GBM heterogeneity as a function of variable epidermal growth factor receptor variant III activity

Abnormal activation of the epidermal growth factor receptor (EGFR) due to a deletion of exons 2-7 of EGFR (EGFRvIII) is a common alteration in glioblastoma (GBM). While this alteration can drive gliomagenesis, tumors harboring EGFRvIII are heterogeneous. To investigate the role for EGFRvIII activation in tumor phenotype we used a neural progenitor cell-based murine model of GBM driven by EGFR signaling and generated tumor progenitor cells with high and low EGFRvIII activation, pEGFRHi and pEGFRLo. In vivo, ex vivo, and in vitro studies suggested a direct association between EGFRvIII activity and increased tumor cell proliferation, decreased tumor cell adhesion to the extracellular matrix, and altered progenitor cell phenotype. Time-lapse confocal imaging of tumor cells in brain slice cultures demonstrated blood vessel co-option by tumor cells and highlighted differences in invasive pattern. Inhibition of EGFR signaling in pEGFRHi promoted cell differentiation and increased cell-matrix adhesion. Conversely, increased EGFRvIII activation in pEGFRLo reduced cell-matrix adhesion. Our study using a murine model for GBM driven by a single genetic driver, suggests differences in EGFR activation contribute to tumor heterogeneity and aggressiveness.


INTRODUCTION
Glioblastoma (GBM) is the most common primary malignant brain tumor in adults and is characterized by intertumoral and intratumoral heterogeneity [1][2][3][4], including differences in genomic and epigenomic alterations [5,6], invasive phenotype [7], and RTK signaling pathway activity [8]. Although gene expression analysis of bulk tumor tissue suggests tumors can be divided into discrete transcriptionally-defined subgroups [6,9], the analysis of multiple samples from the same tumor frequently reveals striking intratumoral diversity in gene copy number, mutations, and stem cell signature [2][3][4]10]. GBM heterogeneity endows this cancer with an ability to adapt to and evade the anti-tumor effects of nearly any therapy. Thus, despite the development and Research Paper www.impactjournals.com/oncotarget application of targeted therapeutics based on an improved understanding of GBM molecular characteristics [11][12][13][14][15], tumor heterogeneity continues to confound efforts for improving GBM treatment outcomes.
The most common RTK alteration in GBM is of the epidermal growth factor receptor (EGFR) and altered EGFR signaling is considered a driver of malignant characteristics in up to 45% of tumors. EGFR is frequently amplified in GBM [11] and approximately 20-30% of amplified tumors express a constitutively active variant of EGFR, EGFRvIII or EGFR*, which contains a deletion of exons 2-7 [11,[16][17][18][19][20]. Several cancers are driven by altered EGFR signaling, yet associations between total receptor levels, activation levels, and clinical outcome have been mixed [21][22][23]. In GBM, increased expression of EGFR or EGFRvIII are also not clearly associated with decreased overall survival [18,[24][25][26][27] and a link between EGFR activation and outcome in GBM is yet to be determined.
EGFR signaling is important to the normal biology of neural stem cells (NSCs), and promotes the transition from a quiescent to an activated state [28]. In the murine postnatal brain, EGFR expression in NSCs is reduced upon neuronal lineage commitment. Aberrant activation of EGFR in neural and glial progenitor cells, either via genetic manipulations or by treatment with exogenous ligand, induces aberrant proliferation of cells with glial features, and can lead to the formation of tumor-like lesions [29][30][31][32].
The constitutive activation of EGFR conferred by EGFRvIII results in altered signaling relative to ligandmediated activation of the full-length receptor with respect to activated downstream pathways, intracellular signaling molecules, and epigenetic regulators [33][34][35]. Complex cross-talk between signaling pathways in human GBM and GBM cell lines can complicate the investigation of activation of a single RTK. Using an immune competent murine model for glioma, based on the genetic manipulation of neural progenitor cells and driven by EGFRvIII, we identify the activity level of EGFRvIII as a determinant of tumor cell aggressiveness.

Generating tumor progenitor cells with differences in EGFRvIII activation
To generate tumor progenitor cell lines exhibiting divergent EGFRvIII activation, we used a model for high grade astrocytoma developed in mice [36]. In this model neural progenitor cells (NPCs) are isolated from the subventricular zone of Ink4a/Arf null mice and transduced with the human EGFRvIII. This and similar models making use of neural stem cells and EGFRvIII overexpression are known to produce tumors heterogeneous in growth and survival [37,38]. Lysates from EGFRvIII-transduced cells derived from the same pool of parental NPCs revealed two lines with divergent EGFRvIII activity as determined by Western blotting, denoted line A and line B ( Figure  1A). While both cell lines had increased EGFR activity relative to parental NPCs, only line A with high EGFRvIII activity had a decrease in expression of the full-length endogenous EGFRwt. To generate tumor progenitor cells, the two EGFRvIII-transduced lines were transplanted intracranially into syngeneic mice and the resulting tumors were propagated as tumorspheres in minimal essential media in vitro as pEGFR Hi and pEGFR Lo , respectively ( Figure 1A).
Following in vivo passage and selection of tumor progenitor cells, the relative differences in EGFR activity were preserved. pEGFR Hi had increased abundance of phosphorylated EGFRvIII, as evidenced at Y1173 and Y1068 tyrosine residues compared to pEGFR Lo (6.5-and 2.86-fold p-EGFR/total, respectively; p<0.005, n=4). Differences in EGFRvIII activation were not due to differences in total expression of EGFRvIII or differences in cell surface expression of EGFRvIII ( Figure 1B and 1C). pEGFR Hi also had increased STAT3 activation (6.89fold; p>0.001, n=4), based on Y705 tyrosine residue phosphorylation ( Figure 1B).

EGFRvIII activity associated with more aggressive tumors and gene expression signature
Orthotopic transplants of pEGFR Hi and pEGFR Lo revealed significant differences in tumor growth in vivo. While both groups developed signs of tumor by 42 days post transplant, the median survival of mice harboring pEGFR Hi cells was 21 days versus 31 days for pEGFR Lo cells (p<0.0001, n=6) ( Figure 2A). The more aggressive pEGFR Hi resulted in greater tumor burden than pEGFR Lo even at comparable median survival time points. Enumeration of ZsGreen tagged tumor cells by fluorescence-activated cell sorting (FACS), revealed an 8.7-fold increase in pEGFR Hi versus pEGFR Lo tumor cells at respective median survival time points (p=0.001, n=6) ( Figure 2B) and increased tumor area (p<0.05) ( Figure  2C).
To investigate in vivo gene expression differences associated with increased EGFRvIII activity, fluorescentlytagged tumor cells were isolated by FACS from tumors and differential gene expression was determined by microarray analysis at median survival ±1 day ( Figure 2D). KEGG pathway annotation of differentially expressed genes identified enrichment of processes related to proliferation and DNA repair in pEGFR Hi tumors. Conversely, processes associated with cell-matrix interactions and the glycocalyx were enriched in pEGFR Lo tumor cells, including 'cell adhesion molecules', 'chondroitin sulfate biosynthesis', and 'heparan sulfate biosynthesis'. In addition, processes associated with a wider range of differentiation phenotypes, such as 'axon guidance' and 'long-term potentiation', were also enriched in pEGFR Lo ( Figure 2E).
Increased in vivo tumor burden, as defined by increased tumor cell number ( Figure 2B), increased tumor area ( Figure 2C), and enrichment of genes involved in DNA replication ( Figure 2E) suggested increased proliferative capacity in pEGFR Hi versus pEGFR Lo cells. In vitro, pEGFR Hi had increased incorporation of EdU (p<0.01) and greatly reduced doubling time compared to pEGFR Lo ( Figure 2F and 2G), reflecting an increased rate of proliferation. No differences in apoptosis were appreciated based on immunohistochemical analysis of cleaved caspase 3 in tumors (data not shown).
While EGFRvIII lacks a ligand-binding domain and can be activated independent of EGF ligand, EGFRvIII can act as a substrate for EGF-activated fulllength EGFR [39]. To assess the role for EGF ligandmediated signaling in the two lines, cell growth was compared in the absence of exogenous EGF ligand. While pEGFR Hi maintained a high rate of proliferation, the growth of pEGFR Lo was decreased three-fold in the absence of EGF ( Figure 2H). These data suggest that the higher expression levels of full-length EGFR in pEGFR Lo may confer some degree of ligand-mediated signaling in these cells that is not appreciated in pEGFR Hi . progenitor cells with divergent EGFR activation from Ink4a/Arf-null neural progenitor cells and EGFRvIII transduction. After transduction, EGFRvIII-transduced cells were assayed for EGFR expression and activity level using western blot and were subsequently passaged through syngeneic mice as intracranial tumors to create the pEGFR Hi and pEGFR Lo lines. B. Western blot analysis of pEGFR Hi (Hi) and pEGFR Lo (Lo) demonstrating differences in EGFRvIII phosphorylation at pY1068 and pY1173 and Stat3 phosphorylation at pY705. C. Analysis of EGFRvIII surface levels on pEGFR Hi (Hi) and pEGFR Lo (Lo) cells using flow cytometry. Solid histograms represent negative controls. Data representative of triplicate experiments. www.impactjournals.com/oncotarget

Blocked differentiation in EGFRvIII highly activated tumor progenitor cells
As compared to pEGFR Hi cells, pEGFR Lo cells had enriched expression of pathways associated with a more differentiated phenotype. Given that activation of EGFR can drive proliferation of a neural progenitor cells we examined the expression of a panel of genes associated with neural stem and progenitor cell maturation.
Real-time quantitative PCR demonstrated increased expression of the immature stem cell markers Prom1 and Id1 in pEGFR Hi ( Figure 3A). Conversely, expression of genes associated with cellular differentiation, such as Tubb3 and Ascl1 (Mash1), were upregulated more than 40-fold in pEGFR Lo . Olig2 and Dlx2 transcripts, genes commonly expressed by progenitor cells, were also more highly expressed in pEGFR Lo compared to pEGFR Hi ( Figure 3A). Consistent with a more undifferentiated phenotype, more than 95% of EGFRvIII-activated pEGFR Hi tumor cells expressed Prominin-1 on their cell surface ( Figure 3B). In contrast, less than 2% of pEGFR Lo tumor cells expressed Prominin-1 ( Figure 3B).
In cell differentiation assays increased EGFRvIII activity in pEGFR Hi cells was associated with reduced differentiation potential relative to in pEGFR Lo cells. The pEGFR Hi cells retained expression of nestin, a marker of immature neural stem and progenitor cells, and had decreased expression of glial (O4 and GFAP) and neuronal lineage markers (TUJ1/βIII-Tubulin) as compared to pEGFR Lo cells ( Figure 3C). Even after 7 days of culture under differentiation conditions pEGFR Hi cells appeared largely undifferentiated, and displayed a small round cellappearance with robust expression of nestin ( Figure 3D). pEGFR Lo cultures, in contrast, contained a significant population of GFAP-expressing cells, indicative of glial differentiation, and exhibited a bipolar or multipolar morphology ( Figure 3E).
To determine whether the lower differentiation potential of pEGFR Hi cells was due to high EGFR activation, we examined differentiation in the presence or absence of the EGFR tyrosine kinase inhibitor erlotinib. After 7 days of culture, pEGFR Hi cells with EGFR inhibition exhibited increased GFAP expression and altered cell morphology, with more elongate cells, as compared to control-treated pEGFR Hi cells, which retained their rounded, clumped morphology ( Figure  3F-3H).

Multicellular co-option of the vasculature in EGFRvIII activated tumors
Both pEGFR Hi and pEGFR Lo tumors were highly infiltrative with involvement of the striatum, cerebral white matter, and cortex ( Figure 4A and 4B), yet the pattern of tumor infiltration was different. At the invasive front, pEGFR Hi tumors formed dense multicellular invasive cords of tumor cells ( Figure 4C and 4E). In contrast, pEGFR Lo tumors appeared diffusely infiltrative primarily as single cells ( Figure 4D and 4F). The pattern of invasion was highly stable as 78% (18/23) of pEGFR Hi and 84% (16/19) of pEGFR Lo secondary intracranial tumors retained the pattern ( Figure 4G). At the invasive tumor front, colabeling of tumor cells and blood vessels revealed blood vessel co-option in both pEGFR Hi and pEGFR Lo tumors ( Figure 4H and 4I). Multicellular clusters of pEGFR Hi tumor cells were often identified in close association with the vasculature.
To model these invasive behaviors ex vivo we injected tagged tumor cells into freshly isolated brain slices and performed confocal imaging of fluorescentlytagged tumor cells and lectin-labeled blood vessels. At 24 hours both pEGFR Hi and pEGFR Lo cells readily invaded into brain slices and showed preferential cooption of blood vessels ( Figure 5A and 5B). Similar to their respective in vivo patterns of invasion, pEGFR Hi cells generated multicellular clusters ( Figure 5A) while pEGFR Lo cells tended to extend as single cells along the vasculature ( Figure 5B). Quantification demonstrated nearly twice as many blood vessel-associated tumor cells per 100 μm length of vessel in pEGFR Hi compared to pEGFR Lo ( Figure 5C). To distinguish multicellular invasion from local tumor cell proliferation in the highly proliferative pEGFR Hi cells, we performed time-lapse confocal imaging of tumor cells in brain slice cultures over 24 hours. Both pEGFR Hi and pEGFR Lo tumor cells invaded along blood vessels, but only pEGFR Hi tumor cells invaded as multicellular groups of cells ( Figure 5D, Suppl video 1 (pEGFR Hi ) and 2 (pEGFR Lo )).

Decreased cell adhesion and invasion in EGFRvIII highly activated tumor cells
Differences in in vivo and ex vivo invasion exhibited by pEGFR Hi and pEGFR Lo tumor cells suggested EGFRvIII activity may influence cell-ECM interactions. Tumor cells plated on cell culture plates pre-coated with laminin, collagen IV, or fibronectin revealed a marked difference in cell adhesion between pEGFR Hi and pEGFR Lo . Increased EGFRvIII activity was associated with a reduced ability to adhere to all substrates ( Figure 6A-6C). To assess how decreased cell adhesion may influence invasion in a 3D matrix we utilized an in vitro matrigel invasion assay in which the primary components are laminin, collagen IV, and enactin/nidogen-1 [40,41]. While pEGFR Lo tumor cells were highly invasive even by 16 hours, pEGFR Hi showed very limited invasion even after 24 hours ( Figure 6D-6E). Thus, the increased EGFRvIII activity in pEGFR Hi was associated with reduced adhesion and reduced invasion in vitro.
To investigate whether reduced adhesion was due to increased EGFRvIII activity, cell adhesion assays were performed while modulating EGFR activity. Reduction of EGFR activity in pEGFR Hi cells, using a 2-hour treatment with erlotinib, demonstrated a 2.35-fold increase in adhesion to collagen IV (p<0.01, n=6) ( Figure 7A-7B). In contrast, adhesion was decreased when EGFR activity was acutely increased in pEGFR Lo cells using the phosphatase inhibitor sodium orthovanadate (NaOV) (Figure 7C-7D). Importantly, pEGFR Lo cell adhesion could be restored upon inhibition of EGFR in NaOV-treated cells ( Figure  7E). This increase in pEGFR Lo adhesion was accompanied www.impactjournals.com/oncotarget

DISCUSSION
GBM are characterized by their heterogeneity, including tumor genetics [6], invasive phenotype [7], and RTK signaling activation [8]. This heterogeneity is an important prognostic factor in GBM and presents a great therapeutic challenge [42]. Murine models for GBM provide a controlled genetic background to study individual oncogenic events [43], however, there has not been a major focus on modeling and studying the impact of differences in RTK activity levels [37,38,44,45]. The potential importance of RTK activity on disease is highlighted by elegant studies examining the cross-talk between constitutive versus ligand-dependent EGFR signaling. Ligand-induced activation of endogenous EGFR phosphorylates EGFRvIII and promotes STAT3 activation, but can also disrupt EGFRvIII-induced NFkB signaling [39,46]. Moreover, ligand-independent activation of EGFR results in non-canonical EGFR signaling through the formation of an EGFR, TBK1, and IRF3 complex, downstream effects of which include reduced sensitivity to chemotherapy [47]. Using a murine model for glioma, we demonstrate EGFRvIII activity-dependent regulation of cell differentiation and cell adhesion and associate these differences with altered animal survival and in vivo tumor growth and invasion. Our results suggest that the activity of the EGFR signaling pathway, in addition to the specific genetic alteration, is an important determinant of tumor aggressiveness.
The more aggressive pEGFR Hi tumors exhibited both increased proliferation and a more cohesive pattern of invasion as compared to pEGFR Lo tumors. While increased proliferation is a known driver of malignancy, differences in invasive pattern may also contribute. Indeed, tumor progression following antiangiogenic therapy has been associated with a perivascular pro-invasive phenotype [48,49].
Diffuse invasion of tumors cells is a characteristic feature of human GBM and contributes to poor prognosis, yet the pattern of invasion is heterogeneous [7] suggesting the contribution of multiple factors. Even within a single patient several patterns can often be observed and differences in cell adhesion, proteolysis, and migration may all contribute. Blood vessel co-option was prominent in pEGFR Hi and pEGFR Lo tumors, yet pEGFR Hi invaded as multicellular clusters while the less aggressive pEGFR Lo invaded as single cells. Single cell invasion of pEGFR Lo was associated with increased in vitro adhesion and increased in vivo expression of genes implicated in cellmatrix interactions, including integrins and proteoglycans. In pEGFR Hi , inhibition of EGFR increased cell-matrix adhesion and promoted a more differentiated phenotype. While the identification of factors downstream of EGFR   that mediate cohesive invasion is a focus of future studies, it is notable that by gene expression array fibronectin (FN1) expression was elevated 5.8-fold in pEGFR Hi tumors. FN1 has previously been implicated in cohesive invasion in glioma and gene knockdown increased cellmatrix adhesion [50]. In addition, differences in protease activity or activation of other receptor tyrosine kinases such as MET can have profound influences on tumor cell invasion [49,51].
Abnormal activation of EGFR in neural stem and progenitor cells promotes proliferation and is detrimental to neuronal cell maturation [29][30][31][32]. pEGFR Hi exhibited profound alterations in progenitor cell phenotype with high expression of Prominin-1 and decreased ability to differentiate, which was at least partially dependent on EGFR activation. In immortalized neural stem cells, EGFRvIII expression has previously been associated with decreased differentiation, increased focal adhesion disassembly, and perivascular invasion in vivo [52]. In addition, a CD133expressing subset of EGFRvIII + glioma cells was shown to have increased self-renewal and tumorigenicity [53]. While several factors, including homozygous deletion of Ink4a/Arf -/- [37], likely contribute to the overall modest degree of cell differentiation that we observed, our data strengthen the link between abnormal EGFR signaling pathway activity and aberrations in cell differentiation.
EGFRvIII is known to be expressed at varying levels within a tumor [54,55] and the regulation of extrachromosomal EGFRvIII DNA is a major mechanism of treatment resistance [56]. Our data, in a genetically controlled background, suggest that the level of EGFR activation, not just the total protein levels, is an important determinant of tumor cell phenotype. Given the complexity of EGFR signaling and the multiple feedback loops and regulatory mechanisms [57], several factors may contribute to differences in EGFR activity and ultimately cellular behavior. These include differences in phosphatase activity, cell differentiation [58], receptor endocytosis, and the activity of other receptor tyrosine kinases [57]. Our data emphasize the importance of signaling pathway activity, in addition to the genetic status of EGFR, in tumor cell behavior and aggressiveness.

Cell culture conditions and reagents
Murine neural stem cell and tumorspheres were cultured as described previously [36,59]. Three dimensional invasion assays were performed as previously described [60]. EGF-dependent growth was calculated over three consecutive passages comparing cells cultured in defined neurosphere media including fibroblast growth factor in the presence and absence of added EGF (20 ng/ ml). For adhesion assays, wells of a 96-well plate were coated with 2 μg/cm 2 Laminin (Sigma L2020), 10 μg/ cm 2 Collagen IV (Sigma C5533) or 5 μg/cm 2 Fibronectin (Sigma F1141). 1-5x10 4 cells were plated per well and incubated for 2 hours. For assays involving erlotinib and sodium orthovanadate (NaOV) drug treatments, 1 μM erlotinib, 50 μM NaOV, or 500 μM NaOV was added to cell suspension for the 2 hour duration of the assay. For adhesion assays using co-treatment of erlotinib and NaOV, 1 μM erlotinib and 250 μM NaOV was used. Protein lysates were prepared as previously described [36]. Protein lysates for studying effects of erlotinib and NaOV on EGFR phosphorylation were acquired after 2 hours of incubation in sphere suspension cultures. Images acquired for invasion and adhesion assays were acquired using a DMI IL LED Leica microscope (Leica, Germany) and quantified using ImageJ 1.48v (http://imagej.nih.gov/ ij/). All antibodies used are described in Table 1.

Tumor cell isolation and RNA isolation
Tumor bearing brains were harvested from six mice with each tumor type and the cerebellum and olfactory bulbs were removed. The brains were diced finely with a razor, dissociated using Worthington's Papain Dissociation System (LK003150) and passed through a 70um filter to obtain a single cell suspension. To remove myelin and red blood cells the cell suspension was re-suspended in 35% Percoll TM (GE Healthcare) and underlaid with 70% Percoll TM . Following a 30 minute centrifugation purified cells were removed from the interface between the 35% and 70% layers and counted for viability. ZsGreen+ tumor cells were isolated by FACS using a BD FACSAria III into TRIzol Reagent (Ambion Life Technologies, Japan), using Dapi to exclude dead cells. Sorted cells were re-suspended in TRIzol Reagent (Ambion Life Technologies, Japan) and RNA was purified using the RNeasy Mini Kit (Qiagen, Germany) according to the manufacturers protocol.

In vivo transplantation
Dissociated cells were transplanted as previously described previously [36,59]. FVB mice (Charles River Laboratories, Wilmington, MA) were used for tumor microarray profiling experiments.

Immunohistochemistry
Murine tumor tissue for immunohistochemical analysis was collected at first sign of tumor growth by perfusing mice with 4% paraformaldehyde and postfixed overnight in 4% PFA. Following postfixation tissue was rinsed in PBS and stored in 70% ethanol until further processing. Murine tumor paraffin embedded sections were immunostained for hEGFR (Dako) to label EGFRvIII-expressing tumor cells. Immunohistochemistry was performed according to standard protocols on the Ventana Medical Systems Benchmark XT (Ventana Medical Systems Inc. Tucson, AR, USA). Histological analysis was performed on hematoxylin and eosin stained sections. Tumor area was analyzed on hEGFR-labeled sections and quantified in ImageJ 1.48v (http://imagej. nih.gov/ij/).

Ex vivo invasion and time-lapse imaging
Brains from P7-P21 FVB mice were dissected, put in cold aCSF, sectioned using a HM650V vibratome (Thermo Fisher scientific, San Jose, CA) at thickness of 190 μm. Slices were incubated in RT aCSF for 30 mins prior to transfer to wells with a MilliCell-CM 40 μm membranes (Millipore, Bedford, MA). For live imaging, ZsGreen expressing cells were injected underneath the slice using a Hamilton syringe (Hamilton, Reno, NV), for confocal analysis cells were deposited on top of slices using P10 pipette. For live imaging slices were imaged after 2 hours of incubation at 37°C, using a Zeiss Observer Z1 equipped with a Zeiss Observer XL stage incubator (37°C, 5% CO 2 , humidified air) (Carl Zeiss, Germany) and a Yokogawa spinning disc scanning unit (Yokogawa electric, Japan). Movies assembled using Zeiss imaging software (Carl Zeiss) at a single plane from images acquired over 6-8 hours. For confocal analysis, slices were fixed using 4% paraformaldehyde in phosphatebuffer saline 24 hours after tumor cell deposition, for 4-6 hours. A proteinase K antigen retrieval step was used to image hEGFR labeled tumor cells, followed by 1 hour of block/permabilization, 24-72 hours incubation in primary anti-human EGFR antibody antibody (Dako, Denmark, Clone H11) at 4°C, and 2 hours incubation in secondary antibodies. Slices were washed in PBS 3-5 times after each antibody incubation step. Slices were mounted onto Superfrost plus glass slides and mounted using Dako

Differentiation assay
Cells were prepared as a single cell suspension in normal culture media, 500-1000 cells per well were added to a laminin/poly-d-lysine coated 8-well culture slide (BD Biocoat 354688, BD biosciences) in 300 μL growth media. The culture slides were incubated overnight and the media carefully removed and replaced with 300 μL of neurobasal (Thermo Fisher) + 2% FBS to induce differentiation. For studies using erlotinib, erlotinib (1 μM) or DMSO was included in differentiation media. Culture slides were incubated for 3 to 7 days, then fixed in 4% PFA. After immunolabeling, six images per well were taken on a Zeiss Axioimager M1 or a Zeiss LSM780 confocal microscope (Carl Zeiss), counted for positivity in ImageJ 1.48v and averaged.

Visualization of f-actin cytoskeleton
pEGFR Lo cells were fixed in 4% PFA 2 hours after seeding cells onto laminin/poly-d-lysine coated 8-well culture slides (BD biosciences). F-actin was visualized using Alexa fluor-546 phalloidin according to the manufacturer's instructions (Thermo Fisher). Cells were treated with DMSO, 250 μM NaOV, or 250 μM NaOV + 1 μM erlotinib for the 2 hour duration of the assay.

Flow cytometry
Cell surface PE-conjugated mouse Prom1 antibody (Miltenyi Biotec, France, Clone MB9-3G8) and antihuman EGFR antibody (Dako, Denmark, Clone H11) were used to analyze Prominin-1 and EGFRvIII surface expression, respectively. Incubation using EGFR was followed by fluorescently tagged secondary incubation. Data was acquired using a FACSCalibur or FACSCanto flow cytometer (BD biosciences). For EdU incorporation assay spheres were cultured for 2 days, incubated with 2 μM EdU for 2 hours, and labeled using Click-IT EdU Assay kit (Thermo Fisher). Cells were analyzed using a FACSCalibur flow cytometer (BD biosciences). Data was analyzed using FlowJo 10 (Tree star, San Carlos, CA).

RNA isolation, cDNA synthesis and RT-PCR
Cells isolated from mouse brains or dissociated tumorspheres from culture were prepared as a single cell suspension and re-suspended in an appropriate volume of RLT buffer from RNeasy mini kit (Qiagen, Germany). RNA was isolated as per the manufacturers instructions and quantified using a Nanodrop (Nanodrop Technologies Inc., Wilmington, DE) Spectrophotometer. Complementary DNA was synthesized using Oligo(dT) primers and Superscript III (18080-051, Thermo Fisher Scientific). Primers were synthesized by IDT (IDT Inc, Coralville, IA). The following primers were used: Id1 (Forward (Fw):

Microarray profiling and analysis
Expression profiling was conducted by the Gladstone Institute Genomics Core Facility using Affymetrix Mouse Gene 1.0 ST Array (901169). Raw expression data was RMA normalized and Log2 transformed using Affymetrix Expression Console Software. The microarray data have been deposited in the GEO database (accession number GSE87332) and described in accordance with MAIME guidelines. Multiple Experiment Viewer Software (MeV; http://www. tm4.org/mev/) was used to perform Significance Analysis of Microarray (SAM) to determine differentially expressed genes between the two tumor types. A 1% false discovery rate and 2-fold expression difference was applied. The subsequent list of differentially expressed genes were analyzed for enriched Kegg Pathways using the Database for Annotation, Visualization and Integrated Discovery (DAVID).

Ethical statement
All animal work was approved by the UCSF institutional care and use committee (IACUC), approval number AN105263.

Statistics
All statistics were performed using GraphPad Prism 6.0 (GraphPad Software, San Diego, CA). The Mann-Whitney Rank Sum or the Student's t-test were used to determine differences between experimental groups as appropriate. The log-rank test was used to compare groups www.impactjournals.com/oncotarget in the Kaplan-Meier survival analysis. Chi-squared test was used to compare the frequency of invasive phenotypes in Figure 1C.