All of the samples were classified according to World Health Organization guidelines (grade II, III or IV for gliomas), and the classification of Sainte Anne Hospital (low grade oligodendroglioma or type A, high grade oligodendroglioma or type B, glioblastomas and malignant glio-neuronal tumors, 20 . The biopsies were collected by a pathologist in the surgical room from July 2002 to July 2005. All patients were 18 years old or older, had signed a written agreement for participation to the research project after having being informed of the goals, potential interest of the research and methods. This biomedical research was conducted according to the declaration of Helsinki, to the French laws, and was approved by the institutional review board of Ste Anne Hospital, Paris. Anatomopathological diagnosis classified tumors as glioblastoma multiformis (n = 6), Malignant Glio-Neuronal Tumors (MGNTs, n = 15), medulloblastoma (n = 2), ganglioglioma (n = 1), oligodendroglioma (n = 23). Immunolabeling of formalin-fixed, paraffin-embedded tissue sections was performed as previously described 20 .

Viable fragments were transferred to a beaker containing 0.25% trypsin in 0.1 mM EDTA (4:1), and slowly stirred at 37°C for 30-60 min. Dissociated cells were plated in 75 cm² tissue culture flasks plated at 2500-5000 cells/cm² in Dulbecco's modified Eagle's: F-12 medium (1:1) containing the N2, G5 (containing FGF and EGF) and B27 supplements (all from Invitrogen, France). After 2 to 47 days in culture, spheres bloomed from clusters of adherent cells. They were dissociated in single-cell suspension each week with a renewal of two third of their culture medium.

Dissociated sphere-derived cells were plated in 18-mm diameter wells in 1 ml volumes at a density of 200 000 cells/well. Cell proliferation/viability was evaluated using the WST1 kit from Roche according to the manufacturer's protocol. We verified that the results were the same as those obtained by counting the absolute number of viable cells using trypan blue exclusion. Clonality was evaluated in 96-well plates seeded at a cell density of 1-200 cells/well/0.1 ml. The number of wells containing at least one secondary sphere was evaluated after 3-4 weeks of culture.

Single-cell suspensions were incubated with fluorochrome-conjugated monoclonal antibodies: fluorescein (FITC)-coupled CD15, FITC-CD11b, FITC-CD45, phycoerythrin (PE)-coupled CD34, PE-Thy-1 (CD90), PE-CD133 and phycoerythrin cyanin 5 (PC5)-coupled CD34, PC5-CD56, and allophycocyanin (APC)-coupled CD133. CD133 antibodies were from Miltenyi (France), all others from Beckman-Coulter. Dead cells were excluded from the analysis by 7-AAD. Acquisition was performed on a Facscalibur (Becton Dickinson) and viable cells were analyzed with Cellquest software.

Cells were plated at a density of 5.000 cells/cm² onto polyornithine-coated glass coverslips and onto Lab-Tek™ II Chamber Slide™ System (Nalge Nunc International) http://www.nuncbrand.com/page.aspx?ID=234 in the presence of FCS or B27 (Invitrogen) for 2 to 14 days. Multiple immunofluorescence assays for neural antigens were performed as previously described 17 20 .

The theoretical and practical aspects of real-time quantitative RT-PCR using the ABI Prism 7900 Sequence Detection System (Perkin-Elmer Applied Biosystems) have been described in detail elsewhere 21 . Briefly, total RNA is reverse-transcribed before real-time PCR amplification. Quantitative values are obtained from the threshold cycle (Ct) number at which the increase in the signal associated with exponential growth of PCR products begins to be detected using PE Biosystems analysis software, according to the manufacturer's manuals. We also quantified transcripts of TBP gene, which encodes the TATA box-binding protein as the endogenous RNA control, and each sample was normalized on the basis of its TBP content.

Results, expressed as N-fold differences in target gene expression relative to the TBP gene, termed Ntarget, were determined by the formula: Ntarget = 2^{ΔCt _sample} , where ΔCt value of the sample was determined by subtracting the Ct value of the target gene from the Ct value of the TBP gene.

The Ntarget values of the samples were subsequently normalized to a "basal mRNA level", i.e. normalized to the smallest amount of target gene mRNA detectable and quantifiable by real-time quantitative RT-PCR assays (target gene Ct value = 35; Ntarget value = 1).

The nucleotide sequences of primers for TBP and the 4 target genes are available on request. The thermal cycling conditions comprised an initial denaturation step at 95°C for 10 min and 50 cycles at 95°C for 15 s and 65°C for 1 min. Experiments were performed with duplicates for each data point.

Cells were treated with medium containing 10 μg/ml Colchicine for 1 to 2 hours, and resuspended in hypotonic 1% sodium citrate at room temperature for 30 minutes. The cells were then washed in a methanol-acetic acid (3:1, v/v) fixative solution for 30 minutes, and spread onto clean dry slides. Q-banding staining was then performed, and 10 metaphases were analyzed for each sample.

Cells were washed three times with PBS for 10 min with gentle shaking, prior to lyses in buffer containing 8 M urea, 4% CHAPS and 40 mM Tris. The protein pellets were dissolved in isoelectric focusing buffer and quantified using the Micro BCA protein assay kit (Pierce). First-dimension isoelectric focusing was performed with the Ettan IPGphor system (Amersham Biosciences) at 20°C with a maximum current setting of 50 mA/strip. Immobiline DryStrip gels (IPG stripes, Amersham Biosciences) with a pH gradient of 3-10 or 4-7 and a length of 13 cm were rehydrated in 250 μl sample solution, containing 80 μg proteins. Prior to SDS-PAGE analysis, IPG stripes were equilibrated in second-dimension equilibration buffer (6 M urea, 2% SDS, 50 mM Tris-HCl, pH 8.8, 30% glycerol) containing 1% DTT (Sigma) to reduce the disulfide bonds of the proteins. The second dimension was carried out using 12.5% gradient polyacrylamide gels at constant 20 mA current per gel. Comparison of the protein maps was achieved using the PD-Quest software according to the manufacturer instructions (BioRad).

In order to preserve tissue architecture and improve sample handling as well as the versatility of the tests that may be applied, we developed a two-step strategy for the 49 surgical samples of our survey. Tumor samples were first sliced in the surgical room in fragments smaller than 1 mm³, and placed on surgical sponges bathed with culture medium. Tumor fragments were either dissociated within 2 hours after surgery (28 out of 49, 58%), or incubated in organotypic conditions for a maximal time of 2 weeks prior dissociation. Tumor fragments cultured on surgical sponges are viable and conserve the tumor architecture for up to 6 weeks (additional file 1). In both cases, the dissociated cells were plated at low density (<5 × 10³ cells/cm²) in a serum-free medium containing EGF and FGF2 (G5 supplement).

In twelve of the fifteen cultures of MGNTs, floating cellular spheres were observed after 14 to 60 days in culture, and could be amplified from 6 months up to 52 months. In contrast, none of the other 23 tumors classified as World Health Organization grade II or III oligodendrogliomas (n = 9 and 14, respectively) generated spheres up to 90 days, the time at which no viable cells remained in the cultures. Similarly, two benign lesions, one Taylor Dysplasia and one adult pilocytic astrocytoma, did not yield any sphere. A third in vitro behavior pattern was observed for cells derived from three of the six glioblastoma multiformis studied, two adult medulloblastomas and one adult ganglioglioma. In these cases, spheres developed but stopped dividing after 2 to 5 months of culture, and disappeared suggesting limited auto-renewal capabilities (Table 1).

These data extended previous reports of absence of EGF/FGF2-responsive clonogenic precursors in oligodendrogliomas 2 3 4 22 . They revealed the unique ability of MGNTs to yield cellular spheres that could be amplified for very long periods.

MGNT cells were moderately proliferative but highly clonogenic. The average proliferation rate, determined using cultures derived from MGNT 1, 6 and 7 (TG1, TG6 and TG7), was equal to 236% ± 27% within 48 hours. In two separate experiments, the percentage of cells initiating-spheres, determined by plating 500 dissociated cells in 6-well plates and assessing the number of secondary spheres, reached 25-30%. Similar results were obtained when cells were plated at a density of 200 cells per well in 96-well plates (TG1, 28.9% ± 8.0; TG6, 15.5% ± 7.0; TG7, 15.6% ± 5.0, mean ± sem, n = 3) (Table 2). Finally, 7 to 10% of the cells seeded at one cell/well generated at least one secondary sphere after one month in culture (TGI, 9 ± 3%, n = 4, TG6, 7 ± 1%, n = 3, TG7, 7 ± 2%, mean ± sem, n = 3). These properties were reproducibly observed for TG1-cells from week 40 to week 100, the latest time tested.

Analysis was conducted with a panel of antibodies directed against antigens characteristic of stem/progenitor cells of the CNS and other tissues, and of cells of the neuronal and glial lineages (Figure 1A-D). Numerous cells expressed nestin and the bHLH-transcription factor Olig2 (Figure 1A, 1C and 1I), normally expressed in stem/progenitor cells. A majority of cells were immunoreactive for GFAP, an astroglial marker (Figure 1B and 1I), whereas a small population was immunoreactive for the neuronal marker ß3-tubulin (Figure 1D and 1I). Each individual sphere exhibited cellular heterogeneity.

When bFGF and EGF were removed and replaced by serum, most but not all spheres underwent changes characteristic of differentiation: they adhered rapidly to the plastic substrate, the cells flattened and acquired a fusiform shape. Cellular heterogeneity both between spheres and within each sphere was maintained in these conditions, as shown by immunocytochemical analysis (Figure 1E-G and 1I). When treated with 0.5% FCS for 14 days, all cells exhibited nuclear Olig2-immunolabeling (Figure 1G and 1I). In the presence of B27, a cocktail designed to promote neuron survival and development in cultures of normal nervous tissue, 30% of the cells within the spheres acquired the appearance of neuron-like cells with a rich neurite-like branching, and ß3-tubulin expression (Figure 1H and 1I). Interestingly, a fraction of the differentiated cells co-expressed neuronal and glial markers as in the original tumor (Figure 1F and 1I).

The cell-surface phenotype of sphere-derived cells from different MGNT cultures (TG1, TG6, TG7) was repeatedly characterized using flow cytometry and remained stable over 6 months and up to 52 months for TG1, the latest time tested (Figure 2). CD56/NCAM, CXCR4, CD90/Thy-1, and VLA2, which are known to be express by neural progenitors, were detected in over 90% of the cells. A second group of proteins was detected in less than 10% of the cells. They included CD15/Lex/ssea1, described on neural stem cells 23 , and CD34 synthesized by stem/progenitor cells from several tissues, but which so far has not been reported on neural cells. CD34 is known to be also expressed by endothelial cells, that may originate from neural stem cells in normal brains 24 . However, no expression was detected for the endothelial-specific markers CD31 and VE-cadherin by MGNT-derived cultured cells. Similarly, none of the hematopoietic markers CD45, CD3, CD11b, CD14, CD19, CD33, CD36, CD38 was expressed, thus excluding that CD34 expression reflected contamination by endothelial or hematopoietic progenitors. Likewise, we did not detect expression of the cytokine receptors c-kit or Flt3, or CD44 and CD93. In addition, we looked for CD133-immunoreactive cells that have been reported to represent up to 50% of dissociated tumor cells examined by flow cytometry 1 2 3 4 5 6 . In agreement with these data, we found that 20 to 30% of cells isolated immediately after tumor resection were positive for this antigen (n = 4). Surprisingly, only a few cultured cells expressed CD133, suggesting a rapid down-regulation in vitro. Furthermore, CD133+ as well as CD133- cells were able to form floating spheres as well as to initiate tumors upon brain grafting (see below). One may hypothesize that CD133 expression is not a characteristic of a subpopulation of cells endowed with stem-like properties at least in MGNTs, but rather depends on their environment.

To determine if the sphere-forming capability segregated according to CD34 or CD15 expression, we sorted the negative and positive fractions from TG1, and TG6 cultures (week 35 to 90), and determined their ability to generate spheres in culture. The ability to form spheres segregated neither with CD15 nor with CD34 expression (Table 3). A side population was identified using the Hoechst 3342 dye. Cells excluding the dye amounted to 2-10% of the cellular population (2 independent experiments). The capacity of the cells to form spheres (in both primary and secondary assays) was independent of their ability to exclude the dye, suggesting that this property depends from the cell activation state rather than its identity.

RT-PCR analysis was additionally used to seek for transcripts usually observed in stem cells and considered as involved in their self-renewal capacities. Oct 4, Nanog, and hTERT transcripts were observed in all MGNT-derived cells forming spheres analyzed (n = 3) (Table 4). Of note, CD133 transcripts were always close to the limit of detection (Table 4).

Molecular characterization of cells-forming spheres was further performed using 2D-SDS-PAGE. The stability of the proteomic profile of a given culture was determined using TG1 cells at two different passages (week 40 and week 160). The proteomic profiles of cells isolated from the spheres were also compared between different MGNTs. Analysis after silver staining revealed a very similar protein pattern for TG1 at both passages examined, as well as between cells derived from distinct patients (TG6, TG7) (Figure 3). Comparison of the protein maps using computer-assisted analysis (PD-Quest software) showed a high coefficient of correlation (0,84 ± 0.06). A differential subtractive approach comparing protein patterns from cells collected from MGNT-derived spheres and a frozen micro-dissected fragment of the original tumor, allowed the MALDI-TOF identification of a group of proteins enriched in cultured cells (Table 5). The eight potential biomarkers identified included the mannose-6-phosphate receptor binding protein 1 reported on early stem cells 25 , and two proteins associated with multidrug resistance, Sorcin being the most strikingly over-expressed. Sorcin mRNA and protein levels have been reported in astrocytoma and are positively correlated with the malignancy of the tumor 26 . These data are consistent with the possibility that Sorcin expression signs for an enhanced occurrence of TICs in high-grade gliomas.

Altogether, these results demonstrate that cells forming cells derived from MGNT exhibit a common and stable surface antigen and proteomic profile, as well as some features of stem/progenitor cells in long-term cultures.

Karyotype analysis and in vivo behavior upon orthotopic grafting were performed in order to determine whether the cell-forming spheres derived from MGNTs were indeed tumoral and behaved as tumor-initiating cells (TICs).

Karyotype analysis was performed on cells collected from spheres grown from 4 tumors (TG1, TG6, TG7, TG10). Gains were observed for each cell examined on chromosomes 1q, 5, 7, 9q, 12, and 14, whereas losses were observed on chromosomes 9p and 18q (Fig 4).

To determine the presence of TICs, MGNT-derived sphere cells from four different MGNTs were grafted into the brain of nude mice (2 mice each, 200-1000 living cells per graft). The recipients were sacrificed from 12 to 24 weeks post-graft. Tissue analysis after hematoxylin/eosin and immunolabeling with antibodies that recognized only the human form of the glial marker vimentin identified human cells in 7 out of 8 mice. In six cases out of seven, the grafts had developed with cell proliferation (as evidenced by Mcm2 labeling), and migration over large distances from the injection site (Fig 5). Several migrating cells were observed along the neural fiber axis (Fig 5).

Taken altogether these data demonstrate that long-term cultured MGNT-derived floating spheres contain tumor cells with neural stem/progenitor phenotype, thus fulfilling the criteria of tumor initiating cells (TICs).

Temozolomide, an oral alkylating agent, is the main drug used for high grade gliomas. The therapeutic concentration given to patients is around 60 μM. Cells cultured from tumors TG1 at three different passages, TG10 and TG16 were exposed for 48h to different concentrations of temozolomide. Cells from all cultures were highly resistant to temozolomide, 60% of the cells being still viable in presence of 1000 μM temozolomide (Table 6). No difference in temozolomide resistance was noted between long-term cultured spheres obtained from FACS-sorted CD133+ and CD133- cells. In addition, removal of growth factors for 10 days or induction of differentiation for 7 days in the presence of 10% FCS did not change TIC resistance to TMZ.