Molecular diagnostics of gliomas: state of the art

Several of the molecular alterations detected in gliomas may have diagnostic and/or prognostic implications, as they are associated with histologically defined tumor types or malignancy grades. However, for most of the molecular changes this does not justify a designation as glioma biomarker, because biomarkers should provide unique diagnostic, prognostic, or predictive information exceeding that reached by mere histological classification. A meaningful diagnostic biomarker should be helpful in the classification of tumors with ambiguous histological features or allow for a clinically useful subdivision of tumors within a given histological tumor type. A useful prognostic biomarker should correlate with disease-free and overall survival, ideally providing information beyond that obtained by established prognostic parameters, such as patient age, clinical performance status, extent of resection, and WHO grade. A predictive biomarker, finally, should provide valuable information on the response to a given therapy, which will help to stratify patients into distinct therapeutic groups to allow for the optimal (“personalized”) treatment. In this regard, the number of molecular biomarkers in neurooncology to date is limited to a few alterations, namely combined deletions of the chromosome arms 1p and 19q in oligodendroglial tumors, MGMT hypermethylation in glioblastomas and anaplastic gliomas, IDH1 and IDH2 mutations in diffuse gliomas, as well as BRAF aberrations in pilocytic astrocytomas (Table 1). More detailed information on the respective testing methods and the information conveyed by these markers is provided in the subsequent paragraphs of this review. In addition, we will review some of the recent advancements concerning the understanding of the biological mechanisms of treatment resistance in high-grade gliomas, such as treatment-associated somatic mutations of the mismatch repair gene MSH6 [15, 57, 139]. We will also discuss the potential role of predictive testing for responsiveness towards targeted therapies, such as the assessment of the EGF receptor status in adult high-grade gliomas or the BRAF status in pilocytic astrocytomas.

Loss of 1p and 19q is the genetic hallmark of oligodendroglial tumors [68, 81]. Frequencies reported vary due to interobserver variability in the distinction of oligoastrocytomas from oligodendrogliomas on the one hand and astrocytomas on the other, as well as the techniques used and the loci investigated [68]. Overall, losses of 1p and 19q are detected in up to 80% of oligodendrogliomas (WHO grade II) and approximately 60% of anaplastic oligodendrogliomas (WHO grade III), whereas 30–50% of oligoastrocytomas, 20–30% of anaplastic oligoastrocytomas, and <10% of diffuse astrocytic gliomas, including the glioblastomas, carry this aberration. The observation that usually both 1p and 19q are completely lost is explained by an unbalanced translocation t(1;19)(q10;p10), of which the 1p–19q derivative is lost, whereas the 1q–19p derivative is retained during cell replication [42, 62]. There is a strong association between 1p/19q codeletion and classical oligodendroglial features on histology (e.g., perinuclear halo and chicken-wire vascular pattern). However, morphology alone cannot predict the 1p/19q status [40, 109] and up till now the diagnosis of oligodendroglial neoplasms is based on the morphological criteria [81].

In 1998, it was first reported that 1p (and combined 1p/19q) loss predicts better response to chemotherapy and longer survival in anaplastic oligodendroglioma patients [17]. Since then, many subsequent studies have been performed, including three prospective randomized phase III trials that corroborated 1p/19q deletion as a powerful prognostic marker in patients with WHO grade III gliomas. Importantly, these studies also indicated that the prognostic power was independent of the type of adjuvant therapy, that is radiotherapy, chemotherapy or combined radio-/chemotherapy [16, 122, 135]. Nevertheless, discussion remains regarding the predictive (response to therapy) versus prognostic (independent of therapy) nature of this marker. Retrospective data on oligodendroglial tumor patients not receiving any radio- or chemotherapy after their initial surgical treatment revealed that 1p/19q loss was not associated with longer progression-free survival [131]. These data would suggest that 1p/19q loss characterizes a group of gliomas that is more sensitive to genotoxic therapy in general, i.e., radio- and alkylating chemotherapy, and, when treated, are associated with significantly longer survival.

Given the undisputed prognostic significance of 1p/19q loss in oligodendroglial tumor patients receiving adjuvant therapy, many institutions have now established diagnostic testing for this aberration. However, while 1p/19q loss is associated with more favorable prognosis of patients receiving adjuvant treatment, it needs to be emphasized that this marker is of limited help for making treatment decisions, such as radio- versus chemotherapy. One should also be aware that the prognostic relevance of 1p/19q loss may be less pronounced in the presence of other, prognostically unfavorable genetic alterations [121]. In addition, different types of 1p losses have been identified that have distinct prognostic implications. For example, oligodendroglial tumors carrying partial terminal or interstitial 1p losses are associated with shorter patient survival when compared with tumors with combined complete 1p/19q losses [33, 125] (Figs. 2, 3). Thus, molecular testing for 1p loss alone by studying just distal markers on 1p36 may pick up cases with less favorable prognosis when compared with oligodendroglial tumors having a complete 1p/19q co-deletion.

Techniques used and the loci investigated for 1p/19q testing differ widely among institutions, depending primarily on the local expertise, existing laboratory equipment and preferences of the neuropathologist and molecular biologist involved (Table 2). Most commonly, loss of heterozygosity (LOH) [59, 102] or (fluorescent) in situ hybridization ((F)ISH) [47, 114] analyses are used. FISH may be preferred by pathologists as tissue characteristics are retained and no corresponding blood sample is required. LOH analysis is most commonly available, but requires comparative evaluation of the same set of loci in DNA extracted from normal cells of the patient, usually peripheral blood leukocytes. In addition, one should be aware that “pseudo-LOH” may be detected in some tumors displaying allelic imbalances due to copy number gain rather than loss of one allele, e.g. the frequent gain of 19q in primary glioblastomas. On the other hand, LOH analysis may be scaled up more easily than FISH to the analysis of multiple loci along each chromosome, which would help to avoid detection of false-positive cases with partial 1p loss or isolated 19q loss, and may even (partly) be automated in larger molecular diagnostic units. Multiplex ligation-dependent probe amplification (MLPA) is another methodical option, as it allows detection of copy number changes of up to 45 loci in a single experiment [63, 64, 111]. Similar to LOH analysis, MLPA only requires standard laboratory facilities for PCR and capillary gel electrophoreses that are widely available in molecular laboratories, but has the advantage that a blood sample of the patient is not required.

The MGMT (O ⁶-methylguanine-DNA methyltransferase) gene at 10q26 is frequently silenced by promoter hypermethylation in diffuse gliomas and this has been pinpointed as an epigenetic mechanism reducing MGMT expression levels. Importantly, an association between MGMT promoter methylation and the response of malignant gliomas to alkylating chemotherapy using nitrosourea compounds [32], temozolomide [48], or a combination of both [53] has been observed. Based on MGMT promoter methylation studies in a subpopulation of patients involved in the EORTC/NCIC 22981/26981 trial [48, 117], Hegi et al. reported that patients treated with radiotherapy and temozolomide, and whose tumors had a methylated MGMT promoter (which is seen in approximately 40% of primary glioblastomas) survive significantly longer when compared with patients whose tumors lacked MGMT promoter methylation [48]. In this land-mark paper, MGMT promoter methylation did not significantly influence survival in patients treated with radiotherapy alone, suggesting that the MGMT hypermethylation is predictive for favorable response to chemotherapy.

Mechanistically, a predictive power of the MGMT promoter status can be explained by the fact that MGMT encodes a DNA repair protein that removes alkyl groups from the O ⁶ position of guanine, which are introduced by alkylating chemotherapeutic agents, such as temozolomide [39]. Chemotherapy-induced DNA alkylation triggers cytotoxicity and apoptosis, while high MGMT expression in the tumor cells counteracts the cytotoxic effects and thereby may cause treatment failure. MGMT promoter methylation, however, impedes the transcription of the gene resulting in a lack of MGMT mRNA and protein expression. Glioblastoma cells with MGMT promoter hypermethylation thus respond better to temozolomide, as they lack the ability to repair the therapy-induced DNA damage.

Because the current standard of care for glioblastoma patients involves radiotherapy plus concomitant and adjuvant temozolomide treatment [117], testing for MGMT promoter methylation by means of methylation-specific polymerase (MSP) chain reaction analysis or other methods is now increasingly performed, not only for patients in clinical trials but also in the routine diagnostic setting. The advantage of MGMT methylation testing is that—if detected—a methylation signal specifically results from the neoplastic glial cells. In contrast, attempts to replace MGMT promoter methylation testing by a technically more simple immunohistochemical analysis of MGMT protein expression did not prove reliable for diagnostic purposes in most studies, as the immunohistochemical assays may be disturbed by a variable content of contaminating non-neoplastic cells, such as microglial cells, lymphocytes, reactive astrocytes and vascular cells, which retain MGMT expression also in MGMT-hypermethylated gliomas [34]. The same applies for expression analyses based on Western blotting or reverse transcription-PCR, as well as biochemical assays detecting enzymatic activity. All these methods may be confounded by contaminating non-neoplastic cells in the investigated tissue homogenates. Nevertheless, several studies reported on significant prognostic associations using such assays on glioma tissues [7, 22, 61, 78].

Although data from the EORTC/NCIC 22981/26981 trial [80] and another large prospective patient cohort [132] found that MGMT promoter methylation was predictive for longer survival only in those patients who received temozolomide, a recent paper reported that MGMT promoter methylation is also predictive of response to radiotherapy and linked to longer survival in the absence of adjuvant chemotherapy [104]. Although this association is somewhat debatable in glioblastomas, prospective trials on anaplastic (WHO grade III) gliomas clearly indicated a prognostic role of MGMT promoter methylation independent from the type of adjuvant treatment, i.e., alkylating chemotherapy or radiotherapy [123, 135]. The reason for a prognostic role of MGMT promoter methylation in patients just receiving radiotherapy remains unclear, and is rather unexpected from a functional point of view. One may speculate that the strong association of MGMT hypermethylation with 1p/19q codeletion [9, 88, 124] and IDH1 mutation [108] (Fig. 3) may reflect a global molecular constellation in anaplastic gliomas that per se is associated with higher sensitivity to cytotoxic therapy and more favorable outcome. Even further, the MGMT-hypermethylated anaplastic gliomas may belong to a group of gliomas characterized by multiple hypermethylated genes [91], potentially including yet unknown genes mediating radioresistance.

Similar to the findings in adult glioma patients, MGMT promoter methylation [110], and reduced MGMT protein expression [100] have been reported to be associated with longer survival of pediatric malignant glioma patients. In ependymal tumors, MGMT promoter methylation is less common when compared with diffuse astrocytic and oligodendroglial tumors [13, 74]. Other malignant brain tumors, such as medulloblastomas [31] and anaplastic meningiomas [12] rarely demonstrate MGMT promoter methylation.

Mutations in the gene encoding the human cytosolic NADPH-dependent isocitrate dehydrogenase (IDH1), an enzyme that participates in the citric acid cycle, were originally identified in 2008 employing large scale sequencing analysis of 22 glioblastomas [95]. All mutations were found in the evolutionarily conserved residue R132 that is located in the substrate-binding site of IDH1. The mutations were detected preferentially in glioblastomas of young patients and in secondary glioblastomas [95]. Soon after, multiple studies corroborated these findings and additionally revealed that somatic IDH1 mutations are present in the vast majority of low-grade diffuse (WHO grade II) and anaplastic (WHO grade III) astrocytic, oligodendroglial, and mixed oligodendroglial neoplasms [2, 90, 137]. In a small subset of WHO grade II and III, diffuse gliomas and secondary glioblastomas that lack IDH1 mutations, the gene of the mitochondrial isoform IDH2 was found to harbor mutations affecting the analogous amino acid (R172) [137]. In contrast, IDH1 or IDH2 mutations are rare in primary glioblastomas and restricted to only individual cases of other primary brain tumors [2, 90, 137].

Already the original paper of Parsons et al. [95] recognized that IDH1 mutations were associated with prolonged overall survival in glioblastoma patients. Subsequent studies analyzing the whole spectrum of diffuse gliomas underscored the association between IDH1 mutation and better outcome, multivariate analyses often revealing IDH1 mutation as a strong and independent favorable prognostic marker [30, 108, 123, 132, 135]. So far, there is no evidence that the type of IDH1 mutation (R132H vs. non-R132H) affects patient survival [41]. In contrast to the strong prognostic significance, several studies reported a lack of predictive significance of IDH1 mutations in gliomas for response to (chemo)therapy [30, 123].

At the genetic level, the presence of IDH1 mutations in diffuse gliomas is strongly correlated with TP53 mutation or 1p/19q deletions [58, 123, 129]. Analysis of multiple glioma biopsies from the same patient revealed that IDH1 mutations do not occur after the acquisition of either a TP53 mutation or loss of 1p/19q [129]. In another study, all 128 gliomas with a complete co-deletion of 1p/19q harbored an IDH1 or IDH2 mutation [77]. These mutations thus seem to represent a very early event, affecting a common glial precursor cell, and may in fact form a prerequisite for the t(1;19) translocation leading to 1p/19q co-deletion and oligodendroglial tumor development. The presence of IDH1 mutations in gliomas was also reported to be tightly associated with MGMT promoter methylation and inversely correlated with the loss of chromosome 10 and EGFR amplification [108, 123]. This latter finding further underscores the notion that primary and secondary glioblastomas are genetically distinct entities, despite their histological similarities [129].

In up to 90% of the diffuse gliomas with mutated IDH1, the mutation is of the R132H (G395A) type [2, 90, 95, 137]. One study of more than 1,000 diffuse WHO grade II and III gliomas showed that R132C IDH1 mutations are associated with astrocytic neoplasms, while IDH2 mutations predominantly occur in oligodendroglial tumors [46]. Another study revealed that non-R132H IDH1 mutations are rare in classic oligodendrogliomas with 1p/19q loss and occur at significantly higher frequency in other grade II and III gliomas, including those with TP53 mutations [41]. Interestingly, all five examined astrocytomas of patients with Li-Fraumeni syndrome carried R132C IDH1 mutations, indicating that glial/glioma precursor cells with a germline TP53 mutation carry an increased risk to acquire such a non-R132H mutation [130].

IDH1 mutations have now also been reported in a subset of (anaplastic) gangliogliomas [56, 116], supratentorial primitive neuroectodermal tumors [2] and in gliomatosis cerebri [112]. IDH mutations are rare or absent in other glial tumors, such as pilocytic astrocytomas [75, 129] and ependymomas [129]. Importantly, the vast majority of high-grade (WHO grade III and IV) gliomas in the pediatric age group lacks IDH1 mutation as well, corroborating the notion that these pediatric neoplasms are fundamentally different from their adult counterparts [1, 2, 96]. IDH1 mutations are reported to occur in some non-central nervous system tumors, e.g., a subset of acute myeloid leukemias with a normal karyotype [23, 103], and occasionally in prostate cancer, B-ALL [23], and paraganglioma [37]. In the vast majority of non-glial tumors, however, such mutations are absent [8].

Under physiological conditions, IDH1 and IDH2 are involved in multiple metabolic processes, such as lipid synthesis, cellular defense against oxidative stress, oxidative respiration, and signal transduction. There are different hypotheses on the role of IDH1 and IDH2 mutations in gliomagenesis, including an effect on the stabilization of hypoxia inducible factor 1 (HIF-1), upregulation of (other) genes involved in angiogenesis, glucose transport and glycolysis, and inhibition of developmental apoptosis [5, 103, 141]. However, recent studies suggest that the heterozygous IDH1 and IDH2 mutations in gliomas do not just result in a loss of function, that is a reduced ability to catalyze conversion of isocitrate to α-ketoglutarate [137], but also confer an enzymatic gain of function, in particular the ability to catalyze the NADPH-dependent reduction of α-ketoglutarate to 2-hydroxyglutarate (2HG). Indeed, gliomas harboring IDH1 mutations demonstrate markedly elevated levels of 2HG, and this ‘onco-metabolite’ may contribute to the oncogenesis of gliomas [26]. Similarly, IDH1 and IDH2 mutations dramatically increase 2HG in acute myeloid leukemia [43, 128]. Because of the specificity of the IDH1 and IDH2 mutations in gliomas and of their metabolic effects, there is hope that these aberrations provide new avenues for anti-cancer therapies [36]. However, further study of the (exact effect of) IDH1 and IDH2 mutations is needed to seize this opportunity.

Various methods are applied for the detection of IDH1 and/or IDH2 mutations in clinical glioma specimens (Table 2), including single-strand conformation polymorphism analysis, direct sequencing [71, 77, 96], PCR-restriction fragment length polymorphism-based assays [14, 56, 86], DNA pyrosequencing and real-time PCR with post-PCR fluorescence melting curve analysis assays [46, 55]. Especially, these latter assays allow for rapid, inexpensive, and sensitive analysis of IDH1 and IDH2 mutations in routinely processed (i.e. formalin-fixed, paraffin-embedded) tumor tissue, even in samples with a low tumor cell content [14, 35, 55]. Only recently, specific, and robust monoclonal antibodies were established that can be used for immunohistochemical analysis of gliomas bearing the IDH1 R132H mutation [19, 72]. Testing for mutations in IDH1 or IDH2 can now easily and effectively be performed in a clinical setting and thereby enhance the diagnostic accuracy, especially, in cases where traditional methods are insufficient to reach a definitive diagnosis. For example, IDH1 and IDH2 mutation analysis might be helpful in case of a differential diagnosis of diffuse glioma versus pilocytic astrocytoma or ependymoma, and for discrimination between primary and secondary glioblastoma [56, 75]. Using immunohistochemistry with the anti-IDH1R132H antibody, even individual cells of gliomas with the IDH1 R132H mutation (e.g. in the periphery of diffuse infiltrative gliomas) can be detected and differentiated from non-neoplastic glial cells [19]. Of note, lack of staining with this antibody does not rule out the presence of an IDH1 mutation (about 10% of the IDH1 mutated gliomas carry a non-R132H mutation) nor of an IDH2 mutation. Given the diagnostic and prognostic importance as well as the robustness and relative ease of IDH1 and IDH2 mutation testing, it is very likely that determination of these mutations will soon become a part of the routine diagnostic assessment of gliomas. Furthermore, it needs to be investigated to which extent the IDH1 and IDH2 status may influence the classification and subsequent treatment of diffuse gliomas. For example, one may address interesting questions like (1) should IDH1 and IDH2 wild-type anaplastic gliomas be biologically considered as glioblastomas that consequently would require more aggressive treatment when compared with IDH1 or IDH2 mutant anaplastic gliomas, or (2) should the rare, prognostically favorable IDH1 mutant primary glioblastoma be regarded as a separate entity distinct from the “ordinary” IDH1 wild-type primary glioblastoma.

The epidermal growth factor receptor gene (EGFR) at 7p12 is the most frequently amplified and overexpressed gene in primary glioblastomas, affecting approximately 40% of these tumors [81, 138]. EGFR rearrangements are also frequent, the most common variant being EGFRvIII consisting of an 801-bp in-frame deletion of exons 2–7 that results in a constitutively activated truncated receptor protein lacking the ligand-binding domain [118]. EGFRvIII represents about half of the rearrangements and is identified in 20–30% of unselected primary glioblastomas and 50–60% of the EGFR-amplified glioblastomas [38]. Identification of EGFR amplification and rearrangements, such as EGFRvIII, are highly indicative for high-grade malignancy and, therefore, may provide diagnostic as well as prognostic information [64, 81]. In fact, detection of EGFR amplification/EGFRvIII in anaplastic or low-grade gliomas strongly suggests that these tumors are more malignant than indicated by their histopathology and an unfavorable impact on the prognosis has been described for these patients [66, 67, 115]. In primary glioblastoma such prognostic association is less obvious [132] although other studies reported on EGFR/EGFRvIII aberrations as poor prognostic factor [27, 89, 97, 113].

Detection of EGFR aberrations also may be relevant from a therapeutic point of view as inhibition of the EGFR pathway bears the potential of restoring apoptosis, thereby increasing the sensitivity to adjuvant therapies. Increased EGFRvIII signaling was indeed associated with a generally poor response to radiation and chemotherapy [4, 20, 134]. However, a benefit resulting from the combined treatment by EGFR inhibition with standard therapies (temozolomide and radiation therapy) is disputed and as yet the clinical benefit of the use of EGFR inhibitors in glioblastomas has been rather disappointing, that is progression-free survival was not prolonged and only a small subset of individual patients responded [45, 101]. Attempts to identify additional biomarkers predictive of response to EGFR-related therapies suggested that tumors with EGFRvIII and intact PTEN [85] or with EGFR amplification (but not EGFRvIII) and low levels of phosphorylated AKT [44] were more likely to respond to the small molecule tyrosine kinase inhibitors erlotinib or gefitinib. The EGFRvIII mutant, as not being present in non-neoplastic tissues, also may serve as an attractive target for immunotherapy [79, 107]. Recent studies reported that the anti-EGFRvIII peptide vaccine CDX-110 increased progression-free and overall survival in EGFRvIII-positive glioblastoma patients when added to radiochemotherapy [49‐51]. Unfortunately, although an antibody specifically recognizing EGFRvIII has been developed for immunohistochemical application, existing patents currently prohibit the use of this antibody for clinical purposes [136, 140]. EGFRvIII analysis alternatively can be performed by reverse transcription-PCR analysis using primers located in exons 1 and 9 [79]. Furthermore, MLPA analysis allows for the detection of EGFR rearrangements by the simultaneous and semi-quantitative copy number analysis of multiple small DNA fragments encompassing different EGFR exons [64, 111], with available assays detecting EGFRvIII and different other types of rearrangements and being applicable to routinely processed formalin-fixed and paraffin-embedded tissue samples [64].

Aberrant activation of the BRAF proto-oncogene at 7q34, most commonly by gene duplication and fusion or less frequently by point mutation, has only recently been identified as the characteristic genetic aberration in pilocytic astrocytomas. BRAF gene alterations are detectable in 60–80% of pilocytic astrocytomas, but are infrequent in diffusely infiltrating astrocytic gliomas [3, 69, 70, 98]. Thus, testing for BRAF gene alterations might be helpful in the sometimes difficult differential diagnosis between pilocytic astrocytomas and low-grade diffuse astrocytomas [75]. Detection of the BRAF‐KIAA1549 fusion gene can be either accomplished by FISH analysis or specialized RT-PCR assays [69, 75] (Table 2). As mentioned above, low-grade diffuse astrocytoma—in contrast to pilocytic astrocytoma—would contain frequent mutations in the IDH1 gene and as such testing for both markers in combination could finally turn the scales for one of the two entities. However, IDH1 and IDH2 mutations are generally rare in pediatric astrocytomas, including the diffusely infiltrating tumors [2, 137], which implies that the significance of diagnostic testing for these mutations may be lower in pediatric glioma patients.

The frequent BRAF alterations in pilocytic astrocytomas may have additional clinical implications as a novel therapeutic target. Tumors with BRAF duplication or activating mutation show aberrant signaling via the BRAF pathway. In vitro studies revealed that both the stable silencing of BRAF through shRNA lentiviral transduction and pharmacological inhibition of MEK1/2, the immediate downstream phosphorylation target of BRAF, blocked proliferation and arrested growth of cultured glioma cells [98]. Thus, pharmacological inhibition of the MAPK pathway may serve as a potential treatment option in pediatric astrocytoma patients, as exemplified in a recent case report [106].

Ependymomas also occur relatively frequently in the pediatric age group. Histological grading of ependymomas is difficult and there appears to be a less-stringent association between tumor grade and prognosis in ependymoma when compared with astrocytoma patients. However, recent data suggest that in addition to age at diagnosis, gain of 1q and homozygous CDKN2A deletion are independent indicators of unfavorable prognosis, whereas gains of chromosomes 9, 15q, and 18 and loss of chromosome 6 are associated with excellent survival for pediatric and adult patients with intracranial ependymomas [76]. Based on these findings, the authors developed a molecular staging system comprising three genetic risk groups. Thus, the analysis of genetic markers in addition to established clinical and histopathologic variables may significantly improve outcome prediction of ependymoma patients and help to stratify patients into distinct risk groups.

There are two main trends in glioma research that have already yielded and probably will yield further molecular biomarkers of clinical impact: the use of large-scale profiling techniques and, mainly driven by the need to work cost-effectively and to increase sample numbers, the formation of large research networks, such as The Cancer Genome Atlas Research Network (TCGA) or, more recently, the International Cancer Genome Consortium (ICGC) [60].

The success of large scale profiling approaches is exemplified by the first detection of IDH1 mutations in a study that sequenced 20,661 genes in 22 human glioblastoma samples [95]. In addition, this study confirmed a set of glioblastoma candidate genes that mainly functioned within the p53, pRb1, and Pi3k/Pten signaling pathways [95]. A publication that was launched contemporaneously by the TCGA consortium on the integrative analysis of DNA copy number, gene expression, and DNA methylation profiling in 206 human glioblastomas similarly reported that the most important pathways that are aberrant in gliomas are the pRb1/Cdk4/cyclin D/Cdkn2A/B pathway, the p53 pathway, and the receptor tyrosin kinase/Ras/Pi3k pathway [18]. Each of these pathways was confirmed to be disrupted in more than three quarters of glioblastomas, meaning that in most tumors two or all three of these pathways were involved. Sequencing of glioblastomas for mutations in 601 selected genes additionally revealed three previously unrecognized mutations that occurred with significant frequency, namely NF1 gene mutations in 14%, ERBB2 gene mutations in 8% and PIK3R1 mutations in nearly 10% of glioblastomas [18]. These novel findings may well have impact on future treatment strategies. As an example, PIK3R1 encodes the regulatory protein p85a subunit of Pi3k and the response to Pi3k inhibitors may depend on whether the tumors bear mutations in this specific gene or not.

Except for the identification of novel individual gene alterations, the signatures produced by high-throughput profiling techniques themselves might convey clinically relevant information. In this regard, it was shown that gene expression-based classification of morphologically ambiguous high-grade gliomas correlates better with prognosis than the histological classification [92]. Furthermore, molecular classification of gliomas on the basis of genomic profiles obtained by array-CGH closely parallels histological classification and is able to distinguish, with few exceptions, between different astrocytoma grades, as well as between primary and secondary glioblastomas [105]. Phillips et al. [99] first reported on three prognostic subclasses of high-grade astrocytomas, namely the proneural (PN), proliferative (Prolif) and mesenchymal (Mes) tumor subclasses that resemble distinct stages in neurogenesis. The proneural tumor signature displayed neuronal lineage markers and was associated with longer survival, while the proliferative and mesenchymal tumor signatures were both linked to shorter survival. Upon recurrence, malignant gliomas frequently shift toward the mesenchymal subclass. A recent study from the TCGA consortium reported on a robust gene expression-based molecular classification of glioblastomas into proneural, neural, classical, and mesenchymal subtypes [126]. Furthermore, it was found that genetic aberrations and expression of EGFR, NF1, and PDGFRA/IDH1 each define the classical, mesenchymal, and proneural subtypes, respectively. Response to aggressive therapy was found to differ by subtype, with the greatest benefit being observed in the classical subtype and no benefit in the proneural subtype. Interestingly, promoter DNA methylation profiling in 272 TCGA glioblastomas revealed that a subset of patients had concerted hypermethylation at a large number of loci, indicating the existence of a glioma-CpG island methylator phenotype (G-CIMP) [91]. Further investigations showed that G-CIMP tumors were more common among low-grade gliomas, displayed a proneural expression signature, frequently carried IDH1 mutations and were associated with significantly longer survival. Other authors also found a tight association between the IDH1 mutation status and gene expression profiles, suggesting two major pathomechanisms in diffuse astrocytic gliomas characterized either by IDH1 mutation and a proneural expression profile, found mostly in diffuse and anaplastic astrocytomas as well as secondary glioblastomas, or by lack of IDH1 mutation and a mesenchymal/proliferative expression profile [120]. In addition, IDH1 mutant glioblastomas have been reported to be frequently accompanied by telomerase-independent alternative lengthening of telomeres (ALTs), suggesting that such ALT + tumors belong to the less aggressive, ‘proneural’ glioblastomas [83].

Although these novel glioma signatures currently incur a lot of attention, their routine assessment by means of large-scale mRNA expression profiling is not suitable due to the limited availability and high costs of this approach. However, a recent study reported that immunhistochemical expression analysis of a nine gene signature, which is applicable to routinely processed tissue samples, may be sufficient to predict glioblastoma outcome [25]. Nevertheless, it remains to be proven that any of these prognostic gene signatures yields clinically relevant data beyond the information provided by the analysis of IDH1 mutation, MGMT promoter methylation and 1p/19q deletion.

A different approach by which large-scale profiling techniques may add in the identification of novel glioma biomarkers is the investigation of defined glioma cell subpopulations. Malignant gliomas are highly heterogeneous and their conventional light-microscopic diagnosis is based on the recognition of certain histological features, such as pathological vessel formation, presence/absence of necroses or an infiltrative growth pattern [81]. Bearing in mind that all these microscopic appearances are equivalents of defined biologic processes, it appears intriguing to dissect such tumor cell subpopulations and to assess their molecular signatures with two closely related goals: first, to foster understanding of the basic processes underlying glioma biology and secondly, to use these “subprofiles” to identify innovative biomarkers that are of relevance for the tumor’s clinical appraisal as a whole. In this regard, Dong et al. assessed the molecular profiles of perinecrotic palisades in comparison to non-palisading tumor cells distant from necrosis [29]. In conjunction with preceding studies, the authors found that the genes most commonly differentially expressed in these palisades conveyed response to hypoxic environmental conditions [10, 11]. Interestingly, a set of five RNAs (POFUT2, PTDSR, PLOD2, ATF5, and HK2) were not only differentially expressed in three micro-dissected glioblastomas, but also provided prognostic information in an independent set of glioblastoma patients. Thus, it appears feasible to derive tissue biomarkers that provide ancillary prognostic and predictive information from the study of defined subpopulations of tumor cells. Of course, this approach is not restricted to pseudopalisading tumor cells, but may be extended to other glioma cell subpopulations, e.g. tumor cells selected for their invasive or stemness properties [6, 28, 54].

Large-scale profiling approaches have also advanced the understanding of the molecular mechanisms that underlie treatment resistance in high-grade gliomas. Investigating gliomas that had recurred after treatment with alkylating agents, Hunter et al. identified somatic mutations of the mismatch repair gene MSH6 in a large-scale sequencing approach of the functional domains of 518 protein kinases [57]. These findings were further evaluated in an independent panel of 46 clinically well-documented glioblastomas [15]. Indeed, in recurrent glioblastomas, the rate of MSH6 mutations was significantly increased adding further evidence to a potential causal link between MSH6 deficiency and treatment resistance.

Interestingly, a subsequent publication reported that MSH6 mutations arise in glioblastomas during temozolomide therapy and mediate temozolomide resistance [139]. In vitro modeling through exposure of a MSH6 wild-type glioblastoma line to temozolomide resulted in resistant clones with one clone showing an MSH6 mutation, namely Thr(1219)Ile, which had also been noted in two treated glioblastomas of the TCGA cohort [18]. Moreover, knockdown of MSH6 in the glioblastoma cell line U251 increased resistance to temozolomide cytotoxicity and reconstitution restored cytotoxicity in MSH6-null glioma cells. These findings indicate that MSH6 mutations and/or mutations in other DNA mismatch repair genes are selected in glioblastomas during temozolomide therapy and that patients who initially responded to a frontline therapy, i.e., particularly patients with MGMT-hypermethylated tumors, may develop treatment resistance by acquiring a hypermutator phenotype involving frequent mutations in DNA mismatch repair genes. As a perspective, combination of alkylating chemotherapy with molecular strategies targeting DNA mismatch repair-deficient cells may help in preventing or minimizing treatment resistance of gliomas.

Although undoubtedly representing a relevant novel discovery, MSH6 mutations may be just the tip of an iceberg of molecular changes that are associated with treatment response. Furthermore, these mutations appear not be linked to high-level microsatellite instability in gliomas [82]. Other studies have reported that alteration of the base excision repair pathway may sensitize glioma cells to temozolomide treatment and suggested inhibition of poly(ADP-ribose)polymerase as a promising therapeutic approach [21, 84, 119]. Furthermore, large-scale heterochromatin reorganization has been observed in glioma cells following treatment with temozolomide and carmustine, suggesting that treatment efficacy may implicate a first event characterized by changes in heterochromatin organization and, conversely, treatment failure may be associated with the aberrant euchromatinization of novel, yet to be identified chemotherapy resistance genes [94]. Another study aiming at identifying molecular profiles specific of treatment resistance to temozolomide identified a “glioma stem cell” or “self-renewal” expression signature as a predictor of poor survival [89]. This signature proved an independent prognostic factor also in multivariate analyses adjusted for the MGMT promoter methylation status and contained HOX genes as well as the putative glioma stem cell marker prominin-1 (CD133). Thus, in access to the identification of individual candidate genes like MSH6, large-scale profiling approaches might help to uncover more complex molecular profiles associated with treatment resistance. These profiles, although still preliminary, may help in identifying “pathways of therapy failure” that eventually could be specifically targeted.

Molecular and translational glioma research has significantly advanced the understanding of glioma pathogenesis and identified a number of diagnostic, prognostic and/or predictive molecular markers that currently are on their way into clinical application. In fact, the antibody against the IDH1 R132H mutation is already used in many neuropathology laboratories, e.g. as a useful diagnostic marker for the differential diagnosis of diffusely infiltrating gliomas versus reactive astrogliosis [25]. Furthermore, both 1p/19q deletion and MGMT promoter methylation are presently being used to stratify patients into different clinical trials, each testing for the efficacy of different drugs or administration schemes in comparison to the respective standard protocol. In case that these studies will proof successful, it is to be foreseen that molecular assessment of the relevant markers will have to be implemented into the routine diagnostic setting outside of clinical trials. Clear cut-off levels for each molecular assay have to be developed and appropriate quality measures have to be established to ensure comparable sensitivity and specificity of molecular test results across different laboratories. Histological control of the tissue specimens used for molecular testing also is an important issue that requires an experienced neuropathologist to avoid false-negative test results due to inappropriate samples with too low tumor cell content.

In the near future, novel insights into the pathogenesis of gliomas are to be expected from ongoing, large-scale collaborative profiling studies addressing the complexity of genetic, epigenetic, transcriptomic, and miRNA changes by high-resolution array-based techniques or deep sequencing approaches. It appears very likely that these studies will uncover novel molecular markers that may further refine the diagnostic assessment of gliomas. However, the future role of molecular diagnostics in neurooncology, in particular concerning the value of predictive markers will also depend on the development and availability of novel therapeutic alternatives to allow for more sophisticated patient-tailored treatment choices based on molecular profiles.