Plasma sampling

Glioblastoma (GBM) patients (GPs) were diagnosed and treated at the Department of Neurosurgery, Ljubljana University Medical Centre, Slovenia. The study was approved by the Republic of Slovenia National Medical Ethics Committee (Documents: 149/05/08 and 90/01/11). Upon obtaining signed informed consents, the blood samples were collected from 16 healthy individuals (HIs) recruited by the Blood Transfusion Centre of Slovenia in Ljubljana, and 16 GPs with a histopathological diagnosis of GBM (GPs). The inclusion criterion for HIs was their age (20–60 years) and their gender (half men, half women). The subjects with known acute and chronic diseases, pregnancy and those smoking and/or taking oral contraception or other drugs, were excluded from the study [23]. GPs had to fulfil the following inclusion criteria: age between 20 and 80 years and a confirmed diagnosis of a malignant glioma. The exclusion criteria for them were metastases and pregnancy. Each blood sample taken was screened for viral and bacterial infection markers of blood-transmittable diseases, i.e. HIV, hepatitis B, hepatitis C and syphilis, and in addition, the body mass index (BMI) was calculated for each individual enrolled.

Plasma samples of HIs were collected three times in the morning after overnight fasting, within a period of one month. The GP plasma samples were obtained prior to surgery, as well as 24 hours and one week after the surgery. Only three samples, collected 24 hours after the surgery, were included in the analysis. Another GP was included at the first occurrence of the disease and then once again at its remission, which was detected ten months after the first sampling. All GPs received pre- and postoperative corticosteroid therapy.

MicroRNA extraction

Peripheral venous blood was drawn into the BD Vacutainer CPT Cell Preparation Tubes containing Sodium Citrate (Becton Dickinson Biosciences, USA). Upon centrifugation, the plasma aliquots were frozen and stored at -80°C until RNA extraction was performed using the mirVana Paris kit (Ambion, USA) according to a slightly modified manufacturer's instructions and with applying additional purification steps (see S1 File. MiRNA extraction protocol). The quality and quantity of the total RNA including small RNAs was assessed using a NanoDrop 8000 Spectrophotometer (Thermo Scientific, USA) and an Agilent 2100 Bioanalyzer, using the SmallRNA kit (Agilent Technologies, USA). The total RNA extracts were kept at -80°C until further processing.

MicroRNA profiling

MiRNA expression profiling was performed using the nCounter miRNA Expression Assay Kits (NanoString Technologies, USA) at VIB Nucleomics (Belgium), allowing us to assay for a total of 734 human (654) and human-associated viral (80) miRNAs. The digital nCounter miRNA profiling technology is capable of accurately discriminating between miRNAs at a single-base resolution in a complex mixture. The system provides a direct digital readout of each miRNA needing only a small amount of the total RNA (100 ng), without requiring cDNA synthesis or enzymatic reactions. The method involves: mixing the total RNA with pairs of capture and reporter probes tailored to specifically recognize each miRNA present, hybridizing and washing away excess probes, immobilizing probe-bound miRNAs on a solid surface, and finally, scanning the different colour-coded bar tags on the reporter probes [24].

Quantitative real-time PCR

The total RNA that was extracted as described (see MiRNA extraction), was used as a template for cDNA synthesis, qPCR reactions and data analyses, which were all performed according to the manufacturer's protocols. The total RNA (1–13 ng) was reverse transcribed using the TaqMan MicroRNA Reverse Transcription Kit (Applied Biosystems, USA) to quantify all the selected miRNAs, and with the High Capacity cDNA Reverse Transcription Kit (Applied Biosystems, USA) to assess the amounts of 18S rRNA. In these experiments, TaqMan 2x Universal PCR Master Mix, No AmpErase UNG and TaqMan MicroRNA Assays (Inventoried Assays, Applied Biosystems, USA) were used for the quantitative real-time PCR analyses of the following pre-selected miRNAs: ebv—miR—BART2—3p (Assay ID: 006174), ebv—miR—BART9—3p (Assay ID: 007435), hsa—miR—302c—3p (Assay ID: 000533), hsv1—miR—H1—5p (Assay ID: 464923_mat), hsa—miR—193a—3p (Assay ID: 002250), hsa—miR—503—5p (Assay ID: 001048) and hsa—miR—383—5p (Assay ID: 000573).

According to the different approaches that have been used in previously published experiments, three standard endogenous reference controls, i.e. the eukaryotic 18S rRNA TaqMan endogenous control (Assay ID: 4333760F), the small nuclear RNA RNU6B (Assay ID: 001093) and the hsa—miR—16 (Assay ID: 000391), were included in the analyses [25,26]. The RNU6B and hsa—miR—16 were analysed using the protocol and reagents for miRNAs, while in case of 18S rRNA, a 2x TaqMan Gene Expression Master Mix was used (Applied Biosystems, USA). Selected samples were processed in the set-up for quantitative PCR analysis using a LightCycler 480 Real-Time PCR System (Roche Applied Systems, USA), under the following conditions: enzyme activation at 95°C for 10 min, 40 cycles of denaturation (95°C, 15 sec) and the same number of steps of annealing and extension (60°C, 60 sec). The analysis of the resulting data was performed as described by Gruden et al. [23].

Statistical analysis

The principal miRNA data analysis was performed using R software (http://www.r-project.org).

Since the sample volumes in the study were not completely balanced and the extraction efficiency varied from sample to sample, a proper normalization of results was eminent. Therefore a number of normalization protocols, including housekeeping stable miRNAs, the total miRNA content and the sum of positive controls, were tested. Based on the results obtained, normalization to the total miRNA content was chosen as the optimal data correction method. To reduce the problem with multiple testing in statistical models, the data were filtered for the nondetectable miRNAs (an expression value below the background in at least 80% of all samples) and also for hsa—miR—720 that is likely a fragment of a tRNA and was therefore removed from the miRBase [27]. Statistical analyses were performed on 284 miRNAs that met the upper criteria.

Hierarchical clustering analysis (HCA), a principal component analysis (PCA) and an analysis of variance were applied in order to get an overview of the sample variability, using the Multiexperiment Viewer software (MeV), where the criteria for data filtration were less strict [28]. Significantly altered miRNA plasma levels were statistically identified by the t—test or, in case of related samples, by the paired t—test.

In the survival analyses, the correlation between miRNAs detected in the plasma samples of GPs and their different survival times following the disease diagnosis was determined by the Kaplan-Meier statistics [29]. The optimal signal cut-off value that provided the most significant distinction between the sample groups was defined by stratifying the data according to six predetermined cut-off ratios: 0.25:0.75, 0.35:0.65, 0.45:0.55, 0.55:0.45, 0.65:0.35, and 0.75:0.25. The statistical significance of the differences (p—values) among these pre-defined group pairs was determined using the log-rank analysis. The relationships between the groups with the lowest p—values were represented by Kaplan-Meier graphs, following the application of corresponding functions within the R programme (R version 2.15.1, libraries KMsurv 0.1–5, knitr 1.2.10, patchDVI 1.9, survival 2.37–4).

Development of the microRNA analysis workflow in SegMine

To allow the analysis of miRNA datasets and their targets we extended the SegMine methodology with special workflow components (widgets) which enable the analysis of miRNAs. Firstly, miRNA expression data in a comma separated format can be loaded by using the Data parser widget. Secondly, the Ranker widget ranks microRNAs according to their ability to distinguish between the two sets of samples (control and treatment sets) using the ReliefF algorithm [30]. Thirdly, the list of ranked microRNAs is translated into a ranked list of targets (genes). For this task the SegMine integrates three microRNA target databases: MiRTarBase with experimentally validated targets and two algorithmically predicted target databases, DIANA-microT-CDS and Targetscan [31–33]. The translation process is performed by voting where microRNAs contribute their ranks to their target genes. If available, target probabilities are used in the process, e.g. in the DIANA-microT-CDS the target probabilities are present for all the targets, while the Targetscan contains aggregate probabilities for only some of them. The translation step is followed by the unmodified SegMine workflow which performs a search for enriched gene sets with the SEGS algorithm, clustering of SEGS rules describing gene sets, and Biomine link discovery between the biological entities (e.g. selected genes, clusters of genes, rules, clusters of rules) [34]. This extension also enables adding the aggregated values from voting to the graph in order to easily identify important targets. Finally, the reverse inference to identify microRNAs which contributed to the observed target is also supported (if available, the target probabilities are taken into account).

We have taken the following steps using the extended SegMine to interpret the differences in miRNA levels between the HI older than 40 years (HIo) and GP groups: 1) the dataset was filtered for miRNAs with absolute logFC (log fold change) values lower than 0.2; 2) ranked miRNAs were translated into a ranked list of targeted genes based on the data from the miRTarBase—section Validated targets [31]; 3) SEGS rules with the corrected p-value ≤ 0.05 were taken into further analysis; 4) hierarchical clustering of gene ontology rules with Ward linkage criteria considering a cutoff line at 11.23 was used to generate six rule clusters that were subsequently analyzed individually, and 5) the resulting Biomine sub-graphs were comprehensively reviewed prior to focusing on the gene hubs (important nodes with a large number of edges).

MicroRNA profiles in the plasma samples of healthy individuals (HIs)

MiRNA profiles in plasma samples were determined by using the nCounter technology. When conceiving the experimental design, special attention was paid to the sampling and sample processing in order to preserve miRNA stability and enable the standardization of the results. From the initial 734 miRNAs, 284 were included in the analyses following their normalization and omitting those below the detection threshold. If not specified otherwise, all the data were statistically evaluated by the t—test with a cut-off value of p = 0.01. In the group of HIs, the intra- and inter- individual variabilities of miRNAs were analyzed by using multivariate statistical methods (HCA, PCA). The coefficients of variation (CVs) represent the measure of physiological miRNA expression variability. Noteworthy, the assessed miRNA profiles showed greater intra—than inter-individual physiological variability, as presented in Fig 1.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. The comparison of intra- and inter-individual miRNA expression variability in the plasma samples of 16 HIs.

Variabilities are represented as boxplots showing miRNA expression coefficients of variation (CVs) for each individual and a complete dataset.

https://doi.org/10.1371/journal.pone.0125791.g001

The influence of BMI on individual miRNA profiles was tested in two groups of HIs, i.e. those with normal (≤24) and those having high (>24) BMI values. Only hsa—miR—323a—5p was significantly differently expressed, being elevated in a group of HIs with the higher BMI. Nevertheless, this miRNA was close to the limit of detection with only a few signals being positive in the group of HIs with the higher BMI (see S2 Table. The overview of the results). Therefore hsa—miR—323a—5p, despite emerging as differentially expressed, should only be potentially assumed to have a valid correlation with the elevated BMI.

Four miRNAs (hsa—miR—208a, hsa—miR—142—3p, hsa—miR—23a—3p and hsa—miR—1) were found to be age dependent (age < 40 versus age > 40 years) with the p—value < 0.05. All of them appeared decreased in the older subgroup of HIs. In a search of dissimilarities in miRNA expression between healthy men and healthy women, no differentially expressed miRNAs were found considering the p—value < 0.01. When the p—value < 0.05 was taken into account, ten miRNAs were found to be differentially expressed. Hsa—miR—1255a, hsa—miR—204, hsa—miR—651, hsa—miR—760 and kshv—miR—K12-4—5p were significantly increased, whereas hsa—miR—151—5p, hsa—miR—199a—5p, hsa—miR—361—5p, hsa—miR—423—3p and hsa—miR—423—5p, were significantly decreased within the male group of healthy individuals (HI).

Results of this part of research show trends that must be interpreted cautiously due to limited number of participants tested.

Identification of differently expressed microRNAs in the plasma samples of GPs and HIs

In the search for novel biomarkers specific to GPs; plasma miRNA profiles of GPs and HIs were compared. Some miRNAs were detected exclusively in one of the two tested groups. Still, only those that were found to be under the detection threshold in less than 20% of samples within both groups were further analyzed. Differences in plasma miRNA levels within the two groups were statistically analyzed with the t—test or, in the case of related samples, (GP samples, obtained before and after surgery) by using the paired t—test, always with a cut-off value of p = 0.01 (adjusted p-value). To assimilate and neutralize the data with respect to age, the analyses were performed between all the GPs and only those HIs being older than 40 years (HIo). To avoid false positive results due to plasma contamination with erythrocytes, blood mononuclear cells and/or platelets, we systematically examined the results for the presence of miRNAs, characteristic for the hematopoietic system, based on the previous findings of several research groups [14,35,36]. These results are presented in Table 1. Together, 102 miRNAs, out of which 41 down-regulated and 61 up-regulated, were found with significantly different plasma levels in GPs, compared to the HIo group. These also included 11 viral miRNAs. After only miRNAs with at least a two-fold difference in levels detected in the HIo versus the GP group were considered among all the significantly different ones, 19 more potential candidates for miRNA biomarkers were revealed. The most strongly down-regulated miRNA in GPs was the hsa—miR—383 (logFC = −1.85), whereas the most strongly up-regulated one was the hsa—miR—603 (logFC = 2.68).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. MiRNAs with altered plasma levels in the GPs group vs. the HIo group.

https://doi.org/10.1371/journal.pone.0125791.t001

The screening analysis was performed using the nCounter technology. MiRNAs with down—regulated plasma levels in the GP vs. the HIo group have the negative logFC values (fold change) and those up—regulated the positive logFC values (t—test, between subjects; just alpha; p<0.01), limited by the absolute logFC value of 1.0. The third column presents which miRNAs have already been correlated to GBM (glioblastoma) according to the miRTarBase [31]. Previous detection of explicit miRNA in the blood cells is marked in the fourth column [14,35,36]. Viral miRNAs are not included in the miRTarBase, therefore their correlation with GBM cannot be confirmed. A complete list of differentially expressed miRNAs is presented in S1 Table. A complete list of differentially expressed miRNAs. Legend: GP—patient with glioblastoma multiforme, HIo—healthy individual, older than 40 years, NA—not applicable;—and + label previously reported detection in GBM or in blood cells.

Plasma microRNAs expression and GP survival

To determine whether the survival of GPs following GBM diagnosis associates with any particular plasma miRNA level, the log-rank analysis was performed. This enabled us to define the signal values for each miRNA that discriminated best between the short-term and long—term survivors. The analysis revealed 24 miRNAs significantly correlated with the survival of GPs. In a group of long-term survivors, 20 of them were found to be significantly up—regulated: hsa—miR—101—3p, hsa—miR—1260a, hsa—miR—145—5p, hsa—miR—185—5p, hsa—miR—18a—5p, hsa—miR—1913, hsa—miR—29c—3p, hsa—miR—302c—3p, hsa—miR—30a—5p, hsa—miR—30d—5p, hsa—miR—30e—5p, hsa—miR—483—5p, hsa—miR—484, hsa—miR—493—3p, hsa—miR—525—3p, hsa—miR—548d—3p, hsa—miR—548d—5p, hsa—miR—566, hsa—miR—592, hsa—miR—620, and four appeared to be significantly down-regulated: hsa—miR—124—3p, hsa—miR—155—5p, hsa—miR—514a—3p and hsa—miR—653—5p. The Kaplan-Meier graphs were generated for each of these 24 miRNAs (see S1 Fig. Kaplan-Meier graphs of 24 plasma miRNAs having an impact on GP survival (p-value < 0.05)). Red curves symbolize patients with miRNA signal intensities above the cut-off value, while the black curves represent those below it. Coloured areas mark the confidence intervals. It was noteworthy that no viral miRNA was found to be associated with the patients’ survival.

All MiRNAs with prognostic potential, according to the log-rank survival analysis, did not also hold diagnostic value, as they were not significantly differentially expressed in the GP and HIo plasma samples. However, ten miRNAs overlapped in this respect and were therefore considered to be both diagnostic and prognostic, which makes them particularly appealing. These were hsa—miR—302c—3p, hsa—miR—592, hsa—miR—484, hsa—miR—1260a, hsa—miR—493—3p, hsa—miR—514a—3p, hsa—miR—145—5p, hsa—miR—30a—5p, hsa—miR—124—3p and hsa—miR—483—5p.

Analysis of the effects in microRNA changes at the level of cellular processes

The SegMine methodology was developed to enable the semantic analyses of gene expression data [37]. This was accomplished by integrating an advanced algorithm for gene set enrichment analysis (SEGS) and link the discovery within heterogeneous biological databases (Biomine) into a coherent data analysis workflow [38], [34]. We extended the SegMine methodology for the analysis of miRNA expression data at the level of cellular processes by linking miRNAs with their targets through available miRNA targets’ databases.

Six SEGS rule clusters describing the differences between the HIo and GP groups were generated using the miRNA analysis workflow within SegMine (Fig 2). As anticipated, the main miRNA clusters were linked with the gene ontology terms describing the cell proliferation processes or the ones involved in cancer signalling (Fig 2). Link discovery with Biomine revealed a generation of strong hubs, highly regulated by miRNAs that were found significantly differentially expressed between the HIo and GP groups. A thorough inspection of the Biomine sub-graphs revealed that the majority of gene hubs are cell cycle- or transcription-associated. Interestingly, one fraction of the nodes, having limited edges with very low weights, emerged to be linked to the main cluster. The fraction was found to cover the cytosolic processes, since the included genes code for numerous ribosomal proteins (see S2 Fig. The SegMine sub-graph showing a cluster of ribosomal proteins).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Hierarchical clusters of rules for the validated targets of miRNAs detected in the plasma samples of GPs.

Hierarchical clustering of the top 100 statistically significant rules (p≤0.05) is presented. The SegMine rules were derived from genes, representing validated targets of the GBM-related plasma miRNAs. Euclidian distance and Ward’s linkage criteria were used to compute the hierarchy.

https://doi.org/10.1371/journal.pone.0125791.g002

The intersection of the two Segmine clusters exposed proteins involved in the retinoblastoma (RB) signalling pathway; one of the most common and important mutated or deregulated pathways in GBM (Fig 3).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. A Biomine sub-graph showing the Retinoblastoma (Rb) signalling pathway.

Genes, involved in the RB signalling pathway are marked with an orange colour. The genes in the subgraph are covered by the rules from clusters 5 and 6 of the hierarchical clustering.

https://doi.org/10.1371/journal.pone.0125791.g003

The genes most frequently targeted by miRNAs, differentially expressed in the plasma samples of GPs and HIos and/or correlated to GP survival were: vascular endothelial growth factor A (VEGFA), heat shock 70kDa protein 1B (HSPA1B), actin beta (ACTB), heat shock protein 90kDa alpha (cytosolic), class A member 1 (HSP90AA1), insulin-like growth factor 1 receptor (IGF1R), cyclin D1 (CCND1), as well as phosphatase and tensin homolog (PTEN) (Fig 4).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 4. Genes most frequently targeted by miRNAs, correlated to the presence of GBM or patient survival.

The presented genes are validated targets of miRNAs differentially expressed in the plasma samples of GPs and the members of the HIo subgroup and/or correlated to patient survival according to the results of analyses of this study, obtained by using the miRTarBase [32]. VEGFA—vascular endothelial growth factor A, HSPA1B—heat shock 70kDa protein 1B, ACTB—actin beta, HSP90AA1—heat shock protein 90kDa alpha (cytosolic), class A member 1, IGF1R—insulin-like growth factor 1 receptor, CCND1—cyclin D1, PTEN—phosphatase and tensin homolog, BCL2—B-cell CLL/lymphoma 2, CCNE1—cyclin E1, CDK4—cyclin-dependent kinase 4, PPIA—peptidylprolyl isomerase A (cyclophilin A), TUBA1B—tubulin, alpha 1b, WEE1—WEE1 G2 checkpoint kinase, CCND2—cyclin D2, CDK2—cyclin-dependent kinase 2, CDK6—cyclin-dependent kinase 6, BIRC5—baculoviral IAP repeat containing 5, EP300—E1A binding protein p300, RRM2—ribonucleotide reductase M2.

https://doi.org/10.1371/journal.pone.0125791.g004

A reverse inference to identify miRNAs which have contributed to the observed target, supported by the SegMine methodology, provided us with a list of miRNAs, most frequently targeting genes included in the sub-graphs. Those targeting at least ten genes in the Biomine sub-graphs were: hsa—miR—320a, hsa—miR—484, hsa—miR—124—3p, hsa—miR—106b—5p, hsa—miR—103a—3p and hsa—miR—503—5p. All of them appeared correlated to the presence of GBM, with hsa—miR—484 and hsa—miR—124—3p being also associated with GP survival.

Viral microRNAs found in the plasma samples of GPs and HIs

To find out whether viral miRNAs might have a role in GBM, a comparative analysis between GP and HIo plasma samples was performed for 80 human-associated viral miRNAs included in the nCounter miRNA expression assay. Eleven viral miRNAs were differentially expressed in the plasma samples of GPs, when compared to those of HIos. Six of them were Epstein Barr virus (EBV) miRNAs (ebv—miR—BART2-3p, ebv—miR—BART2-5p, ebv—miR—BART6-3p, ebv—miR—BART9, ebv—miR—BART15 and ebv—miR—BHRF1-3), 2 Herpes Simplex Virus (HSV) miRNAs (hsv1—miR-H1 and hsv1—miR-H4-3p), 1 Kaposi Sarkoma Human Virus (KSHV) miRNA (kshv—miR-K12-7) and two Human Cytomegalovirus (CMV) miRNAs (hcmv—miR-US5-2 and hcmv—miR-US33-3p).

Nine of them were up-regulated (ebv—miR—BART15, ebv—miR—BART2—5p, ebv—miR—BART6—3p, ebv—miR—BART9, ebv—miR—BHRF1-3, hcmv—miR—US5-2, hsv1—miR—H1 and kshv—miR—K12-7) and two (ebv—miR—BART2—3p and hsv1—miR—H4—5p) down-regulated in a group of GPs, compared to HIos (Fig 5).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 5. Log2 expression values of 11 viral miRNAs, differentially expressed in the GP and HIo plasma samples.

The bars represent the mean values and x each analysed plasma sample.

https://doi.org/10.1371/journal.pone.0125791.g005

These results suggest that viral miRNAs may play an important role in the tumorigenesis of GBM.

Verification of nCounter technology results by real-time PCR (RT-qPCR)

RT-qPCR-based methods are mostly used for the validation of the results obtained with different microarrays and other high-throughput methods. For the confirmatory qPCR analysis, seven selected, differentially expressed miRNAs, ebv—miR—BART2—3p, ebv—miR—BART9—3p, hsa—miR—302c—3p, hsv1—miR—H1—5p, hsa—miR—193a—3p, hsa—miR—503—5p and hsa—miR—383—5p, were used. In addition, three standard endogenous reference controls, i.e. 18S rRNA, RNU6B and hsa—miR—16, were also included. No, or only a few signals were detected for ebv—miR—BART2—3p, ebv—miR—BART9—3p, hsa—miR—302c—3p, hsv1—miR—H1—5p, hsa—miR—193a—3p, hsa—miR—383—5p and RNU6B, while in the case of hsa—miR—503—5p, the results were not reproducible due to the high variability of the replicates. Reproducible results with the corresponding C_q values were obtained only for the endogenous reference controls 18S rRNA and hsa—miR—16. Although RT-qPCR methods for miRNA quantification are relatively inexpensive, commonly available and allow measurements of very small quantities of miRNAs, still the amounts of circulating miRNAs in the peripheral blood are often below their limit of detection [21]. Therefore, due to the low concentrations of circulating miRNAs found in the GP’s and HI’s plasma, RT—qPCR-based detection does not result in appropriate quantifications.