Background
Prostate cancer (PCa) is the most incident male cancer in western countries, constituting the second most common cause of cancer and the sixth leading cause of death by cancer among men worldwide [
1]. For 2012, it was estimated that PCa alone accounted for 420,000 newly diagnosed cancer cases and 101,000 of all cancer-related deaths in European men [
2]. PCa is age-related and very heterogeneous, both molecularly and clinically, ranging from relatively indolent to highly aggressive. It is typically asymptomatic at its earliest stages, when adequate treatment is mostly curative, in contrast with its late diagnosis, which usually impairs a curative-intent therapeutic strategy [
3]. This led to the widespread use of serum PSA as screening tool for PCa. However, it is now commonly accepted that this entailed overdiagnosis and overtreatment, justifying the strong recommendation against PCa screening and prompting the search for more effective biomarkers [
4].
DNA methylation is a chemically stable and easily quantified alteration [
5]. We and others have previously reported on the use of quantitative promoter methylation of several protein-coding genes for early diagnosis and prognostication of PCa [
6]. Although several gene methylation panels have been then developed [
7,
8], both sensitivity and specificity must be perfected to allow for clinical translation.
MicroRNAs, a class of small (19–25 nucleotides) non-coding RNA, are involved in virtually all cellular processes and frequently deregulated in cancer cells [
9], although its abrogation due to aberrant promoter methylation has been seldom reported [
10]. Because this epigenetic alteration is likely to be highly cancer-specific, it might constitute an effective cancer biomarker. Thus, we aimed to explore the potential of microRNA-coding genes promoter methylation as diagnostic and prognostic biomarkers in PCa. Therefore, after genome-wide screening, a set of putative tumor-suppressor microRNAs (miR-34b/c, miR-129-2, miR-152, miR-193b, miR-663a and miR-1258) with increased promoter methylation levels in PCa compared to normal prostate tissues was identified and further validated in clinical samples.
Methods
Patients and samples collection
For the purposes of this study, three independent cohorts of PCa patients were defined.
PCa tissue samples were prospectively collected from 180 patients with clinically localized disease, consecutively diagnosed and submitted to radical prostatectomy (RP) from 2001 to 2006, at Portuguese Oncology Institute of Porto (Cohort #1). Fifteen control samples were obtained from cystoprostatectomy specimens with bladder cancer, not harbouring PCa nor prostatic involvement by urothelial carcinoma (morphologically normal prostate tissue, MNPT). After collection, tissue samples were fresh-frozen at −80 °C and subsequently cut in a cryostat for DNA extraction. Prostate biopsy samples were collected from 74 PCa suspects (elevated serum PSA), referred to Portuguese Oncology Institute - Porto from 2001 to 2003 (Cohort #2). In addition to standard diagnostic cores, a core was collected from the most suspicious area, fresh-frozen at −80 °C and subsequently cut in a cryostat for DNA extraction.
Voided urine samples from 95 PCa patients were collected from 1999 to 2002 (Cohort #3). The control set is composed of urine samples collected from 17 healthy donors and 29 patients without urological malignancy. Samples were centrifuged at 4,000 rpm for 20 min, washed in PBS 1× and the pellets were frozen at −80 °C.
Clinical data was retrieved from clinical charts. Survival data was collected for patients of Cohort #1 and of Cohort #2. Disease-specific survival (DSS) time was calculated as the time elapsed since diagnosis until death or the last follow-up. Disease-free survival (DFS) was calculated from the date of the radical prostatectomy or other curative treatment to the date of biochemical relapse, date of last follow-up, or death if relapse-free.
All patients enrolled (Tables
1 and
2) signed informed consent. This study was approved by institutional review board (CES-IPOPFG-EPE 019/08 and CES-IPOPFG-EPE 205/2013).
Table 1
Clinical and pathological data of tissue and urine samples used in this study
Clinicopathological data | MNPT | PCaa
| Controls | PCab
|
Patients, n
| 15 | 180 | 46 | 95 |
Median age, years (range) | 63 (45–80) | 65 (49–74) | 61 (58–77) | 64 (45–80) |
Median PSA (ng/mL) (range) | - | 8.3 (3.4-23.0) | - | 8.8 (3.5-20.4) |
Pathological Stage |
pT2 (%) | - | 96 (53.3) | - | 46 (48.4) |
pT3 (%) | - | 84 (46.7) | - | 49 (51.6) |
Gleason score |
<7 (%) | - | 56 (31.1) | - | 37 (39.0) |
≥7 (%) | - | 124 (68.9) | - | 58 (61.0) |
Table 2
Clinical and pathological data of Cohort #2 (prostate biopsies)
Median age, years (range) | 68 (49–85) |
Median PSA (ng/mL) (range) | 18.22 (4.52-542) |
Clinical stage |
T2 (%) | 48 (64.9) |
T3/T4 (%) | 26 (35.1) |
Gleason score |
<7 (%) | 30 (40.5) |
≥7 (%) | 44 (59.5) |
Follow up |
Median (months) (range) | 114.9 (10.3–170.1) |
Patients without remission (%) | 3 (4) |
Biochemical recurrence (%) | 29 (39.2) |
Death due to PCa (%) | 13 (17.6) |
Therapy |
Surgery (%) | 17 (23.0) |
ADT (%) | 35 (47.3) |
Radiotherapy (%) | 4 (5.4) |
ADT + Radiotherapy (%) | 17 (23.0) |
Radiotherapy + Criotherapy (%) | 1 (1.3) |
CAPRA Score |
Low-risk (0–2) | 7 (9.5) |
Intermediate-risk (3–5) | 26 (35.1) |
High-risk (6–10) | 41 (55.4) |
Nucleic acid isolation, bisulphite treatment, HumanMethylation 450 BeadChip and qMSP analysis
DNA extracted by phenol-chloroform as described elsewhere[
11] was chemically modified using sodium bisulfite with EZ DNA Methylation-Gold™ Kit (Zymo Research, CA, USA) according to manufacturer’s protocol.
HumanMethylation450 BeadChip (Illumina, USA) allowed for gene methylation profiling of tissue samples (5 controls and 25 tumors), using 500 ng of bisulphite-converted DNA, according to manufacturer’s instructions. DNA methylation levels were depicted as beta-values ranging from 0–1.
Validation of all candidates was performed by quantitative methylation using KAPA SYBR FAST qPCR Kit (Kapa Biosystems, MA, USA). All reactions were run in triplicates in 384-well plates using Roche LightCycler 480 II, with
β-actin (
ACTB) as internal reference gene for normalization. Primer sequences (Additional file
1: Table S1) were designed using Methyl Primer Express 1.0 and purchased from Sigma-Aldrich (MO, USA).
Statistical analysis
For HumanMethylation 450 BeadChip data, a threshold intensity with P-value ≤ 0.01 was considered for further analysis. To identify consistently differentially methylated CpG sites, Wilcoxon rank sum paired test was performed for normalized beta-values. P-values were adjusted using false discovery rate, and CpGs with P-values <0.05 were selected.
In Cohort #1, pathological variables were categorized [Gleason score (GS): <7 and ≥7; pathological stage: pT2 and pT3]. Kruskall-Wallis and Mann–Whitney U tests allowed for comparisons among three or more groups and between two groups, respectively. For multiple comparisons P values were adjusted according to Bonferroni’s correction. Spearman nonparametric correlation was performed to ascertain association between methylation and PSA serum levels.
In Cohort #1 and Cohort #3, receiver operator characteristics (ROC) curves were constructed by plotting true positive rate (sensitivity) against false positive rate (1-specificity) and area under the curve (AUC) was calculated to assess diagnostic performance. Biomarker validity estimates [specificity, sensitivity, positive predictive value, negative predictive value and accuracy] were determined using as cut-off the highest value obtained through ROC curve analysis [sensitivity + (1-specificity)].
In Cohort #1 and Cohort #2, DSS and DFS curves were built using Kaplan–Meier method and the prognostic significance of clinicopathological variables (clinical stage, GS and serum PSA in both cohorts, and CAPRA Score in Cohort #2) was assessed using log-rank test. CAPRA score values were categorized as 0–2 (low-risk), 3–5 (intermediate risk) and 6–10 (high-risk) [
12]. To test the prognostic significance of miR-34b/c and miR-129-2 promoter methylation, samples were categorized based on methylation levels of each miR (using percentile 75 as threshold) [
11]. A Cox-regression model comprising all variables (multivariable analysis) was constructed. SPSS Statistics 20 (IBM, NY, USA) was used for all statistical analyses and graphics were assembled using GraphPad 5 Prism (GraphPad Software, CA, USA).
P values <0.05 were considered statistically significant.
Discussion
PCa remains one of the most prevalent neoplasms and a leading cause of morbidity and mortality in men. Although PSA screening has decreased the number of men diagnosed with metastatic PCa, this was accomplished at the cost of overdiagnosis and overtreatment of a sizeable proportion of men carrying indolent/non-life threatening tumors [
13]. Thus, a strong recommendation against serum PSA-based PCa screening has been issued [
14], prompting the search for more effective biomarkers allowing for better risk stratification of PCa suspects. Herein, we aimed to tackle this clinical quest through discovery and preliminary validation of novel biomarkers for PCa detection and prognostication, using methylation analysis of microRNAs gene promoters.
Owing to our previous experience in DNA methylation analysis of PCa [
6,
11], we searched for altered methylation patterns at the promoter regions of microRNAs deregulated in PCa. This information was then used to develop novel biomarkers, instead of microRNA expression levels, as previously attempted by other researchers [
15]. Indeed, DNA methylation is easier to assess than microRNA expression, it is more specific and, importantly, more stable. Moreover, because microRNAs downregulation in cancer is more common than upregulation, it seemed likely that aberrant promoter methylation might constitute an underlying mechanism, similar to protein-coding genes [
16]. Although several strategies might be used to identify microRNAs putatively downregulated due to promoter hypermethylation, high-throughput technologies such as methylation-array analysis are able to simultaneously pinpoint putative candidates [
17] and the reliability of the results might be readily assessed through analysis of well-known hypermethylated loci. Indeed, results of the methylation array experiments confirmed the high prevalence of
GSTP1 and
APC promoter methylation (data not shown), as we previously demonstrated in PCa [
18]. To increase the likelihood of finding robust candidate biomarkers, we used stringent conditions based on high fold-variation of methylation levels between cancerous and non-cancerous tissue samples. From methylation-array analysis, six candidate microRNAs, putatively deregulated by promoter hypermethylation were identified. MiR-1258, miR-193b and miR-34b/c were the most promising candidates, displaying substantial PCa-specificity compared with other urinary tract tumors, an attractive feature for testing in bodily fluids. MiR-129-2 and miR-663a showed modest results and their inability to discriminate PCa from bladder cancer rendered it unsuitable for testing in urine samples.
Association between promoter methylation levels in tumor tissue samples and standard clinicopathological variables was also assessed. Higher miR-129-2 promoter methylation levels associated with higher GS and stage, suggesting prognostic value. MiR-34b/c, miR-663a and miR-1258 methylation levels also associated with pathological stage, but higher diagnostic performance underscores the potential for detecting PCa at early stages instead of prognostication, as we previously reported for
EFEMP1 promoter methylation [
19]. Nevertheless, in this series of radical prostatectomies (Cohort #1) higher miR-129-2 methylation conveyed independent prognostic information, although only for DSS. Importantly, these results are in line with previous observations concerning the association of higher gene promoter methylation levels with clinicopathological features of more aggressive disease [
11,
20].
Urine is a key sample to evaluate DNA methylation biomarkers for PCa, as it is readily collected and biomarkers are diluted to a smaller extent than in plasma, providing higher sensitivity[
21]. Nevertheless, the amount of DNA potentially deriving from prostatic cells is variable, usually low, entailing the use of a panel with limited number of biomarkers. Thus, only miR-34b/c, miR-193b and mir-1258, were tested in urine samples (Cohort #3). From these, Mir-193b was previously shown to be aberrantly methylated in PCa cell lines as well as in primary tumors, but no data is available regarding its performance as PCa detection biomarker [
22,
23]. Indeed, Mir-193b performed best, with high AUC, sensitivity, specificity and PPV, whereas miR-34b/c performance was more modest.
Intriguingly, miR-1258, which showed the best performance in tissue samples (Cohort #1), displayed a strikingly different result in urines as its methylation levels were higher in controls than in PCa patients. The reason for this discrepant result is not immediately apparent, but it might be due to high miR-1258 promoter methylation in non-epithelial cells, such as leucocytes, which are relatively more abundant in urine than in tumor tissue samples. Moreover, median miR-1258 promoter methylation levels in urines from PCa were substantially inferior to those of miR-193b, impairing the robustness of the assay. It should be recalled that, contrarily to other studies, the urine samples we used were not collected following DRE or prostatic massage, which are usually employed in an attempt to increase sensitivity. Studies dealing with PCa biomarkers in urine vary in the method of urine collection and the real impact of prostatic massage has never been evaluated [
24]. It could be argued that the distance from the peripheral zone to the urinary tract flow may render urinary based tests less sensitive, which would be an important issue since most malignancies arise from this zone. Nevertheless, studies on
PCA3 did not find a difference in the levels of this biomarker between patients with peripheral versus transitional zone PCa [
25,
26].
Currently, the performance of serum PSA and urinary
PCA3, the only biomarkers approved for clinical use is rather limited. The reported performance of serum PSA as PCa biomarker is somewhat modest, with AUC ranging from 0.54 to 0.70 [
27,
28]. Even other serum PSA-derived measurements, like PSA-density, free PSA percentage and PSA-velocity have not significantly improved performance [
28]. Nonetheless,
PCA3, which was reported to perform better than serum PSA both in urine and ejaculates but has not been approved for population-based screening, displays AUCs varying from 0.66 to 0.79 [
27‐
30]. Additionally, although miRs’ expression has been extensively investigated in liquid biopsies, available data for urine samples is rather limited. Nevertheless, an AUC of 0.74 was reported for miR-107 [
31] and simultaneous quantification of miR-107 and miR-574-3p in urine showed an AUC of 0.83, for PCa cancer detection [
32]. We should emphasize that in our dataset, urinary miR-193b promoter methylation (AUC = 0.96) outperformed not the only currently approved clinical biomarkers, but also the previously mentioned miRs, constituting a promising tool for non-invasive PCa detection.
Because a major goal of this study was to discriminate clinically aggressive from indolent PCa, it was critical to test the prognostic value of microRNAs in a pre-therapeutic setting, which was accomplished in series of prospectively collected prostate biopsies (Cohort #3). In univariable analysis, most standard clinicopathological parameters associated with DFS and DSS, clinically validating this dataset. The same was demonstrated for higher miR-129-2 and miR-34b/c promoter methylation levels. The CAPRA score, however, only associated with DSS but not DFS. This was unexpected as its determination at diagnosis associated with DFS in patients with clinically localized disease submitted to RP [
12]. Notwithstanding, our prostate biopsy series included PCa at diverse clinical stages, submitted to different therapeutic modalities (RP, radiotherapy, androgen-deprivation therapy), which might explain the apparent flaw of CAPRA score. In multivariable analysis, only clinical stage, amongst all clinicopathological parameters, retained independent prognostic value, both for DFS and DSS. Remarkably, high miR-34b/c promoter methylation levels also predicted shorter DFS and DSS, suggesting that it might constitute a useful PCa prognostic biomarker. These results suggest that high miR-34b/c promoter methylation levels identify clinically aggressive PCa, irrespective of disease extent at diagnosis.
It should be acknowledged that in spite of the excellent diagnostic performance of miR-193b promoter methylation in urine, additional patient sets must be tested. Furthermore, a larger cohort of patients submitted to biopsy and subjected to different therapies is required to further validate our observations. Ultimately, we plan to develop a multiplex assay to simultaneously assess miR-193b, miR-129-2 and miR34b/c promoter methylation, allowing for diagnostic and prognostic assessment of PCa suspects in a single analysis.
Acknowledgments
The authors are grateful to Mª Conceição Martins, BSc for the skilful technical support and would like to acknowledge to the Departments of Urology and Laboratory Medicine of the Portuguese Oncology Institute of Porto for their collaboration in urine collection.