For several types of brain tumors, including a subgroup of primary GBM, CSC were found to express CD133. CD133⁺, but not CD133^-, tumor cells were able to reconstitute the initial tumor in vivo when injected into immune-deficient nude mice 910. However, recent reports indicate that this initially proposed model may represent an oversimplification and stem cell-specificity of the epitope detected by the antibody AC133 (i.e. glycosylated prominin, CD133 11) has been questioned 12. GBM cells may acquire CD133 after xenotransplanation 13, conversely CD133⁺ and CD133^- cells within CSC lines may have similar tumorigenic potential 1415. In addition, CD133 does not appear to be essential for stem cell-like properties, as subgroups of GBM driven by CD133^- CSC have recently been identified 161718. Thus, stem cell-specific markers other than CD133 were sought for. In recent years new markers (e.g. CD15/SSEA-1, integrin α6) have been described, but there is no consensus on the optimal markers for CSC in GBM 18192021. The CSC hypothesis states that tumor relapses are driven by CSC having escaped multimodal therapy. Possible explanations for treatment failure include insufficient drug delivery or the fact that the treatment targets only more differentiated tumor cells (the tumor bulk), while sparing the small subpopulation of CSC (e.g. via CSC specific mechanisms to escape chemotherapy-induced cell death) 82223. The CSC hypothesis further predicts that only therapies that efficiently eliminate the CSC fraction of a tumor are able to induce long-term responses and thereby halt tumor progression.

However, stem cell-specific therapies, although preventing further growth of the tumor, will not result in an impressive shrinkage of the lesion in vivo but in a persisting period of stable disease that may be followed by a late reduction of tumor volume 8. Because CSC constitute only a rare subpopulation within a tumor, a therapeutic agent selectively depleting CSC will not substantially reduce the overall viability of tumor cells, but may efficiently inhibit their proliferation. In a clinical context, the CSC hypothesis challenges the classical oncological response criteria and questions the evaluation of therapeutic approaches by their effects on the tumor bulk 24.

In vitro, these implications raise questions about viability assays used to determine the survival of all tumor cells (e.g. metabolic assays like 3-(4,5-dimethylthiazole-2-yl)-2,5-diphenyltetrazoliumbromide (MTT) or water soluble tetrazolium (WST), apoptosis/necrosis assays like terminal deoxynucleotidyl transferase dUTP nick end labeling (TUNEL) or propidium iodide). Such assays may fail to detect the stem cell-specific effects of a drug. Sequential in vivo transplantation experiments are the gold-standard to detect the small population of CSC within the whole of tumor cells 25. Still, this experimental approach is only feasible to answer specific experimental questions but does not allow screening experiments. Several additional techniques that may help to better investigate the response of CSC to chemotherapy are currently evaluated (e.g. immediate testing of sorted tumor cells, treatment of tumor bearing mice, monitoring of DNA in CSC, mixed culture of CSC and cells without stem cell properties). At the moment, the combination of experiments investigating clonogenicity, stem cell marker expression, and differentiation of tumor cells may constitute a feasible set of screening experiments for stem cell toxicity in vitro (summarized in 12). Recently described new cell culture methods that might help to avoid the above mentioned time-consuming experimental approaches remain controversial and require further scientific evaluation 2627.

To date, two groups of alkylating agents are commonly used in the clinic: TMZ (8-Carbamoyl-3-methylimidazo (5, 1-d)-1, 2, 3, 5-tetrazin-4(3H)-one) and nitrosoureas (BCNU - carmustine, ACNU - nimustine, CCNU - lomustine; referred to as CNUs). In contrast to CNUs, TMZ undergoes rapid non-enzymatic conversion at physiologic pH to its reactive compound 5-3-(

m

t

i

c

The most frequent sites of DNA base alkylation by CNUs are the N⁷-position of guanine and the N³-position of adenine 3435. In addition, CNUs mediate their cytotoxic effect by chloroethylation at the O⁶-position of guanine which produces an N¹-deoxyguanosinyl-N³-deoxycytidyl crosslink that is poorly repaired and causes DNA strand break during mitosis. The type of cell death caused by CNU-induced DNA damage is dependent on the p53 status of the tumour cells. In p53 wildtype cells CNUs induce apoptotic cell death while they trigger necrotic cell death in p53 deficient cells 36. Thus, CNUs are also active if a functional DNA repair system is lacking, whereas TMZ requires a well-established DNA repair system to be effective. The mechanisms of action of TMZ and CNUs suggest that differences in the DNA repair system of CSC and tumor cells without stem cell properties are important to understand any differential sensitivities of CSC to alkylating agents, and how these may arise.

In addition, it may help to understand a putative differential susceptibility of CSC to CNUs and TMZ. To present, three studies have investigated the DNA repair systems in GBM CSC. McCord et al. 37 compared the radioresistance of CSC lines and classical serum cultured cell lines. CSC lines showed a substantially impaired DNA repair resulting in a reduced radioresistance of CSC lines. Within a subset of CSC lines, the CD133⁺ cells were more resistant to irradiation with 2.5 Gy as compared to CD133^- non-stem cells. These data indicate that the mechanisms mediating CSC radioresponse differ from those in the traditional model. The report by McCord et al. complements an earlier report by Bao et al. 22 showing CSC resist radiotherapy due to a more efficient DNA system than non-stem cells. After irradiation, CSC activate the DNA damage checkpoint more effectively than tumor cells without stem cell properties due to the activation of Chk1 and Chk2 checkpoint kinases 22. However, the differential activation of the DNA repair system in CSC has been questioned by a recent study 38 and it remains to be clarified if the more efficient DNA repair system postulated by Bao et al. 22 constitutes a common property of CSC. Given that TMZ requires an efficient DNA repair (namely the mismatch repair system) to exert its cytotoxic actions, a detailed knowledge of the DNA repair system in CSC as compared to non-CSC will be crucial to help our understanding how TMZ affects the survival of CSC. The functional role of the DNA repair system in CSC and non-CSC may be further complicated by recent data suggesting different molecular subtypes within GBM 39. Because the subgroups also differ with respect to p53 mutations, the intertumoral heterogeneity will further increase the complexity of TMZ resistance and susceptibly.

Although results derived from serum-cultured glioma cell lines may be biased by multiple new mutations induced during long-term culture in serum-containing medium 4041, several authors have described stem cell-like cells within these cell lines 4243. Therefore, experimental data collected in the "pre-CSC era" may provide some clues to the effects of TMZ on CSC and the mechanisms of action. The induction of tumor cell death has been considered as the most important contribution of a therapeutic drug to the overall treatment efficacy. Roos et al. 44 showed that TMZ-induced receptor-mediated or mitochondrial apoptosis in glioma cells was depending on their p53 status. Irrespective of the p53 status, after long-term incubation with TMZ (for up to 144 h), dose-dependent apoptosis occurred at late time points starting after 72-96 h. However, even after continuous exposure to TMZ doses of up to 500 μM, at least 30% viable cells remained. Similar results were described by Hermisson et al. 45 investigating the effects of TMZ on a panel of different glioma cell lines. Even high concentrations of TMZ (up to 1290 μM) resulted in the survival of up to 40% of the cells, indicating cytostatic rather than cytotoxic actions of TMZ in these assays. Interestingly, colony-formation seemed to be much more sensitive to treatment with TMZ even after short-term incubation for only 24 h. The EC₅₀ values for the prevention of clonogenic growth ranged between 7 μM and 511 μM and correlated with the MGMT status of the cell lines. Hirose et al. 33 treated p53 wild-type cells with 100 μM TMZ for 3 h. The population of apoptotic cells did not exceed 10% but clonogenicity was dose-dependently decreased or completely inhibited. The results of these three publications are representative of multiple reports from glioma and other brain tumor cell lines, reporting almost uniformly a much stronger effect of TMZ on clonogenic growth than on the overall survival of glioma cell lines in vitro 33464748.

CNUs are more toxic to p53 mutant cells as compared to TMZ. As reported by Roos et al. 36, ACNU-induced cross-links or double strand breaks are repaired in p53 wild-type cells but accumulate in p53 mutant cells. In addition, the repair genes xpc and ddb2 were not up-regulated in p53 mutant cells in response to ACNU indicating a DNA repair defect in the cells causing hypersensitivity to CNUs. On a functional level, colony-formation was much more sensitive to treatment with CNUs than cell viability which did not fall below 30%. In addition, cell death was a late occurring event 36. Similar findings were reported by Bobola et al. 47, Ashley et al. 49, and Iwado et al. 48. Together, the majority of reports suggest that CNUs preferentially eliminate clonogenic cells but hardly affect the overall viability. Still, this effect was less pronounced than with TMZ.

In summary, there is a consensus in the literature that the clonogenic growth monitored by colony-forming assays is much more sensitive to treatment with either TMZ or CNUs in vitro as compared to the overall cell death in vitro. These findings do not suggest an increased resistance of clonogenic cells (including CSC) towards TMZ or other alkylating agents.

However, none of the papers investigated CSC in detail or used assays specific for CSC. As non-stem cells may also show clonogenic growth without being able to proliferate infinitely, the results have to be interpreted with caution 50. In addition, clonogenic growth is the sum of all events occurring after treatment (including a.o. cell cycle arrest, cell death) and therefore does not allow definite conclusions on the cause of clonogenic cell inactivation.

Several recent studies have focused on the effects of chemotherapy on CSC. The CSC hypothesis led to the development of a new cell culture model for GBM. GBM cell lines derived from freshly resected tumor specimens and cultured in serum-free medium supplemented with EGF and bFGF - conditions optimized for the growth of neural stem cells - mirror the phenotype and genotype of primary tumors more closely than serum-cultured cell lines do 4041. Thus, experiments with this new cell culture model may yield more valid results on the efficacy of therapeutic agents. The relevance of conventional cell lines (cultured under mainly serum-based media conditions) as means to investigate CSC is disputed. Although the results may be biased by multiple new mutations induced during long-term culture in serum-containing media, these studies are also discussed here. Table 1 gives an overview of studies discussed and indicates the possible technical limitations and key findings of each study.

Eramo et al. were the first to investigate the chemoresistance of GBM CSC. A marked resistance was observed of GBM CSC lines to different chemotherapeutic agents, amongst them TMZ. GBM CSC lines were treated with 250 μM TMZ for 48 h and minor cell death was determined as a measure of treatment efficacy 51. Using the same culture model, Clement et al. observed a dose-dependent decrease of proliferation when CD133⁺ CSC lines were treated with up to 100 μM TMZ, whereas only a minor increase of cell death was observed at similar concentrations 52. The authors compared the effects of TMZ with the inhibition of the hedgehoc-GLI1 (shh) signaling pathway and found synergistic activity of shh signaling and TMZ. Notably, administration of each of these substances inhibited cell proliferation, although to a different degree. The effects of TMZ treatment on the tumorigenicity of GBM CSC were not investigated. Beier et al. described that TMZ may selectively deplete clonogenic and tumorigenic cells in a dose-dependent manner whereas it hardly affected overall viability. When CD133⁺ CSC lines derived from primary astrocytic GBM were treated with up to 500 μM TMZ over different periods of time (2 d and 42 d), there was a dose-dependent reduction of CD133⁺ cells (by 80%), cell proliferation (by up to 100%), as well as clonogenicity and tumorigenicity (by up to 100%), whereas cell death did not exceed 6%. Cells with stem cell-like properties were selectively depleted irrespective of the CD133 or MGMT status. However, MGMT expressing CSC lines required up to 100 fold higher TMZ concentrations. In summary, the authors interpreted their results as selective depletion of CSC from CSC lines by TMZ 53. Additional confirmatory experiments (e.g. treatment of tumor bearing mice or characterization of tumor material direct after resection) have not been performed. In addition, the authors only investigated three different TMZ concentrations.

In contrast to these reports, Liu et al. used CSC lines that have been established using serum and switched to serum-free culture conditions before the experiments. The authors found that sorted CD133⁺ CSC showed significant resistance to chemotherapeutic agents (including, amongst others, TMZ). The viability of CD133⁺ cells was significantly less decreased than the viability of CD133^- tumor cells when treated with up to 2000 μM TMZ for 48 h. Notably, even this concentration did not induce cell death in more than 50% of the cells 54. Ghods et al. investigated chemoresistant and aggressive CSC-like cells within the serum cultured 9L gliosarcoma rat cell line. Viability of 9L cells grown in serum-containing medium as a monolayer and 9L cells grown after serum-deprivation as spheres were compared. The 9L cells grown as spheres were more chemoresistant than cells grown as monolayers. Thus, the authors concluded that CSC were more chemoresistant 55. Bleau et al. investigated the side population phenotype cells within murine glioma CSC lines derived from PDGF-induced gliomas after TMZ treatment. In line with a report by Chua et al. 56, the number of side population phenotype cells increased, especially in cells lacking PTEN expression, when CSC lines were treated with up to 100 μM TMZ for up to 2 weeks 57. This translated into an increased tumorigenicity of the long-term TMZ treated cell lines. Unfortunately, neither the MGMT status of the CSC nor the effects of TMZ on stem cell properties were determined. However, the different tumorigenicity of treated and untreated CSC indicated that TMZ increased the number of CSC. Blough et al. investigated the sensitivity of 20 GBM-derived CSC lines cultured under standard conditions 58. They found that 9 out of 20 CSC lines were susceptible to TMZ, while 11 were resistant. The expression of MGMT, but not methylation of the MGMT promoter, indicated TMZ response in vitro. A correlation of CD133 expression and resistance towards TMZ was not reported 59. TMZ substantially eliminated the proportion of CD133⁺ stem cell-like cells in a study by Lamoral-Theys et al. using oligodendroglioma HS683 cells. HS683 cells were cultured using serum, but not using medium conditions optimized for CSC. The authors found that long-term treatment of HS683 cells with increasing concentrations of TMZ (from 0.1 μM to 1000 μM) resulted in a substantially decreased tumorigenicity of long-term treated cells, despite a wash-out phase of 8 weeks after TMZ treatment. Notably, the long-term treatment reduced the tumorigenicity and CD133 expression much more effectively than short-term treatment, indicating that the dosing schema influences the depletion of CSC in vitro. The MGMT expression remained unchanged in both groups 60. Villalva et al. investigated effects of STAT3 inhibition and TMZ using CSC lines. They found that both, inhibition of STAT3 and TMZ, substantially inhibited the clonogenic growth of CSC lines and both substances showed a strong synergy 61. Similar results were reported by Hsieh et al. who investigated the effects of IGF-1 and shh signaling in CSC lines 62.

Based on the high toxicity of TMZ to neurosphere-forming cells within CSC lines, Mihaliak et al. 63 developed a neurosphere recovery assay that was used to investigate the resistance of GBM in the patients. Interestingly and in line with data by Blough et al. 59 and Beier al. 53, they found that the clinically relevant concentrations were able to inhibit the growth of neurospheres in a subgroup of CSC lines and reduced their tumorigenicity. In contrast to the data published by Beier et al., TMZ completely depleted clonogenic cells in only 1 out of 5 CSC lines investigated. NOTCH inhibition successfully inhibited the recovery of neurospheres after successful depletion of clonogenic cells in vitro 64.

Two recent studies investigated the susceptibility of CSC derived from the core and the periphery of GBM in vivo: Pistollato et al. 65 focused on the effects of the intratumoral hypoxic gradient on the distribution, phenotype and TMZ resistance of CSC. They established and compared CSC lines from the core, the intermediate area, and the periphery of GBM. CSC lines from the periphery showed a higher susceptibility to TMZ induced cell death as compared to CSC from the core. Conversely, cells in the core frequently overexpressed CD133 and MGMT. In summary, the authors postulated that the hypoxic gradient drives the distribution of stem cells. A complementary report by Glas et al. 66 confirmed the differential responses of CSC lines derived from the core and the periphery in a subgroup of GBM, however in most of the samples, there was no uniform difference in the response of the corresponding CSC lines towards TMZ, CNUs, or radiation (alone or in combination).

To present, only one paper has thoroughly investigated the effects of CNUs on CSC lines 67. Based on an increased expression of the stem cell marker CD133 and a neurosphere-like growth pattern of resistant cells, the authors concluded that BCNU increased the proportion of CSC in the cell lines investigated.

Due to differences in experimental approaches, these studies are difficult to compare. Given that serum-cultured cell lines rapidly acquire multiple mutations and substantially differ from the original tumor, they may provide conflicting results as compared to studies using medium conditions favoring the growth of CSC. However, most recent studies used CSC lines continuously cultured under serum-free conditions. In summary, Eramo et al., Bleau et al., and Pistollato et al. reported an increased resistance of CSC while Beier et al., Bleau et al., Mihaliak et al., Gilbert et al., and Clement et al. described experimental data relating to CSC and TMZ which is consistent with an increased susceptibility of CSC towards TMZ. The data presented by Glas et al., do not allow unambiguous conclusions on the susceptibility of CSC towards chemotherapy (overview in Table 1).

The data presented in the above-mentioned papers allows several conclusions and points towards open questions: (I) All papers uniformly reported that MGMT protein expression is associated with a high resistance of CSC to TMZ. However the role of other factors known to modulate chemoresistance has not been investigated in detail. (II) Whenever investigated in detail, neurosphere-forming cells without MGMT expression were susceptible to TMZ. (III) Different TMZ schedules and concentrations may result in conflicting experimental results. (IV) Environmental factors, like hypoxia in the core of GBM, may contribute to the resistance of CSC towards TMZ. (V) Several signaling pathways, e.g. Shh, IGF-1/PI-3 kinase, NOTCH, or STAT3 interfere with TMZ resistance. (VI) There is no consensus on adequate methodologies to investigate CSC in vitro.

A set of papers (listed in Table 1) on CSC and TMZ focused on the factors modulating chemoresistance of CSC to alkylating agents. In vitro, approximately half of all GBM cell lines resisted to TMZ concentrations of 50 μM 455359. They obviously activated mechanisms of intrinsic chemoresistance (Figure 1). To date, MGMT is the best characterized and the most important modulator of chemoresistance in GBM 68697071. There is consensus in the literature that MGMT expression substantially increases the resistance of CSC 53545965. However, there is conflicting data regarding overexpression of MGMT in the stem cell compartment 5354. The expression of MGMT in GBM CSC results in a 10-fold increase of TMZ-resistance 4553. The resistance of MGMT expressing CSC 59 fits well to the clinical observations that patients without methylation of the MGMT promoter rarely survive longer than 2 years 69. In these tumors, all cells may have an obvious intrinsic resistance to TMZ. However, relapsing GBM maintained by CSC without intrinsic resistance to TMZ are difficult to understand, assuming an intrinsic susceptibility of CSC.

The role of multi-drug resistance proteins is controversial. Although expressed in GBM cells, it remains unknown whether TMZ is actually transported by these proteins. Although TMZ incubation increased the proportion of ABCG1 expressing cells (i.e. side population cells) 56, TMZ was not a substrate for the ABCG1 transporter in murine glioma cells 72. Schaich et al. reported that MDR1 (ABCB1) mediated TMZ resistance and was an independent predictor for TMZ responsiveness 73.

TMZ requires an intact mismatch repair system to cause toxic double stranded breaks. Thus, mutations in the major crucial component of the mismatch repair system mutS homolog 6 (MSH6) mediated TMZ resistance especially in recurrent GBM after TMZ-based radiochemotherapy 7475. A well-characterized mechanism of resistance inhibits cell death induced by double stranded breaks. Signalling cascades involved include mutations of p53 and Poly(ADP-ribose)polymerase (PARP) signalling and mutations modulating the apoptotic cascade that executes double stranded break-induced apoptosis 4476. Additional but less well-characterized proteins contributing to chemoresistance include protein glutathione-S-transferase. To date, none of these mechanisms of chemoresistance in CSC lines have been investigated in detail.

In addition, the fate of CSC after TMZ treatment in vitro and in vivo remains unknown, although a loss of stem cell properties in GBM CSC lines and a pronounced sensitivity of colony-formation to treatment with TMZ has been observed in serum cultured cell lines 3345464748. In susceptible GBM, TMZ can induce only long lasting cell cycle arrest (G2/M arrest) 3377, senescence 7879, or autophagy 8081. However, despite the mismatch of minor cell death and massively reduced colony-formation, apoptotic cell death 33444882 always occurs.

Under experimental conditions, TMZ may completely eliminate CSC, suggesting that a selective depletion in vitro is feasible 53. In contrast, the survival of patients treated with TMZ did not stabilize 7 suggesting that residual CSC survived in vivo (Figure 1).

The blood-brain-barrier constitutes a specific hurdle for drugs in the brain that reduces the activity of chemotherapy due to decreased concentrations in the brain parenchyma. In approximately 50% of all glioma cells lines, clonogenic cells and/or GBM CSC were highly susceptible to TMZ concentrations of 50 μM in vitro. Still, TMZ concentrations of 5 μM did not deplete clonogenic cells in vitro irrespective of cell culture model or MGMT status 455359. The actual concentrations achieved in the plasma and the brain parenchyma of the patients are therefore important to understand if and where within a GBM TMZ concentrations suffice to eliminate tumor cells, irrespective of stem cell properties. The maximum TMZ concentrations in the plasma of patients are well established and range between 27 μM and 50 μM 83. In contrast, the maximum TMZ concentration in the brain is still unknown. PET studies suggested a concentration of 10-18 μM in the normal parenchyma, however, this methodology does not differentiate between intravasal and interstitial TMZ 84. In CSF, maximal TMZ concentrations ranged from 0.5-10 μM with most patients achieving concentrations around 5 μM (i.e. 1 μg/ml) 85. These values were confirmed by a study reporting on interstitial TMZ concentrations of 3 μM 86. Although detailed studies are lacking, it appears plausible that tumour cells within the contrast-enhancing lesions (i.e. regions of impaired blood-brain barrier) are exposed to concentrations approximating plasma concentrations. In contrast, invading tumours cells are likely to be exposed to concentrations measured in the healthy brain parenchyma (approximately 3-5 μM). Unfortunately, similar studies for CNUs are lacking. Knowing the susceptibility of CSC in vitro, these studies strongly suggest that the incomplete penetration of the blood-brain barrier by TMZ may constitute a major factor for chemoresistance of invading CSC. In line with this idea, a recent report analyzing the recurrence patterns of GBM could show that MGMT non-methylated GBM tend to relapse at the site of primary occurrence 87 whereas MGMT-methylated GBM showed a significantly higher proportion of distant metastases. However, this report remains controversial 88 and the concept described does not explain why susceptible GBM also develop recurrences develop from contrast enhancing lesions under ongoing TMZ therapy.

Extrinsic factors of the microenvironment contribute to the chemoresistance of haematological and solid tumors (reviewed in 89). In these tumors, three different mechanisms of environmentally-mediated chemoresistance have been described: direct cell-cell interactions, local secretion of cytokines like IL-6 or SFD-1, or micro-environmental factors such as hypoxia 8990. Although these mechanisms may substantially contribute to chemoresistance of GBM there is only indirect evidence to suggest that these mechanisms may be relevant for GBM CSC. Villalva et al. showed that the activation of STAT3 increases the resistance of CSC to TMZ 61. Given the expression of IL-6 protein family members like LIF in GBM 91, it is likely that they contribute to chemoresistance of GBM CSC.

The CSC hypothesis predicted that the selective elimination of CSC would halt the perpetual growth and tumorigenicity of tumor cells. However, this hierarchy was recently called into question 1450. It was suggested that the "stemness" of tumor cells may be induced by niche factors such as hypoxia 65929394. Conversely, these data imply that tumor cells may acquire CSC properties. Several lines of evidence support this concept: Wang et al. were the first to show that CD133-negative tumors may acquire CD133 in a xenotransplantation model 13. More recently, Chen et al. reported that different types of CSC within one cell line. Within a given cell line, CSC may give rise to other subtypes suggesting a less stringent cellular hierarchy 14. In melanoma, Morrison et al. showed that these tumors may entirely lack a cellular hierarchy 95. The generation of new CSC from differentiated cells may therefore represent a new mechanism for therapeutic resistance after selective elimination of CSC 8. However, additional experimental data is required to confirm this concept.

There is no consensus in the literature on the effects of chemotherapeutic drugs on GBM CSC in vivo or in vitro. Still, the available experimental data clearly indicate that resistance of GBM and derived CSC against TMZ is much more complex than expected. The initial predictions of the CSC hypothesis remain true under specific experimental conditions but multiple additional aspects have to be considered to understand why GBM almost invariably relapse. This review summarizes several known aspects of chemoresistance; additional contributors are likely to be discovered. A more comprehensive view will help to better understand the multifaceted factors that contribute to the survival or reoccurrence of CSC causing fatal relapses in the patient.