MicroRNA panel in serum reveals novel diagnostic biomarkers for prostate cancer

Study design, patients, and samples

Subjects enrolled in the study were recruited from the Third Affiliated Hospital of Soochow University, Jiangsu Cancer Hospital, and the First Affiliated Hospital of Nanjing Medical University between 2016 and 2017. The Institutional Review Boards approved all procedures of the First Affiliated Hospital of Nanjing Medical University (Ethical Application Ref: 2016-SRFA-149). We obtained the information of TNM stage, differentiation degree, and clinical characteristics from patients’ records (according to the seventh edition American Joint Committee on Cancer (AJCC)). All the PC patients were pathologically confirmed. We drew blood samples when patients did not receive any therapeutic procedures like surgery, endocrine therapy, or chemotherapy. NCs enrolled in our study were healthy volunteers who took the routine physical examination at the First Affiliated Hospital of Nanjing Medical University. Each participant provided the written informed consent.

A total of 86 PC patients and 86 normal controls were enrolled in our study. As shown in Fig. 1, we designed a four-stage experimental process. Several factors were taken into account in sample selection and distribution: (i) the purpose of each separate stage; (ii) the sequence of sample collection in practical operation; (iii) the balance of non-experimental factors among four different sets; (iv) the basic principle of experiment design—control, randomization, replication, and balance. Also, factors such as gender, age, TNM stage, and pathological type might have confounding effects on circulating miRNA expression in PC. Therefore, these potential interference factors were as evenly distributed across four sets as possible to avoid selection bias. In the initial screening phase, we randomly chose twenty peripheral serum samples from PC patients, ten from NCs, and then synthesized two PC sample pools and 1 NC sample pool for miRNA profiling (we pooled every ten samples as one pool sample).

Then, to verify the reproducibility and reliability of the results acquired from the screening phase, serum samples from 28 PC patients and 28 NCs were used to confirm the dysregulated miRNAs in the training stage. The miRNAs identified by the training stage were validated in the following testing stage in 32 PC patients and 32 NCs. After the three steps above, we set an external validation phase to further assess the three-miRNA signature’s diagnostic value with 26 PC patients and 26 NCs. As shown in Table 1, no significant difference in the distribution of age between PC and NC samples was found in any stage. As previously described in our team’s another study (Shan et al., 2018), the identified miRNAs were further validated in 28 pairs of formalin-fixed paraffin-embedded (FFPE) PC tissues and matched adjacent normal tissue specimens from the same surgery patients. Additionally, we analyzed miRNAs’ expression in exosomes in 24 PC patients and 24 NCs to investigate the miRNAs’ potential form in the peripheral circulation.

We collected 5 ml of venous blood in a vacuum blood collection tube and then put them in a refrigerator at 4 °C. These blood collections were processed within 12 h using a two-step centrifugation method (1,500 rpm*10 min, then 12,000 rpm*2 min) at 4 °C. Next, we carefully aspirated and separated the supernatant to obtain the serum, aliquoted it into clean, enzyme-free EP tubes to prevent repeated freezing and thawing. These serum samples were then stored in a −80 °C refrigerator for further analysis.

Exosomes isolation

According to the manufacturer’s protocol, exosomes were isolated from serum using ExoQuick™ (System Biosciences, Mountain View, Calif). Briefly, 200 µl serum was mixed with 100 µl ExoQuick exosome precipitation solution. Then we precipitated exosome pellets by centrifugation at 13,000 rpm for 2 min. The obtained exosome samples were next lysed into 200 µl RNAse-free water for further analysis.

RNA extraction

As previously described (Zhou et al., 2017b), total RNA was isolated from 200 µl serum or exosome samples using mirVana Paris Kit (Ambion, Austin, TX, USA) under the manufacturer’s protocol instruction. We added 5 µl of synthetic C. elegans miR-39 (5nM/L, RiboBio, Guangzhou, China) to each sample after denaturing the solution (Ambion, Austin, TX, USA) for sample-to-sample normalization. We extracted total RNA from FFPE specimens via the High Pure FFPE RNA Micro Kit (Ambion Austin, TX, USA). Then, we dissolved RNA in 100 µl RNAse-free water and stored these RNA samples at −80 °C until use. Nanodrop 2000 Thermo scientific spectrophotometer (NanoDrop Technologies, Wilmington, DE, USA) was used to evaluate the total RNA’s concentration and purity.

Quantitative reverse transcription polymerase chain reaction (qRT-PCR)

The experimental process was mentioned in the previous article (Zhou et al., 2017b). MiRNA was reverse transcribed to cDNA using the miRCURY Locked Nucleic Acid (LNATM) Universal Reverse Transcription (RT) microRNA PCR, Polyadenylation, and cDNA synthesis kit (Exiqon miRNA qPCR panel; Vedbaek, Denmark). We then scanned the microarrays on a 7900HT real-time PCR system (Applied Biosystems, Foster City, CA, USA) with Exiqon miRCURY-Ready-to-Use-PCR-Human-panel-I + II-V1.M (Exiqon miRNA qPCR panel; Vedbaek, Denmark). The amplification of miRNA was performed via specific primers of reverse transcription (RT) and polymerase chain reaction (PCR) (RiboBio, Guangzhou, China). The RT reaction was performed at 42 °C for 60 min followed by 70 °C for 10 min; the PCR reaction was performed at 95 °C for 20 s, followed by 40 cycles of 95 °C for 10 s, 60 °C for 20 s and then 70 °C for 10 s on a 7900HT real-time PCR system. qRT-PCR was performed to amplify and detect the miRNAs with SYBR Green Dye (SYBR^® Premix Ex TaqTM II; TaKaRa, Dalian, China). The expression of miRNAs in serum samples and exosomes were calculated using the comparative 2^−ΔΔCt method normalized to exogenous reference miRNA (cel-miR-103a-3p), ΔCt=CtmiRNA-Ctcel-miR-103a-3p. Tissue miRNAs were determined by the 2^−ΔΔCt method relative to cel-miR-39 and RNU6B (U6) (Livak & Schmittgen, 2001).

Statistical analysis

Mann–Whitney test was used to analyze the differential expression of miRNAs between PC patients and NCs. Moreover, the association between miRNAs and the clinical characteristics was estimated using one-way ANOVA or χ² test. Sample size calculation has been done using software PASS. A sample of 12 from the positive group and 12 from the negative group achieves 91% power to detect a difference of 0.3 between the area under the ROC curve (AUC) under the null hypothesis of 0.5 and an AUC under the alternative hypothesis of 0.8 using a two-sided z-test at a significance level of 0.15. The receiver operating characteristic (ROC) curves and the area under the ROC curve (AUC) calculation, along with Cox’s regression models, were applied to estimate the diagnostic and prognostic value of identified miRNAs. A logistic regression model for PC prediction was used to the data of the training and validation stages. SPSS (version 19.0, IBM, Armonk, New York, USA) and GraphPad Prism 7 (GraphPad Software, USA) were used to performed statistical analyses. Differences were considered significant when P-value < 0.05.

Bioinformatics analysis

Pathway enrichment analyses of miRNA gene targets and differentially expressed genes were performed using DIANA-mirPath v.3.0 (http://www.microrna.gr/miRPathv3) (Vlachos et al., 2015). We conducted Gene Ontology (GO) (The Gene Ontology Consortium, 2019) analysis and Kyoto Encyclopedia of Genes and Genomes (KEGG) (Kanehisa & Goto, 2000) analysis to predict possible biological processes and potential signal pathways. To further confirm our results, we downloaded prostate adenocarcinoma data from The Cancer Genome Atlas (TCGA) and performed a similar analysis.

Discovery of candidate miRNAs from the screening phase

To filter candidate miRNAs in the screening phase, we screen the expression levels of 168 miRNAs in the 2 PC sample pools and the NC sample pool using Exiqon miRCURY-Ready-to-Use-PCR-Human-panel-I + II-V1.M on the qRT-PCR platform. Screening criteria for candidate miRNAs were as previously described (Zhou et al., 2017b). Only the miRNAs with cycle threshold (Ct) values <37 and 5 lower than negative control in the panel were eligible for further analysis. A miRNA was considered as a candidate miRNA if it showed more than 1.5-fold or less than 0.67-fold altered expression in both two pooled PC samples compared to the NC pool sample. As shown in Table 2, a total of 37 miRNAs was found to exhibit dysregulated expression in PC and chosen for the further training stage.

Confirmation of identified miRNAs by qRT-PCR

The expression of the 37 candidate miRNAs selected via the screening phase was further analyzed in 28 PC patients and 28 NCs using qRT-PCR in the training phase. 4 of the 37 miRNAs were further confirmed. During the testing stage, all four miRNAs were subsequently analyzed in 32 PC patient samples and 32 NCs. 3 of the 4 show consistently up-regulated expression and met our standards (Table 3). We then assessed the three miRNAs (miR-146a-5p, miR-24-3p, and miR-93-5p) in 26 PC patient samples and 26 NCs in an external validation stage and found that the result was the same as above. What’s more, all the three miRNAs had significantly higher expression levels in peripheral serum of PC patients as compared with NCs when the results of the three stages were combined (Fig. 2).

Diagnostic value of the candidate miRNAs

To evaluate the three miRNAs’ performance in PC diagnosis, we conducted ROC curves on each of the three miRNAs. Taken three phases into consideration (training phase, testing phase and external validation phase), the AUC for serum miR-146a-5p, miR-24-3p and miR-93-5p was 0.767 (95% confidence interval (CI): 0.696–0.839), 0.767 (95% CI [0.694–0.840]) and 0.781 (95% CI [0.710–0.852]), respectively (Fig. S1). Therefore, each of these three miRNAs had the ability to distinguish PC patients from healthy controls. In addition, when these three miRNAs were treated as a whole, we found that this 3-miRNA signature had a better diagnostic performance. As shown in Fig. 3A, combining the three stages, the AUC of this miRNA panel was 0.819 (95% CI [0.754–0.885]; sensitivity 75.6%, specificity 85.7%). Meanwhile, the diagnostic value of the 3-miRNA panel was assessed in a combination of the former two cohorts (the training and testing stage). The AUC was 0.831 (95% CI [0.757–0.905]; sensitivity = 73.3%, specificity = 80.0%; Fig. 3B). In the external validation stage, similar to the previous training and testing phases, the three identified miRNAs showed a consistent tendency of upregulation, and the panel still performed quite well in discriminating PC patients from NCs with the AUC being AUC of 0.814 (95% CI [0.687–0.942]; sensitivity = 76.9% and specificity = 83.3%; Fig. 3C). The correlation between the serum PSA values, TNM stages, and the three serum miRNAs levels was also analyzed, respectively. However, no significant results were observed. Therefore, data were not presented.

External validation in public databases

We further verified the diagnostic value of the 3-miRNA signature in four public datasets: The Cancer Genome Atlas Prostate Adenocarcinoma (TCGA-PRAD) dataset (http://cancergenome.nih.gov/), the GSE113740 dataset, the GSE113486 dataset, and the GSE112264 dataset. The first TCGA-PRAD dataset represents data from tissues, while the other three datasets contain data of serum samples. In the TCGA-PRAD dataset, the 3-miRNA marker performed well in distinguishing between PC tumor tissues (n = 498) and normal tissues (n = 51). As shown in supplementary Table S1, the corresponding AUC of the 3-miRNA signature was 0.979 (95% CI [0.967–0.991], P < 0.001, specificity: 94%, sensitivity: 94.1%). While in the three datasets of serum samples (GSE113740, GSE113486 and GSE112264), the AUC of the 3-miRNA panel to distinguish PC patients from healthy people was 0.806 (95% CI [0.733–0.880], P < 0.001, specificity:68.7%, sensitivity:84.0%), 0.943 (95% CI [0.901–0.986], P < 0.001, specificity: 100%, sensitivity: 42.5%) and 0.950 (95% CI [0.925–0.976], P < 0.001, specificity: 92.7%, sensitivity: 86.2%), respectively (Fig. 4). To summarize, the selected miRNA panel’s diagnostic value was successfully verified in the four public datasets.

Comparison with previously identified signatures

To further verify the diagnostic capabilities of the 3-miRNA marker, we collected six types of miRNA signatures in serum reported by previous studies and then conducted their ROC curve analysis in the four public datasets mentioned above. As shown in Table S1 and Fig. S2, the 3-miRNA signature has the highest AUC value in the TCGA-PRAD dataset compared to other miRNA panels. In the three GEO datasets, the performance of the 3-miRNA panel was only behind the 12-miRNA signature proposed by Bryant et al.. In conclusion, the 3-miRNA panel selected in this study is superior to other miRNA biomarkers in the diagnosis of PC (Porzycki et al., 2018; Moltzahn et al., 2011; Chen et al., 2012; Bryant et al., 2012; Yaman Agaoglu et al., 2011; Paunescu et al., 2019).

Prognostic value of the three miRNAs for PC

Using data from the TCGA-PRAD dataset, we constructed Cox regression models to estimate the correlation of several influence factors and overall survival (OS) and the three serum miRNAs’ prognostic value for PC. However, none of the three miRNAs were significantly associated with prognosis with P > 0.05 (Table S2).

Expression of the miRNAs in tissue samples

We next detected the expression levels of the three miRNAs in 28 pairs of tissue samples by qRT-PCR. As shown in Fig.S3, the results were not statistically significant in this stage. However, when we analyzed the three miRNAs in the un-matched tissue samples with larger sample size through The Cancer Genome Atlas (TCGA), as shown in Fig. 5, miR-146a-5p and miR-93-5p were found to be consistent with those in serum. In contrast, miR-24-3p was found down-regulated compared with NCs, which was opposite to the expression level of miR-24-3p in serum.

Expression of the identified miRNAs in serum exosomes

Additionally, the expression levels of miRNAs in exosomes were assessed in 24 PC patient samples and 24 NCs by qRT-PCR to explore the potential existing form of the identified miRNAs in PC patients’ serum. Compared to NCs, all the three miRNAs were up-regulated, but only miR-93-5p was statistically significant in serum exosomes (Fig. 6).

Bioinformatics analysis of the identified miRNAs

We assessed these experimentally identified miRNAs with DIANA-miRPathv3.0, a web-server aiming for pathway analysis. By performing the Kyoto Encyclopedia of Genes and Genomes (KEGG) and Gene Ontology (GO) analyses, we analyzed the validated miRNAs’ potential roles that might take part. As shown in Fig.S4A, KEGG analysis showed that these identified miRNAs might participate in some pathways such as AMPK signaling pathway, pathway in cancer, MAPK signaling pathway, TGF-beta signaling pathway, p53 signaling pathway, and so on. As demonstrated in Fig.S4B, GO analysis showed that in RNA metabolic activity, cell cycle, cell proliferation, cell death, and many other biological processes, the three miRNAs should not be ignored.