A Novel miR-146a-POU3F2/SMARCA5 Pathway Regulates Stemness and Therapeutic Response in Glioblastoma

Glioblastomas (GBM, WHO grade IV gliomas) are the most aggressive and lethal primary malignant tumors of the central nervous system in adults (1). Recently, the WHO has classified GBM and lower grade gliomas based on the neomorphic mutation status of isocitrate dehydrogenase (IDH) 1 or 2, and irrespective of grades, gliomas with wild-type (wt) IDH1/2 have worse clinical outcome compared with those with mutant IDH1/2 (1). More than 90% of GBMs have wt-IDH1/2 and irrespective of IDH status patients with newly diagnosed GBM receive the standard of care with maximal safe resection, followed by adjuvant radiation and temozolomide (TMZ) chemotherapy (1–3). Despite the aggressive multimodality treatments, the median survival remains within the range of 12 to 15 months after diagnosis (4). This grim patient outcome is largely attributed to rapid tumor growth, invasion of vital brain structures, and evolution of intrinsic and acquired treatment resistance clones among other factors (4). However, we do not have a clear understanding of the molecular signaling networks driving disease progression and treatment resistance. Although a number of prognostic biomarkers, including promoter methylation of O⁶-methylguanine-DNA methyltransferase (MGMT), neomorphic mutation of IDH1/2, amplification, and/or gain-of-function mutation of EGFR, loss of function mutation/deletion of TP53 or PTEN, and CpG island methylation phenotype (5, 6), have been identified and characterized, the molecular mechanisms underlying the activation of signaling pathways driving the treatment resistance and tumor recurrence have remained poorly understood posing a great challenge in developing truly effective therapeutic intervention. To this end, a number of studies demonstrate that a posttreatment increase in GSC population markedly contributes to tumor recurrence and therapeutic resistance mechanisms (7–9). GSCs possess potential of self-renewal and multipotent differentiation, and are responsible for tumor initiation and maintenance (7, 10–13). These GSC functions are largely mediated by deregulation of Wnt/β-catenin (14), Notch (15), NF-κB (16), and JAK–STAT (17) signaling pathways among others resulting in an aberrant expression of downstream signature molecules that drive the resistance and recurrence mechanisms in GBM. To this end, we sought to identify molecular biomarkers/signatures of prognostic values and significance that drive treatment resistance and recurrence in GBM, and to target these drivers in preclinical GBM models for the development of novel therapeutic interventions.

miRNAs play regulatory roles through silencing the expression of target genes by posttranscriptional degradation and translational repression. Evidence shows that miRNAs are involved in regulating the progression and therapeutic resistance mechanisms in a variety of malignancies including GBM by modulating cancer stem cell functions (18). Previous studies have shown differential miRNA expression in GBM tumor tissues; moreover, tumors harboring altered miRNA expression had differential prognoses and treatment outcomes (19–22). Therefore, the differentially expressed miRNAs may serve as biomarkers and/or drivers of prognosis and treatment response in GBM. Very few studies have investigated the expression of miRNA in primary and recurrent GBM (19, 23). Hence, we undertook a high-throughput molecular profiling approach to identify miRNAs that correlate with clinical outcomes of patients with IDH1/2 wt GBM. We identified miR-146a as a tumor suppressor miRNA in GBM. Subsequently, we have identified and characterized two direct targets of miR-146a, POU domain, class 3, transcription factor 2 (POU3F2) and SWI/SNF Related, Matrix Associated, Actin Dependent Regulator of Chromatin, Subfamily A, Member 5 (SMARCA5), which are found to play key roles in tumor growth and treatment response.

We used two patient cohorts in this study. Briefly, a total of 268 (n = 268) FFPE tumor specimens without IDH1R132H mutation from patient cohort 1 with newly diagnosed GBM were used in the study. All of the patients underwent tumor biopsy or resection at the University of Utrecht (the Netherlands, cohort 1) from 2005 to 2014. A total of 11 pairs (n = 11) of fresh frozen tumor specimens from patient cohort 2 with newly diagnosed GBM and matched recurrent GBM were used as a validation set in the study. All of the patients underwent tumor resection at University Hospitals of Cleveland (cohort 2) from 2008 to 2015 and provided written informed consent to participate in brain tumor research. The study was approved by The Ohio State University, University Hospital of Cleveland, and University Medical Center Utrecht institutional review boards.

Tumor samples were processed to generate miRNA expression data using the NanoString human v3a array, which contains 798 miRNA probes. DNA (250 ng) input was used per sample and the MGMT-STP27 model was used to calculate MGMT promoter methylation status from Illumina EPIC data (24).

Normal human astrocytes (NHA) were purchased from Lonza. Human GBM cell lines U251, T98G, LN229, U87MG, LN18, were purchased from ATCC. No further authentication was performed for NHAs and ATCC cell lines. U87MG/EGFRvIII cells were provided by Dr. Deliang Guo (The Ohio State University). Primary GBM patient-derived GBM30-luc cells were kindly provided by Dr. Balveen Kaur (The Ohio State University). Primary GBM patient-derived cells (08-387 and 3359) were kindly provided by Dr. Jeremy Rich (UC San Diego) and originally isolated from tumor resections in accordance with approved institutional review board protocols. GBM12 and GBM43 were kindly provided by Mayo Clinic. These cell lines were authenticated by DNA profiling. U251, T98G, LN229, U87MG, LN18, and U87MG/EGFRvIII cells were grown in DMEM medium (Gibco) supplemented with 10% (v/v) FBS (Invitrogen), 100 U/mL penicillin/streptomycin (Sigma), and maintained in a humidified atmosphere with 5% CO₂ at 37°C. GBM30, 08-387, 3359, GBM12, and GBM43 cells were cultured in neurobasal medium with EGF (20 ng/mL), FGF (20 ng/mL), B27 (1×), and GlutaMax (1×), and sodium pyruvate (1×) in a humidified atmosphere with 5% CO₂ at 37°C. All cells were periodically tested for mycoplasma contamination using Universal Mycoplasma Detection Kit (ATCC30-1012K).

For mRNA expression analysis, total RNA was extracted from GBM cells using TRizol reagent as described previously (24). The primers are listed in Supplementary Table S1. For miRNA expression analysis, RNA (100 ng) was reverse transcribed using the TaqMan Advanced MicroRNA Assay Kit (Applied Biosystems) with miRNA-specific primers. Primers of Taqman MicroRNA Assays for hsa-miR-146a (Assay ID: 000468) and RNU6B (Assay ID: 001093) were purchased from Invitrogen. Relative expression level of miR-146a was calculated by normalization with RNU6B expression in GBM cells. All the experiments were performed in triplicate.

GBM cells were plated and cultured on six-well plates. At 50% confluence, 10 μmol/L DAC (Sigma) was added to the medium on days 1 and day 3. Cells were then harvested for total RNA isolation and RT-qPCR analysis. Bisulfite modification, methylation-specific PCR (MSP), and bisulfate sequencing (BS) were carried out as described previously (25). Primer pairs for MSP and BS are in Supplementary Table S1.

Protein extraction and Western blot analysis were performed as described previously (26). The antibodies used in this study are listed in Supplementary Table S2. Polyclonal goat anti-rabbit antibody (Cell Signaling Technology) and Western Blotting Detection System (Millipore) were used for exposure.

For functional studies, a specific miR-146 mimic/inhibitor (Invitrogen) and a respective negative control (Invitrogen) were transfected into the GBM cells. For siRNA transfection, specific POU3F2 1/2 and SMARCA5 1/2 siRNAs (50 nmol/L; Dharmacon) and a negative control (Dharmacon) were used. For POU3F2 overexpression, pcDNA3.1+/C-(K)DYK-POU3F2 plasmid (GenScript) was used. Tet-On shmimic-inducible miR-146a construct was purchased from Dharmacon. All siRNAs were transfected into cells using Lipofectamine 2000 transfection reagent (Invitrogen).

For cell proliferation assay, 24 hours posttransfection, cells (1,000 cells/well) were seeded in 96-well plates and measured by CellTiter-Glo (Promega) after indicated time points as described previously (24). Cell invasion assays were performed as described previously (24). All the experiments were performed in triplicate.

GBM cells (100 cells/well) were used for colony formation assay as described previously (24). All the experiments were performed in triplicate. A total of 300 cells were mixed with neurobasal medium in a humidified atmosphere with 5% CO₂ at 37°C. The number of spheres was counted 2 to 3 weeks after cell seeding. All the experiments were performed in triplicate. To evaluate the frequency of sphere-forming cells (SFCf), cells were plated in 96-well plates in a limiting dilution manner (1, 5, 10, 20 cells/well) using FACS. The number of wells containing spheres was counted after 2 weeks, and the SFCf was calculated using the ELDA software http://bioinf.wehi.edu.au/software/elda/index.html (27).

Twenty-four hours posttransfection, cells (2,000 cells/well) were seeded in 96-well plates, then treated with different dose of TMZ, 72 hours after treatment, cell viability was measured by CellTiter-Glo (Promega).

An Annexin V-fluorescence isothiocyanate (FITC) Apoptosis Detection Kit (Invitrogen) was used for apoptosis assay. At 72 hours miR-146a overexpression, 10⁶ cells were resuspended in 200 μL binding buffer with 10 μL Annexin-V-FITC and 5 μL propidium iodide, and incubated in the dark for another 30 minutes. Finally, cells were assessed using flow cytometric analysis. All the experiments were performed in triplicate.

The 3′UTR of POU3F2 and SMARCA5 was synthesized, annealed, and then inserted into the Xho l and Not I sites of the pCheck2-reporter luciferase vector downstream of the stop codon of the gene for luciferase. To induce mutagenesis, the sequences complementary to the binding sites of miR-146a in the 3′UTR was replaced by the mutated sites (Supplementary Table S1). The constructs were confirmed by sanger sequencing. Cells were cotransfected with the wild type or mutated construct, pCheck2 plasmid, and equal amount of negative control or miR-146a mimic. Luciferase assays were performed using the Dual Luciferase Reporter Assay System (Promega).

Six- to eight-week-old athymic nude mice were obtained from the Target Validation Shared Resource at the Ohio State University. All animal procedures were approved by the Subcommittee on Research Animal Care at The Ohio State University. To determine the frequency of tumor-initiating cells (TICf) using the limiting dilution assay, three cell doses (1 × 10⁵, 1 × 10⁴, 1 × 10³) of each sample were injected subcutaneously into athymic nude mice. Mice were monitored for up to 4 weeks postinjection, and the tumor number per group within this period was used to calculate the TICf using aforementioned ELDA software (27). For the intracranial xenograft models, GBM cells (1 × 10⁵ 08-387/miR-146a/luc and GBM30/miR-146a/luc in 2 μL PBS), transduced with Tet-On inducible miR-146a and GFP-tagged luciferase plasmids, were implanted into the brain of the mouse. Doxycycline (Dox) grain-based diet (Thermo Fisher Scientific) was administered 1 day after injection. For drug treatment studies, mice were treated with vehicle or 20 mg/kg of TMZ resuspended in vehicle by oral gavage once a day for 5 days, starting at day 5 postinjection. Luciferin (Perkin Elmer) solution (100 mg/kg) was used for imaging after tumor formation. The IVIS Lumina II imaging platform from the Ohio State University (OSU) Small Animal Imaging Core was used to detect and quantify the signal. Mice were sacrificed when they became moribund and the tumor tissues were harvested for miR-146a expression detection.

Analysis of Nanostring data was carried out in R (24). Cox regression was used to identify the association between expression of miRNAs (continuous) and overall survival (OS) with age as a covariable. miR-146a was then median dichotomized and the log-rank test was employed to visualize the association between the expression and OS. Other statistical analyses were performed using the software package SPSS 23.0 (SPSS). Descriptive statistics, that is, means ± SD, are shown on the Figures. Two sample t tests or ANOVA were performed for data analysis for experiments with two groups or more than two groups' comparisons. Spearman correlation analyses was applied to analyze the association between expression of miR-146a and POU3F2/SMARCA5. The publicly available CGGA datasets were directly analyzed from the CGGA Data Portal at http://www.cgga.org.cn/. The detailed information of the RNA-seq experiments and software used can be found at the CGGA Data Portal at http://www.cgga.org.cn/. P values were calculated two-sided. P value less than 0.05 was defined as statistically significant.

To investigate a potential association between miRNA expression and GBM patient outcomes, miRNA expression profiling in GBM tumor specimens of a cohort of 268 patients with wt-IDH1/2 was conducted using Nanostring v3 technology (24). miR-146a was identified to be one of the top miRNAs, which upon univariable analysis with continuous expression values [hazard ratio (HR) = 0.658; 95% confidence interval (CI), 0.534–0.810; P < 0.001; Supplementary Table S3] showed decreased miR-146a was associated with worse OS. Moreover, multivariable analysis (MVA) showed that independent of clinical variables, which include age, sex, KPS, and treatment (HR = 0.657; 95% CI, 0.527–0.712; P < 0.001; Supplementary Table S3), decreased miR-146a was significantly associated with worse prognosis. Radiotherapy (RT) plus concomitant and adjuvant TMZ has been the standard of care for patients with GBM for over a decade, as opposed to RT alone, as the addition of TMZ resulted in longer OS and progression-free survival (PFS; ref. 28). Importantly, this benefit was observed in patients harboring a methylated MGMT promoter (29). We also found that miR-146a continued to be a strong independent prognostic factor in a multimarker MVA with MGMT promoter methylation status (HR = 0.586; 95% CI, 0.450–0.764; P < 0.001; Supplementary Table S4). The mediandichotomized miR-146a expression also correlated with OS (Fig. 1A). These findings were further validated in two miRNA-array datasets (TCGA: https://www.cancer.gov/tcga and CGGA: http://www.cgga.org.cn/) by MVA (Supplementary Tables S5 and S6). To further explore if miR-146a expression played a role in tumor recurrence, we compared miR-146a expression levels in matched GBM primary and recurrent tumor pairs by Nanostring technology. We found significantly decreased expression of miR-146a in recurrent tumors (Fig. 1B, cohort 1, n = 30, 15 pairs), which was then validated by RT-qPCR (Fig. 1C, cohort 1, n = 28, 14 pairs). To further validate this observation, 11 pairs of GBM primary and recurrent tumor tissues from another patient cohort (cohort 2) were analyzed by RT-qPCR. Herein, miR-146a expression was undetectable in three pairs (data not shown), but six out of the remaining eight pairs showed significantly lower expression of miR-146a in recurrent tumors compared with the matched primary tumors (Fig. 1D, cohort 2, n = 16, eight pairs). Taken together, these data indicate that miR-146a acts as a favorable prognostic biomarker in GBMs, and miR-146a expression is significantly suppressed in recurrent patients with GBM further suggesting that miR-146a may play a role in the treatment response mechanisms in GBM.

To determine the endogenous expression of miR-146a in GBM cells, we performed RT-qPCR in NHAs, six established GBM cell lines (U251, T98G, LN229, U87MG/EGFRvIII, U87MG, and LN18), and three patient-derived primary GBM cell lines (08-387, 3359, and GBM30). We found that the majority of GBM cell lines expressed significantly lower levels of miR-146a, except LN229, compared with the NHA cell line (Fig. 1E). This result suggests that miR-146a is significantly downregulated in GBM cells, which is consistent with a previous study demonstrating that miR-146a expression is lower in GBM tumors compared with adjacent normal tissues (30). Promoter hyper-methylation leading to transcriptional silencing of miRNAs has been found in a variety of cancer (31). Specifically, Wang and colleagues have reported that demethylation of miR-146a promoter by 5-Aza-2′-deoxycytidine (DAC) correlates with delayed progression of castration-resistant prostate cancer (32). Therefore, we explored if miR-146a was silenced by promoter methylation in GBM and found that DAC treatment significantly increased the miR-146a expression in eight out of nine cell lines examined, with exception of LN229 cell line that expressed a relatively higher endogenous level of miR-146a (Fig. 1F). The methylation status of miR-146a was further determined by methylation-specific PCR (MSP) in all nine GBM cell lines. As shown in Fig. 1G, seven of eight GBM cells showed partial methylation of the miR-146a promoter region with downregulated miR-146a expression. Consistent with DAC data, miR-146a promoter in LN229 cells was found to be unmethylated. To confirm the MSP results and further evaluate the methylation status of miR-146a in GBM cell lines, BS was performed for eight CpG sites (−183, −160, −150, −142, −138, −136, −132, and −128) of the promoter region near the transcription start site. Consistent with the MSP results, a high level of methylation was found in seven of eight cell lines with downregulated miR-146a expression (Fig. 1H). The MSP result prompted us to analyze the methylation status of miR-146a in GBM tumor tissues. We selected tumor tissues with relatively low expression levels of miR-146a in cohort 1 and, as expected, methylation of miR-146a was found in 9 of 10 tumor tissues (Fig. 1I). These results suggest that hyper-methylation of CpG islands on the miR-146a promoter contributes, in part, to the downregulation of miR-146a expression in GBM.

Given that high expression of miR-146a was associated with better OS in patients with GBM and it was downregulated in GBM cells, we hypothesized that miR-146a might play an important role in restraining GBM progression. To investigate the effects of miR-146a on cell proliferation and invasion in vitro, U87MG/EGFRvIII, GBM30, and 08-387 cells, which expressed low endogenous levels of miR-146a, were transfected with microRNA negative control (NC) or miR-146a mimic. On the other hand, LN229 cells, which expressed a relatively high level of miR-146a, were transfected with NC or a miR-146a inhibitor. We found that overexpression of miR-146a significantly inhibited proliferation of U87MG/EGFRvIII, GBM30, and 08-387 cells (Fig. 2A–C), which was reproduced in additional patient-derived xenografts (PDX) lines GBM12 and GBM43 with endogenous low expression of miR-146a (Supplementary Figs. S1A and S1B). In consistence with this, inhibition of miR-146a increased the proliferation of LN229 cells (Fig. 2D). In addition to rapid proliferation, invasion is a defining hallmark of GBM cells (18). Accordingly, we determined the role of miR-146a on GBM cell invasion using a trans-well Matrigel assay, which showed that overexpression of miR-146a suppressed invasion of GBM cells in vitro (Fig. 2E; Supplementary Fig. S2A), whereas inhibition of miR-146a increased invasion (Fig. 2F; Supplementary Fig. S2B). Cell proliferation and invasion data were then substantiated by using additional GBM cell lines (U87 and LN18; Supplementary Figs. S3A–S3C). Notably, we observed that overexpression of miR-146a in primary GBM cells significantly inhibited sphere formation, a characteristic feature of these cells (Fig. 2G and H). In consistence with previous studies, activation of NF-ĸB and ERK1/2 was found to be inhibited by overexpression of miR-146a in GBM cells (Supplementary Fig. S4). To study the role of miR-146a on tumor growth in vivo, we generated two tetracycline (doxycycline/Dox) inducible stable cell lines termed 08-387/miR-146a and GBM30/miR-146a. Our data show that expression of miR-146a was induced after Dox treatment and proliferation was inhibited in both cell lines (Supplementary Figs. S5A–S5D). Furthermore, an athymic nude xenograft mouse model was established, in which stable 08-387/miR-146a and GBM30/miR-146a cells were implanted into the brain. Decreased tumor growth was observed by IVIS imaging in Dox-induced group of mice that had higher expression of miR-146a (Fig. 2I–2K) and an overall survival of these mice was prolonged as revealed by the Kaplan–Meier plots (Fig. 2L and M). Representative H&E staining for the tumors are shown in Supplementary Fig. S5E. Dox-induced expression of miR-146a in vivo was further confirmed by qPCR (Fig. 2N and O). Taken together, these data suggest that overexpression of miR-146a inhibits tumor growth and prolongs survival of the tumor cell implanted mice.

TMZ is the most widely used chemotherapy in patients with GBM. To determine if miR-146a sensitizes GBM cells to TMZ, miR-146a was overexpressed in U87MG/EGFRvIII and 08-387 cell lines and cell viability was measured. We found that overexpression of miR-146a could significantly enhance TMZ-induced cell killing (Fig. 3A and B). In addition, fewer colonies were formed in the miR-146a overexpressing group compared with the control group, after TMZ treatment, which was consistent with the outcome of the viability experiments (Fig. 3C and D). To further explore the effects of miR-146a on TMZ-induced apoptosis, we performed Annexin V apoptosis assay and measured caspase activation. As shown in Fig. 3E–G, overexpression of miR-146a markedly enhanced TMZ-induced apoptosis in 08-387 cells. Similar results were obtained in GBM30 cells (Supplementary Figs. S6A and S6B). These findings demonstrate that miR-146a overexpression enhanced TMZ response and potentiated TMZ-induced apoptosis in GBM cells. To further assess the function of miR-146a in TMZ response in vivo, an athymic nude xenograft mouse model was employed. Mice were randomized into four treatment groups, 10 mice per group: (i) Dox(−)/TMZ(−), (ii) Dox(−)/TMZ(+), (iii) Dox(+)/TMZ(−), and (iv) Dox(+)/TMZ(+). Mice were treated, as indicated, with TMZ by oral gavage for 5 days, and euthanized when displayed tumor-associated morbidity. Mice in Dox(−)/TMZ(−) group showed rapid tumor growth (23–25 days, median survival, 25 days), Dox(−)/TMZ(+) and Dox(+)/TMZ(−) groups initially slowed tumor growth, but tumor still grew (33–52 and 33–53 days, respectively, median survival, 38 days). Mice in Dox(+)/TMZ(+) group had significant smaller tumors (Fig. 3H) and had longer survival times (50–63 days, median survival, 55 days, Fig. 3I). Thus, miR-146a not only inhibited tumor growth, but also enhanced TMZ sensitivity both in vitro and in vivo in GBM.

GSCs are found to be primary drivers of tumor recurrence and therapeutic resistance (33). Because miR-146a was found to be downregulated in recurrent GBM tumors (Fig. 1B–D), accordingly, we hypothesized that overexpression of miR-146a would downregulate stemness in GSCs. To test this hypothesis, we first determined the expression of miR-146a in GSCs and non-GSCs (NGSC), which were derived from human brain tumor specimens using CD133 selection (34). Consistent with prior reports, the GSC-enriched fractions expressed high level of Olig2 and Oct4 (Fig. 4A), which are markers of multipotent progenitors. qPCR data revealed that GSCs expressed miR-146a at markedly lower levels than the matched NGSCs, in both 08-387 and 3359 cells (Fig. 4B). Interestingly, after a search of the GEO Profiles database (35), lower expression of miR-146a was also found in GBM stem-like cells (Gene Expression database: GSE23806, P = 0.0304; Supplementary Fig. S7; ref. 36), which confirmed our findings. Further, overexpression of miR-146a resulted in reduced proliferation of GSCs (Fig. 4C and D). Similarly, miR-146a overexpression reduced sphere formation frequency and sphere size by in vitro limiting dilution and sphere formation assays (Fig. 4E–H), as well as in vivo tumorigenicity by frequency of tumor-initiating cells (TICf; Fig. 4I–J), indicating that overexpression of miR-146a inhibited expansion of GSCs. Moreover, additional stemness markers were assessed at mRNA levels, which revealed that overexpression of miR-146a significantly downregulated expression of SALL, SOX2, c-Myc, Nestin, and Oct4 in 08-387 GSCs (Supplementary Fig. S8). These data suggest that overexpression of miR-146a reduced the stemness of GBM cells.

To identify putative target mRNAs of miR-146a, bioinformatics analyses were performed employing different miRNA target prediction tools. The neural transcription factor BRN2 (encoded by the POU3F2 gene), and SMARCA5 among others, were found as novel potential targets of miR-146a (Supplementary Fig. S9). Gene expression analysis using the TCGA datasets (http://gepia.cancer-pku.cn/) revealed that both POU3F2 and SMARCA5 were highly expressed across different cancer types compared with normal tissues (Supplementary Figs. S10A and S11A), particularly in GBM (Fig. 5A and B). Notably, high expression of both of POU3F2 and SMARCA5 correlated with worse OS of glioma patients (Supplementary Figs. S10B and S11B), and these were negatively correlated with miR-146a expression (Supplementary Figs. S10C and S11C), as expected. To determine whether POU3F2 and SMARCA5 are putative targets, different GBM cells were transfected with miR-146a mimic, which resulted in significant reduction of both POU3F2 and SMARCA5 expression in U87MG, U87MG/EGFRvIII, T98G, GBM30, and 08-387 cell lines at the mRNA level (Fig. 5C and D). To confirm changes in expression of POU3F2 and SMARCA5 in the Dox-inducible stable 08-387/miR-146a cell model (Fig. 5E), we treated 08-387/miR-146a cells with different dose of Dox, reduction of POU3F2 and SMARCA5 at both mRNA and protein levels was observed (Fig. 5F and G). To exclude the off-target effect, we detected expression of TRAF6, which is a known miR-146a target (37), and as expected, reduction of TRAF6 was observed when miR-146a was overexpressed by either miR-146a mimics (Supplementary Fig. S12A) or Dox treatment (Supplementary Fig. S12B). To further investigate whether POU3F2 and SMARCA5 are direct and specific targets of miR-146a, luciferase reporter containing wild-type/mutated 3′UTR of POU3F2 and SMARCA5 were constructed (Fig. 5H). Wild-type or mutated 3′UTR reporters of POU3F2 and SMARCA5 were cotransfected with NC/miR-146a mimics into U87MG/EGFRvIII, GBM30, and 08-387 cells, respectively. Consistent reduction in luciferase activity by miR-146a was observed only with wild-type 3′UTR construct, but not with the mutant construct (Fig. 5I and J). Collectively, these data suggest that miR-146a suppressed POU3F2 and SMARCA5 by directly targeting their 3′-UTRs in GBM cells.

Once we found that tumor suppressive and TMZ-sensitizing functions of miR-146a were mediated by knocking down the expression of its direct targets POU3F2 and SMARCA5, we sought to determine the individual relative contributions of these stemness-related transcription factors to miR-146a functions. Among the primary GBM cell lines, we determined that 08-387 expressed relatively high levels of both POU3F2 and SMARCA5, which was consistent with a relatively low expression of miR-146a. We observed that 08-387 cells' proliferation, invasion, and TMZ-response were significantly reduced by siRNA-mediated knockdown of either POU3F2 or SMARCA5 expression (Fig. 6A–F). We noted that knockdown of either target was sufficient to produce effects on cell proliferation, invasion, and TMZ-response comparable with the effects produced by overexpression of miR-146a. To substantiate this notion, miR-146a was inhibited in LN229 cells that expressed higher level of miR146a (Fig. 1E), which were then cotransfected with POU3F2 siRNA and/or SMARCA5 siRNA (Fig. 6G). Subsequent functional assays demonstrated that miR-146a inhibition significantly induced cell proliferation and invasion, which were significantly reversed by knocking down either POU3F2 or SMARCA5 expression (Fig. 6H and I). In addition, miR-146a did not demonstrate a tumor suppressive phenotype when POU3F2/SMARCA5 was overexpressed (Supplementary Figs. S13A–S13C). Thus, either knocking down POU3F2 or SMARCA5 can mimic the effect of miR-146a in GBM, which provoked us to investigate the correlation between POU3F2 and SMARCA5 in patients with GBM. First, we analyzed mRNA expression of POU3F2 and SMARCA5 from both TCGA (https://www.cbioportal.org/) and CGGA (http://www.cgga.org.cn/) datasets, and found that POU3F2 and SMARCA5 were positively correlated (Supplementary Figs. S14A and S14B and S15A and S15B). To confirm the finding in vitro, we knocked down either POU3F2 or SMARCA5 in 08-387 cells (Supplementary Figs. S15C and S15D), and found that expression of SMARCA5 and POU3F2 was decreased at both protein levels (Supplementary Figs. S15C and S15D). Collectively, these data confirmed that the tumor suppressive potential of miR-146a is largely mediated by the miR-146a/POU3F2/SMARCA5 axis (Fig. 6I) in GBM.

Rapid growth, invasion, and resistance to RT and TMZ are largely attributed to the GSC population in GBM tumors. GSCs are capable of executing these pathologic functions through the activation of a number of signal transduction pathways, including the PI3Kinase, Wnt/β-catenin, Notch, NF-κB, and Jak-Stat among others. Activation of these pathways results in aberrant expression of a variety of neural stem cell- and GSC-related genes, which contribute to the maintenance of proliferation, invasion, RT/TMZ-resistance functions in GSCs. Many of these regulatory genes happen to be oncogenes and tumor suppressor genes, which are posttranscriptionally suppressed by tumor suppressive miRNAs and onco-miRNAs, respectively (38). By NanoString analysis of miRNA expression in human GBM specimens, we have identified and validated that miR-146a expression was correlated with favorable patient overall survival. Its expression was partly suppressed by its promoter methylation in primary GBM, and further it was noted that expression of this miRNA was significantly lower in recurrent GBM indicating that miR-146a may have further significance in tumor progression or recurrence. These findings led to the hypothesis that miR-146 could be a tumor suppressor miRNA in GBM and that it might negatively control therapeutic resistance mechanisms likely by suppressing the GSC population and/or their functions. It is worth noting that miR-146a functions as a tumor suppressor miRNA in a number of cancers including NSCLC (39) and prostate cancer (40), and as an onco-miRNA in others that include cervical, head and neck, and hepatocellular cancers (41–43). Kim and colleagues reported that miR-146a was one of the miRNAs that involved in long survival GBM subclass, but did not provided definitive evidence (44). To test our first hypothesis, we took two reciprocal approaches: first, overexpression in primary PDX lines that expressed relatively lower levels of endogenous miR-146; and second, inhibition of miR-146a in cells, namely LN229 that expressed significantly higher level of miR-146. In these reciprocally engineered cell models, we measured cell proliferation, invasion, and stemness-markers in vitro, and tumor growth and overall mouse survival in orthotopically implanted xenograft tumors. Our results clearly suggest that miR-146a functions as a tumor suppressive miR in GBM. This observation is in good agreement with another recent report that shows that ectopic expression of miR-146a inhibits glioma development (45).

Next, we tested if miR-146a sensitizes GBM to TMZ and RT. To this end, our in vitro data clearly indicate that miR-146 overexpression sensitized PDX lines to TMZ in vitro, and in vivo which, in turn, significantly prolonged the mouse survival. The PDX cell line models employed here, were sensitive to low dose (2Gy) of ionizing radiation limiting the evaluation of RT-sensitizing potential of miR-146a.

Once we found that tumor suppressive and TMZ-sensitizing functions of miR-146a were mediated by knocking down the expression of its direct targets POU3F2 and SMARCA5, we sought to determine the relative contributions of these stemness-related transcription factors to miR-146a functions in GBM. POU3F2 (BRN2), a master regulator of neuronal differentiation, is a POU-domain transcription factor well described in developmental biology (46). It has been reported that inhibiting BRN2 expression led to significantly reduced proliferation, migration, and invasion in SCLC, prostate cancer, as well as melanoma (47–50). Importantly, targeting BRN2 is a strategy to treat or prevent neuroendocrine differentiation in prostate cancer (50). SMARCA5 (hSNF2H) is a member of SWI/SNF family, and contains helicase and ATPase activities. hSNF2H promotes tumor growth in ovarian cancer and glioma (51, 52). It has also been shown that miR-100 affects stem cell self-renewal and cell proliferation in part by targeting SMARCA5 (53). Interestingly, Zhou and colleagues reported a reverse correlation between expression of miR-146a and SMARCA5 in a bladder tumor cell model (54), however, they did not provide evidence about the regulation between miR-146a and SMARCA5. Among the primary GBM cell lines we examined, 08-387 expressed relatively high level of both POU3F2 and SMARCA5, which was consistent with a relatively low expression of miR-146a. We observed that 08-387 cells' proliferation, invasion, and TMZ-response were significantly reduced by siRNA-mediated knockdown of either POU3F2 or SMARCA5 expression. We noted, with surprise, that knockdown of either target was sufficient to produce effects on cell proliferation, invasion, and TMZ-response comparable to the effects produced by overexpression of miR-146a. To gain an insight of these unexpected observations, we conducted a series of experiments which revealed that POU3F2 and SMARCA5 positively regulate each other in the absence of miR-146a involvement/modulation. The study has potential limitations. First, miR-146a expression was tested in a small sample size for primary and recurrent GBM, a large patient cohort is needed. Second, the PDX cell line models employed here, were sensitive to low dose (2 Gy) of ionizing radiation limiting the evaluation of RT-sensitizing potential of miR-146a. Third, given the fact that POU3F2 is a transcriptional regulator, it might regulate expression of SMARCA5 through direct binding or indirectly through other targeted genes. Further studies are needed to investigate the mechanism of the regulation between POU3F2 and SMARCA5 to identify the novel therapeutic strategies for GBM.

In summary, focusing on the tumor suppressive functions of miR-146a in GBM, this study revealed the following novel findings: (i) Decreased miR-146a expression is associated with shorter OS independent of MGMT methylation status in GBM; (ii) Promoter methylation-induced silencing of miR-146a drives tumor progression and therapeutic resistance in glioblastoma; (iii) miR-146a inhibits the stemness of GBM cells via directly targeting POU3F2 and SMARCA5 whose expression were positively correlated in GBM. Data from our human cohorts and in vitro/vivo mechanistic analysis strongly implicate that miR-146a plays a significant role in the progression and therapeutic sensitivity in GBM, making it or POU3F2 and SMARCA5 to be potentially attractive and novel therapeutic targets.

J.S. Barnholtz-Sloan reports grants from NIH/NCI during the conduct of the study. No disclosures were reported by the other authors.

T. Cui: Formal analysis, validation, investigation, visualization, methodology, writing-original draft, writing-review and editing. E.H. Bell: Investigation, writing-review and editing. J. McElroy: Investigation. K. Liu: Investigation. E. Sebastian: Investigation. B. Johnson: Investigation. P.M. Gulati: Investigation. A.P. Becker: Investigation. A. Gray: Investigation. M. Geurts: Investigation. D. Subedi: Investigation. L. Yang: Investigation. J.L. Fleming: Investigation. W. Meng: Investigation. J.S. Barnholtz-Sloan: Investigation. M. Venere: Investigation. Q.-E. Wang: Investigation. P.A. Robe: Investigation. S.J. Haque: Investigation, writing-review and editing. A. Chakravarti: Supervision, project administration.

We thank Dr. Jeremy Rich (UC San Diego, USA) for providing primary GBM patient-derived cells. We also thank The Ohio State University (OSU) Comprehensive Cancer Center Small Animal Imaging Core, The Ohio State University (OSU) Genomics Shared Resource (GSR), The Ohio State University (OSU) Target Validation Shared Resource (TVSR), and The Ohio State University (OSU) Comprehensive Cancer Center Pathology Core Facility supported in part by grant no. P30 CA016058, NCI. This work was also supported by NCI [R01CA169368 (to A. Chakravarti), R01CA11522358 (to A. Chakravarti), R01CA1145128 (to A. Chakravarti), R01CA108633 (to A. Chakravarti), R01CA188228 (to A. Chakravarti, K. Liu, and J.S. Barnholtz-Sloan), 1RC2CA148190 (to A. Chakravarti), and U10CA180850-01 (to A. Chakravarti); A Brain Tumor Funders Collaborative Grant (to A. Chakravarti); Ohio State University Comprehensive Cancer Center Award (to A. Chakravarti), and the T&P Bohnenn Fund for Neuro-Oncology Research (grant to P.A. Robe).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.