld J, Samson DJ, Hasselblad V, et al . Single-therapy androgen suppression in men with advanced prostate cancer: a systematic review and meta-analysis. Ann Intern Med 2000; 132: 566–577. 5. Parmar H, Phillips RH, Lightman SL, et al . Randomised controlled study of orchidectomy vs long-acting D-Trp-6-LHRH microcapsules in advanced prostatic carcinoma. Lancet 1985; 2: 1201–1205. 6. Moreau JP, Delavault P, Blumberg J. Luteinizing hormonereleasing hormone agonists in the treatment of prostate cancer: a review of their discovery, development, and place in therapy. Clin Ther 2006; 28: 1485–1508. 7. Tyrrell CJ, Kaisary AV, Iversen P, et al . A randomised comparison of ‘Casodex’ (bicalutamide) 150 mg monotherapy versus castration in the treatment of metastatic and locally advanced prostate cancer. Eur Urol 1998; 33: 447–456. 8. Iversen P, Tyrrell CJ, Kaisary AV, et al . Bicalutamide monotherapy compared with castration in patients with nonmetastatic locally advanced prostate cancer: 6.3 years of follow-up. J Urol 2000; 164: 1579–1582. 9. Heidenreich A, Aus G, Bolla M, et al . EAU guidelines on prostate cancer. Eur Urol 2008; 53: 68–80. 10. Anderson J. The role of antiandrogen monotherapy in the treatment of prostate cancer. BJU Int 2003; 91: 455–461. 11. Palmberg C, Koivisto P, Visakorpi T, et al . PSA decline is an independent prognostic marker in hormonally treated prostate cancer. Eur Urol 1999; 36: 191–196. 12. Attard G, Richards J, de Bono JS. New strategies in metastatic prostate cancer: targeting the androgen receptor signaling pathway. Clin Cancer Res 2011; 17: 1649–1657. 13. Tomlins SA, Rhodes DR, Perner S, et al . Recurrent fusion of TMPRSS2 and ETS transcription factor genes in prostate cancer. Science 2005; 310: 644–648. 14. Yu J, Yu J, Mani RS, et al . An integrated network of androgen receptor, polycomb, and TMPRSS2-ERG gene fusions in prostate cancer progression. Cancer Cell 2010; 17: 443–454. 15. Halabi S, Vogelzang NJ, Ou SS, et al . Progression-free survival as a predictor of overall survival in men with castrate-resistant prostate cancer. J Clin Oncol 2009; 27: 2766–2771. 16. Scher HI, Beer TM, Higano CS, et al . Antitumour activity of MDV3100 in castration-resistant prostate cancer: a phase 1–2 study. Lancet 2010; 375: 1437–1446. 17. de Bono JS, Logothetis CJ, Molina A, et al . Abiraterone and increased survival in metastatic prostate cancer. N Engl J Med 2011; 364: 1995–2005. 18. Tuominen VJ, Isola J. The application of JPEG2000 in virtual microscopy. J Digit Imaging 2009; 22: 250–258. 19. Erkkilä T, Lehmusvaara S, Ruusuvuori P, et al . Probabilistic analysis of gene expression measurements from heterogeneous tissues. Bioinformatics 2010; 26: 2571–2577. 20. Tolonen TT, Kujala PM, Laurila M, et al . Routine dual-color immunostaining with a 3-antibody cocktail improves the detection of small cancers in prostate needle biopsies. Hum Pathol 2011; 42: 1635–1642. 21. Saramäki OR, Harjula AE, Martikainen PM, et al . TMPRSS2:ERG fusion identifies a subgroup of prostate cancers with a favorable prognosis. Clin Cancer Res 2008; 14: 3395–3400. J Pathol (2012) www.thejournalofpathology.com S Lehmusvaara et al. 22. Zhang B, Kirov S, Snoddy J. WebGestalt: an integrated system for exploring gene sets in various biological contexts. Nucleic Acids Res 2005; 33: W741–748. 23. Waltering KK, Helenius MA, Sahu B, et al . Increased expression of androgen receptor sensitizes prostate cancer cells to low levels of androgens. Cancer Res 2009; 69: 8141–8149. 24. Urbanucci A, Sahu B, Seppälä J, et al . Overexpression of androgen receptor enhances the binding of the receptor to the chromatin in prostate cancer. Oncogene 2011; [Epub ahead of print doi: 10.1038/onc.2011.401.] 25. Massie CE, Lynch A, Ramos-Montoya A, et al . The androgen receptor fuels prostate cancer by regulating central metabolism and biosynthesis. EMBO J 2011; 30: 2719–2733. 26. He HH, Meyer CA, Shin H, et al . Nucleosome dynamics define transcriptional enhancers. Nat Genet 2010; 42: 343–347. 27. Barski A, Cuddapah S, Cui K, et al . High-resolution profiling of histone methylations in the human genome. Cell 2007; 129: 823–837. 28. Mostaghel EA, Page ST, Lin DW, et al . Intraprostatic androgens and androgen-regulated gene expression persist after testosterone suppression: therapeutic implications for castration-resistant prostate cancer. Cancer Res 2007; 67: 5033–5041. 29. Holzbeierlein J, Lal P, LaTulippe E, et al . Gene expression analysis of human prostate carcinoma during hormonal therapy identifies androgen-responsive genes and mechanisms of therapy resistance. Am J Pathol 2004; 164: 217–227. 30. Heintzman ND, Stuart RK, Hon G, et al . Distinct and predictive chromatin signatures of transcriptional promoters and enhancers in the human genome. Nat Genet 2007; 39: 311–318. 31. Gery S, Sawyers CL, Agus DB, et al . TMEFF2 is an androgenregulated gene exhibiting antiproliferative effects in prostate cancer cells. Oncogene 2002; 21: 4739–4746. 32. Rubin MA, Varambally S, Beroukhim R, et al . Overexpression, amplification, and androgen regulation of TPD52 in prostate cancer. Cancer Res 2004; 64: 3814–3822. 33. Nelson PS, Clegg N, Arnold H, et al . The program of androgenresponsive genes in neoplastic prostate epithelium. Proc Natl Acad Sci USA 2002; 99: 11890–11895. 34. Shanmugam I, Cheng G, Terranova PF, et al . Serum/ glucocorticoid-induced protein kinase-1 facilitates androgen receptor-dependent cell survival. Cell Death Differ 2007; 14: 2085–2094. 35. Hayes SA, Zarnegar M, Sharma M, et al . SMAD3 represses androgen receptor-mediated transcription. Cancer Res 2001; 61: 2112–2118. 36. Glynne-Jones E, Harper ME, Seery LT, et al . TENB2, a proteoglycan identified in prostate cancer that is associated with disease progression and androgen independence. Int J Cancer 2001; 94: 178–184. 37. Wang R, Xu J, Mabjeesh N, et al . PrLZ is expressed in normal prostate development and in human prostate cancer progression. Clin Cancer Res 2007; 13: 6040–6048. 38. Massie CE, Adryan B, Barbosa-Morais NL, et al . New androgen receptor genomic targets show an interaction with the ETS1 transcription factor. EMBO Rep 2007; 8: 871–878. 39. Setlur SR, Mertz KD, Hoshida Y, et al . Estrogen-dependent signaling in a molecularly distinct subclass of aggressive prostate cancer. J Natl Cancer Inst 2008; 100: 815–825. 40. Barwick BG, Abramovitz M, Kodani M, et al . Prostate cancer genes associated with TMPRSS2–ERG gene fusion and prognostic of biochemical recurrence in multiple cohorts. Br J Cancer 2010; 102: 570–576. 41. Klezovitch O, Risk M, Coleman I, et al . A causal role for ERG in neoplastic transformation of prostate epithelium. Proc Natl Acad Sci USA 2008; 105: 2105–2110. SUPPORTING INFORMATION ON THE INTERNET The following supporting information may be found in the online version of this article: Figure S1. Example images from FISH and IHC assays of TMPRSS2 –ERG fusion. Table S1. More detailed information of the two clinical sample sets that were used in this study. Table S2. Pathological characterization of patients involved in the endocrine trial. Table S3. Sequences of qRT–PCR primers that were used to validate the bioinformatic analyses. Table S4. More detailed list of genes with reduced expression after bicalutamide and/or goserelin treatments. Their expression after androgen induction in prostate cancer cell lines is indicated, as well as the AR binding sites close to their TSS. Table S5. List of genes with increased expression after bicalutamide and/or goserelin treatments. Table S6. Distribution of the TMPRSS2–ERG fusion-positive and -negative samples in the treatment and control groups. Table S7. List of genes with increased expression in the F+ compared to the F− cases in the control and treatment groups. AR and ERG binding sites in VCaP cell line are also indicated for each gene. Table S8. List of genes with decreased expression after the endocrine treatment in the F+ and F− cases. Similarly to Table S6, AR and ERG binding sites in the VCaP cell line are indicated for these genes. Copyright  2012 Pathological Society of Great Britain and Ireland. Published by John Wiley & Sons, Ltd. www.pathsoc.org.uk J Pathol (2012) www.thejournalofpathology.com