Early reports have linked stemness properties of GSCs to CD133 expression and suggested that tumorigenic cells in GBM were restricted to the CD133⁺ population [20,21]. For this reason CD133 marker has been widely used in the past to define cancer stem cells, notably in human malignant gliomas. However, recent studies revealed that a subpopulation of CD133⁻ cells isolated from GBM could also be tumorigenic and exhibit similar stemness features [22,23], raising some caution about the use of CD133 to reliably evaluate GBM cell stemness. These observations prompted us to include both CD133⁺ and CD133⁻ GSCs in the present study. The methodology for isolation and characterization of these cells has been previously described in details by our laboratory [24]. In addition, tumorigenicity and stemness properties of GBM derived stem cells were assessed by xenograft experiments in nude mice (Supplementary Figures S1 and S2). Cells isolated from 10 primary GBM tumors (Table 1 and Supplementary Table S1) were cultured in serum-free medium (DMEM/F12 or NBE supplemented with EGF and βEGF) for approximately one week and proliferated as non-adherent multi-cellular spheres (neurospheres). These culture conditions enable tumor cells to retain the molecular characteristics of the primary tumor [24]. When kept in specific differentiation media (10% fetal bovine serum), these cells became elongated and adherent to the surface of the culture flask and were able to differentiate into the three main cell lineages found in the central nervous system: neurons, astrocytes and oligodendrocytes as they expressed beta-tubulin, GFAP, S100, and O4. An index of differentiation was calculated as the ratio between GFAP and Nestin mRNA expression in differentiated cells vs. neurospheres (Table 1). Although heterogeneity was present within the data, the index obtained for all GBM cell lines were correlated with the differentiated phenotypes observed by confocal microscopy (Supplementary Figure S3).

Sensitivity to temozolomide was determined for all GSC lines (GBM-1 to -10) as described in Materials and Methods. Dose-response curves were established after 5 days of treatment and IC₅₀, IC₂₀, and IC₁₀ values were determined (Table 2). Under stem-like conditions, all cell lines were affected by TMZ and cytotoxicity varied among the cells with IC₅₀ ranging from 367 μM to 1480 μM (median IC₅₀ 630 μM). Clonogenic survival assays performed with GBM1 and GBM2 using IC₅₀ dose confirmed these results (data not shown). These data corroborate the previous study from Blough et al. [19] showing great variations in response to TMZ at clinical doses (below 50 μM). When cultured in differentiation medium, cells exhibited different IC₅₀ values (354 μM to 3950 μM) (median IC₅₀ 568.5 μM), with one highly resistant line. Spearman’s correlation test showed no relationship between IC₅₀ in stem-like condition and after differentiation. The dose-response curves obtained in this experiment indicate that TMZ doses achieved in clinical applications (5–50 μM; 1–10 μg/mL) have a corresponding efficacy below IC₁₀ (i.e., 10% of cell death) after 5 days of treatment for both neurospheres and differentiated cells. Interestingly, comparison of the IC₅₀ values indicates that TMZ sensitivity of GSCs was slightly increased when cells were cultured in differentiation medium (median IC₅₀ 568 μM vs. 630 μM), with seven out of 10 lines showing higher sensitivity after commitment. These results contrast with previous work from Beier et al. suggesting that temozolomide preferentially depletes the stem cell subpopulation in GBM (CD133⁺ and CD133⁻) but spares more differentiated cells [25]. The most likely explanation for the discrepancy comes from the fact that these authors did not analyse the effect of temozolomide under fully differentiating conditions (FBS-containing medium).

MGMT-methylation status of the tumor is an independant prognostic factor for GBM patients treated with an alkylating agent such as TMZ [7,25]. Moreover, the epigenetic inactivation of the MGMT gene is associated with a better response to chemotherapy [4,5]. By using a quantitative pyrosequencing approach [26], we determined the methylation status of the MGMT gene promoter region (Table 3). This quantitative methylation assay detects the level of methylation of 5 CpG sites located in the first exon of the MGMT gene and is of high prognostic value compared to MS-PCR and qMS-PCR [26]. CpG positions assessed in this pyrosequencing assay overlap the region analyzed by qMS-PCR and represent a stable indicator of the overall methylation level. Table 3 shows the methylation status of each CpG island for the 10 tumoral samples and corresponding GSCs in stem-like or differentiated condition. The median methylation level for all five CpG sites tested was 2% in the tumors, 3.5% in neurospheres and 10% in differentiated cells. Statistical analysis of median CpG methylation levels showed significant positive correlations between stem-like and differentiated cells (p = 0.0003), between tumor and stem-like cells (p = 0.0007), between tumor and differentiated cells (p = 0.0005) (Table 4). These observations indicate that the overall tumor methylation pattern is conserved among GSCs, independently of their differentiation status. Comparison of median CpG levels reveals a higher MGMT promoter methylation when cells were kept in differentiation medium (10% vs. 3.5% in stem-like conditions). Nevertheless, statistical analysis between median CpG levels in stem-like conditions vs. differentiating conditions did not show significant difference, most likely due to the great variability between cell lines. Further analysis of MGMT transcript expression revealed no significant difference when cells were maintained in either conditions, although highly variable levels were observed between cell lines (data not shown). Interestingly, the methylation status of the MGMT promoter was correlated with the expression levels of MGMT transcript in both stem-like and differentiating conditions (p < 0.01 by the Spearman’s correlation test, data not shown).

Correlations of TMZ sensitivity with MGMT methylation status were analyzed in GSCs in stem-like conditions and after differentiation. As shown in Table 5 and in Figure 1, no correlations were found between IC₅₀ of TMZ and MGMT methylation in neurospheres, either at individual CpG site, or with the median CpG value. These observations are in accordance with previous results from Blough et al. [19]. However, when cells were kept in differentiation promoting conditions a significant negative correlation was found, despite the reduced number of lines available (p = 0.03 with median CpG value). Hence, for the first time our data demonstrate that GSCs engaged in the differentiation process are more sensitive to TMZ and this sensitivity is correlated to the methylation status of MGMT.

In accordance with our results, a recent analysis of gene expression signatures associated with TMZ resistance in malignant glioma also identified MGMT level as a predictor of TMZ response in GBM cells grown under differentiating conditions [27]. Together these results indicate that the differentiation status of GBM cells should be taken into account when considering the predictive value of MGMT methylation in response to TMZ.

Although GBM tumors contain in majority differentiated cells, statistical analysis of the five CpG islands showed no relationship between MGMT promoter methylation in tumors and GSCs sensitivity to TMZ (Table 5 and Figure 1B). Interestingly, the comparison of median p values tend to suggest a better predictive value of the methylation status of MGMT in highly differentiated tumors. One possible explanation for the lack of correlation could be due to the cellular heterogeneity of the tumor. To answer this question we determined the differentiation level of the 10 tumors included in this study by calculating the ratio GFAP vs. nestin expression (Supplementary Table S2). Results obtained indicate that capacity for differentiation is highly heterogeneous between tumors, with extensive 500-fold variations in differentiation levels. Hence, tumor heterogeneity might be one explanation for the lack of correlation between MGMT status in this set of tumors and TMZ sensitivity of differentiated cells. These data uphold previous immunohistological observations showing intra- and intertumoral heterogeneity in the distribution of anaplastic and differentiated cells in glioblastomas [28]. The methylation status of the MGMT gene promoter is a promising molecular predictive marker, but conflicting studies with controversial data have hampered its clinical use. Our results tend to indicate a major effect of the tumor differentiation status in such studies, which could explain the discrepancies observed in the literature. Further studies on larger cohorts of tumors stratified by their differentiation status are needed to establish the relationship between MGMT promoter methylation and TMZ sensitivity in GBM tumors.

Notwithstanding of the degree of differentiation of the tumors, long-term follow-up of patients in whom GSCs have been isolated revealed a longer overall survival for those with higher MGMT methylation levels. Patients in whom GBM3, 5, 6 and 9 have been isolated had a median survival period of 420 days while other patients had a 306 days median survival. To minimize the risk of selection bias associated with this reduced cohort, all patients included in this study were consecutively treated at our hospital over a 3 years period, were aged-matched and carried wild-type IDH1 and IDH2 genes. Although statistical analysis cannot be performed with a reduced number of patients, these data are in line with previous reports [29] and suggest a prognostic significance for MGMT methylation in GBM tumors.

Tumor samples were obtained from 10 adult GBM patients (GBM1 to GBM10) and processed within 30 min after surgical resection (patients characteristics are summarized in Table 1). These patients were all initially oriented after surgery towards the treatment protocol described by Stupp et al. [30]. Written informed consent forms were obtained from each patient. Primary cultures were derived from tumor samples, cultured and characterized as described previously [24].

Nestin, GFAP, and MGMT mRNA expression were determined by Taqman gene expression assays (Applied Biosystems, Foster City, USA) as described previously [24] and normalized to GAPDH mRNA levels present in the same cDNA sample. Relative changes in nestin, GFAP and MGMT mRNA amounts were determined by the 2^−ΔΔCt method.

Doubling times of cultures and cytotoxicity of Temozolomide (Interchim, Montluçon, France) were determined at passage 10 using the XTT cell proliferation test (Roche, Basel, Switzerland) as previously described [27]. Before analyses, cells from neurospheres were dissociated and seeded in 96- well plates at a density of 5 × 10⁴ cells per well and the quantification of viable cells was performed on a MRXII (Dynex Technologies, Chantilly, VA, USA) at 450 nm. The 50% inhibitory concentrations (IC₅₀) of Temozolomide were determined after treatment with increasing amounts for five days.

DNA was extracted from frozen tumor samples and cells in culture using the Qi-Amp DNA minikit or the DNeasy tissue kit (Qiagen, Hilden, Germany) in accordance to the manufacturer’s instructions. Quantification was performed on a Nanovue spectrophotometer (General Healthcare, Brentford, UK). 500 ng of genomic DNA were bisulfited using the EZ DNA methylation Gold kit according to protocol (Zymo Research, Orange, CA, USA). Universal methylated DNA (Millipore, Billerica, MA, USA) were included as negative and positive controls.

The pyrosequencing methylation assay was performed with the PyroMark^TM MGMT kit (Qiagen, Hilden, Germany) on a Q24 MDx system (Qiagen, Hilden, Germany), according to the manufacturer’s protocol. The PyroMark^TM MGMT kit detects the level of methylation of 5 CpG sites located in the first exon (+17 to +39) of the MGMT gene. A cytosine not followed by a guanine, and which was therefore not methylated, was used as an internal control for completion of the bisulfite treatment. Universal methylated DNA (Millipore, Watford, UK) and unmethylated DNA were included as control. The mean methylation value of the positive and negative controls were 97% and 3% respectively.

Correlation analyses were done by using the Spearman’s rank correlation test. Statistical significance was determined according to the Spearman’s rank correlation coefficient (rho). For n = 10, a rho value > |0.6483| was considered to indicate statistical significance (p < 0.05).