German Cancer Consortium (DKTK), Heidelberg, Germany
Clinical Cooperation Unit Neurooncology, German Cancer Research Center (DKFZ), Heidelberg, Germany
Department of Neurooncology, Neurology Clinic and National Center for Tumor Diseases, University of Heidelberg, Heidelberg, Germany
Address correspondence and reprint requests to Wolfgang Wick, Clinical Cooperation Unit Neurooncology, German Cancer Research Center (DKFZ), Im Neuenheimer Feld 280, D-69120 Heidelberg, Germany. E-mail: firstname.lastname@example.org
Human glioblastomas may be hierarchically organized. Within this hierarchy, glioblastoma-initiating cells have been proposed to be more resistant to radiochemotherapy and responsible for recurrence. Here, established stem cell markers and stem cell attributed characteristics such as self-renewal capacity and tumorigenicity have been profiled in primary glioblastoma cultures to predict radiosensitivity. Furthermore, the sensitivity to radiotherapy of different subpopulations within a single primary glioblastoma culture was analyzed by a flow cytometric approach using Nestin, SRY (sex-determining region Y)-box 2 (SOX2) and glial fibrillary acidic protein. The protein expression of Nestin and SOX2 as well as the mRNA levels of Musashi1, L1 cell adhesion molecule, CD133, Nestin, and pleiomorphic adenoma gene-like 2 inversely correlated with radioresistance in regard to the clonogenic potential. Only CD44 protein expression correlated positively with radioresistance. In terms of proliferation, Nestin protein expression and Musashi1, pleiomorphic adenoma gene-like 2, and CD133 mRNA levels are inversely correlated with radioresistance. Higher expression of stem cell markers does not correlate with resistance to radiochemotherapy in the cancer genome atlas glioblastoma collective. SOX2 expressing subpopulations exist within single primary glioblastoma cultures. These subpopulations predominantly form the proliferative pool of the primary cultures and are sensitive to irradiation. Thus, profiling of established stem cell markers revealed a surprising result. Except CD44, the tested stem cell markers showed an inverse correlation between expression and radioresistance.
Markers used to define glioma-initiating cells (GIC) are generally not defining a more resistant, but rather a more sensitive group of glioma cells. An exemption is CD44 expression. Also proliferation of the GIC culture itself was not systematically associated with radiosensitivity or – resistance, but a SOX-2 positive, proliferative subgroup within a GIC culture is showing the highest radiosensitivity.
For many years, glioblastoma, the most frequent and aggressive primary brain tumor in adults, has been regarded as a clonal malignancy and was modeled in vitro by established glioma cell lines kept in serum-containing medium (SCM). Evidence emerged that primary glioma cell cultures may better reflect the genetic and biological features of glioblastoma (Bjerkvig et al. 2005). At present, the search for glioblastoma subpopulations that are distinct from the main tumor and responsible for tumor initiation and progression as well as resistance to therapy is ongoing. Such cells are usually termed cancer stem cells (CSC) or glioblastoma-initiating cells (GIC), reflecting rather a concept than unequivocal evidence, and only a few examples exist of these cells being a therapeutic target (Zhu et al. 2014). GICs are defined by their properties to self-renew, to express stem cell markers, and most importantly to be highly tumorigenic and able to recapitulate a phenocopy of the tumor of origin in immunocompromised mice when transplanted orthotopically in low cell numbers.
We aimed at profiling different GIC cultures and different subgroups within single GIC cultures to evaluate how the various stem cell markers and techniques to detect stem cells, which are the product of intense research over the last few years (Singh et al. 2003; Bao et al. 2006, 2008; Suva et al. 2009; Fukaya et al. 2010; Rasper et al. 2010; Thon et al. 2010; Zheng et al. 2010), correlate with resistance to radiotherapy.
Furthermore, we tried to detect the stem cell subpopulation within single GIC cultures, to analyze whether they form the radioresistant pool of the GIC culture.
The analysis was done on different levels. The clonogenic potential of the different GIC cultures was estimated by the limiting dilution assay (LDA), proliferation was measured by bromodeoxyuridine (BrdU)- or 5-ethynyl-2 deoxyuridine (EdU)-uptake. Radioresistance was assessed by the relative decline in proliferation and clonogenic potential after irradiation with 4 Gy. These results were correlated with the protein expression of the stem cell markers CD133, CD15, CD44, Nestin, and SRY (sex-determining region Y)-box 2 (SOX2) measured by flow cytometry as well as the mRNA expression levels of various stem cell markers.
To confirm our data, the same markers were analyzed in a radiochemotherapy-treated cancer genome atlas (TCGA) collective of glioblastoma patients. Finally, with the help of flow cytometry, we evaluated whether stem cell subpopulations, behaving differentially from the non-stem cells, exist within a single GIC culture. Exemplarily, the in vitro data were verified in vivo by orthotopical xenotranplantation of glioma-initiating cells in nude mice.
Material and methods
Tumor samples were obtained from adult patients diagnosed with glioblastoma after informed consent. GIC cultures were established from freshly dissected tumor tissue with a success rate of 1/4 tumors. Tumor and neurosphere cultures were cultured as described (Hemmati et al. 2003). Cells were seeded in neural sphere cell medium (NSCM) containing Dulbecco's modified Eagle's medium:F12 medium enriched with B27 supplement, basic fibroblast growth factor (bFGF) (20 ng/mL), epidermal growth factor (EGF) (20 ng/mL), and leukemia inhibitory factor (20 ng/mL) (Hemmati et al. 2003). To propagate cells in culture they were split mechanically. For all the experiments requiring a single-cell suspension spheres were split with accutase (PromoCell, Heidelberg, Germanny). To obtain adherent, differentiated cells, tumor cells were cultured in SCM medium (10% fetal calf serum) on poly-l-lysine-coated tissue flasks for 15 days. Alternatively, differentiation was induced by withdrawal of EGF and bFGF, culture on a poly-l-lysine-coated surface and supplementation of recombinant bone morphogenetic protein 2 (10–50 ng/mL) or ciliary neurotrophic factor (CNTF) (50 ng/mL), (R&D systems, Minneapolis, MN, USA) with and without 5-Azacytidin (3 μM) (AXXORA, Lausen Switzerland) 3 days prior to fixation and subsequently analyzed by immunofluorescence microscopy and flow cytometry (Lee et al. 2008). Glioblastoma origin was confirmed by comparative genomic hybridization (Toedt et al. 2011) indicating typical genomic alterations on chromosomes 7q, 10p, 17q (Holland et al. 2010). A subset of the GIC cultures (WJ and PJ) was analyzed by Illumina Human Methylation 450 array performed from DNA extracted from cell lines. The array data were used to calculate a low-resolution copy number profile as previously described (Sturm et al. 2012).
Human astrocytes (ScienCell Research Laboratories, Carlsbad, CA, USA) were kept in astrocyte medium (ScienCell Research).
Quantitative Real-Time PCR (qRT-PCR)
Total RNA was extracted using a RNA purification system (Qiagen, Hilden, Germany) and treated with RNase-free DNase I to remove genomic DNA (Roche, Mannheim, Germany). cDNA was prepared from 5 μg of total RNA using the Superscript RNase H–Reverse Transcriptase (Invitrogen, Karlsruhe, Germany) and random hexamers (Sigma-Aldrich, Taufkirchen, Germany). For qRT-PCR, gene expression was measured in an ABI Prism 7000 Sequence Detection System (Applied Biosystems, Foster City, CA, USA) with SYBR Green Master Mix (Eurogentec, Cologne, Germany) and primers at optimized concentrations (Opitz et al. 2009). Primers (Sigma-Aldrich) were selected to span exon–exon junctions if possible. Standard curves were generated for each gene and the amplification was 90–100% efficient. Relative quantification of gene expression was determined by comparison of threshold values. All results were normalized to glyceraldehyde-3-phosphate dehydrogenase. The sequences for the housekeeping gene glyceraldehyde-3-phosphate dehydrogenase and the genes evaluated were as follows:
Clonogenic capacity and sensitivity toward radiotherapy
To assess the clonogenic capacity LDA was performed as described (Eirew et al. 2010). Shortly, cells were dissociated with accutase. Afterward, 24 wells of a 96-well microwell plate were each plated with 300, 50, 8, and 1 cells in 0.2 mL of NSCM. After 3 weeks, microwell plates were analyzed for wells showing clones and clonal frequency as well as the SEM was calculated with L-Calc free online software (STEMCELL Technologies, Cologne, Germany). Stem cell frequency was expressed by 1/the minimum amount of cells necessary to form a colony. To evaluate the radioresistance LDA was also performed after irradiation at 4 Gray (Gy).
Experiments to assess differential radiosensitivity were also performed within subfractions of GIC cultures kept in NSCM and analyzed by flow cytometry. Proliferation assessed by BrdU incorporation and cell cycle distribution (DNA content) of different GIC subfractions were measured 72 h after irradiation at 4 and 8 Gy.
Immunocytochemistry and flow cytometry
Immunocytochemistry of neurosphere cultures was performed as described (Geschwind et al. 2001). Neurospheres were fixed in ice-cold methanol or in 4% paraformaldehyde and immunostained with rabbit anti-Nestin (1 : 200; Chemicon, Temecula, CA, USA), mouse anti- TuJ1 (1 : 100; Chemicon), rabbit anti-glial fibrillary acidic protein (GFAP; 1 : 500; Chemicon) followed by Alexa fluorophore-conjugated secondary antibodies (1 : 100; Molecular Probes, Karlsruhe, Germany).
Furthermore, CD133 and CD15 expression were evaluated by flow cytometry with anti-CD133/1-Phycoerythrin (PE) or CD133/2-PE antibody and CD15-Viogreen (Miltenyi Biotec, Bergisch-Gladbach, Germany). CD44, SOX2, Nestin expression, and the amount of cycling cells were evaluated using flow cytometry after staining with anti-CD44-Alexa700, anti-human SOX2-PE, anti-human Nestin-647, and anti-Ki67-Alexa64 antibodies purchased from BD Biosciences (San Jose, CA, USA). Nuclei were counter-stained with 4,6-diamidino-2-phenylindole, and 7-aminoactinomycin.
To assess proliferation cells were incubated for 3–12 h with 10 μM BrdU or EdU at 37°C and 5% CO2. Afterward the cells were fixed and stained using the BrdU-Flow and EdU-Flow Kit from BD Biosciences.
Cells were analyzed in a BD-FACS Canto II flow cytometer, final data were processed with the help of FlowJo flow cytometry analysis software. To translate flow cytometric expression data for statistic evaluation, the mean fluorescence value, after staining with the respective isotype antibody, was log-transformed and subtracted from the log-transformed mean fluorescence value of the specific antibody. This transformation translates the shift seen when performing overlays in the graphically presented flow data. Values smaller than 0.12 corresponding to a specific fluorescence index, SFI, (Hueber et al. 2003) smaller than 1.3 were regarded as no specific staining for a marker, and therefore no marker expression.
Animal experiments and preparation of mouse brains for histology
All animal work was performed in accordance with the German animal protection law (Approving institution: Regierungspräsidium Karlsruhe). Tumorigenicity was determined by injecting 50 to 105 glioblastoma-derived neurosphere cells suspended in phosphate-buffered saline orthotopically into the right striatum of 6–12 weeks old athymic female mice (CD1 nu/nu, Charles River, Sulzfeld, Germany) by a stereotactic procedure. To reduce pain animals were anesthetized with xylazine and ketamine. Neurological symptoms were assessed daily. Symptomatic animals were rapidly sacrificed to prevent pain. The brains of the killed animals were recultured as described for the fresh human tumor specimens. Recultured cells were reimplanted at 50–1000 cells after new spheres had formed to assess whether cells could be serially transplanted. Cryostat transverse brain sections (8 μm) were stained with hematoxylin-eosin (H&E) or with anti-human Nestin antibody (Chemicon International, Billerica, USA) and analyzed by AxioVision software (Carl Zeiss, Jena, Germany). To evaluate the in vivo sensitivity toward irradiation, 2 × 105 GIC were orthotopically implanted in five animals per group. Seven days later cerebral irradiation (6 Gy) was performed in the experimental groups. For local irradiation, brains of nude mice were irradiated using electrons from a standard Linac radiation source. Positioning and shielding of the animals were achieved by a lead/plastic device that allows the exact application of the radiation with a 90% isodose to the targeted 7 × 7 mm brain section, sparing the throat of the mice (Tabatabai et al. 2006).
To assess the influence of mRNA expression of defined stem cell markers on survival in isocitrate dehydrogenase wild-type patients, methylation (Illumina HumanMethylation27 BeadChip, n = 294 samples and Illumina HumanMethylation450 BeadChip, n = 126 samples; Illumina, San Diego, CA, USA), mRNA expression (z-score transformed) and clinical data were obtained from the database of TCGA (http://cancergenome.nih.gov) and from cBioPortal (Cerami et al. 2012) at Jan 15 2013.
To detect the gliomas CpG island methylator phenotype (G-CIMP) (and exclude G-CIMP+ patients from further analysis), unsupervised hierarchical clustering of methylation data was performed as described previously (Wiestler et al. 2013). Briefly, probes: (i) targeting the X and Y chromosomes, (ii) containing a single nucleotide polymorphism within 5 base pairs of and including the CpG site and (iii) not mapping uniquely to the human reference genome (hg19), allowing for one mismatch, were removed. The 1500 (Illumina HumanMethylation27 BeadChip) and 8000 (Illumina HumanMethylation450 BeadChip) most variable (by SD) probes were kept and unsupervised hierarchical clustering was performed for each platform.
Patients (n = 134) included in this analysis were: (i) treatment-naive at the time of tissue extraction and (ii) initially treated with radio- and chemotherapy (temozolomide) as in the EORTC-26981-22981 trial (Stupp et al. 2005). Disease-free survival as reported by the TCGA was the primary endpoint.
The association of mRNA expression and disease-free survival was assessed through univariate Cox proportional hazards regression models. Univariable p values were adjusted for multiple testing using Benjamini–Hochberg correction in order to control the false discovery rate (Benjamini 1995). Analyses were carried out using R (Version 3.01) (R Development Core Team 2011, http://cran.r-mirror.de/.).
Statistical significance was assessed by Student's t-test (Excel, Microsoft, Seattle, WA, USA) at p < 0.05 (significant) or p < 0.01 (highly significant). All in vitro experiments reported here were performed at least twice. Flow cytometry data were repeated at least twice, except tumor Ma-1 which stopped growing in cell culture. For the assessment of a monotonic relationship between the expression (both on protein and mRNA level) of defined stem cell factors and proliferation and clonogenicity, Spearman's rank correlation test was used.
Sphere forming cultures were established from freshly dissected tumor tissue, which was cultivated after dissociation in neurosphere medium. All GICs formed neurospheres under this condition (Fig. 1a).
Exposure to SCM for 15 days or bone morphogenetic protein 2 in NSCM without growth factors for 3 days led to differentiation of the sphere cultures measured by up-regulation of GFAP and the neuronal marker βIII-tubulin as shown by immunofluorescence microscopy and flow cytometry for a better quantification (Fig. 1b–d, Figure S1). These differentiation approaches neither abrogated the expression of the neural stem cell marker Nestin nor the progenitor and stem cell marker SOX2 as observed before (Gursel et al. 2011).
Cells cultured in neurosphere medium formed orthotopic brain tumors at low cell numbers. As few as 50 cells/mouse brain were enough to form deeply infiltrating tumors mimicking the growth pattern of human glioblastomas (Fig. 1e and Figure S1d+e). The time of tumor formation varied from ~ 90 to more than 200 days in the different GIC cultures after implantation of thousand cells (Figure S1e). Interestingly, differentiation in SCM for 15 days did not abrogate tumorigenicity (Figure S1d+e).
Finally, serial repassaging after explantation of the tumors, which had formed in CD1 nu/nu mice, again resulted in gliomas. This detailed analysis was performed for the GIC cultures shown in Table 1. The data presented and the data published before (Lemke et al. 2012) provide evidence that the cells cultured exhibit stem cell characteristics.
Table 1. Comparison of GIC cultures T269, T1, T325, and Ma-1 in regard to established stem cell marker expression and functional stem cell characteristics
Tumorigenicity 1000 cells
Tumor formation: aggressiveness (days until symptomatic)
In vivo passaging
% of Nestin positive cells
% of Sox2 positive cells
Longterm cell culture
Comparison of tumorigenicity after orthotopic implantation of 103 cells, aggressiveness of tumor growth (assessed by the interval, in which the animals got symptomatic), infiltration and in vivo passaging capacity of tumor cells, multilineage differentiation and long-term cell culture, describing the capacity of a sphere culture to be passaged for more than ten times. The expression of SOX2, Nestin, and CD133 was analyzed by flow cytometry. Tumor Ma-1, which grew very aggressively in vivo stopped proliferating in vitro. Hence, Ma-1 did not fulfil all cancer stem cell criteria (n.p. = not possible). (Stronger infiltration with tumor cells reaching the contralateral hemisphere was marked with ++).
102 ± 8.5
225 ± 31.7
87 ± 4.7
233 ± 7.5
Comparative genomic hybridization (CGH) analysis was performed for T1, T325, T269, WJ, PJ, ZH161, ZH305, KNG002, and S24 and demonstrated the glioblastoma origin of the GIC cultures examined (exemplarily in Figure S2–S4).
Stem cell marker profiling helped to predict the radiosensitivity of GIC cultures in the limiting dilution assay
As a high clonogenic potential by itself is supposed to be a characteristic stem cell feature (Bjerkvig et al. 2005), we performed LDA of 10 different GIC cultures without and after irradiation at 4 Gy. The clonogenic potential expressed by one divided by the amount of cells necessary to form at least one new colony varied from 0.0115 in tumor MM to 0.88 in S24. In other words, between different GIC cultures 1.1 (S24) to 86 (MM) cells were necessary to form one new colony (Fig. 2a). We than evaluated the radioresistance of the different GIC cultures by dividing the amount of cells necessary to form at least one colony after irradiation with 4 Gy by the number of cells necessary without irradiation. In tumor KNG002, it took nearly 80x more cells after irradiation in the LDA to form at least one new colony while in tumor T325 only 1.5x the amount of cells were necessary (Fig. 2a). Tumors with a stronger reduction in the clonogenic potential after irradiation were regarded as more radiosensitive compared with the tumors where the difference was smaller. Interestingly, the clonogenic potential without therapy did not predict sensitivity to irradiation at 4 Gy.
In the next step, different stem cell markers were profiled in the GIC cultures to examine whether the stem cell marker expression correlates with radioresistance in the LDA. Flow cytometric expression analysis was performed on the stem cell markers SOX2, Nestin, CD133, CD15, and CD44 in the ten GIC cultures evaluated in the LDA before. With the help of Spearman's rank correlation test, we could show that SOX2 and Nestin expression inversely correlate while CD44 expression positively correlates with radioresistance in the LDA (p ≥ 0.05) (Fig. 2b and c).
To gain a broader look on more stem cell markers and assess whether qPCR analysis suffices to predict radioresistance, we further analyzed the mRNA levels of CD133, Nestin, Aldehyde dehydrogenase 1, ATP-binding cassette, sub-family G member 2, L1CAM (L1 cell adhesion molecule), Musashi1, enhancer of zeste homolog 2, and PLAGL2 (Pleiomorphic adenoma gene-like 2) in 9 of the GIC cultures examined in the LDA before. Statistical analysis revealed that CD133, Nestin, Musashi1, PLAGL2, and L1CAM inversely correlated with radioresistance in the LDA.
Stem cell marker profiling can help to predict the radiosensitivity of GIC cultures proliferation
We next evaluated the proliferation rate without and with 4 Gy irradiation of GIC cultures to examine another parameter, which might reflect radioresistance in vitro. Therefore, GIC cultures were treated for 6 hours with EdU and incorporation was measured by flow cytometry (Fig. 3a). During this time, tumor MM showed the weakest EdU uptake in only ~ 13% of the population, while tumor PJ had the strongest proliferation rate with in EdU uptake in ~ 41% of the cells. After irradiation we could observe a general decline of the proliferation rate. Tumor 325 which was regarded as a very radioresistant tumor showed a relative proliferation of ~ 94% after irradiation, while radiosensitive tumors such as T1 and KNG002 went down to ~19% of their proliferation rate without irradiation.
We did not observe a correlation between proliferation of the unsorted GICs and increased susceptibility to radiotherapy (Fig. 3a). Aiming again to examine whether radioresistance in regard to the proliferation rate was predictable by stem cell marker profiling, we correlated the relative proliferation rate at 4 Gy with the expression of the aforementioned stem cell markers in flow cytometry and qPCR. On the protein level, Nestin inversely correlated with radioresistance (Fig. 3b), on the mRNA level, Musashi1, PLAGL2, and CD133 expression showed an inverse correlation with radioresistance (Fig. 3c).
In vitro radiosensitivity correlates with response to radiotherapy in a xenograft model
Mounting evidence suggests that tumor stem cells occupy a special perivascular and/or perinecrotic niche. This niche is supposed to be necessary for GIC to maintain their stem cell phenotype (Hambardzumyan et al. 2008b; Ricci-Vitiani et al. 2010; Seidel et al. 2010; Seoane 2010). To strengthen the in vitro data on radiosensitivity, we performed an in vivo experiment allowing the cells to grow in an orthotopic environment. Exemplarily, T269 which belonged to the group of the radiosensitive tumors in the LDA and proliferation assay and the radioresistant T325 were implanted each at 2 × 105 cells in five CD1 nu/nu mice per group. T269-bearing mice became symptomatic 71 days after implantation. In contrast, animals orthotopically implanted with T325 cells became symptomatic around day 192. Interestingly, cranial irradiation with 6 Gy prolonged survival for 36 days in the T269 model, whereas the survival difference between irradiated and untreated animals implanted with T325 cells was not significant (Fig. 4). We concluded that the in vitro data were predictive for the in vivo response to irradiation.
Stem cell marker expression does not correlate with radiochemoresistance in the TCGA data base
To translate our data to a larger cohort which better reflects the high genetic heterogeneity of glioblastoma and its four defined subclasses (Verhaak et al. 2010) we reevaluated our results in the TCGA glioblastoma patients collective. The influence of mRNA expression of the defined stem cell markers on disease-free survival of radiochemotherapy-treated patients with isocitrate dehydrogenase wild-type glioblastomas was assessed. Methylation, mRNA expression, and clinical data were obtained from the database of TCGA (http://cancergenome.nih.gov) and from cBioPortal. G-CIMP positive patients were excluded from further analysis to have a more homogeneous group of primary glioblastomas. Patients (n = 134) included in this analysis were treatment-naive at the time of tissue extraction and initially treated with radio- and chemotherapy as in the EORTC 26981/22981 trial (Stupp et al. 2005). Disease-free survival as reported by the TCGA was the primary endpoint. This analysis could not demonstrate a correlation between the examined stem cell markers and survival of the patients (Table 2).
Table 2. Stem cell marker expression does not correlate with radiochemo-resistance in a TCGA collective
Stem cell marker mRNA expression of a TCGA collective of radiochemotherapy-treated glioblastoma Patients (n = 134) was correlated with the risk (hazard ratio, HR) to have a shorter disease-free survival. No marker shows a significant influence on disease-free survival after treatment (CI, Confidence interval; p, level of significance).
Evaluation of GIC culture subpopulation based on intracellular marker expression
So far, the experiments performed demonstrated that higher expression of stem cell markers correlated inversely with radioresistance, with the exception of CD44. Since glioblastomas are highly heterogeneous tumors and a lot more factors interfere with radiotherapy such as p53-status and expression of checkpoint kinases, we were interested to see whether there are subpopulations within a single GIC culture showing a differential response to radiotherapy. Stem cells are supposed to divide asymmetrically which should allow identifying different subpopulations within GIC cultures (Chen et al. 2010; Lathia et al. 2011). Hence, GIC cultures might be composed of more and less differentiated cells that might be masked by mere expression analysis of the whole population. As CD133, CD44, and CD15 were only detected in subset of the GIC cultures, we focused on intracellular stem cell markers.
Therefore, we looked at the expression of Nestin, a marker for neural stem and progenitor cells (Singh et al. 2004; Strojnik et al. 2007), which correlated inversely with radioresistance in the LDA and proliferation assay. The GIC cultures tested showed a monophasic Gaussian distribution of Nestin expression with a homogeneous BrdU-uptake leading us to the conclusion that cells with a higher Nestin expression did not proliferate better than low Nestin expressing cells. Furthermore, all the GICs tested were nearly 100% positive for Nestin. Hence, Nestin did not seem suitable for distinguishing subfractions within individual GIC cultures (Figure S5). Likewise, the more differentiated astrocytic marker GFAP did not allow to differentiate subfractions within GIC cultures in vitro (data not shown).
Finally, we examined the expression of the transcription factor SOX2, a progenitor, and stem cell marker, which demonstrated to be central for the tumorigenicity of GIC cultures. Studies in mice have implicated SOX2 as one of three factors in regulating pluripotency in embryonic stem cells (Fong et al. 2008; Gangemi et al. 2009; Ikushima et al. 2009). In contrast to non-neoplastic human astrocytes, SOX2-mRNA was expressed in GIC cultures (Fig. 5a) and could be detected in glioblastoma tissue sections (Fig. 5b). Interestingly, on the protein level GIC cultures tested partly formed subfraction with a SOX2 positive and a SOX2 negative fraction. T269 was nearly 100% positive for SOX2 expression, while the SOX2positive fraction was the smallest in T325 (Fig. 5c).
The SOX2 subpopulations mainly form the proliferating pool of GIC cultures and are sensitive to irradiation
Combining BrdU incorporation analysis and SOX2-staining in flow cytometry revealed that SOX2positive -cells better proliferated than SOX2negative-cells. T269, for example, demonstrated a maximal proliferating fraction of 29.5% of cells entering the S-phase within 6 hours. Of these 29.5% proliferating cells, 28.8% were SOX2positive and 0.7% were SOX2negative (Fig. 5c; left). The SOX2positive fraction formed 95.6% of all cells in this tumor. T325 was composed of ~ 40% SOX2positive-cells. Again, the proliferating cells, 2.49% in this example, were in ~98% SOX2 positive (Fig. 5c; right).
To further characterize the SOX2positive subpopulation, SOX2 and Ki67 were co-stained to assess the cycling cells. Here, ~ 64% of T325 expressed SOX2. About 10.6% of the SOX2positive cells were cycling detected by a positive Ki67 staining. Only 0.2% of the 36% SOX2negative cells showed a weak positivity for Ki67 (Fig. 5d). We concluded that the cycling cells can be almost exclusively found in the SOX2positive fraction.
To finally assess whether SOX2positive cells are sensitive to irradiation, we irradiated five different GIC cultures with 4 and 8 Gy and analyzed the proliferation capacity 72 h after treatment. All 5 GICs treated showed a dose-dependent reduction in the proliferation rate after irradiation (Fig. 6). T269 e.g., was composed of a nearly 100% SOX2positive fraction, of which ~ 26% took up BrdU within 6 h. After irradiation at 8 Gy, T269 was still composed of nearly 100% Sox2positive cells but only ~ 6% were still proliferating.
The SOX2positive cells which formed predominantly the proliferating pool of the 5 GIC cultures were sensitive to irradiation in terms of their proliferation capacity (Fig. 6). Irradiation did not reduce the amount of SOX2-positive cells 72 h after irradiation. SOX2negative cells showed also a reduced proliferation rate after irradiation but they formed only a minority of the proliferating fraction.
The aim of this study was to examine whether markers used to identify CSC are of value to predict radiosensitivity in primary glioma cultures obtained from patient biopsies. We were interested to find out whether profiling of stem cell characteristics and markers of different GIC cultures or profiling of stem cell markers within a single GIC culture would help predict radiosensitivity of different GIC cultures or a stem cell subgroup within a single GIC culture. So far, CSC are regarded highly relevant for recurrence after treatment in glioblastoma patients (Bao et al. 2006; Beier et al. 2008; Hambardzumyan et al. 2008a,b).
First, we cultivated freshly dissected tumor tissue in neurosphere cell medium until sphere formation was observed. We performed several experiments demonstrating that the cultures fulfilled stem cell criteria and were of glioma origin. The GIC cultures (characterized in detail in Table 1) formed tumors in low cell numbers, showed a multilineage differentiation capacity, were propagated in culture for many passages (except tumor Ma-1), and demonstrated a high clonogenic potential. Surprisingly, differentiation in serum-containing medium without growth factor supplementation did not abrogate the expression of the stem cell markers Nestin and SOX2 (Fig. 1, Figure S1) nor the tumorigenic potential in several tumors (Figure S1). We did not systematically examine whether the growth pattern, like the capacity to infiltrate, changed after the differentiation approach. The fact that tumors still formed argued against a terminal differentiation by the strategies applied. Similarly, Jiang et al. demonstrated in an experimental glioma model in neonatal Gtv-a Arf(−/−) mice – induced by platelet-derived growth factor-B (Jiang et al. 2011), a lingering tumorigenicity of their sphere cultures after induction of differentiation with fetal bovine serum (FBS). According to the authors, this was because of the high plasticity of their GIC cultures.
We next characterized the clonogenic potential in ten GIC cultures with the LDA (Fig. 2). A high clonogenic potential of the unsorted GICs was not associated with a higher radioresistance in the LDA. We concluded that clonogenicity which by itself is related to stemness (Bjerkvig et al. 2005), does not permit the prediction of radioresistance.
However, the relative reduction in the clonogenicity may be used as a parameter for radioresistance, allowing comparison between the different GIC cultures examined. As a second parameter to evaluate radioresistance, we determined the relative decline of the proliferation rate 72 h after irradiation with 4 Gy measured bei EdU incorporation.
Correlating radioresistance in the LDA and proliferation assay with the protein expression of the stem cell markers CD133, CD15, CD44, Nestin, and SOX2 as well as the mRNA levels of Nestin, Musashi1, L1Cam, ATP-binding cassette, sub-family G member 2, Aldehyde dehydrogenase 1, CD133, PLAGL2, and enhancer of zeste homolog 2 revealed an unexpected result. Only CD44 protein expression correlated with radioresistance in the LDA, while Nestin and SOX2 protein expression as well as Nestin, L1CAM, Musashi1, PLAGL2, and CD133 mRNA levels correlated inversely with radioresistance in the LDA. The proliferation analysis showed a negative correlation for radioresistance and Nestin protein expression as well as Musashi1, PLAGL2, and CD133 mRNA levels. The absolute proliferation rate of the different GIC cultures without irradiation was not predictive for radiotherapy sensitivity. We concluded that there is no trivial association between irradiation effects and proliferation.
These unexpected results are in line with the data published before by Beier et al. (Beier et al. 2008), who examined chemotherapy sensitivity in CD133 positive and negative CSC cultures. CD133 was originally suggested to be a key marker for GIC growing as spheroids (Singh et al. 2003, 2004; Lottaz et al. 2010). It is expressed in neural stem cells and has been attributed a role in the development of the central nervous system (Uchida et al. 2000). After the initial paper by Singh et al. (2003), several studies have identified stem cells by their expression of CD133. CD133 and related CD133positive cells were associated with a phenotype of resistance toward radiotherapy (Bao et al. 2006; Rich 2007). Anyway, in the publication from Beier et al., CD133 negative cells were associated with chemoresistance. This is in line with a growing number of publications showing glioma initiation by CD133negative cells that even gave rise to CD133positive cells (Lee et al. 2006b; Wang et al. 2008; Chen et al. 2010) calling the importance of CD133 as a stem cell marker in question.
Our in vitro results, which allow the prediction of radiotherapy sensitivity, were strengthened by the in vivo experiment. Tumor T269, in vitro classified as a more radiosensitive tumor than T325, proved to be also more radiosensitive in orthotopic xenotranplanted nude mice treated at 6 Gy (Fig. 4).
Hence, the microenvironment, which is provided in the mouse, did not alter the differential in vitro radiosensitivity of the GIC lines. Considering that GICs are supposed to differentiate into endothelial cells and vessel-like structures (Ricci-Vitiani et al. 2010), one should assume that GICs are capable, to some extent, of creating their niche in the mouse brain. Therefore, we concluded that radiosensitivity measured in vitro by LDA and proliferation analysis mirrors the sensitivity of our GIC lines realistically and is more than an in vitro artifact. Although, it is unclear whether cells implanted 7 days before irradiation, as scheduled in the experiment, are capable of establishing a niche of their own for the stem cells. At least, the in vivo experiment, covering a time frame of more than 200 days, underscored the fact that the irradiation effects measured after 72 h in vitro reflect more than a casual snapshot.
To argue against the small number of GIC cultures tested (n = 10), we tried to translate our concept into a clinical context by correlating the mRNA level of accepted stem cell markers in a TCGA glioblastoma collective with their sensitivity to radiochemotherapy measured by disease-free survival. Here, no stem cell marker tested predicted resistance or sensitivity to therapy (Table 2). There are several possible explanations for this lack of translation. The data in the TCGA are generated by tumor and bystander cells in an unknown ratio, and in vitro culturing of GICs by itself is prone to amplifying subtypes of glioblastomas (Laks et al. 2009). Thus, our experimental findings were not strengthened by this analysis. Yet, the widely accepted stem cell concept arguing that stem cell marker defined cells are more resistant, and therefore the basis for recurrent disease was also not supported. An example of a more robust concept for bona fide stem cells might be given by stem cells driven by the nuclear receptor tailless (Tlx) (Zhu et al. 2014).
Finally, as glioblastomas are supposed to be hierarchically organized and stem cells divide asymmetrically, we established a multiparameter flow cytometric approach to examine cell cycle distribution and proliferation rate in different subfractions within a single GIC culture. We focussed on the intracellular markers Nestin and SOX2 trying to detect more radioresistant CSC subfractions. Nestin was not suitable to detect subpopulation in vitro within different GIC cultures since its expression was distributed evenly and all the GIC cultures tested were nearly 100% positive for Nestin (Figure S5). SOX2, on the other hand, differentiated subfractions within the GIC cultures tested. The amount of SOX2positive cells varied from more than 95% to ~ 17% (Fig. 6). SOX2positive cells almost exclusively expressed Ki67, which is a marker for cycling cells (Fig. 5d). But most importantly, the SOX2positive proliferating subpopulation was sensitive to irradiation (Fig. 6). This was finally not surprising as we could demonstrate that the SOX2positive cells formed mainly the proliferating fraction of the cells. Irradiation did not hamper SOX2 expression itself within the first 72 h after irradiation, but clearly reduced the amount of proliferating cells in five different GIC cultures within this subfraction.
These results explain why silencing of SOX2 attenuates tumorigenicity (Gangemi et al. 2009) and is in line with the assumption that SOX2 cooperates with cylin D1 in cell cycle progression (Oppel et al. 2011).
The only positive correlation between stem cell marker expression and radioresistance was detected for CD44 which is in line with published data showing that CD44 promotes tumor cell resistance to reactive oxygen species-induced and cytotoxic agent-induced stress by attenuating activation of the Hippo signaling pathway (Xu et al. 2010).
To conclude, within the last decade, the CSC concept has gained a lot of attention in the glioma field. Many markers and techniques have been published to detect the CSC within glioma samples. We tested some of these markers and techniques to prove whether they are capable of predicting resistance to radiotherapy, a feature that is attributed to CSC. Our profiling strategies revealed that the majority of the significantly correlating stem cell markers tested show an inverse correlation with radioresistance, except CD44. Kim et al. came to a similar conclusion when profiling samples of different radiochemotherapy-treated glioblastoma patients for CD133, CD15, and Nestin expression, which did not correlate with a better survival rate (Kim et al. 2011). Furthermore, SOX2 emerged as a marker useful for the detection of subfractions within single GIC cultures, predominantly comprising cycling and radiosensitive cells.
It is likely that CSC are not sufficiently characterized by single markers. It is also likely that surface markers in solid tumors, analyzed after dissociation of the tumor, do not represent a biology-related phenotype, but are prone to change by the manipulation, may shift over time or in response to the microenvironment. Considering also that we have achieved in only around 30–50% of all the patient-derived tissue to establish a GIC culture, it might well be that the in vitro data are simply not appropriate to mirror the in vivo conditions. But taking all these limitations into account, and being aware of the fact that tumor stem cells derived from glioblastomas cultured in bFGF and EGF are probably the best in vitro model we have (Lee et al. 2006a), these data call into question the widely accepted theory that stem cells are the source of therapy resistance. At least the markers published to detect stem cells should not be used without skepticism. Finally, our data point out the necessity of further evaluating the role and therapeutic options of CD44 in radioresistance of glioblastomas.
Acknowledgments and conflict of interest disclosure
This work was supported within the Brain Tumor Network BTNplus (FKZ 01GS0883) of the National Genome Research Network (NGFNplus) (Federal Ministry of Education and Research, BMBF) (MW, WW, BR), NCCR Neuro (MW), and the Charitable Hertie Foundation (WW). We also like to thank for the support by the Microscopy Core Facility of the German Cancer Research Center Heidelberg. We thank Marcos Tatagiba, Clinic for Neurosurgery, University of Tuebingen and Nikolaj Hopf, Department of Neurosurgery, Katharinenhospital Stuttgart for the provision of freshly dissected human tumor tissue and Peter Lichter, German Cancer Research Center (DKFZ) for contributing to primary culture validation.
All experiments were conducted in compliance with the ARRIVE guidelines. The authors declare that they have no conflicts of interest.