Role of microRNAs Located on Chromosome Arm 10q in Malignant Gliomas

Abstract

Deletions of chromosome arm 10q are found in most glioblastomas and subsets of lower grade gliomas. Mutations in the PTEN gene at 10q23.3 are restricted to less than half of the 10q‐deleted gliomas, suggesting additional glioma‐associated tumor suppressors on 10q. We investigated 64 astrocytic gliomas of different malignancy grades for aberrant expression of 16 microRNAs (miRNAs) on 10q. Thereby, we identified four miRNAs (miR‐107, miR‐146b‐5p, miR‐346, miR‐1287‐5p) whose expression was frequently down‐regulated in anaplastic astrocytomas and/or glioblastomas. DNA methylation analyses revealed 5′‐CpG site hypermethylation of miR‐346 in more than two‐thirds of primary glioblastomas, while aberrant 5′‐CpG site methylation of miR‐146b‐5p was frequent in IDH1‐mutant astrocytomas and secondary glioblastomas. Overexpression of either of the four miRNAs in glioma cell lines reduced cell proliferation and/or increased caspase‐3/7 activity. Expression analyses of miRNA overexpressing glioma cells and 3′‐untranslated region luciferase reporter gene assays revealed evidence that these miRNAs post‐transcriptionally regulate expression of glioma‐relevant genes, including CDK6 (miR‐107), EGFR (miR‐146b‐5p, miR‐1287‐5p), TERT and SEMA6A (miR‐346), all of which are overexpressed in malignant gliomas in situ. In summary, we show that the 10q‐located miRNAs miR‐107, miR‐146b‐5p, miR‐346 and miR‐1287‐5p are frequently down‐regulated in malignant gliomas and thereby may support overexpression of important glioma growth‐promoting genes.

Introduction

Astrocytic gliomas are the most frequent glial tumors of the central nervous system 34. The most malignant astrocytic glioma, glioblastoma, accounts for approximately 50% of all gliomas and approximately 16% of all primary brain tumors 34. Glioblastomas are divided into the more common primary glioblastomas, which preferentially arise in elderly patients without any pre‐existing lesion of lower malignancy, and the less common secondary glioblastomas, which develop by progression from pre‐existing lower grade gliomas and preferentially manifest in younger patients 32. Primary and secondary glioblastomas are characterized by distinct molecular profiles: Secondary glioblastomas frequently carry mutations in IDH1 or IDH2, TP53, and ATRX, while primary glioblastomas typically lack IDH1 or IDH2 mutations but frequently demonstrate activating TERT promoter mutation, PTEN mutation, as well as gene amplifications, most commonly affecting the EGFR gene 23, 26, 32. Despite multimodal aggressive treatment, including neurosurgery followed by radiation therapy with concomitant and adjuvant temozolomide chemotherapy 39, the prognosis of glioblastoma patients remains poor with median overall survival of 16 months in clinical trials 10, 19, 20 and less than 10 months in population‐based studies 37. Therefore, a better knowledge of the molecular alterations driving glioma growth is important to identify novel therapeutic targets and predictive biomarkers. Recent studies have characterized the genetic, epigenetic and transcriptional landscapes in glioblastomas 5, 40. However, the precise functional roles of many aberrations are still unclear, including the most frequent chromosomal alteration in both primary and secondary glioblastomas, that is, loss of one copy of the long arm of chromosome 10 32. Mutations in the PTEN tumor suppressor gene at 10q23.3 are present in 25–40% of primary glioblastomas 5, 32. The lack of PTEN mutations in more than 50% of primary glioblastomas and most secondary glioblastomas points to one or more additional tumor suppressor gene(s) on 10q. Large‐scale sequencing of glioblastomas revealed occasional mutations in various 10q genes; however, none of these were regarded as “driver mutations” 5.

Here, we report on microRNAs (miRNAs) located on 10q that show reduced expression in high‐grade astrocytic gliomas of World Health Organization (WHO) grades III or IV. MiRNAs are small non‐coding, single‐stranded RNA molecules of 20–22 nucleotide length that play important roles in the post‐transcriptional regulation of gene expression 21. By binding to complementary sequences in the 3′ untranslated region (3′‐UTR) of mRNAs, miRNAs cause translational repression and/or destabilization of the respective transcripts 21. Aberrant expression of miRNAs is increasingly recognized in many cancers including gliomas, in which miRNAs either act as proto‐oncogenes or tumor suppressor genes 6, 21.

Our study revealed four miRNAs (miR‐107, miR‐146b‐5p, miR‐346 and miR‐1287‐5p) with significantly reduced expression in anaplastic astrocytomas WHO grade III and/or glioblastomas WHO grade IV when compared with non‐neoplastic brain tissue. These candidate miRNAs were selected for further investigations to characterize the role of epigenetic DNA modifications in their down‐regulation, to study miRNA effects on glioma cell proliferation and apoptosis, and to identify miRNA‐regulated target genes.

Materials and Methods

Patients and tissue samples

Tumor tissues were selected from the archive of the Department of Neuropathology, Heinrich‐Heine‐University, Düsseldorf, Germany, and investigated as approved by the institutional review board (study number 2767). All tumors were classified according to WHO criteria 29. Tissue samples were snap frozen and stored at −80°C. Only samples with a histologically estimated tumor cell content of at least 80% were used for molecular analysis. DNA and RNA were extracted as reported 44. The tumor series used for PCR‐based miRNA expression screening consisted of six diffuse astrocytomas WHO grade II, eight anaplastic astrocytomas WHO grade III, and 10 secondary and 40 primary glioblastomas WHO grade IV. Clinicopathological and selected molecular characteristics of the respective patient cohort are provided in Supporting Information Table S1. Twelve commercially available RNA samples from non‐neoplastic human brain tissues (Ambion, Austin, TX, USA; BioChain, Hayward, CA, USA; Clontech, Mountain View, CA, USA; Stratagene, Cedar Creek, TX, USA) were used as non‐neoplastic reference samples (Supporting Information Table S2). Three different commercially available genomic DNA samples (BioChain) from human brain tissue served as reference for DNA methylation analyses.

Real‐time reverse transcription PCR analyses

We investigated 16 of the 18 miRNAs (miR‐107, ‐146b‐5p, ‐202‐3p, ‐202‐5p, ‐346, ‐548e‐3p, ‐605‐5p, ‐606, ‐607, ‐608, ‐609, ‐936, ‐1254, ‐1287‐5p, ‐1296‐5p, ‐2110) that were mapped to 10q in the NCBI36/hg18 assembly (http://www.genome.ucsc.edu). Reverse transcription of mature miRNAs was carried out with 100 ng of total RNA as template using stem‐loop primers specific for each investigated miRNA and Superscript® II reverse transcriptase (Life Technologies, Darmstadt, Germany). Real‐time PCR was performed with commercial TaqMan® miRNA assays (Life Technologies, Assay IDs: miR‐107 #000443, miR‐146b‐5p #001097, miR‐202‐3p #001012, miR‐202‐5p #001013, miR‐346 #000553, miR‐548e‐3p #002881, miR‐605‐5p #001568, miR‐606 #001569, miR‐607 #001570, miR‐608 #001571, miR‐609 #001573, miR‐936 #002179, miR‐1254 #002818, miR‐1287‐5p #002828, miR‐1296‐5p #002908, miR‐2110 #121216_mat) on a StepOnePlus™ Real‐Time PCR system (Life Technologies). Expression changes relative to a calibrator sample (Universal Human Reference RNA, Stratagene, La Jolla, CA, USA) were calculated with the 2^−ΔΔCt method 28 using let‐7a‐5p (Assay ID: #000377) as reference transcript. For miR‐107, miR‐146b‐5p, miR‐346 and miR‐1287‐5p additional TaqMan® experiments were performed using U6 snRNA (Assay ID: #001973) as reference. Expression levels of miRNA target genes as well as the host genes of miR‐107 (PANK1) and miR‐346 (GRID1) were determined by real‐time reverse transcription‐polymerase chain reaction (RT‐PCR) using SYBRGreen® incorporation and the StepOnePlus™ system with ARF1 (ADP‐ribosylation factor 1) as reference. For primer sequences see Supporting Information Table S3.

DNA methylation analyses

Sodium bisulfite treatment of 1 μg genomic DNA was performed with the EZ DNA Methylation‐Gold™ kit (Zymo Research, Freiburg, Germany). For methylation analysis of the CpG island (named “CpG: 47” in the UCSC genome browser GRCh37/hg19 assembly) located in intron 1 of GRID1, a 270 base‐pair (bp) amplicon between nucleotides 88,023,000 and 88,023,269 (chr10; UCSC genome browser, GRCh37/hg19 assembly) was amplified from sodium bisulfite‐modified DNA using HotStar Taq DNA polymerase (Qiagen, Hilden, Germany). For methylation analysis of the CpG island (named “CpG: 356” in the UCSC genome browser, GRCh37/hg19 assembly) located in the 5'genomic region of GRID1, a 223‐bp fragment between nucleotides 88,126,110 and 88,126,331 (chr10; UCSC genome browser, GRCh37/hg19 assembly) was amplified. In the 5'genomic region of miR‐146b‐5p, we investigated methylation patterns in and around three putative SP1 transcription factor‐binding sites predicted by the NSITE program (Version 2.2004, Softberry Inc., http://linux1.softberry.com/cgi‐bin/programs/promoter/nsite.pl). The three sites are located between nucleotides 104,195,360–104,195,369 (SP1_1), nucleotides 104,195,–104,195,466 (SP1_2) and nucleotides 104,195,479–104,195,488 (SP1_3) (chr10; GRCh37/hg19 assembly), and were investigated for methylated CpG sites in two amplicons. The first 98‐bp amplicon encompassed SP1_1 and the second 207‐bp amplicon covered SP1_2 and SP1_3. For methylation analysis of the CpG island (named “CpG: 93” in the UCSC genome browser, GRCh37/hg19 assembly) located in the 5'genomic region of PANK1 (miR‐107), eight CpG‐sites between nucleotides 91,404,454 and 91,404,492 were investigated. Preparation of PCR products for pyrosequencing was performed as described 15. Pyrosequencing was carried out on a PyroMark Q24 system using the PyroMark Q24 software (Qiagen). For primer sequences, see Supporting Information Table S3.

Cell culture experiments

Glioma lines A172, T98G, CCF‐STTG1, U87MG, U138MG, U251MG, and U118MG were obtained from American Type Culture Collection (ATCC, Manassas, VA, USA). TP365MG glioma cells were kindly provided by Dr. V. Peter Collins (Cambridge, UK). LN‐18 and LNT‐229 3 were kindly provided by Dr. Michael Weller (Zürich, Switzerland) with permission of Dr. Monika Hegi (Lausanne, Switzerland). LN‐229 cells with stable overexpression of wild‐type epidermal growth factor receptor (EGFR; LN‐229WTE 14) were kindly provided by Dr. William Weiss (San Francisco, CA, USA). Cell lines were authenticated by genotyping at the Leibniz Institute DSMZ—German Collection of Microorganisms and Cell Cultures (Braunschweig, Germany). All cell lines were grown as monolayer cultures in Dulbecco's modification of Eagle's medium supplemented with 10% heat‐inactivated fetal bovine serum and penicillin/streptomycin (Life Technologies) in 5% CO₂ in a humidified incubator at 37°C. MiRNAs were overexpressed in these glioma cell lines by transient transfection of either 10 nM, 25 nM or 50 nM of pre‐miR™ miRNA precursor molecules (pre‐miR; Life Technologies) using lipofectamine 2000 transfection reagent (Life Technologies). Control cells were transfected with the pre‐miR™ miRNA Precursor Molecules‐Negative Control #1 (pre‐NC1; Life Technologies). Transfection efficiency was determined using a Cy3‐labeled pre‐miR™ miRNA precursor molecule negative control #1 (Life Technologies). Transfection experiments were done in quintuplicate and repeated twice.

To screen for DNA methylation‐ and histone acetylation‐dependent miRNA expression, cells were incubated in medium with 5‐aza‐2′‐deoxycytidine (5‐AZA‐dCR) and trichostatin A (TSA) under two different conditions 1: 500 nM 5‐AZA‐dCR for 48 h, followed by 500 nM 5‐AZA‐dCR and 1 μM TSA for 24 h 2; 1 μM 5‐AZA‐dCR for 48 h, followed by 1 μM 5‐AZA‐dCR and 1 μM TSA for 24 h. RNA was extracted using Trizol (Life Technologies).

Analyses of cell proliferation and apoptotic activity

A commercially available bromodeoxyuridine (BrdU) incorporation assay (Cell proliferation ELISA, BrdU, chemiluminescence; Roche, Mannheim, Germany) was used to determine the effects of pre‐miRNA on glioma cell proliferation. The BrdU labeling solution was added 72 h after transfection and incubated for 2 h at 37°C. Caspase‐3/7 activity of pre‐miRNA‐ or control‐transfected cells was determined 72 h after transfection using the luminescence‐based Caspase‐Glo® 3/7 Assay System (Promega, Mannheim, Germany) and a Beckman Coulter PARADIGM™ detection platform for measurement of luminescence signals. Each experimental group in the individual functional tests consisted of five replicates. Each assay was performed three times.

3′‐UTR luciferase reporter gene assays

Fragments of the 3′ untranslated regions (3′‐UTR) of CDK6, EGFR, TERT and SEMA6A were PCR amplified with primers encompassing the putative binding sites of the respective miRNAs. PCR products were cloned between the XhoI and NotI restriction sites downstream of the Renilla luciferase coding sequence of the psiCHECK^™‐2 vector (Promega). The psiCHECK™‐2 vector contains the Firefly luciferase gene whose robust expression from a HSV‐TK promoter allows for normalization of Renilla luciferase expression. Sequences with mutated binding sites were also cloned into the psiCHECK™‐2 vector and used for verification of miRNA binding specificity (Supporting Information Figure S1). All constructs were sequenced and 100 ng of the psiCHECK^TM‐2 vector either containing the wild‐type or the mutated binding sites was co‐transfected into T98G glioma cells together with 100 nM of the respective pre‐miRNA or negative control pre‐miR‐NC1. Fourty‐eight hours after transfection, Firefly and Renilla luciferase activities were measured using the Dual‐Glo® Luciferase Assay System (Promega). The experiments consisted of five technical replicates each and each transfection was repeated 2–3 times. For primer sequences, see Supporting Information Table S3.

Western blot analyses

Glioma cells were transfected with 50 nM of pre‐miRNA or pre‐NC1 24 h after seeding in 6‐well culture plates. Protein extraction was performed 72 h post‐transfection. Cells were washed twice with cold PBS and lysed with 400 μL lysis buffer (50 mmol/L Tris‐HCl at pH 8.0, 150 mmol/L NaCl, 0.5%, TritonX‐100, 0.5% deoxycholate). Cell lysates were centrifuged and supernatants were separated on 8% or 12% polyacrylamide gels. Separated proteins were transferred to PROTRAN BA 83 nitrocellulose membranes (Schleicher & Schuell BioScience, Dassel, Germany) using a Trans Blot Cell Blot module (Bio‐Rad, München, Germany). Membranes were blocked with 5% milk powder and 0.01% Tween 20 in tris‐buffered saline and probed with antibodies against cyclin dependent kinase 6 (CDK6; 1:1000, Cell Signaling, Frankfurt, Germany, #3136), EGFR (1:500, Dako, Glostrup, Denmark, #M7298), or alpha‐tubulin (1:1000, Sigma, Taufkirchen, Germany, #T9026). Antibody binding was detected by anti‐mouse antibody linked to horseradish peroxidase, followed by incubation with Immobilon Western HRP Substrate Luminol reagent and peroxide solution (MerckMillipore, Darmstadt, Germany). Chemiluminescence was recorded with the LAS‐3000 mini system (Fujifilm Life Science, Stanford, CT, USA). Protein levels were densitometrically determined using the Image J software (rsbweb.nih.gov/ij/) and normalized to alpha‐tubulin. Each experiment was performed in triplicate.

Association of miRNA expression with common molecular aberrations and patient survival

We investigated the association of miR‐107, miR‐146b‐5p and miR‐346 expression with selected genetic alterations (EGFR amplification, PTEN mutation, IDH1 mutation), transcriptional subtype 45 and methylation subtype 5 in glioblastomas using The Cancer Genome Atlas (TCGA) dataset (http://cancergenome.nih.gov/). Data on miR‐1287‐5p are not available in this data base. Expression data, copy number data and mutation calls from TCGA were downloaded via the cBio Cancer Genomics Portal (www.cbioportal.org) 7, 18 using the open‐source R package “cgsdr” (version 1.1.30) and the statistical computing language R (version 2.15.0). The TCGA glioblastomas were already classified concerning the methylation and transcriptional subtype 5. To investigate the association of miR‐107, miR‐146b‐5p and miR‐346 expression with patient survival, we divided the TCGA glioblastoma cohort into tumors with low and high miRNA expression using the median miRNA level as cutoff. Survival data were obtained from Brennan et al 5. Kaplan–Meier survival probability estimates were calculated using the R‐Package “Survival” (version 2.37‐7).

IDH1 and IDH2 mutations in our own patient cohort were determined by sequencing 42. TERT promoter mutations affecting the two hotspots at nt 1,295,228 (C228T) and nt 1,295,250 (C250T) were detected by sequencing. TERT gene copy number gains were assessed by duplex‐PCR analysis. Copy number of TERT was normalized to the copy number of APRT (NM_000485.2). Copy number increases of more than fivefold relative to constitutional DNA were considered as evidence for high‐level copy number gain indicative of gene amplification. Target/reference gene ratio between twofold and fivefold the ratio obtained for constitutional DNA were considered as low‐level gene copy number gains 36. Investigation of 10 blood‐derived DNA samples showed a mean target/reference gene ratio of 1.0 and a standard deviation of 0.5. For primer sequences, see Supporting Information Table S3.

Statistical analyses

Kruskal–Wallis test in combination with Dunn's post‐hoc test was applied to compare miRNA or mRNA expression levels between different tumor groups and normal tissue samples. Two‐sided Student's t‐test was applied to compare methylation levels between different groups. Paired Student's t‐test was used to compare the mean luminescence values between pre‐miRNA and pre‐NC1 transfected glioma cells. The correlation of miRNA expression with host gene mRNA expression or expression of predicted miRNA targets was analyzed using Pearson regression analysis. MiRNA expression differences between molecularly defined glioma subgroups were investigated with two‐sided Student's t‐test (two subgroups) or ANOVA combined with Tukey post‐hoc tests (>2 subgroups). Bonferroni correction was applied to avoid alpha‐error inflation. Logrank tests were used to determine significance in survival analyses. For all statistical tests, a P‐value of <0.05 was considered as significant.

Results

Expression profiling of miRNAs located on 10q

We investigated 64 astrocytic gliomas and 12 normal brain tissue samples for aberrant expression of 16 miRNAs located on 10q (for localization of the 16 miRNAs relative to the PTEN locus, see Supporting Information Figure S2) using stem loop real‐time PCR assays. Thereby, we identified four miRNAs (miR‐107, ‐146b‐5p, ‐346 and miR‐1287‐5p) that showed significantly lower expression levels in astrocytic gliomas of WHO grade III and/or IV relative to non‐neoplastic brain tissue (Figure 1). The other miRNAs either were not expressed (miR‐202‐3p, ‐202‐5p, ‐548e‐3p, ‐606, ‐607, ‐608, ‐609, ‐936) or did not display significantly reduced expression in gliomas (miR‐605‐5p, ‐1254, ‐1296‐5p, ‐2110). Expression of miR‐107 was significantly reduced in diffuse astrocytomas (mean: 0.28, standard deviation (SD): 0.10; P < 0.05), anaplastic astrocytomas (mean: 0.19, SD: 0.08; P < 0.001) and primary glioblastomas (mean: 0.22, SD: 0.18; P < 0.001) as compared with non‐neoplastic brain tissue (mean: 1.56, SD: 0.89). Significantly lower expression of miR‐146b‐5p was detected in anaplastic astrocytomas (mean: 0.09, SD: 0.07; P < 0.05) relative to non‐neoplastic brain tissue (mean: 0.29, SD: 0.18). The mean expression levels of miR‐146b‐5p in diffuse astrocytomas (mean: 0.14, SD: 0.13) and secondary glioblastomas (mean: 0.10, SD: 0.06) also were reduced but these differences did not reach significance. A significantly lower expression of miR‐346 was detected in primary glioblastomas (mean: 1.46, SD: 1.66; P < 0.001) relative to non‐neoplastic brain tissue (mean: 16.07, SD: 18.14). MiR‐1287‐5p exhibited significantly lower expression in primary (mean: 0.22, SD: 0.30; P < 0.01) and secondary glioblastomas (mean: 0.01, SD: 0.01; P < 0.001) as well as diffuse astrocytomas (mean: 0.12, SD: 0.14; P < 0.05) compared with non‐neoplastic brain tissue (mean: 0.70, SD: 0.29).

Figure 1.

Expression of miR‐107, miR‐146b‐5p, miR‐346 and miR‐1287‐5p in astrocytic gliomas of different WHO grades. Relative miRNA expression levels were determined by real‐time RT‐PCR using let‐7a as reference. Box plots are depicted indicating normalized expression values according to Tukey's method. Asterisks indicate significant expression differences compared with non‐neoplastic brain tissue samples (*P < 0.05; **P < 0.01, ***P < 0.001; Kruskal–Wallis test and Dunn's post‐hoc test). AII, diffuse astrocytoma, WHO grade II; AAIII, anaplastic astrocytoma, WHO grade III; sGBIV, secondary glioblastoma, WHO grade IV; pGBIV, primary glioblastoma, WHO grade IV; NB, non‐neoplastic brain tissue; n, number of investigated tissue samples.

We performed additional analyses for the four candidate miRNAs using U6 snRNA as reference transcript. These experiments confirmed the results obtained by normalization against let‐7a (Supporting Information Figure S3A). Analyses of a distinct, partially overlapping cohort of 80 gliomas and nine normal brain tissue samples using TaqMan^® low‐density arrays and normalization against four reference miRNAs (miR‐30a‐d) additionally confirmed the data for miR‐107 and miR‐146b‐5p (Supporting Information Figure S3B). MiR‐107 and miR‐346 are located in protein‐coding genes (PANK1 and GRID1), expression analyses of which revealed reduced mRNA levels in gliomas in line with the corresponding miRNAs (Supporting Information Figure S4A). Regression analyses demonstrated significant co‐regulation of miR‐346 and GRID1 but not miR‐107 and PANK1 in primary glioblastomas (Supporting Information Figure S4B).

Investigation of constitutive expression levels of the four miRNAs in established glioma cell lines revealed significantly lower mean expression levels relative to mean expression in non‐neoplastic brain samples for miR‐146b‐5p and miR‐346 (Supporting Information Figure S5). Mean expression levels of miR‐107 and miR‐1287‐5p also tended to be lower in the glioma cell lines; however, the differences did not reach statistical significance because of larger standard deviations (Supporting Information Figure S5).

Aberrant DNA methylation in the miR‐346 and miR‐146b‐5p 5′‐genomic regions

To elucidate whether DNA methylation is involved in the down‐regulation of the four 10q miRNAs, we investigated their expression with and without treatment with the demethylating agent 5‐aza‐2′‐deoxycytidine (5‐AZA‐dCR) and the histone deacetylase inhibitor trichostatin A (TSA) in four glioma cell lines (TP365MG, A172, U138MG, T98G). While miR‐107 and miR‐1287‐5p expression did not increase in any of the 5‐AZA‐dCR/TSA‐treated cell lines relative to untreated controls, expression of miR‐146b‐5p significantly increased in TP365MG, A172, and T98G cells following 5‐AZA‐dCR/TSA treatment (Figure 2A). Expression of miR‐346 also significantly increased upon treatment in T98G cells and showed similar trends in U138MG and TP365MG cells (Figure 3A).

Figure 2.

miR‐146b‐5p expression in 5‐aza‐2′‐deoxycytidine (5‐AZA‐dCR) and trichostatin A (TSA) treated glioma cell lines (A) as well as methylation of CpG sites located in putative SP1 transcription factor‐binding sites associated with the miR‐146b‐5p locus (B). (A) Relative expression levels of miR‐146b‐5p were determined by real‐time RT‐PCR using let‐7a as reference gene. Expression of the miRNA in the 5‐AZA‐dCR/TSA samples was calculated relative to miRNA expression in the controls (Co). A + T1: cell lines were grown in medium supplemented with 500 nM 5‐AZA‐dCR for 48 h, washed and grown for another 24 h in medium with 500 nM 5‐AZA‐dcr and 1 µM TSA; A + T2: like A + T1, except for a higher 5‐AZA‐dCR concentration (1 µM). Asterisks indicate significant difference when compared with the respective non‐treated controls (*P < 0.05; **P < 0.01; paired t‐test). Bars represent the mean, and error bars represent the standard deviation of three independent experiments, except for U138MG A + T2 with two independent experiments. (B) Methylation analysis of the seven CpG sites located from nt 104,195,358 to nt 104,195,483 on chr10 (UCSC genome browser, GRCh37/hg19 assembly) was performed by pyrosequencing of two PCR products generated from sodium bisulfite‐modified DNA. The three CpG sites located within each of the three putative SP1‐binding sequences are labeled by hachures. The methylation status at each CpG site was gray scale‐coded as follows: white square, 0–25% methylated alleles; light gray square, 26–50% methylated alleles; gray square, 51–75% methylated alleles; black square, 76–100% methylated alleles. methCo, methylation positive control (CpGenome™ Universal Methylated DNA, Merck Millipore, Darmstadt, Germany); AII, diffuse astrocytoma WHO grade II; AAIII, anaplastic astrocytoma WHO grade III; sGBIV, secondary glioblastoma WHO grade IV; pGBIV, primary glioblastoma WHO grade IV; NB, non‐neoplastic brain tissue; Meth% (mean), mean of the percentages of methylated allele derived from pyrosequencing at the seven investigated CpG sites. IDH mut: gray square, IDH1 or IDH2 mutation detected; white square, no IDH1 or IDH2 mutation detected. TERT mut: gray square, TERT promoter mutation detected; white square, no TERT promoter mutation detected. Note a significant difference in the mean percentage of methylated alleles between primary glioblastoma (mean: 5% methylated alleles) and secondary glioblastoma (mean: 30% methylated alleles; P < 0.01; Student's t‐test, two‐sided).

Figure 3.

miR‐346 expression in 5‐aza‐2′‐deoxycytidine (5‐AZA‐dCR) and trichostatin A (TSA) treated glioma cell lines as well as methylation of the miR‐346‐associated CpG island in astrocytic gliomas. (A) Relative expression levels of miR‐346 were determined by real‐time RT‐PCR using let‐7a as reference. Expression of miR‐346 in the A + T1 and A + T2 samples was calculated relative to the expression in the controls (Co). A + T1: Cell lines were grown in medium supplemented with 500 nM 5‐AZA‐dCR for 48 h, washed and grown for another 24 h in medium with 500 nM 5‐AZA‐dCR and 1 µM TSA; A + T2: like A + T1, except for a higher 5‐AZA‐dCR concentration (1 µM). Asterisks indicate significant difference as compared with the respective non‐treated (‐) controls (***P < 0.001; paired t‐test). Bars represent the mean, and error bars represent the standard deviation of three independent experiments, except for U138MG A + T2 with two independent experiments. (B) Methylation analysis of the 9 CpG sites of the CpG: 47 island (chr10; UCSC genome browser, GRCh37/hg19 assembly) located between nucleotides 88,023,052 and 88,023,105 in intron 1 of GRID1 was performed by pyrosequencing of a PCR product generated from sodium bisulfite‐modified DNA. The methylation status at each CpG site was gray scale‐coded as followed: white square, 0–25% methylated alleles; light gray square, 26–50% methylated alleles; gray square, 51–75% methylated alleles; black square, 76–100% methylated alleles; methCo, methylation positive control (CpGenome™ Universal Methylated DNA, Merck Millipore, Darmstadt, Germany); AII, diffuse astrocytoma WHO grade II; AAIII, anaplastic astrocytoma WHO grade III; sGBIV, secondary glioblastoma WHO grade IV; pGBIV, primary glioblastoma WHO grade IV; NB, non‐neoplastic brain tissues. Meth% (mean), mean of the percentages of methylated allele derived from pyrosequencing at the nine investigated CpG sites. IDH mut: gray square, IDH1 or IDH2 mutation detected; white square, no IDH1 or IDH2 mutation detected. Note a significant difference in the mean percentage of methylated alleles between primary glioblastoma (mean: 48% methylated alleles) and secondary glioblastoma (mean: 27% methylated alleles; P < 0.01; Student's t‐test, two‐sided).

To investigate the methylation status of CpG sites in the genomic regions of miR‐107, miR‐146b‐5p, and miR‐346 in gliomas (Figure 2B and 3B) and in three glioma cell lines (Supporting Information Figure S6), we performed pyrosequencing of sodium bisulfite‐modified DNA. These analyses showed no aberrant methylation of the 5′‐CpG island associated with miR‐107/PANK1 in any of the investigated gliomas. In contrast, methylation analysis of seven CpG sites spanning three putative SP1 factor‐binding sites located in the 5′‐genomic region of miR‐146b‐5p revealed significantly higher percentages of methylated alleles in diffuse (P < 0.05) and anaplastic astrocytomas (P < 0.01) as well as secondary glioblastomas (P < 0.01) when compared with primary glioblastomas (Figure 2B). In all three investigated cell lines, the percentage of methylated CpG sites at the miR‐146b‐5p locus decreased significantly in 5‐AZA‐dCR‐treated cells compared with control‐treated cells (Supporting Information Figure S6A).

Investigation of the two CpG islands located upstream (CpG: 356, Supporting Information Figure S7) and downstream (CpG: 47, Figure 3B) of miR‐346 showed a higher mean percentage of methylated alleles in primary glioblastomas than in secondary glioblastomas (P < 0.01) and non‐neoplastic brain tissue samples (P < 0.05). In addition, we investigated the latter CpG island for CpG methylation in 5‐AZA‐dCR‐treated TP365MG, A172, and T98G glioma cells and identified a significant decrease in the methylated alleles in 5‐AZA‐dCR‐treated T98G cells (Supporting Information Figure 6B) compared with control‐treated cells, in line with the significantly increased expression of miR‐346 in this cell line after 5‐AZA‐dCR/TSA treatment.

Effect of candidate miRNA overexpression on proliferation and apoptosis of glioma cells

To investigate potential effects of the four candidate miRNAs on proliferation and apoptosis of glioma cells, we transfected A172 and T98G cells with different concentrations of miRNA precursor molecules or scrambled control molecules, and searched for in vitro functional effects by measuring caspase‐3/7 activity and BrdU incorporation (Figure 4). In A172 cells, overexpression of miR‐107, miR‐146‐5p, and miR‐346 significantly increased caspase‐3/7 activity. Overexpression of miR‐1287‐5p mimics also increased caspase‐3/7 activity, however, at borderline significance. Caspase‐3/7 activity was not significantly altered by overexpression of any of the four miRNAs in T98G cells. Overexpression of any of the four miRNAs had no effect on BrdU uptake in A172 cells. In contrast in T98G cells, overexpression of miR‐146b‐5p, miR‐346, and miR‐1287‐5p caused significant decreases in BrdU uptake compared with control‐transfected cells. A similar trend was observed following transfection of miR‐107 mimics.

Figure 4.

In vitro effects of miR‐107, miR‐146b‐5p, miR‐346 or miR‐1287‐5p overexpression on apoptotic activity (A) and proliferation (B) of glioma cells. A172 and T98G cells were transfected with either 10 nM, 25 nM or 50 nM pre‐miR™ precursor molecules (pre‐107, pre‐146‐5p, pre‐346, pre‐1287‐5p) or negative control pre‐miR™ miRNA precursors (pre‐NC1; Life Technologies). Apoptotic activity was determined by a luminescence‐based caspase‐3/7 assay. Proliferation was measured using a luminescence‐based BrdU incorporation assay. Dots represent the mean of five replicates. P‐values were calculated using paired t‐tests. Mean fold‐expression changes (FC) are also provided.

In addition, we transfected U87MG cells with precursor molecules of the four candidate miRNAs. Overexpression of miR‐107, miR‐146b‐5p and miR‐346 did neither affect proliferation nor apoptotic activity in U87MG cells, whereas transfection of U87MG cells with miR‐1287‐5p precursors resulted in a significant decrease in BrdU uptake and a significant increase in caspase‐3/7 activity compared with control‐transfected cells (Supporting Information Figure S8).

Identification of target genes of miR‐107, miR‐146b‐5p, miR‐346, and miR‐1287‐5p

We specifically focused on four predicted targets with documented or presumed tumor‐promoting activity in malignant gliomas, that is, CDK6, EGFR, telomerase reverse transcriptase (TERT) and semaphorin 6A (SEMA6A). To validate predicted miRNA:mRNA interactions, we determined mRNA expression of these targets in pre‐miRNA‐transfected versus control‐transfected cells (Figure 5A). Thereby, we found that CDK6 (miR‐107), EGFR (miR‐146b‐5p and miR‐1287‐5p), and SEMA6A (miR‐346) were down‐regulated by overexpression of the respective miRNAs. Overexpression of miR‐346 resulted in reduced TERT mRNA expression in LNT‐229 but not in LN‐18 and CCF‐STTG1 cells. Regulation at the protein level was confirmed for CDK6 and EGFR by Western blotting (Figure 5B,C). In addition, we performed luciferase reporter gene assays to demonstrate that all four targets are directly regulated by binding of the respective miRNA to the 3′‐UTR of their transcripts (Figure 5D).

Figure 5.

Validation of predicted miR‐107, miR‐146b‐5p, miR‐346 and miR‐1287‐5p target genes. (A) Results of mRNA analyses. Twenty‐four hours after seeding, cells were transfected with 25 nm pre‐miR™ precursor molecules (pre‐107, pre‐146b‐5p, pre‐346, pre‐1287‐5p; Life Technologies) or negative control pre‐miR™ miRNA precursors (pre‐NC1, Life Technologies). Total RNA was extracted 48 h after transfection and mRNA levels of the putative targets were determined by RT‐PCR using ARF1 (CDK6, EGFR), B2M (SEMA6A) or ACTG1 (TERT) as reference genes and universal human reference RNA (Stratagene) as calibrator. Data represent the mean of three independent transfection experiments. Error bars show standard deviations. Asterisks indicate significant difference as compared with the respective pre‐NC1 controls (paired t‐test, one‐sided). (B–C) Results of Western blot analyses. Cells were transfected with 50 nM pre‐miR™ precursor molecules (pre‐107, pre‐146b‐5p, pre‐1287‐5p) or negative control pre‐miR™ miRNA precursors (pre‐NC1). The expression levels of CDK6 (B), EGFR (C), α‐tubulin—as a loading control—were determined 72 h post‐transfection. Approximate sizes of the protein bands in kilo‐dalton (kDa) are indicated on the right. Below each exemplary Western blot, the results of protein quantification from three independent experiments are depicted. P‐values were calculated by a two‐sided paired t‐test. (D) Results of 3′‐UTR luciferase gene reporter assays. Relative luciferase activity in T98G cells were determined 48 h post‐transfection with 100 nm pre‐miR™ precursor molecules (pre‐107, pre‐146b‐5p, pre‐346, pre‐1287‐5p; Life Technologies) or negative control pre‐miR™ miRNA precursors (pre‐NC1, Life Technologies) in combination with either wild‐type (‐3′WT) or deleted/mutated (‐3'del, ‐3'mut) psiCHECK2 constructs. Renilla luciferase activity was normalized to F irefly luciferase activity. Vertical bars represent the mean of three independent experiments with each performed in quintuplicate. Asterisks indicate significant differences compared with pre‐NC1 transfected cells (paired t‐test, two‐sided).*P < 0.05; **P < 0.01; ***P < 0.001; ****P < 0.0001.

Increased expression of miRNA targets in malignant gliomas

Increased expression of CDK6, EGFR, and SEMA6A in human glioma samples relative to non‐neoplastic brain tissue was demonstrated using microarray datasets from our previous study 42 (Figure 6A). Increased expression of TERT in gliomas was confirmed by real‐time RT‐PCR analysis (Figure 6A). Comparison of miR‐346 expression with activating TERT promoter mutation in our patient cohort revealed lower miR‐346 expression in TERT promoter‐mutant compared with wild‐type samples (Figure 6B, Supporting Information Figure S9). The two cases (Figure 6A) with the highest TERT expression values (2^−ΔΔCt of 3.2 and 17.2) had no detectable promoter mutation but exhibited evidence for TERT low‐level copy number gain (tumor A151) or TERT gene amplification (tumor GB113), respectively (Figure 6C).

Figure 6.

mRNA expression levels of the 10q‐miRNA targets CDK6, EGFR, SEMA6A, and TERT in astrocytic glioma. (A) The data for CDK6, EGFR, and SEMA6A expression were obtained from a published microarray dataset 42. Box plots are depicted indicating normalized expression values according to Tukey's method; n, number of tissue samples investigated. The mRNA expression levels of TERT were determined by real‐time RT‐PCR using ARF1 as reference. Asterisks indicate significant expression differences compared with non‐neoplastic brain tissue samples (*P < 0.05; **P < 0.01; Kruskal–Wallis test and Dunn's post‐hoc test). AII, diffuse astrocytoma WHO grade II; AAIII, anaplastic astrocytoma WHO grade III; sGBIV, secondary glioblastoma WHO grade IV; pGBIV, primary glioblastoma WHO grade IV; NB, non‐neoplastic brain tissue. (B) MiR‐346 expression in TERT‐mutant and wild‐type gliomas. MiR‐346 expression is significantly different between TERT promoter‐mutant compared with TERT promoter wild‐type glioma samples (Mann–Whitney test, ****P < 0.0001). (C) The two cases (GB113 and A151) with highest TERT expression in the absence of a detectable TERT promoter mutation showed evidence for a low‐level copy number gain of TERT (A151) or TERT amplification (GB113). APRT was used as a reference gene in the duplex PCR analysis. NB, non‐neoplastic brain tissue; NTC, no template control; bp, base pairs.

Candidate miRNA expression and common molecular alterations in glioblastomas

Analysis of miRNA expression in relation to IDH1/2 mutation in our patient cohort revealed decreased miR‐146b‐5p and miR‐1287‐5p levels in IDH1/2‐mutant compared with IDH1/2 wild‐type glioblastomas (Figure 7A). In contrast, miR‐107 and miR‐346 expression was higher in IDH1/2‐mutant compared with IDH1/2 wild‐type tumors (Figure 7A). Interrogation of the TCGA glioblastoma dataset confirmed that miR‐146b‐5p expression was significantly decreased in IDH1‐mutant compared with wild‐type tumors (Figure 7B), whereas miR‐107 and miR‐346 expression levels were not significantly related to IDH1 mutation in the TCGA dataset (Supporting Information Table S4). Further analyses of TCGA glioblastomas showed no significant differences of miR‐107, miR‐146b‐5p, and miR‐346 expression related to PTEN mutation. Likewise, EGFR amplification was not associated with distinct expression of these three miRNAs (Supporting Information Table S4). The TCGA glioblastoma cohort revealed no association between miR‐107 and miR‐346 expression and deletion of the respective miRNA loci. MiR‐146b‐5p expression was even higher in glioblastomas with deletion of the miRNA locus (Supporting Information Table S4), a finding compatible with miR‐146b‐5p being aberrantly methylated and down‐regulated preferentially in IDH1‐mutant gliomas but not in primary glioblastomas (Figure 7A,B). However, the increased expression of miR‐146b‐5p in glioblastomas with deleted miR‐146b‐5p locus was not statistically significant after Bonferroni correction.

Figure 7.

Relationship between miRNA expression and IDH mutation status in our institutional glioma series (A) and the TCGA glioblastoma cohort (B). (A) MiR‐107 (P < 0.0001) and miR‐346 (P < 0.001) expression is increased, whereas the expression of miR‐146b‐5p (P < 0.01) and miR‐1287‐5p (P < 0.05) is decreased in astrocytic gliomas with IDH1/IDH2 mutation (IDH mut) compared with IDH1/IDH2 wild‐type (IDH wt) tumors (Mann–Whitney test). Box plots are depicted indicating normalized expression values according to Tukey's method; n, number of tissue samples investigated. (B) MiR‐146b‐5p expression is decreased in IDH1‐mutant (IDH1 mut) as compared with IDH1 wild‐type (IDH1 wt) glioblastomas of the TCGA cohort. MiRNA expression z‐scores and IDH1 mutation data were obtained from 158 TCGA glioblastoma cases via the cBio Cancer Genomics portal. P‐values were calculated using a two‐sided Student's t‐test and corrected for multiple testing (Bonferroni‐correction for 17 comparisons). IDH1 ‐mutant glioblastomas carry missense mutations affecting arginine 132; IDH1 wild‐type glioblastomas carry the regular arginine at codon 132; n, number of investigated cases.

Expression of miR‐107, miR‐146b‐5p, and miR‐346 did not significantly differ between proneural, neural, mesenchymal and classical glioblastoma subtypes 45 in the TCGA cohort. There was no association of miR‐107, miR‐146b‐5p and miR‐346 expression with any of the six molecular subgroups of glioblastoma defined by distinct methylation patterns 5. Within the TCGA cohort, univariate analyses showed no differences in progression‐free and overall survival of glioblastoma patients according to high or low expression of miR‐107, miR‐146b‐5p or miR‐346 (Supporting Information Table S4).

Discussion

We screened astrocytic gliomas for aberrant expression of 16 miRNAs located on 10q and identified four miRNAs (miR‐107, miR‐146b‐5p, miR‐346 and miR‐1287‐5p) with significantly lower expression in high‐grade gliomas compared with non‐neoplastic brain tissue. Among these, aberrant expression of miR‐107 und miR‐146b‐5p has been implicated in various cancers including gliomas 16, 24. In contrast, miR‐346 and miR‐1287‐5p have not been related to glioma pathogenesis before and data on potential roles of these two miRNAs in other cancers are limited 35, 46, 50.

Our study revealed decreased expression of miR‐107 in diffuse and anaplastic astrocytomas as well as in primary glioblastomas relative to non‐neoplastic brain tissue. We further substantiated a tumor‐suppressive function of miR‐107 in gliomas, which in part is mediated by direct regulation of CDK6. These results are in line with other studies reporting miR‐107 as a tumor suppressor in various cancers 11, 16, 27, 41 including gliomas 8 and glioma stem cells 9. In addition to CDK6, NOTCH2 has been reported as miR‐107‐regulated target in gliomas 8. In line with our finding that miR‐107 overexpression induced apoptosis in A172 cells, He et al 22 reported that down‐regulation of miR‐107 inhibits apoptosis in glioma cells by up‐regulating the anti‐apoptotic protein SALL4. In contrast to pancreatic cancer 27, we found no evidence for epigenetic silencing of miR‐107 or its host gene PANK1 in gliomas. Thus, it seems that transcriptional down‐regulation of miR‐107 by DNA hypermethylation is a cell‐type specific mechanism. In colon carcinoma and glioma cells, miR‐107 may be up‐regulated by wild‐type p53 8, 49. One may thus speculate that TP53 mutation, which is frequent in astrocytic gliomas, could contribute to the observed down‐regulation of miR‐107 in these tumors. However, this hypothesis would need to be substantiated in further studies.

The expression of miR‐146b‐5p was significantly decreased in anaplastic astrocytomas and secondary glioblastomas, but not in primary glioblastomas. Both in our own cohort as well as in the TCGA dataset, miR‐146b‐5p expression was significantly lower in IDH1‐mutant relative to IDH1 wild‐type gliomas, suggesting a link between IDH1 mutation and reduced miR‐146b‐5p expression, possibly because of the IDH1 mutation‐associated CpG island hypermethylator phenotype 30. In line with this hypothesis, we found that miR‐146b‐5p expression increased in glioma cells treated with 5‐AZA‐dCR and TSA. While the mouse genome harbors a CpG island in the 5′‐region of miR‐146 43, no CpG island maps to the 5′‐region upstream of the human miR‐146b‐5p locus. However, three putative SP1 transcription factor‐binding sites map within 1 kb upstream of miR‐146b‐5p. Binding of the SP1 transcription factor to one of these three consensus‐binding motifs was recently shown by electrophoretic mobility shift assays 1. The analysis of seven CpG sites in and around these SP1‐binding sites revealed higher methylation levels in secondary glioblastomas compared with primary glioblastomas, further substantiating that aberrant DNA methylation may reduce miR‐146b‐5p expression in gliomas.

Previous studies have reported on inhibitory effects of miR‐146b‐5p on the expression of matrix metalloprotease genes (MMPs) and EGFR 24, 48. We validated the direct binding of miR‐146b‐5p to the EGFR 3′‐UTR as reported 24. Thus, reduced expression of miR‐146b‐5p may facilitate increased expression of EGFR and other target genes, thereby promoting glioma cell proliferation, as shown by Katakowski et al 24 and our data, and facilitating migration/invasion of glioma cells 48. In breast cancer, miR‐146b‐5p also has been linked to regulation of NF‐κB 4. Investigation of the effect of miR‐146b‐5p overexpression on apoptotic activity of glioma cells revealed increased apoptotic activity in A172 cells following transfection of miR‐146b‐5p mimics, suggesting that miR‐146b‐5p down‐regulation may also have anti‐apoptotic effects in gliomas.

Interestingly, we found that another miRNA on 10q, namely miR‐1287‐5p, directly targets EGFR. MiR‐1287‐5p exhibited decreased expression in both primary and secondary glioblastomas. EGFR overexpression is detectable in 60‐70% of glioblastomas, including all tumors with EGFR gene amplification 47. EGFR amplification and overexpression in glioblastomas are frequently associated with monosomy of chromosome 10 12. Thus, reduced expression of two EGFR‐inhibitory miRNAs on 10q (miR‐146b‐5p and miR‐1287‐5p) may represent a mechanistic link between 10q loss and EGFR overexpression in malignant gliomas.

MiR‐346 exhibited significantly decreased expression in primary glioblastomas. Previous studies have linked miR‐346 to the regulation of leukemia inhibitory factor (LIF) in stem cell differentiation 33. Overexpression of miR‐346 was reported in a follicular thyroid cancer cell line 46. Here, we found that miR‐346 is frequently down‐regulated in primary glioblastomas by aberrant CpG methylation, which correlated inversely with the expression of miR‐346 and its host gene GRID1. Moreover, overexpression of miR‐346 decreased proliferation of T98G and increased apoptotic activity in A172 glioma cells, which suggests a cell line‐dependent glioma‐suppressive role. Using 3′‐UTR luciferase reporter gene assays, we also experimentally validated TERT and SEMA6A as direct targets of miR‐346 that are frequently up‐regulated in gliomas. Recent studies reported on frequent TERT promoter mutation in glioblastomas that increase promoter activity and hence transcription 2, 31. TERT mRNA levels were up to sixfold higher in glioblastomas carrying TERT promoter mutations when compared with wild‐type tumors 2. Interestingly, we found lower miR‐346 expression in TERT‐mutant glioblastomas, suggesting that epigenetic silencing of this particular miRNA may additionally facilitate TERT up‐regulation. The other validated miR‐346 target, SEMA6A, encodes a member of the semaphorin family of transmembrane signaling molecules and showed increased expression in pre‐B acute lymphoblastic leukemia 17 and renal cancer 13. SEMA6A signals by binding to plexinA4, inhibition of which decreased proliferation and tumorigenicity of U87MG glioma cells 25. In addition, SEMA6A also promotes angiogenesis by modulating VEGF signaling 38. Thus, SEMA6A up‐regulation facilitated by miR‐346 silencing may promote glioma growth at different levels.

In summary, we provide evidence that the frequent loss of chromosome arm 10q in high‐grade astrocytic gliomas may promote glioma growth not only by the well‐established deletion and mutation of the PTEN tumor suppressor gene at 10q23.3 but also by reduced expression of several miRNAs located on this chromosome arm, including miR‐107, miR‐146b‐5p, miR‐346 and miR‐1287‐5p. Down‐regulation of these miRNAs can facilitate overexpression of several growth‐promoting miRNA‐regulated genes including CDK6, EGFR, TERT, and SEMA6A, which in turn may contribute to glioma progression.

Conflict of Interest

G.R. has received research grants from Roche and Merck as well as honoraria for advisory boards from Amgen, Merck Serono and Roche. The other authors declare no conflict of interest.

Supporting information

Figure S1. Putative miRNA:mRNA interaction sites. Shown are the parts of the 3'UTR regions of CDK6, EGFR, SEMA6A and TERT that harbor the putative binding sites of the respective miRNAs. The seed‐sequences (nucleotides 2‐8) of the respective miRNAs are underlined. For investigation of the EGFR:miR‐1287‐5p interaction the cloned 3'UTR fragment encompassed all four putative miR‐1287‐5p binding sites. In the mutant vector constructs of CDK6_miR‐107, EGFR_miR‐146b‐5p, SEMA6A_miR‐346 the nucleotides complementary to the seed‐sequence were deleted. In the mutant vector construct of TERT_miR‐346 and EGFR_miR‐1287‐5p four nucleotides complementary to the seed‐sequence were mutated as indicated. nt, nucleotide.

Figure S2. Localization of the 16 investigated miRNAs relative to the PTEN locus on chromosomal arm 10q according to the NCBI36/hg18 assembly (http://www.genome.ucsc.edu).

Figure S3. (A) Results of stem‐loop RT‐PCR‐based expression analyses of miR‐107, miR‐146b‐5p, miR‐346, and miR‐1287‐5p in astrocytic gliomas of different WHO grades with relative miRNA expression levels normalized to the expression level of U6 snRNA. (B) Results of miRNA‐expression profiling of miR‐107 and miR‐146b‐5p in a distinct, partially overlapping cohort of astrocytic gliomas using TaqMan® Array MicroRNA Cards (Applied Biosystems, Foster City, CA, USA) and four miRNAs (miR‐30a‐5p, miR‐30b, miR‐30c, miR‐30d) with stable expression across the data set as reference for data normalization. Box plots are depicted indicating normalized expression values according to Tukey's method. Asterisks indicate significant expression differences compared with non‐neoplastic brain tissue samples (*P < 0.05; **P < 0.01, ***P < 0.001; Kruskal–Wallis test and Dunn's post‐hoc test). AII, diffuse astrocytoma, WHO grade II; AAIII, anaplastic astrocytoma, WHO grade III; sGBIV, secondary glioblastoma, WHO grade IV; pGBIV, primary glioblastoma, WHO grade IV; NB, non‐neoplastic brain tissue; n, number of investigated tissue samples.

Figure S4.  PANK1 and GRID1 mRNA expression in astrocytic gliomas. (A) Relative PANK1 and GRID1 mRNA levels were determined by real‐time RT‐PCR using ARF1 as reference gene and universal human reference RNA as calibrator. Box plots are depicted indicating normalized expression values according to Tukey's method. Asterisks indicate significant expression differences compared with non‐neoplastic brain tissue samples (*P < 0.05; **P < 0.01; ***P < 0.001; Kruskal–Wallis test and Dunn's post‐hoc test). AII, diffuse astrocytoma, WHO grade II; AAIII, anaplastic astrocytoma, WHO grade III; sGBIV, secondary glioblastoma, WHO grade IV; pGBIV, primary glioblastoma, WHO grade IV; NB, non‐neoplastic brain tissue; n, number of investigated tissue samples. B) Correlation of miR‐107 expression with PANK1 mRNA levels (left) and of miR‐346 expression with GRID1 mRNA levels (right) in primary glioblastoma samples.

Figure S5. Constitutive expression of miR‐107, miR‐146b‐5p, miR‐346, and miR‐1287‐5p in established glioma cell lines (A172, CCF‐STTG1, T98G, U87MG, U138MG, U251MG, TP365MG, U118MG) in relation to non‐neoplastic brain tissue samples as assessed by stem‐loop RT‐PCR. In case of miR‐346 and miR‐1287‐5p, all cell lines except for U251MG were investigated. MiRNA‐expression profiling of miR‐107 and miR‐146b‐5p was performed using TaqMan® Array MicroRNA Cards (Applied Biosystems, Foster City, CA, USA) and four miRNAs (miR‐30a‐5p, miR‐30b, miR‐30c, miR‐30d) with stable expression across the data set as reference for data normalization. Relative expression levels of miR‐346 and miR‐1287‐5p were determined by real‐time RT‐PCR using let‐7a as reference. Bars represent the mean, and error bars represent the standard deviation of the expression levels of the investigated samples. Asterisks indicate significant expression differences (**P < 0.01, ***P < 0.001; Student's t‐test, two‐sided).

Figure S6. Methylation analysis of CpG sites located either in the 5′‐ genomic region of miR‐146b‐5p (A) or in the miR‐346‐associated CpG island (B) by pyrosequencing of PCR products derived from bisulfite‐modified DNA of 5‐aza‐2′‐deoxycytidine (5‐AZA‐dCR) treated glioma cell lines. (A) Seven CpG sites from nt 104,195,358 to nt 104,195,483 on chr10 (UCSC genome browser, GRCh37/hg19 assembly) located in putative SP1 transcription factor‐binding sites associated with the miR‐146b‐5p locus were investigated for methylation. (B) Nine CpG sites of the CpG: 47 island (chr10; UCSC genome browser, GRCh37/hg19 assembly) located between nucleotides 88,023,052 and 88,023,105 in intron 1 of GRID1 were analyzed for methylation. Bars represent the mean, and error bars represent the standard deviation of three independent experiments. dark gray bars, non‐treated control cells; light gray bars, cells treated with 500 nM 5‐AZA‐dCR for 6 days. Asterisks indicate significant differences in methylation frequency (*P < 0.05; **P < 0.01; paired t‐test, two‐sided).

Figure S7. Parts of the CpG‐365 island in the 5'genomic region of GRID1 found frequently methylated in primary glioblastomas. Methylation analysis of the eight CpG sites located from nucleotide 88,126,156 to nucleotide 88,126,194 on chromosome 10 was performed by pyrosequencing of a PCR product generated from sodium bisulfite‐modified DNA. The methylation status at each CpG site was gray scale‐coded as follows: white square, 0–25% methylated alleles; light gray square, 26‐50% methylated alleles; gray square, 51–75% methylated alleles; black square, 76–100% methylated alleles. methCo, methylation positive control (CpGenome™ Universal Methylated DNA, MerckMillipore, Darmstadt, Germany); AII, diffuse astrocytoma WHO grade II; AAIII, anaplastic astrocytoma WHO grade III; sGBIV, secondary glioblastoma WHO grade IV; pGBIV, primary glioblastoma WHO grade IV; NB, non‐neoplastic brain tissue. Note a significant difference in the percentage of methylated alleles between primary glioblastoma (mean: 48% methylated alleles) and secondary glioblastoma (mean: 14% methylated alleles) (P < 0.001; Student's t‐test, two‐sided).

Figure S8. In vitro effects of miR‐1287‐5p overexpression on apoptotic activity (A) and proliferation (B) of U87MG glioma cells. U87MG cells were transfected with 10 nM, pre‐miR™ precursor molecules (pre‐1287‐5p) or negative control pre‐miR™ miRNA precursors (pre‐NC1) (Life Technologies). Apoptotic activity was determined by a luminescence‐based caspase‐3/7 assay. Proliferation was measured using a luminescence‐based BrdU incorporation assay. Dots represent the mean of 5 replicates. P‐values were calculated using paired t‐tests. Mean fold‐expression changes (FC) are also provided.

Figure S9. MiR‐346 expression levels in TERT‐mutant and wild‐type primary glioblastomas. MiR‐346 expression is significantly different between TERT promoter‐mutant compared with TERT promoter wild‐type glioma samples (unpaired t‐test, *P < 0.05).

Table S1. Clinicopathological and selected molecular data of the investigated patient cohort.

Table S2. Product information of the 12 non‐neoplastic brain tissue samples used for miRNA and mRNA expression analyses.

Table S3. Overview of primer sequences.

Table S4. Results of TCGA data analyses. This table lists the number of cases per molecularly defined subgroup, the median miRNA expression levels (z‐score)/the median survival times (months) within the subgroups as well as uncorrected P‐values to indicate statistical significance of the difference in miRNA expression/median survival time between subgroups (t‐test or ANOVA/logrank test). The P‐value that indicates statistical significance after Bonferroni‐correction is printed in bold.