Global Reactivation of Epigenetically Silenced Genes in Prostate Cancer

Read the Perspective on this article by Jerónimo, p. 1053

Aberrant DNA hypermethylation of CpG islands in the promoter region of genes is well established as a common mechanism for the transcriptional silencing of tumor suppressor genes in cancer cells and, thus, as an alternative mechanism of functional inactivation (1). The GSTP1, p16^INK4a, CDH1, APC, and RASSF1A tumor suppressor genes as well as a number of other cancer genes have been identified as hypermethylated with associated loss of expression in prostate cancer (2). By definition, a candidate gene approach has resulted in the examination of only a limited number of genes for epigenetic alteration. Many other tumor suppressor and cancer genes important in prostate tumorigenesis likely remain to be identified. A global approach to the identification of epigenetically silenced genes in prostate tumor cells could provide methylation signatures for early detection and for predictive classification studies, identify novel targets for therapy, and lead to further elucidation of the biology of this disease.

One global approach to the identification of epigenetically silenced genes in tumor cells is based on the reversal of epigenetic silencing by drugs such as 5-aza-2-deoxycytidine (5Aza-dC) resulting in reexpression analyzed by well-annotated gene expression arrays. This approach can preferentially identify reexpression of epigenetically silenced genes over methylated CpG islands that do not affect transcription. A proportion of the reexpressed genes will have been silenced by promoter hypermethylation in the untreated tumor cell lines (3–6).

In the present study, we examined the global reactivation of epigenetically silenced genes in prostate cancer by analysis of a gene expression microarray with RNA isolated from four prostate tumor cell lines after treatment with 5Aza-dC and trichostatin A (TSA). Through intuitive selection of upregulated genes followed by validation, we have evidence that at least 20 of 45 genes examined are hypermethylated in prostate cancer, and thus, our screen preferentially selected for epigenetically silenced genes. We report here four genes newly identified as hypermethylated in primary prostate tumor specimens. Informed analysis of function combined with a pathway and network database analysis supports the relevance of these genes to prostate cancer.

Four prostate cancer cell lines, LNCaP, DU-145, PC-3, and MDA2b, were obtained directly from the American Type Culture Collection and were cultured according to the company's recommendations. The four prostate cancer cell lines were treated with 5Aza-dC (Sigma) and TSA (Wako) in a combined treatment. 5Aza-dC was dissolved in PBS as a 5 mmol/L stock solution and stored in aliquots at −80°C. TSA was dissolved in absolute ethanol as a 330 μmol/L stock solution and stored at −20°C. Cells were exposed to 5Aza-dC to a final concentration of 5 μmol/L at 0, 24, and 72 hours over two cell divisions by counting the cells, and then treated with TSA to a final concentration of 500 nmol/L during the 24 hours before RNA extraction. Mock (untreated) cells were cultured with an equivalent volume of PBS alone and, for the 24 hours before RNA extraction, with an equivalent volume of ethanol.

Total RNA used for microarray analysis was isolated from treated and mock-cultured cells using TRIzol reagent (Invitrogen) and purified using an RNeasy Mini Kit (Qiagen), combined with DNase treatment. RNA quality was confirmed by the ratio of 28S and 18S rRNA after agarose gel electrophoresis. Total RNA was reverse transcribed using oligo(dT)₂₄ primer and Superscript II reverse transcriptase (Invitrogen) for 1.5 hours at 42°C. cDNA was labeled with Cy3 or Cy5 (Amersham Biosciences) and then hybridized according to the manufacturer's instructions for 18 hours at 42°C on slides with oligonucleotides from 14,802 genes, which were processed and spotted in the DNA Microarray Facility at Fox Chase Cancer Center from 15k human oligonucleotide microarray (MWG-Biotech, Inc.). The Gene list, Gene ID, and Template files for human 15k oligo A can be viewed at http://www.ocimumbio.com/web/arrays/download.asp.

Each cell line was hybridized twice, with a dye-flip replicate using reversed labeling. The hybridized slides were scanned using a GMS 428 scanner (Affymetrix) to generate high-resolution images for both Cy3 and Cy5 channels. Image analysis was done using the ImaGene software (BioDiscovery, Inc.). The spots were identified using an optimized segmentation algorithm. Spots of poor quality, as well as spots with signal levels indistinguishable from the background, were flagged as bad spots. The image data were extracted and analyzed using the Functional Genomics Data Pipeline (7).

Normalization and background correction were done using LOWESS with a width of 0.7 and local background correction. The PC-3 flip array data were removed as quality control measures and its MA plot suggested a poor-quality array. Our goal was to identify genes that changed expression due to methylation in the maximum number of cell lines, as this provided the greatest probability of in vivo methylation. We therefore categorized the genes based on their fold change, determining a cutoff of 2-fold (i.e., ratio = 2, log-ratio = 1) between treated and untreated cell lines using the mean value of the dye-flips. We did not correct for disparate values between the dye-flip readings, as we took one large value as indicative of potential methylation in that cell line, and our goal was to count the number of cell lines with such potential methylation. We ranked the genes based on the number of cell lines with potential methylation. We noted several well-known methylation targets, which we took as partial validation of our approach, and pursued the novel genes further.

After Institutional Review Board approval and informed consent, primary prostate tumor tissues were obtained from surgical specimens resected at Fox Chase Cancer Center followed by pathologic review and dissection of tumor cell–enriched areas. In this study, we used histologically normal tissue from a prostate with no evidence of cancer obtained from cystoprostatectomy of age-matched bladder cancer patients. DNA was extracted from fresh-frozen tissues or paraffin blocks using a standard technique of digestion with proteinase K followed by phenol-chloroform extraction (8).

Genomic DNA (1 μg) in a volume of 50 μL was denatured by NaOH (0.2 mol/L) for 10 minutes at 37°C and then modified by hydroquinone and sodium bisulfite treatment at 50°C for 17 hours under a mineral oil layer. Modified DNA was purified using Wizard DNA Clean-Up system (Promega). Modification was completed by NaOH (0.3 mol/L) treatment for 5 minutes at room temperature, followed by glycogen, ammonium acetate, and ethanol precipitation (9). Pellets were resuspended in water and stored at −20°C for immediate use or at −80°C for longer-term storage.

A typical 200- to 400-bp fragment containing the promoter CpG island was PCR amplified with bisulfite-modified prostate cancer cell line DNA, normal prostate tissue DNA, and normal lymphocyte DNA for each gene analyzed. The PCR product was loaded into a 1.5% agarose gel, cut out, and cleaned (Qiagen). We used a single PCR amplification for each of the KRT7, TACSTD2, SLC15A3, GADD45b, IFI30, ANXA2, and AQP3 genes. The primers used for each gene analyzed are given in Supplementary Table S2.

Primer sequences to methylated DNA sequence were designed together with an internal TaqMan probe labeled with FAM and MGBNFQ for SLC15A3, KRT7, GADD45b, IFI30, or TACSTD2. In vitro methylated normal human genomic DNA, confirmed by bisulfite sequencing to show methylation for the gene to be analyzed, was used as a positive control. The concentration of this DNA was determined and a series of dilutions made for the standard curve. Unmethylated sequence of the ACTINβ gene was used as a normalizing control. The percentage of methylated alleles was calculated for a gene based on the standard curve. An Applied Biosystems 7500 Real-Time PCR machine was used for PCR and data were analyzed with SDS 1.3.1 software. AQP3 and ANXA2 were assessed by conventional gel-based methylation-specific PCR (MSP). Primer and probe sequences are given in Supplementary Table S2.

We analyzed differential expression on a 14,802 human gene microarray between RNA from four mock (untreated) prostate cancer cell lines and the same cell lines treated with 5Aza-dC over two cell doubling times and a single dose of TSA 24 hours before harvesting. We combined the 5Aza-dC treatment with TSA because there is a reported synergistic effect on demethylation of DNA (10). A total of 2,997 genes were upregulated at least 2-fold in one or more of the four prostate cancer cell lines analyzed compared with the mock cells. The list of genes was then ranked in descending order from upregulation in all four cell lines to upregulation in one cell line only (Supplementary Table S1). Further data can be found at http://www.ncbi.nlm.nih.gov/geo/ upon publication. The list of ranked upregulated genes was then prioritized for validation by examination of expression status in the normal cell counterpart compared with the tumor cell and the presence and location of a CpG island in the promoter region (11). In addition, known imprinted genes (e.g., IGF2), poorly annotated genes, or genes previously identified by us as upregulated but not to have cancer-specific methylation (e.g., TGM2 and GAGE7; ref. 11) were excluded. Accordingly, we selected the first 45 genes that had higher or equal expression in normal prostate compared with prostate cancer cells according to the Cancer Genome Anatomy Project (CGAP) Serial Analysis Gene Expression (SAGE) database and that contained a CpG island within the promoter by the criteria of Takai and Jones (12) through WebGene analysis of the genomic sequence. Genes that showed no expression in normal cells or did not have a CpG island were excluded from immediate study (Table 1). Importantly, we noted that three of the most frequently methylated genes identified to date in prostate cancer [i.e., GSTP1 (13), PDLIM4 (14), and IGFBP3 (15)] were included in this 45-gene list.

We then examined the published literature through the GeneCard (http://www.genecards.org) and PubMed databases, which revealed that 15 genes were previously reported to be hypermethylated in prostate cancer (Table 1). These 15 genes were excluded from immediate analysis. We therefore prioritized the remaining genes for study, and the promoter methylation status of 23 of this set of genes was first validated by direct bisulfite sequencing of the untreated prostate tumor cell line DNA. Eight genes showed extensive methylation (i.e., in a majority of CG dinucleotides; Fig. 1 and Supplementary Fig. S1) in at least one prostate tumor cell line (Table 1). To examine if hypermethylation was specific to neoplastic prostate cells, we performed direct bisulfite sequencing of a normal prostate tissue DNA obtained from the cystoprostatectomy (for bladder cancer) of an age-matched male with no clinical or histopathologic evidence of prostate cancer and also a normal lymphocyte DNA. With the exception of BASP1, all genes were unmethylated in the normal prostate tissue DNA (Fig. 1; Supplementary Fig. S1) and unmethylated in the normal lymphocyte DNA.

Because there is evidence that transformed tumor cell lines can have more gene methylation than the patient tumor specimen counterparts (16), we next examined the frequency and timing of hypermethylation of the KRT7, TACSTD2, SLC15A3, GADD45b, and IFI30 genes by quantitative real-time MSP of DNA from a set of 19 prostatic intraepithelial neoplasia (PIN) and 35 prostate tumor tissue specimens from patients (Table 2; Fig. 2). The tumors consisted of 18 low and intermediate grade (Gleason ≤7), 16 high grade (Gleason ≥8), and one of unknown grade. Thirteen tumors were stage I or II and 22 tumors were stage III or IV. Hypermethylation of the ANXA2 and AQP3 genes was examined by conventional gel-based MSP and verified by bisulfite direct sequencing in a subset of 12 PIN and 20 prostate tumor tissue specimen DNA described above.

The keratin 7 (KRT7) gene was hypermethylated in 10 of 19 (53%) PIN and 19 of 35 (54%) tumors. There was a significant difference (P = 0.037) in frequency between 13 of 18 Gleason ≤7 and 5 of 16 Gleason ≥8, but not between stage I-II and stage III-IV prostate tumors (Table 2). Cytokeratins are a subfamily of intermediate filament proteins characterized by biochemical diversity being represented in human epithelial tissues by at least 20 different polypeptides. Expression of particular cytokeratins is often cell type specific. KRT7 is expressed in the ductal epithelium of the genitourinary tract. Expression also varies by course of terminal differentiation (17). Hypermethylation of KRT7 might indicate lineage differentiation or provide a growth advantage through cell-cell adhesion. A recent study reported a subgroup of clear cell renal cell carcinomas with KRT7 expression being associated with genetic stability, a distinct global expression signature, and a more indolent clinical course (18). A network of molecular interactions and canonical pathways for KRT7 formed by Ingenuity Pathways Analysis (IPA) was extensive. Consequently, an example of interaction with one cancer relevant pathway (Wnt signaling) only is shown in Supplementary Fig. S2.

SLC15A3, also known as PHT2 or PTR3, is a member of the SLC15 family of electrogenic transporters that use the proton-motive force for uphill transport of short-chain peptides and peptido-mimetics into a variety of cells (19). Whereas the function of SLC15A3 in cancer cells is unclear, IPA identified interaction of SLC15A3 with the NF-κB and the p38 mitogen-activated protein kinase signaling pathways as well as other cancer-related pathways (Supplementary Fig. S2). We found SLC15A3 to be methylated in 7 of 19 (37%) PIN and 23 of 35 (66%) prostate tumors (Table 2; Fig. 2).

Tumor-associated calcium signal transducer-2 (TACSTD2) or TROP-2 is a cell surface glycoprotein. Its function remains largely unknown; however, it is phosphorylated by protein kinase C, and cross-linking TACSTD2 with antibodies causes a transient increase in intracellular calcium levels, implying that it has a role in signal transduction (20). A recent report identified two subpopulations of basal cells based on TROP2 expression and reported TROP2 to be a marker of human prostate basal cells with stem cell characteristics (21). Prostate tumors with TACSTD2/TROP2 hypermethylation and associated loss of expression may identify a distinct subgroup in terms of tumor lineage or tumor behavior. IPA of TACSTD2 is shown in Supplementary Fig. S2, and interestingly, TACSTD2 interacts with SMARCA4/BRG1, reported to be mutated in tumor cell lines including DU-145 (22) and a component of the SWI-SNF chromatin remodeling complex. In our study, we found TACSTD2 to be unmethylated in 19 PIN but hypermethylated in 6 of 35 tumors (P = 0.08; Table 2; Fig. 2). TACSTD2 has been reported to be hypermethylated in glioblastomas (23). Current opinion is that patients diagnosed with PIN are not at a higher risk of prostate cancer than are patients diagnosed with benign tumors (24). Thus, a gene methylated in prostate tumors but not in PIN might be a useful marker for the differentiation of PIN from other more aggressive lesions (e.g., intraductal carcinoma). TACSTD2 may therefore have utility in emerging methylation-based prostate cancer tests (25–28).

Promoter hypermethylation of the growth arrest and DNA damage–inducible, β (GADD45b, also referred to as hypothetical protein dkfzp566b133) gene was found by quantitative MSP in one prostate tumor only (Gleason 9, stage IV). Direct bisulfite sequencing of the tumor DNA confirmed this result (data not shown). GADD45 is involved in the regulation of cell cycle arrest and apoptosis and inhibition of cell growth. GADD45 responds to environmental stresses by mediating activation of the p38/c-jun NH₂-terminal kinase pathway (29). A selective IPA of the 10 pathways most strongly associated with GADD45b is shown in Supplementary Fig. S2. GADD45b has been reported to be hypermethylated in hepatocellular tumors (30). Another GADD45 gene, GADD45a, was previously reported as hypermethylated in prostate tumors (31).

The IFI30, ANXA2, and AQP3 genes were unmethylated in all PIN and prostate tumor DNA examined. It is possible that analysis of a larger number, or a particular subtype, of primary prostate tumor specimens may yet reveal methylation of these genes. Of the four prostate tumor cell lines, three lines were established from distant metastases and the other, LNCaP, from a lymph node metastasis (32). The source of the cell lines taken together with our observation that these genes were unmethylated in primary specimens highlights that the prostate tumor cell lines may be unrepresentative of the disease found in the population. It will be interesting to see if any of these three genes are subsequently identified as aberrantly methylated in other types of cancer.

Evidence for the specificity and potential of our screen was provided by the inclusion of the GSTP1 gene known to be hypermethylated in the LNCaP and PC-3 cell lines (13), as well as in the majority of primary prostate tumors, and two other genes, PDLIM4 (14) and IGFBP3 (15), reported to be frequently methylated in primary prostate cancer in the 45 selected genes. Because an oligonucleotide probe on an array may not discriminate between alternative splice forms of the same gene, genes that do not display loss of expression of all isoforms with promoter hypermethylation (e.g., RASSF1A) will likely not appear upregulated by a global reactivation approach (33). There was evidence from the literature for epigenetic regulation of several other genes in the 45 studied. The references for such genes are given in Table 1. In our list (Table 1), there were also additional genes reported by several groups to be upregulated after epigenetic reactivation but with no evidence of methylation in prostate tumor cells (e.g., DUSP1; refs. 31, 34). Such genes may be upregulated by demethylation and reactivation of another gene or may result from a stress response to the drug treatment. In general, our study compares well to the “hit rate,” that is, the ratio of genes hypermethylated among genes upregulated after demethylation reported in other global reactivation studies (5, 6, 31). We did not place emphasis on the higher fold of reactivation because we considered this to be arbitrary given the current limited understanding of the degree of reexpression necessary to restore the normal function of a particular gene.

The average total number of genes methylated with functional significance in the prostate tumor cell is unknown but might be reasonably estimated as several hundreds (35, 36). Several prostate cancer global methylation studies have been reported. Yu et al. compared the results from PC-3, DU-145, and LNCaP lines on an oligonucleotide-based methylation array to an expression array (37) and reported a number of genes methylated by array analysis, some of which were replicated by nonquantitative conventional MSP analysis but not by direct bisulfite sequencing. Lodygin et al. performed a global epigenetic reactivation of the same three prostate tumor cell lines and identified a number of known, as well as novel, genes methylated in prostate cancer from the subset of 50 genes examined (31). Several of the genes in the ranked selection by Lodygin were also in our selected list. Their study did not perform quantitative MSP or examine normal cells for imprinted or tissue-specific methylated genes. Hoque et al. also studied epigenetic reactivation in the same three lines as well as the 22Rv1 line. They selected 45 reactivated genes, of which 9 were previously reported to be hypermethylated in cancer cells; 16 of the remaining 36 were methylated in the cell lines and 8 in primary prostate tumors of which 3 did not have methylation in normal prostate cells (38). Chung et al. used methylated CpG island amplification coupled with representational difference analysis of PC-3, DU-145, and LNCaP, and validation by combined bisulfite restriction analysis and pyrosequencing identified six novel methylated genes in primary prostate tumors that had significantly higher methylation than in adjacent normal prostate tissue (39).

In summary, we report for the first time four novel genes with promoter hypermethylation in primary prostate tumor specimens but that are unmethylated in normal cells. One of these genes, TACSTD2/TROP2, is a potential marker for detection or diagnosis of prostate cancer rather than for PIN only, as well as a prostate stem cell marker (21). The specificity of our screen for hypermethylated genes is further supported by the reactivation of GSTP1 and other genes known to be frequently methylated in prostate cancer in our selection. Further mining of the data provided here (Supplementary Table S1) as well as emerging technologies will increase our knowledge of the prostate hypermethylome. Such studies should lead to further understanding of the biology of prostate tumorigenesis and the identification of further hypermethylated genes as candidate markers for the diagnosis and prognosis of prostate cancer.

P. Cairns is a paid consultant to Oncomethylome Sciences. The terms of this arrangement are being managed by Fox Chase Cancer Center in accordance with its conflict of interest policies. The other authors disclosed no potential conflicts of interest.

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.