Temozolomide Preferentially Depletes Cancer Stem Cells in Glioblastoma

Introduction

Human glioblastoma (GBM) is a highly aggressive brain tumor with a median survival of 14.6 months despite aggressive multimodal therapy ( 1). GBM are among the first solid cancers in which tumor cells with stem cell-like features, i.e., so-called cancer stem cells (CSC), were identified. They constitute a rare subpopulation of tumor cells that later differentiate into progenitor-like tumor cells or differentiated tumor cells ( 2). For several different brain tumors, including a subgroup of primary GBM, CD133⁺ but not CD133⁻ cells were found to be able to reconstitute the initial tumor in vivo when as little as 100 cells were injected into nude mice ( 3– 5). In a therapeutic setting, the proportion of CD133⁺ cells (normally 5–30%) was enriched after irradiation in vivo and in vitro, resulting in an increased tumorigenicity of the remaining cells ( 4). However, CD133 does not seem to be essential for the stem cell–like properties of CSC ( 3, 6, 7): We have recently described a novel subgroup of primary GBM that contain no CD133⁺ tumor cells. Instead, we have characterized CD133⁻ CSC, which resemble CD133⁺ CSC in terms of differentiation profile, tumorigenicity, and clonogenicity but show distinct molecular and cellular properties ( 3).

Temozolomide is the most commonly used chemotherapeutic agent in the therapy of GBM and is usually well-tolerated ( 1, 8). It achieves its cytotoxic effect mainly by methylating the O⁶ position of guanine. This adduct can be removed by the DNA repair protein O⁶-methylguanine-DNA- methyltransferase (MGMT), which is expressed in a subgroup of GBM. Consequently, temozolomide displays its highest efficacy against tumors lacking MGMT expression due to a methylated MGMT promoter ( 9, 10). However, although temozolomide significantly increases the proportion of patients surviving for >2 years, long-term survivors are still rare ( 1). The biological properties of CSC provide a possible explanation for the failure of chemotherapy in the long term: CSC from several tumor entities overexpress multidrug resistance proteins ( 11, 12), which may protect them against cytotoxic drugs that kill progenitor and differentiated cells ( 2). CSC would then give rise to recurrent tumors. However, this model has not yet been confirmed experimentally for GBM.

We have now investigated the effect of temozolomide on CD133⁺ and CD133⁻ CSC lines using an established in vitro model of GBM stem cell lines ( 5, 13, 14). Surprisingly, we found that temozolomide preferentially eliminates CSC but spares more differentiated tumor cells.

Materials and Methods

GBM samples. Native tumor tissue samples from GBM resections were provided by the local department of Neurosurgery. All tumors were histologically classified according to the WHO classification of central nervous system tumors ( 15). Experiments were approved by the local ethics committee (University of Regensburg, No. 05/105, June 14, 2005).

Culture of primary GBM cells and spheres. Samples were stored in sterile saline at 4°C and processed within 24 h after resection. They were washed and dissociated by mechanical and enzymatic means (1% trypsine, which does not alter CD133 cell surface expression). Erythrocytes were lysed using NH₄Cl. Trypan blue staining confirmed >80% viability after the procedure. Tumor cells were then grown in stem cell–permissive DMEM-F12 medium supplemented with 20 ng/mL of each human recombinant epidermal growth factor, human recombinant basic fibroblast growth factor (both from R and D Systems), and human leukemia inhibitory factor (Chemicon), and 2% B27 (Life Technologies). These culture conditions enable tumor cells to retain the molecular characteristics of the primary tumor with only minor changes in differentiation, expression pattern, and genetic mutation profile ( 3, 14). If not otherwise stated, 50% of the medium was substituted twice weekly. Temozolomide (Schering Plough; stock solution 150 mmol/L in DMSO) and 6-benzylguanine (6-BG; Sigma-Aldrich; stock solution 0.1 mmol/L in methanol) were diluted with stem cell-permissive medium. Thus, cells were treated with stem cell–permissive medium containing either temozolomide (5, 50, and 500 μmol/L) or the DMSO control (0.3%) over the periods indicated. Metabolic activity was measured using 10% AlamarBlue (Biosource) that was added to the cell culture medium. The absorbance was determined after 16 h at 590 and 540 nm. The lactate dehydrogenase (LDH) assay (Promega; Cytotox96R) was performed according to the instruction manual.

Flow cytometry. For Carboxy-Fluorescein diacetate Succinimidyl Ester (CFSE) staining, cells were labeled for 10 min with 20 μmol/L CFSE (Sigma Aldrich) and counterstained with CD133/2-PE (1:100; Miltenyi Biotech) or isotype control antibody. To label dead cells, 0.1 μg/mL 7-Amino-Actinomycin D (7-AAD; Sigma Aldrich) was applied for 10 min. To analyze cell cycle and BrdUrd incorporation, cells were incubated with 10 μmol/L BrdUrd for 12 h. Cells were fixed with 0.5% p-formaldehyde, incubated with 3 N HCl supplemented with 0.5% Tween, and double stained with anti-BrdUrd (1:25; Roche; secondary antibody: anti-mouse Cy5, 1:50) and 75 μmol/L propidium iodide (Sigma Aldrich). Flow cytometric data were analyzed using WinMDI 2.8 ⁷ and results were confirmed using ModFit LT 3.1 (Verity).

Magnetic and fluorescence activated cell separation. Cells were dissociated and resuspended in PBS containing 0.5% bovine serum albumin and 2 mmol/L EDTA. For magnetic labeling, CD133/1 Micro Beads were used (Miltenyi Biotech). Positive magnetic cell separation (MACS) was performed using several MACS columns in series.

For fluorescence-activated cell separation (FACS), cells were stained with CD133/2-PE (1:10; Miltenyi Biotech) or isotype control antibody (1:10, mIgG1κ-PE; Caltag Laboratories) and sorted on a BD FACSAria (Becton Dickinson).

Immunocytochemistry. A sphere derived from one tumor cell was dissociated, treated with either temozolomide or DMSO (control) over the indicated period, and stained as previously described ( 3). Primary antibodies used were anti–β-tubulin III (1:2,000; Promega), anti–glial fibrillary acidic protein (1:1,000; anti-GFAP; DAKO), anti-galactocerebroside C (1:500), and anti-nestin (1:500; both from Chemicon). Stainings were visualized using anti-mouse Alexa Fluor 488 (1:100; Molecular Probes) or anti-rabbit rhodamine antibody conjugates (1:100; Dianova). Nuclei were counterstained using 4′,6-diamidino-2-phenylindole (1:10,000; Sigma-Aldrich).

In vivo tumor model. Tumor cells were suspended in 2 μL PBS and intracranially injected into T-lymphocyte–deficient NMRI^(nu/nu) mice as described previously ( 3, 16). All procedures were conducted in accordance with German laws governing animal care (Az. 621-2531.1-04/03). Tumor samples were snap frozen. Ten-micrometer sections were stained with H&E using standard protocols.

Immunohistochemistry. Fresh frozen sections (10 μm) of mouse brain were fixed with acetone. Endogenous peroxidase activity was blocked by immersing the slide in 3% H₂O₂ solution for 10 min at room temperature. Nonspecific binding of the primary antibody was blocked by incubating the slide with 0.2% fish skin buffer for 30 min at room temperature. Sections were stained using anti–ki67 antibody (1:50; Novocastra) or isotype control antibody for 1 h at room temperature. Staining was visualized using biotinylated secondary antibody and diaminobenzidine (ABC-kit Vectastain Standard and DAB kit, both from Vector Laboratories).

Magnetic resonance imaging. Three months after injection of tumor cells, mice were anesthetized as previously described ( 17) and tumor-like lesions were detected using a clinical magnetic resonance scanner at a field strength of 3.0 Tesla (Siemens Magnetom Allegra; Siemens Medical Solutions). The magnetic resonance imaging (MRI) of the mouse brain was carried out using a dedicated four-channel rat head phased-array coil assembly (Rapid Biomed). Initial scout images of the brain in three planes were acquired in 14 s using the localizer sequence with TR/TE values of 50/12.4 ms, respectively. The images were acquired using a TR/TE value of 4,000/73 ms using 4 averages and 1 concatenation with an acquisition time of 5 min and 8 s. The acquired images had an in-plane resolution of 200 μm with a slice thickness of 0.7 mm. Tumor volume was determined using DatInf Measure 2.0.08. ⁸

MGMT promoter methylation–specific PCR. The MGMT promoter methylation status was determined in CD133⁺ and CD133⁻ glioma CSC lines by using methylation-specific PCR as described elsewhere ( 18). A total of 200 ng of genomic DNA was treated with sodium bisulfite using the EZ DNA Methylation-Gold kit (HISS Diagnostics). The primer sequences used to detect methylated MGMT promoter sequences were 5′-GTTTTTAGAACGTTTTGCGTTTCGAC-3′ and 5′-CACCGTCCCGAAAAAAAACTCCG-3′. This primer combination allows for the amplification of a 122-bp fragment from methylated DNA. The primer sequences used to detect unmethylated MGMT promoter sequences were 5′-TGTGTTTTTAGAATGTTTTGTGTTTTGAT-3′ and 5′-CTACCACCATCCCAAAAAAAAACTCCA-3′. This primer combination allows for the amplification of a 129-bp fragment from unmethylated DNA. Each PCR product was separated on 2% agarose gels. As positive control sample, we used genomic DNA from the A172 glioma cell line (obtained from American Type Culture Collection), which carries a completely methylated MGMT promoter. Genomic DNA extracted from peripheral blood leukocytes served as unmethylated control sample. In addition, a control reaction without any template DNA was performed together with each PCR experiment.

Quantitative real-time PCR. RNA was extracted with the RNeasy kit (QIAGEN GmbH) and cDNA was synthesized using the Reverse Transcription System (Promega GmbH). Real-time PCR was performed with the QuantiTect SYBR Green PCR kit (QIAGEN) using the Stratagene Mx3000P Real-Time PCR system (Stratagene Europe). The following primer pairs used were as follows: MGMT, 5′-GCCGGCTCTTCACCATCCCG-3′ (forward) and 5′-GCTGCAGACCACTCTGTG GCACG-3′ (reverse); 18s-RNA, 5′-CGGCTACCACATCCAAGGAA-3′ (forward) and 5′-GCTGGAATTACCGCGGCT-3′ (reverse). PCR reactions were run as triplicates on 96-well plates. Temperature profile is as follows: 95°C (15 min) followed by 45 cycles with 95°C (20 s) to 58°C (30 s). The quality of the products was controlled by a melting curve. For quantification, standard curves were established by amplification of serial dilutions of cDNA for both, the target gene and the endogenous reference (18s-RNA).

Western blot analysis. For Western blot analysis of MGMT protein expression, 40 μg of total cell lysate were separated by 12% SDS-PAGE and transferred onto nitrocellulose membranes by electroblotting. The membranes were blocked in 5% milk powder in 0.02% Tween PBS (TPBS) for 30 min and probed with antibodies against MGMT (1:1,000; a kind gift from Prof. B. Kaina, University of Mainz, Mainz, Germany) overnight at 4°C or β-actin (1:4,000; Abcam) for 1 h at room temperature. After 1 h of incubation with horseradish peroxidase–coupled secondary goat anti-rabbit or rabbit anti-mouse antibodies (Pierce) diluted 1:25,000 in TPBS, specific antibody binding was detected by using the enhanced chemiluminescence technique and the Fuji LAS 3000 imaging system. Recombinant human MGMT (Biomol) and total cell lysate from Jurkat cells were used as positive controls for the Western blot experiments.

Statistics. Data analysis was performed using JMP 5.1. The statistical tests used are indicated in the Figure Legends. All data are given as mean and SE. All experiments were performed with the CD133⁺ cell lines R11, R18, R28, R44, and the CD133⁻ cell line R8.

Results

We investigated four CD133⁺ (R11, R18, R28, and R44) and one CD133⁻ (R8) CSC lines derived from primary astrocytic GBMs ( 3). In these cell lines, only a small subgroup of tumor cells (2–4%) fulfilled all stem cell criteria ( 3), whereas the proportion of CD133⁺ cells ranged from 0% to 40%. The cells were cultured under medium conditions that recently proved to be effective to preserve molecular tumor characteristics compared with the primary tumor with only minor changes in differentiation, expression pattern, and genetic mutation profile even after long term culture ( 3, 14).

Temozolomide does not induce cell death but efficiently inhibits proliferation in CD133⁺ CSC lines. Similar to “classic” FCS-cultured glioma cell lines ( 12, 19– 21), all CD133⁺ CSC lines tested were resistant to temozolomide-induced cell death at maximum concentrations reached in the spinal fluid (5 μmol/L temozolomide; ref. 22) and in the plasma (50 μmol/L temozolomide; refs. 22, 23). At higher concentrations, treatment with temozolomide for 7 days led to a reduction in the number of metabolically active, i.e., viable cells, compared with solvent-treated cells ( Fig. 1A, left ). To differentiate between an actual decline in cell numbers, i.e., cell death, and impaired proliferation, we compared the metabolic activity of control-treated cells after 2 days with temozolomide-treated cells after 7 days. This showed that 5 days of additional culture in the presence of 500 μmol/L temozolomide induced neither an increase nor a reduction in metabolic activity, which strongly suggests that the tumor cells had stopped proliferating ( Fig. 1A, right).

To analyze cell death after temozolomide treatment, we used the LDH assay and CFSE. LDH is a cytoplasmatic enzyme that is released if a cell is dying. The LDH activity in the supernatant was only moderately increased after incubation with temozolomide in higher concentrations for 7 days ( Fig. 1B, left). In addition, we labeled CD133⁺ CSC lines with CFSE, a cytoplasmatic dye that is evenly distributed between daughter cells upon cell division but leaks if the cell is dying ( 24). The proportion of cells that had completely lost CFSE increased from 29% to 35% after 7 days incubation with maximal concentrations of temozolomide, suggesting that the substance only marginally induced cell death ( Fig. 1B, right). Conversely, temozolomide significantly reduced the average calculated number of cell divisions per day from 0.65 to 0.16 ( Fig. 1C, left) and the BrdUrd incorporation ( Fig. 1C and D, right) indicating impaired proliferation of the cell lines. Depending on the cell line, temozolomide induced either G₂-M arrest ( Fig. 1D) or G₂ peak loss (data not shown). In all cell lines and in line with previous publications ( 25), the proportion of cells in the sub-G₀ peak indicating apoptotic cell death did not exceed 8% ( Fig. 1D).

Temozolomide depletes CD133⁺ tumor cells in CD133⁺ CSC lines. In CD133⁺ CSC lines, only a small subpopulation of CD133⁺ CSC but not CD133⁻ tumor cells maintains the growth of the cell lines ( 4, 5, 26, 27). Thus, we analyzed the proportion of CD133⁺ tumor cells in temozolomide-treated CD133⁺ CSC lines. Whereas the proportion of viable cells declined by ∼8% after incubation with temozolomide, the proportion of CD133-expressing cells decreased in a time- and dose-dependent manner by up to 80% ( Fig. 2A ; data not shown). To monitor cellular proliferation and cell death of CD133⁺ and CD133⁻ cells, we labeled CD133⁺ CSC lines with CFSE and costained against CD133. When treated with temozolomide, CD133⁻ cells remained, whereas CD133⁺ cells were largely depleted ( Fig. 2B). Surprisingly, these data show that CD133⁺ cells proliferate more rapidly than CD133⁻ tumor cells (on average 0.9 versus 0.36 cell divisions per day; Fig. 2B, inset).

Temozolomide depletes CSC in CD133⁺ CSC lines. We have previously found that only 2% to 10% of CD133⁺ GBM cells possess the potential for sphere formation ( 3, 28). Likewise, others have proposed that only a subset of CD133⁺ tumor cells actually represent CSC ( 28, 29). Accordingly, a selective effect of temozolomide on CSC cannot be reliably assessed in the considerably larger CD133⁺ population. We therefore investigated sphere formation, which is a measure for the presence of CSC ( 30, 31) and observed that temozolomide-treated tumor cells dose-dependently lost their clonogenic potential ( Fig. 2C, left). This loss of stem-cell properties was observed at similar concentrations that were required to deplete CD133⁺ cells ( Fig. 2A). Treatment with temozolomide resulted in a reduced number of equally sized tumor spheres compared with control conditions ( Fig. 2C, right). This suggests that temozolomide does not inhibit the growth of progenitor and differentiated cells derived from CSC that have resisted the challenge with temozolomide. Because both, reversible cell cycle arrest and depletion of CSC, would impair tumor sphere growth, we replaced the continuous exposure to temozolomide by pulse treatment for 2, 4, and 7 days. After 42 days, the number of spheres derived from CD133⁺ CSC lines was determined. This revealed that incubation with 50 to 500 μmol/L temozolomide for 2 days was sufficient to completely abolish sphere formation ( Fig. 2D).

MGMT modulates temozolomide-mediated depletion of CSC in CD133⁺ CSC lines. Although the overall effects of temozolomide did not differ between the tested CSC lines, the temozolomide concentration required to deplete sphere-forming cells varied ( Fig. 3A ) and correlated with expression of MGMT mRNA ( Fig. 3B). The MGMT expression in the CSC line R28 was associated with a significantly higher proportion of CD133⁺ cells remaining after incubation with 50 μmol/L temozolomide for 7 days compared with the cell line R11 that was virtually devoid of MGMT mRNA ( Fig. 3C). In pulse-treated cells, incubation with 50 μmol/L temozolomide for 2 days eliminated all CSC in the MGMT-negative CSC line R11. In contrast, in the MGMT-positive cell line R28, 10-fold higher concentrations (500 μmol/L) only incompletely eliminated CSC with mainly differentiated tumor cells remaining ( Fig. 3D). As expected, the MGMT inhibitor 6-BG sensitized stem cell-like cells in this cell line to the deleterious effects of temozolomide, whereas the inhibitor had no effect on the MGMT-negative CSC line R11 ( Fig. 4A ; data not shown).

MGMT promoter methylation and protein expression in CD133⁺ CSC lines. Because the MGMT mRNA status correlated with the susceptibility of glioma cells toward temozolomide, we hypothesized that differential MGMT promoter methylation and MGMT protein expression in CSC might be responsible for the selective depletion of stem cell–like cells from CSC lines. Whereas the MGMT promoter methylation pattern did not differ between CD133⁺ and CD133⁻ tumor cells ( Fig. 4B), the MGMT protein expression was more pronounced in the CD133⁻ subset of tumor cells ( Fig. 4C). Accordingly, the increased susceptibility of CSC toward temozolomide is not due to a differential MGMT promoter methylation but is enhanced by differential MGMT protein expression in MGMT-positive cell lines.

Temozolomide treatment affects the differentiation potential of CD133⁺ CSC. We have previously shown that GBM-derived CSC can differentiate into cells expressing markers of all three neural lineages ( 3, 16). Therefore, we analyzed the expression of both stem cell and differentiation markers in solvent-treated or temozolomide-treated CD133⁺ CSC lines. As expected, the proportion of cells expressing the neural stem cell marker nestin declined upon treatment with temozolomide, whereas the percentage of GFAP⁺ cells (astrocytic lineage), β-tubulin-III⁺ cells (neuronal lineage), and GalC⁺ cells (oligodendroglial lineage) remained largely stable or even increased ( Fig. 4D).

Temozolomide also depletes CSC in CD133⁻ CSC lines. We have recently described a novel type of CD133⁻ GBM CSC ( 3). They show molecular profiles that are distinct from those of CD133⁺ GBM CSC and give rise to CD133⁻ CSC lines that showed adherently but no neurosphere-like growing spheres. Similar to CD133⁺ CSC lines, CD133⁻ CSC lines were resistant to temozolomide-induced cell death, whereas tumor cell proliferation was significantly inhibited ( Fig. 5A ; data not shown). Because there are no specific markers for CSC in CD133⁻ CSC lines, we determined the proportion of colony-forming cells as an indirect measure for the number of CSC. Upon treatment with 50 μmol/L temozolomide, the CD133⁻ CSC largely lost their clonogenic potential ( Fig. 5B). Incubation with 500 μmol/L temozolomide for 2 days caused no acute overall toxicity but completely eliminated clonogenic cells ( Fig. 5A and C). This may be interpreted as a loss of cells with stem cell-like properties, which is corroborated by a massive decline in the nestin-positive population ( Fig. 5D). In contrast, the fraction of the more differentiated subsets remained stable.

Temozolomide abolishes the tumorigenicity of MGMT-negative CD133⁺ CSC. Finally, we investigated whether the loss of clonogenicity in vitro corresponded to a reduced tumorigenicity of CD133⁺ CSC lines in vivo: CSC cultures (R11, MGMT negative; R28, MGMT positive) were incubated with increasing doses of temozolomide for 2 days, and 200,000 viable cells (assessed by trypan blue exclusion) were injected intracranially into nude mice. Again, temozolomide-treated cells were found to be completely viable (Supplementary Fig. S1). Twelve weeks after tumor inoculation, the tumor size was determined using 3T MRI. Pretreatment with temozolomide dose-dependently reduced the tumor size, i.e., the tumorigenicity of injected tumor cells ( Fig. 6A ). The results were reproduced with the CSC line R11 injected into additional 11 animals (Supplementary Table S1).

Animals were then sacrificed and brains were histologically analyzed. Again, we observed a dose-dependent massive decrease in tumor size irrespective of the MGMT status of the CSC lines injected ( Fig. 6B). Notably, lesions formed by the MGMT-positive CSC line (R28) and the MGMT-negative CSC line (R11) after pretreatment with 50 and 500 μmol/L temozolomide showed a highly different histologic phenotype. Although the MGMT-positive CSC line formed an infiltrating tumor with high cellularity, lesions formed by the pretreated MGMT-negative CSC line showed low cellularity, scar-like margins ( Fig. 6C), and were virtually devoid of proliferating cells ( Fig. 6D). This suggests that the injected pretreated MGMT-negative tumor cells survived in vivo but were not tumorigenic. The impaired tumorigenicity was seen at similar temozolomide concentrations that (a) eliminated CD133⁺ cells and (b) reduced clonogenicity in vitro and, thus, corroborates the concept of temozolomide preferentially eliminating CSC in GBM.

Discussion

Here, we describe that temozolomide preferentially depletes GBM cells with stem cell properties, whereas differentiated tumor cells are resistant to the agent. This finding clearly contradicts the predictions of the CSC hypothesis ( 2) and, therefore, warrants a rethinking of fundamental paradigms of glioma biology and therapy.

One part of the explanation may be that MGMT, the major mediator of temozolomide resistance in GBM, is overexpressed in CD133⁻ cells ( Fig. 4C). However, the selectivity of the drug for CSC must have further reasons because temozolomide also induces the depletion of MGMT-positive CSC as well as the differential depletion of CSC in MGMT-negative CSC lines ( Fig. 3C). Considering that temozolomide requires the formation of base mismatch and, therefore, cell cycle initiation to be effective, different rates of proliferation could also account for the differential susceptibility of stem cells and nonstem cells toward temozolomide. In this context, our finding that the CD133⁺ compartment (which also comprises the actual CSCs) proliferates faster than the CD133⁻ compartment ( Fig. 2B) is certainly of interest. Moreover, this surprising observation raises the question whether the rapid deregulated growth of GBM might be due to rapidly dividing GBM stem cells. However, a definite answer to this question would require the identification of unambiguous markers for CSC in GBM. Because these are not available to date, we cannot exclude the possibility that CSC are—as previously suggested—a small slowly proliferating subset ( 32) that is hardly detectable within the considerably larger population of CD133⁺ cells.

On a more functional level, we could show that temozolomide completely abolishes clonogenicity and tumorigenicity of CD133-positive and CD133-negative CSC lines ( Figs. 5B and C, and 6). It may therefore be hypothesized that temozolomide selectively kills CSC, which constitute such a small subset of the CD133⁺ cells (<10%) that the elimination of these cells only becomes apparent after prolonged culture ( Fig. 1). Alternatively, temozolomide might induce differentiation ( 33) or senescence ( 25) in stem cell–like tumor cells. Again, a conclusive answer to this question will only be possible once an unambiguous marker for CSC has been identified.

Our results clearly contradict recent reports that describe higher resistance of CD133⁺ cells to chemotherapy compared with CD133⁻ cells ( 19, 34). However, most of these studies use short-term metabolic assays to assess the viability of the tumor cells after treatment with extremely high ( 21, 25, 29, 35) temozolomide concentrations (500–2,000 μmol/L). Thus, the observed (still) low rates of metabolic inhibition (∼20%) may be unspecific and not indicative for cell death. Moreover, these authors did not use culture conditions known to maintain the stemness of CSC ( 3– 5, 14, 29, 30, 36– 41) but rather cultured their cells in medium supplemented with FCS. These conditions may not be compatible with the maintenance of the “stem cell state” because they are known to induce a multitude of genetic and transcriptional modifications and lead to the differentiation of CSC ( 14, 36, 39). The importance of suitable culture conditions for CSC-based assays is further confirmed by Clement and colleagues ( 29) who treated GBM CSC lines with 25 μmol/L temozolomide under defined serum-free culture conditions. In line with our results, these authors found that temzolomide induced a 50% to 80% decrease of proliferation in 3 of 4 tested GBM CSC lines with a marginal increase of caspase-3–positive tumor cells. In addition, they also describe slow rates of recovery 20 days after temozolomide treatment, which actually suggests an effect of temozolomide on CSC.

Although these data provide evidence for a selective depletion of CSC in GBM by temozolomide, we nevertheless found that all investigated CSC are completely resistant toward the cytotoxic effects of 5 μmol/L temozolomide, which corresponds to the peak concentration measured in the cerebrospinal fluid of patients ( 22). This resistance to temozolomide is in line with previous data ( 20) and may be due to the highly developed capacity of GBM cells to pump extraneous substances such as fluorochromes or chemotherapeutics out of the cell via ATP-binding cassette transporters (including MDR-1, MDP, and BCRP-1; refs. 11, 42). In addition, also MGMT-negative glioma cells may dispose of DNA repair mechanisms ( 25, 43) that counteract the cytotoxic effects when this alkylating agent is given at low doses. However, although concentrations required to eliminate MGMT-positive CSC (∼500 μmol/L) cannot be reached in vivo (which easily explains the failure of temozolomide-based chemotherapy in the respective patients), CSC from MGMT-negative tumors could be eliminated by 50 μmol/L temozolomide already. Although the exact temozolomide concentrations in the tumor cells are unknown, such concentrations have already been measured in the plasma of patients after standard treatment—and they might also be achieved in those regions of the tumor where the blood-brain barrier has been destroyed. Thus, an intriguing question arises: Why do patients with MGMT-negative tumors die despite the susceptibility of CSC to temozolomide? Considering the impaired blood-brain barrier that would enable temozolomide concentrations of up to 50 μmol/L to be achieved in the tumor, we speculate that glioma cells may hide in otherwise healthy brain tissue in which temozolomide concentrations of 5 μmol/L are likely. Accordingly, recurrent GBM in MGMT-negative tumors would derive from tumor stem cells that have invaded healthy brain tissues before therapy and were thus protected by the blood-brain barrier.

Importantly, our data may not only contribute to a better understanding of CSC in GBM, they also bear implications for the design and interpretation of clinical trials: In GBM, CSC might selectively be depleted (by cell cycle arrest, senescence, cell death, or differentiation), whereas the remaining tumor cells survive with reduced proliferative capacity. Thus, the tumor would not shrink under ongoing chemotherapy but remain stable for a prolonged period—which is exactly what is seen in many clinical responders. Moreover, the hypothesis that those CSC that drive a progressing tumor may not have been exposed to sufficiently high concentrations of temozolomide previously would justify further cycles of chemotherapy with the drug. Finally, our data support recent publications that suggest that dose-intensified temozolomide might be more effective than the conventional dosing scheme ( 10, 44– 46). Thus, optimized temozolomide dosing schemes might even completely eradicate CSC from MGMT-negative tumors, thereby providing a cure (or permanent tumor containment) for affected patients by using a well-established chemotherapeutic agent.

Disclosure of Potential Conflicts of Interest

P. Hau: Speakers honoraria. U. Bogdahn: Speakers honoraria. The other authors disclosed no potential conflicts of interest.