Functional Epigenomics Identifies Genes Frequently Silenced in Prostate Cancer Free

Prostate cancer is the second leading cause of cancer deaths in men in the U.S. For the year 2004, it has been estimated that 230,000 new cases of prostate cancer will be diagnosed and 30,000 deaths related to prostate cancer will occur (1). The molecular basis of prostate cancer initiation and progression is still poorly understood, particularly due to the heterogeneity of primary tumors.

Inactivation of tumor-suppressive genes by either genetic or epigenetic mechanisms contributes to cancer formation. Epigenetic inactivation of genes in cancer cells is largely based on transcriptional silencing by aberrant CpG methylation of CpG-rich promoter regions (2, 3). Epigenetic silencing has been reported for a number of genes in prostate cancer. For example, aberrant CpG methylation in prostate cancer was found for GSTP1 in ∼90%, for RASSF1A in ∼63%, and for RARβ2 in ∼79% of the analyzed samples (4–6). Other examples of genes frequently silenced in prostate cancer are APC, MGMT, and MDR1 (7–9). Inhibition of DNA-methyltransferase activity by 5-aza-2′deoxycytidine leads to reversion of CpG-methylation and re-expression of silenced genes. As transcriptional silencing mediated by CpG methylation involves the recruitment of histone deacetylase activity (10), the effect of 5-aza-2′deoxycytidine is augmented by the histone deacetylase–inhibitor trichostatin A. Silenced genes which are induced by a combined treatment with 5-aza-2′deoxycytidine and trichostatin A can be identified by microarray analysis in a genome-wide manner. The identification of genes specifically silenced by CpG methylation may contribute to a better understanding of the etiology of prostate cancer. Furthermore, CpG methylation promises to be useful as a diagnostic tool because the detection and quantification of specific CpG methylation patterns of DNA in biopsies or body fluids is feasible (11). The use of a panel of CpG methylation markers in combination with standard histologic review of needle biopsies was shown to increase the sensitivity of prostate cancer diagnosis (12).

Moreover, aberrant CpG methylation may potentially be used as a prognostic marker to identify prostate cancer, which will progress to symptomatic or metastatic disease. Here, we describe the results of a microarray-based, genome-wide screen for genes epigenetically silenced by CpG methylation in prostate cancer.

Cell culture and tissue samples. The cell lines Du-145, LNCaP, PC3, PPC1, and TSU-Pr1 were maintained in RPMI 1640 supplemented with 10% fetal bovine serum (FBS) and antibiotics (Invitrogen, Karlsruhe, Germany). The TSU-Pr1 cell line was originally thought to be derived from a metastatic prostate carcinoma, but was later shown to be a variant of the bladder carcinoma cell line T24 (13). The cell line LAPC-4 was kept in RPMI 1640 in the presence of 20% FBS. Human benign prostate hyperplasia (BPH) cells immortalized with SV40 large T antigen (BPH1) were obtained from the German Collection of Microorganisms and Cell Cultures and cultured in RPMI 1640 supplemented with 20% FBS, 20 ng/mL testosterone, 50 μg/mL transferrin, 50 ng/mL sodium selenite, 50 μg/mL insulin, and a mixture of trace elements (Invitrogen). Human prostate epithelial cells (Cambrex Bio Science, Walkersville, MD) were cultured in prostate epithelial cell growth medium (PrEGM, Cambrex Bio Science) on dishes coated with collagen type I (BioCoat, BD Falcon, Heidelberg, Germany). HCT116 colon cancer cells were maintained in 10% FBS McCoy's medium (Invitrogen). Archival samples of primary prostate carcinoma (Gleason Sum 5-10) and cancer-free samples of prostate were obtained from the Institute of Pathology, Ludwig-Maximilians University, Munich, and represent consecutive cases from the year 2001. The samples had been fixed in formalin and embedded in paraffin. All patients had undergone surgery at the same institution. In all cases, two board-certified pathologists agreed on the diagnosis of prostate cancer.

Laser-assisted tissue microdissection. Archival specimens of primary prostate cancer and tumor-free prostate tissue were deparaffinized in xylene and briefly stained with H&E. Microdissection and laser-pressure catapulting was done using a MicroBeam system (PALM, Bernried, Germany). Material obtained from two to three parallel sections (∼10³ cells) was pooled for subsequent DNA isolation.

Microarray analysis. LNCaP, PC3, and Du-145 cells were seeded at low density 24 hours before treatment with 1 μmol/L 5-aza-2′deoxycytidine (Sigma, St. Munich, Germany) and/or 300 nmol/L trichostatin A (Sigma). Total RNA was isolated from cell lines using the RNAgent kit (Promega, Madison, WI). Biotin-labeled cRNA (15 μg) was hybridized to U133A oligonucleotide arrays, analyzed with a GeneChip Scanner 3000 and differential gene expression identified with the Microarray Suite 4.0 software (Affymetrix, Santa Clara, CA).

RT-PCR analysis. Total RNA (5 μg) was reverse-transcribed using the SuperScript kit (Invitrogen). Quantitative PCR of EF1α was used to normalize the employed cDNAs (data not shown). Two units Platinum Taq polymerase (Invitrogen) were used per reaction. Primer sequences and RT-PCR conditions are provided in Supplemental Table 3.

Isolation and bisulfite treatment of genomic DNA. Genomic DNA was isolated by overnight incubation in 100 μg/mL proteinase K (Sigma) and 0.1% SDS (Sigma) at 55°C with subsequent phenol/chloroform extraction and isopropanol precipitation. Herring sperm DNA (1 μg; Promega) was added as a carrier to DNA obtained from laser-microdissected tissue. DNA (2 μg) was denatured in 0.2 mol/L NaOH for 10 minutes at 37°C in 50 μL total volume. 30 μL of 10 mmol/L hydroquinone (Sigma) and 520 μL of 3.5 mol/L sodium bisulfite (pH 5.0; Sigma) was added. After 16 hours at 50°C, the DNA was purified and incubated in 0.3 mol/L NaOH for 5 minutes at room temperature. After ethanol precipitation, the DNA was dissolved in 40 μL Tris-EDTA. For methylation-specific PCR, 2 μL or amplification of templates for bisulfite-sequencing 5 μL of this solution were used.

Genomic bisulfite sequencing. Bisulfite-treated genomic DNA was used as a template to amplify fragments of 400 to 1,000 bp with a high CpG-content around the transcription start site using oligonucleotides listed in Supplemental Table 1. After 5 minutes' incubation at 95°C, 39 to 41 PCR-cycles were done for 20 seconds at 95°C, 30 seconds at annealing temperature, and 60 to 90 seconds at 72°C. Five units of Platinum Taq polymerase (Invitrogen) were used per 100 μL reaction. Gel-purified PCR-products were subcloned in a TOPO-TA vector (Invitrogen). For each gene, at least six individual clones were sequenced on both strands using M13 primers and BigDye terminator, and analyzed on a 3700 capillary sequencer (Applera, Darmstadt, Germany).

Methylation-specific PCR-analysis. Methylation-specific PCR was done in a total volume of 20 μL using 3 units Platinum Taq-polymerase (Invitrogen) per reaction and oligonucleotides listed in Supplemental Table 2 (14). After denaturation at 95°C for 5 minutes, 40 PCR-cycles were done when genomic DNA obtained from cell lines and 45 cycles when microdissected DNA was used as a template. Amplified fragments were separated by 8% polyacrylamide gel electrophoresis and detected by ethidium bromide staining.

Immunohistochemistry. Six-micrometer sections were deparaffinized in xylene, rehydrated in a decreasing ethanol series and boiled for 30 minutes in ProTaq IV buffer (Biocyc, Luckenwalde, Germany) for antigen retrieval. Anti-SFRP1 antibodies (Santa Cruz, Heidelberg, Germany) were used with Vectastain Elite avidin-biotin complex method kit (Vector Laboratories, Burlingame, CA). After counterstaining with hematoxylin, the images were acquired on an Axiovert 200 M microscope (Carl Zeiss, Oberkochen, Germany) coupled to a DXC-390P CCD camera (Sony, Tokyo, Japan) using a PALMRobo V2.1.1 software (PALM).

Protein detection by immunofluorescence. Cells were cultured on CELLocate slides (Eppendorf, Hamburg, Germany) and fixed in 3.7% paraformaldehyde solution for 20 minutes, permeabilized in 0.2% Triton X (Sigma) in PBS for 15 minutes, and blocked in FBS for 30 minutes. Primary antibodies specific for β-catenin (clone 19, Transduction Laboratories, Lexington, KY) diluted in PBS with 10% FBS and 0.05% Tween 20 were added for 1 hour and detected with a Cy3-conjugated donkey anti-mouse antibody (Jackson ImmunoResarch Laboratories, West Grove, PA). The images were acquired using a fluorescent Axiovert 200 M microscope (Carl Zeiss) and Metamorph software (Universal Imaging, Downingtown, PA).

Western blot analysis. Western blotting analysis was done as described previously (15). Antibodies used were directed against α-tubulin (Santa Cruz), vesicular stomatitis virus (Sigma), and phospho-ERK1/2 and ERK1/2 (Cell Signaling Technology, Frankfurt, Germany). Secondary horseradish peroxidase–conjugated anti-mouse and anti-rabbit antibodies (Promega) were used at a dilution of 1:5,000.

Transfection and luciferase reporter assay. The plasmids pGL3-OT, pGL3-OF, and pcDNA3.1-His-WNT1 have been described previously (16). LNCaP, PC3, Du-145 and HCT116 cells were plated at medium density in 12-well plates 24 hours before transfection. Three constructs were cotransfected using Lipofectamine 2000 reagent (Invitrogen): (a) 0.5 μg of pGL3-OT or pGL3-OF; (b) 0.5 μg of pcDNA3.1-His-WNT1 or pcDNA3.1-His-A (Invitrogen), (c) 50 ng of pCMV-β-gal (Promega). Transfections were done in triplicate. After 36 hours, cells were assayed for luciferase activity using a Luciferase Assay System kit (Promega) and for β-galactosidase activity with a Galacto-Light kit (Tropix, Bedford, MA) on a MicroLumatPlus LB96V luminometer (EG&G Berthold, Bad Wildbad, Germany).

Generation and analysis of transgenic cell lines. For stable expression of Dickkopf 3 (DKK3) or SFRP1, the retroviral vector pLXSN (BD Clontech, Heidelberg, Germany), which was modified by insertion of an IRES-EGFP fragment derived from the plasmid pIRES-EGFP2 (BD Clontech), was used. PC3 cells were retrovirally infected using pLXSN-IRES-EGFP2, pL-DKK3vsv-IRES-EGFP2, or pL-SFRP1-IRES-EGFP2. Seventy-two hours after infection, green fluorescent protein–positive cells were sorted by fluorescence-activated cell sorting and expanded. For the assessment of colony formation, cells were seeded at low density in six-well plates (2,000 cells per well) and grown for 10 days. Cells were fixed in 1% formaldehyde and stained with crystal violet. Apoptosis was assessed by propidium iodide staining and flow cytometry as described previously (15).

Epigenetic analysis of prostate cancer cell lines. Conditions for optimal transcriptional induction after combined treatment with 5-aza-2′deoxycytidine and trichostatin A were determined by methylation-specific PCR of the promoter regions of GSTP1, RASSF1A, and p16 in several prostate cancer cell lines. Thereby, the metastatic prostate cancer cell lines PC3, LNCaP, and DU-145 were identified as most suitable for this analysis (data not shown). For re-expression of silenced genes, the cell lines were exposed to 1 μmol/L 5-aza-2′deoxycytidine for 72 hours and to 300 nmol/L trichostatin A for the final 24 hours. Treatment with 300 nmol/L trichostatin A for 24 hours served as a control. Total RNA was isolated, converted to biotinylated cRNA and hybridized to oligonucleotide arrays representing ∼18,400 individual transcripts. Each microarray analysis was done in duplicate. Efficient demethylation of CpG-dinucleotides was confirmed by methylation-specific PCR analysis of selected promoters (Fig. 1A). The microarray analysis revealed that several hundred transcripts were induced in the cells exposed to 5-aza-2′deoxycytidine and trichostatin A when compared with cells treated with trichostatin A alone (data not shown). GSTP1, a gene previously shown to be silenced by CpG methylation in prostate cancer (4), was induced 1.87-fold in LNCaP cells. Therefore, an induction of at least 1.8-fold was chosen as the minimal requirement for further analysis of candidate genes. The induction of GSTP1 was confirmed by RT-PCR (Fig. 1B) and Northern blot analysis (data not shown). Exemplary confirmations of results obtained by microarray analysis were done by RT-PCR for 10 different genes (Fig. 1B). In addition, the expression of RASSF1A and p16, which are known to be induced after demethylation, was analyzed by RT-PCR (Fig. 1B). We detected the re-expression of several imprinted genes (e.g., IGF2) and of genes silenced by CpG methylation in somatic tissues (e.g., MAGE). Furthermore, IFN-responsive genes, which have been previously reported to be activated by 5-aza-2′deoxycytidine treatment (17), were found to be induced (data not shown). Genes belonging to these three classes were excluded from further analysis. Re-expressed genes with known putatively tumor-suppressive functions (e.g., involvement in DNA-repair, negative cell cycle regulation, induction of apoptosis, detoxification, differentiation, or transcriptional regulation) were examined for the presence of CpG islands in their promoters. In total, 50 genes met these criteria (listed in Table 1).

The CpG methylation pattern of candidate promoters. The CpG methylation pattern of GSTM1 and DKK3 was determined in the cell lines LNCaP and PC3, respectively, and compared with the pattern found in primary prostate cancer cells isolated by laser microdissection of paraffin-embedded prostate sections from two patients (Fig. 1C). The CpG methylation pattern found in prostate cancer cell lines reflected the pattern found in primary tumors and is therefore appropriate for the design of methylation-specific PCR primers used in the analysis of primary tumors. In total, the promoters of 44 induced genes were subjected to analysis by bisulfite sequencing in prostate cancer cell lines which have been used for the microarray analysis (Table 1). Extensive CpG methylation was detected in the 5′ regions of DDB2, DKK3, GPX3, GSTM1, and HPGD, whereas cyclooxygenase 2 (COX2), CUTL2, p57, and RIS1 displayed focal CpG methylation (Fig. 1D). In nontransformed, primary prostate epithelial cells, these genes did not display significant CpG methylation (Fig. 1D). Thirty-five of the 44 genes analyzed by bisulfite sequencing did not show detectable CpG methylation, although these genes were significantly up-regulated after the combined 5-aza-2′deoxycytidine/trichostatin A treatment (listed in Table 1). Presumably, these genes were induced secondary to the up-regulation of other genes silenced by CpG methylation or due to nonspecific activation of transcriptional pathways by the combined 5-aza-2′deoxycytidine/trichostatin A treatment (18, 19). It is remotely possible that the CpG dinucleotides responsible for the silencing of these genes were not in the region analyzed by bisulfite sequencing in this study. However, as indicated in Fig. 1D, the analyzed regions covered several hundred base pairs around the transcription start site, which display a high CpG content and are expected to be the main targets for CpG methylation.

Validation of candidate genes for epigenetic silencing. The obtained CpG methylation patterns were then used to assign the positions of methylation-specific PCR primers (indicated in Fig. 1D). The respective methylation-specific PCR primers were tested for their specificity and sensitivity (data not shown). Reliable methylation-specific PCR conditions could not be established for CUTL2 only. For analysis of genes previously known to be silenced by CpG methylation in other tumor types, the respective published methylation-specific PCR primers were tested and used for methylation-specific PCR analysis. Thirteen genes, which displayed significant CpG methylation as determined by bisulfite sequencing or methylation-specific PCR in one of the cell lines, PC3, LNCaP, or Du-145, were examined by methylation-specific PCR in a panel of five prostate cancer cell lines, one cell line established from BPH1, in primary prostate epithelial cells and in the bladder carcinoma cell line (TSU-Pr1; Fig. 2, left). This analysis revealed that the genes initially identified in selected prostate cancer cell lines also displayed CpG methylation in other prostate cancer cell lines and occasionally in the cell line BPH1. Eight of the 13 analyzed genes did not display CpG methylation in primary prostate epithelial cells. Therefore, CpG methylation of these eight genes is a specific feature of cancerous prostate epithelial cells. APOD, DDB2, GSTM1, and RIS1 displayed partial CpG methylation in prostate epithelial cells (Fig. 2). However, the degree of CpG methylation of DDB2, APOD, and GSTM1 seemed to be significantly elevated in most of the prostate cancer cell lines when compared with normal prostate epithelial cells (Fig. 2), suggesting a prostate cancer–specific increase in CpG methylation of these genes and potential subsequent silencing. CpG methylation of SFRP1, GPX3, p57, HPGD, GSTM1, and APOD was also detected in the bladder carcinoma cell line TSU-Pr1.

In order to determine whether CpG methylation correlates with reduced or absent expression of the respective genes, we analyzed the expression levels of nine genes which showed selective or preferential CpG methylation in prostate cancer. RT-PCR analysis of cDNAs obtained from normal prostate epithelial cells derived from four healthy donors and four prostate cancer cell lines revealed several distinct patterns of mRNA expression (Fig. 2, right). The CpG methylation interrogated by the methylation-specific PCR primers used here largely correlated with reduced gene expression in the case of SFRP1, DKK3, GPX3, COX2, GSTM1, APOD, and p57, whereas the detected CpG methylation of DDB2 and HPGD was not accompanied by decreased gene expression. In the case of HPGD, the expression was even induced in three prostate cancer samples, which clearly showed CpG methylation. These results indicate that CpG methylation of a promoter should not be interpreted as proof of its transcriptional repression.

Analysis of CpG methylation in primary prostate cancer samples. The CpG methylation status of the genes showing CpG methylation–mediated silencing in prostate cancer cell lines was determined in 41 primary prostate cancer samples obtained after radical (37 cases) or transurethral (4 cases) resection. In addition, the genes DDB2 and HPGD were included in this analysis, although we had not detected a correlation between CpG methylation and down-regulation of mRNA expression for these genes. Nonetheless, the detection of prostate cancer-specific CpG methylation in the promoter of these genes may be useful for diagnostic applications. Prostatic tissue samples contain several cell types. Non-neoplastic epithelial cells, stromal cells, lymphocytic infiltrates, and blood cells are present in close proximity to prostate cancer cells. Therefore, laser microdissection was employed to isolate prostate cancer cells. Similarly, non-neoplastic prostate epithelial cells were isolated from samples obtained from 9 patients with benign prostate hyperplasia, which did not present prostate cancer and were in the same age group as the 41 prostate cancer patients. Genomic DNA was isolated from these samples and subjected to methylation-specific PCR analysis (Fig. 3A and B). The analyzed genes showed CpG methylation at medium to high frequencies in prostate cancer cells. CpG methylation was detected for SFRP1 in 34 (83%), COX2 in 32 (78%), DKK3 in 28 (68%), GPX3 in 38 (93%), p57 in 23 (56%), HPGD in 30 (73%), and DDB2 in 34 (83%) of the 41 prostate cancer samples analyzed. Predominant methylation of GSTM1 was detected in 24 (58%) of 41 cases. We had also identified consistent silencing of 14-3-3σ in prostate cancer in this screen which was described elsewhere (20). The silencing of 14-3-3σ in the set of 41 prostate cancers analyzed here is depicted for comparison (Fig. 3B). For SFRP1 and COX2, no CpG methylation was detected in BPH derived from nine different patients, suggesting that the CpG methylation of these genes is specific for neoplastic prostate epithelial cells. CpG methylation was detected for GPX3 in two and for DKK3 in one of nine analyzed BPH samples (Fig. 3B). For GSTM1 CpG methylation was detected in the five BPH samples analyzed (Fig. 3B). However, the CpG methylation of GSTM1 was elevated in the majority of prostate cancer samples, whereas in the non-neoplastic BPH cells, equal signals for the PCR products representing methylated and unmethylated GSTM1 alleles were detected. Furthermore, the nonmethylated GSTM1 allele was not detected in several samples of prostate cancers. No obvious correlation between staging information and CpG methylation was detected by cluster analysis (data not shown).

In order to evaluate the tumor-specificity of the CpG methylation detected in prostate cancer, we examined the CpG methylation status in human diploid fibroblasts derived from neonatal skin, in stroma isolated from five cancer-free prostate specimens, and in peripheral blood mononuclear cells from six individuals (Fig. 3B). The p57 gene did not show CpG methylation in stromal cells. Partial CpG methylation of SFRP1 and COX2 was evident in only one of five analyzed stroma samples. One of six peripheral blood mononuclear cell samples showed partial methylation of SFRP1 and p57 genes. By contrast, DKK3 and GPX3 displayed CpG methylation in most of the stromal samples and blood cells.

Loss of SFRP1 expression in prostate cancer. In order to determine whether the CpG methylation of SFRP1 affects the expression of the respective gene product in primary tumors, the level of SFRP1 protein expression was determined by immunohistochemistry in prostate cancer samples derived from 39 different patients (representative example shown in Fig. 4A). In non-neoplastic prostate glands, most of the luminal cells were positive for SFRP1 with a characteristic granular cytoplasmic and apical membrane staining, whereas prostate cancer cells were devoid of SFRP1 staining. A prominent down-regulation (>50% of reduction) or complete loss of SFRP1 protein was detected in 29 of 39 prostate cancer samples (data not shown).

Functional analysis of DKK3 and SFRP1. Epigenetic inactivation of the members of the SFRP gene family presumably contributes to activation of WNT signaling in colorectal cancer (16). DKK3 negatively regulates the β-catenin pathway in osteosarcoma cells (21). Therefore, we asked whether silencing of SFRP1 and/or DKK3 expression in prostate carcinoma cells is associated with constitutive activation of WNT/β-catenin signaling. Unexpectedly, β-catenin was localized at the cell membrane and absent from the nucleus in PC3 and Du145 cells (Fig. 4B), which show significant silencing of DKK3 and SFRP1 (Fig. 2). As nuclear β-catenin is a hallmark of an activated WNT/β-catenin pathway, this pathway is presumably not active in prostate cancer. Furthermore, cotransfection with a WNT1 expression construct did not result in activation of a TCF-reporter in PC3 and Du145 cells (Fig. 4C). In contrast, HCT116 colon cancer cell lines, which harbor an activated WNT/APC/TCF4 pathway (22), showed nuclear β-catenin localization (data not shown) and strong activation of the wild-type but not the mutant TCF reporter by WNT1 coexpression (Fig. 4C).

Recently, it has been reported that WNT1 activates the mitogen-activated protein kinase (MAPK) pathway (23). Therefore, we tested whether the WNT antagonists, DKK3 and SFRP1, inhibit the MAPK pathway. In exponentially growing PC3 cells ectopically expressing DKK3, the level of ERK1 and ERK2 phosphorylation, which is indicative of MAPK activity, was diminished (Fig. 5A and B). Furthermore, PC3 cells ectopically expressing DKK3 or SFRP1 showed a significant decrease in colony formation (Fig. 5C), which was due to a decrease in colony size. The rate of spontaneous apoptosis was not affected by ectopic expression of DKK3 or SFRP1 as determined by flow cytometry (data not shown). Therefore, loss of DKK3 and SFRP1 expression by epigenetic silencing presumably promotes the proliferation of prostate epithelial cells.

The high frequency of CpG methylation we observed in primary tumors suggests that the differences in CpG methylation which were initially detected in prostate cancer cell lines did not occur during in vitro cultivation of prostate cancer cell lines. Whether the CpG methylation of distinct genes is the cause or consequence of their down-regulation during prostate cancer initiation or progression remains to be shown. In addition, the functional relevance of the down-regulation of these genes has to be determined in the future. However, the high frequency of the epigenetic silencing events identified in this study suggests that these genes may be causally involved in prostate cancer development. In the following paragraphs, we will discuss potential functional consequences of the epigenetic silencing events detected here.

SFRP1 negatively regulates the WNT pathway, which is frequently activated in cancer by mutations in the APC and β-catenin genes (24, 25). Recently, SFRP1 was shown to undergo both genetic and epigenetic alterations in colon and bladder cancers (26, 27). Frequent hypermethylation of SFRP1 was recently detected in colorectal cancer (28), where the loss of SFRP1 correlated with an increased β-catenin/TCF4 activity and ectopic expression of SFRP1 led to reduced β-catenin/TCF4 activity. However, in prostate cancer, we could not detect activation of the β-catenin/TCF4 pathway.

Therefore, loss of SFRP1 may contribute to activation of other oncogenic signaling pathways in prostate cancer. We have previously observed an induction of SFRP1 mRNA in human primary fibroblasts undergoing replicative senescence (15), suggesting a possible role of SFRP1 in the terminal proliferation arrest of senescent cells. While this paper was in preparation, CpG methylation of SFRP1 in <5% of prostate cancers was reported (7), whereas we detected CpG methylation of SFRP1 in 83% of prostate cancers. This difference may potentially be caused by the different types of materials that were analyzed. Florl et al. (7) isolated DNA from pieces of tissue containing prostate cancer and other cell types. Here, prostate cancer cells isolated by laser microdissection were analyzed.

DKK3 negatively regulates the WNT-pathway. The zonal distribution of DKK3 expression in the adrenal gland suggests that DKK3 is involved in zonal differentiation or growth (29). DKK3 was identified as a gene which is down-regulated upon immortalization of primary human cells (30). Reduced expression of DKK3 was also found in non–small cell lung cancer and renal clear cell carcinoma (31, 32). Overexpression of DKK3 inhibited growth, invasion, and motility of Saos-2 osteosarcoma cells by modulating WNT-signaling (21). Interestingly, the DKK3 transcript was shown to be up-regulated in senescent primary prostate epithelial cells (33). The epigenetic silencing of DKK3 by CpG methylation observed here might therefore contribute to de-differentiation and immortalization of prostate cancer cells.

Our data suggest that the WNT/β-catenin pathway is not activated in prostate cancer cell lines (Du-145 and PC3) with silenced SFRP1 and DKK3 genes. This is in agreement with a recent comprehensive study of 101 cases of primary prostate cancer: none of the tumors showed nuclear β-catenin staining (34). However, genetic alterations of β-catenin or APC were detected in a subset of advanced prostate cancers and were associated with a resistance to apoptosis (35, 36). Loss of SFRP1 and DKK3 expression may activate alternative signaling pathways. Recently, it was reported that transactivation of the EGF receptor by WNT ligands, which results in MAPK activation, is inhibited by SFRP1 and DKK1 (23). Our data indicate that DKK3 may also have an inhibitory effect on MAPK signaling.

p57/KIP2 belongs to a family of conserved CDK inhibitors, which negatively regulate the cell cycle. Ectopic expression of p57 suppresses cell transformation, whereas cells lacking p57 show increased cell proliferation and decreased differentiation (37–39). p57 expression is decreased in prostate cancer cell lines and primary prostate epithelial cells immortalized by HPV16 E7 (40). Expression of p57 induces a senescence-like phenotype in prostate cancer cells (40), suggesting that down-regulation of p57 may be required for immortalization of prostate cells. The p57 gene is located on chromosome 11p15.5, a region implicated in both sporadic cancers and the Beckwith-Wiedemann cancer syndrome. Mutations of p57 have rarely been detected in human tumors (41). Epigenetic silencing of p57 was also detected in gastric, hepatocellular, pancreatic carcinomas, and acute myeloid leukemia (42, 43). The down-regulation of p57 in bladder cancer involves several mechanism including loss of heterozygosity and hypermethylation (44).

Glutathione peroxidase 3 (GPX3) catalyzes the reduction of peroxides by glutathione and protects cells against oxidative damage. The prostate cancer–specific silencing of GPX3 may lead to an impaired defense against endogenous and exogenous genotoxic compounds, which could potentially result in an increased rate of mutation in critical genes. Down-regulation of GPX3 expression upon transition from normal to neoplastic prostate tissue was recently detected by microarray analysis of microdissected primary prostate cancer (45). Presumably, this down-regulation of GPX3 in prostate cancer is caused by the epigenetic silencing of GPX3 detected here.

Glutathione S-transferases are active in the detoxification of a wide variety of toxins and carcinogens. The common null-allele of GSTM1 shows a weak association with lung cancer (reviewed in ref. 46). However, no significant association of GSTM1 polymorphisms or deletion with prostate cancer have been reported. The tumor-specific hypermethylation of GSTM1 identified here may explain the decreased expression of GSTM1 in prostate cancer detected in three previous studies (45, 47, 48).

COX2 catalyzes the synthesis of prostaglandin H2, a precursor of other prostanoids, and has been implicated in inflammation and carcinogenesis (reviewed in ref. 49). While this manuscript was in preparation, CpG-methylation of the COX2 gene in prostate cancer was reported independently (8, 9). In the study by Kang et al. (8), the CpG methylation of COX2 was detected in 22% of primary prostate cancer, whereas Yegnasubramanian et al. (9) detected hypermethylation of COX2 in 88% of prostate cancer when they applied a quantitative method to analyze a large collection of samples. Interestingly, the hypermethylation of COX2 was associated with a higher risk of recurrence of prostate cancer (9).

The lack of a correlation between CpG methylation and grade or stage of the prostate cancer is in agreement with other recent studies (7–9). These studies and our data suggest that silencing of genes by CpG methylation occurs at an early stage of prostate cancer development. This may have implications for the use of these CpG methylation events for detection of early prostate cancer lesions. The detection of aberrant CpG methylation has several significant advantages when compared with protein- or RNA-based tumor markers (50). According to our analysis, the CpG methylation of SFRP1, COX2, and p57 has the highest specificity for prostate cancer and is therefore of higher relevance for potential diagnostic applications. However, this conclusion requires validation in larger cohorts of patients in the future.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.

We thank Reinhard Hoffman for help with the microarray analysis and Anja Heyer for expert assistance with immunohistochemistry, Wolfgang Klinkert for assistance with flow cytometry, Hiromu Suzuki for WNT1 and SFRP1 encoding constructs, and Axel Ullrich for antibodies and cell lines. Heiko Hermeking is supported by the Max-Planck-Society, the Deutsche Krebshilfe/Dr. Mildred-Scheel-Stiftung and the Rudolf-Bartling-Stiftung.