**2.4. Epigenetics and molecular profiling**

The DNA repair enzyme O-methylguanine-DNA methyltransferase (MGMT) removes alkyl groups from the O6 position. Methylation of the MGMT promoter region results in decreased MGMT activity, which in turn results in decreased tumor resistance to alkylating agent therapy with TMZ, and is therefore a predictive molecular marker [36]. Usually, MGMT is determined in formalin-fixated paraffin-embedded specimens, and approximately 40% of all primary GBM carry a methylated promoter [18]. Less than 15% of gliosarcomas have a methylated phenotype [65]. In pediatric glioblastomas, approximately half of the tumors are methylated, mainly due to the association between H3F3A mutations and the methylated MGMT profile, but the prognostic and predictive role of MGMT methylation in children remains a matter of contro‐ versy [62, 68]. MGMT analysis is essential for almost all clinical studies and one of the most requested molecular analyses in routine neuropathology practice.

Unsupervised hierarchical clustering of GBM identified four tumor subclasses called proneu‐ ral (characterized by mostly IDH/Tp53 mutations, PDGFRα amplifications, PI3K pathway dysregulation), classic (large-scale EGFR amplification, PTEN, and CDKN2A loss), mesen‐ chymal (NF1, Tp53, and CDKN2A alterations), and neuronal (currently no specific genetic alterations). As expected from the IDH mutation data, the patients with a proneural tumor profile were younger and showed the best outcome [60]. The proneural group exhibited a robust hypermethylated glioma-CpG island methylator phenotype (G-CIMP) that is also present in most secondary GBM indicating a common gliomagenesis for IDH-mutated GBMs [85]. Glioma stratification according the IDH status therefore has widely replaced the previous clinic-based separation of primary and secondary GBM. Computational modeling to predict the temporal sequence of driver events during tumorigenesis indicates that most non-GCIMP mesenchymal GBMs arise from a PDGFA-driven proneural-like precursor though additional NF1 loss [86]. However, assignment with established glioblastoma subtype classifiers becomes difficult in cases with substantial tumor heterogeneity and may change further during patient treatment [73].

Glioblastoma intratumoral heterogeneity contributes to therapy resistance. This is exemplified in a whole genome sequencing report that showed not only extensive mutational and copynumber heterogeneity within the primary tumor but also uncovered the recurrence of a double-minute chromosome converging on the KIT/PDGFRA/PI3K/mTOR axis, superseding the IDH1 mutation in dominance in a mutually exclusive manner [87]. Despite targeted therapy with imatinib, the patient succumbed to progressive disease. Another good example is the recent discovery that EGFRvIII mutant cells are expressed only in a fraction of GBM cells of EGFR amplified tumors and enhance the proliferative activity of their neighboring EGFR wild-type tumor cells though cytokine secretion [88]. Because the majority of GBMs exhibit the activation of three or more RTKs, this highlights the need for the combined approach of several specific inhibitors for successful treatment.

Although there is increasing knowledge of divergent molecular alterations in histologically similarly appearing glioblastoma specimens, clinical decision-making on molecular alterations of glioblastoma subtypes is still limited. Notable exceptions include the determination of chromosomal allelic losses in 1p/19q in younger GBM patients, as such tumors respond far better to a combined Procarbazine-CCNU-Vincristine (PCV) therapy regimen [89]. The 1p/19q co-deletion is strongly associated with oligodendroglial tumor morphology and will become a diagnostic marker to be reassigned in future glioblastoma with this signature into the anaplastic oligodendroglioma group [26]. Up to 75% of co-deleted tumors also show either additional IDH1 or IDH2 mutations [90]. This combined molecular signature is so robust and remains visible in tumor recurrences, even in cases with increased intratumoral heterogeneity, and overlaps with "proneural" expression profile of the TCGA genomic landscape [91].

## **2.5. Conclusion**

mately 70% [81]. Alpha-thalassemia C-linked mental retardation (ATRX) mutations are found in approximately 30% of pediatric GBM and in 6% of adult glioblastoma. Interestingly, in pediatric GBM, ATRX mutations occur around a hotspot near the carboxy-terminal helicase,

ATRX and its binding partner DAXX (death-associated protein) belong to a complex with a role in regulating chromatin remodeling, nucleosome assembly and telomere maintenance. ATRX mutant tumors are associated with alternative lengthening of telomeres, the so-called ALT phenotype [78]. Because nuclear ATRX is diminished in tumors with the ALT phenotype, ATRX immunohistochemistry has become useful in identifying potential IDH mutants, H3F3A alterations or secondary GBM (usually showing ATRX loss and being mutually exclusive of LOH 1p/19q) [83]. Furthermore, retrospective analysis of ATRX in samples from the NOA-04

The DNA repair enzyme O-methylguanine-DNA methyltransferase (MGMT) removes alkyl groups from the O6 position. Methylation of the MGMT promoter region results in decreased MGMT activity, which in turn results in decreased tumor resistance to alkylating agent therapy with TMZ, and is therefore a predictive molecular marker [36]. Usually, MGMT is determined in formalin-fixated paraffin-embedded specimens, and approximately 40% of all primary GBM carry a methylated promoter [18]. Less than 15% of gliosarcomas have a methylated phenotype [65]. In pediatric glioblastomas, approximately half of the tumors are methylated, mainly due to the association between H3F3A mutations and the methylated MGMT profile, but the prognostic and predictive role of MGMT methylation in children remains a matter of contro‐ versy [62, 68]. MGMT analysis is essential for almost all clinical studies and one of the most

Unsupervised hierarchical clustering of GBM identified four tumor subclasses called proneu‐ ral (characterized by mostly IDH/Tp53 mutations, PDGFRα amplifications, PI3K pathway dysregulation), classic (large-scale EGFR amplification, PTEN, and CDKN2A loss), mesen‐ chymal (NF1, Tp53, and CDKN2A alterations), and neuronal (currently no specific genetic alterations). As expected from the IDH mutation data, the patients with a proneural tumor profile were younger and showed the best outcome [60]. The proneural group exhibited a robust hypermethylated glioma-CpG island methylator phenotype (G-CIMP) that is also present in most secondary GBM indicating a common gliomagenesis for IDH-mutated GBMs [85]. Glioma stratification according the IDH status therefore has widely replaced the previous clinic-based separation of primary and secondary GBM. Computational modeling to predict the temporal sequence of driver events during tumorigenesis indicates that most non-GCIMP mesenchymal GBMs arise from a PDGFA-driven proneural-like precursor though additional NF1 loss [86]. However, assignment with established glioblastoma subtype classifiers becomes difficult in cases with substantial tumor heterogeneity and may change further during patient

Glioblastoma intratumoral heterogeneity contributes to therapy resistance. This is exemplified in a whole genome sequencing report that showed not only extensive mutational and copy-

while they are widely distributed across the gene in adult GBM [82].

clinical trial showed a survival benefit of ATRX mutant tumors [84].

requested molecular analyses in routine neuropathology practice.

**2.4. Epigenetics and molecular profiling**

14 Neurooncology - Newer Developments

treatment [73].

Glioblastomas have an extensive variety of histological appearances and divergent immuno‐ histochemical and molecular profiles, making diagnosis somewhat difficult for those who are not familiar in working with brain tumors. The histological classification of diffuse gliomas based on the latest WHO grading scheme is a prerequisite to optimal decision-making regarding patient treatment. In addition to core features (microvascular proliferations, necrosis, and secondary structures of Scherer), the clinically relevant pattern and variants (gliosarcoma, giant cell glioblastoma, epithelioid, small cell GBM, and GBM with PNET component) should be clearly depicted in neuropathology reports. Immunohistochemistry and molecular biology have contributed to an improved classification and were shown in some cases to be of prognostic value. A panel of different antibodies is very helpful in securing the diagnosis and avoids potential differential diagnostic pitfalls. The advantages and limitations of the most commonly used antibodies, such as GFAP, WT1, MAP2, MIB-1, P53, IDH1R132H, and ATRX, in GBM have been outlined. The GBM subtype, patient's age, tumor location, and staining results subsequently guide a staged approach to therapeutically relevant molecular analysis, such as the 1p19q codeletion, MGMT promoter methylation, H3F3A screening, TERT promoter and IDH hotspot mutations. The classic concept of primary and secondary glioblas‐ tomas has been challenged by the discovery of clinically divergent molecular GBM cohorts, providing a good example of "convergent evolution showing a similar phenotype of geno‐ typically different tumor cells" [92]. The implementation of these additional molecular markers into routine diagnoses has already started, including the routine determination of MGMT gene promoter methylation status to guide therapy and the re-classification of tumors for appropriate treatment according to LOH1p/19q analysis, and it is expected to further evolve. The heterogenous landscape within and across GBMs underscores the difficulty in developing multimodal targeted therapies and is also a challenge to stratify patients for clinical trials. However, the recent identification of recurring driver mutations as illustrated here provides a rationale to identify tumor-specific peptides and antibody targets that may improve glioblastoma treatment.
