MicroRNAs (miRNAs) are small RNA molecules that regulate gene expression via posttranscriptional inhibition of protein synthesis. They play a vital role in tumorigenesis. To characterize the diagnostic potential of miRNAs in prostate cancer, a leading cause of cancer mortality, we performed screening of miRNA expression profiles. We used commercially available microarrays to establish miRNA expression profiles from a cohort of 20 cancer samples. The expression of selected miRNAs was analyzed by quantitative real-time PCR and the identity of miRNA expressing cells was determined by miRNA in situ hybridization. We identified 25 miRNAs that showed a significant differential expression in cancer samples. The comparison with previously published data generated by deep sequencing of cDNA libraries of small RNA molecules revealed a concordance rate of 47% among miRNAs identified with both techniques. The differential expression of miRNAs miR-375, miR-143 and miR-145 was validated by quantitative PCR. MiRNA in situ hybridization revealed that the differential expression is cancer-cell associated. A combination of three miRNAs correctly classified tissue samples with an accuracy of 77.6% with an area under the receiver–operator characteristic curve of 0.810. Our data extend the knowledge about the deregulation of miRNAs in prostate cancer. The differential expression of several miRNAs is highly consistent using independent cohorts of tumor samples, different tissue preservation methods and different experimental methods. Our results indicate that combinations of miRNAs are promising biomarkers for the diagnosis of prostate cancer.
MicroRNAs (miRNAs) are small, noncoding RNAs of about 21–25 nucleotides that bind to partially complementary sites in the 3′ untranslated region of mRNAs.1, 2 By this, miRNAs can induce the sequence-specific degradation of target mRNAs, mRNA destabilization by 5′ decapping, 3′ deadenylation or inhibition of protein synthesis.3–5 MiRNAs are involved in various biological processes like cell differentiation, cell cycle control, cell growth and immune responses.6 The first report of miRNA deregulation in tumor biology described the downregulation of miR-15/16 in B-cell chronic lymphatic leukemias.7 Both miRNAs target the mRNA for B-cell lymphoma 2 (BCL2) and reduction or deletion of these miRNAs leads to elevated levels of Bcl-2 and reduced apoptosis of B-cells.8 Ever since, a differential expression of miRNAs has been observed in a variety of human malignancies including breast,9, 10 colon,11 pancreas,12 liver13 and ovary.14 Furthermore, an association of miRNA expression and clinical outcome has been described for various human cancers.15–20
Concerning prostate cancer, several miRNA profiling studies have been published,21–26 but the results regarding the deregulation of particular miRNA genes were highly inconsistent.27 We have previously published deep sequencing results showing the deregulation of miRNAs in prostate cancer.28 To gain further insight into the properties of miRNAs as diagnostic biomarkers, we extended our investigations to an independent and larger series of prostate cancer specimens. MiRNA expression profiling was performed using commercial miRNA microarrays. The localization of miRNA expressing cells was evaluated by in situ miRNA hybridization using tissue microarrays constructed from matched cancerous and noncancerous prostate tissue samples.
Material and Methods
Patients and tissue samples
The use of tumor samples for molecular analysis was approved by the respective local ethical review boards of the institutions where tissue samples were collected. Matched tissue samples from prostate cancer and adjacent noncancerous tissue were prepared from prostatectomy specimens from men with so far untreated prostate cancer between 1994 and 1999. Hematoxylin and Eosin stained tissue sections were reviewed by a surgical pathologist (A.H.). Gleason scoring,29 histological diagnosis and TNM classification was performed according to the guidelines of the Union International Contre le Cancer (UICC) 2002.30 None of the patients had detectable distant metastases at the time of surgery. Patient set 1 consisted of 50 individuals collected by the Department of Pathology, University Münster, Germany, between 1994 and 1996. Formalin-fixed, paraffin-embedded (FFPE) tissue was used, as it has been described that formalin fixation has no significant impact on miRNA levels.31 Tissue sections were prepared in 10 μm slices and microdissected by an experienced uropathologist. By this, we generated samples of malignant tissue and matched nonmalignant control tissue. Patient set 2 consisted of 26 cryopreserved samples collected by the Department of Pathology, University Regensburg, Germany, between 2005 and 2006. These samples were macrodissected to ensure a cancer cell content of >70% in the tumor samples and the absence of cancer cells in the corresponding normal tissue. For each patient, the following clinicopathological characteristics were gathered: age at diagnosis, tumor classification according to the UICC 2002 TNM classification system30 and Gleason score of the tumor.29 Detailed patients' characteristics are summarized in Table 1.
Tissue microarrays were constructed from all 76 prostate cancer samples using a method described elsewhere.32 After review by a surgical pathologist, representative areas of tumor and adjacent nontumor tissue were marked for the extraction of biopsy cores. For every case, a single tissue spot of tumor and adjacent normal tissue was present on the tissue microarrays.
RNA from frozen tissue was extracted with Trizol reagent (Invitrogen, Karlsruhe, Germany) according to the manufacturer's instructions. RNA from FFPE tissue was extracted using the MasterPure Complete DNA and RNA Purification kit (Epicentre Biotechnologies, Madison, WI) according to the manufacturer's instructions. All RNA preparations were treated with RNase-free DNase I (Roche, Mannheim, Germany). RNA yield and A260/280 ratio was measured using a NanoDrop 1000 spectrophotometer (Thermo Scientific, Wilmington, DE).
Microarray experiments were performed using 1-color hybridizations on GeneChip miRNA microarrays V1.0 (Affymetrix, Santa Clara, CA) according to the manufacturer's instructions. This array contained 7,815 sequence specific probes covering miRNAs from 71 organisms including 678 human miRNAs as listed in miRBase v11.0 (http://www.mirbase.org). Hybridized microarrays were scanned and signal intensity data was analyzed with Partek software, version 6.2 (Partek, St. Louis, MO). Microarray data files were subjected to robust multi-array average (RMA) background correction and quartile data normalization. Analysis of variance (ANOVA) was used to identify differentially expressed miRNAs. Principal component analysis (PCA)33 was used to compress multidimensional data to three dimensions while maintaining the variance of the dataset. Gene expression measures are available at GEO (http://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE23022).
Quantification of miRNAs was carried out applying a two-step reaction34 using TaqMan miRNA assays and TaqMan reagents (Applied Biosystems, Foster City, CA) according to the manufacturer's protocols. Briefly, 10 ng of total RNA was reverse transcribed using the TaqMan miRNA reverse transcription kit (Applied Biosystems) with miRNA-specific stem–loop primers. The quantitative PCR reactions were performed in the StepOne plus real-time PCR system (Applied Biosystems) using sequence-specific primers and fluorescence labeled probes. All reactions were measured in triplicates in a final volume of 10 μL. Cycling conditions were chosen according to the manufacturer's protocols. To allow a relative quantification of miRNA expression levels, we included RNA prepared from the reference cell lines LNCaP35 or PNF0828 in every reaction plate and all samples were analyzed using probes for specific miRNAs and the endogenous reference RNA RNU6B. The cell line LNCaP was purchased from the German Collection of Microorganisms and Cell Cultures (DSMZ). The cells were authenticated by the DSMZ with multiplex PCR of minisatellite markers. Cells were passaged for <6 months after receipt. To minimize experimental variation, matched samples of tumor and normal tissues were always analyzed on the same reaction plate. Calculation of relative miRNA expression levels by applying the ΔΔCt method36 was performed using the StepOne software V2.0 (Applied Biosystems).
In situ miRNA hybridization
In situ detection of miRNAs was performed as described.37, 38 Tissue microarray slices of 5 μm were used. Briefly, tissue sections were deparaffinized, rehydrated and deproteinated using proteinase K (50 μg/mL). Sections were prehybridized in prehybridizing solution (50% formamide, 5× SSC, 50 μg/mL Heparin, 50 μg/mL yeast tRNA, 2% blocking powder, 0.1% Tween 20 and 5 mM EDTA) for 2 hr before hybridization. Probes (LNA modified and Digoxigenin labeled oligonucleotides; Exiqon, Vedbaek, Denmark) complementary to miR-375, miR-200c, miR-143 and miR-145 were diluted to 50 μM in hybridization solution (50% formamide, 20mM Tris-HCL pH 8.0, 0.3 M NaCl, 10% Dextran Sulfate, 0.5 mg/mL yeast tRNA, 1× Denhardts solution). Tissue sections were incubated with diluted probes over night at a temperature of 21°C below the calculated melting temperature of the probe. After posthybridization washes at hybridization temperatures, bound probes were detected by enzyme coupled antibodies (Alkaline phosphatase conjugated anti-Digoxigenin Fab fragments, Roche, Mannheim, Germany) and subsequent color reaction using the NBT/BCIP reagent (Roche). Tissue sections were counterstained with nuclear fast red staining solution (Sigma, Deisenhofen, Germany). Microscopy was performed with an Olympus BX60 microscope equipped with a XC30 camera system (Olympus, Hamburg, Germany). CellP Software (Olympus) was used for image acquisition.
Data analysis and statistical methods
Statistical analysis of miRNA expression data was performed with SPSS 18.0 (SPSS Incorporated, Chicago, IL) and GraphPad Prism 4.0 (Graph Pad software, La Jolla, CA). All statistical tests were performed as two-sided and p values <0.05 were considered as significant. Receiver–operator characteristics (ROC) and binary logistic regression analyses were used to determine the potential of miRNA expression signatures to discriminate between normal and malignant tissue samples as well as to test the contribution of individual miRNA markers. In the combined dataset, we applied a stepwise logistic regression analysis using the forward inclusion and backward elimination method with the likelihood ratio as inclusion or exclusion criteria (inclusion p = 0.05; exclusion p = 0.1) to determine the optimal combination of miRNAs with the best performance to distinguish between normal and malignant tissue. Pearson correlation was used to test the correlation of miRNA expression levels and major Gleason component.
Patient clinicopathological characteristics
There were significant differences between the two patient cohorts regarding the pathological tumor (pT) stage of the tumors, the Gleason sum and the primary Gleason component. Patient set 1 was mainly characterized by tumors with a Gleason score of 7 while patient set 2 was characterized by a higher proportion of pT3b tumors and tumors with Gleason sores of 8 or 9 (Table 1).
MiRNA microarray expression data
Microarray expression analysis was performed with a subset of 20 matched tumor and normal tissue samples from patient set 1. As it has been described that the differential expression of several tumor biomarkers is more pronounced in dedifferentiated cases,39, 40 we selected the 20 cases with the highest Gleason score for microarray analysis. ANOVA test statistics identified a group of 72 miRNAs that displayed a significant differential expression between tumor and nontumor tissues (p < 0.05) (Supporting Information Table S1). This group of 72 miRNAs was used for a PCA to reduce the complexity of multidimensional data to a three-dimensional display (Fig. 1a). Tumor and nontumor tissue samples showed a tendency to form distinct clusters, but these were not clearly separated from each other. We then performed unsupervised hierarchical clustering using all 72 miRNAs differentially expressed. Clustering was performed using the euclidean distance measure and average linkage. The standardized intensity measure was cropped at −2 and 2 (Fig. 1b). Only one tumor sample was assigned to the cluster of normal samples and two normal tissue samples were assigned to the tumors. Additionally, another software was used for hierarchical clustering to test if the results were consistent. The Multi Experiment Viewer v4.641 was used to perform clustering with different distance metrics (Supporting information Fig. S2). It became evident that two of the normal samples (cases 5 and 15) as well as three tumor samples (cases 16, 29 and 31) were repeatedly misclassified. It remained unclear if this was due to impurities in the tissue preparations or variation in miRNA expression.
We selected suitable miRNAs for validation by applying a correction for multiple testing. As a correction by the Bonferroni method42 is not suitable for large datasets, we performed a correction to limit the false discovery rate.43 At a false discovery rate of 0.05, 25 miRNAs showed a significant differential expression (p < 0.05 ANOVA; Table 2).
Table 2. MiRNAs differentially expressed in cancer cells as compared to normal tissue
We have recently published results of a deep sequencing analysis of miRNA expression profiles in prostate cancer. A total of 33 miRNAs were differentially expressed with the miRNAs miR-375, miR-200c, miR-143 and miR-145 exhibiting the most pronounced deregulation.28 We compared the results from microarray analyses to the results of the deep sequencing approach. Of 25 miRNAs that sustained correction for multiple testing, 19 had also been detected by deep sequencing before. Seven miRNAs (miR-375, miR-200c, miR-106a, miR-106b, let-7a, miR-21 and miR-20a) were reconfirmed to be upregulated, two miRNAs (miR-145 and miR-221) were reconfirmed to be downregulated and one miRNA (miR-101) displayed an inverse expression pattern when comparing the two methods. Nine miRNAs were identified by deep sequencing but did not show a differential expression of more than 1.5-fold. This corresponds to a concordance rate of 47% among the 19 miRNAs detected with both methods. The remaining six miRNAs identified by microarray analyses were not detected by deep sequencing although both analyses were based on the identical version of miRBase (v11.0).
For validation by quantitative real-time (qRT-PCR) PCR, we selected the miRNAs miR-375 (upregulated 3.9-folds) and miR-200c (upregulated 2.9-folds) that displayed the most pronounced overexpression in tumor tissues. Furthermore, we selected the miRNAs miR-145 (downregulated 1.4-fold) and miR-143 (downregulated 1.2-fold). MiR-143 was not initially identified as deregulated by microarray but deep sequencing data from our own group showed a downregulation of this miRNA.
Validation of miRNA expression by qRT-PCR analysis
Relative expression of miRNAs miR-375, miR-200c, miR-143 and miR-145 was analyzed in both patient sets consisting of 26 and 50 cases. We found that the miRNAs miR-375, miR-143 and miR-145 displayed a significant differential expression between the matched tumor and normal samples. MiR-375 was upregulated 1.64-fold in set 1 and 1.97-fold in set 2. MiR-143 was downregulated 1.96-fold and 2.78-folds. MiR-145 was downregulated 2.44-folds and 3.45-folds, respectively (p < 0.05, paired t test). MiR-200c did show a significant differential expression only in patient set 1, where it displayed an upregulation of 1.27-fold (Supporting Information Table S2 and Fig. S2).
In situ miRNA hybridization
As tumors are heterogeneous containing not only malignant but also stromal and other cells we performed in situ detection of miRNAs to examine the distribution of miRNA expressing cells in prostate cancer tissue sections. For this, LNA modified oligonucleotide probes against miRNAs miR-375, miR-200c, miR-143 and miR-145 (Exiqon) were used (Fig. 2).
We noted that miR-375 and miR-143 resulted in a distinct cytoplasmic staining of epithelial cells, whereas surrounding nonepithelial tissue did not show a staining signal. MiR-200c and miR-145 displayed a tendency toward a perinuclear staining. It is of note that miR-145 was also detectable in endothelial cells of blood vessels and to a certain extent in stromal tissue, where single cells displayed nuclear staining. In corresponding tumor samples, the expression of miR-375 and miR-200c was limited to cancer cells. The expression of miR-143 and miR-145 in tumor tissue displayed more heterogeneity. Although some cases with retained glandular structures still showed a weak cancer cell-specific miRNA expression, no miRNA expression was seen in the majority of samples. MiR-145 could still be detected in endothelial cells of blood vessels, while stromal cells expressing miR-145 were absent. Experiments using an unspecific (scramble) control probe resulted in a weak staining pattern that did not correspond to any staining pattern seen with specific detection probes. Furthermore, we perfomed immunohistochemistry using antibodies against E-Cadherin and Myosin VI. MiRNA expressing cells were also stained with E-Cadherin specific antibodies, showing that miRNA expression is strongest in epithelial cells. Myosin VI is also expressed in epithelial cells but the staining intensity is enhanced in tumor samples compared to normal tissue (Supporting Information Fig. S4).
Diagnostic capabilities of miRNAs
As specific miRNAs were found to be differentially expressed in the majority of tumor cases analyzed, we tested whether miRNA expression patterns are capable of distinguishing between malignant and nonmalignant tissue. MiR-375, miR-200c, miR-143 and miR-145 were used as potential markers in a binary logistic regression analysis to determine their diagnostic properties. As described before, we expected the differential expression of miRNAs to be more pronounced in dedifferentiated samples. Therefore, we defined patient set 2 containing a higher proportion of dedifferentiated tumors as a discovery set and all results were subsequently tested in patient set 1 designated as the test set.
In the discovery set, the best positive discriminating factor was miR-375 and the best negative discriminating factor was miR-145 with 67.3 and 71.2% of the samples classified correctly (Table 3). MiR-200c was no significant discriminating factor and was therefore excluded from further analysis. We used full logistic regression to build a predictive model incorporating miR-375, miR-143and miR-145. In the discovery set, the resulting model was able to correctly classify 90.4% of the samples and in the test set the rate of correctly classified samples reached 73.0%.
Table 3. Receiver–operator characteristics and classification properties of miRNAs
Next, we tested the performance of the three selected miRNA markers in a mixed cohort. We combined the discovery and test sets to form a combined patient set of 76 individuals. Even in this mixed patient set, the three miRNA markers correctly classified 77.6% of the samples (Table 3). To further characterize the capability of the selected miRNAs to distinguish between tumor and nontumor samples, we performed a ROC analysis in the discovery set, the test set (Fig. 3) and the combined patient sets (Supporting Information Fig. S3). For each miRNA and the combination of miR-375, miR-143 and miR-145, the resulting area under the ROC curve (AUC), the 95% confidence interval, the p value and the proportion of correctly classified samples as determined by binary logistic regression are shown in Table 3. In all cases, a combination of the three miRNA markers performed better at discriminating between tumor and nontumor samples than any single miRNA marker. In the test set, the discriminatory capability was less accurate (Fig. 3b). This was probably due to differences in the composition of the patient sets. Finally, we validated the selected miRNA markers in an independent patient cohort. A previously published dataset44 was used. When applying the same analytical scheme, the combination of miR-375, miR-143 and miR-145 could correctly classify 70.8% of all samples. The ROC analysis resulted in an AUC of 0.854, p = 0.003.
As it has been proposed that the ratio of miRNA expression may be more suitable for diagnostic or prognostic purposes than the expression of any single miRNA,45, 46 we calculated the ratio of miRNA expression of the best positive predictive miRNA, miR-375, and the best negative predictive miRNA, miR-145, or the ratio of miR-375 and the second best negative discriminator miR-143. Similar to our previous approach, we first analyzed the expression ratios in the designated discovery set. An optimal cutoff value was determined and subsequently applied to the test set. The miRNA expression ratio of miR-375/miR-145 showed a median of 24.1 (range 0.57–153.8) in normal tissue samples and a median of 224.2 (range 19.84–2,763) in tumor tissues. The ROC curve had an AUC of 0.920 (95% confidence interval 0.847–0.993, p < 0.0001). The optimal cutoff value was determined according to the likelihood ratio calculated for every possible value. With a miR-375/miR-145 cutoff value of 103, we reached 73.08% sensitivity and 96.15% specificity. In the test set, this cutoff value resulted in 84% sensitivity but only 60% specificity.
The miR-375/miR-143 ratio in the discovery set showed a median of 8.77 (range 0.22–26.56) in normal tissue samples and a median of 46.69 (range 7.36–309.0) in tumor tissues. The calculated ROC curve displayed an AUC of 0.932 (95% confidence interval 0.861–1.000, p < 0.001). With a cutoff value of 22.6, this resulted in 83.33% sensitivity and 95.83% specificity. In the test set, this resulted in 52.0% sensitivity and 76.0% specificity.
Association with clinicopathological features
We next analyzed whether expression levels of miR-375, miR-200c, miR-143 and miR-145 correlated with clinicopathological features. In the discovery patient set, no correlation of miRNA expression levels to pT stage or the Gleason sum was found. However, we discovered a negative correlation between the expression level of miR-143 and the predominant differentiation pattern of the respective tumors expressed by the primary Gleason component. Hereby, poorly differentiated tumors expressed a reduced amount of miR-143 (p = 0.043, Kruskal–Wallis test). This held true even in the extended cohort of patient sets 1 and 2 combined (p = 0.006, Kruskal–Wallis test). This result is concordant with previous reports,39, 40 showing that the deregulation of specific miRNAs is more pronounced in dedifferentiated tumors.
To date several studies have been published addressing the differential expression of miRNAs in prostate cancer21–26, 44 but the results were highly inconsistent. This might be due to differences in sample composition or the detection platform used. Moreover, the results were partially not validated by RT-PCR. In our study, we analyzed the expression of 678 human miRNAs in 20 matched tissue samples of histologically confirmed prostate cancer tissue and adjacent nonmalignant tissue using commercial miRNA microarrays. A total of 72 miRNAs displayed a significant differential expression in tumor tissue compared to nontumor tissue. Hierarchical clustering using two independent software programs and different clustering algorithms revealed that a number of tumor samples clustered with normal samples and vice versa. Most likely this result reflects a degree of heterogeneity in miRNA expression, although we cannot exclude a certain degree of impurities in the tissue preparations used.
After correction for multiple testing, a group of 25 miRNAs was identified. Our data do only partially overlap with any miRNA screening study published so far. A total of 16 of 25 miRNAs identified in our study (Let-7a, miR-101, miR-106a, miR-106b, miR-141, miR-145, miR-17, miR-182, miR-200c, miR-20a, miR-21, miR-214, miR-221, miR-222, miR375 and miR-93) are in concordance with at least one other publication.21–26, 44
We have previously published miRNA expression data from prostate cancer tissue generated by deep sequencing of cDNA libraries.28 When comparing the miRNA microarray expression data with these results, 19 miRNAs have been detected independently with both methods. Seven of 19 miRNAs were reconfirmed to be upregulated, two miRNAs were reconfirmed to be downregulated and one miRNA displayed an inverse expression pattern when comparing the two methods. This corresponds to a concordance rate of 47% among those miRNAs detected with both techniques.
Given the fact that the previous deep sequencing analysis was conducted using pooled RNA preparations from tumor and unmatched healthy control tissue while the microarray analysis was performed using microdissected archival tissue, these differences might be due to population-based effects or inherent differences in the detection platforms.
For validation, we selected miR-375, miR-200c, miR-143 and miR-145 on the basis of the degree of differential expression as well as previously generated miRNA expression data. Their expression was analyzed in two independent patient sets of 26 and 50 tissue pairs by qRT-PCR. Except for miR-200c, whose differential expression could only be confirmed in one patient set, the other three miRNAs could be confirmed in both sets. MiR-375 was upregulated in 62 of 76 (82%) cases, miR-200c in 50 of 76 (66%), miR-143 was downregulated in 59 of 76 (78%) and miR-145 in 72 of 76 (95%) cases analyzed. Upregulation of miR-375 and miR-200c as well as downregulation of miR-143 and miR-145 has previously been described in prostate cancer.21, 22, 25, 44 We have shown that miRNAs miR-143 and miR-145 are both able to target the 3′ untranslated region of MYO6.28 Myosin VI overexpression in prostate cancer has already been demonstrated by immunohistochemistry yielding a correlation to tumor aggressiveness as determined by the Gleason score.47 In accordance with these findings, we found an inverse correlation of the expression of miR-143 and the primary Gleason component with more dedifferentiated tumors expressing less miR-143. For miR-145, we did not observe any similar correlation with Gleason sum or Gleason components.
One major concern in studies addressing the comparison of tumor and nontumor tissue is the composition of the tissue samples used for screening. It is of great importance to rule out that differences in miRNA expression are merely due to a different proportion of, e.g., stromal cells in the tissue preparations. For the miRNA microarray experiments, we used microdissected tissue samples. Additionally, we performed miRNA in situ hybridization experiments, which supported the hypothesis that miRNA expression is indeed cancer cell-associated.
By ROC analyses we could demonstrate that miRNAs are capable of distinguishing between tumor and nontumor tissue. The best discriminating miRNA miR-145 could correctly classify 71% of tissue samples. When combining miR-375, miR-143 and miR-145, we reached a correct classification rate of almost 78%. This shows that miRNAs could serve as valuable biomarkers in the diagnosis of prostate cancer.
To elucidate the application of miRNAs as diagnostic biomarkers, especially in a clinical setting, we used miRNA expression ratios as described elsewhere.46 The miR-375/miR-143 ratio resulted in 83% sensitivity and 95% specificity in one patient set. This suggests that miRNA expression ratios may hold a significant potential as clinical diagnostic biomarkers. As it is also possible to assess miRNA expression patterns in urine48 or blood serum,49 the tissue-derived expression patterns of the most promising miRNAs should be analyzed in this type of clinical material to evaluate their diagnostic or prognostic properties.
MiRNA expression profiling is becoming more and more interesting regarding the potential target genes regulated by these molecules. A comprehensive analysis of miRNA target genes could lead to the elucidation of pathways deregulated in cancer cells and subsequent identification of therapeutic targets. So far, only a limited number of miRNA target genes have been validated and the bioinformatic prediction of potential miRNA target genes remains crucial. We performed a bioinformatic analysis using miRecords that allows searching with 11 target prediction algorithms using the target selection criteria described44 (Supporting Information Table S3). Predicted target genes of the miRNAs deregulated in prostate cancer include regulators of epithelial-mesenchymal transition ZEB1 and ZEB2, the matrix-metalloproteinase inhibitor RECK, the proto-oncogene n-MYC and cell cycle regulator CDK6. Furthermore, we queried the DIANA mirPath database50 to identify target genes located in signaling pathways. MiR-200c is predicted to regulate CDK2 and E2F3 indicating a role for this miRNA in cell cycle control. Interestingly, both miR-200c and miR-375 are predicted to interact with PDPK1, which plays a vital role in the activation of the AKT signaling pathway. MiR-143 may also have an impact on signaling pathways as two predicted target genes, KRAS and MAPK1, are located in the mitogen-activated protein kinase signaling pathway. MiR-145 target genes are not located in described signaling pathways but by targeting FOXO1, this miRNA may participate in the induction of a prosurvival phenotype of prostate cancer cells. Thorough analyses of these and other predicted target genes of miRNAs in prostate cancer are necessary to elucidate the functional consequences of miRNA deregulation.
Taken together, our study provides evidence that miRNA profiling using differentially preserved tissue samples and different experimental platforms are able to generate consistent and reliable results. MiRNAs as single biomarkers or in combination could be useful in the diagnosis of prostate cancer and further studies should be conducted to examine their diagnostic capabilities in a clinical setting.
The authors thank Dr. Gerhard Unteregger, Department of Urology, University of Saarland Medical School, for providing PNF08 normal human prostate fibroblasts. This research was funded by a grant from Wilhelm-Sander-Stiftung (grant number 2007.025.01; to B.W. and F.G.) and by the German Federal Ministry of Education and Research (grant NGFN-Plus number 01GS0801/4; to J. S. and F.G.).