OUP user menu

A new alternative mechanism in glioblastoma vascularization: tubular vasculogenic mimicry

Soufiane El Hallani, Blandine Boisselier, Florent Peglion, Audrey Rousseau, Carole Colin, Ahmed Idbaih, Yannick Marie, Karima Mokhtari, Jean-Léon Thomas, Anne Eichmann, Jean-Yves Delattre, Andrew J. Maniotis, Marc Sanson
DOI: http://dx.doi.org/10.1093/brain/awq044 973-982 First published online: 7 April 2010


Glioblastoma is one of the most angiogenic human tumours and endothelial proliferation is a hallmark of the disease. A better understanding of glioblastoma vasculature is needed to optimize anti-angiogenic therapy that has shown a high but transient efficacy. We analysed human glioblastoma tissues and found non-endothelial cell-lined blood vessels that were formed by tumour cells (vasculogenic mimicry of the tubular type). We hypothesized that CD133+ glioblastoma cells presenting stem-cell properties may express pro-vascular molecules allowing them to form blood vessels de novo. We demonstrated in vitro that glioblastoma stem-like cells were capable of vasculogenesis and endothelium-associated genes expression. Moreover, a fraction of these glioblastoma stem-like cells could transdifferentiate into vascular smooth muscle-like cells. We describe here a new mechanism of alternative glioblastoma vascularization and open a new perspective for the antivascular treatment strategy.

  • glioblastoma
  • angiogenesis
  • vasculogenic mimicry
  • stem cell


Glioblastomas are the most frequent and malignant primary brain tumours in adults and have a poor prognosis despite surgery and conventional radio-chemotherapy. Histologically, glioblastomas are highly angiogenic and characterized by microvascular proliferations (Louis et al., 2007). Anti-vascular endothelial growth factor therapy has had significant efficacy in glioblastomas with nearly 50% of responders, but acquired antiangiogenic resistance may occur (Vredenburgh et al., 2007; Kreisl et al., 2009). A better understanding of tumour vascularization is needed to optimize antivascular therapy. Most research has focussed on the role of angiogenesis involving endothelial cell sprouting and recruitment of new vessels into a tumour from pre-existing vasculature (Ausprunk and Folkman, 1977; Brat and Van Meir, 2001). However, it is known that alternative vascularization mechanisms may occur in brain tumours, resulting in vascular co-option of vessels (Holash et al., 1999), angioblast vasculogenesis (Santarelli et al., 2006), intussusceptive microvascular growth (Kurz et al., 2003) and vasculogenic mimicry. The term vasculogenic mimicry describes the formation of fluid-conducting channels by highly invasive and genetically dysregulated tumour cells. Two distinctive types of vasculogenic mimicry have been reported in tumours. Vasculogenic mimicry of the patterned matrix type in no way resembles blood vessels morphologically: the channels are composed of a basement membrane, lined by tumour cells in their external superficies and no endothelial cells are found on their inner wall despite blood plasma and red blood cells flowing through the channels (Maniotis et al., 1999). In contrast, vasculogenic mimicry of the tubular type may be confused morphologically with endothelial cell-lined blood vessels. Independent observers have previously reported the existence of non-endothelial cell-lined blood channels in tumours (Timar and Toth, 2000; Shirakawa et al., 2002; Folberg and Maniotis, 2004; Van der Schaft et al., 2005) and identified tumour cells as the lining cells of their luminal surface (Liu et al., 2002). Vasculogenic mimicry of the patterned matrix type results in the ability of tumour cells to express endothelium-associated genes that are also involved in embryonic vasculogenesis (Seftor et al., 2002). Such plastic properties could be associated with cancer stem cells, a subpopulation of undifferentiated tumour cells that present with a marked capacity for proliferation, self-renewal, multiple lineage differentiation and tumour initiation (Bonnet and Dick, 1997).

Cancer stem cells have been identified in brain tumours, including glioblastomas (Singh et al., 2003; Galli et al., 2004; Yuan et al., 2004). The CD133 stem-cell marker has been used prospectively to isolate a small fraction of glioblastoma cells with enhanced stem-cell properties. Vasculogenic mimicry of the patterned matrix type was previously reported in human glioblastoma tissues (Yue et al., 2005) and human glioma cell-line xenografts (Niclou et al., 2008). However, occurrence of vasculogenic mimicry of the tubular type in glioblastomas is still unknown and the vasculogenic capacity of brain tumour stem cells has not been proven. Here, we immunohistochemically analysed human glioblastoma tissues and found non-endothelial cell-lined vessels in a subset of glioblastomas. Combining fluorescent in situ hybridization and immunophenotyping, we confirmed that these non-endothelial cell-lined vessels are formed by primary tumour cells. We demonstrated in vitro that CD133+ glioblastoma stem-like cells (GSC) were capable of vasculogenesis and vascular smooth muscle-like cell differentiation.

Materials and methods

Glioblastoma tissue preparation

Formalin-fixed, paraffin-embedded tissue sections (5 µm) from 40 glioblastomas (World Health Organization classification of brain tumours) (Louis et al., 2007) were de-paraffinized twice with xylene. The slides were subsequently hydrated in a series of ethanol solutions (100, 90 and 70%), washed with phosphate-buffered saline and treated with antigen retrieval solution (citrate buffer pH9.0; Dako Cytomation, France) at 96°C for 20 min and then cooled at room temperature for 15 min. Clinical characteristics and sample size of the glioblastoma tissues are shown in Table 1.

View this table:
Table 1

Clinical characteristics of glioblastoma tissues

Median age (range)61.2 (38–85)
M/F26/14 = 1.8
Median Karnofsy performance Status (range)80 (30–100)
Median survival (range)13.3 months (0.5; 185.4)
Chemotherapy and radiotherapy19
Supportive care7
Partial removal9
Complete removal31


For immunochemistry staining we used the Envision + Dual Link System-HRP (DAB +) (Dako Cytomation, France). Slides were incubated for 10 min with 3% hydrogen peroxide in distilled water to block the endogenous peroxidase activity. After three washes with phosphate-buffered saline, they were incubated for 30 min with 5% bovine serum albumin in phosphate-buffered saline to block non-specific antibody binding. Serial sections were then covered with monoclonal mouse anti-human CD34 or collagen-IV antibody (IgG1, 1:50, Dako Cytomation, France), incubated for 1 h at room temperature, washed in phosphate-buffered saline three times, and incubated subsequently in Envision Polymer for 30 min. Substrate-chromogen solution (DAB) was applied for 5 min and reactions were stopped by distilled water washes. Finally, slides were counterstained with Mayer’s haematoxylin for 1 min and cover-slipped with a permanent mounting medium. To highlight matrix-associated vascular channels of glioblastomas, tissues were stained following the Periodic Acid-Schiff (PAS) procedure before counterstaining with Mayer’s haematoxylin. These sections were viewed under direct light microscopy.

Fluorescent in situ hybridization and immunofluorescence

Sections were immersed for 2 min each in 70, 80 and 100% ethanol and air-dried on a hot plate (45°C) prior to denaturation at 72°C in 70% formamide/2× saline sodium citrate for 5 min. Slides were then placed in 70, 80 and 100% ethanol for 1 min each and then air-dried on a hot plate prior-to-probe application. The EGFR fluorescent in situ hybridization probe mix (Dako Cytomation, France) was used according to the manufacturer’s instructions. The Texas Red-labelled DNA probe that binds to the EGFR gene on chromosome 7q11.2 was incubated at 82°C for 5 min and overnight in a humid chamber at 45°C. Following post-hybridization washes with 2× saline sodium citrate, sections were incubated in blocking solution (2% foetal calf serum, 1% bovine serum albumin, 0.1% Triton X-100 and 0.05% Tween-20 in phosphate-buffered saline) for 1 h. Monoclonal mouse anti-human α-SMA antibody (IgG2a, 1:50, Dako Cytomation, France) was applied to sections and left overnight at 4°C. Sections were washed three times with phosphate-buffered saline and subsequently incubated with Alexa 488-conjugated goat anti-mouse antibody (1:1000; Molecular Probes, Invitrogen, France) as secondary reagent. Finally, slides were counterstained with 4′,6-diamidino-2-phenylindole (DAPI) (Sigma-Aldrich, France), washed in phosphate-buffered saline and mounted in antifade medium Fluoromount-G (Interchim, France). This method provided clear immunofluorescence staining and fluorescent in situ hybridization signals. Slides were examined under a Zeiss AxioImager.Z1 microscope. Fluorescence images were captured using AxioCam MRm camera and analysed with AxioVision Rel. 4.6 software (Carl Zeiss).

Cell cultures

Culture of primary glioblastoma stem-like cells and sphere forming assay

Glioblastoma samples were provided by the local Department of Neurosurgery from patients who had given written and informed consent, as approved by the local research ethics boards at the Salpetriere Hospital. Histological analyses were done by the Department of Neuropathology. Samples were washed with Hanks’ balanced salt solution (Invitrogen, France), dissected, sectioned and enzymatically dissociated with both 5 mg/ml of Trypsin (Sigma-Aldrich, France) and 200 U/ml of DNAse (Sigma-Aldrich, France) for 10 min at 37°C. Erythrocytes were lysed using NH4Cl. The cells were seeded into T75 flasks at 10 000 cells/cm2. The culture medium (stem cells medium) consisted of Dulbecco’s modified Eagle’s medium/F12 (Invitrogen, France) supplemented with 20 ng/ml of epidermal growth factor, 20 ng/ml of basic fibroblast growth factor (both from Sigma-Aldrich, France), B27 (1:50; Invitrogen, France) and 1% penicillin–streptomycin. Cultures were incubated in 5% CO2 at 37°C. After 3 days of culture, CD133 Microbead Kit (Miltenyi Biotech, France) was used to isolate the CD133+ tumour cell population according to the manufacturer’s instructions. Sorted cells were resuspended in neurosphere medium and maintained in 5% CO2 at 37°C. Formed primary spheres were harvested, dissociated enzymatically into single cells and plated at a density of 5000 cells/cm2 in the presence neurosphere medium. Cultures were fed by changing half of the medium every 3 days. Subsphere-forming assay (also called passage) was repeated every 10 days.

Human cerebral microvascular endothelial cells culture

Immortalized human cerebral microvascular endothelial cells were obtained from Dr Pierre Olivier Couraud (Institut Cochin, France) and cultured with endothelial basal medium EBM-2 (Lonza, France) supplemented with 5% foetal calf serum, 1 ng/ml basic fibroblast growth factor, 10 mM HEPES and 1% penicillin–streptomycin.

Human vascular smooth muscle cells culture

Immortalized human vascular smooth muscle cells were obtained from Dr Luc Mouthon (Institut Cochin, France) and cultured with Dulbecco’s modified Eagle’s medium supplemented with 10% foetal calf serum, 10 mM HEPES and 1% penicillin–streptomycin.

U87 cell-line culture

U87 cells were purchased from ATCC and cultured with Dulbecco’s modified Eagle’s medium supplemented with 10% foetal calf serum and 1% penicillin–streptomycin.

Primary rat astrocytes culture

Primary rat astrocytes were prepared and cultured as previously described (Etienne-Manneville, 2006).

Lentiviral infection

Green fluorescent protein lentivirus vector construction and virus production were performed by Dr Philippe Ravassard (Pierre and Marie Curie University) as previously described (Russ et al., 2008). Dissociated GSCs were infected with green fluorescent protein-expressing retroviral vector. Labelled cells were selected using a fluorescence-activated cell sorter (FACS Aria, BD Biosciences). The efficiency of transduction was over 80%.

In vitro angiogenesis assay

Vasculogenic tube formation was tested using a commercial Matrigel assay kit (BD Biosciences, France). Green fluorescent protein-expressing GSC were dissociated and resuspended at 6 × 104 cells/ml in endothelial basal medium containing 2% foetal calf serum. Wells of 24-well tissue culture plates were coated with Matrigel (0.1 ml/well, BD Biosciences, France) which was allowed to polymerase at 37°C for 30 min. The indicated cell suspension was then plated at 0.5 ml/well onto the surface of Matrigel and incubated at 37°C. Cells were photographed using Nikon Eclipse TE2000U fluorescence inverted microscope.

Differentiation assay

For neural differentiation, primary glioblastoma spheres were plated onto sterile multiwell glass slide coated with poly-l-ornithine (Sigma-Aldrich, France) in neurosphere medium lacking epidermal growth factor and basic fibroblast growth factor but supplemented with 10% foetal bovine serum. Cells were fixed after 7 days of differentiation culture with 4% paraformaldehyde for 15 min, permeabilized with phosphate-buffered saline/0.1% Triton X-100, blocked with phosphate-buffered saline/3% bovine serum albumin for 20 min and immunostained for 1 h with primary antibodies against nestin (1:200; mouse monoclonal IgG1; Santa Cruz Biotechnology, Germany), Sox2 (1:200; mouse monoclonal IgG2a; R&D Systems, France), glial fibrillary acidic protein (1:400; rabbit polyclonal; Dako Cytomation, France) and neuronal class III beta-tubulin-1 (1:500; mouse monoclonal IgG2a; Covance, France). After washes, appropriate secondary antibodies were incubated for 1 h (1:1000; Alexa 594 goat anti-mouse IgG1, Alexa 488 goat anti-mouse IgG2a, Alexa 488 goat anti-rabbit and Alexa 594 goat anti-mouse IgG2a from Molecular Probes, Invitrogen, France). For smooth muscle differentiation, we plated dissociated green fluorescent protein-expressing GSC in Dulbecco’s modified Eagle’s medium supplemented with 10% foetal bovine serum. After 7 days of differentiation culture, cells were fixed sequentially with 4% paraformaldehyde (15 min at room temperature) and methanol (10 min at –20°C). Monoclonal mouse anti-human α-SMA immunostaining was performed as described above. DAPI (Sigma-Aldrich, France) was used for nuclei staining. Slides were mounted in antifade medium Fluoromount-G (Interchim, France) and examined under a Zeiss AxioImager.Z1 microscope.

Western blot

Proteins were run on a Laemmli-type 10% sodium dodecyl sulphate–polyacrylamide gel electrophoresis and electroblotted onto nitrocellulose. Blocking of nitrocellulose and immunostaining was performed in a buffer containing 10 mM Tris (pH 7.4), 150 mM NaCl, 5% dry milk and 0.1% Tween-20. The membranes were stained with either the antibodies directed against Nestin (1:5000; VPA5922; AbCyst; France), glial fibrillary acidic protein (1:500; C-19; Santa Cruz Biotechnology, Germany), neuronal class III beta-tubulin-1 (1:5000; Covance; France) or Tubulin (1:2000; MCA77; Serotec, France). Thereafter, the nitrocellulose sheets were incubated with the appropriate secondary antibody horseradish–peroxidase conjugate. The antibody binding was detected using Pierce ECL western blotting substrate according to the manufacturer's instructions (Thermo Scientific, France).

Reverse transcriptase–polymerase chain reaction

Total RNA was extracted from GSC-A, GSC-B, human cerebral microvascular endothelial cells and human vascular smooth muscle cells using RNable (Eurobio, France) and verified by electrophoresis (Agilent 2100 Bioanalyzer). cDNA was synthesized with 200 units of Moloney Murine Leukemia Virus (M-MLV) Reverse Transcriptase (Invitrogen, France) in 15 µl of 1× first-strand buffer (Promega, France) and 2 mmol/l deoxynucleotide triphosphates in the presence of 40 units RNase inhibitor RNasin (Promega), 0.5 µg random primers (Promega) and 1 µg total RNA. Primer sequences synthesized by Invitrogen (France) and used for polymerase chain reaction amplifications are listed in Table 2. The conditions were as follows: 5 min at 94°C for denaturation, followed by 30 s at 94°C, 1 min at 60°C and 1 min 30 s at 72°C for 35 cycles and 7 min at 72°C for final elongation. The reverse transcriptase polymerase chain reaction products were electrophoretically analysed in 1% agarose and visualized by ethidium bromide staining.

View this table:
Table 2

Primers used for reverse transcriptase polymerase chain reaction analysis

GeneForward primerReverse primer


We used CD34 staining to identify the endothelium in glioblastoma tissue sections and PAS staining to determine the basement membrane of tumour blood vessels. Tumour vessels showed positive reaction for CD34 on their luminal surface and PAS-positive reaction in their wall (Fig. 1A). Microvascular proliferations, typically formed by multilayered tufts of endothelial cells, were easily recognized by the CD34–PAS dual staining (Fig. 1B). However, in a subset of glioblastomas, we could find the existence of PAS-positive tubular structures containing red blood cells but lined by CD34 cells in the luminal surface. They varied morphologically from large (Fig. 1C) to small channels (Fig. 1E and G). In order to test if these CD34 cell-lined tubular structures were blood vessels, we immunostained the consecutive histological sections with collagen-IV and found positive reaction in their wall (Fig. 1F and H). A longitudinally sectioned blood vessel with distinctive CD34+ and CD34 portions was found in one case (Fig. 1D) suggesting a possible anastomosis between CD34+ and CD34 vessels. These non-endothelial cell-lined vessels were markedly present in several zones of the section in 12.5%, isolated as one or two channels in 45% and totally absent in 42.5% of our glioblastoma tissue series.

Figure 1

Non-endothelial blood vessels in glioblastoma. Endothelial cells are detected with anti-CD34 immunohistochemistry staining (dark brown) and vascular basement membrane with PAS staining (purple magenta) in (A) tubular blood vessel and (B) microvascular proliferation of glioblastomas. (C) Large vessel containing red blood cells stained positively with PAS but negatively with CD34. (D) Longitudinally sectioned blood vessel presents distinctive CD34+ and CD34 portions. (E) and (G) CD34 cell-lined vessels stain with collagen IV in the respective serial sections (F) and (H). Scale bars are 50 µm.

We then hypothesized that CD34 vessels were lined by tumour cells. We selected paraffin sections from glioblastomas that present several CD34 cell-lined vessels (according to our histological analysis) and EGFR amplification (according to our Array Comparative Genomic Hybridization (CGHa) database) (Idbaih et al., 2008). We used α-SMA monoclonal antibody to identify vascular smooth muscle cells and fluorescent in situ hybridization detection of EGFR amplification to identify tumour cells. As shown in Fig. 2A, EGFR amplification was detected as double minutes by red hybridization signals in a large proportion of cells present in glioblastoma tissues. α-SMA monoclonal antibody specifically stained vascular smooth muscle cells (Fig. 2B and C). As expected, nuclei of the luminal surface did not harbour EGFR amplification in CD34+ cell-lined vessels (Fig. 2C). Interestingly, CD34 blood vessels were bordered by EGFR-amplified cells meaning that they were tumour cells (Fig. 2D). Moreover, some of these tumour cell-lined vessels showed positive α-SMA staining in their wall (Fig. 2E and F) suggesting that they could eventually display the smooth muscle component. Therefore, the non-endothelial cell-lined vessels are channels formed by primary tumour cells.

Figure 2

The tubular type vasculogenic mimicry in glioblastoma. (A) Nuclei are stained with DAPI (blue), fluorescent in situ hybridization EGFR probe (red) label tumour cells carrying EGFR amplification as double minutes (display multiple red signals). (B) Vascular smooth muscle cells are detected by anti-α-SMA immunofluorescence staining. (C) Nuclei of the vessel luminal surface do not harbour EGFR amplification. (D) Tumour cells carrying EGFR amplifications are visible in the luminal surface of a glioblastoma vessel. (E) The wall of an EGFR amplified cell-lined blood vessel presents α-SMA staining. Details of area delineated in (E) are enlarged in (F). Scale bars are 25 µm.

We established two CD133+ glioblastoma stem-like cell cultures (GSC-A and GSC-B) that demonstrated growth into tumour spheres (Fig. 3A). They were generated from solid primary adult glioblastomas carrying EGFR amplification and showed conservative genomic profile in culture (Fig. 3B). Undifferentiated tumour spheres immunostained for Nestin and Sox2 (characteristic neural stem-cell markers) and revealed multilineage potential (expression of glial fibrillary acidic protein for astrocytes and neuronal class III beta-tubulin-1 for neurons) at the differentiation assay (Fig. 3C). In addition, the ability of GSC to express markers from neural stem-cell lineage was validated by western blot (Fig. 3D).

Figure 3

Characterization of glioblastoma stem-like cells. (A) Phase contrast (a) and fluorescent (b) microscopy of undifferentiated green fluorescent protein-expresing GSC growing into tumour spheres in neurosphere medium. Phase contrast microscopy of differentiated GSC at the differentiation assay (c). Scale bars are 100 µm. (B) Comparative genomic hybridization of GSC demonstrating tumour genomic alterations. Each bacterial artificial chromosome spotted on the comparative genomic hybridization array is represented by a dot. bacterial artificial chromosomes are ordered on the x-axis according to their position in the genome. For each chromosome the telomere of the short arm is on the left and the telomere of the long arm is on the right. The y-axis corresponds to fluorescence ratio. Yellow, green and red indicate genomic copy number normal, loss and gain, respectively. Genetic alteration includes complete chromosome 10 loss, gain of chromosome 7 with EGFR amplification (arrow). (C) Immunostaining of tumour cells for neural stem-cell markers (Nestin and Sox2) in the sphere, then astrocytic [glial fibrillary acidic protein (GFAP) in green staining] and neuronal (neuronal class III beta-tubulin-1 in red staining) markers by the differentiated cells around tumour sphere at Day 7. Scale bars are 50 µm. (D) Expression of Nestin, glial fibrillary acidic protein, neuronal class III beta-tubulin-1 and Tubulin by undifferentiated (sphere) and differentiated (diff) GSC were analysed by western blot as described. Extracts from primary rat astrocytes (Astro) and U87 were used as controls. TUJ-1 = neuronal class III beta-tubulin-1.

To investigate the capacity of different GSC to display vasculogenesis in vitro, we used 3D Matrigel tube formation assays for direct comparison between GSC-A (derived from glioblastomas that contained tumour cell-lined vessels) and GSC-B (derived from glioblastomas that did not contain tumour cell-lined vessels at the histological analysis). GSC–A underwent a dramatic reorganization and formed efficiently a vasculogenic network of tubular structures within 2 days. In contrast, GSC-B formed hardly any structures even after 7 days (Fig. 4A). Undifferentiated and dissociated GSC-A were then cultured with 10% foetal calf serum after basic fibroblast growth factor and epidermal growth factor withdrawal. After cell differentiation, 1% of GSC-A displayed a flattened morphology and showed α-SMA-stained stress fibres (Fig. 4B). We went on to perform semi-quantitative reverse transcriptase polymerase chain reaction and found that several molecules involved in normal vasculogenesis and vasculogenic mimicry phenomenon were expressed in GSC (including EphA2, Laminin 5γ2, tissue factor pathway inhibitor-1, Neuropilin-2 and Endoglin) while major endothelial markers (CD34, CD31) were still lacking (Fig. 4C). Laminin 5γ2 expression was lower in GSC-B comparing to GSC-A. Vascular smooth muscle cell markers, α-SMA and Desmin, were expressed in both differentiated GSC.

Figure 4

Glioblastoma stem-like cell vasculogenesis. (A) GSC-A and GSC-B are labelled with green fluorescent protein. GSC-A forms tubular structures in 3D culture on Matrigel while green fluorescent protein-expressing GSC-B does not. Scale bars are 250 µm. (B) Green fluorescent protein labelled-GSC-A exhibits a flattened morphology and α-SMA immunofluorescence after differentiation. Scale bars are 20 µm. (C) Expression of endothelial and vascular smooth muscle cell-related genes is measured by semiquantitative reverse transcriptase polymerase chain reaction in GSC-A and GSC-B compared to human cerebral microvascular endothelial cells and human vascular smooth muscle cells. The housekeeping gene ALAS is used as control.


Here we describe, for the first time, the presence of tubular vessels formed by tumour cells in a subset of glioblastomas. Similar structures presented as ‘tumour cell-lined sinuses’ or ‘blood lakes’ have been observed in other human tumours (Kellner, 1941; Warren and Shubik, 1966; Radnot and Antal, 1979; Sato et al., 1982; Konerding et al., 1989). However, these reports failed to go beyond this phenomenon to prove the vascular nature of these structures, because collections of extravasated erythrocytes could sculpt a path through tumour cells that becomes a conduit for intratumoral blood flow and appears like a blood sinus (Warren and Shubik, 1966; Nasu et al., 1999).

Positive identification of presumptive tumour cell-lined blood vessels in cancers of non-vascular origin should then consider several criteria to argue against the claim that these channels are artefacts (McDonald and Foss, 2000). First, the tubular structures in question must be shown to be blood vessels. Vascular basement membrane, an important structural component of blood vessels, can be identified using immunohistochemistry for collagen-IV as a basal lamina antigen (Franciosi et al., 2007). We showed positive collagen-IV staining in the wall of our tubular structures. Second, the absence of endothelial cell marker is not sufficient. Tumour cells considered to be in contact with the vascular lumen must be positively identified. EGFR amplification is one of the most frequent genetic changes in primary glioblastomas (Idbaih et al., 2009) and our GSC showed to contain and maintain EGFR amplification, making this a specific marker of tumour cells in glioblastoma tissues. We identified EGFR amplification in the lining-cells of the non-endothelial vessels proving that they were definitely tumour cells. Third, though technically difficult to locate, the interface of tumour cell-lined vessels and endothelial cell-lined vessels would show that the channels are part of the tumour vasculature. We were able to find this interface in one glioblastoma section. Therefore, major criteria are met to provide convincing arguments that the presumptive tubular structures are blood vessels formed by primary tumour cells in glioblastomas.

Several observations reported that tumour cells are located in the walls of tumour blood vessels and form a part of the vessel surface while the remaining part is covered by endothelium. This is known as ‘mosaic vessels’ where tumour cells undergo intravasation into the lumen and stay temporally in the vessel wall (Chang et al., 2000). The tumour cell-lined vessels that we observed in glioblastomas could be the result of a complete invasion of the vessel wall by tumour cells, perhaps the endpoint of mosaicism. However, from our data we strongly suggest that GSC are capable of forming blood vessels de novo. They also have the ability to express endothelium-associated genes including EphA2 (ephrin receptor), Laminin 5γ2 (basement membrane component) and Neuropilin-2 (vascular endothelial growth factor receptor co-receptor). These molecules are required for the formation and maintenance of blood vessels (Serini et al., 2006; Hess et al., 2007; Geretti et al., 2008). At the same time, they are well known to play important roles in nervous system development (Culley et al., 2001; Wilkinson, 2001; Fujisawa et al., 2004). The anatomical structure similarities between the nervous and vascular systems are striking. Over the past decade, it has become apparent that neural and vascular guidance pathways share common signalling mechanisms, including ephrins and neuropilins in particular (Eichmann et al., 2005). It is therefore likely that a subpopulation of glioblastoma cells with neurodevelopmental features makes use of these common neural and vascular patterning tools to develop a proper blood vessel network. Nevertheless, GSCs merely mimic the function of vessels as they do not transdifferentiate into endothelial cells with respect to the lack of major endothelial markers expression. Thus, the term ‘tubular vasculogenic mimicry’ is appropriate to describe the formation of these vascular channels by tumour cells.

More surprisingly, we found that a fraction of GSC have the capacity to transdifferentiate into smooth muscle-like cells that may constitute part of the tumour cell-lined vessel wall as the essential muscular component. This is consistent with the report that rat foetal brain stem cells can give rise to smooth muscle cells (Tsai et al., 2000; Song et al., 2009), which have an identical contractile function to vascular smooth muscle cells (Oishi et al., 2002). From our findings, we suggest that GSC are more plastic than previously thought, providing more evidence for their mesenchymal differentiation potential (Tso et al., 2006; Ricci-Vitiani et al., 2008; Rieske et al., 2009).

Traditional anti-vascular therapies aimed at endothelial cells are not effective in blocking tubular network formation by tumour cells (Van der Schaft et al., 2004). Since the vascularization of glioblastomas is heterogeneous, designing a therapeutic approach that targets only angiogenic vessels might result in incomplete therapy. It would be prudent to target tumour cell-lined vessels because they may participate to the antiangiogenic resistance by providing an alternative pathway for glioblastoma vascularization. For instance, Laminin can be a potent target to block both endothelial and non-endothelial tumour vascularization. Down-regulation of Laminin 5γ2 resulted in the complete inability of aggressive melanoma cells to form vasculogenic-like networks in tri-dimensional culture (Seftor et al., 2001) which is consistent with the lack of vasculogenic ability in GSC-B that showed low levels of Laminin 5γ2 expression.

In conclusion, we describe a new alternative mechanism in glioblastoma vascularization. This finding provides a better comprehension of tumour vascularization and cancer stem-cell plasticity, and has important implications in the treatment strategy. One should evaluate the overall contribution of such tumour cell-formed vessels to glioblastoma blood flow and determine their sensitivity to current antiangiogenic therapies using quantitative methods with appropriate sampling. Also, understanding the influence of the microenvironment in determining the vascular fate of GSC may provide new perspectives on tumour cell plasticity that could be exploited for novel strategies in cancer differentiation therapy.


Institut National du Cancer grant (PL 046); Institut de France (Fondation Energie); Association de Recherche contre les Tumeurs Cerebrales (ARTC); Ligue Nationale Contre le Cancer support to S.E.H.


  • Abbreviations:
    glioblastoma stem-like cells
    Periodic Acid-Schiff


View Abstract