Background
Glioblastoma multiforme (GBM) is the most common and aggressive primary adult brain tumor. Despite advances in treatment modalities, the prognosis of GBM patients is very poor[
1]. Therefore, it is urgent to develop new diagnostic and prognostic tools and treatments which may be beneficial for improving the clinical management of GBM. Currently, tumor stratifications relying on molecular profiles are increasingly prevalent and important. Furthermore, molecular and genetic profiling studies have identified several prognostic and predictive markers for GBM[
2,
3].
MicroRNAs (miRNAs or miRs), which are endogenous non-coding small RNAs, post-transcriptionally regulate gene expression through inhibition of translation or degradation of target mRNAs[
4]. MiRNAs are aberrantly expressed in a variety of tumor types and exert important regulations on tumor biology via acting as oncogenes or tumor suppressors[
5]. Recently, several studies indicate that expressions of miRNAs are associated with patients’ survival and are able to function as prognostic and predictive indicators[
6,
7]. Moreover, it has been confirmed that miRNA expression profiles are more accurate to classify tumors than mRNA profiles[
8]. However, in the selection of miRNA markers for GBM prognosis, the applying of following aspects, such as small dataset, explanatory variables, single miRNA analysis, pre-selection of miRNAs and use of approaches, finally lead to a variety set of different miRNA markers.
The main purpose of this study is to identify specific miRNA markers that are closely associated with tumor progression and survivals for GBM patients by analyzing significantly altered miRNAs in a large dataset. Another goal is to investigate the availability and rationality of interactions of interested miRNAs as prognostic and predictive indictors for clinical outcome of GBM patients. In this study, we found that miR-326, miR-130a, miR-155, miR-210 and 4 miRNA-interactions could function as prognostic and predictive markers for survival of GBM patient.
Discussion
In this study, we identified 38 differentially expressed miRNAs from the most significantly altered miRNAs using data from TCGA dataset. Kaplan-Meier survival and Cox multivariate proportional hazard model confirmed that the expression of miR-326, miR-130a, miR-155 and miR-210 were correlated with OS and PFS of GBM patients and were verified for the first time as independent predictors for GBM patient survival. More importantly, interactions between miR-323/miR-326/miR-329 and miR-130a/miR-155/miR-210 were also significantly related with clinical outcome and were more sensitive to discriminate and predict survival time of patients. Moreover, interactions of miR-323 and miR-130a, miR-326 and miR-155, miR-326 and miR-210 and of miR-329 and miR-130 were also confirmed as independent prognostic indicators to clinical outcome of GBM patients. In addition, the availability and rationality of these interactions as independent prognostic and predictive indicators were supported by integrated analysis of network, biological processes and correlation analysis with confirmed GBM gene markers. Our results suggest a potential application of miRNA profiles and their interactions in development and improvement of prognostic tools and treatments.
Presently, except that presurgical prognosis relies largely on age and Karnofsky Performance Status (KPS), no convincing prognostic and predictive factors have been prevalent in clinical management of GBM patients, although several prognostic and predictive markers or models have been proposed or developed, such as MGMT promoter methylation[
25], BRAF fusions and IDH1 mutations[
26], subclassification based on gene expression[
2], Immunohistochemical analysis[
27] and Volume-Age-KPS (VAK) prognostic model related with MR-imaging[
28]. Notably, due to development of molecular and gene profiles, molecular stratification for patients’ outcome are increasingly emphasized, which leads to the extensive investigation and exploration of molecular markers.
MicroRNAs, as a family of small non-coding RNAs which are negatively involved in gene regulations, have been recognized as important intervention targets and predictive tools for several diseases because of the stability and convenience of miRNA detection[
29‐
31]. Actually, several study groups have identified a pool of miRNA signatures for clinical outcome prediction. Through screening expression profiles of 200 miRNAs from 84 astrocytoma samples, miR-106a, miR-181b and miR-21 were identified as diagnostic and prognostic markers in defining the signature of astrocytomas and predicting the post-surgical outcome[
6]. In another study including 38 GBM samples, miR-21, miR-181c, miR-195, and miR-196b were associated with survival of GBM patients[
32]. Using TCGA dataset with 253 individuals, 23 and 19 miRNAs were defined to be associated with OS and PFS, respectively[
7]. Also, in another publication with 222 GBM samples, a risk score, formulated on the basis of expression signatures of 10 miRNAs, was associated with GBM patient survival, which was suggested to predict GBM patient survival[
33]. On one hand, all these studies indicated that miRNAs were thoroughly involved in GBM biology and several miRNAs could act as predictive and classified indicators for GBM clinical outcome. On the other hand, one concern has been aroused that all the identified miRNAs were almost totally different among these publications, which may be due to different uses of approaches or pre-selections of target miRNAs and so on. In this study, through calculating, sorting and overlapping mean Z-score values in GBM samples from three separate batches, we obtained the most altered miRNAs, which ensured that these candidate miRNAs were more specific and accurate to distinguish expression differences between GBM and normal brain tissues. Herein, the candidate miRNAs in this article were more convincing and feasible for further potential application in clinical practice.
This study did not follow the conventional training and validation test analysis. However, selection bias was yielded and validation of our findings was supported though corroborations as follows. First of all, all miRNAs were selected from the top most altered and overlapped miRNAs which were sorted according to mean Z-scores originated from 3 independent batches. Then, all expression levels of interested miRNAs have been validated on miRNAMap and other independent miRNA detections, which could be considered as external validations. Furthermore, biological function of the interested miRNAs and their target genes were analyzed, which may directly reflect the roles of miRNAs in tumor progression.
Among the 6 interested miRNAs, miR-326 was reported to inhibit GBM cell growth, whereas miR-155 was shown to promote GBM proliferation[
34,
35], which could be explained by that the target genes of miR-326 and miR-155 were mostly related with apoptosis (Figure
4A). However, according to our knowledge, there is no study reporting the associations between OS and PFS and miR-326/miR-155, while our result for the first time showed that high level of miR-326 and low level of miR-155 were significantly associated with long OS and PFS. Likewise, we first found that low levels of miR-323 and miR-329 correlated with long OS, and high level of miR-130a and low level of miR-210 were linked with extended either OS or PFS. These survival analyses indicated that miR-326 and miR-130a functioned as tumor suppressors while the others as oncogenes. However, it should be noted that expression levels of miR-323/miR-329 were elevated in no-recurrent and longer TTP patients, which were not consistent with oncogenic roles of miR-323/miR-329. Several reasons may be responsible for this inconsistence. Initially, it has been confirmed that on average one miRNA has approximately 100 target sites, regulating a large fraction of protein-coding genes involving in several biological processes, such as cell proliferation, apoptosis, and cell motion etc.[
36]. Second, putative targets of miR-323 and miR-329 incorporated a family of molecules associated with cell migration and adhesion as shown in Figure
4A. Furthermore, the recurrence of GBM is related with these migration and adhesion genes[
37]. Herein, miR-323/miR-329 may be involved in migration inhibition in non-recurrent patients through elevation of their expression levels. This inconsistence also occurred to miR-130a, which was shown to not only inhibit tumor suppressor RUNX3 in hepatocellular carcinoma[
38] but also suppress proto-oncogene MET in lung cancer. This may be due to extensive distribution of predictive targets of miR-130a[
39]. To date, there is no functional study related with miR-323, miR-329, miR-130a and miR-210 in GBM.
The complication of biological function of these miRNAs also indicated that it may be more reasonable to study their interactions, because of the multifactorial nature of the disease, and the distinguishing feature of miRNAs that an average miRNA has approximately 100 target sites and regulates a large fraction of protein-coding genes, which form a regulatory network[
36,
40]. OS and PFS analysis showed that the two-miRNA interaction were more sensitive and accurate to discriminate and predict the survival time in relative to one single miRNA. For instance, the longest gap of mean survival time of OS and PFS occurred on miR-130a, with 8.1 months and 5.4 months, respectively. However, the longest gap of mean survival of OS and PFS was 20.9 months and 15.8 months related with the interaction of miR-326 and miR-130a. Moreover, this interaction effect made the stratification of patients’ survival more detailed and specific. For example, the mean OS of patients with high and low miR-130a was 26.6 months and 18.5 months respectively, whereas the corresponding survival of patients with both low miR-323 and miR-130a was 30.9months, and that with both low miR-323 and high miR-130a was 14.4 months. Therefore, the interaction analysis of miRNAs may provide new views on diagnosis and prognosis of GBM patients.
In summary, we identify miR-326, miR-130a, miR-155 and miR-210 markers related with survival of GBM. More importantly, we determine the availability and rationality of 4 miRNA-interactions as more specific and accurate prognostic and predictive indicators to clinical outcome of GBM patients, implying the application for diagnostic and prognostic tools and treatments.
Competing interests
All authors declared no conflicts of interest.
Authors’ contributions
QSW designed the study, carried out data analysis and drafted the manuscript. LS performed bioinformatics analysis. HD, FYM and TY participated in the collection and analysis of the data. PY conceived of the study and participated in its design and coordination and helped to draft the manuscript. All authors read and approved the final manuscript.