Genetic Diversity of Glioblastoma Multiforme : Impact on Future Therapies

Glioblastoma multiforme (GBM; WHO grade IV) is the most malignant type of glioma and in addition the most abundant malignant cancer of the adult human brain. Despite progress in diagnosis, surgery and chemotherapy, the median survival time of patients suffering from GBM is approximately 15 months (Stupp et al., 2005). The five years survival time is less than 5% (CBTRUS, 2010). Because glioblastoma cells show a highly infiltrating growth into the brain tissue, a total resection is not possible. In addition, glioblastoma cells are remarkably resistant to chemotherapy and ionizing radiation. In addition, the association of a portion of these cells with hypoxic and necrotic areas within the tumor increases their resistance. Gliobastoma multiforme tumors can be classified: 1. by histopathology (WHO) in conventional glioblastomas (93%), giant cell glioblastoma (5%) and gliosarcoma (2%); 2. by pathogenesis in primary GBM (90%) and secondary GBM (10%); 3. by gene expression analysis in (I) classical, (II) mesenchymal, (III) proneural or (IV) neural type of GBM; 4. by genomic analysis in subgroups harboring specific mutations and/or altered gene dosage/chromosome number. Conventional glioblastomas constitute approximately 93% of all glioblastomas and can be divided into primary or secondary tumors: primary glioblastomas represent approximately 90% and develop de novo, whereas the incidence of secondary glioblastomas that arise from astroyctomas WHO grade II and III is in the range of 5 to 10%. Primary and secondary glioblastomas differ in their genetic defects: for example 39% of primary glioblastomas harbor an amplification of the EGF receptor (EGFR) locus, whereas in secondary glioblastomas no amplification was detected. Mutations within the p53 gene are more abundant in secondary glioblastomas. Unconventional glioblastomas include giant cell glioblastomas, gliosarcomas and other rare types (for details see section 4).

Unlike oncogenes, tumor suppressor genes generally follow the "two hit" model, which implies that both alleles of a particular tumor suppressor gene have to be inactivated before an effect is manifested.If only one allele is inactivated, the second correct allele can still produce the correct protein.Whereas mutant oncogene alleles are typically dominant, mutant tumor suppressor genes are usually recessive.The mutational activation of oncogenes induces loss of heterozygosity and genomic instability in mammalian cells.These results have used to formulate the oncogene-induced replication stress model (for review see: Halazonetis et al., 2008).In precancerous lesions with intact p53 gene, the oncogene-induced DNA damage leads to p53-dependent apoptosis and/or senescence.After the function of p53 is lost, cells are able to escape its apoptotic and/or senescence effects, and the precancerous lesion is predestinated to become cancerous (Gorgoulis et al., 2005;Bartkova et al., 2005;Bartkova et al., 2006;Di Micco et al., 2006).DNA damage has an important role in promoting polyploidization.If cells with altered DNA enter mitosis, defects in chromosomal segregation and cytokinesis occur (for review see: Chow and Poon, 2010).The gene for the tumor suppressor protein p53 is mutated in about half of human cancers.It was shown recently by several groups, that eliminating p53 function by mutation leads to dramatically increased reprogramming efficiency of differentiated cells into induced pluripotent stem cells.Important for the field of cancer biology is the report of Mizuno et al. (2010), demonstrating that breast and lung cancers harboring TP53 mutations exhibit stem cell-like transcriptional signatures.These data suggest a role for active p53 in preventing the emergence of cancerous stem-like cells during tumor progression.Since TP53 mutations often arise in a late stage of tumor progression, when many cancer cells with different genetic alterations coexist, some cancer cells may be susceptible to reprogramming to generate stem-like cancer cells, leading to further tumor progression and cellular heterogeneity.

Genetic diversity of glioblastoma multiforme
DNA sequencing and gene dosage analysis of GBM revealed a high number of shared as well as individual-specific mutations, deletions and amplifications of DNA sequences.A hallmark of many primary GBMs is the loss of one copy of chromosome 10 harboring the locus for the PTEN tumor suppressor gene and/or amplification of the EGF receptor locus at chromosome 7.As a consequence, the Akt signalling pathway is often overactivated in GBM.Array comparative genomic hybridization (CGH) analyses revealed, that primary glioblastomas can be divided into three major genetic subgroups, i.e. tumors with chromosome 7 gain and chromosome 10 loss, tumors with chromosome 10 loss and tumors without copy number changes in chromosomes 7 or 10 (Misra et al., 2005).Parsons et al. (2008) sequenced 20661 genes coding for proteins in 22 GBM samples and 1 normal sample.They observed that 685 genes contained at least 1 non-silent somatic mutation.94% of these alterations were single base substitutions that were uniformly distributed among the 21 GBM samples, resulting in an average of 47 mutations per GBM.About 15% of the missense mutations were predicted to have a significant effect on protein function.The same 22 GBM samples were analysed for copy number alterations through hybridization of DNA samples to single nucleotide polymorphism (SNP) arrays, leading to the identification of 147 amplifications and 134 homozygous deletions.Parsons et al. (2008) next studied the probabilities that the mutations were either "driver" or "passenger".Driver mutations may provide a selective advantage to the cancer cell, whereas passenger mutations arise by the instability of the tumor genome and have no effect on tumor growth.Analysis of all data was used to identify GBM candidate cancer genes that were likely drivers, pointing to alterations in several signaling pathways: CDKN2A (altered in 50% of GBMs); TP53, EGFR, and PTEN (altered in 30 to 40%); NF1, CDK4, and RB1 (altered in 12 to 15%); and PIK3CA and PIK3R1 (altered in 8 to 10%).By analysing additional gene members within signaling pathways affected by these genes, the authors identified alterations of critical genes in the RB1 pathway (RB1, CDK4, and CDKN2A; altered in 68% of GBMs), TP53 pathway (TP53, MDM2, and MDM4; altered in 64%), and the PI3K/PTEN pathway (PIK3CA, PIK3R1, PTEN, and IRS1; altered in 50%).Mutations in the NF1 gene (coding for neurofibromatosis-related protein NF-1 or neurofibromin 1, a stimulator of GTPase activity of ras proteins) were observed in 16 of 105 GBMs (15%).Mutations in the IDH1 gene (coding for the citric acid cycle enzyme isocitrate dehydrogenase 1) were reported in 18 of 149 GBMs (12%).The Cancer Genome Atlas Research Network ( 2008) study analysed 91 GBM samples and found 453 non-silent somatic mutations in 223 genes.Affected signaling pathways include TP53, PTEN, NF1, EGFR, ERBB2, RB1, NF1, PIK3R1, and PIK3CA.High-level amplifications were observed frequently for EGFR, CDK4, PDGFR, MDM2, and MDM4 genes, whereas homozygous deletion events were often associated with CDKN2A/B and PTEN genes.In this study, GBMs from patients treated with temozolomide and/or lomustine were analysed for mutations.Treatment with alkylating agents resulted in more than tenfold increase in the number of mutations, that was dependent on the methylation status of the gene for the DNA repair enzyme O 6 -methylguanine-DNA methyltransferase (MGMT).Bredel et al. (2009) published "A Network Model of a Cooperative Genetic Landscape in Brain Tumors".The authors demonstrate that a multigene risk scoring model based on gene dosis and expression of 7 landscape genes (POLD2, CYCS, MYC, AKR1C3, YME1L1, ANXA7, and PDCD4) is associated with the overall length of survival in 189 glioblastoma samples.Yadav et al. (2009) reported that loss of function of ANXA7 (annexin 7) stabilizes the EGFR protein and increases EGFR signaling in glioblastoma cells.ANXA7 haploinsuffiency doubles the tumorigenic potential of glioblastoma cells.The heterozygous loss of ANXA7 in about 75% of GBM in the Cancer Genome Atlas Research Network study (2008) plus the observed infrequent ANXA7 mutation in about 6% of GBM is indicative for its role as a haploinsuffiency gene.A multigene predictor model of outcome in GBM based on expression analysis of 9 genes was published by Colman et al. (2009).Verhaak et al. (2010) used gene expression analysis to divide GBM into 4 subtypes: I. Classical, II.Mesenchymal, III.Proneural, and IV.Neural.The reproducibility of this classification was demonstrated in an independent validation set.To get insight into the genomic events, the authors used copy number and sequence data from the Cancer Genome Atlas Research Network (2008).I. Classical subtype of GBM (21% of core samples): Neural precursor and stem cell markers NES, as well as Notch and Sonic hedgehog signaling pathways were highly expressed in the Classical subtype.Chromosome 7 amplification paired with chromosome 10 loss was seen in 100% of the Classical subtype.High level EGF receptor (EGFR) gene amplification was observed in 97% of the Classical subtype and infrequently in other subtypes.Even though TP53 is the most frequently mutated gene in GBM (Cancer Genome Atlas Research Network, 2008), there was a distinct lack of TP53 mutations in the Classical subtype samples sequenced.Deletion events at 10q23 harboring the PTEN locus were observed in 100% of the Classical subtype.Focal 9p21.3 homozygous deletion targeting CDKN2A (encoding for both p16INK4A and p14 ARF) was frequent and co-occurred with EGFR amplification in 94% of the Classical subtype.
II. Mesenchymal subtype of GBM (32% of core samples): The Mesenchymal subtype displayed expression of mesenchymal markers as described earlier (Phillips et al. 2006) III.Proneural subtype of GBM (31% of core samples): The Proneural group showed high expression of oligodendrocytic genes, underlining its status as an atypical GBM subtype.The majority of TP53 mutations and TP53 loss of heterozygosity were found in Proneural samples.The classic GBM signature, chromosome 7 amplification associated with chromosome 10 loss was less prevalent and occurred in only 54% of the Proneural subtype.Focal amplifications of the locus at 4q12 harboring the PDGF Receptor A (PDGFRA) gene were seen in all subtypes of GBM but at a much higher rate (35%) in Proneural samples.11 of the 12 observed mutations in the isocitrate dehydrogenase 1 gene (IDH1) were found in this class.
IV. Neural subtype of GBM (16% of core samples): The Neural subtype was typified by the expression of neuron markers.

Subtypes and clinical correlations
Three of four tumors classified as secondary GBMs were found in the Proneural group.The Proneural subtype was associated with younger age, PDGFRA abnormalities, IDH1 and TP53 mutations, all of which have been associated with secondary GBM in earlier studies (Arjona et al., 2006;Furnari et al., 2007;Kleihues and Ohgaki, 1999;Watanabe et al., 1996;Yan et al., 2009).Verhaak et al. (2010) concluded that tumors did not change class at recurrence, because recurrent tumors were found in all subtypes (Murat et al., 2008).
Although statistically not significant, there was a trend towards longer survival for patients with a Proneural signature.Aggressive treatment significantly reduced mortality in Classical and Mesenchymal subtypes, had a less pronounced effect in the Neural subtype and did not alter survival in the Proneural subtype.There was no association of GBM subtype with methylation status of the DNA repair gene MGMT, which has been positively linked to therapy response (all data and conclusions from Verhaak et al., 2010).Bredel et al. (2010) reported that NFKBIA (nuclear factor of κ-light polypeptide gene enhancer in B-cells inhibitor-), an inhibitor of EGFR signaling pathway, is often deleted in GBM (Table 4).Most deletions occur in non-classical subtypes of GBM.Deletion and low expression of NFKBIA were reported to be associated with unfavorable outcomes.The authors present a two-gene model based on the expression of NFKBIA and MGMT that is strongly associated with the clinical course of GBM (Table 5).To add an additional level of complexity, gene dosage analysis of separate tumor areas derived from one GBM revealed area-specific genomic imbalances (see chapter 5).

Correlation between genetic and histopathologic diversity in GBM
Glioblastomas are morphologically highly heterogeneous and in addition the histological features often vary in different areas of one tumor.Currently three distinct common histopathological variants of GBM are recognized by the actual World Health Organization classification scheme, including conventional glioblastoma, giant cell glioblastoma, and gliosarcoma.Despite of lack of any histopathological difference, primary (de novo) and secondary (with an evidence of a lower-grade precursor) conventional glioblastomas harbor distinct molecular genetic abnormalities: Primary glioblastomas are characterized by relatively high frequencies of EGFR amplification, PTEN deletion, and CDKN2A (p16) loss, whereas secondary glioblastomas often contain TP53 mutations, especially those involving codons 248 and 273 or G:C->A:T mutations at CpG sites (Ohgaki et al., 2004).
Even within the conventional glioblastoma category, the cellular composition is heterogeneous and may include small or fibrillary, gemistocytic, granular, lipidized and occasional giant cells or oligodendroglial components.According to the predomination of one of these cell types indicating patterns of differentiation, the WHO distinguishes respective subtypes of glioblastoma such as small cell glioblastoma, glioblastoma with granular cell astrocytoma features, glioblastoma with lipidized cells; whereas giant cell glioblastoma is recognized as a distinct clinicopathologic entity (Kleihues et al., 2007;Miller and Perry, 2007).Small cell astrocytoma is an aggressive histologic variant being often misdiagnosed as anaplastic oligodendroglioma because of considerable morphologic similarities.Despite of histological overlap clinicopathologic and genetic features are distinct: there are no small cell astrocytomas harboring 1p/19q codeletions, whereas vIII mutant form of EGFR, EGFR amplification and 10q deletions are present in 50%, 69% and 97% of small cell astrocytomas, respectively (Perry et al., 2004).
Once thought to represent a reactive component, gemistocytes have been found to harbor TP53 mutations and cytogenetic abnormalities (chromosome 7p gains and 10q losses); therefore, they are now thought to represent a true neoplastic component (Kros et al., 2000).
In rare cases granular cells may predominate and create the impression of a granular cell tumor.Similar to astrocytomas with non-granular cytology, these tumors may also harbor TP53 mutations, high-frequency loss of heterozygosity at 9p, 10q, and 17p, and less frequent loss of heterozygosity at 1p and 19q (Castellano-Sanchez et al., 2003).Brat et al. (2002) reported the largest series of such tumors to date (22 cases, including 4 grade II, 7 grade III, and 11 grade IV tumors) and found that these tumors were more aggressive than nongranular cell astrocytomas of the same grade.
Glioblastoma with oligodendroglioma component is an astrocytoma WHO-grade IV containing oligodendroglial areas varying in size and frequency (Kleihues et al., 2007).Despite of oligodendroglial component and in contrast to rather frequent codeletions in WHO grade III anaplastic oligodendroglioma (approximately 85%), deletion of either 1p (24%), 19q (43%), or combined 1p/19q (22%) is relatively infrequent in glioblastoma with oligodendroglioma component (Miller and Perry, 2007).Giant cell glioblastoma is a rare variant that constitutes up to 5% of glioblastoma and is recognized as a distinct clinicopathologic entity in the WHO 2007 classification.Although occasional giant cells may be found in conventional glioblastoma, these cells are a predominating cytologic component in giant cell glioblastoma.As the name implies, the tumor cells are markedly enlarged and bizarre, often appear often multi-nucleated and tumor masses are typically well-circumscribed.It occurs in younger patients (fifth decade) (Kleihues et al., 2007).
Despite showing a very poor prognosis giant cell glioblastoma appears to carry a slightly better prognosis than conventional glioblastoma (Burger and Vollmer, 1980;Margetts and Kalyan-Raman, 1989;Shinojima et al., 2004), perhaps because of a less infiltrative behaviour.
Gliosarcoma constitutes roughly 2% of GBMs and is also recognized as a distinct clinicopathologic entity in the WHO 2007 classification.These tumors are characterized by their well-circumscribed, biphasic tissue pattern with clearly distinguishable areas of glial and mesenchymal differentiation.The glial component of gliosarcoma may display any of the aforementioned cytologic attributes and is typically immunoreactive for GFAP.The mesenchymal component is GFAP-negative and may also carry a wide variety of morphologic appearances, with evidence of differentiation along fibroblastic, cartilaginous, osseous, smooth and striated muscle, and adipose lines (Kleihues et al., 2007).There is a cytogenetic and molecular evidence for a monoclonal origin of both components (Actor et al., 2002;Paulus et al., 1994;Reis et al., 2000).Exept for the infrequent EGFR amplification, gliosarcomas are genetically similar to primary glioblastomas: they harbor likewise low frequency of TP53 mutations (up to 24%) and similar rates of PTEN deletions (38%) as well as deletions of p16 INK4a gene (at the CDKN2A locus, 9p21) in roughly 37% (Actor et al., 2002;Reis et al., 2000).

Area-specific genomic Imbalances in glioblastoma multiforme
Using flow cytometry data analysis Hoshino et al. (1978) reported that different tumor regions of one glioblastoma showed a highly variable distribution of ploidy.By use of a DNA fingerprinting technique Misra et al. (2000) analysed genetic alterations within two or three tumor areas from seven glioblastomas.In all cases except one, different areas of one tumor displayed different fingerprints, indicating a striking extent of intratumoral genetic heterogeneity.Conventional comparative genomic hybridization (CGH) was used to study the intratumoral patterns of genomic imbalance in Glioblastoma multiforme (Harada et al., 1998;Jung et al., 1999).Array comparative genomic hybridization was utilized by Nobusawa et al. (2010) to study in detail tumor area-specific genomic imbalances.Genetic alterations common to all the areas analyzed within a single tumor included gains at chromosomes 1q32.1 (PIK3C2B, MDM4), 4q11-q12 (KIT, PDGFRA), 7p12.1-11.2(EGFR), 12q13.3-12q14.1 (GLI1, CDK4), and 12q15 (MDM2), and loss of chromosomes at 9p21.1-24.3(p16 INK4a /p14 ARF = CDKN2a), 10p15.3-q26.3(PTEN, etc.), and 13q12.11-q34(SPRY2, RB1).These alterations are likely to be causative in the pathogenesis of glioblastomas (driver mutations).Additionally, the authors reported numerous tumor area-specific genomic imbalances, which may be either nonfunctional (passenger mutations) or functional, but constitute secondary events reflecting clonal selection and/or progressive genomic instability, a hallmark of glioblastomas.Area specific-evolution of genomic imbalances in GBM may be comparable to the genetic evolution and genomic instability of metastatic pancreas cancer that has been studied in detail recently (Campbell et al., 2010;Yachida et al., 2010).Loeper et al. (2001) reported that frequent mitotic errors occur in genetically microheterogenous glioblastomas.The authors used fluorescent in situ hybridization (FISH) to study chromosome numbers in a series of 24 glioblastomas.All examined chromosomes showed mitotic instability indicated by numerical aberrations within significant amounts of tumor cells.For chromosomes 10 and 17 only monosomy was observed, whereas chromosome 7 showed trisomy/polysomy.In contrast to other chromosomes displaying monosomy as well as trisomy, copy number changes of chromosomes 7, 10 and 17 seem to be the result of selection in favor of the respective aberration (Loeper et al., 2001).In this context it is interesting to note that neural stem and progenitor cells in the subventricular zone of mouse postnatal brain are frequently aneuploid (Kaushal et al., 2003) and that chromosome segregation defects contribute to aneuploidy in normal neural progenitor cells of the mouse cerebral cortex (Yang et al., 2003).Studies in mice and human demonstrate that chromosomal mosaicism is a prominent feature of neural stem cells, whereas interchromosomal translocations or partial chromosomal deletions or insertions are extremely rare (for review see: Peterson et al., 2008).Glioblastoma stem cells share several properties with neural stem cells, i.e. the growth in floating spheres under serum-free conditions, the expression of the stem cell marker nestin and the differentiation into neural cells like astrocytes or neurons.This similarity in marker expression and behaviour has led to the hypothesis that glioblastoma stem cells may be derived from NSCs (Berger et al., 2004;Sanai et al., 2005).
Recent research makes clear that GBMs do not behave as a whole; local heterogeneity may arise because the tumor regionally adapts to the microenvironment.The influence of microenvironment-induced stimuli may be the force behind clonal selection and acquisition of area specific genomic imbalances in GBM.In addition, regional genomic alterations may be associated with the development of resistance to irradiation and/or chemotherapy, resulting in tumor recurrence and/or progression.

Mechanisms leading to genetic alterations in glioblastoma multiforme
Analysis of copy number alterations showed an average of 7 amplifications and 6 homozygous deletions per GBM.In addition, an average of 47 mutations was reported (Parsons et al., 2008).
A characteristic feature of GBM is a chromosomal instability (CIN) phenotype distinguished by the loss or gain of complete chromosomes, for example by the gain of chromosome 7 and/or loss of chromosome 10.These chromosome copy number changes can be explained by merotelic spindle attachment that is associated with bipolar but more often with multipolar mitosis.Multipolar spindle pole coalescence in cells with supernumerary centrosomes has been reported as a major source of chromosomal misattachment and chromosome missegregation in colorectal cancer cell lines (Silkworth et al., 2009).
Obviously, specific chromosome aberrations may be associated with growth advantage for clonal populations of cancer cells (for example by the loss of tumor suppressor genes).
In addition to its well-defined role in signal transduction at the plasma membrane, recent results have identified PTEN as a new guardian of the genome (for review see: Yin and Shen, 2008).Pten-deficient mouse embryo fibroblasts revealed an increased frequency of mitotic centromere-associated chromosomal instability as well as spontaneous DNA doublestrand breaks (Shen et al., 2007).Li Li et al. (2008) developed a mouse model by infecting PTEN-/-neural precursor cells with an EGFRvIII expressing retrovirus and found that EGFRvIII expression and PTEN loss synergistically induced chromosomal instability and glial tumors.Interestingly, polyploidization of mammalian hepatocytes occurs through failed cytokinesis and is followed by a process that was called reductive mitoses (Duncan et al. 2010).The authors postulate a dynamic model of hepatocyte polyploidization, ploidy reversal and aneuploidy (ploidy conveyor) and propose that this mechanism evolved to generate genetic diversity and permits adaptation of hepatocytes to xenobiotic and nutritional injury.
Several studies point to a link between centrosome amplification, chromosomal instability and the development of cancer (for review see: D´Assoro et al., 2002).Cells in resected high grade gliomas and cultured glioblastoma cells have been reported to exhibit often centrosome amplifications (Loh et al., 2010) and the centrosomal protein -tubulin is overexpressed and shows altered subcellular localization in GBM (Katsetos et al., 2006;Loh et al., 2010).Multipolar mitoses were occasionally observed in time lapse recordings of cultured glioblastoma cells (Hegedüs et al., 2000).Our laboratory used long term life cell imaging to study mitoses in a newly established glioblastoma cell line and found that cytokinesis defects followed by multipolar mitosis may be an important mechanism that is used by glioblastoma cells to reduce ploidy and generate viable daughter cells (our unpublished results).

Epigenetics in glioblastoma multiforme
To add another level of complexity, many types of cancer cells carry aberrant epigenetic modifications.Changes in epigenetic marks (caused or not caused by genetic alterations) may have an fundamental impact on tumor development and/or tumor progression.Epigenetic markers in human gliomas have been reviewed by Hesson et al. (2008).Two groups have studied in detail DNA methylation profiles in GBM (Etcheverry et al. 2010;Nousmehr et al., 2010).Hypermethylation at a large number of genetic loci occurred in a subgroup (proneural group) of glioblastoma patients and was associated with improved outcome (Nousmehr et al., 2010).
Epigenetic mechanisms like methylation of DNA have already an impact on chemotheraypy of GBM.Temozolomide (TMZ, Temodal®) is an orally administered alkylating drug that is often used for chemotherapy of GBM.O 6 -Methylguanine-DNA methyltransferase (MGMT) is a DNA repair enzyme that specifically removes promutagenic alkyl DNA adducts from the O 6 position of guanine residues in DNA which are induced by alkylating agents like temozolomide (Goth and Rajewsky, 1974;Margison and Kleihues, 1975).Loss of MGMT expression may be caused by transcriptional silencing through hypermethylation of its CpG islands (Esteller et al., 1999;Qian & Brent, 1997), is frequently (45% to 75%) present in glioblastomas (Bello et al., 2004;Kamiryo et al., 2004;Nakamura et al., 2001) and results in improved survival of glioblastoma patients treated with the alkylating agent temozolomide (Fukushima et al., 2009;Hegi et al, 2005;Hegi et al., 2008).On the other hand not all glioblastoma patients with MGMT promoter methylation respond to alkylating agents and in addition responding GBMs cannot avoid eventual recurrence (Fukushima et al., 2009;Hegi et al., 2008).MGMT promoter methylation appears to occur with a higher frequency in secondary than in primary glioblastoma (Bello et al., 2004;Nakamura et al., 2001) but there is no evidence about its correlation with other histopathologic subtypes.However, prospective randomized studies of EORTC (European Organisation for Research and Treatment of Cancer) and NCIC (National Cancer Institute of Canada) trial have revealed a significant prolongation of progression free and overall survival for patients with newly diagnosed glioblastoma treated by the concomitant and adjuvant temozolomide and irradiation.By this means median survival has been increased over one year (Stupp et al., 2005(Stupp et al., , 2009)).
The methylation status of the MGMT gene promoter is being used as a biomarker for the potential benefit of the addition of temozolomide to the therapy because its epigenetic silencing has been identified as a strong and independent predictive factor of treatment response for anaplastic glioma patients undergoing chemotherapy with alkylating agents (Hegi et al., 2005;Wick et al., 2009).3 to 5% of the GBM patients survive for more than 3 years.MGMT hypermethylation was reported to be significantly more frequent in the longterm survivor group (Krex et al., 2007).
The assumption that DNA methylation of CpG island on the MGMT promoter represses consecutively transcriptional activity of the MGMT gene and expression of MGMT protein has been used to explain the correlation between the positive promoter methylation status and favorable treatment response after chemotherapy with temozolomide (Kaina et al., 2007).However, studies that were performed to validate a relationship between MGMT promoter methylation and protein expression have yielded contradictory results in brain tumors as well as in other neoplasms (Brell et al., 2011).While some studies report a significant correlation between MGMT protein expression analyzed by immunohistochemistry (IHC) and MGMT promoter status measured by methylation-specific polymerase chain reaction (MSP) in glioblastoma and brain metastases of various origin (Ingold et al., 2009;Spiegl-Kreinecker et al., 2009;Tang et al., 2011), other studies failed to detect correlations between both parameters (Brell et al., 2005(Brell et al., , 2011;;Christmann et al., 2010;Preusser et al., 2008).
In addition, there is increasing evidence that MGMT mRNA expression, unlike MGMT protein expression, could be a better predictor for tumor sensitivity to alkylating agents than MGMT methylation status (Everhard et al., 2009;Kreth et al., 2011).Kreth et al. (2011) provide not only evidence that the degree of MGMT mRNA expression is highly correlated with the MGMT promoter methylation status, but also that low MGMT mRNA expression is strongly predictive for prolonged time to progression, treatment response, and length of survival.Furthermore, the authors found that in case of discordance the patients with methylated tumors combined with high MGMT mRNA expression did significantly worse than those with low transcriptional activity or unmethylated tumors with low MGMT mRNA expression.
Finally Kreth et al. (2011) assume methylation-independent pathways of MGMT expression regulation; however, the exact role of DNA-methyltransferases DNMT1 and DNMT3b that are likely to be involved in methylation of CpG islands of MGMT gene promoter remains unclear.
In the Cancer Genome Atlas Research Network ( 2008) study, GBMs from patients treated with temozolomide and/or lomustine were analysed for mutations.Treatment with alkylating agents resulted in a more than tenfold increase in the number of mutations that was dependent on the methylation status of the MGMT gene.This phenotype seems to be caused by mutations in the MSH6 gene (Cahill et al., 2007;Hunter et al., 2006;Yip et al., 2009) and other genes of the DNA mismatch repair pathway (Cancer Genome Atlas Research Network, 2008).The loss of the mismatch repair protein MSH6 in GBM is associated with tumor progression during temozolomide treatment (Cahill et al., 2007;Hunter et al., 2006;Yip et al., 2009).

Stem cells in glioblastoma multiforme
The cancer stem cell concept that is of importance for the genesis of many types of cancer receives increasing credit also in the field of GBM.A minor population of GBM cancer stem cells, which may be derived from genetically altered neural stem cells, is presumed to generate transit amplifying cells with high mitotic activity.Because these stem cells appear to have a low mitotic activity, they are difficult to target by radiotherapy and conventional chemotherapy.
For recent reviews on glioblastoma stem cells the reader is referred to Huang et al. (2010) and McLendon and Rich (2010).Ignatova et al. (2002) firstly described cells with stem-like properties in human cortical glial tumors.Singh et al. (2003) used the cell surface marker CD133 to isolate a clonogenic population of cells showing stem-like features in medulloblastomas and pilocytic astrocytomas.These cells were declared as tumor stem cells based on their capabilities of self-renewal and multilineage differentiation.Galli et al. (2004) and Yuan et al. (2004) confirmed these findings for glioblastomas.Bao et al. (2006a) selected CD133+ cells from glioblastoma biopsies that were capable of forming tumorspheres in vitro, demonstrated self-renewal and multilineage differentiation and resulted in tumours after transplantation into nude mice.In contrast, CD133-cells did not form tumorspheres in vitro und were not tumorigenic in nude mice.CD133+ cells proved to be a minor population of cells in GBM biopsies.Clinical studies suggested that the percentage of CD133+ cells (Zhang et al., 2008;Zeppernick et al., 2008) or the rate of tumorsphere formation in vitro (Laks et al., 2009;Panosyan et al., 2010) can be used to predict overall survival time of patients.However, it should be noted, that contrary results also exist (Phi et al., 2009;Kim et al., 2011).In recurrent glioblastomas the percentage of CD133+ cells is increased strongly when compared with primary glioblastomas (Pallini et al., 2010).Surprisingly, the increase in expression of CD133 after tumor recurrence was associated significantly with longer survival.Thon et al. (2008) described a correlation between the amount of CD133+ cells within the tumor mass and the WHO grade of glioma (WHO grade II, III and IV).Bao et al. (2006a) demonstrated that CD133+ cells constitutively expressed DNA repair genes at much higher levels that CD133-cells, mediating resistance to X-irradiation in CD133+ cells.
Brain tumor stem cells seem to be localized in a perivascular niche (Bao et al., 2006b;Calabrese et al., 2007;Shen et al., 2008) and low oxygen tension (hypoxia) is associated with its undifferentiated state.In glioblastomas, cancer stem cells express much higher levels of VEGF than non-stem cancer cells and show increased angiogenic potential in vivo (Bao et al., 2006b;Li et al., 2009).Because VEGF expression is under control of transcription factors of the hypoxia inducible factor (HIF) family, one should note that expression of HIF2 is unique to glioma stem cells and correlated with poor patient survival, whereas HIF1 is found in all malignant cells (Li et al., 2009).It has been reported that different human cancers (GBM, colorectal carcinoma, and NSCL carcinoma) converge at the HIF2 oncogenic axis (Franovic et al., 2009).The authors propose that inhibition of HIF2 may be of broad clinical interest in the treatment of cancers with different genetic signatures.Hjelmeland et al. (2010) published recently that acidic stress promotes a glioma stem cell phenotype by induction of HIF2 and other glioma stem cell markers.The authors suggest that an increase in intratumoral pH may be of benefit for targeting the stem cell phenotype.Three recent papers demonstrate that stem-like cells in GBM are able to differentiate into endothelial cells and may give rise to tumor endothelium (Ricci-Vitiani et al., 2010;Thon et al., 2008;Wang et al., 2010).These results define a novel mechanism for cancer vasculogenesis and may help to explain the failure of currently used inhibitors of angiogenesis.Glioma stem cells as targets for novel strategies of treatment have been recently reviewed (Dietrich et al., 2010;Gilbert & Ross, 2011).

Promising targets for chemotherapy of glioblastoma multiforme
GBMs are highly infiltrative tumors that show resistance to conventional chemotherapy.Many chemotherapeutic agents are not able to reach the tumor in sufficient doses, because the blood brain barrier is at least partially intact in these tumors.Most mitotic inhibitors used in clinic impair the function of mitotic spindles by targeting tubulins that are basic components of microtubules.Because microtubules in non-mitotic cells are also affected, these compounds often exhibit significant side effects (for example neurotoxicity).Future therapies of GBM may involve small molecules that inhibit the activity of aurora kinases A or B, polo kinases or the mitotic kinesin Eg5, all proteins that have specific functions in different phases of mitosis (for review see: Kaestner and Bastians, 2010;Sudakin and Yen, 2007).Pharmaceutical companies are on the way to develop selective inhibitors that target these proteins.Phase I and II studies on different forms of solid cancers are currently underway to study newly developed mitosis inhibitors and may also open the way for a more efficient therapy of GBM.Interestingly, it has been reported that in glioblastoma expression of aurora kinases A (Barton et al., 2010) and B (Zeng et al., 2007) were both associated with poor prognosis and may be targets for therapy.Among several other proteins also histone deacetylases (HDACs) may be promising targets for future therapy of GBM (Argyriou and Kalofonos, 2009).ABC transporters play an important role in the development of multidrug resistance.The role of ABC transporters in the resistance network of glioblastoma was reviewed by Bleau et al. (2009).
The humanized monoclonal antibody against vascular endothelial growth factor (Bevacizumab, Avastin®) has been approved by the FDA for treatment of GBM.Although targeting the tumor vasculature with Bevacizumab reduced the number of cancer-like stem cells in orthotopic brain tumor xenografts (Calabrese et al., 2007), a recent phase II study indicates that bevacizumab does not affect median survival of patients with recurrent GBM (Pope et al, 2011).
Potential targets for directed therapy of GBM may include extracellular matrix proteins of the perivascular niche that influence proliferation and/or migration of cancer stem cells.
Targeting integrin 6 has recently been shown to inhibit self-renewal, proliferation, and tumor formation capacity of glioblastoma stem cells (Lathia et al., 2010).Cilengitide (Impetreve®) is a cyclic pentapeptide harboring a RGD sequence.RGD sequences present on extracellular matrix proteins mediate the binding to integrins, a class of cell surface receptors.Cilengitide is a selective inhibitor of ν 3/5 integrins and currently under study as an inhibitor of angiogenesis in several types of solid cancer.Cilengitide monotherapy was well tolerated and exhibited modest antitumor activity among patients with recurrent GBM in a randomized phase II study (Reardon et al., 2008).Also targeting glioma stem cells the neural cell adhesion molecule L1CAM has been reported to suppress glioma growth (Bao et al., 2008).Glioblastoma cells display complex surface structures with numerous microvilli and filopodia that resist the actions of cytolytic effector lymphocytes (Hoa et al., 2010).It should also be noted that gliomas are accompanied by numerous microglia/macrophages.As was recently reported, inhibition of microglia/macrophage activation may represent a new and effective strategy to suppress proliferation of glioma cells (Zhai et al., 2011).Subtypes of breast cancer or leukemia can be efficiently treated by inhibiting the one excessively activated signal transduction pathway that is linked to malignancy.For GBM a monocausal therapy by inhibition of a single overactivated signaling pathway seems to be less promising, because cells or even regions with different genetic defects coexist in one tumor.A personalized therapy based on analysis of the individual genetic defects is not yet in sight for GBM.

Summary and perspective
Many types of cancer cells evolve through a multistep process in which genetic aberrations accumulate and finally lead to cells exhibiting aberrant gene expression programs.GBM has been considered as a system/network disease (Fathallah-Shaykh, 2010), because its phenotypes appear to be generated by several interconnected aberrant signal transduction pathways as well as numerous molecular abnormalities, thereby resulting in uncontrolled mitosis and migration of GBM cancer cells.In GBM local heterogeneity arises as the tumor regionally adapts to microenvironmental cues.Future molecular therapies of GBM should target its Achilles' heels: the elimination of the small intratumoral subpopulation of cells that exert stem cell properties and the inhibition of mitosis within the population of transit amplifying cells, which is responsible for forming the tumor mass.

Acknowledgments
We thank the Maria-Pesch Stiftung and the Nolting-Stiftung for support.

References
Actor, B.; Cobbers, J.M., Buschges, R., Wolter, M., Knobbe, C.B., Lichter, P., Reifenberger, G. & Weber, R.G. (2002).Comprehensive analysis of genomic alterations in gliosarcoma and its two tissue components.Genes Chromosomes Cancer, Vol.34, No.4, (December 2002), pp. 416-427, ISSN 1045-2257 Giant cell g l i o b l a s t o m a s h a v e a n i n c r e a s e d e x p r e s s i o n o f A u r o r a K i n a s e B ; c o m b i n e d w i t h T P 53 mutations this may be responsible in induce cytokinesis defects and the development of multinucleated cells (

Table 4 .
Relationship of four molecular subtypes of glioblastoma to gene-dosage profiles for NFKBIA and EGFR across 188 glioblastomas.NFKBIA deletions are rare in classical (6%) and more common in non-classical (32%) glioblastomas.
Verhaak et al. (2010)pe a degree of mutual exclusivity between NFKBIA deletion and EGFR amplification was suggested: NFKBIA deletion or EGFR amplification were observed in 53%, whereas concomitant occurrence (NFKBIA deletion together with EGFR amplification) was observed only in 5%.All data and conclusions are fromBredel et al. (2010).Glioblastoma subtypes were classified according toVerhaak et al. (2010).

Table 5 .
Bredel et al. (2010)on of the clinical course of GBM with expression of NFKBIA and expression/methylation status of the promoter of O 6 -methylguanine-DNA methyltransferase (MGMT).All data and conclusions are fromBredel et al. (2010).

Table 6 .
Temme et al., 2010).Genetic profile of the common histopathological glioblastoma subtypes.Similar tendencies are indicated by grayscale.Modified from