Global Methylation Profiling for Risk Prediction of Prostate Cancer

This article is featured in Highlights of This Issue, p. 2721

Prostate cancer is the most frequently diagnosed cancer in men in Western countries (1). Current available therapies are surgery, radiotherapy, and androgen ablation. The response rate is as high nearly 90%; however, most of these recur or become refractory and androgen-independent (2). Existing approaches for diagnosing prostate cancer features digital rectal examination screening and serum prostate-specific antigen (PSA) determination (3). However, because of their limited sensitivity and specificity, these methods cannot reliably identify early-stage prostate cancer (4). The clinicopathologic features predictive of prostate cancer consist of Gleason score, tumor-node-metastasis (TNM) stage, surgical margin status, preoperative serum PSA levels, and GPSM (Gleason score, preoperative PSA, seminal vesicle involvement, and marginal status) score. GPSM score is a prognostic model using the weighed sum of the pathologic Gleason score, preoperative PSA, seminal vesicle involvement, and marginal status to predict biochemical progression after radical prostatectomy (5, 6). Several preoperative nomograms developed in the past decade have been based on clinicopathologic variables (6–8). These factors are helpful but far from perfect due to significant clinical heterogeneity of the disease.

Among the different types of biomarkers discovered (9), DNA methylation markers stand out for their potential to provide a unique combination of specificity, sensitivity, high information content, and applicability to a wide variety of clinical specimens (10). DNA promoter hypermethylation can lead to gene silencing (11); on the other hand, global hypomethylation, inducing genomic instability, contributes to cell transformation. Both events, either individually or in cooperation, result in the development and progression of cancer (12). Perhaps one of the most important features of DNA methylation profiling in cancer is that these profiles are both tissue- and tumor-type–specific (13). This can enable the identification of biomarkers for diagnosis, prognosis, and response to treatment and to identify potential biologic pathways which are disrupted in tumor development (14). GSTP1 is the most frequently investigated gene in prostate cancer epigenetics (15–18). Only a limited panel of genes has been identified so far, including those from our studies that are methylated at a high frequency in prostate cancer (16, 19–21). The limited value of established prognostic markers demands identification of new molecular parameters of interest in predicting the prognosis of patients with prostate cancer. We hypothesize that DNA methylation changes that occur very early during the development of cancer before the manifestation of clinical symptoms could have a substantial impact as reliable early detection biomarkers and those methylation changes that can distinguish the aggressive forms of prostate cancer which are life-threatening from indolent tumors could be established as new prognostic biomarkers for making effective treatment decisions to improve the cure rates and reduce the mortality due to neoplasia.

The objective of this study was to identify biologically and clinically relevant methylated genes as biomarkers characteristic of prostate cancer versus benign tissues, recurrent versus nonrecurrent tissues, clinical recurrent versus biochemical recurrent tissues, and systemic versus local recurrence using methylation microarrays for diagnosis and prediction of prostate cancer progression. The methylation profiles were generated from 36 systemic recurrence, 44 local recurrence, 43 biochemical recurrence, and 75 nonrecurrent patients. In addition, 40 matched normal prostatic tissues were also used. Methylation of genes that predict the risk of recurrence was analyzed by univariate and multivariate analyses in addition to the binary group comparisons. We further validated the methylation of genes in 20 independent cases of systemic recurrence, local recurrence, biochemical recurrence, and nonrecurrent patients.

Surgically resected prostate cancer tissue specimens were obtained from patients who had undergone radical prostatectomy for prostate cancer at Mayo Clinic, with Institutional Review Board approval. All the tissues were collected from nonpretreated patients and frozen at −80°C as described earlier (22). Hematoxylin and eosin–stained sections were prepared before and after slides were cut for DNA isolation, to ensure that tumor was present in the tissues used for analysis with minimal number of disturbing stromal cells. The pathologic status was confirmed before processing, and we included tumor samples with more than 70% cancer and excluded non-age–matched samples for methylation profiling. Cancerous regions were identified and marked in the sections. The sections were then macrodissected from the slide using a 21-gauge sterile needle to separate tumor from nontumor cells. For diagnosis of prostate cancer, case (tumor) and control (normal adjacent tissues) age-matched (birth year) samples were used. For prediction of progression, cases (tissues from recurrent patients within 5 years of prostatectomy) and controls (tissues from patients without any recurrence for 7 years following radical prostatectomy) matched on age, Gleason score, stage, and preoperative PSA were used. All patients with postoperative biochemical failure (men with initial PSA increase without further clinical progression for 5 years following radical prostatectomy, PSA level ≥ 0.2 ng/mL) were identified as PSA failure. Clinical recurrence is defined as men with local and systemic recurrence from 90 days to 5 years following the PSA increase. Local recurrence is defined on the basis of digital rectal examination, transrectal ultrasonography, and biopsy. Systemic progression is defined as evidence of metastatic disease on a bone scan. Clinical and pathologic information, including Gleason score, preoperative PSA, TNM stage, ploidy, extraprostatic extension, regional lymph node involvement, positive margins of resection, and GPSM score of patients were obtained from the Mayo SPORE prostate cancer tumor registry. Separately collected prostate tissue samples matched with the cancer tissues (obtained from the same patients) were used as matched normal controls.

Before processing for DNA isolation, quality control check of the samples was conducted to ensure that there was no DNA damage of the samples during the preservation and storage at −80°C. Genomic DNA was isolated from 10 tissue sections of 20-μm thickness using the ZR Genomic DNA II Kit (Zymo Research Corp.). Using a Zymo EZ DNA methylation kit (Zymo Research Corp.) 1.0 μg of the DNA was subjected to sodium bisulfite modification as described (16, 21, 22). Universal methylated human DNA and nonmethylated DNA (Zymo Research Corp.) were used as positive and negative controls for bisulfite modification. DNA quantity was then determined using both NanoDrop and the Agilent BioAnalyzer 2100 with DNA 1000 and 7500 chips.

DNA methylation analysis was conducted using the Illumina Infinium Human Methylation27 BeadChip. Each HumanMethylation27 BeadChip consisted of 12 arrays and up to 4 bead chips were processed simultaneously. Cases and controls were randomly placed within the bead chips. The bead chip allowed the interrogation of 27,578 CpG sites representing 14,495 protein-coding gene promoters, which includes about 1,000 cancer-associated genes. The standard protocol provided by Illumina was used for DNA methylation analysis as described (23).

Raw data were imported into GenomeStudio (version 2009.1), and a methylation profile (including average β, detection P value, and signal intensities for signal A and B) was generated for each sample. The methylation status for each CpG site was measured by comparing fluorescent signal from the methylated allele with the sum of the fluorescent signal from both methylated and unmethylated alleles. Detailed quality assessment was conducted by examining the distribution of average β, number of reliably detected CpGs for each sample, principal components analysis plot, and unsupervised clustering. The differentially methylated CpGs were identified using one-way ANOVA with a batch-effect included in the model to adjust for any potential batch-effect. Analysis was first conducted between all tumors and normal samples and, subsequently, stratified analyses between tumors with recurrence and tumors without recurrence; tumors with systematic recurrence and tumors with local recurrence; and combined tumors with systematic and local recurrence and tumors with biochemical recurrence. Multiple testing was adjusted using false discovery rate q value as described (24). CpGs with a P value <0.05 and mean methylation difference between 2 contrast groups greater than 5% were initially selected as candidate markers. All statistical analyses were conducted using β-value as a continuous variable unless specified otherwise. Unsupervised and supervised hierarchical clustering analyses were conducted with the heatmap.2 function in the gplots library. Unsupervised clustering was used to characterize methylation patterns in an unbiased fashion, as conducted in other studies using methylation arrays. Supervised clustering analysis was used to further investigate methylation differences observed in unsupervised clustering. Additional evidence to support the delineation of clusters was obtained through unsupervised principal component analysis. The frequency and level of CpG methylation across different clusters were compared using a 2-sample proportion test based on both binarized and continuous β-values. The association of clinicopathologic and molecular variables with each cluster was analysed using continuous β-values and the 2-sample proportion t test. To estimate the HR of recurrence as a result of methylation change, we conducted a univariate analysis using Cox proportional hazard model. To evaluate whether the methylation can have an independent predictive value for recurrence, we incorporated Gleason score and pathologic stage into the Cox proportional model for multivariate analysis. All statistical analyses were conducted in R version 2.7.1 (The R Foundation for Statistical Computing) at 5% significance level unless otherwise stated. Where applicable, Bonferroni correction was applied to adjust for multiple testing. Differentially expressed genes were evaluated for biologic functions using Ingenuity Pathway Analysis (Ingenuity). Sensitivity and specificity of the genes were determined by receiver operator characteristic (ROC) curve analysis using the GraphPad Prism 4 Software. The area under the ROC curve and the methylation threshold yielding the optimal sensitivity and specificity were calculated for each DNA methylation marker.

Pyrosequencing assays were developed to confirm the DNA methylation of genes observed in the microarray profiling. We selected the genes from each group and validated the methylation levels in an independent set of 20 samples representing each group. One microgram of genomic DNA was bisulphite-converted as described above. For pyrosequencing, a 2-step PCR was carried out using specially designed forward and reverse primers using PyroMark Assay Design 2.0 software (Qiagen). The sequencing primers were designed to analyze the target CpG site used in the Infinium methylation microarrays. PCR primers, annealing temperature used for PCR amplification, and the sequencing primers are listed in Table 3. One of the 2 primers in the PCR amplification of the target region was biotinylated to further select the strand for pyrosequencing. PCR amplifications were carried out in a final volume of 25 μL using PyroMark PCR kit per manufacturer's procedure (25). For pyrosequencing analysis, 10 μL of the PCR product and 3.5 nmol of the target-specific sequencing primer, PyroMark Gold reagents, and the PyroMark MD instrument (Qiagen) were used following the supplier's protocol. A gene was considered as methylated if the methylation density was greater than 10%. The CpG site methylation data were obtained from 20 cases and 20 controls for each group. ROC curves were generated for each DNA methylation marker; these were summarized by area under the curve (AUC). Significance of methylation of the individual CpG site as a marker for the given group was analyzed using Student t test. P value >0.05 was considered statistically significant.

To determine the prevalence of methylation changes in a spectrum of prostate cancer disease ranging from indolent tumor to aggressive metastatic disease, we measured the methylation in a total of 238 samples (198 prostate cancer tissues and 40 matched normal tissues) using Illumina Infinium Methylation27 bead arrays. Clinicopathologic characteristics of 198 prostate cancer tissues are listed in Table 1. Each Human Methylation27 Bead Array enabled analysis of 12 samples. Investigation of the quantile-normalization at signal intensities of the total 238 samples visualized using a principal components analysis plot revealed 2 distinct sets of samples. Samples run on 4 microarrays (n = 46, which include 40 tumor and 6 normal) showed batch-effect from that of the other microarrays (n = 192, which include 158 tumor and 34 normal; ref. 26). The average β signal intensities from the 4 arrays were higher from the rest of the arrays. Therefore, we first analyzed the data obtained from 192 samples, then analyzed the results in the 46 samples with the batch-effect. In our first analysis, we grouped 192 prostate tissues on the basis of clinical outcome. For group 1, we compared tumor tissues (n = 158) versus normal tissues (n = 34) to identify the markers for early detection of cancer. For group 2, we compared tissues from patients with disease recurrence (n = 98) versus tissues from without disease recurrence patients (n = 60) to identify markers associated with recurrence of cancer. For group 3, we compared tissues from patients with clinical recurrence (n = 59) versus biochemical recurrence (n = 39) to distinguish increasing PSA without metastases or symptoms from metastasis. For group 4, we compared tissues from patients with systemic recurrence (n = 23) versus those with local recurrence (n = 36) to identify the markers associated with metastatic spread. The genes were initially selected using a P value <0.05 and a mean difference between 2 contrast groups greater than 5%. We further narrowed down the selection using a combination of P value and fold change in methylation between the 2 comparison groups. With a fold change cutoff of >2.0 and a P value of <8.67E-25, there are 147 genes in group 1. In groups 2, 3, and 4 with a fold change cutoff of >1.5 and with P values of <0.002, <0.05, and <0.01 respectively, there are 75, 16, and 68 genes significantly methylated. The list of genes and their major biologic functions are shown (Supplementary Table S1A–S1D). Methylation levels of 25 genes that showed increased frequency of methylation in cases compared with controls in the 1, 2, and 4 groups, and 7 genes for group 3, were depicted in the cluster diagram (Fig. 1A–D). To assess the potential use of hypermethylation of genes as molecular markers of prostate cancer, we determined the optimal sensitivity and specificity of the individual genes. The sensitivity and specificity of the 25 most differentially methylated genes for group 1, 2, and 4, and 7 genes for group 3, are shown in Table 2A–D. Genes listed in Table 2A were highly sensitive and specific to distinguishing prostate cancer from normal prostate tissues. Each of these markers had area under the ROC curve of more than 0.9 and sensitivities and specificities between 87% and 100%. Genes listed in Table 2B were methylated in prostate cancer tissues of patients with recurrence of the disease after primary therapy compared with those of patients without disease recurrence. These genes had an area under the ROC curve between 0.63 and 0.86, and sensitivities and specificities between 52% and 86%. There were fewer genes that were methylated in patients with clinical recurrence, compared with patients with only biochemical recurrence. The area under the ROC curve for these genes was between 0.66 and 0.70, with sensitivities and specificities between 47% and 84%. Finally, we assessed the genes that were methylated in patients with systemic recurrence compared with those with local recurrence. The genes with systemic recurrence showed an area under the ROC curve between 0.68 and 0.93, with sensitivities and specificities between 55% and 91%.

Risk of recurrence was estimated for methylation change by univariate analysis using Cox proportional hazard model by pooling all recurrence types [pooled (n = 98), which include biochemical (n = 39), local (n = 36), and systemic recurrence (n = 23); as the survival curves were almost similar for the three recurrence types]. The event was prostate cancer recurrence and those without the event (nonrecurrence, n = 60) were censored at the time of last follow-up. The methylation average β-values were transformed to the scale of 1 to 20 by multiplying with 20 so that each unit of increment is equivalent to an increase of 5% methylation ratio. With a P value of <9.0E-05, the top 128 genes associated with recurrence from the univariate analysis are shown (Supplementary Table S2). It is intriguing to find that there are 17 of 25 genes presented in the Table 2B from the binary recurrence versus nonrecurrence analysis that are among the 128 genes from the univariate Cox model of analysis. These findings suggest that our binary analysis includes most of the genes associated with recurrence. To determine whether these methylation changes could have an independent predictive value for recurrence, we incorporated Gleason score and pathologic stage into the above mentioned Cox model and conducted multivariate analysis. With a P value <9.0E-05, there are 183 genes that predict recurrence risk (Supplementary Table S3). Similar to the univariate analysis, we found that 16 of 25 genes presented in the Table 2B from the binary recurrence versus nonrecurrence analysis are among the 183 genes from the multivariate Cox model of analysis which indicates that these genes could be independent predictors of recurrence regardless of Gleason score and pathologic stage. Taken together, our data suggest that methylation of these genes could play a vital role in prostate cancer progression.

Because of the batch-effects observed in the 4 microarrays containing 46 samples, we analyzed these separately. There were 6 matched normal, 15 nonrecurrent, 4 biochemical recurrent, 8 local recurrent, and 13 systemic recurrent cases in these samples. Because of the small number of tissues representing different groups, the data obtained from these 46 samples were used to analyze the sensitivity and specificity of the genes shown in Table 2A–D. These results were almost similar to those seen with the batch of the above 192 samples (Supplementary Table S4A–S4D]).

Because of the impracticability of validating all genes identified in the microarray, we focused on 14 genes to confirm their methylation in different groups of prostate cancer tissues. These genes were selected because of their increased frequency of methylation in cases compared with the controls in the respective groups. To confirm the methylation data, we further validated the methylation of genes in an independent set of 20 samples representing each group by pyrosequencing method. These samples were matched to the samples used in microarray analysis based on the age, PSA levels, Gleason score, TNM stage, and GPSM score. Clinicopathologic characteristics of a total of 80 prostate cancer tissues, which included 20 from each group, are listed in Supplementary Table S5. The bisulfite-converted DNA was used for PCR amplification using specific primers, and the PCR product was analyzed by pyrosequencing using gene-specific sequencing primers (Table 3). We selected the genes HIF3A, HAAO, RARβ, and GSTP1 for group 1; CRIP1, RUNX3, HS3ST2, FLNC, and RASGRF2 for group 2; PHLDA3, TNFRSF10D, RASGRF2, and ZNF135 for group 3; and BCL11B, POU3F3, and RASGRF2 for group 4. The same CpG site of the gene used in the microarray experiments was selected as the target CpG site for all the genes except RUNX3 and FLNC, to design the PCR primers and sequencing primers for pyrosequencing using the PSQ Assay Design 2.0 Software. For the genes RUNX3 and FLNC, PyroMark CpG assays (Qiagen) were used because of the difficulty in designing the primers for the target site used in microarray experiments by the PSQ Assay Design. Pyrosequencing was conducted in triplicates for each sample, and the representative pyrogram of the genes from each group of cases was shown (Fig. 2A–D). We determined the optimal sensitivity and specificity of methylation by ROC analysis. This analysis revealed that HIF3A, HAAO, RARβ, and GSTP1 had ROC AUC of greater than 0.99, with sensitivities and specificities of 84% to 100% for detecting prostate cancer. CRIP1, RUNX3, HS3ST2, FLNC, and RASGRF2 had AUC of greater than 0.66, with sensitivities and specificities of 55% to 75% for prediction of recurrence. PHLDA3, RASGRF2, and TNFRSF10D had an AUC of greater than 0.69, with sensitivities and specificities of 60% to 75% for prediction of clinical recurrence, whereas the ZNF135 showed lower AUC of 0.55, and sensitivity and specificity of 50%. BCL11B, POU3F3, and RASGRF2 had an AUC of greater than 0.70, with sensitivities and specificities of 60% to 75% for prediction of systemic recurrence (Table 4). Except for the ZNF135, which is less consistent with the microarray data, validation of all other genes supports the findings obtained from the microarray experiments, although the magnitude of gene methylation was not always the same using these 2 different methods.

We comprehensively examined the methylation of genes to identify biomarkers for detection of cancer and to predict clinical outcome of the disease in a total of 238 prostate tissues using Methylation27 BeadChips. Our methylation data from the Illumina Infinium microarray platform were subjected to batch-effects. The 46 samples ran on 4 beadchips had higher average detection P values (26), which had a significant impact on downstream analyses and result interpretation. Therefore, the data from these 46 samples were analyzed separately from the rest of 192 samples to remove the batch-effect on the analysis. Because of the low number of samples per group in the 46 samples, we used this data set to compare the methylation levels of genes selected from the 192 sample set. The results obtained from 46 samples were analogous to the 192 samples, although the power for statistical analysis was lower (Supplementary Table S4A–S4D).

To identify the genes for diagnosis of prostate cancer, we initially compared the prostate cancer tissues and normal adjacent tissues. There were 147 differentially hypermethylated genes in the tumor tissues with a P value <0.8.67E-25 and a fold change of greater than 2. To narrow down the list of genes as biomarkers for early detection, we concentrated on the top 25 hypermethylated genes (Fig. 1A), all of which show very high sensitivity and specificity for detection of prostate cancer (Table 2A). In agreement with earlier work for detection of prostate cancer, many of the genes found to be hypermethylated in our study were previously reported to be methylated in prostate cancer tissues (21, 27, 28). Many of these genes function are associated with protection of cells against oxidative damage. An increased vulnerability to genome-damaging stresses from electrophiles and oxidants could cause wide range of DNA lesions including base modification that attribute to inactivation of these genes by hypermethylation permitting prostate carcinogenesis (29). Several studies have shown PTGS2 methylation as a prognostic biomarker for prostate cancer (28, 30, 31). In our study, PTGS2 was significantly hypermethylated in tumor versus normal and systemic recurrence versus local recurrence groups with a P < 0.01 (Supplementary Table S6). Our data suggest that aberrant methylation of these genes occurred in the early stage of carcinogenesis and persisted even during the progression of cancer to metastasis. Because it is impractical to validate all genes identified in the current study, we selected GSTP1, RARβ, HIF3A, and HAAO genes from group 1 tumor tissues versus normal tissues for pyrosequencing. Many reports, including ours, have shown aberrant methylation of GSTP1 and RARβ in prostate cancer (21, 32). Our validation by pyrosequencing confirmed the methylation of GSTP1 and RARβ in prostate cancer. Because GSTP1 methylation is not specific to prostate cancer, we suggest that combination of multiple markers can be used to sensitively and uniquely identify prostate cancer. In exploratory analysis, we validated HIF3A, which plays a central role in the response of cells to hypoxia (33) and HAAO, which is associated with microsatellite stability (34). We found for the first time highly significant hypermethylation of HIF3A and HAAO genes (Table 2A) in prostate cancer tissues, which indicates that methylation of these genes could be an early event in prostate cancer development. Our results indicate that methylation of these genes is very homogeneous in different patients, consistently maintained during the progression to metastasis, and can be considered as effective biomarkers for prostate cancer detection.

Although many published studies have assessed the performance of candidate biomarkers in predicting time to relapse of prostate cancer following radical prostatectomy (35), very few markers were identified with limited sensitivity and specificity that can effectively identify those patients with prostate cancer with a high risk of early clinical progression or prostate cancer–specific mortality (15, 28, 36). Predicting the probability of recurrence will enable us to distinguish between patients who have an indolent cancer not requiring any form of treatment and patients with biochemical relapse who may have a more aggressive course requiring early intervention. To predict the relapse and identify the risk of progression, we compared all recurrence types with nonrecurrence. Genes found to be associated with risk of recurrence both by univariate and multivariate analyses (Supplementary Tables S2 and S3) were also significantly methylated in the recurrence versus nonrecurrence group comparison (Table 2B). Our univariate and binary analysis revealed similar set of genes associated with recurrence and that are further supported by multivariate analysis which suggests that these genes could be considered as independent predictive biomarkers of recurrence. We selected CRIP1, FLNC, RASGRF2, RUNX3, and HS3ST2 from group 2 for validation by pyrosequencing. In our previous study, we showed methylation of FLNC as markers for recurrence (21). FLNC is an actin cross-linking protein, our microarray analysis and pyrosequencing further confirm the methylation of FLNC as a biomarker for recurrence. RUNX3 (37) is reported to be methylated at 48% and 84% in different cohorts of patients with prostate cancer (15). HS3ST2 controls the rate of production of the critical active site 3-O-sulfate group on the antithrombin-binding site of anticoagulant heparin sulfate (38), CRIP1 is involved in intestinal zinc transport (39) and RASGRF2, is a tumor suppressor gene that plays a vital role in lymphocyte proliferation, T-cell signaling responses, and lymphomagenesis (40, 41). Tumor tissues of recurrent patients showed increase in the methylation of these genes. We show here for the first time that RASGRF2 is a significantly methylated gene in all 3 groups of recurrent prostate tissues. Our validation supports the findings from microarrays that there is an increase in the methylation of these genes in the prostate cancer tissues of patients who developed recurrence in contrast to patients with nonrecurrence. Because these genes are involved in signaling and transcription pathways, their inhibition by promoter methylation may plausibly have a role in prostate cancer progression.

To predict the outcome of hormone-refractory patients with prostate cancer who progress to clinical recurrence from PSA recurrence, we selected PHLDA3, RASGRF2, TNFRS10D, and ZNF135 from group 3, clinical recurrence versus biochemical recurrence patients for validation by pyrosequencing. TNFRSF10D is a TRAIL receptor with a truncated death domain, a biomarker to predict cancer invasion, inflammation progression (42), PHLDA3, a tumor suppressor protein is a p53-regulated repressor of AKT (43) and ZNF135 is a zinc finger protein located on 19q13 involved in cellular proliferation and differentiation (44). We have shown, for the first time, methylation of these genes in clinical recurrence cases. Except for ZNF135, which showed relatively low sensitivity, the other 3 genes showed significant increase in methylation in the patients with clinical recurrence, which supports the microarray data. Methylation of these genes could be considered as a supplementary approach to the histopathologic markers of cancerous lesions.

Local recurrence without systemic progression can be actively treated by salvage external beam radiotherapy (sEBRT; ref. 45). Systemic progression patients will not respond well to sEBRT. In our study, we have seen relatively high mortality in the systemic recurrence group. To confirm the genes associated with the risk of systemic progression (Table 2D), we validated BCL11B, POU3F3, and RASGRF2. BCL11B was repressed by methylation in chronic lymphocytic leukemia (46) and POU3F3 (47) was shown to be methylated in B-cell lymphoma (48). The sensitivity and specificity of RASGRF2 methylation by pyrosequencing are similar to that of microarray results. BCL11B and POU3F3 methylation is relatively low in the validation samples than in the microarray samples. This could be due to the heterogeneity of the metastatic prostate cancer tissues. For example, one previous study reported methylation of 16 genes in 91 metastatic prostate cancer tissues (28). The results from this study reveal that methylation of the genes does not correlate with the anatomic metastatic sites. The role of these genes in the metastatic progression need to be evaluated in future studies.

The data presented here are of significance because we provide a large unbiased list of potential targets of DNA methylation in a broad spectrum of prostate cancer tissues. It is possible that several of these genes could have potential tumor suppressor properties, and the study of these genes may provide further information on their roles in metastatic progression. It is therefore possible that systemic analysis of these genes may help in the development of new prognostic markers for prostate cancer and guide therapeutic timing in these patients. It is known that DNA methyltransferase inhibitors act, in part, through reversing aberrant DNA methylation and restoring gene expression. The genes identified in our study would be helpful in evaluating the outcome in the ongoing clinical trials with DNA methylation inhibitors (49).

In summary, our results revealed an expanded list of genes as biomarkers for diagnosis of prostate cancer. Although we validated a few genes from each group, future studies will evaluate the efficacy of other genes from each group, and a combination of multiple genes could serve as unique biomarkers for prediction of prostate cancer progression. We postulate that the alterations in methylation that occur in the early stages of tumor development (Table 2A) are very homogeneous and persist through the progression of the disease. These genes can be considered as the early detection biomarkers of cancer. Genes that were differentially methylated in the recurrent patients (Table 2B–D) appear to be more heterogeneous. During the metastatic progression, tumor cells acquire epigenetic changes that are distinct to each individual tumor. Taken together, these observations suggest that no single gene can likely predict the progression of all advanced prostate tumors. The altered methylation of genes associated with the recurrence of cancer can be considered as a supplementary approach to the histopathologic workup. A multidimensional approach, including clinicopathologic and molecular parameters, needs to be established for prediction of prostate cancer progression.

No potential conflicts of interests were disclosed.

This work supported by the grants: American Cancer Society RSG-09-175-01-CCE and U.S. Department of Defense W81XWH-09-1-0216.

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.