Truncated Glioma-Associated Oncogene Homolog 1 (tGLI1) Mediates Mesenchymal Glioblastoma via Transcriptional Activation of CD44

Glioblastoma (GBM) is the most common and most lethal brain tumor in adults, accounting for 15% of all brain tumors. GBM prognosis remains poor, with a median survival of 14 months and less than 5% of patients surviving 5 years after diagnosis (1, 2). Extensive genomic analyses have divided GBMs into three to four distinct subtypes with slight variations (3, 4). The proneural and mesenchymal subtypes of GBM are commonly delineated across two classification systems. Mesenchymal GBMs tend to respond poorly to chemotherapy and radiation, and are associated with a worse prognosis (3–5).

Recent studies of GBM identified subpopulations of tumor cells exhibiting stem cell–like properties, including the ability to self-renew, persistently proliferate, and differentiate into multiple cell lineages (6). The subpopulation of stem-like cells, or glioma stem cells (GSC), is a key driver of tumor initiation, recurrence, and chemoresistance (7). Recently, patient-derived GSCs were identified to contain two distinct and mutually exclusive subtypes termed “proneural” and “mesenchymal” (8, 9). CD44 was identified as a key marker of mesenchymal GSCs (8). High CD44 expression in GBM is predictive of poorer overall survival and is associated with increased GBM invasion, proliferation, and chemoresistance (10). Mesenchymal GSCs were found to be more aggressive and radioresistant; therefore, understanding the signaling pathways controlling the mesenchymal GSC phenotype is the key to developing targeted treatments for mesenchymal GSCs in GBM.

The Sonic Hedgehog (SHH) pathway plays an essential role in GSC function (11). The glioma-associated oncogene homolog 1 (GLI1) family of zinc-finger transcription factors function as the terminal effectors of the SHH signaling pathway (12). Our laboratory discovered truncated GLI1 (tGLI1) as a novel alternatively spliced, gain-of-function variant of GLI1, with a 41 amino acid deletion spanning the entire exon 3 and part of exon 4, which retains all of the known functional domains of GLI1 (13). We reported that tGLI1 regulates known GLI1 target genes to a similar degree as wild-type GLI1 (13) but gained the ability to transcriptionally activate genes not regulated by GLI1, including CD24, HPA1, TEM7, VEGFA, VEGFC, and VEGFR2, thus promoting cancer cell growth, migration, invasion, and angiogenesis (13–17). We also showed that tGLI1 is detectable only in cell lines, patient-derived xenografts, and primary specimens of GBM, but is undetectable in normal brain tissue or other normal human tissues (18). The tumor-specific expression pattern of tGLI1 in invasive breast cancer was reported in our previous study (14). Other groups have confirmed our findings that tGLI1 is detectable in triple-negative breast cancer (19) and malignant gliomas (20). Metastatic hepatoma cells also express tGLI1 (21).

GLI1 has been shown to drive transcription of pluripotency markers in cancer stem cells; however, the role that gain-of-function tGLI1 plays in cancer stem cell function remains unknown (11, 22). It is also unknown whether tGLI1 promotes GBM progression using orthotopic xenograft mouse models. To address these gaps of knowledge, we initiated the current study that includes two different animal models and found that tGLI1 drives the formation of larger, more proliferative, and more highly vascularized tumors than GLI1 in orthotopic GBM and GSC xenograft mouse models. Mechanistic studies provided evidence linking tGLI1 to mesenchymal GBM and GSCs. Our results also established tGLI1 as a novel transcriptional activator of the CD44 gene, a known marker and regulator of mesenchymal GSCs and also other cancer types.

U373MG cells were obtained from ATCC and cultured according to their recommendations. Luciferase-expressing G48LL2 cells were developed by Dr. Waldemar Debinski (23). BTCOE 4810/4795 cell lines were developed and validated from patient tumors at Wake Forest Brain Tumor Center of Excellence, Winston-Salem, NC (24). Patient-derived GSCs (proneural GSC-11 and GSC-23; mesenchymal GSC-20 and GSC-28) were kind gifts from Drs. Erik Sulman and Krishna Bhat at The University of Texas MD Anderson Cancer Center (Houston, TX; ref. 8). GSCs were passaged under neurosphere-forming conditions in serum-free DMEM/F12 growth medium supplemented with B27, FGF, and EGF in order to preserve stem-like properties. All cells used were tested every 6 months for Mycoplasma; only cells free of Mycoplasma were used in the studies. Normal brain tissue microarray (BNC17011) and glioma tissue microarray (GL2083) were purchased from US Biomax. Additional GBM samples were from Wake Forest Brain Tumor Center of Excellence (24). We have obtained written informed consent from the patients; the studies were approved by Wake Forest Institutional Review Board and were conducted in accordance with recognized ethical guidelines.

The open reading frame for tGLI1 and GLI1 were cloned into the lentiviral expressing vector pCDH-CMV-MCS-EF1-RFP-Puro (CD516B-2) by System Biosciences. Plasmids were sequenced to confirm insertion of tGLI1 and GLI1 open reading frames. Lentiviral packaging was performed using a third-generation lentiviral pPACK packaging system from System Biosciences. The pPACK packaging plasmids along with the tGLI1 or GLI1 lentiviral expressing plasmid were cotransfected into 293TN cells for 48 to 96 hours, and the cell culture medium was subsequently collected. Viral particles were concentrated, titers measured, and used to infect G48LL2 and GSC-28 cells. Infected cells were treated with puromycin and FACS sorted for RFP-positive cells.

Female nude mice 6 to 8 weeks of age were used. Actively growing G48LL2 or GSC-28 cells were injected at a concentration of 1 to 5 × 10⁵ cells in 5 μL PBS into the right frontal lobe. For the GSC-28 in vivo study, immediately prior to inoculation into mouse brain, GSC-28 neurospheres were harvested and disassociated to single-cell suspension. Mice were anesthetized with a ketamine/xylazine mixture to the coronal suture through a scalp incision according to an approved Institutional Animal Care and Use Committee protocol. For bioluminescent imaging, xenograft-bearing mice were injected with d-luciferin intraperitoneally at 100 mg/kg body weight and then imaged weekly using PerkinElmer IVIS100 imager.

A tGLI1 activation signature (tGAS) was generated by averaging the median-centered expression of the six tGLI1 target genes (CD24, VEGFA, VEGFC, HPA1, TEM7, VEGFR2; refs. 13–17). Gene set enrichment analysis was performed by generating the Gene MatriX file (.gmx) by using published signatures for AngioMatrix (25), The Cancer Genome Atlas (TCGA) GBM subtypes (4), the Phillips GBM subtypes (3), and the GSC subtypes (8). The Gene Cluster Text file (.gct) was generated from the TCGA GBM dataset. The Categorical Class file (.cls) was generated based on the tGAS score or the GLI1 expression of patients in the TCGA GBM dataset. The number of permutations for gene set enrichment analysis (GSEA) was set to 1,000, and we used the TCGA gene list as the chip platform. For generation of heat maps, patients were divided by a high or low tGAS score and the genes included in the map were genes within the published signatures for the indicated GBM subtypes (3, 4). Heat maps were generated using Morpheus software developed by the Broad Institute, Cambridge, MA.

A tGLI1-targeting locked nucleic acid (LNA) oligonucleotide (oligo) and a nontargeting control oligo were custom made by Exiqon. Sequence for the control LNA oligo is C⁺T⁺G⁺T⁺C*T*T*C*A*G*T*T*C⁺A⁺A⁺C. BLAST analysis did not reveal binding of the control oligo to any genes. Sequence for the tGLI1-targeting LNA oligo is C⁺A⁺A⁺C⁺T*T*G*A*C*T*T*C*T⁺G⁺T⁺C.

Phosphorothioated bases are indicated by *, whereas LNA bases are labeled by ⁺. Both modifications make the oligos resistant to nuclease-mediated degradation. U373MG GBM cells with high endogenous tGLI1 expression were transfected for 48 hours using Lipofectamine 2000 (Invitrogen) and then seeded for the neurosphere formation assay. tGLI1 and GLI1 mRNA expression was determined by RT-qPCR.

Data are presented as mean ± SE. The Student t test, Pearson correlation, univariate/multivariate Cox proportional hazards tests, and one-way ANOVA were performed using SigmaPlot version 11.0.

Whether tGLI1 plays a role in promoting GBM intracranial growth has not been investigated. Herein, we generated isogenic cell lines from a low-passage G48LL2 human GBM cell line stably expressing empty vector, GLI1, or tGLI1 (Fig. 1A). Expression levels for GLI1 and tGLI1 in these cell lines are similar to those found in GBM specimens (Fig. 1A, bottom). The isogenic lines were implanted intracranially into female nude mice, and tumor growth was tracked by weekly bioluminescence imaging. Results showed that GBM cells expressing tGLI1 formed larger tumors (Fig. 1B). Representative images are shown in Fig. 1C and D. Analysis of brain sections using IHC indicated increased proliferation index (Ki-67 IHC) and increased microvessel density (CD31 IHC) in tGLI1-expressing tumors (Fig. 1E–H). Representative IHC images are shown in Fig. 1I. Using custom GLI1- and tGLI1-specific antibodies that we developed and validated (17), we further examined GBM specimens and normal brain tissues for GLI1 and tGLI1 expression, and found tGLI1 to be highly expressed in GBM samples but undetectable in normal brain, whereas GLI1 is expressed in both samples (Fig. 1J and K). Results in Fig. 1 demonstrate that tGLI1 promotes GBM tumor growth in the brain microenvironment and that tGLI1 is expressed in GBM, but not in normal brain.

We asked whether tGLI1 activity can be used as a prognostic indicator for GBM. To indicate tGLI1 activation status, we created tGAS using expression levels of its six target genes (13–17). We then analyzed a GEO dataset (GSE4290) consisting of mRNA expression profiles of 23 normal brains, 45 grade II gliomas, 31 grade III gliomas, and 81 grade IV gliomas (GBM) for tGAS, and found GBMs to have the highest tGAS scores (Fig. 2A). In contrast, GLI1 levels were not significantly different among four groups (Fig. 2B). Next, we examined whether tGAS was associated with clinical outcome for patients with GBM using a TCGA GBM database and univariate/multivariate analyses, and observed that high tGAS resulted in high hazard ratios (HR; Fig. 2C), suggesting that tGLI1 activation is independently associated with poor overall survival of patients with GBM. Kaplan–Meier analysis further showed that patients with GBM with high tGAS were associated with worse overall survival compared to patients with low tGAS (Fig. 2D). In contrast, GLI1 mRNA was not associated with overall survival (Fig. 2E).

Results in Fig. 1H indicated that tGLI1 promoted GBM vascularity. To confirm this observation, we correlated tGAS with vascularity markers, and found tGAS, but not GLI1, to significantly associate with CD31 and VE-cadherin (Fig. 2F–I). GSEA with the AngioMatrix signature that has been shown to be associated with tumor vascularity in GBM (25) further indicated that GBMs with high tGAS, but not high GLI1, had significant enrichment with the AngioMatrix signature (Fig. 2J–M). Results in Fig. 2 demonstrate that tGLI1, but not GLI1, is associated with poor overall survival and enhanced tumor angiogenesis in patients with GBM.

We examined whether the extent of tGLI1 activation differs among GBM subtypes. First, we divided the GBM cohort (TCGA) into two groups with high tGAS or low tGAS. We then used GSEA to determine the degrees of enrichment with established gene signatures for the four TCGA GBM subtypes, namely, mesenchymal, proneural, neural, and classical subtypes (4). As shown in Fig. 3A, GBM tumors with high tGAS were enriched for the TCGA mesenchymal gene signature, but not for the signatures for the other three subtypes. Interestingly, GBM tumors with low tGAS were significantly enriched for the TCGA classical signature, which is less aggressive than the mesenchymal subtype. Furthermore, we analyzed the same GBM cohort for the enrichment for another set of GBM subtype–specific gene signatures, namely, Phillips mesenchymal, proneural, and proliferation signatures (3). Consistent with the results of Fig. 3A, GBMs with high tGAS were enriched for the Phillips mesenchymal gene signature, but not the Phillips proneural or proliferation signatures (Fig. 3B). Mesenchymal GBMs had the highest tGAS scores among the four TCGA subtypes (Fig. 3C). In contrast, high GLI1 expression was not associated with either TCGA or Phillips mesenchymal subtype, but associated with the TCGA classical subtype (Fig. 3D). GBMs with low GLI1 were enriched with both TCGA proneural and Phillips proneural gene signatures (Fig. 3E). These findings indicate that tGLI1 activation is a hallmark of mesenchymal GBM.

Mesenchymal GBM is associated with poor patient survival and multidrug resistance—key clinical aspects of GBM thought to be driven by the cancer stem cell subpopulation (3, 5, 26). The role of GLI1 in GSC function has been studied; however, the role that tGLI1 plays in GSCs is unknown. We first compared monolayer culture with GSC-containing neurospheres for tGLI1 and GLI1 expression, and found tGLI1, but not GLI1, to be enriched in neurospheres (Fig. 4A). We further found that tGLI1-overexpressing cells formed significantly more neurospheres (Fig. 4B and C). Furthermore, we specifically knocked down tGLI1 expression in U373MG GBM cells with high endogenous tGLI1 expression to further determine the importance of tGLI1 in GBM. As shown in Fig. 4D, we designed and used a tGLI1-targeting LNA phosphorothioated oligo that has been chemically modified to resist nuclease-mediated degradation. As shown in Fig. 4D (left), the tGLI1-targeting LNA oligo specifically knocked down tGLI1 expression but had no effect on GLI1. Importantly, GBM cells with tGLI1 knockdown showed a reduced ability to form neurospheres compared to cells with the nontargeting control LNA oligo (Fig. 4D, right). Blast analysis did not reveal binding of the control oligos to any genes. These results further support our conclusion that endogenous tGLI1 is essential for the neurosphere-forming ability of GBM cells.

To identify the mechanism by which tGLI1 promotes a GBM stem cell phenotype, we determined the effects of tGLI1 on known stem cell–related genes and found that tGLI1-overexpressing GBM cells showed increased expression of CD44, along with a decrease in Sox2, Nanog, OCT4, and CD133 expression (Fig. 4E; Supplementary Table S1). Flow cytometry confirmed increased CD44⁺ cells in tGLI1-expressing GBM cells but no significant change in the CD133⁺ population (Fig. 4F). CD44 is regarded as the marker for the mesenchymal GSCs, whereas CD133 is the marker for the proneural GSCs (8). Using the chromatin immunopreciptation (ChIP) assay, we detected preferential binding of tGLI1, but not GLI1, to three regions of the CD44 gene promoter (Fig. 4G; Supplementary Fig. S1), which show no homology to consensus GLI1-binding sequences. Using the luciferase assay, we showed that tGLI1 transactivated the CD44 promoter in two SHH-stimulated GBM cell lines and HEK293 cells (Fig. 4H), and that SHH enhanced tGLI1-mediated activation of the CD44 gene promoter (Fig. 4I; Supplementary Table S2). Results in Fig. 4 indicate a novel, important role that tGLI1 plays in GBM stem–like cells and in the transcriptional activation of CD44.

Recent studies classified GSCs isolated from GBM specimens into the proneural or mesenchymal subtypes based on gene expression profiles (8, 9). CD44 was defined as the marker for the mesenchymal GSCs, whereas CD133 was for the proneural GSCs. Because tGLI1 binds to and transactivates the CD44 promoter and tGLI1 is preferentially activated in mesenchymal GBM, we determined whether tGLI1 expression is associated with the mesenchymal subtype of GSC. Here, we examined patient-derived GSCs, two mesenchymal and two proneural GSC neurospheres, and found the mesenchymal GSC neurospheres to express higher levels of tGLI1 and lower levels of GLI1 compared to proneural GSC neurospheres (Fig. 5A). We also observed that mesenchymal GSC neurospheres had higher tGAS than the proneural GSC neurospheres (Fig. 5B). We further analyzed a TCGA GBM dataset for the relationship between tGAS and GSC signatures, and observed that GBMs with high tGAS are enriched for the mesenchymal GSC signature, but not the proneural GSC signature (Fig. 5C, left). In contrast, GLI1 was not enriched for either GSC signature (Fig. 5C, right).

We found that tGLI1 ectopic expression enhanced the neurosphere-forming capability of both proneural and mesenchymal GSCs (Fig. 5D and E), and led to an increase in CD44 expression and a decrease in CD133 expression at both the mRNA and protein levels (Fig. 5F and G). Further analysis of the four different GSCs indicated a pattern where high tGLI1-expressing mesenchymal GSC neurospheres expressed higher levels of CD44 and lower levels of CD133 compared with low tGLI1-expressing proneural GSC neurospheres (Fig. 5H). Consistent with these observations, analysis of the TCGA GBM dataset revealed a positive correlation between tGAS and CD44, but not tGAS and CD133 (Fig. 5I). Collectively, results in Fig. 5 demonstrate that tGLI1 is preferentially expressed and activated in the mesenchymal GSC neurospheres over proneural GSC neurospheres, enhances expression of CD44, a mesenchymal GSC marker, and positively correlates with CD44 expression in a GBM cohort.

We have shown that tGLI1 ectopic expression promoted GBM growth in an orthotopic mouse model and that tGLI1 made GSCs form greater numbers of neurospheres. In light of these observations, we aimed to validate our GSC findings in vivo. We first generated three isogenic mesenchymal GSC-28 isogenic lines with stable expression of control, GLI1, or tGLI1 lentiviral vectors (Fig. 6A). Of note, GSC-28 neurospheres express high endogenous tGLI1 (see Fig. 5A). Expression levels for GLI1 and tGLI1 in these stable GSC lines are similar to those found in GBM specimens, as shown by IHC (Fig. 6A, bottom). GSC neurospheres were harvested and disassociated into single-cell suspension and were then implanted into the mouse brains; mice were imaged weekly. As shown in the growth curves and representative tumor images in Fig. 6B–D, increased tGLI1 expression rendered GSC-28 cells more aggressive in growths. Mice bearing tGLI1-expressing GSC-28 xenografts had a shortened survival time (Fig. 6E). IHC analyses further demonstrated that tGLI1-expressing GSC-28 tumors had the highest proliferative index (Ki-67 IHC) and microvessel density (CD31 IHC; Fig. 6F–J). In summary, these results indicate that tGLI1 renders GSCs more aggressive in growth and angiogenesis in the brain microenvironment.

We made the following important novel observations in this study: (i) tGLI1 promotes the neurosphere-forming ability of GBM and GSCs in vitro, and their intracranial growth and angiogenesis in vivo; (ii) tGLI1 is highly expressed in GBM but undetectable in normal brain, whereas GLI1 is equally expressed in both tissues; (iii) tGLI1 is predominantly activated in the mesenchymal subtype of GBM and GSCs that are more aggressive among different subtypes (4); (iv) patients with GBM with high tGLI1 activity in their tumors had shortened overall survival and increased tumor angiogenesis compared with those with low tGLI1 activity; and (v) tGLI1 functions as a transcriptional activator of CD44. By reporting these findings, our study advances the biological understanding of GBM and GSCs and transcriptional regulation of an important stem cell marker/mediator, CD44.

Our bioinformatics analyses revealed tGLI1 activation as a hallmark of mesenchymal GBM and GSCs, which is an important finding. We used two independent mesenchymal gene signatures and found both signatures to be highly enriched in GBMs and GSCs with high tGAS, an indicator for tGLI1 transcriptional activity. Of note, the classification of GBM into distinct subtypes was first reported by Phillips and colleagues in 2006, with three distinct subtypes (3), whereas a more recently study in 2010 by Verhaak and colleagues classified GBM into four distinct subtypes (4). Both studies identified a common subtype named the mesenchymal subtype according to the expression of genes associated with a mesenchymal cell phenotype. Notably, several earlier studies identified mediators and markers for the mesenchymal subtype of GBM and GSCs. For example, RTVP-1 was found to express at a higher level in mesenchymal GBM associated with tumor recurrence and poor clinical outcome (27). RTVP-1 overexpression induced mesenchymal differentiation of human neural stem cells, whereas silencing RTVP-1 inhibited the mesenchymal transformation and stemness of GSCs. An important future task is to explore the potential cross-talk between RTVP-1 and tGLI1 in regulating mesenchymal GSCs.

It is also important future task to investigate the potential interactions between tGLI1 and TNFα/NF-κB in light of the observations that both are enriched in CD44⁺ mesenchymal GSCs and that the proneural GSCs can undergo differentiation to a mesenchymal state in a TNFα/NF-κB–dependent manner (8, 28). Most recently, S100A4 was reported as a novel biomarker of GSCs that is enriched in cells with tumor-initiating and sphere-forming abilities; selective ablation of S100A4-expressing cells blocked tumor growth in vitro and in vivo (29). Whether tGLI1 crosstalks with S100A4, thereby promoting mesenchymal GSCs is unknown, but this line of future research is warranted.

Mesenchymal GBM and GSCs are not only more tumorigenic but also more resistant to radiation therapy (8). Given the ability of tGLI1 to promote mesenchymal GBM and GSCs, it is possible that tGLI1-expressing cells are more resistant to radiation therapy. Although the direct role of tGLI1 in radiation resistance has not been reported, it has been shown that tumor cells with hyperactive SHH-GLI1 signaling are more resistant to radiation therapy (30–32). Because tGLI1 functions as a gain-of-function GLI1 and tGLI1 has a higher propensity than GLI1 to promote GSCs in vitro and in vivo, we speculate that tGLI1⁺ GSCs are more resistant to radiation therapy compared to GLI1⁺ GSCs, which could be tested in future studies.

CD44 is regarded as a marker for cancer stem cells for a number of cancers, including breast cancer, pancreatic cancer, and GBM (8, 33, 34), and is associated with tumor initiation and progression (33, 35–37). However, emerging evidence suggests that CD44 contributes to the stem cell phenotype via various mechanisms, such as osteopontin signaling and promotion of HIF2α activity (38). However, transcriptional regulation of the CD44 gene is still not well understood. It has been reported that the EGR1 transcription factor can induce CD44 expression in antigen-stimulated B cells (39). NF-κB has been shown to upregulate CD44 expression in GBM (8) and breast cancer (40). NF-κB cooperates with AP-1 to bind to a cis-element of the CD44 promoter, leading to CD44 expression in breast cancer in a cell type–specific manner (41). Y-box binding protein-1 transcription factor induces CD44 expression in breast cancers (42). Surprisingly, the FOXO3 tumor suppressor has been shown to induce CD44 expression in pancreatic cancer (43). Our discovery that tGLI1 directly activates CD44 gene expression sheds important new light into the molecular mechanisms contributing to high CD44 expression in GSCs and possibly cancer stem cells in other tumor types.

Our present and previous studies uncovered that tGLI1 is expressed in a tumor-specific fashion in GBM and breast cancer (13–17). Expression of tGLI1 in malignant gliomas (20) and breast cancer (19) has been confirmed by other groups. Interestingly, a recent study detected tGLI1 in metastatic hepatoma cells (21). The mechanisms for the tumor-specific expression of tGLI1 are, however, still not elucidated. Since the splicing machinery is highly dysregulated in cancers (44), it is likely that the splicing factors that synthesize tGLI1 are aberrantly overexpressed in GBM, breast cancer, and metastatic hepatomas. Identification of these splicing factors constitutes an important task that could lead to strategies that inhibit tumor progression through inhibiting tGLI1 production.

Evidence from our laboratory and those of other groups suggests that tGLI1 may be regarded as a novel therapeutic target for multiple cancer types (13–17, 19, 21, 45). This notion is supported by the observations that tGLI1 is expressed tumor specifically and that tGLI1 plays an important role in tumor growth, angiogenesis and cancer stem cell renewal. We further speculate that targeting tGLI1 rather than GLI1 could minimize normal tissue toxicity because tGLI1 is only detected in cancer tissues, whereas GLI1 is expressed in both cancerous and normal tissues. Smoothened inhibitors can inhibit both tGLI1 and GLI1; however, their efficacy is only modest in GBM and breast cancer due to their inability to suppress noncanonical smoothened-independent activation of tGLI1 and GLI1 (46). Our study provides novel insights into the biology of GBM and GSCs, particularly those belonging to the mesenchymal subtype, defines tGLI1 as a novel mediator of GBM growth and GSC self-renewal, and establishes tGLI1 as a novel transcriptional activator of CD44.

No potential conflicts of interest were disclosed.

Conception and design: R.L. Carpenter, H.-W. Lo

Development of methodology: R.L. Carpenter, D. Zhu

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): T.K. Rimkus, R.L. Carpenter, S. Sirkisoon, D. Zhu, S.B. Tatter

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): T.K. Rimkus, R.L. Carpenter, D. Zhu, M.D. Chan, G.J. Lesser, W. Debinski

Writing, review, and/or revision of the manuscript: T.K. Rimkus, R.L. Carpenter, B.C. Pasche, M.D. Chan, G.J. Lesser, S.B. Tatter, K. Watabe, W. Debinski, H.-W. Lo

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): R.L. Carpenter, D. Zhu

Study supervision: R.L. Carpenter, H.-W. Lo

The authors thank Drs. Erik Sulman and Krishna Bhat at The University of Texas MD Anderson Cancer Center (Houston, TX) for gifting us the GSC cells. The authors also acknowledge the financial support from the NIH: R01NS087169 to H.-W. Lo; T32CA079448 to R.L. Carpenter; R01NS087169-3S1 to H.-W. Lo and S.R. Sirkisoon; P30CA012197 to B.C. Pasche; and from the U.S. Department of Defense: W81XWH-17-1-0044 to H.-W. Lo.

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.