Dysregulated methylation at imprinted genes in prostate tumor tissue detected by methylation microarray

Procedures for participant recruitment have been described previously [10]. Briefly, study subjects were identified via the Yale Cancer Center Rapid Case Ascertainment system and all patients consented to the donation of tissue to the Yale-New Haven Hospital (YNHH) tissue bank. Samples were obtained according to protocols approved by the Research Ethics Board from YNHH, New Haven County, Connecticut and the Connecticut Department of Public Health Human Investigations Committee. Tissue sections from seventeen pairs of formalin-fixed paraffin-embedded (FFPE) prostate cancers and corresponding adjacent normal tissue specimens, obtained from patients who had undergone surgery between 2005 and 2009 at YNHH, were mounted on slides and examined by an expert pathologist. Gleason grades varied between specimens, with a composite score ranging from 6 to 9. No subjects who had received either chemo- or radio-therapy were included in the study.

Sections of tumor and adjacent normal tissue were pathologically reviewed, manually microdissected, and collected into 1.5 ml microtubes. The DNeasy Blood & Tissue Kit (Qiagen, Valencia, CA) was used to isolate genomic DNA according to the manufacturer’s protocols.

Equal amounts of DNA from tumor and matching adjacent normal tissue from twelve subjects (a sufficient amount of DNA was unavailable for five of the seventeen subjects) were combined by tissue type for CpG methylation microarray analysis. Methylation of imprinted genes was assessed using the Illumina Infinium HumanMethylation27 Array (Illumina, Inc., San Diego, CA). The CpG sites of the imprinted genes were located within promoter regions, ranging from 3 to 1,495 bp upstream of the transcription start site (average distance: 426 ± 373 bp). A methylation index (β) was obtained for each site, which is a continuous variable ranging between 0 and 1 representing the ratio of the intensity of the methylated-probe signal to the total locus signal intensity (a β value of 0 corresponds to no methylation while a value of 1 corresponds to 100% methylation at the specific CpG locus measured). Complete array data have been uploaded to the Gene Expression Omnibus (GEO) database (http://​www.​ncbi.​nlm.​nih.​gov/​geo/​; accession number GSE26319).

In order to confirm methylation microarray results we carried out quantitative DNA methylation analysis using Sequenom’s EpiTYPER assay (Sequenom, Inc., San Diego, CA). Methylation levels at WT1 were analyzed using tumor and adjacent normal tissue DNA from five subjects (non-pooled) for which DNA was of sufficient concentration. Analysis was conducted using pre-validated primers from the Sequenom’s Imprinting EpiPanel designed to target imprinting control regions of known imprinted genes (Amplicon WT1.ALT.TRANSCRIPT_05; Forward primer: GTAGGGGTTAGGGGAGGTAAAGT; Reverse primer: CCCAATCACAATACAACTACAATCA). Average methylation levels in tumor versus normal tissue were compared for individual CpG sites and for overall WT1 methylation.

We used the Oncomine expression profiling database version 4.4 (http://​www.​oncomine.​org; accessed March 3rd, 2012) to search for expression array comparisons between prostate cancer tissue and normal tissue (either from adjacent normal tissue or healthy controls) in five genes with the strongest evidence of LOI in our dataset. We searched the database for expression differences in human prostate cancer using the gene symbol as the keyword (e.g. “DLK1”).

The Infinium methylation data were analyzed using Illumina's GenomeStudio software, which employs a custom model to yield a DiffScore and p-value for each CpG site based on a comparison of the mean methylation level in tumor tissue versus that of adjacent normal tissue. To control for multiple comparisons, adjustments were made in order to obtain an adjusted P value (designed as the false discovery rate, or Q value) for each observation using the method originally proposed by Benjamini-Hochberg [11]. CpG sites were defined as differentially methylated if the Q values obtained were < 0.05. Average methylation levels derived from SEQUENOM EpiTYPER analysis were compared between tumor and adjacent normal samples using a two sample t-test.

Based on our analysis of a pooled sample of 12 pairs of prostate tumor and adjacent normal tissues, our results demonstrate an overall disruption of methylation patterns of imprinted genes in tumor tissue. Among 397 CpG sites analyzed in 56 imprinted genes that were covered by the Illumina Infinium HumanMethylation27 array, the methylation index (β) in adjacent normal tissues was 0.453 on average, and in prostate cancer tissue methylation was significantly higher at 52 sites (13.1%) and significantly lower at 17 (4.3%) sites (P < 0.05), jointly spanning 28 unique genes (Table 1). Average percent changes in the methylation index were calculated for each of the 56 imprinted genes, in which percent changes were averaged across the set of corresponding CpG sites for each gene (mean number of CpG sites/gene: 7.1 ± 6.2). Of the 56 genes, 13 demonstrated a higher average methylation of at least 5%, and 9 demonstrated a lower average methylation of at least 5% (Figure 1).

Results from five genes were particularly notable based on the magnitude of methylation change from adjacent normal tissue and the consistency in the direction of these changes across multiple CpG sites (Figure 2). CpG site cg17412258, 212 bp upstream of the DLK1 transcription start site, showed 4.32 fold higher methylation in the tumor samples relative to the adjacent normal tissue samples, with a 5% higher average methylation index for the other CpG sites. SLC22A18 demonstrated similar methylation changes, where 3.75 fold higher methylation was observed at CpG site cg03336167 with a tendency toward higher methylation at remaining sites. Three CpG sites in the promoter region of PLAGL1 showed statistically significant higher methylation (P < 0.05), including 3.27 fold higher methylation at cg17895149. Four CpG sites in the proximal TP73 promoter region demonstrated statistically significant higher methylation indices (1.44, 1.74, 1.97, and 2.28 fold higher methylation) (P < 0.05), while one site further upstream from the transcription start site (917 bp) showed statistically significantly lower methylation (P < 0.05). WT1 demonstrated the greatest consistency in methylation aberration, with increased methylation at 16 of 20 CpG sites. This increased methylation was statistically significant (P < 0.05) at eight CpG sites including 7.67 fold higher methylation at cg22533573. All five of the identified genes (DLK1, PLAGL1, SLC22A18, TP73, and WT1) have presumed tumor suppressing functions and taken together demonstrate a strong tendency towards higher methylation at the CpG sites assessed.

We sought to confirm the hypermethylation observed at WT1 using SEQUENOM’s EpiTYPER quantitative methylation assay. Results were consistent with the prior analysis by microarray, demonstrating roughly 2–4 fold hypermethylation across all eight CpG sites analyzed in the WT1 imprinting control region (including cg22533573, as discussed above) in comparing individual tumor and adjacent normal tissue pairs. Averaged across the eight analyzed CpG sites and five tumor-normal tissue pairs with adequate DNA for this confirmation, we observed a 2.04-fold increase in methylation in tumor tissue relative to adjacent normal tissue (average normal methylation level: 25.2%; average tumor methylation level: 51.4%; P = 0.0104).

We used the Oncomine expression profiling database to search for expression array comparisons between prostate cancer tissue and prostate tissue from healthy controls in DLK1, PLAGL1, SLC22A18, TP73, and WT1; expression data for all available studies are presented in Figure 3. Six of fourteen studies reported statistically significant (P < 0.05) expression changes for DLK1, demonstrating lower expression of the gene in prostate cancer tissues relative to healthy control tissues in four studies and higher expression in two. These studies reported expression fold changes of −4.547 (n = 34) [12], -2.216 (n = 89) [13], -1.496 (n = 102) [14], -1.242 (n = 88) [15], 1.124 (n = 57) [16], and 1.350 (n = 21) [17], respectively. Of 38 studies, 27 reported significantly lower expression of PLAGL1 in prostate cancer tissue relative to healthy controls. These studies reported lower expression fold changes ranging from 1.143 [18] to 1.995 [17], and no study reported higher expression of the gene in tumor tissue. Five of thirteen studies were identified that showed significantly lower expression for SLC22A18, with fold changes of −1.734 (n = 13) [19], -1.411 (n = 21) [17], -1.400 (n = 26) [20], -1.341 (n = 52) [21], and −1.157 (n = 57) [16]. Four out of 22 studies showed significant WT1 expression changes, including two studies demonstrating lower expression fold changes of −1.315 (n = 102) [14], and −1.106 (n = 35) [22], and two studies showing significantly higher expression with a fold change of 1.327 (n = 21) [17] and 2.878 (n = 21) [17]. Of nine studies, one (n = 35) [22] demonstrated a significant change in expression of TP73, with a −1.218 fold change.