A tailored next-generation sequencing panel identified distinct subtypes of wildtype IDH and TERT promoter glioblastomas.

Central nervous system tumors are classified based on an integrated diagnosis combining histology and molecular characteristics, including IDH1/2 and H3-K27M mutations, as well as 1p/19q codeletion. Here, we aimed to develop and assess the feasibility of a glioma-tailored 48-gene next-generation sequencing (NGS) panel for integrated glioma diagnosis. We designed a glioma-tailored 48-gene NGS panel for detecting 1p/19q codeletion and mutations in IDH1/2, TP53, PTEN, PDGFRA, NF1, RB1, CDKN2A/B, CDK4, and the TERT promoter (TERTp). We analyzed 106 glioma patients (grade II: 19 cases, grade III: 23 cases, grade IV: 64 cases) using this system. The 1p/19q codeletion was detected precisely in oligodendroglial tumors using our NGS panel. In a cohort of 64 grade Ⅳ gliomas, we identified 56 IDH-wildtype glioblastomas. Within these IDH-wildtype glioblastomas, 33 samples (58.9%) showed a mutation in TERTp. Notably, PDGFRA mutations and their amplification were more commonly seen in TERTp-wildtype glioblastomas (43%) than in TERTp-mutant glioblastomas (6%) (P = .001). Hierarchical molecular classification of IDH-wildtype glioblastomas revealed 3 distinct groups of IDH-wildtype glioblastomas. One major cluster was characterized by mutations in PDGFRA, amplification of CDK4 and PDGFRA, homozygous deletion of CDKN2A/B, and absence of TERTp mutations. This cluster was significantly associated with older age (P = .021), higher Ki-67 score (P = .007), poor prognosis (P = .012), and a periventricular tumor location. We report the development of a glioma-tailored NGS panel for detecting 1p/19q codeletion and driver gene mutations on a single platform. Our panel identified distinct subtypes of IDH- and TERTp-wildtype glioblastomas with frequent PDGFRA alterations.

ataxia telangiectasia‐mutated serine/threonine kinase gene

ATRX chromatin remodeler gene

B‐Raf proto‐oncogene, serine/threonine kinase gene;

cyclin dependent kinase 4 gene

cyclin dependent kinase inhibitor 2A/B gene

comparative genomic hybridization

The Consortium to Inform Molecular and Practical Approaches to CNS Tumor Taxonomy

chromogenic in situ hybridization

central nervous system

copy number variation

epidermal growth factor receptor gene

formalin‐fixed and paraffin‐embedded

far upstream element binding protein 1 gene

glioblastoma

isocitrate dehydrogenase

mitogen‐activated protein kinase 4 gene

multiplex ligation‐dependent probe amplification

neurofibromin 1 gene

next‐generation sequencing

platelet derived growth factor receptor alpha gene

phosphatase and tensin homolog

retinoblastoma transcriptional corepressor 1 gene

telomerase reverse transcriptase gene

telomerase reverse transcriptase gene promoter

tumor protein p53 gene

World Health Organization

INTRODUCTION

The standard method for diagnosing CNS tumors has changed from a histology‐based approach to an integrated histology and molecular characteristics‐based approach following the implementation of the revised 2016 WHO classifications and their subsequent update by The Consortium to Inform Molecular and Practical Approaches to CNS Tumor Taxonomy (cIMPACT‐NOW). , , , , , , In this integrated diagnosis, genetic mutations in isocitrate dehydrogenase 1 or 2 (IDH1 or IDH2), codeletion of chromosomal arms 1p and 19q (1p/19q codeletion), and the H3‐K27M mutation must be evaluated. Other numerous genetic mutations are known to play an important role in gliomas such as mutations in PTEN and the epidermal growth factor receptor (EGFR) gene, which may serve as diagnostic, prognostic, and therapeutic biomarkers. , ,

Currently, diagnostic neuropathology laboratories test for selected biomarkers individually, including mutations in IDH1/2, the ATRX chromatin remodeler gene (ATRX), the telomerase reverse transcriptase gene (TERT), H3‐K27M, and the B‐Raf proto‐oncogene, serine/threonine kinase gene (BRAF), as well as 1p/19q codeletion. This involves various testing methods, including immunohistochemistry with mutation‐specific antibodies such as those against IDH1‐R132H, BRAF‐V600E, and H3‐K27M, , , conventional Sanger sequencing or pyrosequencing of tumor DNA for detection of mutations, FISH, CGH, chromogenic in situ hybridization (CISH), PCR‐based microsatellite analysis, real‐time comparative quantitative PCR; and multiplex ligation‐dependent probe amplification for detection of 1p/19q codeletion. , , , , Immunohistochemical (IHC) assays for IDH1‐R132H, BRAF‐V600E, and H3‐K27M are relatively sensitive and specific, but there is a non‐trivial rate of discordance. In addition, a significant proportion of patients harbor other mutations in these genes that are not detectable by IHC, potentially limiting the utility of IHC screening. In the case of 1p/19q codeletion, the most common diagnostic tool is FISH. However, this test can yield false‐positive results because commercially available FISH probes are typically designed to target the 1p36 and 19q13 regions and cannot discriminate between whole‐ and partial‐arm chromosome loss. High‐throughput array technologies such as CGH is a “whole‐genome” array that detects DNA CNVs at multiple sites simultaneously. However, it is cost‐prohibitive and therefore not suitable for routine molecular diagnoses in the clinic. To solve these problems, NGS is increasingly used in the diagnostic routine of leading oncology centers. NGS panels allow simultaneous assessment of several mutations and CNVs.

Several targeted NGS panels are available commercially; however, most are designed to identify important alterations in a broad spectrum of cancers but not glioma specifically. , Several customized gene panels have been established but they are unable to detect whole chromosome loss of 1p and 19q or do not cover enough driver genes. , , , , Therefore, we constructed a glioma‐tailored 48‐gene NGS panel for detecting 1p/19q codeletion and mutations as a routine molecular diagnostic tool for gliomas. Our NGS panel integrates molecular barcode technology into a single gene‐specific, primer‐based target enrichment process, with clear discrimination of false positives from true positives. This results in both greater sensitivity and greater accuracy in the detection of variants. Molecular barcoding technology aims to reduce the impact of enrichment and sequencing artifacts and has the potential to improve mutation detection accuracy. ,

This novel assay paves the way toward simultaneous detection of both allelic imbalances and mutations in small amounts of DNA retrieved from FFPE tissue samples for glioma subtype diagnostics. In this study, we describe the application of a glioma‐tailored 48‐gene NGS panel‐based analysis of 106 gliomas and evaluate the feasibility by routine molecular diagnostics. Our panel identified distinct subtypes of IDH‐ and TERTp‐wildtype glioblastomas with frequent PDGFRA alterations.

MATERIALS AND METHODS

CNS tumor samples

In total, 106 FFPE tumor tissue samples from 106 patients were selected from the CNS tumor tissue bank at Kagoshima University Hospital. In 5 patients, 2 samples were obtained at different time points and in 1 patient, 3 samples were obtained at different time points. The study was approved by the Institutional Review Board of Kagoshima University and it complied with the Helsinki Declaration. Informed consent was obtained from each patient. Resected tumors were fixed with phosphate‐buffered 10% formalin within 24 h and routinely processed for paraffin embedding, followed by sectioning for hematoxylin and eosin staining. All tumors were originally classified according to the WHO classification of 2016. The tumor series consisted of 9 diffuse grade II astrocytomas (A II); 14 grade III anaplastic astrocytomas (AA III); 58 grade IV glioblastomas (GBM IV), including 2 secondary glioblastomas; 10 grade II oligodendrogliomas (O II); 9 grade III anaplastic oligodendrogliomas (AO III); and 6 grade IV diffuse midline gliomas (DMG IV). Supporting Information Table S1 provides an overview of the relevant clinical and histological data of the 106 patients investigated. All tissues were histologically evaluated by board‐certified pathologists (MK, TH, and AT) to ensure an estimated tumor cell content of 30% or more, from which DNA was extracted. In 9 samples, the results were obtained from stereotactic biopsy specimens and in 97 samples the results were obtained in larger tumor resection specimens. In all patients, when analyzing CNVs, we sequenced leukocyte DNA for comparison against the matched tumor DNA.

DNA extraction and quantification

For DNA preparation from FFPE samples, we used the Maxwell 16 FFPE Tissue LEV DNA Purification kit (Promega) according to the manufacturer's instruction. Leukocyte DNA was extracted from samples from 106 patients with the DNeasy Blood & Tissue Kit (QIAGEN). Afterward, the concentration of DNA was measured using a Qubit 3.0 Fluorometer dsDNA BR assay kit (Life Technologies), DNA quality was monitored by QIAseq DNA QuantiMIZE kits (QIAGEN). The extracted DNA was diluted to a concentration of 5‐10 ng/μL as a template, and PCR was performed using the QIAseq DNA QuantiMIZE kits.

Design of the glioma‐tailored 48‐gene NGS panel

The customized gene panel for 48 glioma‐associated genes was created using the QIAseq (QIAGEN) targeted DNA panels (Figure 1). These 48 genes included the most commonly mutated genes in gliomas as well as genes previously suggested as diagnostically relevant molecular markers, such as IDH1/2, ATRX, the capicua transcriptional repressor gene (CIC), the TERT promoter (TERTp), and BRAF (Figure 1). For detection of 1p/19q codeletion, we analyzed the CNVs of chromosome 1p loci 9 genes, chromosome 1q loci 5 genes, chromosome 19p loci 5 genes, and chromosome 19q loci 5 genes (Figure 1). The final NGS panel consisted of 1954 primer pairs covering (99.95%) the coding sequences of 48 genes, and TERTp regions.

FIGURE 1

Glioma‐tailored 48‐gene next‐generation sequencing (NGS) panel design

Library preparation and NGS

Blood and FFPE DNA were treated to construct the NGS library using the QIAseq Custom Brain Tumor Panel (QIAGEN) (glioma‐tailored 48‐gene NGS panel), also applied to produce the NGS library. In total, 40 (blood) and 100‐200 (FFPE) ng DNA for the QIAseq Custom Brain Tumor Panel were used for library construction and were processed with a MiSeq sequencer (Illumina) after dilution with hybridization buffer to a DNA concentration of 20 pmol/L.

Analysis of NGS data for sequence variants and copy number changes

Amplicon sequences were aligned to the human reference genome GRCh37 (hg19) in the target region of the sequence. Data were analyzed using the QIAGEN Web Portal service (https://www.qiagen.com./us/shop/genes‐and‐oathways/data‐analysis‐center‐overview‐page/).

FISH

A PDGFRA amplification test was performed on a 4‐μm FFPE tissue section. The FISH probes used were bacterial artificial chromosome clones RP11‐231C18 (PDGFRA gene, 4q12) and control probes (4p11, CHR4‐10‐GR); (all from Empire Genomics). All FISH experimental procedures were performed according to the manufacturer's instructions. Preparations were counterstained with 4,6‐diamidino‐phenyl‐indole (DAPI).

Array‐CGH

Genomic DNA was extracted from whole blood samples using a DNeasy Blood & Tissue Kit (QIAGEN) and 1 μg of each sample was labeled using an Agilent SureTag DNA Labeling Kit (Agilent Technologies) and hybridized on an Agilent SurePrint G3 Human CGH Microarray 4 × 180 k (design ID 085723; Agilent Technologies), following the manufacturer's protocol.

Analysis of array data

Slides were scanned using an Agilent Autofocus Dynamic Scanner (Agilent Technologies) and quantified using Agilent's Feature Extraction software (v10.10.0.23). Quantitative data were loaded into CytoGenomics Software 5.0 (Agilent Technologies) and analyzed using the Aberration Detection Method 2 (ADM2) statistical algorithm at a threshold of 6.0 to identify genomic intervals with copy number changes. To reduce false‐positive calls, a filter was applied to define the minimum log2 ratio (0.25), and the minimum number of probes.

Statistical analyses

Unsupervised average linkage hierarchical clustering was applied to the NGS data obtained from the tumors based on Jaccard's matching coefficient to calculate distances. This analysis was performed using R open source statistical computing language (v3.5.3) and the integrated development environment RStudio (v0.99.484) as well as the R packages nmf (v0.20.6), mass (v7.3‐51.5) and stats (v3.2.2). Cluster analysis was performed using Euclidean distance and Ward.D2 linkage. Other statistical analyses were performed using a JMP Pro v13 software (SAS Institute). Differences were considered significant at P < .05.

RESULTS

Detection of 1p/19q codeletion

A codeletion of 1p/19q was identified when all of the following conditions were satisfied: a copy number loss of all genes on 1p (9 genes) and 19q (5 genes); (average CNVs of chromosome 1q loci 5 genes)/(average CNVs of chromosome 1p loci 9 genes) ≥2; and (average CNVs of chromosome 19p loci 5 genes)/(average CNVs of chromosome 19q loci 5 genes) ≥ 2. To validate this definition of 1p/19q codeletion, we used 6 oligodendroglial tumor samples and 2 glioblastoma (GBM) samples that had previously been tested by array‐CGH. Representative data of our 48‐gene NGS panels for detection of the 1p/19q codeletion are shown in Figure 2. No 1p/19q codeletions were observed in the GBM samples (100% specificity) and concordant positive results were obtained for all oligodendroglial samples (100% sensitivity) (Figures 2, S1 and S2). Thus, agreement was observed between both tests.

FIGURE 2

Detection of 1p/19q codeletion. A patient with anaplastic oligodendroglioma with IDH‐mutation and 1p/19q codeletion. For detection of 1p/19q codeletion, we analyzed copy number variations (CNVs) of chromosome 1p loci (9 genes), chromosome 1q loci (5 genes), chromosome 19p loci (5 genes), and chromosome 19q loci (5 genes). To validate 1p/19q codeletion detection, oligodendroglial tumor samples were tested by array‐comparative genomic hybridization (CGH)

Mutation analysis in all gliomas

In total, we included 106 samples from 106 patients. The results of the NGS panel for the 106 samples were organized following the 2016 WHO classification system. The clinical data for each case are listed in Table S1. Table S2 summarizes the mutations and CNVs in all gliomas.

In 106 tumors, the most commonly mutated genes were TERTp, the tumor protein p53 gene (TP53), IDH1, PTEN, the neurofibromin 1 gene (NF1), and ATRX (Figure 3A). Interestingly, 6.6% (7/106) of cases showed no mutations in any of the regions examined by the NGS assay. No mutations were detected in 5.2% (3/58) of GBMs, 21.4% (3/14) of anaplastic astrocytomas, and 11.1% (1/9) of diffuse astrocytomas. IDH1 mutations were identified in 33 of the 44 (75%) diffuse and anaplastic gliomas, including astrocytic and oligodendroglial tumors, as well as secondary GBMs. These mutations corresponded to IDH1‐R132H, and no IDH2 mutations were detected. TERTp mutations were most commonly detected in oligodendroglial tumors (OD II: 90%; AO III: 100%), followed by IDH‐wildtype GBMs (58.9%) (Table S2). Codeletion of 1p/19q was detected in all oligodendroglial tumors. Mutations of CIC and the far upstream element binding protein 1 gene (FUBP1) were detected in 68.4% and 57.9%, respectively, of the oligodendroglial tumors with 1p/19q codeletion. The most commonly mutated genes in GBMs with wildtype IDH were TERTp, TP53, PTEN, the retinoblastoma (RB) transcriptional corepressor 1 gene (RB1), NF1, and the platelet derived growth factor receptor alpha gene (PDGFRA) (Figure S3A). In astrocytomas (WHO grade II/III), the incidence of IDH1, TP53, and ATRX mutations was significantly higher, at 52.2%, 56.5%, and 47.8%, respectively. The majority of astrocytomas (WHO grade II/III) with an IDH1 mutation also showed a TP53 mutation (100%) and frequent ATRX mutations (83.3%). The incidence of TERTp and EGFR mutations in grade III astrocytomas was higher than those in grade II TERTp and PTEN mutations were more common in GBMs, than in grade II/III astrocytomas. However, there were fewer EGFR and ATRX mutations in GBMs than in grade II/III astrocytomas (Table S2). We identified mutations in the extracellular domain of EGFR in 100% (2/2) of GBMs. We also identified mutations in the extracellular domain and kinase domain of EGFR in 80% (4/5) and 20% (1/5) of grade II/III astrocytomas, respectively. In DMG cases, the most commonly mutated genes were TP53 and NF1, while TERTp mutations were not detected (Table S2).

FIGURE 3

Frequency of genetic alterations of all gliomas. Frequency of mutations (A), amplification (B), and loss (C) in each gene of glioma samples

Copy number analysis in all gliomas

All of the 106 cases (100%) showed evidence of CNVs in one of the NGS panel genes. The most common genes showing evidence of amplification were EGFR, PDGFRA, and the cyclin dependent kinase 4 gene (CDK4) (Figure 3B). The most common genes showing evidence of loss were PTEN, RB1, FUBP1, the cyclin dependent kinase inhibitor 2A/B gene (CDKN2A/B), the mitogen‐activated protein kinase 4 gene (MAP2K4), and the ataxia telangiectasia‐mutated (ATM) serine/threonine kinase gene (ATM) (Figure 3C). In IDH‐wildtype GBM cases, the most common genes showing evidence of amplification were EGFR, CDK4, and PDGFRA (Figure S3B), while the most common genes showing evidences of loss were PTEN, RB1, CDKN2A/B, ATM, MAP2K4, NF1 (Figure S3C). The representative mutual exclusivity was observed in the pairs of PDGFRA amplification/mutation and EGFR amplification/mutation. In contrast, only 4.3% and 0% of astrocytomas (WHO grades II/III) showed evidence of PDGFRA amplification or NF1 loss, respectively. EGFR and CDK4 amplification, and CDKN2A/B homozygous deletion occurred in grade III astrocytomas, but was not detected in any of the grade II astrocytomas. In DMG cases, the most common genes showing evidence of amplification were PDGFRA and FGFR, while the most common genes showing evidences of loss were RB1, MAP2K4, PTEN, ATM, and NF1 (Table S2).

Unsupervised hierarchical cluster analysis of all gliomas

To investigate the potential of the NGS panel data to molecularly classify tumors, we performed an unsupervised hierarchical cluster analysis taking into account sequence changes and CNVs detected in the 106 investigated gliomas. This analysis revealed 3 distinct groups of gliomas with mutations in IDH1, TERTp, CIC, FUBP1, PTEN, TP53, and ATRX, 1p/19q codeletion, PTEN and RB1 loss, CDKN2A/B homozygous deletion, and EGFR, CDK4, and PDGFRA amplification. One major cluster consisted of 61 primarily astrocytomas characterized predominantly by mutations of TERTp, PTEN, and ATRX, loss of PTEN and RB1, and amplification of EGFR. A second major cluster, including 26 primarily astrocytomas, was characterized by mutations in IDH1, ATRX, TP53, and PDGFRA, as well as amplification of CDK4 and PDGFRA. The third major cluster, including 19 primarily oligodendroglial tumors, was characterized by mutations in IDH1, TERTp, FUBP1, and CIC, as well as 1p/19q codeletion (Figure S4).

Genetic and clinical features of wildtype IDH GBM

We identified 56 IDH‐wildtype cases of GBM with available molecular data. Among this cohort, the average age of patients was 63.35 y old (range: 22‐88). Genetic alterations in TERTp were detected in 33 tumors (58.9%) and the remaining 23 patients (41.1%) had TERTp‐wildtype GBM. Importantly, the average age of patients with TERTp‐wildtype and TERTp‐mutant GBMs was not significantly different (62.87 vs 63.69 y, P = .426). We examined the genetic correlation between TERTp‐wildtype vs mutant tumors (Table 1). PDGFRA mutations and amplification were more common in TERTp‐wildtype GBMs (10/23, 43%) than in TERTp‐mutant GBMs (2/33, 6%). Conversely, EGFR mutations and amplification were more commonly seen in TERTp‐mutant GBMs (11/33, 33%) than in TERTp‐wildtype GBMs (3/23, 13%). We noted that 7/23 (30%) of the TERTp‐wildtype GBMs harbored CDK4 amplification, compared with 5/33 (15%) of TERTp‐mutant GBMs. Moreover, PTEN mutations and/or loss were detected in 29/33 (88%) of TERTp‐mutant GBMs, while only 13/23 (57%) of TERTp‐wildtype GBMs had PTEN mutation and/or loss (Table 1).

TABLE 1

Comparison of IDH‐wildtype glioblastomas according to TERT promoter mutation status

	TERTp‐wild (n = 23)	TERTp‐mutant (n = 33)	P‐value
MTOR	1 (4%)	0 (0%)	.179
JAK1	1 (4%)	0 (0%)	.179
FUBP1	1 (4%)	0 (0%)	.179
PDGFRA mutation and/or amplification	10 (43%)	2 (6%)	.001*
EGFR mutation and/or amplification	3 (13%)	11 (33%)	.076
BRAF	1 (4%)	0 (0%)	.179
CDKN2A + CDKN2B mutation and/or homozygous deletion	9 (39%)	14 (42%)	.805
PTEN mutation and/or loss	13 (57%)	29 (88%)	.008*
ATM	1 (4%)	0 (0%)	.179
CDK4 amplification	7 (30%)	5 (15%)	.173
MDM2 amplification	1 (4%)	2 (6%)	.777
RB1 mutation and/or loss	12 (52%)	19 (58%)	.689
TP53 mutation and/or loss	13 (57%)	14 (42%)	.298
NF1 mutation and/or loss	5 (22%)	10 (30%)	.473
ATRX loss	2 (9%)	6 (18%)	.306

Groups were compared by chi‐square (χ2) tests.

P < .05 was considered statistically significant.

Next, we performed an unsupervised hierarchical cluster analysis on the 56 IDH‐wildtype GBMs. This analysis revealed 3 distinct groups of IDH‐wildtype GBMs, with mutations in TERTp, PTEN, RB1, NF1, TP53, and PDGRFA, loss of PTEN, RB1, and NF1, homozygous deletion of CDKN2A/B, as well as EGFR, CDK4, and PDGFRA amplification. One major cluster (Group A) was characterized by mutations of TERTp and NF1, loss of PTEN and NF1, amplification of EGFR, and a lack of TP53 mutations. A second major cluster (Group B) was characterized by mutations in TERTp, PTEN, TP53, and RB1 as well as a lack of CDKN2A/B homozygous deletion. The third major cluster (Group C) was characterized by mutations in TP53 and PDGFRA, amplification of CDK4 and PDGFRA, homozygous deletion of CDKN2A/B, and a lack of TERTp and PTEN mutations (Figure 4).

FIGURE 4

Results of unsupervised hierarchical clustering analysis of the glioma‐tailored 48‐gene next‐generation sequencing (NGS) panel data obtained in 56 IDH‐wildtype glioblastomas

In addition, we compared clinical features among Groups A, B, and C (Table 2). The average Ki‐67 score of Group C was 45.17%, which was significantly higher than that of Group A (P = .007) but similar to that of Group B (Table 2). Although there was a trend for patients in Group C toward a periventricular tumor location, there was no statistical significance (P = .073). We did not detect any differences in gender and Karnofsky performance status. The average age of Group C was 71.4, meaning these patients were significantly older than that of Groups A and B (P = .021). Interestingly, when we excluded Group C, the average age of patients with TERTp‐wildtype GBMs was 48.56 y, which was significantly younger than that of patients with TERTp‐mutant GBMs (63.75 y, P = .009). The median time of overall survival was 65 mo for Group A, 13 mo for Group B, and 19 mo for Group C. The overall survival was significantly shorter in Groups B and C compared with that in Group A (P = .012) (Figure S5A). Moreover, the overall survival was not significantly different when we compared TERTp‐wildtype and TERTp‐mutant statuses (P = .298) (Figure S5B). However, when we excluded Group C from this analysis, the survival was significantly shorter for patients with TERTp‐mutant GBM, compared with that for patients with TERTp‐wildtype GBM (P = .042) (Figure S5C).

TABLE 2

Clinical features of IDH‐wildtype glioblastomas according to subtypes by unsupervised hierarchical cluster analysis

	Group A (n = 26)		Group B (n = 15)		Group C (n = 15)		P‐value
	Cases	Ratio (%)	Cases	Ratio (%)	Cases	Ratio (%)
Age	58.04 ± 15.81		64.53 ± 15.34		71.4 ± 10.26		.021*
Gender
Male	15	57.69	8	53.33	8	53.33	.948
Female	11	42.31	7	46.67	7	46.67
Location
Periventricular	8	30.77	4	26.67	10	66.67	.073
Subcortical	17	65.38	11	73.33	5	33.33
Infratentorial	1	3.85	0	0	0	0
Karnofsky performance status
0‐70	7	26.79	7	46.67	7	46.67	.309
80‐100	19	73.08	8	53.33	8	53.33
Ki‐67 (%)	30.08 ± 11.33		45.4 ± 19.69		45.17 ± 22.50		.007*
Representative genetic features
TERTp	Mutant		Mutant		Wild
PDGFRA	Intact		Intact		amp/mut
PTEN	Mutant		Mutant		Wild
TP53	Wild		Mutant		Mutant
CDKN2A/B	Homozygous deletion		Intact		Homozygous deletion

Groups were compared by chi‐square (χ2) tests.

P < .05 was considered statistically significant.

Validation of PDGFRA gene amplification by FISH

To validate PDGFRA gene amplifications performed by our NGS panel, we conducted FISH on 10 selective GBM cases, comprising 5 defined by NGS as showing PDGFRA amplification, and 5 defined as lacking PDGFRA amplification. FISH analysis showed that there was no PDGFRA amplification in the GBM samples defined by NGS as lacking PDGFRA amplification, and concordant positive results in GBM samples defined by NGS as showing PDGFRA amplification (Figure S6).

EGFR and PDGFRA alterations in large validation cohort

We retrieved molecular characteristics of the GBM cohort from a previous publication, and having excluded H3F3A, IDH1/2 and BRAF V600E‐mutant cases, we analyzed 468 cases conclusively diagnosed as IDH‐wildtype GBM using cBioPortal for Cancer Genomics (https://cbioportal.org). Genetic alterations in TERTp were detected in 89% of IDH‐wildtype GBM. Mutual exclusivity was observed in the pairs of PDGFRA alterations and EGFR alterations (Figure S7).

DISCUSSION

In this study, we constructed a glioma‐tailored 48‐gene NGS panel for detecting 1p/19q codeletion and driver gene mutations as a routine molecular diagnostic tool for gliomas in a single platform. Zacher et al and Na et al reported that 1p/19q codeletion could be detected by their NGS panels. , However, their panels did not include whole chromosome arms (1p and 19q) and did not allow for the distinction between whole and partial chromosomal loss. In this study, we integrated the loss of heterozygosity from the CNV analysis of genes on chromosome 1p (9 genes), 1q (5 genes), 19p (5 genes), and 19q (5 genes) to identify the complete codeletion of 1p/19q. The copy number loss of 1p/19q genes detected in NGS was compared with CGH and the results were concordant in cases of oligodendroglial tumors. We believe that our method accurately detects 1p and 19q whole chromosome arm deletion. For the detection of diagnostic DNA copy number changes, our glioma‐tailored 48‐gene NGS panel reliably revealed complete 1p/19q codeletion.

Half of the IDH‐wildtype grade II/III astrocytomas exhibited unfavorable genetic features such as alterations of EGFR, PTEN, RB1, and TERTp in our cohort study. Some of IDH‐mutant grade II/III astrocytomas exhibited homozygous deletion of CDKN2A/B or amplification of CDK4. Moreover, a recent report suggested that homozygous deletion of CDKN2A/B and amplification of PDGFRA and CDK4 are related to poor prognoses in IDH‐mutant astrocytic gliomas. , , Thus, our NGS panel offers feasible molecular stratifications for risk.

Recent reports have indicated that 70%‐80% of GBM genomes harbor either C228T or C250T mutations in the promoter region of TERT. , , , , The IDH‐wildtype GBM bearing wildtype TERTp is associated with prolonged overall survival compared with those carrying mutations in TERTp. , , In our cohort, 58.9% of IDH‐wildtype GBM showed mutations in TERTp, which is substantially less frequent than in previous reports. However, another report from Japan indicated that 58%‐59% of GBM had mutations in TERTp, , suggesting that TERTp mutations may be less frequent in Japan than in other countries. Previous reports have shown that patients with the TERTp‐wildtype GBM were significantly younger, on average, than those with TERTp‐mutant GBM, , , , contrasting with our results. However, when we analyzed 301 cases conclusively diagnosed with IDH‐wildtype GBM, part of a previously published Japanese large cohort, the average age of patients with TERTp‐wildtype and TERTp‐mutant GBMs was statistically comparable (61.02 vs 63.36 y, P = .065), in line with our findings.

Another study has shown that a portion of IDH‐ and TERTp‐wildtype GBM utilizes distinct genetic mechanisms of telomere maintenance driven by an alternative lengthening of telomerase positive subgroup displaying alterations in ATRX or SMARCAL1, and TERT structural rearrangements. However, our NGS panel could not detect SMARCAL1 alteration and TERT structural variants, which constitutes a limitation of our NGS panel.

In this study, our analyses showed that the TERTp‐wildtype subgroup of IDH‐wildtype GBM had a distinct genomic profile, being significantly enriched for PDGFRA mutations and/or amplification compared with TERTp‐mutant GBM. Moreover, TERTp‐mutant GBMs are enriched for PTEN mutations and/or loss compared with TERTp‐wildtype GBM. Approximately 15% of IDH‐wildtype GBM had amplification of PDGFRA, which is compatible with our findings. Recent reports have indicated that 14.4%‐26% of GBM genomes harbor mutations in EGFR. , However, the frequency of EGFR mutations in GBMs from our cohorts (3.4%) was much lower. This discrepancy may be due to the inclusion of a Japanese cohort.

Using hierarchical molecular classification of IDH‐wildtype GBM, we revealed 3 distinct groups. One major cluster (Group C) was characterized by mutations in TP53 and PDGFRA, amplification of CDK4 and PDGFRA, homozygous deletion of CDKN2A/B, and a lack of TERTp and PTEN mutations. Interestingly, Group C was significantly associated with older age, despite the absence of TERTp mutations. No previous studies have reported the correlations between TERTp‐wildtype status and PDGFRA alteration, thus we hypothesized that Group C in our cohort is a distinct subgroup of IDH‐wildtype GBM. Interestingly, in our cohorts, there was no difference in the average age of patients with TERTp‐wildtype GBM or TERTp‐mutant GBM. However, when we excluded Group C, patients with TERTp‐wildtype GBM were significantly younger than those with TERTp‐mutant GBM, in accordance with previous studies. , , , No distinct difference in survival was observed for patients with TERTp‐wildtype GBM or TERTp‐mutant GBM in our cohorts. However, when we excluded Group C, survival was significantly shorter in patients with TERTp‐mutant GBM than in patients with TERTp‐wildtype GBM, in accordance with previous studies. , , Therefore, the clinical characteristics of Group C might extend to a specific subgroup of Japanese cohorts. In addition, Group C was significantly associated with a higher Ki‐67 score. In previous reports on GBMs, the Ki‐67 score was significantly higher in tumors with CDKN2A homozygous deletions, which have a deleterious effect on cell cycle control. We speculated that the high Ki‐67 score of Group C might correlate with the dysregulation of cell cycles due to CDKN2A/B deletion and CDK4 amplification. Moreover, Group C was significantly associated with poor prognosis. In previous reports, PDGFRA was defined as one of the molecular markers of GBM proneural subtypes. Curiously, IDH‐wildtype proneural tumors had the worst prognosis among all GBM subtypes. However, first‐line bevacizumab plus standard‐of‐care therapy conferred a significant overall survival advantage for patients with proneural IDH‐wildtype tumors. Thus bevacizumab might be more effective in Group C than in other groups, which needs to be validated in a future study.

In summary, we report on the establishment of a glioma‐tailored 48‐gene NGS panel for detecting 1p/19q codeletion and driver mutations as a routine molecular diagnostic tool of gliomas. This study identified alterations of PDGFRA as co‐occurring hallmarks of TERTp‐wildtype GBM, potentially reflecting the unique molecular etiology and clinical features of these tumors. If further validated, our findings may have significant implications for the subclassification of IDH‐wildtype GBM. Such subclassification are likely to provide more precise information to patients and may influence bedside decisions.

DISCLOSURE

The authors declare no conflict of interest.

Figure S1

Click here for additional data file.

Figure S2

Click here for additional data file.

Figure S3

Click here for additional data file.

Figure S4

Click here for additional data file.

Figure S5

Click here for additional data file.

Figure S6

Click here for additional data file.

Figure S7

Click here for additional data file.

Table S1

Click here for additional data file.

Table S2

Click here for additional data file.