Aberrant DOCK2, GRASP, HIF3A and PKFP Hypermethylation has Potential as a Prognostic Biomarker for Prostate Cancer

Abstract

Prostate cancer (PCa) is a clinically heterogeneous disease and currently, accurate diagnostic and prognostic molecular biomarkers are lacking. This study aimed to identify novel DNA hypermethylation markers for PCa with future potential for blood-based testing. Accordingly, to search for genes specifically hypermethylated in PCa tissue samples and not in blood cells or other cancer tissue types, we performed a systematic analysis of genome-wide DNA methylation data (Infinium 450K array) available in the Marmal-aid database for 4072 malignant/normal tissue samples of various types. We identified eight top candidate markers (cg12799885, DOCK2, FBXO30, GRASP, HIF3A, MOB3B, PFKP, and TPM4) that were specifically hypermethylated in PCa tissue samples and hypomethylated in other benign and malignant tissue types, including in peripheral blood cells. Potential as diagnostic and prognostic biomarkers was further assessed by the quantitative methylation specific PCR (qMSP) analysis of 37 nonmalignant and 197 PCa tissue samples from an independent population. Here, all eight hypermethylated candidates showed high sensitivity (75⁻94%) and specificity (84⁻100%) for PCa. Furthermore, DOCK2, GRASP, HIF3A and PKFP hypermethylation was significantly associated with biochemical recurrence (BCR) after radical prostatectomy (RP; 197 patients), independent of the routine clinicopathological variables. DOCK2 is the most promising single candidate marker (hazard ratio (HR) (95% confidence interval (CI)): 1.96 (1.24⁻3.10), adjusted p = 0.016; multivariate cox regression). Further validation studies are warranted and should investigate the potential value of these hypermethylation candidate markers for blood-based testing also.

Keywords: DNA methylation; biomarker; diagnosis; epigenetics; prognosis; prostate cancer.

Figure 1

Selection of samples and biomarker candidates. Tissue sample selection for bioinformatics analyses. We selected male samples (n = 4072) with known disease status to ensure specificity for prostate cancer (PCa) versus other cancers and non-malignant tissues.

Figure 2

Flow chart of biomarker discovery process using 450K array data from the Marmal-aid database. Bioinformatics analyses: To identify CpG sites that may also be suitable for a future blood-based testing, we initially excluded CpG sites that displayed signs of methylation in blood cells (n = 876 samples). Next, we used a 2-stringed discovery approach to identify: 1) “Pan-cancer” biomarker candidates; we selected CpG sites with hypermethylation in PCa tissue samples and excluded CpG sites with hypermethylation in normal prostate tissue. 2) PCa-specific biomarker candidates; we selected CpG sites with hypermethylation in PCa tissue samples and excluded CpG sites with hypermethylation in other cancer types and/or in normal prostate tissue samples. This resulted in the selection of four “pan-cancer” and seven “PCa-specific” CpG sites suitable for the qMSP assay design.

Figure 3

Methylation levels (ß-values) for the 11 selected biomarker candidates in the blood cell samples (n = 876), normal prostate tissue samples (n = 81), other normal tissue samples (n = 634), prostate cancer tissue samples (n = 187) and other cancer tissue samples (n = 2294). Other normal/cancer sample types included 14 different groups (AML (acute myeloid leukemia)/ALL (acute lymphoblastic leukemia), bladder, CNS (central nervous system), colorectal, head and neck, liver, lung, lymphoma, kidney, melanoma, pancreatic, sarcoma, stomach and thyroid). * p < 0.001 (Wilcoxon Mann–Whitney test). The colored boxes indicate the 25–75th percentiles and the black horizontal lines indicate the median. Top whiskers are the 3rd quartile + 1.5 interquartile range and the bottom whiskers are the 1st quartile – 1.5 interquartile range. Round dots indicate outliers.

Figure 4

Diagnostic potential of selected methylation biomarker candidates. Receiver operating characteristics (ROC) curve analysis of the radical prostatectomy (RP) tissue samples (n = 197) as compared to the nonmalignant prostate tissue samples (28 AN and 9 BPH). AUC: Area under the curve.

Figure 5

Box plots show the DNA methylation levels (normalized to ALUC4) for each top candidate marker as compared to the CAPRA-S risk score. CAPRA-S scores ranged from 0–2 (low), 3–5 (intermediate) and ≥6 (high). p-value, Wilcoxon Mann–Whitney test. The colored boxes indicate the 25–75th percentiles and the black lines indicate the median. The top whiskers are the 3rd quartile + 1.5 interquartile range and the bottom whiskers are the 1st quartile–1.5 interquartile range. Dots indicate outliers.

Figure 6

Kaplan–Meier plots with biochemical recurrence (BCR) as an endpoint in the Danish RP cohort. For each marker candidate, the patients were divided into low and high methylation subgroups based on the ROC curve analysis of the BCR status at the 36 months follow-up after RP. p-value from the log-rank test.

Figure 7

External validation of prognostic potential of DOCK2 hypermethylation in The Cancer Genome Atlas (TCGA) cohort. Kaplan–Meyer plots with biochemical recurrence (BCR) as an endpoint. Division of patients into low and high methylation groups was based on the ROC curve analysis of the BCR status at 36 months after RP. P-value from the log-rank test.

Figure 8

Kaplan–Meier plots with biochemical recurrence (BCR) as an endpoint in the subgroup of Danish PCa patients with D’Amico high-risk. Patients were divided into low and high methylation groups for DOCK2 based on the ROC curve analysis of the BCR status at 36 months post-RP. p-value from the log-rank test.