MNK Inhibition Disrupts Mesenchymal Glioma Stem Cells and Prolongs Survival in a Mouse Model of Glioblastoma

This article is featured in Highlights of This Issue, p. 893

Glioblastoma is the most common and deadliest primary brain tumor (1). Despite surgical resection, chemotherapy and radiation, there are no effective treatments for glioblastoma multiforme (2). A subpopulation of cancer stem cells, referred to as tumor-initiating cells (TIC) or glioma stem cells (GSC), has been identified in glioblastoma multiforme and other high-grade gliomas (3–6). GSCs expressing a mesenchymal gene signature are particularly resistant to therapy, grow more rapidly than other subtypes, and express specific cancer stem cell markers (e.g., CD44; refs. 7–9). Developing strategies to target this resistant subpopulation of cells may lead to improved clinical outcomes.

Protein synthesis is a highly regulated process that contributes to oncogenesis and therapeutic resistance in glioblastoma multiforme and other cancers (10–12). MNKs regulate protein synthesis through phosphorylation of the eukaryotic translation initiation factor 4E (eIF4E), a member of the eIF4F cap-binding complex (13, 14). Phosphorylation of eIF4E by MNKs leads to translation of a subset of oncogenic transcripts (15). Inhibition of MNKs with small-molecule inhibitors or knockdown of MKNK1 and MKNK2 disrupts growth of glioblastoma multiforme cells and prevents tumor growth in vivo (16, 17). However, few clinically relevant MNK inhibitors are available and none have been shown to disrupt the growth of glioblastoma multiforme tumors in intracranial mouse models of the disease (10).

Merestinib (LY2801653) is a novel multikinase inhibitor, with potent in vitro activity against MNKs, MET, and other protein kinases (18–21). The compound has shown significant antitumor activity in several xenograft mouse models of non–small cell lung cancer and other solid tumors, including one subcutaneous xenograft model of glioblastoma multiforme (20). In this study, we sought to investigate MNKs as potential targets in GSCs. Our study suggests an important role for the MNK inhibitor, merestinib, as it inhibits MNK signaling in glioblastoma multiforme cells and GSCs, blocks growth of GSCs as neurospheres, and improves overall survival in an intracranial xenograft mouse model. These findings suggest a mesenchymal-specific role for MNKs in glioblastoma multiforme and highlight a particular vulnerability of mesenchymal GSCs for pharmacologic MNK inhibition.

Our results show that merestinib blocks phosphorylation of eIF4E in established glioblastoma multiforme cell lines and patient-derived GSCs. Analysis of data from The Cancer Genome Atlas (TCGA) reveals that the MKNK1 and MKNK2 genes are overexpressed in glioblastoma multiforme from the mesenchymal subtype. Furthermore, in glioblastoma multiforme, MKNK1 expression correlates with CD44, a mesenchymal GSC marker. Using patient-derived mesenchymal GSCs, we found that merestinib disrupts cancer stem cell viability and frequency, as determined by neurosphere formation and extreme limiting dilution analysis (ELDA). Finally, in an intracranial xenograft mouse model of glioblastoma multiforme, merestinib inhibited MNK signaling and improved overall survival.

Glioblastoma multiforme cell lines were grown in DMEM supplemented with FBS (10%) and gentamycin (0.1 mg/mL). U87 cells were authenticated by short tandem repeat (STR) analysis in January 2016 (Genetica DNA Laboratories). The isolation of patient-derived glioma stem cells and generation of GSC lines (83Mes, MD30, and GBM43) has been described previously (8, 22). GSCs were cultured in DMEM/F12 supplemented with EGF (20 ng/mL), bFGF (20 ng/mL), heparin (5 μg/mL), B27 (2%), and gentamycin (0.1 mg/mL). Merestinib was provided by Eli Lilly & Company and dissolved in DMSO for in vitro studies. For in vivo studies, merestinib was first dissolved in PEG400, followed by sonication and addition of 20% Captisol in water.

Cells were harvested and washed three times with cold PBS by centrifugation. Cell pellets were lysed with phosphorylation lysis buffer (50 mmol/L Hepes, 150 mmol/L NaCl, 1 mmol/L MgCl₂, 0.5% Triton, 10% glycerol, 0.5% sodium deoxycholate, pH 7.9) supplemented freshly with phosphatase and protease inhibitors (Roche). Protein concentrations were measured by Bradford assay (Bio-Rad) using the Synergy HT plate reader and Gen5 software (BioTek Instruments). Equal concentrations of whole-cell lysates were separated by SDS-PAGE (Bio-Rad) and transferred by semidry transfer to Immobilon PVDF membranes (Millipore). Membranes were blocked with 5% BSA in 1× TBST and incubated with primary antibodies overnight. Primary antibodies against phospho-eIF4E (Ser209) and eIF4E were obtained from Cell Signaling Technology and used at a dilution of 1:1,000. Following primary antibody incubation, membranes were washed three times with 1× TBST and incubated with anti-rabbit (GE Healthcare) or anti-mouse (Bio-Rad) horseradish peroxidase (HRP)-conjugated secondary antibodies for 1 hour. Membranes were then washed three times with 1× TBST and developed with WesternBright ECL HRP substrate (Advansta) and autoradiography film (Denville Scientific).

For polysomal fractionation, cell lysates were separated with a 10% to 50% sucrose gradient as described previously (23). Polysomal fractions were pooled and RNA was purified using the AllPrep RNA/Protein Kit (Qiagen). Specific primers for CCND1, CCND2, BCL2, and GAPDH (Thermo Fisher) were used for qRT-PCR. GAPDH was used for normalization.

The unprocessed Affymetrix exon-array datasets for 419 glioblastoma multiforme samples were downloaded from the TCGA data portal (https://tcga-data.nci.nih.gov/tcga). We followed the data preprocessing procedure described in our recent study (24). Samples underwent subtyping into one of four molecular classes of glioblastoma multiforme (classical, mesenchymal, proneural, or neural), as described previously (25). We used an isoform-based classifier to obtain the patient subtype information (24). Unpaired t tests were used to determine whether MKNK1, MKNK2, or MET were differentially expressed between different glioblastoma multiforme subtypes.

RNASeqV2 level 3–released gene level expression data for RNA sequencing (RNA-seq) were downloaded for glioblastoma multiforme and low-grade gliomas (LGG) from TCGA. The data processing and quality control were done by the Broad Institute's TCGA workgroup. The reference gene transcript set was based on the HG19 UCSC gene standard track. MapSplice (26) was used to perform the alignment and RSEM (27) to perform the quantitation. Unpaired t tests were used to determine whether MKNK1 was differentially expressed between LGG and glioblastoma multiforme. The upper quartile normalized RSEM count estimates were base-10 log transformed before t tests.

MKNK1 gene expression data were downloaded from the GBM Bio Discovery Portal (GBM-BioDP) as previously described using the Verhaak Core dataset (25, 28). Survival analysis of TCGA patients was performed using the multigene prognostic index from the GBM-BioDP. For survival analysis, gene expression data for MKNK1 and CD44 (Agilent G4502A_07) for patients from the Verhaak Core were used.

To determine cell viability following treatment with merestinib, the Cell Proliferation Reagent WST-1 (Roche) was used according to the manufacturer's instructions. Briefly, U87, 83Mes, MD30, or GBM43 cells were seeded into 96-well plates at a density of 3,000 cells per well in the presence of DMSO or merestinib at the indicated concentrations. After 5 days of incubation at 37°C in 5% CO₂, the WST-1 reagent was added and viability was quantified using a Synergy HT plate reader and the Gen5 software (BioTek).

To assess colony formation, the CytoSelect 96-Well Cell Transformation Assay (Cell Biolabs) was used according to the manufacturer's instructions. Briefly, U87, 83Mes, MD30, or GBM43 cells were seeded in soft agar at a density of 2,500 cells per well in the presence of DMSO or merestinib (10 μmol/L). After 7 days of incubation at 37°C in 5% CO₂, colony formation was quantified by solubilizing soft agar, lysing cells, and incubating cell lysates with the CyQUANT GR Dye (Cell Biolabs), followed by analysis with the Synergy HT plate reader and the Gen5 software (BioTek).

For analysis of apoptosis, the BD Pharmingen FITC Annexin V Apoptosis Detection Kit I (BD Biosciences) was used according to the manufacturer's instructions. Briefly, U87 cells were seeded into 6-well plates and incubated at 37°C in 5% CO₂ until reaching 70% confluence. Cells were then incubated with DMSO or merestinib (10 μmol/L) for 48 hours. Following treatment, cells were harvested using trypsin, washed three times with PBS, and stained with propidium iodide (PI) staining solution and FITC Annexin V. Stained samples were analyzed by flow cytometry and FlowJo 10 for Mac.

Cells were seeded into round-bottom 96-well plates (Greiner Bio-One) containing DMSO or merestinib at the indicated cell numbers by flow cytometry using forward- and side scatter, single-cell sorting as described previously (8). After 7 days, cells were stained with 0.1 μg/mL acridine orange as described previously (29). Neurospheres were imaged using the Cytation 3 Cell Imaging Multi-Mode Reader with a 4× objective. For ELDA, neurosphere diameters were measured using the Cytation 3 software. Neurospheres measuring ≥100 μm in diameter were scored positively for sphere formation for ELDA and analyzed using the ELDA online software (http://bioinf.wehi.edu.au/software/elda/; ref. 30).

All animal studies were carried out in accordance with the Institutional Animal Care and Use Committee of the Northwestern University (Chicago, IL). Luciferase-expressing U87 cells were intracranially injected (100,000 cells/μL with a total injection volume of 2 μL/animal) into 5- to 6-week-old athymic nu/nu female mice (Taconic Biosciences). Bioluminescence imaging was used to monitor tumor growth as described previously (31). At 17 days postinjection of tumor cells, mice were randomized into control and treatment groups according to intracranial tumor bioluminescence values. Mice were treated with vehicle control or merestinib at a dose of 12 mg/kg, twice daily (5 days of treatment and 2 days of rest) for 2 weeks. Mice were monitored until required euthanasia due to indication of neurologic compromise from increasing tumor burden or at 55 days after tumor cell injection.

After sacrificing mice, resected brains were harvested for hematoxylin and eosin (H&E) staining and immunohistochemical analysis. Brains were fixed with 10% buffered formalin overnight. Brains were then embedded in paraffin and sectioned for H&E staining and immunohistochemical analysis. Sections were processed using the BOND-MAX Automated IHC/ISH Stainer and its Polymer Detection System (Leica Biosystems). The Bond Dewax Solution (AR9222) was used at 72°C. The Bond Epitope Retrieval Solution 1 (AR9961) was used for 20 minutes at 100°C. Samples were pretreated with 3% hydrogen peroxide for 5 minutes. For detection of eIF4E phosphorylation, the eIF4E (Ser209) antibody [EP2151Y] (Abcam) was used at a 1:2,000 dilution for 15 minutes. The post primary polymer penetration enhancer reagent was added for 8 minutes, followed by the polymer poly-HRP secondary antibody for 8 minutes. Hematoxylin was used for 5 minutes. eIF4E phosphorylation (Ser209) was semiquantified by light microscopic analysis, in which a board-certified neuropathologist (C. Horbinski) ranked the tumors from strongest to weakest, followed by Spearman rank order correlation. Mitoses were scored per 10 high-power fields (600×). Both phospho-eIF4E and mitotic index were analyzed while blinded to treatment group.

Unless otherwise specified, statistics were performed using GraphPad Prism 6.0 for Mac.

Recent advances in cancer genomics have identified at least 4 distinct molecular subtypes of glioblastoma multiforme: classical, mesenchymal, neural, and proneural (25, 32, 33). These subtypes were initially categorized using gene expression profiling, which identified expression signatures similar to those found during normal neurogenesis. Our recent studies have revised subtype classification using isoform-level gene expression values, which provide more robust subtype classifications with greater prognostic significance (24). We sought to use our subtype classifications to understand how the MNK genes (MKNK1 and MKNK2) are differentially expressed in glioblastoma multiforme. Using RNA-seq data from TCGA, we discovered that MKNK1 and MKNK2 are significantly overexpressed in mesenchymal subtype glioblastoma multiforme when compared with other subtypes (Fig. 1A and B). MET, a previously identified mesenchymal gene and target of merestinib, was also enriched in the mesenchymal cohort (Fig. 1C; refs. 19, 34).

Given the importance of MNK1 in the maintenance of glioblastoma multiforme survival under various conditions (11, 17), we further explored the relationship between MNK1, molecular subtype, and glioma grade. Using data from three different gene arrays, we validated that MKNK1 is overexpressed in the mesenchymal subtype (Fig. 1D–F). We also found that MKNK1 expression increases with glioma grade and is highest in glioblastoma multiforme, when compared with grade 2 or grade 3 gliomas (Fig. 1G). These data align with findings by others indicating that MKNK1 expression is increased in glioblastoma multiforme when compared with normal human astrocytes or patients with oligodendroglioma or anaplastic astrocytoma (17). Furthermore, across all four glioblastoma multiforme subtypes, MKNK1 and the mesenchymal GSC marker, CD44, are positively correlated and predict poor prognosis when overexpressed concurrently (Fig. 1H and I; ref. 8). Taken together, these findings indicate an important role for MNK1 signaling in the mesenchymal subtype of glioblastoma multiforme and maintenance of mesenchymal GSCs, suggesting MNKs are promising targets in this glioblastoma multiforme subtype.

To study the effectiveness of MNK inhibition in glioblastoma multiforme, we employed one established cell line (U87) and three patient-derived GSC cell lines (83Mes, GBM43, and MD30; refs. 7–9, 22). The patient-derived cell lines were grown as nonadherent neurospheres in serum-free medium and were designated "glioma stem cells" (GSC). GSCs have unique properties, including the ability to form neurospheres, and are enriched for the mesenchymal GSC markers aldehyde dehydrogenase and CD44 (Supplementary Fig. S1). We have extensively characterized these GSCs in previous publications.

Given the potential role for MNKs in glioblastoma multiforme, we sought to study the effect of merestinib on MNK-mediated protein phosphorylation and mRNA translation in the established glioblastoma multiforme cell line, U87, and in GSCs. The phosphorylation of eIF4E on serine 209 was blocked by merestinib in U87 as well as the patient-derived GSC lines 83Mes and GBM43 (Fig. 2A–C). As eIF4E phosphorylation is important for active mRNA translation, we sought to determine the effects of merestinib on this process. To this end, we analyzed monosomal and polysomal fractions. Treatment of U87, 83Mes, and GBM43 cells with the inhibitor resulted in an altered translational profile, as demonstrated by an increase in the monosomal (40S, 60S, and 80S) peaks and a decrease in the polysomal peaks (Fig. 2D–F). Further analysis of U87 and 83Mes profiles shows a decrease in the area under the polysomal curve upon treatment, indicating a decrease in global protein synthesis (Fig. 2G and H). When transcripts for cyclins D1 and D2 undergoing active translation were analyzed, we found that mRNA levels for both these cyclins were significantly decreased in the polysomal fractions of merestinib-treated samples (Fig. 2I and J), suggesting that merestinib is a potent inhibitor of the translation of these oncogenic mRNAs. Taken together, these findings indicate that merestinib treatment inhibits MNK activity and protein synthesis in established glioblastoma multiforme cells and GSCs.

We next sought to identify the effects of merestinib on glioblastoma multiforme cells. For these studies, the established glioblastoma multiforme cell line, U87, was used. Cell viability, anchorage-independent growth in soft agar, and apoptosis were assessed following treatment with merestinib (Fig. 3). The inhibitor exhibited suppressive effects on cell viability and anchorage-independent growth in U87 cells (Fig. 3A and B). In addition, there was an increase in apoptosis following 48-hour incubation with the inhibitor (Fig. 3C). We then measured the effect of merestinib on cell viability, anchorage-independent growth, and neurosphere formation on GSCs. Merestinib treatment decreased cell viability in a dose-dependent manner in 83Mes, MD30, and GBM43 GSCs, with IC₅₀ values of 4.3, 4.9, and 3.2 μmol/L, respectively (Fig. 4A). Similarly, merestinib disrupted malignant transformation as measured by anchorage-independent growth in soft agar (Fig. 4B). We next examined whether merestinib could disrupt neurosphere formation in 83Mes and MD30 GSCs. When increasing numbers of cells were seeded, merestinib disrupted neurosphere size across most cell densities (Fig. 4C–F). Furthermore, ELDA of neurospheres was used to determine effects on stem cell frequencies. Merestinib led to a significant decrease in the GSC frequency in 83Mes and MD30 GSCs. Specifically, the stem cell frequencies in 83Mes GSCs dropped from 1 in 3.45 for DMSO to 1 in 30.24 cells for merestinib (Fig. 4G). Similarly, the stem cell frequencies for MD30 GSCs dropped from 1 in 16.6 for DMSO to 1 in 288.7 cells for merestinib (Fig. 4H). These results indicate a significant decrease in the cancer stem cell populations. Taken together, these results strongly suggest that merestinib disrupts these tumor-initiating cells in mesenchymal subtype gliomas.

In further studies, we tested the inhibitor in an intracranial xenograft glioblastoma multiforme mouse model. Nude mice were injected with luciferase-expressing U87 cells, and after tumor formation, mice were treated with either merestinib or control vehicle for two weeks and monitored for a total of 55 days. Comparison of control and merestinib treated mice shows a trend toward decreased tumor volume (Fig. 5A and B). Furthermore, merestinib treatment significantly prolonged survival when compared with the vehicle control (Fig. 5C). Analysis of tumors by IHC demonstrated a decrease in eIF4E phosphorylation, indicating that merestinib was able to inhibit MNK signaling in vivo (Fig. 5D). Rank order analysis of merestinib-treated or control samples demonstrated a significant decrease in eIF4E phosphorylation in the treated group (Fig. 5E). Finally, the number of mitoses per 10 high-power fields showed a trend toward reduced proliferation in treated samples when compared with controls (Fig. 5F).

Most glioblastoma multiforme patients die from tumor recurrence after standard therapy. Therefore, better treatment options for malignant brain tumors are desperately needed. New therapies for glioblastoma multiforme and other high-grade gliomas must include targeting of resistant GSC populations to prevent tumor recurrence and improve clinical outcomes. Furthermore, treatment strategies should be tailored for particular molecular subtypes. Mesenchymal subtype glioblastoma multiforme are among the most aggressive forms of the tumor with a median survival as low as 11.8 months (25). Mesenchymal tumors and GSCs are enriched with particular molecular markers (e.g., MET, CD44) and exhibit increased proliferation rates, increased radiation resistance, and poorer overall survival when compared with other subtypes (35). Developing therapies to target mesenchymal GSCs is an attractive approach that has the potential to improve outcomes in glioblastoma multiforme and other high-grade gliomas.

In this study, we present several novel findings. First, we use our previously described isoform-level, gene expression classification system (24) to provide evidence that the MNK genes (MKNK1 and MKNK2) are differentially expressed across glioblastoma multiforme subtypes and are most highly expressed in mesenchymal glioblastoma multiforme. Expanding upon this finding, we demonstrated that MKNK1 expression is increased in glioblastoma multiforme as compared with grade 2 and grade 3 gliomas. MKNK1 expression also correlated with the previously identified mesenchymal GSC gene, CD44, and simultaneous upregulation of both genes predicts poor prognosis in glioblastoma multiforme. These findings led us to explore the potential of a multikinase inhibitor, merestinib, which has shown potent in vitro activity against MNKs (20), as a therapy to eliminate mesenchymal GSCs. Our findings demonstrate that merestinib blocks MNK signaling in an established glioblastoma multiforme cell line and three GSC lines. Furthermore, merestinib significantly reduced global protein synthesis and inhibited translation of the oncogenic mRNAs, CCND1 and CCND2, and produced a small decrease in BCL2 mRNA in glioblastoma multiforme cells and GSCs. The inhibitor demonstrated activity in viability, soft agar, and apoptosis assays using the established glioblastoma multiforme cell line, U87. Furthermore, merestinib showed potent activity against GSC viability, growth in soft agar, and neurosphere growth, indicating an inhibition on the growth potential of these cells. Using ELDA, we also demonstrated that merestinib disrupts the sphere-forming potential of two mesenchymal GSC lines, indicating a suppression of the stem cell population. Finally, in an intracranial xenograft mouse model of glioblastoma multiforme, we found that merestinib disrupts MNK signaling in vivo and significantly prolongs survival in mice.

Merestinib is a multikinase inhibitor with activity against both MNK1 and MNK2 and has demonstrated potent antineoplastic effects in several solid tumors (18–20). For the first time, our study provides detailed analysis of this inhibitor in glioblastoma multiforme and establishes an important effect on cancer stem cells. The striking effect of the inhibitor on GSC growth, malignant transformation and neurosphere formation is particularly interesting, as it raises the possibility of a promising approach for targeting the tumor-initiating cancer stem cell population. Although MNK1 and MNK2 are the only serine/threonine kinases inhibited by merestinib, it is important to note that the inhibitor affects other kinases (20). In vitro assays have shown that the inhibitor blocks the activity of other protein kinases known to be important for the growth of glioblastoma multiforme and GSCs. In particular, the receptor tyrosine kinases MET and AXL, which have also been identified as drivers of mesenchymal GSCs (8, 9), are inhibited by merestinib at low doses (20). Therefore, the diversity of merestinib targets could prove beneficial, preventing the development of resistance by redundant or parallel signaling pathways (36–38).

Our findings provide strong evidence for targeting the MNK axis in glioblastoma multiforme tumors. However, the specific mechanism by which MNKs support the growth of GSCs remains elusive. Some studies have suggested that MNK1 positively regulates expression of the GSC-promoting factors TGFβ and Sema3C (17). In mesenchymal GSCs, TGFβ signaling has been shown to regulate proliferation, invasion, and promote immune evasion (17, 39). Furthermore, GSC secretion of the soluble factor, Sema3C, promotes cancer stem cell maintenance through activation of Rac1 signaling (40). Similarly, MNK signaling plays a key role in the maintenance of stem cell populations in other cancer types. In leukemia, the MNK–eIF4E signaling cascade has been implicated in the maintenance of leukemic precursors in blast crisis chronic myeloid leukemia (41). In line with these findings, we have shown that MNK inhibition with cercosporamide, an antifungal compound found to be a potent inhibitor of MNK1 and MNK2 activity (42), disrupts colony formation in primary leukemic progenitors from patients with acute myeloid leukemia (43). Taken together, these observations strongly suggest a role for MNKs in the maintenance of therapy-resistant cancer stem cell populations. Further studies are warranted and may show important clinical–translational implications for the treatment of glioblastoma multiforme.

No potential conflicts of interest were disclosed.

Conception and design: J.B. Bell, F.D. Eckerdt, L.C. Platanias

Development of methodology: J.B. Bell, F.D. Eckerdt, Y. Bi, A.A. Alvarez, C. Horbinski, C.D. James, L.C. Platanias

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): J.B. Bell, F.D. Eckerdt, K. Alley, L.P. Magnusson, H. Hussain, Y. Bi, A.D. Arslan, J. Clymer, S.-Y. Cheng, I. Nakano, C. Horbinski

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): J.B. Bell, F.D. Eckerdt, K. Alley, Y. Bi, A.A. Alvarez, C. Horbinski, R.V. Davuluri, C.D. James, L.C. Platanias

Writing, review, and/or revision of the manuscript: J.B. Bell, F.D. Eckerdt, K. Alley, H. Hussain, A.A. Alvarez, S. Goldman, S.-Y. Cheng, C. Horbinski, R.V. Davuluri, L.C. Platanias

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): L.P. Magnusson, H. Hussain, C. Horbinski

Study supervision: L.P. Magnusson, C.D. James, L.C. Platanias

This work was supported by the NIH grants CA155566, CA77816, and CA121192, and by grant I01CX000916 from the Department of Veterans Affairs. J.B. Bell was supported in part by NIH/NCI training grant T32 CA09560 and MSTP NIH training grant T32 GM008152. A.D. Arslan and A.A. Alvarez were supported in part by NIH/NCI training grant T32 CA070085.

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.