Main

A recent concept in brain tumor biology is that brain tumors arise from cancer stem cells (CSCs) that are CD133 positive (CD133(+)). It has been reported that a small number of CD133(+) glioblastoma multiforme (GBM) cells are able to recapitulate the original tumor in vivo, whereas millions of CD133-negative (CD133(−)) cells could not produce brain tumor masses.1, 2, 3, 4, 5, 6 However, accumulating evidence suggests that CD133(−) GBM cells can also regenerate heterogenous tumors in vivo,7, 8 and generation of the huge and rapidly growing tumors by only CD133(+) CSCs would be difficult because more than 50% of GBM patients have few CD133(+) cells.9

As a majority of neurogenic astrocytes in the adult brain are not recognized by a CD133 antibody,8 it is likely that CD133 might be newly expressed in GBM CSCs that are derived from CD133(−) adult neural stem cells (NSCs) or terminally differentiated brain cells, such as astrocytes, neurons, and oligodendrocytes. Given that the gene expression profile is changed when GBM recurs after treatments,10 it is plausible that new CD133 expression may occur if the characteristics of CSCs are changed or if some CSCs are selected by treatment. Furthermore, the wide-range variation in CD133(+) cell ratio (0.1–50% in GBM patients)1, 2, 3, 4, 5, 6 also suggests the existence of other GBM CSCs that do not express CD133.

Therefore, we hypothesize that there are several kinds of CSCs in the tumor mass of GMB, and these diverse CSCs possess diverse cellular and molecular characteristics, such as tumor initiation potential, chemo-radiation resistance, and gene expression profile.

MATERIALS AND METHODS

Tumor Specimens and Primary Tumor Cultures

Following informed consent, tumor samples classified as GBMs based on World Health Organization criteria11 were obtained from patients undergoing surgical treatment at the Samsung Medical Center (Seoul, Korea) in accordance with the appropriate Institutional Review Boards. Tumors were enzymatically dissociated into single cells and red blood cells were removed by differential centrifugation. Dissociated cells were cultured in the ‘NBE’ conditions consisting of NeuroBasal Media (Invitrogen, Carlsbad, CA, USA), N2 and B27 supplements (Invitrogen), as well as human recombinant bFGF and EGF (50 ng/ml each; R&D Systems, Minneapolis, MN, USA). To differentiate cultured cells, cells were plated onto poly-L-lysine/laminin mixture-precoated culture dish (Invitrogen) and were subjected to growth in DMEM with 10% fetal bovine serum (10% FBS/DMEM; Cambrex, East Rutherford, NJ, USA). Immunofluorescence assay was performed using following antibodies: Tuj1 (Chemicon, Billerica, MA, USA), GFAP (Sigma, St Louis, MN, USA), and O4 (Chemicon).

Fluorescence-Activated Cell Sorting and Intracranial Cell Transplantation

Dissociated GBM cells were labeled with 10 μl anti-CD133/2-PE antibody (Miltenyi Biotec, Bergisch Gladbach, Germany) per 106 cells. CD133(+) cell ratios were analyzed, and CD133(+) and CD133(−) populations were sorted using a FACSAria machine (BD Biosciences, San Jose, CA, USA). A total of 5 × 103–2 × 104 purified cells suspended in 5 μl HBSS were stereotactically injected into 6-week-old NOD/SCID mouse brains (2 mm left and 1 mm anterior to the bregma, 2 mm deep).

Specimens and Immunohistochemistry

As shown neurological symptom, mice were sacrificed and their brains were removed. The brains were processed for paraffin or frozen section. Paraffin sections were stained with hematoxylin and eosin as per standard histopathological technique and the tumor volume was recorded (largest width2 × largest length × 0.5). Frozen sections were fixed with acetone and then immunohistochemistry was performed as described previously12 using following antibodies: CD31 (BD Pharmingen, San Diego, CA, USA), PCNA (Dako, Glostrup, Denmark), and p-Akt (Cell signaling technology, Danvers, MA, USA).

RNA Expression Array and Data Analysis

Total RNAs were isolated using TRIZOL reagent (Invitrogen) and further purified using RNeasy Mini Kit (Qiagen, Hilden, Germany). Samples were processed and hybridized to Affymetrix U-133 plus2 GeneChip Arrays according to the Affymetrix GeneChip Expression Analysis Technical Manual (Affymetrix, Santa Clara, CA, USA). All arrays were confirmed to be within acceptable minimal quality control parameters. The gene expression CEL files were normalized using Robust Multichip Averaging procedure and PM-MM difference model was used to obtain the expression values. The hierarchical cluster analysis was performed using R package software 2.6.0 using the Euclidean distance with complete linkage.

RESULTS

GBM Patients Divided into CD133-High and CD133-Low GBM Groups Show Different Clinical Characteristics

To evaluate our hypothesis, we first divided GBM patients into a CD133-high group (CD133(+) cell ratio ≥3%, n=7) and a CD133-low group (CD133(+) cell ratio <3%, n=13) by fluorescence-activated cell sorting (FACS) analysis with anti-CD133 antibody (Supplementary Figure 1, Table 1). We analyzed the patients’ clinical characteristics using MRI scan data (Figure 1a) to determine whether these groups possess similar or different clinical characteristics. As compared to CD133-high GBMs, CD133-low GBMs have tendency to be localized within the deeper structures of the brain and to show more invasive growth patterns and ventricle involvement (Figure 1b). However, there was no significant difference in sphere-forming capacity between cells derived from CD133-high and CD133-low GBMs, as cultured under ‘NBE’ conditions (NeuroBasal Media supplemented with N2, B27, EGF, and bFGF) (Table 1).

Table 1 Summary of patients and their clinical characteristics
Figure 1
figure 1

CD133-high and CD133-low GBM patient groups show different clinical characteristics. (a) Representative MRI scan data of four CD133-low patients (patient nos. 1, 3, 4, 5) and four CD133-high patients (patient nos. 16, 17, 18, 20) divided by CD133-positive cell ratio. (b) Comparisons of tumor location, invasiveness or multiplicity, and ventricle involvement between CD133-high patients and CD133-low patients. *P<0.05. (c) Comparison of disease progression rates between 11 CD133-low and seven CD133-high GBM patients after chemotherapy and radiotherapy.

CD133-Low and CD133-High GBMs Show Different Clinical Outcomes

As CD133-low GBMs showed more invasive morphologies in MRI scan, we tested whether disease progression after treatment is different between CD133-high and CD133-low GBM patients. The rate of disease progression after chemotherapy and radiotherapy was relatively higher in the CD133-low GBMs (45.5%) compared to CD133-high GBMs (28.6%) (Figure 1c). Therefore, these results indicate that CD133-high and CD133-low GBM patients have different clinical characteristics and outcomes.

CD133-Low and CD133-High GBMs Show Different Gene Expression Profiles

Next, we compared gene expression patterns between subgroups of patients whose GBMs had the lowest and highest CD133(+) cell ratios, termed ‘CD133-lowest’ (CD133(+) cell ratio <0.5%, n=3) and ‘CD133-highest’ (CD133(+) cell ratio >15%, n=4). Following cDNA microarray analysis, we performed a hierarchical clustering of 34 selected genes that are used for subclassification of GBM, such as proneuronal, mesenchymal, and proliferative subtypes.10, 11 As shown in Figure 2a, CD133-lowest GBMs showed a relative upregulation of genes that have been previously linked to the ‘mesenchymal’ or ‘proliferative’ subtype, which has worse clinical outcome, whereas CD133-highest GBMs showed higher expression of genes that are known to be upregulated in the ‘proneuronal’ GBM subtype.10 These results were further validated by real-time RT-PCR, showing that expression of DLL3 and SOX8 (proneuronal genes) is decreased, while YKL40 (a mesenchymal gene) was upregulated in the CD133-lowest GBMs (Figure 2b).

Figure 2
figure 2

CD133-lowest and CD133-highest GBMs show different gene expression profiles. (a) Heatmaps showing expression profiles of the genes classified into ‘proneuronal’, ‘mesenchymal’, and ‘proliferative’ subtypes in three CD133-lowest and four CD133-highest GBM tumor masses. (b) The expression levels of five selected genes (CD133, DLL3, SOX8, YKL40, and E2F7) in three CD133-lowest and four CD133-highest GBM tumor masses were validated by real-time RT-PCR. *P<0.05. (c) Heatmaps showing the expression profiles of the genes classified into ‘proneuronal’, ‘mesenchymal’, and ‘proliferative’ subtypes in two CD133-high GBM tumor masses (patient nos. 16 and 19) and their cognate cells, which are cultured in ‘NBE’ conditions for 10 days after acute dissociation from tumor masses.

Previous data showed that when dissociated GBM cells are cultured in ‘NBE’ conditions, which were originally designed to culture normal NSCs, they become very similar to normal NSCs and thereby recapitulate the genotype, gene expression patterns, and in vivo biology of GBMs.13 Interestingly, when cells dissociated from CD133-high GBM patients (no. 16 or 19) (CD133(+) cell ratios=19.90 and 45.54%, respectively) were subjected to growth in ‘NBE’ conditions for 10 days, expression of genes characteristic of the ‘proneuronal’ GBM subtype was markedly decreased, whereas expression of ‘mesenchymal’ or ‘proliferative’ GBM subtype-related genes increased (Figure 2c). However, both CD133(+) and CD133(−) GBM cells grown in 10% FBS/DMEM conditions for 10 days were prone to differentiate into various brain cells that express GFAP (astrocyte), Tuj1 (neuron), or O4 (oligodendrocyte) (data not shown). These results indicate that the gene expression pattern of CSCs can change depending on in vivo and in vitro conditions, without alteration of their differentiation property.

Both CD133-Low and CD133-High GBMs Show Stem Cell-Like Characteristics In Vitro

We compared the sphere-forming abilities of the CD133-low or CD133-high GBMs in the ‘NBE’ condition and found that 9 in 13 CD133-low GBMs (69%) made many spheres within a week (Figure 3a, Table 1). This frequency was comparable with that of CD133-high GBMs (six in seven, 86%). Even the three CD133-lowest GBMs (patient nos. 1, 2, 3; CD133(+) cell ratio <0.2%) made spheres well in vitro (Table 1). Spheres formed by either CD133-low or CD133-high GBMs had similar morphology and size (Figure 3a). When the sphere cells were forced to differentiate in 10% FBS/DMEM, most cells from both CD133-low and CD133-high GBMs showed morphology of differentiated cells and expressed differentiated neural cell markers such as Tuj1, GFAP, and O4 (Figure 3b; a representative data from GBM patient no. 12). These results suggest that cells from CD133-low GBMs also have stem cell-like characteristics in vitro.

Figure 3
figure 3

Both CD133-low and CD133-high GBMs show stem cell-like characteristics in vitro. (a) Neurosphere-forming ability in the ‘NBE’ condition was compared between cells dissociated from CD133-high and CD133-low GBMs. The representative photos of spheres derived from CD133-high and CD133-low GBMs (left panel). (b) Sphere cells cultured in the ‘NBE’ condition were differentiated with 10% FBS. Representative photos present expressions of Tuj1, GFAP, and O4 determined by immunofluorescence assay.

Both CD133(−) and CD133(+) GBM Cells Possess Tumor-Initiating Potential

As it has been recently documented that some GBM patients have CD133(−) CSCs and others have CD133(+),7 and low-grade gliomas possess multiple and spatially distinct clonal populations,14 we isolated CD133(+) and CD133(−) cells from tumor specimens of a number of GBM patients using FACS analysis (Figure 4a; a representative data from GBM patient no. 4) to compare tumor-initiating potential. When CD133(+) or CD133(−) cells purified by FACS sorting were injected into the brains of NOD/SCID mice (n=3–5), tumors were produced in about half of the injected brains, indicating that both cell types harbor tumor-initiating potential (CD133(+)=7 in 17, CD133(−)=11 in 20). In addition, both CD133(+) and CD133(−) cells purified from four out of six human GBM patients were able to generate tumor masses that show similar GBM histological features (Table 2, Figure 4b). Furthermore, tumor volumes in the mice injected with sorted CD133(+) or CD133(−) cells were not significantly different, and neurological symptoms were detected at almost the same time point from injections (Table 2). However, although the volume and overall histology of tumor masses produced by injection of sorted CD133(+) or CD133(−) cells were relatively similar, the tumor masses from CD133(−) cells contained more microvessels and PCNA(+)-proliferating cells (Figure 4c), whereas those from CD133(+) cells were found to have significantly higher levels of phosphorylated Akt (pAkt, Figure 4c). These results suggest that there are multiple kinds of tumor-initiating cells, such as CD133(−) and CD133(+) cells, in every GBM patient.

Figure 4
figure 4

Both CD133(−) and CD133(+) GBM cells possess tumor-initiating potential. (a) Representative FACS data showing CD133(+) and CD133(−) cell populations isolated from one of 20 GBM patients using FACS. (b) When orthotopically injected into NOD/SCID mice, both CD133(+) and CD133(−) cells derived from one GBM patient produced tumor masses. Representative photos showing brain tumor tissue sections stained with hematoxylin and eosin. (c) Representative photos showing CD31-positive, PCNA-positive, and pAkt-positive cells in the tumor masses derived from CD133(−) and CD133(+) cells. Quantitative data showing microvessel numbers, PCNA(+) and pAkt(+) cell numbers in tumors derived from CD133(−) and CD133(+) cells. *P<0.05.

Table 2 Summary of tumor incidence, tumor volume, and the onset of first neurological symptom of mice injected with CD133(+) and CD133(−) cells isolated from six GBM patients

DISCUSSION

A new paradigm in brain tumor biology that only CD133(+) cancer cells harbor a tumor-initiating potential raises an attractive hypothesis that GBMs can be cured if one eradicates CD133(+) CSCs, which occupy small portion of GMB cells. 3, 4 However, the story would not be simple, as GBMs have been known to possess various genetic alterations and to consist of heterogeneous cell populations,15, 16 arguing that CSCs in GBMs might also be heterogeneous and possess different genetic alterations. In support of above postulation, it has been recently documented that some GBM patients have CD133(−) CSCs and others have CD133(+).7 Even low-grade gliomas, which show more relatively homogeneous histology than GBMs, have been reported to possess multiple and spatially distinct clonal populations.14 Furthermore, in the present study, we demonstrated that there are CSCs in both CD133(+) and CD133(−) cell population originated from one GBM patient, and both CD133(−) and CD133(+) cells enabled the formation of tumor masses that showed similar volumes at 3 months post-injections.

Consistent with a previous report that GBM patients can be divided into groups that present different prognosis,10 we found that GBM patients can be classified into two groups (CD133-low and CD133-high group) according to their CD133(+) cell ratios, and CD133-low GBMs (CD133(+) cell ratio <3%) showed more aggressive morphologies as determined by MRI scan and unique gene expression patterns (‘mesenchymal’ or ‘proliferative’ subtype) that are related to worse prognosis.10, 11 It is interesting to note that it has been reported that gene expression profiles change from the ‘proneuronal’ to ‘mesenchymal’ subtype in GBMs recurred after treatment.10 As CD133(−) CSCs represented the ‘mesenchymal’ type, it might be plausible that CD133(−) CSCs might be more resistant to radiotherapy and chemotherapy compared to CD133(+) CSCs, which are already recognized to be markedly resistant to conventional anticancer therapies.17, 18

When GBM cells dissociated from GBM patients are cultured in the ‘NBE’ condition, which is originally designed to culture the normal NSCs, they harbor extensive similarities to normal NSCs and thereby recapitulate the genotype, gene expression patterns, and in vivo biology of GBMs.13 As GBM cells dissociated from CD133-highest GBMs (CD133(+) cell ratios=19.90 or 45.54%), which presented ‘proneuronal’ subtype of gene expression, were cultured in the ‘NBE’ condition for 10 days, their gene expression patterns were changed from ‘proneuronal’ subtype to ‘mesenchymal’ or ‘proliferative’ subtypes that were observed from CD133(−) CSCs, arguing that CD133(−) CSCs might have advantages for survival or proliferation in the ‘NBE’ condition.

Taken together, our data show that there are at least two kinds of CSCs (CD133(+) and CD133(−)) in each GBM patient and that they have different biological and clinical characteristics. An understanding of all GBM CSC types and their clinical implications, as well as development of new GBM CSC-specific markers, should be a crucial step toward more effective therapeutic modalities against human GBMs.