• Open Access

Tumor-Specific Activation of the C-JUN/MELK Pathway Regulates Glioma Stem Cell Growth in a p53-Dependent Manner§


  • Author contributions: C.G., K.J., Y.N., H.K., and S.G.: collection and/or assembly of data; Y.K.B.: collection and/or assembly of data and manuscript writing; I.N.: conception and design, data analysis and interpretation, financial support, and manuscript writing. *C.G. and Y.K.B.S. contributed equally to this article.

  • Disclosure of potential conflicts of interest is found at the end of this article.

  • §

    First published online in STEM CELLSEXPRESS January 17, 2013.


Accumulated evidence suggests that glioma stem cells (GSCs) may contribute to therapy resistance in high-grade glioma (HGG). Although recent studies have shown that the serine/threonine kinase maternal embryonic leucine-zipper kinase (MELK) is abundantly expressed in various cancers, the function and mechanism of MELK remain elusive. Here, we demonstrate that MELK depletion by shRNA diminishes the growth of GSC-derived mouse intracranial tumors in vivo, induces glial fibrillary acidic protein (+) glial differentiation of GSCs leading to decreased malignancy of the resulting tumors, and prolongs survival periods of tumor-bearing mice. Tissue microarray analysis with 91 HGG tumors demonstrates that the proportion of MELK (+) cells is a statistically significant indicator of postsurgical survival periods. Mechanistically, MELK is regulated by the c-Jun NH(2)-terminal kinase (JNK) signaling and forms a complex with the oncoprotein c-JUN in GSCs but not in normal progenitors. MELK silencing induces p53 expression, whereas p53 inhibition induces MELK expression, indicating that MELK and p53 expression are mutually exclusive. Additionally, MELK silencing-mediated GSC apoptosis is partially rescued by both pharmacological p53 inhibition and p53 gene silencing, indicating that MELK action in GSCs is p53 dependent. Furthermore, irradiation of GSCs markedly elevates MELK mRNA and protein expression both in vitro and in vivo. Clinically, recurrent HGG tumors following the failure of radiation and chemotherapy exhibit a statistically significant elevation of MELK protein compared with untreated newly diagnosed HGG tumors. Together, our data indicate that GSCs, but not normal cells, depend on JNK-driven MELK/c-JUN signaling to regulate their survival, maintain GSCs in an immature state, and facilitate tumor radioresistance in a p53-dependent manner. STEM CELLS 2013;31:870–881


The current first-line therapy for patients with glioblastoma multiforme (GBM) consists of maximal surgical resection followed by radiotherapy and Temozolomide chemotherapy. Even with intensive multimodal treatment, the median overall survival for patients with GBM is 14.6 months, and the 2-year survival rate is 26.5% [1, 2]. The limited efficacy of current therapies for GBM highlights the importance of increasing understanding of the physiology and pathology of this devastating disease. GBM is composed of heterogeneous types of tumor cells. Thus, the identification of cancer stem cells (CSCs) in GBM (glioma stem cells; GSCs) has attracted substantial attention as a therapeutic cellular target due to their relative resistance to current therapies [3–10]. Because regulatory signaling pathways are broadly shared by somatic and CSCs, molecularly targeted therapies would require the specific targeting of critical genes/pathways in GSCs but not in their somatic counterparts, neural stem/progenitor cells (NPCs).

Maternal embryonic leucine-zipper kinase (MELK) is a member of the snf1/AMPK family of serine/threonine kinases [11, 12]. Recent investigations, including ours, have largely, if not solely, focused on the mRNA expression and in vitro functional characterization of MELK. The data generated have suggested that MELK mRNA is elevated in various organ-specific stem cells and cancers [13–16]. In GBM, we previously reported that the expression of MELK mRNA is elevated in patient-derived GSCs and that both siRNA-mediated MELK knockdown and the pharmacological inhibition of a MELK-mediated pathway induced GSC apoptosis both in vitro and in vivo. In contrast, targeting MELK did not significantly affect the survival of normal NPCs derived from mouse brains in vitro raising the possibility that MELK is required for the survival of GSCs but not for somatic noncancer cells [15, 17]. The clinical relevance of MELK as a therapeutic target has been demonstrated by the inverse correlation of MELK mRNA levels with patients' survival periods [15, 18]. Furthermore, upregulated MELK mRNA is not restricted to cancer in the brain. Elevated MELK mRNA has been observed in tumor tissue samples derived from breast, colorectal, lung, and ovarian tumors [13, 14, 19–22]. In addition, MELK knockdown decreases the in vivo growth of transformed fibroblasts in a subcutaneous xenograft model, presenting the first evidence of the function of MELK in cancer in vivo [23]. In the context of MELKs' role in tumorigenesis, Melk-expressing mouse mammary cells possessed higher tumor-initiating potential, and the lentiviral delivery of MELK shRNA reduced mouse mammary tumorigenesis in vivo [13]. Although accumulating evidence suggests that MELK is an attractive molecular target, its protein expression is still poorly characterized, and MELKs' role in the maintenance of the stem cell state in GBM has not yet been clarified. Furthermore, targeted therapies for MELK in cancer have not been developed.

Signaling pathways that regulate the function of MELK in cancer remain poorly characterized. Biochemical analysis has found that exogenously expressed murine Melk binds to the zinc-finger-like Zpr9, which results in the enhancement of B-Myb transcription activities in murine cell lines [24]. The association of MELK with the cell cycle-regulated kinase CDC25b has also been reported, suggesting that MELK signaling is related to the G2/M progression as a mitotic kinase [20, 25, 26]. In Xenopus, mitogen-activated protein kinase (MAPK) phosphorylates MELK and enhances its kinase activity, specifically during mitosis [27]. Notably, c-Jun NH(2)-terminal kinases (JNKs) are an evolutionally conserved subgroup of MAPKs mediated by the upstream MAPKs–MKK4 and MKK7. The JNK pathway is required for the regulation of cell proliferation, apoptosis, and inflammatory responses in cancers, including high-grade gliomas (HGGs) [28–32]. Immediate downstream targets of JNKs include the oncoprotein c-JUN. The stimuli activated by JNKs trigger the phosphorylation of c-JUN, leading to c-JUN transcriptional activation in cancer. Interestingly, a recent study demonstrated that JNK-2 and c-JUN-specific siRNA decrease the expression of the stem cell-associated membrane protein Notch2 in patient-derived GSCs and that JNK2 knockdown diminishes the in vivo growth of glioma cell lines [33]. In this study, we sought to determine the functional roles of MELK in GSCs in vivo and to identify the signaling mechanism that orchestrates MELK signaling in GSCs and the normal counterparts, NPCs.



All tumor samples were obtained as approved by the Institutional Review Board at OSU (IRB No. 2005C0075) or UCLA, as described previously [17, 34-36]. The mice were experimentally used in accordance with the Institutional Animal Care and Use Committee guidelines at OSU under the approved protocol (2009A0241).

Tissue Culture

Surgery at OSU was performed by E.A. Chiocca and I. Nakano. Neurosphere (NS) cultures derived from the obtained tumor specimens were prepared as previously described [17, 34-36].

Tissue Microarray

Tissue microarray (TMA) consisting of three to six representative 0.6-mm cores from formalin-fixed, paraffin-embedded tissue blocks was generated in the Department of Pathology and Laboratory Medicine at OSU. HGG tissue samples were collected from patients who underwent surgery at OSU Medical Center. After immunohistochemistry, tissues too small and/or crushed were eliminated, and 91 samples were processed to the quantitative analysis using Immunoratio software (http://imtmicroscope.uta.fi/immunoratio/).

Cell Lysis and Immunoblotting

The following antibodies were used for Western blotting: anti-MELK (rabbit, 1:1,000, Sigma-Aldrich, St. Louis, MO), (rabbit, 1:1,000, Sigma-Aldrich http://www.sigmaaldrich.com/united-states.html), anti-c-JUN (rabbit, 1:1,000, Cell Signaling Technology, Danvers, MA (rabbit, 1:1,000, Cell Signaling Technology, http://www.cellsignal.com/), aantiphosphorylated JNK (rabbit, 1:1,000, Cell Signaling Technology, Danvers, MA, Anti-tubulin (Millipore, Billerica, MA) (mouse, 1:1,000, Millipore, http://www.millipore.com/offices/cp3/billerica, Billerica, MA), anti-HDAC1 (rabbit, 1:1,000, Sigma-Aldrich, St. Louis, MO), (rabbit, 1:1,000, Sigma-Aldrich), and anti-GAPDH (rabbit, 14C10, Cell Signaling Technology, Danvers, MA) (rabbit, 14C10, Cell Signaling Technology) antibodies.


Protein interaction was tested by coimmunoprecipitation. Either anti-Flag beads (Sigma-Aldrich, St. Louis, MO) beads (Sigma) or anti-MELK antibody cross-linked with protein A/G beads (Santa Cruz, Santa Cruz, CA) (Santa Cruz) was incubated with 500 μg of cell lysates at 4°C over night. Sixteen hours postincubation, protein was eluted from beads and subjected to SDS gel electrophoresis using NuPage gel apparatus (Invitrogen, Grand Island, NY) (Invitrogen) and immunoblotted for required protein.

Xenotransplantation of Tumor Spheres into Mice

Six- to eight-week-old female Athymic nude mice (nu/nu) mice (NCI/NIH, Bethesda, MD) were anesthetized with intraperitoneal administration of ketamine and Xylazine (Ketamine 75 mg/kg, Xylazine 5 mg/kg). Infected U87 or GBM528 spheres (or NSs) were dissociated and certain numbers of cells were stereotactically transplanted into the right striatum. After specified duration of time, the mice underwent intracardiac perfusion-fixation with 4% paraformaldehyde. Brains were removed and retrieved to create frozen sections. The serial section slides were stained using hematoxylin and eosin, Anti Ki-67 (mouse, Dako) (mouse, 1:1,000, DAKO, http://www.dako.com/us//?setCountry=true&purl=/) or glial fibrillary acidic protein (Anti GFAP (rabbit, Dako, Carpinteria, CA)) (rabbit, 1:1,000, DAKO, Z0334). Photos of sections with largest tumors were taken with BX51 fluorescent microscope (Olympus America, Inc.). These images were imported into ImageJ software for quantitative analysis.


Values are given as means ± SD, unless noted otherwise in the figure legend. The number of replicates is noted in the figure or legends. Absent error bars in the bar graphs signify SD values are smaller than the graphic symbols. Comparison of mean values between multiple groups was evaluated by the Tukey HSD test for Post-ANOVA pair-wise comparisons in a one-way ANOVA unless noted otherwise in the figure legend. Comparison of mean values between two groups was evaluated by chi-square test or t test. All statistical tests were two-sided. For all statistical methods, a p-value less than .05 was considered significant. For more detailed materials and methods, see supporting information.


MELK Protein Is Abundantly Expressed in HGGs and Is a Postsurgical Prognostic Indicator of HGG

To characterize MELK (+) cells in gliomas and normal brains, we collected 91 primary surgical specimens, including WHO Grade II glioma (n = 13), Grade III glioma (n = 34), GBM (n = 32), and adjacent normal brain (n = 12), and assessed MELK expression in these tumors using immunohistochemistry (Fig. 1A). MELK expression was predominantly found in the nuclei of tumor cells and was substantially higher in Grade III malignant glioma and grade IV GBM compared with normal brains or Grade II glioma (Fig. 1A). When Grade III and GBM were combined into the HGG group, a statistically significant elevation of MELK protein expression was observed in comparison to the level of MELK in the low-grade glioma (supporting information Fig. S1). MELK expression was not exclusive for GSC; non-GSCs also expressed MELK at lower and varying levels. To analyze the prognostic impact of MELK immunohistochemistry, we investigated whether MELK protein expression predicts the postsurgical survival periods of HGG patients. Fifty-seven tumors were divided into three groups based on MELK immunostaining intensities. Patients with higher MELK levels had shorter postsurgical survival periods (median survival of 15.8 months in the MELKhigh group vs. 31.3 months in the MELKlow group) (Fig. 1B). These data suggest that MELK is a prognostic factor in the postsurgical survival of HGG patients, indicating the clinical relevance of MELK as a therapeutic target in GBM.

Figure 1.

MELK protein is highly expressed in high-grade glioma. (A): Upper panels: representative images of immunohistochemistry showing that MELK expression is higher in GBM compared with Grade II glioma and normal brain. Lower panels: TMA slides, containing 91 samples (12 adjacent normal brain, 13 Grade II glioma, 34 Grade III glioma, and 32 GBM), were evaluated by MELK immunohistochemistry. Each dot represents the average of the percentage of MELK (+) cells in each sample. Bar = 100 μm; NB: normal brain; Gr.II: Grade II glioma; Gr.III: Grade III glioma. (B): Patients with high-grade glioma were divided into three groups according to their MELK expression levels, and their postsurgical survival rate was evaluated using Kaplan-Meier analysis. There is a significant difference between the high MELK expression group and the low MELK expression group. The p-value is .041 and the overall p-value is .03. (C): Graph representing the relative signal intensities of MELK mRNA expression determined by transcriptome microarray using normal astrocytes (n = 3) and glioma sphere samples (n = 30: triplicated samples from 10 indicated patients). (D): Upper panel: immunocytochemistry of dissociated GBM157 spheres shows that MELK is coexpressed with Nestin and SOX2, but not with NeuN or GFAP. Bar = 5 μm. Lower panel: the quantification of the above immunocytochemistry data showing the percentage of SOX2, Nestin, GFAP, and NeuN (+) cells that are also immunoreactive for MELK. (E): Western blot analysis with three GBM sphere samples (GBM157, GBM146, and GBM528) shows that the expression of MELK and CD133 is higher in tumorigenic GBM spheres (GSCs) compared with their differentiated progeny (non-GSCs), whereas GFAP expression is higher in non-GSCs. GAPDH was used as an internal control. Three independent experiments were performed on each GBM sphere samples and representative results are shown. Abbreviations: GBM, glioblastoma multiforme; GFAP, glial fibrillary acidic protein; GSC, glioma stem cell; MELK, maternal embryonic leucine-zipper kinase; NB, normal brain.

MELK Protein Expression Is Enriched in GSCs

To assess the MELK expression in GBM spheres (GSCs) (n = 30; triplicated samples from 10 patients) compared to normal astrocytes (n = 3), we measured the signal intensities in our transcriptome microarray results with these samples (Fig. 1C). MELK mRNA was significantly higher in GBM NSs as opposed to normal astrocytes. To specify the cell types of MELK (+) cells, we costained MELK (+) cells derived from GBM patient samples with the cell type-specific markers by immunocytochemistry (Fig. 1D; supporting information Fig. S2A). Majority of MELK (+) GBM cells were colabeled with the immature neural cell markers (Nestin and SOX2) but not with the lineage-committed differentiation markers (GFAP and NeuN). Immunocytochemistry of cryosectioned GBM spheres displayed a similar staining pattern (supporting information Fig. S2B). Immunoblotting of the cell lysates from GBM samples confirmed a higher expression of MELK in tumorigenic GSCs in all three samples (GBM157, 146, and 528) compared to their sister cultures, in which tumorigenic potential is lost, either by withdrawal of growth factors from culture medium or under serum-containing medium (Fig. 1E and data not shown) [17, 21, 36]. These data suggest that MELK is predominantly expressed in GBM cells harboring stem cell characteristics.

MELK Silencing Abrogates GSC Self-Renewal In Vitro and Tumor Growth In Vivo

We next investigated the function of MELK in GSCs in vitro and in vivo. To this end, MELK shRNA was introduced to GBM528 spheres using a lentivirus-mediated gene delivery system (Fig. 2A; supporting information Fig. S3A). We designed three MELK-specific shRNA vectors. Western blot analysis exhibited a strong suppression of MELK expression with the two vectors. To assess the effect of these vectors on GSCs in vitro, we used sphere formation assay as a surrogate measurement for the self-renewal of GSCs [17, 21, 36]. MELK depletion significantly abrogated the primary sphere-forming ability of GBM528 cells (Fig. 2A; supporting information Fig. S3B). GSC self-renewal was also investigated by a second round of sphere formation, yielding evidence of a further enhanced inhibitory effect by MELK depletion. Data were normalized by natural log transformation. Linear mixed effect model which considering observational dependencies across groups of no-target shRNA and MELK-targeting shRNA (shMELK) were used to model the effects of tumor type, shRNA, and their interaction. The effect of shMELK on the tumor type was tested by interaction contrast. The result indicated that the fold of change between primary and secondary sphere formation was 4.4 with shMELK (95% confidence interval: 2.8, 6.8) and 1.3 with no-target (95% confidence interval: 0.8, 2.1), and the ratio of the two was 3.3 which was significantly greater that 1 (95% confidence interval: 1.9, 5.5; p = .0009).

Figure 2.

MELK knockdown attenuates glioma stem cell (GSC) growth in vitro and in vivo. (A): Western blot analysis (left) shows the elimination of MELK by shMELK in comparison to no-target shRNA control. Phase-contrast images represent the effect of MELK elimination on in vitro sphere formation by MELK knockdown. The graph (right) shows that shMELK significantly decreases primary and secondary sphere formation compared with the no-target control. Data represent three independent experiments. (B): Representative microMRI images (left) of four individual mouse intracranial tumors derived from stereotactic xenografting GBM528 spheres (105 cells) after they were infected with a lentiviral vector that carried either no-target shRNA or shMELK. The arrowheads indicate the location of tumors in the control group, and the arrows indicate the injection tract in the shMELK group. The graph (right) indicates the statistically significant difference of the tumor sizes (n = 11). p ≤ .001. (C): Left panels: representative images of immunohistochemistry for no-target shRNA- and MELK shRNA-infected GBM528-derived tumors, stained with Hematoxylin & Eosine, GFAP, and Ki67. Necrotic areas within tumors are shown as arrows. Right panels: histogram represents the number of GFAP (+) and Ki67 (+) cells in no-target shRNA and MELK shRNA-infected GBM528-derived tumors. Bar = 100 μm. p < .05. (D): The Kaplan-Meier curve showing the survival of mice harboring intracranial tumors derived from GBM528 spheres infected with no-target or MELK shRNA. p < .002. Abbreviations: GBM, glioblastoma multiforme; GFAP, glial fibrillary acidic protein; MELK, maternal embryonic leucine-zipper kinase; NS, neurosphere; shMELK, shRNA-mediated knockdown.

We then tested whether the in vitro data are relevant to the in vivo function of MELK. We examined the effect of MELK depletion on the in vivo growth of GBM528 sphere-derived mouse intracranial tumors. T2-weighted MRI images at day 90 postimplantation showed significantly suppressed intracranial tumor growth by the implantation of MELK shRNA-transduced GBM528 spheres into the brains of immunocompromised mice compared with no-target shRNA-transduced spheres (Fig. 2B). Heterointense signals on T2-weighted images in the control group indicated massive hemorrhagic lesions in the tumors.

MELK depletion in GBM528 resulted in a reduced frequency of tumor formation in immunocompromised mouse brains, suggesting that MELK is likely required for tumor initiation (supporting information Fig. S4A). More significantly, a reduction of subsequent tumor sizes by MELK shRNA was observed (supporting information Fig. S4B). The brains implanted with the control tumor cells displayed gross evidence of highly angiogenic and hemorrhagic tumors, in stark contrast with the brains implanted with MELK shRNA-infected tumor cells (supporting information Fig. S4C). Histopathological analysis of the control brains revealed large, highly proliferative, vascular-rich tumors with widespread necrosis and hemorrhage, indicating that the tumors were WHO Grade IV GBM (incidence: 7/7). In contrast, the brains with MELK shRNA-transduced tumors displayed no noticeable necrosis or endothelial proliferation, indicating that the tumors were histopathologically WHO Grade III glioma but not GBM (incidence: 4/4) (Fig. 2C). When these tumors were immunostained with the differentiated glial marker GFAP, we observed a marked increase of the GFAP signal in the MELK shRNA-transduced tumors (Fig. 2C). Immunoreactivity to a proliferation maker Ki67 was markedly diminished in tumors with MELK knockdown. Collectively, these data indicate that MELK is required for the maintenance of the immature state of GBM cells in vivo. In addition, the overall survival of the control mice was significantly reduced compared with MELK shRNA-transduced tumors (Fig. 2D). However, all MELK shRNA-transduced tumors subsequently killed the mice due to tumor burden. We questioned whether the growth of the MELK shRNA-transduced tumors was due to the presence of a subset of GSCs that could grow without MELK or to a subset of transplanted GSC escapes from lentiviral MELK depletion. Western blot analysis of MELK with the two sets of tumors at the time of death indicated that MELK expression in MELK shRNA-transduced tumors returned to the basal level, indicating that the latter is the case (Supporting Information Fig. S5). Furthermore, MELK silencing-derived inhibition of glioma cell growth in vivo was confirmed with a conventional GBM cell line U87 (supporting information Fig. S6). Animals harboring MELK-depleted U87 cells survived significantly longer than the control mice (median survival of 14.5 days vs. 21.3 days). Collectively, these data suggest that MELK causally contributes to GSC self-renewal in vitro, and tumor growth and maintenance of the stem cell state of GSCs in vivo.

JNK Signaling Regulates MELK Expression in GSCs

We conducted a closer observation of the MELK immunohistochemistry data with surgical specimens of GBM and noticed an accumulation of MELK (+) cells at the perivascular areas in tumors (Fig. 3A). This observation was in accordance with the hypothesis of a perivascular niche for GSCs [9, 37-40]. We therefore hypothesized that some soluble factors from vascular endothelial cells may stimulate MELK expression. To address this hypothesis, we cocultured GBM157 spheres with human umbilical vascular endothelial cells (HUVECs) and evaluated the change in MELK expression by Western blot analysis (Fig. 3B). Coculture of GBM157 spheres with HUVECs markedly increased MELK expression in the GBM157 spheres. The elevated MELK expression appeared to be stimulated by soluble factors because the increase of MELK expression was also observed when GBM157 spheres were stimulated with the conditioned medium of HUVECs alone. In particular, the stimulation of GBM157 spheres with basic fibroblast growth factor (bFGF), but not with vascular endothelial growth factor, resulted in an increase of MELK signals detected by Western blot (Fig. 3B).

Figure 3.

JNK2 regulates nuclear MELK expression in a tumor-specific manner. (A): Immunohistochemistry of two GBM specimens demonstrates that MELK (brown) is preferentially expressed in the perivascular compartment of GBM tissues. The nuclei are counterstained with Hematoxylin in blue. Bar = 50 μm. V: tumor vessels. (B): Western blot analysis indicates that MELK expression is elevated in GBM spheres that were either cocultured with EC for 2 days (left) or treated with the CM from EC for 6 hours (right). The MELK expression level was also elevated when GBM spheres were stimulated with bFGF, but not with VEGF, for 6 hours. GAPDH was used as an internal control. Each experiment performed three times independently. (C): Upper panel: table representing the Western blot results showing the effect on MELK expression by 6-hour treatment with 10 different pathway inhibitors or vehicle (DMSO) using four GBM sphere samples (GBM146, GBM157, GBM205, and GBM206) and three normal sphere samples (16wf, 1105, and 1106). JNKiII diminished MELK expression in GBM spheres (highlighted in red), but not in normal spheres. Lower panel: the MELK mRNA expression showed no apparent changes when GBM146 spheres were treated with JNKiII. (D): Graph represents the Western blot data for MELK and phosphorylated form of JNK (p-JNK) expression in GBM sphere samples (n = 5) compared with normal sphere samples (n = 3). The values indicate the relative expression. Western blot with each sample was analyzed three times independently. (E): Immunocytochemistry shows that nuclear MELK expression is eliminated by JNKiII treatment in GBM146 spheres, but not in f16w (normal) spheres after 6 hours of treatment. Bar = 10 μm (left panels). Western blot indicates that the JNKiII-induced reduction of MELK expression occurs in the nucleus, but not in the cytoplasm of GBM146 spheres after 6 hours of treatment (right panels). Both immunocytochemistry and Western blot were performed three times independently. (F): Western blot shows that overexpression of the constitutively active form (CA) of JNK2, but not JNK1, increases MELK expression in GBM146 spheres (GSCs). Overexpression of CA JNK2 does not increase MELK expression in GBM146 cells propagated in serum-containing medium (non-GSCs). CTRL: empty vector. GAPDH was used as an internal control. Experiments repeated two times. Abbreviations: bFGF, basic fibroblast growth factor; CM, conditioned media; EC, endothelial cells; GBM, glioblastoma multiforme; GSC, glioma stem cell; JNKiII, JNK inhibitor II; MELK, maternal embryonic leucine-zipper kinase; NS, neurosphere; VEGF, vascular endothelial growth factor.

We then sought to determine which signaling pathways regulate MELK activity in GSCs. According to a previous study by Badouel et al. [20], MAPK can phosphorylate the Xenopus ortholog of MELK (xMELK), leading to the regulation of cell cycle progression. To dissect the MELK-associated signaling pathways in somatic versus glioma cells, we examined the effects of 10 different inhibitors of the downstream proteins of bFGF signaling on MELK protein expression using both GBM-derived and normal brain-derived spheres (Fig. 3C). Western blotting demonstrated that only the JNK inhibitor (JNKiII) resulted in a noticeable difference in the level of MELK when GBM and normal spheres were compared. The treatment of GBM spheres with JNKiII suppressed MELK protein expression in GBM spheres, but no obvious change was observed in normal spheres (Fig. 3C). These results were confirmed with four GBM samples (GBM146, GBM157, GBM205, and GBM206) and three normal sphere samples (f16w, 1105, and 1106) in a dose-dependent manner (Fig. 3C; supporting information Figs. S7, S8). In contrast, MELK RNA expression showed no apparent changes following the same treatment with two GBM samples (Fig. 3C; supporting information Fig. S9), suggesting that the effect of JNK on MELK is a post-transcriptional event.

We compared the expression of MELK and phosphorylated JNK (p-JNK) in five GBM spheres (GBM146, GBM157, GBM205, GBM206, and GBM1600) and three normal spheres (f16w, 1105, and 1106) (Fig. 3D). All five GBM sphere samples had substantially higher MELK and p-JNK expression than normal spheres. Because JNKs consist of multiple isoforms, the individual isoforms may have different effects on MELK expression. Western blot analysis of normal spheres (f16w) and two GBM sphere samples (GBM146 and GBM157) demonstrated the GBM sphere-specific activation of JNK2/3, but not JNK1 (supporting information Fig. S10).

Immunocytochemistry with MELK antibody demonstrated that the reduction of MELK in GBM sphere cells occurs specifically in the nucleus and not in the cytoplasm (Fig. 3E, left panel). This result was further confirmed by Western blot after subcellular fractionation of treated GBM cells (Fig. 3E, right panel). Next, we overexpressed mitogen kinase kinase 7 (MKK7)-driven constitutively active forms of JNK1α1 and JNK2α1 in GBM146 sphere cells and evaluated the change in MELK expression. Western blot analysis showed that JNK2α1, but not JNK1α1, increased MELK expression in GSCs, but not in non-GSCs (Fig. 3F; supporting information Fig. S11). Collectively, these data suggest that JNK2 is an upstream regulator of MELK, specifically in the nuclei of GSCs but not in normal cells.

MELK Is a Direct Target of the Oncoprotein C-JUN in GSCs But Not in Normal Progenitors

The downstream targets of JNK2 include the oncoprotein/transcription factor c-JUN. Affymetrix Human Genome U133A Array indicates that comparison of MELK and c-JUN expression profile shows statistically significant correlation of the expression of these two genes in GBM patients (Fig. 4A). Similar to the intratumoral regions with high MELK expression, immunohistochemistry detected the preferential expression of c-JUN at the perivascular tumor areas in the GBMs (Fig. 4B). Immunocytochemistry of dissociated GBM146 spheres demonstrated that MELK colocalizes with c-JUN in the nuclei of GBM cells (Fig. 4B).

Figure 4.

The downstream target of JNK, c-JUN oncoprotein, mediates MELK expression and binds MELK in GSCs, but not in normal progenitors. (A):MELK and c-JUN expressions from gene expression omnibus profile were compared for their correlation in their expression in GBM patients. Pearson correlation coefficient was 0.34 with p-value = .0006. (B): Upper panels: representative images of immunohistochemistry demonstrate the perivascular expression of MELK and c-Jun in the same sample of GBM. Arrows indicate the same vessel in serial sections. Bar = 50 μm. Lower panels: representative images of immunohistochemistry show the nuclear colocalization of MELK and c-JUN in dissociated GBM146 spheres. Bar = 5 μm. The proportion of MELK (+) cells in c-JUN (+) cells in GBM146 spheres is shown in the bottom. (C): Upper panels: Western blot analysis with 293T cells transfected with control vector (CTRL) or MELK-Flag vector following immunoprecipitation with anti-Flag antibody displays the binding of exogenously expressed MELK to c-JUN. Middle panels: Western blot following immunoprecipitation with MELK antibody demonstrates the binding of endogenous MELK to c-JUN in GBM spheres (GBM146), but not in normal spheres (f16w). Lower panels: Western blot analysis with 293T cells overexpressing MELK-Flag vector (WT) or MELK-Flag vector that is devoid of kinase activity (D150A) following immunoprecipitation with anti-Flag antibody. These results are the representation of three independent experiments. (D): Western blot demonstrates that two c-Jun siRNAs targeting different regions (sic-JUN #1 and #2) decrease the amount of MELK protein, whereas MELK siRNA (siMELK) does not significantly alter c-JUN expression. GAPDH was used as an internal control. Experiment was repeated three times. (E): Western blot demonstrates that the dominant-negative form (DN) of c-JUN diminishes MELK expression in GBM146 spheres (GSCs), but not in GBM146 cells propagated in serum-containing medium (non-GSCs). (S): short time exposure, (L): long time exposure. GAPDH was used as an internal control. Abbreviations: GBM, glioblastoma multiforme; GSC, glioma stem cell; MELK, maternal embryonic leucine-zipper kinase.

We then examined whether MELK physically associates with c-JUN. Exogenously expressed MELK protein tagged with the Flag epitope was pulled down with the Flag antibody, and Western blot analysis with c-JUN antibody was performed (Fig. 4C, upper panels). The detection of a band corresponding to c-JUN indicated that exogenously expressed MELK is bound to c-JUN. Binding of endogenous MELK to c-JUN in GSCs was also confirmed by immunoprecipitation of GBM157 protein samples with MELK antibody (Fig. 4C, middle panels). Notably, this physical interaction between MELK and c-JUN was not detected in normal spheres (f16w) (Fig. 4C, middle panels), indicating that the MELK/c-JUN complex is likely tumor specific. The expression of kinase-deficient MELK (D150A) decreased this interaction, suggesting that the interaction is MELK-kinase dependent (Fig. 4C, lower panels).

Based on these observations, we predicted that MELK and c-JUN might have transcriptional interaction. Using siRNA-mediated gene knockdown, we observed that c-JUN depletion strongly suppressed MELK protein expression, whereas MELK knockdown resulted in a marginal change in c-JUN expression (Fig. 4D). These data suggested that the presence of c-JUN is required for MELK, but not vice versa. We next investigated whether the inactivation of c-JUN reduces MELK expression. Overexpression of the dominant-negative form of c-JUN diminished MELK expression in GSCs, but not in non-GSCs (Fig. 4E; supporting information Fig. S11).

MELK Signaling on the Survival of GBM Cells Is p53 Dependent

One of the major downstream targets of the oncoprotein c-JUN is the tumor suppressor gene p53. The loss of p53 by either mutation or deletion is associated with glioma initiation and progression and CSC maintenance in a number of malignancies [41–43]. Thus, we examined whether MELK signaling interacts with p53. As expected, siRNA-mediated MELK depletion increased the expression of both p53 mRNA and protein in a statistically significant manner in p53-intact GBM spheres (Fig. 5A; supporting information Fig. S14). Next, we measured the promoter activity of p53 under the expression of MELK in 293T cells and U87 cells lines. MELK overexpression suppressed the promoter activity of p53 (Fig. 5B left panel; supporting information Fig. S12). In turn, depletion of MELK by MELK shRNA targeting its 3′ untranslated region (UT) increased the p53 promoter activity in U87 cells. Of note, exogenous restoration of MELK by overexpression of the MELK coding region in MELK 3′UT shRNA-treated U87 cells reversed the phenotype (Fig. 5B right panel; supporting information Fig. S12). These data suggest that MELK negatively regulate p53 activity. Conversely, treatment of U87 cells with p53 inhibitor, pifithrin, paradoxically increased the mRNA level of MELK (Fig. 5C, left panel). Similarly, depletion of p53 by siRNA in U87 cells increased the expression of MELK mRNA (Fig. 5C, middle panel; supporting information Fig. S13). This data were also supported using two isogenic U87 cell lines with the single difference of p53 status (U87 MG as the p53-intact line and U87E6 as the p53-inactivated line) (Fig. 5C, right panel). We found that MELK expression was markedly elevated in p53-inactivated cells compared with p53-intact cells and that MELK silencing induces the upregulation of p53 protein in p53-intact cells but not in p53-inactivated cells (Fig. 5C, right panel). Furthermore, the MELK silencing-mediated induction of p53 expression was also observed with the patient-derived glioma sphere sample GBM1600 (p53-intact sample), but not in GBM2313 (p53-deleted sample) (supporting information Fig. S14). Taken together, these data indicate the mutual inhibitory regulation of MELK and p53 in glioma cells.

Figure 5.

MELK downregulation increases p53 expression and induces apoptosis in a p53-dependent manner. (A): Left panels: Representative images of RT-PCR displaying that MELK knockdown by siRNA (siMELK) increases p53 mRNA expression in GBM157 spheres. Middle panels: representative images of Western blot displaying that the p53 protein level is also increased by MELK knockdown. siRNA for GFP (siGFP) was used as a control. Right panel: quantitative analysis of three independent experiments demonstrating the significance. p < .01. (B): Left panels: graph indicating the p53 promoter activity in 293T cells overexpressing MELK or GFP (control), determined by luciferase reporter assay. p-Value ≤ .001. Right panels: 293T cells cotransfected with p53 reporter plasmid together with MELK shRNA (shMELK) (targeting 3′ untranslated region: UT) or shMELK (3′UT) plus MELK expression plasmids. Forty-eight hours post-transfection, cells were measured for the p53 promoter activity. p-Value for shMELK with respect to nontarget shRNA is ≤.001. p-Value for shMELK+MELK with respect to shMELK is ≤.001. All the experiments were performed in triplicates. (C): Left panel: graph indicating relative MELK expression determined by qRT-PCR in U87 cells treated with DMSO or 40 μM of p53 inhibitor pifithrin for 48 hours. Experiment was done in triplicates. p = .0006. Middle panel: U87 cells were transfected with control or p53 siRNA. Seventy-two hours post-transfection, mRNA level of MELK was measured by qRT-PCR. p-Value is .0002. Experiment was done three times independently. Right panel: representative images of Western blot using U87MG (p53-intact) and U87E6 (p53-inactivated) cells following lentiviral infection with no-target shRNA and shRNA for MELK (shMELK). MELK silencing induces p53 and p21 expression in U87MG cells, but not in U87E6 cells. (D): Left panel: U87 cells infected with nontarget or MELK shRNA. Twenty-four hours postinfection, cells were treated with 80 μM of pifithrin for another 24 hours and subjected to apoptosis assay with Annexin V labeling (n = 3). p-Value was .002 for shMELK and shMELK plus pifithrin treatment. Right panel: U87 cells transfected with siRNA for MELK (siMELK) and/or siRNA for p53 (sip53). siRNA for GFP was used as a control. Effect on apoptosis is measured by Annexin V labeling (n = 3). p-Value <.001 between siMELK and siMELK+sip53 treatment that are positive for Annexin V. Abbreviations: GFP, green fluorescent protein; MELK, maternal embryonic leucine-zipper kinase.

Activation of p53 plays a central role in cellular senescence and apoptosis [44]. To understand the physiological role of MELK-p53 pathway more in depth, we performed apoptosis analysis by Annexin V labeling in U87 cells with or without inactivation of p53 using pifithrin or siRNA-mediated p53 silencing under the background of MELK depletion (Fig. 5D). First, induction of apoptosis by MELK depletion was confirmed by the increased percentage of Annexin V (+) U87 cells. This increased rate of apoptosis was partially but significantly diminished by both pharmacological p53 inactivation with pifithrin (Fig. 5D, left panel) and siRNA-mediated p53 silencing (Fig. 5D, right panel), suggesting the critical role of p53 in the MELK pathway for the maintenance of GBM cell survival. Collectively, these data suggest that the MELK signaling in GBM cell survival is p53 dependent.

MELK Inhibits Radiation-Induced Apoptosis of GSCs and MELK (+) Cells Are Accumulated in Recurrent HGGs After Therapy Failure

Because GSCs are considered relatively refractory to current therapies, including radiation, we hypothesized that MELK causally contributes to the survival of GSCs following radiation insult. We measured the expression of MELK mRNA in irradiated GBM 528 spheres at different time points. Upregulation of MELK expression was observed as early as 1 hour postradiation, suggesting that MELK is a stress-induced kinase (Fig. 6A upper panel). Likewise, c-JUN level increased in radiation-treated GBM528 spheres (Fig. 6A, middle panel). To rule out the possibility that the increase of MELK and c-JUN is secondarily due to increased GSC population by radiation, we investigated the expression of a stem cell marker, CD133, with the same sample set (Supporting Information Fig. S15). Unlike MELK and c-JUN, CD133 expression did not exhibit any noticeable changes during this short duration following radiation treatment. We then performed Western blot to verify MELK and c-JUN expression in radiation-treated GSCs, both in vitro and in vivo (Fig. 6A, lower panels). Radiation treatment increased MELK and c-JUN expression in GBM528 spheres in vitro in a dose-dependent manner and GBM528-derived mouse intracranial tumors in vivo. The radiation treatment-induced increase of MELK expression was further observed in two patient-derived GBM sphere samples (GBM1600 and GBM2313) (supporting information Fig. S14).

Figure 6.

MELK facilitates survival of radiation-treated glioma stem cells and recurrent high-grade glioma tumors after therapy failure express elevated levels of MELK. (A): Upper and middle panels: GBM 528 spheres were treated with radiation (5 Gy) and the MELK and c-JUN mRNA levels were measured by qRT-PCR. Lower right panel: GBM 528 spheres treated with indicated doses of radiation and subjected to Western blot to assess the expression of MELK and c-JUN. Lower right panel: mice harboring GBM528 sphere-derived intracranial tumors were treated with whole brain radiation (5 Gy) at 2 weeks post-transplantation. Treated tumors were collected at indicated time points, and Western blot analysis was performed to determine the expression of c-JUN and MELK in tumors. (S): short time exposure, (L): long time exposure. Numbers below each Western blot represents the quantification data of the respective bands. The data were confirmed by three independent experiments. (B): GBM157 spheres transfected with siRNA for MELK (siMELK) or MELK overexpression vector were subjected to radiation treatment (5 Gy). The proportion of apoptosis at 48 hours postradiation treatment was analyzed by flow cytometry using Annexin V antibody and propidium iodide. siRNA for GFP (siGFP) and GFP vector were used as a control. The data were confirmed by three independent experiments. (C): Upper panels: graph indicating the proportion of MELK (+) cells, determined by immunohistochemistry, in primary (newly diagnosed) and recurrent GBM tissues in unmatched (left) and matched (right) cases. For unmatched cases p = .0086 and for matched cases p = .038. Lower panels: representative images for immunohistochemistry with two matched patient samples with primary and recurrent GBM tumors. The arrows indicate vascular sclerosis by radiation-induced damage in recurrent tumors. Bars = 50 μm. Abbreviations: GBM, glioblastoma multiforme; GFP, green fluorescent protein; MELK, maternal embryonic leucine-zipper kinase.

Next, we sought to determine the function of MELK in radiation-treated GSCs. We performed overexpression and silencing experiments with GBM157 spheres (Fig. 6B). Radiation treatment of GBM157 spheres increased the Annexin V (+) apoptotic cell population from 3.8% to 14.3%. MELK silencing in radiated GBM157 spheres resulted in a further increase of the apoptotic cell population to 34.0% (Fig. 6B). Conversely, the overexpression of MELK decreased the radiation treatment-induced apoptosis in GBM cells from 47.4% to 29.6% (Fig. 6B). Taken together, these data suggest that MELK plays a critical role in GSC survival after radiation insult.

Lastly, to determine the clinical relevance of the roles of MELK in the establishment of therapy resistance in HGG, we collected 71 HGG tissues that we treated at OSU. The samples included 41 newly diagnosed untreated (primary) tumors and 30 recurrent tumors. All patients with recurrent tumors underwent conventional treatments, including postsurgical radiotherapy and Temozolomide chemotherapy, prior to secondary surgery. MELK immunohistochemistry of these samples showed a statistically significant elevation of MELK expression in the recurrent tumors (Fig. 6C). Direct comparison of MELK expression with the matched patients, who underwent two surgeries at OSU for newly diagnosed tumors and subsequent recurrent tumors after therapy failure (n = 10), further confirmed these results (Fig. 6C).


In this study, we demonstrated a set of novel clinical and molecular mechanistic findings: (a) MELK protein is elevated in HGGs and GSCs derived from these tumors; (b) MELK immunoreactivity is a prognostic indicator of the postsurgical survival periods of HGG patients; (c) MELK is required for the growth of GSCs both in vitro and in vivo; (d) the JNK/c-JUN pathway regulates nuclear MELK in GSCs, but not in normal cells; (e) c-JUN physically associates with MELK, and this protein complex formation is dependent on the kinase activity of MELK; (f) p53 and MELK expressions in GBM cells are mutually exclusive and p53 negatively regulates MELK action on GSCs; (g) radiation treatment increases the expression of MELK and c-JUN in GSCs in vitro and in vivo; and (h) the expression of MELK is elevated in relapsed and recurrent GBM tumors compared with untreated GBM tumors. Therefore, the JNK-driven c-JUN/MELK interaction represents a critical mechanism for controlling GSC survival, self-renewal, radioresistance, and tumorigenesis in a p53-dependent manner.

Due to the differences in intracellular circuitry, the dependence of CSCs on specific oncogenes may differ from that of non-CSCs and normal cells in the original tissues [45–47]. A recent study has demonstrated that a MAPK, JNK, plays a role in the maintenance of self-renewal and tumor initiation in GBM cells, as evidenced by data demonstrating that treatment with a JNK inhibitor decreases the tumor stem-like cells in GBM [48]. In this study, we found that the inhibition of JNK diminishes nuclear MELK post-transcriptionally in GSCs, but not in normal cells (Fig. 3E). We further identified that MELK binds c-JUN, the downstream target of JNK, in GSCs, but not in normal progenitors (Fig. 4C). This binding of c-JUN to MELK was diminished when the kinase activity of MELK was nullified (Fig. 4C), indicating that the kinase activity of MELK is required for its interaction with c-JUN. The tumor-specific MELK interaction with JNK/c-JUN is likely one of the reasons for the selective apoptosis that occurs as a result of MELK inhibition in GSCs, but not in normal progenitors. Further investigation would elucidate the mechanism of the tumor-specific regulation of MELK by JNK/c-JUN.

The oncogenic transcription factor c-JUN composes the activator protein-1 (AP-1) complex that binds to promoter sites of a variety of downstream target genes for their transactivation [49, 50]. The negative regulation of tumor suppressor gene p53 is one of the major downstream signaling by AP-1 in human cancers. In GBM, alteration of p53 signaling is observed in as many as 87%. In this study, we found that the expression of p53 exhibited an inverse correlation with MELK expression: MELK silencing increased p53 expression, whereas p53 inhibition increased MELK expression (Fig. 5A, 5C). Also, the promoter activity of p53 decreased under MELK overexpression, whereas MELK shRNA exhibited the opposite result (Fig. 5B; supporting information Fig. S12). Given that MELK and p53 appear to be mutually exclusively expressed in glioma cells, it is likely that MELK negatively regulates p53 activity and vice versa. Regarding the functional interaction of MELK and p53, MELK depletion-mediated GBM cell apoptosis is, at least partially, rescued by p53 inhibition (Fig. 5D). Various mitotic kinases (e.g., MDM2) are known to regulate p53 signaling through transcriptional activation/inhibition and/or the formation of a protein complex with p53 protein in cancers [51]. Similar to the present data regarding MELK action on p53, c-JUN expression is also known to be mutually exclusive with p53 expression, and these two proteins physically interact with each other [50]. Similar to c-JUN, MELK may form complex with p53, thereby bind to the p53 promoter for autoregulation of p53 activity in GSCs. Alternatively, c-JUN may act as a substrate for the kinase activity of MELK to affect p53 signaling. Further studies would clarify these possibilities.

This study provided several pieces of evidence for the clinical relevance of MELK as a target for GBM. First, experimental radiation treatment strongly upregulated MELK protein in GBM spheres in vitro and GBM sphere-derived mouse tumors in vivo (Fig. 6A). These data indicate that MELK is likely a stress-induced kinase. Second, further supporting this hypothesis, the blockage of MELK upregulation in radiation-treated GSCs through shRNA-mediated silencing resulted in an increase of their radiation-induced apoptosis. In turn, MELK overexpression resulted in a partial rescue by preventing GSC apoptosis (Fig. 6B). Third, the statistically significant elevation of MELK immunoreactivities in recurrent HGG tumors after radiation/chemotherapy was observed in our clinical samples. These biological and clinical data indicate that MELK causally contributes to the survival of GSCs, particularly when they suffer from radiation insult.


In this study, we provide the first evidence of the novel and specific roles of c-JUN/MELK signaling in GSCs. shRNA-mediated MELK silencing suppressed multiple rounds of tumor sphere formation in vitro (self-renewal), induced the differentiation of GSCs into a GFAP (±) glial lineage in vivo (maintenance of immature state), and attenuated growth of xenografted GSC-derived tumors in mouse brains (tumorigenesis). The protein complex formation of MELK with the oncoprotein c-JUN was identified in GSCs but not the normal counterparts, NPCs. The c-JUN/MELK signaling is regulated by the bFGF/JNK pathway in a p53-dependent manner in GSCs. MELK immunohistochemistry is an indicator of the postsurgical prognosis of HGG patients, further supporting the clinical relevance of the MELK signaling in cancers. Targeting MELK appears to be a promising therapeutic approach for the treatment of devastating cancers including HGG. Given that MELK expression is highly variable between cases, it is mandated to determine which subset of HGG patients may meaningfully benefit from molecularly targeted therapies for MELK. Future studies would clarify the molecular signaling of MELK, c-JUN, and their associated pathways in cancers and CSCs.


We thank Drs. Harley Kornblum (UCLA) and Antonio E. Chiocca (Brigham and Women's Hospital, Harvard) for discussion and constructive criticisms, Dr. Paul Mischel (UCLA) for kindly providing U87MG and U87E6 cell lines, and Dr. Jennifer Zhang for kindly providing the JNK and c-JUN constructs for this study. We are thankful to Dr. Mark Drew for sharing the FACS facility and Dr. Xiaokui Mo for help on statistical analysis. We also thank Ms. Y. Kimberly for the expert editorial assistance and other members of our lab for their continuous help. This work was supported by the American Cancer Society (MRSG-08-108-01), the National Brain Tumor Foundation, Vincent J. Sgro/The American Brain Tumor Association, and the start-up fund from the Ohio State University, the Department of Neurological Surgery (to I.N.).


The authors indicate no potential conflicts of interest.