Modulation of cerebral endothelial cell function by TGF-β in glioblastoma: VEGF-dependent angiogenesis versus endothelial mesenchymal transition.

Glioblastoma are among the most angiogenic tumors. The molecular mechanisms that control blood vessel formation by endothelial cells (EC) in glioblastoma remain incompletely understood. Transforming growth factor-β (TGF-β) is a key regulatory cytokine that has proinvasive and stemness-maintaining autocrine properties in glioblastoma and confers immunosuppression to the tumor microenvironment. Here we characterize potential pro- and anti-angiogenic activities of TGF-β in the context of glioblastoma in vitro, using human brain-derived microvascular endothelial cells (hCMEC/D3) and glioblastoma-derived endothelial cells (GMEC) as model systems. We find that TGF-β induces vascular endothelial growth factor (VEGF) and placental growth factor (PlGF) mRNA expression and protein release in a TGF-β receptor (TβR) II / activin-like kinase (ALK)-5-dependent manner under normoxia and hypoxia, defining potential indirect proangiogenic activity of TGF-β in glioblastoma. In parallel, exogenous TGF-β has also inhibitory effects on EC properties and induces endothelial-mesenchymal transition (EndMT) in hCMEC and GMEC. Accordingly, direct inhibition of endogenous TGF-β/ALK-5 signalling increases EC properties such as tube formation, von-Willebrand factor (vWF) and claudin (CLDN) 5 expression. Yet, the supernatant of TGF-β-stimulated hCMEC and GMEC strongly promotes EC-related gene expression and tube formation in a cediranib-sensitive manner. These observations shed light on the complex pro- and anti-angiogenic pathways involving the cross-talk between TGF-β and VEGF/PLGF signalling in glioblastoma which may involve parallel stimulation of angiogenesis and EndMT in distinct target cell populations.


INTRODUCTION
Blood supply is an essential prerequisite for solid tumor growth and therefore a bona fide target for intervention. The processes by which tumors initiate and maintain their blood supply are complex and have remained incompletely understood. Vasculogenesis is defined as the initial migration and differentiation of endothelial precursor cells to form blood vessels de novo. Subsequent to this, angiogenesis occurs by the formation of new blood vessels from pre-existing ones, also termed sprouting angiogenesis. Angiogenesis in glioblastoma is characterized by a neovasculature that has an erratic basement membrane, reduced pericyte coverage, decreased astrocytic end feet association with blood vessels, and altered expression of adhesion molecules on endothelial cells (EC) [1]. This abnormal vascularization causes excessive leakiness and promotes edema formation. www.impactjournals.com/oncotarget One traditional view states that hypoxic tumor cells release vascular endothelial growth factor (VEGF) which in turn counteracts tumor hypoxia by mediating the growth of new blood vessels. However, it is unlikely that all blood vessel formation in glioblastoma is triggered by hypoxiamediated VEGF signaling, notably early in the disease course when there is probably little hypoxia. Thus, alternative pathways regulating tumor-associated angiogenesis beyond hypoxia-driven VEGF synthesis are likely to exist.
Transforming growth factor (TGF)-β is a prototype member of a cytokine family involved in most cancerassociated processes. In glioblastoma, immunosuppression and autocrine control of migration, invasiveness and stemness have received most attention. However, TGF-β has also been implicated in angiogenesis in various model systems including physiological and pathological angiogenesis. Direct TGF-β signalling on endothelial cells has been linked mainly to two type I receptors, activin-like receptor kinases (ALK)-1 and ALK-5. These two TGFβR I-mediated pathways have been attributed to opposing effects in response to TGF-β: ALK-1 on interaction with TβRII induced Smad 1/5 phosphorylation to induce endothelial cell proliferation [2] and migration [3] whereas ALK-5 recruited by TβRII induces Smad 2/3 phosphorylation, thereby inhibiting the above-mentioned effects [3]. TGF-β family ligands can interact with coreceptors represented by endoglin (CD105), which is highly expressed in proliferating endothelial cells, and betaglycan (TβRIII). In the present study, we investigated the interaction and dependence of these key signaling pathways and their role for glioma angiogenesis, using hCMEC and GMEC as models.
After confirming the expression of the TGF-β receptors and ligand, we transiently silenced TβRII, ALK-1, ALK-5, endoglin or TβRIII in the absence or presence of TGF-β. Silencing was confirmed at the mRNA and protein levels wherever feasible ( Figure 1E-1F). The specificity of silencing ALK-1 versus ALK-5 was verified by RT-PCR (Supplementary Figure 1B-1C). Interestingly, exposure to exogenous TGF-β decreased TβRIII and increased ALK-5 at mRNA level and endoglin expression at mRNA and protein levels. Constitutive pSmad2 was faintly detectable and not induced by receptor silencing. Constitutive pSmad1/5 levels increased mildly with silencing of endoglin, but not with silencing of the other receptors ( Figure 1E-1F). TGF-β strongly induced pSmad2, and this induction was attenuated by the silencing of TβRII, ALK-1, ALK-5 or endoglin, although not TβRIII. TGF-β strongly decreased pSmad1/5 levels, and this effect was attenuated somewhat by the silencing of ALK-1, ALK-5, TβRII, TβRIII or endoglin, but robustly by SD-208. TβRII levels were strongly reduced at 72 h after exposure to TGF-β ( Figure 1F).
Under these experimental conditions, exposure of hCMEC to TGF-β did not induce significant loss of viability assessed by trypan blue dye exclusion, cell cycle arrest assessed by PI staining or features of apoptosis as assessed by annexin V / propidium labeling (Supplementary Figure 2).

TGF-β induces VEGF mRNA expression and protein release in hCMEC under normoxic and hypoxic conditions
We next investigated the effect of recombinant TGF-β on VEGF expression in hCMEC at increasing concentrations for 24 h either at normoxia or under hypoxic conditions (1% O 2 ). VEGF mRNA expression and protein release were increased approximately 10fold when the cells were shifted to hypoxia. Exposure to TGF-β increased VEGF mRNA expression and protein release in a concentration-and time-dependent manner, more significantly even in hypoxia than under normoxic conditions (Figure 2A-2D). The lowest concentration of TGF-β (0.1 ng/ml) did not increase VEGF protein in normoxia whereas the increase was significant in hypoxia. The induction of VEGF mRNA expression was not restricted to a specific VEGF isoform ( Figure 2E). Hypoxia increased PlGF mRNA expression, moreover, the combination of hypoxia and TGF-β resulted in a further enhancement compared to hypoxia or TGF-β alone ( Figure 2F). PlGF release was increased by TGF-β or hypoxia, and the combination was more effective than either treatment alone ( Figure 2G).
The important signaling function of ALK-5 was confirmed by pharmacological studies using SD-208 which demonstrated no effect on constitutive VEGF release, but strong activity against TGF-β-evoked VEGF gene transcription and protein release ( Figure 3C-3D). This effect was not restricted to specific VEGF mRNA isoforms (Supplementary Figure 3). Under hypoxic conditions, SD-208 even inhibited constitutive VEGF release. The effects of TGF-β and accordingly SD-208 were overall more prominent for PlGF than for VEGF ( Figure 3E-3F).

Transcriptional regulation of TGF-β-evoked VEGF release
To elucidate the transcriptional regulation of VEGF by TGF-β, 2 major transcription factors known to have multiple response elements on the promoter region of VEGF, namely, specificity protein (SP1) and early growth response (EGR)1 were screened for basal mRNA expression in hCMEC, HUVEC and two glioma cell lines, U87MG and LN-18 (Supplementary Figure 4A-4B). All cells expressed SP1 and EGR1. SP1 and EGR1 mRNA expression decreased with TGF-β exposure at 1-6 h in hCMEC, however, protein levels were unaffected (Supplementary Figure 4C). Moreover, the silencing of SP1 did not alter VEGF mRNA expression or protein release (Supplementary Figure 4D). Hence, these data do not support the idea that these transcription factors are involved in the TGF-β-evoked increase of VEGF in hCMEC in normoxia.

TGF-β induces endothelial mesenchymal transition (EndMT) and TGF-β-stimulated supernatant increases tube formation via VEGFR signaling in hCMEC
We noted that exposure to TGF-β caused changes in the morphology of hCMEC, making the cells more elongated and spindle-like in a SD-208-sensitive, thus presumably ALK-5-dependent manner ( Figure 4A). Interestingly, however, SD-208 alone induced tube formation, endothelial cell sprouting and expression of the tight junction marker, CLDN5. Conversely, exogenous TGF-β blocked the formation of tubes and sprouts and decreased vWF and CLDN5 mRNA expression. Pretreatment with SD-208 rescued tube formation, sprouting, vWF and CLDN5 mRNA expression ( Figure  4B-4D). Further, TGF-β increased integrin β3 and β5, snail family zinc finger (SNAI) 1 and N-cadherin mRNA expression and protein levels and pretreatment with SD-208 attenuated this increase ( Figure 4D-4F). Altogether, the pattern of decreased vWF [4], CLDN5 [5,6] with a concomitant increase in β3 [7] and β5 integrin expression [8], SNAI1 [5,9] and N-cadherin [10] are prominent signs of mesenchymal transition. To dissect the apparent opposing pro-and anti-angiogenic effects of TGF-β on hCMEC, we changed the experimental paradigm and isolated the effect of TGF-β-induced VEGF on tube formation and cell sprouting by using supernatants of hCMEC treated with TGF-β for 96 h. These were added to hCMEC that had been pre-treated with vehicle (DMSO) or SD-208. Under these conditions, TGF-β-stimulated supernatant increased tube formation and sprouting. SD-208 super-induced these pro-angiogenic effects of TGFβ-stimulated supernatants, consistent with antagonistic activity of endogenous or supernatant-derived TGF-β and illustrating how TGF-β and VEGF counterbalance the endothelial phenotype. Cediranib, a pan-VEGFR inhibitor, decreased constitutive and TGF-β-stimulated supernatantinduced tube formation in the absence or presence of SD-208 ( Figure 4G).

Endothelial properties of glioblastoma-derived endothelial cells (GMEC)
Finally we wished to confirm the principle observations summarized above in freshly isolated CD31positive ex vivo endothelial cells (ZHE-459, ZHE-464, ZHE-483-2) from glioblastoma tissue. Their endothelial nature was verified by CD31, CD34, and vWF expression and by their tube formation capacity (Supplementary Figure 5A-5B). CD31 mRNA expression in the CD31positive ZHE-464 cells at P5 was reduced approximately 50 times compared to that on the day of isolation whereas CD31-negative cells did not express CD31 even after 5 passages in endothelial medium, DMEM or neurobasal medium (Supplementary Figure 5C).

Figure 4: TGF-β induces EndMT in hCMEC and TGF-β-stimulated hCMEC supernatant increases tube formation. A.
hCMEC were seeded in full medium, treated with TGF-β (10 ng/ml) or SD-208 (1 μM) for 96 h, and examined by phase contrast microscopy. B. Tube formation was quantified by assessing the number of tubes (n), tube length (mm) and number of branching points (n). C. Endothelial cell sprouting was assessed by counting numbers of sprouts (n). D. mRNA expression of vWF, CLDN5, β3 and β5 integrins, SNAI1 and N-cadherin was assessed by qRT-PCR under the same conditions as in A, for 72 h. Data are expressed as mean and SD (n = 3) ( + p < 0.05, ++ p < 0.01, +++ p < 0.001, one-way ANOVA followed by Tukey's post hoc test, effect of TGF-β, ΦΦ p < 0.01, ΦΦΦ p < 0.001, effect of SD-208). E. Protein levels of SNAI1 and N-cadherin were assessed by immunoblot under the same conditions as A at 96 h. F. Protein levels of αvβ3 and αvβ5 was assessed by flow cytometry under the same conditions as in A, at 96 h. G. hCMEC supernatant treated with TGF-β for 96 h was concentrated and added with or without cediranib (1 μM) to hCMEC pretreated with DMSO or 1 μM SD-208 for 1 h. Tube formation was assessed after 12 h, by determining the number of tubes and tube length. Data are expressed as mean and SD (n = 3) (ΦΦ p < 0.01, ΦΦΦ p < 0.001, one-way ANOVA followed by Tukey's post hoc test, effect of SD-208, + p < 0.05, ++ p < 0.01, +++ p < 0.001, effect of TGF-β-stimulated supernatant, #p < 0.05, ###p < 0.001, effect of cediranib). www.impactjournals.com/oncotarget

DISCUSSION
Angiogenesis is an important step in the multi-step formation of solid cancers including glioblastoma. It requires a crosstalk between tumor and endothelial cells that is likely to involve much more complex interactions than merely hypoxic tumor cell-derived VEGF acting on endothelial cells. Targeting blood vessel formation in glioblastoma has been the favored therapeutic strategy in clinical drug development in the first decade of this century. Yet, the failure of multiple drugs acting not only on VEGF or its receptors, but also on alternative targets including integrins has imposed a brake on further efforts of targeting angiogenesis and instead called for a better understanding of the processes that govern glioblastomaassociated blood vessel formation [11].
Primary resistance to or escape from anti-angiogenic therapy in glioblastoma is likely to be mediated by compensatory activation of multiple pro-angiogenic pathways and shaped by complex interactions between tumor cells and their microenvironment [12]. VEGF remains to be the major molecule thought to drive endothelial cell growth, migration, vessel dilation and permeability. VEGF, VEGF-B, VEGF-C, VEGF-D and PlGF represent a family of related growth factor molecules which bind to VEGFR1-3 with varying affinity. VEGF binding to VEGFR-2 is critically required for vascular development and sprouting, maturation and differentiation of endothelial cells [13].
TGF-β is a multifunctional cytokine that has been strongly associated with the malignant phenotype of glioblastoma and become an attractive target for pharmacological intervention [14,15] Here we report that hCMEC express all known TGF-β receptors and coreceptors (Figure 1). Their constitutive VEGF release was largely unaffected by TGF-β receptor gene silencing except for the effect of TβRIII coreceptor silencing ( Figure 3A), suggesting that TβRIII exerts activity beyond TGF-β signaling in this paradigm [16]. Exogenous TGF-β induced VEGF mRNA synthesis and protein release in hCMEC in a time-and concentration-dependent manner ( Figure 2). The inhibition of VEGF release by ALK-5 gene silencing or pharmacological inhibition using SD-208 led to the identification of ALK-5 as the principal type 1 receptor mediating the VEGF induction by TGF-β (Figure 3), consistent with the observation that ALK-5 -/mouse endothelial cells obtained from E9.5 embryos shows reduced VEGF mRNA expression [17].
TGF-β has been shown to increase PlGF mRNA and protein release in keratinocytes [18] and in retinal epithelium [19]. We show that TGF-β induces PlGF expression and release in endothelial cells in normoxic (Figure 2, Figure 3 and Figure 6) and hypoxic conditions ( Figure 3). We further show that ALK5 mediates the effects of TGF-β on PlGF and that hypoxia may superinduce the PlGF system in the context of glioblastoma.
We observed that VEGFR-2 protein levels in hCMEC and GMEC decreased under hypoxia ( Figure 5). Up-regulation of VEGFR-1 and down-regulation of VEGFR-2 at the mRNA and protein levels have been reported previously in HUVEC exposed to hypoxia [20,21]. The promoter regions of VEGFR-1 that have HIF response elements (HRE) were identified whereas no obvious HRE on the VEGFR-2 promoter region was detected. It was speculated that the regulation of VEGFR-2 in hypoxia was mediated by an unindentified paracrine soluble factor [20] or through post-transcriptional mechanisms [21]. In response to VEGF-A, hypoxic human dermal microvascular endothelial cells (HDMEC) displayed a reduction in VEGFR-2 phosphorylation compared with normoxia, indicating reduced VEGF-Ainduced receptor activation in hypoxia [22].
VEGF and TGF-β signalling may induce a plethora of gene expression changes observed in glioblastoma vessels in vivo [26]. The dual effects of TGF-β on biological aspects of angiogenesis characterized here (Figure 8) makes the consequences of therapeutic targeting of either pathway difficult to predict. It must be acknowledged that therapeutic targeting of VEGF receptors by cediranib has failed in glioblastoma [27] whereas bevacizumab, while convincingly delaying disease progression, also did not impact overall survival [28,29]. Yet, it has remained uncertain to what extent these drugs truly reached their desired site of action which must include the tumor microenvironment. Blood brain barrier permeability may have limited the efficacy of cediranib and bevacizumab would a priori only be able to scavenge humoral VEGF, but probably not be able to prevent autocrine signaling that is likely to be relevant in glioblastoma, too [30]. Accordingly, control of VEGF synthesis and release is still a valid target in glioblastoma.

Isolation of human glioblastoma-derived endothelial cells
The net weight of the glioma tissue was measured followed by dissociation with 10 mg/ml collagenase/ dispase (11097113001, Roche, Basel, Switzerland) and rotation with gentleMACS C-tubes (130-093-237, Miltenyi biotech, Cologne, Germany). Erythrocytes were lysed by resuspending and incubating the cells in a 3:1 ratio of water:PBS. Cells were then counted and 20 μl of FcR blocking reagent and 20 μl of CD31 microbeads (130-091-935, Miltenyi biotech) were added per 10 7 cells. After 15 min rotating incubation at 4°C, the cells were passed through a 40 μm cell strainer to remove cell aggregates. Cells were then passed through an LS column placed in a magnetic field of a MACS separator. The cells that pass through (CD31-negative population) were again passed through another LS column to get a purer CD31-negative population. These cells were divided into 3 fractions and grown in Neurobasal medium (NBM) supplemented with B-27 (20 μl/ml) and Glutamax (10 μl/ ml) (Invitrogen, Carlsbad, CA, USA), FGF and EGF (20 ng/ml each; Peprotech, Rocky Hill, PA, USA), DMEM supplemented with 10% FCS and 1% L-glutamine (Gibco) or endothelial medium (EM). The CD31-positive cells which remained in the column were forcefully washed out and cultured in collagen type I-coated dishes until confluency and passaged once more before use. For ZHE-483-2, CD45-positive cells were depleted using LD columns (130-042-901, Miltenyi biotech), and then from the CD45-negative population, CD31-positive cells were isolated using the above procedure.

Real-time (RT) PCR
Total RNA was prepared with the NucleoSpin System (Macherey-Nagel, Düren, Germany) and cDNA was transcribed using Superscript II reverse transcriptase (Bio-rad, Munich, Germany). For real-time PCR, cDNA amplification was monitored using SYBR Green chemistry on the 7300 Real time PCR System (Applied Biosystems, Zug, Switzerland). The conditions for these PCR reactions were: 40 cycles, 95°C/15 sec, 60°C/1 min, using the primers specified in Supplementary Table 1. Arf1 transcript levels were used as a reference for relative quantification of mRNA expression levels using the ΔCT method because they were observed to be stable within each cell line under the experimental conditions reported here.

ELISA
hCMEC or GMEC were seeded at 2.6 × 10 4 cells/ cm 2 . Indicated treatments were performed in basal medium without serum and supplements for indicated time periods. Supernatants were harvested and cellular debris was removed by centrifugation 1200 rpm for 10 min. Supernatants were then concentrated by centrifugation at 4000 g for 30 min using the Amicon ® Ultra 3K device 3,000 NMWL (Millipore). Protein concentrations in supernatants were measured by BCA assay (Pierce Biotechnology, Rockford, IL, USA) and equal amounts of protein were used for VEGF-A ELISA (BMS277/2TEN, eBioscience, Vienna, Austria) and PlGF ELISA (RAB0404, Sigma). Pure serum-free medium without cells was used as a negative control for all ELISA.

Angiogenesis-related in vitro assays
hCMEC were cultured in full medium with TGF-β or SD-208 or both for 96 h before being detached using accutase (PAA Laboratories, Cölbe, Germany) and transferred to growth factor-reduced matrigel (356230, BD Biosciences, Franklin Lakes, NJ, USA) in 96 well plates at 15,000 cells/50 μl. Tube formation was assessed at 4, 12 and 24 h and photographs were taken using a Carl Zeiss microscope (Axiovert 100, Göttingen, Germany). Tube length (mm), number of tubes (n) and number of branching points (n) were calculated as average of 3 images per condition from independent experiments using the Adobe Acrobat X Pro (Adobe, Ireland).
For the sprouting assay, the conditions were same as for tube formation except that 2, 000 cells/50 μl were transferred to 96 well plates containing growth factorreduced matrigel and were left in the matrigel for 7 days. Then the number of sprouts formed by the cells was counted for 3 images per condition from independent experiments using ImageJ 1.40g software (NIH).
For analysis of cell death, cells were grown in 6-well plates and treated with TGF-β (10 ng/ml) for 72 h. Cells were then harvested, annexin (Anx) V-fluorescein isothiocyanate (1:100) and propidium iodide (PI; 50 μg/ ml) were added, and fluorescence in a total of 10,000 events (cells) per condition was recorded in the flow cytometer. Annexin V-or PI-positive cells were counted as dead cells, the remaining cells were designated the surviving cell fraction. Loss of viability was also confirmed by trypan blue dye exclusion [31].

Statistical analysis
Data are commonly derived from experiments performed at least twice in duplicates with similar results. Quantitative data are represented as mean ± standard deviation (SD) from the mean. Statistical significance was assessed using the student t-test, one-way ANOVA with Tukey's post hoc test or a two-way ANOVA wherever applicable. A p value below 0.05 was considered significant. All statistical analyses were performed using Prism 5 (GraphPad Software, La Jolla, CA, USA) at p < 0.05, p < 0.01, p < 0.001 or p < 0.0001.