Feedback circuitry between miR-218 repression and RTK activation in Glioblastoma

Abstract

Receptor tyrosine kinase (RTK) pathway signaling plays a central role in the growth and progression of Glioblastoma (GBM), a highly aggressive group of brain tumors. We recently reported that miR-218 repression, an essentially uniform feature of human GBM, directly promotes RTK hyperactivation by increasing the expression of key positive signaling effectors, including EGFR, PLCγ1, PIK3CA and ARAF (1). However, enhanced RTK signaling is known to activate compensatory inhibitory feedback mechanisms in both normal and cancer cells. We demonstrate here that miR-218 repression in GBM cells also increases the abundance of additional upstream and downstream signaling mediators, including PDGFRα, RSK2, and S6K1, which collectively function to alleviate inhibitory RTK feedback regulation. In turn, RTK signaling suppresses miR-218 expression via STAT3, which binds directly to the miR-218 locus, along with BCLAF1, to repress its expression. These data identify novel interacting feedback loops by which miR-218 repression promotes increased RTK signaling in high-grade gliomas.

Introduction

Glioblastoma (GBM) are the most common malignant brain tumors, with a median patient survival of 12–14 months following aggressive surgery, radiation and chemotherapy (2–5). Extensive research has begun to elucidate the molecular mechanisms underlying GBM initiation and progression. Large-scale genomic analyses initiated by The Cancer Genome Atlas (TCGA) and other groups (6, 7) classified GBM into multiple subtypes, of which Proneural and Mesenchymal GBM comprise the two major classes. Human gliomas typically exhibit recurring abnormalities in genes encoding the tumor suppressors RB and p53, the receptor tyrosine kinases (RTKs) epidermal growth factor receptor (EGFR) and platelet-derived growth factor receptor (PDGFR), and downstream RTK pathway components, including the kinase AKT, mitogen-activated protein kinase (MAPK), and the guanosine triphosphatase RAS (8). Furthermore, activating mutations of phosphatidyl-inositol-3 kinase (PI3K) are observed in nearly all GBM patients (9). We recently demonstrated that RTK signaling in GBM is regulated by the microRNA miR-218 (1), whose abundance is decreased relative to normal brain tissue, in all GBM patient samples tested (10). Specifically, miR-218 binds to and inhibits the expression of mRNAs encoding critical signaling effectors, including EGFR and PLCγ1, thus miR-218 repression promotes RTK signaling in GBM.

To ensure homeostatic regulation in healthy cells, RTK signaling is tightly controlled by negative feedback mechanisms. For example, mTORC1 hyperactivation impairs PI3K stimulation by decreasing amounts of IRS1/2 and PDGFR (9, 11), whereas mTORC1 inhibition increases PDGFR accumulation and MAPK or AKT activation in GBM (9, 12). Moreover, a characteristic feature of RSK2-deficient mice is increased and sustained ERK activation in skeletal muscle, suggesting that RSK2 may also contribute to negative feedback control of upstream ERK signaling (13). Multiple studies have shown that RTK pathway signaling is subject to substantial feedback regulation in GBM and other cancers (12, 14). Because miR-218 repression increases the abundance of various RTK pathway components (1), we investigated whether miR-218-dependent modulation of RTK signaling also engages inhibitory feedback control. Our data reveal that miR-218 repression reduced inhibitory feedback regulation by increasing key effectors, including RSK2, S6K1 and PDGFRα, thereby sustaining enhanced RTK pathway activity in GBM cells. We further define a novel mechanism whereby RTK signaling, in turn, promotes miR-218 repression in a STAT3-dependent manner, revealing the existence of complex, interacting feedback loops regulating RTK activation in GBM.

Results

Identification of ribosomal S6 kinases as novel miR-218 targets

We previously demonstrated that miR-218 impairs the expression of multiple RTK signaling pathway components, including EGFR and PLCγ1, and that miR-218 repression promotes a cumulative RTK activation cascade in GBM (1). Because EGFR and PLCγ1 signaling converge on ERK and mTORC1 (Fig. 1A), we investigated the role of miR-218 in regulating these pathways. Transfection of T98G and U373 glioblastoma cells with a miR-218 mimic reduced the phosphorylation of RSK and S6K1, downstream effectors of ERK and mTORC1, respectively (Fig. 1B). Total RSK and S6K1 protein abundance was also reduced by miR-218 expression (Fig. 1B), suggesting that miR-218 inhibits their corresponding mRNAs directly. Indeed, miR-218 mimic repressed the activity of luciferase reporter gene constructs containing either RSK2 or S6K1 3’UTRs, an effect lost upon mutation of the corresponding miR-218 seed sequences (Fig. 1C and fig. S1A). Furthermore, transient or stable miR-218 expression reduced RSK2 and S6K1 abundance in GBM cells and glioma stem like cells (GSCs), respectively (Fig. 1, D and E; fig. S1B). We performed intracranial orthotopic xenograft assays by transplanting T3691 GSCs expressing either scrambled (SCR) or miR-218 RNAs, and observed reduced RSK2 and S6K1 transcripts in Tum-3691-218 brain tumor sections (Tum-3691-SCR compared with Tum-3691-218 tumors) (Fig. 1F). Of note, RSK2 and S6K1 mRNAs are increased in human GBM patient samples (n=30) compared to 12 normal brain tissues (Fig. 1G), and immunohistochemical assays demonstrated increased RSK2 and S6K1 protein accumulation in human GBM tissues (35/35 cases and 25/25 cases, respectively) (Fig. 1H). We then compared RSK2 and S6K1 expression between Proneural and Mesenchymal GBM subtypes. S6K1 mRNA abundance was higher in both Mesenchymal and Proneural GBMs relative to normal brain tissue samples (see fig. S1C and D). However, we did not observe dramatic differences in RSK2 expression between normal and both GBM subtypes. There could be numerous reasons for this observation. For example, miR-218 mediated regulation of RSK2 is likely to be only one of multiple factors that control the expression of this gene. It should be noted that, although RSK2 transcript accumulation is similar between normal and GBM subtypes, we did see a remarkable and consistent increase of RSK2 protein in GBM patient samles, compared to normal brain tissue (Figure 1H). Interestingly, RSK2 mRNA levels are higher in Mesenchymal when compared to Proneural tumors (fig. S1D). This is noteworthy, as we have previously shown that miR-218 repression is also more pronounced in Mesenchymal than Proneural GBM (1), indicating an inverse correlation between the two, consistent with miR-218 regulation of RSK2. Finally, siRNA-mediated RSK2 or S6K1 knockdown promoted staurosporine-mediated cell death (fig. S1, E to H), as did miR-218 mimic in GBM cells (fig. S1I). Because staurosporine, a protein kinase inhibitor, induces apoptosis, these results indicate that miR-218, RSK2, and S6K1 are in a common pathway. In addition to our previous studies (1), the data reveal a pattern in which miR-218 controls multiple RTK pathway components, including both cell surface receptors and downstream effectors.

Fig. 1. miR-218 directly regulates RSK2 and S6K1.

(A) Schematic depiction of the miR-218-RTK signaling pathway. (B) Western blot for phosphorylated and total RSK and S6K1 in T98G and U373 cells transfected with miR-218 mimics, serum-starved for 24 hours, then stimulated with serum and EGF for 30 min. (C) Luciferase reporter assay with 3’UTRs of RSK2 and S6K1 seed sequences in T98G cells upon expression of a negative control (Neg) or miR-218 mimic. Reporters containing the corresponding mutated seed sequences (Mut) were used as controls. (D) Western blot for RSK2 and S6K1 abundance in U87MG, T98G and U373 cells 72 hours after transfection with miR-218 mimic. (E) Left: RSK2 and S6K1 protein accumulation after stable expression of pre-miR-218 in T3691 GSCs. Right: Densitometry of three assays assessing changes in RSK2 and S6K1 abundance in miR-218 treated relative to control T3691 cells. (F) Analysis of RSK2 and S6K1 mRNA isolated from intracranial orthotopic xenografts derived from T3691 GSCs expressing miR-218 (Tum-3691-218) relative to those expressing a scrambled RNA (SCR) (Tum-3691-SCR). (n=3 independent tumors). (G) qRT-PCR of RSK2 and S6K1 expression in GBM patient samples (red; G1–G30) and normal brain tissue (green; B1–B12). (H) Immunohistochemical analysis of RSK2 and S6K1 abundance in GBM (n=60) and normal brain tissue microarrays (n=40). Scale bar, 20 µm. (*) p<0.05, (**) p<0.005 and (***) p<0.0005. Data mean ± SEM from n as stated at each panel.

miR-218 modulates RTK feedback regulatory mechanisms by directly regulating PDGFRα

Regulatory feedback mechanisms that modulate RTK activity have a role in GBM and other cancers (12, 14). Because both S6K1 and RSK2 enhance ERK activity, we tested whether miR-218-mediated S6K1 and RSK2 inhibition might result in compensatory ERK re-activation. miR-218 mimic did not increase the phosphorylation of ERK in GBM cells (Fig. 2A), although S6K1 or RSK2 knockdown by siRNAs or shRNAs did, as expected (Fig. 2B and fib. S2A). These data suggest that miR-218 regulates additional targets that uncouple RTK signaling from established regulatory feedback control.

Fig. 2. Multi-level regulation of RTK members by miR-218.

(A) Abundance of phosphorylated ERK1 or ERK2 (ERK1/2) in T98G and U373 cells transfected with an miR-218 mimic. (B) Western blot for phosphorylated ERK1/2 in U87MG and U373 cells after stable knockdown of S6K1 and RSK2 by shRNA. (C) Western blot to examine the phosphorylation of S6K1 (at Thr³⁸⁹) and AKT [at Ser⁴⁷³ (S473)] and PDGFRα in U87MG and U373 cells exposed to AZD8055 (500 µM) for 1 or 24 hours (hrs). (D) Western blot for PDGFRα abundance in U373 cells transfected with a negative or miR-218 mimic and exposed to AZD8055 for 1 or 24 hours. (E) Western blot for PDGFRα abundance in U87MG and U373 cells transfected with a negative control (−) or miR-218 mimic (+). (F) Relative amount of PDGFRα mRNA isolated from Tum-3691-SCR and Tum-3691-21 intracranial xenografts (n=3 tumors each). (G) Expression of PDGFRα, S6K1 and RSK2 mRNA in U373 cells transfected with a control or miR-218 inhibitor. *p<0.05, **p<0.005 and ***p<0.0005. Data are means ± SEM from 3 experiments.

Because miR-218 suppresses the abundance of S6K1 (an mTORC1 substrate) and Rictor (a critical mTORC2 component) (15), we investigated the role of overall mTOR signaling in these phenotypes. Because hyperactivation of AKT and mTORC1 signaling suppresses PDGFRα accumulation as a compensatory mechanism (9, 11), we investigated whether miR-218 regulates PDGFRα activity. To mimic the combined effects of miR-218 on mTORC1 and mTORC2, we used a dual mTOR kinase inhibitor (AZD8055). Treatment of GBM cells with AZD8055 reduced phosphorylation of S6K1 (residue Thr³⁸⁹) and AKT (residue Ser⁴⁷³) at 1 and 24 hours, indicating inhibition of mTORC1 and mTORC2 activity, respectively (Fig. 2C). As expected, AZD8055 increased PDGFRα protein in U373 cells (Fig. 2C), but this effect was largely blocked by miR-218 (Fig. 2D). Similarly, shRNA-mediated inhibition of S6K1 coupled with miR-218 treatment reduced the levels of PDGFRα and phosphorylated AKT when compared to shS6K1 or miR-218 mimic alone (fig. S2B). S6K1 suppression by itself was not sufficient to increase PDGFRα abundance in U373 cells (fig. S2B).

Because PDGFRα was reduced in response to miR-218 mimic (Fig. 2D, and fig. S2B), we hypothesized that PDGFRα could also be a direct miR-218 target. PDGFRα 3’UTR luciferase reporter activity, mRNA and protein were all substantially repressed by miR-218 (Fig. 2E and fig. S2, C and D). Furthermore, PDGFRα mRNAs were decreased in orthotopic Tum-3691-218 tumors compared to Tum-3691-SCR samples (Fig. 2F). Overall, the expression of PDGFRα mRNA in GBM patient samples was greater than that in normal brain tissue (fig. S2E). These data indicate that PDGFRα is a direct miR-218 target. Combined with our aforementioned data, the results so far suggest that increased PDGFRα protein in miR-218-deficient GBM cells may reduce feedback inhibition from enhanced S6K1 and RSK2 signaling, thereby promoting sustained ERK activation (Fig. 2A). Finally, miR-218 inhibition modestly enhanced the expression of RSK2, S6K1, and PDGFRα (Fig. 2G), highlighting the importance of miR-218 repression in overriding the RTK feedback circuit. Notably, combined knockdown of RSK2 and PDGFRα or S6K1 and PDGFRα is not sufficient to abrogate the feedback ERK reactivation mechanism (fig. S2F). These data indicate that inhibition of only two miR-218 targets cannot quench feedback reactivation, likely due to compensation by other RTK pathway components, which are also miR-218 targets (1).

miR-218 repression promotes RTK activity by attenuating feedback inhibitory mechanisms

We next investigated whether miR-218 deficiency overrides feedback inhibitory mechanisms more generally in GBM cells. To test this hypothesis, we used a phospho-kinase antibody array to screen the activity of RTK signaling pathway components. Our previous studies demonstrated that miR-218-mediated effects were enhanced by ischemic or chemotherapeutic stresses that mimic the GBM tumor microenvironment (1). Temozolomide (TMZ) is a widely used GBM chemotherapeutic agent. Consequently, we challenged U87MG cells stably expressing scrambled (U87-SCR) or miR-218 (U87-218) with serum free medium (SFM) or a combination of SFM and TMZ for 72 hours, and evaluated the phosphorylation status of various RTK signaling members (Fig. 3, A and B; table S1). The phosphorylation of numerous RTK signaling components was significantly reduced in U87-218 cells compared to that in U87-SCR controls (Fig. 3, A to D). The reduction in RTK component phosphorylation was more pronounced when cells were exposed to SFM and TMZ combined than when cells were exposed to SFM alone (Fig. 3, A to D). These results indicate that miR-218 repression increases the basal activity of multiple RTK pathway effectors to override feedback inhibitory mechanisms, favoring sustained RTK signaling in GBM (Fig. 3E).

Fig. 3. Phospho-proteomic analysis of miR-218 expressing cells under microenvironmental stress.

(A and B) Phospho-kinase array performed with lysates from U87MG cells expressing scrambled control (SCR) or miR-218 cultured with (A) SFM or (B) SFM +TMZ for 72 hours. Boxed areas correspond to the phosphorylated proteins that were modulated by increased miR-218 abundance. The kinases corresponding to these codes are listed in table S1. (C, D) Quantification of phopsho-proteins from the proteomic array (average of duplicate spots). (E) Schematic model describing the role of miR-218 in regulating components of RTK feedback signaling. Yellow shade represents previously unidentified miR-218 targets *p<0.05, **p<0.005 and ***p<0.0005. Data are means ± SEM from 2 experiments.

RTK activation, in turn, suppresses miR-218 expression in GBM

Given that low miR-218 resulted in a cumulative increase in RTK pathway flux, we next investigated whether RTK activation might regulate miR-218 in a reciprocal fashion in GBM. We hypothesized that active RTK signaling would repress miR-218 transcription. Pharmacological inhibition of RTK activity using a pan-RTK inhibitor (sunitinib malate), a PDGFR, VEGF and FGFR inhibitor (SU6668), or a EGFR inhibitor (AG-1478) increased the expression of miR-218 in U87MG, U373 cells and T4121 GSCs (Fig. 4, A and B). Similarly, U87MG cells expressing a plasmid encoding the constitutively active EGFR variant III (EGFRvIII) significantly reduced miR-218 expression (Fig. 4C). This effect was lost when a “dead kinase” (DK) EGFRvIII construct (16) was used (Fig. 4C). Similar results were obtained using LN18 and U373 cells (Fig.4D, and fig. S4A). In addition, we determined that ligand-induced activation of the EGFR signaling pathway (by treatment with EGF) reduced miR-218 expression in U373 and LN18 cells (Fig. 4E). Analysis of TCGA data from primary human GBM revealed that miR-218 transcription is lower in samples with high EGFR abundance (Fig. 4F) compared to samples with low EGFR. In aggregate, these findings indicate that RTK signaling represses the miR-218 locus.

Fig. 4. RTK-mediated miR-218 suppression in GBM.

(A) miR-218 expression in T4121 GSCs and U87MG cells treated with a pan-RTK inhibitor (RTKi) for 24 hours. (B) miR-218 expression in T4121 GSCs and U373 cells treated with SU668 and AG-1478 for 24 hours. (C) miR-218 expression in U87MG cells exhibiting low (L), medium (M) or high (H) abundance of EGFRvIII, confirmed by Western blot. A construct expressing a kinase-deficient EGFR vIII (DK) was used as the control. (D) miR-218 expression in LN18 cells expressing a DK or EGFRvIII (vIII) plasmid. (E) miR-218 expression in LN18 and U373 cells stimulated with EGF for 24 hours. (F) miR-218 expression in GBM patient samples from TCGA that were stratified into two groups based on EGFR abundance (high, N=250; low, n=121). (G) Data depicting the relative amount of methylation at the SLIT2 locus using individual methylation probes spanning the locus, intronic sequences included. (H) Abundance of DNMT1, EZH2 and SUZ12 in cells expressing EGFRvIII. *p<0.05, **p<0.005 and ***p<0.0005. Data are means ± SEM from ≥3 experiments.

To explore the mechanism(s) of RTK signaling-mediated effects on miR-218 expression, we examined basic miR-218 transcriptional regulation in GBM cells. miR-218 is encoded within intronic sequences of the SLIT2 and SLIT3 genes. We observed a significantly higher positive correlation between miR-218 and SLIT2 expression relative to miR-218 and SLIT3 expression in GBM patient samples from the TCGA (fig. S3, A and B). Basal miR-218 mRNA abundance also correlated more closely with SLIT2 than SLIT3 in various GBM cell lines (fig. S3C), indicating that miR-218 predominantly originates from an intronic sequence within SLIT2 in this context. Both host genes are silenced by methylation in GBM (17, 18), suggesting a potentially epigenetic basis for miR-218 suppression by RTK signaling. TCGA data revealed variable DNA methylation patterns across the SLIT2 locus, with substantial methylation intensity observed at the position corresponding to miR-218, and to a lesser extent at the SLIT2 transcription start site (Fig. 4G). Because previous studies showed that DNA methyltransferase 1 (DNMT1) and histone methyltransferase EZH2 play an important role in SLIT2 suppression by methylation (19, 20), it seemed likely that DNMT1 and EZH2 result in the same repression of the miR-218 locus. Of note, treatment with the demethylating agent 5-azacytidine or an EZH2 inhibitor (DZNep) produced comparable increases in miR-218 and SLIT2 transcript levels (fig. S4, B and C), suggesting that DNMT1 and EZH2 exert epigenetic silencing along the SLIT2/miR-218 genomic region.

We then further investigated whether EGFRvIII activation contributes to epigenetic miR-218 silencing by regulating these epigenetic modifying factors, which mediate miR-218 silencing. Indeed, high EGFRvIII activation modestly increased DNMT1 protein accumulation in U87MG and LN18 cells (Fig. 4H), although the abundance of EZH2 and SUZ12 was not significantly altered. To define EGFR downstream effectors that potentially regulate DNMT1, we screened a cohort of inhibitors (such as those blocking AKT, ERK, STAT3 and JNK activity) targeting various RTK pathway members, and identified STAT3 and AKT as potential intermediates that promote DNMT1 expression (fig. S4D). Furthermore, shRNA-mediated STAT3 knockdown reduced DNMT1 mRNA and protein (fig. S4, E and F), consistent with a recent study demonstrating STAT3 induces DNMT1 in malignant T lymphocytes (21). Of note, EZH2 abundance was also reduced by AKT and STAT3 inhibition (fig. S4, D to F). Our data suggest a model wherein RTK signaling represses the miR-218 locus, probably through increased methylation, in a STAT3-dependent manner.

STAT3 dependent miR-218 repression in GBM

To test our model directly, we inhibited STAT3 in LN18 cells with shRNAs, which resulted in higher miR-218 expression (Fig. 5A). STAT3 knockdown did not alter expression of the host gene SLIT2 (Fig. 5B), but instead increased the abundance of primary-miR-218 (pri-miR-218) transcripts in LN18 cells, as well as in T3832, T4302 and T4121 GSCs (Fig. 5C and fig. S5, A and B). This differential regulation appeared to be independent of DNA methylation status (or EZH2 activity), because miR-218 and SLIT2 exhibited concurrent expression changes upon treatment with inhibitors of DNMT1 or EZH2 (fig. S4, B and C). Analysis of the SLIT2 genomic locus revealed a potential STAT3 binding site 4 kilobases downstream of the miR-218 transcriptional start site (designated the “ miR-218 +4kb” site) (http://genome.ucsc.edu/cgi-bin/hgGateway). Accordingly, chromatin immunoprecipitation (ChIP) analysis detected robust STAT3 binding to the miR-218 +4kb site (Fig. 5D) suggesting that STAT3 directly represses the miR-218 locus. Ectopic expression of STAT3 in U373 cells significantly reduced primary and mature miR-218 miRNAs (Fig. 5E). Therefore, our data suggest that miR-218 expression can be regulated by DNA methylation as part of the SLIT2 locus and independently by STAT3 repression, and that RTK signaling impacts particularly the methylation-independent regulation of miR-218 expression.

Fig. 5. STAT3 acts as a repressor in regulating miR-218 in GBM.

(A) miR-218 and (B) SLIT2 mRNA abundance upon stable inhibition of STAT3 in LN18 cells. (C) pri-miR-218 expression after STAT3 knockdown by shRNA in LN18 cells. (D) Schematic of the potential STAT3 binding site near the miR-218 genomic locus (top), and ChIP analysis for STAT3 at the 4 kb site downstream of miR-218 (bottom). 18s rRNA was used as the control. (E) pri-miR-218 or miR-218 expression in LN18 cells transfected with STAT3. (F) Schematic of the DNA pull down assay using biotinylated oligos spanning the STAT3-binding site adjacent to the miR-218 locus (top), and the subsequent Western blot for STAT3 and BCLAF1 (bottom). (G and H) pri-miR-218 expression upon BCLAF1 knockdown by each of two shRNAs in (G) T3691 and (H) T4302 GSCs. (I) ChIP analysis for BCLAF1 enrichment at the miR-218+4kb genomic locus. 18s rRNA was used as the control. (J) ChIP assay for BCLAFI at the miR-218+4kb site in LN18 cells infected with control (shSCR) or STAT3 shRNA (shSTAT3). (K) Proposed model of STAT3-dependent regulation of miR-218 by epigenetic and non-epigenetic mechanisms. *p<0.05, **p<0.005 and ***p<0.0005. #p=0.06. Data are means ± SEM from three independent experiments.

To identify other factors that interact with STAT3 to mediate miR-218 repression, DNA pull-down assays coupled with proteomic analysis were performed. Briefly, a 100 bp fragment spanning the miR-218 +4kb site was biotin labeled (Fig. 5F), incubated with LN18 nuclear extracts, and precipitated with streptavidin-agarose beads. After confirming the presence of STAT3 in the precipitate by immunoblot analysis (Fig. 5F), LC-MS/MS analysis was performed to identify additional bound proteins. From the list of >200 peptides detected by this analysis, we focused on proteins with known repressor functions, including PHB2, BRG1, CEBPZ, XRN2, and BCLAF1. To assess the role of these targets in mediating miR-218 suppression, we silenced their expression individually using shRNAs, and analyzed effects on pri-miR-218 transcription. BCL2-associated transcription factor 1 (BCLAF1) knockdown specifically increased pri-miR-218 but not SLIT2 expression in GSCs (Fig.5, G and H; fig. S5C), similar to the results obtained by STAT3 knockdown. We further confirmed increased enrichment of BCLAF1 bound to the miR-218+4kb region using the biotin-streptavidin DNA pull-down assay, as well as ChIP using a BCLAF1 antibody (Fig. 5, F and I). Finally, we silenced STAT3 to specifically investigate its role in BCLAF1 enrichment at the miR-218+4kb region, and determined that the presence of BCLAF1 is completely STAT3-dependent (Fig. 5J). Coimmunoprecipitation experiments failed to reveal a direct association between STAT3 and BCLAF1, suggesting that STAT3 is necessary for BCLAF1 recruitment to this site, but that other factors must also be involved. In summary, as shown in Fig. 5K, activated RTK signaling suppresses miR-218 expression by increasing the binding of STAT3 and BCLAF1 to the miR-218+4kb region. Their binding attenuates miR-218-mediated suppression of RTK pathway, and therefore forms a self-reinforced feedback loop.

Discussion

Genome-wide analyses of GBM patient samples have identified cumulative genetic and epigenetic changes during tumor development (22–24). During this multistep process, tumor suppressor genes are silenced, oncogenes are activated, and miRNA expression is altered to facilitate tumor cell survival and growth. miRNAs are of great interest because of their ability to target numerous genes, and their use in discovering biological mechanisms. Our experiments reveal that miR-218 impacts feedback regulatory components of RTK signaling (9, 11, 25), and that low miR-218 expression increases the abundance of multiple RTK pathway components, including cell surface receptors and downstream effectors. Together, these effects serve to overcome the homeostatic feedback inhibitory mechanisms that normally operate in cells with active RTK signaling. This regulatory phenomenon may have important implications for RTK signal transduction; for example, mTORC1 activation has been shown to initiate a negative feedback loop stemming from S6K1, leading to PI3K and MAPK inhibition (12). Therefore, GBM cells with high mTORC1 activity could experience inhibitory checks at the AKT and ERK levels. Our data reveal a previously unreported mechanism whereby low miR-218 expression increases baseline abundance of S6K1 and RSK2 to alleviate feedback inhibition and sustain RTK pathway activity in GBM cells.

In addition to these downstream targets, miR-218 also inhibits PDGFRα (26), an important feedback regulatory component of the PI3K-AKT signaling pathway. Previous studies have indicated that hyperactivation of PI3K-AKT significantly suppresses PDGFR expression, in an mTOR-dependent manner (11). Because PI3K-AKT signaling is highly active in GBM (27), this could lead to PDGFR inhibition as a compensatory mechanism for maintaining tumor microenvironmental homeostasis, especially during depletion of O₂ and nutrients. Our results indicate that reduced miR-218 abundance enables tumor cells to modestly increase the amount of PDGFR, without triggering strong feedback inhibitory responses. In a similar fashion, miR-296 expressed in primary tumor endothelial cells has been shown to reduce the abundance of hepatocyte growth factor-regulated tyrosine kinase substrate (HGS), resulting in increased VEGFR2 and PDGFR accumulation, which are targets of HGS-mediated degradation (28). These studies indicate the importance of miRNAs in fine-tuning the expression of RTK regulatory components to sustain signaling without activating inhibitory feedback responses.

miR-218 is downregulated in all gliomas when compared to normal brain tissue samples, but the repression is more pronounced in mesenchymal GBM when compared to proneural subtypes (1). Our results indicate that STAT3 mediates a repressive effect on miR-218 expression in GBM, and identify a specific STAT3 binding site closely linked to the miR-218 locus. STAT3 is predominantly known as a transcriptional activator, but multiple studies have shown that STAT3 can also act as a repressor (29, 30). Nevertheless, relevant proteins required for STAT3’s repressive activity have not been previously identified. Of note, STAT3 is a master regulator of a mesenchymal gene signature in gliomas (31), and we propose that STAT3 binding could be responsible for the highly repressed state of miR-218 in mesenchymal GBM. We also identified increased recruitment of BCLAF1 to the STAT3 binding region. BCLAF1 is a transcriptional repressor that interacts with antiapoptotic members of the BCL2 family (32). Of note, recent studies have indicated that BCLAF1 also interacts with BRCA1, and mediates the formation of a BRCA1-mRNA splicing complex during DNA damage (33). By direct interaction with BCLAF1, BRCA1 recruits the splicing machinery and favors pre-mRNA splicing of BRCA1 target genes, which promotes transcript stability (33). It is possible that an intronic region spanning the miR-218 locus is also preferentially spliced out through a BCLAF1-dependent co-transcriptional splicing event (33). Collectively, this study identifies complex regulatory feedback loops by which miR-218 repression increases the abundance of multiple RTK pathway components in GBM, thereby overcoming inhibitory feedback mechanisms that would otherwise restrain RTK signaling. In turn, increased RTK signaling increases the expression of the gene encoding STAT3, which directly binds to the miR-218 locus, along with BCLAF1, and represses its expression to reinforce low miR-218 abundance in GBM (Fig. 5K).

Materials and Methods

Cell culture reagents and treatments

GSCs were grown in Neurobasal medium, and supplemented with B27 1:50 (Invitrogen, Carlsbad, CA), epidermal growth factor (Sigma; 20 ng/ml) and basic fibroblast growth factor (20 ng/ml), as previously described (34). U87MG, U373, LN18 and T98G (ATCC) cells were cultured in DMEM containing 10% FBS and antibiotics. U87MG DK and EGFRvIII cells were generously provided by Dr. Matthew J. Lazzara (University of Pennsylvania). Staurosporine (Cell Signaling) was used induce cell death at 1–2 µM for four hrs. To create chemotherapeutic microenvironmental stress, we combined serum starvation with Temozolomide (TMZ; 250 µm) for 72 hrs. To activate MAPK signaling pathway, human epidermal growth factor (hEGF) was used at a concentration of 100ng/mL for 30 min. A dual mTOR kinase inhibitor, AZD8055 (Selleck Chemicals), was used at a final concentration of 500 nM. The RTK inhibitors sunitinib malate and AG-1478 hydrochloride (Tocris Bioscience) were used at a final concentration of 10 µM for 24 hrs. RTK inhibitor SU6668 (Tocris Bioscience) was used at a final concentration of 10 µM in T4121 cells and 50 µM in U373 cells for 24 hrs. For inhibiting methylation, 5 azacytidine was used at 10 µM final concentration for 6 days and dosing medium was changed every two days. EZH2 inhibitor (3-deazaneplanocin A; DZNep) was used for 24 hrs at a final concentration of 10 µM.

GBM patient samples

GBM patient tissue sections (n=30) and control specimens (n=12) were obtained from formalin-fixed paraffin embedded tissues (Department of Pathology and Laboratory Medicine, U. Pennsylvania). The control sections consisted of temporal lobectomy specimens obtained from patients with intractable epilepsy and showed histopathologic evidence of mild to focally moderate gliosis but no lesions. All the GBM blocks contained more than 95% tumor cells.

Western blot, immunohistochemistry (IHC) and immunofluorescence (IF)

Based on the appropriate experimental time points, cells were lysed by standard procedure in lysis buffer containing a protease inhibitor cocktail. Protein lysates (50 µg) were resolved on SDS-PAGE gels followed by immunoblot detection and visualization with ECL kit (PerkinElmer). Antibodies and concentrations used for immunoblots are as follows: BCLAF1 (Bethyl laboratories #A300-608; 1:1000), DNMT1 (abcam #ab1353; 1:1000), Cell Signaling antibodies used at 1:1000 dilution- rabbit RSK (#9355), rabbit p-RSK, rabbit S6K1 (#9202), rabbit p-S6K1 (#9208), rabbit β-tubulin (#2146), rabbit RSK2 (#9340), rabbit ERK1/2 (#4695), rabbit p-ERK1/2 (#9101), rabbit p-AKT(S473) (#9271), rabbit AKT (#9272), rabbit PDGFRα (#3164), rabbit EGFR (#4267), rabbit p-EGFR (#3777), rabbit EZH2 (#5246), rabbit SUZ12 (#3737), rabbit p-STAT3 (#9145) and mouse STAT3 (#9139). For IHC analysis, brain glioblastoma and normal tissue array (40 cases/80 cores) was purchased from Biomax (GL806). The paraffin embedded tissue array processing for IHC analysis was performed as described earlier (1). Antibodies and the concentrations used for IHC and IF are as follows: rabbit S6K1 (1:50, Cell Signaling), rabbit RSK2 (1:100, Cell Signaling).

RNA from orthotopic brain tumor xenografts

All experiments in mice were approved by the University of Pennsylvania Institutional Animal Care and Use Committee (IACUC) and performed in accordance with NIH guidelines. Intracranial xenograft tumor assays were performed in Nu/Nu mice, as described (35). Briefly, GSC T3691 tumor spheres (50,000 cells; T3691-SCR or T3691-218) in a total volume of 5 µL were implanted into the right frontal lobes of nude mice. The animals were sacrificed on day 25 post implantation upon exhibition of neurological symptoms. RNA was isolated from frozen brain tissue sections.

Luciferase reporter assay

The 3’ untranslated region (UTR) of RSK2, S6K1 and PDGFRα were cloned into pMIR-REPORT miRNA expression reporter vector (ABI). Twenty four hrs after transfection with negative or miR-218 mimics in T98G cells, the respective plasmids were transfected into cells with Fugene 6 transfection reagent (Roche) in 24-well plates. Renilla Luciferase plasmid was co-transfected and utilized to normalize the firefly Luciferase values expressed from the pMIR-REPORT expression vector. Luciferase assays were performed using the dual luciferase protocol (Promega) at 48 hrs post transfection of the corresponding plasmids containing the 3’UTR regions. For testing the role of specific seed sequences in the 3’UTR of miR-218 target genes, mutations were introduced into the miR-218 binding seed sequences and compared to the wild type sequence (table S2). The duplex sequences spanning the miR-218 binding sites for the respective target genes were cloned into pMIR-REPORT expression vector for further luciferase analysis.

Quantitative real time PCR

miRNeasy mini kit (Qiagen) was used to isolate total RNA, and the RNA was reverse transcribed into cDNA using High Capacity RNA-to-cDNA kit (ABI). Analysis of miR-218 expression was performed using the TaqMan MicroRNA Reverse Transcription Kit (ABI) according to the instructions. Taqman primers (purchased from Applied Biosystems) were used to measure the abundance of all the transcripts and analyses were performed on the ABI 7900HT system (ABI). All target mRNA expression was normalized to that of 18s. Sequences for 18s SYBR green primer set are given as follows: h18sF – GAATTCCCAGTAAGTGCGGG; h18sR – GGGCAGGGACTTAATCAACG. The qRT–PCR data reflect the average mRNA abundance from three independent RNA extractions and reverse transcription reactions with error bars showing standard error of the mean.

Transient or stable overexpression of miR-218, siRNAs, or shRNAs

For cell culture studies, cells were transfected with a control or a mature miR-218 mimic (Dharmacon) with HiPerfect reagent (Invitrogen). For viral transduction, cells were transduced with lentiviral particles bearing pCDH-miR-218, pLKO1-shRSK2 and pLKO1-shS6K1 plasmids generated in HEK-293T cells. 293T cells were transfected with pCDH-miR-218/SCR (System Biosciences), pLKO1-shRSK2 and pLKO1-shS6K1 (Open Biosystems) and viral packaging plasmids, according to the Fugene reagent protocol (Roche). For siRNA experiments, siRNAs were purchased from Dharmacon (siRSK2: ON-TARGETPLUS SMARTPOOL, HUMAN RPS6KA3; siS6K1: ON-TARGETPLUS SMARTPOOL, HUMAN RPS6KB1). Similar to miRNA mimics, cells were transfected with “negative” or siRNA against gene of interest. The efficiency of siRNAs and shRNAs was tested by western blotting.

Phospho-kinase array

The human phospho-kinase array was purchased from R&D systems. U87-SCR or U87-218 cells exposed to serum free medium (SFM) or SFM and TMZ for 72 hrs were solubilized in Lysis Buffer 6, and the concentration of the proteins was determined according to standard BCA assay. A total of 45 antibodies (table S1) printed onto two separate nitrocellulose membranes were used for this study. Briefly, the membranes were blocked with Array buffer 1, then incubated with cell lysates overnight at 4°C. The membranes were washed and then incubated with the detection antibody cocktail for 2 hrs at room temperature. The membranes were washed further, and chemireagent mix was used for autoradiographic detection of protein spots.

Biotinylated DNA pulldown analysis

The proteins bound to specific oligos (genomic region with STAT3 binding site that is kb downstream to miR-218 locus [http://genome.ucsc.edu/cgi-bin/hgGateway]) were isolated based on a protocol reported by Wu (36). Briefly, nuclear proteins were extracted according to standard methods and concentration of the sample was determined. The duplex biotinylated oligos (table S3) were incubated with streptavidin-agarose bead suspension and nuclear proteins for 2 hrs. The mixture was centrifuged and the supernatant was removed. The beads were washed followed by suspension in 2× Laemmli sample buffer, and incubated at 95°C for 5 minutes. The beads in Laemmli buffer were centrifuged and the supernatant was resolved on an SDS page gel for western blotting for indicated proteins.

LC-MS/MS analysis

The protein sample extracted using the biotinylated DNA pull down assay (as described above) was ran on a resolving gel. The LC-MS/MS analysis was performed at the Proteomics core facility at the Wistar Institute. Briefly, the gel band was excised and digested with trypsin, and the digests were analyzed by LC-MS/MS on an LTQ-Orbitrap XL mass spectrometer. MS/MS spectra generated were searched against a human database (UniProt.org) using SEQUEST. Only proteins identified by 2 or more peptides (high-confidence) were included, and common contaminants (eg. keratins) were removed.

Chromatin-immunoprecipitation (ChIP)

ChIP assays were completed using Imgenex QuikChIP kit from Novis Biologicals. Briefly, five million LN18 cells were cross-linked with 1% formaldehyde for 10 minutes at 37°C. The cross-linking reaction was stopped by glycine, followed by cell lysis using SDS buffer containing protease inhibitor cocktail. Cell lysates were sonicated into DNA fragments of 200–1000 bp in size, and immunoprecipitated with IgG, STAT3 (Cell Signaling) or BCLAF1 (Bethyl Laboratories) antibodies, followed by incubation with DNA/Protein A/G agarose. The immunoprecipitated DNA fragments were washed with a series of buffers, de-crosslinked at 65°C, and purified using QIAquick PCR purification kit (Qiagen). Semi-quantitative PCR was performed using the following primers: 218-STAT3+4K: 5’-TGCCTTCTGGCAAGCATTGT-3’ and 5’-TCCAAGGGTCTCCAGGAGTAT-3’; 18s: 5’-GAATTCCCAGTAAGTGCGGG-3’ and 5’-GGGCAGGGACTTAATCAACG-3’.

TCGA microarray data analysis

The available data for gene expression, miRNA expression and methylation were downloaded from the TCGA Data portal (https://tcga-data.nci.nih.gov/tcga/). Microarray data was analyzed using Partek software (Partek Inc.). In the case of gene expression, data generated on Affymetrix microarray platform HT_HG-U133A for 385 tumor and 10 normal samples was subjected to GCRMA normalization (GCRMA background correction, quantile normalization, log2 transformation and Median polish probeset summarization). For miRNA expression, we used previously normalized TCGA Level2 data from 426 tumor samples and 10 normal samples run on Agilent’s miRNA microarray. For methylation profile, we imported TCGA Level2 data for 111 tumor samples from Human Methylation 450 into Partek and analyzed the intensity of the methylation probes.

Statistics

Statistical significance was assessed using a Student's unpaired t-test for generating the p-values. p<0.05 was considered statistically significant.