Introduction

The cancer stem cell hypothesis proposes that tumors contain a small subset of cells with stem-like properties, which have the exclusive ability to self-renew and sustain the growth of the tumor (Clarke et al., 2006). Tumor stem cells were first identified in leukemia, and more recently also discovered in various solid tumors, including cancers of the brain such as glioblastomas, medulloblastomas and ependymomas (Ignatova et al., 2002; Hemmati et al., 2003; Singh et al., 2003, 2004; Galli et al., 2004; Kondo et al., 2004; Tunici et al., 2004; Yuan et al., 2004; Taylor et al., 2005; Bao et al., 2006; Lee et al., 2006). Brain tumor stem cells are defined by (i) the capacity to self-renew, (ii) the ability to initiate brain tumors upon orthotopic implantation and (iii) multipotency, that is, the capacity to differentiate into cells with a neuronal, astrocytic or oligodendroglial phenotype (Pilkington, 2005; Vescovi et al., 2006). In addition, brain tumor stem cells are characterized by the expression of neural stem cell antigens and the ability to grow as nonadherent spheres termed 'neurospheres' when cultured in the presence of epidermal growth factor and fibroblast growth factor under serum-free conditions. Thus, stem-like cells in brain tumors share many characteristics with normal neural stem cells, supporting the hypothesis that brain tumors can arise from neural stem or progenitor cells (Pilkington, 2005; Vescovi et al., 2006). Recent studies further showed that glioblastoma cells cultured under neural stem cell conditions can display heterogeneous growth characteristics and molecular profiles, suggesting that they may either arise from different cell types or from similar cells that have acquired different genetic alterations (Galli et al., 2004; Beier et al., 2007).

Reports that brain tumor stem cells are resistant to conventional radiation and pharmacological treatments (Bao et al., 2006; Liu et al., 2006) have been taken to suggest that development of more effective therapies for glial malignancies would be facilitated by the existence of in vitro and in vivo models that faithfully recapitulate the stem cell component of these lesions. Using the neurosphere culture method (Svendsen et al., 1998; Hemmati et al., 2003; Galli et al., 2004; Tunici et al., 2004; Lee et al., 2006), we established glioblastoma stem-like (GS) cell lines from nine different glioblastomas, currently maintained for up to 3 years in vitro. Although established under identical conditions, our panel of glioblastomas gave rise to two distinct subtypes of long-term cultures. Four cell lines shared similar gene expression patterns, associated with neural development and displayed a full stem-like phenotype, whereas five cell lines shared a different gene expression pattern and displayed only a restricted stem-like phenotype.

Results

Establishment of GS cell cultures and genomic profiling

Nineteen glioblastomas were dissociated into single cells and plated at clonal density (Gritti et al., 1996) in neurosphere medium. Nine of these cultures became expandable for more than eight passages, whereas 10 did not survive the first or second split. SNP chip analyses of loci that are frequently affected by gains or losses in human glioblastomas indicated no significant differences in the frequency of copy number alterations between tumors that generated expandable cultures and those that did not (Supplementary Table 1).

All nine expandable GS cell cultures were derived from adult patients (Table 1). Eight of these patients underwent their first surgery for a newly diagnosed glioblastoma. The GS-2 tumor was an exception, being the second recurrence in a long-term survivor who was subjected to two prior surgeries, radiation and chemotherapy for a glioblastoma with oligodendroglial features. Nuclear accumulation of the p53 protein in at least 10% of the tumor cells was present in four tumors (GS-1, GS-2, GS-4 and GS-6) (Table 1). Mutations in the TP53 gene were found in five tumors (Table 1). Amplification of the EGFR gene was detected in two tumors, copy number losses at the PTEN locus in all nine tumors and copy number loss at the CDKN2A locus in six tumors (Supplementary Table 1).

Table 1 - Patient and tumor characteristics.

Full table

All primary GS cultures were composed of single cells, which gave rise to clonal spheres within 1–2 weeks. To date, eight GS cell lines have been grown for at least 50 passages. An exception was GS-6, which was lost after passage 10. The morphology of the GS cell lines differed considerably. Four cultures (GS-3, GS-5, GS-8 and GS-9) grew entirely spherically with no attached cells (Figure 1). In contrast, GS-1, GS-4 and GS-6 grew mostly adherently, only occasionally forming spheres that rarely detached from the culture plate. GS-2 and GS-7 grew semiadherent, forming spheres that tended to adhere to the culture plate (Figure 1).

Figure 1.

Growth pattern of glioblastoma cultures established under neural stem cell conditions. In primary culture, cells in all glioblastoma stem-like (GS) cell cultures formed some true spheres that were derived from single cells, as well as some spherically aggregated clusters of cells. After passaging, the cultures either assumed an entirely spherical growth pattern (GS-3, GS-5, GS-8, GS-9) or a semiadherent growth pattern with spheres that mostly attached to the culture dish (GS-2-, GS-7), or a predominantly adherent growth pattern with only occasional formation of spheres that mostly remained attached (GS-1, GS-4, GS-6).

Full figure and legend (132K)

Gene expression profiling identifies different clusters of GS cell lines

To analyse gene expression profiles of GS cell lines, Affymetrix HG-U133 Plus 2.0 microarray analyses were performed. After transcripts called 'absent' in all samples and transcripts displaying less than a ratio of 3 between the highest and lowest value were removed from the data set, a subset of 22 373 transcripts remained. Cluster analysis of this subset resulted in a dendrogram clearly separating cell lines GS-3, GS-5, GS-8 and GS-9 (cluster-1) from GS-1, GS-2, GS-4, GS-6 and GS-7 (cluster-2) (Figure 2a). Thus, the four cell lines that displayed entirely spherical growth in vitro also shared a common gene expression pattern, distinct from that of the cell lines with semiadherent or adherent growth. The expression profiles proved to be stable when the analyses were repeated up to 14 passages later (Figures 2a and b).

Figure 2.

Gene expression analysis of glioblastoma stem-like (GS) cell cultures using Affymetrix HG-U133 Plus 2.0 chips. In total, 23373 transcripts were called 'present' and displayed a threefold range in intensity. Cluster analysis of these transcripts separated cell lines GS-3, GS-5, GS-8 and GS-9 (cluster-1) from GS-1, GS-2, GS-4, GS-6 and GS-7 (cluster-2). All GS cell cultures except for GS-6 were analysed in duplicate (a). Analysis of differentially expressed transcripts showed that 392 transcripts were overexpressed in cluster-1 and 374 transcripts were overexpressed in cluster-2. Two replicates for each cell line were analysed except for GS-6. Gene expression profiles for individual GS cell lines remained stable over multiple passages (b). Significant associations of differentially expressed genes in GS cell cultures with Gene Ontology terms. Using the FatiGo web tool, two terms were identified as significant, both with higher expression in cluster-1 (adjusted P-value <0.05). Of the 392 cluster-1 genes, 59 were annotated with Gene Ontology term at level 5, and 14 of these (23.7% ) belonged to 'nervous system development.'. Fifty genes were annotated at level 7, and 6 of these (12.0% ) belonged to the term 'neuropeptide signaling pathway' (c).

Full figure and legend (258K)

We next determined which genes were most strongly differentially expressed between the two sample clusters. Using SAM software with a false discovery rate cutoff of 0.02, 392 transcripts were significantly overexpressed in the first cluster and 374 transcripts were overexpressed in the second cluster (Figure 2b and Supplementary Data). We used the Marmite web tool (Al-Shahrour et al., 2006) to detect differential distributions of terms in PubMed between transcripts overexpressed in either cluster. Only 8 of the genes overexpressed in cluster-1 cell lines belonged to the Marmite entity 'glioblastoma,' whereas 27 of the genes overexpressed by cluster-2 cell lines belonged to this entity (adjusted P-value <0.05) (Supplementary Table 2). Thus, genes overexpressed by the cluster-2 cell lines were significantly more often the subject of previously published glioblastoma-related studies than genes overexpressed by cluster-1 cell lines.

Using the FatiGo web tool (Al-Shahrour et al., 2004), we searched for significant associations of transcripts overexpressed in either the cluster-1 or cluster-2 cell lines with terms in the Gene Ontology database. Only two terms were significant, both with higher expression in cluster-1 cell lines (adjusted P-value <0.05). These terms were 'nervous system development' and 'neuropeptide signaling pathway,' belonging to the Gene Ontology category 'biological process,' levels 5 and 7, respectively (Figure 2c and Supplementary Table 3).

The involvement of differentially expressed genes in biological pathways was analysed using the FatiGO+ web tool for testing KEGG (Kyoto Encyclopedia of Genes and Genomes) pathways (Kanehisa et al., 2004; Al-Shahrour et al., 2006). Of the 392 transcripts overexpressed in the cluster-1 cell lines, 14 were annotated to a total of 20 different pathways, and of the 374 transcripts overexpressed in the cluster-2 cell lines, 36 were annotated to 43 pathways. The most obvious differences between the 2 groups were that of the 14 annotated transcripts from the first group, 2 (14.3% ) were part of the 'Notch signaling pathway' (DLL3 and HES5), whereas none from the second group belonged to this pathway. Conversely, of the 36 annotated transcripts from the second group, 5 were part of the pathway 'ECM-receptor interaction' (ITGB1, COL1A2, SDC4, ITGB5, CD44), whereas none from the first group belonged to this pathway. All other differences between differentially expressed transcripts annotated in the KEGG pathway database were less pronounced (data not shown).

To validate selected genes that were highly overexpressed in either the cluster-1 or cluster-2 cell lines and were identified by Gene Ontology or pathway analysis, we performed quantitative PCR analyses. Highest expression levels of genes overexpressed in the cluster-1 cell lines were confirmed by this analysis, including BCAN, CROC4, ELAVL3, OLIG2, DLL3, ROBO2, MAP2, BAI1 and NES (nestin) (Figure 3). Conversely, strongest expression of MET, LOX and CAV-1 was confirmed in the cluster-2 cell lines and CD44 was most strongly expressed in cell lines GS-1 and GS-4.

Figure 3.

Real-time PCR analysis of selected genes overexpressed in cluster-1 or cluster-2. Relative amounts of target mRNA were normalized to RPL-13A as internal control, values were calibrated according to the C_T method and relative quantity values (RQv) were calculated. In the inset, the mean overexpression in the two clusters is listed; for genes represented more than once on the arrays, the highest of the fold values is shown.

Full figure and legend (188K)

GS cell lines have different capacity for differentiation

Glioma stem cells are characterized by the expression of stem cell markers and the capacity for multilineage differentiation. CD133, nestin and Sox2 (SRY-related HMG-box gene 2) are frequently used as markers for an undifferentiated state. Analysing CD133 expression by flow cytometry, we detected between 1.9 and 6.4% CD133⁺ cells in the cluster-1 cell lines (Figure 4a). In contrast, cluster-2 cell lines contained no detectable CD133⁺ cells. Cell line GS-2, which is unusual since it was derived from a long-term survivor of a glioblastoma with oligodendroglial features, contained 1% CD133⁺ cells.

Figure 4.

Flow cytometric analysis of undifferentiated and differentiated glioblastoma stem-like (GS) cells. Mechanically dissociated undifferentiated cells were incubated with antibodies against CD133, nestin or SRY-related HMG-box gene 2 (Sox2), and the percentage of immunoreactive cells was analysed (a). Over 6 days of differentiation in medium containing 10% fetal calf serum, 1 mM retinoic acid and 10 mg ml^{- 1} cAMP, immunoreactivity for neurofilament (NF), galactocerebroside (GalC) and glial fibrillary acidic protein (GFAP) was analysed. Results for GS-9 are shown as a representative example. Asterisks in the lower panel indicate a significant increase in the percentage of immunoreactive cells compared with day 0 (before differentiation) (b). Maximal increase in the percentage of immunoreactive cells determined on day 4 (d4) or day 6 (d6) of differentiation compared with day 0. All values listed differ significantly from those measured on day 0 (P<0.05) (c). Values in (a–c) are meanss.d. of triplicate determinations.

Full figure and legend (214K)

Using fluorescence immunocytochemistry, we detected strong nestin expression in most undifferentiated GS cell lines (Figure 5a). By flow cytometry, 20.8–66.7% of the cells were found to be nestin⁺ in the different GS cultures (Figure 4a). GS-1 and GS-4 displayed relatively low proportions of nestin⁺ cells compared with the other cell lines. Sox2 was also detectable in all GS cell cultures and typically localized to the nuclei (Figure 5b). By flow cytometry, 1.4–16.2% of Sox2⁺ cells were detected in the different GS cell lines, and cluster-2 lines, except for GS-2, displayed relatively low proportions of Sox2⁺ cells compared with most cluster-1 cultures, although GS-3 contained the lowest proportion of Sox2⁺ cells (Figure 4a).

Figure 5.

Fluorescence immunocytochemistry of glioblastoma stem-like (GS) cells before and after differentiation. Undifferentiated spheres contained numerous nestin immunoreactive cells (a), and fewer cells showed nuclear reactivity for SRY-related HMG-box gene 2 (Sox2) (b). After 3 days of differentiation, many cells expressed microtubule-associated protein 2 (MAP2) (c) and neurofilament (NF) (d). Neurofilament and nestin were often coexpressed, displaying different sublocalizations within individual cells (e). Only after 12 days of differentiation, individual cells stained clearly for galactocerebroside (GalC), showing a typical oligodendrocytic morphology with multiple long processes (f). Undifferentiated spheres contained glial fibrillary acidic protein (GFAP) immunoreactive cells (g), and also many of the attached cells after 3 days of differentiation were GFAP positive (h). Stainings for NF and GFAP (i), MAP2 and GFAP (j) as well as GalC and GFAP (k) were mutually exclusive, whereas nestin and GFAP (l) were frequently coexpressed (all analysed after 3 days).

Full figure and legend (607K)

We next assessed the multipotency of GS cell lines by exposing them to medium containing 10% fetal calf serum, retinoic acid and cAMP, which induces neuronal and glial differentiation. After overnight incubation in differentiation medium, most cells attached to the culture plate. After 3 days, fluorescence immunohistochemistry typically showed numerous cells immunoreactive for the neuronal markers MAP2 and neurofilament (NF) (Figures 5c and d). Notably, many of the cells that expressed NF were multinucleated. During normal neural development, nestin is downregulated in the process of terminal neuronal differentiation and becomes replaced by NFs (Dahlstrand et al., 1992). We, therefore, analysed the coexpression of nestin and NFs in GS cells and found frequent coexpression of both during early differentiation (Figure 5e). In contrast, coexpression was rare during later stages of differentiation (data not shown), suggesting that nestin is replaced by NFs over time in GS cells. The proportion of NF⁺ cells was quantified by flow cytometry on day 0 (before differentiation), and after 4 and 6 days of differentiation. The greatest increase in NF⁺ cells was typically observed on day 4 (Figure 4b), and was significant for all GS cell lines analysed, ranging from 21.4 to 541.4% of the cells (Figure 4c). Notably, after 4 days of differentiation the proportion of NF⁺ cells decreased again, a phenomenon that has similarly been noted by others for class III -tubulin as well as for GFAP (glial fibrillary acidic protein), and was interpreted as a capacity of brain tumor stem cells to resist differentiation (Zhang et al., 2006).

To assess oligodendroglial differentiation, we analysed the expression of GalC (galactocerebroside) (Ranscht et al., 1982). After 3 days of differentiation, numerous cells were suspected to be faintly immunoreactive by fluorescence immunocytochemistry; however, the signals were so weak that they were mostly indistinguishable from background (not shown). After 6 days, still no clear signals were detectable, but after 12 days individual cells with strong GalC immunoreactivity became clearly distinguishable, typically displaying an oligodendrocytic morphology with small cell bodies and multiple delicate processes (Figure 5f). In contrast to these findings, all cell lines displayed a significant increase in the proportion of GalC⁺ cells after 4 and/or 6 days of differentiation by flow cytometry, ranging from 54.6 to 977.0% (Figures 4b and c). Thus, upregulation of GalC obviously occurs at an earlier time point than detectable by immunocytochemistry. Interestingly, cluster-2 cell lines except for GS-2 showed the lowest potential for oligodendroglial differentiation compared with the other cell lines.

GFAP is a marker for astrocytic differentiation but is also expressed by neural stem cells (Doetsch et al., 1999). GFAP was already expressed in many of the undifferentiated cultures (Figure 5g). Under differentiating conditions, GFAP expression increased in all cluster-1 cell lines, but it either decreased or remained unaltered in the cluster-2 lines (Figure 5h; Figures 4b and c), exemplifying the phenotypic diversity between these two groups of cultures.

To analyse whether GFAP was expressed exclusively in differentiated cells or could be coexpressed with other differentiation markers, we performed double-staining experiments. Immunoreactivity for NF and GFAP or MAP2 and GFAP was not found in the same cells (Figures 5i and j). Likewise, no coexpression of GFAP and GalC (after 12 days) was detected (Figure 5k). In contrast, cells that expressed GFAP frequently showed co-immunoreactivity for nestin during early differentiation (Figure 5l), but less so during later stages (6 or 12 days, data not shown), suggesting that nestin can become downregulated in favor of GFAP during astrocytic differentiation.

Subcolony formation and proliferation rate

Self-renewal is a critical feature of neural stem cells and cancer stem cells (Pilkington, 2005; Vescovi et al., 2006). We performed limiting dilution assays to analyse whether single GS cells had the capacity to form new subspheres or colonies. Subsphere formation was observed for all cluster-1 cell lines as well as for the partly adherent cell lines GS-2 and GS-7 with a frequency between 16.9 and 53.8% . Adherent cell lines GS-1 and GS-4 were also clonogenic (21.1 and 19.7% , respectively), but formed adherent colonies rather than spheres.

We further analysed whether cluster-1 cell lines differed from cluster-2 cell lines in their proliferation rate. There were no consistent differences between the two groups (Supplementary Figure 1).

In vivo growth of glioma spheres

To address whether GS cell lines were tumorigenic in vivo, cells were xenografted into the striatum of nude mice. For each cell line, three animals were injected with 150 000 cells and at least three with 1000 cells. All mice that received cluster-1 cell lines or the GS-2 line developed tumors, regardless of the cell number injected (Table 2). The other cell lines displayed either no tumor take or reduced tumorigenicity. Only one of three and one of ten mice developed a tumor when 1000 or 150 000 GS-7 cells were injected respectively. Similarly, only one mouse that received 150 000 GS-1 cells developed a tumor, which took 18 months to become symptomatic (Table 2). GS-4 cells were not tumorigenic after 18 months of observation. We found no association between survival time (Table 2) and the proliferation rate of GS cell lines in vitro (Supplementary Figure 1).

Table 2 - In vivo tumorigenicity.

Full table

The morphology of tumors varied between the different GS lines. Whereas cluster-1 cell lines always formed highly infiltrative tumors and also the single GS-1 tumor grew diffusely, tumors originating from GS-2 and GS-7 grew comparatively well circumscribed (Figures 6a–g). In addition, tumors derived from GS-2 and GS-7 tended to expand extracerebrally. In these extracerebral components, occasional necrotic foci were detectable, but no true pseudopalisading was found. In all other tumors necroses were absent. Likewise, no true microvascular proliferations, which are characteristic of human glioblastomas, were found in any tumor, despite occasional prominent vascular structures with thickened endothelium.

Figure 6.

Tumor growth in vivo. Glioblastoma stem-like (GS) cell lines were xenografted into the caudate/putamen of nude mice, and animals were killed when they developed symptoms. Diffusely growing tumors such as those derived from GS-5 and GS-8 caused enlargement of the brain and hyperemia (a). Tumors derived from GS-5 (b) and GS-8 (c) infiltrated the brain diffusely, including the corpus callosum and contralateral hemisphere. Tumors derived from GS-2 grew well circumscribed (d). Diffuse single cell infiltration for GS-5 tumors (e) and GS-8 tumors (f) was confirmed at higher magnification, whereas GS-2 tumors (g) showed comparatively sharply delineated borders. Immunohistochemically, nestin (h), SRY-related HMG-box gene 2 (Sox2) (i), glial fibrillary acidic protein (GFAP) (j), microtubule-associated protein 2 (MAP2) (k) and CNPase (l) were detected in all tumors.

Full figure and legend (904K)

Frozen sections of the tumors were immunostained for nestin, Sox2, GFAP, MAP2, GalC and CNPase. Except for GalC, all antigens were detected in all tumors analysed (Figures 6h–l). Nestin, Sox2 and GFAP were typically present diffusely throughout the tumors (Figures 6h–j), whereas MAP2 was frequently detected in small groups of cells (Figure 6k). As GalC is a marker characteristic of advanced oligodendroglial differentiation, which might not be supported by the microenvironment in murine brain, we also stained for CNPase, a marker for earlier oligodendroglial lineage selection (Glaser et al., 2005), which was detectable in all tumors but was typically restricted to individual cells (Figure 6l).

Discussion

Distinct subtypes of glioblastoma have been identified by molecular genetic studies (Ohgaki and Kleihues, 2007) and gene expression analyses (Phillips et al., 2006), and some indications of heterogeneity have also been obtained for glioblastoma stem cell-enriched cultures (Galli et al., 2004; Beier et al., 2007). From our analysis of nine different glioblastoma cultures established under neural stem cell conditions, we propose that two major phenotypes exist. The first type (cluster-1) is characterized by the expression of neurodevelopmental genes and displays a full stem-like phenotype with spherical growth in vitro, expression of CD133, broad neuro-glial differentiation capacity and 100% tumorigenicity with invasive growth in vivo. The second type (cluster-2) displays an expression signature enriched for extracellular matrix-related genes and only a restricted stem-like phenotype, fulfilling only part of the criteria considered typical of glioblastoma stem cells.

Initially, we took 19 glioblastomas into culture, but 10 cultures did not survive the first or second split. This success rate in establishing permanent cultures is similar to the 50% rate achieved by others (Galli et al., 2004). No DNA copy number alterations were unique to tumors that either did or did not exhibit the capacity to establish expandable cultures. Similarly, no genomic gains or losses were identified in the primary tumors that correlated with the phenotype of the cultures. Interestingly, however, tumors that gave rise to cluster-1 GS cell lines never displayed nuclear accumulation of p53, whereas most tumors that generated cluster-2 lines did, suggesting that alterations in the p53 pathway tend to be associated with a more restricted stem-like phenotype.

Some evidence of heterogeneity among glioblastoma stem cell-enriched cultures was reported previously (Galli et al., 2004; Beier et al., 2007). Galli et al. (2004) identified three fast- and three slow-growing cell lines, showing by reverse transcription–PCR analysis of 21 selected genes that the former shared similar gene expression patterns, whereas the latter displayed more variable patterns. Heterogeneity of glioblastoma-derived stem cell-enriched cultures was further found by Beier et al. (2007), who discovered different growth characteristics and gene expression patterns mostly related to major histocompatibility complex class II proteins. In contrast to and in expansion of these studies, our present study suggests that the expression of neurodevelopmental genes as opposed to extracellular matrix genes may be crucial for the full stem-like phenotype.

Among the markers overexpressed by the cluster-1 cell lines were elements of the Notch signaling pathway, including DLL3, HES6 and NOTCH4. Notch signaling plays a key role in maintaining neural precursor cells in a multipotent dividing state and has been implicated in contributing to the stem-like character of both glioma and medulloblastoma models (Kageyama and Ohtsuka, 1999; Fan et al., 2006; Shih and Holland, 2006). Our findings suggest that increased activation of the Notch pathway contributes to the full stem-like phenotype of our cell lines. Interestingly, many of the markers that distinguish between the cluster-1 and cluster-2 cultures have been previously reported as markers of the prognostically significant 'proneural' and 'mesenchymal' subtypes of human glioblastoma (Phillips et al., 2006). However, whether cluster-1 and cluster-2 culture phenotypes arise from tumors with distinct clinical outcomes can not be established on the basis of the small sample population included in the current study.

Cluster-1 cell lines, which all grew spherically, contained the highest proportion of CD133⁺ cells, ranging from 1.9 to 6.4% , whereas only one of the semiadherent and none of the adherent cultures demonstrated detectable CD133⁺ cells (GS-2, 1% ). This association between growth pattern and CD133 immunoreactivity is consistent with previous studies on cells sorted from primary tumors (Singh et al., 2003; Bao et al., 2006) and with observations from unsorted CD133⁺ and CD133^- glioblastoma-derived cultures (Beier et al., 2007). In addition, we found an association with the growth pattern of orthotopic grafts. Cluster-1 GS cell lines, which contained CD133⁺ cells, were 100% tumorigenic and grew highly invasive but non-necrotic, whereas the remaining cultures showed greatly restricted tumorigenicity, more compact growth and occasionally necroses. The in vivo phenotype of cluster-1 cell lines is thus largely consistent with previous reports on orthotopic grafts of glioblastoma neurosphere cultures (Galli et al., 2004; Tunici et al., 2004; Yuan et al., 2004; Lee et al., 2006; Piccirillo et al., 2006). In the more compact growing cluster-2 culture-derived tumors, necroses were confined to parts that grew exophytically outside the brain, suggesting that a relative lack of cooptable host vasculature (Holash et al., 1999) in the extracerebral intracranial compartment may account for the propensity of tumors to necrotize in this location. Taken together, the striking differences in the in vivo growth characteristics between cluster-1 and cluster-2 cell cultures provide further support for the existence of two distinct phenotypes.

Although most of the cluster-2 cell lines showed no detectable CD133⁺ cells, they, however, contained cells expressing the stem/progenitor markers Sox2 and nestin. Furthermore, these lines were clonogenic, with GS-2 and GS-7 even forming new subspheres, and giving rise to increasing numbers of NF⁺ and GalC⁺ cells under differentiating conditions. In contrast to the cluster-1 cell lines, however, they were incapable of producing increased proportions of GFAP⁺ cells. Although it cannot be excluded by our findings that some degree of astrocytic differentiation also occurred in cluster-2 cell lines during the 6-day observation period, they obviously showed a reduced propensity for differentiation along the astrocytic lineage relative to cluster-1 cell lines. Thus, despite their clonogenicity and expression of Sox2 and nestin, the cluster-2 cell lines displayed only restricted differentiation capacity.

To conclude, on the basis of gene expression profiling and by neurobiological criteria, we identified clear subtypes among human glioblastoma cultures established under neural stem cell conditions. Whereas some cell lines fulfilled all criteria described previously for human glioma stem cell cultures, others showed only a restricted stem cell phenotype. The extent to which these differences reflect variations in stem-like cells present between individual human glioblastoma cases and/or represent distinct emerging lineages remains to be established. With the stem cells being the prime suspect responsible for the inevitable recurrence of glioblastomas, our findings suggest that for these critical cellular components of the tumor, very diverse strategies may have to be developed.

Materials and methods

Establishment and growth of glioma sphere cultures

Nineteen fresh glioblastoma samples were obtained from patients operated in the Department of Neurosurgery, University Medical Center Hamburg-Eppendorf and were processed as described by Galli et al. (2004) with modifications. Samples were cut into <1 mm³ fragments, washed with Hanks' balanced salt solution (Invitrogen, Carlsbad, CA, USA), and digested with 1 mg ml^{- 1} collagenase/dispase (Roche, Mannheim, Germany) for 30 min at 37 °C. Digested fragments were filtered using a 70 m cell mesh (Sigma-Aldrich, St Louis, MO, USA), and the cells were seeded into T25 flasks at 2500–5000 cells cm^{- 2}. The culture medium (neurosphere medium) consisted of neurobasal medium (Invitrogen) with B27 supplement (20 l ml^{- 1}; Invitrogen), Glutamax (10 l ml^{- 1}; Invitrogen), fibroblast growth factor-2 (20 ng ml^{- 1}; Peprotech, Rocky Hill, NJ, USA), epidermal growth factor (20 ng ml^{- 1}; Peprotech) and heparin (32 IE ml^{- 1}; Ratiopharm, Ulm, Germany). Growth factors and heparin were renewed twice weekly. Spheres were split by mechanical dissociation when they reached a size of 200–500 m.

SNP chip analysis

Affymetrix 100K SNP chips (Affymetrix, Santa Clara, CA, USA) were used to analyse relative copy number changes in 9 tumors that generated GS cell cultures and in 10 tumors that did not generate expandable cultures. The CNAT SPACN algorithm provided by Affymetrix was applied to estimate copy number changes. Amplification of a locus was defined as a mean copy number estimate of >4 for SNPs within the locus or, in the case of platelet-derived growth factor receptor- and CDK4, for SNPs immediately flanking both sides of the locus. Copy number loss was defined as a statistically significant reduction (P<0.05) in CNAT SPACN values across the set of 10 SNPs within or closest to the locus.

Analysis of TP53 mutations

Tumor specimens as well as GS cell lines were analysed for mutations in all exons of the TP53 gene by direct sequencing in both directions with primers located in the flanking introns (primer sequences are available upon request).

Differentiation of glioma spheres

Mechanically dissociated cells were transferred into eight-chamber Lab Tec-slides (Nalge Nunc International, Rochester, NY, USA) at 5000 cells per well for immunocytochemistry or into six-well chambers (Nalge Nunc) at 25 000 cells per well for flow cytometry. Differentiation medium was added, consisting of neurobasal medium with B27 supplement (20 l ml^{- 1}), Glutamax (10 l ml^{- 1}), 100 U ml^{- 1} penicillin, 100 g ml^{- 1} streptomycin, 0.25 g ml^{- 1} fungizone, 10% fetal calf serum (all from Invitrogen), 1 mM retinoic acid and 10 mg ml^{- 1} cAMP (both from Sigma-Aldrich). On days 1, 3, 4, 6 and 12, cells were analysed by flow cytometry or immunocytochemistry as described below.

Immunocytochemistry and immunohistochemistry

Cells or frozen sections prepared from the mouse brain were fixed with 4% paraformaldehyde. Cells or tissue sections were permeabilized with 3% Triton X-100 in phosphate-buffered saline (PBS) (except for GalC staining) and blocked with 5% horse serum. Primary antibodies were mouse anti-nestin (1:200; Chemicon, Temecula, CA, USA), rabbit anti-nestin (1:200; Chemicon), rabbit anti-GFAP (1:40; Dako, Glostrup, Denmark), mouse anti-GalC, (1:100; Chemicon), mouse anti-MAP2 (1:50; Chemicon), mouse anti-NF (1:500; Sternberger Monoclonals Inc., Lutherville, MD, USA) and mouse anti-Sox2 (1:50; R&D Systems, Minneapolis, MN, USA). After incubation for 90 min the slides were washed with 5% horse serum. Secondary antibodies, goat anti-rabbit-Alexa 546 (1:300; Invitrogen) and donkey anti-mouse-Alexa 488 (1:300; Invitrogen), were added for 30 min. Slides were mounted using Vectashield Hard Set mounting medium with DAPI (Vector Laboratories, Burlingame, CA, USA). For double staining, mouse antibodies were added first for 90 min, followed by the addition of the rabbit antibody for 60 min and simultaneous detection with secondary antibodies.

Immunohistochemistry for p53 was routinely performed in the Department of Neuropathology, University Medical Center Hamburg-Eppendorf. Briefly, paraffin sections of tumors were dewaxed and stained using a mouse monoclonal antibody against p53 (clone DO-7, Dako, Glostrup, Denmark), followed by detection using the Envision kit (Dako).

Flow cytometry

Cells were dissociated mechanically. For detection of intracellular antigens, cells were washed with 0.1% saponin/PBS (incubation buffer), fixed with 4% paraformaldehyde and permeabilized using 0.5% saponine/PBS. After washing with incubation buffer, cells were incubated with mouse anti-nestin (1:200; Chemicon), rabbit anti-GFAP (1:50; Dako), mouse anti-GalC (1:50; Chemicon), mouse anti-MAP2 (1:50; Chemicon), mouse anti-NF (1:25; Dako), mouse anti-Sox2 (1:25; R&D Systems), rabbit IgG (1:250; Dako) or mouse IgG₁ (1:20; BD Biosciences, San Jose, CA, USA) in 0.1% saponin/PBS for 30 min at 4 °C. After washing, FITC-conjugated swine anti-rabbit (1:40; Dako) or FITC-conjugated goat anti-mouse (1:20; Dako) was added and incubated for 30 min at 4 °C. Cells were counterstained with CyStain DNA 1 step (Partec, Münster, Germany). Analysis was performed using a PAS Particle Analysing System (Partec).

For detection of CD133, unfixed dissociated cells were adjusted to a volume of 200 l in PBS. Either 10 l anti-CD133/1-PE and 10 l anti-CD133/2-PE (both Miltenyi Biotech, Bergisch Gladbach, Germany) or 10 l mouse IgG1-PE (Santa Cruz, Santa Cruz, CA, USA) as negative control were added together with 20 l FcR-blocking reagent (Miltenyi Biotec), and incubated for 30 min at 4 °C. After washing with PBS, cells were analysed within 1 h by flow cytometry.

Differences in the proportion of immunoreactive cells during differentiation were analysed using the unpaired t-test or the Mann–Whitney rank-sum test and the SigmaStat program (version 2.0).

Single colony formation analysis

The formation of subspheres or colonies from single cells was analysed using a limiting dilution assay. Mechanically dissociated cells were seeded into 96-well plates at a theoretical density of 1 cell per well in neurosphere medium. After overnight culture, microscopic observation was utilized to identify wells that contained a single cell. Colony formation was scored 14 days after initial seeding. The percentage of cells that formed spheres was determined, and meanss.d. were calculated by pooling the results of three different experiments.

Proliferation analysis

Cells were seeded into 24-well plates in neurosphere medium (2000 cells per well). Cell counts were determined on days 1, 4 and 7 after seeding.

Analysis of tumorigenicity in vivo

In vivo tumors were generated as described previously (Kunkel et al., 2001). Briefly, 150 000 or 1000 dissociated tumor cells in 5 l neurobasal medium were injected into the basal ganglia of 6- to 8-week-old anesthetized NMRI-nu/nu mice. Animals were killed when they developed weight loss greater than 10% or when they developed neurological symptoms.

Microarray analysis

Total RNA was extracted from cells using the RNeasy Protect Mini Kit (Qiagen, Hilden, Germany). For each GS cell culture (with the exception of GS-6), two samples taken from different passages were utilized for expression profiling. Genomic DNA contamination was removed through an on-column DNase digestion step. Affymetrix HG-U133 Plus 2.0 chips were employed for expression profiling. The data have been submitted to the GEO database (Barrett et al., 2007), accession number GSE8049. Quality control of data sets was performed according to the manufacturer's manual (www.affymetrix.com/support/technical/manual/expressionmanual.affx). The background-adjusted, normalized intensities were computed using the gcRMA package (Wu et al., 2004). Gene detection was performed using the PANP package (http://bioconductor.org/packages/bioc/vignettes/panp/inst/doc/panp.pdf) for the R program (version 2.3.1). For cluster analysis, agglomerative clustering was performed utilizing Cluster 3.0 (de Hoon et al., 2004). Data were analysed for differentially expressed genes by repeated permutation testing using the SAM software (Tusher et al., 2001) with the false discovery rate cutoff set to 0.02. The Marmite web tool (Al-Shahrour et al., 2006) was used to find differential distributions of bioentities extracted from PubMed between groups of differentially expressed genes. The FatiGO web tool was used for functional analysis of differentially expressed genes based on the Gene Ontology database (Al-Shahrour et al., 2004). The involvement of genes in defined biological pathways was analysed using the FatiGO+ web tool (Al-Shahrour et al., 2006) for testing Kyoto Encyclopedia of Genes and Genomes pathways (Kanehisa et al., 2004).

Real-time PCR analysis

RNA was extracted from cultured cells using Trizol-Reagent (Invitrogen), and cDNA was synthesized using random primers (MWG Biotech, Ebersberg, Germany) and Superscript II reverse transcriptase (Invitrogen). Gene expression analyses were performed using TaqMan Gene Expression Assays (Applied Biosystems, Foster City, CA, USA). For primers, PCR conditions and relative quantity analysis, see Supplementary Information.

References

1. Al-Shahrour F, Diaz-Uriarte R, Dopazo J. (2004). FatiGO: a web tool for finding significant associations of Gene Ontology terms with groups of genes. Bioinformatics 20: 578–580. | Article | PubMed | ISI | ChemPort |
2. Al-Shahrour F, Minguez P, Tarraga J, Montaner D, Alloza E, Vaquerizas JM et al. (2006). BABELOMICS: a systems biology perspective in the functional annotation of genome-scale experiments. Nucleic Acids Res 34: W472–W476. | Article | PubMed | ISI | ChemPort |
3. Bao S, Wu Q, McLendon RE, Hao Y, Shi Q, Hjelmeland AB et al. (2006). Glioma stem cells promote radioresistance by preferential activation of the DNA damage response. Nature 444: 756–760. | Article | PubMed | ISI | ChemPort |
4. Barrett T, Troup DB, Wilhite SE, Ledoux P, Rudnev D, Evangelista C et al. (2007). NCBI GEO: mining tens of millions of expression profiles—database and tools update. Nucleic Acids Res 35: D760–D765. | Article | PubMed | ISI | ChemPort |
5. Beier D, Hau P, Proescholdt M, Lohmeier A, Wischhusen J, Oefner PJ et al. (2007). CD133(+) and CD133(- ) glioblastoma-derived cancer stem cells show differential growth characteristics and molecular profiles. Cancer Res 67: 4010–4015. | Article | PubMed | ChemPort |
6. Clarke MF, Dick JE, Dirks PB, Eaves CJ, Jamieson CH, Jones DL et al. (2006). Cancer stem cells—perspectives on current status and future directions: AACR Workshop on cancer stem cells. Cancer Res 66: 9339–9344. | Article | PubMed | ChemPort |
7. Dahlstrand J, Zimmerman LB, McKay RD, Lendahl U. (1992). Characterization of the human nestin gene reveals a close evolutionary relationship to neurofilaments. J Cell Sci 103(Part 2): 589–597. | PubMed | ISI | ChemPort |
8. de Hoon MJ, Imoto S, Nolan J, Miyano S. (2004). Open source clustering software. Bioinformatics 20: 1453–1454. | Article | PubMed | ISI | ChemPort |
9. Doetsch F, Caille I, Lim DA, Garcia-Verdugo JM, Alvarez-Buylla A. (1999). Subventricular zone astrocytes are neural stem cells in the adult mammalian brain. Cell 97: 703–716. | Article | PubMed | ISI | ChemPort |
10. Fan X, Matsui W, Khaki L, Stearns D, Chun J, Li YM et al. (2006). Notch pathway inhibition depletes stem-like cells and blocks engraftment in embryonal brain tumors. Cancer Res 66: 7445–7452. | Article | PubMed | ChemPort |
11. Galli R, Binda E, Orfanelli U, Cipelletti B, Gritti A, De Vitis S et al. (2004). Isolation and characterization of tumorigenic, stem-like neural precursors from human glioblastoma. Cancer Res 64: 7011–7021. | Article | PubMed | ISI | ChemPort |
12. Glaser T, Perez-Bouza A, Klein K, Brustle O. (2005). Generation of purified oligodendrocyte progenitors from embryonic stem cells. FASEB J 19: 112–114. | PubMed | ChemPort |
13. Gritti A, Parati EA, Cova L, Frolichsthal P, Galli R, Wanke E et al. (1996). Multipotential stem cells from the adult mouse brain proliferate and self-renew in response to basic fibroblast growth factor. J Neurosci 16: 1091–1100. | PubMed | ISI | ChemPort |
14. Hemmati HD, Nakano I, Lazareff JA, Masterman-Smith M, Geschwind DH, Bronner-Fraser M et al. (2003). Cancerous stem cells can arise from pediatric brain tumors. Proc Natl Acad Sci USA 100: 15178–15183. | Article | PubMed | ChemPort |
15. Holash J, Maisonpierre PC, Compton D, Boland P, Alexander CR, Zagzag D et al. (1999). Vessel cooption, regression, and growth in tumors mediated by angiopoietins and VEGF. Science 284: 1994–1998. | Article | PubMed | ISI | ChemPort |
16. Ignatova TN, Kukekov VG, Laywell ED, Suslov ON, Vrionis FD, Steindler DA. (2002). Human cortical glial tumors contain neural stem-like cells expressing astroglial and neuronal markers in vitro. Glia 39: 193–206. | Article | PubMed | ISI |
17. Kageyama R, Ohtsuka T. (1999). The Notch–Hes pathway in mammalian neural development. Cell Res 9: 179–188. | Article | PubMed | ISI | ChemPort |
18. Kanehisa M, Goto S, Kawashima S, Okuno Y, Hattori M. (2004). The KEGG resource for deciphering the genome. Nucleic Acids Res 32: D277–D280. | Article | PubMed | ISI | ChemPort |
19. Kondo T, Setoguchi T, Taga T. (2004). Persistence of a small subpopulation of cancer stem-like cells in the C6 glioma cell line. Proc Natl Acad Sci USA 101: 781–786. | Article | PubMed | ChemPort |
20. Kunkel P, Ulbricht U, Bohlen P, Brockmann MA, Fillbrandt R, Stavrou D et al. (2001). Inhibition of glioma angiogenesis and growth in vivo by systemic treatment with a monoclonal antibody against vascular endothelial growth factor receptor-2. Cancer Res 61: 6624–6628. | PubMed | ChemPort |
21. Lee J, Kotliarova S, Kotliarov Y, Li A, Su Q, Donin NM et al. (2006). Tumor stem cells derived from glioblastomas cultured in bFGF and EGF more closely mirror the phenotype and genotype of primary tumors than do serum-cultured cell lines. Cancer Cell 9: 391–403. | Article | PubMed | ISI | ChemPort |
22. Liu G, Yuan X, Zeng Z, Tunici P, Ng H, Abdulkadir IR et al. (2006). Analysis of gene expression and chemoresistance of CD133+ cancer stem cells in glioblastoma. Mol Cancer 5: 67. | Article | PubMed | ChemPort |
23. Ohgaki H, Kleihues P. (2007). Genetic pathways to primary and secondary glioblastoma. Am J Pathol 170: 1445–1453. | Article | PubMed | ChemPort |
24. Phillips HS, Kharbanda S, Chen R, Forrest WF, Soriano RH, Wu TD et al. (2006). Molecular subclasses of high-grade glioma predict prognosis, delineate a pattern of disease progression, and resemble stages in neurogenesis. Cancer Cell 9: 157–173. | Article | PubMed | ISI | ChemPort |
25. Piccirillo SG, Reynolds BA, Zanetti N, Lamorte G, Binda E, Broggi G et al. (2006). Bone morphogenetic proteins inhibit the tumorigenic potential of human brain tumour-initiating cells. Nature 444: 761–765. | Article | PubMed | ISI | ChemPort |
26. Pilkington GJ. (2005). Cancer stem cells in the mammalian central nervous system. Cell Prolif 38: 423–433. | Article | PubMed | ChemPort |
27. Ranscht B, Clapshaw PA, Price J, Noble M, Seifert W. (1982). Development of oligodendrocytes and Schwann cells studied with a monoclonal antibody against galactocerebroside. Proc Natl Acad Sci USA 79: 2709–2713. | Article | PubMed | ChemPort |
28. Shih AH, Holland EC. (2006). Notch signaling enhances nestin expression in gliomas. Neoplasia 8: 1072–1082. | Article | PubMed | ChemPort |
29. Singh SK, Clarke ID, Terasaki M, Bonn VE, Hawkins C, Squire J et al. (2003). Identification of a cancer stem cell in human brain tumors. Cancer Res 63: 5821–5828. | PubMed | ISI | ChemPort |
30. Singh SK, Hawkins C, Clarke ID, Squire JA, Bayani J, Hide T et al. (2004). Identification of human brain tumour initiating cells. Nature 432: 396–401. | Article | PubMed | ISI | ChemPort |
31. Svendsen CN, ter Borg MG, Armstrong RJ, Rosser AE, Chandran S, Ostenfeld T et al. (1998). A new method for the rapid and long term growth of human neural precursor cells. J Neurosci Methods 85: 141–152. | Article | PubMed | ISI | ChemPort |
32. Taylor MD, Poppleton H, Fuller C, Su X, Liu Y, Jensen P et al. (2005). Radial glia cells are candidate stem cells of ependymoma. Cancer Cell 8: 323–335. | Article | PubMed | ISI | ChemPort |
33. Tunici P, Bissola L, Lualdi E, Pollo B, Cajola L, Broggi G et al. (2004). Genetic alterations and in vivo tumorigenicity of neurospheres derived from an adult glioblastoma. Mol Cancer 3: 25. | Article | PubMed | ChemPort |
34. Tusher VG, Tibshirani R, Chu G. (2001). Significance analysis of microarrays applied to the ionizing radiation response. Proc Natl Acad Sci USA 98: 5116–5121. | Article | PubMed | ChemPort |
35. Vescovi AL, Galli R, Reynolds BA. (2006). Brain tumour stem cells. Nat Rev Cancer 6: 425–436. | Article | PubMed | ISI | ChemPort |
36. Wu Z, Irizarry RA, Gentleman R, Martinez-Murillo F, Spencer F. (2004). A model-based background adjustment for oligonucleotide expression arrays. J Am Stat Assoc 99: 909–917. | Article | ISI |
37. Yuan X, Curtin J, Xiong Y, Liu G, Waschsmann-Hogiu S, Farkas DL et al. (2004). Isolation of cancer stem cells from adult glioblastoma multiforme. Oncogene 23: 9392–9400. | Article | PubMed | ISI | ChemPort |
38. Zhang QB, Ji XY, Huang Q, Dong J, Zhu YD, Lan Q. (2006). Differentiation profile of brain tumor stem cells: a comparative study with neural stem cells. Cell Res 16: 909–915. | Article | PubMed | ChemPort |

Acknowledgements

We thank Professor Dr Christian Hagel for access to neuropathological information and to viewing tumor slides. We also thank Regina Fillbrandt, Svenja Zapf and Dorothea Zirkel for expert technical assistance and Steve Guerrero and Pete Haverty for their advice and assistance in analysis of SNP chip data. KL and MW are supported by the Deutsche Forschungsgemeinschaft (LA1300/3-1) and by the Erich und Gertrud Roggenbuck-Stiftung. HG is supported by a scholarship of the FAZIT-Stiftung. Material funds were generously provided by the Rudolf-Bartling-Stiftung and the Georg and Jürgen Rickertsen Stiftung.

Supplementary Information accompanies the paper on the Oncogene website (http://www.nature.com/onc).