Main

Prostate cancer (PCa) is the most frequent cancer in males in Europe and North America. Digital rectal examination (DRE) and serum prostate-specific antigen (PSA) monitoring followed by serial biopsy and histological analysis is the cornerstone of PCa screening/diagnosis. However, although very high PSA values indicate a high probability of cancer detection by biopsy, lower values poorly discriminate PCa from inflammatory processes and benign hyperplasia (Ploussard et al, 2013; Hayes and Barry, 2014). On the basis of the results of two large epidemiological studies showing a very small or no effect of PSA testing on PCa mortality (Andriole et al, 2012; Schroder et al, 2012) and the negative consequences of PSA screening, which include overdiagnosis, overtreatment and treatment complications, the US Preventive Services Task Force recommended against the use of PSA testing for PCa screening (Aly et al, 2015; Johnson et al, 2015).

These considerations have engendered an interest in the identification of novel markers of PCa with improved specificity compared with PSA. Circulating microRNAs (miRNAs; Bartel, 2004; Waltering et al, 2011; Kim and Kim, 2013) may represent excellent biomarkers because of their ease of detection, stability in biological fluids and minimal invasiveness of the test. Different studies identified specific miRNA signatures associated with PCa (Mahn et al, 2011; Zhang et al, 2011; Chen et al, 2012; Kelly et al, 2015; Mihelich et al, 2015), although a clear consensus has not yet emerged. Discrepant findings may in part reflect differences in pre-analytical sample processing and data normalisation/analysis.

In the present study, we explored the efficacy of circulating miRNA as biomarkers for PCa. To circumvent problems linked to data normalisation, we calculated ratios of upregulated and downregulated miRNAs. Results indicated that the miR-106a/miR-130b and miR-106a/miR-223 ratios are highly sensitive and more specific than PSA in discriminating localised PCa from benign prostatic hyperplasia (BPH).

Materials and methods

Patients

A total of 67 patients were recruited; 36 were positive for PCa upon biopsy (BIO+) and 31 were negative (BIO−). BIO+ and BIO− patients were comparable in terms of mean age (65.2 and 66.7 years, respectively) and PSA (8.5 and 8.4 ng ml−1, respectively). All the BIO+ patients had localised PCa, and all of the BIO− patients were diagnosed with BPH. Additional clinical characteristics of the patients are described in the Supplementary Materials and Methods and Supplementary Table S1.

Quantitative RT–PCR

RNA was extracted from plasma as described in the Supplementary Materials and Methods. Five microlitres of total RNA were used for first-strand cDNA synthesis in a 15-μl reaction volume, using the TaqMan miRNA Reverse Transcription kit and miRNA-specific stem-loop primers (Thermo Fisher Scientific, Foster City, CA, USA). cDNAs (2.5 μl) were amplified for 45 cycles using TaqMan miRNA primers and probes (Thermo Fisher Scientific) and LightCycler 480 PCR Master Mix (Roche Diagnostics GmbH, Mannheim, Germany). No-RT and no-template negative controls were included. The amplification reactions were performed in a LightCycler 480 II (Roche). Signals were quantified using the second derivative maximum method (Software Version 1.5, Roche). All amplification curves were carefully examined, and products with poor curves were discarded. Ct values <40 were obtained for >99% of the reaction samples.

Data normalisation

As candidate normalisers we tested spiked-in cel-miR-39 as well as human miR-24, which exhibited low variability in our data set (data not shown) and was used as a normaliser in other studies (Bianchi et al, 2011).

Relative expression of target miRNAs was calculated as ΔCtmiR=CtmiR−Ctnormalizer. Differential expression of each miRNA was then evaluated as ΔΔCtmiR=ΔCtmiR(patient)− i.e., by subtracting the mean ΔCt value for the BIO− group from the ΔCt of each BIO+ patient.

To circumvent biases linked to the choice of an appropriate reference, we also used the miRNA ratio approach (Boeri et al, 2011). The Ct value of each miRNA was converted to the corresponding expression level (2−Ct). Twenty-one different miRNA ratios were calculated as

Statistical analysis

The two-tailed Wilcoxon rank-sum test was used to identify miRNAs significantly different between the BIO+ and BIO− groups, whereas the one-tailed Wilcoxon rank-sum test was used for miRNA ratios. Post hoc power analysis was performed using a nonparametric resampling technique (10 000 replications). A univariate logistic regression model was built to evaluate the ability of each miRNA ratio to predict the BIO+ outcome. The two most significant miRNA ratios were also used in a bivariate logistic regression model. Odds ratios and 95% confidence intervals were estimated for the fitted logistic regression models. Receiver operating characteristic (ROC) curves were plotted and the area under the ROC curve (AUC) was estimated to compare the two most significant miRNA ratios. Optimal thresholds were determined using the Youden’s J statistic. Sensitivity, specificity, negative predictive value, positive predictive value and accuracy were determined. Statistical analyses were performed in the R environment, using a customised code and the pROC package for ROC curve analysis.

Results

Circulating miRNA in PCa and BPH patients

We examined 36 patients with localised PCa confirmed by serial biopsy (BIO+) and 31 individuals diagnosed with BPH (BIO−). Clinical assessment of the patients included age, PSA, DRE, Gleason scoring and T stage (Supplementary Table S1). Twelve miRNAs were selected among those reported to be involved in PCa (Supplementary Table S2). To facilitate their application to large-scale testing, we focused on miRNAs that could be detected with a single round of qRT–PCR (i.e., without a pre-amplification step).

Several studies underscored the critical importance of data normalisation when measuring circulating miRNAs. Pilot tests with the cel-miR-39 spike-in reference showed a wide variability in Ct values (Supplementary Figure S1), possibly due to differences in plasma RNAase levels, as previously reported (Mahn et al, 2011). We thus normalised each miRNA using as internal reference miR-24, which was relatively constant in our samples and was used as a normaliser in other studies (Bianchi et al, 2011). Results showed that miR-106a, miR-20a, miR-223 and miR-21 were significantly different between BIO+ and BIO− individuals (Figure 1).

Figure 1
figure 1

Boxplots of the distribution of –ΔΔCt values for circulating miR-20a, miR-21, miR-93, miR-106a, miR-130b and miR-223 in patients with localised PCa vs biopsy-negative individuals. Each dot represents a patient sample. –ΔΔCt values (see Materials and methods) were calculated using endogenous miR-24 as a reference. Differences between BIO+ and BIO− samples were analysed using two-tailed Wilcoxon rank-sum test. miR-106a, miR-223, miR-20a and miR-21 showed P-values<0.05.

Full size image

miRNA ratios distinguish PCa (BIO+) from BPH (BIO−) patients

To overcome the bias linked to data normalisation using a calibrator RNA, we next calculated the ratios between the expression values of miRNA pairs listed in Supplementary Table S3. Fourteen ratios were found to be significantly different between the BIO+ and BIO− groups with P<0.05. We focused on the miR-106a/miR-130b and miR-106a/miR-223 ratios (Figure 2A), which ranked highest in terms of statistical significance (P-values were <0.0001) and showed a statistical power >0.99 (see Supplementary Table S3 for details). As shown in Figure 1, miR-106a was upregulated in BIO+ samples, whereas miR-130b and miR-223 were downregulated.

Figure 2
figure 2

Circulating miR-106a/miR-130b and miR-106a/miR-223 ratios predict the presence of localised prostate cancer. (A) Relative expression levels are shown as 2−ΔCt (see Materials and methods). Both miRNA ratios were significantly higher (one-tailed Wilcoxon rank-sum test) in PCa patients compared with biopsy-negative patients (P=6.97 × 10−6 for miR-106a/miR-130b; P=8.5 × 10−5 for miR-106a/miR-223). (B) Receiver operating characteristic (ROC) curves generated for univariate (red and blue traces) and bivariate (black trace) logistic regression models against the ROC curve of PSA (green trace). The miRNA ratios were better than PSA for discriminating PCa from BIO− samples. Areas under ROC curves (AUC) are shown in parentheses. (C) Proposed miRNA-based diagnostic flowchart for PCa diagnosis.

Full size image

These miRNA ratios were next evaluated by univariate logistic regression analysis. Resulting areas under ROC curves (AUC) were 0.81 and 0.77 for miR-106a/miR-130b and miR-106a/miR-223, respectively (Figure 2B); bivariate logistic regression analysis using both miRNA ratios as predictors showed an AUC of 0.84 (Figure 2B). All these values were considerably higher than the AUC of 0.56 obtained for PSA, a value consistent with other studies (Guzel et al, 2015). Odds ratios are reported in Supplementary Tables S4 and S5.

Determining threshold values for miRNA ratios

Statistical analysis indicated optimal 2−ΔCt threshold values of 10.413 for miR-106a/miR-130b and 0.176 for miR-106a/miR-223. To facilitate application in clinical practice, results of the bivariate logistic regression model were transformed into a miRNA score=0.1855 × (miR-106a/miR-130b)+2.3162 × (miR-106a/miR-223), where the miRNA ratios indicate 2−ΔCt values. The optimal threshold for this miRNA score was 2.142. With these cutoff values, specificity and sensitivity were 0.87 and 0.72 (miR-106a/130b), 0.65 and 0.81 (miR-106a/223), and 0.81 and 0.83 (miRNA score), respectively.

A diagnostic flowchart for PCa screening

On the basis of these findings, we propose a flowchart for the selection of patients to be subjected to biopsy (Figure 2C). Patients with PSA >4 ng ml−1 (right-hand portion of the chart) would first be tested for miR-106a/miR-130b; patients with a 2−ΔCt>10.413 would be eligible for biopsy, whereas patients below this threshold would be tested for the miR-106a/miR-223 ratio. Patients with a miR-106a/miR-223 2−ΔCt>0.176 would be biopsied, whereas patients below this threshold would not.

The left-hand portion of the flowchart refers to patients with PSA<4 ng ml−1, but DRE+. In our data set, among the six patients with these characteristics, the miR-106a/miR-130b ratio correctly distinguished all four BIO+ from the two BIO− patients, generating no false-positive or false-negative results. Table 1 compares the diagnostic performance of the proposed biomarkers with PSA.

Table 1 Assessment of the performance of different diagnostic strategies
Full size table

Conclusions

Recent large-scale studies found no or very small benefit for periodic PSA and DRE screening after up to 13 years of follow-up (Andriole et al, 2012; Schroder et al, 2012); identification of other biomarkers to aid early diagnosis and biopsy decision is thus in high demand.

In this study, we explored the efficacy of circulating miRNA as biomarkers for PCa screening. Our findings suggest that a single miRNA might be poorly informative, especially considering the known biases linked to data normalisation. We thus propose to calculate ratios between one upregulated miRNA and one downregulated miRNA in the same patient. Results showed that the miR-106a/miR-130b and miR-106a/miR-223 ratios discriminated between localised PCa and BPH patients with a specificity much superior to PSA. A diagnostic flowchart that employs two miR ratios in combination with standard PSA (Figure 2C) should considerably increase the specificity and also improve the sensitivity compared with PSA testing alone (Table 1).

The miRNA ratio approach described in the present study may be applied to large-scale screening, as it requires measuring only three circulating miRNAs and overcomes the need for a normaliser miRNA. The increased specificity of this assay compared with PSA alone could help to reduce the costs and morbidity of unnecessary repeated biopsies. Future studies should test the validity of this approach in an ample prospective cohort of patients.