We tested DNA methylation in more than 480,000 CpG sites using the HumanMethylation450 Beadchip in PCa samples and normal tissues. We successfully identified 28,735 differentially methylated CpG sites in PCa. Many of the identified DMCs were independent replications of previous findings of differential DNA methylation in PCa 
. In the first, 98.5% (1,157 out of 1,175) of common CpG sites that were present in our DMC list and the HM27 BeadChip showed significant methylation differences in the previous data set 
. In the second, a total of 1,014 out of 2,481 (40.9%) cancer-specific methylated regions that were identified using the M-NGS method in PCa, overlapped with our DMCs 
. When considering the huge differences in the practical CpG coverage of these two methods (M-NGS vs
. 1.7%), the concordance rate of 40.9% can be considered very high.
The distribution of DMCs in the HM450 showed a similar pattern to the distribution of DMCs in a colon cancer cell line 
. Sandoval et al. found that most hypermethylated CpG sites were located in CGI, CGI shore, and CGI shelf regions, while more than half of hypermethylated CpG sites were indentified in proximal promoter regions. Our data was consistent with that observation, showing that 69.5% of hyper-DMCs were located in CGI, CGI shore, and CGI shelf regions, while 52.4% of hyper-DMCs were indentified in proximal promoter regions.
We have noticed that a total of 6,907 CpG sites assayed by the HM450 BeadChip happen to be located within single nucleotide polymorphism (SNP) loci. A total of 336 CpG sites among the 28,735 DMCs that we identified were also located in SNP loci as indicated in Table S3
. Therefore special attention is required to interpret any observed associations with these 336 DMCs.
We chose to limit our analysis to the proximal promoter region because this region is well characterized for its effects of DNA methylation on gene silencing. This approach also allowed us to evaluate associations with gene expression changes based on an independent PCa data set, based on the assumption that inverse correlations between promoter methylation and gene expression may plausibly indicate a functional result of differentially methylated genes identified in PCa. This approach identified a total of 122 genes including seven known DNA methylation genes in PCa. As positive confirmation of our approach, GSTP1
, which is the most thoroughly studied hypermethylated gene in PCa was found to be hypermethylated in this analysis (). A total 7 out of 15 CpG sites in the HM450 BeadChip were identified as hyper-DMCs (range of mean methylation difference: 0.27 ~ 0.48) in the GSTP1
gene (Table S3
). Messenger RNA of the GSTP1
gene has been previously shown to be significantly down-regulated in primary PCa, and metastatic tumors show further down-regulation 
The aldehyde oxidase 1 (AOX1
) gene is involved in various metabolic pathways, including drug metabolism and generation of reactive oxygen species 
. Reduced AOX1 protein expression in chronic pancreatitis and an absence of AOX1 protein expression in pancreatic cancer have been reported 
. In addition, decreased AOX1 protein expression was detected in hepatocellular carcinoma and this deregulation of AOX1 expression was associated with tumor stage and metastatic status 
. Aberrant DNA hypermethylation of the AOX1
promoter region was recently reported in colon cancer and PCa 
. Within the context of these prior published data, our data further supports a role for aberrant DNA methylation in the AOX1
promoter of PCa tumors, while also providing the most comprehensive coverage of DMCs in this gene to date.
The observed differences of DNA methylation in the AOX1 promoter between PCa samples and normal tissues were remarkable. This promoter region showed a very consistent methylation pattern in 34 CpG sites spanning 400 base pairs in each phenotype. This genomic location is amenable to evaluation via other assays, such as Pyrosequencing or methylation-specific PCR.
The majority of PCa tumors (92.6%; sensitivity) were observed to have positive DNA methylation, with a mean methylation cut-off of 0.3 and 94.3% of normal tissues showing negative DNA methylation with the same cut-off. This sensitivity of AOX1
methylation was similar to other well-known methylated genes including GSTP1
that were tested using Pyrosequencing methods 
. This high sensitivity of AOX1
promoter methylation and huge differences of DNA methylation between PCa samples and normal tissues suggest the potential utility of this gene as a biomarker for PCa diagnosis. However, additional studies will be required before assessing the feasibility of this gene as a biomarker. Studies of AOX1
promoter methylation status in control groups and/or using biofluid samples such as serum and urine may be needed.
Our analysis also associated promoter methylation with reduced AOX1
gene expression in PCa samples compared with normal tissues and further decreased expression of this gene in metastatic tumor samples (Figure S5A
). This AOX1
gene expression was associated with Gleason score in PCa (P
0.012). PCa samples with higher Gleason scores showed lower AOX1
gene expression than samples with lower Gleason PCa (Figure S5A
). The HM27 methylation data showed hypermethylation in samples with higher Gleason PCa than in samples with lower Gleason PCa (P
0.039, Figure S5B
). However, our bisulfite sequencing data on the AOX1
promoter region did not show any association with clinicopathological parameters including age at the time of surgery, Gleason score, and TNM stage in PCa or normal samples (data not shown). The precise molecular mechanism of deregulated AOX1
in human carcinogenesis remains unknown and may need to be explored in the future.
Interestingly, AOX1 converts 5-hydroxyindoleacetaldehyde to 5-hydroxyindoleacetate, which is a significantly down-regulated metabolite in metastatic prostate cancers compared to primary tumor samples (Figure S6
. This metabolite 5-hydroxyindoleacetate is produced by two major enzymes, aldehyde dehydrogenase family proteins (ALDH2
etc.) and aldehyde oxidase (AOX1
) in tryptophan metabolic pathway 
. The expression of ALDH2
were significantly down-regulated in primary tumor samples and further down-regulated in metastatic tumors, similar to AOX1
expression (Figure S7A
, B, and C) 
. Expression of MAOB
, which catalyzes the conversion of serotonin to 5-hydroxyindoleacetaldehyde was also significantly down-regulated (Figure S7D
). This data supports the hypothesis that some portion of the tryptophan metabolic pathway is deregulated in PCa development and progression. However, the biological effects of down-regulated AOX1
or 5-hydroxyindoleacetate on PCa development may need to be further studied.
gene encodes spondin2, which is an extracellular matrix protein. Spondin2 protein has multiple biological functions including neuronal development and innate immune response 
. Significantly increased expression of SPON2
mRNA and protein has been previously reported in PCa 
. The biological function of the SPON2
gene in human carcinogenesis is not yet elucidated, but it was used as the target of antibody-based radiotherapy in PCa 
. This is the first time aberrant SPON2
hypomethylation was identified and an inverse correlation was observed between SPON2
methylation and SPON2
Our comparison of DNA methylation in PCa samples and normal tissues identified more DMCs than did the other comparisons groups we examined such as the Gleason score PCa groups. Comparison of PCa progression groups showed relatively small changes in global DNA methylation and these relatively small changes were observed in three studies that used a BeadChip technique, including the current study 
. Such evaluations may require more precise detection methods with much larger CpG coverage to detect DNA methylation changes in PCa progression.
The HM450 BeadChip tested 18 different CpG sites in the PPARGC1A
gene. A Gleason-associated DMC (cg12691631) was identified in a CpG site that is located within the promoter region of the PPARGC1A
gene. Among 18 CpG sites on the HM450, eight additional CpG sites showed weak associations with Gleason score (range of raw P
~ 0.03) and all of these CpG sites showed higher methylation in higher Gleason PCa than lower Gleason PCa. The protein encoded by this PPARGC1A
gene is a transcriptional coactivator that regulates the genes involved energy metabolism 
. This protein is known to interact with other nuclear receptors such as estrogen receptor α and androgen receptor 
. Furthermore, a genetic variant in this gene was associated with breast cancer 
. One CpG site in the HM27 data also showed weak association with Gleason score (P
0.01 with higher methylation in higher Gleason PCa). This may need to be confirmed amongst a larger number of samples.
Even though DMCs from non-promoter regions were excluded from this study, future studies should not ignore the importance of these DMCs. DNA methylation in non-promoter regions may play functional roles 
. DNA methylation of intragenic regions may regulate gene expression by functioning as alternative promoters 
. However, the correlation between DNA methylation of intragenic regions and gene expression is not clear because contradictory results have been obtained in different cells or tissues 
. It is also not clear whether DNA methylation of intragenic regions is a consequence of other molecular mechanisms 
. A total of 204 DMCs (81 genes) passed the same selection steps and it is possible that DMCs in non-promoter regions will have diagnostic potential as a biomarker. Therefore, many DMCs in these non-promoter regions need to be evaluated in the future.
In this study, we used only the gene-level expression data, without considering alternative promoters, in order to avoid over-complication of the analysis. However, it should be possible to perform exon-level analysis using higher coverage HM450 data and exon expression data. Such analysis may provide interesting results on the effect of DNA methylation on alternative splicing or alternative promoters within a gene.
In this study, we identified 122 differentially methylated genes that were functionally deregulated in prostate cancer. A quarter of the genes in this set (32 genes) were newly identified as aberrantly methylated genes in PCa. Some of these genes may have the potential to distinguish between tumor versus normal tissues, and thus could serve as a diagnostic biomarker.