Vascular Endothelial Growth Factor Receptor 2 (VEGFR-2) Plays a Key Role in Vasculogenic Mimicry Formation, Neovascularization and Tumor Initiation by Glioma Stem-like Cells

Human glioblastomas (GBM) are thought to be initiated by glioma stem-like cells (GSLCs). GSLCs also participate in tumor neovascularization by transdifferentiating into vascular endothelial cells. Here, we report a critical role of GSLCs in the formation of vasculogenic mimicry (VM), which defines channels lined by tumor cells to supply nutrients to early growing tumors and tumor initiation. GSLCs preferentially expressed vascular endothelial growth factor receptor-2 (VEGFR-2) that upon activation by VEGF, mediated chemotaxis, tubule formation and increased expression of critical VM markers by GSLCs. Knockdown of VEGFR-2 in GSLCs by shRNA markedly reduced their capacity of self-renewal, forming tubules, initiating xenograft tumors, promoting vascularization and the establishment of VM. Our study demonstrates VEGFR-2 as an essential molecule to sustain the “stemness” of GSLCs, their capacity to initiate tumor vasculature, and direct initiation of tumor.


Introduction
Glioblastoma (GBM) is one of the most vascularized tumors with increased microvasculature as a hallmark in pathology [1]. Previous studies of cancer vascularization focused mainly on the angiogenesis that is developed through sprouting from pre-existing vessels and the vasculogenesis that is established via recruitment of endothelial progenitor cells (EPCs) from the bone marrow [2]. Anti-vascular endothelial growth factor (VEGF) therapy has achieved limited efficacy in GBM due to rapidly developed resistance [3,4]. Therefore, better understanding of the mechanisms of tumor vascularization may improve the efficacy of antiantiangiogenic therapy.
Vascularization in brain tumors is a complex process that involves vessel co-option [5], angioblast vasculogenesis [6], intussusceptive microvascular growth [7] and vasculogenic mimicry (VM). VM defines the ability of highly invasive tumor cells to form fluid-conducing channels. The presence of VM in malignant tumors is associated with increased patient mortality [8,9]. Morphologically, the channels of VM consist of a basement membrane with lining of tumor cells in the external wall, without endothelial cells (ECs) on the inner wall despite the presence of blood flow in the channels [10]. Microarray analysis of VM-positive tissues from aggressive melanoma reveals increased expression of genes associated with undifferentiated embryoniclike phenotype [11], suggesting the participation of cancer stem cells (CSCs), a subpopulation of tumor cells that possess the capacity of self-renewal, multi-lineage differentiation, tumor initiation and resistance to chemo-or radio-therapy [12][13][14][15].
CSCs have been identified in brain tumors, including GBM [15][16][17][18]. Glioma stem-like cells (GSLCs) can be enriched from established cell lines or primary tumor tissues by using CD133 positive selection or generating neurospheres in serum-free media containing growth factors [12,[19][20][21]. We and others have demonstrated that GSLCs actively interact with vascular niche in the tumor and promote angiogenesis through the release of VEGF to recruit and stimulate the proliferation of host ECs [19][20][21]. Moreover, recent studies have shown that GSLCs may directly participate in the formation of tumor vessel by transdifferentiating into vascular EC-like cells [22,23]. The ability of GSLCs to acquire an EC-like phenotype and directly form tumor vasculature represents a novel mechanism of tumor vascularization [24,25]. In addition, GSLCs may play a role in the formation of VM, which is important for growing tumors to obtain nutrients during the early stage of progression [26,27]. In this study, we report that GSLCs expressed higher levels of VEGF receptor 2 (VEGFR-2), which was essential for controlling the selfrenewal, tumorigenicity and the formation of new vessels and VM in tumors by GSLCs.

Primary Human Glioma Samples
Glioma specimens were obtained from patients in the Department of Neurosurgery, Southwest Hospital, Third Military Medical University, Chongqing, and Tiantan Hospital, Beijing, TEM: Transmission electron microscopy of VM in GSLC initiated xenograft tumors. A vascular channel is lined by a thin basal lamina (red arrow) corresponding to the walls of the channel seen by conventional light microscopy. No endothelial cells line the tubule as compared to xenograft tumor formed by U87 cells (upper panel). Scale bar = 5 mm. (D) ECs in a human GBM section detected by anti-CD34 (black color). Vascular basement membrane with PAS staining (purple magenta) and tumor cells are labeled with anti-nestin antibody (brown color). a, Black arrows show tubular blood vessels stained with PAS. b, Double-staining of CD34 and PAS shows vessels (red arrows); PAS-positive tubules are lined by CD34 2 cells (black arrow). c, Tumor vessels containing red blood cells are positive for nestin but negative for CD34 (red arrow). Black arrow shows CD34 + blood vessels. China with written consent. The tumors were classified independently by at least two pathologists according to 2007 WHO classification of central nervous system tumors [30]. All experiments were approved by the institutional ethics committee.
Double immunofluorescence staining on frozen sections was used for detecting the co-expression of CD31 and GFAP or Laminin B2 (1:100 dilution. 610722. mouse monoclonal, BD Biosciences, USA). The cells were incubated with anti-human CD31 overnight at 4uC. After revealing CD31 by Alexa 549, the cells were incubated with goat anti-human GFAP at 4uC overnight. GFAP in cells was detected by FITC-conjugated goat anti-mouse IgG. Isotype-matched IgGs were used as negative controls. Similar procedures were used for double immunofluorescence staining of laminin B2 and GFAP, VEGF and VEGFR-2 as well as CD133 and VEGFR-2.
CD34-PAS or nestin-PAS dual-staining on paraffin sections of human glioma specimens [31], and xenograft tumors formed byU87 cells and GSLCs was performed as follows: 4-5 mm paraffin sections were stained with anti-CD34 (1:200 dilution. 550390, monoclonal mouse anti-human, BD Biosciences, USA) and anti-nestin antibodies. The sections were then stained with PAS and counterstained with Mayer's hematoxylin followed by examination under light microscopy. ALP staining was visualized with a DAB Kit (DakoCytomation) and horseradish peroxidase staining was visualized with AEC (Maixin-Bio, China). Sections were counterstained with hematoxylin. To define matrix-associated vascular channels, GBM tissues were stained with PAS and hematoxylin. The sections were examined under light microscopy.
For kinase phosphorylation, floating spheres were kept in the differentiation medium without bFGF and EGF for 24 h before stimulation with 10 ng/ml VEGF for different times. Floating spheres were then laid on poly-lysine coated cover glasses and fixed in 4% formaldehyde. After washing in 0.01 M PBS and blocked with 5% BSA or normal goat serum for 1 h at 37uC in a humidified chamber, the cells were incubated with anti-human phospho-VEGFR-2 (Tyr1175)(1:100 dilution. #2478, Cell Signaling Technology, USA ) followed by goat anti-rabbit IgG Cy3. Nuclei were counterstained with DAPI.

Analysis of Microvessel Density (MVD)
MVD was assessed with anti-CD31 immunostaining (1:100, BD Biosciences, USA), and a vessel was defined as any CD31 positive staining as previously described [32]. Briefly, after immunostaining, microscopic scanning of the entire tumor section under low power fields (6 40) of a laser scanning microscopy (Zeiss laser scanning microscopy 510 META, Carl Zeiss AG, Germany) was performed to identify hot spots that are the areas of highest neovascularization. Individual image was then taken under high power fields (6200) to count CD31 positive cells or cell clusters in a defined area. In order to determine the mean number of MVD within a tumor, the number of CD31 positive cells or cell clusters was determined in five different areas with adjacent fields by two independent pathologists who had no knowledge of the sample identity. Final results of MVD was calculated by Image J 1.43u (NIH, USA).

RT-PCR and Real-time Quantitative (q) RT-PCR
Total RNA was extracted from U87 GBM cells and GSLCs with an RNeasy mini kit (Qiagen, Valencia, CA). Primer sequences and PCR conditions are listed in Table S1. RT-PCR products were electrophoresed in 1% agarose and visualized by ethidium bromide staining. qRT-PCR was performed on an ABI PRISM 7900HT sequence detection system (Applied Biosystems, Norwalk, CA) using SYBR Green PCR Master Mix (Applied Biosystems). Primer sequences and qRT-PCR conditions are listed in Table S2.
Standard curves were generated and the relative amount of target mRNA was normalized against b-actin mRNA.

Western Blotting
GSLCs cultured in differentiation medium for 24 h were stimulated with VEGF. The cells were lysed on ice in 100 ml RIPA buffer with a protease and phosphatase inhibitor cocktail (Pierce Biotechnology, PA, USA). After centrifugation at 10,0006g at 4uC for 20 min, equal amount of proteins underwent 4% to 12% Bis-Tris SDS-PAGE (Invitrogen). The proteins were then blotted onto Immobilon-P PVDF membranes (Millipore, CT, USA). After blocking with 3% nonfat dry milk for 1 h at room temperature, the membranes were incubated with primary antibodies in PBS containing 0.01% Tween 20 overnight at 4uC. The phosphorylated proteins were visualized by staining the membranes with horseradish peroxidase-conjugated secondary antibody for 1 h followed by incubation with Super Signal Chemo-luminescent Substrate Stable Peroxide Solution (Pierce) and detection with BIOMAX-MR film (Eastman Kodak). When necessary, the membranes were stripped with Restore Western Blot Stripping Buffer (Pierce) and reprobed with antibodies against other proteins.

Chemotaxis
Cell chemotaxis was measured in 48-well chemotaxis chambers (NeuroProbe, Cabin John, MD). A 27 ml aliquot of chemoattractants was placed in the wells of the lower compartment and 50 ml of tumor cells (at 2610 6 /ml) were placed in the wells of the upper compartment. Two compartments of the chamber were separated by a 10 mm pore-sized polycarbonate filter (GE Osmonics Labstore) coated with 50 mg/ml collagen type I (BD). After incubation at 37uC for 240 min, the filters were removed, stained, and cells that migrated across the filters were counted under light microscopy. The results were expressed either as the mean number (6 SE) of migrated cells in three high-powered fields (4006) in triplicates or as chemotaxis index that represents the fold increase in cell migration in response to stimulants over medium control.

In vitro Tubule Formation
Vasculogenic tubule formation by GBM cells was tested on the reduced growth factor matrix (Matrigel, BD). Eight-chamber slides (Lab-Tek, 155411) were pre-coated with Matrigel (0.1 ml/well, BD) and incubated at 37uC for 30 min. Floating spheres of GSLCs were resuspended in the EC basal medium-2 (EBM-2, Lonza) containing 2% FCS with or without 10 ng/ml VEGF (PeproTech) and 1 mg/ml VEGFR-2 blocking antibody (sc-74002, Santa Cruz Biotechnology, USA). After 4 days, the tubules formed by GSLCs were counted under light microscopy. Tubule formation was analyzed by Image J software [33].

Transmission Electron Microscopy (TEM)
U87 and GSLC xenograft tumors were fixed in 2.5% buffered glutaraldehyde and were post-fixed in a solution of 1% osmium tetroxide, dehydrated, and embedded. Thin sections were stained with uranyl acetate-lead citrate and examined with a Hitachi H-7500 transmission electron microscope.

Statistical Analyses
All experiments were conducted at least three times with reproducible results. Results presented were from representative experiments. Where applicable, data were expressed as the mean 6 SE. Statistical significance of the difference between testing and control groups was analyzed with SPSS10.0 software. When two groups were compared, the unpaired Student's t test was used. When multiple groups were evaluated, one-way ANOVA was used. P,0.05 was considered statistically significant.

GSLC-derived EC-like Cells form VM and Neovasculature in Xenograft Tumors and Primary Human GBM
Our previous studies have demonstrated the existence of GSLCs in the human GBM cell line U87 [19][20][21]. GSLCs were enriched in serum-free culture medium as evidenced by sphere formation (Fig. S1A). GSLCs obtained were positive for CD133 and a neural precursor marker nestin as well as multiple stem cell markers including Oct4, Nanog and Notch1. Additionally, GSLCs cultured in differentiation medium expressed GFAP for glial cells, neuronal class III beta-tubulin and microtubule associated protein 2 (MAP2) for neurons (Fig. S1B, C). Moreover, GSLCs exhibited markedly increased capacity to initiate tumors when implanted s.c. in nude mice. Also, co-expression of CD133 and VEGFR-2 was detected in the tumors formed by U87 GBM cells and GSLCs ( Fig. S2A-C). However, GSLC-derived tumors contained increased number of CD133 + /VEGFR-2 + cells as compared to the tumors formed by parental U87 GBM cells (Fig. S2C, right). Therefore, U87 GBM cells contain GSLCs with typical stem cell properties and the expression of an EC marker VEGFR-2.
We and others have shown that CSCs produce higher levels of VEGF [6,[16][17][18][19] to induce the formation of capillary-like structures by ECs [16]. To investigate the capability of GSLCs to initiate new blood vessels in tumor, we compared the sources of ECs in both GSLC-and the parental U87 GBM cell-derived xenograft tumors. Figure 1 shows that both U87-and GSLCderived tumors contained murine EC-like cells as stained by an anti-mouse CD31 antibody (Fig. 1A, left). Both U87-and GSLCderived tumors also contained human EC-like cells as stained by an anti-human CD31 antibody (Fig. 1A, middle). However, GSCLderived tumors contained a higher number of human EC-like cells than parental U87-derived tumors. In GSLC-derived tumors, 34% of CD31 + cells were of the human origin, whereas almost all CD31 + cells in the tumors formed by parental U87 GBM cells were from mice (Fig. 1A, right).
The presence of human EC-like cells in GSLC-derived tumors was confirmed by co-staining of tumor tissues in which a proportion of CD31 + cells also expressed the glial marker GFAP (Fig. 1B, top), consistent with the findings in primary human GBM [34]. We further observed the co-expression of human laminin B2 and GFAP in GSLC-derived tumors (Fig. 1B, bottom). These observations suggest the presence of GSLC-derived VM in GSLCxenograft tumors. To confirm this, we detected red blood cells containing PAS-positive tubular structures lined by GBM cells with mitosis in GSLC-derived tumors. Moreover, to definitely demonstrated the formation of VM formed by tumor cells, by transmission electron microscopy we show in Fig1 C that VM in GSLC formed xenograft tumors was lined by mural cells of tumor, but not endothelial origin. In contrast, parental U87 cell-derived tumors contained vessels with PAS-positive reaction only in the basement membrane, which were coated with endothelial cells as defected in both IHC and TEM (Fig. 1C).
To identify VM in primary human GBM sections, we used PAS and antibodies against CD34 and nestin to stain the tumor blood vessels. PAS-positive lumina were detected in human GBM sections (Fig. 1Da, black arrows). Tumor vessels were positive for both CD34 and PAS (Fig. 1Db, red arrows). In the same tumor sections, we detected PAS-positive tubular structures lined by CD34 2 cells in the luminal surface (Fig. 1Db, black arrow). To examine whether these CD34 2 cells were tumor cells, we doublestained CD34 and nestin. Figure 1Dc shows that CD34 2 celllined tubular structures were nestin-positive and contained red blood cells (Fig. 1Dc, red arrow). In addition, PAS-positive tubular structures containing red blood cells were lined by nestin positive cells in the luminal surface (Fig. 1Dd, black arrow). The interface of tumor cell-lined vessels and EC-lined vessels was also found in human GBM sections (Fig. 1E), with some tubular structures formed by both tumor cells (red arrow, brown color) and ECs (black arrow, black color). Thus, non-EC-lined vessels in GSLC-derived murine xenograft tumors are VM formed by tumor cells derived from human GSLCs and such VM can also be found in human primary GBM.

GSLCs Preferentially Express VEGFR-2
We next attempted to identify genes that may contribute to the VM-forming and vasculogenic property of GSLCs. Several genes involved in vasculogenesis and VM formation were expressed in GSLCs, including VEGFR-2, VE-cadherin, EphA2 and laminin 5c2 ( Fig. 2A). In particular, VEGFR-2 was more highly expressed in GSLCs at both mRNA and protein levels as compared to the parental U87 GBM cells ( Fig. 2A, B). Notably, neither GSLCs nor U87 GBM cells expressed VEGFR1 ( Fig. 2A, insert). Stimulation of GSLCs with VEGF upregulated the expression of the genes for VEGFR-2 (Fig. 2C) and VE-cadherin (Fig. 2D), consistent with the expression of both CD133 and VEGFR-2 in cells of primary human grade IV (12/15), III (6/10), II (1/7) glioma specimens (Fig. 2E). Therefore, VEGFR-2 is expressed by GSLCs derived from both GBM cell lines and primary GBM tissues.

VEGFR-2 Expressed by GSLCs is Functional
VEGFR-2 expressed by GSLCs is functional, because VEGF induced significant chemotaxis of GSLCs and the response was more potent than the parental U87 GBM cells that expressed lower level of VEGFR-2 (Fig. 3A, left). Also, a VEGFR-2 mAb significantly reduced GSLC chemotaxis response induced by VEGF (Fig. 3A, right). In addition, GSLCs formed a higher number of tubules than the parental U87 GBM cells in the absence or presence of VEGF (Fig. 3B). The formation of tubulelike structures by GSLCs in response to VEGF was inhibited by the VEGFR-2 mAb (Fig. 3B, right and C). Further, VEGF stimulation increased the expression of VEGFR-2 by GSLCs (Fig. 3D). These results indicate an important role of VEGFR-2 in mediating the VM-forming and vasculogenic potential of GSLCs.
Since VEGFR-2 phosphorylation on key tyrosine residues is necessary for activation of down-stream signaling molecules, we examined the phosphorylation of Y1059 and Y1175 in the Cterminal domain of VEGFR-2 [35]. There was a constitutive level of Y1059 phosphorylation in VEGFR-2 in GSLCs, which was not affected by VEGF stimulation [data not shown], suggesting selfphosphorylation at Y1059. In contrast to Y1059, Y1175 in VEGFR-2 was rapidly and transiently phosphorylated when GSLCs were stimulated with VEGF (Fig. 4A). The phosphorylation of Y1175 in VEGFR-2 was associated with increased activation of ERK1/2 and p38 MAPKs as well as AKT (Fig. 4B,  C, D), which are known to be coupled to VEGFR-2 signaling pathway. The possibility that VEGFR-2 on GSLCs may be activated by ligand present in the tumor environment was confirmed in the xenograft tumors initiated by GSLCs in which tumor cells expressed high levels of both VEGFR-2 and VEGF (Fig. 4E). These results indicate that functional VEGFR-2 expressed by GSLCs mediates important GSLC functions such as increased cell migration and transdifferentiation into cells to form VM in tumor.

VEGFR-2 is Critical for the Self-renewal and Tubule Formation by GSLCs
To further clarify the function of VEGFR-2 in GSLCs, we used shRNA to knockdown VEGFR-2. Figure 5A shows that transduction of VEGFR-2 shRNA into the GBM cell line U87 reduced VEGFR-2 mRNA expression. Knockdown of VEGFR-2 markedly reduced the chemotaxis of U87 GBM cells in response to VEGF (Fig. S3B) and also reduced the proliferation of U87 GBM cells in vitro as compared with the wide type (WT) and mock knockdown cells (Fig. S3C). When subcutaneously injected into nude mice, VEGFR-2 knockdown GBM cells showed a markedly reduced tumorigenicity (Fig. S3D) with increased mouse survival (Fig. S3E).
Since we have shown that VEGFR-2 is more highly expressed in GSLCs enriched from parental U87 cells, we hypothesized that the marked suppression of the growth and tumorigenicity of U87 GBM cells by VEGFR-2 knockdown may be caused by a selective effect on GSLCs. We therefore examined the capacity of VEGFR-2 knockdown U87 GBM cells to form floating spheres, an indication of GSLC enrichment and self-renewal. The number of floating spheres derived from VEGFR-2 knockdown U87 GBM cells was significantly reduced, indicating that VEGFR-2 knockdown impaired the enrichment and self-renewal of GSLCs (Fig. 5A). Figure 5B shows that VEGFR-2 expression was absent in the GSLCs enriched from VEGFR-2 knockdown parental U87 cells. However, these GSLCs retained the expression of stem cell markers CD133 and Oct4 (Fig. 5C, top). When cultured in differentiation media, the cells were capable of converting into GFAP + astrocytic and MAP-2 + neuronal cells (Fig. 5C, bottom).
Despite the expression of stem cell markers and the capacity of multi-lineage differentiation, VEGFR-2 knockdown reduced the expression of VM-associated marker VE-cadherin by GSLCs (Fig. 5D) and markedly inhibited their formation of tubules (Fig. 5E, top and Fig. 5F) in the presence or absence of VEGF stimulation (Fig. 5E, F). This is in contrast to VEGFR-2 expressing GSLCs, which maintained the capacity to form tubules and were highly responsive to VEGF stimulation (Fig. 5E, F). These results indicate that VEGFR-2 is essential for the self-renewal and formation of tubular structures by GSLCs.

VEGFR-2 is Required for New Vascularization and Tumor Initiation by GSLCs
The reduced self-renewal and tubule formation by VEGFR-2 knockdown GSLCs and the reduced tumorigenicity of VEGFR-2-knockdown U87 parental cells led us to further test the tumorigenicity of GSLCs and their contribution to tumor vascularization in vivo. As compared to VEGFR-2-positive GSLCs, VEGFR-2-knockdown GSLC-formed tumors grew much more slowly in nude mice with prolonged mouse survival (Fig. 6A, B). Additionally, VEGFR-2 knockdown GSLC-derived xenograft tumors contained fewer vessels formed by cells of human origin as demonstrated by observations in which human CD31-positive vessels were hardly visible in the tumors formed by VEGFR-2 knockdown GSLCs. Rather, these tumors contained a large number of murine CD31-positive vessels (Fig. 6C). In contrast, VEGFR-2-positive GSLC-formed tumors grew much more rapidly and contained vessels stained for both human and murine CD31 (Fig. 6A-C). In addition, CD133positive cells were markedly reduced in the tumors formed by VEGFR-2-knockdown GSLCs as compared to the tumors derived from VEGFR-2 positive GSLCs (Fig. 6D). These results indicate that VEGFR-2 actively participates in vasculogenesis and self-renewal of GSLCs in xenograft tumors.
To evaluate the role of VEGFR-2 in VM formation by GSLCs, we stained human GFAP and lamininB2 in tumors formed by GSLCs with or without VEGFR-2 shRNA. VM within xenografts were measured by quantitative image analysis [36] to assess the density of lamininB2 immunoreactivity per cross-section area. In tumors formed by GSLCs with VEGFR2 shRNA, VM was inhibited by 60% (Fig. 6E) compared with tumors originated from mock transfected GSLCs [1.2%60.5% versus 3.0%60.9% (n = 5)]. These results indicate that VEGFR-2 is critical for VM formation by GSLCs and the rapid growth of tumors initiated by GSLCs.

Discussion
In this study, we demonstrated a critical role of VEGFR2 in the formation of neovascularization such as VM and new vessels in growing tumors initiated by GSLCs both in xenograft models and in primary GBMs. We observed the presence of tubular vessels formed by tumor cells in human GBMs in which the interface of tumor cell-lined vessels and EC-lined vessels was detected. Tumor cells located in the walls of blood vessels became a part of the vessel surface with the remaining part covered by ECs, a typical structure of ''mosaic vessels'' [37]. Such ''mosaic vessels'' have been interpreted as the result of incorporation of the vessel wall by tumor cells [37]. Our study suggests that such tumor cells are likely originated from CSCs with the capability of multi-lineage differentiation including transdifferentiation into EC-like cells that are directly involved in the different phases of neovascularization in growing tumors. In addition to being closely located near microvessels to promote angiogenesis by producing VEGF [19,21], our present study shows that GSLCs express ECassociated genes coding for EphA2 (ephrin receptor), VEcadherin, laminin5c2, and VEGFR-2. These molecules are not only required for the formation and maintenance of blood vessels [38,39], but are also critical for nervous system development [40], in which neural and vascular guidance pathways share common signaling mechanisms [41]. Therefore, GSLCs with neurodevelopmental potential may utilize common neural and vascular patterning properties to develop blood vessel networks in tumors.
Our study also provided the evidence for the capacity of CD133 + GSLCs to form VM and EC-lined vessel by preferentially expressing VEGFR-2, which is required for the self-renewal of GSLCs and tumor initiation. VEGFR-2 (also known as KDR or Flk-1) is a high-affinity receptor for VEGF and is the most important molecule for angiogenesis and vasculogenesis by mediating almost all cellular response to VEGF [42]. Moreover, VEGFR-2 is not only expressed in blood and lymph vessel ECs to promote their recruitment and proliferation [43], but is also expressed in normal stem cells to mediate vasculogenesis [44,45]. In addition, VEGFR-2 is expressed in tumor cells from patients with colorectal cancer (CRC) [46] and its activation by VEGF in an autocrine and paracrine manner promotes tumor growth [35].
We found that VEGFR-2 is expressed by human GBM cells with a much higher level in CD133 + GSLCs, suggesting a link between VEGFR-2 and the biological behavior of GSLCs. Indeed, it has been demonstrated that VEGFR-2 is a critical GSLCdependent VM biomarker for predisposition of human malignant gliomas to increased patient mortality [9]. Furthermore, functional VEGFR-2 is required by GSLCs not only for VM formation and transdifferentiation into EC, but also for more rapid tumor growth. We speculated that the VM may provide the sufficient nutrition to promote the tumor growth in early date before the appearance of EC-lined vessels. This provides a potential explanation for previously established correlation between the presence of VM in GBM and poor patient prognosis [9].
The demonstration that VEGFR-2 is required for the growth of tumors initiated by GSLCs defines a novel function of VEGFR-2 to control a variety of GSLC properties including self-renewal, multi-lineage differentiation, tumor initiation and angiogenesis [19][20][21]47,48]. Such properties are also used by GSLCs for evasion of host anti-tumor immunity [49,50]. Although it is believed that tumor neovasculature is derived from pre-existing blood vessels or bone-marrow-derived circulating EC progenitors [51,52], the frequency of BM-derived ECs is very low in tumor neovasculature [53], suggesting that tumor vasculature may be derived either from adjacent existing vessels or from a subpopulation of tumor cells with EC trans-differentiation potential [10,54]. Our study provides direct evidence that VEGFR-2 + / CD133 + GSLCs serve as progenitors for VM formation and EClined vessels. This conclusion is supported by the facts that VEGFR-2 + /CD133 + GSLCs are capable of differentiating into EC and EC-like cells; that some CD133 + GSLCs constitutively express both VEGFR-2 and VEGF, which promote their differentiation into EC-like cells in an autocrine or paracrine manner; that blood vessels and VM in xenograft tumors initiated by VEGFR-2 + /CD133 + GSLCs are mainly derived from transplanted human tumor cells; and that the vasculature in xenograft tumors formed by GSLCs containing VEGFR-2 shRNA was formed mostly by mouse-derived ECs.
In the final stage of the preparation of our manuscript, a same research group reported the capacity of glioblastoma cells, and GSLC to form VM in tumors [26,27], which was independently confirmed by our study. The observation of GSLCs as progenitors for VM formation may explain the reason for generally ineffective anti-angiogenic cancer therapy in the clinic [55]. Recently, the proof-of-the principle has been established for GSLCs as the potential therapeutic targets. The discovery of the novel role of VEGFR-2 in GSLCs for tumorigenicity and vasculogenesis is important for design of additional GSLC-targeting therapies. In this respect, shRNA knockdown of VEGFR-2 reduced the selfrenewal, tumor initiation and vascularization, in particular VM formation by GSLCs, has shown promising utilization of such approach for clinical therapies. In addition, the signaling molecules coupled to VEGFR-2 in GSLCs are also relevant to therapeutic design. For instance, VE-cadherin is an EC-specific adherence junction component involved in the formation of complex with VEGFR-2. Elimination of VE-cadherin disrupts VEGF-induced signal transduction necessary for EC survival [56]. In our study, VEGFR-2 knockdown in GSLCs diminished the expression of VE-cadherin in association with reduced formation of VM. Further studies focusing on VEGFR-2-dependent protumorigenic and vasculogenic interactions of GSLCs with VEGFproducing malignant or nonmalignant host cells in the tumor microenvironment will provide insight into the mechanistic basis of tumor development and therapy design.  Chemotaxis of U87 GBM cells with VEGFR-2 knockdown in response to 10 ng/ml VEGF (left) or to different doses of VEGF (right). * Indicates significantly reduced chemotaxis of VEGFR-2 knockdown U87 cells in response to VEGF as compared with mock cell chemotaxis (p,0.05). (C) Proliferation of U87 GBM cells with VEGFR-2 shRNA. * Indicates significantly reduced proliferation of U87 cells with VEGFR-2 shRNA. (D) Xenograft tumor growth in nude mice. U87 GBM cells with VEGFR-2 shRNA (2610 6 ) or U87 GBM cells with mock shRNA (2610 6 ) were subcutaneously injected into nude mice (5 mice/group). Tumor growth was monitored up to 52 days. *Indicates significantly reduced growth of tumors formed by U87 cells containing VEGFR-2 shRNA (p,0.05). (E) Survival rate of mice with xenograft tumors derived from U87 GBM cells with VEGFR-2 shRNA or mock shRNA (5 mice/group). * Indicates significantly prolonged survival of mice bearing VEGFR-2 knockdown U87 cells (p,0.05).

(TIF)
Table S1 Primers used for RT-PCR. RT-PCR was performed with 0.3 mg total RNA for each sample. The conditions were 30 min at 50uC for reverse transcription, 5 min at 95uC for denaturation and then followed by amplification and extension. (DOC) Table S2 Primers used for qRT-PCR. All reactions were completed in a 20 ml reaction volume in triplicate and the amplification consisted of 15 min at 42uC for reverse transcription, 5 min at 95uC for denaturization followed by 30 s at 95uC, 1 min at 57uC and 1 min at 72uC for 35 cycles and 10 min at 72uC for extension. (DOC)