Glioblastomas (GBM) are lethal brain tumors where poor outcome is attributed to cellular heterogeneity, therapeutic resistance, and a highly infiltrative nature. These characteristics are preferentially linked to GBM cancer stem cells (GSC), but how GSCs maintain their stemness is incompletely understood and the subject of intense investigation. Here, we identify a novel signaling loop that induces and maintains GSCs consisting of an atypical metalloproteinase, ADAMDEC1, secreted by GSCs. ADAMDEC1 rapidly solubilizes FGF2 to stimulate FGFR1 expressed on GSCs. FGFR1 signaling induces upregulation of ZEB1 via ERK1/2 that regulates ADAMDEC1 expression through miR-203, creating a positive feedback loop. Genetic or pharmacologic targeting of components of this axis attenuates self-renewal and tumor growth. These findings reveal a new signaling axis for GSC maintenance and highlight ADAMDEC1 and FGFR1 as potential therapeutic targets in GBM.
Cancer stem cells (CSC) drive tumor growth in many cancers including GBM. We identified a novel sheddase, ADAMDEC1, which initiates an FGF autocrine loop to promote stemness in CSCs. This loop can be targeted to reduce GBM growth.
This article is highlighted in the In This Issue feature, p. 1469
Glioblastoma (GBM) is a uniformly fatal disease with a median survival of approximately 20 months after diagnosis (1–3). GBM represents a prototypical example of a highly heterogeneous tumor with multiple distinct, identifiable molecular subclasses within a single tumor (4, 5). Frequent tumor recurrence and poor outcome are thought to be a consequence of resident GBM cancer stem cell (GSC) populations resistant to current therapies (6–8). Thus, GSCs are a candidate population for tumor recurrence (9–12). A recent study demonstrates that GBM contains GSC populations that produce non-stem tumor cell (NSTC) progenies, with only GSCs capable of propagating tumor formation (13). How GSCs are maintained across the changing tumor microenvironment within hypoxic, vascular, or invasive niches (14) remains unclear.
Tumor progression is frequently linked to the secretion of metalloproteinases that enable tissue invasion and intravasation by cancer cells via extracellular matrix (ECM) degradation. This also causes a release of trophic factors to stimulate tumor growth, dispersal, and modulation of inflammatory responses. Members of the ADAM family of zinc-dependent proteinases contribute to GBM therapeutic resistance and invasiveness (15, 16), as well as to the regulation of GSCs (17–19). ADAMDEC1 is a soluble member of this family that is novel in mammals. It has restricted hydrolytic capacity due to a substitution of an active site residue (20) and selectively solubilizes growth factors from immobile precursor forms (21). Whether ADAMDEC1 solubilizes additional ligands, and whether this contributes to GBM growth and progression, has yet to be determined.
Trophic factors from the tumor microenvironment are essential to GBM growth and GSC maintenance. FGF2 is crucial in normal neural development and stem-cell function, and is a known oncogenic factor in GBM (22). FGF2 promotes glioma growth and vascularization (23) and GSC self-renewal (24). Nevertheless, how FGF2 specifically contributes to GSC functions is incompletely understood.
We now find ADAMDEC1 initiates FGF2 signaling through FGFR1 and ERK1/2 that then mediates GSC self-renewal and maintenance through induction of the stem-cell transcription factor ZEB1. ZEB1 additionally induces expression of ADAMDEC1 through miR-203, completing a positive feedback loop and contributing to GSC maintenance in the tumor microenvironment. Importantly, we show that pharmacologic intervention at the level of FGF2 can disrupt this loop and may constitute a translational therapeutic strategy.
Given the role for some of the ADAM family of metalloproteinases in GBM growth and progression (17, 19), we assessed the entire family for members that are elevated in GBM and may contribute to tumor growth. We interrogated the bioinformatics database GlioVis (25) for mRNA expression levels of ADAM family members across multiple GBM datasets (Fig. 1A). When normalized to brain expression levels, ADAMDEC1 emerged among the top proteases upregulated in GBM. Further investigation of these candidates using The Cancer Genome Atlas (TCGA) and LeeY datasets revealed that ADAMDEC1 was the only protease where increased expression associated with poorer prognosis of patients with GBM (Fig. 1B; Supplementary Fig. S1A and S1B). TCGA data also demonstrate that ADAMDEC1 mRNA levels increase with glioma grade (Fig. 1C). We performed IHC on patient specimens to test whether this extends to the protein level and found elevated ADAMDEC1 in GBM (Fig. 1D). Fluorescence immunostaining in patient-derived xenograft models revealed that ADAMDEC1 was expressed by tumor cells, and not host microglia (Fig. 1E).
We evaluated a role for ADAMDEC1 in maintaining stemness by first enriching GSC and NSTC populations by CD133 expression. Physical separation of the cell types showed increased cellular protein expression of ADAMDEC1in CD133+ cells compared with their CD133− counterparts across multiple patient-derived cell lines (Fig. 2A). Furthermore, testing conditioned media (CM) from GSC and NSTC cultures revealed that GSCs exclusively secrete ADAMDEC1. To evaluate a functional role for ADAMDEC1 in GSC maintenance, we knocked down ADAMDEC1 expression using lentiviral delivery of short hairpin RNA (shRNA) constructs. We found that ADAMDEC1 knockdown reduced expression of the stem cell–associated transcription factor SOX2 and increased expression of astrocytic differentiation marker GFAP (Fig. 2B). ADAMDEC1 knockdown also decreased in vitro sphere formation (Fig. 2C) and proliferation (Fig. 2D) of primary patient-derived GBM cells compared with nontargeted controls. To further scrutinize the relevance of ADAMDEC1, we orthotopically implanted ADAMDEC1 knockdown cells into immunocompromised mice, and observed a significant increase in survival of tumor-bearing mice compared with controls (Fig. 2E; Supplementary Fig. S2A). These data demonstrate ADAMDEC1 is a key regulator of GSCs.
As ADAMDEC1 is a sheddase capable of processing cytokines, we next determined whether ADAMDEC1 promoted GBM growth and progression via cytokine release. We treated GSCs and matching NSTCs with recombinant (r) ADAMDEC1 for 48 hours, after which CM was collected and multiple cytokines were evaluated using anti-cytokine bead-based flow cytometry (Fig. 2F). This experiment showed a dose-dependent increase of soluble FGF2 in the culture media with increasing amounts of rADAMDEC1. Pretreatment of cells with proteolytic enzymes blocked this effect, indicating that rADAMDEC1 released FGF2 from the ECM, rather than inducing FGF2 secretion from cells (Supplementary Fig. S2B). In contrast, GROα release was unaffected by rADAMDEC1. We evaluated FGF2 release over time using ELISA to find that GSC, but not NSTC, cultures released FGF2 within minutes following treatment with rADAMDEC1(Fig. 2G). Finally, ADAMDEC1 knockdown resulted in reduced activation of FGFR signaling, as demonstrated by Western blotting using a pan-phospho-FGFR antibody, whereas rADAMDEC1 treatment increased FGFR phosphorylation (Fig. 2H).
We next sought to define how FGF2 acts on GSCs by testing whether FGF2 correlated with the GSC-associated transcription factors ZEB1, SOX2, or OLIG2 (26). Using TCGA gene-expression data, we found each of these transcription factors correlated with FGF2 (Fig. 3A; Supplementary Fig. S3). To validate these correlations, we used patient-derived GBM cells that had been cultured in EGF only, and treated these cultures with recombinant FGF2. We found that rFGF2 dose-dependently induced expression of ZEB1, SOX2, and OLIG2 (Fig. 3B). In functional assays, rFGF2 treatment increased sphere formation (Fig. 3C). Conversely, a small-molecule inhibitor identified in a screen to block the interaction between FGF2 and FGF receptors [2-naphthalenesulfonic acid (NSC-65575); ref. 27] reduced clonogenicity and sphere formation (Fig. 3D; Supplementary Fig. S4A), but not viability (Supplementary Fig. S4B and S4C), of patient-derived GBM cells. Together, these results implicate FGF2 in GSC activation.
FGF2 can bind to FGFR1–4; therefore, we used a bioinformatic approach to determine whether any FGFRs show association with outcome of patients with GBM. We performed hierarchical cluster analysis of TCGA data using FGF2, FGFR1–4, ZEB1, SOX2, and OLIG2 as determinants. This revealed the existence of three principal clusters (Fig. 3E), with cluster 1 comprising approximately 30% of samples and showing association of FGF2, FGFR1–3, and stem-cell transcription factors (ZEB1, OLIG2, and SOX2). Cluster 2 was the smallest and showed low expression of FGF2, FGFR1–3, and stem-cell transcription factors. Conversely, this was the only cluster where FGFR4 was present. Cluster 3 constitutes most (>50%) of the samples, and shows high expression of ZEB1, SOX2, and OLIG2, but not of FGF2 or its receptors. Gene set enrichment analysis (GSEA) revealed an overlap of cluster 1 with signatures of the classic and mesenchymal molecular subclasses (4), whereas cluster 2 showed positive correlation with mesenchymal and negative correlation with all other signatures, and cluster 3 was enriched for neural and proneural subclass signatures (Supplementary Fig. S5).
Because data in the TCGA repository may be affected by the presence of nontumor cells in patient specimens, we validated our cluster analysis in the Human Glioblastoma Cell Culture (HGCC) dataset, which is derived from patient cell lines (28). Consistently, we found all three clusters represented. In this dataset, FGFR1 was most strongly associated with cluster 1, which was also most enriched for the mesenchymal subclass (Fig. 3E). This demonstrates an association of FGF2 and stemness-associated transcription factors in a significant fraction of GBM specimens and that considerable heterogeneity exists among these samples.
We used Kaplan–Meier analysis of the TCGA Glioblastoma Multiforme (provisional) dataset to further investigate which FGFR may be most relevant for patient outcome. This analysis confirmed that high expression of FGFR1 is associated with poor outcome, whereas for FGFR2 the relationship was inverse of this with augmented FGFR2 expression correlating to a better prognosis (Fig. 3F). No significant correlation was apparent for FGFR3, FGFR4 and patient survival. Of note, combinatorial analysis for FGFR1 and ADAMDEC1 showed a strong association with survival.
We next determined the FGFR expression in patient-derived GBM cells and found that FGFR1–3 were expressed, but that FGFR4 was consistently absent (Fig. 4A), as suggested by our cluster analyses (Fig. 3E). Using flow cytometry to quantify FGFR1–3 expression on GBM cells (Fig. 4B; Supplementary Fig. S6A and S6B), we found FGFR1 present on a small subset of cells (1%–5%), consistent with a stem-cell population. In contrast, FGFR2 was expressed on a large fraction (30%–50%) of cells. FGFR3 expression showed large variance, ranging from 5% to 25%. To test the functional relationship of different FGFR subtypes with GSCs, we performed lentiviral shRNA-mediated knockdown of FGFR1–3 in patient-derived GBM cells (Fig. 4C; Supplementary Fig. S6C). We found that only FGFR1 loss resulted in a decrease of sphere-forming capacity (Fig. 4D), which was accompanied by a decrease in ZEB1, SOX2, and OLIG2 expression (Fig. 4E) and a decrease in proliferation (Supplementary Fig. S6D). Loss of FGFR1 or ZEB1, but not of FGFR2 or FGFR3, abolished the FGF2-mediated increase in self-renewal (Supplementary Fig. S6E and S6F). Importantly, FGFR1 knockdown increased survival of tumor-bearing mice after orthotopic transplantation (Fig. 4F), and reduced tumorigenesis upon limiting-dilution orthotopic transplantation (Fig. 4G), revealing an approximate 6-fold enrichment of GSCs in control versus shFGFR1 cells.
To further substantiate this pivotal role of FGFR1 in GSCs, we performed rescue experiments. Targeted expression of full-length FGFR1 increased expression of ZEB1, SOX2, and OLIG2 in control cells, and restored ZEB1, SOX2, and OLIG2 expression in FGFR1 knockdown cells (Fig. 4H). Concomitantly, overexpression of FGFR1 increased sphere formation in control and FGFR1 knockdown cells (Fig. 4I). In contrast, ZEB1 knockdown negated the effects of FGFR1 overexpression on levels of SOX2 and OLIG2 expression, as well as on sphere formation, indicating that ZEB1 is downstream of FGFR1 and upstream of SOX2 and OLIG2.
As FGFR1 is functionally relevant for stem-cell maintenance in GBM, we hypothesized that this receptor may identify a GSC population. To test this hypothesis, we first quantified FGFR1 expression under culture conditions conducive to GSC maintenance (sphere cultures supplemented with mitogens) or differentiation (adherent cultures with growth-factor withdrawal, supplemented with serum or BMP4). We found that FGFR1 expression is high in GSC cultures, whereas FGFR1 levels decrease under differentiation conditions (Fig. 5A; Supplementary Fig. S7A). The expression of the stem-cell transcription factors ZEB1, SOX2, and OLIG2 followed the same pattern. In contrast, FGFR2 and FGFR3 expression increased in differentiation conditions (Fig. 5A). To test whether FGFR1 may be a marker of GSCs, we isolated FGFR1-expressing cells via flow cytometry (Fig. 5B). Using an independent FGFR1 antibody, we confirmed increased FGFR1 expression in the FGFR1+ fraction. Likewise, expression of ZEB1, SOX2, and OLIG2 was enriched in the FGFR1+ fraction (Fig. 5B). Next, we tested FGFR1+ cells in functional assays for stemness immediately after FACS. We found that FGFR1+ cells showed greater clonogenicity than FGFR1− cells in a limiting dilution experiment (Fig. 5C). Finally, we used limiting dilution orthotopic xenografts to determine the stem-cell frequency of FGFR1+ and FGFR1− cells (Fig. 5D). Extreme limiting dilution analysis (29) demonstrated an approximate 5-fold enrichment of GSCs in FGFR1+ cells, consistent with our results in limiting dilution of FGFR1-knockdown cells (Fig. 4G). Together, these data strongly support that (i) FGFR1 transduces FGF2 signal to induce GSC self-renewal, (ii) ZEB1, SOX2, and OLIG2 are downstream targets of FGF2–FGFR1 signaling, and (iii) FGFR1 is a surface marker of GSCs.
Our results indicate that GSCs secrete ADAMDEC1, causing release of FGF2 that induces GSC maintenance via FGFR1 signaling to ZEB1. We next sought to address how the expression of ADAMDEC1 in GSCs might be regulated. Speculating that ZEB1 may induce ADAMDEC1 expression, we tested whether ADAMDEC1 forms a positive feedback loop in GSCs with FGFR1 and ZEB1. Indeed, FGF2 treatment induced expression of ADAMDEC1 compared with EGF stimulation (Fig. 6A). We tested whether this effect on ADAMDEC1 expression was mediated by FGFR1, and found that FGFR1 knockdown reduced ADAMDEC1 expression, whereas targeted expression of FGFR1 increased ADAMDEC1 levels (Fig. 6B) and potentiated cell viability after ADAMDEC1 loss (Supplementary Fig. S7B). ZEB1 knockdown also decreased ADAMDEC1 protein expression (Fig. 6C), whereas overexpression of ZEB1 increased ADAMDEC1 levels (Fig. 6D) and rescued its expression after knockdown of ZEB1 (Fig. 6E) or FGFR1 (Fig. 6F). Although a recent study assessing ZEB1-binding sites in the genome of GSCs found no evidence for ZEB1 binding to the ADAMDEC1 promoter (30), ZEB1 frequently exerts its transcriptional effects through repression of miRNAs such as miR-200c or miR-203 (31), which we have previously described in GSCs (7). Using miRWalk (32), we identified predicted binding sites for miR-203 in the coding sequence of ADAMDEC1 (Fig. 6G), but not for miR-200c. We therefore overexpressed miR-203 and miR-200c in GBM cells, and found that only miR-203 decreased expression of ADAMDEC1 (Fig. 6H), whereas both miRNAs repressed ZEB1, as described previously (31). On the basis of these studies, we conclude that ZEB1 can regulate ADAMDEC1 through repression of miR-203.
Finally, we sought to identify how FGFR1 signals to ZEB1. A phospho-protein array revealed ERK1/2, STAT3, AKT, and p38 as candidate targets in our GBM cells after FGF2 stimulation (data not shown). Western blotting for phosphorylation of these candidate signaling molecules in control and FGFR1 knockdown cells demonstrated that only ERK1/2 phosphorylation was attenuated after depletion of FGFR1 (Fig. 7A). This indicates the ERK signaling cascade as effector pathway for mediating stemness upon FGFR1 activation. To validate this, we used an ERK1/2 inhibitor (SCH772984; ref. 33; Fig. 7B), and confirmed that this molecule could completely block ERK1/2 activation after treatment with FGF2 (Fig. 7C). Importantly, this inhibitor decreased expression of ZEB1 and ADAMDEC1 within 30 minutes of treatment (Fig. 7D), indicating that ERK1/2 may also directly regulate ADAMDEC1 expression.
Together, our data support the existence of a positive feedback loop that activates stemness in GBM (Fig. 7E). ADAMDEC1 is secreted by GSCs, and releases FGF2 from the ECM in the tumor microenvironment. FGF2 binds to FGFR1 on the surface of GSCs, where activation of this receptor leads to increased ERK1/2 signaling, resulting in expression of the downstream targets ADAMDEC1, ZEB1, SOX2, and OLIG2. ZEB1 mediates FGF2–FGFR1 effects on stemness and regulates expression of ADAMDEC1 through miR-203, thereby completing the loop.
The mechanisms by which GSCs maintain their stemness across different niches in the tumor landscape, including hypoxic, vascular, invasive niches, is incompletely understood and a matter of intense investigation (14). Our study identifies a new feedback loop that enables GSCs to access FGF2 in the tumor microenvironment through secretion of ADAMDEC1. ADAMDEC1 is a novel member of the ADAM family of metalloproteinases due to the absence of a transmembrane domain and altered catalytic domain, features shared by no other ADAM family member (20, 34, 35), that result in a unique secreted, soluble protease with distinct ligand specificity (20, 21). In addition, ADAMDEC1 was shown to modulate apical membrane extrusion of epithelial cells (36). Here, we demonstrate that ADAMDEC1 is highly expressed in glioma with enhanced expression correlating to increasing tumor grade, and ADAMDEC1 expression correlated with patient survival. Furthermore, we demonstrate that ADAMDEC1 was enriched in GSC populations and regulated cell proliferation, sphere formation, and tumorigenesis. ADAMDEC1 was recently shown to mediate the cleavage and release of active EGF, an important GSC trophic factor (21), although in our data FGF2 had a stronger effect on GSC stemness than EGF alone.
FGF2 binds to all four members of the FGF receptor family, with splice isoforms mediating binding affinity (37). FGF2–FGFR1 signaling promotes glioma growth and radioresistance (38, 39) and higher FGFR1 expression is associated with poor outcome for these tumors (40). A recent study indicated a link between FGFR1 and ZEB1 in GBM (41), but definitive evidence for FGFR1 regulating stemness including limiting-dilution in vivo transplantation was lacking. This is presented in our study. Conversely, FGFR2 expression is reduced in GBM compared with low-grade glioma (42), and higher FGFR2 levels are associated with improved survival (43). In contrast, a recent study found that FGFR2 signaling mediates GBM radioresistance (44). FGFR3 was the second-most differentially expressed gene between infiltrating and tumor-core GBM cells in a recent single-cell study (45), and FGFR3–TACC3 gene fusions have been identified as oncogenic drivers of GBM growth (46).
FGF2 activates FGFR1 on the GSC cell surface, which in turn induces expression of stem-cell transcription factors. Crucially, FGFR1 was the only FGFR that we found to be functionally associated with tumor cell sphere formation in culture, a hallmark of stemness. The relevance of FGFR1 for GSCs is further supported by increased survival after loss of FGFR1, as well as increased tumorigenicity of FGFR1+ cells. We and others have shown that ZEB1 is a key regulator of stemness, invasion, and chemoresistance in GBM (7, 12, 30), and consistently we find that ZEB1 mediates the stemness effects of FGF2–FGFR1 signaling. Singh and colleagues have recently demonstrated that ZEB1, SOX2, and OLIG2 form an autonomous transcriptional loop in GBM, and can regulate their expression reciprocally (26). Our data support these findings, as the FGF2–FGFR1 complex activates expression of SOX2 and OLIG2 in our system as well, and ZEB1, SOX2, and OLIG2 are linked in hierarchical cluster analysis of TCGA data. It is tempting to speculate due to the apparent codependency of the transcription factors ZEB1, SOX2, and OLIG2 that interference with any one member of this circuit may disrupt stemness in GSCs. Our data indicate this may be the case, as loss of ZEB1 was sufficient to abrogate the effects of FGF2 stimulation and increased FGFR1 expression on GSCs.
We additionally found that ZEB1 regulates expression of ADAMDEC1, creating a feedback loop that would enable GSCs to thrive in the central nervous system, as FGF2 is highly prevalent across the brain. We find that in the TCGA dataset, approximately 30% of patients with GBM are characterized by increased expression of FGF2, FGFRs, and ZEB1, and thus may benefit from therapeutic intervention aimed at disrupting this feedback loop. Importantly, we show that pharmacologic intervention of FGF2 binding to its cognate receptors can block stemness in GSCs, indicating that this may be an exploitable fulcrum for future therapies. This may be particularly relevant for those patients for whom cluster analysis shows higher expression of components of this feedback loop. Currently, clinical trials are under way for FGF receptor tyrosine kinase inhibitors (47). Although these compounds are selective for FGFR1–3 over FGFR4, no compounds exist at present that are selective for FGFR1 over FGFR2–4. Furthermore, trials for FGFR inhibitors in glioma recruit only patients with amplification or mutation of FGFR genes, which constitute only approximately 3% to 5% of all patients with GBM (47). Our data indicate that a much larger fraction of patients could benefit from anti-FGFR therapies by targeting treatment-resistant GSCs, should such trials be successful. Taken as a whole, our data identify a novel GSC signaling axis, ADAMDEC1–FGF2–FGFR1–ZEB1, and present a druggable, translational point of fragility.
Primary Human Glioblastoma Cells
Human GBM cells were cultured as described previously (7, 12). Briefly, for GSC sphere culture, 5 × 104 cells/mL were plated in N2 medium (Thermo Fisher Scientific) supplemented with 2% BSA (Thermo Fisher Scientific) containing 20 ng/mL recombinant human EGF (PeproTech). For some experiments, recombinant human FGF2 (PeproTech) was added at concentrations ranging from 5 ng/mL to 80 ng/mL. For differentiation experiments, spheres were dissociated into single cells and plated in DMEM/F12 (Thermo Fisher Scientific) supplemented with 10% FBS or 25 ng/mL recombinant human BMP4 (PeproTech).
Cell lines from the HGCC repository (28) were cultured in Neurobasal Medium (Thermo Fisher Scientific) supplemented with B27 (Thermo Fisher Scientific) and 20 ng/mL EGF and FGF2. Spheres were passaged when they reached an average diameter of 150 μm.
In some experiments, previously established GBM xenografts obtained from Duke University and the University of Florida were used and maintained as described previously (48, 49). Tissue was digested with papain (Worthington) as described previously (49), and dissociated cells were allowed to recover overnight prior to use. Thereafter, dissociated cells were sorted on the basis of CD133 expression using magnetic beads (Miltenyi Biotec). CD133-positive cancer stem cells were maintained in Neurobasal Medium (Life Technologies) supplemented with penicillin/streptomycin (50 U/mL final concentration), l-glutamine (2 mmol/L), B27 (Life Technologies), sodium pyruvate (1 mmol/L), EGF (20 ng/mL, R&D Systems), and FGF2 (20 ng/mL, R&D Systems). CD133− NSTCs were cultured in DMEM supplemented with 10% FBS and penicillin/streptomycin (50 U/mL).
Plasmids and Lentiviral Transduction
Control and knockdown ZEB1, FGFR1, FGFR2, and FGFR3 plasmids were purchased from Dharmacon. Plasmids for expression of miR-203 and miR-200c were a gift from Thomas Brabletz (University of Erlangen, Germany). Control and knockdown ADAMDEC1 plasmids were purchased from Sigma. Different clones for each shRNA plasmid were tested and the best knockdowns were selected to produce lentiviral particles. A plasmid for overexpression of FGFR1 was a gift from Dominic Esposito (Addgene plasmid #70367) and cloned into an expression vector (pHIV-IRES-mRFP) using the Gateway system (Invitrogen). Lentiviral particles were generated by cotransfecting HEK293T cells with second-generation packaging plasmids (psPAX2 and pMD2.G) using Lipofectamine 3000 (Invitrogen). Medium containing lentiviral particles was collected at 48 hours and 72 hours after transfection. Viral supernatants were combined and filtered with a 0.45-μm pore size filter, followed by ultracentrifugation at 185,000 rcf with a L8-70M Ultracentrifuge (Beckman). Pelleted viral particles were diluted in 200 μL of N2 medium. Concentrated viral particles were aliquoted and stored at −80°C.
For lentiviral transduction, 1 × 105 cells were preincubated for 1 to 2 hours in N2 medium without antibiotics and 1 μg/μL of polybrene (Santa Cruz Biotechnology) to increase transduction efficiency. Eighteen hours after transduction, medium was replaced with complete N2 medium and growth factors.
The small-molecule inhibitors of FGF2, NSC-47762, NSC-58057, NSC-65575, and NSC-65576 (27), were obtained from the Developmental Therapeutics Program, Division of Cancer Treatment and Diagnosis, NCI, NIH (Bethesda, MD).
Sphere-Forming Frequency Assays
Limiting dilution analysis was carried out, following 14-day incubation, in a 96-well format with 24 wells of each dilution; 1, 5, 10, and 20 cells/well. Cells were sorted using the BD FACSAria II Flow Cytometer. Limiting dilution plots and stem-cell frequencies were calculated using ELDA analysis (http://bioinf.wehi.edu.au/software/elda/index.html; ref. 29).
Sphere-forming assays were performed as described previously (7), with the following modifications: 100 to 200 single cells were seeded per well in 96-well plates, with 6 replicates per condition. Cells were cultured in 80 μL of N2 medium supplemented with 20 ng/mL of EGF and/or 30 ng/mL of FGF2 as specified in the text. The number of spheres with a diameter greater than 70 μm was quantified on a GelCount analyzer (Oxford Optronix) 5 days after seeding.
For colony-forming cell assays, human GBM cells were cultured in N2 supplemented with 20 ng/mL of EGF, 30 ng/mL FGF2, and collagen (Stem Cell Technologies) at a ratio of 1:3 (collagen:medium). Cells were supplemented with fresh growth factors twice a week. Colonies greater than 200 μm in diameter were counted two weeks after plating on a GelCount analyzer.
Flow Cytometry and Cell Sorting
Immunostaining and flow cytometry was performed as described previously (12). Data were acquired on a BD LSR Fortessa (BD Biosciences), using FACSDiva software (BD Biosciences) and analyzed with a FlowJo ver. 8.8.7 (Tree Star, Inc). For cell sorting, stained single-cell suspensions were purified on a BD FACSAria Fusion (BD Biosciences) and immediately used for downstream experiments.
Protein Isolation and Western Blotting
Protein isolation and quantification were performed as described previously (7, 21). For Western blotting, 5 to 20 μg of sample was mixed with an equal volume of 2× Laemmli buffer (Bio-Rad) containing β-mercaptoethanol, and denatured at 95°C for 5 minutes. Following separation on a Mini-Protean 4% to 15% Bis-Tris Gel (Bio-Rad), proteins were transferred onto polyvinylidene difluoride membranes using a Mini Trans-Blot Turbo Transfer System (Bio-Rad). Membranes were blocked and probed for primary and secondary antibodies as described previously (ref. 7; see Supplementary Table S1 for antibody information), and visualized using Clarity Western ECL substrate (Bio-Rad) on a ChemiDoc MP imaging system (Bio-Rad). Results were normalized to GAPDH or ACTNB as housekeeping proteins.
Female SCID mice ages 4 to 6 weeks were used for orthotopic xenografts. Animal care and handling and all procedures were performed according to NIH, the Federation of European Laboratory Animal Science Associations (FELASA), and institutional guidelines and approved by the UK home office (PPL30/3331) and the Institutional Animal Care and Use Committee of the Cleveland Clinic Foundation (protocol 2012-0752). Intracranial tumor transplants were performed as described previously (7, 50). Depending on the experiment, 1,000 to 50,000 cells were stereotactically implanted in 5 μL of media devoid of growth factors/supplements. Mice were maintained under isoflurane anesthesia during procedures. Mice were monitored daily for the development of neurologic signs and body-weight loss. Animals at endpoint were transcardially perfused using 2% paraformaldehyde, and the brains were removed for histology.
Tissue Preparation and Immunofluorescence
Post-fixed brains were cryoprotected, embedded, and sectioned as described previously (51). Sections were prepared for immunofluorescence or IHC staining, mounted, and coverslipped using standard protocols.
TCGA, NCI, Gravendeel, Murat, and Kamoun mRNA datasets were analyzed via the online tool GlioVis (http://gliovis.bioinfo.cnio.es/). Further analysis of survival correlations was performed using the Xena platform (https://www.biorxiv.org/content/early/2018/05/18/326470; https://xena.ucsc.edu/welcome-to-ucsc-xena/).
Hierarchical Cluster Analysis
HGCC cell line expression data (28) was clustered using the hierarchical clustering module (52) from GenePattern (https://cloud.genepattern.org; ref. 53) with Pearson correlation as distance measure, using row centering and normalization. Z-score values from the Glioblastoma Mutiforme (TCGA, Provisional) Tumor Samples with mRNA data (U133 microarray only; 528 samples) dataset was downloaded from cBioportal (54) and clustered with Pearson correlation as distance measure, using the same GenePattern module as before.
GSEA was performed on our identified clusters using the GSEA module from GenePattern (55); this was done using the Glioblastoma Mutiforme (TCGA, Provisional) Tumor Samples with mRNA data (U133 microarray only) dataset downloaded from the UCSC Xena Platform website. Clusters were compared one versus the rest using the Verhaak Glioblastoma Proneural, Classical, Neural, and Mesenchymal (4) gene sets from MsigDB (56). The results for the TCGA dataset here are in whole or part based upon data generated by the TCGA Research Network (http://cancergenome.nih.gov/).
Images were acquired using a Leica TCS-SP8-AOBS inverted confocal microscope (Leica Microsystems).
Statistical analyses were performed in GraphPad Prism version 7, using statistical tests as indicated in the text. In all analyses, P values < 0.05 were deemed significant.
Disclosure of Potential Conflicts of Interest
No potential conflicts of interest were disclosed.
Conception and design: J.S. Hale, D.J. Silver, T.M. McIntyre, K. Forsberg-Nilsson, J.D. Lathia, F.A. Siebzehnrubl
Development of methodology: J.S. Hale, D.J. Silver, R. Chen, T.M. McIntyre, G. Colombo, F.A. Siebzehnrubl
Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): A. Jimenez-Pascual, J.S. Hale, J. Pugh, D.J. Silver, D. Bayik, G. Roversi, T.J. Alban, S. Rao, T.M. McIntyre, K.O. Holmberg
Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): A. Jimenez-Pascual, J.S. Hale, D.J. Silver, D. Bayik, G. Roversi, K.O. Holmberg, K. Forsberg-Nilsson, J.D. Lathia, F.A. Siebzehnrubl
Writing, review, and/or revision of the manuscript: A. Jimenez-Pascual, J.S. Hale, D.J. Silver, D. Bayik, G. Roversi, S. Rao, T.M. McIntyre, G. Taraboletti, K. Forsberg-Nilsson, J.D. Lathia, F.A. Siebzehnrubl
Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): A. Jimenez-Pascual, J.S. Hale, R. Chen, J.D. Lathia
Study supervision: J.S. Hale, J.D. Lathia, F.A. Siebzehnrubl
Other (performing experiments): A. Kordowski
Funding was provided by a Medical Research Council grant (MR/S007709/1), Tenovus Cancer Care (TIG2015/L19), and a Cancer Research UK Cardiff Centre Development Fund Award (to F.A. Siebzehnrubl), the Lisa Dean Moseley Foundation (to J.D. Lathia and T.M. McIntyre), and the Cleveland Clinic and Case Comprehensive Cancer Center (to J.D. Lathia). D. Bayik was supported by Case Comprehensive Cancer Center T32CA059366. We thank members of the Lathia and Siebzehnrubl laboratories for insightful comments on the project and manuscript. We also thank Amanda Mendelsohn (Cleveland Clinic) for illustration assistance.
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.