Introduction

Glioblastoma multiforme (GBM) is a severe type of brain cancer characterized by its large growth potential and very poor clinical outcome. It is one of the most aggressive and incurable types of cancer reflected in a median survival of less than 1 year of all GBM cases and a 5-year survival rate of less than 5 % [1, 2]. GBM affects 2–3 per 100,000 persons per year making it a rare type of cancer, but still, it accounts for 16 % of all brain tumors and 54 % of all clinically diagnosed gliomas in the USA [2]. The clinical presentation of GBM depends on the location of the tumors and generally involves focal neurological deficits, headaches, and seizures. Tumors are most commonly found in the frontal lobes of the supratentorial compartments; however, they are not restricted to these areas, as illustrated by GBM tumors found in other parts of the central nervous system (CNS), such as the spinal cord and brainstem [3].

GBM can be subdivided into de novo-occurring tumors, termed primary GBM, or tumors developed from lower-grade astrocytomas, termed secondary GBM. The most prominent occurring subtype of GBM is the primary tumors. These tumors are often characterized by amplification or overexpression of the epidermal growth factor receptor (EGFR) (40–60 % of all primary GBM tumors) combined with genetic alterations in the EGFR gene, which results in mutated forms of this receptor [1]. This is opposed to secondary GBM, which is characterized by progressive addition of mutations in p53, platelet-derived growth factor receptor, and the retinoblastoma gene [4–6]. Nevertheless, this distinctive division of mutations into the different GBM subtypes is not definitive [3].

Determination of disease prognosis is most often based on histological classification combined with information on patient age and tumor size and location. These factors have all been defined as indicators of patient survival and treatment outcome, but due to the sustained poor overall survival of GBM patients, new arrays of prognostic indicators have been requested to aid in the clinical decision making [1]. In recent years, several molecular biomarkers have been characterized including chromosomal aberrations, methylation status of the methyl guanine methyl transferase (MGMT) promoter, mutations in important genes (isocitrate dehydrogenase 1 (IDH1), EGFR, and p53), and dysregulation of microRNAs [7]. Loss of heterozygosity in chromosomes 9p and 10q is associated with decreased survival, while co-deletion of 1p and 19q correlates with better treatment response and longer survival [7]. Hypermethylation of the MGMT promoter leads to lower expression levels of MGMT, which sensitizes GBM tumors to chemotherapeutic treatment and thus is associated with a significantly better patient outcome [8, 9]. Improvement of the disease condition is also observed in patients with mutation in the IDH1 gene, which is most often found in secondary GBM. Furthermore, the expressional profile of specific microRNA signatures also correlates with overall survival, time to progression, and treatment success [10–12].

The Fundamentals of MicroRNAs

MicroRNA (miRNA) is a class of non-coding single-stranded RNA comprised of approximately 22 nucleotides with the ability to negatively regulate gene expression posttranscriptionally [13, 14]. miRNAs bind to the 3′ untranslated regions (UTRs), and sometimes 5′UTRs, of their messenger RNA (mRNA) targets, to which they exhibit imperfect complementarity, hence, enabling one miRNA to inhibit translation of multiple genes [15, 16]. The first miRNA was discovered in 1993 in Caenorhabditis elegans and denoted lin-4 [17]. Later, upon the discovery of let-7, found to be conserved in several species, miRNA regulation was recognized as an omnipresent phenomenon in eukaryotic organisms [18, 19]. miRNAs are acknowledged as crucial micro-modulators of normal cellular homeostasis, and accordingly, dysregulation of miRNAs have been associated with a wide range of pathological conditions, such as cancer [20], cardiovascular disease [21, 22], and autoimmune [23] and neurodegenerative disorders [24]. Expression of miRNAs in pathological specimens or biofluids, compared to non-pathologic samples, is subject to great scientific efforts [25]. This poses interesting perspectives in terms of novel diagnostic and prognostic approaches and is inherently the initial step in uncovering the role of individual miRNAs in the context of different diseases, eventually paving the way for novel miRNA-based therapies.

MicroRNA Biogenesis

To understand the context of miRNA as a potential prognostic tool in patients with GBM, the essential steps in the biogenesis of miRNAs and the modes by which they exert their repression on downstream targets are summarized (see Fig. 1).

Fig. 1
figure 1

The biogenesis of miRNA requires RNA polymerase II/III for the transcription of pri-miRNA. The pri-miRNA product is then cleaved by the Drosha-DGCR8 complex into pre-miRNA. The pre-miRNA is exported to the cytoplasm by Exportin-5 in the presence of Ran-GTP co-factor. Once in the cytoplasm, the pre-miRNA is cleaved by the Dicer-TRBP complex into a miRNA duplex, which is unwound into two products: a guide strand bound to Ago2, which is incorporated into the RISC, and a passenger strand, which is degraded. Finally, the miRNA binds to its target mRNAs resulting in mRNA target cleavage, translational repression, or mRNA decay. A more novel fate of the miRNAs is the selective secretion via microvesicles or exosomes. Ran = Ras-related nuclear protein; GTP = guanosine-5′-triphosphate; TRBP = TAR (HIV-1) RNA binding protein; Ago2 = Argonaute protein 2; RISC = RNA-induced silencing complex

Full size image

The linear biogenesis of miRNA begins with the transcription of miRNA genes by RNA polymerase II/III, giving rise to a primary transcript called pri-miRNA, which is subsequently polyadenylated and capped. The transcript then folds into a hairpin-loop structure via intrastrand base-pairing [26]. This structure is cleaved by the Drosha/DGCR8 complex to become pre-miRNA and transported out of the nucleus by Exportin-5 in a Ran-GTP-dependent process [27]. In the cell cytoplasm, the RNAse-III enzyme known as Dicer cleaves the pre-miRNA of which only one strand (known as guide strand) is incorporated into the RNA-induced silencing complex (RISC), the cytoplasmic effector machine of miRNA. The passenger strand is subsequently degraded [28]. The RISC is comprised of Dicer, double-stranded RNA-binding factor, and Argonaut protein 2 (Ago2). The posttranscriptional RNA silencing is facilitated via imperfect complementary binding of miRNA attached to RISC, to the respective mRNA 3′UTR, resulting in translational inhibition [29]. Additionally, miRNAs are selectively excreted via lipoproteins or microvesicles, potentially functioning as a mode of intercellular communication. This last notion is important in relation to the nature of sampling material in the sense that plasma miRNA patterns might be a useful diagnostic and/or prognostic marker of ongoing pathological processes [30, 31]. For a more comprehensive review of miRNA biogenesis, please refer to Winter et al. [26].

MicroRNA Expression in Glioblastoma Multiforme

miRNAs can be regarded as cancer biomarkers when their variation in expression identifies the cancerous state. To date, almost all tumor tissue analyzed by miRNA profiling has provided distinct miRNA profiles compared to normal tissue [32]. These differential profiles can be further associated with prognostic factors and disease progression [33–35]. In GBM, the number of studies pertaining to miRNA expression and functional characterization has grown and miRNA signatures are refining GBM classification, differentiating between the different grades and stages, providing key regulatory links to disrupted signaling pathways such as those facilitating cell growth. This has lead to a more in depth understanding about GBM pathology [36].

Early studies show that miRNA expression in tumor samples seems lower, and this could be a function of cellular differentiation status [32, 37]. It appears that the most common dysregulation of miRNA in GBM is observed to be overexpression, based on the systematic literature review published by Møller et al. Here, for example, miR-17, miR-21, miR-93, miR-221, and miR-222 have been intensively investigated with respect to both their expression and functionality, but the functional properties of the vast majority remains completely unknown [38]. The most extensively investigated miRNA is miR-21, which is consistently reported to be overexpressed in GBM in a grade-specific manner [12, 39–68]. At least for GBM, miR-21 appears to be the major anti-apoptotic and pro-survival factor that is linked to shorter progression-free survival [12, 69, 70].

Expression profiling of miRNA in patient tissue and investigation of their putative function using in vitro primary cultures and in vivo studies have provided an insight not only into the genes that are regulated by respective miRNA, but also the pathways that are disrupted, many of which are hallmarks of GBM biology (reviewed by Lakomy et al. [12]). The pattern of miRNA expression, whether its up or downregulation, is now becoming a recognized tool in addition to gene expression profiling to stratify GBM patients into different groups [36]. Here, the miRNA cohorts are smaller and miRNA signatures pertaining to overall or progression-free survival are starting to evolve, albeit they are still very much dependent on the individual patient history, tumor size, age, and treatment regimen.

Overall and Progression-Free Survival as Clinical Endpoints in Glioblastoma Multiforme

In the literature, both overall survival (OS) and progression-free survival (PFS) are widely used end points to assess the predictive factor of a given miRNA signature; however, the two terms do not provide equal information [71–73]. When evaluating a treatment response, OS is used as a measure of the end result including the complete disease history and possible other factors affecting the lifespan. The PFS is more specific in its measure of the effect of a specific treatment in the form of tumor control or radiographic response. Reviewing the literature and trying to draw conclusions are therefore challenging when both OS and PFS are applied [74]. The response assessment criteria for GBM has been developed over the course of several decades as a result of technology advances in imaging and expanded knowledge on tumor biology. Before 1990, the Levin and WHO Oncology Response Criteria, which primarily was based on contrast-enhanced computer tomography, was the standard assessment methods [75]. These were substituted by the standardized McDonald Criteria, which took into account that contrast enhancement could be affected by clinical factors such as the use of corticosteroids [76]. The McDonald Criteria incorporated the clinical assessment (neurology status) of the patient in the designation of response to therapy as being a complete response, partial response, or stable or progressive disease. With the arrival of magnetic resonance imaging and new therapeutic options, the response assessment criteria was developed further and standardized with regard to all aspects of imaging, timing, and evaluation techniques. Especially the introduction of bevacizumab, a monoclonal antibody targeting VEGF-A and a resulting increased risk of pseudoprogression interpreted as disease progression stimulated the modification and lead to the Revised Assessment in Neuro-Oncology (RANO) Criteria in 2010 [77–80]. Because of this development over the last 20 years, caution should be taken in the comparison of particular PFS data.

Aim of the Review

This review is aimed at providing an up-to-date account of the miRNA expression profiles in tumor tissue associated with prognosis and survival in GBM. It is meant as an updatable list of studies and signatures that have been linked to the progression of GBM, to give an account of the miRNAs, which have been reported to be suitable as a prognostic factor for short- or long-term survival. Plasma miRNA expression has also been associated with survival in GBM, but this is not within the scope of this review [25]. Based on the literature, the studies are stratified into those based on publicly available databases and those conducted on independent tissue cohorts. These data sets have been extensively reviewed and combined to derive a signature or pattern of miRNAs, which has a prognostic potential. The miRNAs reported to have a protective or risk-associated profile have been highlighted in relation to GBM. Finally, the studies that have reported a miRNA signature with respect to prognosis have been compared to find common miRNA profile across the different studies.

Methodology and Delimitations

A Medline database search on “microRNA, glioblastoma, survival, prognosis and progression” (typed: “(microRNA OR miRNA) AND (glioblastoma OR glioma) AND (survival OR prognosis OR progression)”) was performed (date of last search entry: November 26, 2013). The results contained a total of 270 papers; 125 of these were chosen based on title and abstract content. Of the remaining, 100 papers were cell culture studies and 45 were reviews or review like and were therefore excluded. A total of 125 papers were reviewed for miRNA expression level in GBM correlated to survival and/or progression, 25 involved database studies, and 35 contained studies on GBM tissue (not database-derived). The miRNA profiles, often presented in the form of signatures, were extracted from the papers. This review summarizes the studies investigating miRNAs in GBM and explores their correlation to clinical outcome and highlights the functional characteristics of the miRNAs linked to protection (i.e., longer survival) or risk (i.e., shorter survival). The miRNAs that are included in the signature of more than one study and involved in the progression of GBM have been identified, and their functional role, if known, is discussed.

Prognostic MicroRNA Signatures in Glioblastoma Multiforme

MicroRNA Signatures Derived from Database Mining

A total of 25 studies were based on database entries. For the individual studies, the database accessed, cohort size, cohort factor, and normalization methodology along with miRNA signature were documented (Table 1). The majority of the studies used The Cancer Genome Atlas (TCGA) (http://cancergenome.nih.gov). However, four studies used the Chinese Glioma Genome Atlas (CGGA) (http://www.cgga.org.cn), which uses the Illumina Human v2.0 miRNA Expression BeadChip microarray platform [11, 81–83]. The study by Ma et al evaluated two large cohorts of data, the CGGA with 198 samples containing low-grade gliomas and 91 GBMs and an additional cohort of 128 samples, with low-grade gliomas and 68 GBMs to validate the array data. High expression of miR-196b was conferring poor prognosis when stratifying the patients into high miR-196b expression and low miR-196b expression groups [82]. Following a similar experimental setup, Wu et al. looked at 91 GBM patients taken from the array data and validated their findings in a cohort of 60 GBM patients. Here, they focused on miR-328, showing that a low level of expression was conferring poor prognosis [83]. The TCGA dataset has also been used for developing a new method for predicting the outcome based on miRNA expression; however, only one of the studies provided the miRNA identified [84].

Table 1 Studies performed on dataset obtained from public databases
Full size table

While a few studies gave rise to a defined multiple-miRNA signature, eight of the studies looked at a single miRNA. For example, two studies evaluated the functionally well-characterized miRNA, miR-10b. Gabriely et al. showed that miR-10b was expressed in GBM tissue while not present in normal brain tissues. Using TCGA data, they investigated the association between the expression of miR-10b and clinical outcome and found that miR-10b correlated with survival although with stratified conditions, the association was insignificant. When the correlation with survival for miR-10b was assessed together with miR-10a, however, the association with survival was significant regardless of stratification; hence, high levels of miR-10 conferred poor survival [85]. Guessous et al. also found a correlation between high levels of miR-10b and poor survival by analyzing the TCGA data and further reported on a functional role of miR-10b in GBM stem cells [86].

Since the majority of the studies use the TCGA dataset in analyzing the expression of miRNAs in GBM, the platform for generating the data was the same. The only differences seen were in the downstream analysis, other clinically prognostic factors, and the type of filtering applied. Expression analysis was conducted on Agilent 8 × 15 K Human miRNA microarrays, with data available at four levels. The first level is the raw non-normalized data from the array (level 1), and the second level (level 2) is the processed normalized signal. The third level (level 3) is the segmented data, assemblies of the processed data from single samples, and grouped by probed loci to form larger contiguous regions. The fourth level (level 4) is the summary, a quantified association across classes of samples and associations based on molecular abnormalities, sample characteristics, and/or clinical variability. Not all studies state the level of data they use, but most use the third level. Only one study, looking at a ten-miRNA signature, used level 1 and quantile-normalized the expression data. Here, they segregated patients in to high- and low-risk groups and identified seven miRNAs associated with high risk of disease progression and three miRNAs that were found to be protective [87]. A more elaborate study by Delfino et al. used quantile-normalized data (although it does not specify the level of the data set analyzed) and identified 45 miRNAs in the TCGA data across race, gender, recurrence, and therapy linked to survival [71]. Using level 2 data with multiarray algorithm normalization, Zinn et al. looked at 78 patients and included a Volume (tumor volume), Age, and Karnofsky Performance Score (VAK) classification to dichotomize the patients into VAK A (good prognosis) and VAK B (poor prognosis). A total of five miRNAs were associated with short-term survival (miR-566, miR-505, miR-345, miR-484, and miR-92b), and three miRNAs were associated with long-term survival (miR-511, miR-369-3p, and miR-655) [88].

Though the normalization was not standard for all studies along with the variation in cohort size, most of them used normalized data from the TCGA database and therefore had the same material. The cohort size, however, ranges from 170 to 580 (mean = 329 ± 121) and is a function of the cohort factors that are investigated. For example, factors such as age, grade, MGMT methylation, chemotherapy regimen, IDH1 mutation, or grade sub-classification are just some examples of where the investigation is based on prior knowledge of clinical data correlated with expression and survival [11, 36, 83, 89, 90]. In addition to the TCGA, a number of the studies have used independent GBM tissue validation cohorts for identifying differentially expressed miRNAs with respect to cohort factors [82, 83, 91].

MicroRNA Signatures Derived from Independent Tissue Cohorts

The studies performed on independent sample sets can generally be characterized as being validation of database findings, validation of literature findings, or novel array-based determination of miRNA profiles of clinical interest in GBM (Table 2). A total of 35 studies identified miRNA signatures associated to survival, and many of these have used tissues to validate signatures previously found in datasets described in the last section and contained in Table 1. The majority of studies (n = 30) use PCR-based methods when validating miRNA expression, while several studies use different types of arrays. The PCR-based methods require normalization, and most of the studies use RNU6B, though there are a few studies that use others, such as hsa-miR-16 or RNU5A [51, 92].

Table 2 Studies performed on independent tissue cohorts
Full size table

The starting material used in the studies was either tissue or formalin-fixed paraffin-embedded (FFPE) tissue with a variable cohort size (min = 12, max = 253, mean = 91 ± 59). With regard to sample preparation, de Biase et al. have shown that there is no difference in the miRNA expression obtained from tissue and FFPE tissue, and some studies also use both types to validate their findings [93]. Twenty-five studies focused on single-miRNA candidates, while the remainder focused on expression profiles of several miRNAs (min = 1, max = 30, mean = 3 ± 5).

While most of the studies focus on smaller miRNAs signatures, Niyazi et al. present a larger cohort of miRNA as a putative survival signature. They used a top-down approach, where they filtered the miRNAs based on the variance in expression across the samples and chose the 30 most dysregulated miRNAs. These miRNAs were used to stratify the samples into two patterns, which correlated with short- and long-term survival [94]. This approach was also applied in several database studies, limiting the number of miRNAs down to a specific signature [51, 84, 90, 91]. Others looked at pre-selected miRNAs already linked to GBM pathogenesis in the literature [92, 95, 96]. Zhang et al. found that miR-221 and miR-222 expression was significantly increased in high-grade gliomas compared with low grade, positively correlated with degree of glioma infiltration. This corresponded well to the fact that overexpression of miR-221 and miR-222 increased cell invasion [12]. In addition, Quintavalle et al. showed that miR-221 and miR-222 were upregulated in GBM patients and that they target MGMT mRNA thereby inducing greater temozolomide-mediated cell death [92].

Of all the studies, 12 of them utilized both databases and independent tissue cohorts. They all link an expression of one or more miRNAs to survival; however, some studies categorize a given miRNA to be protective or risk-associated. Three studies used hazard ratio to assess whether a specific miRNA was protective or risk-associated, while Wang et al. used a Significance analysis of microarray (SAM) and Li et al. used the Cox-regression coefficient to designate the miRNAs [11, 84, 87, 90]. Interestingly, large variations can be found in the choice of control tissues across the individual studies, ranging from purchased RNA from normal brains to tissues from epilepsy patients or patients with cerebral trauma. Such differences in control tissues might also be a factor in the incoherency between the miRNA signatures found in the different studies (Table 2). In addition, only few studies specify their use of the terms OS and PFS, which also makes direct comparison difficult.

MicroRNA Reported to be Protective or Risk-Associated

A number of studies provided miRNA signatures associated with survival or progression in GBM and reported that individual miRNAs of these signatures could be regarded as either protective or risk-associated. These miRNAs and their functional role in GBM pathogenesis and progression are presented in Table 3.

Table 3 miRNAs reported to be protective or risk-associated
Full size table

Clinically Protective MicroRNAs

In the group of the protective miRNAs (n = 22), only two miRNAs, miR-544 and miR-1227, have not been described previously in relation to miRNA alterations in GBM pathogenesis. Eleven of the protective miRNAs were significantly increased in studies comparing GBM specimens to normal brain tissue, while only six were significantly downregulated. Surprisingly, three of these miRNAs are well described as miRNAs with an oncogenic potential and have several validated targets considered to be tumor suppressor genes. This includes the extensively investigated miR-17-5p, which in vitro has been shown to increase angiogenesis and growth when overexpressed and decrease viability and proliferation when inhibited, making it unlikely that this miRNA, at least solely, should be considered protective [97]. As to miR-19a, miR-19b, and miR-106a, there is currently an inconsistency in the literature regarding their role in GBM development. The functional data available on miR-106a shows that overexpression by transfection of GBM cell lines causes a significant decrease in proliferation and an increase in apoptosis, likely mediated by the suppression of E2F1, supporting the notion of it being tumor suppressive [98]. More in line with what would be expected, miRNAs with previously investigated tumor suppressive capabilities are present on the list whereby miR-128a and miR-181d are most notable. miR-128 has been investigated in 13 studies demonstrating its wide range of oncogenic mRNA targets and its ability to inhibit angiogenesis and proliferation and even to significantly decrease total tumor volume in vivo [12, 40–42, 48, 56, 62, 66, 67, 99–103]. Similarly, although, less extensively investigated is miR-181d, which has been shown to target the oncogenes Bcl-2 and K-Ras whereby apoptosis is increased and proliferation decreased. miR-181d transfection is demonstrated to decrease in vivo tumor size and has been shown to increase the susceptibility to the chemotherapeutic agent, temozolomide [104, 105].

Risk-Associated MicroRNAs

Within the cohort of miRNAs described as risk-associated (n = 22), nine have not been previously associated with miRNA modulation in GBM. Of the 13 miRNAs mentioned in the literature, ten are overexpressed in GBM specimen, three are underexpressed, and four have been functionally characterized. miR-34a is well studied in numerous GBM cell lines and shown to increase cell differentiation and decrease total tumor volume in a xenograft mouse model of GBM [106, 107]. The less investigated miR-146b is similarly known to decrease in vitro invasiveness, migration, proliferation, and tumor volume in mice [108, 109]. Both miR-34a and miR-146b are, in terms of isolated functional characteristics, not associated with risk of GBM progression (Table 3). The oncogenic miRNAs, miR-221 and miR-222, clinically associated with risk, have been studied in relation to a diverse list of cancers including GBM. They inhibit a number of common gene targets such as PUMA and P57 both involved in apoptosis. When overexpressed in vitro, both miR-221 and miR-222 potentiate classic cancer hallmarks, i.e., proliferation, angiogenesis, and invasion. In vivo studies have revealed that miR-221 or miR-222 overexpression is associated with increased tumor growth, a situation that can be reversed with administration of corresponding antagomirs [110, 111].

Out of 44 miRNAs reported to be protective or risk-associated, only eight were not previously described as significantly modulated in GBM samples. This demonstrates a relatively broad coverage in terms of the miRNAs investigated purely to assess miRNA modulation in GBM pathogenesis without correlating the data to clinical outcome (Table 3). No general patterns apply to these cohorts of protective and risk-associated miRNAs, as such, several miRNAs, which are described as oncogenic from a functional standpoint, are present within the cohort of protective miRNAs and vice versa. This comparison between in vitro functionality and clinical implication of GBM-related miRNAs illustrates that although a specific miRNA may have a specific set of functional characteristics when artificially over or underexpressed in isolated in vitro models, this is not necessarily a good indicator for the multifactorial clinical progression of GBM. For more elaborate details of the functional characterization of miRNAs involved in GBM, please refer to Møller et al. [38].

MicroRNAs Included in Several Signatures

Based on the multiple-miRNA signatures identified from both tissue and database studies pertaining to survival, the miRNAs found in multiple studies were identified (Table 4). Most of the miRNA found in signatures are specific for the given study. Thirteen of the miRNAs were identified in more than one study; however, no miRNAs were identified in more than three studies. Through miRNA array analysis, Niyazi et al. found a 30-miRNA signature in an independent cohort, which divided the samples into short- and long-term survival [94]. Furthermore, Zhang et al. and Srinivasan et al. used similar methods and the same database, but the overlap between these studies was poor [87, 91].

Table 4 miRNA signatures correlating with survival in GBM
Full size table

It is striking that the database studies do not reveal better coherency; however, this could be attributed to the cohort factors studied or the filtering of the miRNAs during the analysis. The tissue studies represent independent cohorts; however, many of the database studies have validated their results in independent cohorts, whereby the cohort factors could be the prime source for the lack of overlap. Additionally, Ilhan-Mutlu et al. chose to investigate seven well-characterized miRNAs (miR-10b, miR-21, miR-181b, miR-181c, miR-195, miR-221, miR-222) and found that none of them correlated with survival [95] contradicting other studies [87, 112]. Therefore, the 13 miRNAs identified in more than one signature could be more applicable in their prediction of survival and of great interest in relation to GBM prognosis.

Functional Analysis of the MicroRNAs Included in Several Signatures

The majority of the 13 miRNAs included in more than one signature have been functionally characterized in GBM and associated with the expression of validated target genes (Table 5). The most well-characterized miRNA in GBM is miR-21, which functions as an oncogenic miRNA. miR-21 has numerous validated target genes that it represses in GBM and therefore it is interesting that this miRNA is included in two signatures. The target genes of miR-21 include genes associated with proliferation (e.g., PTEN and PDCD4), invasiveness (e.g. TIMP3 and RECK), and susceptibility to chemo and radiation therapy (e.g., hMSH2), factors, which are all characteristics of GBM tumors [44, 48, 59, 70]. The same characteristics of GBM tumor growth are also modulated by miR-221 and miR-222, both of which appear in three signatures. Being less well characterized than miR-21, miR-221 and miR-222 still have several validated target genes including some important tumor suppressor genes such as P27, P57, TIMP3, and Cx43 [96, 111, 113, 114].

Table 5 Functional characteristics of miRNAs found in several signatures
Full size table

In addition to the oncogenic miRNAs identified in more than one signature, different tumor suppressor miRNAs were also found in several signatures. miR-195 has validated target genes, including some cyclins and E2F3, which are associated with cell proliferation [115, 116]. Hence, a low expression of this miRNA should in theory correlate with a favorable clinical outcome, which is in fact reflected in the clinical data [12]. Another interesting miRNA shown in more than one signature is miR-136, which has very little functional characterization, but the current validated target genes include the important oncogene, Bcl-2 [117]. Several of the 13 miRNAs (miR-155, miR-17-5p, miR-181b, miR-195, miR-20a, miR-21, miR-221, and miR-222) are known to modulate the mesenchymal mode of migration and invasion (MMMI), which is an important characteristic of GBM cells [38, 118]. Three of the miRNAs identified in more than one signature have no functional characterization and could possibly reveal numerous relevant target genes to substantiate the importance of the 13 miRNAs in future determination of patient prognosis.

Perspectives on MicroRNA-Based Therapies for the Treatment of Glioblastoma Multiforme

Given the fact that several miRNA signatures associated with OS or PFS have been identified and that these miRNAs have functional characteristics with importance in GBM progression, a therapeutic concept taking advantage of such correlations seems inherent. The use of miRNA-based therapies in the treatment of GBM is still in its primary phases with exciting basic research being published frequently [119].

Approaches for utilizing miRNAs in such treatment regimens includes both inhibition of oncogenic miRNAs (e.g., miR-21) or overexpression of tumor suppressor miRNAs (e.g., miR-146b) with different types of carriers to facilitate delivery directly to the tumor tissue [120, 121]. Systemic administration of a liposome-encapsulated tumor suppressor miRNA, miR-7, led to a significant tumor size reduction in a xenograft mouse model of GBM. In addition, several key oncogenes were downregulated upon the tumor suppressor miRNA delivery [122]. Another more sophisticated type of lipid-based delivery was exploited by Griveau et al. where locked nucleic acid miRNA inhibitors against miR-21 conferred increased radiosensitivity in U87MG cells [120]. miRNA carriers have also been generated with polymer-based technology, using poly(amido amine) to encapsulate miR-7 for delivery to U251 cells, which resulted in a higher transfection efficiency than liposomal delivery [123].

Of particular interest in solving the problems with efficient drug delivery to the brain, both in malignancies and neurodegenerative diseases are the use of exosomes as drug carriers [124]. Exosomes are endogenous vesicular structures with a diameter ranging from 40 to 120 nm produced by all cells in the body [125]. They are characterized by expression of specific proteins in the membrane (especially tetraspanins) and their ability to deliver proteins, mRNA and miRNAs [126]. The delivered mRNAs and miRNAs are fully functional and can be translated into protein or inhibit mRNA targets in the recipient cells [127, 128].

The potential of exosomes to deliver functional RNAs to cells was utilized by Alvarez-Erviti et al., who provided interesting evidence as to how exosomes might be used to deliver drugs across the ever troubling blood-brain barrier. Immature dendritic cells were transfected to produce exosomes that expressed a neuron-specific targeting peptide on their surfaces to facilitate specific delivery of the exosome cargo. These exosomes successfully delivered both GAPDH- and BACE1-siRNA across the blood-brain barrier resulting in specific gene silencing in the neuronal tissue [129]. Using a somewhat similar approach, Ohno et al. showed that exosomes targeted to EGFR could deliver the tumor suppressor miRNA, let-7a, to a xenograft breast cancer model after intravenous administration. Furthermore, let-7a suppressed the growth of the tumor underscoring the relevance of using exosomal delivery in malignant diseases [130].

Evidence is now emerging showing that exosomal delivery of interfering RNAs could be relevant in the treatment of GBM. GBM cell lines were shown to be resistant to treatment with anti-miRs against the oncogenic miRNA, miR-9, described in Table 3. However, if these GBM cells were co-cultured with anti-miR-transfected mesenchymal stem cells (MSCs) or cultured in the presence of anti-miR-transfected MSC-exosomes, miR-9 was significantly downregulated. This decrease in miR-9 expression made the GBM cells more susceptible to treatment with the chemotherapeutic drug, temozolomide [131]. Katakowski et al. also produced exosomes in MSCs, which were transfected with a miR-146b expression vector. The resulting miR-146b-containing exosomes were injected into xenograft GBM tumors, leading to a significant reduction in tumor volume compared to vehicle-treated controls [121]. Interestingly, it has previously been shown that miR-146b negatively correlates with survival in GBM [87]. The use of exosomes in the treatment of GBM may have a great potential and should be substantiated with more evidence including choice of relevant miRNA cargo and direct targeting of GBM cells to facilitate intravenous administration.

Concluding Remarks

This review presents the studies investigating the expression of specific miRNAs or miRNA signatures with respect to their correlation to clinical progression of GBM. A large part of the studies utilize data from the same databases (TCGA or CGGA), but they do not necessarily reveal the same results. This is because the extracted data and the filtering based on clinical information differ across individual studies, which makes comparison difficult (Table 1). The studies using individual tissue cohorts also reveal different miRNA signatures with only some consistency between them. Such varying results may be caused by several factors, including miRNAs investigated, type of array platform utilized, cohort size, and especially the choice of control tissue. Comparing miRNA expression data to control tissue obtained from another type of diseased brain (i.e., epilepsy) might be problematic because it may induce variations in the miRNA expression data compared to studies using non-diseased normal brain tissue. Furthermore, imperfect description of terminology with regard to OS and PFS may also add complexity to the comparison of the different miRNA signatures. Several studies report some miRNAs to have a protective or risk-associated profile with respect to their correlation with clinical outcome in GBM. Interestingly, several of these miRNAs have validated functions in vitro and in vivo, which are opposite to the way that they should mediate either protection or risk. Therefore, the in vitro and in vivo studies available for numerous miRNAs are not necessarily good indicators for the multifactorial clinical progression of GBM (Table 3). However, many of the miRNAs reported to be either protective or risk-associated or the miRNAs included in several signatures do in fact have validated targets and functional characteristics, which are in line with their correlation to clinical progression or survival of GBM (Tables 3, 4, and 5). Having been associated with disease progression or survival in several studies, these miRNAs may be valuable for future determination of patient prognosis and could possibly serve as targets for miRNA-based therapies, which hold a great potential in the treatment of this severe malignant disease.