EB1-dependent long survival of glioblastoma-grafted mice with the oral tubulin-binder BAL101553 is associated with inhibition of tumor angiogenesis

Glioblastoma (GBM) are aggressive brain tumors with limited treatment options. Cancer stem-like cells (CSLCs) contribute to GBM invasiveness, representing promising targets. BAL101553, a prodrug of BAL27862, is a novel small molecule tubulin-binding agent, promoting tumor cell death through spindle assembly checkpoint activation, which is currently in Phase 1/2a in advanced solid tumor patients including GBM. This study aimed to evaluate long-term daily oral BAL101553 treatment of mice orthotopically grafted with GBM CSLCs (GBM6) according to EB1 expression-level, and to decipher its mechanism of action on GBM stem cells. Oral treatment with BAL101553 for 100 days provoked a large EB1 expression level-dependent survival benefit, together with a decrease in tumor growth and brain invasion. Formation of vascular structures by the fluorescent GBM6-GFP-sh0 cells, mimicking endothelial vascular networks, was observed in the brains of control grafted mice. Following BAL101553 treatment, vessels were no longer detectable, suggesting inhibition of the endothelial trans-differentiation of GBM stem cells. In vitro, BAL27862 treatment resulted in a switch to the endothelial-like phenotype of GBM6 towards an astrocytic phenotype. Moreover, the drug inhibited secretion of VEGF, thus preventing normal endothelial cell migration activated by CSLCs. The decrease in VEGF secretion was confirmed in a human GBM explant following drug treatment. Altogether, our data first confirm the potential of EB1 expression as a response-predictive biomarker of BAL101553 in GBM we previously published and add new insights in BAL101553 long-term action by counteracting CSLCs mediated tumor angiogenesis. Our results strongly support BAL101553 clinical studies in GBM patients.


INTRODUCTION
Glioblastoma (GBM) are the most frequent and aggressive primary brain tumors in adults with a median survival of only 15 months with the current standard of care [1]. The high heterogeneity due to functionally diverse cell types, hypervascularisation and the infiltrative nature of GBM tumor cells contributes to resistance to chemo-and radiotherapy [2,3]. Cancer stem cells represent a subpopulation of cells within GBM that are www.oncotarget.com Oncotarget, 2020, Vol. 11, (No. 8), pp: 759-774 Research Paper www.oncotarget.com characterized by increased resistance to chemotherapy and radiotherapy, suggesting that stem cells are likely responsible for failure of treatment and high recurrence rates [4][5]. Cancer stem cells are capable of infinite selfrenewal and multi-potential differentiation [6]. Moreover, GBM stem cells can initiate very aggressive tumor formation when implanted in xenograft models. Therefore, GBM stem cells are considered as a relevant target for GBM therapy, and the elimination of these cells is crucial in treating GBM.
GBM stem cells have been reported to directly contribute to the tumor vasculature through transdifferentiation into endothelial cells [7][8][9]. GBM stem cellderived endothelial cells possess the same vasculogenic activity as do endothelial cells [10,11] and are able to organize into capillary-like tubes in 3D Matrigel cultures [12]. Furthermore, the connection between neural stem cells and the endothelial compartment seems to be critical in GBM, where cancer stem cells closely interact with the vascular niche and promote angiogenesis through the release of vascular endothelial growth factor a (VEGFa) and stromal-derived factor 1 [13,14].
BAL27862 was previously shown to be a potent reversible microtubule (MT) destabilizer in vitro, when used at high concentrations, binding the colchicine site of the beta-tubulin [15]. It is a very potent inhibitor of tumor cell growth and promoter of cell death, whose activity is associated with activation of the spindle assembly checkpoint. Moreover, the drug is efficiently distributed into the brain and the tumor, with anticancer activity in GBM models, thus providing a strong rationale for its use for GBM treatment [16]. We recently demonstrated the anticancer activity of BAL27862 (in vitro) and its prodrug BAL101553 (xenografts) on the GBM cancer stem-like cell (CSLC) model GBM6. A survival advantage was observed after short-term intravenous treatment with indications of a reversal of GBM6 CSLC characteristics both in vivo and in vitro [16]. The A2B5 + GBM6 cell line was established from CSLCs isolated from a GBM patient. GBM6 cells display a mesenchymal phenotype with a high tumorigenicity and infiltrative pattern of migration in vivo [17,18]. Moreover, we previously reported that microtubule (MT) +End-binding 1-protein (EB1) overexpression correlates with GBM progression and poor survival in a large cohort of GBM patients [19]. Importantly, the level of EB1 expression in GBM6 cells strongly influenced BAL27862/BAL101553 response (even at sub-cytotoxic concentrations in vitro) as drug treatment was less potent in EB1-downregulated GBM6 than in EB1-expressing control cells [16]. BAL101553, administered both orally and intravenously, is currently undergoing phase 1/2a clinical evaluation in patients with solid tumors including GBM.
This study aims to evaluate long-term daily oral BAL101553 treatment of mice orthotopically engrafted with GBM6, according to EB1 expression level, as well as to further decipher the mechanism of action of the drug on GBM stem cells. Daily oral treatment with BAL101553 (5 administrations a week for 100 days) provoked a dramatic survival benefit, together with a decrease in tumor growth and brain invasion, which was EB1 expression level dependent. Moreover, the drug counteracted tumor vessel formation in the brains of treated mice. In vitro, the drug switched the endothelial-like phenotype of CSLCs towards an astrocytic phenotype, and it inhibited secretion of VEGF; thus, preventing normal endothelial cell migration activated by CSLCs. Our study shows for the first time that BAL101553 treatment counteracts tumor angiogenesis by acting on CSLCs in an EB1-dependent manner and provides new insights into the therapeutic targeting of CSLCs. Altogether, our data strongly support clinical studies with long-term administration of BAL101553 in GBM patients.

Long-term daily oral BAL101553 treatment enhances survival, reduces tumor growth and invasion in mice orthotopically grafted with GBM6; an effect potentiated by EB1 expression
To analyze long-term oral BAL101553 treatment in vivo, control GBM6-GFP-sh0 cells and EB1-downregulated GBM6-GFP-shEB1 cells were stereotaxically grafted into the subventricular zone of nude mice at day 0, and animals were orally treated with 30 mg/kg BAL101553 or vehicle control from day 35 to135 (5 administrations / week, 8 mice/group). Animals were monitored each day for weight loss, ataxia, and periorbital hemorrhage. Treatment was well-tolerated as evidenced by an absence of weight loss ( Figure 1A and 1B) (Supplementary Figure 1A and 1B) or change in animal well-being as compared with vehicle-treated animals. When treated with vehicle, animals grafted with GBM6-GFP-shEB1 cells had prolonged survival time, compared to animals grafted with GBM6-GFP-sh0 cells (215.5 and 170 days for GBM6-GFP-shEB1 and GBM6-GFP-sh0 tumor-bearing animals, respectively) ( Figure 1C and 1D, Table 1), confirming the bad prognostic value of EB1 (19). A significant long-term survival benefit was observed in the groups of animals treated with BAL101553. Indeed, median overall survival was extended by 326.5 days and 155 days for GBM6-GFP-sh0 and GBM6-GFP-shEB1 tumor-bearing animals, respectively, as compared with vehicle-treated animals. Moreover, death of tumor-grafted animals was observed at ages included in their normal life span [20]. Importantly, BAL101553 was more efficient if EB1 was not down-regulated (survival gain of 326.5 days, p<0.0001).
Serial sagittal sections of the brains were collected at the end of drug treatment (135 days post-grafting) to analyze tumor invasion (Supplementary Figure 1C).
Pictures taken from single sections were used for a 3D reconstruction of brains and quantification ( Figure 1E). GBM6-GFP-sh0 cells from vehicle-treated mice largely invaded the brain. Indeed, large and invasive tumors were detected invading the cortex, the striatum and the contralateral hemisphere ( Figure 1E top left panel, Supplementary Figure 1C, top left panel). In contrast, brain invasion was less pronounced in EB1 downregulated tumor grafted mice ( Figure 1E, top right panel, Supplementary Figure 1C, top right panel). Likewise, the volume of EB1-expressing tumor (sh0) was 7.1x10 -2 mm 3 , while that of shEB1 tumor was only 1.9x10 -2 mm 3 . Importantly, the volume of EB1-expressing tumor (sh0) after BAL101553 treatment was 1.6x10 -2 mm 3 , resulting in a clear reduction of the tumor volume (-76%) and in tumor spreading indicating a suppressive effect of BAL101553 on tumor growth and brain invasion in vivo ( Figure 1E

In vivo, BAL101553 induces loss of stem cell properties and inhibits tumor vasculature
In order to investigate whether BAL101553 could alter stem-cell properties in vivo, we first quantified the proportion of CSLCs in tumors under treatment. GFP+ tumor cells were sorted by FACs and tumor cells expressing the stem cell marker A2B5 were quantified ( Figure 2A). More than 80% of tumor cells expressed A2B5 at the time of grafting (day 0). At day 75, BAL101553 treatment shifted the proportion of tumor cells in favor of A2B5cells (-77.8 ± 2.3%; p<0.05 and -74.3 ± 2.6%; p<0,001, for GBM6-GFP-sh0 and GBM6-GFP-shEB1 tumors, respectively ( Figure 2A). However, at day 135, the decrease in the proportion of A2B5 + cells was only maintained in GBM6-GFP-sh0 tumors (-75.3 ± 4.7%; p<0,001; Figure 2B). These data show that BAL101553 treatment reduced efficiently the proportion of CSLCs in GBM6 orthotopic tumor-bearing mice in EB1-expressing tumors, as long as the drug was administered. To determine whether BAL effect results from a dramatic reduction in tumor volume rather than a specific effect on GBM CSLCs, we then analyzed distribution of tumor CSLCs within brains by immunohistochemistry analysis with an anti-GFP antibody. Importantly, formation of patterned tubule networks by the GBM6-GFP-sh0 cells mimicking endothelial-lined vascular networks was observed inside tumors ( Figure 2C). Staining with human anti-CD31 antibody confirmed the presence of vessels containing the human stem cells. Importantly, no staining with anti-GFP and human anti-CD31 antibodies was observed in brain isolated from normal mouse. These results suggest that CSLCs acquire endothelial properties to participate to the tumor vasculature. Following BAL101553 treatment, we have detected in GBM6-shEB1 tumors the presence of several structures stained with anti-GFP and human anti-CD31 antibodies, contrary to GBM6-sh0 tumors, suggesting that the drug blocked CSLCs-acquired endothelial properties in an EB1 expression-level dependent manner.

In vitro, BAL27862 treatment inhibits GBM6 endothelial differentiation, potentiated by EB1 expression
We then decided to investigate the potential antiangiogenic mechanism of the drug on GBM6 cells in vitro. We first observed that GBM6 cells could participate to vascular structures under appropriate cell culture conditions in vitro. Indeed, GBM6 cells acquired an endothelial phenotype when co-cultured with human endothelial cells, as a typical capillary network including fluorescent cells (GBM6-GFP cells) was observed 48 h after seeding GBM6-GFP-sh0 and GBM6-GFP-shEB1 cells with HMEC-1 on Matrigel ( Figure 3A and 3B).
To determine whether cell-cell contact between HMEC-1 and GBM6 cells was necessary for tube formation, GBM6-GFP-sh0 and shEB1 were seeded alone on Matrigel. Interestingly, a capillary network was formed 48 h later, suggesting that GBM6 cells can differentiate into endothelial-like cells ( Figure 3C and 3F). A favorable microenvironment was required, since GBM6 cells did not form tubes when cultured on a poly-DL-ornithine substrate (Supplementary Figure 2A). No statistically significant difference in the length of tubes was observed between GBM6-GFP-sh0 and shEB1 when co-cultured with HMEC-1 or grown alone, suggesting that the level of expression of EB1 did not influence GBM6 differentiation towards an endothelial phenotype. A difference between control GBM6-GFP-sh0 and GBM6-GFP-shEB1 cells was only observed when spheres of GBM6 were seeded on a preformed vascular network ( Figure 3D, Supplementary Video 1). Indeed, GBM6-GFP-sh0 but not GBM6-GFP-shEB1 cells migrated along tubes, confirming the promigratory role of EB1 in CSLCs as previously described [16].
As the endothelial phenotype is characterized by specific biomarkers, the expression of CD31 and ERG in GBM6 cells after BAL27862 treatment was analyzed. Cells were harvested non-enzymatically from Matrigel at the same time as tube length measurement (48 h). GBM6-GFP-sh0 and GBM6-GFP-shEB1 expressed CD31 and ERG, thus confirming endothelial differentiation of GBM6 when cultured on Matrigel ( Figure 3G and 3H) but not on poly-DL-ornithine (Supplementary Figure 2F and 2G). Importantly, exposure to a low concentration of BAL27862 (6 nM) triggered a significant decrease in ERG and CD31 expression level in GBM6-GFP-sh0 cells ( Figure 3G and 3H). For example, ERG expression was decreased around 3 times as quantified by western blotting and the percentage of CD31-expressing GBM6-GFP-sh0 cells was decreased by 80.3 ± 5.2 % after drug treatment (FACs analysis) ( Figure 3H). In addition, an increase in the expression of GFAP was detected in control GBM6-GFP-sh0 cells following drug treatment, which is typical of the astrocytic differentiation ( Figure 3G). Altogether, these results show that sub-cytotoxic concentrations of BAL27862 switch the endothelial-like differentiation of GBM6 cells towards an astrocytic phenotype in an EB1 expression-level dependent manner.

BAL27862 suppressed microtubule dynamic instability
In order to decipher the mechanism of action of BAL27862 on GBM6 cells, we first performed indirect immunofluorescence of α-tubulin in GBM6-sh0 and GBM6-shEB1 cells. We observed that contrary to high concentration (1 µM), the low concentrations of BAL27862 that inhibited endothelial differentiation (i.e. 6 nM), did not depolymerize MT network in cells suggesting that BAL27862 mainly acts by stabilizing MT dynamics ( Figure 4A). We then performed measurements of MT plus-end dynamics, at steady state, in an in vitro reconstituted system using dynamic MT and EB3-GFP as a plus-end tracker ( Table 2). Nanomolar concentrations of BAL27862 (75-100 nM) suppressed MT dynamics by decreasing the MT growth rate and increasing time and distance-based catastrophe frequencies. Such stabilizing effect of BAL27862 on MT dynamic instability parameters in a reconstituted system is consistent with the effect of several other members of the MTA family, including Taxanes, Epothilones and Vinca-alkaloids [21,22]. In addition, we previously demonstrated the crucial role of MT dynamics alterations by MTA on VEGF signaling disruption in endothelial cells resulting in inhibition of angiogenesis and migration (19,23), leading us to investigate the role of BAL27862 on VEGF secretion.

BAL27862 inhibits VEGF secretion by GBM6 and human GBM tissue
Besides differentiation of CSLCs to become part of the tumor vasculature as endothelial cells, CSLCs can also exert paracrine effects on endothelial cell by secreting soluble factors to stimulate tumor angiogenesis [13,14]. To examine this, we employed an in vitro model of angiogenesis in which HMEC-1 cells were induced to migrate when stimulated by angiogenic factors [24] (Supplementary Figure 3). First, we measured the concentration of VEGF in the culture medium of GBM6-GFP-sh0 and shEB1 grown on Matrigel in the lower chamber after BAL27862 treatment. As shown in Figure  5A, a low, non-cytotoxic concentration of BAL27862 inhibited secretion of VEGF by 57 ± 15% (p<0.05) and 25 ± 18% in GBM6-GFP-sh0 and GBM6-GFP-shEB1 cells, respectively. However, mRNA levels of VEGF were not altered by BAL27862 ( Figure 5B). Then an upper chamber containing HMEC-1 cells was inserted and the percentage of HMEC-1 cells migrating to the lower chamber containing GBM6 cells was quantified. In experiments with vehicletreated GBM6 cells, HMEC-1 cells triggered pronounced migration, while the increased migration of HMEC-1 cells was significantly prevented when GBM6 cells were pretreated with BAL27862 ( Figure 5C and 5D). In control experiments where GBM6 were not seeded in the lower chamber, HMEC-1 cells were unable to migrate, meaning that a soluble factor released by GBM6 was necessary for migration of endothelial cells ( Figure 5C). When GBM6  were cultured on poly-DL-ornithine, no VEGF was secreted and no migrating HMEC-1 cell was scored (Supplementary Figure 4A-4C). Finally, we confirm that VEGF secretion by GBM6 cells was involved in endothelial cell migration by using siRNA to deplete VEGF in GBM6-GFP-sh0 and shEB1 seeded on the lower chamber ( Figure 5E). As shown in Figures 5F and 5G, the number of HMEC-1 cells migrating to the lower chamber was decreased after VEGF down-regulation (-77.3 ± 4.9%; p<0.05 and -60.8 ± 14.0%; p<0.05, for GBM6-GFP-sh0 and GBM6-GFP-shEB1 cells, respectively). Altogether, these results reveal that sub-cytotoxic concentrations of BAL27862 inhibit VEGF secretion by GBM6 cells and consequently suppress GBM6-induced migration of endothelial cells. These effects were lower in EB1-down regulated stem cells. Finally, BAL27862 decreased VEGF secretion in a human GBM explant culture after 3 days of exposure, as shown in vitro with GBM6 ( Figure 5H).

DISCUSSION
In this study, we have demonstrated that long-term daily, oral treatment with BAL101553 strongly enhanced survival of mice when orthopically grafted with human GBM CSLCs. Indeed, a 100 day-treatment was sufficient to provoke a survival benefit that was almost one year longer than observed in vehicle-treated mice. This result emphasizes our previous work showing that a short-term treatment with BAL101553 administered intravenously (on days 30, 33 and 36 after tumor implantation) induced a moderate EB1-dependent survival benefit in tumor-bearing mice. BAL101553 activity was higher in control CSLCs than in EB1 down-regulated CSLCs, thus confirming the role of EB1 expression as a potential biomarker for drug response [16,19]. Post-mortem analysis of brains indicated that tumor growth was reduced and tumor spreading was limited under treatment, while it invaded the surrounding parenchyma in vehicle-treated mice. However, the tumor was not totally eradicated suggesting that additional treatment in a maintenance therapy schedule may have maintained a higher level of tumor growth inhibition; perhaps, with an intermittent schedule to limit resistant cell emergence [25]. The good tolerability and oral bioavailability of BAL101553 are important advantages when considering long-term treatments.
Our study shows for the first time that BAL101553 exerts anti-angiogenic effects in vivo and in vitro by acting on GBM CSLCs. Furthermore, as also reported for other GBM stem models, we demonstrated that GBM6 cells have the ability to activate the angiogenic switch; a critical transition point for tumor growth [26][27][28]. To decipher the anti-angiogenic mechanism of the drug, experiments were designed on Matrigel to mimic in vitro the process of CSLC trans-differentiation into endothelial cells in particular microenvironmental conditions [9].
Cell cultures on poly-D-ornithine were conducted as negative controls. BAL27862 treatment caused a switch of GBM6 endothelial differentiation towards an astroglial phenotype, consistent with our previous observations [16]. Furthermore, it altered the contribution of CSLCs to normal endothelial cell behavior by decreasing VEGF secretion by GBM6 cells and, thereby, reducing endothelial cell migration and tube formation. A decrease in VEGF secretion induced by the drug was also observed in human GBM explants that contain around 30% of cancer stem cells. This point is of interest, as VEGF secretion is known to initiate angiogenesis [29]. Our results confirm the angiogenic dependency of GBM [30]; however, anti-angiogenic therapeutics were unsuccessful so far. The administration of bevacizumab, an anti-VEGFa antibody, to GBM patients improved progression-free survival but failed to show a survival benefit, because many patients rapidly developed resistance [31]. The resistance mechanism(s) is not yet elucidated. The ability of CSLCs to trans-differentiate into endothelial cells that display low sensitivity to bevacizumab or other anti-VEGFR therapies has been suggested [32,33]. Inhibition of the endothelial trans-differentiation process of CSLCs by BAL101553 could potentially overcome antiangiogenic therapy resistance in GBM, by acting upstream in the signaling cascade of angiogenesis, as compared with the anti-angiogenic drugs currently marketed. Several MTA, such as Vinca-alkaloids and combretastatin A4 and its derivatives have been described for antiangiogenic and antivascular properties by inhibiting the formation of endothelial capillary -like structures and disrupting preformed tubular structures from endothelial cells [34][35][36]. BAL 27862 shares these properties. Importantly, our study is the first showing that suppressing MT dynamics leads to reduction in VEGF secretion, which mechanism remains to decipher but probably impact VEGF trafficking as no effect was observed on mRNA levels. Furthemore, BAL27862 exerts anti-angiogenic activity by tackling CSLCs, such effect was observed at low concentrations that were not cytotoxic and did not alter vessel formation by normal endothelial cells. Moreover, the dose used for in vivo experiments was well-tolerated, and no sign of vascular toxicity such as hemorrhage was detected.
Our results highlight EB1-dependent effects of BAL101553, which were observed mostly in control sh0 cells, suggesting that the mechanism of sensitization to BAL101553 by EB1 in angiogenesis is related to stabilization of MT dynamics. Through its C-terminal part, EB1 interacts with numerous partners at growing MT ends to form molecular networks that contribute to the regulation of MT dynamic instability and regulate MT functions [37]. In addition, the amount of EB1 influences the way MT dynamic instability is altered [38] and posttranslational modifications of EB1 also modulate the regulation of MT dynamics, proliferation /migration and subsequent MTA response [23,38,39]. Interestingly, www.oncotarget.com VEGF suppressed MT dynamics in living HUVECs, induced EB1 C-terminal detyrosination and increased EB1 comet length. Vinflunine, a MTA of the Vinca-alkaloid family, decreased the level of the detyrosinated form of the C-terminal of EB1, increased the native tyrosinated form of EB1 and completely abolished the effect of VEGF on EB1 comets [23]. Vasohibins were identified as intrinsic factors in endothelial cells with some inhibitory effect on angiogenesis [40,41]. However, they are overexpressed upon VEGF treatment of endothelial cells, suggesting that they have a more complex role in angiogenesis. Interestingly, vasohibins have been recently found as the first enzymes able to detyrosinate the C-terminal EEY sequence of tubulin and EB1 [42]. All these new findings highlight questions on the functionality of the tyrosination cycle of tubulin and EB1 in cancer stem cells and the role of cytoskeleton and drugs targeting MT in angiogenesis. Moreover, it has been recently proposed that the loss of Golgi-MT attachment resulting from EB binding to MT minus end in mutant EB1/EB3 cells might affect cell polarity and thus cell motility in 3D matrices [43]. Functional consequences of EB1 binding to the minus end of MT remain to be understood.
Altogether, these results show that BAL101553 counteracts the formation of brain tumor vessels by inhibiting GBM stem cell trans-differentiation in tumorderived endothelial cells, as well as VEGF secretion; thus, cutting off the tumor from blood and nutrition supply and potentially adding to the direct anti-tumor cell effect. A high level of EB1 expression in CSLCs potentiates the drug effects, further supporting the potential of EB1 expression as a BAL101553 response-predictive biomarker in GBM.

Compounds
BAL27862 and BAL101553 were provided by Basilea Pharmaceutica International Ltd (Basel, Switzerland). BAL27862 is the active moiety of the lysine prodrug BAL101553, which was developed to improve solubility without the need of excipients associated with side effects. Hence, the prodrug BAL101553 was used for all animal experiments. As the cleavage of the prodrug BAL101553 is incomplete in cell culture medium, the active moiety BAL27862 was used for in vitro assessments.

Explant culture of human glioblastoma tissue
GBM tissue samples from a GBM patient were collected after surgery and placed in Dulbecco's modified Eagle's medium (DMEM) supplemented with 0.5% fetal calf serum (FCS), 1% penicillin-streptomycin and 1% sodium pyruvate (Gibco-Invitrogen, Cergy Pontoise) [44]. Tissues were cut into 500 μm pieces in defined serum-free stem cell medium [16] and plated on 12-well plates precoated with poly-(L)-lysine (10 μg/mL; Sigma). Medium was supplemented with 0.4% methycellulose (Sigma). Explants cultures were then incubated at 37°C in a humidified atmosphere of 5% CO2. After 72 h of culture, explants were treated with 2, 6 or 20 nM of BAL27862. After 72 h of treatment, explants were processed for analysis of VEGFa protein secretion using an ELISA kit (Sigma-Aldrich).

Cytotoxic assay
Cells (5000 cells/well) were seeded on poly-DLornithine (Sigma-Aldrich) coated 96-well plates (10 µg/ mL) and allowed to grow for 24 h before treatment with BAL27862. Growth inhibition of cells was measured after 72 h by using a sulforhodamine B assay kit (Sigma-Aldrich) as described previously [19].

In vitro FACS analysis
GBM6 cells were seeded onto 6-well plates previously coated with Matrigel or poly-DL-ornithine and incubated with BAL27862 or vehicle for 48 h. Cells were then harvested and incubated with CD31-PE antibody (clone AC128, reference 130-092-653) (Miltenyi Biotec). Cells were fixed and analyzed using flow cytometry (FACS Calibur™, BD Biosciences). A total of 100,000 events were counted for each sample and data were recorded with CellQuest Pro Software (BD Biosciences) and analyzed using FlowJo software (Tree Star Inc., San Carlos, CA) choosing the Dean-Jet-Fox model analysis.

Endothelial cell capillary-like tube formation assay
Matrigel (Corning) was thawed at 4°C and 48well plates were coated with Matrigel, which was then allowed to solidify for 1 h at 37°C before cell seeding. Cells (3x10 4 GBM6 or HMEC-1 alone, or 15x10 3 GBM6 with 15x10 3 HMEC-1 for co-culture experiments) were then added in medium with BAL27862 or vehicle. For analysis of GBM6 invasion on a well-organized capillarylike network, HMEC-1 were first seeded on Matrigel and incubated without BAL27862 or vehicle for 15 h, then 10 GBM6 spheres were added concomitantly with BAL27862 or vehicle. Cells were allowed to undergo morphogenesis and form capillary-like structures and photographs were taken at indicated time points using the 4X objective of a Juli TM Stage live-cell imaging analyzer (NanoEnTek), and the total length of capillary tubes formed was measured using Image-Pro Plus software.
In order to analyze the potential changes in protein expression, GBM6 cells (3×10 5 ) were seeded onto 6-well plates previously coated with Matrigel and harvested at 48 h. For western blot analysis, cells were incubated with a Cell Recovery Solution (BD Biosciences) for 1 h at 4°C under agitation to allow complete dissolution of the Matrigel, then pelleted, washed with cold PBS and finally lysed as described in the western blotting section. For FACs analysis, cells were incubated with 100 units/well of Dispase (Corning) for 2 h at 37°C under agitation to allow complete dissolution of the Matrigel, then pelleted and washed with PBS.

Transwell migration
To assay GBM6-induced migration of endothelial cells, a co-culture assay using migration chambers as described by Tsujii et al was used [24]. Assays were performed in 24-well transwell chambers (8.0 μm diameter pore). GBM6 cells (6x10 4 ) were exposed to BAL27862 or vehicle (control) for 48 h on Matrigel. For VEGFa down-regulation experiments, cells were seeded 24 h after siRNA transfection. After treatment, the medium from the lower chamber was replaced with fresh medium without BAL27862 or vehicle, and the VEGFa protein released was determined using an ELISA kit (Sigma-Aldrich). Alternatively, an insert containing HMEC-1 cells (5x10 4 ) in MCDB-131 medium without SVF was then placed into the lower chamber. After 24 h, cell migration was quantified as described [19].

Indirect immunofluorescence analysis
GBM6 cells were grown on 8-well chamber slides (Labtek, Thermo Fisher Scientific), precoated for 1 h with poly-DL-ornithine (Sigma-Aldrich) (10µg/ml), to be treated for 48 h with BAL27862. As previously described [19], cells were incubated with an anti-α-tubulin (clone DM1A; Sigma-Aldrich) primary antibody, and then with Alexa488 secondary antibody (Invitrogen). Staining was observed using either a Leica DM-IRBE microscope. Images were acquired using Metamoph software and were processed using Image J software.

In vitro plus end tracking assay of GFP-EB3 and measurements of MT dynamics:
The assay was performed by growing MTs in the presence of 13 μM purified tubulin (Cytoskeleton, Denver, CO) from GMPCPP-stabilized (Sigma-Aldrich, St. Louis, MO) and paclitaxel-stabilized MT seeds, which were immobilized on glass coverslips using Poly-l-Lysine PEGbiotin (SuSoS, Dübendorf, CH)-streptavidin (Invitrogen, Carlsbad, CA, Lonza, Basel, CH) links, as described in [45]. BAL 27862 was added at the beginning of MT polymerization.

RNA extraction
Total RNA was extracted using QIAamp® RNA Blot Mini kit (Qiagen, Courtaboeuf, France) according to the manufacturer's instructions. RNA samples with no evidence of ribosomal peak degradation and RIN values of 6-10 were used for real-time quantitative PCR analyses [46] after treatment with 1U ribonuclease-free deoxyribonuclease (Roche Applied Science, Meylan, France) at 37°C for 15 minutes.

Animal experiments
All experimental procedures and animal care were carried out in conformity with the guidelines of the French Government and approved by the Regional Committee for Ethics on Animal Experiments (authorization number 0100903). Six to 8 weeks old female athymic nude mice were obtained from Harlan Laboratories France and stereotactically grafted with GBM6-GFP-sh0 or shEB1 cells as previously described [16]. Four groups (n=8) were randomized for GBM6-GFP-sh0-and GBM6-GFP-shEB1-bearing mice respectively, receiving daily oral administrations of BAL101553 (30 mg/kg in 10 ml/kg application volume) or vehicle solution (10 ml/kg, control groups) from day 35-135 after glioma implantation (5 administrations / week). Animals were monitored each day for weight loss, ataxia, and periorbital hemorrhage. Animals were euthanized when affected by hemiplegia or 20% weight loss. At 75, 105 and 135 days after glioma implantation, three animals of each group were sacrificed for analysis of tumor volumes by 3D-reconstruction, for FACS analysis [16] and for immunohistochemistry staining.

Immunohistochemistry
GFP or CD31 immunohistochemistry was carried out in tumor area, using an anti-GFP primary antibody (Abcam) or an anti-human CD31 primary antibody (clone JC/70A) (Dako) and avidin-biotin-peroxidase method. Isotype control antibody (BD Biosciences) was used at the same concentration as primary antibody. Brains from non-grafted mice were used as control.

3D reconstruction
For each brain, a complete collection of serial cryosections was imaged by automatic scanning microscope and processed by For3D as described [48, 49] using a Miraxmidi slide scanner (Zeiss, Jena, Germany). ImageJ and homemade Matlab functions were used to render brains and tumors in 3D. Autofluorescence was used to detect and segment the whole brain tissue. GBM cells expressing GFP were segmented from autofluorescence by subtracting background in Image J and applying higher threshold values. Matlab was used to identify individual volumes of the 3D GBM tumors within each brain, as described [48,49].

Statistical analysis
Data are presented as mean ± S.E.M. At least, three independent experiments were performed for each condition in cellular studies. In vitro data were analyzed using the two-tailed Student's t test. Survival medians were estimated by the Kaplan-Meier product limit method. The log-rank test was used to compare survival rates by univariate analysis. Reported p-values are two-sided, and p<0.05 was considered statistically significant. Asterisks indicate significant level vs control * , p < 0.05; ** , p < 0.005; *** , p < 0.001. Statistical analyses were performed with GraphPad 5.0 statistical software.

Author contributions
Raphaël Bergès: designed the project, performed experiments and collected data, analyzed the data, wrote the manuscript. Aurélie Tchoghandjian: performed experiments and collected data. Arnauld Sergé: provided technical support. Stéphane Honoré: provided technical support. Dominique Figarella-Branger: analyzed the data. Felix Bachmann: analyzed the data. Heidi A. Lane: analyzed the data. Diane Braguer: designed the project, analyzed the data, wrote the manuscript.

ACKNOWLEDGMENTS AND FUNDING
This work has been carried out thanks to the support of the ITMO Cancer AVIESAN as part of the Cancer Plan no. PC201419 and Basilea. The authors wish to thank Jessica Cappai and Nathalie Baeza-Kallee for technical help and the animal facility "centre d'exploration fonctionnelle scientifique" (CEFOS) agreement E 13-055-5, campus Timone Aix-Marseille University.