Glioblastoma TCGA Mesenchymal and IGS 23 Tumors are Identifiable by IHC and have an Immune-phenotype Indicating a Potential Benefit from Immunotherapy

For many years, the backbone of first-line therapy for glioblastoma has been maximal and safe surgery followed by radiotherapy with concomitant and adjuvant temozolomide, although recently the addition of tumor-treating fields has shown some additional survival benefit (1, 2). Prognosis is highly dependent on clinical and molecular factors, including IDH mutation status and MGMT promoter methylation (3, 4).

Molecular subtyping of glioblastoma may help determine prognosis and identify the optimal therapy for specific populations, as it has been shown in other diseases as breast cancer (5, 6). While several systems have been proposed for classifying glioblastoma (7–20), two classifications have been used in clinical trials to explore treatment sensitivities: The Cancer Genome Atlas (TCGA) and the Instrinsic Glioma Subtypes (IGS; refs. 21–25). The latest TCGA classification identified three subtypes—classical, proneural, and mesenchymal—and suggested that there was intratumoral heterogeneity because some cases activated more than one transcriptional signature (13). The TCGA analysis also confirmed a different pattern of activation of the immune microenvironment, especially in mesenchymal tumors, which had an increased presence of macrophages/microglia, M1 and M2 macrophages, and neutrophil gene signatures (13). The IGS classification (16, 26) segregates gliomas of all histologic grades into seven different clusters. Glioblastoma tumors are distributed mainly in clusters 18, 22, and 23 but can also be seen in other clusters, where prognosis is better. While the TCGA analyses were performed mainly in fresh-frozen specimens, the IGS used both fresh-frozen and formalin-fixed, paraffin-embedded (FFPE) tissues. Parallelisms have been drawn between the two classifications: for example, samples assigned to IGS clusters 9 and 17 are generally assigned to the TCGA proneural subtype, those in IGS 18 to the classical subtype, and those in IGS 23 to the mesenchymal subtype (23).

The identification of molecular subtypes by IHC has been explored in several studies, as it would allow these subtypes to be integrated in routine pathologic diagnosis (27–31) and be used in customizing treatment. In addition, recent findings on tumor–immune cell interaction and determinants of immunogenicity have highlighted the role of immunotherapies (32). The importance of gene fusions in solid tumors has been appreciated only recently, largely due to high-throughput technologies, such as RNA sequencing (RNA-seq), and has led to important therapeutic implications (33).

We have explored the TCGA and IGS glioblastoma molecular subtypes by analyzing FFPE samples from a uniformly treated population of newly diagnosed patients with glioblastoma. Our objectives were to characterize the subtypes by IHC and to explore molecular alterations, such as immune-phenotypical patterns and gene fusions, that could identify different subtypes as candidates for selected therapies.

This study was approved by the Institutional Review Board of the Hospital Germans Trias i Pujol and by the Ethics Committees of all the participating institutions and their Biobanks and was conducted in accordance with the ethical standards as laid down in the 1964 Declaration of Helsinki and its later amendments. All patients or their representatives gave their written informed consent to participate in this study.

From 2004 to 2014, the Glioma Catalonia (GLIOCAT) project (34) collected clinical data from 432 consecutive patients with glioblastoma from six institutions, all of whom had received the standard first-line treatment (surgery followed by radiotherapy with concurrent and adjuvant temozolomide). The pathologic diagnosis confirming glioblastoma was performed by several pathologists before patients were included in the project. Once selected for inclusion, each case was anonymized and given a number to identify it across all data. Presurgical and postsurgical MRIs were uploaded on a shared platform linking each case to their clinical and molecular data. Demographic and treatment characteristics were recorded. DNA was extracted for the study of MGMT methylation status if it had not previously been assessed. Tissue microarrays (TMA) were built for IHC analyses of all samples with sufficient tissue. RNA was extracted and assessed for quality before sending it for RNA-seq.

DNA was extracted from two 15-μm sections of FFPE tissue using the QIAamp DNA Mini Kit (QIAGEN GmbH), following the manufacturer's protocol. In cases with less than 50% of tumor cells, the tumor tissue was macrodissected manually. Then 500 ng of extracted DNA was subjected to bisulfite treatment using the EZ DNA Methylation-Gold Kit (Zymo Research Corporation). DNA methylation patterns in the CpG island of the MGMT gene were determined by methylation-specific PCR using primers specific for either methylated or modified nonmethylated DNA, as described previously (35).

Between two and four cores of each case were obtained, depending on the amount of tissue available. High necrotic areas were not selected. TMAs were constructed using a Veridiam Tissue Array Instrument (El Cajon), model VTA-100, using a 1-mm-diameter needle.

Consecutive 4-μm-thick sections were obtained and hematoxylin–eosin staining was done in sections 1, 20, and 40 to evaluate the persistence of the tumor at each spot.

Sixteen antibodies were preplanned, selected on the basis of the literature (IDH1-R132H, p53, Olig2, Ki67, c-Met, p16, CD44, D2-40, nestin, SOX2, YKL40, EGFR, PDL1, H3K27M, H3.3 G34R, and PTEN) and four (IDO1, TCIRG1, SHC1, RUNX3) were selected on the basis of RNA-seq results showing differential expression in the TCGA glioblastoma molecular subtypes: classical versus mesenchymal, classical versus proneural, and mesenchymal versus proneural. The differential expression analysis was performed as follows. Using the first 38 sequenced samples classified according to the Support Vector Machine (SVM) algorithm, we performed a supervised analysis with DESeq2 (RRID:SCR_015687; ref. 36; Supplementary Data File S1a–c; Fig. 1) and a further analysis between pairs (classical/mesenchymal, classical/proneural, and mesenchymal/proneural) for the top 50 differentially expressed genes (Supplementary Fig. S1A–S1C). Two of these 50 genes (TCIRG1 and SHC1) and an additional two not in the top 50 (RUNX3 and IDO1) were consistently related to the mesenchymal subtype. These four genes were selected for analysis by IHC (Supplementary Table S1). Primary antibody characteristics, Research Resource Identifiers (RRID), pretreatment conditions, dilutions, and scoring systems used are shown in Supplementary Table S2. Immunoreactivity was scored on different cell sublocations using a quantitative or semiquantitative scale. Cores with less than 50% of tumor were considered nonevaluable, and samples with more than 50% of nonevaluable cores were considered noninformative.

BenchmarK XT (Ventana Medical System Inc.), Bond Max (Leica Microsystems GmbH), and Dako (Dako Cytomation) automated immunolabeling systems were used for different antibodies.

The highest-quality RNA samples were further anonymized and sent to the Centro Nacional de Análisis Genómico (CNAG-CRG, Barcelona, Spain) for analysis by RNA-seq. Before performing RNA-seq analysis in the FFPE samples, we compared results in four cases of paired FFPE and fresh-frozen samples and confirmed the similarity of results in the two types of tissue samples (34).

The RNA extraction of FFPE tumor specimens was performed on five 15-μm-deep tissue cuts using the RNeasy FFPE Kit (Qiagen) according to the manufacturer's recommendations. RNA quantity and purity were measured with the NanoDrop ND-1000 spectrophotometer (Thermo Fisher Scientific) and the Qubit RNA HS Assay Kit (Invitrogen) used with the Qubit Fluorometer (Thermo Fisher Scientific). RNA was analyzed with the 2100 Bioanalyzer (Agilent) and quality was assessed by Eukaryote Total RNA Nano Bioanalyzer assay (Agilent) according to the “DV200” standard—the percentage of RNA fragments >200 nucleotides—as recommended by Illumina for FFPE samples.

FFPE samples were sequenced in different batches and we first selected the highest-quality RNAs from patients with the most complete clinical and radiological data. However, as many samples did not have RNA with the minimal quality requirements, we subsequently included cases without available MRIs and, finally, those without informative results from the MGMT methylation analyses.

The starting input material for the construction of the RNA library was DNA-free total RNA from FFPE samples, obtained using the KAPA RNA HyperPrep Kit with RiboErase (Roche Kapa Biosystems), according to the manufacturer's protocol. Briefly, the ribosomal RNA (rRNA) was depleted from 0.8 to 1 μg of total RNA using RiboErase. rRNA-depleted RNA samples were fragmented in the presence of divalent ions at 65°C for 1 minute due to the low quality of the initial total RNA. Following RNA fragmentation (first- and second-strand synthesis), the Illumina-compatible bar-coded adapters were ligated at a concentration of 1.5 μmol/L. Libraries were enriched with 15 cycles of PCR using KAPA HiFi HotStart ReadyMix (Kapa-Roche). The size and quality of the libraries were assessed in a High Sensitivity DNA Bioanalyzer assay (Agilent).

The libraries were sequenced on HiSeq2000 (Illumina) in paired-end mode with a read length of 2 × 76 bp using TruSeq SBS Kit v4. Each sample was sequenced in a fraction of a sequencing v4 flow cell lane, following the manufacturer's protocol. Image analysis, base calling, and quality scoring of the run were processed using Real Time Analysis (RTA 1.18.66.3) software and followed by the generation of FASTQ sequence files by CASAVA (RRID:SCR_001802).

RNA-seq paired-end reads were mapped against the human reference genome (GRCh38) using STAR (RRID:SCR_015899, version 2.5.2a; ref. 37) with ENCODE parameters for long RNA. Genes were quantified using RSEM (RRID:SCR_013027, version 1.2.28; ref. 38) and GENCODE (RRID:SCR_014966, version 24). Mapping quality metrics were calculated with GEMtools (http://gemtools.github.io/) and RSeQC (RRID:SCR_005275; ref. 39). Normalization and transformation of the expression data for plotting and analysis were done with the variance stabilizing transformation function of DESeq2 (RRID:SCR_015687; ref. 36). Principal component analysis was performed with the R prcomp function and the R ggplot2 library (RRID:SCR_014601).

The TCGA classification of glioblastoma molecular subtypes (11–13) was performed with the GlioVis portal (40), which contains over 6,500 tumor samples of approximately 50 expression datasets and facilitates classification with three algorithms: SVM, K-nearest neighbor (KNN), and single-sample gene set enrichment Analysis (ssGSEA). Glioma CpG island methylator phenotype (G-CIMP) status was predicted using the GlioVis SVM algorithm. The GlioVis glioblastoma TCGA cohort was selected as training dataset for both glioblastoma molecular subtype and G-CIMP predictions. The IGS classification of glioblastoma molecular subtypes was done with the R clusterRepro package (13, 16), using the centroids for IGS0, IGS9, IGS16, IGS17, IGS18, IGS22, and IGS23 subtypes as described by Gravendeel (16, 24, 25).

We applied immune profiling algorithms to our dataset to characterize in silico the cellular composition of the intratumoral immune infiltrates. We used two different approaches. First, we used the ImmunoPhenogram tool (https://tcia.at/tools/toolsMain) to calculate the z-scores for MHC, checkpoint (CP) molecules, effector cell (EC), and suppressor cell (SC; ref. 32). We then performed a cell-type deconvolution analysis with CIBERSORT (mode: absolute; ref. 41) with the 22 immune cell type signature (LM22) as reference dataset, to qualitatively enumerate the immune cell types found within the tumors and differences among the subtypes and patient populations.

We then screened for fusions that could characterize each subtype. For fusion detection, RNA-seq reads were mapped against the reference human genome with STAR-fusion version 0.7.0 with default options using GENCODE version 24. Previously described fusions were selected as final candidates to describe subtypes (42).

Categorical variables were compared with the χ² or Fisher exact test. Clinical data were merged with molecular and IHC results for statistical analyses. IHC values were presented as positive or negative, median values with 25%–75% percentiles or in quartiles if the median was near 0. Differences in protein expression among subtypes were compared with the Kruskal–Wallis test. The Spearman coefficient was used to estimate the correlation between gene and protein expression. Analyses were performed with R software v3.4.2 and SPSS v24 (IBM, RRID:SCR_002865).

The datasets generated during this study are available from the corresponding author on reasonable request. Molecular data underlying the findings described in the manuscript are fully available without restriction from the Bioproject Sequence Read Archive (http://www.ncbi.nlm.nih.gov/bioproject/PRJNA613395).

Tumor samples were obtained from 329 of the 432 patients with glioblastoma registered in the GLIOCAT project (34). After multiple RNA extractions from each sample, 357 RNA libraries were prepared and RNA-seq results were obtained for 151 tumor samples. Twenty seven of these samples were not included in subsequent analyses: 18 because the RNA was degraded and/or contaminated and nine because they were obtained at second surgeries at recurrence. The remaining 124 samples were included in further analyses. In addition, 17 TMAs were prepared and 280 samples were analyzed by IHC. One hundred and thirteen patients had both RNA-seq results and IHC data (Supplementary Table S3).

On the basis of the RNA-seq results, the tumor samples were classified into the three TCGA molecular subtypes (13), using three GlioVis algorithms SVM, KNN, and ssGSEA (Supplementary Data File S1d; ref. 40). There was agreement among all three algorithms on the subtype classification of 82 samples, which were considered a separate subgroup in further analyses (Supplementary Table S3). Samples were also classified according to the IGS molecular subtypes and information about TCGA subtype, IGS subtype, MGMT status, and presence or absence of fusions is shown in Fig. 2. Most tumors classified as IGS cluster 18 aligned with the TCGA classical subtype and those classified as IGS cluster 23 aligned with the TCGA mesenchymal subtype, especially for those samples that called the same subtype group by all three algorithms used. IGS 9 and IGS 17 aligned mostly with proneural tumors.

We tested 20 antibodies to analyze samples classified in the TCGA molecular subtypes by each of the three algorithms (Supplementary Table S1e). No patients had histone H3F3A mutations. Five of the preplanned antibodies (EGFR, SOX2, IDH1-R132H, Olig2, and Ki67) showed statistically significant different patterns of expression among subtypes, which confirmed parallelisms between the classical and the IGS 18 and the mesenchymal and the IGS 23 subtypes (Table 1). The differential expression study provided information about the genes that were expressed differentially in the molecular subtypes. Of the 50 genes with the greatest differential expression (Supplementary Data File S1a–c; Supplementary Fig. S1), the mesenchymal subtype showed the highest significant adjusted P values to be differentiated from the other subtypes. We selected four of the genes for IHC analysis (Supplementary Table S1). A correlation was observed between gene and protein expression for all the antibodies selected (Supplementary Fig. S2).

Two of these genes (TCIRG1 and SHC1) clearly identified the mesenchymal subtype, as classified by all three algorithms, as well as the IGS 23 cluster. In summary, IHC revealed that the TCGA classical subtype was identified by high expression of EGFR and low expression of PTEN, while the mesenchymal subtype and IGS 23 cluster were identified by low expression of SOX2 and high expression of SHC1 and TCIRG1. In addition, the proneural subtype was identified by frequent positive expression of IDH1 and high expression of Olig2 and Ki67. Most of these patterns of expression were shared by IGS 18 and IGS 23 tumors (Table 1; Fig. 3; Supplementary Fig. S3). Supplementary Fig. S4 shows representative staining for each of these antibodies.

We then explored differences in immune cell infiltration among the TCGA subtypes (Fig. 4) and found no significant differences in immunophenoscores (IPS). Nevertheless, when we broke down the scores into MHC, immune CP, ECs, and SCs, we found that mesenchymal, independently of the algorithm used, and IGS 23 tumors had a higher positive EC score and a more negative SC score than classical or proneural tumors. Mesenchymal and IGS 23 tumors also had a more negative CP score indicating that they express lower levels of immune CP molecules.

The cell-type deconvolution analysis showed that mesenchymal tumors were highly enriched in M2 macrophages, resting memory CD4⁺ T cells, and activated dendritic cells, and that IGS 23 tumors showed the same pattern (Fig. 5).

Among the 124 samples with informative RNA-seq results, 25 (20.1%) had gene fusions (Fig. 1). Fusions were more prevalent in the classical subtype: 12 samples with fusions (63.1%) were classified as classical by all three methods, while six were classified as classical by one or two methods. Fifteen of the 25 samples with fusions (60%) were classified as IGS 18. Four samples (3.2%) harbored more than one fusion, mainly EGFR with other genes.

The most frequent gene involved in fusions was EGFR, which occurred in 14 samples (11.2%), most of which were IGS 18 and/or TCGA classical tumors (Supplementary data file S1f).

We have examined the TCGA (13) and IGS (16) molecular subtypes in FFPE tumor samples from a uniformly treated cohort of 124 newly diagnosed patients with glioblastoma. We explored the feasibility of classifying tumors using IHC, which can easily be integrated into clinical practice, and we analyzed biological differences, such as immune-phenotyping patterns and gene fusions, that could identify different subtypes as candidates for selected therapies.

We found that samples were classified differently depending on the GlioVis algorithm (40) used for the TCGA subtypes. Nevertheless, we confirmed previous observations (23) of a high overlapping between the TCGA mesenchymal subtype and the IGS cluster 23 and between the TCGA classical subtype and the IGS cluster 18, which was especially evident for those samples that called only one subtype. We also identified an alignment of IGS clusters 9 and 17 with the proneural subtype.

Several previous groups have explored the use of IHC to predict glioblastoma subtypes (27–31) and have proposed different IHC panels for identifying the TCGA subtypes but none analyzed their cases based on results at the RNA level. By basing our IHC analyses on the differential expression observed in our RNA-seq data, we were able to select antibodies directed to specific genes.

Our IHC results showed that the molecular subtypes had different patterns of protein expression. Certain proteins were underexpressed or overexpressed in each particular subtype regardless of the algorithm used to classify the sample. In line with previous reports (27–29, 31), the classical subtype was consistently identified by IHC by overexpression of EGFR, as has previously been described at the molecular level (11, 12). The IGS cluster 18 showed a similar IHC pattern. In addition, gene fusions were predominant in these subtypes and most of the EGFR fusions were detected in tumors classified as classical by at least one of the algorithms. Proneural tumors were clearly identified by high expression of Olig2 and Ki67 and frequent IDH1-R132H mutations. Both IDH1 and Olig2 have previously been related to the proneural subtype (10, 11, 43).

Importantly, the differential expression analysis allowed us to identify some genes that were highly and differentially expressed in tumors mainly classified into the mesenchymal subtype by most of the algorithms. IHC targeting two of these genes, SHC1 and TCIRG1, was able to identify these tumors. The same IHC pattern was observed in tumors classified as IGS cluster 23. To the best of our knowledge, neither SHC1 nor TCIRG1 antibodies have previously been described as a marker of the mesenchymal subtype. SHC1 is a RAS-pathway protein that interacts with EGFR and receptors with protein tyrosine kinase activity; it has been related to angiogenesis and cytokine-mediated signaling pathways, among other functions (https://www.uniprot.org/uniprot/P29353). TCIRG1 is involved in several biological processes, including B-cell differentiation, cellular response to cytokine stimulus, Ig production and mediated immune response, inflammatory response, IFNγ secretion, memory T-cell activation and differentiation, and Th1 cell activation (https://www.uniprot.org/uniprot/A0A024R5E5; ref. 44).

In line with previous observations, immune-phenotyping also revealed differences according to glioblastoma molecular subtype. IPS were not uniformly different across the subtypes but we found that mesenchymal tumors—and the overlapping IGS cluster 23—are more infiltrated by immune ECs and contain a lower number of SCs. This pattern has been related to immunotherapy response in melanoma (32). Moreover, as previously reported (13), these tumors are enriched in M2 macrophages, resting memory CD4⁺ T cells, and activated dendritic cells. Therefore, this group of tumors, which are frequently classified as mesenchymal and IGS 23, seem to present a stronger host antitumor response, making them potentially ideal candidates for immunotherapy approaches, such as anti-M2 and/or dendritic vaccination (45–48). The association of increased levels of M2 macrophages in mesenchymal tumors has been described by Wang and colleagues (13), but in contrast with their findings, we could not confirm an increased expression of dendritic cells in the classical subtype.

Our study has some limitations, including its retrospective nature and the use of FFPE specimens for the TCGA subtyping. Importantly, however, this is the first study based on RNA-seq analysis of FFPE samples in a sizeable cohort (n = 124) of newly diagnosed, uniformly treated patients with glioblastoma with available paired specimens for IHC analysis. The subtypes differed in regard to protein expression, gene fusions, and immune-phenotype and could be identified by protein expression studied by IHC. Our results lead us to postulate therapeutic implications that warrant further investigation in larger series and clinical trials. Glioblastoma is a heterogeneous disease with different biologic and metabolic pathways and there is a subset of tumors that are frequently classified as mesenchymal or IGS cluster 23. These tumors may be identified with IHC in routine clinical practice by high TCIRG1 and SHC1 expression and low SOX2 expression and could well be optimal candidates for immunotherapy.

C. Carrato reports personal fees from AbbVie (advisory board) outside the submitted work. M. Martinez-Garcia reports personal fees and nonfinancial support from Roche and Pfizer, as well as personal fees from Celgene and Pierre Fabre outside the submitted work. T. Ribalta reports grants from Hospital Clinic of Barcelona during the conduct of the study; there are no other relationships/conditions/circumstances. J. Capellades reports personal fees from Eisai (honoraria as speaker), UCB Pharma (honoraria as speaker), and nonfinancial support from Medtronic (travel expenses) outside the submitted work. A.M. Muñoz-Marmol reports personal fees from AbbVie Inc (consulting), Roche (talk), Pfizer (talk), and Merck Sharp & Dohme España (talk), as well as other from Amgen (reactives for biomarker testing) and Novartis (reactives for biomarker testing) outside the submitted work. B. Bellosillo reports grants from Fundació LaMarató TV3 and personal fees from Qiagen during the conduct of the study; B. Bellosillo also reports personal fees from AstraZeneca, Biocartis, Merck-Serono, Novartis, Pfizer, and BMS, grants and personal fees from Hoffman-La Roche, and personal fees and nonfinancial support from Thermo Fisher Scientific outside the submitted work. N. de la Iglesia reports grants from Marató TV3 Foundation during the conduct of the study. C. Balana reports grants from Fundació LaMarató TV3 (project 665/C/2013) during the conduct of the study, as well as personal fees from Karyiopharm (advisory role), Abbvie (advisory role), and Celgene (advisory role) outside the submitted work. No potential conflicts of interest were disclosed by the other authors.

C. Carrato: Conceptualization, data curation, formal analysis, investigation, methodology, writing-original draft, writing-review and editing. F. Alameda: Conceptualization, data curation, formal analysis, supervision, funding acquisition, investigation, methodology, writing-original draft, project administration, writing-review and editing. A. Esteve-Codina: Software, formal analysis, investigation, methodology, writing-original draft, writing-review and editing. E. Pineda: Conceptualization, resources, data curation, formal analysis, investigation, methodology, writing-review and editing. O. Arpí: Formal analysis, investigation, methodology. M. Martinez-Garcia: Conceptualization, resources, investigation, methodology. M. Mallo: Resources, software, formal analysis, visualization, writing-review and editing. M. Gut: Resources, formal analysis, supervision, validation, investigation, methodology, writing-review and editing. R. Lopez-Martos: Conceptualization, formal analysis, investigation, writing-review and editing. S. Del Barco: Resources, data curation, investigation, writing-review and editing. T. Ribalta: Data curation, investigation, methodology, writing-review and editing. J. Capellades: Conceptualization, data curation, software, formal analysis, investigation, methodology, writing-review and editing. J. Puig: Conceptualization, data curation, formal analysis, investigation, writing-review and editing. O. Gallego: Resources, data curation, investigation, writing-review and editing. C. Mesia: Resources, data curation, investigation, writing-review and editing. A.M^a. Muñoz-Marmol: Methodology, writing-review and editing. I. Archilla: Data curation, formal analysis, investigation, writing-review and editing. M. Arumí: Data curation, investigation, writing-review and editing. J.M. Blanc: Data curation, formal analysis, validation, investigation, writing-review and editing. B. Bellosillo: Formal analysis, investigation, writing-review and editing. S. Menendez: Resources, investigation, methodology, writing-review and editing. A. Esteve: Software, formal analysis, writing-review and editing. S. Bagué: Formal analysis, investigation, writing-review and editing. A. Hernandez: Data curation, investigation, writing-review and editing. J. Craven-Bartle: Investigation, methodology, writing-review and editing. R. Fuentes: Data curation, investigation, writing-review and editing. N. Vidal: Data curation, investigation, writing-review and editing. I. Aldecoa: Data curation, investigation, writing-review and editing. N. de la Iglesia: Conceptualization, formal analysis, funding acquisition, methodology, writing-original draft, writing-review and editing. C. Balana: Conceptualization, resources, data curation, formal analysis, supervision, funding acquisition, investigation, visualization, methodology, writing-original draft, project administration, writing-review and editing.

This study was funded by the Fundació La Marató TV3 (665/C/2013; http://www.ccma.cat/tv3/marato/projectes-financats/2012/231/), PIs: Carmen Balana, Francesc Alameda, and Nuria de la Iglesia. Anna Esteve-Codina is funded by ISCIII/MINECO (PT17/0009/0019) and co-funded by FEDER. Silvia Menendez is supported by the Health Department of the Generalitat de Catalunya (PERIS SLT006/17/00040), Carmen Balana is funded by ISCIII/AES (INT19/00032).

The authors thank the biobanks of the participating institutions: Fundació Institut Mar d'Investigacions Mèdiques, Institut d'Investigacions Biomèdiques August Pi i Sunyer (IDIBAPS), Institut d'Investigació en Ciències de la Salut Germans Trias i Pujol (IGTP), Fundació Institut de Recerca de l'Hospital de la Santa Creu i St. Pau (IIB Sant Pau), and the Xarxa de Bancs de Tumors sponsored by Pla Director d'Oncologia de Catalunya (XBTC). We also thank Natalia Garcia-Balaña for her role as data manager and in coordinating the group. We are grateful for the collaboration of all the professionals of the GLIOCAT Group.

Other Clinical Investigators of the GLIOCAT Group that contributed to this study:

Marc Dabad¹, Avelina Tortosa², Teresa Pujol³, Laura Oleaga³, Cristian de Quintana-Schmidt⁴, Miquel Gil-Gil⁵, Izaskun Valduvieco⁶, Anna Martinez-Cardús⁷, Marta Domenech⁷, Anna Estival⁷, Salvador Villa⁸, Angels Camins⁹, Jordi Marruecos¹⁰, Sira Domenech¹¹

¹CNAG-CRG, Centre for Genomic Regulation, Barcelona Institute of Science and Technology, Barcelona, Spain.

²Laboratori de Quimio-resistència i Cáncer, School of Medicine and Health Sciences, University of Barcelona, Department of Fundamental Care and Medical-Surgical Nursing, Bellvitge Biomedical Research Institute (IDIBELL), Bellvitge, Spain.

³Radiology Department, Hospital Clínic, Barcelona, Spain.

⁴Neurosurgery Department, Hospital de Sant Pau, Barcelona, Spain.

⁵Neuro-Oncology Unit & Medical Oncology Department, Institut Catala d'Oncologia (ICO), Institut de Investigació Bellvitge (IDIBELL), L'Hospitalet, Barcelona, Spain.

⁶Radiation Therapy Department, Hospital Clínic, Barcelona, Spain.

⁷Institut Catala d'Oncologia (ICO) Badalona, Badalona Applied Research Group in Oncology (B-ARGO Group), Institut Investigació Germans Trias i Pujol (IGTP), Badalona, Spain.

⁸Radiation Oncology Department, Institut Catala d'Oncologia (ICO), Badalona, Spain.

⁹Radiology Department, Institut de Diagnòstic per la Imatge, Hospital de Bellvitge, L'Hospitalet de LLobregat, Barcelona, Spain.

¹⁰Radiation Oncology Department, Institut Catala d'Oncologia (ICO), Girona, Spain.

¹¹Radiology Department, Institut de Diagnòstic per la Imatge, Hospital Germas Trias i Pujol, Badalona, Spain.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.