Diagnostic and prognostic implications of microRNA profiling in prostate carcinoma

Abstract

This study aimed to investigate the microRNA (miRNA) profile in prostate carcinoma tissue by microarray analysis and RT-qPCR, to clarify associations of miRNA expression with clinicopathologic data and to evaluate the potential of miRNAs as diagnostic and prognostic markers. Matched tumor and adjacent normal tissues were obtained from 76 radical prostatectomy specimens. Twenty-four tissue pairs were analyzed using human miRNA microarrays for 470 human miRNAs. Differentially expressed miRNAs were validated by TaqMan RT-qPCR using all 76 tissue pairs. The diagnostic potential of miRNAs was calculated by receiver operating characteristics analyses. The prognostic value was assessed in terms of biochemical recurrence using Kaplan–Meier and Cox regression analyses. Fifteen differentially expressed miRNAs were identified with concordant fold-changes by microarray and RT-qPCR analyses. Ten microRNAs (hsa-miR-16, hsa-miR-31, hsa-miR-125b, hsa-miR-145, hsa-miR-149, hsa-miR-181b, hsa-miR-184, hsa-miR-205, hsa-miR-221, hsa-miR-222) were downregulated and 5 miRNAs (hsa-miR-96, hsa-miR-182, hsa-miR-182*, hsa-miR-183, hsa-375) were upregulated. Expression of 5 miRNAs correlated with Gleason score or pathological tumor stage. Already 2 microRNAs classified up to 84% of malignant and nonmalignant samples correctly. Expression of hsa-miR-96 was associated with cancer recurrence after radical prostatectomy and that prognostic information was confirmed by an independent tumor sample set from 79 patients. That was shown with hsa-miR-96 and the Gleason score as final variables in the Cox models build in the 2 patient sets investigated. Thus, differential miRNAs in prostate cancer are useful diagnostic and prognostic indicators. This study provides a solid basis for further functional analyses of miRNAs in prostate cancer.

MicroRNAs (miRNAs) are small noncoding RNAs with a length of approximately 22 nucleotides. They regulate gene expression by mRNA cleavage and at posttranscriptional level by translational suppression. They play important roles in various biological and metabolic processes, including development, differentiation, signal transduction, cell maintenance, diseases and cancers. Bioinformatic predictions indicate that miRNAs regulate ∼30% of all protein coding genes.1 It is estimated that approximately 1,000 miRNAs exist in the vertebrate genome.2 So far, 695 human miRNAs are registered at miRBase release 11.0 (http://microrna.sanger.ac.uk/).

The first association of miRNAs and tumor biology has been described by downregulation of miR-15/-16 in B cell chronic lymphocytic leukemias.3 Ever since, miRNA deregulation has been observed in a large variety of tumors including pancreas,4 breast,5 ovary,6 colon,7, 8 lung8 and other solid tumors.8, 9 For some of these tumors, an association of miRNA expression with clinical data and outcome was described.4, 10

To date, few articles investigated miRNA regulation in prostate cancer. Only 5 studies examined miRNA expression in more than 10 samples11–15 with highly inconsistent results.16 Moreover, only few of these microarray data were validated by quantitative real-time reverse transcription PCR (RT-qPCR). A solid analysis of miRNAs in prostate cancer was pending, so far. Also, known associations of miRNA expression with clinicopathological data in prostate cancer was only observed for hsa-miR-146a and hsa-miR-184 for highgrade tumors.17 Only 1 study investigated the association of miRNAs with clinical follow-up data and found a characteristic expression pattern.14

In this study, we aimed to investigate the miRNA profile in malignant prostate tissue in comparison to matched normal prostate tissue by microarray analysis and RT-qPCR, to analyze associations of miRNA expression with clinicopathological data, to evaluate the potential of miRNAs as diagnostic and/or prognostic markers and thus to provide a solid data basis for further functional analyses of tumor-relevant miRNAs.

Material and Methods

Patients and tissue samples

Tumor tissue and normal adjacent tissue from 2 groups of 76 and 79 men with untreated prostate carcinoma were collected after radical prostatectomy between 2001 and 2005 (Table 1). Samples of patient set 1 were used for microarray analyses and RT-PCR measurements of miRNAs, while samples of the patient set 2 served afterwards to validate the prognostic implication of hsa-miR-96 expression discovered with the samples of patient set 1. The selection of samples was done by the availability of cryo-preserved tissues. For each patient, the following clinicopathological information was gathered: age, preoperative prostate-specific antigen (PSA), tumor classification according to the UICC 2002 TNM System,18 tumor grading according to Gleason based on the whole specimen,19 follow-up time after surgery and PSA concentration during follow-up.

Table 1. Patients and tumor characteristics

The study was approved by the ethical board of the hospital. Fresh prostate tissues were sampled directly after surgical removal of the gland. One full frontal section, which was grossly tumor suspicious, was deep frozen in liquid nitrogen. A diagnostic H&E section was prepared to verify tumor content and margin status and to identify areas of normal and tumor tissue. These regions of interest were punch biopsied, and then another frozen section was made to ascertain tumor content and to assign a Gleason score. Only cases with more than 90% tumor tissue were considered for further analysis.

RNA extraction

Frozen matched malignant and nonmalignant samples were collected in RNAlater Stabilization reagent (Qiagen GmbH, Hilden, Germany). RNA was extracted with the miRNeasy Mini Kit (Qiagen GmbH, Hilden, Germany). RNA yield and A260/280 ratio were monitored with a NanoDropND-100 spectrometer (NanoDrop Technologies, Wilmington, DE), and RNA integrity numbers were measured with the 2100 Bioanalyzer (Agilent Technologies, Santa Clara, CA). Only RNA extracts with RNA integrity number values >6 were included in further analysis.

MiRNA microarray experiments

Microarray experiments were carried out as recently described using 1-color hybridizations on human catalog 8-plex 15K miRNA microarrays (AMADID 016436; Agilent) encoding probes for 470 human and 64 viral miRNAs from the Sanger database v9.1 (Supporting Information S1).20 Gene expressions measures are available at GEO (http://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE14857).

MiRNA quantitative real-time RT-PCR

For detection of mature miRNAs, a two-step RT-qPCR was carried out using the TaqMan MiRNA Assay (Applied Biosystems, Foster City, CA) according to the manufacturer's protocol as previously described (Supporting Information S2).20 Samples were measured in triplicates, and a no template control and 2 interplate controls were carried along in each PCR run. To minimize the analytical variation, paired malignant and nonmalignant samples were always analyzed on 1 PCR plate. Analytical precision was assessed by intrarun measurements (n = 5) for each miRNA and ranged from 0.24 to 0.71% for crossing-point (Cp) values between 24.61 and 30.6 (Table in Supporting Information S2).

GenEX software (MultiD Analyses AB, Göteborg, Sweden) was used to analyze and normalize the RT-qPCR data.21 This program allows the correction of PCR efficiencies, the compensation for differences between runs by normalizing with interplate calibrators and the normalization with endogenous reference genes. Formulas for normalization are given in the Supporting Information S2. In addition, fold changes in gene expression in tumor samples normalized to an endogenous reference gene and relative to the normalized expression in nonmalignant samples were calculated (Supporting Information S2). As explained later, hsa-miR-130b was used as endogenous reference gene.

Data analysis and statistical methods

Microarray data analysis was performed as previously reported (Supporting Information S1).20

Statistical analyses of RT-qPCR data were performed with SPSS version 17.0 (SPSS, Chicago, IL), GraphPad Prism version 5.01 (GraphPad Software, La Jolla, CA) and MedCalc version 10.3.2 (MedCalc Software, Mariakerke, Belgium). Kolmogorov–Smirnof normality test, Wilcoxon test with combined Monte-Carlo analysis of 10,000 random samples, Mann–Whitney U test and Spearman rank correlation coefficients were used. All tests were performed two-tailed and p-values <0.05 were considered statistically significant. Receiver operating characteristics (ROC) curves with combined bootstrap analysis of 10,000 random samples were calculated to determine the potential of miRNAs to discriminate between malignant and nonmalignant samples.22 Binary logistic regression with subsequent cross validation was performed to identify the best discriminating combinations of miRNAs and to calculate the percentage overall correct classifications. For disease progression analyses, the Kaplan–Meier approach (log-rank test) and Cox proportional hazard regression analysis were used.

Sample size determinations were performed using GraphPad Statmate, version 2.0 and MedCalc on the basis of a two-sided alpha error of 5% and a power of 80%. Considering a proportional difference of about 0.40 in Kaplan–Meier curves as observed in patients after radical prostatectomy depending on the main risk factors Gleason score and tumor stage23, 24 about 25 subjects in each group have to be investigated. We decided to study at least 70 patients in the follow-up to realize a power of at least 80%, because the sample size between the groups of tumor recurrence or no recurrence could not be predicted.

Results

Patients and tumor characteristics

The main information is summarized in Table 1. The clinical and tumor characteristics did not differ between the 2 patient sets.

Microarray expression data

Analysis was performed on 24 matched malignant and normal adjacent tissue samples as described previously.20 A group of 78 candidate reporters was detected as differently expressed and was used for principal component analysis (PCA) on intensity profile levels to compress the multidimensional data to lower dimensions. The PCA plot (Fig. 1a) visualizes that samples aggregate to groups which are not clearly separated from each other. Four malignant samples are assigned to the normal group, whereas 1 normal adjacent tissue sample is assigned to the group of malignant samples.

Figure 1. Unsupervised analysis of 24 matched malignant and adjacent normal prostate tissue samples based on miRNA expression profiles using the 8-plex 15 K miRNA microarray (Agilent AMADID 016436). (a) Identification of malignant samples (green) and normal samples (pink) by PCA. (b) 2D cluster analysis across intensity profiles (left) and miRNA reporters (top). Hierarchical trees (top and left) and a heat map (bottom) are displayed. Green squares encode for downregulated and red items for upregulated miRNAs.

Furthermore, a hierarchical 2D-Cluster analysis using the similarity measure Pearson correlation and the heuristic criteria average link was applied without any statistical cuts on all 48 intensity profiles and 78 reporters derived from the primary data analysis step (Fig. 1b). The z-score was cropped to −3 to +3 in the heat map. A dendrogram of miRNA reporters is displayed on top, and the intensity profile dendrogram is displayed on the left of the heatmap. The heatmap assorts to clusters but a sharp segregation of up and downregulated miRNAs is not possible. The dendrogram of intensity profiles affirms the PCA results as the same 4 malignant samples are assigned the nonmalignant samples, and 1 nonmalignant is assigned to the malignant samples.

To render the analysis more robust and to reduce the high number of differentially identified results, reporters comprising individual miRNAs were combined into single values and a cut-off at 1.5-fold absolute change between malignant and matched nonmalignant ratio experiments was applied. In all 24 sample pairs, we identified 10 miRNAs that were regulated over 1.5-fold change (Table 2); 4 miRNAs were downregulated and 6 miRNAs were upregulated in prostate cancer specimen. Furthermore, we identified 5 candidates (hsa-miR-149, hsa-miR-181b, hsa-miR-368, hsa-miR-524*, hsa-miR-634), which passed selection criteria in at least 12 of 24 sample pairs. We combined both results to assign miRNAs for validation by RT-qPCR and identified 15 candidates (Table 2).

Table 2. miRNAs differentially expressed in malignant to matched nonmalignant tissue samples of prostate cancer

Validation of differentially expressed miRNA by RT-qPCR analysis

On the basis of these microarray experiments and an additional literature search9, 12, 13, 17, 25, 26 that identified 4 differentially expressed miRNAs (hsa-miR-125b, hsa-miR-145, hsa-miR-184, hsa-miR-373), the expression levels of 18 miRNAs and 3 putative reference genes (RNU6B, Z30, hsa-miR-16) were subsequently validated by RT-qPCR measurements (Table 2).

At first, to determine if expression of these miRNAs is detectable by RT-qPCR 2 RNA pools (malignant and nonmalignant) were prepared containing equal amounts of RNA from 12 samples per group used in the microarray analysis (Supporting Information S2). We excluded miRNAs from further analyses, if Cp values were >35 in both pools (hsa-miR-368, hsa-miR-373, hsa-miR-634). Reference gene candidates were excluded if delta Cp was greater 0.6 (Z30). Therefore, 15 miRNAs and 2 references were measured in all 76 individual samples (Table 2).

The putative reference gene, hsa-miR-16, showed a 1.16-fold lower expression in tumor tissue (Wilcoxon test, p = 0.0003) and was therefore excluded, whereas the second reference gene candidate RNU6B as well as hsa-miR-130b primarily not considered as reference gene showed no significant expression changes (p = 0.478 and p = 0.905, respectively). An equivalence test with an accepted Cp value difference of 0.2 showed that only hsa-miR-130b could be regarded equivalently expressed in malignant and nonmalignant tissue and was therefore used to normalize subsequent miRNA analyses.27

After data analysis of Cp values by GenEX and the normalization to the reference gene hsa-miR-130b, all measured miRNAs showed a significantly different expression between normal and tumor tissue (Table 2). Ten miRNAs were downregulated, and 5 miRNAs were upregulated.

The correlation between the expressions of the measured miRNAs was analyzed by Spearman rank correlation (Supporting Information Table SI). The correlation coefficients (r_s) ranged from −0.027 to 0.876 (p = 0.749 to p < 0.0001) with 1 major cluster including hsa-miR-96, hsa-miR-182, hsa-miR-182*, hsa-miR-375 with r_s ranging from 0.679 to 0.876 (p < 0.0001). Except for hsa-miR-375, these miRNAs are located in the same genomic region on chromosome 7. Expression of hsa-miR-221 and hsa-miR-222 also correlated strongly with each other (r_s = 0.848, p < 0.0001). Both miRNAs are expressed in close distance on chromosome X.

Association of miRNA expression with clinicopathologic data

Significant correlations with Gleason score was only observed for hsa-miR-31 (r_s = −0.277; p = 0.016), hsa-miR-96 (r_s = 0.267; p = 0.020) and hsa-miR-205 (r_s = −0.245; p = 0.033) (Supporting Information Table SII). Only hsa-miR-125b (r_s = −0.269; p = 0.019), hsa-miR-205 (r_s = −0.274; p = 0.017) and hsa-miR-222 (r_s = −0.232; p = 0.044) significantly correlated with tumor stage (Supporting Information Table SII).

MiRNAs as diagnostic markers

ROC analyses were performed to evaluate the capability of miRNAs to discriminate between normal and tumor tissue using samples from patient set 1 (Table 3; Supporting Information Fig. S1). The best single miRNA was hsa-miR-205 with an area under the ROC curve (AUC) of 0.82 and a correct overall classification of 72%. By combining all miRNAs using the logistic regression approach, correct classification was increased to 82% (AUC = 0.86) in comparison to hsa-miR-205, but the AUC did not differ significantly (p = 0.29). Logistic regression with all miRNAs using a backward elimination approach resulted in a pattern of 6 miRNAs (hsa-miR-96, hsa-miR-149, hsa-miR-181b, hsa-miR-182, hsa-miR-205, hsa-miR-375) and an overall correct classification of 80% (AUC = 0.88). If the best discriminating up and downregulated single miRNAs (hsa-miR-183, hsa-miR-205) were combined, an overall correct classification of 84% (AUC = 0.88) was obtained with a tendency to an improved discriminative power compared with hsa-miR-205 alone (p = 0.062) but not different to that with all miRNAs (p = 0.486) (Supporting Information Fig. S1).

Table 3. Performance of miRNAs to discriminate between malignant and nonmalignant tissue samples from prostate cancer of patient set 1

MiRNAs as prognostic markers

Seventy-five clinically characterized prostate cancer cases had complete follow-up data and were included in the disease recurrence analysis. Median follow-up time was 50 months (range: 1–93). The time of recurrence as the primary end point of this analysis was defined as the first postoperative PSA value >0.1 μg/l PSA confirmed by at least 1 subsequent rising value after the patients had reached an undetectable PSA level (detection limit <0.04 μg/l) after surgery. Twelve patients experienced a biochemical relapse according to this criterion.

Performance of prognostic parameters was tested by Kaplan–Meier analysis and Cox proportional hazard regression. The recurrence-free interval was significantly reduced with increasing pT stage (p = 0.006) and Gleason score (p = 0.004) showing that our study group was representative (Figs. 2a and 2b; Table 4). The Gleason scores of the whole specimens and those of the tissue sections used for the miRNA analysis were strongly correlated (intraclass correlation coefficient of 0.73). For analysis, miRNA expressions were dichotomized by the median. The recurrence-free interval was significantly decreased in patients with high hsa-miR-96 expression in the tumor samples (p = 0.038) (Fig. 2c; Table 4). All clinicopathologic parameters and miRNA variables were assessed regarding their prognostic performance in multivariate Cox regression full and stepwise reduced models (Table 4). Hsa-miR-96 remained as miRNA prognostic indicator alone together with the Gleason score in the selective model. Although hsa-miR-96 was not a fully independent variable (p = 0.052), the −2 log-likelihood value was significantly reduced in comparison to the beginning −2 log-likelihood (χ² = 14.22, p = 0.001, df = 2) and in comparison to a model with Gleason score alone (χ² = 4.48, p = 0.034, df = 1) showing the essential contribution of the hsa-miR-96 expression in this model (Table 4). The c-index, which gives the probability that the Cox regression model predicts the correct order of survival times,28 was 0.85.

Figure 2. Kaplan–Meier analysis of recurrence free survival according to (a) pathological stage, (b) Gleason score and (c) hsa-miR-96 expression. The time of recurrence was defined as the first postoperative PSA value >0.1 μg/l PSA confirmed by at least 1 subsequent rising value after the patients had reached an undetectable PSA level (detection limit <0.04 μg/l) after surgery.

Table 4. Univariate and multivariate Cox proportional hazard analysis of clinicopathologic parameters and differentially expressed miRNAs in patients with prostate cancer with regard to the recurrence-free interval after radical prostatectomy

The prognostic impact of hsa-miR-96 was validated in a second cohort of 79 patients (Table 1). In univariate Kaplan–Meier analysis, cases with higher hsa-miR-96 expression had again significantly higher risk for biochemical recurrence (p = 0.039) as well as cases with high Gleason score (p = 0.004) and high pT stage (p < 0.0001). As Gleason score and hsa-miR-96 were identified as covariates for a survival model in the previous sample set, this model was applied on the validation cohort. Again, the −2 log-likelihood was significantly reduced by this model (χ² = 21.9, p < 0.0001, df = 2) and in comparison to the model with Gleason score alone (χ² = 4.26, p = 0.039, df = 1). The hazard ratios were 3.55 (95% CI: 0.95–13.3, p = 0.06) for hsa-miR-96 and 6.16 (95% CI: 2.26–16.8, p = 0.0004) for the Gleason score. These data confirmed the prognostic significance of hsa-miR-96 already shown in the study with the first set of samples, although hsa-miR-96 was again not clearly indentified as independent variable.

However, it should be mentioned that taking together the data of the 2 sets both the Gleason score and hsa-miR-96 were independent prognostic factors in the Cox regression model. That applies both to the models established with the Gleason scores assessed in the whole specimens and to those build with the Gleason scores of the samples for the miRNA analysis because of the strong correlation between the 2 Gleason score values as mentioned earlier (whole specimen: Gleason score HR: 3.51, 95% CI: 1.81–6.77; p = 0.0002 and hsa-miR-96 HR: 3.20, 95% CI: 1.17–8.71; p = 0.023. Sample: Gleason score HR: 3.55, 95% CI: 1.98–6.36; p < 0.0001 and hsa-miR-96 HR: 4.38, 95% CI: 1.57–12.2; p = 0.005).

The predictive power of Cox regression model for hsa-miR-96 was additionally evaluated by comparing the observed survival curves with the expected survival curves.29 Because the observed and expected plots for both cohorts as well as these analyzed together are close to one another, it can be concluded from this graphical goodness-of-fit test that the proportional hazards assumption of the Cox regression model is satisfied (Supporting Information Fig. S2).

Target search and recognition

In silico identification of putative targets is an inevitable basis for further functional studies. The most widely-used algorithms are miRanda, picTar and TargetScan. We searched targets at miRecords (http://mirecords.umn.edu/miRecords) that allows searching with 11 algorithms in 1 step. The following criteria were used to refine results: (i) a putative target had to be identified consistently by miRanda, picTar and TargetScan and (ii) with at least 2 additional algorithms (Supporting Information Table SIII).

Discussion

We started our study by screening expression of 470 miRNAs by microarray in 24 matched pairs of histologically confirmed tumor tissue and normal adjacent tissue. MiRNA regulation was subsequently validated in RT-qPCR analyses of 76 tissue pairs and additionally in 79 tumor samples for the prognostic information of hsa-miR-96. Concordance between both platforms was good, but 1 miRNA (hsa-miR-130b), which was observed as regulated miRNA in the microarray analysis, was found equally expressed in RT-qPCR and could therefore be used as reference gene.

To date, there are only 5 studies of miRNA expression in prostate cancer in more than 10 samples with largely divergent results. No miRNAs were found concordantly expressed in all 5 studies11–15 and also our data do hardly overlap with that of other studies (Supporting Information Fig. S3).11, 13

Four downregulated miRNAs in our study (hsa-miR-16, hsa-miR-31, hsa-miR-181b, hsa-miR-184) were found to be upregulated in other miRNA studies in prostate cancer.11, 15 However, there are some discrepancies in study design that could promote these contradictory results. For example, the results are based only upon microarray measurements that partially detected pre-miRNAs,15 and the expressions were not validated by RT-PCR.11, 15 Furthermore, nonmatching normal tissue from healthy individuals was used, whereas we used paired normal adjacent tissue from the same prostate. We observed that the expression of miRNAs is partly heterogeneous in prostate cancer samples. This is in contrast to miRNA expression in other cancer types e.g. clear cell renal cell carcinoma.20 Therefore, using nonmatched tissues may significantly bias results because of interindividual variability of miRNA expression.

Downregulation of miR-16 has also been observed in prostate cancer samples13 and in primary cells received from patients with pT2 and pT3 tumors.30 Reconstitution of hsa-miR-16 levels in LNCaP cells and primary tumor cells led to growth arrest and apoptosis. BCL2 was identified as a direct target of hsa-miR-16.31 Downregulation of hsa-miR-31 has been observed in other cancer types yet, as human gastric tumors.32 Hsa-miR-181b has been found to be downregulated in glioblastomas33 but upregulated in colorectal cancer.34 The role of hsa-miR-184 in tumor cells is also contradictory in other studies. An increased expression of hsa-miR-184 with increasing Gleason score has been observed.17 Hsa-miR-184 was strongly upregulated in lung carcinoma and its inhibition led to decreased cell proliferation.35 In contrast, this miRNA was found to be strongly downregulated in MYCN-amplified tumors in comparison to other neuroblastoma subgroups.36

Our results show the upregulation of hsa-miR-96, hsa-miR-182* and hsa-miR-183 in prostate cancer for the first time (Table 2). These miRNAs also highly correlated with hsa-miR-182 by being transcribed as a single polycistronic primary-miRNA.37 Hsa-miR-96 was also found upregulated in primary cells from chronic myeloid leukemia38 and in hepatocellular tumors.39 Moreover, we found hsa-miR-96 to be associated with Gleason score and biochemical relapse. The Cox regression analysis in the 2 patient sets, in the first identification set and the second validation set, proved that increased hsa-miR-96 remained as prognostic indicator alone together with the Gleason score in the prediction model for disease recurrence (Table 4), although it was not a fully independent factor (p = 0.052). In addition, the results of the Kleinbaum approach29 and the c-index28 of 0.85 support the importance of hsa-miR-96 in this respect. The c-index corresponds to the AUC of the ROC curve and a model showing a c-index greater than 0.80 has been considered to provide useful information in predicting the outcome of patients.28 A characteristic expression pattern of miRNAs was recently reported in patients with biochemical relapse within 2 years after prostatectomy when compared with patients without disease recurrence after 10 years.14 Using a subset of 16 miRNAs, the authors achieved an overall correct classification of 80% between the patients in these 2 groups. A nonsignificant differential expression of hsa-miR-135b and hsa-miR-194 in patients with early prostate cancer recurrence was observed, but the Cox regression analysis was not performed. Thus, all these results support the view that miRNAs could become promising prognostic markers.

All other miRNAs are in concordance with prior studies in prostate cancer. Hsa-miR-125b, hsa-miR-145, hsa-miR-149, hsa-miR-205, hsa-miR-221 and hsa-miR-222 were downregulated in at least 1 other study, whereas hsa-miR-182 and hsa-miR-375 were upregulated in 1 other study each, respectively.

ROC analyses of miRNA expressions provided good tissue classification results. Even a single miRNA classified 72% of the samples correctly and combining hsa-miR-205 and hsa-miR-183 increased correctly classified samples to 84%. The results highlight that miRNAs have a great potential to serve as diagnostic markers. Expression profiles of mature miRNA have been published to be capable to classify poorly differentiated tumors.40 In addition, it has been shown that free circulating miRNAs are detectable in plasma of patients with prostate cancer highlighting their eligibility in clinical practice.41 Detection of miRNAs may also be possible in urine of patients with prostate cancer as it is applied for the PCA3 test.42

Some of the miRNAs are of special interest concerning their putative targets. To identify putative targets and to provide a foundation for functional analyses, we performed in silico analysis (Supporting Information Table SIII). Until now, only few miRNA targets have been validated in prostate cancer. Hsa-miR-125b, which we found to be downregulated and correlated with pathological stage, inhibits BAK1, an inducer of apoptosis which is increased in prostate cancer tissue and in androgen-independent cell lines.43, 44 Hsa-miR-221 and hsa-miR-222 expression negatively correlated with p27 expression in PC3 cells derived tumor xenografts45 and their upregulation promoted growth of LNCaP cells.46 However, the role of these 2 miRNAs is still contradictory as their downregulation was frequently observed in prostate cancer.11, 13, 14 Furthermore, decreased growth of LNCaP cells after ectopic expression of hsa-miR-221 or hsa-miR-222 was ascertained.14 One frequently identified target of hsa-miR-221 and hsa-miR-222 by in silico target research was ADP-ribosylation factor 4 (ARF4). ARF4 interacts with EGFR, thereby enhancing oncogenic processes.47 On the other hand, the loss of hsa-miR-205 shown in malignant prostate tissue in our study (Table 2) and by others11 as well as in other cancer types48, 49 is suggested a hallmark of epithelial-mesenchymal transition. Hsa-miR-205 targets SIP1 and ZEP which promote this transition.50 This target search for hsa-miR-205 also showed that one of the most frequently observed targets was the RAB11 family interacting protein 1 (class I) (Supporting Information Table SIII). This protein promotes migration of tumor cells.51 Taken together, these data strongly suggest that hsa-miR-205 may be involved in prostate cancer progression.

Although this study was conducted according to the “Reporting Recommendations for Tumor Marker Prognostic Studies”52 some limitations merit discussion. First, number of patients with biochemical relapse was relatively small, but corresponded to the failure rate generally observed.23, 24 Nevertheless, the prognostic value of Gleason score and pT stage could be reproduced indicating representativeness of our cohort. Further, recurrence-free interval after radical prostatectomy was significantly decreased with increasing hsa-miR-96 expression and that was confirmed in an independent second patient set. Therefore, the risk of a type II error can be neglected as far as possible. Because differences in miRNA expression between malignant and nonmalignant prostate cancer tissue are highly significant (p = 0.012 to p < 0.0001), type I errors could be similarly excluded for diagnostic predictions. The study is limited by its retrospective nature; however, all measurements were performed in a blinded manner.

In summary, this study on miRNA profile in prostate cancer provides clear evidence that miRNAs can be used as diagnostic and even prognostic markers in this disease which merits further prospective studies to verify these important findings. In addition, the differential miRNA profile affords a solid basis for further functional analyses of miRNAs in prostate cancer.