Medical Express

ISSN (print): 2318-8111

ISSN (online): 2358-0429

Issue: 4.5 - 6 Articles

Back to summary


How to cite


Galatro TF, Sola P, Moretti IF, Miura FK, Marie SMOSKN, Lerario AM. Correlation between molecular features and genetic subtypes of Glioblastoma: critical analysis in 109 cases. MEDICALEXPRESS 2017;4(5):M170504



Correlation between molecular features and genetic subtypes of Glioblastoma: critical analysis in 109 cases

Thais F Galatro1; Paula Sola1; Isabele F Moretti1; Flavio K Miura1; Sueli M Oba-ShinjoI, Suely KN Marie1,2; Antonio M Lerario3,4

1. Universidade de Sao Paulo, Faculdade de Medicina FMUSP, Department of Neurology, Laboratory of Molecular and Cellular Biology, LIM15, São Paulo, SP, Brazil
2. University of São Paulo, Center for Studies of Cellular and Molecular Therapy (NAP-NETCEM-NUCEL), São Paulo, SP, Brazil
3. University of Michigan, Division of Internal Medicine Metabolism, Endocrinology and Diabetes, Ann Arbor, MI, USA
4. Universidade de Sao Paulo, Faculdade de Medicina, Bioinformatics coordinator of SELA (NGS facility core), São Paulo, SP, BR


Received in July 28 2017.
First Review in September 6 2017.
Accepted in October 3 2017.


OBJECTIVE: Glioblastoma, the most common and lethal brain tumor, is also one of the most defying forms of malignancies in terms of treatment. Integrated genomic analysis has searched deeper into the molecular architecture of GBM, revealing a new sub-classification and promising precision in the care for patients with specific alterations.
METHOD: Here, we present the classification of a Brazilian glioblastoma cohort into its main molecular subtypes. Using a high-throughput DNA sequencing procedure, we have classified this cohort into proneural, classical and mesenchymal sub-types. Next, we tested the possible use of the overexpression of the EGFR and CHI3L1 genes, detected through immunohistochemistry, for the identification of the classical and mesenchymal subtypes, respectively.
RESULTS: Our results demonstrate that genetic identification of the glioblastoma subtypes is not possible using single targeted mutations alone, particularly in the case of the Mesenchymal subtype. We also show that it is not possible to single out the mesenchymal cases through CHI3L1 expression.
CONCLUSION: Our data indicate that the Mesenchymal subtype, the most malignant of the glioblastomas, needs further and more thorough research to be ensure adequate identification.

Keywords: glioblastoma, classification, DNA sequence analysis, CHI3L1, EGFR.


OBJETIVO: O glioblastoma (GBM), o tumor cerebral mais comum e mais letal, é também um dos tipos de tumores de mais difícil tratamento. Análises genômicas integradas têm contribuído para um melhor entendimento da arquitetura molecular dos GBMs, revelando uma nova subclassificação com a promessa de precisão no tratamento de pacientes com alterações específicas. Neste estudo, nós apresentamos a classificação de uma casuística brasileira de GBMs dentro dos principais subtipos do tumor.
MÉTODO: Usando sequenciamento de DNA em larga escala, foi possível classificar os tumores em proneural, clássico e mesenquimal. Em seguida, testamos o possível uso da hiperexpressão de EGFR e CHI3L1 para a identificação dos subtipos clássico e mesenquimal, respectivamente.
RESULTADOS: Nossos resultados deixam claro que a identificação genética dos subtipos moleculares de GBM não é possível utilizando-se apenas um único tipo de mutação, em particular nos casos de GBMs mesenquimais. Da mesma forma, não é possível distinguir os casos mesenquimais apenas com a expressão de CHI3L1.
CONCLUSÃO: Nossos dados indicam que o subtipo mesenquimal, o mais maligno dos GBMs, necessita de uma análise mais aprofundada para sua identificação.

Palavras-chave: glioblastoma, classificação, análise de sequência de DNA, CHI3L1, EGFR



Glioblastoma multiforme (GBM) is the most common and most malignant form of brain tumor in adults. Glioblastomas belong to the glioma group of tumors. The World Health Organization1 classifies gliomas according to their resemblance to their cell of origin, along with histological and molecular features. Glioblastomas, the most prevalent within the glioma group, are extremely aggressive grade IV tumors, exhibiting high mitotic rates, micro-vascular proliferation and necrosis; they also present the poorest prognosis, with a median survival of 15 months from the time of diagnosis.2 Due to their invasive nature, complete surgical resection is very difficult to achieve. The presence of residual tumor cells results in recurrence and malignant progression, albeit at different intervals.

Glioblastomas can be further divided into two subgroups: primary GBMs, which arise de novo, and secondary GBMs, which result from the progression of a lower grade tumor.3 These two clinical forms of GBM have different and extensively characterized molecular features.3 The past decade has seen the rise of high-throughput sequencing techniques that provide in-depth knowledge of molecular alterations in tumor cells. GBM has been one of the most molecularly profiled tumors by several groups, including The Cancer Genome Atlas (TCGA) Networks.4-10 These studies have singled out specific determinant mutations within the newly identified subtypes: proneural, neural, classical and mesenchymal. Proneural tumors present alterations in the IDH1, PDGRFA and TP53 genes; classical tumors display mainly the EGFR mutation/amplification and the presence of the EGFRvIII oncogenic variant; mesenchymal tumors show RB1 and NF1 mutations as their main features. The neural subtype presents overexpression of neural markers, but display no specific genetic alterations. Patient overall survival was also correlated with GBM molecular subtypes, with the mesenchymal subtype presenting the worst prognosis. However, the reproducibility, clinical relevance, and functional basis of these subclasses remain to be established.

In this study, we have developed a NGS-based gene panel to detect and analyze several genes commonly mutated in GBM. We classified a series of Brazilian GBM cases based on the somatic mutation signatures previously established for the three main subtypes: proneural, classical and mesenchymal. We validated our genetic findings through orthogonal methods and immunohistochemistry. In this multi-institution cohort of Brazilian patients, GBM subtype distribution and associated patient outcomes corroborate previously published results and further validates the clinical relevance of GBM subtyping. NGS-based approaches for GBM classification are fast, reliable, and therefore, may have value as a diagnostic tool that may add to the clinical decision-making process.



Included patients and ethical statement. One hundred and nine GBM samples were obtained during therapeutic surgery of patients treated by the Neurosurgery Group of the Department of Neurology at Hospital das Clínicas at the School of Medicine of the University of São Paulo, in the period of 2000 to 2014. Written informed consent was obtained from all patients according to the ethical guidelines approved by the institutional Ethics Committee. (case # 0599/10). Complete patient information, along with the current WHO grading system1 and clinical findings are presented in Supplemental Table 1, annexed to this article.



GBM sampling and diagnosis. GBM diagnosis was confirmed by neuropathologists from the Division of Pathological Anatomy of the same institution, according to the WHO grading system available at the time.11 Samples were macrodissected and immediately snap-frozen in liquid nitrogen upon surgical removal. A 4µm-thick cryosection of each sample was analyzed under a light microscope after hematoxylin-eosin staining for assessment of cellular debris as well as necrotic and non-neoplastic areas; this was followed by removal from the frozen block by microdissection prior to DNA extractions.

Targeted gene panel sequencing. Based on previous literature assessing GBM molecular sub classification,6,8,9 40 genes were targeted for capture and deep sequencing. This customized panel included genes such as NF1, RB1, EGFR, TP53, PTEN, IDH1 and PDGFRA, whose alterations have all been previously associated with the three main GBM subtypes: proneural, classical and mesenchymal. Using the SureDesign tool (Agilent Technologies Inc., USA), the targeted RNA capture enrichment baits were designed to include coding exon regions and 50 bp from both the 3’ end and 5’ end of the flanking intronic sequence. The purpose was to incorporate possible splicing site mutations in our analysis. A total of 973 regions from the 40 genes were targeted for a final capture size of 2.79 Mega base pairs (Mb).

A target-enrichment DNA library was constructed using the Agilent SureSelect XT Target Enrichment Kit (Agilent Technologies Inc.), following the recommended protocol. Two hundred nanograms of tumor DNA was sheared using a E220 focused-ultrasonicator (Covaris, USA) to generate DNA fragments with a mean peak around 150bp. Indexed and adaptor-ligated libraries were multiplexed and paired-end sequenced on Illumina NextSeq 500 platform (Illumina, USA). An additional library was built from peripheral blood DNA from all the patients pooled in equal amounts.

Bioinformatics analysis. Sequencing data was generated as 150-bp paired-end reads using the Illumina Next-Seq platform. Raw data was aligned to the hg38 assembly of the human genome using the BWA mapping software.12 Aligned reads were coordinate-sorted with the bamsort tool from Biobambam2.13 Variant calling was performed simultaneously in all the tumor samples with the reference genome–free algorithm Platypus,14 and in the germline pool with freebayes, with the parameters properly set to call pooled data (pooled-discrete and ploidy parameter set to 132, which corresponds to the number of alleles present in the pool). The resulting VCFs file were annotated with SnpEff and SnpSIFT.15

Tissue microarray (TMA) construction and Immunohistochemistry. Two representative areas of each tumor were chosen by neuropathologists and marked both on HE sections and on the original paraffin block. Each of the 0.6 mm-diameter three cores of tumor tissue was extracted from the marked area of each donor block using an arraying machine (MTA-1, Beecher Instruments Inc., USA). The cores were inserted into a TMA recipient block in predetermined sites. Sections of 3µm-thickness were cut from the TMA block. A representative TMA section was initially stained with HE to assess the suitability of each core, and all other sections were paraffin coated and stored at -20°C until use.

For immunohistochemical detection, serial TMA sections were deparaffinized, rehydrated, treated for endogenous peroxidase blocking and subjected to antigen retrieval. Slides were immersed in 10 mM citrate buffer, pH 6.0 and incubated at 122°C for 3 min using an electric pressure cooker (BioCare Medical, USA). Specimens were then blocked and further incubated with anti-human CHI3L1 (mouse monoclonal, clone AT4A3; Abcam, United Kingdom) and with anti-human EGFR (mouse monoclonal, clone 31G7; Thermo Fisher Scientific, USA) at 16-20°C for 16 hours. Development of the reaction was performed with a commercial kit (Novolink; Novocastra, United Kingdom) at room temperature, using diaminobenzidine and Harris hematoxylin for nuclear staining. Optimization using positive controls suggested by the manufacturer of each antibody was performed in order to obtain optimal dilution. Two observers (TFG and IFM) evaluated staining intensity of tissue sections independently. A semi- quantitative scoring system considering both intensity of staining and the respective percentage of stained cells was applied as follows: score 0: no cells stained; score 1: 10–25% cells stained; score 2: 26–50% cells stained; score 3: 51–75% cells stained; score 4: 76–100% cells stained. Only cases with positive cell staining with scores > 2 were considered as positive. Digital photomicrographs of representative fields were captured and processed using PICASA 3 (Google, USA).

Statistical analysis. Survival data was assessed by comparing the Kaplan-Meier survival curves using the log-rank (Mantel Cox) test. A p-value < 0.05 was considered statistically significant. For this analysis, we included 77 cases with complete clinical follow-up. Calculations were performed using SPSS, version 23.0 (IBM, USA).



Molecular classification of GBM. We used the mutational profile to classify tumors into three major molecular subtypes: proneural, classical, and mesenchymal. We defined the proneural subtype by the presence of IDH1, PDGFRA, and TP53 mutations. The classical subtype was defined by EGFR alterations (amplification and mutations), PTEN mutations, and the presence of the oncogenic variant EGFRVIII. EGFR amplification results was based on a previous publication from our group.16 Finally, for the definition of the mesenchymal subtype, we used mutations in NF1 and RB1. For each patient, we used DNA from blood leukocytes for germline alteration subtraction. None of the found alteration were present in these germline controls. Aside from this control, we considered in this analysis a) known pathogenic mutations, such as the IDH1 R132H, b) variants that were not present in population databases (1000kg/ExAC), c) variants with predicted pathogenicity (nonsense or frameshift, or a missense mutation with high SIFT and polyphen scores), d) variants not present in our germline pool. All the tumors which did not carry a mutation in the genes described above and met the defined criteria were classified as "others". Given the lack of a specific genetic profile defining the neural subtype, we were not able to identify this subclass in our dataset. The complete list of genetic alteration can be found in Supplemental Table 2.



The targeted NGS analysis was performed on 109 GBM samples. A 20x coverage was achieved in ~97% of the targeted regions of the GBM samples. The mean coverage of the germline pool was 6063.67x.

Following the proposed criteria, we were able to classify 89 of the 109 analyzed GBM samples within the three major subtypes. Genetic alterations typical of the classical subtype were the most prevalent, accounting for 45.9% of the classified cases. Samples with alterations inherent to the mesenchymal subtype corresponded to 19.3% of the cases; alterations particular to the proneural subtype contributed 16.5% of the cases. This information is diagrammatically shown in Figure 1. A preview of survival rates are also shown in this figure.


Figure 1 - Distribution of the molecularly classified GBM cases. Depiction of the main genetic alterations of the cases molecularly classified in our GBM cohort. Classification was achieved using a customized gene panel, followed by exome sequencing (refer to text for details). The color gradient at the bottom depicts patient overall survival (blue indicates long; purple short survival as shown in the color gradient at the left); gray represents cases without a complete clinical follow-up. amp: amplification; mut: mutation.


The overall survival of GBM patients according to our molecular classification is shown in Figure 2: patients classified as mesenchymal GBM had a shorter overall survival in comparison to other subgroups (medians of 8, 9, and 13 months for the mesenchymal, classical, and proneural, respectively; log-rank p < 0.05).


Figure 2: Survival curve of the molecularly classified GBM cases. Kaplan-Meier of 77 GBM cases detailing median survival (log rank test, p=0.128) times according to our molecular classification. Patients alive at the time of the last clinical follow-up were censored from this analysis. m, months.


Correlation between molecular classification of GBM and immunohistochemical markers to differentiate between classical and mesenchymal subtypes. Concomitant with the molecular classification of GBMs, we also assessed the protein expression levels of CHI3L1 (YKL-40) and EGFR. CHI3L1 has been previously associated with the mesenchymal subtype of GBM, while EGFR overexpression is a known characteristic of the classical subtype. Immunohistochemistry staining was performed in a subset of our molecularly classified samples, comprising a cohort of 40 cases (8 mesenchymal, 9 proneural, 20 classical and 3 others). Figure 3 shows the results and Figure 4 depicts examples of stained tissue microarray sections. While only one of the mesenchymal samples showed negative levels of CHI3L1, cases positive for CHI3L1 (scoring above 2) comprised samples of the three subtypes (6 mesenchymal, 6 proneural and 10 classical cases were positive for CHI3L1). For EGFR, all proneural and 6 mesenchymal cases proved to be negative for this marker. Despite presenting genetic characteristics of classical subtypes, 4 of these cases did not present positivity for EGFR immunostaining. A complete listing of immunohistochemistry score can be found in Supplemental Table 2.


Figure 3: Scoring of immunohistochemical analysis for CHI3L1 and EGFR in GBM cases. Each GBM sample, previously classified with the molecular approach, was scored for the classical (EGFR, blue) and mesenchymal (CHI3L1, pink) immunohistochemical markers. Scoring ranged from 0-4, in which score 0: no cells stained; score 1: 10–25% cells stained, score 2: 26–50% cells stained; score 3: 51–75% cells stained; score 4: 76–100% cells stained. EGFR staining was prevalent in classical samples, although classical cases did not present significant staining of the marker, while other mesenchymal samples presented comparable levels of EGFR. CHI3L1 was present in high levels in cases from all three GBM subtypes.



Figure 4: Representative immunohistochemical sections for the evaluation of CHI3L1 and EGFR in molecularly classified GBM samples. Proneural (A-D), mesenchymal (E-H) and classical (I-L) GBM representative cases were stained for CHI3L1 and EGFR, whose overexpression has been associated with mesenchymal and classical subtypes, respectively. CHI3L1 expression was widespread among GBM samples, and not exclusive of mesenchymal cases. Both proneural (A) and classical (I) samples displayed high levels of this marker. Despite being negative in proneural cases (D), EGFR overexpression failed to differentiate between mesenchymal and classical cases, with both groups showing positive and negative cases. The reaction was performed in paraffin embedded tissue sections with a commercial polymer kit (Novolink; Novocastra, UK), using diaminobenzidine as developer and Harris hematoxylin for nuclear counterstaining. 400x magnification for all images. Scale bar 10µm.



The advent of next generation sequencing and large-scale molecular analysis over the past decade revealed that molecular alterations predict patients’ responses to treatment, overall survival and clinical outcome. A new light has been shed on the high level of GBM heterogeneity and new sub classifications have emerged. For GBM, several studies have singled out specific determinant mutations of the main, newly identified subtypes: proneural, classical and mesenchymal.4–10 Their potential to aid the choice of better treatment options is a step forward toward precision medicine.

We have performed a somatic mutation analysis in our GBM cohort utilizing a customized gene panel that comprises all the coding regions and splicing regions the most commonly mutated genes described in GBM. Our NGS panel allowed us to classify approximately ~85% of our samples according to molecular subtype using a single assay; this would have been very laborious and time-consuming using traditional methods, such as Sanger sequencing.17 The results we obtained, regarding the percentage of each GBM subtype, as well as their association with patient overall survival, is in accordance with what has been published so far.5–9

Since the first documentation of the molecular subclasses of GBMs (available by 2010), these have proved to be fundamental for new discoveries that may ultimately lead to better clinical approaches and precision medicine. For instance, Davis et al18 used the classification to compare genetic alterations in cultured brain-tumor initiating cells. This model system is important to study treatment options and GBM biology. Natash et al19 explored the role of oncostatin, a cytokine present in the microenvironment of GBM and its signaling pathway, which is associated with poor prognosis. They related such features to the aggressive nature in the mesenchymal GBM subtype. Chen and Xu20 have recently developed an algorithm that matches FDA approved drugs to the molecular subtype of GBMs, based on the genetic alterations of each subtype. Thus, it is important to specifically differentiate classical from mesenchymal GBMs: classical GBMs are amenable to treatment by specific tyrosine kinase inhibitors and adjuvant antibodies;21 in contrast, mesenchymal GBMs have only one potentially useful drug, currently introduced as a phase 1 clinical trial (clinical trial identifier: NCT02272270). Therefore, new practical proposals, more suitable for major diagnostic centers, have been looked for. The exemplified studies highlight the usefulness of GBM classification in improving current knowledge, biological understanding and diagnosis and treatment options for this tumor.

However, and because the basis for such classification is the genetic approach, requiring the implementation of molecular biology techniques, other more feasible methodologies have been searched for. A recent study revealed that MRI derived quantitative volumetric tumor phenotype features only moderately predict the molecular GBM subtypes, suggesting that subtypes do not generally alter the size composition of tumor areas.22 Therefore, the more doable proteomic IHC-based approach, still based on the reported molecular findings, has been the preferred technique for patient classification.

Both proneural and classical GBM subtypes present genetic point mutations or alterations (IDH1 R132H and EGFRvIII, respectively) that have made it possible to develop, over recent years, specific antibodies to achieve a cheaper and more suitable approach in major diagnostic centers.23,24 Nonetheless, the alteration of EGFRvIII was found in only 28.4% of our GBM cohort, while classical cases corresponded to 45.9% of all cases. For those extra cases, EGFR protein overexpression still excluded 4 out of 23 samples with a classical genetic subtype.

The mesenchymal subtype presents a high level of heterogeneity of alterations within the usually mutated genes of this subtype (NF1 and RB1). CHI3L1 (also known as YKL-40) is a secreted lectin, related to the regulation of hypoxia-induced injury response and has been previously associated with the mesenchymal GBM.6,8 Previous studies have proposed CHI3L1 as an additional marker to be used as a diagnosis tool through an immunohistochemical approach for the sub classification of GBMs.25,26 Yet, our results indicate that the higher expression of CHI3L1 is not an event exclusive of the mesenchymal cases, because high percentages of both proneural and classical samples displayed equivalent high levels.



Our results indicate the need for a genetic approach to further classify GBMs, so that a higher number of patients can profit from the precision provided by such classification in terms of treatment options.



The authors sincerely thank doctors and residents from the Discipline of Neurosurgery of the Department of Neurology at Hospital das Clínicas of School of Medicine, University of São Paulo, from the Neurosurgery group at Sírio Libanês Hospital and, particularly, Drs André M Bianco and Egmond Alves, for the therapeutic and diagnostic procedures of all patients included in this study. We also thank the physicians and technicians at the Division of Pathological Anatomy of the same institutions, particularly Dr Sergio Rosemberg, for the histological classification of tumor samples and tissue section processing. The authors thank São Paulo Research Foundation (FAPESP), grants #2013/07704- 3, #2013/06315-3, #2013/02162-8, #2014-50137-5 and #2016/15652-1, CAPES-NUFFIC (062/15), Conselho Nacional de Pesquisa (CNPq 305730/2015-0) and Faculdade de Medicina FMUSP for financial support.



AML, SKNM and SMOS conceived the study. AML performed the bioinformatics analysis. TFG and PS performed the experiments. SKNM and FKM followed-up the patients. SKMN and AML provided supervision. AML, SKNM and TFG wrote the manuscript. All authors contributed to the editing of the paper.



The authors declare no competing interests.



1. Louis DN, Perry A, Reifenberger G, von Deimling A, Figarella-Branger D, Cavenee WK, et al. The 2016 World Health Organization Classification of Tumors of the Central Nervous System: a summary. Acta Neuropathol. 2016;131(6):803–20. DOI:10.1007/s00401-016-1545-1.

2. Wen PY, Kesari S. Malignant gliomas in adults. N Engl J Med. 2008;359(5):492–507. DOI:10.1056/NEJMra0708126.

3. Ohgaki H, Kleihues P. The definition of primary and secondary glioblastoma. Clin Cancer Res Off J Am Assoc Cancer Res. 2013;19(4):764–72. DOI:10.1158/1078-0432.CCR-12-3002.

4. Cancer Genome Atlas Research Network, Brat DJ, Verhaak RGW, Aldape KD, Yung WKA, Salama SR, et al. Comprehensive, Integrative Genomic Analysis of Diffuse Lower-Grade Gliomas. N Engl J Med. 2015(26);372 (26):2481–98. DOI:10.1056/NEJMoa1402121.

5. Ceccarelli M, Barthel FP, Malta TM, Sabedot TS, Salama SR, Murray BA, et al. Molecular Profiling Reveals Biologically Discrete Subsets and Pathways of Progression in Diffuse Glioma. Cell. 2016;164(3):550–63. DOI:10.1016/j.cell.2015.12.028.

6. Verhaak RGW, Hoadley KA, Purdom E, Wang V, Qi Y, Wilkerson MD, et al. Integrated genomic analysis identifies clinically relevant subtypes of glioblastoma characterized by abnormalities in PDGFRA, IDH1, EGFR, and NF1. Cancer Cell. 2010;17(1):98–110. DOI:10.1016/j.ccr.2009.12.020.

7. Stieber D, Golebiewska A, Evers L, Lenkiewicz E, Brons NHC, Nicot N, et al. Glioblastomas are composed of genetically divergent clones with distinct tumourigenic potential and variable stem cell-associated phenotypes. Acta Neuropathol. 2014;127(2):203–19. DOI:10.1007/s00401-013-1196-4.

8. Phillips HS, Kharbanda S, Chen R, Forrest WF, Soriano RH, Wu TD, et al. Molecular subclasses of high-grade glioma predict prognosis, delineate a pattern of disease progression, and resemble stages in neurogenesis. Cancer Cell. 2006;9(3):157–73. DOI:10.1016/j.ccr.2006.02.019.

9. Brennan CW, Verhaak RGW, McKenna A, Campos B, Noushmehr H, Salama SR, et al. The somatic genomic landscape of glioblastoma. Cell. 2013;155(2):462–77. DOI:10.1016/j.cell.2013.09.034.

10. Parsons DW, Jones S, Zhang X, Lin JC-H, Leary RJ, Angenendt P, et al. An integrated genomic analysis of human glioblastoma multiforme. Science. 2008;321(5897):1807–12. DOI:10.1126/science.1164382.

11. Louis DN, Ohgaki H, Wiestler OD, Cavenee WK, Burger PC, Jouvet A, et al. The 2007 WHO classification of tumours of the central nervous system. Acta Neuropathol. 2007 Aug;114(2):97-109. DOI:10.1007/s00401-007-0243-4

12. Li H, Durbin R. Fast and accurate short read alignment with Burrows- Wheeler transform. Bioinforma Oxf Engl. 2009;25(14):1754–60. DOI:10.1093/bioinformatics/btp324.

13. Tischler G, Leonard S. biobambam: tools for read pair collation based algorithms on BAM files. Source Code Biol Med. 2014;9:13. DOI:10.1186/1751-0473-9-13.

14. Platypus, a reference genome–free algorithm that rapidly calls variants in clinical sequencing data. SciBX Sci-Bus Exch. 2014;7. DOI:10.1038/scibx.2014.936.

15. Cingolani P, Patel VM, Coon M, Nguyen T, Land SJ, Ruden DM, et al. Using Drosophila melanogaster as a Model for Genotoxic Chemical Mutational Studies with a New Program, SnpSift. Front Genet. 2012;3:35. DOI:10.3389/fgene.2012.00035.

16. Carvalho PO, Uno M, Oba-Shinjo SM, Rosemberg S, Wakamatsu A, da Silva CC, et al. Activation of EGFR signaling from pilocytic astrocytomas to glioblastomas. Int J Biol Markers. 2014;29(2):e120-128. DOI:10.5301/JBM.5000045.

17. Le Gallo M, Lozy F, Bell DW. Next-Generation Sequencing. Adv Exp Med Biol. 2017;943:119–48. DOI:10.1007/978-3-319-43139-0_5.

18. Davis B, Shen Y, Poon CC, Luchman HA, Stechishin OD, Pontifex CS, et al. Comparative genomic and genetic analysis of glioblastoma-derived brain tumor-initiating cells and their parent tumors. Neuro-Oncol. 2016;18(3):350– 60. DOI:10.1093/neuonc/nov143.

19. Natesh K, Bhosale D, Desai A, Chandrika G, Pujari R, Jagtap J, et al. Oncostatin-M differentially regulates mesenchymal and proneural signature genes in gliomas via STAT3 signaling. Neoplasia. 2015;17(2):225–37. DOI:10.1016/j.neo.2015.01.001.

20. Chen Y, Xu R. Drug repurposing for glioblastoma based on molecular subtypes. J Biomed Inform. 2016;64:131–8. DOI:10.1016/j.jbi.2016.09.019.

21. Padfield E, Ellis HP, Kurian KM. Current Therapeutic Advances Targeting EGFR and EGFRvIII in Glioblastoma. Front Oncol. 2015;5:5. DOI:10.3389/fonc.2015.00005.

22. Grossmann P, Gutman DA, Dunn WD, Holder CA, Aerts HJWL. Imaging- genomics reveals driving pathways of MRI derived volumetric tumor phenotype features in Glioblastoma. BMC Cancer. 2016;16:611. DOI:10.1186/s12885-016-2659-5.

23. Kato Y. Specific monoclonal antibodies against IDH1/2 mutations as diagnostic tools for gliomas. Brain Tumor Pathol. 2015;32(1):3–11. DOI:10.1007/s10014-014-0202-4.

24. Gupta P, Han S-Y, Holgado-Madruga M, Mitra SS, Li G, Nitta RT, et al. Development of an EGFRvIII specific recombinant antibody. BMC Biotechnol. 2010;10:72. DOI:10.1186/1472-6750-10-72.

25. Conroy S, Kruyt FAE, Joseph JV, Balasubramaniyan V, Bhat KP, Wagemakers M, et al. Subclassification of Newly Diagnosed Glioblastomas through an Immunohistochemical Approach. PLoS ONE 2014;9(12):e115687. DOI:10.1371/journal.pone.0115687.

26. Joseph JV, Conroy S, Pavlov K, Sontakke P, Tomar T, Eggens-Meijer E, et al. Hypoxia enhances migration and invasion in glioblastoma by promoting a mesenchymal shift mediated by the HIF1α–ZEB1 axis. Cancer Lett. 2015;359(1):107–16. DOI:10.1016/j.canlet.2015.01.010.